1 /*
   2  * Copyright (c) 2016, 2024, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2016, 2024 SAP SE. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "precompiled.hpp"
  27 #include "asm/macroAssembler.inline.hpp"
  28 #include "c1/c1_Compilation.hpp"
  29 #include "c1/c1_LIRAssembler.hpp"
  30 #include "c1/c1_MacroAssembler.hpp"
  31 #include "c1/c1_Runtime1.hpp"
  32 #include "c1/c1_ValueStack.hpp"
  33 #include "ci/ciArrayKlass.hpp"
  34 #include "ci/ciInstance.hpp"
  35 #include "gc/shared/collectedHeap.hpp"
  36 #include "memory/universe.hpp"
  37 #include "nativeInst_s390.hpp"
  38 #include "oops/objArrayKlass.hpp"
  39 #include "runtime/frame.inline.hpp"
  40 #include "runtime/safepointMechanism.inline.hpp"
  41 #include "runtime/sharedRuntime.hpp"
  42 #include "runtime/stubRoutines.hpp"
  43 #include "utilities/macros.hpp"
  44 #include "utilities/powerOfTwo.hpp"
  45 #include "vmreg_s390.inline.hpp"
  46 
  47 #define __ _masm->
  48 
  49 #ifndef PRODUCT
  50 #undef __
  51 #define __ (Verbose ? (_masm->block_comment(FILE_AND_LINE),_masm) : _masm)->
  52 #endif
  53 
  54 //------------------------------------------------------------
  55 
  56 bool LIR_Assembler::is_small_constant(LIR_Opr opr) {
  57   // Not used on ZARCH_64
  58   ShouldNotCallThis();
  59   return false;
  60 }
  61 
  62 LIR_Opr LIR_Assembler::receiverOpr() {
  63   return FrameMap::Z_R2_oop_opr;
  64 }
  65 
  66 LIR_Opr LIR_Assembler::osrBufferPointer() {
  67   return FrameMap::Z_R2_opr;
  68 }
  69 
  70 int LIR_Assembler::initial_frame_size_in_bytes() const {
  71   return in_bytes(frame_map()->framesize_in_bytes());
  72 }
  73 
  74 // Inline cache check: done before the frame is built.
  75 // The inline cached class is in Z_inline_cache(Z_R9).
  76 // We fetch the class of the receiver and compare it with the cached class.
  77 // If they do not match we jump to the slow case.
  78 int LIR_Assembler::check_icache() {
  79   return __ ic_check(CodeEntryAlignment);
  80 }
  81 
  82 void LIR_Assembler::clinit_barrier(ciMethod* method) {
  83   assert(!method->holder()->is_not_initialized(), "initialization should have been started");
  84 
  85   Label L_skip_barrier;
  86   Register klass = Z_R1_scratch;
  87 
  88   metadata2reg(method->holder()->constant_encoding(), klass);
  89   __ clinit_barrier(klass, Z_thread, &L_skip_barrier /*L_fast_path*/);
  90 
  91   __ load_const_optimized(klass, SharedRuntime::get_handle_wrong_method_stub());
  92   __ z_br(klass);
  93 
  94   __ bind(L_skip_barrier);
  95 }
  96 
  97 void LIR_Assembler::osr_entry() {
  98   // On-stack-replacement entry sequence (interpreter frame layout described in frame_s390.hpp):
  99   //
 100   //   1. Create a new compiled activation.
 101   //   2. Initialize local variables in the compiled activation. The expression stack must be empty
 102   //      at the osr_bci; it is not initialized.
 103   //   3. Jump to the continuation address in compiled code to resume execution.
 104 
 105   // OSR entry point
 106   offsets()->set_value(CodeOffsets::OSR_Entry, code_offset());
 107   BlockBegin* osr_entry = compilation()->hir()->osr_entry();
 108   ValueStack* entry_state = osr_entry->end()->state();
 109   int number_of_locks = entry_state->locks_size();
 110 
 111   // Create a frame for the compiled activation.
 112   __ build_frame(initial_frame_size_in_bytes(), bang_size_in_bytes());
 113 
 114   // OSR buffer is
 115   //
 116   // locals[nlocals-1..0]
 117   // monitors[number_of_locks-1..0]
 118   //
 119   // Locals is a direct copy of the interpreter frame so in the osr buffer
 120   // the first slot in the local array is the last local from the interpreter
 121   // and the last slot is local[0] (receiver) from the interpreter
 122   //
 123   // Similarly with locks. The first lock slot in the osr buffer is the nth lock
 124   // from the interpreter frame, the nth lock slot in the osr buffer is 0th lock
 125   // in the interpreter frame (the method lock if a sync method)
 126 
 127   // Initialize monitors in the compiled activation.
 128   //   I0: pointer to osr buffer
 129   //
 130   // All other registers are dead at this point and the locals will be
 131   // copied into place by code emitted in the IR.
 132 
 133   Register OSR_buf = osrBufferPointer()->as_register();
 134   { assert(frame::interpreter_frame_monitor_size() == BasicObjectLock::size(), "adjust code below");
 135     int monitor_offset = BytesPerWord * method()->max_locals() +
 136       (2 * BytesPerWord) * (number_of_locks - 1);
 137     // SharedRuntime::OSR_migration_begin() packs BasicObjectLocks in
 138     // the OSR buffer using 2 word entries: first the lock and then
 139     // the oop.
 140     for (int i = 0; i < number_of_locks; i++) {
 141       int slot_offset = monitor_offset - ((i * 2) * BytesPerWord);
 142       // Verify the interpreter's monitor has a non-null object.
 143       __ asm_assert_mem8_isnot_zero(slot_offset + 1*BytesPerWord, OSR_buf, "locked object is null", __LINE__);
 144       // Copy the lock field into the compiled activation.
 145       __ z_lg(Z_R1_scratch, slot_offset + 0, OSR_buf);
 146       __ z_stg(Z_R1_scratch, frame_map()->address_for_monitor_lock(i));
 147       __ z_lg(Z_R1_scratch, slot_offset + 1*BytesPerWord, OSR_buf);
 148       __ z_stg(Z_R1_scratch, frame_map()->address_for_monitor_object(i));
 149     }
 150   }
 151 }
 152 
 153 // --------------------------------------------------------------------------------------------
 154 
 155 address LIR_Assembler::emit_call_c(address a) {
 156   __ align_call_far_patchable(__ pc());
 157   address call_addr = __ call_c_opt(a);
 158   if (call_addr == nullptr) {
 159     bailout("const section overflow");
 160   }
 161   return call_addr;
 162 }
 163 
 164 int LIR_Assembler::emit_exception_handler() {
 165   // Generate code for exception handler.
 166   address handler_base = __ start_a_stub(exception_handler_size());
 167   if (handler_base == nullptr) {
 168     // Not enough space left for the handler.
 169     bailout("exception handler overflow");
 170     return -1;
 171   }
 172 
 173   int offset = code_offset();
 174 
 175   address a = Runtime1::entry_for (C1StubId::handle_exception_from_callee_id);
 176   address call_addr = emit_call_c(a);
 177   CHECK_BAILOUT_(-1);
 178   __ should_not_reach_here();
 179   guarantee(code_offset() - offset <= exception_handler_size(), "overflow");
 180   __ end_a_stub();
 181 
 182   return offset;
 183 }
 184 
 185 // Emit the code to remove the frame from the stack in the exception
 186 // unwind path.
 187 int LIR_Assembler::emit_unwind_handler() {
 188 #ifndef PRODUCT
 189   if (CommentedAssembly) {
 190     _masm->block_comment("Unwind handler");
 191   }
 192 #endif
 193 
 194   int offset = code_offset();
 195   Register exception_oop_callee_saved = Z_R10; // Z_R10 is callee-saved.
 196   Register Rtmp1                      = Z_R11;
 197   Register Rtmp2                      = Z_R12;
 198 
 199   // Fetch the exception from TLS and clear out exception related thread state.
 200   Address exc_oop_addr = Address(Z_thread, JavaThread::exception_oop_offset());
 201   Address exc_pc_addr  = Address(Z_thread, JavaThread::exception_pc_offset());
 202   __ z_lg(Z_EXC_OOP, exc_oop_addr);
 203   __ clear_mem(exc_oop_addr, sizeof(oop));
 204   __ clear_mem(exc_pc_addr, sizeof(intptr_t));
 205 
 206   __ bind(_unwind_handler_entry);
 207   __ verify_not_null_oop(Z_EXC_OOP);
 208   if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
 209     __ lgr_if_needed(exception_oop_callee_saved, Z_EXC_OOP); // Preserve the exception.
 210   }
 211 
 212   // Perform needed unlocking.
 213   MonitorExitStub* stub = nullptr;
 214   if (method()->is_synchronized()) {
 215     // C1StubId::monitorexit_id expects lock address in Z_R1_scratch.
 216     LIR_Opr lock = FrameMap::as_opr(Z_R1_scratch);
 217     monitor_address(0, lock);
 218     stub = new MonitorExitStub(lock, true, 0);
 219     __ unlock_object(Rtmp1, Rtmp2, lock->as_register(), *stub->entry());
 220     __ bind(*stub->continuation());
 221   }
 222 
 223   if (compilation()->env()->dtrace_method_probes()) {
 224     ShouldNotReachHere(); // Not supported.
 225 #if 0
 226     __ mov(rdi, r15_thread);
 227     __ mov_metadata(rsi, method()->constant_encoding());
 228     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit)));
 229 #endif
 230   }
 231 
 232   if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
 233     __ lgr_if_needed(Z_EXC_OOP, exception_oop_callee_saved);  // Restore the exception.
 234   }
 235 
 236   // Remove the activation and dispatch to the unwind handler.
 237   __ pop_frame();
 238   __ z_lg(Z_EXC_PC, _z_common_abi(return_pc), Z_SP);
 239 
 240   // Z_EXC_OOP: exception oop
 241   // Z_EXC_PC: exception pc
 242 
 243   // Dispatch to the unwind logic.
 244   __ load_const_optimized(Z_R5, Runtime1::entry_for (C1StubId::unwind_exception_id));
 245   __ z_br(Z_R5);
 246 
 247   // Emit the slow path assembly.
 248   if (stub != nullptr) {
 249     stub->emit_code(this);
 250   }
 251 
 252   return offset;
 253 }
 254 
 255 int LIR_Assembler::emit_deopt_handler() {
 256   // Generate code for exception handler.
 257   address handler_base = __ start_a_stub(deopt_handler_size());
 258   if (handler_base == nullptr) {
 259     // Not enough space left for the handler.
 260     bailout("deopt handler overflow");
 261     return -1;
 262   }  int offset = code_offset();
 263   // Size must be constant (see HandlerImpl::emit_deopt_handler).
 264   __ load_const(Z_R1_scratch, SharedRuntime::deopt_blob()->unpack());
 265   __ call(Z_R1_scratch);
 266   guarantee(code_offset() - offset <= deopt_handler_size(), "overflow");
 267   __ end_a_stub();
 268 
 269   return offset;
 270 }
 271 
 272 void LIR_Assembler::jobject2reg(jobject o, Register reg) {
 273   if (o == nullptr) {
 274     __ clear_reg(reg, true/*64bit*/, false/*set cc*/); // Must not kill cc set by cmove.
 275   } else {
 276     AddressLiteral a = __ allocate_oop_address(o);
 277     bool success = __ load_oop_from_toc(reg, a, reg);
 278     if (!success) {
 279       bailout("const section overflow");
 280     }
 281   }
 282 }
 283 
 284 void LIR_Assembler::jobject2reg_with_patching(Register reg, CodeEmitInfo *info) {
 285   // Allocate a new index in table to hold the object once it's been patched.
 286   int oop_index = __ oop_recorder()->allocate_oop_index(nullptr);
 287   PatchingStub* patch = new PatchingStub(_masm, patching_id(info), oop_index);
 288 
 289   AddressLiteral addrlit((intptr_t)0, oop_Relocation::spec(oop_index));
 290   assert(addrlit.rspec().type() == relocInfo::oop_type, "must be an oop reloc");
 291   // The null will be dynamically patched later so the sequence to
 292   // load the address literal must not be optimized.
 293   __ load_const(reg, addrlit);
 294 
 295   patching_epilog(patch, lir_patch_normal, reg, info);
 296 }
 297 
 298 void LIR_Assembler::metadata2reg(Metadata* md, Register reg) {
 299   bool success = __ set_metadata_constant(md, reg);
 300   if (!success) {
 301     bailout("const section overflow");
 302     return;
 303   }
 304 }
 305 
 306 void LIR_Assembler::klass2reg_with_patching(Register reg, CodeEmitInfo *info) {
 307   // Allocate a new index in table to hold the klass once it's been patched.
 308   int index = __ oop_recorder()->allocate_metadata_index(nullptr);
 309   PatchingStub* patch = new PatchingStub(_masm, PatchingStub::load_klass_id, index);
 310   AddressLiteral addrlit((intptr_t)0, metadata_Relocation::spec(index));
 311   assert(addrlit.rspec().type() == relocInfo::metadata_type, "must be an metadata reloc");
 312   // The null will be dynamically patched later so the sequence to
 313   // load the address literal must not be optimized.
 314   __ load_const(reg, addrlit);
 315 
 316   patching_epilog(patch, lir_patch_normal, reg, info);
 317 }
 318 
 319 void LIR_Assembler::emit_op3(LIR_Op3* op) {
 320   switch (op->code()) {
 321     case lir_idiv:
 322     case lir_irem:
 323       arithmetic_idiv(op->code(),
 324                       op->in_opr1(),
 325                       op->in_opr2(),
 326                       op->in_opr3(),
 327                       op->result_opr(),
 328                       op->info());
 329       break;
 330     case lir_fmad: {
 331       const FloatRegister opr1 = op->in_opr1()->as_double_reg(),
 332                           opr2 = op->in_opr2()->as_double_reg(),
 333                           opr3 = op->in_opr3()->as_double_reg(),
 334                           res  = op->result_opr()->as_double_reg();
 335       __ z_madbr(opr3, opr1, opr2);
 336       if (res != opr3) { __ z_ldr(res, opr3); }
 337     } break;
 338     case lir_fmaf: {
 339       const FloatRegister opr1 = op->in_opr1()->as_float_reg(),
 340                           opr2 = op->in_opr2()->as_float_reg(),
 341                           opr3 = op->in_opr3()->as_float_reg(),
 342                           res  = op->result_opr()->as_float_reg();
 343       __ z_maebr(opr3, opr1, opr2);
 344       if (res != opr3) { __ z_ler(res, opr3); }
 345     } break;
 346     default: ShouldNotReachHere(); break;
 347   }
 348 }
 349 
 350 
 351 void LIR_Assembler::emit_opBranch(LIR_OpBranch* op) {
 352 #ifdef ASSERT
 353   assert(op->block() == nullptr || op->block()->label() == op->label(), "wrong label");
 354   if (op->block() != nullptr)  { _branch_target_blocks.append(op->block()); }
 355   if (op->ublock() != nullptr) { _branch_target_blocks.append(op->ublock()); }
 356 #endif
 357 
 358   if (op->cond() == lir_cond_always) {
 359     if (op->info() != nullptr) { add_debug_info_for_branch(op->info()); }
 360     __ branch_optimized(Assembler::bcondAlways, *(op->label()));
 361   } else {
 362     Assembler::branch_condition acond = Assembler::bcondZero;
 363     if (op->code() == lir_cond_float_branch) {
 364       assert(op->ublock() != nullptr, "must have unordered successor");
 365       __ branch_optimized(Assembler::bcondNotOrdered, *(op->ublock()->label()));
 366     }
 367     switch (op->cond()) {
 368       case lir_cond_equal:        acond = Assembler::bcondEqual;     break;
 369       case lir_cond_notEqual:     acond = Assembler::bcondNotEqual;  break;
 370       case lir_cond_less:         acond = Assembler::bcondLow;       break;
 371       case lir_cond_lessEqual:    acond = Assembler::bcondNotHigh;   break;
 372       case lir_cond_greaterEqual: acond = Assembler::bcondNotLow;    break;
 373       case lir_cond_greater:      acond = Assembler::bcondHigh;      break;
 374       case lir_cond_belowEqual:   acond = Assembler::bcondNotHigh;   break;
 375       case lir_cond_aboveEqual:   acond = Assembler::bcondNotLow;    break;
 376       default:                         ShouldNotReachHere();
 377     }
 378     __ branch_optimized(acond,*(op->label()));
 379   }
 380 }
 381 
 382 
 383 void LIR_Assembler::emit_opConvert(LIR_OpConvert* op) {
 384   LIR_Opr src  = op->in_opr();
 385   LIR_Opr dest = op->result_opr();
 386 
 387   switch (op->bytecode()) {
 388     case Bytecodes::_i2l:
 389       __ move_reg_if_needed(dest->as_register_lo(), T_LONG, src->as_register(), T_INT);
 390       break;
 391 
 392     case Bytecodes::_l2i:
 393       __ move_reg_if_needed(dest->as_register(), T_INT, src->as_register_lo(), T_LONG);
 394       break;
 395 
 396     case Bytecodes::_i2b:
 397       __ move_reg_if_needed(dest->as_register(), T_BYTE, src->as_register(), T_INT);
 398       break;
 399 
 400     case Bytecodes::_i2c:
 401       __ move_reg_if_needed(dest->as_register(), T_CHAR, src->as_register(), T_INT);
 402       break;
 403 
 404     case Bytecodes::_i2s:
 405       __ move_reg_if_needed(dest->as_register(), T_SHORT, src->as_register(), T_INT);
 406       break;
 407 
 408     case Bytecodes::_f2d:
 409       assert(dest->is_double_fpu(), "check");
 410       __ move_freg_if_needed(dest->as_double_reg(), T_DOUBLE, src->as_float_reg(), T_FLOAT);
 411       break;
 412 
 413     case Bytecodes::_d2f:
 414       assert(dest->is_single_fpu(), "check");
 415       __ move_freg_if_needed(dest->as_float_reg(), T_FLOAT, src->as_double_reg(), T_DOUBLE);
 416       break;
 417 
 418     case Bytecodes::_i2f:
 419       __ z_cefbr(dest->as_float_reg(), src->as_register());
 420       break;
 421 
 422     case Bytecodes::_i2d:
 423       __ z_cdfbr(dest->as_double_reg(), src->as_register());
 424       break;
 425 
 426     case Bytecodes::_l2f:
 427       __ z_cegbr(dest->as_float_reg(), src->as_register_lo());
 428       break;
 429     case Bytecodes::_l2d:
 430       __ z_cdgbr(dest->as_double_reg(), src->as_register_lo());
 431       break;
 432 
 433     case Bytecodes::_f2i:
 434     case Bytecodes::_f2l: {
 435       Label done;
 436       FloatRegister Rsrc = src->as_float_reg();
 437       Register Rdst = (op->bytecode() == Bytecodes::_f2i ? dest->as_register() : dest->as_register_lo());
 438       __ clear_reg(Rdst, true, false);
 439       __ z_cebr(Rsrc, Rsrc);
 440       __ z_brno(done); // NaN -> 0
 441       if (op->bytecode() == Bytecodes::_f2i) {
 442         __ z_cfebr(Rdst, Rsrc, Assembler::to_zero);
 443       } else { // op->bytecode() == Bytecodes::_f2l
 444         __ z_cgebr(Rdst, Rsrc, Assembler::to_zero);
 445       }
 446       __ bind(done);
 447     }
 448     break;
 449 
 450     case Bytecodes::_d2i:
 451     case Bytecodes::_d2l: {
 452       Label done;
 453       FloatRegister Rsrc = src->as_double_reg();
 454       Register Rdst = (op->bytecode() == Bytecodes::_d2i ? dest->as_register() : dest->as_register_lo());
 455       __ clear_reg(Rdst, true, false);  // Don't set CC.
 456       __ z_cdbr(Rsrc, Rsrc);
 457       __ z_brno(done); // NaN -> 0
 458       if (op->bytecode() == Bytecodes::_d2i) {
 459         __ z_cfdbr(Rdst, Rsrc, Assembler::to_zero);
 460       } else { // Bytecodes::_d2l
 461         __ z_cgdbr(Rdst, Rsrc, Assembler::to_zero);
 462       }
 463       __ bind(done);
 464     }
 465     break;
 466 
 467     default: ShouldNotReachHere();
 468   }
 469 }
 470 
 471 void LIR_Assembler::align_call(LIR_Code code) {
 472   // End of call instruction must be 4 byte aligned.
 473   int offset = __ offset();
 474   switch (code) {
 475     case lir_icvirtual_call:
 476       offset += MacroAssembler::load_const_from_toc_size();
 477       // no break
 478     case lir_static_call:
 479     case lir_optvirtual_call:
 480     case lir_dynamic_call:
 481       offset += NativeCall::call_far_pcrelative_displacement_offset;
 482       break;
 483     default: ShouldNotReachHere();
 484   }
 485   if ((offset & (NativeCall::call_far_pcrelative_displacement_alignment-1)) != 0) {
 486     __ nop();
 487   }
 488 }
 489 
 490 void LIR_Assembler::call(LIR_OpJavaCall* op, relocInfo::relocType rtype) {
 491   assert((__ offset() + NativeCall::call_far_pcrelative_displacement_offset) % NativeCall::call_far_pcrelative_displacement_alignment == 0,
 492          "must be aligned (offset=%d)", __ offset());
 493   assert(rtype == relocInfo::none ||
 494          rtype == relocInfo::opt_virtual_call_type ||
 495          rtype == relocInfo::static_call_type, "unexpected rtype");
 496   // Prepend each BRASL with a nop.
 497   __ relocate(rtype);
 498   __ z_nop();
 499   __ z_brasl(Z_R14, op->addr());
 500   add_call_info(code_offset(), op->info());
 501 }
 502 
 503 void LIR_Assembler::ic_call(LIR_OpJavaCall* op) {
 504   address virtual_call_oop_addr = nullptr;
 505   AddressLiteral empty_ic((address) Universe::non_oop_word());
 506   virtual_call_oop_addr = __ pc();
 507   bool success = __ load_const_from_toc(Z_inline_cache, empty_ic);
 508   if (!success) {
 509     bailout("const section overflow");
 510     return;
 511   }
 512 
 513   // CALL to fixup routine. Fixup routine uses ScopeDesc info
 514   // to determine who we intended to call.
 515   __ relocate(virtual_call_Relocation::spec(virtual_call_oop_addr));
 516   call(op, relocInfo::none);
 517 }
 518 
 519 void LIR_Assembler::move_regs(Register from_reg, Register to_reg) {
 520   if (from_reg != to_reg) __ z_lgr(to_reg, from_reg);
 521 }
 522 
 523 void LIR_Assembler::const2stack(LIR_Opr src, LIR_Opr dest) {
 524   assert(src->is_constant(), "should not call otherwise");
 525   assert(dest->is_stack(), "should not call otherwise");
 526   LIR_Const* c = src->as_constant_ptr();
 527 
 528   unsigned int lmem = 0;
 529   unsigned int lcon = 0;
 530   int64_t cbits = 0;
 531   Address dest_addr;
 532   switch (c->type()) {
 533     case T_INT:  // fall through
 534     case T_FLOAT:
 535       dest_addr = frame_map()->address_for_slot(dest->single_stack_ix());
 536       lmem = 4; lcon = 4; cbits = c->as_jint_bits();
 537       break;
 538 
 539     case T_ADDRESS:
 540       dest_addr = frame_map()->address_for_slot(dest->single_stack_ix());
 541       lmem = 8; lcon = 4; cbits = c->as_jint_bits();
 542       break;
 543 
 544     case T_OBJECT:
 545       dest_addr = frame_map()->address_for_slot(dest->single_stack_ix());
 546       if (c->as_jobject() == nullptr) {
 547         __ store_const(dest_addr, (int64_t)NULL_WORD, 8, 8);
 548       } else {
 549         jobject2reg(c->as_jobject(), Z_R1_scratch);
 550         __ reg2mem_opt(Z_R1_scratch, dest_addr, true);
 551       }
 552       return;
 553 
 554     case T_LONG:  // fall through
 555     case T_DOUBLE:
 556       dest_addr = frame_map()->address_for_slot(dest->double_stack_ix());
 557       lmem = 8; lcon = 8; cbits = (int64_t)(c->as_jlong_bits());
 558       break;
 559 
 560     default:
 561       ShouldNotReachHere();
 562   }
 563 
 564   __ store_const(dest_addr, cbits, lmem, lcon);
 565 }
 566 
 567 void LIR_Assembler::const2mem(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info, bool wide) {
 568   assert(src->is_constant(), "should not call otherwise");
 569   assert(dest->is_address(), "should not call otherwise");
 570 
 571   LIR_Const* c = src->as_constant_ptr();
 572   Address addr = as_Address(dest->as_address_ptr());
 573 
 574   int store_offset = -1;
 575 
 576   if (dest->as_address_ptr()->index()->is_valid()) {
 577     switch (type) {
 578       case T_INT:    // fall through
 579       case T_FLOAT:
 580         __ load_const_optimized(Z_R0_scratch, c->as_jint_bits());
 581         store_offset = __ offset();
 582         if (Immediate::is_uimm12(addr.disp())) {
 583           __ z_st(Z_R0_scratch, addr);
 584         } else {
 585           __ z_sty(Z_R0_scratch, addr);
 586         }
 587         break;
 588 
 589       case T_ADDRESS:
 590         __ load_const_optimized(Z_R1_scratch, c->as_jint_bits());
 591         store_offset = __ reg2mem_opt(Z_R1_scratch, addr, true);
 592         break;
 593 
 594       case T_OBJECT:  // fall through
 595       case T_ARRAY:
 596         if (c->as_jobject() == nullptr) {
 597           if (UseCompressedOops && !wide) {
 598             __ clear_reg(Z_R1_scratch, false);
 599             store_offset = __ reg2mem_opt(Z_R1_scratch, addr, false);
 600           } else {
 601             __ clear_reg(Z_R1_scratch, true);
 602             store_offset = __ reg2mem_opt(Z_R1_scratch, addr, true);
 603           }
 604         } else {
 605           jobject2reg(c->as_jobject(), Z_R1_scratch);
 606           if (UseCompressedOops && !wide) {
 607             __ encode_heap_oop(Z_R1_scratch);
 608             store_offset = __ reg2mem_opt(Z_R1_scratch, addr, false);
 609           } else {
 610             store_offset = __ reg2mem_opt(Z_R1_scratch, addr, true);
 611           }
 612         }
 613         assert(store_offset >= 0, "check");
 614         break;
 615 
 616       case T_LONG:    // fall through
 617       case T_DOUBLE:
 618         __ load_const_optimized(Z_R1_scratch, (int64_t)(c->as_jlong_bits()));
 619         store_offset = __ reg2mem_opt(Z_R1_scratch, addr, true);
 620         break;
 621 
 622       case T_BOOLEAN: // fall through
 623       case T_BYTE:
 624         __ load_const_optimized(Z_R0_scratch, (int8_t)(c->as_jint()));
 625         store_offset = __ offset();
 626         if (Immediate::is_uimm12(addr.disp())) {
 627           __ z_stc(Z_R0_scratch, addr);
 628         } else {
 629           __ z_stcy(Z_R0_scratch, addr);
 630         }
 631         break;
 632 
 633       case T_CHAR:    // fall through
 634       case T_SHORT:
 635         __ load_const_optimized(Z_R0_scratch, (int16_t)(c->as_jint()));
 636         store_offset = __ offset();
 637         if (Immediate::is_uimm12(addr.disp())) {
 638           __ z_sth(Z_R0_scratch, addr);
 639         } else {
 640           __ z_sthy(Z_R0_scratch, addr);
 641         }
 642         break;
 643 
 644       default:
 645         ShouldNotReachHere();
 646     }
 647 
 648   } else { // no index
 649 
 650     unsigned int lmem = 0;
 651     unsigned int lcon = 0;
 652     int64_t cbits = 0;
 653 
 654     switch (type) {
 655       case T_INT:    // fall through
 656       case T_FLOAT:
 657         lmem = 4; lcon = 4; cbits = c->as_jint_bits();
 658         break;
 659 
 660       case T_ADDRESS:
 661         lmem = 8; lcon = 4; cbits = c->as_jint_bits();
 662         break;
 663 
 664       case T_OBJECT:  // fall through
 665       case T_ARRAY:
 666         if (c->as_jobject() == nullptr) {
 667           if (UseCompressedOops && !wide) {
 668             store_offset = __ store_const(addr, (int32_t)NULL_WORD, 4, 4);
 669           } else {
 670             store_offset = __ store_const(addr, (int64_t)NULL_WORD, 8, 8);
 671           }
 672         } else {
 673           jobject2reg(c->as_jobject(), Z_R1_scratch);
 674           if (UseCompressedOops && !wide) {
 675             __ encode_heap_oop(Z_R1_scratch);
 676             store_offset = __ reg2mem_opt(Z_R1_scratch, addr, false);
 677           } else {
 678             store_offset = __ reg2mem_opt(Z_R1_scratch, addr, true);
 679           }
 680         }
 681         assert(store_offset >= 0, "check");
 682         break;
 683 
 684       case T_LONG:    // fall through
 685       case T_DOUBLE:
 686         lmem = 8; lcon = 8; cbits = (int64_t)(c->as_jlong_bits());
 687         break;
 688 
 689       case T_BOOLEAN: // fall through
 690       case T_BYTE:
 691         lmem = 1; lcon = 1; cbits = (int8_t)(c->as_jint());
 692         break;
 693 
 694       case T_CHAR:    // fall through
 695       case T_SHORT:
 696         lmem = 2; lcon = 2; cbits = (int16_t)(c->as_jint());
 697         break;
 698 
 699       default:
 700         ShouldNotReachHere();
 701     }
 702 
 703     if (store_offset == -1) {
 704       store_offset = __ store_const(addr, cbits, lmem, lcon);
 705       assert(store_offset >= 0, "check");
 706     }
 707   }
 708 
 709   if (info != nullptr) {
 710     add_debug_info_for_null_check(store_offset, info);
 711   }
 712 }
 713 
 714 void LIR_Assembler::const2reg(LIR_Opr src, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
 715   assert(src->is_constant(), "should not call otherwise");
 716   assert(dest->is_register(), "should not call otherwise");
 717   LIR_Const* c = src->as_constant_ptr();
 718 
 719   switch (c->type()) {
 720     case T_INT: {
 721       assert(patch_code == lir_patch_none, "no patching handled here");
 722       __ load_const_optimized(dest->as_register(), c->as_jint());
 723       break;
 724     }
 725 
 726     case T_ADDRESS: {
 727       assert(patch_code == lir_patch_none, "no patching handled here");
 728       __ load_const_optimized(dest->as_register(), c->as_jint());
 729       break;
 730     }
 731 
 732     case T_LONG: {
 733       assert(patch_code == lir_patch_none, "no patching handled here");
 734       __ load_const_optimized(dest->as_register_lo(), (intptr_t)c->as_jlong());
 735       break;
 736     }
 737 
 738     case T_OBJECT: {
 739       if (patch_code != lir_patch_none) {
 740         jobject2reg_with_patching(dest->as_register(), info);
 741       } else {
 742         jobject2reg(c->as_jobject(), dest->as_register());
 743       }
 744       break;
 745     }
 746 
 747     case T_METADATA: {
 748       if (patch_code != lir_patch_none) {
 749         klass2reg_with_patching(dest->as_register(), info);
 750       } else {
 751         metadata2reg(c->as_metadata(), dest->as_register());
 752       }
 753       break;
 754     }
 755 
 756     case T_FLOAT: {
 757       Register toc_reg = Z_R1_scratch;
 758       __ load_toc(toc_reg);
 759       address const_addr = __ float_constant(c->as_jfloat());
 760       if (const_addr == nullptr) {
 761         bailout("const section overflow");
 762         break;
 763       }
 764       int displ = const_addr - _masm->code()->consts()->start();
 765       if (dest->is_single_fpu()) {
 766         __ z_ley(dest->as_float_reg(), displ, toc_reg);
 767       } else {
 768         assert(dest->is_single_cpu(), "Must be a cpu register.");
 769         __ z_ly(dest->as_register(), displ, toc_reg);
 770       }
 771     }
 772     break;
 773 
 774     case T_DOUBLE: {
 775       Register toc_reg = Z_R1_scratch;
 776       __ load_toc(toc_reg);
 777       address const_addr = __ double_constant(c->as_jdouble());
 778       if (const_addr == nullptr) {
 779         bailout("const section overflow");
 780         break;
 781       }
 782       int displ = const_addr - _masm->code()->consts()->start();
 783       if (dest->is_double_fpu()) {
 784         __ z_ldy(dest->as_double_reg(), displ, toc_reg);
 785       } else {
 786         assert(dest->is_double_cpu(), "Must be a long register.");
 787         __ z_lg(dest->as_register_lo(), displ, toc_reg);
 788       }
 789     }
 790     break;
 791 
 792     default:
 793       ShouldNotReachHere();
 794   }
 795 }
 796 
 797 Address LIR_Assembler::as_Address(LIR_Address* addr) {
 798   if (addr->base()->is_illegal()) {
 799     Unimplemented();
 800   }
 801 
 802   Register base = addr->base()->as_pointer_register();
 803 
 804   if (addr->index()->is_illegal()) {
 805     return Address(base, addr->disp());
 806   } else if (addr->index()->is_cpu_register()) {
 807     Register index = addr->index()->as_pointer_register();
 808     return Address(base, index, addr->disp());
 809   } else if (addr->index()->is_constant()) {
 810     intptr_t addr_offset = addr->index()->as_constant_ptr()->as_jint() + addr->disp();
 811     return Address(base, addr_offset);
 812   } else {
 813     ShouldNotReachHere();
 814     return Address();
 815   }
 816 }
 817 
 818 void LIR_Assembler::stack2stack(LIR_Opr src, LIR_Opr dest, BasicType type) {
 819   switch (type) {
 820     case T_INT:
 821     case T_FLOAT: {
 822       Register tmp = Z_R1_scratch;
 823       Address from = frame_map()->address_for_slot(src->single_stack_ix());
 824       Address to   = frame_map()->address_for_slot(dest->single_stack_ix());
 825       __ mem2reg_opt(tmp, from, false);
 826       __ reg2mem_opt(tmp, to, false);
 827       break;
 828     }
 829     case T_ADDRESS:
 830     case T_OBJECT: {
 831       Register tmp = Z_R1_scratch;
 832       Address from = frame_map()->address_for_slot(src->single_stack_ix());
 833       Address to   = frame_map()->address_for_slot(dest->single_stack_ix());
 834       __ mem2reg_opt(tmp, from, true);
 835       __ reg2mem_opt(tmp, to, true);
 836       break;
 837     }
 838     case T_LONG:
 839     case T_DOUBLE: {
 840       Register tmp = Z_R1_scratch;
 841       Address from = frame_map()->address_for_double_slot(src->double_stack_ix());
 842       Address to   = frame_map()->address_for_double_slot(dest->double_stack_ix());
 843       __ mem2reg_opt(tmp, from, true);
 844       __ reg2mem_opt(tmp, to, true);
 845       break;
 846     }
 847 
 848     default:
 849       ShouldNotReachHere();
 850   }
 851 }
 852 
 853 // 4-byte accesses only! Don't use it to access 8 bytes!
 854 Address LIR_Assembler::as_Address_hi(LIR_Address* addr) {
 855   ShouldNotCallThis();
 856   return Address(); // unused
 857 }
 858 
 859 // 4-byte accesses only! Don't use it to access 8 bytes!
 860 Address LIR_Assembler::as_Address_lo(LIR_Address* addr) {
 861   ShouldNotCallThis();
 862   return Address(); // unused
 863 }
 864 
 865 void LIR_Assembler::mem2reg(LIR_Opr src_opr, LIR_Opr dest, BasicType type, LIR_PatchCode patch_code,
 866                             CodeEmitInfo* info, bool wide) {
 867 
 868   assert(type != T_METADATA, "load of metadata ptr not supported");
 869   LIR_Address* addr = src_opr->as_address_ptr();
 870   LIR_Opr to_reg = dest;
 871 
 872   Register src = addr->base()->as_pointer_register();
 873   Register disp_reg = Z_R0;
 874   int disp_value = addr->disp();
 875   bool needs_patching = (patch_code != lir_patch_none);
 876 
 877   if (addr->base()->type() == T_OBJECT) {
 878     __ verify_oop(src, FILE_AND_LINE);
 879   }
 880 
 881   PatchingStub* patch = nullptr;
 882   if (needs_patching) {
 883     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 884     assert(!to_reg->is_double_cpu() ||
 885            patch_code == lir_patch_none ||
 886            patch_code == lir_patch_normal, "patching doesn't match register");
 887   }
 888 
 889   if (addr->index()->is_illegal()) {
 890     if (!Immediate::is_simm20(disp_value)) {
 891       if (needs_patching) {
 892         __ load_const(Z_R1_scratch, (intptr_t)0);
 893       } else {
 894         __ load_const_optimized(Z_R1_scratch, disp_value);
 895       }
 896       disp_reg = Z_R1_scratch;
 897       disp_value = 0;
 898     }
 899   } else {
 900     if (!Immediate::is_simm20(disp_value)) {
 901       __ load_const_optimized(Z_R1_scratch, disp_value);
 902       __ z_la(Z_R1_scratch, 0, Z_R1_scratch, addr->index()->as_register());
 903       disp_reg = Z_R1_scratch;
 904       disp_value = 0;
 905     }
 906     disp_reg = addr->index()->as_pointer_register();
 907   }
 908 
 909   // Remember the offset of the load. The patching_epilog must be done
 910   // before the call to add_debug_info, otherwise the PcDescs don't get
 911   // entered in increasing order.
 912   int offset = code_offset();
 913 
 914   assert(disp_reg != Z_R0 || Immediate::is_simm20(disp_value), "should have set this up");
 915 
 916   bool short_disp = Immediate::is_uimm12(disp_value);
 917 
 918   switch (type) {
 919     case T_BOOLEAN: // fall through
 920     case T_BYTE  :  __ z_lb(dest->as_register(),   disp_value, disp_reg, src); break;
 921     case T_CHAR  :  __ z_llgh(dest->as_register(), disp_value, disp_reg, src); break;
 922     case T_SHORT :
 923       if (short_disp) {
 924                     __ z_lh(dest->as_register(),   disp_value, disp_reg, src);
 925       } else {
 926                     __ z_lhy(dest->as_register(),  disp_value, disp_reg, src);
 927       }
 928       break;
 929     case T_INT   :
 930       if (short_disp) {
 931                     __ z_l(dest->as_register(),    disp_value, disp_reg, src);
 932       } else {
 933                     __ z_ly(dest->as_register(),   disp_value, disp_reg, src);
 934       }
 935       break;
 936     case T_ADDRESS:
 937       __ z_lg(dest->as_register(), disp_value, disp_reg, src);
 938       break;
 939     case T_ARRAY : // fall through
 940     case T_OBJECT:
 941     {
 942       if (UseCompressedOops && !wide) {
 943         __ z_llgf(dest->as_register(), disp_value, disp_reg, src);
 944         __ oop_decoder(dest->as_register(), dest->as_register(), true);
 945       } else {
 946         __ z_lg(dest->as_register(), disp_value, disp_reg, src);
 947       }
 948       __ verify_oop(dest->as_register(), FILE_AND_LINE);
 949       break;
 950     }
 951     case T_FLOAT:
 952       if (short_disp) {
 953                     __ z_le(dest->as_float_reg(),  disp_value, disp_reg, src);
 954       } else {
 955                     __ z_ley(dest->as_float_reg(), disp_value, disp_reg, src);
 956       }
 957       break;
 958     case T_DOUBLE:
 959       if (short_disp) {
 960                     __ z_ld(dest->as_double_reg(),  disp_value, disp_reg, src);
 961       } else {
 962                     __ z_ldy(dest->as_double_reg(), disp_value, disp_reg, src);
 963       }
 964       break;
 965     case T_LONG  :  __ z_lg(dest->as_register_lo(), disp_value, disp_reg, src); break;
 966     default      : ShouldNotReachHere();
 967   }
 968 
 969   if (patch != nullptr) {
 970     patching_epilog(patch, patch_code, src, info);
 971   }
 972   if (info != nullptr) add_debug_info_for_null_check(offset, info);
 973 }
 974 
 975 void LIR_Assembler::stack2reg(LIR_Opr src, LIR_Opr dest, BasicType type) {
 976   assert(src->is_stack(), "should not call otherwise");
 977   assert(dest->is_register(), "should not call otherwise");
 978 
 979   if (dest->is_single_cpu()) {
 980     if (is_reference_type(type)) {
 981       __ mem2reg_opt(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()), true);
 982       __ verify_oop(dest->as_register(), FILE_AND_LINE);
 983     } else if (type == T_METADATA || type == T_ADDRESS) {
 984       __ mem2reg_opt(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()), true);
 985     } else {
 986       __ mem2reg_opt(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()), false);
 987     }
 988   } else if (dest->is_double_cpu()) {
 989     Address src_addr_LO = frame_map()->address_for_slot(src->double_stack_ix());
 990     __ mem2reg_opt(dest->as_register_lo(), src_addr_LO, true);
 991   } else if (dest->is_single_fpu()) {
 992     Address src_addr = frame_map()->address_for_slot(src->single_stack_ix());
 993     __ mem2freg_opt(dest->as_float_reg(), src_addr, false);
 994   } else if (dest->is_double_fpu()) {
 995     Address src_addr = frame_map()->address_for_slot(src->double_stack_ix());
 996     __ mem2freg_opt(dest->as_double_reg(), src_addr, true);
 997   } else {
 998     ShouldNotReachHere();
 999   }
1000 }
1001 
1002 void LIR_Assembler::reg2stack(LIR_Opr src, LIR_Opr dest, BasicType type, bool pop_fpu_stack) {
1003   assert(src->is_register(), "should not call otherwise");
1004   assert(dest->is_stack(), "should not call otherwise");
1005 
1006   if (src->is_single_cpu()) {
1007     const Address dst = frame_map()->address_for_slot(dest->single_stack_ix());
1008     if (is_reference_type(type)) {
1009       __ verify_oop(src->as_register(), FILE_AND_LINE);
1010       __ reg2mem_opt(src->as_register(), dst, true);
1011     } else if (type == T_METADATA || type == T_ADDRESS) {
1012       __ reg2mem_opt(src->as_register(), dst, true);
1013     } else {
1014       __ reg2mem_opt(src->as_register(), dst, false);
1015     }
1016   } else if (src->is_double_cpu()) {
1017     Address dstLO = frame_map()->address_for_slot(dest->double_stack_ix());
1018     __ reg2mem_opt(src->as_register_lo(), dstLO, true);
1019   } else if (src->is_single_fpu()) {
1020     Address dst_addr = frame_map()->address_for_slot(dest->single_stack_ix());
1021     __ freg2mem_opt(src->as_float_reg(), dst_addr, false);
1022   } else if (src->is_double_fpu()) {
1023     Address dst_addr = frame_map()->address_for_slot(dest->double_stack_ix());
1024     __ freg2mem_opt(src->as_double_reg(), dst_addr, true);
1025   } else {
1026     ShouldNotReachHere();
1027   }
1028 }
1029 
1030 void LIR_Assembler::reg2reg(LIR_Opr from_reg, LIR_Opr to_reg) {
1031   if (from_reg->is_float_kind() && to_reg->is_float_kind()) {
1032     if (from_reg->is_double_fpu()) {
1033       // double to double moves
1034       assert(to_reg->is_double_fpu(), "should match");
1035       __ z_ldr(to_reg->as_double_reg(), from_reg->as_double_reg());
1036     } else {
1037       // float to float moves
1038       assert(to_reg->is_single_fpu(), "should match");
1039       __ z_ler(to_reg->as_float_reg(), from_reg->as_float_reg());
1040     }
1041   } else if (!from_reg->is_float_kind() && !to_reg->is_float_kind()) {
1042     if (from_reg->is_double_cpu()) {
1043       __ z_lgr(to_reg->as_pointer_register(), from_reg->as_pointer_register());
1044     } else if (to_reg->is_double_cpu()) {
1045       // int to int moves
1046       __ z_lgr(to_reg->as_register_lo(), from_reg->as_register());
1047     } else {
1048       // int to int moves
1049       __ z_lgr(to_reg->as_register(), from_reg->as_register());
1050     }
1051   } else {
1052     ShouldNotReachHere();
1053   }
1054   if (is_reference_type(to_reg->type())) {
1055     __ verify_oop(to_reg->as_register(), FILE_AND_LINE);
1056   }
1057 }
1058 
1059 void LIR_Assembler::reg2mem(LIR_Opr from, LIR_Opr dest_opr, BasicType type,
1060                             LIR_PatchCode patch_code, CodeEmitInfo* info, bool pop_fpu_stack,
1061                             bool wide) {
1062   assert(type != T_METADATA, "store of metadata ptr not supported");
1063   LIR_Address* addr = dest_opr->as_address_ptr();
1064 
1065   Register dest = addr->base()->as_pointer_register();
1066   Register disp_reg = Z_R0;
1067   int disp_value = addr->disp();
1068   bool needs_patching = (patch_code != lir_patch_none);
1069 
1070   if (addr->base()->is_oop_register()) {
1071     __ verify_oop(dest, FILE_AND_LINE);
1072   }
1073 
1074   PatchingStub* patch = nullptr;
1075   if (needs_patching) {
1076     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1077     assert(!from->is_double_cpu() ||
1078            patch_code == lir_patch_none ||
1079            patch_code == lir_patch_normal, "patching doesn't match register");
1080   }
1081 
1082   assert(!needs_patching || (!Immediate::is_simm20(disp_value) && addr->index()->is_illegal()), "assumption");
1083   if (addr->index()->is_illegal()) {
1084     if (!Immediate::is_simm20(disp_value)) {
1085       if (needs_patching) {
1086         __ load_const(Z_R1_scratch, (intptr_t)0);
1087       } else {
1088         __ load_const_optimized(Z_R1_scratch, disp_value);
1089       }
1090       disp_reg = Z_R1_scratch;
1091       disp_value = 0;
1092     }
1093   } else {
1094     if (!Immediate::is_simm20(disp_value)) {
1095       __ load_const_optimized(Z_R1_scratch, disp_value);
1096       __ z_la(Z_R1_scratch, 0, Z_R1_scratch, addr->index()->as_register());
1097       disp_reg = Z_R1_scratch;
1098       disp_value = 0;
1099     }
1100     disp_reg = addr->index()->as_pointer_register();
1101   }
1102 
1103   assert(disp_reg != Z_R0 || Immediate::is_simm20(disp_value), "should have set this up");
1104 
1105   if (is_reference_type(type)) {
1106     __ verify_oop(from->as_register(), FILE_AND_LINE);
1107   }
1108 
1109   bool short_disp = Immediate::is_uimm12(disp_value);
1110 
1111   // Remember the offset of the store. The patching_epilog must be done
1112   // before the call to add_debug_info_for_null_check, otherwise the PcDescs don't get
1113   // entered in increasing order.
1114   int offset = code_offset();
1115   switch (type) {
1116     case T_BOOLEAN: // fall through
1117     case T_BYTE  :
1118       if (short_disp) {
1119                     __ z_stc(from->as_register(),  disp_value, disp_reg, dest);
1120       } else {
1121                     __ z_stcy(from->as_register(), disp_value, disp_reg, dest);
1122       }
1123       break;
1124     case T_CHAR  : // fall through
1125     case T_SHORT :
1126       if (short_disp) {
1127                     __ z_sth(from->as_register(),  disp_value, disp_reg, dest);
1128       } else {
1129                     __ z_sthy(from->as_register(), disp_value, disp_reg, dest);
1130       }
1131       break;
1132     case T_INT   :
1133       if (short_disp) {
1134                     __ z_st(from->as_register(),  disp_value, disp_reg, dest);
1135       } else {
1136                     __ z_sty(from->as_register(), disp_value, disp_reg, dest);
1137       }
1138       break;
1139     case T_LONG  :  __ z_stg(from->as_register_lo(), disp_value, disp_reg, dest); break;
1140     case T_ADDRESS: __ z_stg(from->as_register(),    disp_value, disp_reg, dest); break;
1141       break;
1142     case T_ARRAY : // fall through
1143     case T_OBJECT:
1144       {
1145         if (UseCompressedOops && !wide) {
1146           Register compressed_src = Z_R14;
1147           __ oop_encoder(compressed_src, from->as_register(), true, (disp_reg != Z_R1) ? Z_R1 : Z_R0, -1, true);
1148           offset = code_offset();
1149           if (short_disp) {
1150             __ z_st(compressed_src,  disp_value, disp_reg, dest);
1151           } else {
1152             __ z_sty(compressed_src, disp_value, disp_reg, dest);
1153           }
1154         } else {
1155           __ z_stg(from->as_register(), disp_value, disp_reg, dest);
1156         }
1157         break;
1158       }
1159     case T_FLOAT :
1160       if (short_disp) {
1161         __ z_ste(from->as_float_reg(),  disp_value, disp_reg, dest);
1162       } else {
1163         __ z_stey(from->as_float_reg(), disp_value, disp_reg, dest);
1164       }
1165       break;
1166     case T_DOUBLE:
1167       if (short_disp) {
1168         __ z_std(from->as_double_reg(),  disp_value, disp_reg, dest);
1169       } else {
1170         __ z_stdy(from->as_double_reg(), disp_value, disp_reg, dest);
1171       }
1172       break;
1173     default: ShouldNotReachHere();
1174   }
1175 
1176   if (patch != nullptr) {
1177     patching_epilog(patch, patch_code, dest, info);
1178   }
1179 
1180   if (info != nullptr) add_debug_info_for_null_check(offset, info);
1181 }
1182 
1183 
1184 void LIR_Assembler::return_op(LIR_Opr result, C1SafepointPollStub* code_stub) {
1185   assert(result->is_illegal() ||
1186          (result->is_single_cpu() && result->as_register() == Z_R2) ||
1187          (result->is_double_cpu() && result->as_register_lo() == Z_R2) ||
1188          (result->is_single_fpu() && result->as_float_reg() == Z_F0) ||
1189          (result->is_double_fpu() && result->as_double_reg() == Z_F0), "convention");
1190 
1191   __ z_lg(Z_R1_scratch, Address(Z_thread, JavaThread::polling_page_offset()));
1192 
1193   // Pop the frame before the safepoint code.
1194   __ pop_frame_restore_retPC(initial_frame_size_in_bytes());
1195 
1196   if (StackReservedPages > 0 && compilation()->has_reserved_stack_access()) {
1197     __ reserved_stack_check(Z_R14);
1198   }
1199 
1200   // We need to mark the code position where the load from the safepoint
1201   // polling page was emitted as relocInfo::poll_return_type here.
1202   __ relocate(relocInfo::poll_return_type);
1203   __ load_from_polling_page(Z_R1_scratch);
1204 
1205   __ z_br(Z_R14); // Return to caller.
1206 }
1207 
1208 int LIR_Assembler::safepoint_poll(LIR_Opr tmp, CodeEmitInfo* info) {
1209   const Register poll_addr = tmp->as_register_lo();
1210   __ z_lg(poll_addr, Address(Z_thread, JavaThread::polling_page_offset()));
1211   guarantee(info != nullptr, "Shouldn't be null");
1212   add_debug_info_for_branch(info);
1213   int offset = __ offset();
1214   __ relocate(relocInfo::poll_type);
1215   __ load_from_polling_page(poll_addr);
1216   return offset;
1217 }
1218 
1219 void LIR_Assembler::emit_static_call_stub() {
1220 
1221   // Stub is fixed up when the corresponding call is converted from calling
1222   // compiled code to calling interpreted code.
1223 
1224   address call_pc = __ pc();
1225   address stub = __ start_a_stub(call_stub_size());
1226   if (stub == nullptr) {
1227     bailout("static call stub overflow");
1228     return;
1229   }
1230 
1231   int start = __ offset();
1232 
1233   __ relocate(static_stub_Relocation::spec(call_pc));
1234 
1235   // See also Matcher::interpreter_method_reg().
1236   AddressLiteral meta = __ allocate_metadata_address(nullptr);
1237   bool success = __ load_const_from_toc(Z_method, meta);
1238 
1239   __ set_inst_mark();
1240   AddressLiteral a((address)-1);
1241   success = success && __ load_const_from_toc(Z_R1, a);
1242   if (!success) {
1243     bailout("const section overflow");
1244     return;
1245   }
1246 
1247   __ z_br(Z_R1);
1248   assert(__ offset() - start <= call_stub_size(), "stub too big");
1249   __ end_a_stub(); // Update current stubs pointer and restore insts_end.
1250 }
1251 
1252 void LIR_Assembler::comp_op(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Op2* op) {
1253   bool unsigned_comp = condition == lir_cond_belowEqual || condition == lir_cond_aboveEqual;
1254   if (opr1->is_single_cpu()) {
1255     Register reg1 = opr1->as_register();
1256     if (opr2->is_single_cpu()) {
1257       // cpu register - cpu register
1258       if (is_reference_type(opr1->type())) {
1259         __ z_clgr(reg1, opr2->as_register());
1260       } else {
1261         assert(!is_reference_type(opr2->type()), "cmp int, oop?");
1262         if (unsigned_comp) {
1263           __ z_clr(reg1, opr2->as_register());
1264         } else {
1265           __ z_cr(reg1, opr2->as_register());
1266         }
1267       }
1268     } else if (opr2->is_stack()) {
1269       // cpu register - stack
1270       if (is_reference_type(opr1->type())) {
1271         __ z_cg(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
1272       } else {
1273         if (unsigned_comp) {
1274           __ z_cly(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
1275         } else {
1276           __ z_cy(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
1277         }
1278       }
1279     } else if (opr2->is_constant()) {
1280       // cpu register - constant
1281       LIR_Const* c = opr2->as_constant_ptr();
1282       if (c->type() == T_INT) {
1283         if (unsigned_comp) {
1284           __ z_clfi(reg1, c->as_jint());
1285         } else {
1286           __ z_cfi(reg1, c->as_jint());
1287         }
1288       } else if (c->type() == T_METADATA) {
1289         // We only need, for now, comparison with null for metadata.
1290         assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "oops");
1291         Metadata* m = c->as_metadata();
1292         if (m == nullptr) {
1293           __ z_cghi(reg1, 0);
1294         } else {
1295           ShouldNotReachHere();
1296         }
1297       } else if (is_reference_type(c->type())) {
1298         // In 64bit oops are single register.
1299         jobject o = c->as_jobject();
1300         if (o == nullptr) {
1301           __ z_ltgr(reg1, reg1);
1302         } else {
1303           jobject2reg(o, Z_R1_scratch);
1304           __ z_cgr(reg1, Z_R1_scratch);
1305         }
1306       } else {
1307         fatal("unexpected type: %s", basictype_to_str(c->type()));
1308       }
1309       // cpu register - address
1310     } else if (opr2->is_address()) {
1311       if (op->info() != nullptr) {
1312         add_debug_info_for_null_check_here(op->info());
1313       }
1314       if (unsigned_comp) {
1315         __ z_cly(reg1, as_Address(opr2->as_address_ptr()));
1316       } else {
1317         __ z_cy(reg1, as_Address(opr2->as_address_ptr()));
1318       }
1319     } else {
1320       ShouldNotReachHere();
1321     }
1322 
1323   } else if (opr1->is_double_cpu()) {
1324     assert(!unsigned_comp, "unexpected");
1325     Register xlo = opr1->as_register_lo();
1326     Register xhi = opr1->as_register_hi();
1327     if (opr2->is_double_cpu()) {
1328       __ z_cgr(xlo, opr2->as_register_lo());
1329     } else if (opr2->is_constant()) {
1330       // cpu register - constant 0
1331       assert(opr2->as_jlong() == (jlong)0, "only handles zero");
1332       __ z_ltgr(xlo, xlo);
1333     } else {
1334       ShouldNotReachHere();
1335     }
1336 
1337   } else if (opr1->is_single_fpu()) {
1338     if (opr2->is_single_fpu()) {
1339       __ z_cebr(opr1->as_float_reg(), opr2->as_float_reg());
1340     } else {
1341       // stack slot
1342       Address addr = frame_map()->address_for_slot(opr2->single_stack_ix());
1343       if (Immediate::is_uimm12(addr.disp())) {
1344         __ z_ceb(opr1->as_float_reg(), addr);
1345       } else {
1346         __ z_ley(Z_fscratch_1, addr);
1347         __ z_cebr(opr1->as_float_reg(), Z_fscratch_1);
1348       }
1349     }
1350   } else if (opr1->is_double_fpu()) {
1351     if (opr2->is_double_fpu()) {
1352     __ z_cdbr(opr1->as_double_reg(), opr2->as_double_reg());
1353     } else {
1354       // stack slot
1355       Address addr = frame_map()->address_for_slot(opr2->double_stack_ix());
1356       if (Immediate::is_uimm12(addr.disp())) {
1357         __ z_cdb(opr1->as_double_reg(), addr);
1358       } else {
1359         __ z_ldy(Z_fscratch_1, addr);
1360         __ z_cdbr(opr1->as_double_reg(), Z_fscratch_1);
1361       }
1362     }
1363   } else {
1364     ShouldNotReachHere();
1365   }
1366 }
1367 
1368 void LIR_Assembler::comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst, LIR_Op2* op) {
1369   Label    done;
1370   Register dreg = dst->as_register();
1371 
1372   if (code == lir_cmp_fd2i || code == lir_ucmp_fd2i) {
1373     assert((left->is_single_fpu() && right->is_single_fpu()) ||
1374            (left->is_double_fpu() && right->is_double_fpu()), "unexpected operand types");
1375     bool is_single = left->is_single_fpu();
1376     bool is_unordered_less = (code == lir_ucmp_fd2i);
1377     FloatRegister lreg = is_single ? left->as_float_reg() : left->as_double_reg();
1378     FloatRegister rreg = is_single ? right->as_float_reg() : right->as_double_reg();
1379     if (is_single) {
1380       __ z_cebr(lreg, rreg);
1381     } else {
1382       __ z_cdbr(lreg, rreg);
1383     }
1384     if (VM_Version::has_LoadStoreConditional()) {
1385       Register one       = Z_R0_scratch;
1386       Register minus_one = Z_R1_scratch;
1387       __ z_lghi(minus_one, -1);
1388       __ z_lghi(one,  1);
1389       __ z_lghi(dreg, 0);
1390       __ z_locgr(dreg, one,       is_unordered_less ? Assembler::bcondHigh            : Assembler::bcondHighOrNotOrdered);
1391       __ z_locgr(dreg, minus_one, is_unordered_less ? Assembler::bcondLowOrNotOrdered : Assembler::bcondLow);
1392     } else {
1393       __ clear_reg(dreg, true, false);
1394       __ z_bre(done); // if (left == right) dst = 0
1395 
1396       // if (left > right || ((code ~= cmpg) && (left <> right)) dst := 1
1397       __ z_lhi(dreg, 1);
1398       __ z_brc(is_unordered_less ? Assembler::bcondHigh : Assembler::bcondHighOrNotOrdered, done);
1399 
1400       // if (left < right || ((code ~= cmpl) && (left <> right)) dst := -1
1401       __ z_lhi(dreg, -1);
1402     }
1403   } else {
1404     assert(code == lir_cmp_l2i, "check");
1405     if (VM_Version::has_LoadStoreConditional()) {
1406       Register one       = Z_R0_scratch;
1407       Register minus_one = Z_R1_scratch;
1408       __ z_cgr(left->as_register_lo(), right->as_register_lo());
1409       __ z_lghi(minus_one, -1);
1410       __ z_lghi(one,  1);
1411       __ z_lghi(dreg, 0);
1412       __ z_locgr(dreg, one, Assembler::bcondHigh);
1413       __ z_locgr(dreg, minus_one, Assembler::bcondLow);
1414     } else {
1415       __ z_cgr(left->as_register_lo(), right->as_register_lo());
1416       __ z_lghi(dreg,  0);     // eq value
1417       __ z_bre(done);
1418       __ z_lghi(dreg,  1);     // gt value
1419       __ z_brh(done);
1420       __ z_lghi(dreg, -1);     // lt value
1421     }
1422   }
1423   __ bind(done);
1424 }
1425 
1426 // result = condition ? opr1 : opr2
1427 void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type,
1428                           LIR_Opr cmp_opr1, LIR_Opr cmp_opr2) {
1429   assert(cmp_opr1 == LIR_OprFact::illegalOpr && cmp_opr2 == LIR_OprFact::illegalOpr, "unnecessary cmp oprs on s390");
1430 
1431   Assembler::branch_condition acond = Assembler::bcondEqual, ncond = Assembler::bcondNotEqual;
1432   switch (condition) {
1433     case lir_cond_equal:        acond = Assembler::bcondEqual;    ncond = Assembler::bcondNotEqual; break;
1434     case lir_cond_notEqual:     acond = Assembler::bcondNotEqual; ncond = Assembler::bcondEqual;    break;
1435     case lir_cond_less:         acond = Assembler::bcondLow;      ncond = Assembler::bcondNotLow;   break;
1436     case lir_cond_lessEqual:    acond = Assembler::bcondNotHigh;  ncond = Assembler::bcondHigh;     break;
1437     case lir_cond_greaterEqual: acond = Assembler::bcondNotLow;   ncond = Assembler::bcondLow;      break;
1438     case lir_cond_greater:      acond = Assembler::bcondHigh;     ncond = Assembler::bcondNotHigh;  break;
1439     case lir_cond_belowEqual:   acond = Assembler::bcondNotHigh;  ncond = Assembler::bcondHigh;     break;
1440     case lir_cond_aboveEqual:   acond = Assembler::bcondNotLow;   ncond = Assembler::bcondLow;      break;
1441     default:                    ShouldNotReachHere();
1442   }
1443 
1444   if (opr1->is_cpu_register()) {
1445     reg2reg(opr1, result);
1446   } else if (opr1->is_stack()) {
1447     stack2reg(opr1, result, result->type());
1448   } else if (opr1->is_constant()) {
1449     const2reg(opr1, result, lir_patch_none, nullptr);
1450   } else {
1451     ShouldNotReachHere();
1452   }
1453 
1454   if (VM_Version::has_LoadStoreConditional() && !opr2->is_constant()) {
1455     // Optimized version that does not require a branch.
1456     if (opr2->is_single_cpu()) {
1457       assert(opr2->cpu_regnr() != result->cpu_regnr(), "opr2 already overwritten by previous move");
1458       __ z_locgr(result->as_register(), opr2->as_register(), ncond);
1459     } else if (opr2->is_double_cpu()) {
1460       assert(opr2->cpu_regnrLo() != result->cpu_regnrLo() && opr2->cpu_regnrLo() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
1461       assert(opr2->cpu_regnrHi() != result->cpu_regnrLo() && opr2->cpu_regnrHi() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
1462       __ z_locgr(result->as_register_lo(), opr2->as_register_lo(), ncond);
1463     } else if (opr2->is_single_stack()) {
1464       __ z_loc(result->as_register(), frame_map()->address_for_slot(opr2->single_stack_ix()), ncond);
1465     } else if (opr2->is_double_stack()) {
1466       __ z_locg(result->as_register_lo(), frame_map()->address_for_slot(opr2->double_stack_ix()), ncond);
1467     } else {
1468       ShouldNotReachHere();
1469     }
1470   } else {
1471     Label skip;
1472     __ z_brc(acond, skip);
1473     if (opr2->is_cpu_register()) {
1474       reg2reg(opr2, result);
1475     } else if (opr2->is_stack()) {
1476       stack2reg(opr2, result, result->type());
1477     } else if (opr2->is_constant()) {
1478       const2reg(opr2, result, lir_patch_none, nullptr);
1479     } else {
1480       ShouldNotReachHere();
1481     }
1482     __ bind(skip);
1483   }
1484 }
1485 
1486 void LIR_Assembler::arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest,
1487                              CodeEmitInfo* info, bool pop_fpu_stack) {
1488   assert(info == nullptr, "should never be used, idiv/irem and ldiv/lrem not handled by this method");
1489 
1490   if (left->is_single_cpu()) {
1491     assert(left == dest, "left and dest must be equal");
1492     Register lreg = left->as_register();
1493 
1494     if (right->is_single_cpu()) {
1495       // cpu register - cpu register
1496       Register rreg = right->as_register();
1497       switch (code) {
1498         case lir_add: __ z_ar (lreg, rreg); break;
1499         case lir_sub: __ z_sr (lreg, rreg); break;
1500         case lir_mul: __ z_msr(lreg, rreg); break;
1501         default: ShouldNotReachHere();
1502       }
1503 
1504     } else if (right->is_stack()) {
1505       // cpu register - stack
1506       Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
1507       switch (code) {
1508         case lir_add: __ z_ay(lreg, raddr); break;
1509         case lir_sub: __ z_sy(lreg, raddr); break;
1510         default: ShouldNotReachHere();
1511       }
1512 
1513     } else if (right->is_constant()) {
1514       // cpu register - constant
1515       jint c = right->as_constant_ptr()->as_jint();
1516       switch (code) {
1517         case lir_add: __ z_agfi(lreg, c);  break;
1518         case lir_sub: __ z_agfi(lreg, -c); break; // note: -min_jint == min_jint
1519         case lir_mul: __ z_msfi(lreg, c);  break;
1520         default: ShouldNotReachHere();
1521       }
1522 
1523     } else {
1524       ShouldNotReachHere();
1525     }
1526 
1527   } else if (left->is_double_cpu()) {
1528     assert(left == dest, "left and dest must be equal");
1529     Register lreg_lo = left->as_register_lo();
1530     Register lreg_hi = left->as_register_hi();
1531 
1532     if (right->is_double_cpu()) {
1533       // cpu register - cpu register
1534       Register rreg_lo = right->as_register_lo();
1535       Register rreg_hi = right->as_register_hi();
1536       assert_different_registers(lreg_lo, rreg_lo);
1537       switch (code) {
1538         case lir_add:
1539           __ z_agr(lreg_lo, rreg_lo);
1540           break;
1541         case lir_sub:
1542           __ z_sgr(lreg_lo, rreg_lo);
1543           break;
1544         case lir_mul:
1545           __ z_msgr(lreg_lo, rreg_lo);
1546           break;
1547         default:
1548           ShouldNotReachHere();
1549       }
1550 
1551     } else if (right->is_constant()) {
1552       // cpu register - constant
1553       jlong c = right->as_constant_ptr()->as_jlong_bits();
1554       switch (code) {
1555         case lir_add: __ z_agfi(lreg_lo, c); break;
1556         case lir_sub:
1557           if (c != min_jint) {
1558                       __ z_agfi(lreg_lo, -c);
1559           } else {
1560             // -min_jint cannot be represented as simm32 in z_agfi
1561             // min_jint sign extended:      0xffffffff80000000
1562             // -min_jint as 64 bit integer: 0x0000000080000000
1563             // 0x80000000 can be represented as uimm32 in z_algfi
1564             // lreg_lo := lreg_lo + -min_jint == lreg_lo + 0x80000000
1565                       __ z_algfi(lreg_lo, UCONST64(0x80000000));
1566           }
1567           break;
1568         case lir_mul: __ z_msgfi(lreg_lo, c); break;
1569         default:
1570           ShouldNotReachHere();
1571       }
1572 
1573     } else {
1574       ShouldNotReachHere();
1575     }
1576 
1577   } else if (left->is_single_fpu()) {
1578     assert(left == dest, "left and dest must be equal");
1579     FloatRegister lreg = left->as_float_reg();
1580     FloatRegister rreg = right->is_single_fpu() ? right->as_float_reg() : fnoreg;
1581     Address raddr;
1582 
1583     if (rreg == fnoreg) {
1584       assert(right->is_single_stack(), "constants should be loaded into register");
1585       raddr = frame_map()->address_for_slot(right->single_stack_ix());
1586       if (!Immediate::is_uimm12(raddr.disp())) {
1587         __ mem2freg_opt(rreg = Z_fscratch_1, raddr, false);
1588       }
1589     }
1590 
1591     if (rreg != fnoreg) {
1592       switch (code) {
1593         case lir_add: __ z_aebr(lreg, rreg);  break;
1594         case lir_sub: __ z_sebr(lreg, rreg);  break;
1595         case lir_mul: __ z_meebr(lreg, rreg); break;
1596         case lir_div: __ z_debr(lreg, rreg);  break;
1597         default: ShouldNotReachHere();
1598       }
1599     } else {
1600       switch (code) {
1601         case lir_add: __ z_aeb(lreg, raddr);  break;
1602         case lir_sub: __ z_seb(lreg, raddr);  break;
1603         case lir_mul: __ z_meeb(lreg, raddr);  break;
1604         case lir_div: __ z_deb(lreg, raddr);  break;
1605         default: ShouldNotReachHere();
1606       }
1607     }
1608   } else if (left->is_double_fpu()) {
1609     assert(left == dest, "left and dest must be equal");
1610     FloatRegister lreg = left->as_double_reg();
1611     FloatRegister rreg = right->is_double_fpu() ? right->as_double_reg() : fnoreg;
1612     Address raddr;
1613 
1614     if (rreg == fnoreg) {
1615       assert(right->is_double_stack(), "constants should be loaded into register");
1616       raddr = frame_map()->address_for_slot(right->double_stack_ix());
1617       if (!Immediate::is_uimm12(raddr.disp())) {
1618         __ mem2freg_opt(rreg = Z_fscratch_1, raddr, true);
1619       }
1620     }
1621 
1622     if (rreg != fnoreg) {
1623       switch (code) {
1624         case lir_add: __ z_adbr(lreg, rreg); break;
1625         case lir_sub: __ z_sdbr(lreg, rreg); break;
1626         case lir_mul: __ z_mdbr(lreg, rreg); break;
1627         case lir_div: __ z_ddbr(lreg, rreg); break;
1628         default: ShouldNotReachHere();
1629       }
1630     } else {
1631       switch (code) {
1632         case lir_add: __ z_adb(lreg, raddr); break;
1633         case lir_sub: __ z_sdb(lreg, raddr); break;
1634         case lir_mul: __ z_mdb(lreg, raddr); break;
1635         case lir_div: __ z_ddb(lreg, raddr); break;
1636         default: ShouldNotReachHere();
1637       }
1638     }
1639   } else if (left->is_address()) {
1640     assert(left == dest, "left and dest must be equal");
1641     assert(code == lir_add, "unsupported operation");
1642     assert(right->is_constant(), "unsupported operand");
1643     jint c = right->as_constant_ptr()->as_jint();
1644     LIR_Address* lir_addr = left->as_address_ptr();
1645     Address addr = as_Address(lir_addr);
1646     switch (lir_addr->type()) {
1647       case T_INT:
1648         __ add2mem_32(addr, c, Z_R1_scratch);
1649         break;
1650       case T_LONG:
1651         __ add2mem_64(addr, c, Z_R1_scratch);
1652         break;
1653       default:
1654         ShouldNotReachHere();
1655     }
1656   } else {
1657     ShouldNotReachHere();
1658   }
1659 }
1660 
1661 void LIR_Assembler::intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr thread, LIR_Opr dest, LIR_Op* op) {
1662   switch (code) {
1663     case lir_sqrt: {
1664       assert(!thread->is_valid(), "there is no need for a thread_reg for dsqrt");
1665       FloatRegister src_reg = value->as_double_reg();
1666       FloatRegister dst_reg = dest->as_double_reg();
1667       __ z_sqdbr(dst_reg, src_reg);
1668       break;
1669     }
1670     case lir_abs: {
1671       assert(!thread->is_valid(), "there is no need for a thread_reg for fabs");
1672       FloatRegister src_reg = value->as_double_reg();
1673       FloatRegister dst_reg = dest->as_double_reg();
1674       __ z_lpdbr(dst_reg, src_reg);
1675       break;
1676     }
1677     default: {
1678       ShouldNotReachHere();
1679       break;
1680     }
1681   }
1682 }
1683 
1684 void LIR_Assembler::logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst) {
1685   if (left->is_single_cpu()) {
1686     Register reg = left->as_register();
1687     if (right->is_constant()) {
1688       int val = right->as_constant_ptr()->as_jint();
1689       switch (code) {
1690         case lir_logic_and: __ z_nilf(reg, val); break;
1691         case lir_logic_or:  __ z_oilf(reg, val); break;
1692         case lir_logic_xor: __ z_xilf(reg, val); break;
1693         default: ShouldNotReachHere();
1694       }
1695     } else if (right->is_stack()) {
1696       Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
1697       switch (code) {
1698         case lir_logic_and: __ z_ny(reg, raddr); break;
1699         case lir_logic_or:  __ z_oy(reg, raddr); break;
1700         case lir_logic_xor: __ z_xy(reg, raddr); break;
1701         default: ShouldNotReachHere();
1702       }
1703     } else {
1704       Register rright = right->as_register();
1705       switch (code) {
1706         case lir_logic_and: __ z_nr(reg, rright); break;
1707         case lir_logic_or : __ z_or(reg, rright); break;
1708         case lir_logic_xor: __ z_xr(reg, rright); break;
1709         default: ShouldNotReachHere();
1710       }
1711     }
1712     move_regs(reg, dst->as_register());
1713   } else {
1714     Register l_lo = left->as_register_lo();
1715     if (right->is_constant()) {
1716       __ load_const_optimized(Z_R1_scratch, right->as_constant_ptr()->as_jlong());
1717       switch (code) {
1718         case lir_logic_and:
1719           __ z_ngr(l_lo, Z_R1_scratch);
1720           break;
1721         case lir_logic_or:
1722           __ z_ogr(l_lo, Z_R1_scratch);
1723           break;
1724         case lir_logic_xor:
1725           __ z_xgr(l_lo, Z_R1_scratch);
1726           break;
1727         default: ShouldNotReachHere();
1728       }
1729     } else {
1730       Register r_lo;
1731       if (is_reference_type(right->type())) {
1732         r_lo = right->as_register();
1733       } else {
1734         r_lo = right->as_register_lo();
1735       }
1736       switch (code) {
1737         case lir_logic_and:
1738           __ z_ngr(l_lo, r_lo);
1739           break;
1740         case lir_logic_or:
1741           __ z_ogr(l_lo, r_lo);
1742           break;
1743         case lir_logic_xor:
1744           __ z_xgr(l_lo, r_lo);
1745           break;
1746         default: ShouldNotReachHere();
1747       }
1748     }
1749 
1750     Register dst_lo = dst->as_register_lo();
1751 
1752     move_regs(l_lo, dst_lo);
1753   }
1754 }
1755 
1756 // See operand selection in LIRGenerator::do_ArithmeticOp_Int().
1757 void LIR_Assembler::arithmetic_idiv(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr temp, LIR_Opr result, CodeEmitInfo* info) {
1758   if (left->is_double_cpu()) {
1759     // 64 bit integer case
1760     assert(left->is_double_cpu(), "left must be register");
1761     assert(right->is_double_cpu() || is_power_of_2(right->as_jlong()),
1762            "right must be register or power of 2 constant");
1763     assert(result->is_double_cpu(), "result must be register");
1764 
1765     Register lreg = left->as_register_lo();
1766     Register dreg = result->as_register_lo();
1767 
1768     if (right->is_constant()) {
1769       // Convert division by a power of two into some shifts and logical operations.
1770       Register treg1 = Z_R0_scratch;
1771       Register treg2 = Z_R1_scratch;
1772       jlong divisor = right->as_jlong();
1773       jlong log_divisor = log2i_exact(right->as_jlong());
1774 
1775       if (divisor == min_jlong) {
1776         // Min_jlong is special. Result is '0' except for min_jlong/min_jlong = 1.
1777         if (dreg == lreg) {
1778           NearLabel done;
1779           __ load_const_optimized(treg2, min_jlong);
1780           __ z_cgr(lreg, treg2);
1781           __ z_lghi(dreg, 0);           // Preserves condition code.
1782           __ z_brne(done);
1783           __ z_lghi(dreg, 1);           // min_jlong / min_jlong = 1
1784           __ bind(done);
1785         } else {
1786           assert_different_registers(dreg, lreg);
1787           NearLabel done;
1788           __ z_lghi(dreg, 0);
1789           __ compare64_and_branch(lreg, min_jlong, Assembler::bcondNotEqual, done);
1790           __ z_lghi(dreg, 1);
1791           __ bind(done);
1792         }
1793         return;
1794       }
1795       __ move_reg_if_needed(dreg, T_LONG, lreg, T_LONG);
1796       if (divisor == 2) {
1797         __ z_srlg(treg2, dreg, 63);     // dividend < 0 ? 1 : 0
1798       } else {
1799         __ z_srag(treg2, dreg, 63);     // dividend < 0 ? -1 : 0
1800         __ and_imm(treg2, divisor - 1, treg1, true);
1801       }
1802       if (code == lir_idiv) {
1803         __ z_agr(dreg, treg2);
1804         __ z_srag(dreg, dreg, log_divisor);
1805       } else {
1806         assert(code == lir_irem, "check");
1807         __ z_agr(treg2, dreg);
1808         __ and_imm(treg2, ~(divisor - 1), treg1, true);
1809         __ z_sgr(dreg, treg2);
1810       }
1811       return;
1812     }
1813 
1814     // Divisor is not a power of 2 constant.
1815     Register rreg = right->as_register_lo();
1816     Register treg = temp->as_register_lo();
1817     assert(right->is_double_cpu(), "right must be register");
1818     assert(lreg == Z_R11, "see ldivInOpr()");
1819     assert(rreg != lreg, "right register must not be same as left register");
1820     assert((code == lir_idiv && dreg == Z_R11 && treg == Z_R10) ||
1821            (code == lir_irem && dreg == Z_R10 && treg == Z_R11), "see ldivInOpr(), ldivOutOpr(), lremOutOpr()");
1822 
1823     Register R1 = lreg->predecessor();
1824     Register R2 = rreg;
1825     assert(code != lir_idiv || lreg==dreg, "see code below");
1826     if (code == lir_idiv) {
1827       __ z_lcgr(lreg, lreg);
1828     } else {
1829       __ clear_reg(dreg, true, false);
1830     }
1831     NearLabel done;
1832     __ compare64_and_branch(R2, -1, Assembler::bcondEqual, done);
1833     if (code == lir_idiv) {
1834       __ z_lcgr(lreg, lreg); // Revert lcgr above.
1835     }
1836     if (ImplicitDiv0Checks) {
1837       // No debug info because the idiv won't trap.
1838       // Add_debug_info_for_div0 would instantiate another DivByZeroStub,
1839       // which is unnecessary, too.
1840       add_debug_info_for_div0(__ offset(), info);
1841     }
1842     __ z_dsgr(R1, R2);
1843     __ bind(done);
1844     return;
1845   }
1846 
1847   // 32 bit integer case
1848 
1849   assert(left->is_single_cpu(), "left must be register");
1850   assert(right->is_single_cpu() || is_power_of_2(right->as_jint()), "right must be register or power of 2 constant");
1851   assert(result->is_single_cpu(), "result must be register");
1852 
1853   Register lreg = left->as_register();
1854   Register dreg = result->as_register();
1855 
1856   if (right->is_constant()) {
1857     // Convert division by a power of two into some shifts and logical operations.
1858     Register treg1 = Z_R0_scratch;
1859     Register treg2 = Z_R1_scratch;
1860     jlong divisor = right->as_jint();
1861     jlong log_divisor = log2i_exact(right->as_jint());
1862     __ move_reg_if_needed(dreg, T_LONG, lreg, T_INT); // sign extend
1863     if (divisor == 2) {
1864       __ z_srlg(treg2, dreg, 63);     // dividend < 0 ?  1 : 0
1865     } else {
1866       __ z_srag(treg2, dreg, 63);     // dividend < 0 ? -1 : 0
1867       __ and_imm(treg2, divisor - 1, treg1, true);
1868     }
1869     if (code == lir_idiv) {
1870       __ z_agr(dreg, treg2);
1871       __ z_srag(dreg, dreg, log_divisor);
1872     } else {
1873       assert(code == lir_irem, "check");
1874       __ z_agr(treg2, dreg);
1875       __ and_imm(treg2, ~(divisor - 1), treg1, true);
1876       __ z_sgr(dreg, treg2);
1877     }
1878     return;
1879   }
1880 
1881   // Divisor is not a power of 2 constant.
1882   Register rreg = right->as_register();
1883   Register treg = temp->as_register();
1884   assert(right->is_single_cpu(), "right must be register");
1885   assert(lreg == Z_R11, "left register must be rax,");
1886   assert(rreg != lreg, "right register must not be same as left register");
1887   assert((code == lir_idiv && dreg == Z_R11 && treg == Z_R10)
1888       || (code == lir_irem && dreg == Z_R10 && treg == Z_R11), "see divInOpr(), divOutOpr(), remOutOpr()");
1889 
1890   Register R1 = lreg->predecessor();
1891   Register R2 = rreg;
1892   __ move_reg_if_needed(lreg, T_LONG, lreg, T_INT); // sign extend
1893   if (ImplicitDiv0Checks) {
1894     // No debug info because the idiv won't trap.
1895     // Add_debug_info_for_div0 would instantiate another DivByZeroStub,
1896     // which is unnecessary, too.
1897     add_debug_info_for_div0(__ offset(), info);
1898   }
1899   __ z_dsgfr(R1, R2);
1900 }
1901 
1902 void LIR_Assembler::throw_op(LIR_Opr exceptionPC, LIR_Opr exceptionOop, CodeEmitInfo* info) {
1903   assert(exceptionOop->as_register() == Z_EXC_OOP, "should match");
1904   assert(exceptionPC->as_register() == Z_EXC_PC, "should match");
1905 
1906   // Exception object is not added to oop map by LinearScan
1907   // (LinearScan assumes that no oops are in fixed registers).
1908   info->add_register_oop(exceptionOop);
1909 
1910   // Reuse the debug info from the safepoint poll for the throw op itself.
1911   __ get_PC(Z_EXC_PC);
1912   add_call_info(__ offset(), info); // for exception handler
1913   address stub = Runtime1::entry_for (compilation()->has_fpu_code() ? C1StubId::handle_exception_id
1914                                                                     : C1StubId::handle_exception_nofpu_id);
1915   emit_call_c(stub);
1916 }
1917 
1918 void LIR_Assembler::unwind_op(LIR_Opr exceptionOop) {
1919   assert(exceptionOop->as_register() == Z_EXC_OOP, "should match");
1920 
1921   __ branch_optimized(Assembler::bcondAlways, _unwind_handler_entry);
1922 }
1923 
1924 void LIR_Assembler::emit_arraycopy(LIR_OpArrayCopy* op) {
1925   ciArrayKlass* default_type = op->expected_type();
1926   Register src = op->src()->as_register();
1927   Register dst = op->dst()->as_register();
1928   Register src_pos = op->src_pos()->as_register();
1929   Register dst_pos = op->dst_pos()->as_register();
1930   Register length  = op->length()->as_register();
1931   Register tmp = op->tmp()->as_register();
1932 
1933   CodeStub* stub = op->stub();
1934   int flags = op->flags();
1935   BasicType basic_type = default_type != nullptr ? default_type->element_type()->basic_type() : T_ILLEGAL;
1936   if (basic_type == T_ARRAY) basic_type = T_OBJECT;
1937 
1938   // If we don't know anything, just go through the generic arraycopy.
1939   if (default_type == nullptr) {
1940     address copyfunc_addr = StubRoutines::generic_arraycopy();
1941 
1942     if (copyfunc_addr == nullptr) {
1943       // Take a slow path for generic arraycopy.
1944       __ branch_optimized(Assembler::bcondAlways, *stub->entry());
1945       __ bind(*stub->continuation());
1946       return;
1947     }
1948 
1949     // Save outgoing arguments in callee saved registers (C convention) in case
1950     // a call to System.arraycopy is needed.
1951     Register callee_saved_src     = Z_R10;
1952     Register callee_saved_src_pos = Z_R11;
1953     Register callee_saved_dst     = Z_R12;
1954     Register callee_saved_dst_pos = Z_R13;
1955     Register callee_saved_length  = Z_ARG5; // Z_ARG5 == Z_R6 is callee saved.
1956 
1957     __ lgr_if_needed(callee_saved_src, src);
1958     __ lgr_if_needed(callee_saved_src_pos, src_pos);
1959     __ lgr_if_needed(callee_saved_dst, dst);
1960     __ lgr_if_needed(callee_saved_dst_pos, dst_pos);
1961     __ lgr_if_needed(callee_saved_length, length);
1962 
1963     // C function requires 64 bit values.
1964     __ z_lgfr(src_pos, src_pos);
1965     __ z_lgfr(dst_pos, dst_pos);
1966     __ z_lgfr(length, length);
1967 
1968     // Pass arguments: may push as this is not a safepoint; SP must be fix at each safepoint.
1969 
1970     // The arguments are in the corresponding registers.
1971     assert(Z_ARG1 == src,     "assumption");
1972     assert(Z_ARG2 == src_pos, "assumption");
1973     assert(Z_ARG3 == dst,     "assumption");
1974     assert(Z_ARG4 == dst_pos, "assumption");
1975     assert(Z_ARG5 == length,  "assumption");
1976 #ifndef PRODUCT
1977     if (PrintC1Statistics) {
1978       __ load_const_optimized(Z_R1_scratch, (address)&Runtime1::_generic_arraycopystub_cnt);
1979       __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
1980     }
1981 #endif
1982     emit_call_c(copyfunc_addr);
1983     CHECK_BAILOUT();
1984 
1985     __ compare32_and_branch(Z_RET, (intptr_t)0, Assembler::bcondEqual, *stub->continuation());
1986 
1987     __ z_lgr(tmp, Z_RET);
1988     __ z_xilf(tmp, -1);
1989 
1990     // Restore values from callee saved registers so they are where the stub
1991     // expects them.
1992     __ lgr_if_needed(src, callee_saved_src);
1993     __ lgr_if_needed(src_pos, callee_saved_src_pos);
1994     __ lgr_if_needed(dst, callee_saved_dst);
1995     __ lgr_if_needed(dst_pos, callee_saved_dst_pos);
1996     __ lgr_if_needed(length, callee_saved_length);
1997 
1998     __ z_sr(length, tmp);
1999     __ z_ar(src_pos, tmp);
2000     __ z_ar(dst_pos, tmp);
2001     __ branch_optimized(Assembler::bcondAlways, *stub->entry());
2002 
2003     __ bind(*stub->continuation());
2004     return;
2005   }
2006 
2007   assert(default_type != nullptr && default_type->is_array_klass() && default_type->is_loaded(), "must be true at this point");
2008 
2009   int elem_size = type2aelembytes(basic_type);
2010   int shift_amount;
2011 
2012   switch (elem_size) {
2013     case 1 :
2014       shift_amount = 0;
2015       break;
2016     case 2 :
2017       shift_amount = 1;
2018       break;
2019     case 4 :
2020       shift_amount = 2;
2021       break;
2022     case 8 :
2023       shift_amount = 3;
2024       break;
2025     default:
2026       shift_amount = -1;
2027       ShouldNotReachHere();
2028   }
2029 
2030   Address src_length_addr = Address(src, arrayOopDesc::length_offset_in_bytes());
2031   Address dst_length_addr = Address(dst, arrayOopDesc::length_offset_in_bytes());
2032   Address src_klass_addr = Address(src, oopDesc::klass_offset_in_bytes());
2033   Address dst_klass_addr = Address(dst, oopDesc::klass_offset_in_bytes());
2034 
2035   // Length and pos's are all sign extended at this point on 64bit.
2036 
2037   // test for null
2038   if (flags & LIR_OpArrayCopy::src_null_check) {
2039     __ compareU64_and_branch(src, (intptr_t)0, Assembler::bcondZero, *stub->entry());
2040   }
2041   if (flags & LIR_OpArrayCopy::dst_null_check) {
2042     __ compareU64_and_branch(dst, (intptr_t)0, Assembler::bcondZero, *stub->entry());
2043   }
2044 
2045   // Check if negative.
2046   if (flags & LIR_OpArrayCopy::src_pos_positive_check) {
2047     __ compare32_and_branch(src_pos, (intptr_t)0, Assembler::bcondLow, *stub->entry());
2048   }
2049   if (flags & LIR_OpArrayCopy::dst_pos_positive_check) {
2050     __ compare32_and_branch(dst_pos, (intptr_t)0, Assembler::bcondLow, *stub->entry());
2051   }
2052 
2053   // If the compiler was not able to prove that exact type of the source or the destination
2054   // of the arraycopy is an array type, check at runtime if the source or the destination is
2055   // an instance type.
2056   if (flags & LIR_OpArrayCopy::type_check) {
2057     assert(Klass::_lh_neutral_value == 0, "or replace z_lt instructions");
2058 
2059     if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
2060       __ load_klass(tmp, dst);
2061       __ z_lt(tmp, Address(tmp, in_bytes(Klass::layout_helper_offset())));
2062       __ branch_optimized(Assembler::bcondNotLow, *stub->entry());
2063     }
2064 
2065     if (!(flags & LIR_OpArrayCopy::src_objarray)) {
2066       __ load_klass(tmp, src);
2067       __ z_lt(tmp, Address(tmp, in_bytes(Klass::layout_helper_offset())));
2068       __ branch_optimized(Assembler::bcondNotLow, *stub->entry());
2069     }
2070   }
2071 
2072   if (flags & LIR_OpArrayCopy::src_range_check) {
2073     __ z_la(tmp, Address(src_pos, length));
2074     __ z_cl(tmp, src_length_addr);
2075     __ branch_optimized(Assembler::bcondHigh, *stub->entry());
2076   }
2077   if (flags & LIR_OpArrayCopy::dst_range_check) {
2078     __ z_la(tmp, Address(dst_pos, length));
2079     __ z_cl(tmp, dst_length_addr);
2080     __ branch_optimized(Assembler::bcondHigh, *stub->entry());
2081   }
2082 
2083   if (flags & LIR_OpArrayCopy::length_positive_check) {
2084     __ z_ltr(length, length);
2085     __ branch_optimized(Assembler::bcondNegative, *stub->entry());
2086   }
2087 
2088   // Stubs require 64 bit values.
2089   __ z_lgfr(src_pos, src_pos); // int -> long
2090   __ z_lgfr(dst_pos, dst_pos); // int -> long
2091   __ z_lgfr(length, length);   // int -> long
2092 
2093   if (flags & LIR_OpArrayCopy::type_check) {
2094     // We don't know the array types are compatible.
2095     if (basic_type != T_OBJECT) {
2096       // Simple test for basic type arrays.
2097       if (UseCompressedClassPointers) {
2098         __ z_l(tmp, src_klass_addr);
2099         __ z_c(tmp, dst_klass_addr);
2100       } else {
2101         __ z_lg(tmp, src_klass_addr);
2102         __ z_cg(tmp, dst_klass_addr);
2103       }
2104       __ branch_optimized(Assembler::bcondNotEqual, *stub->entry());
2105     } else {
2106       // For object arrays, if src is a sub class of dst then we can
2107       // safely do the copy.
2108       NearLabel cont, slow;
2109       Register src_klass = Z_R1_scratch;
2110       Register dst_klass = Z_R10;
2111 
2112       __ load_klass(src_klass, src);
2113       __ load_klass(dst_klass, dst);
2114 
2115       __ check_klass_subtype_fast_path(src_klass, dst_klass, tmp, &cont, &slow, nullptr);
2116 
2117       store_parameter(src_klass, 0); // sub
2118       store_parameter(dst_klass, 1); // super
2119       emit_call_c(Runtime1::entry_for (C1StubId::slow_subtype_check_id));
2120       CHECK_BAILOUT2(cont, slow);
2121       // Sets condition code 0 for match (2 otherwise).
2122       __ branch_optimized(Assembler::bcondEqual, cont);
2123 
2124       __ bind(slow);
2125 
2126       address copyfunc_addr = StubRoutines::checkcast_arraycopy();
2127       if (copyfunc_addr != nullptr) { // use stub if available
2128         // Src is not a sub class of dst so we have to do a
2129         // per-element check.
2130 
2131         int mask = LIR_OpArrayCopy::src_objarray|LIR_OpArrayCopy::dst_objarray;
2132         if ((flags & mask) != mask) {
2133           // Check that at least both of them object arrays.
2134           assert(flags & mask, "one of the two should be known to be an object array");
2135 
2136           if (!(flags & LIR_OpArrayCopy::src_objarray)) {
2137             __ load_klass(tmp, src);
2138           } else if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
2139             __ load_klass(tmp, dst);
2140           }
2141           Address klass_lh_addr(tmp, Klass::layout_helper_offset());
2142           jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2143           __ load_const_optimized(Z_R1_scratch, objArray_lh);
2144           __ z_c(Z_R1_scratch, klass_lh_addr);
2145           __ branch_optimized(Assembler::bcondNotEqual, *stub->entry());
2146         }
2147 
2148         // Save outgoing arguments in callee saved registers (C convention) in case
2149         // a call to System.arraycopy is needed.
2150         Register callee_saved_src     = Z_R10;
2151         Register callee_saved_src_pos = Z_R11;
2152         Register callee_saved_dst     = Z_R12;
2153         Register callee_saved_dst_pos = Z_R13;
2154         Register callee_saved_length  = Z_ARG5; // Z_ARG5 == Z_R6 is callee saved.
2155 
2156         __ lgr_if_needed(callee_saved_src, src);
2157         __ lgr_if_needed(callee_saved_src_pos, src_pos);
2158         __ lgr_if_needed(callee_saved_dst, dst);
2159         __ lgr_if_needed(callee_saved_dst_pos, dst_pos);
2160         __ lgr_if_needed(callee_saved_length, length);
2161 
2162         __ z_llgfr(length, length); // Higher 32bits must be null.
2163 
2164         __ z_sllg(Z_ARG1, src_pos, shift_amount); // index -> byte offset
2165         __ z_sllg(Z_ARG2, dst_pos, shift_amount); // index -> byte offset
2166 
2167         __ z_la(Z_ARG1, Address(src, Z_ARG1, arrayOopDesc::base_offset_in_bytes(basic_type)));
2168         assert_different_registers(Z_ARG1, dst, dst_pos, length);
2169         __ z_la(Z_ARG2, Address(dst, Z_ARG2, arrayOopDesc::base_offset_in_bytes(basic_type)));
2170         assert_different_registers(Z_ARG2, dst, length);
2171 
2172         __ z_lgr(Z_ARG3, length);
2173         assert_different_registers(Z_ARG3, dst);
2174 
2175         __ load_klass(Z_ARG5, dst);
2176         __ z_lg(Z_ARG5, Address(Z_ARG5, ObjArrayKlass::element_klass_offset()));
2177         __ z_lg(Z_ARG4, Address(Z_ARG5, Klass::super_check_offset_offset()));
2178         emit_call_c(copyfunc_addr);
2179         CHECK_BAILOUT2(cont, slow);
2180 
2181 #ifndef PRODUCT
2182         if (PrintC1Statistics) {
2183           NearLabel failed;
2184           __ compareU32_and_branch(Z_RET, (intptr_t)0, Assembler::bcondNotEqual, failed);
2185           __ load_const_optimized(Z_R1_scratch, (address)&Runtime1::_arraycopy_checkcast_cnt);
2186           __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
2187           __ bind(failed);
2188         }
2189 #endif
2190 
2191         __ compareU32_and_branch(Z_RET, (intptr_t)0, Assembler::bcondEqual, *stub->continuation());
2192 
2193 #ifndef PRODUCT
2194         if (PrintC1Statistics) {
2195           __ load_const_optimized(Z_R1_scratch, (address)&Runtime1::_arraycopy_checkcast_attempt_cnt);
2196           __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
2197         }
2198 #endif
2199 
2200         __ z_lgr(tmp, Z_RET);
2201         __ z_xilf(tmp, -1);
2202 
2203         // Restore previously spilled arguments
2204         __ lgr_if_needed(src, callee_saved_src);
2205         __ lgr_if_needed(src_pos, callee_saved_src_pos);
2206         __ lgr_if_needed(dst, callee_saved_dst);
2207         __ lgr_if_needed(dst_pos, callee_saved_dst_pos);
2208         __ lgr_if_needed(length, callee_saved_length);
2209 
2210         __ z_sr(length, tmp);
2211         __ z_ar(src_pos, tmp);
2212         __ z_ar(dst_pos, tmp);
2213       }
2214 
2215       __ branch_optimized(Assembler::bcondAlways, *stub->entry());
2216 
2217       __ bind(cont);
2218     }
2219   }
2220 
2221 #ifdef ASSERT
2222   if (basic_type != T_OBJECT || !(flags & LIR_OpArrayCopy::type_check)) {
2223     // Sanity check the known type with the incoming class. For the
2224     // primitive case the types must match exactly with src.klass and
2225     // dst.klass each exactly matching the default type. For the
2226     // object array case, if no type check is needed then either the
2227     // dst type is exactly the expected type and the src type is a
2228     // subtype which we can't check or src is the same array as dst
2229     // but not necessarily exactly of type default_type.
2230     NearLabel known_ok, halt;
2231     metadata2reg(default_type->constant_encoding(), tmp);
2232     if (UseCompressedClassPointers) {
2233       __ encode_klass_not_null(tmp);
2234     }
2235 
2236     if (basic_type != T_OBJECT) {
2237       if (UseCompressedClassPointers)         { __ z_c (tmp, dst_klass_addr); }
2238       else                                    { __ z_cg(tmp, dst_klass_addr); }
2239       __ branch_optimized(Assembler::bcondNotEqual, halt);
2240       if (UseCompressedClassPointers)         { __ z_c (tmp, src_klass_addr); }
2241       else                                    { __ z_cg(tmp, src_klass_addr); }
2242       __ branch_optimized(Assembler::bcondEqual, known_ok);
2243     } else {
2244       if (UseCompressedClassPointers)         { __ z_c (tmp, dst_klass_addr); }
2245       else                                    { __ z_cg(tmp, dst_klass_addr); }
2246       __ branch_optimized(Assembler::bcondEqual, known_ok);
2247       __ compareU64_and_branch(src, dst, Assembler::bcondEqual, known_ok);
2248     }
2249     __ bind(halt);
2250     __ stop("incorrect type information in arraycopy");
2251     __ bind(known_ok);
2252   }
2253 #endif
2254 
2255 #ifndef PRODUCT
2256   if (PrintC1Statistics) {
2257     __ load_const_optimized(Z_R1_scratch, Runtime1::arraycopy_count_address(basic_type));
2258     __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
2259   }
2260 #endif
2261 
2262   __ z_sllg(tmp, src_pos, shift_amount); // index -> byte offset
2263   __ z_sllg(Z_R1_scratch, dst_pos, shift_amount); // index -> byte offset
2264 
2265   assert_different_registers(Z_ARG1, dst, dst_pos, length);
2266   __ z_la(Z_ARG1, Address(src, tmp, arrayOopDesc::base_offset_in_bytes(basic_type)));
2267   assert_different_registers(Z_ARG2, length);
2268   __ z_la(Z_ARG2, Address(dst, Z_R1_scratch, arrayOopDesc::base_offset_in_bytes(basic_type)));
2269   __ lgr_if_needed(Z_ARG3, length);
2270 
2271   bool disjoint = (flags & LIR_OpArrayCopy::overlapping) == 0;
2272   bool aligned = (flags & LIR_OpArrayCopy::unaligned) == 0;
2273   const char *name;
2274   address entry = StubRoutines::select_arraycopy_function(basic_type, aligned, disjoint, name, false);
2275   __ call_VM_leaf(entry);
2276 
2277   if (stub != nullptr) {
2278     __ bind(*stub->continuation());
2279   }
2280 }
2281 
2282 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
2283   if (dest->is_single_cpu()) {
2284     if (left->type() == T_OBJECT) {
2285       switch (code) {
2286         case lir_shl:  __ z_sllg (dest->as_register(), left->as_register(), 0, count->as_register()); break;
2287         case lir_shr:  __ z_srag (dest->as_register(), left->as_register(), 0, count->as_register()); break;
2288         case lir_ushr: __ z_srlg (dest->as_register(), left->as_register(), 0, count->as_register()); break;
2289         default: ShouldNotReachHere();
2290       }
2291     } else {
2292       assert(code == lir_shl || left == dest, "left and dest must be equal for 2 operand form right shifts");
2293       Register masked_count = Z_R1_scratch;
2294       __ z_lr(masked_count, count->as_register());
2295       __ z_nill(masked_count, 31);
2296       switch (code) {
2297         case lir_shl:  __ z_sllg (dest->as_register(), left->as_register(), 0, masked_count); break;
2298         case lir_shr:  __ z_sra  (dest->as_register(), 0, masked_count); break;
2299         case lir_ushr: __ z_srl  (dest->as_register(), 0, masked_count); break;
2300         default: ShouldNotReachHere();
2301       }
2302     }
2303   } else {
2304     switch (code) {
2305       case lir_shl:  __ z_sllg (dest->as_register_lo(), left->as_register_lo(), 0, count->as_register()); break;
2306       case lir_shr:  __ z_srag (dest->as_register_lo(), left->as_register_lo(), 0, count->as_register()); break;
2307       case lir_ushr: __ z_srlg (dest->as_register_lo(), left->as_register_lo(), 0, count->as_register()); break;
2308       default: ShouldNotReachHere();
2309     }
2310   }
2311 }
2312 
2313 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, jint count, LIR_Opr dest) {
2314   if (left->type() == T_OBJECT) {
2315     count = count & 63;  // Shouldn't shift by more than sizeof(intptr_t).
2316     Register l = left->as_register();
2317     Register d = dest->as_register_lo();
2318     switch (code) {
2319       case lir_shl:  __ z_sllg (d, l, count); break;
2320       case lir_shr:  __ z_srag (d, l, count); break;
2321       case lir_ushr: __ z_srlg (d, l, count); break;
2322       default: ShouldNotReachHere();
2323     }
2324     return;
2325   }
2326   if (dest->is_single_cpu()) {
2327     assert(code == lir_shl || left == dest, "left and dest must be equal for 2 operand form right shifts");
2328     count = count & 0x1F; // Java spec
2329     switch (code) {
2330       case lir_shl:  __ z_sllg (dest->as_register(), left->as_register(), count); break;
2331       case lir_shr:  __ z_sra  (dest->as_register(), count); break;
2332       case lir_ushr: __ z_srl  (dest->as_register(), count); break;
2333       default: ShouldNotReachHere();
2334     }
2335   } else if (dest->is_double_cpu()) {
2336     count = count & 63; // Java spec
2337     Register l = left->as_pointer_register();
2338     Register d = dest->as_pointer_register();
2339     switch (code) {
2340       case lir_shl:  __ z_sllg (d, l, count); break;
2341       case lir_shr:  __ z_srag (d, l, count); break;
2342       case lir_ushr: __ z_srlg (d, l, count); break;
2343       default: ShouldNotReachHere();
2344     }
2345   } else {
2346     ShouldNotReachHere();
2347   }
2348 }
2349 
2350 void LIR_Assembler::emit_alloc_obj(LIR_OpAllocObj* op) {
2351   if (op->init_check()) {
2352     // Make sure klass is initialized & doesn't have finalizer.
2353     // init_state needs acquire, but S390 is TSO, and so we are already good.
2354     const int state_offset = in_bytes(InstanceKlass::init_state_offset());
2355     Register iklass = op->klass()->as_register();
2356     add_debug_info_for_null_check_here(op->stub()->info());
2357     if (Immediate::is_uimm12(state_offset)) {
2358       __ z_cli(state_offset, iklass, InstanceKlass::fully_initialized);
2359     } else {
2360       __ z_cliy(state_offset, iklass, InstanceKlass::fully_initialized);
2361     }
2362     __ branch_optimized(Assembler::bcondNotEqual, *op->stub()->entry()); // Use long branch, because slow_case might be far.
2363   }
2364   __ allocate_object(op->obj()->as_register(),
2365                      op->tmp1()->as_register(),
2366                      op->tmp2()->as_register(),
2367                      op->header_size(),
2368                      op->object_size(),
2369                      op->klass()->as_register(),
2370                      *op->stub()->entry());
2371   __ bind(*op->stub()->continuation());
2372   __ verify_oop(op->obj()->as_register(), FILE_AND_LINE);
2373 }
2374 
2375 void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
2376   Register len = op->len()->as_register();
2377   __ move_reg_if_needed(len, T_LONG, len, T_INT); // sign extend
2378 
2379   if (UseSlowPath ||
2380       (!UseFastNewObjectArray && (is_reference_type(op->type()))) ||
2381       (!UseFastNewTypeArray   && (!is_reference_type(op->type())))) {
2382     __ z_brul(*op->stub()->entry());
2383   } else {
2384     __ allocate_array(op->obj()->as_register(),
2385                       op->len()->as_register(),
2386                       op->tmp1()->as_register(),
2387                       op->tmp2()->as_register(),
2388                       arrayOopDesc::base_offset_in_bytes(op->type()),
2389                       type2aelembytes(op->type()),
2390                       op->klass()->as_register(),
2391                       *op->stub()->entry(),
2392                       op->zero_array());
2393   }
2394   __ bind(*op->stub()->continuation());
2395 }
2396 
2397 void LIR_Assembler::type_profile_helper(Register mdo, ciMethodData *md, ciProfileData *data,
2398                                         Register recv, Register tmp1, Label* update_done) {
2399   uint i;
2400   for (i = 0; i < VirtualCallData::row_limit(); i++) {
2401     Label next_test;
2402     // See if the receiver is receiver[n].
2403     Address receiver_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)));
2404     __ z_cg(recv, receiver_addr);
2405     __ z_brne(next_test);
2406     Address data_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)));
2407     __ add2mem_64(data_addr, DataLayout::counter_increment, tmp1);
2408     __ branch_optimized(Assembler::bcondAlways, *update_done);
2409     __ bind(next_test);
2410   }
2411 
2412   // Didn't find receiver; find next empty slot and fill it in.
2413   for (i = 0; i < VirtualCallData::row_limit(); i++) {
2414     Label next_test;
2415     Address recv_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)));
2416     __ z_ltg(Z_R0_scratch, recv_addr);
2417     __ z_brne(next_test);
2418     __ z_stg(recv, recv_addr);
2419     __ load_const_optimized(tmp1, DataLayout::counter_increment);
2420     __ z_stg(tmp1, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)), mdo);
2421     __ branch_optimized(Assembler::bcondAlways, *update_done);
2422     __ bind(next_test);
2423   }
2424 }
2425 
2426 void LIR_Assembler::setup_md_access(ciMethod* method, int bci,
2427                                     ciMethodData*& md, ciProfileData*& data, int& mdo_offset_bias) {
2428   Unimplemented();
2429 }
2430 
2431 void LIR_Assembler::store_parameter(Register r, int param_num) {
2432   assert(param_num >= 0, "invalid num");
2433   int offset_in_bytes = param_num * BytesPerWord;
2434   check_reserved_argument_area(offset_in_bytes);
2435   offset_in_bytes += FrameMap::first_available_sp_in_frame;
2436   __ z_stg(r, offset_in_bytes, Z_SP);
2437 }
2438 
2439 void LIR_Assembler::store_parameter(jint c, int param_num) {
2440   assert(param_num >= 0, "invalid num");
2441   int offset_in_bytes = param_num * BytesPerWord;
2442   check_reserved_argument_area(offset_in_bytes);
2443   offset_in_bytes += FrameMap::first_available_sp_in_frame;
2444   __ store_const(Address(Z_SP, offset_in_bytes), c, Z_R1_scratch, true);
2445 }
2446 
2447 void LIR_Assembler::emit_typecheck_helper(LIR_OpTypeCheck *op, Label* success, Label* failure, Label* obj_is_null) {
2448   // We always need a stub for the failure case.
2449   CodeStub* stub = op->stub();
2450   Register obj = op->object()->as_register();
2451   Register k_RInfo = op->tmp1()->as_register();
2452   Register klass_RInfo = op->tmp2()->as_register();
2453   Register dst = op->result_opr()->as_register();
2454   Register Rtmp1 = Z_R1_scratch;
2455   ciKlass* k = op->klass();
2456 
2457   assert(!op->tmp3()->is_valid(), "tmp3's not needed");
2458 
2459   // Check if it needs to be profiled.
2460   ciMethodData* md = nullptr;
2461   ciProfileData* data = nullptr;
2462 
2463   if (op->should_profile()) {
2464     ciMethod* method = op->profiled_method();
2465     assert(method != nullptr, "Should have method");
2466     int bci = op->profiled_bci();
2467     md = method->method_data_or_null();
2468     assert(md != nullptr, "Sanity");
2469     data = md->bci_to_data(bci);
2470     assert(data != nullptr,                "need data for type check");
2471     assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
2472   }
2473 
2474   // Temp operands do not overlap with inputs, if this is their last
2475   // use (end of range is exclusive), so a register conflict is possible.
2476   if (obj == k_RInfo) {
2477     k_RInfo = dst;
2478   } else if (obj == klass_RInfo) {
2479     klass_RInfo = dst;
2480   }
2481   assert_different_registers(obj, k_RInfo, klass_RInfo);
2482 
2483   if (op->should_profile()) {
2484     Register mdo = klass_RInfo;
2485     metadata2reg(md->constant_encoding(), mdo);
2486     NearLabel not_null;
2487     __ compareU64_and_branch(obj, (intptr_t) 0, Assembler::bcondNotEqual, not_null);
2488     // Object is null; update MDO and exit.
2489     Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
2490     int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
2491     __ or2mem_8(data_addr, header_bits);
2492     __ branch_optimized(Assembler::bcondAlways, *obj_is_null);
2493     __ bind(not_null);
2494 
2495     NearLabel update_done;
2496     Register recv = k_RInfo;
2497     __ load_klass(recv, obj);
2498     type_profile_helper(mdo, md, data, recv, Rtmp1, &update_done);
2499     Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
2500     __ add2mem_64(counter_addr, DataLayout::counter_increment, Rtmp1);
2501     __ bind(update_done);
2502   } else {
2503     __ compareU64_and_branch(obj, (intptr_t) 0, Assembler::bcondEqual, *obj_is_null);
2504   }
2505 
2506   Label *failure_target = failure;
2507   Label *success_target = success;
2508 
2509   // Patching may screw with our temporaries,
2510   // so let's do it before loading the class.
2511   if (k->is_loaded()) {
2512     metadata2reg(k->constant_encoding(), k_RInfo);
2513   } else {
2514     klass2reg_with_patching(k_RInfo, op->info_for_patch());
2515   }
2516   assert(obj != k_RInfo, "must be different");
2517 
2518   __ verify_oop(obj, FILE_AND_LINE);
2519 
2520   // Get object class.
2521   // Not a safepoint as obj null check happens earlier.
2522   if (op->fast_check()) {
2523     if (UseCompressedClassPointers) {
2524       __ load_klass(klass_RInfo, obj);
2525       __ compareU64_and_branch(k_RInfo, klass_RInfo, Assembler::bcondNotEqual, *failure_target);
2526     } else {
2527       __ z_cg(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
2528       __ branch_optimized(Assembler::bcondNotEqual, *failure_target);
2529     }
2530     // Successful cast, fall through to profile or jump.
2531   } else {
2532     bool need_slow_path = !k->is_loaded() ||
2533                           ((int) k->super_check_offset() == in_bytes(Klass::secondary_super_cache_offset()));
2534     intptr_t super_check_offset = k->is_loaded() ? k->super_check_offset() : -1L;
2535     __ load_klass(klass_RInfo, obj);
2536     // Perform the fast part of the checking logic.
2537     __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1,
2538                                      (need_slow_path ? success_target : nullptr),
2539                                      failure_target, nullptr,
2540                                      RegisterOrConstant(super_check_offset));
2541     if (need_slow_path) {
2542       // Call out-of-line instance of __ check_klass_subtype_slow_path(...):
2543       address a = Runtime1::entry_for (C1StubId::slow_subtype_check_id);
2544       store_parameter(klass_RInfo, 0); // sub
2545       store_parameter(k_RInfo, 1);     // super
2546       emit_call_c(a); // Sets condition code 0 for match (2 otherwise).
2547       __ branch_optimized(Assembler::bcondNotEqual, *failure_target);
2548       // Fall through to success case.
2549     }
2550   }
2551 
2552   __ branch_optimized(Assembler::bcondAlways, *success);
2553 }
2554 
2555 void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
2556   LIR_Code code = op->code();
2557   if (code == lir_store_check) {
2558     Register value = op->object()->as_register();
2559     Register array = op->array()->as_register();
2560     Register k_RInfo = op->tmp1()->as_register();
2561     Register klass_RInfo = op->tmp2()->as_register();
2562     Register Rtmp1 = Z_R1_scratch;
2563 
2564     CodeStub* stub = op->stub();
2565 
2566     // Check if it needs to be profiled.
2567     ciMethodData* md = nullptr;
2568     ciProfileData* data = nullptr;
2569 
2570     assert_different_registers(value, k_RInfo, klass_RInfo);
2571 
2572     if (op->should_profile()) {
2573       ciMethod* method = op->profiled_method();
2574       assert(method != nullptr, "Should have method");
2575       int bci = op->profiled_bci();
2576       md = method->method_data_or_null();
2577       assert(md != nullptr, "Sanity");
2578       data = md->bci_to_data(bci);
2579       assert(data != nullptr,                "need data for type check");
2580       assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
2581     }
2582     NearLabel done;
2583     Label *success_target = &done;
2584     Label *failure_target = stub->entry();
2585 
2586     if (op->should_profile()) {
2587       Register mdo = klass_RInfo;
2588       metadata2reg(md->constant_encoding(), mdo);
2589       NearLabel not_null;
2590       __ compareU64_and_branch(value, (intptr_t) 0, Assembler::bcondNotEqual, not_null);
2591       // Object is null; update MDO and exit.
2592       Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
2593       int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
2594       __ or2mem_8(data_addr, header_bits);
2595       __ branch_optimized(Assembler::bcondAlways, done);
2596       __ bind(not_null);
2597 
2598       NearLabel update_done;
2599       Register recv = k_RInfo;
2600       __ load_klass(recv, value);
2601       type_profile_helper(mdo, md, data, recv, Rtmp1, &update_done);
2602       Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
2603       __ add2mem_64(counter_addr, DataLayout::counter_increment, Rtmp1);
2604       __ bind(update_done);
2605     } else {
2606       __ compareU64_and_branch(value, (intptr_t) 0, Assembler::bcondEqual, done);
2607     }
2608 
2609     add_debug_info_for_null_check_here(op->info_for_exception());
2610     __ load_klass(k_RInfo, array);
2611     __ load_klass(klass_RInfo, value);
2612 
2613     // Get instance klass (it's already uncompressed).
2614     __ z_lg(k_RInfo, Address(k_RInfo, ObjArrayKlass::element_klass_offset()));
2615     // Perform the fast part of the checking logic.
2616     __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, success_target, failure_target, nullptr);
2617     // Call out-of-line instance of __ check_klass_subtype_slow_path(...):
2618     address a = Runtime1::entry_for (C1StubId::slow_subtype_check_id);
2619     store_parameter(klass_RInfo, 0); // sub
2620     store_parameter(k_RInfo, 1);     // super
2621     emit_call_c(a); // Sets condition code 0 for match (2 otherwise).
2622     __ branch_optimized(Assembler::bcondNotEqual, *failure_target);
2623     // Fall through to success case.
2624 
2625     __ bind(done);
2626   } else {
2627     if (code == lir_checkcast) {
2628       Register obj = op->object()->as_register();
2629       Register dst = op->result_opr()->as_register();
2630       NearLabel success;
2631       emit_typecheck_helper(op, &success, op->stub()->entry(), &success);
2632       __ bind(success);
2633       __ lgr_if_needed(dst, obj);
2634     } else {
2635       if (code == lir_instanceof) {
2636         Register obj = op->object()->as_register();
2637         Register dst = op->result_opr()->as_register();
2638         NearLabel success, failure, done;
2639         emit_typecheck_helper(op, &success, &failure, &failure);
2640         __ bind(failure);
2641         __ clear_reg(dst);
2642         __ branch_optimized(Assembler::bcondAlways, done);
2643         __ bind(success);
2644         __ load_const_optimized(dst, 1);
2645         __ bind(done);
2646       } else {
2647         ShouldNotReachHere();
2648       }
2649     }
2650   }
2651 }
2652 
2653 void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
2654   Register addr = op->addr()->as_pointer_register();
2655   Register t1_cmp = Z_R1_scratch;
2656   if (op->code() == lir_cas_long) {
2657     Register cmp_value_lo = op->cmp_value()->as_register_lo();
2658     Register new_value_lo = op->new_value()->as_register_lo();
2659     __ z_lgr(t1_cmp, cmp_value_lo);
2660     // Perform the compare and swap operation.
2661     __ z_csg(t1_cmp, new_value_lo, 0, addr);
2662   } else if (op->code() == lir_cas_int || op->code() == lir_cas_obj) {
2663     Register cmp_value = op->cmp_value()->as_register();
2664     Register new_value = op->new_value()->as_register();
2665     if (op->code() == lir_cas_obj) {
2666       if (UseCompressedOops) {
2667                  t1_cmp = op->tmp1()->as_register();
2668         Register t2_new = op->tmp2()->as_register();
2669         assert_different_registers(cmp_value, new_value, addr, t1_cmp, t2_new);
2670         __ oop_encoder(t1_cmp, cmp_value, true /*maybe null*/);
2671         __ oop_encoder(t2_new, new_value, true /*maybe null*/);
2672         __ z_cs(t1_cmp, t2_new, 0, addr);
2673       } else {
2674         __ z_lgr(t1_cmp, cmp_value);
2675         __ z_csg(t1_cmp, new_value, 0, addr);
2676       }
2677     } else {
2678       __ z_lr(t1_cmp, cmp_value);
2679       __ z_cs(t1_cmp, new_value, 0, addr);
2680     }
2681   } else {
2682     ShouldNotReachHere(); // new lir_cas_??
2683   }
2684 }
2685 
2686 void LIR_Assembler::breakpoint() {
2687   Unimplemented();
2688   //  __ breakpoint_trap();
2689 }
2690 
2691 void LIR_Assembler::push(LIR_Opr opr) {
2692   ShouldNotCallThis(); // unused
2693 }
2694 
2695 void LIR_Assembler::pop(LIR_Opr opr) {
2696   ShouldNotCallThis(); // unused
2697 }
2698 
2699 void LIR_Assembler::monitor_address(int monitor_no, LIR_Opr dst_opr) {
2700   Address addr = frame_map()->address_for_monitor_lock(monitor_no);
2701   __ add2reg(dst_opr->as_register(), addr.disp(), addr.base());
2702 }
2703 
2704 void LIR_Assembler::emit_lock(LIR_OpLock* op) {
2705   Register obj = op->obj_opr()->as_register();  // May not be an oop.
2706   Register hdr = op->hdr_opr()->as_register();
2707   Register lock = op->lock_opr()->as_register();
2708   if (LockingMode == LM_MONITOR) {
2709     if (op->info() != nullptr) {
2710       add_debug_info_for_null_check_here(op->info());
2711       __ null_check(obj);
2712     }
2713     __ branch_optimized(Assembler::bcondAlways, *op->stub()->entry());
2714   } else if (op->code() == lir_lock) {
2715     assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
2716     // Add debug info for NullPointerException only if one is possible.
2717     if (op->info() != nullptr) {
2718       add_debug_info_for_null_check_here(op->info());
2719     }
2720     __ lock_object(hdr, obj, lock, *op->stub()->entry());
2721     // done
2722   } else if (op->code() == lir_unlock) {
2723     assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
2724     __ unlock_object(hdr, obj, lock, *op->stub()->entry());
2725   } else {
2726     ShouldNotReachHere();
2727   }
2728   __ bind(*op->stub()->continuation());
2729 }
2730 
2731 void LIR_Assembler::emit_load_klass(LIR_OpLoadKlass* op) {
2732   Register obj = op->obj()->as_pointer_register();
2733   Register result = op->result_opr()->as_pointer_register();
2734 
2735   CodeEmitInfo* info = op->info();
2736   if (info != nullptr) {
2737     add_debug_info_for_null_check_here(info);
2738   }
2739 
2740   if (UseCompressedClassPointers) {
2741     __ z_llgf(result, Address(obj, oopDesc::klass_offset_in_bytes()));
2742     __ decode_klass_not_null(result);
2743   } else {
2744     __ z_lg(result, Address(obj, oopDesc::klass_offset_in_bytes()));
2745   }
2746 }
2747 void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
2748   ciMethod* method = op->profiled_method();
2749   int bci          = op->profiled_bci();
2750   ciMethod* callee = op->profiled_callee();
2751 
2752   // Update counter for all call types.
2753   ciMethodData* md = method->method_data_or_null();
2754   assert(md != nullptr, "Sanity");
2755   ciProfileData* data = md->bci_to_data(bci);
2756   assert(data != nullptr && data->is_CounterData(), "need CounterData for calls");
2757   assert(op->mdo()->is_single_cpu(),  "mdo must be allocated");
2758   Register mdo  = op->mdo()->as_register();
2759   assert(op->tmp1()->is_double_cpu(), "tmp1 must be allocated");
2760   Register tmp1 = op->tmp1()->as_register_lo();
2761   metadata2reg(md->constant_encoding(), mdo);
2762 
2763   Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
2764   // Perform additional virtual call profiling for invokevirtual and
2765   // invokeinterface bytecodes
2766   if (op->should_profile_receiver_type()) {
2767     assert(op->recv()->is_single_cpu(), "recv must be allocated");
2768     Register recv = op->recv()->as_register();
2769     assert_different_registers(mdo, tmp1, recv);
2770     assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
2771     ciKlass* known_klass = op->known_holder();
2772     if (C1OptimizeVirtualCallProfiling && known_klass != nullptr) {
2773       // We know the type that will be seen at this call site; we can
2774       // statically update the MethodData* rather than needing to do
2775       // dynamic tests on the receiver type.
2776 
2777       // NOTE: we should probably put a lock around this search to
2778       // avoid collisions by concurrent compilations.
2779       ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
2780       uint i;
2781       for (i = 0; i < VirtualCallData::row_limit(); i++) {
2782         ciKlass* receiver = vc_data->receiver(i);
2783         if (known_klass->equals(receiver)) {
2784           Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
2785           __ add2mem_64(data_addr, DataLayout::counter_increment, tmp1);
2786           return;
2787         }
2788       }
2789 
2790       // Receiver type not found in profile data. Select an empty slot.
2791 
2792       // Note that this is less efficient than it should be because it
2793       // always does a write to the receiver part of the
2794       // VirtualCallData rather than just the first time.
2795       for (i = 0; i < VirtualCallData::row_limit(); i++) {
2796         ciKlass* receiver = vc_data->receiver(i);
2797         if (receiver == nullptr) {
2798           Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)));
2799           metadata2reg(known_klass->constant_encoding(), tmp1);
2800           __ z_stg(tmp1, recv_addr);
2801           Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
2802           __ add2mem_64(data_addr, DataLayout::counter_increment, tmp1);
2803           return;
2804         }
2805       }
2806     } else {
2807       __ load_klass(recv, recv);
2808       NearLabel update_done;
2809       type_profile_helper(mdo, md, data, recv, tmp1, &update_done);
2810       // Receiver did not match any saved receiver and there is no empty row for it.
2811       // Increment total counter to indicate polymorphic case.
2812       __ add2mem_64(counter_addr, DataLayout::counter_increment, tmp1);
2813       __ bind(update_done);
2814     }
2815   } else {
2816     // static call
2817     __ add2mem_64(counter_addr, DataLayout::counter_increment, tmp1);
2818   }
2819 }
2820 
2821 void LIR_Assembler::align_backward_branch_target() {
2822   __ align(OptoLoopAlignment);
2823 }
2824 
2825 void LIR_Assembler::emit_delay(LIR_OpDelay* op) {
2826   ShouldNotCallThis(); // There are no delay slots on ZARCH_64.
2827 }
2828 
2829 void LIR_Assembler::negate(LIR_Opr left, LIR_Opr dest, LIR_Opr tmp) {
2830   // tmp must be unused
2831   assert(tmp->is_illegal(), "wasting a register if tmp is allocated");
2832   assert(left->is_register(), "can only handle registers");
2833 
2834   if (left->is_single_cpu()) {
2835     __ z_lcr(dest->as_register(), left->as_register());
2836   } else if (left->is_single_fpu()) {
2837     __ z_lcebr(dest->as_float_reg(), left->as_float_reg());
2838   } else if (left->is_double_fpu()) {
2839     __ z_lcdbr(dest->as_double_reg(), left->as_double_reg());
2840   } else {
2841     assert(left->is_double_cpu(), "Must be a long");
2842     __ z_lcgr(dest->as_register_lo(), left->as_register_lo());
2843   }
2844 }
2845 
2846 void LIR_Assembler::rt_call(LIR_Opr result, address dest,
2847                             const LIR_OprList* args, LIR_Opr tmp, CodeEmitInfo* info) {
2848   assert(!tmp->is_valid(), "don't need temporary");
2849   emit_call_c(dest);
2850   CHECK_BAILOUT();
2851   if (info != nullptr) {
2852     add_call_info_here(info);
2853   }
2854 }
2855 
2856 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {
2857   ShouldNotCallThis(); // not needed on ZARCH_64
2858 }
2859 
2860 void LIR_Assembler::membar() {
2861   __ z_fence();
2862 }
2863 
2864 void LIR_Assembler::membar_acquire() {
2865   __ z_acquire();
2866 }
2867 
2868 void LIR_Assembler::membar_release() {
2869   __ z_release();
2870 }
2871 
2872 void LIR_Assembler::membar_loadload() {
2873   __ z_acquire();
2874 }
2875 
2876 void LIR_Assembler::membar_storestore() {
2877   __ z_release();
2878 }
2879 
2880 void LIR_Assembler::membar_loadstore() {
2881   __ z_acquire();
2882 }
2883 
2884 void LIR_Assembler::membar_storeload() {
2885   __ z_fence();
2886 }
2887 
2888 void LIR_Assembler::on_spin_wait() {
2889   Unimplemented();
2890 }
2891 
2892 void LIR_Assembler::leal(LIR_Opr addr_opr, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
2893   assert(patch_code == lir_patch_none, "Patch code not supported");
2894   LIR_Address* addr = addr_opr->as_address_ptr();
2895   assert(addr->scale() == LIR_Address::times_1, "scaling unsupported");
2896   __ load_address(dest->as_pointer_register(), as_Address(addr));
2897 }
2898 
2899 void LIR_Assembler::get_thread(LIR_Opr result_reg) {
2900   ShouldNotCallThis(); // unused
2901 }
2902 
2903 #ifdef ASSERT
2904 // Emit run-time assertion.
2905 void LIR_Assembler::emit_assert(LIR_OpAssert* op) {
2906   Unimplemented();
2907 }
2908 #endif
2909 
2910 void LIR_Assembler::peephole(LIR_List*) {
2911   // Do nothing for now.
2912 }
2913 
2914 void LIR_Assembler::atomic_op(LIR_Code code, LIR_Opr src, LIR_Opr data, LIR_Opr dest, LIR_Opr tmp) {
2915   assert(code == lir_xadd, "lir_xchg not supported");
2916   Address src_addr = as_Address(src->as_address_ptr());
2917   Register base = src_addr.base();
2918   intptr_t disp = src_addr.disp();
2919   if (src_addr.index()->is_valid()) {
2920     // LAA and LAAG do not support index register.
2921     __ load_address(Z_R1_scratch, src_addr);
2922     base = Z_R1_scratch;
2923     disp = 0;
2924   }
2925   if (data->type() == T_INT) {
2926     __ z_laa(dest->as_register(), data->as_register(), disp, base);
2927   } else if (data->type() == T_LONG) {
2928     assert(data->as_register_lo() == data->as_register_hi(), "should be a single register");
2929     __ z_laag(dest->as_register_lo(), data->as_register_lo(), disp, base);
2930   } else {
2931     ShouldNotReachHere();
2932   }
2933 }
2934 
2935 void LIR_Assembler::emit_profile_type(LIR_OpProfileType* op) {
2936   Register obj = op->obj()->as_register();
2937   Register tmp1 = op->tmp()->as_pointer_register();
2938   Register tmp2 = Z_R1_scratch;
2939   Address mdo_addr = as_Address(op->mdp()->as_address_ptr());
2940   ciKlass* exact_klass = op->exact_klass();
2941   intptr_t current_klass = op->current_klass();
2942   bool not_null = op->not_null();
2943   bool no_conflict = op->no_conflict();
2944 
2945   Label update, next, none, null_seen, init_klass;
2946 
2947   bool do_null = !not_null;
2948   bool exact_klass_set = exact_klass != nullptr && ciTypeEntries::valid_ciklass(current_klass) == exact_klass;
2949   bool do_update = !TypeEntries::is_type_unknown(current_klass) && !exact_klass_set;
2950 
2951   assert(do_null || do_update, "why are we here?");
2952   assert(!TypeEntries::was_null_seen(current_klass) || do_update, "why are we here?");
2953 
2954   __ verify_oop(obj, FILE_AND_LINE);
2955 
2956   if (do_null || tmp1 != obj DEBUG_ONLY(|| true)) {
2957     __ z_ltgr(tmp1, obj);
2958   }
2959   if (do_null) {
2960     __ z_brnz(update);
2961     if (!TypeEntries::was_null_seen(current_klass)) {
2962       __ z_lg(tmp1, mdo_addr);
2963       __ z_oill(tmp1, TypeEntries::null_seen);
2964       __ z_stg(tmp1, mdo_addr);
2965     }
2966     if (do_update) {
2967       __ z_bru(next);
2968     }
2969   } else {
2970     __ asm_assert(Assembler::bcondNotZero, "unexpected null obj", __LINE__);
2971   }
2972 
2973   __ bind(update);
2974 
2975   if (do_update) {
2976 #ifdef ASSERT
2977     if (exact_klass != nullptr) {
2978       __ load_klass(tmp1, tmp1);
2979       metadata2reg(exact_klass->constant_encoding(), tmp2);
2980       __ z_cgr(tmp1, tmp2);
2981       __ asm_assert(Assembler::bcondEqual, "exact klass and actual klass differ", __LINE__);
2982     }
2983 #endif
2984 
2985     Label do_update;
2986     __ z_lg(tmp2, mdo_addr);
2987 
2988     if (!no_conflict) {
2989       if (exact_klass == nullptr || TypeEntries::is_type_none(current_klass)) {
2990         if (exact_klass != nullptr) {
2991           metadata2reg(exact_klass->constant_encoding(), tmp1);
2992         } else {
2993           __ load_klass(tmp1, tmp1);
2994         }
2995 
2996         // Klass seen before: nothing to do (regardless of unknown bit).
2997         __ z_lgr(Z_R0_scratch, tmp2);
2998         assert(Immediate::is_uimm(~TypeEntries::type_klass_mask, 16), "or change following instruction");
2999         __ z_nill(Z_R0_scratch, TypeEntries::type_klass_mask & 0xFFFF);
3000         __ compareU64_and_branch(Z_R0_scratch, tmp1, Assembler::bcondEqual, next);
3001 
3002         // Already unknown: Nothing to do anymore.
3003         __ z_tmll(tmp2, TypeEntries::type_unknown);
3004         __ z_brc(Assembler::bcondAllOne, next);
3005 
3006         if (TypeEntries::is_type_none(current_klass)) {
3007           __ z_lgr(Z_R0_scratch, tmp2);
3008           assert(Immediate::is_uimm(~TypeEntries::type_mask, 16), "or change following instruction");
3009           __ z_nill(Z_R0_scratch, TypeEntries::type_mask & 0xFFFF);
3010           __ compareU64_and_branch(Z_R0_scratch, (intptr_t)0, Assembler::bcondEqual, init_klass);
3011         }
3012       } else {
3013         assert(ciTypeEntries::valid_ciklass(current_klass) != nullptr &&
3014                ciTypeEntries::valid_ciklass(current_klass) != exact_klass, "conflict only");
3015 
3016         // Already unknown: Nothing to do anymore.
3017         __ z_tmll(tmp2, TypeEntries::type_unknown);
3018         __ z_brc(Assembler::bcondAllOne, next);
3019       }
3020 
3021       // Different than before. Cannot keep accurate profile.
3022       __ z_oill(tmp2, TypeEntries::type_unknown);
3023       __ z_bru(do_update);
3024     } else {
3025       // There's a single possible klass at this profile point.
3026       assert(exact_klass != nullptr, "should be");
3027       if (TypeEntries::is_type_none(current_klass)) {
3028         metadata2reg(exact_klass->constant_encoding(), tmp1);
3029         __ z_lgr(Z_R0_scratch, tmp2);
3030         assert(Immediate::is_uimm(~TypeEntries::type_klass_mask, 16), "or change following instruction");
3031         __ z_nill(Z_R0_scratch, TypeEntries::type_klass_mask & 0xFFFF);
3032         __ compareU64_and_branch(Z_R0_scratch, tmp1, Assembler::bcondEqual, next);
3033 #ifdef ASSERT
3034         {
3035           Label ok;
3036           __ z_lgr(Z_R0_scratch, tmp2);
3037           assert(Immediate::is_uimm(~TypeEntries::type_mask, 16), "or change following instruction");
3038           __ z_nill(Z_R0_scratch, TypeEntries::type_mask & 0xFFFF);
3039           __ compareU64_and_branch(Z_R0_scratch, (intptr_t)0, Assembler::bcondEqual, ok);
3040           __ stop("unexpected profiling mismatch");
3041           __ bind(ok);
3042         }
3043 #endif
3044 
3045       } else {
3046         assert(ciTypeEntries::valid_ciklass(current_klass) != nullptr &&
3047                ciTypeEntries::valid_ciklass(current_klass) != exact_klass, "inconsistent");
3048 
3049         // Already unknown: Nothing to do anymore.
3050         __ z_tmll(tmp2, TypeEntries::type_unknown);
3051         __ z_brc(Assembler::bcondAllOne, next);
3052         __ z_oill(tmp2, TypeEntries::type_unknown);
3053         __ z_bru(do_update);
3054       }
3055     }
3056 
3057     __ bind(init_klass);
3058     // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
3059     __ z_ogr(tmp2, tmp1);
3060 
3061     __ bind(do_update);
3062     __ z_stg(tmp2, mdo_addr);
3063 
3064     __ bind(next);
3065   }
3066 }
3067 
3068 void LIR_Assembler::emit_profile_inline_type(LIR_OpProfileInlineType* op) {
3069   Unimplemented();
3070 }
3071 
3072 void LIR_Assembler::emit_updatecrc32(LIR_OpUpdateCRC32* op) {
3073   assert(op->crc()->is_single_cpu(), "crc must be register");
3074   assert(op->val()->is_single_cpu(), "byte value must be register");
3075   assert(op->result_opr()->is_single_cpu(), "result must be register");
3076   Register crc = op->crc()->as_register();
3077   Register val = op->val()->as_register();
3078   Register res = op->result_opr()->as_register();
3079 
3080   assert_different_registers(val, crc, res);
3081 
3082   __ load_const_optimized(res, StubRoutines::crc_table_addr());
3083   __ kernel_crc32_singleByteReg(crc, val, res, true);
3084   __ z_lgfr(res, crc);
3085 }
3086 
3087 #undef __