1 /*
   2  * Copyright (c) 2000, 2013, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "c1/c1_Compilation.hpp"
  27 #include "c1/c1_LIRAssembler.hpp"
  28 #include "c1/c1_MacroAssembler.hpp"
  29 #include "c1/c1_Runtime1.hpp"
  30 #include "c1/c1_ValueStack.hpp"
  31 #include "ci/ciArrayKlass.hpp"
  32 #include "ci/ciInstance.hpp"
  33 #include "gc_interface/collectedHeap.hpp"
  34 #include "memory/barrierSet.hpp"
  35 #include "memory/cardTableModRefBS.hpp"
  36 #include "nativeInst_sparc.hpp"
  37 #include "oops/objArrayKlass.hpp"
  38 #include "runtime/sharedRuntime.hpp"
  39 
  40 #define __ _masm->
  41 
  42 
  43 //------------------------------------------------------------
  44 
  45 
  46 bool LIR_Assembler::is_small_constant(LIR_Opr opr) {
  47   if (opr->is_constant()) {
  48     LIR_Const* constant = opr->as_constant_ptr();
  49     switch (constant->type()) {
  50       case T_INT: {
  51         jint value = constant->as_jint();
  52         return Assembler::is_simm13(value);
  53       }
  54 
  55       default:
  56         return false;
  57     }
  58   }
  59   return false;
  60 }
  61 
  62 
  63 bool LIR_Assembler::is_single_instruction(LIR_Op* op) {
  64   switch (op->code()) {
  65     case lir_null_check:
  66     return true;
  67 
  68 
  69     case lir_add:
  70     case lir_ushr:
  71     case lir_shr:
  72     case lir_shl:
  73       // integer shifts and adds are always one instruction
  74       return op->result_opr()->is_single_cpu();
  75 
  76 
  77     case lir_move: {
  78       LIR_Op1* op1 = op->as_Op1();
  79       LIR_Opr src = op1->in_opr();
  80       LIR_Opr dst = op1->result_opr();
  81 
  82       if (src == dst) {
  83         NEEDS_CLEANUP;
  84         // this works around a problem where moves with the same src and dst
  85         // end up in the delay slot and then the assembler swallows the mov
  86         // since it has no effect and then it complains because the delay slot
  87         // is empty.  returning false stops the optimizer from putting this in
  88         // the delay slot
  89         return false;
  90       }
  91 
  92       // don't put moves involving oops into the delay slot since the VerifyOops code
  93       // will make it much larger than a single instruction.
  94       if (VerifyOops) {
  95         return false;
  96       }
  97 
  98       if (src->is_double_cpu() || dst->is_double_cpu() || op1->patch_code() != lir_patch_none ||
  99           ((src->is_double_fpu() || dst->is_double_fpu()) && op1->move_kind() != lir_move_normal)) {
 100         return false;
 101       }
 102 
 103       if (UseCompressedOops) {
 104         if (dst->is_address() && !dst->is_stack() && (dst->type() == T_OBJECT || dst->type() == T_ARRAY)) return false;
 105         if (src->is_address() && !src->is_stack() && (src->type() == T_OBJECT || src->type() == T_ARRAY)) return false;
 106       }
 107 
 108       if (UseCompressedKlassPointers) {
 109         if (src->is_address() && !src->is_stack() && src->type() == T_ADDRESS &&
 110             src->as_address_ptr()->disp() == oopDesc::klass_offset_in_bytes()) return false;
 111       }
 112 
 113       if (dst->is_register()) {
 114         if (src->is_address() && Assembler::is_simm13(src->as_address_ptr()->disp())) {
 115           return !PatchALot;
 116         } else if (src->is_single_stack()) {
 117           return true;
 118         }
 119       }
 120 
 121       if (src->is_register()) {
 122         if (dst->is_address() && Assembler::is_simm13(dst->as_address_ptr()->disp())) {
 123           return !PatchALot;
 124         } else if (dst->is_single_stack()) {
 125           return true;
 126         }
 127       }
 128 
 129       if (dst->is_register() &&
 130           ((src->is_register() && src->is_single_word() && src->is_same_type(dst)) ||
 131            (src->is_constant() && LIR_Assembler::is_small_constant(op->as_Op1()->in_opr())))) {
 132         return true;
 133       }
 134 
 135       return false;
 136     }
 137 
 138     default:
 139       return false;
 140   }
 141   ShouldNotReachHere();
 142 }
 143 
 144 
 145 LIR_Opr LIR_Assembler::receiverOpr() {
 146   return FrameMap::O0_oop_opr;
 147 }
 148 
 149 
 150 LIR_Opr LIR_Assembler::osrBufferPointer() {
 151   return FrameMap::I0_opr;
 152 }
 153 
 154 
 155 int LIR_Assembler::initial_frame_size_in_bytes() {
 156   return in_bytes(frame_map()->framesize_in_bytes());
 157 }
 158 
 159 
 160 // inline cache check: the inline cached class is in G5_inline_cache_reg(G5);
 161 // we fetch the class of the receiver (O0) and compare it with the cached class.
 162 // If they do not match we jump to slow case.
 163 int LIR_Assembler::check_icache() {
 164   int offset = __ offset();
 165   __ inline_cache_check(O0, G5_inline_cache_reg);
 166   return offset;
 167 }
 168 
 169 
 170 void LIR_Assembler::osr_entry() {
 171   // On-stack-replacement entry sequence (interpreter frame layout described in interpreter_sparc.cpp):
 172   //
 173   //   1. Create a new compiled activation.
 174   //   2. Initialize local variables in the compiled activation.  The expression stack must be empty
 175   //      at the osr_bci; it is not initialized.
 176   //   3. Jump to the continuation address in compiled code to resume execution.
 177 
 178   // OSR entry point
 179   offsets()->set_value(CodeOffsets::OSR_Entry, code_offset());
 180   BlockBegin* osr_entry = compilation()->hir()->osr_entry();
 181   ValueStack* entry_state = osr_entry->end()->state();
 182   int number_of_locks = entry_state->locks_size();
 183 
 184   // Create a frame for the compiled activation.
 185   __ build_frame(initial_frame_size_in_bytes());
 186 
 187   // OSR buffer is
 188   //
 189   // locals[nlocals-1..0]
 190   // monitors[number_of_locks-1..0]
 191   //
 192   // locals is a direct copy of the interpreter frame so in the osr buffer
 193   // so first slot in the local array is the last local from the interpreter
 194   // and last slot is local[0] (receiver) from the interpreter
 195   //
 196   // Similarly with locks. The first lock slot in the osr buffer is the nth lock
 197   // from the interpreter frame, the nth lock slot in the osr buffer is 0th lock
 198   // in the interpreter frame (the method lock if a sync method)
 199 
 200   // Initialize monitors in the compiled activation.
 201   //   I0: pointer to osr buffer
 202   //
 203   // All other registers are dead at this point and the locals will be
 204   // copied into place by code emitted in the IR.
 205 
 206   Register OSR_buf = osrBufferPointer()->as_register();
 207   { assert(frame::interpreter_frame_monitor_size() == BasicObjectLock::size(), "adjust code below");
 208     int monitor_offset = BytesPerWord * method()->max_locals() +
 209       (2 * BytesPerWord) * (number_of_locks - 1);
 210     // SharedRuntime::OSR_migration_begin() packs BasicObjectLocks in
 211     // the OSR buffer using 2 word entries: first the lock and then
 212     // the oop.
 213     for (int i = 0; i < number_of_locks; i++) {
 214       int slot_offset = monitor_offset - ((i * 2) * BytesPerWord);
 215 #ifdef ASSERT
 216       // verify the interpreter's monitor has a non-null object
 217       {
 218         Label L;
 219         __ ld_ptr(OSR_buf, slot_offset + 1*BytesPerWord, O7);
 220         __ cmp_and_br_short(O7, G0, Assembler::notEqual, Assembler::pt, L);
 221         __ stop("locked object is NULL");
 222         __ bind(L);
 223       }
 224 #endif // ASSERT
 225       // Copy the lock field into the compiled activation.
 226       __ ld_ptr(OSR_buf, slot_offset + 0, O7);
 227       __ st_ptr(O7, frame_map()->address_for_monitor_lock(i));
 228       __ ld_ptr(OSR_buf, slot_offset + 1*BytesPerWord, O7);
 229       __ st_ptr(O7, frame_map()->address_for_monitor_object(i));
 230     }
 231   }
 232 }
 233 
 234 
 235 // Optimized Library calls
 236 // This is the fast version of java.lang.String.compare; it has not
 237 // OSR-entry and therefore, we generate a slow version for OSR's
 238 void LIR_Assembler::emit_string_compare(LIR_Opr left, LIR_Opr right, LIR_Opr dst, CodeEmitInfo* info) {
 239   Register str0 = left->as_register();
 240   Register str1 = right->as_register();
 241 
 242   Label Ldone;
 243 
 244   Register result = dst->as_register();
 245   {
 246     // Get a pointer to the first character of string0 in tmp0
 247     //   and get string0.length() in str0
 248     // Get a pointer to the first character of string1 in tmp1
 249     //   and get string1.length() in str1
 250     // Also, get string0.length()-string1.length() in
 251     //   o7 and get the condition code set
 252     // Note: some instructions have been hoisted for better instruction scheduling
 253 
 254     Register tmp0 = L0;
 255     Register tmp1 = L1;
 256     Register tmp2 = L2;
 257 
 258     int  value_offset = java_lang_String:: value_offset_in_bytes(); // char array
 259     if (java_lang_String::has_offset_field()) {
 260       int offset_offset = java_lang_String::offset_offset_in_bytes(); // first character position
 261       int  count_offset = java_lang_String:: count_offset_in_bytes();
 262       __ load_heap_oop(str0, value_offset, tmp0);
 263       __ ld(str0, offset_offset, tmp2);
 264       __ add(tmp0, arrayOopDesc::base_offset_in_bytes(T_CHAR), tmp0);
 265       __ ld(str0, count_offset, str0);
 266       __ sll(tmp2, exact_log2(sizeof(jchar)), tmp2);
 267     } else {
 268       __ load_heap_oop(str0, value_offset, tmp1);
 269       __ add(tmp1, arrayOopDesc::base_offset_in_bytes(T_CHAR), tmp0);
 270       __ ld(tmp1, arrayOopDesc::length_offset_in_bytes(), str0);
 271     }
 272 
 273     // str1 may be null
 274     add_debug_info_for_null_check_here(info);
 275 
 276     if (java_lang_String::has_offset_field()) {
 277       int offset_offset = java_lang_String::offset_offset_in_bytes(); // first character position
 278       int  count_offset = java_lang_String:: count_offset_in_bytes();
 279       __ load_heap_oop(str1, value_offset, tmp1);
 280       __ add(tmp0, tmp2, tmp0);
 281 
 282       __ ld(str1, offset_offset, tmp2);
 283       __ add(tmp1, arrayOopDesc::base_offset_in_bytes(T_CHAR), tmp1);
 284       __ ld(str1, count_offset, str1);
 285       __ sll(tmp2, exact_log2(sizeof(jchar)), tmp2);
 286       __ add(tmp1, tmp2, tmp1);
 287     } else {
 288       __ load_heap_oop(str1, value_offset, tmp2);
 289       __ add(tmp2, arrayOopDesc::base_offset_in_bytes(T_CHAR), tmp1);
 290       __ ld(tmp2, arrayOopDesc::length_offset_in_bytes(), str1);
 291     }
 292     __ subcc(str0, str1, O7);
 293   }
 294 
 295   {
 296     // Compute the minimum of the string lengths, scale it and store it in limit
 297     Register count0 = I0;
 298     Register count1 = I1;
 299     Register limit  = L3;
 300 
 301     Label Lskip;
 302     __ sll(count0, exact_log2(sizeof(jchar)), limit);             // string0 is shorter
 303     __ br(Assembler::greater, true, Assembler::pt, Lskip);
 304     __ delayed()->sll(count1, exact_log2(sizeof(jchar)), limit);  // string1 is shorter
 305     __ bind(Lskip);
 306 
 307     // If either string is empty (or both of them) the result is the difference in lengths
 308     __ cmp(limit, 0);
 309     __ br(Assembler::equal, true, Assembler::pn, Ldone);
 310     __ delayed()->mov(O7, result);  // result is difference in lengths
 311   }
 312 
 313   {
 314     // Neither string is empty
 315     Label Lloop;
 316 
 317     Register base0 = L0;
 318     Register base1 = L1;
 319     Register chr0  = I0;
 320     Register chr1  = I1;
 321     Register limit = L3;
 322 
 323     // Shift base0 and base1 to the end of the arrays, negate limit
 324     __ add(base0, limit, base0);
 325     __ add(base1, limit, base1);
 326     __ neg(limit);  // limit = -min{string0.length(), string1.length()}
 327 
 328     __ lduh(base0, limit, chr0);
 329     __ bind(Lloop);
 330     __ lduh(base1, limit, chr1);
 331     __ subcc(chr0, chr1, chr0);
 332     __ br(Assembler::notZero, false, Assembler::pn, Ldone);
 333     assert(chr0 == result, "result must be pre-placed");
 334     __ delayed()->inccc(limit, sizeof(jchar));
 335     __ br(Assembler::notZero, true, Assembler::pt, Lloop);
 336     __ delayed()->lduh(base0, limit, chr0);
 337   }
 338 
 339   // If strings are equal up to min length, return the length difference.
 340   __ mov(O7, result);
 341 
 342   // Otherwise, return the difference between the first mismatched chars.
 343   __ bind(Ldone);
 344 }
 345 
 346 
 347 // --------------------------------------------------------------------------------------------
 348 
 349 void LIR_Assembler::monitorexit(LIR_Opr obj_opr, LIR_Opr lock_opr, Register hdr, int monitor_no) {
 350   if (!GenerateSynchronizationCode) return;
 351 
 352   Register obj_reg = obj_opr->as_register();
 353   Register lock_reg = lock_opr->as_register();
 354 
 355   Address mon_addr = frame_map()->address_for_monitor_lock(monitor_no);
 356   Register reg = mon_addr.base();
 357   int offset = mon_addr.disp();
 358   // compute pointer to BasicLock
 359   if (mon_addr.is_simm13()) {
 360     __ add(reg, offset, lock_reg);
 361   }
 362   else {
 363     __ set(offset, lock_reg);
 364     __ add(reg, lock_reg, lock_reg);
 365   }
 366   // unlock object
 367   MonitorAccessStub* slow_case = new MonitorExitStub(lock_opr, UseFastLocking, monitor_no);
 368   // _slow_case_stubs->append(slow_case);
 369   // temporary fix: must be created after exceptionhandler, therefore as call stub
 370   _slow_case_stubs->append(slow_case);
 371   if (UseFastLocking) {
 372     // try inlined fast unlocking first, revert to slow locking if it fails
 373     // note: lock_reg points to the displaced header since the displaced header offset is 0!
 374     assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
 375     __ unlock_object(hdr, obj_reg, lock_reg, *slow_case->entry());
 376   } else {
 377     // always do slow unlocking
 378     // note: the slow unlocking code could be inlined here, however if we use
 379     //       slow unlocking, speed doesn't matter anyway and this solution is
 380     //       simpler and requires less duplicated code - additionally, the
 381     //       slow unlocking code is the same in either case which simplifies
 382     //       debugging
 383     __ br(Assembler::always, false, Assembler::pt, *slow_case->entry());
 384     __ delayed()->nop();
 385   }
 386   // done
 387   __ bind(*slow_case->continuation());
 388 }
 389 
 390 
 391 int LIR_Assembler::emit_exception_handler() {
 392   // if the last instruction is a call (typically to do a throw which
 393   // is coming at the end after block reordering) the return address
 394   // must still point into the code area in order to avoid assertion
 395   // failures when searching for the corresponding bci => add a nop
 396   // (was bug 5/14/1999 - gri)
 397   __ nop();
 398 
 399   // generate code for exception handler
 400   ciMethod* method = compilation()->method();
 401 
 402   address handler_base = __ start_a_stub(exception_handler_size);
 403 
 404   if (handler_base == NULL) {
 405     // not enough space left for the handler
 406     bailout("exception handler overflow");
 407     return -1;
 408   }
 409 
 410   int offset = code_offset();
 411 
 412   __ call(Runtime1::entry_for(Runtime1::handle_exception_from_callee_id), relocInfo::runtime_call_type);
 413   __ delayed()->nop();
 414   __ should_not_reach_here();
 415   guarantee(code_offset() - offset <= exception_handler_size, "overflow");
 416   __ end_a_stub();
 417 
 418   return offset;
 419 }
 420 
 421 
 422 // Emit the code to remove the frame from the stack in the exception
 423 // unwind path.
 424 int LIR_Assembler::emit_unwind_handler() {
 425 #ifndef PRODUCT
 426   if (CommentedAssembly) {
 427     _masm->block_comment("Unwind handler");
 428   }
 429 #endif
 430 
 431   int offset = code_offset();
 432 
 433   // Fetch the exception from TLS and clear out exception related thread state
 434   __ ld_ptr(G2_thread, in_bytes(JavaThread::exception_oop_offset()), O0);
 435   __ st_ptr(G0, G2_thread, in_bytes(JavaThread::exception_oop_offset()));
 436   __ st_ptr(G0, G2_thread, in_bytes(JavaThread::exception_pc_offset()));
 437 
 438   __ bind(_unwind_handler_entry);
 439   __ verify_not_null_oop(O0);
 440   if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
 441     __ mov(O0, I0);  // Preserve the exception
 442   }
 443 
 444   // Preform needed unlocking
 445   MonitorExitStub* stub = NULL;
 446   if (method()->is_synchronized()) {
 447     monitor_address(0, FrameMap::I1_opr);
 448     stub = new MonitorExitStub(FrameMap::I1_opr, true, 0);
 449     __ unlock_object(I3, I2, I1, *stub->entry());
 450     __ bind(*stub->continuation());
 451   }
 452 
 453   if (compilation()->env()->dtrace_method_probes()) {
 454     __ mov(G2_thread, O0);
 455     __ save_thread(I1); // need to preserve thread in G2 across
 456                         // runtime call
 457     metadata2reg(method()->constant_encoding(), O1);
 458     __ call(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), relocInfo::runtime_call_type);
 459     __ delayed()->nop();
 460     __ restore_thread(I1);
 461   }
 462 
 463   if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
 464     __ mov(I0, O0);  // Restore the exception
 465   }
 466 
 467   // dispatch to the unwind logic
 468   __ call(Runtime1::entry_for(Runtime1::unwind_exception_id), relocInfo::runtime_call_type);
 469   __ delayed()->nop();
 470 
 471   // Emit the slow path assembly
 472   if (stub != NULL) {
 473     stub->emit_code(this);
 474   }
 475 
 476   return offset;
 477 }
 478 
 479 
 480 int LIR_Assembler::emit_deopt_handler() {
 481   // if the last instruction is a call (typically to do a throw which
 482   // is coming at the end after block reordering) the return address
 483   // must still point into the code area in order to avoid assertion
 484   // failures when searching for the corresponding bci => add a nop
 485   // (was bug 5/14/1999 - gri)
 486   __ nop();
 487 
 488   // generate code for deopt handler
 489   ciMethod* method = compilation()->method();
 490   address handler_base = __ start_a_stub(deopt_handler_size);
 491   if (handler_base == NULL) {
 492     // not enough space left for the handler
 493     bailout("deopt handler overflow");
 494     return -1;
 495   }
 496 
 497   int offset = code_offset();
 498   AddressLiteral deopt_blob(SharedRuntime::deopt_blob()->unpack());
 499   __ JUMP(deopt_blob, G3_scratch, 0); // sethi;jmp
 500   __ delayed()->nop();
 501   guarantee(code_offset() - offset <= deopt_handler_size, "overflow");
 502   __ end_a_stub();
 503 
 504   return offset;
 505 }
 506 
 507 
 508 void LIR_Assembler::jobject2reg(jobject o, Register reg) {
 509   if (o == NULL) {
 510     __ set(NULL_WORD, reg);
 511   } else {
 512     int oop_index = __ oop_recorder()->find_index(o);
 513     assert(Universe::heap()->is_in_reserved(JNIHandles::resolve(o)), "should be real oop");
 514     RelocationHolder rspec = oop_Relocation::spec(oop_index);
 515     __ set(NULL_WORD, reg, rspec); // Will be set when the nmethod is created
 516   }
 517 }
 518 
 519 
 520 void LIR_Assembler::jobject2reg_with_patching(Register reg, CodeEmitInfo *info) {
 521   // Allocate a new index in table to hold the object once it's been patched
 522   int oop_index = __ oop_recorder()->allocate_oop_index(NULL);
 523   PatchingStub* patch = new PatchingStub(_masm, PatchingStub::load_mirror_id, oop_index);
 524 
 525   AddressLiteral addrlit(NULL, oop_Relocation::spec(oop_index));
 526   assert(addrlit.rspec().type() == relocInfo::oop_type, "must be an oop reloc");
 527   // It may not seem necessary to use a sethi/add pair to load a NULL into dest, but the
 528   // NULL will be dynamically patched later and the patched value may be large.  We must
 529   // therefore generate the sethi/add as a placeholders
 530   __ patchable_set(addrlit, reg);
 531 
 532   patching_epilog(patch, lir_patch_normal, reg, info);
 533 }
 534 
 535 
 536 void LIR_Assembler::metadata2reg(Metadata* o, Register reg) {
 537   __ set_metadata_constant(o, reg);
 538 }
 539 
 540 void LIR_Assembler::klass2reg_with_patching(Register reg, CodeEmitInfo *info) {
 541   // Allocate a new index in table to hold the klass once it's been patched
 542   int index = __ oop_recorder()->allocate_metadata_index(NULL);
 543   PatchingStub* patch = new PatchingStub(_masm, PatchingStub::load_klass_id, index);
 544   AddressLiteral addrlit(NULL, metadata_Relocation::spec(index));
 545   assert(addrlit.rspec().type() == relocInfo::metadata_type, "must be an metadata reloc");
 546   // It may not seem necessary to use a sethi/add pair to load a NULL into dest, but the
 547   // NULL will be dynamically patched later and the patched value may be large.  We must
 548   // therefore generate the sethi/add as a placeholders
 549   __ patchable_set(addrlit, reg);
 550 
 551   patching_epilog(patch, lir_patch_normal, reg, info);
 552 }
 553 
 554 void LIR_Assembler::emit_op3(LIR_Op3* op) {
 555   Register Rdividend = op->in_opr1()->as_register();
 556   Register Rdivisor  = noreg;
 557   Register Rscratch  = op->in_opr3()->as_register();
 558   Register Rresult   = op->result_opr()->as_register();
 559   int divisor = -1;
 560 
 561   if (op->in_opr2()->is_register()) {
 562     Rdivisor = op->in_opr2()->as_register();
 563   } else {
 564     divisor = op->in_opr2()->as_constant_ptr()->as_jint();
 565     assert(Assembler::is_simm13(divisor), "can only handle simm13");
 566   }
 567 
 568   assert(Rdividend != Rscratch, "");
 569   assert(Rdivisor  != Rscratch, "");
 570   assert(op->code() == lir_idiv || op->code() == lir_irem, "Must be irem or idiv");
 571 
 572   if (Rdivisor == noreg && is_power_of_2(divisor)) {
 573     // convert division by a power of two into some shifts and logical operations
 574     if (op->code() == lir_idiv) {
 575       if (divisor == 2) {
 576         __ srl(Rdividend, 31, Rscratch);
 577       } else {
 578         __ sra(Rdividend, 31, Rscratch);
 579         __ and3(Rscratch, divisor - 1, Rscratch);
 580       }
 581       __ add(Rdividend, Rscratch, Rscratch);
 582       __ sra(Rscratch, log2_intptr(divisor), Rresult);
 583       return;
 584     } else {
 585       if (divisor == 2) {
 586         __ srl(Rdividend, 31, Rscratch);
 587       } else {
 588         __ sra(Rdividend, 31, Rscratch);
 589         __ and3(Rscratch, divisor - 1,Rscratch);
 590       }
 591       __ add(Rdividend, Rscratch, Rscratch);
 592       __ andn(Rscratch, divisor - 1,Rscratch);
 593       __ sub(Rdividend, Rscratch, Rresult);
 594       return;
 595     }
 596   }
 597 
 598   __ sra(Rdividend, 31, Rscratch);
 599   __ wry(Rscratch);
 600 
 601   add_debug_info_for_div0_here(op->info());
 602 
 603   if (Rdivisor != noreg) {
 604     __ sdivcc(Rdividend, Rdivisor, (op->code() == lir_idiv ? Rresult : Rscratch));
 605   } else {
 606     assert(Assembler::is_simm13(divisor), "can only handle simm13");
 607     __ sdivcc(Rdividend, divisor, (op->code() == lir_idiv ? Rresult : Rscratch));
 608   }
 609 
 610   Label skip;
 611   __ br(Assembler::overflowSet, true, Assembler::pn, skip);
 612   __ delayed()->Assembler::sethi(0x80000000, (op->code() == lir_idiv ? Rresult : Rscratch));
 613   __ bind(skip);
 614 
 615   if (op->code() == lir_irem) {
 616     if (Rdivisor != noreg) {
 617       __ smul(Rscratch, Rdivisor, Rscratch);
 618     } else {
 619       __ smul(Rscratch, divisor, Rscratch);
 620     }
 621     __ sub(Rdividend, Rscratch, Rresult);
 622   }
 623 }
 624 
 625 
 626 void LIR_Assembler::emit_opBranch(LIR_OpBranch* op) {
 627 #ifdef ASSERT
 628   assert(op->block() == NULL || op->block()->label() == op->label(), "wrong label");
 629   if (op->block() != NULL)  _branch_target_blocks.append(op->block());
 630   if (op->ublock() != NULL) _branch_target_blocks.append(op->ublock());
 631 #endif
 632   assert(op->info() == NULL, "shouldn't have CodeEmitInfo");
 633 
 634   if (op->cond() == lir_cond_always) {
 635     __ br(Assembler::always, false, Assembler::pt, *(op->label()));
 636   } else if (op->code() == lir_cond_float_branch) {
 637     assert(op->ublock() != NULL, "must have unordered successor");
 638     bool is_unordered = (op->ublock() == op->block());
 639     Assembler::Condition acond;
 640     switch (op->cond()) {
 641       case lir_cond_equal:         acond = Assembler::f_equal;    break;
 642       case lir_cond_notEqual:      acond = Assembler::f_notEqual; break;
 643       case lir_cond_less:          acond = (is_unordered ? Assembler::f_unorderedOrLess          : Assembler::f_less);           break;
 644       case lir_cond_greater:       acond = (is_unordered ? Assembler::f_unorderedOrGreater       : Assembler::f_greater);        break;
 645       case lir_cond_lessEqual:     acond = (is_unordered ? Assembler::f_unorderedOrLessOrEqual   : Assembler::f_lessOrEqual);    break;
 646       case lir_cond_greaterEqual:  acond = (is_unordered ? Assembler::f_unorderedOrGreaterOrEqual: Assembler::f_greaterOrEqual); break;
 647       default :                         ShouldNotReachHere();
 648     }
 649     __ fb( acond, false, Assembler::pn, *(op->label()));
 650   } else {
 651     assert (op->code() == lir_branch, "just checking");
 652 
 653     Assembler::Condition acond;
 654     switch (op->cond()) {
 655       case lir_cond_equal:        acond = Assembler::equal;                break;
 656       case lir_cond_notEqual:     acond = Assembler::notEqual;             break;
 657       case lir_cond_less:         acond = Assembler::less;                 break;
 658       case lir_cond_lessEqual:    acond = Assembler::lessEqual;            break;
 659       case lir_cond_greaterEqual: acond = Assembler::greaterEqual;         break;
 660       case lir_cond_greater:      acond = Assembler::greater;              break;
 661       case lir_cond_aboveEqual:   acond = Assembler::greaterEqualUnsigned; break;
 662       case lir_cond_belowEqual:   acond = Assembler::lessEqualUnsigned;    break;
 663       default:                         ShouldNotReachHere();
 664     };
 665 
 666     // sparc has different condition codes for testing 32-bit
 667     // vs. 64-bit values.  We could always test xcc is we could
 668     // guarantee that 32-bit loads always sign extended but that isn't
 669     // true and since sign extension isn't free, it would impose a
 670     // slight cost.
 671 #ifdef _LP64
 672     if  (op->type() == T_INT) {
 673       __ br(acond, false, Assembler::pn, *(op->label()));
 674     } else
 675 #endif
 676       __ brx(acond, false, Assembler::pn, *(op->label()));
 677   }
 678   // The peephole pass fills the delay slot
 679 }
 680 
 681 
 682 void LIR_Assembler::emit_opConvert(LIR_OpConvert* op) {
 683   Bytecodes::Code code = op->bytecode();
 684   LIR_Opr dst = op->result_opr();
 685 
 686   switch(code) {
 687     case Bytecodes::_i2l: {
 688       Register rlo  = dst->as_register_lo();
 689       Register rhi  = dst->as_register_hi();
 690       Register rval = op->in_opr()->as_register();
 691 #ifdef _LP64
 692       __ sra(rval, 0, rlo);
 693 #else
 694       __ mov(rval, rlo);
 695       __ sra(rval, BitsPerInt-1, rhi);
 696 #endif
 697       break;
 698     }
 699     case Bytecodes::_i2d:
 700     case Bytecodes::_i2f: {
 701       bool is_double = (code == Bytecodes::_i2d);
 702       FloatRegister rdst = is_double ? dst->as_double_reg() : dst->as_float_reg();
 703       FloatRegisterImpl::Width w = is_double ? FloatRegisterImpl::D : FloatRegisterImpl::S;
 704       FloatRegister rsrc = op->in_opr()->as_float_reg();
 705       if (rsrc != rdst) {
 706         __ fmov(FloatRegisterImpl::S, rsrc, rdst);
 707       }
 708       __ fitof(w, rdst, rdst);
 709       break;
 710     }
 711     case Bytecodes::_f2i:{
 712       FloatRegister rsrc = op->in_opr()->as_float_reg();
 713       Address       addr = frame_map()->address_for_slot(dst->single_stack_ix());
 714       Label L;
 715       // result must be 0 if value is NaN; test by comparing value to itself
 716       __ fcmp(FloatRegisterImpl::S, Assembler::fcc0, rsrc, rsrc);
 717       __ fb(Assembler::f_unordered, true, Assembler::pn, L);
 718       __ delayed()->st(G0, addr); // annuled if contents of rsrc is not NaN
 719       __ ftoi(FloatRegisterImpl::S, rsrc, rsrc);
 720       // move integer result from float register to int register
 721       __ stf(FloatRegisterImpl::S, rsrc, addr.base(), addr.disp());
 722       __ bind (L);
 723       break;
 724     }
 725     case Bytecodes::_l2i: {
 726       Register rlo  = op->in_opr()->as_register_lo();
 727       Register rhi  = op->in_opr()->as_register_hi();
 728       Register rdst = dst->as_register();
 729 #ifdef _LP64
 730       __ sra(rlo, 0, rdst);
 731 #else
 732       __ mov(rlo, rdst);
 733 #endif
 734       break;
 735     }
 736     case Bytecodes::_d2f:
 737     case Bytecodes::_f2d: {
 738       bool is_double = (code == Bytecodes::_f2d);
 739       assert((!is_double && dst->is_single_fpu()) || (is_double && dst->is_double_fpu()), "check");
 740       LIR_Opr val = op->in_opr();
 741       FloatRegister rval = (code == Bytecodes::_d2f) ? val->as_double_reg() : val->as_float_reg();
 742       FloatRegister rdst = is_double ? dst->as_double_reg() : dst->as_float_reg();
 743       FloatRegisterImpl::Width vw = is_double ? FloatRegisterImpl::S : FloatRegisterImpl::D;
 744       FloatRegisterImpl::Width dw = is_double ? FloatRegisterImpl::D : FloatRegisterImpl::S;
 745       __ ftof(vw, dw, rval, rdst);
 746       break;
 747     }
 748     case Bytecodes::_i2s:
 749     case Bytecodes::_i2b: {
 750       Register rval = op->in_opr()->as_register();
 751       Register rdst = dst->as_register();
 752       int shift = (code == Bytecodes::_i2b) ? (BitsPerInt - T_BYTE_aelem_bytes * BitsPerByte) : (BitsPerInt - BitsPerShort);
 753       __ sll (rval, shift, rdst);
 754       __ sra (rdst, shift, rdst);
 755       break;
 756     }
 757     case Bytecodes::_i2c: {
 758       Register rval = op->in_opr()->as_register();
 759       Register rdst = dst->as_register();
 760       int shift = BitsPerInt - T_CHAR_aelem_bytes * BitsPerByte;
 761       __ sll (rval, shift, rdst);
 762       __ srl (rdst, shift, rdst);
 763       break;
 764     }
 765 
 766     default: ShouldNotReachHere();
 767   }
 768 }
 769 
 770 
 771 void LIR_Assembler::align_call(LIR_Code) {
 772   // do nothing since all instructions are word aligned on sparc
 773 }
 774 
 775 
 776 void LIR_Assembler::call(LIR_OpJavaCall* op, relocInfo::relocType rtype) {
 777   __ call(op->addr(), rtype);
 778   // The peephole pass fills the delay slot, add_call_info is done in
 779   // LIR_Assembler::emit_delay.
 780 }
 781 
 782 
 783 void LIR_Assembler::ic_call(LIR_OpJavaCall* op) {
 784   __ ic_call(op->addr(), false);
 785   // The peephole pass fills the delay slot, add_call_info is done in
 786   // LIR_Assembler::emit_delay.
 787 }
 788 
 789 
 790 void LIR_Assembler::vtable_call(LIR_OpJavaCall* op) {
 791   add_debug_info_for_null_check_here(op->info());
 792   __ load_klass(O0, G3_scratch);
 793   if (Assembler::is_simm13(op->vtable_offset())) {
 794     __ ld_ptr(G3_scratch, op->vtable_offset(), G5_method);
 795   } else {
 796     // This will generate 2 instructions
 797     __ set(op->vtable_offset(), G5_method);
 798     // ld_ptr, set_hi, set
 799     __ ld_ptr(G3_scratch, G5_method, G5_method);
 800   }
 801   __ ld_ptr(G5_method, Method::from_compiled_offset(), G3_scratch);
 802   __ callr(G3_scratch, G0);
 803   // the peephole pass fills the delay slot
 804 }
 805 
 806 int LIR_Assembler::store(LIR_Opr from_reg, Register base, int offset, BasicType type, bool wide, bool unaligned) {
 807   int store_offset;
 808   if (!Assembler::is_simm13(offset + (type == T_LONG) ? wordSize : 0)) {
 809     assert(!unaligned, "can't handle this");
 810     // for offsets larger than a simm13 we setup the offset in O7
 811     __ set(offset, O7);
 812     store_offset = store(from_reg, base, O7, type, wide);
 813   } else {
 814     if (type == T_ARRAY || type == T_OBJECT) {
 815       __ verify_oop(from_reg->as_register());
 816     }
 817     store_offset = code_offset();
 818     switch (type) {
 819       case T_BOOLEAN: // fall through
 820       case T_BYTE  : __ stb(from_reg->as_register(), base, offset); break;
 821       case T_CHAR  : __ sth(from_reg->as_register(), base, offset); break;
 822       case T_SHORT : __ sth(from_reg->as_register(), base, offset); break;
 823       case T_INT   : __ stw(from_reg->as_register(), base, offset); break;
 824       case T_LONG  :
 825 #ifdef _LP64
 826         if (unaligned || PatchALot) {
 827           __ srax(from_reg->as_register_lo(), 32, O7);
 828           __ stw(from_reg->as_register_lo(), base, offset + lo_word_offset_in_bytes);
 829           __ stw(O7,                         base, offset + hi_word_offset_in_bytes);
 830         } else {
 831           __ stx(from_reg->as_register_lo(), base, offset);
 832         }
 833 #else
 834         assert(Assembler::is_simm13(offset + 4), "must be");
 835         __ stw(from_reg->as_register_lo(), base, offset + lo_word_offset_in_bytes);
 836         __ stw(from_reg->as_register_hi(), base, offset + hi_word_offset_in_bytes);
 837 #endif
 838         break;
 839       case T_ADDRESS:
 840       case T_METADATA:
 841         __ st_ptr(from_reg->as_register(), base, offset);
 842         break;
 843       case T_ARRAY : // fall through
 844       case T_OBJECT:
 845         {
 846           if (UseCompressedOops && !wide) {
 847             __ encode_heap_oop(from_reg->as_register(), G3_scratch);
 848             store_offset = code_offset();
 849             __ stw(G3_scratch, base, offset);
 850           } else {
 851             __ st_ptr(from_reg->as_register(), base, offset);
 852           }
 853           break;
 854         }
 855 
 856       case T_FLOAT : __ stf(FloatRegisterImpl::S, from_reg->as_float_reg(), base, offset); break;
 857       case T_DOUBLE:
 858         {
 859           FloatRegister reg = from_reg->as_double_reg();
 860           // split unaligned stores
 861           if (unaligned || PatchALot) {
 862             assert(Assembler::is_simm13(offset + 4), "must be");
 863             __ stf(FloatRegisterImpl::S, reg->successor(), base, offset + 4);
 864             __ stf(FloatRegisterImpl::S, reg,              base, offset);
 865           } else {
 866             __ stf(FloatRegisterImpl::D, reg, base, offset);
 867           }
 868           break;
 869         }
 870       default      : ShouldNotReachHere();
 871     }
 872   }
 873   return store_offset;
 874 }
 875 
 876 
 877 int LIR_Assembler::store(LIR_Opr from_reg, Register base, Register disp, BasicType type, bool wide) {
 878   if (type == T_ARRAY || type == T_OBJECT) {
 879     __ verify_oop(from_reg->as_register());
 880   }
 881   int store_offset = code_offset();
 882   switch (type) {
 883     case T_BOOLEAN: // fall through
 884     case T_BYTE  : __ stb(from_reg->as_register(), base, disp); break;
 885     case T_CHAR  : __ sth(from_reg->as_register(), base, disp); break;
 886     case T_SHORT : __ sth(from_reg->as_register(), base, disp); break;
 887     case T_INT   : __ stw(from_reg->as_register(), base, disp); break;
 888     case T_LONG  :
 889 #ifdef _LP64
 890       __ stx(from_reg->as_register_lo(), base, disp);
 891 #else
 892       assert(from_reg->as_register_hi()->successor() == from_reg->as_register_lo(), "must match");
 893       __ std(from_reg->as_register_hi(), base, disp);
 894 #endif
 895       break;
 896     case T_ADDRESS:
 897       __ st_ptr(from_reg->as_register(), base, disp);
 898       break;
 899     case T_ARRAY : // fall through
 900     case T_OBJECT:
 901       {
 902         if (UseCompressedOops && !wide) {
 903           __ encode_heap_oop(from_reg->as_register(), G3_scratch);
 904           store_offset = code_offset();
 905           __ stw(G3_scratch, base, disp);
 906         } else {
 907           __ st_ptr(from_reg->as_register(), base, disp);
 908         }
 909         break;
 910       }
 911     case T_FLOAT : __ stf(FloatRegisterImpl::S, from_reg->as_float_reg(), base, disp); break;
 912     case T_DOUBLE: __ stf(FloatRegisterImpl::D, from_reg->as_double_reg(), base, disp); break;
 913     default      : ShouldNotReachHere();
 914   }
 915   return store_offset;
 916 }
 917 
 918 
 919 int LIR_Assembler::load(Register base, int offset, LIR_Opr to_reg, BasicType type, bool wide, bool unaligned) {
 920   int load_offset;
 921   if (!Assembler::is_simm13(offset + (type == T_LONG) ? wordSize : 0)) {
 922     assert(base != O7, "destroying register");
 923     assert(!unaligned, "can't handle this");
 924     // for offsets larger than a simm13 we setup the offset in O7
 925     __ set(offset, O7);
 926     load_offset = load(base, O7, to_reg, type, wide);
 927   } else {
 928     load_offset = code_offset();
 929     switch(type) {
 930       case T_BOOLEAN: // fall through
 931       case T_BYTE  : __ ldsb(base, offset, to_reg->as_register()); break;
 932       case T_CHAR  : __ lduh(base, offset, to_reg->as_register()); break;
 933       case T_SHORT : __ ldsh(base, offset, to_reg->as_register()); break;
 934       case T_INT   : __ ld(base, offset, to_reg->as_register()); break;
 935       case T_LONG  :
 936         if (!unaligned) {
 937 #ifdef _LP64
 938           __ ldx(base, offset, to_reg->as_register_lo());
 939 #else
 940           assert(to_reg->as_register_hi()->successor() == to_reg->as_register_lo(),
 941                  "must be sequential");
 942           __ ldd(base, offset, to_reg->as_register_hi());
 943 #endif
 944         } else {
 945 #ifdef _LP64
 946           assert(base != to_reg->as_register_lo(), "can't handle this");
 947           assert(O7 != to_reg->as_register_lo(), "can't handle this");
 948           __ ld(base, offset + hi_word_offset_in_bytes, to_reg->as_register_lo());
 949           __ lduw(base, offset + lo_word_offset_in_bytes, O7); // in case O7 is base or offset, use it last
 950           __ sllx(to_reg->as_register_lo(), 32, to_reg->as_register_lo());
 951           __ or3(to_reg->as_register_lo(), O7, to_reg->as_register_lo());
 952 #else
 953           if (base == to_reg->as_register_lo()) {
 954             __ ld(base, offset + hi_word_offset_in_bytes, to_reg->as_register_hi());
 955             __ ld(base, offset + lo_word_offset_in_bytes, to_reg->as_register_lo());
 956           } else {
 957             __ ld(base, offset + lo_word_offset_in_bytes, to_reg->as_register_lo());
 958             __ ld(base, offset + hi_word_offset_in_bytes, to_reg->as_register_hi());
 959           }
 960 #endif
 961         }
 962         break;
 963       case T_METADATA:  __ ld_ptr(base, offset, to_reg->as_register()); break;
 964       case T_ADDRESS:
 965 #ifdef _LP64
 966         if (offset == oopDesc::klass_offset_in_bytes() && UseCompressedKlassPointers) {
 967           __ lduw(base, offset, to_reg->as_register());
 968           __ decode_klass_not_null(to_reg->as_register());
 969         } else
 970 #endif
 971         {
 972           __ ld_ptr(base, offset, to_reg->as_register());
 973         }
 974         break;
 975       case T_ARRAY : // fall through
 976       case T_OBJECT:
 977         {
 978           if (UseCompressedOops && !wide) {
 979             __ lduw(base, offset, to_reg->as_register());
 980             __ decode_heap_oop(to_reg->as_register());
 981           } else {
 982             __ ld_ptr(base, offset, to_reg->as_register());
 983           }
 984           break;
 985         }
 986       case T_FLOAT:  __ ldf(FloatRegisterImpl::S, base, offset, to_reg->as_float_reg()); break;
 987       case T_DOUBLE:
 988         {
 989           FloatRegister reg = to_reg->as_double_reg();
 990           // split unaligned loads
 991           if (unaligned || PatchALot) {
 992             __ ldf(FloatRegisterImpl::S, base, offset + 4, reg->successor());
 993             __ ldf(FloatRegisterImpl::S, base, offset,     reg);
 994           } else {
 995             __ ldf(FloatRegisterImpl::D, base, offset, to_reg->as_double_reg());
 996           }
 997           break;
 998         }
 999       default      : ShouldNotReachHere();
1000     }
1001     if (type == T_ARRAY || type == T_OBJECT) {
1002       __ verify_oop(to_reg->as_register());
1003     }
1004   }
1005   return load_offset;
1006 }
1007 
1008 
1009 int LIR_Assembler::load(Register base, Register disp, LIR_Opr to_reg, BasicType type, bool wide) {
1010   int load_offset = code_offset();
1011   switch(type) {
1012     case T_BOOLEAN: // fall through
1013     case T_BYTE  :  __ ldsb(base, disp, to_reg->as_register()); break;
1014     case T_CHAR  :  __ lduh(base, disp, to_reg->as_register()); break;
1015     case T_SHORT :  __ ldsh(base, disp, to_reg->as_register()); break;
1016     case T_INT   :  __ ld(base, disp, to_reg->as_register()); break;
1017     case T_ADDRESS: __ ld_ptr(base, disp, to_reg->as_register()); break;
1018     case T_ARRAY : // fall through
1019     case T_OBJECT:
1020       {
1021           if (UseCompressedOops && !wide) {
1022             __ lduw(base, disp, to_reg->as_register());
1023             __ decode_heap_oop(to_reg->as_register());
1024           } else {
1025             __ ld_ptr(base, disp, to_reg->as_register());
1026           }
1027           break;
1028       }
1029     case T_FLOAT:  __ ldf(FloatRegisterImpl::S, base, disp, to_reg->as_float_reg()); break;
1030     case T_DOUBLE: __ ldf(FloatRegisterImpl::D, base, disp, to_reg->as_double_reg()); break;
1031     case T_LONG  :
1032 #ifdef _LP64
1033       __ ldx(base, disp, to_reg->as_register_lo());
1034 #else
1035       assert(to_reg->as_register_hi()->successor() == to_reg->as_register_lo(),
1036              "must be sequential");
1037       __ ldd(base, disp, to_reg->as_register_hi());
1038 #endif
1039       break;
1040     default      : ShouldNotReachHere();
1041   }
1042   if (type == T_ARRAY || type == T_OBJECT) {
1043     __ verify_oop(to_reg->as_register());
1044   }
1045   return load_offset;
1046 }
1047 
1048 void LIR_Assembler::const2stack(LIR_Opr src, LIR_Opr dest) {
1049   LIR_Const* c = src->as_constant_ptr();
1050   switch (c->type()) {
1051     case T_INT:
1052     case T_FLOAT: {
1053       Register src_reg = O7;
1054       int value = c->as_jint_bits();
1055       if (value == 0) {
1056         src_reg = G0;
1057       } else {
1058         __ set(value, O7);
1059       }
1060       Address addr = frame_map()->address_for_slot(dest->single_stack_ix());
1061       __ stw(src_reg, addr.base(), addr.disp());
1062       break;
1063     }
1064     case T_ADDRESS: {
1065       Register src_reg = O7;
1066       int value = c->as_jint_bits();
1067       if (value == 0) {
1068         src_reg = G0;
1069       } else {
1070         __ set(value, O7);
1071       }
1072       Address addr = frame_map()->address_for_slot(dest->single_stack_ix());
1073       __ st_ptr(src_reg, addr.base(), addr.disp());
1074       break;
1075     }
1076     case T_OBJECT: {
1077       Register src_reg = O7;
1078       jobject2reg(c->as_jobject(), src_reg);
1079       Address addr = frame_map()->address_for_slot(dest->single_stack_ix());
1080       __ st_ptr(src_reg, addr.base(), addr.disp());
1081       break;
1082     }
1083     case T_LONG:
1084     case T_DOUBLE: {
1085       Address addr = frame_map()->address_for_double_slot(dest->double_stack_ix());
1086 
1087       Register tmp = O7;
1088       int value_lo = c->as_jint_lo_bits();
1089       if (value_lo == 0) {
1090         tmp = G0;
1091       } else {
1092         __ set(value_lo, O7);
1093       }
1094       __ stw(tmp, addr.base(), addr.disp() + lo_word_offset_in_bytes);
1095       int value_hi = c->as_jint_hi_bits();
1096       if (value_hi == 0) {
1097         tmp = G0;
1098       } else {
1099         __ set(value_hi, O7);
1100       }
1101       __ stw(tmp, addr.base(), addr.disp() + hi_word_offset_in_bytes);
1102       break;
1103     }
1104     default:
1105       Unimplemented();
1106   }
1107 }
1108 
1109 
1110 void LIR_Assembler::const2mem(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info, bool wide) {
1111   LIR_Const* c = src->as_constant_ptr();
1112   LIR_Address* addr     = dest->as_address_ptr();
1113   Register base = addr->base()->as_pointer_register();
1114   int offset = -1;
1115 
1116   switch (c->type()) {
1117     case T_INT:
1118     case T_FLOAT:
1119     case T_ADDRESS: {
1120       LIR_Opr tmp = FrameMap::O7_opr;
1121       int value = c->as_jint_bits();
1122       if (value == 0) {
1123         tmp = FrameMap::G0_opr;
1124       } else if (Assembler::is_simm13(value)) {
1125         __ set(value, O7);
1126       }
1127       if (addr->index()->is_valid()) {
1128         assert(addr->disp() == 0, "must be zero");
1129         offset = store(tmp, base, addr->index()->as_pointer_register(), type, wide);
1130       } else {
1131         assert(Assembler::is_simm13(addr->disp()), "can't handle larger addresses");
1132         offset = store(tmp, base, addr->disp(), type, wide, false);
1133       }
1134       break;
1135     }
1136     case T_LONG:
1137     case T_DOUBLE: {
1138       assert(!addr->index()->is_valid(), "can't handle reg reg address here");
1139       assert(Assembler::is_simm13(addr->disp()) &&
1140              Assembler::is_simm13(addr->disp() + 4), "can't handle larger addresses");
1141 
1142       LIR_Opr tmp = FrameMap::O7_opr;
1143       int value_lo = c->as_jint_lo_bits();
1144       if (value_lo == 0) {
1145         tmp = FrameMap::G0_opr;
1146       } else {
1147         __ set(value_lo, O7);
1148       }
1149       offset = store(tmp, base, addr->disp() + lo_word_offset_in_bytes, T_INT, wide, false);
1150       int value_hi = c->as_jint_hi_bits();
1151       if (value_hi == 0) {
1152         tmp = FrameMap::G0_opr;
1153       } else {
1154         __ set(value_hi, O7);
1155       }
1156       store(tmp, base, addr->disp() + hi_word_offset_in_bytes, T_INT, wide, false);
1157       break;
1158     }
1159     case T_OBJECT: {
1160       jobject obj = c->as_jobject();
1161       LIR_Opr tmp;
1162       if (obj == NULL) {
1163         tmp = FrameMap::G0_opr;
1164       } else {
1165         tmp = FrameMap::O7_opr;
1166         jobject2reg(c->as_jobject(), O7);
1167       }
1168       // handle either reg+reg or reg+disp address
1169       if (addr->index()->is_valid()) {
1170         assert(addr->disp() == 0, "must be zero");
1171         offset = store(tmp, base, addr->index()->as_pointer_register(), type, wide);
1172       } else {
1173         assert(Assembler::is_simm13(addr->disp()), "can't handle larger addresses");
1174         offset = store(tmp, base, addr->disp(), type, wide, false);
1175       }
1176 
1177       break;
1178     }
1179     default:
1180       Unimplemented();
1181   }
1182   if (info != NULL) {
1183     assert(offset != -1, "offset should've been set");
1184     add_debug_info_for_null_check(offset, info);
1185   }
1186 }
1187 
1188 
1189 void LIR_Assembler::const2reg(LIR_Opr src, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
1190   LIR_Const* c = src->as_constant_ptr();
1191   LIR_Opr to_reg = dest;
1192 
1193   switch (c->type()) {
1194     case T_INT:
1195     case T_ADDRESS:
1196       {
1197         jint con = c->as_jint();
1198         if (to_reg->is_single_cpu()) {
1199           assert(patch_code == lir_patch_none, "no patching handled here");
1200           __ set(con, to_reg->as_register());
1201         } else {
1202           ShouldNotReachHere();
1203           assert(to_reg->is_single_fpu(), "wrong register kind");
1204 
1205           __ set(con, O7);
1206           Address temp_slot(SP, (frame::register_save_words * wordSize) + STACK_BIAS);
1207           __ st(O7, temp_slot);
1208           __ ldf(FloatRegisterImpl::S, temp_slot, to_reg->as_float_reg());
1209         }
1210       }
1211       break;
1212 
1213     case T_LONG:
1214       {
1215         jlong con = c->as_jlong();
1216 
1217         if (to_reg->is_double_cpu()) {
1218 #ifdef _LP64
1219           __ set(con,  to_reg->as_register_lo());
1220 #else
1221           __ set(low(con),  to_reg->as_register_lo());
1222           __ set(high(con), to_reg->as_register_hi());
1223 #endif
1224 #ifdef _LP64
1225         } else if (to_reg->is_single_cpu()) {
1226           __ set(con, to_reg->as_register());
1227 #endif
1228         } else {
1229           ShouldNotReachHere();
1230           assert(to_reg->is_double_fpu(), "wrong register kind");
1231           Address temp_slot_lo(SP, ((frame::register_save_words  ) * wordSize) + STACK_BIAS);
1232           Address temp_slot_hi(SP, ((frame::register_save_words) * wordSize) + (longSize/2) + STACK_BIAS);
1233           __ set(low(con),  O7);
1234           __ st(O7, temp_slot_lo);
1235           __ set(high(con), O7);
1236           __ st(O7, temp_slot_hi);
1237           __ ldf(FloatRegisterImpl::D, temp_slot_lo, to_reg->as_double_reg());
1238         }
1239       }
1240       break;
1241 
1242     case T_OBJECT:
1243       {
1244         if (patch_code == lir_patch_none) {
1245           jobject2reg(c->as_jobject(), to_reg->as_register());
1246         } else {
1247           jobject2reg_with_patching(to_reg->as_register(), info);
1248         }
1249       }
1250       break;
1251 
1252     case T_METADATA:
1253       {
1254         if (patch_code == lir_patch_none) {
1255           metadata2reg(c->as_metadata(), to_reg->as_register());
1256         } else {
1257           klass2reg_with_patching(to_reg->as_register(), info);
1258         }
1259       }
1260       break;
1261 
1262     case T_FLOAT:
1263       {
1264         address const_addr = __ float_constant(c->as_jfloat());
1265         if (const_addr == NULL) {
1266           bailout("const section overflow");
1267           break;
1268         }
1269         RelocationHolder rspec = internal_word_Relocation::spec(const_addr);
1270         AddressLiteral const_addrlit(const_addr, rspec);
1271         if (to_reg->is_single_fpu()) {
1272           __ patchable_sethi(const_addrlit, O7);
1273           __ relocate(rspec);
1274           __ ldf(FloatRegisterImpl::S, O7, const_addrlit.low10(), to_reg->as_float_reg());
1275 
1276         } else {
1277           assert(to_reg->is_single_cpu(), "Must be a cpu register.");
1278 
1279           __ set(const_addrlit, O7);
1280           __ ld(O7, 0, to_reg->as_register());
1281         }
1282       }
1283       break;
1284 
1285     case T_DOUBLE:
1286       {
1287         address const_addr = __ double_constant(c->as_jdouble());
1288         if (const_addr == NULL) {
1289           bailout("const section overflow");
1290           break;
1291         }
1292         RelocationHolder rspec = internal_word_Relocation::spec(const_addr);
1293 
1294         if (to_reg->is_double_fpu()) {
1295           AddressLiteral const_addrlit(const_addr, rspec);
1296           __ patchable_sethi(const_addrlit, O7);
1297           __ relocate(rspec);
1298           __ ldf (FloatRegisterImpl::D, O7, const_addrlit.low10(), to_reg->as_double_reg());
1299         } else {
1300           assert(to_reg->is_double_cpu(), "Must be a long register.");
1301 #ifdef _LP64
1302           __ set(jlong_cast(c->as_jdouble()), to_reg->as_register_lo());
1303 #else
1304           __ set(low(jlong_cast(c->as_jdouble())), to_reg->as_register_lo());
1305           __ set(high(jlong_cast(c->as_jdouble())), to_reg->as_register_hi());
1306 #endif
1307         }
1308 
1309       }
1310       break;
1311 
1312     default:
1313       ShouldNotReachHere();
1314   }
1315 }
1316 
1317 Address LIR_Assembler::as_Address(LIR_Address* addr) {
1318   Register reg = addr->base()->as_register();
1319   LIR_Opr index = addr->index();
1320   if (index->is_illegal()) {
1321     return Address(reg, addr->disp());
1322   } else {
1323     assert (addr->disp() == 0, "unsupported address mode");
1324     return Address(reg, index->as_pointer_register());
1325   }
1326 }
1327 
1328 
1329 void LIR_Assembler::stack2stack(LIR_Opr src, LIR_Opr dest, BasicType type) {
1330   switch (type) {
1331     case T_INT:
1332     case T_FLOAT: {
1333       Register tmp = O7;
1334       Address from = frame_map()->address_for_slot(src->single_stack_ix());
1335       Address to   = frame_map()->address_for_slot(dest->single_stack_ix());
1336       __ lduw(from.base(), from.disp(), tmp);
1337       __ stw(tmp, to.base(), to.disp());
1338       break;
1339     }
1340     case T_OBJECT: {
1341       Register tmp = O7;
1342       Address from = frame_map()->address_for_slot(src->single_stack_ix());
1343       Address to   = frame_map()->address_for_slot(dest->single_stack_ix());
1344       __ ld_ptr(from.base(), from.disp(), tmp);
1345       __ st_ptr(tmp, to.base(), to.disp());
1346       break;
1347     }
1348     case T_LONG:
1349     case T_DOUBLE: {
1350       Register tmp = O7;
1351       Address from = frame_map()->address_for_double_slot(src->double_stack_ix());
1352       Address to   = frame_map()->address_for_double_slot(dest->double_stack_ix());
1353       __ lduw(from.base(), from.disp(), tmp);
1354       __ stw(tmp, to.base(), to.disp());
1355       __ lduw(from.base(), from.disp() + 4, tmp);
1356       __ stw(tmp, to.base(), to.disp() + 4);
1357       break;
1358     }
1359 
1360     default:
1361       ShouldNotReachHere();
1362   }
1363 }
1364 
1365 
1366 Address LIR_Assembler::as_Address_hi(LIR_Address* addr) {
1367   Address base = as_Address(addr);
1368   return Address(base.base(), base.disp() + hi_word_offset_in_bytes);
1369 }
1370 
1371 
1372 Address LIR_Assembler::as_Address_lo(LIR_Address* addr) {
1373   Address base = as_Address(addr);
1374   return Address(base.base(), base.disp() + lo_word_offset_in_bytes);
1375 }
1376 
1377 
1378 void LIR_Assembler::mem2reg(LIR_Opr src_opr, LIR_Opr dest, BasicType type,
1379                             LIR_PatchCode patch_code, CodeEmitInfo* info, bool wide, bool unaligned) {
1380 
1381   assert(type != T_METADATA, "load of metadata ptr not supported");
1382   LIR_Address* addr = src_opr->as_address_ptr();
1383   LIR_Opr to_reg = dest;
1384 
1385   Register src = addr->base()->as_pointer_register();
1386   Register disp_reg = noreg;
1387   int disp_value = addr->disp();
1388   bool needs_patching = (patch_code != lir_patch_none);
1389 
1390   if (addr->base()->type() == T_OBJECT) {
1391     __ verify_oop(src);
1392   }
1393 
1394   PatchingStub* patch = NULL;
1395   if (needs_patching) {
1396     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1397     assert(!to_reg->is_double_cpu() ||
1398            patch_code == lir_patch_none ||
1399            patch_code == lir_patch_normal, "patching doesn't match register");
1400   }
1401 
1402   if (addr->index()->is_illegal()) {
1403     if (!Assembler::is_simm13(disp_value) && (!unaligned || Assembler::is_simm13(disp_value + 4))) {
1404       if (needs_patching) {
1405         __ patchable_set(0, O7);
1406       } else {
1407         __ set(disp_value, O7);
1408       }
1409       disp_reg = O7;
1410     }
1411   } else if (unaligned || PatchALot) {
1412     __ add(src, addr->index()->as_register(), O7);
1413     src = O7;
1414   } else {
1415     disp_reg = addr->index()->as_pointer_register();
1416     assert(disp_value == 0, "can't handle 3 operand addresses");
1417   }
1418 
1419   // remember the offset of the load.  The patching_epilog must be done
1420   // before the call to add_debug_info, otherwise the PcDescs don't get
1421   // entered in increasing order.
1422   int offset = code_offset();
1423 
1424   assert(disp_reg != noreg || Assembler::is_simm13(disp_value), "should have set this up");
1425   if (disp_reg == noreg) {
1426     offset = load(src, disp_value, to_reg, type, wide, unaligned);
1427   } else {
1428     assert(!unaligned, "can't handle this");
1429     offset = load(src, disp_reg, to_reg, type, wide);
1430   }
1431 
1432   if (patch != NULL) {
1433     patching_epilog(patch, patch_code, src, info);
1434   }
1435   if (info != NULL) add_debug_info_for_null_check(offset, info);
1436 }
1437 
1438 
1439 void LIR_Assembler::prefetchr(LIR_Opr src) {
1440   LIR_Address* addr = src->as_address_ptr();
1441   Address from_addr = as_Address(addr);
1442 
1443   if (VM_Version::has_v9()) {
1444     __ prefetch(from_addr, Assembler::severalReads);
1445   }
1446 }
1447 
1448 
1449 void LIR_Assembler::prefetchw(LIR_Opr src) {
1450   LIR_Address* addr = src->as_address_ptr();
1451   Address from_addr = as_Address(addr);
1452 
1453   if (VM_Version::has_v9()) {
1454     __ prefetch(from_addr, Assembler::severalWritesAndPossiblyReads);
1455   }
1456 }
1457 
1458 
1459 void LIR_Assembler::stack2reg(LIR_Opr src, LIR_Opr dest, BasicType type) {
1460   Address addr;
1461   if (src->is_single_word()) {
1462     addr = frame_map()->address_for_slot(src->single_stack_ix());
1463   } else if (src->is_double_word())  {
1464     addr = frame_map()->address_for_double_slot(src->double_stack_ix());
1465   }
1466 
1467   bool unaligned = (addr.disp() - STACK_BIAS) % 8 != 0;
1468   load(addr.base(), addr.disp(), dest, dest->type(), true /*wide*/, unaligned);
1469 }
1470 
1471 
1472 void LIR_Assembler::reg2stack(LIR_Opr from_reg, LIR_Opr dest, BasicType type, bool pop_fpu_stack) {
1473   Address addr;
1474   if (dest->is_single_word()) {
1475     addr = frame_map()->address_for_slot(dest->single_stack_ix());
1476   } else if (dest->is_double_word())  {
1477     addr = frame_map()->address_for_slot(dest->double_stack_ix());
1478   }
1479   bool unaligned = (addr.disp() - STACK_BIAS) % 8 != 0;
1480   store(from_reg, addr.base(), addr.disp(), from_reg->type(), true /*wide*/, unaligned);
1481 }
1482 
1483 
1484 void LIR_Assembler::reg2reg(LIR_Opr from_reg, LIR_Opr to_reg) {
1485   if (from_reg->is_float_kind() && to_reg->is_float_kind()) {
1486     if (from_reg->is_double_fpu()) {
1487       // double to double moves
1488       assert(to_reg->is_double_fpu(), "should match");
1489       __ fmov(FloatRegisterImpl::D, from_reg->as_double_reg(), to_reg->as_double_reg());
1490     } else {
1491       // float to float moves
1492       assert(to_reg->is_single_fpu(), "should match");
1493       __ fmov(FloatRegisterImpl::S, from_reg->as_float_reg(), to_reg->as_float_reg());
1494     }
1495   } else if (!from_reg->is_float_kind() && !to_reg->is_float_kind()) {
1496     if (from_reg->is_double_cpu()) {
1497 #ifdef _LP64
1498       __ mov(from_reg->as_pointer_register(), to_reg->as_pointer_register());
1499 #else
1500       assert(to_reg->is_double_cpu() &&
1501              from_reg->as_register_hi() != to_reg->as_register_lo() &&
1502              from_reg->as_register_lo() != to_reg->as_register_hi(),
1503              "should both be long and not overlap");
1504       // long to long moves
1505       __ mov(from_reg->as_register_hi(), to_reg->as_register_hi());
1506       __ mov(from_reg->as_register_lo(), to_reg->as_register_lo());
1507 #endif
1508 #ifdef _LP64
1509     } else if (to_reg->is_double_cpu()) {
1510       // int to int moves
1511       __ mov(from_reg->as_register(), to_reg->as_register_lo());
1512 #endif
1513     } else {
1514       // int to int moves
1515       __ mov(from_reg->as_register(), to_reg->as_register());
1516     }
1517   } else {
1518     ShouldNotReachHere();
1519   }
1520   if (to_reg->type() == T_OBJECT || to_reg->type() == T_ARRAY) {
1521     __ verify_oop(to_reg->as_register());
1522   }
1523 }
1524 
1525 
1526 void LIR_Assembler::reg2mem(LIR_Opr from_reg, LIR_Opr dest, BasicType type,
1527                             LIR_PatchCode patch_code, CodeEmitInfo* info, bool pop_fpu_stack,
1528                             bool wide, bool unaligned) {
1529   assert(type != T_METADATA, "store of metadata ptr not supported");
1530   LIR_Address* addr = dest->as_address_ptr();
1531 
1532   Register src = addr->base()->as_pointer_register();
1533   Register disp_reg = noreg;
1534   int disp_value = addr->disp();
1535   bool needs_patching = (patch_code != lir_patch_none);
1536 
1537   if (addr->base()->is_oop_register()) {
1538     __ verify_oop(src);
1539   }
1540 
1541   PatchingStub* patch = NULL;
1542   if (needs_patching) {
1543     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1544     assert(!from_reg->is_double_cpu() ||
1545            patch_code == lir_patch_none ||
1546            patch_code == lir_patch_normal, "patching doesn't match register");
1547   }
1548 
1549   if (addr->index()->is_illegal()) {
1550     if (!Assembler::is_simm13(disp_value) && (!unaligned || Assembler::is_simm13(disp_value + 4))) {
1551       if (needs_patching) {
1552         __ patchable_set(0, O7);
1553       } else {
1554         __ set(disp_value, O7);
1555       }
1556       disp_reg = O7;
1557     }
1558   } else if (unaligned || PatchALot) {
1559     __ add(src, addr->index()->as_register(), O7);
1560     src = O7;
1561   } else {
1562     disp_reg = addr->index()->as_pointer_register();
1563     assert(disp_value == 0, "can't handle 3 operand addresses");
1564   }
1565 
1566   // remember the offset of the store.  The patching_epilog must be done
1567   // before the call to add_debug_info_for_null_check, otherwise the PcDescs don't get
1568   // entered in increasing order.
1569   int offset;
1570 
1571   assert(disp_reg != noreg || Assembler::is_simm13(disp_value), "should have set this up");
1572   if (disp_reg == noreg) {
1573     offset = store(from_reg, src, disp_value, type, wide, unaligned);
1574   } else {
1575     assert(!unaligned, "can't handle this");
1576     offset = store(from_reg, src, disp_reg, type, wide);
1577   }
1578 
1579   if (patch != NULL) {
1580     patching_epilog(patch, patch_code, src, info);
1581   }
1582 
1583   if (info != NULL) add_debug_info_for_null_check(offset, info);
1584 }
1585 
1586 
1587 void LIR_Assembler::return_op(LIR_Opr result) {
1588   // the poll may need a register so just pick one that isn't the return register
1589 #if defined(TIERED) && !defined(_LP64)
1590   if (result->type_field() == LIR_OprDesc::long_type) {
1591     // Must move the result to G1
1592     // Must leave proper result in O0,O1 and G1 (TIERED only)
1593     __ sllx(I0, 32, G1);          // Shift bits into high G1
1594     __ srl (I1, 0, I1);           // Zero extend O1 (harmless?)
1595     __ or3 (I1, G1, G1);          // OR 64 bits into G1
1596 #ifdef ASSERT
1597     // mangle it so any problems will show up
1598     __ set(0xdeadbeef, I0);
1599     __ set(0xdeadbeef, I1);
1600 #endif
1601   }
1602 #endif // TIERED
1603   __ set((intptr_t)os::get_polling_page(), L0);
1604   __ relocate(relocInfo::poll_return_type);
1605   __ ld_ptr(L0, 0, G0);
1606   __ ret();
1607   __ delayed()->restore();
1608 }
1609 
1610 
1611 int LIR_Assembler::safepoint_poll(LIR_Opr tmp, CodeEmitInfo* info) {
1612   __ set((intptr_t)os::get_polling_page(), tmp->as_register());
1613   if (info != NULL) {
1614     add_debug_info_for_branch(info);
1615   } else {
1616     __ relocate(relocInfo::poll_type);
1617   }
1618 
1619   int offset = __ offset();
1620   __ ld_ptr(tmp->as_register(), 0, G0);
1621 
1622   return offset;
1623 }
1624 
1625 
1626 void LIR_Assembler::emit_static_call_stub() {
1627   address call_pc = __ pc();
1628   address stub = __ start_a_stub(call_stub_size);
1629   if (stub == NULL) {
1630     bailout("static call stub overflow");
1631     return;
1632   }
1633 
1634   int start = __ offset();
1635   __ relocate(static_stub_Relocation::spec(call_pc));
1636 
1637   __ set_metadata(NULL, G5);
1638   // must be set to -1 at code generation time
1639   AddressLiteral addrlit(-1);
1640   __ jump_to(addrlit, G3);
1641   __ delayed()->nop();
1642 
1643   assert(__ offset() - start <= call_stub_size, "stub too big");
1644   __ end_a_stub();
1645 }
1646 
1647 
1648 void LIR_Assembler::comp_op(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Op2* op) {
1649   if (opr1->is_single_fpu()) {
1650     __ fcmp(FloatRegisterImpl::S, Assembler::fcc0, opr1->as_float_reg(), opr2->as_float_reg());
1651   } else if (opr1->is_double_fpu()) {
1652     __ fcmp(FloatRegisterImpl::D, Assembler::fcc0, opr1->as_double_reg(), opr2->as_double_reg());
1653   } else if (opr1->is_single_cpu()) {
1654     if (opr2->is_constant()) {
1655       switch (opr2->as_constant_ptr()->type()) {
1656         case T_INT:
1657           { jint con = opr2->as_constant_ptr()->as_jint();
1658             if (Assembler::is_simm13(con)) {
1659               __ cmp(opr1->as_register(), con);
1660             } else {
1661               __ set(con, O7);
1662               __ cmp(opr1->as_register(), O7);
1663             }
1664           }
1665           break;
1666 
1667         case T_OBJECT:
1668           // there are only equal/notequal comparisions on objects
1669           { jobject con = opr2->as_constant_ptr()->as_jobject();
1670             if (con == NULL) {
1671               __ cmp(opr1->as_register(), 0);
1672             } else {
1673               jobject2reg(con, O7);
1674               __ cmp(opr1->as_register(), O7);
1675             }
1676           }
1677           break;
1678 
1679         default:
1680           ShouldNotReachHere();
1681           break;
1682       }
1683     } else {
1684       if (opr2->is_address()) {
1685         LIR_Address * addr = opr2->as_address_ptr();
1686         BasicType type = addr->type();
1687         if ( type == T_OBJECT ) __ ld_ptr(as_Address(addr), O7);
1688         else                    __ ld(as_Address(addr), O7);
1689         __ cmp(opr1->as_register(), O7);
1690       } else {
1691         __ cmp(opr1->as_register(), opr2->as_register());
1692       }
1693     }
1694   } else if (opr1->is_double_cpu()) {
1695     Register xlo = opr1->as_register_lo();
1696     Register xhi = opr1->as_register_hi();
1697     if (opr2->is_constant() && opr2->as_jlong() == 0) {
1698       assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "only handles these cases");
1699 #ifdef _LP64
1700       __ orcc(xhi, G0, G0);
1701 #else
1702       __ orcc(xhi, xlo, G0);
1703 #endif
1704     } else if (opr2->is_register()) {
1705       Register ylo = opr2->as_register_lo();
1706       Register yhi = opr2->as_register_hi();
1707 #ifdef _LP64
1708       __ cmp(xlo, ylo);
1709 #else
1710       __ subcc(xlo, ylo, xlo);
1711       __ subccc(xhi, yhi, xhi);
1712       if (condition == lir_cond_equal || condition == lir_cond_notEqual) {
1713         __ orcc(xhi, xlo, G0);
1714       }
1715 #endif
1716     } else {
1717       ShouldNotReachHere();
1718     }
1719   } else if (opr1->is_address()) {
1720     LIR_Address * addr = opr1->as_address_ptr();
1721     BasicType type = addr->type();
1722     assert (opr2->is_constant(), "Checking");
1723     if ( type == T_OBJECT ) __ ld_ptr(as_Address(addr), O7);
1724     else                    __ ld(as_Address(addr), O7);
1725     __ cmp(O7, opr2->as_constant_ptr()->as_jint());
1726   } else {
1727     ShouldNotReachHere();
1728   }
1729 }
1730 
1731 
1732 void LIR_Assembler::comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst, LIR_Op2* op){
1733   if (code == lir_cmp_fd2i || code == lir_ucmp_fd2i) {
1734     bool is_unordered_less = (code == lir_ucmp_fd2i);
1735     if (left->is_single_fpu()) {
1736       __ float_cmp(true, is_unordered_less ? -1 : 1, left->as_float_reg(), right->as_float_reg(), dst->as_register());
1737     } else if (left->is_double_fpu()) {
1738       __ float_cmp(false, is_unordered_less ? -1 : 1, left->as_double_reg(), right->as_double_reg(), dst->as_register());
1739     } else {
1740       ShouldNotReachHere();
1741     }
1742   } else if (code == lir_cmp_l2i) {
1743 #ifdef _LP64
1744     __ lcmp(left->as_register_lo(), right->as_register_lo(), dst->as_register());
1745 #else
1746     __ lcmp(left->as_register_hi(),  left->as_register_lo(),
1747             right->as_register_hi(), right->as_register_lo(),
1748             dst->as_register());
1749 #endif
1750   } else {
1751     ShouldNotReachHere();
1752   }
1753 }
1754 
1755 
1756 void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type) {
1757   Assembler::Condition acond;
1758   switch (condition) {
1759     case lir_cond_equal:        acond = Assembler::equal;        break;
1760     case lir_cond_notEqual:     acond = Assembler::notEqual;     break;
1761     case lir_cond_less:         acond = Assembler::less;         break;
1762     case lir_cond_lessEqual:    acond = Assembler::lessEqual;    break;
1763     case lir_cond_greaterEqual: acond = Assembler::greaterEqual; break;
1764     case lir_cond_greater:      acond = Assembler::greater;      break;
1765     case lir_cond_aboveEqual:   acond = Assembler::greaterEqualUnsigned;      break;
1766     case lir_cond_belowEqual:   acond = Assembler::lessEqualUnsigned;      break;
1767     default:                         ShouldNotReachHere();
1768   };
1769 
1770   if (opr1->is_constant() && opr1->type() == T_INT) {
1771     Register dest = result->as_register();
1772     // load up first part of constant before branch
1773     // and do the rest in the delay slot.
1774     if (!Assembler::is_simm13(opr1->as_jint())) {
1775       __ sethi(opr1->as_jint(), dest);
1776     }
1777   } else if (opr1->is_constant()) {
1778     const2reg(opr1, result, lir_patch_none, NULL);
1779   } else if (opr1->is_register()) {
1780     reg2reg(opr1, result);
1781   } else if (opr1->is_stack()) {
1782     stack2reg(opr1, result, result->type());
1783   } else {
1784     ShouldNotReachHere();
1785   }
1786   Label skip;
1787 #ifdef _LP64
1788     if  (type == T_INT) {
1789       __ br(acond, false, Assembler::pt, skip);
1790     } else
1791 #endif
1792       __ brx(acond, false, Assembler::pt, skip); // checks icc on 32bit and xcc on 64bit
1793   if (opr1->is_constant() && opr1->type() == T_INT) {
1794     Register dest = result->as_register();
1795     if (Assembler::is_simm13(opr1->as_jint())) {
1796       __ delayed()->or3(G0, opr1->as_jint(), dest);
1797     } else {
1798       // the sethi has been done above, so just put in the low 10 bits
1799       __ delayed()->or3(dest, opr1->as_jint() & 0x3ff, dest);
1800     }
1801   } else {
1802     // can't do anything useful in the delay slot
1803     __ delayed()->nop();
1804   }
1805   if (opr2->is_constant()) {
1806     const2reg(opr2, result, lir_patch_none, NULL);
1807   } else if (opr2->is_register()) {
1808     reg2reg(opr2, result);
1809   } else if (opr2->is_stack()) {
1810     stack2reg(opr2, result, result->type());
1811   } else {
1812     ShouldNotReachHere();
1813   }
1814   __ bind(skip);
1815 }
1816 
1817 
1818 void LIR_Assembler::arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest, CodeEmitInfo* info, bool pop_fpu_stack) {
1819   assert(info == NULL, "unused on this code path");
1820   assert(left->is_register(), "wrong items state");
1821   assert(dest->is_register(), "wrong items state");
1822 
1823   if (right->is_register()) {
1824     if (dest->is_float_kind()) {
1825 
1826       FloatRegister lreg, rreg, res;
1827       FloatRegisterImpl::Width w;
1828       if (right->is_single_fpu()) {
1829         w = FloatRegisterImpl::S;
1830         lreg = left->as_float_reg();
1831         rreg = right->as_float_reg();
1832         res  = dest->as_float_reg();
1833       } else {
1834         w = FloatRegisterImpl::D;
1835         lreg = left->as_double_reg();
1836         rreg = right->as_double_reg();
1837         res  = dest->as_double_reg();
1838       }
1839 
1840       switch (code) {
1841         case lir_add: __ fadd(w, lreg, rreg, res); break;
1842         case lir_sub: __ fsub(w, lreg, rreg, res); break;
1843         case lir_mul: // fall through
1844         case lir_mul_strictfp: __ fmul(w, lreg, rreg, res); break;
1845         case lir_div: // fall through
1846         case lir_div_strictfp: __ fdiv(w, lreg, rreg, res); break;
1847         default: ShouldNotReachHere();
1848       }
1849 
1850     } else if (dest->is_double_cpu()) {
1851 #ifdef _LP64
1852       Register dst_lo = dest->as_register_lo();
1853       Register op1_lo = left->as_pointer_register();
1854       Register op2_lo = right->as_pointer_register();
1855 
1856       switch (code) {
1857         case lir_add:
1858           __ add(op1_lo, op2_lo, dst_lo);
1859           break;
1860 
1861         case lir_sub:
1862           __ sub(op1_lo, op2_lo, dst_lo);
1863           break;
1864 
1865         default: ShouldNotReachHere();
1866       }
1867 #else
1868       Register op1_lo = left->as_register_lo();
1869       Register op1_hi = left->as_register_hi();
1870       Register op2_lo = right->as_register_lo();
1871       Register op2_hi = right->as_register_hi();
1872       Register dst_lo = dest->as_register_lo();
1873       Register dst_hi = dest->as_register_hi();
1874 
1875       switch (code) {
1876         case lir_add:
1877           __ addcc(op1_lo, op2_lo, dst_lo);
1878           __ addc (op1_hi, op2_hi, dst_hi);
1879           break;
1880 
1881         case lir_sub:
1882           __ subcc(op1_lo, op2_lo, dst_lo);
1883           __ subc (op1_hi, op2_hi, dst_hi);
1884           break;
1885 
1886         default: ShouldNotReachHere();
1887       }
1888 #endif
1889     } else {
1890       assert (right->is_single_cpu(), "Just Checking");
1891 
1892       Register lreg = left->as_register();
1893       Register res  = dest->as_register();
1894       Register rreg = right->as_register();
1895       switch (code) {
1896         case lir_add:  __ add  (lreg, rreg, res); break;
1897         case lir_sub:  __ sub  (lreg, rreg, res); break;
1898         case lir_mul:  __ mulx (lreg, rreg, res); break;
1899         default: ShouldNotReachHere();
1900       }
1901     }
1902   } else {
1903     assert (right->is_constant(), "must be constant");
1904 
1905     if (dest->is_single_cpu()) {
1906       Register lreg = left->as_register();
1907       Register res  = dest->as_register();
1908       int    simm13 = right->as_constant_ptr()->as_jint();
1909 
1910       switch (code) {
1911         case lir_add:  __ add  (lreg, simm13, res); break;
1912         case lir_sub:  __ sub  (lreg, simm13, res); break;
1913         case lir_mul:  __ mulx (lreg, simm13, res); break;
1914         default: ShouldNotReachHere();
1915       }
1916     } else {
1917       Register lreg = left->as_pointer_register();
1918       Register res  = dest->as_register_lo();
1919       long con = right->as_constant_ptr()->as_jlong();
1920       assert(Assembler::is_simm13(con), "must be simm13");
1921 
1922       switch (code) {
1923         case lir_add:  __ add  (lreg, (int)con, res); break;
1924         case lir_sub:  __ sub  (lreg, (int)con, res); break;
1925         case lir_mul:  __ mulx (lreg, (int)con, res); break;
1926         default: ShouldNotReachHere();
1927       }
1928     }
1929   }
1930 }
1931 
1932 
1933 void LIR_Assembler::fpop() {
1934   // do nothing
1935 }
1936 
1937 
1938 void LIR_Assembler::intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr thread, LIR_Opr dest, LIR_Op* op) {
1939   switch (code) {
1940     case lir_sin:
1941     case lir_tan:
1942     case lir_cos: {
1943       assert(thread->is_valid(), "preserve the thread object for performance reasons");
1944       assert(dest->as_double_reg() == F0, "the result will be in f0/f1");
1945       break;
1946     }
1947     case lir_sqrt: {
1948       assert(!thread->is_valid(), "there is no need for a thread_reg for dsqrt");
1949       FloatRegister src_reg = value->as_double_reg();
1950       FloatRegister dst_reg = dest->as_double_reg();
1951       __ fsqrt(FloatRegisterImpl::D, src_reg, dst_reg);
1952       break;
1953     }
1954     case lir_abs: {
1955       assert(!thread->is_valid(), "there is no need for a thread_reg for fabs");
1956       FloatRegister src_reg = value->as_double_reg();
1957       FloatRegister dst_reg = dest->as_double_reg();
1958       __ fabs(FloatRegisterImpl::D, src_reg, dst_reg);
1959       break;
1960     }
1961     default: {
1962       ShouldNotReachHere();
1963       break;
1964     }
1965   }
1966 }
1967 
1968 
1969 void LIR_Assembler::logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest) {
1970   if (right->is_constant()) {
1971     if (dest->is_single_cpu()) {
1972       int simm13 = right->as_constant_ptr()->as_jint();
1973       switch (code) {
1974         case lir_logic_and:   __ and3 (left->as_register(), simm13, dest->as_register()); break;
1975         case lir_logic_or:    __ or3  (left->as_register(), simm13, dest->as_register()); break;
1976         case lir_logic_xor:   __ xor3 (left->as_register(), simm13, dest->as_register()); break;
1977         default: ShouldNotReachHere();
1978       }
1979     } else {
1980       long c = right->as_constant_ptr()->as_jlong();
1981       assert(c == (int)c && Assembler::is_simm13(c), "out of range");
1982       int simm13 = (int)c;
1983       switch (code) {
1984         case lir_logic_and:
1985 #ifndef _LP64
1986           __ and3 (left->as_register_hi(), 0,      dest->as_register_hi());
1987 #endif
1988           __ and3 (left->as_register_lo(), simm13, dest->as_register_lo());
1989           break;
1990 
1991         case lir_logic_or:
1992 #ifndef _LP64
1993           __ or3 (left->as_register_hi(), 0,      dest->as_register_hi());
1994 #endif
1995           __ or3 (left->as_register_lo(), simm13, dest->as_register_lo());
1996           break;
1997 
1998         case lir_logic_xor:
1999 #ifndef _LP64
2000           __ xor3 (left->as_register_hi(), 0,      dest->as_register_hi());
2001 #endif
2002           __ xor3 (left->as_register_lo(), simm13, dest->as_register_lo());
2003           break;
2004 
2005         default: ShouldNotReachHere();
2006       }
2007     }
2008   } else {
2009     assert(right->is_register(), "right should be in register");
2010 
2011     if (dest->is_single_cpu()) {
2012       switch (code) {
2013         case lir_logic_and:   __ and3 (left->as_register(), right->as_register(), dest->as_register()); break;
2014         case lir_logic_or:    __ or3  (left->as_register(), right->as_register(), dest->as_register()); break;
2015         case lir_logic_xor:   __ xor3 (left->as_register(), right->as_register(), dest->as_register()); break;
2016         default: ShouldNotReachHere();
2017       }
2018     } else {
2019 #ifdef _LP64
2020       Register l = (left->is_single_cpu() && left->is_oop_register()) ? left->as_register() :
2021                                                                         left->as_register_lo();
2022       Register r = (right->is_single_cpu() && right->is_oop_register()) ? right->as_register() :
2023                                                                           right->as_register_lo();
2024 
2025       switch (code) {
2026         case lir_logic_and: __ and3 (l, r, dest->as_register_lo()); break;
2027         case lir_logic_or:  __ or3  (l, r, dest->as_register_lo()); break;
2028         case lir_logic_xor: __ xor3 (l, r, dest->as_register_lo()); break;
2029         default: ShouldNotReachHere();
2030       }
2031 #else
2032       switch (code) {
2033         case lir_logic_and:
2034           __ and3 (left->as_register_hi(), right->as_register_hi(), dest->as_register_hi());
2035           __ and3 (left->as_register_lo(), right->as_register_lo(), dest->as_register_lo());
2036           break;
2037 
2038         case lir_logic_or:
2039           __ or3 (left->as_register_hi(), right->as_register_hi(), dest->as_register_hi());
2040           __ or3 (left->as_register_lo(), right->as_register_lo(), dest->as_register_lo());
2041           break;
2042 
2043         case lir_logic_xor:
2044           __ xor3 (left->as_register_hi(), right->as_register_hi(), dest->as_register_hi());
2045           __ xor3 (left->as_register_lo(), right->as_register_lo(), dest->as_register_lo());
2046           break;
2047 
2048         default: ShouldNotReachHere();
2049       }
2050 #endif
2051     }
2052   }
2053 }
2054 
2055 
2056 int LIR_Assembler::shift_amount(BasicType t) {
2057   int elem_size = type2aelembytes(t);
2058   switch (elem_size) {
2059     case 1 : return 0;
2060     case 2 : return 1;
2061     case 4 : return 2;
2062     case 8 : return 3;
2063   }
2064   ShouldNotReachHere();
2065   return -1;
2066 }
2067 
2068 
2069 void LIR_Assembler::throw_op(LIR_Opr exceptionPC, LIR_Opr exceptionOop, CodeEmitInfo* info) {
2070   assert(exceptionOop->as_register() == Oexception, "should match");
2071   assert(exceptionPC->as_register() == Oissuing_pc, "should match");
2072 
2073   info->add_register_oop(exceptionOop);
2074 
2075   // reuse the debug info from the safepoint poll for the throw op itself
2076   address pc_for_athrow  = __ pc();
2077   int pc_for_athrow_offset = __ offset();
2078   RelocationHolder rspec = internal_word_Relocation::spec(pc_for_athrow);
2079   __ set(pc_for_athrow, Oissuing_pc, rspec);
2080   add_call_info(pc_for_athrow_offset, info); // for exception handler
2081 
2082   __ call(Runtime1::entry_for(Runtime1::handle_exception_id), relocInfo::runtime_call_type);
2083   __ delayed()->nop();
2084 }
2085 
2086 
2087 void LIR_Assembler::unwind_op(LIR_Opr exceptionOop) {
2088   assert(exceptionOop->as_register() == Oexception, "should match");
2089 
2090   __ br(Assembler::always, false, Assembler::pt, _unwind_handler_entry);
2091   __ delayed()->nop();
2092 }
2093 
2094 void LIR_Assembler::emit_arraycopy(LIR_OpArrayCopy* op) {
2095   Register src = op->src()->as_register();
2096   Register dst = op->dst()->as_register();
2097   Register src_pos = op->src_pos()->as_register();
2098   Register dst_pos = op->dst_pos()->as_register();
2099   Register length  = op->length()->as_register();
2100   Register tmp = op->tmp()->as_register();
2101   Register tmp2 = O7;
2102 
2103   int flags = op->flags();
2104   ciArrayKlass* default_type = op->expected_type();
2105   BasicType basic_type = default_type != NULL ? default_type->element_type()->basic_type() : T_ILLEGAL;
2106   if (basic_type == T_ARRAY) basic_type = T_OBJECT;
2107 
2108 #ifdef _LP64
2109   // higher 32bits must be null
2110   __ sra(dst_pos, 0, dst_pos);
2111   __ sra(src_pos, 0, src_pos);
2112   __ sra(length, 0, length);
2113 #endif
2114 
2115   // set up the arraycopy stub information
2116   ArrayCopyStub* stub = op->stub();
2117 
2118   // always do stub if no type information is available.  it's ok if
2119   // the known type isn't loaded since the code sanity checks
2120   // in debug mode and the type isn't required when we know the exact type
2121   // also check that the type is an array type.
2122   if (op->expected_type() == NULL) {
2123     __ mov(src,     O0);
2124     __ mov(src_pos, O1);
2125     __ mov(dst,     O2);
2126     __ mov(dst_pos, O3);
2127     __ mov(length,  O4);
2128     address copyfunc_addr = StubRoutines::generic_arraycopy();
2129 
2130     if (copyfunc_addr == NULL) { // Use C version if stub was not generated
2131       __ call_VM_leaf(tmp, CAST_FROM_FN_PTR(address, Runtime1::arraycopy));
2132     } else {
2133 #ifndef PRODUCT
2134       if (PrintC1Statistics) {
2135         address counter = (address)&Runtime1::_generic_arraycopystub_cnt;
2136         __ inc_counter(counter, G1, G3);
2137       }
2138 #endif
2139       __ call_VM_leaf(tmp, copyfunc_addr);
2140     }
2141 
2142     if (copyfunc_addr != NULL) {
2143       __ xor3(O0, -1, tmp);
2144       __ sub(length, tmp, length);
2145       __ add(src_pos, tmp, src_pos);
2146       __ cmp_zero_and_br(Assembler::less, O0, *stub->entry());
2147       __ delayed()->add(dst_pos, tmp, dst_pos);
2148     } else {
2149       __ cmp_zero_and_br(Assembler::less, O0, *stub->entry());
2150       __ delayed()->nop();
2151     }
2152     __ bind(*stub->continuation());
2153     return;
2154   }
2155 
2156   assert(default_type != NULL && default_type->is_array_klass(), "must be true at this point");
2157 
2158   // make sure src and dst are non-null and load array length
2159   if (flags & LIR_OpArrayCopy::src_null_check) {
2160     __ tst(src);
2161     __ brx(Assembler::equal, false, Assembler::pn, *stub->entry());
2162     __ delayed()->nop();
2163   }
2164 
2165   if (flags & LIR_OpArrayCopy::dst_null_check) {
2166     __ tst(dst);
2167     __ brx(Assembler::equal, false, Assembler::pn, *stub->entry());
2168     __ delayed()->nop();
2169   }
2170 
2171   if (flags & LIR_OpArrayCopy::src_pos_positive_check) {
2172     // test src_pos register
2173     __ cmp_zero_and_br(Assembler::less, src_pos, *stub->entry());
2174     __ delayed()->nop();
2175   }
2176 
2177   if (flags & LIR_OpArrayCopy::dst_pos_positive_check) {
2178     // test dst_pos register
2179     __ cmp_zero_and_br(Assembler::less, dst_pos, *stub->entry());
2180     __ delayed()->nop();
2181   }
2182 
2183   if (flags & LIR_OpArrayCopy::length_positive_check) {
2184     // make sure length isn't negative
2185     __ cmp_zero_and_br(Assembler::less, length, *stub->entry());
2186     __ delayed()->nop();
2187   }
2188 
2189   if (flags & LIR_OpArrayCopy::src_range_check) {
2190     __ ld(src, arrayOopDesc::length_offset_in_bytes(), tmp2);
2191     __ add(length, src_pos, tmp);
2192     __ cmp(tmp2, tmp);
2193     __ br(Assembler::carrySet, false, Assembler::pn, *stub->entry());
2194     __ delayed()->nop();
2195   }
2196 
2197   if (flags & LIR_OpArrayCopy::dst_range_check) {
2198     __ ld(dst, arrayOopDesc::length_offset_in_bytes(), tmp2);
2199     __ add(length, dst_pos, tmp);
2200     __ cmp(tmp2, tmp);
2201     __ br(Assembler::carrySet, false, Assembler::pn, *stub->entry());
2202     __ delayed()->nop();
2203   }
2204 
2205   int shift = shift_amount(basic_type);
2206 
2207   if (flags & LIR_OpArrayCopy::type_check) {
2208     // We don't know the array types are compatible
2209     if (basic_type != T_OBJECT) {
2210       // Simple test for basic type arrays
2211       if (UseCompressedKlassPointers) {
2212         // We don't need decode because we just need to compare
2213         __ lduw(src, oopDesc::klass_offset_in_bytes(), tmp);
2214         __ lduw(dst, oopDesc::klass_offset_in_bytes(), tmp2);
2215         __ cmp(tmp, tmp2);
2216         __ br(Assembler::notEqual, false, Assembler::pt, *stub->entry());
2217       } else {
2218         __ ld_ptr(src, oopDesc::klass_offset_in_bytes(), tmp);
2219         __ ld_ptr(dst, oopDesc::klass_offset_in_bytes(), tmp2);
2220         __ cmp(tmp, tmp2);
2221         __ brx(Assembler::notEqual, false, Assembler::pt, *stub->entry());
2222       }
2223       __ delayed()->nop();
2224     } else {
2225       // For object arrays, if src is a sub class of dst then we can
2226       // safely do the copy.
2227       address copyfunc_addr = StubRoutines::checkcast_arraycopy();
2228 
2229       Label cont, slow;
2230       assert_different_registers(tmp, tmp2, G3, G1);
2231 
2232       __ load_klass(src, G3);
2233       __ load_klass(dst, G1);
2234 
2235       __ check_klass_subtype_fast_path(G3, G1, tmp, tmp2, &cont, copyfunc_addr == NULL ? stub->entry() : &slow, NULL);
2236 
2237       __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type);
2238       __ delayed()->nop();
2239 
2240       __ cmp(G3, 0);
2241       if (copyfunc_addr != NULL) { // use stub if available
2242         // src is not a sub class of dst so we have to do a
2243         // per-element check.
2244         __ br(Assembler::notEqual, false, Assembler::pt, cont);
2245         __ delayed()->nop();
2246 
2247         __ bind(slow);
2248 
2249         int mask = LIR_OpArrayCopy::src_objarray|LIR_OpArrayCopy::dst_objarray;
2250         if ((flags & mask) != mask) {
2251           // Check that at least both of them object arrays.
2252           assert(flags & mask, "one of the two should be known to be an object array");
2253 
2254           if (!(flags & LIR_OpArrayCopy::src_objarray)) {
2255             __ load_klass(src, tmp);
2256           } else if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
2257             __ load_klass(dst, tmp);
2258           }
2259           int lh_offset = in_bytes(Klass::layout_helper_offset());
2260 
2261           __ lduw(tmp, lh_offset, tmp2);
2262 
2263           jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2264           __ set(objArray_lh, tmp);
2265           __ cmp(tmp, tmp2);
2266           __ br(Assembler::notEqual, false, Assembler::pt,  *stub->entry());
2267           __ delayed()->nop();
2268         }
2269 
2270         Register src_ptr = O0;
2271         Register dst_ptr = O1;
2272         Register len     = O2;
2273         Register chk_off = O3;
2274         Register super_k = O4;
2275 
2276         __ add(src, arrayOopDesc::base_offset_in_bytes(basic_type), src_ptr);
2277         if (shift == 0) {
2278           __ add(src_ptr, src_pos, src_ptr);
2279         } else {
2280           __ sll(src_pos, shift, tmp);
2281           __ add(src_ptr, tmp, src_ptr);
2282         }
2283 
2284         __ add(dst, arrayOopDesc::base_offset_in_bytes(basic_type), dst_ptr);
2285         if (shift == 0) {
2286           __ add(dst_ptr, dst_pos, dst_ptr);
2287         } else {
2288           __ sll(dst_pos, shift, tmp);
2289           __ add(dst_ptr, tmp, dst_ptr);
2290         }
2291         __ mov(length, len);
2292         __ load_klass(dst, tmp);
2293 
2294         int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
2295         __ ld_ptr(tmp, ek_offset, super_k);
2296 
2297         int sco_offset = in_bytes(Klass::super_check_offset_offset());
2298         __ lduw(super_k, sco_offset, chk_off);
2299 
2300         __ call_VM_leaf(tmp, copyfunc_addr);
2301 
2302 #ifndef PRODUCT
2303         if (PrintC1Statistics) {
2304           Label failed;
2305           __ br_notnull_short(O0, Assembler::pn, failed);
2306           __ inc_counter((address)&Runtime1::_arraycopy_checkcast_cnt, G1, G3);
2307           __ bind(failed);
2308         }
2309 #endif
2310 
2311         __ br_null(O0, false, Assembler::pt,  *stub->continuation());
2312         __ delayed()->xor3(O0, -1, tmp);
2313 
2314 #ifndef PRODUCT
2315         if (PrintC1Statistics) {
2316           __ inc_counter((address)&Runtime1::_arraycopy_checkcast_attempt_cnt, G1, G3);
2317         }
2318 #endif
2319 
2320         __ sub(length, tmp, length);
2321         __ add(src_pos, tmp, src_pos);
2322         __ br(Assembler::always, false, Assembler::pt, *stub->entry());
2323         __ delayed()->add(dst_pos, tmp, dst_pos);
2324 
2325         __ bind(cont);
2326       } else {
2327         __ br(Assembler::equal, false, Assembler::pn, *stub->entry());
2328         __ delayed()->nop();
2329         __ bind(cont);
2330       }
2331     }
2332   }
2333 
2334 #ifdef ASSERT
2335   if (basic_type != T_OBJECT || !(flags & LIR_OpArrayCopy::type_check)) {
2336     // Sanity check the known type with the incoming class.  For the
2337     // primitive case the types must match exactly with src.klass and
2338     // dst.klass each exactly matching the default type.  For the
2339     // object array case, if no type check is needed then either the
2340     // dst type is exactly the expected type and the src type is a
2341     // subtype which we can't check or src is the same array as dst
2342     // but not necessarily exactly of type default_type.
2343     Label known_ok, halt;
2344     metadata2reg(op->expected_type()->constant_encoding(), tmp);
2345     if (UseCompressedKlassPointers) {
2346       // tmp holds the default type. It currently comes uncompressed after the
2347       // load of a constant, so encode it.
2348       __ encode_klass_not_null(tmp);
2349       // load the raw value of the dst klass, since we will be comparing
2350       // uncompressed values directly.
2351       __ lduw(dst, oopDesc::klass_offset_in_bytes(), tmp2);
2352       if (basic_type != T_OBJECT) {
2353         __ cmp(tmp, tmp2);
2354         __ br(Assembler::notEqual, false, Assembler::pn, halt);
2355         // load the raw value of the src klass.
2356         __ delayed()->lduw(src, oopDesc::klass_offset_in_bytes(), tmp2);
2357         __ cmp_and_br_short(tmp, tmp2, Assembler::equal, Assembler::pn, known_ok);
2358       } else {
2359         __ cmp(tmp, tmp2);
2360         __ br(Assembler::equal, false, Assembler::pn, known_ok);
2361         __ delayed()->cmp(src, dst);
2362         __ brx(Assembler::equal, false, Assembler::pn, known_ok);
2363         __ delayed()->nop();
2364       }
2365     } else {
2366       __ ld_ptr(dst, oopDesc::klass_offset_in_bytes(), tmp2);
2367       if (basic_type != T_OBJECT) {
2368         __ cmp(tmp, tmp2);
2369         __ brx(Assembler::notEqual, false, Assembler::pn, halt);
2370         __ delayed()->ld_ptr(src, oopDesc::klass_offset_in_bytes(), tmp2);
2371         __ cmp_and_brx_short(tmp, tmp2, Assembler::equal, Assembler::pn, known_ok);
2372       } else {
2373         __ cmp(tmp, tmp2);
2374         __ brx(Assembler::equal, false, Assembler::pn, known_ok);
2375         __ delayed()->cmp(src, dst);
2376         __ brx(Assembler::equal, false, Assembler::pn, known_ok);
2377         __ delayed()->nop();
2378       }
2379     }
2380     __ bind(halt);
2381     __ stop("incorrect type information in arraycopy");
2382     __ bind(known_ok);
2383   }
2384 #endif
2385 
2386 #ifndef PRODUCT
2387   if (PrintC1Statistics) {
2388     address counter = Runtime1::arraycopy_count_address(basic_type);
2389     __ inc_counter(counter, G1, G3);
2390   }
2391 #endif
2392 
2393   Register src_ptr = O0;
2394   Register dst_ptr = O1;
2395   Register len     = O2;
2396 
2397   __ add(src, arrayOopDesc::base_offset_in_bytes(basic_type), src_ptr);
2398   if (shift == 0) {
2399     __ add(src_ptr, src_pos, src_ptr);
2400   } else {
2401     __ sll(src_pos, shift, tmp);
2402     __ add(src_ptr, tmp, src_ptr);
2403   }
2404 
2405   __ add(dst, arrayOopDesc::base_offset_in_bytes(basic_type), dst_ptr);
2406   if (shift == 0) {
2407     __ add(dst_ptr, dst_pos, dst_ptr);
2408   } else {
2409     __ sll(dst_pos, shift, tmp);
2410     __ add(dst_ptr, tmp, dst_ptr);
2411   }
2412 
2413   bool disjoint = (flags & LIR_OpArrayCopy::overlapping) == 0;
2414   bool aligned = (flags & LIR_OpArrayCopy::unaligned) == 0;
2415   const char *name;
2416   address entry = StubRoutines::select_arraycopy_function(basic_type, aligned, disjoint, name, false);
2417 
2418   // arraycopy stubs takes a length in number of elements, so don't scale it.
2419   __ mov(length, len);
2420   __ call_VM_leaf(tmp, entry);
2421 
2422   __ bind(*stub->continuation());
2423 }
2424 
2425 
2426 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
2427   if (dest->is_single_cpu()) {
2428 #ifdef _LP64
2429     if (left->type() == T_OBJECT) {
2430       switch (code) {
2431         case lir_shl:  __ sllx  (left->as_register(), count->as_register(), dest->as_register()); break;
2432         case lir_shr:  __ srax  (left->as_register(), count->as_register(), dest->as_register()); break;
2433         case lir_ushr: __ srl   (left->as_register(), count->as_register(), dest->as_register()); break;
2434         default: ShouldNotReachHere();
2435       }
2436     } else
2437 #endif
2438       switch (code) {
2439         case lir_shl:  __ sll   (left->as_register(), count->as_register(), dest->as_register()); break;
2440         case lir_shr:  __ sra   (left->as_register(), count->as_register(), dest->as_register()); break;
2441         case lir_ushr: __ srl   (left->as_register(), count->as_register(), dest->as_register()); break;
2442         default: ShouldNotReachHere();
2443       }
2444   } else {
2445 #ifdef _LP64
2446     switch (code) {
2447       case lir_shl:  __ sllx  (left->as_register_lo(), count->as_register(), dest->as_register_lo()); break;
2448       case lir_shr:  __ srax  (left->as_register_lo(), count->as_register(), dest->as_register_lo()); break;
2449       case lir_ushr: __ srlx  (left->as_register_lo(), count->as_register(), dest->as_register_lo()); break;
2450       default: ShouldNotReachHere();
2451     }
2452 #else
2453     switch (code) {
2454       case lir_shl:  __ lshl  (left->as_register_hi(), left->as_register_lo(), count->as_register(), dest->as_register_hi(), dest->as_register_lo(), G3_scratch); break;
2455       case lir_shr:  __ lshr  (left->as_register_hi(), left->as_register_lo(), count->as_register(), dest->as_register_hi(), dest->as_register_lo(), G3_scratch); break;
2456       case lir_ushr: __ lushr (left->as_register_hi(), left->as_register_lo(), count->as_register(), dest->as_register_hi(), dest->as_register_lo(), G3_scratch); break;
2457       default: ShouldNotReachHere();
2458     }
2459 #endif
2460   }
2461 }
2462 
2463 
2464 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, jint count, LIR_Opr dest) {
2465 #ifdef _LP64
2466   if (left->type() == T_OBJECT) {
2467     count = count & 63;  // shouldn't shift by more than sizeof(intptr_t)
2468     Register l = left->as_register();
2469     Register d = dest->as_register_lo();
2470     switch (code) {
2471       case lir_shl:  __ sllx  (l, count, d); break;
2472       case lir_shr:  __ srax  (l, count, d); break;
2473       case lir_ushr: __ srlx  (l, count, d); break;
2474       default: ShouldNotReachHere();
2475     }
2476     return;
2477   }
2478 #endif
2479 
2480   if (dest->is_single_cpu()) {
2481     count = count & 0x1F; // Java spec
2482     switch (code) {
2483       case lir_shl:  __ sll   (left->as_register(), count, dest->as_register()); break;
2484       case lir_shr:  __ sra   (left->as_register(), count, dest->as_register()); break;
2485       case lir_ushr: __ srl   (left->as_register(), count, dest->as_register()); break;
2486       default: ShouldNotReachHere();
2487     }
2488   } else if (dest->is_double_cpu()) {
2489     count = count & 63; // Java spec
2490     switch (code) {
2491       case lir_shl:  __ sllx  (left->as_pointer_register(), count, dest->as_pointer_register()); break;
2492       case lir_shr:  __ srax  (left->as_pointer_register(), count, dest->as_pointer_register()); break;
2493       case lir_ushr: __ srlx  (left->as_pointer_register(), count, dest->as_pointer_register()); break;
2494       default: ShouldNotReachHere();
2495     }
2496   } else {
2497     ShouldNotReachHere();
2498   }
2499 }
2500 
2501 
2502 void LIR_Assembler::emit_alloc_obj(LIR_OpAllocObj* op) {
2503   assert(op->tmp1()->as_register()  == G1 &&
2504          op->tmp2()->as_register()  == G3 &&
2505          op->tmp3()->as_register()  == G4 &&
2506          op->obj()->as_register()   == O0 &&
2507          op->klass()->as_register() == G5, "must be");
2508   if (op->init_check()) {
2509     __ ldub(op->klass()->as_register(),
2510           in_bytes(InstanceKlass::init_state_offset()),
2511           op->tmp1()->as_register());
2512     add_debug_info_for_null_check_here(op->stub()->info());
2513     __ cmp(op->tmp1()->as_register(), InstanceKlass::fully_initialized);
2514     __ br(Assembler::notEqual, false, Assembler::pn, *op->stub()->entry());
2515     __ delayed()->nop();
2516   }
2517   __ allocate_object(op->obj()->as_register(),
2518                      op->tmp1()->as_register(),
2519                      op->tmp2()->as_register(),
2520                      op->tmp3()->as_register(),
2521                      op->header_size(),
2522                      op->object_size(),
2523                      op->klass()->as_register(),
2524                      *op->stub()->entry());
2525   __ bind(*op->stub()->continuation());
2526   __ verify_oop(op->obj()->as_register());
2527 }
2528 
2529 
2530 void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
2531   assert(op->tmp1()->as_register()  == G1 &&
2532          op->tmp2()->as_register()  == G3 &&
2533          op->tmp3()->as_register()  == G4 &&
2534          op->tmp4()->as_register()  == O1 &&
2535          op->klass()->as_register() == G5, "must be");
2536 
2537   LP64_ONLY( __ signx(op->len()->as_register()); )
2538   if (UseSlowPath ||
2539       (!UseFastNewObjectArray && (op->type() == T_OBJECT || op->type() == T_ARRAY)) ||
2540       (!UseFastNewTypeArray   && (op->type() != T_OBJECT && op->type() != T_ARRAY))) {
2541     __ br(Assembler::always, false, Assembler::pt, *op->stub()->entry());
2542     __ delayed()->nop();
2543   } else {
2544     __ allocate_array(op->obj()->as_register(),
2545                       op->len()->as_register(),
2546                       op->tmp1()->as_register(),
2547                       op->tmp2()->as_register(),
2548                       op->tmp3()->as_register(),
2549                       arrayOopDesc::header_size(op->type()),
2550                       type2aelembytes(op->type()),
2551                       op->klass()->as_register(),
2552                       *op->stub()->entry());
2553   }
2554   __ bind(*op->stub()->continuation());
2555 }
2556 
2557 
2558 void LIR_Assembler::type_profile_helper(Register mdo, int mdo_offset_bias,
2559                                         ciMethodData *md, ciProfileData *data,
2560                                         Register recv, Register tmp1, Label* update_done) {
2561   uint i;
2562   for (i = 0; i < VirtualCallData::row_limit(); i++) {
2563     Label next_test;
2564     // See if the receiver is receiver[n].
2565     Address receiver_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)) -
2566                           mdo_offset_bias);
2567     __ ld_ptr(receiver_addr, tmp1);
2568     __ verify_oop(tmp1);
2569     __ cmp_and_brx_short(recv, tmp1, Assembler::notEqual, Assembler::pt, next_test);
2570     Address data_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)) -
2571                       mdo_offset_bias);
2572     __ ld_ptr(data_addr, tmp1);
2573     __ add(tmp1, DataLayout::counter_increment, tmp1);
2574     __ st_ptr(tmp1, data_addr);
2575     __ ba(*update_done);
2576     __ delayed()->nop();
2577     __ bind(next_test);
2578   }
2579 
2580   // Didn't find receiver; find next empty slot and fill it in
2581   for (i = 0; i < VirtualCallData::row_limit(); i++) {
2582     Label next_test;
2583     Address recv_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)) -
2584                       mdo_offset_bias);
2585     __ ld_ptr(recv_addr, tmp1);
2586     __ br_notnull_short(tmp1, Assembler::pt, next_test);
2587     __ st_ptr(recv, recv_addr);
2588     __ set(DataLayout::counter_increment, tmp1);
2589     __ st_ptr(tmp1, mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)) -
2590               mdo_offset_bias);
2591     __ ba(*update_done);
2592     __ delayed()->nop();
2593     __ bind(next_test);
2594   }
2595 }
2596 
2597 
2598 void LIR_Assembler::setup_md_access(ciMethod* method, int bci,
2599                                     ciMethodData*& md, ciProfileData*& data, int& mdo_offset_bias) {
2600   md = method->method_data_or_null();
2601   assert(md != NULL, "Sanity");
2602   data = md->bci_to_data(bci);
2603   assert(data != NULL,       "need data for checkcast");
2604   assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
2605   if (!Assembler::is_simm13(md->byte_offset_of_slot(data, DataLayout::header_offset()) + data->size_in_bytes())) {
2606     // The offset is large so bias the mdo by the base of the slot so
2607     // that the ld can use simm13s to reference the slots of the data
2608     mdo_offset_bias = md->byte_offset_of_slot(data, DataLayout::header_offset());
2609   }
2610 }
2611 
2612 void LIR_Assembler::emit_typecheck_helper(LIR_OpTypeCheck *op, Label* success, Label* failure, Label* obj_is_null) {
2613   // we always need a stub for the failure case.
2614   CodeStub* stub = op->stub();
2615   Register obj = op->object()->as_register();
2616   Register k_RInfo = op->tmp1()->as_register();
2617   Register klass_RInfo = op->tmp2()->as_register();
2618   Register dst = op->result_opr()->as_register();
2619   Register Rtmp1 = op->tmp3()->as_register();
2620   ciKlass* k = op->klass();
2621 
2622 
2623   if (obj == k_RInfo) {
2624     k_RInfo = klass_RInfo;
2625     klass_RInfo = obj;
2626   }
2627 
2628   ciMethodData* md;
2629   ciProfileData* data;
2630   int mdo_offset_bias = 0;
2631   if (op->should_profile()) {
2632     ciMethod* method = op->profiled_method();
2633     assert(method != NULL, "Should have method");
2634     setup_md_access(method, op->profiled_bci(), md, data, mdo_offset_bias);
2635 
2636     Label not_null;
2637     __ br_notnull_short(obj, Assembler::pn, not_null);
2638     Register mdo      = k_RInfo;
2639     Register data_val = Rtmp1;
2640     metadata2reg(md->constant_encoding(), mdo);
2641     if (mdo_offset_bias > 0) {
2642       __ set(mdo_offset_bias, data_val);
2643       __ add(mdo, data_val, mdo);
2644     }
2645     Address flags_addr(mdo, md->byte_offset_of_slot(data, DataLayout::flags_offset()) - mdo_offset_bias);
2646     __ ldub(flags_addr, data_val);
2647     __ or3(data_val, BitData::null_seen_byte_constant(), data_val);
2648     __ stb(data_val, flags_addr);
2649     __ ba(*obj_is_null);
2650     __ delayed()->nop();
2651     __ bind(not_null);
2652   } else {
2653     __ br_null(obj, false, Assembler::pn, *obj_is_null);
2654     __ delayed()->nop();
2655   }
2656 
2657   Label profile_cast_failure, profile_cast_success;
2658   Label *failure_target = op->should_profile() ? &profile_cast_failure : failure;
2659   Label *success_target = op->should_profile() ? &profile_cast_success : success;
2660 
2661   // patching may screw with our temporaries on sparc,
2662   // so let's do it before loading the class
2663   if (k->is_loaded()) {
2664     metadata2reg(k->constant_encoding(), k_RInfo);
2665   } else {
2666     klass2reg_with_patching(k_RInfo, op->info_for_patch());
2667   }
2668   assert(obj != k_RInfo, "must be different");
2669 
2670   // get object class
2671   // not a safepoint as obj null check happens earlier
2672   __ load_klass(obj, klass_RInfo);
2673   if (op->fast_check()) {
2674     assert_different_registers(klass_RInfo, k_RInfo);
2675     __ cmp(k_RInfo, klass_RInfo);
2676     __ brx(Assembler::notEqual, false, Assembler::pt, *failure_target);
2677     __ delayed()->nop();
2678   } else {
2679     bool need_slow_path = true;
2680     if (k->is_loaded()) {
2681       if ((int) k->super_check_offset() != in_bytes(Klass::secondary_super_cache_offset()))
2682         need_slow_path = false;
2683       // perform the fast part of the checking logic
2684       __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, noreg,
2685                                        (need_slow_path ? success_target : NULL),
2686                                        failure_target, NULL,
2687                                        RegisterOrConstant(k->super_check_offset()));
2688     } else {
2689       // perform the fast part of the checking logic
2690       __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, O7, success_target,
2691                                        failure_target, NULL);
2692     }
2693     if (need_slow_path) {
2694       // call out-of-line instance of __ check_klass_subtype_slow_path(...):
2695       assert(klass_RInfo == G3 && k_RInfo == G1, "incorrect call setup");
2696       __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type);
2697       __ delayed()->nop();
2698       __ cmp(G3, 0);
2699       __ br(Assembler::equal, false, Assembler::pn, *failure_target);
2700       __ delayed()->nop();
2701       // Fall through to success case
2702     }
2703   }
2704 
2705   if (op->should_profile()) {
2706     Register mdo  = klass_RInfo, recv = k_RInfo, tmp1 = Rtmp1;
2707     assert_different_registers(obj, mdo, recv, tmp1);
2708     __ bind(profile_cast_success);
2709     metadata2reg(md->constant_encoding(), mdo);
2710     if (mdo_offset_bias > 0) {
2711       __ set(mdo_offset_bias, tmp1);
2712       __ add(mdo, tmp1, mdo);
2713     }
2714     __ load_klass(obj, recv);
2715     type_profile_helper(mdo, mdo_offset_bias, md, data, recv, tmp1, success);
2716     // Jump over the failure case
2717     __ ba(*success);
2718     __ delayed()->nop();
2719     // Cast failure case
2720     __ bind(profile_cast_failure);
2721     metadata2reg(md->constant_encoding(), mdo);
2722     if (mdo_offset_bias > 0) {
2723       __ set(mdo_offset_bias, tmp1);
2724       __ add(mdo, tmp1, mdo);
2725     }
2726     Address data_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()) - mdo_offset_bias);
2727     __ ld_ptr(data_addr, tmp1);
2728     __ sub(tmp1, DataLayout::counter_increment, tmp1);
2729     __ st_ptr(tmp1, data_addr);
2730     __ ba(*failure);
2731     __ delayed()->nop();
2732   }
2733   __ ba(*success);
2734   __ delayed()->nop();
2735 }
2736 
2737 void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
2738   LIR_Code code = op->code();
2739   if (code == lir_store_check) {
2740     Register value = op->object()->as_register();
2741     Register array = op->array()->as_register();
2742     Register k_RInfo = op->tmp1()->as_register();
2743     Register klass_RInfo = op->tmp2()->as_register();
2744     Register Rtmp1 = op->tmp3()->as_register();
2745 
2746     __ verify_oop(value);
2747     CodeStub* stub = op->stub();
2748     // check if it needs to be profiled
2749     ciMethodData* md;
2750     ciProfileData* data;
2751     int mdo_offset_bias = 0;
2752     if (op->should_profile()) {
2753       ciMethod* method = op->profiled_method();
2754       assert(method != NULL, "Should have method");
2755       setup_md_access(method, op->profiled_bci(), md, data, mdo_offset_bias);
2756     }
2757     Label profile_cast_success, profile_cast_failure, done;
2758     Label *success_target = op->should_profile() ? &profile_cast_success : &done;
2759     Label *failure_target = op->should_profile() ? &profile_cast_failure : stub->entry();
2760 
2761     if (op->should_profile()) {
2762       Label not_null;
2763       __ br_notnull_short(value, Assembler::pn, not_null);
2764       Register mdo      = k_RInfo;
2765       Register data_val = Rtmp1;
2766       metadata2reg(md->constant_encoding(), mdo);
2767       if (mdo_offset_bias > 0) {
2768         __ set(mdo_offset_bias, data_val);
2769         __ add(mdo, data_val, mdo);
2770       }
2771       Address flags_addr(mdo, md->byte_offset_of_slot(data, DataLayout::flags_offset()) - mdo_offset_bias);
2772       __ ldub(flags_addr, data_val);
2773       __ or3(data_val, BitData::null_seen_byte_constant(), data_val);
2774       __ stb(data_val, flags_addr);
2775       __ ba_short(done);
2776       __ bind(not_null);
2777     } else {
2778       __ br_null_short(value, Assembler::pn, done);
2779     }
2780     add_debug_info_for_null_check_here(op->info_for_exception());
2781     __ load_klass(array, k_RInfo);
2782     __ load_klass(value, klass_RInfo);
2783 
2784     // get instance klass
2785     __ ld_ptr(Address(k_RInfo, ObjArrayKlass::element_klass_offset()), k_RInfo);
2786     // perform the fast part of the checking logic
2787     __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, O7, success_target, failure_target, NULL);
2788 
2789     // call out-of-line instance of __ check_klass_subtype_slow_path(...):
2790     assert(klass_RInfo == G3 && k_RInfo == G1, "incorrect call setup");
2791     __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type);
2792     __ delayed()->nop();
2793     __ cmp(G3, 0);
2794     __ br(Assembler::equal, false, Assembler::pn, *failure_target);
2795     __ delayed()->nop();
2796     // fall through to the success case
2797 
2798     if (op->should_profile()) {
2799       Register mdo  = klass_RInfo, recv = k_RInfo, tmp1 = Rtmp1;
2800       assert_different_registers(value, mdo, recv, tmp1);
2801       __ bind(profile_cast_success);
2802       metadata2reg(md->constant_encoding(), mdo);
2803       if (mdo_offset_bias > 0) {
2804         __ set(mdo_offset_bias, tmp1);
2805         __ add(mdo, tmp1, mdo);
2806       }
2807       __ load_klass(value, recv);
2808       type_profile_helper(mdo, mdo_offset_bias, md, data, recv, tmp1, &done);
2809       __ ba_short(done);
2810       // Cast failure case
2811       __ bind(profile_cast_failure);
2812       metadata2reg(md->constant_encoding(), mdo);
2813       if (mdo_offset_bias > 0) {
2814         __ set(mdo_offset_bias, tmp1);
2815         __ add(mdo, tmp1, mdo);
2816       }
2817       Address data_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()) - mdo_offset_bias);
2818       __ ld_ptr(data_addr, tmp1);
2819       __ sub(tmp1, DataLayout::counter_increment, tmp1);
2820       __ st_ptr(tmp1, data_addr);
2821       __ ba(*stub->entry());
2822       __ delayed()->nop();
2823     }
2824     __ bind(done);
2825   } else if (code == lir_checkcast) {
2826     Register obj = op->object()->as_register();
2827     Register dst = op->result_opr()->as_register();
2828     Label success;
2829     emit_typecheck_helper(op, &success, op->stub()->entry(), &success);
2830     __ bind(success);
2831     __ mov(obj, dst);
2832   } else if (code == lir_instanceof) {
2833     Register obj = op->object()->as_register();
2834     Register dst = op->result_opr()->as_register();
2835     Label success, failure, done;
2836     emit_typecheck_helper(op, &success, &failure, &failure);
2837     __ bind(failure);
2838     __ set(0, dst);
2839     __ ba_short(done);
2840     __ bind(success);
2841     __ set(1, dst);
2842     __ bind(done);
2843   } else {
2844     ShouldNotReachHere();
2845   }
2846 
2847 }
2848 
2849 
2850 void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
2851   if (op->code() == lir_cas_long) {
2852     assert(VM_Version::supports_cx8(), "wrong machine");
2853     Register addr = op->addr()->as_pointer_register();
2854     Register cmp_value_lo = op->cmp_value()->as_register_lo();
2855     Register cmp_value_hi = op->cmp_value()->as_register_hi();
2856     Register new_value_lo = op->new_value()->as_register_lo();
2857     Register new_value_hi = op->new_value()->as_register_hi();
2858     Register t1 = op->tmp1()->as_register();
2859     Register t2 = op->tmp2()->as_register();
2860 #ifdef _LP64
2861     __ mov(cmp_value_lo, t1);
2862     __ mov(new_value_lo, t2);
2863     // perform the compare and swap operation
2864     __ casx(addr, t1, t2);
2865     // generate condition code - if the swap succeeded, t2 ("new value" reg) was
2866     // overwritten with the original value in "addr" and will be equal to t1.
2867     __ cmp(t1, t2);
2868 #else
2869     // move high and low halves of long values into single registers
2870     __ sllx(cmp_value_hi, 32, t1);         // shift high half into temp reg
2871     __ srl(cmp_value_lo, 0, cmp_value_lo); // clear upper 32 bits of low half
2872     __ or3(t1, cmp_value_lo, t1);          // t1 holds 64-bit compare value
2873     __ sllx(new_value_hi, 32, t2);
2874     __ srl(new_value_lo, 0, new_value_lo);
2875     __ or3(t2, new_value_lo, t2);          // t2 holds 64-bit value to swap
2876     // perform the compare and swap operation
2877     __ casx(addr, t1, t2);
2878     // generate condition code - if the swap succeeded, t2 ("new value" reg) was
2879     // overwritten with the original value in "addr" and will be equal to t1.
2880     // Produce icc flag for 32bit.
2881     __ sub(t1, t2, t2);
2882     __ srlx(t2, 32, t1);
2883     __ orcc(t2, t1, G0);
2884 #endif
2885   } else if (op->code() == lir_cas_int || op->code() == lir_cas_obj) {
2886     Register addr = op->addr()->as_pointer_register();
2887     Register cmp_value = op->cmp_value()->as_register();
2888     Register new_value = op->new_value()->as_register();
2889     Register t1 = op->tmp1()->as_register();
2890     Register t2 = op->tmp2()->as_register();
2891     __ mov(cmp_value, t1);
2892     __ mov(new_value, t2);
2893     if (op->code() == lir_cas_obj) {
2894       if (UseCompressedOops) {
2895         __ encode_heap_oop(t1);
2896         __ encode_heap_oop(t2);
2897         __ cas(addr, t1, t2);
2898       } else {
2899         __ cas_ptr(addr, t1, t2);
2900       }
2901     } else {
2902       __ cas(addr, t1, t2);
2903     }
2904     __ cmp(t1, t2);
2905   } else {
2906     Unimplemented();
2907   }
2908 }
2909 
2910 void LIR_Assembler::set_24bit_FPU() {
2911   Unimplemented();
2912 }
2913 
2914 
2915 void LIR_Assembler::reset_FPU() {
2916   Unimplemented();
2917 }
2918 
2919 
2920 void LIR_Assembler::breakpoint() {
2921   __ breakpoint_trap();
2922 }
2923 
2924 
2925 void LIR_Assembler::push(LIR_Opr opr) {
2926   Unimplemented();
2927 }
2928 
2929 
2930 void LIR_Assembler::pop(LIR_Opr opr) {
2931   Unimplemented();
2932 }
2933 
2934 
2935 void LIR_Assembler::monitor_address(int monitor_no, LIR_Opr dst_opr) {
2936   Address mon_addr = frame_map()->address_for_monitor_lock(monitor_no);
2937   Register dst = dst_opr->as_register();
2938   Register reg = mon_addr.base();
2939   int offset = mon_addr.disp();
2940   // compute pointer to BasicLock
2941   if (mon_addr.is_simm13()) {
2942     __ add(reg, offset, dst);
2943   } else {
2944     __ set(offset, dst);
2945     __ add(dst, reg, dst);
2946   }
2947 }
2948 
2949 void LIR_Assembler::emit_updatecrc32(LIR_OpUpdateCRC32* op) {
2950   fatal("CRC32 intrinsic is not implemented on this platform");
2951 }
2952 
2953 void LIR_Assembler::emit_lock(LIR_OpLock* op) {
2954   Register obj = op->obj_opr()->as_register();
2955   Register hdr = op->hdr_opr()->as_register();
2956   Register lock = op->lock_opr()->as_register();
2957 
2958   // obj may not be an oop
2959   if (op->code() == lir_lock) {
2960     MonitorEnterStub* stub = (MonitorEnterStub*)op->stub();
2961     if (UseFastLocking) {
2962       assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
2963       // add debug info for NullPointerException only if one is possible
2964       if (op->info() != NULL) {
2965         add_debug_info_for_null_check_here(op->info());
2966       }
2967       __ lock_object(hdr, obj, lock, op->scratch_opr()->as_register(), *op->stub()->entry());
2968     } else {
2969       // always do slow locking
2970       // note: the slow locking code could be inlined here, however if we use
2971       //       slow locking, speed doesn't matter anyway and this solution is
2972       //       simpler and requires less duplicated code - additionally, the
2973       //       slow locking code is the same in either case which simplifies
2974       //       debugging
2975       __ br(Assembler::always, false, Assembler::pt, *op->stub()->entry());
2976       __ delayed()->nop();
2977     }
2978   } else {
2979     assert (op->code() == lir_unlock, "Invalid code, expected lir_unlock");
2980     if (UseFastLocking) {
2981       assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
2982       __ unlock_object(hdr, obj, lock, *op->stub()->entry());
2983     } else {
2984       // always do slow unlocking
2985       // note: the slow unlocking code could be inlined here, however if we use
2986       //       slow unlocking, speed doesn't matter anyway and this solution is
2987       //       simpler and requires less duplicated code - additionally, the
2988       //       slow unlocking code is the same in either case which simplifies
2989       //       debugging
2990       __ br(Assembler::always, false, Assembler::pt, *op->stub()->entry());
2991       __ delayed()->nop();
2992     }
2993   }
2994   __ bind(*op->stub()->continuation());
2995 }
2996 
2997 
2998 void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
2999   ciMethod* method = op->profiled_method();
3000   int bci          = op->profiled_bci();
3001   ciMethod* callee = op->profiled_callee();
3002 
3003   // Update counter for all call types
3004   ciMethodData* md = method->method_data_or_null();
3005   assert(md != NULL, "Sanity");
3006   ciProfileData* data = md->bci_to_data(bci);
3007   assert(data->is_CounterData(), "need CounterData for calls");
3008   assert(op->mdo()->is_single_cpu(),  "mdo must be allocated");
3009   Register mdo  = op->mdo()->as_register();
3010 #ifdef _LP64
3011   assert(op->tmp1()->is_double_cpu(), "tmp1 must be allocated");
3012   Register tmp1 = op->tmp1()->as_register_lo();
3013 #else
3014   assert(op->tmp1()->is_single_cpu(), "tmp1 must be allocated");
3015   Register tmp1 = op->tmp1()->as_register();
3016 #endif
3017   metadata2reg(md->constant_encoding(), mdo);
3018   int mdo_offset_bias = 0;
3019   if (!Assembler::is_simm13(md->byte_offset_of_slot(data, CounterData::count_offset()) +
3020                             data->size_in_bytes())) {
3021     // The offset is large so bias the mdo by the base of the slot so
3022     // that the ld can use simm13s to reference the slots of the data
3023     mdo_offset_bias = md->byte_offset_of_slot(data, CounterData::count_offset());
3024     __ set(mdo_offset_bias, O7);
3025     __ add(mdo, O7, mdo);
3026   }
3027 
3028   Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()) - mdo_offset_bias);
3029   Bytecodes::Code bc = method->java_code_at_bci(bci);
3030   const bool callee_is_static = callee->is_loaded() && callee->is_static();
3031   // Perform additional virtual call profiling for invokevirtual and
3032   // invokeinterface bytecodes
3033   if ((bc == Bytecodes::_invokevirtual || bc == Bytecodes::_invokeinterface) &&
3034       !callee_is_static &&  // required for optimized MH invokes
3035       C1ProfileVirtualCalls) {
3036     assert(op->recv()->is_single_cpu(), "recv must be allocated");
3037     Register recv = op->recv()->as_register();
3038     assert_different_registers(mdo, tmp1, recv);
3039     assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
3040     ciKlass* known_klass = op->known_holder();
3041     if (C1OptimizeVirtualCallProfiling && known_klass != NULL) {
3042       // We know the type that will be seen at this call site; we can
3043       // statically update the MethodData* rather than needing to do
3044       // dynamic tests on the receiver type
3045 
3046       // NOTE: we should probably put a lock around this search to
3047       // avoid collisions by concurrent compilations
3048       ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
3049       uint i;
3050       for (i = 0; i < VirtualCallData::row_limit(); i++) {
3051         ciKlass* receiver = vc_data->receiver(i);
3052         if (known_klass->equals(receiver)) {
3053           Address data_addr(mdo, md->byte_offset_of_slot(data,
3054                                                          VirtualCallData::receiver_count_offset(i)) -
3055                             mdo_offset_bias);
3056           __ ld_ptr(data_addr, tmp1);
3057           __ add(tmp1, DataLayout::counter_increment, tmp1);
3058           __ st_ptr(tmp1, data_addr);
3059           return;
3060         }
3061       }
3062 
3063       // Receiver type not found in profile data; select an empty slot
3064 
3065       // Note that this is less efficient than it should be because it
3066       // always does a write to the receiver part of the
3067       // VirtualCallData rather than just the first time
3068       for (i = 0; i < VirtualCallData::row_limit(); i++) {
3069         ciKlass* receiver = vc_data->receiver(i);
3070         if (receiver == NULL) {
3071           Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)) -
3072                             mdo_offset_bias);
3073           metadata2reg(known_klass->constant_encoding(), tmp1);
3074           __ st_ptr(tmp1, recv_addr);
3075           Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)) -
3076                             mdo_offset_bias);
3077           __ ld_ptr(data_addr, tmp1);
3078           __ add(tmp1, DataLayout::counter_increment, tmp1);
3079           __ st_ptr(tmp1, data_addr);
3080           return;
3081         }
3082       }
3083     } else {
3084       __ load_klass(recv, recv);
3085       Label update_done;
3086       type_profile_helper(mdo, mdo_offset_bias, md, data, recv, tmp1, &update_done);
3087       // Receiver did not match any saved receiver and there is no empty row for it.
3088       // Increment total counter to indicate polymorphic case.
3089       __ ld_ptr(counter_addr, tmp1);
3090       __ add(tmp1, DataLayout::counter_increment, tmp1);
3091       __ st_ptr(tmp1, counter_addr);
3092 
3093       __ bind(update_done);
3094     }
3095   } else {
3096     // Static call
3097     __ ld_ptr(counter_addr, tmp1);
3098     __ add(tmp1, DataLayout::counter_increment, tmp1);
3099     __ st_ptr(tmp1, counter_addr);
3100   }
3101 }
3102 
3103 void LIR_Assembler::align_backward_branch_target() {
3104   __ align(OptoLoopAlignment);
3105 }
3106 
3107 
3108 void LIR_Assembler::emit_delay(LIR_OpDelay* op) {
3109   // make sure we are expecting a delay
3110   // this has the side effect of clearing the delay state
3111   // so we can use _masm instead of _masm->delayed() to do the
3112   // code generation.
3113   __ delayed();
3114 
3115   // make sure we only emit one instruction
3116   int offset = code_offset();
3117   op->delay_op()->emit_code(this);
3118 #ifdef ASSERT
3119   if (code_offset() - offset != NativeInstruction::nop_instruction_size) {
3120     op->delay_op()->print();
3121   }
3122   assert(code_offset() - offset == NativeInstruction::nop_instruction_size,
3123          "only one instruction can go in a delay slot");
3124 #endif
3125 
3126   // we may also be emitting the call info for the instruction
3127   // which we are the delay slot of.
3128   CodeEmitInfo* call_info = op->call_info();
3129   if (call_info) {
3130     add_call_info(code_offset(), call_info);
3131   }
3132 
3133   if (VerifyStackAtCalls) {
3134     _masm->sub(FP, SP, O7);
3135     _masm->cmp(O7, initial_frame_size_in_bytes());
3136     _masm->trap(Assembler::notEqual, Assembler::ptr_cc, G0, ST_RESERVED_FOR_USER_0+2 );
3137   }
3138 }
3139 
3140 
3141 void LIR_Assembler::negate(LIR_Opr left, LIR_Opr dest) {
3142   assert(left->is_register(), "can only handle registers");
3143 
3144   if (left->is_single_cpu()) {
3145     __ neg(left->as_register(), dest->as_register());
3146   } else if (left->is_single_fpu()) {
3147     __ fneg(FloatRegisterImpl::S, left->as_float_reg(), dest->as_float_reg());
3148   } else if (left->is_double_fpu()) {
3149     __ fneg(FloatRegisterImpl::D, left->as_double_reg(), dest->as_double_reg());
3150   } else {
3151     assert (left->is_double_cpu(), "Must be a long");
3152     Register Rlow = left->as_register_lo();
3153     Register Rhi = left->as_register_hi();
3154 #ifdef _LP64
3155     __ sub(G0, Rlow, dest->as_register_lo());
3156 #else
3157     __ subcc(G0, Rlow, dest->as_register_lo());
3158     __ subc (G0, Rhi,  dest->as_register_hi());
3159 #endif
3160   }
3161 }
3162 
3163 
3164 void LIR_Assembler::fxch(int i) {
3165   Unimplemented();
3166 }
3167 
3168 void LIR_Assembler::fld(int i) {
3169   Unimplemented();
3170 }
3171 
3172 void LIR_Assembler::ffree(int i) {
3173   Unimplemented();
3174 }
3175 
3176 void LIR_Assembler::rt_call(LIR_Opr result, address dest,
3177                             const LIR_OprList* args, LIR_Opr tmp, CodeEmitInfo* info) {
3178 
3179   // if tmp is invalid, then the function being called doesn't destroy the thread
3180   if (tmp->is_valid()) {
3181     __ save_thread(tmp->as_register());
3182   }
3183   __ call(dest, relocInfo::runtime_call_type);
3184   __ delayed()->nop();
3185   if (info != NULL) {
3186     add_call_info_here(info);
3187   }
3188   if (tmp->is_valid()) {
3189     __ restore_thread(tmp->as_register());
3190   }
3191 
3192 #ifdef ASSERT
3193   __ verify_thread();
3194 #endif // ASSERT
3195 }
3196 
3197 
3198 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {
3199 #ifdef _LP64
3200   ShouldNotReachHere();
3201 #endif
3202 
3203   NEEDS_CLEANUP;
3204   if (type == T_LONG) {
3205     LIR_Address* mem_addr = dest->is_address() ? dest->as_address_ptr() : src->as_address_ptr();
3206 
3207     // (extended to allow indexed as well as constant displaced for JSR-166)
3208     Register idx = noreg; // contains either constant offset or index
3209 
3210     int disp = mem_addr->disp();
3211     if (mem_addr->index() == LIR_OprFact::illegalOpr) {
3212       if (!Assembler::is_simm13(disp)) {
3213         idx = O7;
3214         __ set(disp, idx);
3215       }
3216     } else {
3217       assert(disp == 0, "not both indexed and disp");
3218       idx = mem_addr->index()->as_register();
3219     }
3220 
3221     int null_check_offset = -1;
3222 
3223     Register base = mem_addr->base()->as_register();
3224     if (src->is_register() && dest->is_address()) {
3225       // G4 is high half, G5 is low half
3226       // clear the top bits of G5, and scale up G4
3227       __ srl (src->as_register_lo(),  0, G5);
3228       __ sllx(src->as_register_hi(), 32, G4);
3229       // combine the two halves into the 64 bits of G4
3230       __ or3(G4, G5, G4);
3231       null_check_offset = __ offset();
3232       if (idx == noreg) {
3233         __ stx(G4, base, disp);
3234       } else {
3235         __ stx(G4, base, idx);
3236       }
3237     } else if (src->is_address() && dest->is_register()) {
3238       null_check_offset = __ offset();
3239       if (idx == noreg) {
3240         __ ldx(base, disp, G5);
3241       } else {
3242         __ ldx(base, idx, G5);
3243       }
3244       __ srax(G5, 32, dest->as_register_hi()); // fetch the high half into hi
3245       __ mov (G5, dest->as_register_lo());     // copy low half into lo
3246     } else {
3247       Unimplemented();
3248     }
3249     if (info != NULL) {
3250       add_debug_info_for_null_check(null_check_offset, info);
3251     }
3252 
3253   } else {
3254     // use normal move for all other volatiles since they don't need
3255     // special handling to remain atomic.
3256     move_op(src, dest, type, lir_patch_none, info, false, false, false);
3257   }
3258 }
3259 
3260 void LIR_Assembler::membar() {
3261   // only StoreLoad membars are ever explicitly needed on sparcs in TSO mode
3262   __ membar( Assembler::Membar_mask_bits(Assembler::StoreLoad) );
3263 }
3264 
3265 void LIR_Assembler::membar_acquire() {
3266   // no-op on TSO
3267 }
3268 
3269 void LIR_Assembler::membar_release() {
3270   // no-op on TSO
3271 }
3272 
3273 void LIR_Assembler::membar_loadload() {
3274   // no-op
3275   //__ membar(Assembler::Membar_mask_bits(Assembler::loadload));
3276 }
3277 
3278 void LIR_Assembler::membar_storestore() {
3279   // no-op
3280   //__ membar(Assembler::Membar_mask_bits(Assembler::storestore));
3281 }
3282 
3283 void LIR_Assembler::membar_loadstore() {
3284   // no-op
3285   //__ membar(Assembler::Membar_mask_bits(Assembler::loadstore));
3286 }
3287 
3288 void LIR_Assembler::membar_storeload() {
3289   __ membar(Assembler::Membar_mask_bits(Assembler::StoreLoad));
3290 }
3291 
3292 
3293 // Pack two sequential registers containing 32 bit values
3294 // into a single 64 bit register.
3295 // src and src->successor() are packed into dst
3296 // src and dst may be the same register.
3297 // Note: src is destroyed
3298 void LIR_Assembler::pack64(LIR_Opr src, LIR_Opr dst) {
3299   Register rs = src->as_register();
3300   Register rd = dst->as_register_lo();
3301   __ sllx(rs, 32, rs);
3302   __ srl(rs->successor(), 0, rs->successor());
3303   __ or3(rs, rs->successor(), rd);
3304 }
3305 
3306 // Unpack a 64 bit value in a register into
3307 // two sequential registers.
3308 // src is unpacked into dst and dst->successor()
3309 void LIR_Assembler::unpack64(LIR_Opr src, LIR_Opr dst) {
3310   Register rs = src->as_register_lo();
3311   Register rd = dst->as_register_hi();
3312   assert_different_registers(rs, rd, rd->successor());
3313   __ srlx(rs, 32, rd);
3314   __ srl (rs,  0, rd->successor());
3315 }
3316 
3317 
3318 void LIR_Assembler::leal(LIR_Opr addr_opr, LIR_Opr dest) {
3319   LIR_Address* addr = addr_opr->as_address_ptr();
3320   assert(addr->index()->is_illegal() && addr->scale() == LIR_Address::times_1 && Assembler::is_simm13(addr->disp()), "can't handle complex addresses yet");
3321 
3322   __ add(addr->base()->as_pointer_register(), addr->disp(), dest->as_pointer_register());
3323 }
3324 
3325 
3326 void LIR_Assembler::get_thread(LIR_Opr result_reg) {
3327   assert(result_reg->is_register(), "check");
3328   __ mov(G2_thread, result_reg->as_register());
3329 }
3330 
3331 #ifdef ASSERT
3332 // emit run-time assertion
3333 void LIR_Assembler::emit_assert(LIR_OpAssert* op) {
3334   assert(op->code() == lir_assert, "must be");
3335 
3336   if (op->in_opr1()->is_valid()) {
3337     assert(op->in_opr2()->is_valid(), "both operands must be valid");
3338     comp_op(op->condition(), op->in_opr1(), op->in_opr2(), op);
3339   } else {
3340     assert(op->in_opr2()->is_illegal(), "both operands must be illegal");
3341     assert(op->condition() == lir_cond_always, "no other conditions allowed");
3342   }
3343 
3344   Label ok;
3345   if (op->condition() != lir_cond_always) {
3346     Assembler::Condition acond;
3347     switch (op->condition()) {
3348       case lir_cond_equal:        acond = Assembler::equal;                break;
3349       case lir_cond_notEqual:     acond = Assembler::notEqual;             break;
3350       case lir_cond_less:         acond = Assembler::less;                 break;
3351       case lir_cond_lessEqual:    acond = Assembler::lessEqual;            break;
3352       case lir_cond_greaterEqual: acond = Assembler::greaterEqual;         break;
3353       case lir_cond_greater:      acond = Assembler::greater;              break;
3354       case lir_cond_aboveEqual:   acond = Assembler::greaterEqualUnsigned; break;
3355       case lir_cond_belowEqual:   acond = Assembler::lessEqualUnsigned;    break;
3356       default:                         ShouldNotReachHere();
3357     };
3358     __ br(acond, false, Assembler::pt, ok);
3359     __ delayed()->nop();
3360   }
3361   if (op->halt()) {
3362     const char* str = __ code_string(op->msg());
3363     __ stop(str);
3364   } else {
3365     breakpoint();
3366   }
3367   __ bind(ok);
3368 }
3369 #endif
3370 
3371 void LIR_Assembler::peephole(LIR_List* lir) {
3372   LIR_OpList* inst = lir->instructions_list();
3373   for (int i = 0; i < inst->length(); i++) {
3374     LIR_Op* op = inst->at(i);
3375     switch (op->code()) {
3376       case lir_cond_float_branch:
3377       case lir_branch: {
3378         LIR_OpBranch* branch = op->as_OpBranch();
3379         assert(branch->info() == NULL, "shouldn't be state on branches anymore");
3380         LIR_Op* delay_op = NULL;
3381         // we'd like to be able to pull following instructions into
3382         // this slot but we don't know enough to do it safely yet so
3383         // only optimize block to block control flow.
3384         if (LIRFillDelaySlots && branch->block()) {
3385           LIR_Op* prev = inst->at(i - 1);
3386           if (prev && LIR_Assembler::is_single_instruction(prev) && prev->info() == NULL) {
3387             // swap previous instruction into delay slot
3388             inst->at_put(i - 1, op);
3389             inst->at_put(i, new LIR_OpDelay(prev, op->info()));
3390 #ifndef PRODUCT
3391             if (LIRTracePeephole) {
3392               tty->print_cr("delayed");
3393               inst->at(i - 1)->print();
3394               inst->at(i)->print();
3395               tty->cr();
3396             }
3397 #endif
3398             continue;
3399           }
3400         }
3401 
3402         if (!delay_op) {
3403           delay_op = new LIR_OpDelay(new LIR_Op0(lir_nop), NULL);
3404         }
3405         inst->insert_before(i + 1, delay_op);
3406         break;
3407       }
3408       case lir_static_call:
3409       case lir_virtual_call:
3410       case lir_icvirtual_call:
3411       case lir_optvirtual_call:
3412       case lir_dynamic_call: {
3413         LIR_Op* prev = inst->at(i - 1);
3414         if (LIRFillDelaySlots && prev && prev->code() == lir_move && prev->info() == NULL &&
3415             (op->code() != lir_virtual_call ||
3416              !prev->result_opr()->is_single_cpu() ||
3417              prev->result_opr()->as_register() != O0) &&
3418             LIR_Assembler::is_single_instruction(prev)) {
3419           // Only moves without info can be put into the delay slot.
3420           // Also don't allow the setup of the receiver in the delay
3421           // slot for vtable calls.
3422           inst->at_put(i - 1, op);
3423           inst->at_put(i, new LIR_OpDelay(prev, op->info()));
3424 #ifndef PRODUCT
3425           if (LIRTracePeephole) {
3426             tty->print_cr("delayed");
3427             inst->at(i - 1)->print();
3428             inst->at(i)->print();
3429             tty->cr();
3430           }
3431 #endif
3432         } else {
3433           LIR_Op* delay_op = new LIR_OpDelay(new LIR_Op0(lir_nop), op->as_OpJavaCall()->info());
3434           inst->insert_before(i + 1, delay_op);
3435           i++;
3436         }
3437 
3438 #if defined(TIERED) && !defined(_LP64)
3439         // fixup the return value from G1 to O0/O1 for long returns.
3440         // It's done here instead of in LIRGenerator because there's
3441         // such a mismatch between the single reg and double reg
3442         // calling convention.
3443         LIR_OpJavaCall* callop = op->as_OpJavaCall();
3444         if (callop->result_opr() == FrameMap::out_long_opr) {
3445           LIR_OpJavaCall* call;
3446           LIR_OprList* arguments = new LIR_OprList(callop->arguments()->length());
3447           for (int a = 0; a < arguments->length(); a++) {
3448             arguments[a] = callop->arguments()[a];
3449           }
3450           if (op->code() == lir_virtual_call) {
3451             call = new LIR_OpJavaCall(op->code(), callop->method(), callop->receiver(), FrameMap::g1_long_single_opr,
3452                                       callop->vtable_offset(), arguments, callop->info());
3453           } else {
3454             call = new LIR_OpJavaCall(op->code(), callop->method(), callop->receiver(), FrameMap::g1_long_single_opr,
3455                                       callop->addr(), arguments, callop->info());
3456           }
3457           inst->at_put(i - 1, call);
3458           inst->insert_before(i + 1, new LIR_Op1(lir_unpack64, FrameMap::g1_long_single_opr, callop->result_opr(),
3459                                                  T_LONG, lir_patch_none, NULL));
3460         }
3461 #endif
3462         break;
3463       }
3464     }
3465   }
3466 }
3467 
3468 void LIR_Assembler::atomic_op(LIR_Code code, LIR_Opr src, LIR_Opr data, LIR_Opr dest, LIR_Opr tmp) {
3469   LIR_Address* addr = src->as_address_ptr();
3470 
3471   assert(data == dest, "swap uses only 2 operands");
3472   assert (code == lir_xchg, "no xadd on sparc");
3473 
3474   if (data->type() == T_INT) {
3475     __ swap(as_Address(addr), data->as_register());
3476   } else if (data->is_oop()) {
3477     Register obj = data->as_register();
3478     Register narrow = tmp->as_register();
3479 #ifdef _LP64
3480     assert(UseCompressedOops, "swap is 32bit only");
3481     __ encode_heap_oop(obj, narrow);
3482     __ swap(as_Address(addr), narrow);
3483     __ decode_heap_oop(narrow, obj);
3484 #else
3485     __ swap(as_Address(addr), obj);
3486 #endif
3487   } else {
3488     ShouldNotReachHere();
3489   }
3490 }
3491 
3492 #undef __