1 /*
   2  * Copyright (c) 2016, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2016 SAP SE. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "precompiled.hpp"
  27 #include "c1/c1_Compilation.hpp"
  28 #include "c1/c1_LIRAssembler.hpp"
  29 #include "c1/c1_MacroAssembler.hpp"
  30 #include "c1/c1_Runtime1.hpp"
  31 #include "c1/c1_ValueStack.hpp"
  32 #include "ci/ciArrayKlass.hpp"
  33 #include "ci/ciInstance.hpp"
  34 #include "gc/shared/collectedHeap.hpp"
  35 #include "gc/shared/barrierSet.hpp"
  36 #include "gc/shared/cardTableModRefBS.hpp"
  37 #include "nativeInst_s390.hpp"
  38 #include "oops/objArrayKlass.hpp"
  39 #include "runtime/sharedRuntime.hpp"
  40 #include "vmreg_s390.inline.hpp"
  41 
  42 #define __ _masm->
  43 
  44 #ifndef PRODUCT
  45 #undef __
  46 #define __ (Verbose ? (_masm->block_comment(FILE_AND_LINE),_masm) : _masm)->
  47 #endif
  48 
  49 //------------------------------------------------------------
  50 
  51 bool LIR_Assembler::is_small_constant(LIR_Opr opr) {
  52   // Not used on ZARCH_64
  53   ShouldNotCallThis();
  54   return false;
  55 }
  56 
  57 LIR_Opr LIR_Assembler::receiverOpr() {
  58   return FrameMap::Z_R2_oop_opr;
  59 }
  60 
  61 LIR_Opr LIR_Assembler::osrBufferPointer() {
  62   return FrameMap::Z_R2_opr;
  63 }
  64 
  65 int LIR_Assembler::initial_frame_size_in_bytes() const {
  66   return in_bytes(frame_map()->framesize_in_bytes());
  67 }
  68 
  69 // Inline cache check: done before the frame is built.
  70 // The inline cached class is in Z_inline_cache(Z_R9).
  71 // We fetch the class of the receiver and compare it with the cached class.
  72 // If they do not match we jump to the slow case.
  73 int LIR_Assembler::check_icache() {
  74   Register receiver = receiverOpr()->as_register();
  75   int offset = __ offset();
  76   __ inline_cache_check(receiver, Z_inline_cache);
  77   return offset;
  78 }
  79 
  80 void LIR_Assembler::osr_entry() {
  81   // On-stack-replacement entry sequence (interpreter frame layout described in interpreter_sparc.cpp):
  82   //
  83   //   1. Create a new compiled activation.
  84   //   2. Initialize local variables in the compiled activation. The expression stack must be empty
  85   //      at the osr_bci; it is not initialized.
  86   //   3. Jump to the continuation address in compiled code to resume execution.
  87 
  88   // OSR entry point
  89   offsets()->set_value(CodeOffsets::OSR_Entry, code_offset());
  90   BlockBegin* osr_entry = compilation()->hir()->osr_entry();
  91   ValueStack* entry_state = osr_entry->end()->state();
  92   int number_of_locks = entry_state->locks_size();
  93 
  94   // Create a frame for the compiled activation.
  95   __ build_frame(initial_frame_size_in_bytes(), bang_size_in_bytes());
  96 
  97   // OSR buffer is
  98   //
  99   // locals[nlocals-1..0]
 100   // monitors[number_of_locks-1..0]
 101   //
 102   // Locals is a direct copy of the interpreter frame so in the osr buffer
 103   // the first slot in the local array is the last local from the interpreter
 104   // and the last slot is local[0] (receiver) from the interpreter
 105   //
 106   // Similarly with locks. The first lock slot in the osr buffer is the nth lock
 107   // from the interpreter frame, the nth lock slot in the osr buffer is 0th lock
 108   // in the interpreter frame (the method lock if a sync method)
 109 
 110   // Initialize monitors in the compiled activation.
 111   //   I0: pointer to osr buffer
 112   //
 113   // All other registers are dead at this point and the locals will be
 114   // copied into place by code emitted in the IR.
 115 
 116   Register OSR_buf = osrBufferPointer()->as_register();
 117   { assert(frame::interpreter_frame_monitor_size() == BasicObjectLock::size(), "adjust code below");
 118     int monitor_offset = BytesPerWord * method()->max_locals() +
 119       (2 * BytesPerWord) * (number_of_locks - 1);
 120     // SharedRuntime::OSR_migration_begin() packs BasicObjectLocks in
 121     // the OSR buffer using 2 word entries: first the lock and then
 122     // the oop.
 123     for (int i = 0; i < number_of_locks; i++) {
 124       int slot_offset = monitor_offset - ((i * 2) * BytesPerWord);
 125       // Verify the interpreter's monitor has a non-null object.
 126       __ asm_assert_mem8_isnot_zero(slot_offset + 1*BytesPerWord, OSR_buf, "locked object is NULL", __LINE__);
 127       // Copy the lock field into the compiled activation.
 128       __ z_lg(Z_R1_scratch, slot_offset + 0, OSR_buf);
 129       __ z_stg(Z_R1_scratch, frame_map()->address_for_monitor_lock(i));
 130       __ z_lg(Z_R1_scratch, slot_offset + 1*BytesPerWord, OSR_buf);
 131       __ z_stg(Z_R1_scratch, frame_map()->address_for_monitor_object(i));
 132     }
 133   }
 134 }
 135 
 136 // --------------------------------------------------------------------------------------------
 137 
 138 address LIR_Assembler::emit_call_c(address a) {
 139   __ align_call_far_patchable(__ pc());
 140   address call_addr = __ call_c_opt(a);
 141   if (call_addr == NULL) {
 142     bailout("const section overflow");
 143   }
 144   return call_addr;
 145 }
 146 
 147 int LIR_Assembler::emit_exception_handler() {
 148   // If the last instruction is a call (typically to do a throw which
 149   // is coming at the end after block reordering) the return address
 150   // must still point into the code area in order to avoid assertion
 151   // failures when searching for the corresponding bci. => Add a nop.
 152   // (was bug 5/14/1999 - gri)
 153   __ nop();
 154 
 155   // Generate code for exception handler.
 156   address handler_base = __ start_a_stub(exception_handler_size());
 157   if (handler_base == NULL) {
 158     // Not enough space left for the handler.
 159     bailout("exception handler overflow");
 160     return -1;
 161   }
 162 
 163   int offset = code_offset();
 164 
 165   address a = Runtime1::entry_for (Runtime1::handle_exception_from_callee_id);
 166   address call_addr = emit_call_c(a);
 167   CHECK_BAILOUT_(-1);
 168   __ should_not_reach_here();
 169   guarantee(code_offset() - offset <= exception_handler_size(), "overflow");
 170   __ end_a_stub();
 171 
 172   return offset;
 173 }
 174 
 175 // Emit the code to remove the frame from the stack in the exception
 176 // unwind path.
 177 int LIR_Assembler::emit_unwind_handler() {
 178 #ifndef PRODUCT
 179   if (CommentedAssembly) {
 180     _masm->block_comment("Unwind handler");
 181   }
 182 #endif
 183 
 184   int offset = code_offset();
 185   Register exception_oop_callee_saved = Z_R10; // Z_R10 is callee-saved.
 186   Register Rtmp1                      = Z_R11;
 187   Register Rtmp2                      = Z_R12;
 188 
 189   // Fetch the exception from TLS and clear out exception related thread state.
 190   Address exc_oop_addr = Address(Z_thread, JavaThread::exception_oop_offset());
 191   Address exc_pc_addr  = Address(Z_thread, JavaThread::exception_pc_offset());
 192   __ z_lg(Z_EXC_OOP, exc_oop_addr);
 193   __ clear_mem(exc_oop_addr, sizeof(oop));
 194   __ clear_mem(exc_pc_addr, sizeof(intptr_t));
 195 
 196   __ bind(_unwind_handler_entry);
 197   __ verify_not_null_oop(Z_EXC_OOP);
 198   if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
 199     __ lgr_if_needed(exception_oop_callee_saved, Z_EXC_OOP); // Preserve the exception.
 200   }
 201 
 202   // Preform needed unlocking.
 203   MonitorExitStub* stub = NULL;
 204   if (method()->is_synchronized()) {
 205     // Runtime1::monitorexit_id expects lock address in Z_R1_scratch.
 206     LIR_Opr lock = FrameMap::as_opr(Z_R1_scratch);
 207     monitor_address(0, lock);
 208     stub = new MonitorExitStub(lock, true, 0);
 209     __ unlock_object(Rtmp1, Rtmp2, lock->as_register(), *stub->entry());
 210     __ bind(*stub->continuation());
 211   }
 212 
 213   if (compilation()->env()->dtrace_method_probes()) {
 214     ShouldNotReachHere(); // Not supported.
 215 #if 0
 216     __ mov(rdi, r15_thread);
 217     __ mov_metadata(rsi, method()->constant_encoding());
 218     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit)));
 219 #endif
 220   }
 221 
 222   if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
 223     __ lgr_if_needed(Z_EXC_OOP, exception_oop_callee_saved);  // Restore the exception.
 224   }
 225 
 226   // Remove the activation and dispatch to the unwind handler.
 227   __ pop_frame();
 228   __ z_lg(Z_EXC_PC, _z_abi16(return_pc), Z_SP);
 229 
 230   // Z_EXC_OOP: exception oop
 231   // Z_EXC_PC: exception pc
 232 
 233   // Dispatch to the unwind logic.
 234   __ load_const_optimized(Z_R5, Runtime1::entry_for (Runtime1::unwind_exception_id));
 235   __ z_br(Z_R5);
 236 
 237   // Emit the slow path assembly.
 238   if (stub != NULL) {
 239     stub->emit_code(this);
 240   }
 241 
 242   return offset;
 243 }
 244 
 245 int LIR_Assembler::emit_deopt_handler() {
 246   // If the last instruction is a call (typically to do a throw which
 247   // is coming at the end after block reordering) the return address
 248   // must still point into the code area in order to avoid assertion
 249   // failures when searching for the corresponding bci. => Add a nop.
 250   // (was bug 5/14/1999 - gri)
 251   __ nop();
 252 
 253   // Generate code for exception handler.
 254   address handler_base = __ start_a_stub(deopt_handler_size());
 255   if (handler_base == NULL) {
 256     // Not enough space left for the handler.
 257     bailout("deopt handler overflow");
 258     return -1;
 259   }  int offset = code_offset();
 260   // Size must be constant (see HandlerImpl::emit_deopt_handler).
 261   __ load_const(Z_R1_scratch, SharedRuntime::deopt_blob()->unpack());
 262   __ call(Z_R1_scratch);
 263   guarantee(code_offset() - offset <= deopt_handler_size(), "overflow");
 264   __ end_a_stub();
 265 
 266   return offset;
 267 }
 268 
 269 void LIR_Assembler::jobject2reg(jobject o, Register reg) {
 270   if (o == NULL) {
 271     __ clear_reg(reg, true/*64bit*/, false/*set cc*/); // Must not kill cc set by cmove.
 272   } else {
 273     AddressLiteral a = __ allocate_oop_address(o);
 274     bool success = __ load_oop_from_toc(reg, a, reg);
 275     if (!success) {
 276       bailout("const section overflow");
 277     }
 278   }
 279 }
 280 
 281 void LIR_Assembler::jobject2reg_with_patching(Register reg, CodeEmitInfo *info) {
 282   // Allocate a new index in table to hold the object once it's been patched.
 283   int oop_index = __ oop_recorder()->allocate_oop_index(NULL);
 284   PatchingStub* patch = new PatchingStub(_masm, patching_id(info), oop_index);
 285 
 286   AddressLiteral addrlit((intptr_t)0, oop_Relocation::spec(oop_index));
 287   assert(addrlit.rspec().type() == relocInfo::oop_type, "must be an oop reloc");
 288   // The NULL will be dynamically patched later so the sequence to
 289   // load the address literal must not be optimized.
 290   __ load_const(reg, addrlit);
 291 
 292   patching_epilog(patch, lir_patch_normal, reg, info);
 293 }
 294 
 295 void LIR_Assembler::metadata2reg(Metadata* md, Register reg) {
 296   bool success = __ set_metadata_constant(md, reg);
 297   if (!success) {
 298     bailout("const section overflow");
 299     return;
 300   }
 301 }
 302 
 303 void LIR_Assembler::klass2reg_with_patching(Register reg, CodeEmitInfo *info) {
 304   // Allocate a new index in table to hold the klass once it's been patched.
 305   int index = __ oop_recorder()->allocate_metadata_index(NULL);
 306   PatchingStub* patch = new PatchingStub(_masm, PatchingStub::load_klass_id, index);
 307   AddressLiteral addrlit((intptr_t)0, metadata_Relocation::spec(index));
 308   assert(addrlit.rspec().type() == relocInfo::metadata_type, "must be an metadata reloc");
 309   // The NULL will be dynamically patched later so the sequence to
 310   // load the address literal must not be optimized.
 311   __ load_const(reg, addrlit);
 312 
 313   patching_epilog(patch, lir_patch_normal, reg, info);
 314 }
 315 
 316 void LIR_Assembler::emit_op3(LIR_Op3* op) {
 317   switch (op->code()) {
 318     case lir_idiv:
 319     case lir_irem:
 320       arithmetic_idiv(op->code(),
 321                       op->in_opr1(),
 322                       op->in_opr2(),
 323                       op->in_opr3(),
 324                       op->result_opr(),
 325                       op->info());
 326       break;
 327     case lir_fmad: {
 328       const FloatRegister opr1 = op->in_opr1()->as_double_reg(),
 329                           opr2 = op->in_opr2()->as_double_reg(),
 330                           opr3 = op->in_opr3()->as_double_reg(),
 331                           res  = op->result_opr()->as_double_reg();
 332       __ z_madbr(opr3, opr1, opr2);
 333       if (res != opr3) { __ z_ldr(res, opr3); }
 334     } break;
 335     case lir_fmaf: {
 336       const FloatRegister opr1 = op->in_opr1()->as_float_reg(),
 337                           opr2 = op->in_opr2()->as_float_reg(),
 338                           opr3 = op->in_opr3()->as_float_reg(),
 339                           res  = op->result_opr()->as_float_reg();
 340       __ z_maebr(opr3, opr1, opr2);
 341       if (res != opr3) { __ z_ler(res, opr3); }
 342     } break;
 343     default: ShouldNotReachHere(); break;
 344   }
 345 }
 346 
 347 
 348 void LIR_Assembler::emit_opBranch(LIR_OpBranch* op) {
 349 #ifdef ASSERT
 350   assert(op->block() == NULL || op->block()->label() == op->label(), "wrong label");
 351   if (op->block() != NULL)  { _branch_target_blocks.append(op->block()); }
 352   if (op->ublock() != NULL) { _branch_target_blocks.append(op->ublock()); }
 353 #endif
 354 
 355   if (op->cond() == lir_cond_always) {
 356     if (op->info() != NULL) { add_debug_info_for_branch(op->info()); }
 357     __ branch_optimized(Assembler::bcondAlways, *(op->label()));
 358   } else {
 359     Assembler::branch_condition acond = Assembler::bcondZero;
 360     if (op->code() == lir_cond_float_branch) {
 361       assert(op->ublock() != NULL, "must have unordered successor");
 362       __ branch_optimized(Assembler::bcondNotOrdered, *(op->ublock()->label()));
 363     }
 364     switch (op->cond()) {
 365       case lir_cond_equal:        acond = Assembler::bcondEqual;     break;
 366       case lir_cond_notEqual:     acond = Assembler::bcondNotEqual;  break;
 367       case lir_cond_less:         acond = Assembler::bcondLow;       break;
 368       case lir_cond_lessEqual:    acond = Assembler::bcondNotHigh;   break;
 369       case lir_cond_greaterEqual: acond = Assembler::bcondNotLow;    break;
 370       case lir_cond_greater:      acond = Assembler::bcondHigh;      break;
 371       case lir_cond_belowEqual:   acond = Assembler::bcondNotHigh;   break;
 372       case lir_cond_aboveEqual:   acond = Assembler::bcondNotLow;    break;
 373       default:                         ShouldNotReachHere();
 374     }
 375     __ branch_optimized(acond,*(op->label()));
 376   }
 377 }
 378 
 379 
 380 void LIR_Assembler::emit_opConvert(LIR_OpConvert* op) {
 381   LIR_Opr src  = op->in_opr();
 382   LIR_Opr dest = op->result_opr();
 383 
 384   switch (op->bytecode()) {
 385     case Bytecodes::_i2l:
 386       __ move_reg_if_needed(dest->as_register_lo(), T_LONG, src->as_register(), T_INT);
 387       break;
 388 
 389     case Bytecodes::_l2i:
 390       __ move_reg_if_needed(dest->as_register(), T_INT, src->as_register_lo(), T_LONG);
 391       break;
 392 
 393     case Bytecodes::_i2b:
 394       __ move_reg_if_needed(dest->as_register(), T_BYTE, src->as_register(), T_INT);
 395       break;
 396 
 397     case Bytecodes::_i2c:
 398       __ move_reg_if_needed(dest->as_register(), T_CHAR, src->as_register(), T_INT);
 399       break;
 400 
 401     case Bytecodes::_i2s:
 402       __ move_reg_if_needed(dest->as_register(), T_SHORT, src->as_register(), T_INT);
 403       break;
 404 
 405     case Bytecodes::_f2d:
 406       assert(dest->is_double_fpu(), "check");
 407       __ move_freg_if_needed(dest->as_double_reg(), T_DOUBLE, src->as_float_reg(), T_FLOAT);
 408       break;
 409 
 410     case Bytecodes::_d2f:
 411       assert(dest->is_single_fpu(), "check");
 412       __ move_freg_if_needed(dest->as_float_reg(), T_FLOAT, src->as_double_reg(), T_DOUBLE);
 413       break;
 414 
 415     case Bytecodes::_i2f:
 416       __ z_cefbr(dest->as_float_reg(), src->as_register());
 417       break;
 418 
 419     case Bytecodes::_i2d:
 420       __ z_cdfbr(dest->as_double_reg(), src->as_register());
 421       break;
 422 
 423     case Bytecodes::_l2f:
 424       __ z_cegbr(dest->as_float_reg(), src->as_register_lo());
 425       break;
 426     case Bytecodes::_l2d:
 427       __ z_cdgbr(dest->as_double_reg(), src->as_register_lo());
 428       break;
 429 
 430     case Bytecodes::_f2i:
 431     case Bytecodes::_f2l: {
 432       Label done;
 433       FloatRegister Rsrc = src->as_float_reg();
 434       Register Rdst = (op->bytecode() == Bytecodes::_f2i ? dest->as_register() : dest->as_register_lo());
 435       __ clear_reg(Rdst, true, false);
 436       __ z_cebr(Rsrc, Rsrc);
 437       __ z_brno(done); // NaN -> 0
 438       if (op->bytecode() == Bytecodes::_f2i) {
 439         __ z_cfebr(Rdst, Rsrc, Assembler::to_zero);
 440       } else { // op->bytecode() == Bytecodes::_f2l
 441         __ z_cgebr(Rdst, Rsrc, Assembler::to_zero);
 442       }
 443       __ bind(done);
 444     }
 445     break;
 446 
 447     case Bytecodes::_d2i:
 448     case Bytecodes::_d2l: {
 449       Label done;
 450       FloatRegister Rsrc = src->as_double_reg();
 451       Register Rdst = (op->bytecode() == Bytecodes::_d2i ? dest->as_register() : dest->as_register_lo());
 452       __ clear_reg(Rdst, true, false);  // Don't set CC.
 453       __ z_cdbr(Rsrc, Rsrc);
 454       __ z_brno(done); // NaN -> 0
 455       if (op->bytecode() == Bytecodes::_d2i) {
 456         __ z_cfdbr(Rdst, Rsrc, Assembler::to_zero);
 457       } else { // Bytecodes::_d2l
 458         __ z_cgdbr(Rdst, Rsrc, Assembler::to_zero);
 459       }
 460       __ bind(done);
 461     }
 462     break;
 463 
 464     default: ShouldNotReachHere();
 465   }
 466 }
 467 
 468 void LIR_Assembler::align_call(LIR_Code code) {
 469   // End of call instruction must be 4 byte aligned.
 470   int offset = __ offset();
 471   switch (code) {
 472     case lir_icvirtual_call:
 473       offset += MacroAssembler::load_const_from_toc_size();
 474       // no break
 475     case lir_static_call:
 476     case lir_optvirtual_call:
 477     case lir_dynamic_call:
 478       offset += NativeCall::call_far_pcrelative_displacement_offset;
 479       break;
 480     case lir_virtual_call:   // currently, sparc-specific for niagara
 481     default: ShouldNotReachHere();
 482   }
 483   if ((offset & (NativeCall::call_far_pcrelative_displacement_alignment-1)) != 0) {
 484     __ nop();
 485   }
 486 }
 487 
 488 void LIR_Assembler::call(LIR_OpJavaCall* op, relocInfo::relocType rtype) {
 489   assert((__ offset() + NativeCall::call_far_pcrelative_displacement_offset) % NativeCall::call_far_pcrelative_displacement_alignment == 0,
 490          "must be aligned (offset=%d)", __ offset());
 491   assert(rtype == relocInfo::none ||
 492          rtype == relocInfo::opt_virtual_call_type ||
 493          rtype == relocInfo::static_call_type, "unexpected rtype");
 494   // Prepend each BRASL with a nop.
 495   __ relocate(rtype);
 496   __ z_nop();
 497   __ z_brasl(Z_R14, op->addr());
 498   add_call_info(code_offset(), op->info());
 499 }
 500 
 501 void LIR_Assembler::ic_call(LIR_OpJavaCall* op) {
 502   address virtual_call_oop_addr = NULL;
 503   AddressLiteral empty_ic((address) Universe::non_oop_word());
 504   virtual_call_oop_addr = __ pc();
 505   bool success = __ load_const_from_toc(Z_inline_cache, empty_ic);
 506   if (!success) {
 507     bailout("const section overflow");
 508     return;
 509   }
 510 
 511   // CALL to fixup routine. Fixup routine uses ScopeDesc info
 512   // to determine who we intended to call.
 513   __ relocate(virtual_call_Relocation::spec(virtual_call_oop_addr));
 514   call(op, relocInfo::none);
 515 }
 516 
 517 // not supported
 518 void LIR_Assembler::vtable_call(LIR_OpJavaCall* op) {
 519   ShouldNotReachHere();
 520 }
 521 
 522 void LIR_Assembler::move_regs(Register from_reg, Register to_reg) {
 523   if (from_reg != to_reg) __ z_lgr(to_reg, from_reg);
 524 }
 525 
 526 void LIR_Assembler::const2stack(LIR_Opr src, LIR_Opr dest) {
 527   assert(src->is_constant(), "should not call otherwise");
 528   assert(dest->is_stack(), "should not call otherwise");
 529   LIR_Const* c = src->as_constant_ptr();
 530 
 531   unsigned int lmem = 0;
 532   unsigned int lcon = 0;
 533   int64_t cbits = 0;
 534   Address dest_addr;
 535   switch (c->type()) {
 536     case T_INT:  // fall through
 537     case T_FLOAT:
 538       dest_addr = frame_map()->address_for_slot(dest->single_stack_ix());
 539       lmem = 4; lcon = 4; cbits = c->as_jint_bits();
 540       break;
 541 
 542     case T_ADDRESS:
 543       dest_addr = frame_map()->address_for_slot(dest->single_stack_ix());
 544       lmem = 8; lcon = 4; cbits = c->as_jint_bits();
 545       break;
 546 
 547     case T_OBJECT:
 548       dest_addr = frame_map()->address_for_slot(dest->single_stack_ix());
 549       if (c->as_jobject() == NULL) {
 550         __ store_const(dest_addr, (int64_t)NULL_WORD, 8, 8);
 551       } else {
 552         jobject2reg(c->as_jobject(), Z_R1_scratch);
 553         __ reg2mem_opt(Z_R1_scratch, dest_addr, true);
 554       }
 555       return;
 556 
 557     case T_LONG:  // fall through
 558     case T_DOUBLE:
 559       dest_addr = frame_map()->address_for_slot(dest->double_stack_ix());
 560       lmem = 8; lcon = 8; cbits = (int64_t)(c->as_jlong_bits());
 561       break;
 562 
 563     default:
 564       ShouldNotReachHere();
 565   }
 566 
 567   __ store_const(dest_addr, cbits, lmem, lcon);
 568 }
 569 
 570 void LIR_Assembler::const2mem(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info, bool wide) {
 571   assert(src->is_constant(), "should not call otherwise");
 572   assert(dest->is_address(), "should not call otherwise");
 573   // See special case in LIRGenerator::do_StoreIndexed.
 574   // T_BYTE: Special case for card mark store.
 575   assert(type == T_BYTE || !dest->as_address_ptr()->index()->is_valid(), "not supported");
 576   LIR_Const* c = src->as_constant_ptr();
 577   Address addr = as_Address(dest->as_address_ptr());
 578 
 579   int store_offset = -1;
 580   unsigned int lmem = 0;
 581   unsigned int lcon = 0;
 582   int64_t cbits = 0;
 583   switch (type) {
 584     case T_INT:    // fall through
 585     case T_FLOAT:
 586       lmem = 4; lcon = 4; cbits = c->as_jint_bits();
 587       break;
 588 
 589     case T_ADDRESS:
 590       lmem = 8; lcon = 4; cbits = c->as_jint_bits();
 591       break;
 592 
 593     case T_OBJECT:  // fall through
 594     case T_ARRAY:
 595       if (c->as_jobject() == NULL) {
 596         if (UseCompressedOops && !wide) {
 597           store_offset = __ store_const(addr, (int32_t)NULL_WORD, 4, 4);
 598         } else {
 599           store_offset = __ store_const(addr, (int64_t)NULL_WORD, 8, 8);
 600         }
 601       } else {
 602         jobject2reg(c->as_jobject(), Z_R1_scratch);
 603         if (UseCompressedOops && !wide) {
 604           __ encode_heap_oop(Z_R1_scratch);
 605           store_offset = __ reg2mem_opt(Z_R1_scratch, addr, false);
 606         } else {
 607           store_offset = __ reg2mem_opt(Z_R1_scratch, addr, true);
 608         }
 609       }
 610       assert(store_offset >= 0, "check");
 611       break;
 612 
 613     case T_LONG:    // fall through
 614     case T_DOUBLE:
 615       lmem = 8; lcon = 8; cbits = (int64_t)(c->as_jlong_bits());
 616       break;
 617 
 618     case T_BOOLEAN: // fall through
 619     case T_BYTE:
 620       lmem = 1; lcon = 1; cbits = (int8_t)(c->as_jint());
 621       break;
 622 
 623     case T_CHAR:    // fall through
 624     case T_SHORT:
 625       lmem = 2; lcon = 2; cbits = (int16_t)(c->as_jint());
 626       break;
 627 
 628     default:
 629       ShouldNotReachHere();
 630   };
 631 
 632   // Index register is normally not supported, but for
 633   // LIRGenerator::CardTableModRef_post_barrier we make an exception.
 634   if (type == T_BYTE && dest->as_address_ptr()->index()->is_valid()) {
 635     __ load_const_optimized(Z_R0_scratch, (int8_t)(c->as_jint()));
 636     store_offset = __ offset();
 637     if (Immediate::is_uimm12(addr.disp())) {
 638       __ z_stc(Z_R0_scratch, addr);
 639     } else {
 640       __ z_stcy(Z_R0_scratch, addr);
 641     }
 642   }
 643 
 644   if (store_offset == -1) {
 645     store_offset = __ store_const(addr, cbits, lmem, lcon);
 646     assert(store_offset >= 0, "check");
 647   }
 648 
 649   if (info != NULL) {
 650     add_debug_info_for_null_check(store_offset, info);
 651   }
 652 }
 653 
 654 void LIR_Assembler::const2reg(LIR_Opr src, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
 655   assert(src->is_constant(), "should not call otherwise");
 656   assert(dest->is_register(), "should not call otherwise");
 657   LIR_Const* c = src->as_constant_ptr();
 658 
 659   switch (c->type()) {
 660     case T_INT: {
 661       assert(patch_code == lir_patch_none, "no patching handled here");
 662       __ load_const_optimized(dest->as_register(), c->as_jint());
 663       break;
 664     }
 665 
 666     case T_ADDRESS: {
 667       assert(patch_code == lir_patch_none, "no patching handled here");
 668       __ load_const_optimized(dest->as_register(), c->as_jint());
 669       break;
 670     }
 671 
 672     case T_LONG: {
 673       assert(patch_code == lir_patch_none, "no patching handled here");
 674       __ load_const_optimized(dest->as_register_lo(), (intptr_t)c->as_jlong());
 675       break;
 676     }
 677 
 678     case T_OBJECT: {
 679       if (patch_code != lir_patch_none) {
 680         jobject2reg_with_patching(dest->as_register(), info);
 681       } else {
 682         jobject2reg(c->as_jobject(), dest->as_register());
 683       }
 684       break;
 685     }
 686 
 687     case T_METADATA: {
 688       if (patch_code != lir_patch_none) {
 689         klass2reg_with_patching(dest->as_register(), info);
 690       } else {
 691         metadata2reg(c->as_metadata(), dest->as_register());
 692       }
 693       break;
 694     }
 695 
 696     case T_FLOAT: {
 697       Register toc_reg = Z_R1_scratch;
 698       __ load_toc(toc_reg);
 699       address const_addr = __ float_constant(c->as_jfloat());
 700       if (const_addr == NULL) {
 701         bailout("const section overflow");
 702         break;
 703       }
 704       int displ = const_addr - _masm->code()->consts()->start();
 705       if (dest->is_single_fpu()) {
 706         __ z_ley(dest->as_float_reg(), displ, toc_reg);
 707       } else {
 708         assert(dest->is_single_cpu(), "Must be a cpu register.");
 709         __ z_ly(dest->as_register(), displ, toc_reg);
 710       }
 711     }
 712     break;
 713 
 714     case T_DOUBLE: {
 715       Register toc_reg = Z_R1_scratch;
 716       __ load_toc(toc_reg);
 717       address const_addr = __ double_constant(c->as_jdouble());
 718       if (const_addr == NULL) {
 719         bailout("const section overflow");
 720         break;
 721       }
 722       int displ = const_addr - _masm->code()->consts()->start();
 723       if (dest->is_double_fpu()) {
 724         __ z_ldy(dest->as_double_reg(), displ, toc_reg);
 725       } else {
 726         assert(dest->is_double_cpu(), "Must be a long register.");
 727         __ z_lg(dest->as_register_lo(), displ, toc_reg);
 728       }
 729     }
 730     break;
 731 
 732     default:
 733       ShouldNotReachHere();
 734   }
 735 }
 736 
 737 Address LIR_Assembler::as_Address(LIR_Address* addr) {
 738   if (addr->base()->is_illegal()) {
 739     Unimplemented();
 740   }
 741 
 742   Register base = addr->base()->as_pointer_register();
 743 
 744   if (addr->index()->is_illegal()) {
 745     return Address(base, addr->disp());
 746   } else if (addr->index()->is_cpu_register()) {
 747     Register index = addr->index()->as_pointer_register();
 748     return Address(base, index, addr->disp());
 749   } else if (addr->index()->is_constant()) {
 750     intptr_t addr_offset = addr->index()->as_constant_ptr()->as_jint() + addr->disp();
 751     return Address(base, addr_offset);
 752   } else {
 753     ShouldNotReachHere();
 754     return Address();
 755   }
 756 }
 757 
 758 void LIR_Assembler::stack2stack(LIR_Opr src, LIR_Opr dest, BasicType type) {
 759   switch (type) {
 760     case T_INT:
 761     case T_FLOAT: {
 762       Register tmp = Z_R1_scratch;
 763       Address from = frame_map()->address_for_slot(src->single_stack_ix());
 764       Address to   = frame_map()->address_for_slot(dest->single_stack_ix());
 765       __ mem2reg_opt(tmp, from, false);
 766       __ reg2mem_opt(tmp, to, false);
 767       break;
 768     }
 769     case T_ADDRESS:
 770     case T_OBJECT: {
 771       Register tmp = Z_R1_scratch;
 772       Address from = frame_map()->address_for_slot(src->single_stack_ix());
 773       Address to   = frame_map()->address_for_slot(dest->single_stack_ix());
 774       __ mem2reg_opt(tmp, from, true);
 775       __ reg2mem_opt(tmp, to, true);
 776       break;
 777     }
 778     case T_LONG:
 779     case T_DOUBLE: {
 780       Register tmp = Z_R1_scratch;
 781       Address from = frame_map()->address_for_double_slot(src->double_stack_ix());
 782       Address to   = frame_map()->address_for_double_slot(dest->double_stack_ix());
 783       __ mem2reg_opt(tmp, from, true);
 784       __ reg2mem_opt(tmp, to, true);
 785       break;
 786     }
 787 
 788     default:
 789       ShouldNotReachHere();
 790   }
 791 }
 792 
 793 // 4-byte accesses only! Don't use it to access 8 bytes!
 794 Address LIR_Assembler::as_Address_hi(LIR_Address* addr) {
 795   ShouldNotCallThis();
 796   return 0; // unused
 797 }
 798 
 799 // 4-byte accesses only! Don't use it to access 8 bytes!
 800 Address LIR_Assembler::as_Address_lo(LIR_Address* addr) {
 801   ShouldNotCallThis();
 802   return 0; // unused
 803 }
 804 
 805 void LIR_Assembler::mem2reg(LIR_Opr src_opr, LIR_Opr dest, BasicType type, LIR_PatchCode patch_code,
 806                             CodeEmitInfo* info, bool wide, bool unaligned) {
 807 
 808   assert(type != T_METADATA, "load of metadata ptr not supported");
 809   LIR_Address* addr = src_opr->as_address_ptr();
 810   LIR_Opr to_reg = dest;
 811 
 812   Register src = addr->base()->as_pointer_register();
 813   Register disp_reg = Z_R0;
 814   int disp_value = addr->disp();
 815   bool needs_patching = (patch_code != lir_patch_none);
 816 
 817   if (addr->base()->type() == T_OBJECT) {
 818     __ verify_oop(src);
 819   }
 820 
 821   PatchingStub* patch = NULL;
 822   if (needs_patching) {
 823     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 824     assert(!to_reg->is_double_cpu() ||
 825            patch_code == lir_patch_none ||
 826            patch_code == lir_patch_normal, "patching doesn't match register");
 827   }
 828 
 829   if (addr->index()->is_illegal()) {
 830     if (!Immediate::is_simm20(disp_value)) {
 831       if (needs_patching) {
 832         __ load_const(Z_R1_scratch, (intptr_t)0);
 833       } else {
 834         __ load_const_optimized(Z_R1_scratch, disp_value);
 835       }
 836       disp_reg = Z_R1_scratch;
 837       disp_value = 0;
 838     }
 839   } else {
 840     if (!Immediate::is_simm20(disp_value)) {
 841       __ load_const_optimized(Z_R1_scratch, disp_value);
 842       __ z_la(Z_R1_scratch, 0, Z_R1_scratch, addr->index()->as_register());
 843       disp_reg = Z_R1_scratch;
 844       disp_value = 0;
 845     }
 846     disp_reg = addr->index()->as_pointer_register();
 847   }
 848 
 849   // Remember the offset of the load. The patching_epilog must be done
 850   // before the call to add_debug_info, otherwise the PcDescs don't get
 851   // entered in increasing order.
 852   int offset = code_offset();
 853 
 854   assert(disp_reg != Z_R0 || Immediate::is_simm20(disp_value), "should have set this up");
 855 
 856   bool short_disp = Immediate::is_uimm12(disp_value);
 857 
 858   switch (type) {
 859     case T_BOOLEAN: // fall through
 860     case T_BYTE  :  __ z_lb(dest->as_register(),   disp_value, disp_reg, src); break;
 861     case T_CHAR  :  __ z_llgh(dest->as_register(), disp_value, disp_reg, src); break;
 862     case T_SHORT :
 863       if (short_disp) {
 864                     __ z_lh(dest->as_register(),   disp_value, disp_reg, src);
 865       } else {
 866                     __ z_lhy(dest->as_register(),  disp_value, disp_reg, src);
 867       }
 868       break;
 869     case T_INT   :
 870       if (short_disp) {
 871                     __ z_l(dest->as_register(),    disp_value, disp_reg, src);
 872       } else {
 873                     __ z_ly(dest->as_register(),   disp_value, disp_reg, src);
 874       }
 875       break;
 876     case T_ADDRESS:
 877       if (UseCompressedClassPointers && addr->disp() == oopDesc::klass_offset_in_bytes()) {
 878         __ z_llgf(dest->as_register(), disp_value, disp_reg, src);
 879         __ decode_klass_not_null(dest->as_register());
 880       } else {
 881         __ z_lg(dest->as_register(), disp_value, disp_reg, src);
 882       }
 883       break;
 884     case T_ARRAY : // fall through
 885     case T_OBJECT:
 886     {
 887       if (UseCompressedOops && !wide) {
 888         __ z_llgf(dest->as_register(), disp_value, disp_reg, src);
 889         __ oop_decoder(dest->as_register(), dest->as_register(), true);
 890       } else {
 891         __ z_lg(dest->as_register(), disp_value, disp_reg, src);
 892       }
 893       break;
 894     }
 895     case T_FLOAT:
 896       if (short_disp) {
 897                     __ z_le(dest->as_float_reg(),  disp_value, disp_reg, src);
 898       } else {
 899                     __ z_ley(dest->as_float_reg(), disp_value, disp_reg, src);
 900       }
 901       break;
 902     case T_DOUBLE:
 903       if (short_disp) {
 904                     __ z_ld(dest->as_double_reg(),  disp_value, disp_reg, src);
 905       } else {
 906                     __ z_ldy(dest->as_double_reg(), disp_value, disp_reg, src);
 907       }
 908       break;
 909     case T_LONG  :  __ z_lg(dest->as_register_lo(), disp_value, disp_reg, src); break;
 910     default      : ShouldNotReachHere();
 911   }
 912   if (type == T_ARRAY || type == T_OBJECT) {
 913     __ verify_oop(dest->as_register());
 914   }
 915 
 916   if (patch != NULL) {
 917     patching_epilog(patch, patch_code, src, info);
 918   }
 919   if (info != NULL) add_debug_info_for_null_check(offset, info);
 920 }
 921 
 922 void LIR_Assembler::stack2reg(LIR_Opr src, LIR_Opr dest, BasicType type) {
 923   assert(src->is_stack(), "should not call otherwise");
 924   assert(dest->is_register(), "should not call otherwise");
 925 
 926   if (dest->is_single_cpu()) {
 927     if (type == T_ARRAY || type == T_OBJECT) {
 928       __ mem2reg_opt(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()), true);
 929       __ verify_oop(dest->as_register());
 930     } else if (type == T_METADATA) {
 931       __ mem2reg_opt(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()), true);
 932     } else {
 933       __ mem2reg_opt(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()), false);
 934     }
 935   } else if (dest->is_double_cpu()) {
 936     Address src_addr_LO = frame_map()->address_for_slot(src->double_stack_ix());
 937     __ mem2reg_opt(dest->as_register_lo(), src_addr_LO, true);
 938   } else if (dest->is_single_fpu()) {
 939     Address src_addr = frame_map()->address_for_slot(src->single_stack_ix());
 940     __ mem2freg_opt(dest->as_float_reg(), src_addr, false);
 941   } else if (dest->is_double_fpu()) {
 942     Address src_addr = frame_map()->address_for_slot(src->double_stack_ix());
 943     __ mem2freg_opt(dest->as_double_reg(), src_addr, true);
 944   } else {
 945     ShouldNotReachHere();
 946   }
 947 }
 948 
 949 void LIR_Assembler::reg2stack(LIR_Opr src, LIR_Opr dest, BasicType type, bool pop_fpu_stack) {
 950   assert(src->is_register(), "should not call otherwise");
 951   assert(dest->is_stack(), "should not call otherwise");
 952 
 953   if (src->is_single_cpu()) {
 954     const Address dst = frame_map()->address_for_slot(dest->single_stack_ix());
 955     if (type == T_OBJECT || type == T_ARRAY) {
 956       __ verify_oop(src->as_register());
 957       __ reg2mem_opt(src->as_register(), dst, true);
 958     } else if (type == T_METADATA) {
 959       __ reg2mem_opt(src->as_register(), dst, true);
 960     } else {
 961       __ reg2mem_opt(src->as_register(), dst, false);
 962     }
 963   } else if (src->is_double_cpu()) {
 964     Address dstLO = frame_map()->address_for_slot(dest->double_stack_ix());
 965     __ reg2mem_opt(src->as_register_lo(), dstLO, true);
 966   } else if (src->is_single_fpu()) {
 967     Address dst_addr = frame_map()->address_for_slot(dest->single_stack_ix());
 968     __ freg2mem_opt(src->as_float_reg(), dst_addr, false);
 969   } else if (src->is_double_fpu()) {
 970     Address dst_addr = frame_map()->address_for_slot(dest->double_stack_ix());
 971     __ freg2mem_opt(src->as_double_reg(), dst_addr, true);
 972   } else {
 973     ShouldNotReachHere();
 974   }
 975 }
 976 
 977 void LIR_Assembler::reg2reg(LIR_Opr from_reg, LIR_Opr to_reg) {
 978   if (from_reg->is_float_kind() && to_reg->is_float_kind()) {
 979     if (from_reg->is_double_fpu()) {
 980       // double to double moves
 981       assert(to_reg->is_double_fpu(), "should match");
 982       __ z_ldr(to_reg->as_double_reg(), from_reg->as_double_reg());
 983     } else {
 984       // float to float moves
 985       assert(to_reg->is_single_fpu(), "should match");
 986       __ z_ler(to_reg->as_float_reg(), from_reg->as_float_reg());
 987     }
 988   } else if (!from_reg->is_float_kind() && !to_reg->is_float_kind()) {
 989     if (from_reg->is_double_cpu()) {
 990       __ z_lgr(to_reg->as_pointer_register(), from_reg->as_pointer_register());
 991     } else if (to_reg->is_double_cpu()) {
 992       // int to int moves
 993       __ z_lgr(to_reg->as_register_lo(), from_reg->as_register());
 994     } else {
 995       // int to int moves
 996       __ z_lgr(to_reg->as_register(), from_reg->as_register());
 997     }
 998   } else {
 999     ShouldNotReachHere();
1000   }
1001   if (to_reg->type() == T_OBJECT || to_reg->type() == T_ARRAY) {
1002     __ verify_oop(to_reg->as_register());
1003   }
1004 }
1005 
1006 void LIR_Assembler::reg2mem(LIR_Opr from, LIR_Opr dest_opr, BasicType type,
1007                             LIR_PatchCode patch_code, CodeEmitInfo* info, bool pop_fpu_stack,
1008                             bool wide, bool unaligned) {
1009   assert(type != T_METADATA, "store of metadata ptr not supported");
1010   LIR_Address* addr = dest_opr->as_address_ptr();
1011 
1012   Register dest = addr->base()->as_pointer_register();
1013   Register disp_reg = Z_R0;
1014   int disp_value = addr->disp();
1015   bool needs_patching = (patch_code != lir_patch_none);
1016 
1017   if (addr->base()->is_oop_register()) {
1018     __ verify_oop(dest);
1019   }
1020 
1021   PatchingStub* patch = NULL;
1022   if (needs_patching) {
1023     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1024     assert(!from->is_double_cpu() ||
1025            patch_code == lir_patch_none ||
1026            patch_code == lir_patch_normal, "patching doesn't match register");
1027   }
1028 
1029   assert(!needs_patching || (!Immediate::is_simm20(disp_value) && addr->index()->is_illegal()), "assumption");
1030   if (addr->index()->is_illegal()) {
1031     if (!Immediate::is_simm20(disp_value)) {
1032       if (needs_patching) {
1033         __ load_const(Z_R1_scratch, (intptr_t)0);
1034       } else {
1035         __ load_const_optimized(Z_R1_scratch, disp_value);
1036       }
1037       disp_reg = Z_R1_scratch;
1038       disp_value = 0;
1039     }
1040   } else {
1041     if (!Immediate::is_simm20(disp_value)) {
1042       __ load_const_optimized(Z_R1_scratch, disp_value);
1043       __ z_la(Z_R1_scratch, 0, Z_R1_scratch, addr->index()->as_register());
1044       disp_reg = Z_R1_scratch;
1045       disp_value = 0;
1046     }
1047     disp_reg = addr->index()->as_pointer_register();
1048   }
1049 
1050   assert(disp_reg != Z_R0 || Immediate::is_simm20(disp_value), "should have set this up");
1051 
1052   if (type == T_ARRAY || type == T_OBJECT) {
1053     __ verify_oop(from->as_register());
1054   }
1055 
1056   bool short_disp = Immediate::is_uimm12(disp_value);
1057 
1058   // Remember the offset of the store. The patching_epilog must be done
1059   // before the call to add_debug_info_for_null_check, otherwise the PcDescs don't get
1060   // entered in increasing order.
1061   int offset = code_offset();
1062   switch (type) {
1063     case T_BOOLEAN: // fall through
1064     case T_BYTE  :
1065       if (short_disp) {
1066                     __ z_stc(from->as_register(),  disp_value, disp_reg, dest);
1067       } else {
1068                     __ z_stcy(from->as_register(), disp_value, disp_reg, dest);
1069       }
1070       break;
1071     case T_CHAR  : // fall through
1072     case T_SHORT :
1073       if (short_disp) {
1074                     __ z_sth(from->as_register(),  disp_value, disp_reg, dest);
1075       } else {
1076                     __ z_sthy(from->as_register(), disp_value, disp_reg, dest);
1077       }
1078       break;
1079     case T_INT   :
1080       if (short_disp) {
1081                     __ z_st(from->as_register(),  disp_value, disp_reg, dest);
1082       } else {
1083                     __ z_sty(from->as_register(), disp_value, disp_reg, dest);
1084       }
1085       break;
1086     case T_LONG  :  __ z_stg(from->as_register_lo(), disp_value, disp_reg, dest); break;
1087     case T_ADDRESS: __ z_stg(from->as_register(),    disp_value, disp_reg, dest); break;
1088       break;
1089     case T_ARRAY : // fall through
1090     case T_OBJECT:
1091       {
1092         if (UseCompressedOops && !wide) {
1093           Register compressed_src = Z_R14;
1094           __ oop_encoder(compressed_src, from->as_register(), true, (disp_reg != Z_R1) ? Z_R1 : Z_R0, -1, true);
1095           offset = code_offset();
1096           if (short_disp) {
1097             __ z_st(compressed_src,  disp_value, disp_reg, dest);
1098           } else {
1099             __ z_sty(compressed_src, disp_value, disp_reg, dest);
1100           }
1101         } else {
1102           __ z_stg(from->as_register(), disp_value, disp_reg, dest);
1103         }
1104         break;
1105       }
1106     case T_FLOAT :
1107       if (short_disp) {
1108         __ z_ste(from->as_float_reg(),  disp_value, disp_reg, dest);
1109       } else {
1110         __ z_stey(from->as_float_reg(), disp_value, disp_reg, dest);
1111       }
1112       break;
1113     case T_DOUBLE:
1114       if (short_disp) {
1115         __ z_std(from->as_double_reg(),  disp_value, disp_reg, dest);
1116       } else {
1117         __ z_stdy(from->as_double_reg(), disp_value, disp_reg, dest);
1118       }
1119       break;
1120     default: ShouldNotReachHere();
1121   }
1122 
1123   if (patch != NULL) {
1124     patching_epilog(patch, patch_code, dest, info);
1125   }
1126 
1127   if (info != NULL) add_debug_info_for_null_check(offset, info);
1128 }
1129 
1130 
1131 void LIR_Assembler::return_op(LIR_Opr result) {
1132   assert(result->is_illegal() ||
1133          (result->is_single_cpu() && result->as_register() == Z_R2) ||
1134          (result->is_double_cpu() && result->as_register_lo() == Z_R2) ||
1135          (result->is_single_fpu() && result->as_float_reg() == Z_F0) ||
1136          (result->is_double_fpu() && result->as_double_reg() == Z_F0), "convention");
1137 
1138   AddressLiteral pp(os::get_polling_page());
1139   __ load_const_optimized(Z_R1_scratch, pp);
1140 
1141   // Pop the frame before the safepoint code.
1142   int retPC_offset = initial_frame_size_in_bytes() + _z_abi16(return_pc);
1143   if (Displacement::is_validDisp(retPC_offset)) {
1144     __ z_lg(Z_R14, retPC_offset, Z_SP);
1145     __ add2reg(Z_SP, initial_frame_size_in_bytes());
1146   } else {
1147     __ add2reg(Z_SP, initial_frame_size_in_bytes());
1148     __ restore_return_pc();
1149   }
1150 
1151   if (StackReservedPages > 0 && compilation()->has_reserved_stack_access()) {
1152     __ reserved_stack_check(Z_R14);
1153   }
1154 
1155   // We need to mark the code position where the load from the safepoint
1156   // polling page was emitted as relocInfo::poll_return_type here.
1157   __ relocate(relocInfo::poll_return_type);
1158   __ load_from_polling_page(Z_R1_scratch);
1159 
1160   __ z_br(Z_R14); // Return to caller.
1161 }
1162 
1163 int LIR_Assembler::safepoint_poll(LIR_Opr tmp, CodeEmitInfo* info) {
1164   AddressLiteral pp(os::get_polling_page());
1165   __ load_const_optimized(tmp->as_register_lo(), pp);
1166   guarantee(info != NULL, "Shouldn't be NULL");
1167   add_debug_info_for_branch(info);
1168   int offset = __ offset();
1169   __ relocate(relocInfo::poll_type);
1170   __ load_from_polling_page(tmp->as_register_lo());
1171   return offset;
1172 }
1173 
1174 void LIR_Assembler::emit_static_call_stub() {
1175 
1176   // Stub is fixed up when the corresponding call is converted from calling
1177   // compiled code to calling interpreted code.
1178 
1179   address call_pc = __ pc();
1180   address stub = __ start_a_stub(call_stub_size());
1181   if (stub == NULL) {
1182     bailout("static call stub overflow");
1183     return;
1184   }
1185 
1186   int start = __ offset();
1187 
1188   __ relocate(static_stub_Relocation::spec(call_pc));
1189 
1190   // See also Matcher::interpreter_method_oop_reg().
1191   AddressLiteral meta = __ allocate_metadata_address(NULL);
1192   bool success = __ load_const_from_toc(Z_method, meta);
1193 
1194   __ set_inst_mark();
1195   AddressLiteral a((address)-1);
1196   success = success && __ load_const_from_toc(Z_R1, a);
1197   if (!success) {
1198     bailout("const section overflow");
1199     return;
1200   }
1201 
1202   __ z_br(Z_R1);
1203   assert(__ offset() - start <= call_stub_size(), "stub too big");
1204   __ end_a_stub(); // Update current stubs pointer and restore insts_end.
1205 }
1206 
1207 void LIR_Assembler::comp_op(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Op2* op) {
1208   bool unsigned_comp = condition == lir_cond_belowEqual || condition == lir_cond_aboveEqual;
1209   if (opr1->is_single_cpu()) {
1210     Register reg1 = opr1->as_register();
1211     if (opr2->is_single_cpu()) {
1212       // cpu register - cpu register
1213       if (opr1->type() == T_OBJECT || opr1->type() == T_ARRAY) {
1214         __ z_clgr(reg1, opr2->as_register());
1215       } else {
1216         assert(opr2->type() != T_OBJECT && opr2->type() != T_ARRAY, "cmp int, oop?");
1217         if (unsigned_comp) {
1218           __ z_clr(reg1, opr2->as_register());
1219         } else {
1220           __ z_cr(reg1, opr2->as_register());
1221         }
1222       }
1223     } else if (opr2->is_stack()) {
1224       // cpu register - stack
1225       if (opr1->type() == T_OBJECT || opr1->type() == T_ARRAY) {
1226         __ z_cg(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
1227       } else {
1228         if (unsigned_comp) {
1229           __ z_cly(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
1230         } else {
1231           __ z_cy(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
1232         }
1233       }
1234     } else if (opr2->is_constant()) {
1235       // cpu register - constant
1236       LIR_Const* c = opr2->as_constant_ptr();
1237       if (c->type() == T_INT) {
1238         if (unsigned_comp) {
1239           __ z_clfi(reg1, c->as_jint());
1240         } else {
1241           __ z_cfi(reg1, c->as_jint());
1242         }
1243       } else if (c->type() == T_OBJECT || c->type() == T_ARRAY) {
1244         // In 64bit oops are single register.
1245         jobject o = c->as_jobject();
1246         if (o == NULL) {
1247           __ z_ltgr(reg1, reg1);
1248         } else {
1249           jobject2reg(o, Z_R1_scratch);
1250           __ z_cgr(reg1, Z_R1_scratch);
1251         }
1252       } else {
1253         fatal("unexpected type: %s", basictype_to_str(c->type()));
1254       }
1255       // cpu register - address
1256     } else if (opr2->is_address()) {
1257       if (op->info() != NULL) {
1258         add_debug_info_for_null_check_here(op->info());
1259       }
1260       if (unsigned_comp) {
1261         __ z_cly(reg1, as_Address(opr2->as_address_ptr()));
1262       } else {
1263         __ z_cy(reg1, as_Address(opr2->as_address_ptr()));
1264       }
1265     } else {
1266       ShouldNotReachHere();
1267     }
1268 
1269   } else if (opr1->is_double_cpu()) {
1270     assert(!unsigned_comp, "unexpected");
1271     Register xlo = opr1->as_register_lo();
1272     Register xhi = opr1->as_register_hi();
1273     if (opr2->is_double_cpu()) {
1274       __ z_cgr(xlo, opr2->as_register_lo());
1275     } else if (opr2->is_constant()) {
1276       // cpu register - constant 0
1277       assert(opr2->as_jlong() == (jlong)0, "only handles zero");
1278       __ z_ltgr(xlo, xlo);
1279     } else {
1280       ShouldNotReachHere();
1281     }
1282 
1283   } else if (opr1->is_single_fpu()) {
1284     if (opr2->is_single_fpu()) {
1285       __ z_cebr(opr1->as_float_reg(), opr2->as_float_reg());
1286     } else {
1287       // stack slot
1288       Address addr = frame_map()->address_for_slot(opr2->single_stack_ix());
1289       if (Immediate::is_uimm12(addr.disp())) {
1290         __ z_ceb(opr1->as_float_reg(), addr);
1291       } else {
1292         __ z_ley(Z_fscratch_1, addr);
1293         __ z_cebr(opr1->as_float_reg(), Z_fscratch_1);
1294       }
1295     }
1296   } else if (opr1->is_double_fpu()) {
1297     if (opr2->is_double_fpu()) {
1298     __ z_cdbr(opr1->as_double_reg(), opr2->as_double_reg());
1299     } else {
1300       // stack slot
1301       Address addr = frame_map()->address_for_slot(opr2->double_stack_ix());
1302       if (Immediate::is_uimm12(addr.disp())) {
1303         __ z_cdb(opr1->as_double_reg(), addr);
1304       } else {
1305         __ z_ldy(Z_fscratch_1, addr);
1306         __ z_cdbr(opr1->as_double_reg(), Z_fscratch_1);
1307       }
1308     }
1309   } else {
1310     ShouldNotReachHere();
1311   }
1312 }
1313 
1314 void LIR_Assembler::comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst, LIR_Op2* op) {
1315   Label    done;
1316   Register dreg = dst->as_register();
1317 
1318   if (code == lir_cmp_fd2i || code == lir_ucmp_fd2i) {
1319     assert((left->is_single_fpu() && right->is_single_fpu()) ||
1320            (left->is_double_fpu() && right->is_double_fpu()), "unexpected operand types");
1321     bool is_single = left->is_single_fpu();
1322     bool is_unordered_less = (code == lir_ucmp_fd2i);
1323     FloatRegister lreg = is_single ? left->as_float_reg() : left->as_double_reg();
1324     FloatRegister rreg = is_single ? right->as_float_reg() : right->as_double_reg();
1325     if (is_single) {
1326       __ z_cebr(lreg, rreg);
1327     } else {
1328       __ z_cdbr(lreg, rreg);
1329     }
1330     if (VM_Version::has_LoadStoreConditional()) {
1331       Register one       = Z_R0_scratch;
1332       Register minus_one = Z_R1_scratch;
1333       __ z_lghi(minus_one, -1);
1334       __ z_lghi(one,  1);
1335       __ z_lghi(dreg, 0);
1336       __ z_locgr(dreg, one,       is_unordered_less ? Assembler::bcondHigh            : Assembler::bcondHighOrNotOrdered);
1337       __ z_locgr(dreg, minus_one, is_unordered_less ? Assembler::bcondLowOrNotOrdered : Assembler::bcondLow);
1338     } else {
1339       __ clear_reg(dreg, true, false);
1340       __ z_bre(done); // if (left == right) dst = 0
1341 
1342       // if (left > right || ((code ~= cmpg) && (left <> right)) dst := 1
1343       __ z_lhi(dreg, 1);
1344       __ z_brc(is_unordered_less ? Assembler::bcondHigh : Assembler::bcondHighOrNotOrdered, done);
1345 
1346       // if (left < right || ((code ~= cmpl) && (left <> right)) dst := -1
1347       __ z_lhi(dreg, -1);
1348     }
1349   } else {
1350     assert(code == lir_cmp_l2i, "check");
1351     if (VM_Version::has_LoadStoreConditional()) {
1352       Register one       = Z_R0_scratch;
1353       Register minus_one = Z_R1_scratch;
1354       __ z_cgr(left->as_register_lo(), right->as_register_lo());
1355       __ z_lghi(minus_one, -1);
1356       __ z_lghi(one,  1);
1357       __ z_lghi(dreg, 0);
1358       __ z_locgr(dreg, one, Assembler::bcondHigh);
1359       __ z_locgr(dreg, minus_one, Assembler::bcondLow);
1360     } else {
1361       __ z_cgr(left->as_register_lo(), right->as_register_lo());
1362       __ z_lghi(dreg,  0);     // eq value
1363       __ z_bre(done);
1364       __ z_lghi(dreg,  1);     // gt value
1365       __ z_brh(done);
1366       __ z_lghi(dreg, -1);     // lt value
1367     }
1368   }
1369   __ bind(done);
1370 }
1371 
1372 // result = condition ? opr1 : opr2
1373 void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type) {
1374   Assembler::branch_condition acond = Assembler::bcondEqual, ncond = Assembler::bcondNotEqual;
1375   switch (condition) {
1376     case lir_cond_equal:        acond = Assembler::bcondEqual;    ncond = Assembler::bcondNotEqual; break;
1377     case lir_cond_notEqual:     acond = Assembler::bcondNotEqual; ncond = Assembler::bcondEqual;    break;
1378     case lir_cond_less:         acond = Assembler::bcondLow;      ncond = Assembler::bcondNotLow;   break;
1379     case lir_cond_lessEqual:    acond = Assembler::bcondNotHigh;  ncond = Assembler::bcondHigh;     break;
1380     case lir_cond_greaterEqual: acond = Assembler::bcondNotLow;   ncond = Assembler::bcondLow;      break;
1381     case lir_cond_greater:      acond = Assembler::bcondHigh;     ncond = Assembler::bcondNotHigh;  break;
1382     case lir_cond_belowEqual:   acond = Assembler::bcondNotHigh;  ncond = Assembler::bcondHigh;     break;
1383     case lir_cond_aboveEqual:   acond = Assembler::bcondNotLow;   ncond = Assembler::bcondLow;      break;
1384     default:                    ShouldNotReachHere();
1385   }
1386 
1387   if (opr1->is_cpu_register()) {
1388     reg2reg(opr1, result);
1389   } else if (opr1->is_stack()) {
1390     stack2reg(opr1, result, result->type());
1391   } else if (opr1->is_constant()) {
1392     const2reg(opr1, result, lir_patch_none, NULL);
1393   } else {
1394     ShouldNotReachHere();
1395   }
1396 
1397   if (VM_Version::has_LoadStoreConditional() && !opr2->is_constant()) {
1398     // Optimized version that does not require a branch.
1399     if (opr2->is_single_cpu()) {
1400       assert(opr2->cpu_regnr() != result->cpu_regnr(), "opr2 already overwritten by previous move");
1401       __ z_locgr(result->as_register(), opr2->as_register(), ncond);
1402     } else if (opr2->is_double_cpu()) {
1403       assert(opr2->cpu_regnrLo() != result->cpu_regnrLo() && opr2->cpu_regnrLo() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
1404       assert(opr2->cpu_regnrHi() != result->cpu_regnrLo() && opr2->cpu_regnrHi() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
1405       __ z_locgr(result->as_register_lo(), opr2->as_register_lo(), ncond);
1406     } else if (opr2->is_single_stack()) {
1407       __ z_loc(result->as_register(), frame_map()->address_for_slot(opr2->single_stack_ix()), ncond);
1408     } else if (opr2->is_double_stack()) {
1409       __ z_locg(result->as_register_lo(), frame_map()->address_for_slot(opr2->double_stack_ix()), ncond);
1410     } else {
1411       ShouldNotReachHere();
1412     }
1413   } else {
1414     Label skip;
1415     __ z_brc(acond, skip);
1416     if (opr2->is_cpu_register()) {
1417       reg2reg(opr2, result);
1418     } else if (opr2->is_stack()) {
1419       stack2reg(opr2, result, result->type());
1420     } else if (opr2->is_constant()) {
1421       const2reg(opr2, result, lir_patch_none, NULL);
1422     } else {
1423       ShouldNotReachHere();
1424     }
1425     __ bind(skip);
1426   }
1427 }
1428 
1429 void LIR_Assembler::arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest,
1430                              CodeEmitInfo* info, bool pop_fpu_stack) {
1431   assert(info == NULL, "should never be used, idiv/irem and ldiv/lrem not handled by this method");
1432 
1433   if (left->is_single_cpu()) {
1434     assert(left == dest, "left and dest must be equal");
1435     Register lreg = left->as_register();
1436 
1437     if (right->is_single_cpu()) {
1438       // cpu register - cpu register
1439       Register rreg = right->as_register();
1440       switch (code) {
1441         case lir_add: __ z_ar (lreg, rreg); break;
1442         case lir_sub: __ z_sr (lreg, rreg); break;
1443         case lir_mul: __ z_msr(lreg, rreg); break;
1444         default: ShouldNotReachHere();
1445       }
1446 
1447     } else if (right->is_stack()) {
1448       // cpu register - stack
1449       Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
1450       switch (code) {
1451         case lir_add: __ z_ay(lreg, raddr); break;
1452         case lir_sub: __ z_sy(lreg, raddr); break;
1453         default: ShouldNotReachHere();
1454       }
1455 
1456     } else if (right->is_constant()) {
1457       // cpu register - constant
1458       jint c = right->as_constant_ptr()->as_jint();
1459       switch (code) {
1460         case lir_add: __ z_agfi(lreg, c);  break;
1461         case lir_sub: __ z_agfi(lreg, -c); break; // note: -min_jint == min_jint
1462         case lir_mul: __ z_msfi(lreg, c);  break;
1463         default: ShouldNotReachHere();
1464       }
1465 
1466     } else {
1467       ShouldNotReachHere();
1468     }
1469 
1470   } else if (left->is_double_cpu()) {
1471     assert(left == dest, "left and dest must be equal");
1472     Register lreg_lo = left->as_register_lo();
1473     Register lreg_hi = left->as_register_hi();
1474 
1475     if (right->is_double_cpu()) {
1476       // cpu register - cpu register
1477       Register rreg_lo = right->as_register_lo();
1478       Register rreg_hi = right->as_register_hi();
1479       assert_different_registers(lreg_lo, rreg_lo);
1480       switch (code) {
1481         case lir_add:
1482           __ z_agr(lreg_lo, rreg_lo);
1483           break;
1484         case lir_sub:
1485           __ z_sgr(lreg_lo, rreg_lo);
1486           break;
1487         case lir_mul:
1488           __ z_msgr(lreg_lo, rreg_lo);
1489           break;
1490         default:
1491           ShouldNotReachHere();
1492       }
1493 
1494     } else if (right->is_constant()) {
1495       // cpu register - constant
1496       jlong c = right->as_constant_ptr()->as_jlong_bits();
1497       switch (code) {
1498         case lir_add: __ z_agfi(lreg_lo, c); break;
1499         case lir_sub:
1500           if (c != min_jint) {
1501                       __ z_agfi(lreg_lo, -c);
1502           } else {
1503             // -min_jint cannot be represented as simm32 in z_agfi
1504             // min_jint sign extended:      0xffffffff80000000
1505             // -min_jint as 64 bit integer: 0x0000000080000000
1506             // 0x80000000 can be represented as uimm32 in z_algfi
1507             // lreg_lo := lreg_lo + -min_jint == lreg_lo + 0x80000000
1508                       __ z_algfi(lreg_lo, UCONST64(0x80000000));
1509           }
1510           break;
1511         case lir_mul: __ z_msgfi(lreg_lo, c); break;
1512         default:
1513           ShouldNotReachHere();
1514       }
1515 
1516     } else {
1517       ShouldNotReachHere();
1518     }
1519 
1520   } else if (left->is_single_fpu()) {
1521     assert(left == dest, "left and dest must be equal");
1522     FloatRegister lreg = left->as_float_reg();
1523     FloatRegister rreg = right->is_single_fpu() ? right->as_float_reg() : fnoreg;
1524     Address raddr;
1525 
1526     if (rreg == fnoreg) {
1527       assert(right->is_single_stack(), "constants should be loaded into register");
1528       raddr = frame_map()->address_for_slot(right->single_stack_ix());
1529       if (!Immediate::is_uimm12(raddr.disp())) {
1530         __ mem2freg_opt(rreg = Z_fscratch_1, raddr, false);
1531       }
1532     }
1533 
1534     if (rreg != fnoreg) {
1535       switch (code) {
1536         case lir_add: __ z_aebr(lreg, rreg);  break;
1537         case lir_sub: __ z_sebr(lreg, rreg);  break;
1538         case lir_mul_strictfp: // fall through
1539         case lir_mul: __ z_meebr(lreg, rreg); break;
1540         case lir_div_strictfp: // fall through
1541         case lir_div: __ z_debr(lreg, rreg);  break;
1542         default: ShouldNotReachHere();
1543       }
1544     } else {
1545       switch (code) {
1546         case lir_add: __ z_aeb(lreg, raddr);  break;
1547         case lir_sub: __ z_seb(lreg, raddr);  break;
1548         case lir_mul_strictfp: // fall through
1549         case lir_mul: __ z_meeb(lreg, raddr);  break;
1550         case lir_div_strictfp: // fall through
1551         case lir_div: __ z_deb(lreg, raddr);  break;
1552         default: ShouldNotReachHere();
1553       }
1554     }
1555   } else if (left->is_double_fpu()) {
1556     assert(left == dest, "left and dest must be equal");
1557     FloatRegister lreg = left->as_double_reg();
1558     FloatRegister rreg = right->is_double_fpu() ? right->as_double_reg() : fnoreg;
1559     Address raddr;
1560 
1561     if (rreg == fnoreg) {
1562       assert(right->is_double_stack(), "constants should be loaded into register");
1563       raddr = frame_map()->address_for_slot(right->double_stack_ix());
1564       if (!Immediate::is_uimm12(raddr.disp())) {
1565         __ mem2freg_opt(rreg = Z_fscratch_1, raddr, true);
1566       }
1567     }
1568 
1569     if (rreg != fnoreg) {
1570       switch (code) {
1571         case lir_add: __ z_adbr(lreg, rreg); break;
1572         case lir_sub: __ z_sdbr(lreg, rreg); break;
1573         case lir_mul_strictfp: // fall through
1574         case lir_mul: __ z_mdbr(lreg, rreg); break;
1575         case lir_div_strictfp: // fall through
1576         case lir_div: __ z_ddbr(lreg, rreg); break;
1577         default: ShouldNotReachHere();
1578       }
1579     } else {
1580       switch (code) {
1581         case lir_add: __ z_adb(lreg, raddr); break;
1582         case lir_sub: __ z_sdb(lreg, raddr); break;
1583         case lir_mul_strictfp: // fall through
1584         case lir_mul: __ z_mdb(lreg, raddr); break;
1585         case lir_div_strictfp: // fall through
1586         case lir_div: __ z_ddb(lreg, raddr); break;
1587         default: ShouldNotReachHere();
1588       }
1589     }
1590   } else if (left->is_address()) {
1591     assert(left == dest, "left and dest must be equal");
1592     assert(code == lir_add, "unsupported operation");
1593     assert(right->is_constant(), "unsupported operand");
1594     jint c = right->as_constant_ptr()->as_jint();
1595     LIR_Address* lir_addr = left->as_address_ptr();
1596     Address addr = as_Address(lir_addr);
1597     switch (lir_addr->type()) {
1598       case T_INT:
1599         __ add2mem_32(addr, c, Z_R1_scratch);
1600         break;
1601       case T_LONG:
1602         __ add2mem_64(addr, c, Z_R1_scratch);
1603         break;
1604       default:
1605         ShouldNotReachHere();
1606     }
1607   } else {
1608     ShouldNotReachHere();
1609   }
1610 }
1611 
1612 void LIR_Assembler::fpop() {
1613   // do nothing
1614 }
1615 
1616 void LIR_Assembler::intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr thread, LIR_Opr dest, LIR_Op* op) {
1617   switch (code) {
1618     case lir_sqrt: {
1619       assert(!thread->is_valid(), "there is no need for a thread_reg for dsqrt");
1620       FloatRegister src_reg = value->as_double_reg();
1621       FloatRegister dst_reg = dest->as_double_reg();
1622       __ z_sqdbr(dst_reg, src_reg);
1623       break;
1624     }
1625     case lir_abs: {
1626       assert(!thread->is_valid(), "there is no need for a thread_reg for fabs");
1627       FloatRegister src_reg = value->as_double_reg();
1628       FloatRegister dst_reg = dest->as_double_reg();
1629       __ z_lpdbr(dst_reg, src_reg);
1630       break;
1631     }
1632     default: {
1633       ShouldNotReachHere();
1634       break;
1635     }
1636   }
1637 }
1638 
1639 void LIR_Assembler::logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst) {
1640   if (left->is_single_cpu()) {
1641     Register reg = left->as_register();
1642     if (right->is_constant()) {
1643       int val = right->as_constant_ptr()->as_jint();
1644       switch (code) {
1645         case lir_logic_and: __ z_nilf(reg, val); break;
1646         case lir_logic_or:  __ z_oilf(reg, val); break;
1647         case lir_logic_xor: __ z_xilf(reg, val); break;
1648         default: ShouldNotReachHere();
1649       }
1650     } else if (right->is_stack()) {
1651       Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
1652       switch (code) {
1653         case lir_logic_and: __ z_ny(reg, raddr); break;
1654         case lir_logic_or:  __ z_oy(reg, raddr); break;
1655         case lir_logic_xor: __ z_xy(reg, raddr); break;
1656         default: ShouldNotReachHere();
1657       }
1658     } else {
1659       Register rright = right->as_register();
1660       switch (code) {
1661         case lir_logic_and: __ z_nr(reg, rright); break;
1662         case lir_logic_or : __ z_or(reg, rright); break;
1663         case lir_logic_xor: __ z_xr(reg, rright); break;
1664         default: ShouldNotReachHere();
1665       }
1666     }
1667     move_regs(reg, dst->as_register());
1668   } else {
1669     Register l_lo = left->as_register_lo();
1670     if (right->is_constant()) {
1671       __ load_const_optimized(Z_R1_scratch, right->as_constant_ptr()->as_jlong());
1672       switch (code) {
1673         case lir_logic_and:
1674           __ z_ngr(l_lo, Z_R1_scratch);
1675           break;
1676         case lir_logic_or:
1677           __ z_ogr(l_lo, Z_R1_scratch);
1678           break;
1679         case lir_logic_xor:
1680           __ z_xgr(l_lo, Z_R1_scratch);
1681           break;
1682         default: ShouldNotReachHere();
1683       }
1684     } else {
1685       Register r_lo;
1686       if (right->type() == T_OBJECT || right->type() == T_ARRAY) {
1687         r_lo = right->as_register();
1688       } else {
1689         r_lo = right->as_register_lo();
1690       }
1691       switch (code) {
1692         case lir_logic_and:
1693           __ z_ngr(l_lo, r_lo);
1694           break;
1695         case lir_logic_or:
1696           __ z_ogr(l_lo, r_lo);
1697           break;
1698         case lir_logic_xor:
1699           __ z_xgr(l_lo, r_lo);
1700           break;
1701         default: ShouldNotReachHere();
1702       }
1703     }
1704 
1705     Register dst_lo = dst->as_register_lo();
1706 
1707     move_regs(l_lo, dst_lo);
1708   }
1709 }
1710 
1711 // See operand selection in LIRGenerator::do_ArithmeticOp_Int().
1712 void LIR_Assembler::arithmetic_idiv(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr temp, LIR_Opr result, CodeEmitInfo* info) {
1713   if (left->is_double_cpu()) {
1714     // 64 bit integer case
1715     assert(left->is_double_cpu(), "left must be register");
1716     assert(right->is_double_cpu() || is_power_of_2_long(right->as_jlong()),
1717            "right must be register or power of 2 constant");
1718     assert(result->is_double_cpu(), "result must be register");
1719 
1720     Register lreg = left->as_register_lo();
1721     Register dreg = result->as_register_lo();
1722 
1723     if (right->is_constant()) {
1724       // Convert division by a power of two into some shifts and logical operations.
1725       Register treg1 = Z_R0_scratch;
1726       Register treg2 = Z_R1_scratch;
1727       jlong divisor = right->as_jlong();
1728       jlong log_divisor = log2_long(right->as_jlong());
1729 
1730       if (divisor == min_jlong) {
1731         // Min_jlong is special. Result is '0' except for min_jlong/min_jlong = 1.
1732         if (dreg == lreg) {
1733           NearLabel done;
1734           __ load_const_optimized(treg2, min_jlong);
1735           __ z_cgr(lreg, treg2);
1736           __ z_lghi(dreg, 0);           // Preserves condition code.
1737           __ z_brne(done);
1738           __ z_lghi(dreg, 1);           // min_jlong / min_jlong = 1
1739           __ bind(done);
1740         } else {
1741           assert_different_registers(dreg, lreg);
1742           NearLabel done;
1743           __ z_lghi(dreg, 0);
1744           __ compare64_and_branch(lreg, min_jlong, Assembler::bcondNotEqual, done);
1745           __ z_lghi(dreg, 1);
1746           __ bind(done);
1747         }
1748         return;
1749       }
1750       __ move_reg_if_needed(dreg, T_LONG, lreg, T_LONG);
1751       if (divisor == 2) {
1752         __ z_srlg(treg2, dreg, 63);     // dividend < 0 ? 1 : 0
1753       } else {
1754         __ z_srag(treg2, dreg, 63);     // dividend < 0 ? -1 : 0
1755         __ and_imm(treg2, divisor - 1, treg1, true);
1756       }
1757       if (code == lir_idiv) {
1758         __ z_agr(dreg, treg2);
1759         __ z_srag(dreg, dreg, log_divisor);
1760       } else {
1761         assert(code == lir_irem, "check");
1762         __ z_agr(treg2, dreg);
1763         __ and_imm(treg2, ~(divisor - 1), treg1, true);
1764         __ z_sgr(dreg, treg2);
1765       }
1766       return;
1767     }
1768 
1769     // Divisor is not a power of 2 constant.
1770     Register rreg = right->as_register_lo();
1771     Register treg = temp->as_register_lo();
1772     assert(right->is_double_cpu(), "right must be register");
1773     assert(lreg == Z_R11, "see ldivInOpr()");
1774     assert(rreg != lreg, "right register must not be same as left register");
1775     assert((code == lir_idiv && dreg == Z_R11 && treg == Z_R10) ||
1776            (code == lir_irem && dreg == Z_R10 && treg == Z_R11), "see ldivInOpr(), ldivOutOpr(), lremOutOpr()");
1777 
1778     Register R1 = lreg->predecessor();
1779     Register R2 = rreg;
1780     assert(code != lir_idiv || lreg==dreg, "see code below");
1781     if (code == lir_idiv) {
1782       __ z_lcgr(lreg, lreg);
1783     } else {
1784       __ clear_reg(dreg, true, false);
1785     }
1786     NearLabel done;
1787     __ compare64_and_branch(R2, -1, Assembler::bcondEqual, done);
1788     if (code == lir_idiv) {
1789       __ z_lcgr(lreg, lreg); // Revert lcgr above.
1790     }
1791     if (ImplicitDiv0Checks) {
1792       // No debug info because the idiv won't trap.
1793       // Add_debug_info_for_div0 would instantiate another DivByZeroStub,
1794       // which is unnecessary, too.
1795       add_debug_info_for_div0(__ offset(), info);
1796     }
1797     __ z_dsgr(R1, R2);
1798     __ bind(done);
1799     return;
1800   }
1801 
1802   // 32 bit integer case
1803 
1804   assert(left->is_single_cpu(), "left must be register");
1805   assert(right->is_single_cpu() || is_power_of_2(right->as_jint()), "right must be register or power of 2 constant");
1806   assert(result->is_single_cpu(), "result must be register");
1807 
1808   Register lreg = left->as_register();
1809   Register dreg = result->as_register();
1810 
1811   if (right->is_constant()) {
1812     // Convert division by a power of two into some shifts and logical operations.
1813     Register treg1 = Z_R0_scratch;
1814     Register treg2 = Z_R1_scratch;
1815     jlong divisor = right->as_jint();
1816     jlong log_divisor = log2_long(right->as_jint());
1817     __ move_reg_if_needed(dreg, T_LONG, lreg, T_INT); // sign extend
1818     if (divisor == 2) {
1819       __ z_srlg(treg2, dreg, 63);     // dividend < 0 ?  1 : 0
1820     } else {
1821       __ z_srag(treg2, dreg, 63);     // dividend < 0 ? -1 : 0
1822       __ and_imm(treg2, divisor - 1, treg1, true);
1823     }
1824     if (code == lir_idiv) {
1825       __ z_agr(dreg, treg2);
1826       __ z_srag(dreg, dreg, log_divisor);
1827     } else {
1828       assert(code == lir_irem, "check");
1829       __ z_agr(treg2, dreg);
1830       __ and_imm(treg2, ~(divisor - 1), treg1, true);
1831       __ z_sgr(dreg, treg2);
1832     }
1833     return;
1834   }
1835 
1836   // Divisor is not a power of 2 constant.
1837   Register rreg = right->as_register();
1838   Register treg = temp->as_register();
1839   assert(right->is_single_cpu(), "right must be register");
1840   assert(lreg == Z_R11, "left register must be rax,");
1841   assert(rreg != lreg, "right register must not be same as left register");
1842   assert((code == lir_idiv && dreg == Z_R11 && treg == Z_R10)
1843       || (code == lir_irem && dreg == Z_R10 && treg == Z_R11), "see divInOpr(), divOutOpr(), remOutOpr()");
1844 
1845   Register R1 = lreg->predecessor();
1846   Register R2 = rreg;
1847   __ move_reg_if_needed(lreg, T_LONG, lreg, T_INT); // sign extend
1848   if (ImplicitDiv0Checks) {
1849     // No debug info because the idiv won't trap.
1850     // Add_debug_info_for_div0 would instantiate another DivByZeroStub,
1851     // which is unnecessary, too.
1852     add_debug_info_for_div0(__ offset(), info);
1853   }
1854   __ z_dsgfr(R1, R2);
1855 }
1856 
1857 void LIR_Assembler::throw_op(LIR_Opr exceptionPC, LIR_Opr exceptionOop, CodeEmitInfo* info) {
1858   assert(exceptionOop->as_register() == Z_EXC_OOP, "should match");
1859   assert(exceptionPC->as_register() == Z_EXC_PC, "should match");
1860 
1861   // Exception object is not added to oop map by LinearScan
1862   // (LinearScan assumes that no oops are in fixed registers).
1863   info->add_register_oop(exceptionOop);
1864 
1865   // Reuse the debug info from the safepoint poll for the throw op itself.
1866   __ get_PC(Z_EXC_PC);
1867   add_call_info(__ offset(), info); // for exception handler
1868   address stub = Runtime1::entry_for (compilation()->has_fpu_code() ? Runtime1::handle_exception_id
1869                                                                     : Runtime1::handle_exception_nofpu_id);
1870   emit_call_c(stub);
1871 }
1872 
1873 void LIR_Assembler::unwind_op(LIR_Opr exceptionOop) {
1874   assert(exceptionOop->as_register() == Z_EXC_OOP, "should match");
1875 
1876   __ branch_optimized(Assembler::bcondAlways, _unwind_handler_entry);
1877 }
1878 
1879 void LIR_Assembler::emit_arraycopy(LIR_OpArrayCopy* op) {
1880   ciArrayKlass* default_type = op->expected_type();
1881   Register src = op->src()->as_register();
1882   Register dst = op->dst()->as_register();
1883   Register src_pos = op->src_pos()->as_register();
1884   Register dst_pos = op->dst_pos()->as_register();
1885   Register length  = op->length()->as_register();
1886   Register tmp = op->tmp()->as_register();
1887 
1888   CodeStub* stub = op->stub();
1889   int flags = op->flags();
1890   BasicType basic_type = default_type != NULL ? default_type->element_type()->basic_type() : T_ILLEGAL;
1891   if (basic_type == T_ARRAY) basic_type = T_OBJECT;
1892 
1893   // If we don't know anything, just go through the generic arraycopy.
1894   if (default_type == NULL) {
1895     Label done;
1896     // Save outgoing arguments in callee saved registers (C convention) in case
1897     // a call to System.arraycopy is needed.
1898     Register callee_saved_src     = Z_R10;
1899     Register callee_saved_src_pos = Z_R11;
1900     Register callee_saved_dst     = Z_R12;
1901     Register callee_saved_dst_pos = Z_R13;
1902     Register callee_saved_length  = Z_ARG5; // Z_ARG5 == Z_R6 is callee saved.
1903 
1904     __ lgr_if_needed(callee_saved_src, src);
1905     __ lgr_if_needed(callee_saved_src_pos, src_pos);
1906     __ lgr_if_needed(callee_saved_dst, dst);
1907     __ lgr_if_needed(callee_saved_dst_pos, dst_pos);
1908     __ lgr_if_needed(callee_saved_length, length);
1909 
1910     // C function requires 64 bit values.
1911     __ z_lgfr(src_pos, src_pos);
1912     __ z_lgfr(dst_pos, dst_pos);
1913     __ z_lgfr(length, length);
1914 
1915     address C_entry = CAST_FROM_FN_PTR(address, Runtime1::arraycopy);
1916 
1917     address copyfunc_addr = StubRoutines::generic_arraycopy();
1918 
1919     // Pass arguments: may push as this is not a safepoint; SP must be fix at each safepoint.
1920 
1921     // The arguments are in the corresponding registers.
1922     assert(Z_ARG1 == src,     "assumption");
1923     assert(Z_ARG2 == src_pos, "assumption");
1924     assert(Z_ARG3 == dst,     "assumption");
1925     assert(Z_ARG4 == dst_pos, "assumption");
1926     assert(Z_ARG5 == length,  "assumption");
1927     if (copyfunc_addr == NULL) { // Use C version if stub was not generated.
1928       emit_call_c(C_entry);
1929     } else {
1930 #ifndef PRODUCT
1931       if (PrintC1Statistics) {
1932         __ load_const_optimized(Z_R1_scratch, (address)&Runtime1::_generic_arraycopystub_cnt);
1933         __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
1934       }
1935 #endif
1936       emit_call_c(copyfunc_addr);
1937     }
1938     CHECK_BAILOUT();
1939 
1940     __ compare32_and_branch(Z_RET, (intptr_t)0, Assembler::bcondEqual, *stub->continuation());
1941 
1942     if (copyfunc_addr != NULL) {
1943       __ z_lgr(tmp, Z_RET);
1944       __ z_xilf(tmp, -1);
1945     }
1946 
1947     // Restore values from callee saved registers so they are where the stub
1948     // expects them.
1949     __ lgr_if_needed(src, callee_saved_src);
1950     __ lgr_if_needed(src_pos, callee_saved_src_pos);
1951     __ lgr_if_needed(dst, callee_saved_dst);
1952     __ lgr_if_needed(dst_pos, callee_saved_dst_pos);
1953     __ lgr_if_needed(length, callee_saved_length);
1954 
1955     if (copyfunc_addr != NULL) {
1956       __ z_sr(length, tmp);
1957       __ z_ar(src_pos, tmp);
1958       __ z_ar(dst_pos, tmp);
1959     }
1960     __ branch_optimized(Assembler::bcondAlways, *stub->entry());
1961 
1962     __ bind(*stub->continuation());
1963     return;
1964   }
1965 
1966   assert(default_type != NULL && default_type->is_array_klass() && default_type->is_loaded(), "must be true at this point");
1967 
1968   int elem_size = type2aelembytes(basic_type);
1969   int shift_amount;
1970 
1971   switch (elem_size) {
1972     case 1 :
1973       shift_amount = 0;
1974       break;
1975     case 2 :
1976       shift_amount = 1;
1977       break;
1978     case 4 :
1979       shift_amount = 2;
1980       break;
1981     case 8 :
1982       shift_amount = 3;
1983       break;
1984     default:
1985       shift_amount = -1;
1986       ShouldNotReachHere();
1987   }
1988 
1989   Address src_length_addr = Address(src, arrayOopDesc::length_offset_in_bytes());
1990   Address dst_length_addr = Address(dst, arrayOopDesc::length_offset_in_bytes());
1991   Address src_klass_addr = Address(src, oopDesc::klass_offset_in_bytes());
1992   Address dst_klass_addr = Address(dst, oopDesc::klass_offset_in_bytes());
1993 
1994   // Length and pos's are all sign extended at this point on 64bit.
1995 
1996   // test for NULL
1997   if (flags & LIR_OpArrayCopy::src_null_check) {
1998     __ compareU64_and_branch(src, (intptr_t)0, Assembler::bcondZero, *stub->entry());
1999   }
2000   if (flags & LIR_OpArrayCopy::dst_null_check) {
2001     __ compareU64_and_branch(dst, (intptr_t)0, Assembler::bcondZero, *stub->entry());
2002   }
2003 
2004   // Check if negative.
2005   if (flags & LIR_OpArrayCopy::src_pos_positive_check) {
2006     __ compare32_and_branch(src_pos, (intptr_t)0, Assembler::bcondLow, *stub->entry());
2007   }
2008   if (flags & LIR_OpArrayCopy::dst_pos_positive_check) {
2009     __ compare32_and_branch(dst_pos, (intptr_t)0, Assembler::bcondLow, *stub->entry());
2010   }
2011 
2012   // If the compiler was not able to prove that exact type of the source or the destination
2013   // of the arraycopy is an array type, check at runtime if the source or the destination is
2014   // an instance type.
2015   if (flags & LIR_OpArrayCopy::type_check) {
2016     assert(Klass::_lh_neutral_value == 0, "or replace z_lt instructions");
2017 
2018     if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
2019       __ load_klass(tmp, dst);
2020       __ z_lt(tmp, Address(tmp, in_bytes(Klass::layout_helper_offset())));
2021       __ branch_optimized(Assembler::bcondNotLow, *stub->entry());
2022     }
2023 
2024     if (!(flags & LIR_OpArrayCopy::src_objarray)) {
2025       __ load_klass(tmp, src);
2026       __ z_lt(tmp, Address(tmp, in_bytes(Klass::layout_helper_offset())));
2027       __ branch_optimized(Assembler::bcondNotLow, *stub->entry());
2028     }
2029   }
2030 
2031   if (flags & LIR_OpArrayCopy::src_range_check) {
2032     __ z_la(tmp, Address(src_pos, length));
2033     __ z_cl(tmp, src_length_addr);
2034     __ branch_optimized(Assembler::bcondHigh, *stub->entry());
2035   }
2036   if (flags & LIR_OpArrayCopy::dst_range_check) {
2037     __ z_la(tmp, Address(dst_pos, length));
2038     __ z_cl(tmp, dst_length_addr);
2039     __ branch_optimized(Assembler::bcondHigh, *stub->entry());
2040   }
2041 
2042   if (flags & LIR_OpArrayCopy::length_positive_check) {
2043     __ z_ltr(length, length);
2044     __ branch_optimized(Assembler::bcondNegative, *stub->entry());
2045   }
2046 
2047   // Stubs require 64 bit values.
2048   __ z_lgfr(src_pos, src_pos); // int -> long
2049   __ z_lgfr(dst_pos, dst_pos); // int -> long
2050   __ z_lgfr(length, length);   // int -> long
2051 
2052   if (flags & LIR_OpArrayCopy::type_check) {
2053     // We don't know the array types are compatible.
2054     if (basic_type != T_OBJECT) {
2055       // Simple test for basic type arrays.
2056       if (UseCompressedClassPointers) {
2057         __ z_l(tmp, src_klass_addr);
2058         __ z_c(tmp, dst_klass_addr);
2059       } else {
2060         __ z_lg(tmp, src_klass_addr);
2061         __ z_cg(tmp, dst_klass_addr);
2062       }
2063       __ branch_optimized(Assembler::bcondNotEqual, *stub->entry());
2064     } else {
2065       // For object arrays, if src is a sub class of dst then we can
2066       // safely do the copy.
2067       NearLabel cont, slow;
2068       Register src_klass = Z_R1_scratch;
2069       Register dst_klass = Z_R10;
2070 
2071       __ load_klass(src_klass, src);
2072       __ load_klass(dst_klass, dst);
2073 
2074       __ check_klass_subtype_fast_path(src_klass, dst_klass, tmp, &cont, &slow, NULL);
2075 
2076       store_parameter(src_klass, 0); // sub
2077       store_parameter(dst_klass, 1); // super
2078       emit_call_c(Runtime1::entry_for (Runtime1::slow_subtype_check_id));
2079       CHECK_BAILOUT();
2080       // Sets condition code 0 for match (2 otherwise).
2081       __ branch_optimized(Assembler::bcondEqual, cont);
2082 
2083       __ bind(slow);
2084 
2085       address copyfunc_addr = StubRoutines::checkcast_arraycopy();
2086       if (copyfunc_addr != NULL) { // use stub if available
2087         // Src is not a sub class of dst so we have to do a
2088         // per-element check.
2089 
2090         int mask = LIR_OpArrayCopy::src_objarray|LIR_OpArrayCopy::dst_objarray;
2091         if ((flags & mask) != mask) {
2092           // Check that at least both of them object arrays.
2093           assert(flags & mask, "one of the two should be known to be an object array");
2094 
2095           if (!(flags & LIR_OpArrayCopy::src_objarray)) {
2096             __ load_klass(tmp, src);
2097           } else if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
2098             __ load_klass(tmp, dst);
2099           }
2100           Address klass_lh_addr(tmp, Klass::layout_helper_offset());
2101           jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2102           __ load_const_optimized(Z_R1_scratch, objArray_lh);
2103           __ z_c(Z_R1_scratch, klass_lh_addr);
2104           __ branch_optimized(Assembler::bcondNotEqual, *stub->entry());
2105         }
2106 
2107         // Save outgoing arguments in callee saved registers (C convention) in case
2108         // a call to System.arraycopy is needed.
2109         Register callee_saved_src     = Z_R10;
2110         Register callee_saved_src_pos = Z_R11;
2111         Register callee_saved_dst     = Z_R12;
2112         Register callee_saved_dst_pos = Z_R13;
2113         Register callee_saved_length  = Z_ARG5; // Z_ARG5 == Z_R6 is callee saved.
2114 
2115         __ lgr_if_needed(callee_saved_src, src);
2116         __ lgr_if_needed(callee_saved_src_pos, src_pos);
2117         __ lgr_if_needed(callee_saved_dst, dst);
2118         __ lgr_if_needed(callee_saved_dst_pos, dst_pos);
2119         __ lgr_if_needed(callee_saved_length, length);
2120 
2121         __ z_llgfr(length, length); // Higher 32bits must be null.
2122 
2123         __ z_sllg(Z_ARG1, src_pos, shift_amount); // index -> byte offset
2124         __ z_sllg(Z_ARG2, dst_pos, shift_amount); // index -> byte offset
2125 
2126         __ z_la(Z_ARG1, Address(src, Z_ARG1, arrayOopDesc::base_offset_in_bytes(basic_type)));
2127         assert_different_registers(Z_ARG1, dst, dst_pos, length);
2128         __ z_la(Z_ARG2, Address(dst, Z_ARG2, arrayOopDesc::base_offset_in_bytes(basic_type)));
2129         assert_different_registers(Z_ARG2, dst, length);
2130 
2131         __ z_lgr(Z_ARG3, length);
2132         assert_different_registers(Z_ARG3, dst);
2133 
2134         __ load_klass(Z_ARG5, dst);
2135         __ z_lg(Z_ARG5, Address(Z_ARG5, ObjArrayKlass::element_klass_offset()));
2136         __ z_lg(Z_ARG4, Address(Z_ARG5, Klass::super_check_offset_offset()));
2137         emit_call_c(copyfunc_addr);
2138         CHECK_BAILOUT();
2139 
2140 #ifndef PRODUCT
2141         if (PrintC1Statistics) {
2142           NearLabel failed;
2143           __ compareU32_and_branch(Z_RET, (intptr_t)0, Assembler::bcondNotEqual, failed);
2144           __ load_const_optimized(Z_R1_scratch, (address)&Runtime1::_arraycopy_checkcast_cnt);
2145           __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
2146           __ bind(failed);
2147         }
2148 #endif
2149 
2150         __ compareU32_and_branch(Z_RET, (intptr_t)0, Assembler::bcondEqual, *stub->continuation());
2151 
2152 #ifndef PRODUCT
2153         if (PrintC1Statistics) {
2154           __ load_const_optimized(Z_R1_scratch, (address)&Runtime1::_arraycopy_checkcast_attempt_cnt);
2155           __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
2156         }
2157 #endif
2158 
2159         __ z_lgr(tmp, Z_RET);
2160         __ z_xilf(tmp, -1);
2161 
2162         // Restore previously spilled arguments
2163         __ lgr_if_needed(src, callee_saved_src);
2164         __ lgr_if_needed(src_pos, callee_saved_src_pos);
2165         __ lgr_if_needed(dst, callee_saved_dst);
2166         __ lgr_if_needed(dst_pos, callee_saved_dst_pos);
2167         __ lgr_if_needed(length, callee_saved_length);
2168 
2169         __ z_sr(length, tmp);
2170         __ z_ar(src_pos, tmp);
2171         __ z_ar(dst_pos, tmp);
2172       }
2173 
2174       __ branch_optimized(Assembler::bcondAlways, *stub->entry());
2175 
2176       __ bind(cont);
2177     }
2178   }
2179 
2180 #ifdef ASSERT
2181   if (basic_type != T_OBJECT || !(flags & LIR_OpArrayCopy::type_check)) {
2182     // Sanity check the known type with the incoming class. For the
2183     // primitive case the types must match exactly with src.klass and
2184     // dst.klass each exactly matching the default type. For the
2185     // object array case, if no type check is needed then either the
2186     // dst type is exactly the expected type and the src type is a
2187     // subtype which we can't check or src is the same array as dst
2188     // but not necessarily exactly of type default_type.
2189     NearLabel known_ok, halt;
2190     metadata2reg(default_type->constant_encoding(), tmp);
2191     if (UseCompressedClassPointers) {
2192       __ encode_klass_not_null(tmp);
2193     }
2194 
2195     if (basic_type != T_OBJECT) {
2196       if (UseCompressedClassPointers)         { __ z_c (tmp, dst_klass_addr); }
2197       else                                    { __ z_cg(tmp, dst_klass_addr); }
2198       __ branch_optimized(Assembler::bcondNotEqual, halt);
2199       if (UseCompressedClassPointers)         { __ z_c (tmp, src_klass_addr); }
2200       else                                    { __ z_cg(tmp, src_klass_addr); }
2201       __ branch_optimized(Assembler::bcondEqual, known_ok);
2202     } else {
2203       if (UseCompressedClassPointers)         { __ z_c (tmp, dst_klass_addr); }
2204       else                                    { __ z_cg(tmp, dst_klass_addr); }
2205       __ branch_optimized(Assembler::bcondEqual, known_ok);
2206       __ compareU64_and_branch(src, dst, Assembler::bcondEqual, known_ok);
2207     }
2208     __ bind(halt);
2209     __ stop("incorrect type information in arraycopy");
2210     __ bind(known_ok);
2211   }
2212 #endif
2213 
2214 #ifndef PRODUCT
2215   if (PrintC1Statistics) {
2216     __ load_const_optimized(Z_R1_scratch, Runtime1::arraycopy_count_address(basic_type));
2217     __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
2218   }
2219 #endif
2220 
2221   __ z_sllg(tmp, src_pos, shift_amount); // index -> byte offset
2222   __ z_sllg(Z_R1_scratch, dst_pos, shift_amount); // index -> byte offset
2223 
2224   assert_different_registers(Z_ARG1, dst, dst_pos, length);
2225   __ z_la(Z_ARG1, Address(src, tmp, arrayOopDesc::base_offset_in_bytes(basic_type)));
2226   assert_different_registers(Z_ARG2, length);
2227   __ z_la(Z_ARG2, Address(dst, Z_R1_scratch, arrayOopDesc::base_offset_in_bytes(basic_type)));
2228   __ lgr_if_needed(Z_ARG3, length);
2229 
2230   bool disjoint = (flags & LIR_OpArrayCopy::overlapping) == 0;
2231   bool aligned = (flags & LIR_OpArrayCopy::unaligned) == 0;
2232   const char *name;
2233   address entry = StubRoutines::select_arraycopy_function(basic_type, aligned, disjoint, name, false);
2234   __ call_VM_leaf(entry);
2235 
2236   __ bind(*stub->continuation());
2237 }
2238 
2239 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
2240   if (dest->is_single_cpu()) {
2241     if (left->type() == T_OBJECT) {
2242       switch (code) {
2243         case lir_shl:  __ z_sllg (dest->as_register(), left->as_register(), 0, count->as_register()); break;
2244         case lir_shr:  __ z_srag (dest->as_register(), left->as_register(), 0, count->as_register()); break;
2245         case lir_ushr: __ z_srlg (dest->as_register(), left->as_register(), 0, count->as_register()); break;
2246         default: ShouldNotReachHere();
2247       }
2248     } else {
2249       assert(code == lir_shl || left == dest, "left and dest must be equal for 2 operand form right shifts");
2250       Register masked_count = Z_R1_scratch;
2251       __ z_lr(masked_count, count->as_register());
2252       __ z_nill(masked_count, 31);
2253       switch (code) {
2254         case lir_shl:  __ z_sllg (dest->as_register(), left->as_register(), 0, masked_count); break;
2255         case lir_shr:  __ z_sra  (dest->as_register(), 0, masked_count); break;
2256         case lir_ushr: __ z_srl  (dest->as_register(), 0, masked_count); break;
2257         default: ShouldNotReachHere();
2258       }
2259     }
2260   } else {
2261     switch (code) {
2262       case lir_shl:  __ z_sllg (dest->as_register_lo(), left->as_register_lo(), 0, count->as_register()); break;
2263       case lir_shr:  __ z_srag (dest->as_register_lo(), left->as_register_lo(), 0, count->as_register()); break;
2264       case lir_ushr: __ z_srlg (dest->as_register_lo(), left->as_register_lo(), 0, count->as_register()); break;
2265       default: ShouldNotReachHere();
2266     }
2267   }
2268 }
2269 
2270 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, jint count, LIR_Opr dest) {
2271   if (left->type() == T_OBJECT) {
2272     count = count & 63;  // Shouldn't shift by more than sizeof(intptr_t).
2273     Register l = left->as_register();
2274     Register d = dest->as_register_lo();
2275     switch (code) {
2276       case lir_shl:  __ z_sllg (d, l, count); break;
2277       case lir_shr:  __ z_srag (d, l, count); break;
2278       case lir_ushr: __ z_srlg (d, l, count); break;
2279       default: ShouldNotReachHere();
2280     }
2281     return;
2282   }
2283   if (dest->is_single_cpu()) {
2284     assert(code == lir_shl || left == dest, "left and dest must be equal for 2 operand form right shifts");
2285     count = count & 0x1F; // Java spec
2286     switch (code) {
2287       case lir_shl:  __ z_sllg (dest->as_register(), left->as_register(), count); break;
2288       case lir_shr:  __ z_sra  (dest->as_register(), count); break;
2289       case lir_ushr: __ z_srl  (dest->as_register(), count); break;
2290       default: ShouldNotReachHere();
2291     }
2292   } else if (dest->is_double_cpu()) {
2293     count = count & 63; // Java spec
2294     Register l = left->as_pointer_register();
2295     Register d = dest->as_pointer_register();
2296     switch (code) {
2297       case lir_shl:  __ z_sllg (d, l, count); break;
2298       case lir_shr:  __ z_srag (d, l, count); break;
2299       case lir_ushr: __ z_srlg (d, l, count); break;
2300       default: ShouldNotReachHere();
2301     }
2302   } else {
2303     ShouldNotReachHere();
2304   }
2305 }
2306 
2307 void LIR_Assembler::emit_alloc_obj(LIR_OpAllocObj* op) {
2308   if (op->init_check()) {
2309     // Make sure klass is initialized & doesn't have finalizer.
2310     const int state_offset = in_bytes(InstanceKlass::init_state_offset());
2311     Register iklass = op->klass()->as_register();
2312     add_debug_info_for_null_check_here(op->stub()->info());
2313     if (Immediate::is_uimm12(state_offset)) {
2314       __ z_cli(state_offset, iklass, InstanceKlass::fully_initialized);
2315     } else {
2316       __ z_cliy(state_offset, iklass, InstanceKlass::fully_initialized);
2317     }
2318     __ branch_optimized(Assembler::bcondNotEqual, *op->stub()->entry()); // Use long branch, because slow_case might be far.
2319   }
2320   __ allocate_object(op->obj()->as_register(),
2321                      op->tmp1()->as_register(),
2322                      op->tmp2()->as_register(),
2323                      op->header_size(),
2324                      op->object_size(),
2325                      op->klass()->as_register(),
2326                      *op->stub()->entry());
2327   __ bind(*op->stub()->continuation());
2328   __ verify_oop(op->obj()->as_register());
2329 }
2330 
2331 void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
2332   Register len = op->len()->as_register();
2333   __ move_reg_if_needed(len, T_LONG, len, T_INT); // sign extend
2334 
2335   if (UseSlowPath ||
2336       (!UseFastNewObjectArray && (op->type() == T_OBJECT || op->type() == T_ARRAY)) ||
2337       (!UseFastNewTypeArray   && (op->type() != T_OBJECT && op->type() != T_ARRAY))) {
2338     __ z_brul(*op->stub()->entry());
2339   } else {
2340     __ allocate_array(op->obj()->as_register(),
2341                       op->len()->as_register(),
2342                       op->tmp1()->as_register(),
2343                       op->tmp2()->as_register(),
2344                       arrayOopDesc::header_size(op->type()),
2345                       type2aelembytes(op->type()),
2346                       op->klass()->as_register(),
2347                       *op->stub()->entry());
2348   }
2349   __ bind(*op->stub()->continuation());
2350 }
2351 
2352 void LIR_Assembler::type_profile_helper(Register mdo, ciMethodData *md, ciProfileData *data,
2353                                         Register recv, Register tmp1, Label* update_done) {
2354   uint i;
2355   for (i = 0; i < VirtualCallData::row_limit(); i++) {
2356     Label next_test;
2357     // See if the receiver is receiver[n].
2358     Address receiver_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)));
2359     __ z_cg(recv, receiver_addr);
2360     __ z_brne(next_test);
2361     Address data_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)));
2362     __ add2mem_64(data_addr, DataLayout::counter_increment, tmp1);
2363     __ branch_optimized(Assembler::bcondAlways, *update_done);
2364     __ bind(next_test);
2365   }
2366 
2367   // Didn't find receiver; find next empty slot and fill it in.
2368   for (i = 0; i < VirtualCallData::row_limit(); i++) {
2369     Label next_test;
2370     Address recv_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)));
2371     __ z_ltg(Z_R0_scratch, recv_addr);
2372     __ z_brne(next_test);
2373     __ z_stg(recv, recv_addr);
2374     __ load_const_optimized(tmp1, DataLayout::counter_increment);
2375     __ z_stg(tmp1, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)), mdo);
2376     __ branch_optimized(Assembler::bcondAlways, *update_done);
2377     __ bind(next_test);
2378   }
2379 }
2380 
2381 void LIR_Assembler::setup_md_access(ciMethod* method, int bci,
2382                                     ciMethodData*& md, ciProfileData*& data, int& mdo_offset_bias) {
2383   Unimplemented();
2384 }
2385 
2386 void LIR_Assembler::store_parameter(Register r, int param_num) {
2387   assert(param_num >= 0, "invalid num");
2388   int offset_in_bytes = param_num * BytesPerWord + FrameMap::first_available_sp_in_frame;
2389   assert(offset_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
2390   __ z_stg(r, offset_in_bytes, Z_SP);
2391 }
2392 
2393 void LIR_Assembler::store_parameter(jint c, int param_num) {
2394   assert(param_num >= 0, "invalid num");
2395   int offset_in_bytes = param_num * BytesPerWord + FrameMap::first_available_sp_in_frame;
2396   assert(offset_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
2397   __ store_const(Address(Z_SP, offset_in_bytes), c, Z_R1_scratch, true);
2398 }
2399 
2400 void LIR_Assembler::emit_typecheck_helper(LIR_OpTypeCheck *op, Label* success, Label* failure, Label* obj_is_null) {
2401   // We always need a stub for the failure case.
2402   CodeStub* stub = op->stub();
2403   Register obj = op->object()->as_register();
2404   Register k_RInfo = op->tmp1()->as_register();
2405   Register klass_RInfo = op->tmp2()->as_register();
2406   Register dst = op->result_opr()->as_register();
2407   Register Rtmp1 = Z_R1_scratch;
2408   ciKlass* k = op->klass();
2409 
2410   assert(!op->tmp3()->is_valid(), "tmp3's not needed");
2411 
2412   // Check if it needs to be profiled.
2413   ciMethodData* md = NULL;
2414   ciProfileData* data = NULL;
2415 
2416   if (op->should_profile()) {
2417     ciMethod* method = op->profiled_method();
2418     assert(method != NULL, "Should have method");
2419     int bci = op->profiled_bci();
2420     md = method->method_data_or_null();
2421     assert(md != NULL, "Sanity");
2422     data = md->bci_to_data(bci);
2423     assert(data != NULL,                "need data for type check");
2424     assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
2425   }
2426 
2427   // Temp operands do not overlap with inputs, if this is their last
2428   // use (end of range is exclusive), so a register conflict is possible.
2429   if (obj == k_RInfo) {
2430     k_RInfo = dst;
2431   } else if (obj == klass_RInfo) {
2432     klass_RInfo = dst;
2433   }
2434   assert_different_registers(obj, k_RInfo, klass_RInfo);
2435 
2436   if (op->should_profile()) {
2437     NearLabel not_null;
2438     __ compareU64_and_branch(obj, (intptr_t) 0, Assembler::bcondNotEqual, not_null);
2439     // Object is null; update MDO and exit.
2440     Register mdo = klass_RInfo;
2441     metadata2reg(md->constant_encoding(), mdo);
2442     Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
2443     int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
2444     __ or2mem_8(data_addr, header_bits);
2445     __ branch_optimized(Assembler::bcondAlways, *obj_is_null);
2446     __ bind(not_null);
2447   } else {
2448     __ compareU64_and_branch(obj, (intptr_t) 0, Assembler::bcondEqual, *obj_is_null);
2449   }
2450 
2451   NearLabel profile_cast_failure, profile_cast_success;
2452   Label *failure_target = op->should_profile() ? &profile_cast_failure : failure;
2453   Label *success_target = op->should_profile() ? &profile_cast_success : success;
2454 
2455   // Patching may screw with our temporaries on sparc,
2456   // so let's do it before loading the class.
2457   if (k->is_loaded()) {
2458     metadata2reg(k->constant_encoding(), k_RInfo);
2459   } else {
2460     klass2reg_with_patching(k_RInfo, op->info_for_patch());
2461   }
2462   assert(obj != k_RInfo, "must be different");
2463 
2464   __ verify_oop(obj);
2465 
2466   // Get object class.
2467   // Not a safepoint as obj null check happens earlier.
2468   if (op->fast_check()) {
2469     if (UseCompressedClassPointers) {
2470       __ load_klass(klass_RInfo, obj);
2471       __ compareU64_and_branch(k_RInfo, klass_RInfo, Assembler::bcondNotEqual, *failure_target);
2472     } else {
2473       __ z_cg(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
2474       __ branch_optimized(Assembler::bcondNotEqual, *failure_target);
2475     }
2476     // Successful cast, fall through to profile or jump.
2477   } else {
2478     bool need_slow_path = !k->is_loaded() ||
2479                           ((int) k->super_check_offset() == in_bytes(Klass::secondary_super_cache_offset()));
2480     intptr_t super_check_offset = k->is_loaded() ? k->super_check_offset() : -1L;
2481     __ load_klass(klass_RInfo, obj);
2482     // Perform the fast part of the checking logic.
2483     __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1,
2484                                      (need_slow_path ? success_target : NULL),
2485                                      failure_target, NULL,
2486                                      RegisterOrConstant(super_check_offset));
2487     if (need_slow_path) {
2488       // Call out-of-line instance of __ check_klass_subtype_slow_path(...):
2489       address a = Runtime1::entry_for (Runtime1::slow_subtype_check_id);
2490       store_parameter(klass_RInfo, 0); // sub
2491       store_parameter(k_RInfo, 1);     // super
2492       emit_call_c(a); // Sets condition code 0 for match (2 otherwise).
2493       CHECK_BAILOUT();
2494       __ branch_optimized(Assembler::bcondNotEqual, *failure_target);
2495       // Fall through to success case.
2496     }
2497   }
2498 
2499   if (op->should_profile()) {
2500     Register mdo = klass_RInfo, recv = k_RInfo;
2501     assert_different_registers(obj, mdo, recv);
2502     __ bind(profile_cast_success);
2503     metadata2reg(md->constant_encoding(), mdo);
2504     __ load_klass(recv, obj);
2505     type_profile_helper(mdo, md, data, recv, Rtmp1, success);
2506     __ branch_optimized(Assembler::bcondAlways, *success);
2507 
2508     __ bind(profile_cast_failure);
2509     metadata2reg(md->constant_encoding(), mdo);
2510     __ add2mem_64(Address(mdo, md->byte_offset_of_slot(data, CounterData::count_offset())), -(int)DataLayout::counter_increment, Rtmp1);
2511     __ branch_optimized(Assembler::bcondAlways, *failure);
2512   } else {
2513     __ branch_optimized(Assembler::bcondAlways, *success);
2514   }
2515 }
2516 
2517 void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
2518   LIR_Code code = op->code();
2519   if (code == lir_store_check) {
2520     Register value = op->object()->as_register();
2521     Register array = op->array()->as_register();
2522     Register k_RInfo = op->tmp1()->as_register();
2523     Register klass_RInfo = op->tmp2()->as_register();
2524     Register Rtmp1 = Z_R1_scratch;
2525 
2526     CodeStub* stub = op->stub();
2527 
2528     // Check if it needs to be profiled.
2529     ciMethodData* md = NULL;
2530     ciProfileData* data = NULL;
2531 
2532     assert_different_registers(value, k_RInfo, klass_RInfo);
2533 
2534     if (op->should_profile()) {
2535       ciMethod* method = op->profiled_method();
2536       assert(method != NULL, "Should have method");
2537       int bci = op->profiled_bci();
2538       md = method->method_data_or_null();
2539       assert(md != NULL, "Sanity");
2540       data = md->bci_to_data(bci);
2541       assert(data != NULL,                "need data for type check");
2542       assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
2543     }
2544     NearLabel profile_cast_success, profile_cast_failure, done;
2545     Label *success_target = op->should_profile() ? &profile_cast_success : &done;
2546     Label *failure_target = op->should_profile() ? &profile_cast_failure : stub->entry();
2547 
2548     if (op->should_profile()) {
2549       NearLabel not_null;
2550       __ compareU64_and_branch(value, (intptr_t) 0, Assembler::bcondNotEqual, not_null);
2551       // Object is null; update MDO and exit.
2552       Register mdo = klass_RInfo;
2553       metadata2reg(md->constant_encoding(), mdo);
2554       Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
2555       int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
2556       __ or2mem_8(data_addr, header_bits);
2557       __ branch_optimized(Assembler::bcondAlways, done);
2558       __ bind(not_null);
2559     } else {
2560       __ compareU64_and_branch(value, (intptr_t) 0, Assembler::bcondEqual, done);
2561     }
2562 
2563     add_debug_info_for_null_check_here(op->info_for_exception());
2564     __ load_klass(k_RInfo, array);
2565     __ load_klass(klass_RInfo, value);
2566 
2567     // Get instance klass (it's already uncompressed).
2568     __ z_lg(k_RInfo, Address(k_RInfo, ObjArrayKlass::element_klass_offset()));
2569     // Perform the fast part of the checking logic.
2570     __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, success_target, failure_target, NULL);
2571     // Call out-of-line instance of __ check_klass_subtype_slow_path(...):
2572     address a = Runtime1::entry_for (Runtime1::slow_subtype_check_id);
2573     store_parameter(klass_RInfo, 0); // sub
2574     store_parameter(k_RInfo, 1);     // super
2575     emit_call_c(a); // Sets condition code 0 for match (2 otherwise).
2576     CHECK_BAILOUT();
2577     __ branch_optimized(Assembler::bcondNotEqual, *failure_target);
2578     // Fall through to success case.
2579 
2580     if (op->should_profile()) {
2581       Register mdo = klass_RInfo, recv = k_RInfo;
2582       assert_different_registers(value, mdo, recv);
2583       __ bind(profile_cast_success);
2584       metadata2reg(md->constant_encoding(), mdo);
2585       __ load_klass(recv, value);
2586       type_profile_helper(mdo, md, data, recv, Rtmp1, &done);
2587       __ branch_optimized(Assembler::bcondAlways, done);
2588 
2589       __ bind(profile_cast_failure);
2590       metadata2reg(md->constant_encoding(), mdo);
2591       __ add2mem_64(Address(mdo, md->byte_offset_of_slot(data, CounterData::count_offset())), -(int)DataLayout::counter_increment, Rtmp1);
2592       __ branch_optimized(Assembler::bcondAlways, *stub->entry());
2593     }
2594 
2595     __ bind(done);
2596   } else {
2597     if (code == lir_checkcast) {
2598       Register obj = op->object()->as_register();
2599       Register dst = op->result_opr()->as_register();
2600       NearLabel success;
2601       emit_typecheck_helper(op, &success, op->stub()->entry(), &success);
2602       __ bind(success);
2603       __ lgr_if_needed(dst, obj);
2604     } else {
2605       if (code == lir_instanceof) {
2606         Register obj = op->object()->as_register();
2607         Register dst = op->result_opr()->as_register();
2608         NearLabel success, failure, done;
2609         emit_typecheck_helper(op, &success, &failure, &failure);
2610         __ bind(failure);
2611         __ clear_reg(dst);
2612         __ branch_optimized(Assembler::bcondAlways, done);
2613         __ bind(success);
2614         __ load_const_optimized(dst, 1);
2615         __ bind(done);
2616       } else {
2617         ShouldNotReachHere();
2618       }
2619     }
2620   }
2621 }
2622 
2623 void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
2624   Register addr = op->addr()->as_pointer_register();
2625   Register t1_cmp = Z_R1_scratch;
2626   if (op->code() == lir_cas_long) {
2627     assert(VM_Version::supports_cx8(), "wrong machine");
2628     Register cmp_value_lo = op->cmp_value()->as_register_lo();
2629     Register new_value_lo = op->new_value()->as_register_lo();
2630     __ z_lgr(t1_cmp, cmp_value_lo);
2631     // Perform the compare and swap operation.
2632     __ z_csg(t1_cmp, new_value_lo, 0, addr);
2633   } else if (op->code() == lir_cas_int || op->code() == lir_cas_obj) {
2634     Register cmp_value = op->cmp_value()->as_register();
2635     Register new_value = op->new_value()->as_register();
2636     if (op->code() == lir_cas_obj) {
2637       if (UseCompressedOops) {
2638                  t1_cmp = op->tmp1()->as_register();
2639         Register t2_new = op->tmp2()->as_register();
2640         assert_different_registers(cmp_value, new_value, addr, t1_cmp, t2_new);
2641         __ oop_encoder(t1_cmp, cmp_value, true /*maybe null*/);
2642         __ oop_encoder(t2_new, new_value, true /*maybe null*/);
2643         __ z_cs(t1_cmp, t2_new, 0, addr);
2644       } else {
2645         __ z_lgr(t1_cmp, cmp_value);
2646         __ z_csg(t1_cmp, new_value, 0, addr);
2647       }
2648     } else {
2649       __ z_lr(t1_cmp, cmp_value);
2650       __ z_cs(t1_cmp, new_value, 0, addr);
2651     }
2652   } else {
2653     ShouldNotReachHere(); // new lir_cas_??
2654   }
2655 }
2656 
2657 void LIR_Assembler::set_24bit_FPU() {
2658   ShouldNotCallThis(); // x86 only
2659 }
2660 
2661 void LIR_Assembler::reset_FPU() {
2662   ShouldNotCallThis(); // x86 only
2663 }
2664 
2665 void LIR_Assembler::breakpoint() {
2666   Unimplemented();
2667   //  __ breakpoint_trap();
2668 }
2669 
2670 void LIR_Assembler::push(LIR_Opr opr) {
2671   ShouldNotCallThis(); // unused
2672 }
2673 
2674 void LIR_Assembler::pop(LIR_Opr opr) {
2675   ShouldNotCallThis(); // unused
2676 }
2677 
2678 void LIR_Assembler::monitor_address(int monitor_no, LIR_Opr dst_opr) {
2679   Address addr = frame_map()->address_for_monitor_lock(monitor_no);
2680   __ add2reg(dst_opr->as_register(), addr.disp(), addr.base());
2681 }
2682 
2683 void LIR_Assembler::emit_lock(LIR_OpLock* op) {
2684   Register obj = op->obj_opr()->as_register();  // May not be an oop.
2685   Register hdr = op->hdr_opr()->as_register();
2686   Register lock = op->lock_opr()->as_register();
2687   if (!UseFastLocking) {
2688     __ branch_optimized(Assembler::bcondAlways, *op->stub()->entry());
2689   } else if (op->code() == lir_lock) {
2690     assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
2691     // Add debug info for NullPointerException only if one is possible.
2692     if (op->info() != NULL) {
2693       add_debug_info_for_null_check_here(op->info());
2694     }
2695     __ lock_object(hdr, obj, lock, *op->stub()->entry());
2696     // done
2697   } else if (op->code() == lir_unlock) {
2698     assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
2699     __ unlock_object(hdr, obj, lock, *op->stub()->entry());
2700   } else {
2701     ShouldNotReachHere();
2702   }
2703   __ bind(*op->stub()->continuation());
2704 }
2705 
2706 void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
2707   ciMethod* method = op->profiled_method();
2708   int bci          = op->profiled_bci();
2709   ciMethod* callee = op->profiled_callee();
2710 
2711   // Update counter for all call types.
2712   ciMethodData* md = method->method_data_or_null();
2713   assert(md != NULL, "Sanity");
2714   ciProfileData* data = md->bci_to_data(bci);
2715   assert(data->is_CounterData(), "need CounterData for calls");
2716   assert(op->mdo()->is_single_cpu(),  "mdo must be allocated");
2717   Register mdo  = op->mdo()->as_register();
2718   assert(op->tmp1()->is_double_cpu(), "tmp1 must be allocated");
2719   Register tmp1 = op->tmp1()->as_register_lo();
2720   metadata2reg(md->constant_encoding(), mdo);
2721 
2722   Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
2723   Bytecodes::Code bc = method->java_code_at_bci(bci);
2724   const bool callee_is_static = callee->is_loaded() && callee->is_static();
2725   // Perform additional virtual call profiling for invokevirtual and
2726   // invokeinterface bytecodes.
2727   if ((bc == Bytecodes::_invokevirtual || bc == Bytecodes::_invokeinterface) &&
2728       !callee_is_static &&  // Required for optimized MH invokes.
2729       C1ProfileVirtualCalls) {
2730     assert(op->recv()->is_single_cpu(), "recv must be allocated");
2731     Register recv = op->recv()->as_register();
2732     assert_different_registers(mdo, tmp1, recv);
2733     assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
2734     ciKlass* known_klass = op->known_holder();
2735     if (C1OptimizeVirtualCallProfiling && known_klass != NULL) {
2736       // We know the type that will be seen at this call site; we can
2737       // statically update the MethodData* rather than needing to do
2738       // dynamic tests on the receiver type.
2739 
2740       // NOTE: we should probably put a lock around this search to
2741       // avoid collisions by concurrent compilations.
2742       ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
2743       uint i;
2744       for (i = 0; i < VirtualCallData::row_limit(); i++) {
2745         ciKlass* receiver = vc_data->receiver(i);
2746         if (known_klass->equals(receiver)) {
2747           Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
2748           __ add2mem_64(data_addr, DataLayout::counter_increment, tmp1);
2749           return;
2750         }
2751       }
2752 
2753       // Receiver type not found in profile data. Select an empty slot.
2754 
2755       // Note that this is less efficient than it should be because it
2756       // always does a write to the receiver part of the
2757       // VirtualCallData rather than just the first time.
2758       for (i = 0; i < VirtualCallData::row_limit(); i++) {
2759         ciKlass* receiver = vc_data->receiver(i);
2760         if (receiver == NULL) {
2761           Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)));
2762           metadata2reg(known_klass->constant_encoding(), tmp1);
2763           __ z_stg(tmp1, recv_addr);
2764           Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
2765           __ add2mem_64(data_addr, DataLayout::counter_increment, tmp1);
2766           return;
2767         }
2768       }
2769     } else {
2770       __ load_klass(recv, recv);
2771       NearLabel update_done;
2772       type_profile_helper(mdo, md, data, recv, tmp1, &update_done);
2773       // Receiver did not match any saved receiver and there is no empty row for it.
2774       // Increment total counter to indicate polymorphic case.
2775       __ add2mem_64(counter_addr, DataLayout::counter_increment, tmp1);
2776       __ bind(update_done);
2777     }
2778   } else {
2779     // static call
2780     __ add2mem_64(counter_addr, DataLayout::counter_increment, tmp1);
2781   }
2782 }
2783 
2784 void LIR_Assembler::align_backward_branch_target() {
2785   __ align(OptoLoopAlignment);
2786 }
2787 
2788 void LIR_Assembler::emit_delay(LIR_OpDelay* op) {
2789   ShouldNotCallThis(); // There are no delay slots on ZARCH_64.
2790 }
2791 
2792 void LIR_Assembler::negate(LIR_Opr left, LIR_Opr dest) {
2793   assert(left->is_register(), "can only handle registers");
2794 
2795   if (left->is_single_cpu()) {
2796     __ z_lcr(dest->as_register(), left->as_register());
2797   } else if (left->is_single_fpu()) {
2798     __ z_lcebr(dest->as_float_reg(), left->as_float_reg());
2799   } else if (left->is_double_fpu()) {
2800     __ z_lcdbr(dest->as_double_reg(), left->as_double_reg());
2801   } else {
2802     assert(left->is_double_cpu(), "Must be a long");
2803     __ z_lcgr(dest->as_register_lo(), left->as_register_lo());
2804   }
2805 }
2806 
2807 void LIR_Assembler::fxch(int i) {
2808   ShouldNotCallThis(); // x86 only
2809 }
2810 
2811 void LIR_Assembler::fld(int i) {
2812   ShouldNotCallThis(); // x86 only
2813 }
2814 
2815 void LIR_Assembler::ffree(int i) {
2816   ShouldNotCallThis(); // x86 only
2817 }
2818 
2819 void LIR_Assembler::rt_call(LIR_Opr result, address dest,
2820                             const LIR_OprList* args, LIR_Opr tmp, CodeEmitInfo* info) {
2821   assert(!tmp->is_valid(), "don't need temporary");
2822   emit_call_c(dest);
2823   CHECK_BAILOUT();
2824   if (info != NULL) {
2825     add_call_info_here(info);
2826   }
2827 }
2828 
2829 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {
2830   ShouldNotCallThis(); // not needed on ZARCH_64
2831 }
2832 
2833 void LIR_Assembler::membar() {
2834   __ z_fence();
2835 }
2836 
2837 void LIR_Assembler::membar_acquire() {
2838   __ z_acquire();
2839 }
2840 
2841 void LIR_Assembler::membar_release() {
2842   __ z_release();
2843 }
2844 
2845 void LIR_Assembler::membar_loadload() {
2846   __ z_acquire();
2847 }
2848 
2849 void LIR_Assembler::membar_storestore() {
2850   __ z_release();
2851 }
2852 
2853 void LIR_Assembler::membar_loadstore() {
2854   __ z_acquire();
2855 }
2856 
2857 void LIR_Assembler::membar_storeload() {
2858   __ z_fence();
2859 }
2860 
2861 void LIR_Assembler::on_spin_wait() {
2862   Unimplemented();
2863 }
2864 
2865 void LIR_Assembler::leal(LIR_Opr addr_opr, LIR_Opr dest) {
2866   LIR_Address* addr = addr_opr->as_address_ptr();
2867   assert(addr->scale() == LIR_Address::times_1, "scaling unsupported");
2868   __ load_address(dest->as_pointer_register(), as_Address(addr));
2869 }
2870 
2871 void LIR_Assembler::get_thread(LIR_Opr result_reg) {
2872   ShouldNotCallThis(); // unused
2873 }
2874 
2875 #ifdef ASSERT
2876 // Emit run-time assertion.
2877 void LIR_Assembler::emit_assert(LIR_OpAssert* op) {
2878   Unimplemented();
2879 }
2880 #endif
2881 
2882 void LIR_Assembler::peephole(LIR_List*) {
2883   // Do nothing for now.
2884 }
2885 
2886 void LIR_Assembler::atomic_op(LIR_Code code, LIR_Opr src, LIR_Opr data, LIR_Opr dest, LIR_Opr tmp) {
2887   assert(code == lir_xadd, "lir_xchg not supported");
2888   Address src_addr = as_Address(src->as_address_ptr());
2889   Register base = src_addr.base();
2890   intptr_t disp = src_addr.disp();
2891   if (src_addr.index()->is_valid()) {
2892     // LAA and LAAG do not support index register.
2893     __ load_address(Z_R1_scratch, src_addr);
2894     base = Z_R1_scratch;
2895     disp = 0;
2896   }
2897   if (data->type() == T_INT) {
2898     __ z_laa(dest->as_register(), data->as_register(), disp, base);
2899   } else if (data->type() == T_LONG) {
2900     assert(data->as_register_lo() == data->as_register_hi(), "should be a single register");
2901     __ z_laag(dest->as_register_lo(), data->as_register_lo(), disp, base);
2902   } else {
2903     ShouldNotReachHere();
2904   }
2905 }
2906 
2907 void LIR_Assembler::emit_profile_type(LIR_OpProfileType* op) {
2908   Register obj = op->obj()->as_register();
2909   Register tmp1 = op->tmp()->as_pointer_register();
2910   Register tmp2 = Z_R1_scratch;
2911   Address mdo_addr = as_Address(op->mdp()->as_address_ptr());
2912   ciKlass* exact_klass = op->exact_klass();
2913   intptr_t current_klass = op->current_klass();
2914   bool not_null = op->not_null();
2915   bool no_conflict = op->no_conflict();
2916 
2917   Label update, next, none, null_seen, init_klass;
2918 
2919   bool do_null = !not_null;
2920   bool exact_klass_set = exact_klass != NULL && ciTypeEntries::valid_ciklass(current_klass) == exact_klass;
2921   bool do_update = !TypeEntries::is_type_unknown(current_klass) && !exact_klass_set;
2922 
2923   assert(do_null || do_update, "why are we here?");
2924   assert(!TypeEntries::was_null_seen(current_klass) || do_update, "why are we here?");
2925 
2926   __ verify_oop(obj);
2927 
2928   if (do_null || tmp1 != obj DEBUG_ONLY(|| true)) {
2929     __ z_ltgr(tmp1, obj);
2930   }
2931   if (do_null) {
2932     __ z_brnz(update);
2933     if (!TypeEntries::was_null_seen(current_klass)) {
2934       __ z_lg(tmp1, mdo_addr);
2935       __ z_oill(tmp1, TypeEntries::null_seen);
2936       __ z_stg(tmp1, mdo_addr);
2937     }
2938     if (do_update) {
2939       __ z_bru(next);
2940     }
2941   } else {
2942     __ asm_assert_ne("unexpect null obj", __LINE__);
2943   }
2944 
2945   __ bind(update);
2946 
2947   if (do_update) {
2948 #ifdef ASSERT
2949     if (exact_klass != NULL) {
2950       __ load_klass(tmp1, tmp1);
2951       metadata2reg(exact_klass->constant_encoding(), tmp2);
2952       __ z_cgr(tmp1, tmp2);
2953       __ asm_assert_eq("exact klass and actual klass differ", __LINE__);
2954     }
2955 #endif
2956 
2957     Label do_update;
2958     __ z_lg(tmp2, mdo_addr);
2959 
2960     if (!no_conflict) {
2961       if (exact_klass == NULL || TypeEntries::is_type_none(current_klass)) {
2962         if (exact_klass != NULL) {
2963           metadata2reg(exact_klass->constant_encoding(), tmp1);
2964         } else {
2965           __ load_klass(tmp1, tmp1);
2966         }
2967 
2968         // Klass seen before: nothing to do (regardless of unknown bit).
2969         __ z_lgr(Z_R0_scratch, tmp2);
2970         assert(Immediate::is_uimm(~TypeEntries::type_klass_mask, 16), "or change following instruction");
2971         __ z_nill(Z_R0_scratch, TypeEntries::type_klass_mask & 0xFFFF);
2972         __ compareU64_and_branch(Z_R0_scratch, tmp1, Assembler::bcondEqual, next);
2973 
2974         // Already unknown: Nothing to do anymore.
2975         __ z_tmll(tmp2, TypeEntries::type_unknown);
2976         __ z_brc(Assembler::bcondAllOne, next);
2977 
2978         if (TypeEntries::is_type_none(current_klass)) {
2979           __ z_lgr(Z_R0_scratch, tmp2);
2980           assert(Immediate::is_uimm(~TypeEntries::type_mask, 16), "or change following instruction");
2981           __ z_nill(Z_R0_scratch, TypeEntries::type_mask & 0xFFFF);
2982           __ compareU64_and_branch(Z_R0_scratch, (intptr_t)0, Assembler::bcondEqual, init_klass);
2983         }
2984       } else {
2985         assert(ciTypeEntries::valid_ciklass(current_klass) != NULL &&
2986                ciTypeEntries::valid_ciklass(current_klass) != exact_klass, "conflict only");
2987 
2988         // Already unknown: Nothing to do anymore.
2989         __ z_tmll(tmp2, TypeEntries::type_unknown);
2990         __ z_brc(Assembler::bcondAllOne, next);
2991       }
2992 
2993       // Different than before. Cannot keep accurate profile.
2994       __ z_oill(tmp2, TypeEntries::type_unknown);
2995       __ z_bru(do_update);
2996     } else {
2997       // There's a single possible klass at this profile point.
2998       assert(exact_klass != NULL, "should be");
2999       if (TypeEntries::is_type_none(current_klass)) {
3000         metadata2reg(exact_klass->constant_encoding(), tmp1);
3001         __ z_lgr(Z_R0_scratch, tmp2);
3002         assert(Immediate::is_uimm(~TypeEntries::type_klass_mask, 16), "or change following instruction");
3003         __ z_nill(Z_R0_scratch, TypeEntries::type_klass_mask & 0xFFFF);
3004         __ compareU64_and_branch(Z_R0_scratch, tmp1, Assembler::bcondEqual, next);
3005 #ifdef ASSERT
3006         {
3007           Label ok;
3008           __ z_lgr(Z_R0_scratch, tmp2);
3009           assert(Immediate::is_uimm(~TypeEntries::type_mask, 16), "or change following instruction");
3010           __ z_nill(Z_R0_scratch, TypeEntries::type_mask & 0xFFFF);
3011           __ compareU64_and_branch(Z_R0_scratch, (intptr_t)0, Assembler::bcondEqual, ok);
3012           __ stop("unexpected profiling mismatch");
3013           __ bind(ok);
3014         }
3015 #endif
3016 
3017       } else {
3018         assert(ciTypeEntries::valid_ciklass(current_klass) != NULL &&
3019                ciTypeEntries::valid_ciklass(current_klass) != exact_klass, "inconsistent");
3020 
3021         // Already unknown: Nothing to do anymore.
3022         __ z_tmll(tmp2, TypeEntries::type_unknown);
3023         __ z_brc(Assembler::bcondAllOne, next);
3024         __ z_oill(tmp2, TypeEntries::type_unknown);
3025         __ z_bru(do_update);
3026       }
3027     }
3028 
3029     __ bind(init_klass);
3030     // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
3031     __ z_ogr(tmp2, tmp1);
3032 
3033     __ bind(do_update);
3034     __ z_stg(tmp2, mdo_addr);
3035 
3036     __ bind(next);
3037   }
3038 }
3039 
3040 void LIR_Assembler::emit_updatecrc32(LIR_OpUpdateCRC32* op) {
3041   assert(op->crc()->is_single_cpu(), "crc must be register");
3042   assert(op->val()->is_single_cpu(), "byte value must be register");
3043   assert(op->result_opr()->is_single_cpu(), "result must be register");
3044   Register crc = op->crc()->as_register();
3045   Register val = op->val()->as_register();
3046   Register res = op->result_opr()->as_register();
3047 
3048   assert_different_registers(val, crc, res);
3049 
3050   __ load_const_optimized(res, StubRoutines::crc_table_addr());
3051   __ not_(crc, noreg, false); // ~crc
3052   __ update_byte_crc32(crc, val, res);
3053   __ not_(res, crc, false); // ~crc
3054 }
3055 
3056 #undef __