1 /*
   2  * Copyright (c) 2000, 2010, 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 (dst->is_register()) {
 104         if (src->is_address() && Assembler::is_simm13(src->as_address_ptr()->disp())) {
 105           return !PatchALot;
 106         } else if (src->is_single_stack()) {
 107           return true;
 108         }
 109       }
 110 
 111       if (src->is_register()) {
 112         if (dst->is_address() && Assembler::is_simm13(dst->as_address_ptr()->disp())) {
 113           return !PatchALot;
 114         } else if (dst->is_single_stack()) {
 115           return true;
 116         }
 117       }
 118 
 119       if (dst->is_register() &&
 120           ((src->is_register() && src->is_single_word() && src->is_same_type(dst)) ||
 121            (src->is_constant() && LIR_Assembler::is_small_constant(op->as_Op1()->in_opr())))) {
 122         return true;
 123       }
 124 
 125       return false;
 126     }
 127 
 128     default:
 129       return false;
 130   }
 131   ShouldNotReachHere();
 132 }
 133 
 134 
 135 LIR_Opr LIR_Assembler::receiverOpr() {
 136   return FrameMap::O0_oop_opr;
 137 }
 138 
 139 
 140 LIR_Opr LIR_Assembler::incomingReceiverOpr() {
 141   return FrameMap::I0_oop_opr;
 142 }
 143 
 144 
 145 LIR_Opr LIR_Assembler::osrBufferPointer() {
 146   return FrameMap::I0_opr;
 147 }
 148 
 149 
 150 int LIR_Assembler::initial_frame_size_in_bytes() {
 151   return in_bytes(frame_map()->framesize_in_bytes());
 152 }
 153 
 154 
 155 // inline cache check: the inline cached class is in G5_inline_cache_reg(G5);
 156 // we fetch the class of the receiver (O0) and compare it with the cached class.
 157 // If they do not match we jump to slow case.
 158 int LIR_Assembler::check_icache() {
 159   int offset = __ offset();
 160   __ inline_cache_check(O0, G5_inline_cache_reg);
 161   return offset;
 162 }
 163 
 164 
 165 void LIR_Assembler::osr_entry() {
 166   // On-stack-replacement entry sequence (interpreter frame layout described in interpreter_sparc.cpp):
 167   //
 168   //   1. Create a new compiled activation.
 169   //   2. Initialize local variables in the compiled activation.  The expression stack must be empty
 170   //      at the osr_bci; it is not initialized.
 171   //   3. Jump to the continuation address in compiled code to resume execution.
 172 
 173   // OSR entry point
 174   offsets()->set_value(CodeOffsets::OSR_Entry, code_offset());
 175   BlockBegin* osr_entry = compilation()->hir()->osr_entry();
 176   ValueStack* entry_state = osr_entry->end()->state();
 177   int number_of_locks = entry_state->locks_size();
 178 
 179   // Create a frame for the compiled activation.
 180   __ build_frame(initial_frame_size_in_bytes());
 181 
 182   // OSR buffer is
 183   //
 184   // locals[nlocals-1..0]
 185   // monitors[number_of_locks-1..0]
 186   //
 187   // locals is a direct copy of the interpreter frame so in the osr buffer
 188   // so first slot in the local array is the last local from the interpreter
 189   // and last slot is local[0] (receiver) from the interpreter
 190   //
 191   // Similarly with locks. The first lock slot in the osr buffer is the nth lock
 192   // from the interpreter frame, the nth lock slot in the osr buffer is 0th lock
 193   // in the interpreter frame (the method lock if a sync method)
 194 
 195   // Initialize monitors in the compiled activation.
 196   //   I0: pointer to osr buffer
 197   //
 198   // All other registers are dead at this point and the locals will be
 199   // copied into place by code emitted in the IR.
 200 
 201   Register OSR_buf = osrBufferPointer()->as_register();
 202   { assert(frame::interpreter_frame_monitor_size() == BasicObjectLock::size(), "adjust code below");
 203     int monitor_offset = BytesPerWord * method()->max_locals() +
 204       (2 * BytesPerWord) * (number_of_locks - 1);
 205     // SharedRuntime::OSR_migration_begin() packs BasicObjectLocks in
 206     // the OSR buffer using 2 word entries: first the lock and then
 207     // the oop.
 208     for (int i = 0; i < number_of_locks; i++) {
 209       int slot_offset = monitor_offset - ((i * 2) * BytesPerWord);
 210 #ifdef ASSERT
 211       // verify the interpreter's monitor has a non-null object
 212       {
 213         Label L;
 214         __ ld_ptr(OSR_buf, slot_offset + 1*BytesPerWord, O7);
 215         __ cmp(G0, O7);
 216         __ br(Assembler::notEqual, false, Assembler::pt, L);
 217         __ delayed()->nop();
 218         __ stop("locked object is NULL");
 219         __ bind(L);
 220       }
 221 #endif // ASSERT
 222       // Copy the lock field into the compiled activation.
 223       __ ld_ptr(OSR_buf, slot_offset + 0, O7);
 224       __ st_ptr(O7, frame_map()->address_for_monitor_lock(i));
 225       __ ld_ptr(OSR_buf, slot_offset + 1*BytesPerWord, O7);
 226       __ st_ptr(O7, frame_map()->address_for_monitor_object(i));
 227     }
 228   }
 229 }
 230 
 231 
 232 // Optimized Library calls
 233 // This is the fast version of java.lang.String.compare; it has not
 234 // OSR-entry and therefore, we generate a slow version for OSR's
 235 void LIR_Assembler::emit_string_compare(LIR_Opr left, LIR_Opr right, LIR_Opr dst, CodeEmitInfo* info) {
 236   Register str0 = left->as_register();
 237   Register str1 = right->as_register();
 238 
 239   Label Ldone;
 240 
 241   Register result = dst->as_register();
 242   {
 243     // Get a pointer to the first character of string0 in tmp0 and get string0.count in str0
 244     // Get a pointer to the first character of string1 in tmp1 and get string1.count in str1
 245     // Also, get string0.count-string1.count in o7 and get the condition code set
 246     // Note: some instructions have been hoisted for better instruction scheduling
 247 
 248     Register tmp0 = L0;
 249     Register tmp1 = L1;
 250     Register tmp2 = L2;
 251 
 252     int  value_offset = java_lang_String:: value_offset_in_bytes(); // char array
 253     int offset_offset = java_lang_String::offset_offset_in_bytes(); // first character position
 254     int  count_offset = java_lang_String:: count_offset_in_bytes();
 255 
 256     __ ld_ptr(str0, value_offset, tmp0);
 257     __ ld(str0, offset_offset, tmp2);
 258     __ add(tmp0, arrayOopDesc::base_offset_in_bytes(T_CHAR), tmp0);
 259     __ ld(str0, count_offset, str0);
 260     __ sll(tmp2, exact_log2(sizeof(jchar)), tmp2);
 261 
 262     // str1 may be null
 263     add_debug_info_for_null_check_here(info);
 264 
 265     __ ld_ptr(str1, value_offset, tmp1);
 266     __ add(tmp0, tmp2, tmp0);
 267 
 268     __ ld(str1, offset_offset, tmp2);
 269     __ add(tmp1, arrayOopDesc::base_offset_in_bytes(T_CHAR), tmp1);
 270     __ ld(str1, count_offset, str1);
 271     __ sll(tmp2, exact_log2(sizeof(jchar)), tmp2);
 272     __ subcc(str0, str1, O7);
 273     __ add(tmp1, tmp2, tmp1);
 274   }
 275 
 276   {
 277     // Compute the minimum of the string lengths, scale it and store it in limit
 278     Register count0 = I0;
 279     Register count1 = I1;
 280     Register limit  = L3;
 281 
 282     Label Lskip;
 283     __ sll(count0, exact_log2(sizeof(jchar)), limit);             // string0 is shorter
 284     __ br(Assembler::greater, true, Assembler::pt, Lskip);
 285     __ delayed()->sll(count1, exact_log2(sizeof(jchar)), limit);  // string1 is shorter
 286     __ bind(Lskip);
 287 
 288     // If either string is empty (or both of them) the result is the difference in lengths
 289     __ cmp(limit, 0);
 290     __ br(Assembler::equal, true, Assembler::pn, Ldone);
 291     __ delayed()->mov(O7, result);  // result is difference in lengths
 292   }
 293 
 294   {
 295     // Neither string is empty
 296     Label Lloop;
 297 
 298     Register base0 = L0;
 299     Register base1 = L1;
 300     Register chr0  = I0;
 301     Register chr1  = I1;
 302     Register limit = L3;
 303 
 304     // Shift base0 and base1 to the end of the arrays, negate limit
 305     __ add(base0, limit, base0);
 306     __ add(base1, limit, base1);
 307     __ neg(limit);  // limit = -min{string0.count, strin1.count}
 308 
 309     __ lduh(base0, limit, chr0);
 310     __ bind(Lloop);
 311     __ lduh(base1, limit, chr1);
 312     __ subcc(chr0, chr1, chr0);
 313     __ br(Assembler::notZero, false, Assembler::pn, Ldone);
 314     assert(chr0 == result, "result must be pre-placed");
 315     __ delayed()->inccc(limit, sizeof(jchar));
 316     __ br(Assembler::notZero, true, Assembler::pt, Lloop);
 317     __ delayed()->lduh(base0, limit, chr0);
 318   }
 319 
 320   // If strings are equal up to min length, return the length difference.
 321   __ mov(O7, result);
 322 
 323   // Otherwise, return the difference between the first mismatched chars.
 324   __ bind(Ldone);
 325 }
 326 
 327 
 328 // --------------------------------------------------------------------------------------------
 329 
 330 void LIR_Assembler::monitorexit(LIR_Opr obj_opr, LIR_Opr lock_opr, Register hdr, int monitor_no) {
 331   if (!GenerateSynchronizationCode) return;
 332 
 333   Register obj_reg = obj_opr->as_register();
 334   Register lock_reg = lock_opr->as_register();
 335 
 336   Address mon_addr = frame_map()->address_for_monitor_lock(monitor_no);
 337   Register reg = mon_addr.base();
 338   int offset = mon_addr.disp();
 339   // compute pointer to BasicLock
 340   if (mon_addr.is_simm13()) {
 341     __ add(reg, offset, lock_reg);
 342   }
 343   else {
 344     __ set(offset, lock_reg);
 345     __ add(reg, lock_reg, lock_reg);
 346   }
 347   // unlock object
 348   MonitorAccessStub* slow_case = new MonitorExitStub(lock_opr, UseFastLocking, monitor_no);
 349   // _slow_case_stubs->append(slow_case);
 350   // temporary fix: must be created after exceptionhandler, therefore as call stub
 351   _slow_case_stubs->append(slow_case);
 352   if (UseFastLocking) {
 353     // try inlined fast unlocking first, revert to slow locking if it fails
 354     // note: lock_reg points to the displaced header since the displaced header offset is 0!
 355     assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
 356     __ unlock_object(hdr, obj_reg, lock_reg, *slow_case->entry());
 357   } else {
 358     // always do slow unlocking
 359     // note: the slow unlocking code could be inlined here, however if we use
 360     //       slow unlocking, speed doesn't matter anyway and this solution is
 361     //       simpler and requires less duplicated code - additionally, the
 362     //       slow unlocking code is the same in either case which simplifies
 363     //       debugging
 364     __ br(Assembler::always, false, Assembler::pt, *slow_case->entry());
 365     __ delayed()->nop();
 366   }
 367   // done
 368   __ bind(*slow_case->continuation());
 369 }
 370 
 371 
 372 int LIR_Assembler::emit_exception_handler() {
 373   // if the last instruction is a call (typically to do a throw which
 374   // is coming at the end after block reordering) the return address
 375   // must still point into the code area in order to avoid assertion
 376   // failures when searching for the corresponding bci => add a nop
 377   // (was bug 5/14/1999 - gri)
 378   __ nop();
 379 
 380   // generate code for exception handler
 381   ciMethod* method = compilation()->method();
 382 
 383   address handler_base = __ start_a_stub(exception_handler_size);
 384 
 385   if (handler_base == NULL) {
 386     // not enough space left for the handler
 387     bailout("exception handler overflow");
 388     return -1;
 389   }
 390 
 391   int offset = code_offset();
 392 
 393   __ call(Runtime1::entry_for(Runtime1::handle_exception_id), relocInfo::runtime_call_type);
 394   __ delayed()->nop();
 395   debug_only(__ stop("should have gone to the caller");)
 396   assert(code_offset() - offset <= exception_handler_size, "overflow");
 397   __ end_a_stub();
 398 
 399   return offset;
 400 }
 401 
 402 
 403 // Emit the code to remove the frame from the stack in the exception
 404 // unwind path.
 405 int LIR_Assembler::emit_unwind_handler() {
 406 #ifndef PRODUCT
 407   if (CommentedAssembly) {
 408     _masm->block_comment("Unwind handler");
 409   }
 410 #endif
 411 
 412   int offset = code_offset();
 413 
 414   // Fetch the exception from TLS and clear out exception related thread state
 415   __ ld_ptr(G2_thread, in_bytes(JavaThread::exception_oop_offset()), O0);
 416   __ st_ptr(G0, G2_thread, in_bytes(JavaThread::exception_oop_offset()));
 417   __ st_ptr(G0, G2_thread, in_bytes(JavaThread::exception_pc_offset()));
 418 
 419   __ bind(_unwind_handler_entry);
 420   __ verify_not_null_oop(O0);
 421   if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
 422     __ mov(O0, I0);  // Preserve the exception
 423   }
 424 
 425   // Preform needed unlocking
 426   MonitorExitStub* stub = NULL;
 427   if (method()->is_synchronized()) {
 428     monitor_address(0, FrameMap::I1_opr);
 429     stub = new MonitorExitStub(FrameMap::I1_opr, true, 0);
 430     __ unlock_object(I3, I2, I1, *stub->entry());
 431     __ bind(*stub->continuation());
 432   }
 433 
 434   if (compilation()->env()->dtrace_method_probes()) {
 435     __ mov(G2_thread, O0);
 436     jobject2reg(method()->constant_encoding(), O1);
 437     __ call(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), relocInfo::runtime_call_type);
 438     __ delayed()->nop();
 439   }
 440 
 441   if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
 442     __ mov(I0, O0);  // Restore the exception
 443   }
 444 
 445   // dispatch to the unwind logic
 446   __ call(Runtime1::entry_for(Runtime1::unwind_exception_id), relocInfo::runtime_call_type);
 447   __ delayed()->nop();
 448 
 449   // Emit the slow path assembly
 450   if (stub != NULL) {
 451     stub->emit_code(this);
 452   }
 453 
 454   return offset;
 455 }
 456 
 457 
 458 int LIR_Assembler::emit_deopt_handler() {
 459   // if the last instruction is a call (typically to do a throw which
 460   // is coming at the end after block reordering) the return address
 461   // must still point into the code area in order to avoid assertion
 462   // failures when searching for the corresponding bci => add a nop
 463   // (was bug 5/14/1999 - gri)
 464   __ nop();
 465 
 466   // generate code for deopt handler
 467   ciMethod* method = compilation()->method();
 468   address handler_base = __ start_a_stub(deopt_handler_size);
 469   if (handler_base == NULL) {
 470     // not enough space left for the handler
 471     bailout("deopt handler overflow");
 472     return -1;
 473   }
 474 
 475   int offset = code_offset();
 476   AddressLiteral deopt_blob(SharedRuntime::deopt_blob()->unpack());
 477   __ JUMP(deopt_blob, G3_scratch, 0); // sethi;jmp
 478   __ delayed()->nop();
 479   assert(code_offset() - offset <= deopt_handler_size, "overflow");
 480   debug_only(__ stop("should have gone to the caller");)
 481   __ end_a_stub();
 482 
 483   return offset;
 484 }
 485 
 486 
 487 void LIR_Assembler::jobject2reg(jobject o, Register reg) {
 488   if (o == NULL) {
 489     __ set(NULL_WORD, reg);
 490   } else {
 491     int oop_index = __ oop_recorder()->find_index(o);
 492     RelocationHolder rspec = oop_Relocation::spec(oop_index);
 493     __ set(NULL_WORD, reg, rspec); // Will be set when the nmethod is created
 494   }
 495 }
 496 
 497 
 498 void LIR_Assembler::jobject2reg_with_patching(Register reg, CodeEmitInfo *info) {
 499   // Allocate a new index in oop table to hold the oop once it's been patched
 500   int oop_index = __ oop_recorder()->allocate_index((jobject)NULL);
 501   PatchingStub* patch = new PatchingStub(_masm, PatchingStub::load_klass_id, oop_index);
 502 
 503   AddressLiteral addrlit(NULL, oop_Relocation::spec(oop_index));
 504   assert(addrlit.rspec().type() == relocInfo::oop_type, "must be an oop reloc");
 505   // It may not seem necessary to use a sethi/add pair to load a NULL into dest, but the
 506   // NULL will be dynamically patched later and the patched value may be large.  We must
 507   // therefore generate the sethi/add as a placeholders
 508   __ patchable_set(addrlit, reg);
 509 
 510   patching_epilog(patch, lir_patch_normal, reg, info);
 511 }
 512 
 513 
 514 void LIR_Assembler::emit_op3(LIR_Op3* op) {
 515   Register Rdividend = op->in_opr1()->as_register();
 516   Register Rdivisor  = noreg;
 517   Register Rscratch  = op->in_opr3()->as_register();
 518   Register Rresult   = op->result_opr()->as_register();
 519   int divisor = -1;
 520 
 521   if (op->in_opr2()->is_register()) {
 522     Rdivisor = op->in_opr2()->as_register();
 523   } else {
 524     divisor = op->in_opr2()->as_constant_ptr()->as_jint();
 525     assert(Assembler::is_simm13(divisor), "can only handle simm13");
 526   }
 527 
 528   assert(Rdividend != Rscratch, "");
 529   assert(Rdivisor  != Rscratch, "");
 530   assert(op->code() == lir_idiv || op->code() == lir_irem, "Must be irem or idiv");
 531 
 532   if (Rdivisor == noreg && is_power_of_2(divisor)) {
 533     // convert division by a power of two into some shifts and logical operations
 534     if (op->code() == lir_idiv) {
 535       if (divisor == 2) {
 536         __ srl(Rdividend, 31, Rscratch);
 537       } else {
 538         __ sra(Rdividend, 31, Rscratch);
 539         __ and3(Rscratch, divisor - 1, Rscratch);
 540       }
 541       __ add(Rdividend, Rscratch, Rscratch);
 542       __ sra(Rscratch, log2_intptr(divisor), Rresult);
 543       return;
 544     } else {
 545       if (divisor == 2) {
 546         __ srl(Rdividend, 31, Rscratch);
 547       } else {
 548         __ sra(Rdividend, 31, Rscratch);
 549         __ and3(Rscratch, divisor - 1,Rscratch);
 550       }
 551       __ add(Rdividend, Rscratch, Rscratch);
 552       __ andn(Rscratch, divisor - 1,Rscratch);
 553       __ sub(Rdividend, Rscratch, Rresult);
 554       return;
 555     }
 556   }
 557 
 558   __ sra(Rdividend, 31, Rscratch);
 559   __ wry(Rscratch);
 560   if (!VM_Version::v9_instructions_work()) {
 561     // v9 doesn't require these nops
 562     __ nop();
 563     __ nop();
 564     __ nop();
 565     __ nop();
 566   }
 567 
 568   add_debug_info_for_div0_here(op->info());
 569 
 570   if (Rdivisor != noreg) {
 571     __ sdivcc(Rdividend, Rdivisor, (op->code() == lir_idiv ? Rresult : Rscratch));
 572   } else {
 573     assert(Assembler::is_simm13(divisor), "can only handle simm13");
 574     __ sdivcc(Rdividend, divisor, (op->code() == lir_idiv ? Rresult : Rscratch));
 575   }
 576 
 577   Label skip;
 578   __ br(Assembler::overflowSet, true, Assembler::pn, skip);
 579   __ delayed()->Assembler::sethi(0x80000000, (op->code() == lir_idiv ? Rresult : Rscratch));
 580   __ bind(skip);
 581 
 582   if (op->code() == lir_irem) {
 583     if (Rdivisor != noreg) {
 584       __ smul(Rscratch, Rdivisor, Rscratch);
 585     } else {
 586       __ smul(Rscratch, divisor, Rscratch);
 587     }
 588     __ sub(Rdividend, Rscratch, Rresult);
 589   }
 590 }
 591 
 592 
 593 void LIR_Assembler::emit_opBranch(LIR_OpBranch* op) {
 594 #ifdef ASSERT
 595   assert(op->block() == NULL || op->block()->label() == op->label(), "wrong label");
 596   if (op->block() != NULL)  _branch_target_blocks.append(op->block());
 597   if (op->ublock() != NULL) _branch_target_blocks.append(op->ublock());
 598 #endif
 599   assert(op->info() == NULL, "shouldn't have CodeEmitInfo");
 600 
 601   if (op->cond() == lir_cond_always) {
 602     __ br(Assembler::always, false, Assembler::pt, *(op->label()));
 603   } else if (op->code() == lir_cond_float_branch) {
 604     assert(op->ublock() != NULL, "must have unordered successor");
 605     bool is_unordered = (op->ublock() == op->block());
 606     Assembler::Condition acond;
 607     switch (op->cond()) {
 608       case lir_cond_equal:         acond = Assembler::f_equal;    break;
 609       case lir_cond_notEqual:      acond = Assembler::f_notEqual; break;
 610       case lir_cond_less:          acond = (is_unordered ? Assembler::f_unorderedOrLess          : Assembler::f_less);           break;
 611       case lir_cond_greater:       acond = (is_unordered ? Assembler::f_unorderedOrGreater       : Assembler::f_greater);        break;
 612       case lir_cond_lessEqual:     acond = (is_unordered ? Assembler::f_unorderedOrLessOrEqual   : Assembler::f_lessOrEqual);    break;
 613       case lir_cond_greaterEqual:  acond = (is_unordered ? Assembler::f_unorderedOrGreaterOrEqual: Assembler::f_greaterOrEqual); break;
 614       default :                         ShouldNotReachHere();
 615     };
 616 
 617     if (!VM_Version::v9_instructions_work()) {
 618       __ nop();
 619     }
 620     __ fb( acond, false, Assembler::pn, *(op->label()));
 621   } else {
 622     assert (op->code() == lir_branch, "just checking");
 623 
 624     Assembler::Condition acond;
 625     switch (op->cond()) {
 626       case lir_cond_equal:        acond = Assembler::equal;                break;
 627       case lir_cond_notEqual:     acond = Assembler::notEqual;             break;
 628       case lir_cond_less:         acond = Assembler::less;                 break;
 629       case lir_cond_lessEqual:    acond = Assembler::lessEqual;            break;
 630       case lir_cond_greaterEqual: acond = Assembler::greaterEqual;         break;
 631       case lir_cond_greater:      acond = Assembler::greater;              break;
 632       case lir_cond_aboveEqual:   acond = Assembler::greaterEqualUnsigned; break;
 633       case lir_cond_belowEqual:   acond = Assembler::lessEqualUnsigned;    break;
 634       default:                         ShouldNotReachHere();
 635     };
 636 
 637     // sparc has different condition codes for testing 32-bit
 638     // vs. 64-bit values.  We could always test xcc is we could
 639     // guarantee that 32-bit loads always sign extended but that isn't
 640     // true and since sign extension isn't free, it would impose a
 641     // slight cost.
 642 #ifdef _LP64
 643     if  (op->type() == T_INT) {
 644       __ br(acond, false, Assembler::pn, *(op->label()));
 645     } else
 646 #endif
 647       __ brx(acond, false, Assembler::pn, *(op->label()));
 648   }
 649   // The peephole pass fills the delay slot
 650 }
 651 
 652 
 653 void LIR_Assembler::emit_opConvert(LIR_OpConvert* op) {
 654   Bytecodes::Code code = op->bytecode();
 655   LIR_Opr dst = op->result_opr();
 656 
 657   switch(code) {
 658     case Bytecodes::_i2l: {
 659       Register rlo  = dst->as_register_lo();
 660       Register rhi  = dst->as_register_hi();
 661       Register rval = op->in_opr()->as_register();
 662 #ifdef _LP64
 663       __ sra(rval, 0, rlo);
 664 #else
 665       __ mov(rval, rlo);
 666       __ sra(rval, BitsPerInt-1, rhi);
 667 #endif
 668       break;
 669     }
 670     case Bytecodes::_i2d:
 671     case Bytecodes::_i2f: {
 672       bool is_double = (code == Bytecodes::_i2d);
 673       FloatRegister rdst = is_double ? dst->as_double_reg() : dst->as_float_reg();
 674       FloatRegisterImpl::Width w = is_double ? FloatRegisterImpl::D : FloatRegisterImpl::S;
 675       FloatRegister rsrc = op->in_opr()->as_float_reg();
 676       if (rsrc != rdst) {
 677         __ fmov(FloatRegisterImpl::S, rsrc, rdst);
 678       }
 679       __ fitof(w, rdst, rdst);
 680       break;
 681     }
 682     case Bytecodes::_f2i:{
 683       FloatRegister rsrc = op->in_opr()->as_float_reg();
 684       Address       addr = frame_map()->address_for_slot(dst->single_stack_ix());
 685       Label L;
 686       // result must be 0 if value is NaN; test by comparing value to itself
 687       __ fcmp(FloatRegisterImpl::S, Assembler::fcc0, rsrc, rsrc);
 688       if (!VM_Version::v9_instructions_work()) {
 689         __ nop();
 690       }
 691       __ fb(Assembler::f_unordered, true, Assembler::pn, L);
 692       __ delayed()->st(G0, addr); // annuled if contents of rsrc is not NaN
 693       __ ftoi(FloatRegisterImpl::S, rsrc, rsrc);
 694       // move integer result from float register to int register
 695       __ stf(FloatRegisterImpl::S, rsrc, addr.base(), addr.disp());
 696       __ bind (L);
 697       break;
 698     }
 699     case Bytecodes::_l2i: {
 700       Register rlo  = op->in_opr()->as_register_lo();
 701       Register rhi  = op->in_opr()->as_register_hi();
 702       Register rdst = dst->as_register();
 703 #ifdef _LP64
 704       __ sra(rlo, 0, rdst);
 705 #else
 706       __ mov(rlo, rdst);
 707 #endif
 708       break;
 709     }
 710     case Bytecodes::_d2f:
 711     case Bytecodes::_f2d: {
 712       bool is_double = (code == Bytecodes::_f2d);
 713       assert((!is_double && dst->is_single_fpu()) || (is_double && dst->is_double_fpu()), "check");
 714       LIR_Opr val = op->in_opr();
 715       FloatRegister rval = (code == Bytecodes::_d2f) ? val->as_double_reg() : val->as_float_reg();
 716       FloatRegister rdst = is_double ? dst->as_double_reg() : dst->as_float_reg();
 717       FloatRegisterImpl::Width vw = is_double ? FloatRegisterImpl::S : FloatRegisterImpl::D;
 718       FloatRegisterImpl::Width dw = is_double ? FloatRegisterImpl::D : FloatRegisterImpl::S;
 719       __ ftof(vw, dw, rval, rdst);
 720       break;
 721     }
 722     case Bytecodes::_i2s:
 723     case Bytecodes::_i2b: {
 724       Register rval = op->in_opr()->as_register();
 725       Register rdst = dst->as_register();
 726       int shift = (code == Bytecodes::_i2b) ? (BitsPerInt - T_BYTE_aelem_bytes * BitsPerByte) : (BitsPerInt - BitsPerShort);
 727       __ sll (rval, shift, rdst);
 728       __ sra (rdst, shift, rdst);
 729       break;
 730     }
 731     case Bytecodes::_i2c: {
 732       Register rval = op->in_opr()->as_register();
 733       Register rdst = dst->as_register();
 734       int shift = BitsPerInt - T_CHAR_aelem_bytes * BitsPerByte;
 735       __ sll (rval, shift, rdst);
 736       __ srl (rdst, shift, rdst);
 737       break;
 738     }
 739 
 740     default: ShouldNotReachHere();
 741   }
 742 }
 743 
 744 
 745 void LIR_Assembler::align_call(LIR_Code) {
 746   // do nothing since all instructions are word aligned on sparc
 747 }
 748 
 749 
 750 void LIR_Assembler::call(LIR_OpJavaCall* op, relocInfo::relocType rtype) {
 751   __ call(op->addr(), rtype);
 752   // The peephole pass fills the delay slot, add_call_info is done in
 753   // LIR_Assembler::emit_delay.
 754 }
 755 
 756 
 757 void LIR_Assembler::ic_call(LIR_OpJavaCall* op) {
 758   RelocationHolder rspec = virtual_call_Relocation::spec(pc());
 759   __ set_oop((jobject)Universe::non_oop_word(), G5_inline_cache_reg);
 760   __ relocate(rspec);
 761   __ call(op->addr(), relocInfo::none);
 762   // The peephole pass fills the delay slot, add_call_info is done in
 763   // LIR_Assembler::emit_delay.
 764 }
 765 
 766 
 767 void LIR_Assembler::vtable_call(LIR_OpJavaCall* op) {
 768   add_debug_info_for_null_check_here(op->info());
 769   __ ld_ptr(O0, oopDesc::klass_offset_in_bytes(), G3_scratch);
 770   if (__ is_simm13(op->vtable_offset())) {
 771     __ ld_ptr(G3_scratch, op->vtable_offset(), G5_method);
 772   } else {
 773     // This will generate 2 instructions
 774     __ set(op->vtable_offset(), G5_method);
 775     // ld_ptr, set_hi, set
 776     __ ld_ptr(G3_scratch, G5_method, G5_method);
 777   }
 778   __ ld_ptr(G5_method, methodOopDesc::from_compiled_offset(), G3_scratch);
 779   __ callr(G3_scratch, G0);
 780   // the peephole pass fills the delay slot
 781 }
 782 
 783 
 784 // load with 32-bit displacement
 785 int LIR_Assembler::load(Register s, int disp, Register d, BasicType ld_type, CodeEmitInfo *info) {
 786   int load_offset = code_offset();
 787   if (Assembler::is_simm13(disp)) {
 788     if (info != NULL) add_debug_info_for_null_check_here(info);
 789     switch(ld_type) {
 790       case T_BOOLEAN: // fall through
 791       case T_BYTE  : __ ldsb(s, disp, d); break;
 792       case T_CHAR  : __ lduh(s, disp, d); break;
 793       case T_SHORT : __ ldsh(s, disp, d); break;
 794       case T_INT   : __ ld(s, disp, d); break;
 795       case T_ADDRESS:// fall through
 796       case T_ARRAY : // fall through
 797       case T_OBJECT: __ ld_ptr(s, disp, d); break;
 798       default      : ShouldNotReachHere();
 799     }
 800   } else {
 801     __ set(disp, O7);
 802     if (info != NULL) add_debug_info_for_null_check_here(info);
 803     load_offset = code_offset();
 804     switch(ld_type) {
 805       case T_BOOLEAN: // fall through
 806       case T_BYTE  : __ ldsb(s, O7, d); break;
 807       case T_CHAR  : __ lduh(s, O7, d); break;
 808       case T_SHORT : __ ldsh(s, O7, d); break;
 809       case T_INT   : __ ld(s, O7, d); break;
 810       case T_ADDRESS:// fall through
 811       case T_ARRAY : // fall through
 812       case T_OBJECT: __ ld_ptr(s, O7, d); break;
 813       default      : ShouldNotReachHere();
 814     }
 815   }
 816   if (ld_type == T_ARRAY || ld_type == T_OBJECT) __ verify_oop(d);
 817   return load_offset;
 818 }
 819 
 820 
 821 // store with 32-bit displacement
 822 void LIR_Assembler::store(Register value, Register base, int offset, BasicType type, CodeEmitInfo *info) {
 823   if (Assembler::is_simm13(offset)) {
 824     if (info != NULL)  add_debug_info_for_null_check_here(info);
 825     switch (type) {
 826       case T_BOOLEAN: // fall through
 827       case T_BYTE  : __ stb(value, base, offset); break;
 828       case T_CHAR  : __ sth(value, base, offset); break;
 829       case T_SHORT : __ sth(value, base, offset); break;
 830       case T_INT   : __ stw(value, base, offset); break;
 831       case T_ADDRESS:// fall through
 832       case T_ARRAY : // fall through
 833       case T_OBJECT: __ st_ptr(value, base, offset); break;
 834       default      : ShouldNotReachHere();
 835     }
 836   } else {
 837     __ set(offset, O7);
 838     if (info != NULL) add_debug_info_for_null_check_here(info);
 839     switch (type) {
 840       case T_BOOLEAN: // fall through
 841       case T_BYTE  : __ stb(value, base, O7); break;
 842       case T_CHAR  : __ sth(value, base, O7); break;
 843       case T_SHORT : __ sth(value, base, O7); break;
 844       case T_INT   : __ stw(value, base, O7); break;
 845       case T_ADDRESS:// fall through
 846       case T_ARRAY : //fall through
 847       case T_OBJECT: __ st_ptr(value, base, O7); break;
 848       default      : ShouldNotReachHere();
 849     }
 850   }
 851   // Note: Do the store before verification as the code might be patched!
 852   if (type == T_ARRAY || type == T_OBJECT) __ verify_oop(value);
 853 }
 854 
 855 
 856 // load float with 32-bit displacement
 857 void LIR_Assembler::load(Register s, int disp, FloatRegister d, BasicType ld_type, CodeEmitInfo *info) {
 858   FloatRegisterImpl::Width w;
 859   switch(ld_type) {
 860     case T_FLOAT : w = FloatRegisterImpl::S; break;
 861     case T_DOUBLE: w = FloatRegisterImpl::D; break;
 862     default      : ShouldNotReachHere();
 863   }
 864 
 865   if (Assembler::is_simm13(disp)) {
 866     if (info != NULL) add_debug_info_for_null_check_here(info);
 867     if (disp % BytesPerLong != 0 && w == FloatRegisterImpl::D) {
 868       __ ldf(FloatRegisterImpl::S, s, disp + BytesPerWord, d->successor());
 869       __ ldf(FloatRegisterImpl::S, s, disp               , d);
 870     } else {
 871       __ ldf(w, s, disp, d);
 872     }
 873   } else {
 874     __ set(disp, O7);
 875     if (info != NULL) add_debug_info_for_null_check_here(info);
 876     __ ldf(w, s, O7, d);
 877   }
 878 }
 879 
 880 
 881 // store float with 32-bit displacement
 882 void LIR_Assembler::store(FloatRegister value, Register base, int offset, BasicType type, CodeEmitInfo *info) {
 883   FloatRegisterImpl::Width w;
 884   switch(type) {
 885     case T_FLOAT : w = FloatRegisterImpl::S; break;
 886     case T_DOUBLE: w = FloatRegisterImpl::D; break;
 887     default      : ShouldNotReachHere();
 888   }
 889 
 890   if (Assembler::is_simm13(offset)) {
 891     if (info != NULL) add_debug_info_for_null_check_here(info);
 892     if (w == FloatRegisterImpl::D && offset % BytesPerLong != 0) {
 893       __ stf(FloatRegisterImpl::S, value->successor(), base, offset + BytesPerWord);
 894       __ stf(FloatRegisterImpl::S, value             , base, offset);
 895     } else {
 896       __ stf(w, value, base, offset);
 897     }
 898   } else {
 899     __ set(offset, O7);
 900     if (info != NULL) add_debug_info_for_null_check_here(info);
 901     __ stf(w, value, O7, base);
 902   }
 903 }
 904 
 905 
 906 int LIR_Assembler::store(LIR_Opr from_reg, Register base, int offset, BasicType type, bool unaligned) {
 907   int store_offset;
 908   if (!Assembler::is_simm13(offset + (type == T_LONG) ? wordSize : 0)) {
 909     assert(!unaligned, "can't handle this");
 910     // for offsets larger than a simm13 we setup the offset in O7
 911     __ set(offset, O7);
 912     store_offset = store(from_reg, base, O7, type);
 913   } else {
 914     if (type == T_ARRAY || type == T_OBJECT) __ verify_oop(from_reg->as_register());
 915     store_offset = code_offset();
 916     switch (type) {
 917       case T_BOOLEAN: // fall through
 918       case T_BYTE  : __ stb(from_reg->as_register(), base, offset); break;
 919       case T_CHAR  : __ sth(from_reg->as_register(), base, offset); break;
 920       case T_SHORT : __ sth(from_reg->as_register(), base, offset); break;
 921       case T_INT   : __ stw(from_reg->as_register(), base, offset); break;
 922       case T_LONG  :
 923 #ifdef _LP64
 924         if (unaligned || PatchALot) {
 925           __ srax(from_reg->as_register_lo(), 32, O7);
 926           __ stw(from_reg->as_register_lo(), base, offset + lo_word_offset_in_bytes);
 927           __ stw(O7,                         base, offset + hi_word_offset_in_bytes);
 928         } else {
 929           __ stx(from_reg->as_register_lo(), base, offset);
 930         }
 931 #else
 932         assert(Assembler::is_simm13(offset + 4), "must be");
 933         __ stw(from_reg->as_register_lo(), base, offset + lo_word_offset_in_bytes);
 934         __ stw(from_reg->as_register_hi(), base, offset + hi_word_offset_in_bytes);
 935 #endif
 936         break;
 937       case T_ADDRESS:// fall through
 938       case T_ARRAY : // fall through
 939       case T_OBJECT: __ st_ptr(from_reg->as_register(), base, offset); break;
 940       case T_FLOAT : __ stf(FloatRegisterImpl::S, from_reg->as_float_reg(), base, offset); break;
 941       case T_DOUBLE:
 942         {
 943           FloatRegister reg = from_reg->as_double_reg();
 944           // split unaligned stores
 945           if (unaligned || PatchALot) {
 946             assert(Assembler::is_simm13(offset + 4), "must be");
 947             __ stf(FloatRegisterImpl::S, reg->successor(), base, offset + 4);
 948             __ stf(FloatRegisterImpl::S, reg,              base, offset);
 949           } else {
 950             __ stf(FloatRegisterImpl::D, reg, base, offset);
 951           }
 952           break;
 953         }
 954       default      : ShouldNotReachHere();
 955     }
 956   }
 957   return store_offset;
 958 }
 959 
 960 
 961 int LIR_Assembler::store(LIR_Opr from_reg, Register base, Register disp, BasicType type) {
 962   if (type == T_ARRAY || type == T_OBJECT) __ verify_oop(from_reg->as_register());
 963   int store_offset = code_offset();
 964   switch (type) {
 965     case T_BOOLEAN: // fall through
 966     case T_BYTE  : __ stb(from_reg->as_register(), base, disp); break;
 967     case T_CHAR  : __ sth(from_reg->as_register(), base, disp); break;
 968     case T_SHORT : __ sth(from_reg->as_register(), base, disp); break;
 969     case T_INT   : __ stw(from_reg->as_register(), base, disp); break;
 970     case T_LONG  :
 971 #ifdef _LP64
 972       __ stx(from_reg->as_register_lo(), base, disp);
 973 #else
 974       assert(from_reg->as_register_hi()->successor() == from_reg->as_register_lo(), "must match");
 975       __ std(from_reg->as_register_hi(), base, disp);
 976 #endif
 977       break;
 978     case T_ADDRESS:// fall through
 979     case T_ARRAY : // fall through
 980     case T_OBJECT: __ st_ptr(from_reg->as_register(), base, disp); break;
 981     case T_FLOAT : __ stf(FloatRegisterImpl::S, from_reg->as_float_reg(), base, disp); break;
 982     case T_DOUBLE: __ stf(FloatRegisterImpl::D, from_reg->as_double_reg(), base, disp); break;
 983     default      : ShouldNotReachHere();
 984   }
 985   return store_offset;
 986 }
 987 
 988 
 989 int LIR_Assembler::load(Register base, int offset, LIR_Opr to_reg, BasicType type, bool unaligned) {
 990   int load_offset;
 991   if (!Assembler::is_simm13(offset + (type == T_LONG) ? wordSize : 0)) {
 992     assert(base != O7, "destroying register");
 993     assert(!unaligned, "can't handle this");
 994     // for offsets larger than a simm13 we setup the offset in O7
 995     __ set(offset, O7);
 996     load_offset = load(base, O7, to_reg, type);
 997   } else {
 998     load_offset = code_offset();
 999     switch(type) {
1000       case T_BOOLEAN: // fall through
1001       case T_BYTE  : __ ldsb(base, offset, to_reg->as_register()); break;
1002       case T_CHAR  : __ lduh(base, offset, to_reg->as_register()); break;
1003       case T_SHORT : __ ldsh(base, offset, to_reg->as_register()); break;
1004       case T_INT   : __ ld(base, offset, to_reg->as_register()); break;
1005       case T_LONG  :
1006         if (!unaligned) {
1007 #ifdef _LP64
1008           __ ldx(base, offset, to_reg->as_register_lo());
1009 #else
1010           assert(to_reg->as_register_hi()->successor() == to_reg->as_register_lo(),
1011                  "must be sequential");
1012           __ ldd(base, offset, to_reg->as_register_hi());
1013 #endif
1014         } else {
1015 #ifdef _LP64
1016           assert(base != to_reg->as_register_lo(), "can't handle this");
1017           assert(O7 != to_reg->as_register_lo(), "can't handle this");
1018           __ ld(base, offset + hi_word_offset_in_bytes, to_reg->as_register_lo());
1019           __ lduw(base, offset + lo_word_offset_in_bytes, O7); // in case O7 is base or offset, use it last
1020           __ sllx(to_reg->as_register_lo(), 32, to_reg->as_register_lo());
1021           __ or3(to_reg->as_register_lo(), O7, to_reg->as_register_lo());
1022 #else
1023           if (base == to_reg->as_register_lo()) {
1024             __ ld(base, offset + hi_word_offset_in_bytes, to_reg->as_register_hi());
1025             __ ld(base, offset + lo_word_offset_in_bytes, to_reg->as_register_lo());
1026           } else {
1027             __ ld(base, offset + lo_word_offset_in_bytes, to_reg->as_register_lo());
1028             __ ld(base, offset + hi_word_offset_in_bytes, to_reg->as_register_hi());
1029           }
1030 #endif
1031         }
1032         break;
1033       case T_ADDRESS:// fall through
1034       case T_ARRAY : // fall through
1035       case T_OBJECT: __ ld_ptr(base, offset, to_reg->as_register()); break;
1036       case T_FLOAT:  __ ldf(FloatRegisterImpl::S, base, offset, to_reg->as_float_reg()); break;
1037       case T_DOUBLE:
1038         {
1039           FloatRegister reg = to_reg->as_double_reg();
1040           // split unaligned loads
1041           if (unaligned || PatchALot) {
1042             __ ldf(FloatRegisterImpl::S, base, offset + 4, reg->successor());
1043             __ ldf(FloatRegisterImpl::S, base, offset,     reg);
1044           } else {
1045             __ ldf(FloatRegisterImpl::D, base, offset, to_reg->as_double_reg());
1046           }
1047           break;
1048         }
1049       default      : ShouldNotReachHere();
1050     }
1051     if (type == T_ARRAY || type == T_OBJECT) __ verify_oop(to_reg->as_register());
1052   }
1053   return load_offset;
1054 }
1055 
1056 
1057 int LIR_Assembler::load(Register base, Register disp, LIR_Opr to_reg, BasicType type) {
1058   int load_offset = code_offset();
1059   switch(type) {
1060     case T_BOOLEAN: // fall through
1061     case T_BYTE  : __ ldsb(base, disp, to_reg->as_register()); break;
1062     case T_CHAR  : __ lduh(base, disp, to_reg->as_register()); break;
1063     case T_SHORT : __ ldsh(base, disp, to_reg->as_register()); break;
1064     case T_INT   : __ ld(base, disp, to_reg->as_register()); break;
1065     case T_ADDRESS:// fall through
1066     case T_ARRAY : // fall through
1067     case T_OBJECT: __ ld_ptr(base, disp, to_reg->as_register()); break;
1068     case T_FLOAT:  __ ldf(FloatRegisterImpl::S, base, disp, to_reg->as_float_reg()); break;
1069     case T_DOUBLE: __ ldf(FloatRegisterImpl::D, base, disp, to_reg->as_double_reg()); break;
1070     case T_LONG  :
1071 #ifdef _LP64
1072       __ ldx(base, disp, to_reg->as_register_lo());
1073 #else
1074       assert(to_reg->as_register_hi()->successor() == to_reg->as_register_lo(),
1075              "must be sequential");
1076       __ ldd(base, disp, to_reg->as_register_hi());
1077 #endif
1078       break;
1079     default      : ShouldNotReachHere();
1080   }
1081   if (type == T_ARRAY || type == T_OBJECT) __ verify_oop(to_reg->as_register());
1082   return load_offset;
1083 }
1084 
1085 
1086 // load/store with an Address
1087 void LIR_Assembler::load(const Address& a, Register d,  BasicType ld_type, CodeEmitInfo *info, int offset) {
1088   load(a.base(), a.disp() + offset, d, ld_type, info);
1089 }
1090 
1091 
1092 void LIR_Assembler::store(Register value, const Address& dest, BasicType type, CodeEmitInfo *info, int offset) {
1093   store(value, dest.base(), dest.disp() + offset, type, info);
1094 }
1095 
1096 
1097 // loadf/storef with an Address
1098 void LIR_Assembler::load(const Address& a, FloatRegister d, BasicType ld_type, CodeEmitInfo *info, int offset) {
1099   load(a.base(), a.disp() + offset, d, ld_type, info);
1100 }
1101 
1102 
1103 void LIR_Assembler::store(FloatRegister value, const Address& dest, BasicType type, CodeEmitInfo *info, int offset) {
1104   store(value, dest.base(), dest.disp() + offset, type, info);
1105 }
1106 
1107 
1108 // load/store with an Address
1109 void LIR_Assembler::load(LIR_Address* a, Register d,  BasicType ld_type, CodeEmitInfo *info) {
1110   load(as_Address(a), d, ld_type, info);
1111 }
1112 
1113 
1114 void LIR_Assembler::store(Register value, LIR_Address* dest, BasicType type, CodeEmitInfo *info) {
1115   store(value, as_Address(dest), type, info);
1116 }
1117 
1118 
1119 // loadf/storef with an Address
1120 void LIR_Assembler::load(LIR_Address* a, FloatRegister d, BasicType ld_type, CodeEmitInfo *info) {
1121   load(as_Address(a), d, ld_type, info);
1122 }
1123 
1124 
1125 void LIR_Assembler::store(FloatRegister value, LIR_Address* dest, BasicType type, CodeEmitInfo *info) {
1126   store(value, as_Address(dest), type, info);
1127 }
1128 
1129 
1130 void LIR_Assembler::const2stack(LIR_Opr src, LIR_Opr dest) {
1131   LIR_Const* c = src->as_constant_ptr();
1132   switch (c->type()) {
1133     case T_INT:
1134     case T_FLOAT:
1135     case T_ADDRESS: {
1136       Register src_reg = O7;
1137       int value = c->as_jint_bits();
1138       if (value == 0) {
1139         src_reg = G0;
1140       } else {
1141         __ set(value, O7);
1142       }
1143       Address addr = frame_map()->address_for_slot(dest->single_stack_ix());
1144       __ stw(src_reg, addr.base(), addr.disp());
1145       break;
1146     }
1147     case T_OBJECT: {
1148       Register src_reg = O7;
1149       jobject2reg(c->as_jobject(), src_reg);
1150       Address addr = frame_map()->address_for_slot(dest->single_stack_ix());
1151       __ st_ptr(src_reg, addr.base(), addr.disp());
1152       break;
1153     }
1154     case T_LONG:
1155     case T_DOUBLE: {
1156       Address addr = frame_map()->address_for_double_slot(dest->double_stack_ix());
1157 
1158       Register tmp = O7;
1159       int value_lo = c->as_jint_lo_bits();
1160       if (value_lo == 0) {
1161         tmp = G0;
1162       } else {
1163         __ set(value_lo, O7);
1164       }
1165       __ stw(tmp, addr.base(), addr.disp() + lo_word_offset_in_bytes);
1166       int value_hi = c->as_jint_hi_bits();
1167       if (value_hi == 0) {
1168         tmp = G0;
1169       } else {
1170         __ set(value_hi, O7);
1171       }
1172       __ stw(tmp, addr.base(), addr.disp() + hi_word_offset_in_bytes);
1173       break;
1174     }
1175     default:
1176       Unimplemented();
1177   }
1178 }
1179 
1180 
1181 void LIR_Assembler::const2mem(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info ) {
1182   LIR_Const* c = src->as_constant_ptr();
1183   LIR_Address* addr     = dest->as_address_ptr();
1184   Register base = addr->base()->as_pointer_register();
1185 
1186   if (info != NULL) {
1187     add_debug_info_for_null_check_here(info);
1188   }
1189   switch (c->type()) {
1190     case T_INT:
1191     case T_FLOAT:
1192     case T_ADDRESS: {
1193       LIR_Opr tmp = FrameMap::O7_opr;
1194       int value = c->as_jint_bits();
1195       if (value == 0) {
1196         tmp = FrameMap::G0_opr;
1197       } else if (Assembler::is_simm13(value)) {
1198         __ set(value, O7);
1199       }
1200       if (addr->index()->is_valid()) {
1201         assert(addr->disp() == 0, "must be zero");
1202         store(tmp, base, addr->index()->as_pointer_register(), type);
1203       } else {
1204         assert(Assembler::is_simm13(addr->disp()), "can't handle larger addresses");
1205         store(tmp, base, addr->disp(), type);
1206       }
1207       break;
1208     }
1209     case T_LONG:
1210     case T_DOUBLE: {
1211       assert(!addr->index()->is_valid(), "can't handle reg reg address here");
1212       assert(Assembler::is_simm13(addr->disp()) &&
1213              Assembler::is_simm13(addr->disp() + 4), "can't handle larger addresses");
1214 
1215       Register tmp = O7;
1216       int value_lo = c->as_jint_lo_bits();
1217       if (value_lo == 0) {
1218         tmp = G0;
1219       } else {
1220         __ set(value_lo, O7);
1221       }
1222       store(tmp, base, addr->disp() + lo_word_offset_in_bytes, T_INT);
1223       int value_hi = c->as_jint_hi_bits();
1224       if (value_hi == 0) {
1225         tmp = G0;
1226       } else {
1227         __ set(value_hi, O7);
1228       }
1229       store(tmp, base, addr->disp() + hi_word_offset_in_bytes, T_INT);
1230       break;
1231     }
1232     case T_OBJECT: {
1233       jobject obj = c->as_jobject();
1234       LIR_Opr tmp;
1235       if (obj == NULL) {
1236         tmp = FrameMap::G0_opr;
1237       } else {
1238         tmp = FrameMap::O7_opr;
1239         jobject2reg(c->as_jobject(), O7);
1240       }
1241       // handle either reg+reg or reg+disp address
1242       if (addr->index()->is_valid()) {
1243         assert(addr->disp() == 0, "must be zero");
1244         store(tmp, base, addr->index()->as_pointer_register(), type);
1245       } else {
1246         assert(Assembler::is_simm13(addr->disp()), "can't handle larger addresses");
1247         store(tmp, base, addr->disp(), type);
1248       }
1249 
1250       break;
1251     }
1252     default:
1253       Unimplemented();
1254   }
1255 }
1256 
1257 
1258 void LIR_Assembler::const2reg(LIR_Opr src, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
1259   LIR_Const* c = src->as_constant_ptr();
1260   LIR_Opr to_reg = dest;
1261 
1262   switch (c->type()) {
1263     case T_INT:
1264     case T_ADDRESS:
1265       {
1266         jint con = c->as_jint();
1267         if (to_reg->is_single_cpu()) {
1268           assert(patch_code == lir_patch_none, "no patching handled here");
1269           __ set(con, to_reg->as_register());
1270         } else {
1271           ShouldNotReachHere();
1272           assert(to_reg->is_single_fpu(), "wrong register kind");
1273 
1274           __ set(con, O7);
1275           Address temp_slot(SP, (frame::register_save_words * wordSize) + STACK_BIAS);
1276           __ st(O7, temp_slot);
1277           __ ldf(FloatRegisterImpl::S, temp_slot, to_reg->as_float_reg());
1278         }
1279       }
1280       break;
1281 
1282     case T_LONG:
1283       {
1284         jlong con = c->as_jlong();
1285 
1286         if (to_reg->is_double_cpu()) {
1287 #ifdef _LP64
1288           __ set(con,  to_reg->as_register_lo());
1289 #else
1290           __ set(low(con),  to_reg->as_register_lo());
1291           __ set(high(con), to_reg->as_register_hi());
1292 #endif
1293 #ifdef _LP64
1294         } else if (to_reg->is_single_cpu()) {
1295           __ set(con, to_reg->as_register());
1296 #endif
1297         } else {
1298           ShouldNotReachHere();
1299           assert(to_reg->is_double_fpu(), "wrong register kind");
1300           Address temp_slot_lo(SP, ((frame::register_save_words  ) * wordSize) + STACK_BIAS);
1301           Address temp_slot_hi(SP, ((frame::register_save_words) * wordSize) + (longSize/2) + STACK_BIAS);
1302           __ set(low(con),  O7);
1303           __ st(O7, temp_slot_lo);
1304           __ set(high(con), O7);
1305           __ st(O7, temp_slot_hi);
1306           __ ldf(FloatRegisterImpl::D, temp_slot_lo, to_reg->as_double_reg());
1307         }
1308       }
1309       break;
1310 
1311     case T_OBJECT:
1312       {
1313         if (patch_code == lir_patch_none) {
1314           jobject2reg(c->as_jobject(), to_reg->as_register());
1315         } else {
1316           jobject2reg_with_patching(to_reg->as_register(), info);
1317         }
1318       }
1319       break;
1320 
1321     case T_FLOAT:
1322       {
1323         address const_addr = __ float_constant(c->as_jfloat());
1324         if (const_addr == NULL) {
1325           bailout("const section overflow");
1326           break;
1327         }
1328         RelocationHolder rspec = internal_word_Relocation::spec(const_addr);
1329         AddressLiteral const_addrlit(const_addr, rspec);
1330         if (to_reg->is_single_fpu()) {
1331           __ patchable_sethi(const_addrlit, O7);
1332           __ relocate(rspec);
1333           __ ldf(FloatRegisterImpl::S, O7, const_addrlit.low10(), to_reg->as_float_reg());
1334 
1335         } else {
1336           assert(to_reg->is_single_cpu(), "Must be a cpu register.");
1337 
1338           __ set(const_addrlit, O7);
1339           load(O7, 0, to_reg->as_register(), T_INT);
1340         }
1341       }
1342       break;
1343 
1344     case T_DOUBLE:
1345       {
1346         address const_addr = __ double_constant(c->as_jdouble());
1347         if (const_addr == NULL) {
1348           bailout("const section overflow");
1349           break;
1350         }
1351         RelocationHolder rspec = internal_word_Relocation::spec(const_addr);
1352 
1353         if (to_reg->is_double_fpu()) {
1354           AddressLiteral const_addrlit(const_addr, rspec);
1355           __ patchable_sethi(const_addrlit, O7);
1356           __ relocate(rspec);
1357           __ ldf (FloatRegisterImpl::D, O7, const_addrlit.low10(), to_reg->as_double_reg());
1358         } else {
1359           assert(to_reg->is_double_cpu(), "Must be a long register.");
1360 #ifdef _LP64
1361           __ set(jlong_cast(c->as_jdouble()), to_reg->as_register_lo());
1362 #else
1363           __ set(low(jlong_cast(c->as_jdouble())), to_reg->as_register_lo());
1364           __ set(high(jlong_cast(c->as_jdouble())), to_reg->as_register_hi());
1365 #endif
1366         }
1367 
1368       }
1369       break;
1370 
1371     default:
1372       ShouldNotReachHere();
1373   }
1374 }
1375 
1376 Address LIR_Assembler::as_Address(LIR_Address* addr) {
1377   Register reg = addr->base()->as_register();
1378   return Address(reg, addr->disp());
1379 }
1380 
1381 
1382 void LIR_Assembler::stack2stack(LIR_Opr src, LIR_Opr dest, BasicType type) {
1383   switch (type) {
1384     case T_INT:
1385     case T_FLOAT: {
1386       Register tmp = O7;
1387       Address from = frame_map()->address_for_slot(src->single_stack_ix());
1388       Address to   = frame_map()->address_for_slot(dest->single_stack_ix());
1389       __ lduw(from.base(), from.disp(), tmp);
1390       __ stw(tmp, to.base(), to.disp());
1391       break;
1392     }
1393     case T_OBJECT: {
1394       Register tmp = O7;
1395       Address from = frame_map()->address_for_slot(src->single_stack_ix());
1396       Address to   = frame_map()->address_for_slot(dest->single_stack_ix());
1397       __ ld_ptr(from.base(), from.disp(), tmp);
1398       __ st_ptr(tmp, to.base(), to.disp());
1399       break;
1400     }
1401     case T_LONG:
1402     case T_DOUBLE: {
1403       Register tmp = O7;
1404       Address from = frame_map()->address_for_double_slot(src->double_stack_ix());
1405       Address to   = frame_map()->address_for_double_slot(dest->double_stack_ix());
1406       __ lduw(from.base(), from.disp(), tmp);
1407       __ stw(tmp, to.base(), to.disp());
1408       __ lduw(from.base(), from.disp() + 4, tmp);
1409       __ stw(tmp, to.base(), to.disp() + 4);
1410       break;
1411     }
1412 
1413     default:
1414       ShouldNotReachHere();
1415   }
1416 }
1417 
1418 
1419 Address LIR_Assembler::as_Address_hi(LIR_Address* addr) {
1420   Address base = as_Address(addr);
1421   return Address(base.base(), base.disp() + hi_word_offset_in_bytes);
1422 }
1423 
1424 
1425 Address LIR_Assembler::as_Address_lo(LIR_Address* addr) {
1426   Address base = as_Address(addr);
1427   return Address(base.base(), base.disp() + lo_word_offset_in_bytes);
1428 }
1429 
1430 
1431 void LIR_Assembler::mem2reg(LIR_Opr src_opr, LIR_Opr dest, BasicType type,
1432                             LIR_PatchCode patch_code, CodeEmitInfo* info, bool unaligned) {
1433 
1434   LIR_Address* addr = src_opr->as_address_ptr();
1435   LIR_Opr to_reg = dest;
1436 
1437   Register src = addr->base()->as_pointer_register();
1438   Register disp_reg = noreg;
1439   int disp_value = addr->disp();
1440   bool needs_patching = (patch_code != lir_patch_none);
1441 
1442   if (addr->base()->type() == T_OBJECT) {
1443     __ verify_oop(src);
1444   }
1445 
1446   PatchingStub* patch = NULL;
1447   if (needs_patching) {
1448     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1449     assert(!to_reg->is_double_cpu() ||
1450            patch_code == lir_patch_none ||
1451            patch_code == lir_patch_normal, "patching doesn't match register");
1452   }
1453 
1454   if (addr->index()->is_illegal()) {
1455     if (!Assembler::is_simm13(disp_value) && (!unaligned || Assembler::is_simm13(disp_value + 4))) {
1456       if (needs_patching) {
1457         __ patchable_set(0, O7);
1458       } else {
1459         __ set(disp_value, O7);
1460       }
1461       disp_reg = O7;
1462     }
1463   } else if (unaligned || PatchALot) {
1464     __ add(src, addr->index()->as_register(), O7);
1465     src = O7;
1466   } else {
1467     disp_reg = addr->index()->as_pointer_register();
1468     assert(disp_value == 0, "can't handle 3 operand addresses");
1469   }
1470 
1471   // remember the offset of the load.  The patching_epilog must be done
1472   // before the call to add_debug_info, otherwise the PcDescs don't get
1473   // entered in increasing order.
1474   int offset = code_offset();
1475 
1476   assert(disp_reg != noreg || Assembler::is_simm13(disp_value), "should have set this up");
1477   if (disp_reg == noreg) {
1478     offset = load(src, disp_value, to_reg, type, unaligned);
1479   } else {
1480     assert(!unaligned, "can't handle this");
1481     offset = load(src, disp_reg, to_reg, type);
1482   }
1483 
1484   if (patch != NULL) {
1485     patching_epilog(patch, patch_code, src, info);
1486   }
1487 
1488   if (info != NULL) add_debug_info_for_null_check(offset, info);
1489 }
1490 
1491 
1492 void LIR_Assembler::prefetchr(LIR_Opr src) {
1493   LIR_Address* addr = src->as_address_ptr();
1494   Address from_addr = as_Address(addr);
1495 
1496   if (VM_Version::has_v9()) {
1497     __ prefetch(from_addr, Assembler::severalReads);
1498   }
1499 }
1500 
1501 
1502 void LIR_Assembler::prefetchw(LIR_Opr src) {
1503   LIR_Address* addr = src->as_address_ptr();
1504   Address from_addr = as_Address(addr);
1505 
1506   if (VM_Version::has_v9()) {
1507     __ prefetch(from_addr, Assembler::severalWritesAndPossiblyReads);
1508   }
1509 }
1510 
1511 
1512 void LIR_Assembler::stack2reg(LIR_Opr src, LIR_Opr dest, BasicType type) {
1513   Address addr;
1514   if (src->is_single_word()) {
1515     addr = frame_map()->address_for_slot(src->single_stack_ix());
1516   } else if (src->is_double_word())  {
1517     addr = frame_map()->address_for_double_slot(src->double_stack_ix());
1518   }
1519 
1520   bool unaligned = (addr.disp() - STACK_BIAS) % 8 != 0;
1521   load(addr.base(), addr.disp(), dest, dest->type(), unaligned);
1522 }
1523 
1524 
1525 void LIR_Assembler::reg2stack(LIR_Opr from_reg, LIR_Opr dest, BasicType type, bool pop_fpu_stack) {
1526   Address addr;
1527   if (dest->is_single_word()) {
1528     addr = frame_map()->address_for_slot(dest->single_stack_ix());
1529   } else if (dest->is_double_word())  {
1530     addr = frame_map()->address_for_slot(dest->double_stack_ix());
1531   }
1532   bool unaligned = (addr.disp() - STACK_BIAS) % 8 != 0;
1533   store(from_reg, addr.base(), addr.disp(), from_reg->type(), unaligned);
1534 }
1535 
1536 
1537 void LIR_Assembler::reg2reg(LIR_Opr from_reg, LIR_Opr to_reg) {
1538   if (from_reg->is_float_kind() && to_reg->is_float_kind()) {
1539     if (from_reg->is_double_fpu()) {
1540       // double to double moves
1541       assert(to_reg->is_double_fpu(), "should match");
1542       __ fmov(FloatRegisterImpl::D, from_reg->as_double_reg(), to_reg->as_double_reg());
1543     } else {
1544       // float to float moves
1545       assert(to_reg->is_single_fpu(), "should match");
1546       __ fmov(FloatRegisterImpl::S, from_reg->as_float_reg(), to_reg->as_float_reg());
1547     }
1548   } else if (!from_reg->is_float_kind() && !to_reg->is_float_kind()) {
1549     if (from_reg->is_double_cpu()) {
1550 #ifdef _LP64
1551       __ mov(from_reg->as_pointer_register(), to_reg->as_pointer_register());
1552 #else
1553       assert(to_reg->is_double_cpu() &&
1554              from_reg->as_register_hi() != to_reg->as_register_lo() &&
1555              from_reg->as_register_lo() != to_reg->as_register_hi(),
1556              "should both be long and not overlap");
1557       // long to long moves
1558       __ mov(from_reg->as_register_hi(), to_reg->as_register_hi());
1559       __ mov(from_reg->as_register_lo(), to_reg->as_register_lo());
1560 #endif
1561 #ifdef _LP64
1562     } else if (to_reg->is_double_cpu()) {
1563       // int to int moves
1564       __ mov(from_reg->as_register(), to_reg->as_register_lo());
1565 #endif
1566     } else {
1567       // int to int moves
1568       __ mov(from_reg->as_register(), to_reg->as_register());
1569     }
1570   } else {
1571     ShouldNotReachHere();
1572   }
1573   if (to_reg->type() == T_OBJECT || to_reg->type() == T_ARRAY) {
1574     __ verify_oop(to_reg->as_register());
1575   }
1576 }
1577 
1578 
1579 void LIR_Assembler::reg2mem(LIR_Opr from_reg, LIR_Opr dest, BasicType type,
1580                             LIR_PatchCode patch_code, CodeEmitInfo* info, bool pop_fpu_stack,
1581                             bool unaligned) {
1582   LIR_Address* addr = dest->as_address_ptr();
1583 
1584   Register src = addr->base()->as_pointer_register();
1585   Register disp_reg = noreg;
1586   int disp_value = addr->disp();
1587   bool needs_patching = (patch_code != lir_patch_none);
1588 
1589   if (addr->base()->is_oop_register()) {
1590     __ verify_oop(src);
1591   }
1592 
1593   PatchingStub* patch = NULL;
1594   if (needs_patching) {
1595     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1596     assert(!from_reg->is_double_cpu() ||
1597            patch_code == lir_patch_none ||
1598            patch_code == lir_patch_normal, "patching doesn't match register");
1599   }
1600 
1601   if (addr->index()->is_illegal()) {
1602     if (!Assembler::is_simm13(disp_value) && (!unaligned || Assembler::is_simm13(disp_value + 4))) {
1603       if (needs_patching) {
1604         __ patchable_set(0, O7);
1605       } else {
1606         __ set(disp_value, O7);
1607       }
1608       disp_reg = O7;
1609     }
1610   } else if (unaligned || PatchALot) {
1611     __ add(src, addr->index()->as_register(), O7);
1612     src = O7;
1613   } else {
1614     disp_reg = addr->index()->as_pointer_register();
1615     assert(disp_value == 0, "can't handle 3 operand addresses");
1616   }
1617 
1618   // remember the offset of the store.  The patching_epilog must be done
1619   // before the call to add_debug_info_for_null_check, otherwise the PcDescs don't get
1620   // entered in increasing order.
1621   int offset;
1622 
1623   assert(disp_reg != noreg || Assembler::is_simm13(disp_value), "should have set this up");
1624   if (disp_reg == noreg) {
1625     offset = store(from_reg, src, disp_value, type, unaligned);
1626   } else {
1627     assert(!unaligned, "can't handle this");
1628     offset = store(from_reg, src, disp_reg, type);
1629   }
1630 
1631   if (patch != NULL) {
1632     patching_epilog(patch, patch_code, src, info);
1633   }
1634 
1635   if (info != NULL) add_debug_info_for_null_check(offset, info);
1636 }
1637 
1638 
1639 void LIR_Assembler::return_op(LIR_Opr result) {
1640   // the poll may need a register so just pick one that isn't the return register
1641 #if defined(TIERED) && !defined(_LP64)
1642   if (result->type_field() == LIR_OprDesc::long_type) {
1643     // Must move the result to G1
1644     // Must leave proper result in O0,O1 and G1 (TIERED only)
1645     __ sllx(I0, 32, G1);          // Shift bits into high G1
1646     __ srl (I1, 0, I1);           // Zero extend O1 (harmless?)
1647     __ or3 (I1, G1, G1);          // OR 64 bits into G1
1648 #ifdef ASSERT
1649     // mangle it so any problems will show up
1650     __ set(0xdeadbeef, I0);
1651     __ set(0xdeadbeef, I1);
1652 #endif
1653   }
1654 #endif // TIERED
1655   __ set((intptr_t)os::get_polling_page(), L0);
1656   __ relocate(relocInfo::poll_return_type);
1657   __ ld_ptr(L0, 0, G0);
1658   __ ret();
1659   __ delayed()->restore();
1660 }
1661 
1662 
1663 int LIR_Assembler::safepoint_poll(LIR_Opr tmp, CodeEmitInfo* info) {
1664   __ set((intptr_t)os::get_polling_page(), tmp->as_register());
1665   if (info != NULL) {
1666     add_debug_info_for_branch(info);
1667   } else {
1668     __ relocate(relocInfo::poll_type);
1669   }
1670 
1671   int offset = __ offset();
1672   __ ld_ptr(tmp->as_register(), 0, G0);
1673 
1674   return offset;
1675 }
1676 
1677 
1678 void LIR_Assembler::emit_static_call_stub() {
1679   address call_pc = __ pc();
1680   address stub = __ start_a_stub(call_stub_size);
1681   if (stub == NULL) {
1682     bailout("static call stub overflow");
1683     return;
1684   }
1685 
1686   int start = __ offset();
1687   __ relocate(static_stub_Relocation::spec(call_pc));
1688 
1689   __ set_oop(NULL, G5);
1690   // must be set to -1 at code generation time
1691   AddressLiteral addrlit(-1);
1692   __ jump_to(addrlit, G3);
1693   __ delayed()->nop();
1694 
1695   assert(__ offset() - start <= call_stub_size, "stub too big");
1696   __ end_a_stub();
1697 }
1698 
1699 
1700 void LIR_Assembler::comp_op(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Op2* op) {
1701   if (opr1->is_single_fpu()) {
1702     __ fcmp(FloatRegisterImpl::S, Assembler::fcc0, opr1->as_float_reg(), opr2->as_float_reg());
1703   } else if (opr1->is_double_fpu()) {
1704     __ fcmp(FloatRegisterImpl::D, Assembler::fcc0, opr1->as_double_reg(), opr2->as_double_reg());
1705   } else if (opr1->is_single_cpu()) {
1706     if (opr2->is_constant()) {
1707       switch (opr2->as_constant_ptr()->type()) {
1708         case T_INT:
1709           { jint con = opr2->as_constant_ptr()->as_jint();
1710             if (Assembler::is_simm13(con)) {
1711               __ cmp(opr1->as_register(), con);
1712             } else {
1713               __ set(con, O7);
1714               __ cmp(opr1->as_register(), O7);
1715             }
1716           }
1717           break;
1718 
1719         case T_OBJECT:
1720           // there are only equal/notequal comparisions on objects
1721           { jobject con = opr2->as_constant_ptr()->as_jobject();
1722             if (con == NULL) {
1723               __ cmp(opr1->as_register(), 0);
1724             } else {
1725               jobject2reg(con, O7);
1726               __ cmp(opr1->as_register(), O7);
1727             }
1728           }
1729           break;
1730 
1731         default:
1732           ShouldNotReachHere();
1733           break;
1734       }
1735     } else {
1736       if (opr2->is_address()) {
1737         LIR_Address * addr = opr2->as_address_ptr();
1738         BasicType type = addr->type();
1739         if ( type == T_OBJECT ) __ ld_ptr(as_Address(addr), O7);
1740         else                    __ ld(as_Address(addr), O7);
1741         __ cmp(opr1->as_register(), O7);
1742       } else {
1743         __ cmp(opr1->as_register(), opr2->as_register());
1744       }
1745     }
1746   } else if (opr1->is_double_cpu()) {
1747     Register xlo = opr1->as_register_lo();
1748     Register xhi = opr1->as_register_hi();
1749     if (opr2->is_constant() && opr2->as_jlong() == 0) {
1750       assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "only handles these cases");
1751 #ifdef _LP64
1752       __ orcc(xhi, G0, G0);
1753 #else
1754       __ orcc(xhi, xlo, G0);
1755 #endif
1756     } else if (opr2->is_register()) {
1757       Register ylo = opr2->as_register_lo();
1758       Register yhi = opr2->as_register_hi();
1759 #ifdef _LP64
1760       __ cmp(xlo, ylo);
1761 #else
1762       __ subcc(xlo, ylo, xlo);
1763       __ subccc(xhi, yhi, xhi);
1764       if (condition == lir_cond_equal || condition == lir_cond_notEqual) {
1765         __ orcc(xhi, xlo, G0);
1766       }
1767 #endif
1768     } else {
1769       ShouldNotReachHere();
1770     }
1771   } else if (opr1->is_address()) {
1772     LIR_Address * addr = opr1->as_address_ptr();
1773     BasicType type = addr->type();
1774     assert (opr2->is_constant(), "Checking");
1775     if ( type == T_OBJECT ) __ ld_ptr(as_Address(addr), O7);
1776     else                    __ ld(as_Address(addr), O7);
1777     __ cmp(O7, opr2->as_constant_ptr()->as_jint());
1778   } else {
1779     ShouldNotReachHere();
1780   }
1781 }
1782 
1783 
1784 void LIR_Assembler::comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst, LIR_Op2* op){
1785   if (code == lir_cmp_fd2i || code == lir_ucmp_fd2i) {
1786     bool is_unordered_less = (code == lir_ucmp_fd2i);
1787     if (left->is_single_fpu()) {
1788       __ float_cmp(true, is_unordered_less ? -1 : 1, left->as_float_reg(), right->as_float_reg(), dst->as_register());
1789     } else if (left->is_double_fpu()) {
1790       __ float_cmp(false, is_unordered_less ? -1 : 1, left->as_double_reg(), right->as_double_reg(), dst->as_register());
1791     } else {
1792       ShouldNotReachHere();
1793     }
1794   } else if (code == lir_cmp_l2i) {
1795 #ifdef _LP64
1796     __ lcmp(left->as_register_lo(), right->as_register_lo(), dst->as_register());
1797 #else
1798     __ lcmp(left->as_register_hi(),  left->as_register_lo(),
1799             right->as_register_hi(), right->as_register_lo(),
1800             dst->as_register());
1801 #endif
1802   } else {
1803     ShouldNotReachHere();
1804   }
1805 }
1806 
1807 
1808 void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result) {
1809 
1810   Assembler::Condition acond;
1811   switch (condition) {
1812     case lir_cond_equal:        acond = Assembler::equal;        break;
1813     case lir_cond_notEqual:     acond = Assembler::notEqual;     break;
1814     case lir_cond_less:         acond = Assembler::less;         break;
1815     case lir_cond_lessEqual:    acond = Assembler::lessEqual;    break;
1816     case lir_cond_greaterEqual: acond = Assembler::greaterEqual; break;
1817     case lir_cond_greater:      acond = Assembler::greater;      break;
1818     case lir_cond_aboveEqual:   acond = Assembler::greaterEqualUnsigned;      break;
1819     case lir_cond_belowEqual:   acond = Assembler::lessEqualUnsigned;      break;
1820     default:                         ShouldNotReachHere();
1821   };
1822 
1823   if (opr1->is_constant() && opr1->type() == T_INT) {
1824     Register dest = result->as_register();
1825     // load up first part of constant before branch
1826     // and do the rest in the delay slot.
1827     if (!Assembler::is_simm13(opr1->as_jint())) {
1828       __ sethi(opr1->as_jint(), dest);
1829     }
1830   } else if (opr1->is_constant()) {
1831     const2reg(opr1, result, lir_patch_none, NULL);
1832   } else if (opr1->is_register()) {
1833     reg2reg(opr1, result);
1834   } else if (opr1->is_stack()) {
1835     stack2reg(opr1, result, result->type());
1836   } else {
1837     ShouldNotReachHere();
1838   }
1839   Label skip;
1840   __ br(acond, false, Assembler::pt, skip);
1841   if (opr1->is_constant() && opr1->type() == T_INT) {
1842     Register dest = result->as_register();
1843     if (Assembler::is_simm13(opr1->as_jint())) {
1844       __ delayed()->or3(G0, opr1->as_jint(), dest);
1845     } else {
1846       // the sethi has been done above, so just put in the low 10 bits
1847       __ delayed()->or3(dest, opr1->as_jint() & 0x3ff, dest);
1848     }
1849   } else {
1850     // can't do anything useful in the delay slot
1851     __ delayed()->nop();
1852   }
1853   if (opr2->is_constant()) {
1854     const2reg(opr2, result, lir_patch_none, NULL);
1855   } else if (opr2->is_register()) {
1856     reg2reg(opr2, result);
1857   } else if (opr2->is_stack()) {
1858     stack2reg(opr2, result, result->type());
1859   } else {
1860     ShouldNotReachHere();
1861   }
1862   __ bind(skip);
1863 }
1864 
1865 
1866 void LIR_Assembler::arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest, CodeEmitInfo* info, bool pop_fpu_stack) {
1867   assert(info == NULL, "unused on this code path");
1868   assert(left->is_register(), "wrong items state");
1869   assert(dest->is_register(), "wrong items state");
1870 
1871   if (right->is_register()) {
1872     if (dest->is_float_kind()) {
1873 
1874       FloatRegister lreg, rreg, res;
1875       FloatRegisterImpl::Width w;
1876       if (right->is_single_fpu()) {
1877         w = FloatRegisterImpl::S;
1878         lreg = left->as_float_reg();
1879         rreg = right->as_float_reg();
1880         res  = dest->as_float_reg();
1881       } else {
1882         w = FloatRegisterImpl::D;
1883         lreg = left->as_double_reg();
1884         rreg = right->as_double_reg();
1885         res  = dest->as_double_reg();
1886       }
1887 
1888       switch (code) {
1889         case lir_add: __ fadd(w, lreg, rreg, res); break;
1890         case lir_sub: __ fsub(w, lreg, rreg, res); break;
1891         case lir_mul: // fall through
1892         case lir_mul_strictfp: __ fmul(w, lreg, rreg, res); break;
1893         case lir_div: // fall through
1894         case lir_div_strictfp: __ fdiv(w, lreg, rreg, res); break;
1895         default: ShouldNotReachHere();
1896       }
1897 
1898     } else if (dest->is_double_cpu()) {
1899 #ifdef _LP64
1900       Register dst_lo = dest->as_register_lo();
1901       Register op1_lo = left->as_pointer_register();
1902       Register op2_lo = right->as_pointer_register();
1903 
1904       switch (code) {
1905         case lir_add:
1906           __ add(op1_lo, op2_lo, dst_lo);
1907           break;
1908 
1909         case lir_sub:
1910           __ sub(op1_lo, op2_lo, dst_lo);
1911           break;
1912 
1913         default: ShouldNotReachHere();
1914       }
1915 #else
1916       Register op1_lo = left->as_register_lo();
1917       Register op1_hi = left->as_register_hi();
1918       Register op2_lo = right->as_register_lo();
1919       Register op2_hi = right->as_register_hi();
1920       Register dst_lo = dest->as_register_lo();
1921       Register dst_hi = dest->as_register_hi();
1922 
1923       switch (code) {
1924         case lir_add:
1925           __ addcc(op1_lo, op2_lo, dst_lo);
1926           __ addc (op1_hi, op2_hi, dst_hi);
1927           break;
1928 
1929         case lir_sub:
1930           __ subcc(op1_lo, op2_lo, dst_lo);
1931           __ subc (op1_hi, op2_hi, dst_hi);
1932           break;
1933 
1934         default: ShouldNotReachHere();
1935       }
1936 #endif
1937     } else {
1938       assert (right->is_single_cpu(), "Just Checking");
1939 
1940       Register lreg = left->as_register();
1941       Register res  = dest->as_register();
1942       Register rreg = right->as_register();
1943       switch (code) {
1944         case lir_add:  __ add  (lreg, rreg, res); break;
1945         case lir_sub:  __ sub  (lreg, rreg, res); break;
1946         case lir_mul:  __ mult (lreg, rreg, res); break;
1947         default: ShouldNotReachHere();
1948       }
1949     }
1950   } else {
1951     assert (right->is_constant(), "must be constant");
1952 
1953     if (dest->is_single_cpu()) {
1954       Register lreg = left->as_register();
1955       Register res  = dest->as_register();
1956       int    simm13 = right->as_constant_ptr()->as_jint();
1957 
1958       switch (code) {
1959         case lir_add:  __ add  (lreg, simm13, res); break;
1960         case lir_sub:  __ sub  (lreg, simm13, res); break;
1961         case lir_mul:  __ mult (lreg, simm13, res); break;
1962         default: ShouldNotReachHere();
1963       }
1964     } else {
1965       Register lreg = left->as_pointer_register();
1966       Register res  = dest->as_register_lo();
1967       long con = right->as_constant_ptr()->as_jlong();
1968       assert(Assembler::is_simm13(con), "must be simm13");
1969 
1970       switch (code) {
1971         case lir_add:  __ add  (lreg, (int)con, res); break;
1972         case lir_sub:  __ sub  (lreg, (int)con, res); break;
1973         case lir_mul:  __ mult (lreg, (int)con, res); break;
1974         default: ShouldNotReachHere();
1975       }
1976     }
1977   }
1978 }
1979 
1980 
1981 void LIR_Assembler::fpop() {
1982   // do nothing
1983 }
1984 
1985 
1986 void LIR_Assembler::intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr thread, LIR_Opr dest, LIR_Op* op) {
1987   switch (code) {
1988     case lir_sin:
1989     case lir_tan:
1990     case lir_cos: {
1991       assert(thread->is_valid(), "preserve the thread object for performance reasons");
1992       assert(dest->as_double_reg() == F0, "the result will be in f0/f1");
1993       break;
1994     }
1995     case lir_sqrt: {
1996       assert(!thread->is_valid(), "there is no need for a thread_reg for dsqrt");
1997       FloatRegister src_reg = value->as_double_reg();
1998       FloatRegister dst_reg = dest->as_double_reg();
1999       __ fsqrt(FloatRegisterImpl::D, src_reg, dst_reg);
2000       break;
2001     }
2002     case lir_abs: {
2003       assert(!thread->is_valid(), "there is no need for a thread_reg for fabs");
2004       FloatRegister src_reg = value->as_double_reg();
2005       FloatRegister dst_reg = dest->as_double_reg();
2006       __ fabs(FloatRegisterImpl::D, src_reg, dst_reg);
2007       break;
2008     }
2009     default: {
2010       ShouldNotReachHere();
2011       break;
2012     }
2013   }
2014 }
2015 
2016 
2017 void LIR_Assembler::logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest) {
2018   if (right->is_constant()) {
2019     if (dest->is_single_cpu()) {
2020       int simm13 = right->as_constant_ptr()->as_jint();
2021       switch (code) {
2022         case lir_logic_and:   __ and3 (left->as_register(), simm13, dest->as_register()); break;
2023         case lir_logic_or:    __ or3  (left->as_register(), simm13, dest->as_register()); break;
2024         case lir_logic_xor:   __ xor3 (left->as_register(), simm13, dest->as_register()); break;
2025         default: ShouldNotReachHere();
2026       }
2027     } else {
2028       long c = right->as_constant_ptr()->as_jlong();
2029       assert(c == (int)c && Assembler::is_simm13(c), "out of range");
2030       int simm13 = (int)c;
2031       switch (code) {
2032         case lir_logic_and:
2033 #ifndef _LP64
2034           __ and3 (left->as_register_hi(), 0,      dest->as_register_hi());
2035 #endif
2036           __ and3 (left->as_register_lo(), simm13, dest->as_register_lo());
2037           break;
2038 
2039         case lir_logic_or:
2040 #ifndef _LP64
2041           __ or3 (left->as_register_hi(), 0,      dest->as_register_hi());
2042 #endif
2043           __ or3 (left->as_register_lo(), simm13, dest->as_register_lo());
2044           break;
2045 
2046         case lir_logic_xor:
2047 #ifndef _LP64
2048           __ xor3 (left->as_register_hi(), 0,      dest->as_register_hi());
2049 #endif
2050           __ xor3 (left->as_register_lo(), simm13, dest->as_register_lo());
2051           break;
2052 
2053         default: ShouldNotReachHere();
2054       }
2055     }
2056   } else {
2057     assert(right->is_register(), "right should be in register");
2058 
2059     if (dest->is_single_cpu()) {
2060       switch (code) {
2061         case lir_logic_and:   __ and3 (left->as_register(), right->as_register(), dest->as_register()); break;
2062         case lir_logic_or:    __ or3  (left->as_register(), right->as_register(), dest->as_register()); break;
2063         case lir_logic_xor:   __ xor3 (left->as_register(), right->as_register(), dest->as_register()); break;
2064         default: ShouldNotReachHere();
2065       }
2066     } else {
2067 #ifdef _LP64
2068       Register l = (left->is_single_cpu() && left->is_oop_register()) ? left->as_register() :
2069                                                                         left->as_register_lo();
2070       Register r = (right->is_single_cpu() && right->is_oop_register()) ? right->as_register() :
2071                                                                           right->as_register_lo();
2072 
2073       switch (code) {
2074         case lir_logic_and: __ and3 (l, r, dest->as_register_lo()); break;
2075         case lir_logic_or:  __ or3  (l, r, dest->as_register_lo()); break;
2076         case lir_logic_xor: __ xor3 (l, r, dest->as_register_lo()); break;
2077         default: ShouldNotReachHere();
2078       }
2079 #else
2080       switch (code) {
2081         case lir_logic_and:
2082           __ and3 (left->as_register_hi(), right->as_register_hi(), dest->as_register_hi());
2083           __ and3 (left->as_register_lo(), right->as_register_lo(), dest->as_register_lo());
2084           break;
2085 
2086         case lir_logic_or:
2087           __ or3 (left->as_register_hi(), right->as_register_hi(), dest->as_register_hi());
2088           __ or3 (left->as_register_lo(), right->as_register_lo(), dest->as_register_lo());
2089           break;
2090 
2091         case lir_logic_xor:
2092           __ xor3 (left->as_register_hi(), right->as_register_hi(), dest->as_register_hi());
2093           __ xor3 (left->as_register_lo(), right->as_register_lo(), dest->as_register_lo());
2094           break;
2095 
2096         default: ShouldNotReachHere();
2097       }
2098 #endif
2099     }
2100   }
2101 }
2102 
2103 
2104 int LIR_Assembler::shift_amount(BasicType t) {
2105   int elem_size = type2aelembytes(t);
2106   switch (elem_size) {
2107     case 1 : return 0;
2108     case 2 : return 1;
2109     case 4 : return 2;
2110     case 8 : return 3;
2111   }
2112   ShouldNotReachHere();
2113   return -1;
2114 }
2115 
2116 
2117 void LIR_Assembler::throw_op(LIR_Opr exceptionPC, LIR_Opr exceptionOop, CodeEmitInfo* info) {
2118   assert(exceptionOop->as_register() == Oexception, "should match");
2119   assert(exceptionPC->as_register() == Oissuing_pc, "should match");
2120 
2121   info->add_register_oop(exceptionOop);
2122 
2123   // reuse the debug info from the safepoint poll for the throw op itself
2124   address pc_for_athrow  = __ pc();
2125   int pc_for_athrow_offset = __ offset();
2126   RelocationHolder rspec = internal_word_Relocation::spec(pc_for_athrow);
2127   __ set(pc_for_athrow, Oissuing_pc, rspec);
2128   add_call_info(pc_for_athrow_offset, info); // for exception handler
2129 
2130   __ call(Runtime1::entry_for(Runtime1::handle_exception_id), relocInfo::runtime_call_type);
2131   __ delayed()->nop();
2132 }
2133 
2134 
2135 void LIR_Assembler::unwind_op(LIR_Opr exceptionOop) {
2136   assert(exceptionOop->as_register() == Oexception, "should match");
2137 
2138   __ br(Assembler::always, false, Assembler::pt, _unwind_handler_entry);
2139   __ delayed()->nop();
2140 }
2141 
2142 
2143 void LIR_Assembler::emit_arraycopy(LIR_OpArrayCopy* op) {
2144   Register src = op->src()->as_register();
2145   Register dst = op->dst()->as_register();
2146   Register src_pos = op->src_pos()->as_register();
2147   Register dst_pos = op->dst_pos()->as_register();
2148   Register length  = op->length()->as_register();
2149   Register tmp = op->tmp()->as_register();
2150   Register tmp2 = O7;
2151 
2152   int flags = op->flags();
2153   ciArrayKlass* default_type = op->expected_type();
2154   BasicType basic_type = default_type != NULL ? default_type->element_type()->basic_type() : T_ILLEGAL;
2155   if (basic_type == T_ARRAY) basic_type = T_OBJECT;
2156 
2157   // set up the arraycopy stub information
2158   ArrayCopyStub* stub = op->stub();
2159 
2160   // always do stub if no type information is available.  it's ok if
2161   // the known type isn't loaded since the code sanity checks
2162   // in debug mode and the type isn't required when we know the exact type
2163   // also check that the type is an array type.
2164   // We also, for now, always call the stub if the barrier set requires a
2165   // write_ref_pre barrier (which the stub does, but none of the optimized
2166   // cases currently does).
2167   if (op->expected_type() == NULL ||
2168       Universe::heap()->barrier_set()->has_write_ref_pre_barrier()) {
2169     __ mov(src,     O0);
2170     __ mov(src_pos, O1);
2171     __ mov(dst,     O2);
2172     __ mov(dst_pos, O3);
2173     __ mov(length,  O4);
2174     __ call_VM_leaf(tmp, CAST_FROM_FN_PTR(address, Runtime1::arraycopy));
2175 
2176     __ br_zero(Assembler::less, false, Assembler::pn, O0, *stub->entry());
2177     __ delayed()->nop();
2178     __ bind(*stub->continuation());
2179     return;
2180   }
2181 
2182   assert(default_type != NULL && default_type->is_array_klass(), "must be true at this point");
2183 
2184   // make sure src and dst are non-null and load array length
2185   if (flags & LIR_OpArrayCopy::src_null_check) {
2186     __ tst(src);
2187     __ br(Assembler::equal, false, Assembler::pn, *stub->entry());
2188     __ delayed()->nop();
2189   }
2190 
2191   if (flags & LIR_OpArrayCopy::dst_null_check) {
2192     __ tst(dst);
2193     __ br(Assembler::equal, false, Assembler::pn, *stub->entry());
2194     __ delayed()->nop();
2195   }
2196 
2197   if (flags & LIR_OpArrayCopy::src_pos_positive_check) {
2198     // test src_pos register
2199     __ tst(src_pos);
2200     __ br(Assembler::less, false, Assembler::pn, *stub->entry());
2201     __ delayed()->nop();
2202   }
2203 
2204   if (flags & LIR_OpArrayCopy::dst_pos_positive_check) {
2205     // test dst_pos register
2206     __ tst(dst_pos);
2207     __ br(Assembler::less, false, Assembler::pn, *stub->entry());
2208     __ delayed()->nop();
2209   }
2210 
2211   if (flags & LIR_OpArrayCopy::length_positive_check) {
2212     // make sure length isn't negative
2213     __ tst(length);
2214     __ br(Assembler::less, false, Assembler::pn, *stub->entry());
2215     __ delayed()->nop();
2216   }
2217 
2218   if (flags & LIR_OpArrayCopy::src_range_check) {
2219     __ ld(src, arrayOopDesc::length_offset_in_bytes(), tmp2);
2220     __ add(length, src_pos, tmp);
2221     __ cmp(tmp2, tmp);
2222     __ br(Assembler::carrySet, false, Assembler::pn, *stub->entry());
2223     __ delayed()->nop();
2224   }
2225 
2226   if (flags & LIR_OpArrayCopy::dst_range_check) {
2227     __ ld(dst, arrayOopDesc::length_offset_in_bytes(), tmp2);
2228     __ add(length, dst_pos, tmp);
2229     __ cmp(tmp2, tmp);
2230     __ br(Assembler::carrySet, false, Assembler::pn, *stub->entry());
2231     __ delayed()->nop();
2232   }
2233 
2234   if (flags & LIR_OpArrayCopy::type_check) {
2235     __ ld_ptr(src, oopDesc::klass_offset_in_bytes(), tmp);
2236     __ ld_ptr(dst, oopDesc::klass_offset_in_bytes(), tmp2);
2237     __ cmp(tmp, tmp2);
2238     __ br(Assembler::notEqual, false, Assembler::pt, *stub->entry());
2239     __ delayed()->nop();
2240   }
2241 
2242 #ifdef ASSERT
2243   if (basic_type != T_OBJECT || !(flags & LIR_OpArrayCopy::type_check)) {
2244     // Sanity check the known type with the incoming class.  For the
2245     // primitive case the types must match exactly with src.klass and
2246     // dst.klass each exactly matching the default type.  For the
2247     // object array case, if no type check is needed then either the
2248     // dst type is exactly the expected type and the src type is a
2249     // subtype which we can't check or src is the same array as dst
2250     // but not necessarily exactly of type default_type.
2251     Label known_ok, halt;
2252     jobject2reg(op->expected_type()->constant_encoding(), tmp);
2253     __ ld_ptr(dst, oopDesc::klass_offset_in_bytes(), tmp2);
2254     if (basic_type != T_OBJECT) {
2255       __ cmp(tmp, tmp2);
2256       __ br(Assembler::notEqual, false, Assembler::pn, halt);
2257       __ delayed()->ld_ptr(src, oopDesc::klass_offset_in_bytes(), tmp2);
2258       __ cmp(tmp, tmp2);
2259       __ br(Assembler::equal, false, Assembler::pn, known_ok);
2260       __ delayed()->nop();
2261     } else {
2262       __ cmp(tmp, tmp2);
2263       __ br(Assembler::equal, false, Assembler::pn, known_ok);
2264       __ delayed()->cmp(src, dst);
2265       __ br(Assembler::equal, false, Assembler::pn, known_ok);
2266       __ delayed()->nop();
2267     }
2268     __ bind(halt);
2269     __ stop("incorrect type information in arraycopy");
2270     __ bind(known_ok);
2271   }
2272 #endif
2273 
2274   int shift = shift_amount(basic_type);
2275 
2276   Register src_ptr = O0;
2277   Register dst_ptr = O1;
2278   Register len     = O2;
2279 
2280   __ add(src, arrayOopDesc::base_offset_in_bytes(basic_type), src_ptr);
2281   LP64_ONLY(__ sra(src_pos, 0, src_pos);) //higher 32bits must be null
2282   if (shift == 0) {
2283     __ add(src_ptr, src_pos, src_ptr);
2284   } else {
2285     __ sll(src_pos, shift, tmp);
2286     __ add(src_ptr, tmp, src_ptr);
2287   }
2288 
2289   __ add(dst, arrayOopDesc::base_offset_in_bytes(basic_type), dst_ptr);
2290   LP64_ONLY(__ sra(dst_pos, 0, dst_pos);) //higher 32bits must be null
2291   if (shift == 0) {
2292     __ add(dst_ptr, dst_pos, dst_ptr);
2293   } else {
2294     __ sll(dst_pos, shift, tmp);
2295     __ add(dst_ptr, tmp, dst_ptr);
2296   }
2297 
2298   if (basic_type != T_OBJECT) {
2299     if (shift == 0) {
2300       __ mov(length, len);
2301     } else {
2302       __ sll(length, shift, len);
2303     }
2304     __ call_VM_leaf(tmp, CAST_FROM_FN_PTR(address, Runtime1::primitive_arraycopy));
2305   } else {
2306     // oop_arraycopy takes a length in number of elements, so don't scale it.
2307     __ mov(length, len);
2308     __ call_VM_leaf(tmp, CAST_FROM_FN_PTR(address, Runtime1::oop_arraycopy));
2309   }
2310 
2311   __ bind(*stub->continuation());
2312 }
2313 
2314 
2315 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
2316   if (dest->is_single_cpu()) {
2317 #ifdef _LP64
2318     if (left->type() == T_OBJECT) {
2319       switch (code) {
2320         case lir_shl:  __ sllx  (left->as_register(), count->as_register(), dest->as_register()); break;
2321         case lir_shr:  __ srax  (left->as_register(), count->as_register(), dest->as_register()); break;
2322         case lir_ushr: __ srl   (left->as_register(), count->as_register(), dest->as_register()); break;
2323         default: ShouldNotReachHere();
2324       }
2325     } else
2326 #endif
2327       switch (code) {
2328         case lir_shl:  __ sll   (left->as_register(), count->as_register(), dest->as_register()); break;
2329         case lir_shr:  __ sra   (left->as_register(), count->as_register(), dest->as_register()); break;
2330         case lir_ushr: __ srl   (left->as_register(), count->as_register(), dest->as_register()); break;
2331         default: ShouldNotReachHere();
2332       }
2333   } else {
2334 #ifdef _LP64
2335     switch (code) {
2336       case lir_shl:  __ sllx  (left->as_register_lo(), count->as_register(), dest->as_register_lo()); break;
2337       case lir_shr:  __ srax  (left->as_register_lo(), count->as_register(), dest->as_register_lo()); break;
2338       case lir_ushr: __ srlx  (left->as_register_lo(), count->as_register(), dest->as_register_lo()); break;
2339       default: ShouldNotReachHere();
2340     }
2341 #else
2342     switch (code) {
2343       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;
2344       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;
2345       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;
2346       default: ShouldNotReachHere();
2347     }
2348 #endif
2349   }
2350 }
2351 
2352 
2353 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, jint count, LIR_Opr dest) {
2354 #ifdef _LP64
2355   if (left->type() == T_OBJECT) {
2356     count = count & 63;  // shouldn't shift by more than sizeof(intptr_t)
2357     Register l = left->as_register();
2358     Register d = dest->as_register_lo();
2359     switch (code) {
2360       case lir_shl:  __ sllx  (l, count, d); break;
2361       case lir_shr:  __ srax  (l, count, d); break;
2362       case lir_ushr: __ srlx  (l, count, d); break;
2363       default: ShouldNotReachHere();
2364     }
2365     return;
2366   }
2367 #endif
2368 
2369   if (dest->is_single_cpu()) {
2370     count = count & 0x1F; // Java spec
2371     switch (code) {
2372       case lir_shl:  __ sll   (left->as_register(), count, dest->as_register()); break;
2373       case lir_shr:  __ sra   (left->as_register(), count, dest->as_register()); break;
2374       case lir_ushr: __ srl   (left->as_register(), count, dest->as_register()); break;
2375       default: ShouldNotReachHere();
2376     }
2377   } else if (dest->is_double_cpu()) {
2378     count = count & 63; // Java spec
2379     switch (code) {
2380       case lir_shl:  __ sllx  (left->as_pointer_register(), count, dest->as_pointer_register()); break;
2381       case lir_shr:  __ srax  (left->as_pointer_register(), count, dest->as_pointer_register()); break;
2382       case lir_ushr: __ srlx  (left->as_pointer_register(), count, dest->as_pointer_register()); break;
2383       default: ShouldNotReachHere();
2384     }
2385   } else {
2386     ShouldNotReachHere();
2387   }
2388 }
2389 
2390 
2391 void LIR_Assembler::emit_alloc_obj(LIR_OpAllocObj* op) {
2392   assert(op->tmp1()->as_register()  == G1 &&
2393          op->tmp2()->as_register()  == G3 &&
2394          op->tmp3()->as_register()  == G4 &&
2395          op->obj()->as_register()   == O0 &&
2396          op->klass()->as_register() == G5, "must be");
2397   if (op->init_check()) {
2398     __ ld(op->klass()->as_register(),
2399           instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc),
2400           op->tmp1()->as_register());
2401     add_debug_info_for_null_check_here(op->stub()->info());
2402     __ cmp(op->tmp1()->as_register(), instanceKlass::fully_initialized);
2403     __ br(Assembler::notEqual, false, Assembler::pn, *op->stub()->entry());
2404     __ delayed()->nop();
2405   }
2406   __ allocate_object(op->obj()->as_register(),
2407                      op->tmp1()->as_register(),
2408                      op->tmp2()->as_register(),
2409                      op->tmp3()->as_register(),
2410                      op->header_size(),
2411                      op->object_size(),
2412                      op->klass()->as_register(),
2413                      *op->stub()->entry());
2414   __ bind(*op->stub()->continuation());
2415   __ verify_oop(op->obj()->as_register());
2416 }
2417 
2418 
2419 void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
2420   assert(op->tmp1()->as_register()  == G1 &&
2421          op->tmp2()->as_register()  == G3 &&
2422          op->tmp3()->as_register()  == G4 &&
2423          op->tmp4()->as_register()  == O1 &&
2424          op->klass()->as_register() == G5, "must be");
2425   if (UseSlowPath ||
2426       (!UseFastNewObjectArray && (op->type() == T_OBJECT || op->type() == T_ARRAY)) ||
2427       (!UseFastNewTypeArray   && (op->type() != T_OBJECT && op->type() != T_ARRAY))) {
2428     __ br(Assembler::always, false, Assembler::pt, *op->stub()->entry());
2429     __ delayed()->nop();
2430   } else {
2431     __ allocate_array(op->obj()->as_register(),
2432                       op->len()->as_register(),
2433                       op->tmp1()->as_register(),
2434                       op->tmp2()->as_register(),
2435                       op->tmp3()->as_register(),
2436                       arrayOopDesc::header_size(op->type()),
2437                       type2aelembytes(op->type()),
2438                       op->klass()->as_register(),
2439                       *op->stub()->entry());
2440   }
2441   __ bind(*op->stub()->continuation());
2442 }
2443 
2444 
2445 void LIR_Assembler::type_profile_helper(Register mdo, int mdo_offset_bias,
2446                                         ciMethodData *md, ciProfileData *data,
2447                                         Register recv, Register tmp1, Label* update_done) {
2448   uint i;
2449   for (i = 0; i < VirtualCallData::row_limit(); i++) {
2450     Label next_test;
2451     // See if the receiver is receiver[n].
2452     Address receiver_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)) -
2453                           mdo_offset_bias);
2454     __ ld_ptr(receiver_addr, tmp1);
2455     __ verify_oop(tmp1);
2456     __ cmp(recv, tmp1);
2457     __ brx(Assembler::notEqual, false, Assembler::pt, next_test);
2458     __ delayed()->nop();
2459     Address data_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)) -
2460                       mdo_offset_bias);
2461     __ ld_ptr(data_addr, tmp1);
2462     __ add(tmp1, DataLayout::counter_increment, tmp1);
2463     __ st_ptr(tmp1, data_addr);
2464     __ ba(false, *update_done);
2465     __ delayed()->nop();
2466     __ bind(next_test);
2467   }
2468 
2469   // Didn't find receiver; find next empty slot and fill it in
2470   for (i = 0; i < VirtualCallData::row_limit(); i++) {
2471     Label next_test;
2472     Address recv_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)) -
2473                       mdo_offset_bias);
2474     load(recv_addr, tmp1, T_OBJECT);
2475     __ br_notnull(tmp1, false, Assembler::pt, next_test);
2476     __ delayed()->nop();
2477     __ st_ptr(recv, recv_addr);
2478     __ set(DataLayout::counter_increment, tmp1);
2479     __ st_ptr(tmp1, mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)) -
2480               mdo_offset_bias);
2481     __ ba(false, *update_done);
2482     __ delayed()->nop();
2483     __ bind(next_test);
2484   }
2485 }
2486 
2487 
2488 void LIR_Assembler::setup_md_access(ciMethod* method, int bci,
2489                                     ciMethodData*& md, ciProfileData*& data, int& mdo_offset_bias) {
2490   md = method->method_data();
2491   if (md == NULL) {
2492     bailout("out of memory building methodDataOop");
2493     return;
2494   }
2495   data = md->bci_to_data(bci);
2496   assert(data != NULL,       "need data for checkcast");
2497   assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
2498   if (!Assembler::is_simm13(md->byte_offset_of_slot(data, DataLayout::header_offset()) + data->size_in_bytes())) {
2499     // The offset is large so bias the mdo by the base of the slot so
2500     // that the ld can use simm13s to reference the slots of the data
2501     mdo_offset_bias = md->byte_offset_of_slot(data, DataLayout::header_offset());
2502   }
2503 }
2504 
2505 void LIR_Assembler::emit_typecheck_helper(LIR_OpTypeCheck *op, Label* success, Label* failure, Label* obj_is_null) {
2506   // we always need a stub for the failure case.
2507   CodeStub* stub = op->stub();
2508   Register obj = op->object()->as_register();
2509   Register k_RInfo = op->tmp1()->as_register();
2510   Register klass_RInfo = op->tmp2()->as_register();
2511   Register dst = op->result_opr()->as_register();
2512   Register Rtmp1 = op->tmp3()->as_register();
2513   ciKlass* k = op->klass();
2514 
2515 
2516   if (obj == k_RInfo) {
2517     k_RInfo = klass_RInfo;
2518     klass_RInfo = obj;
2519   }
2520 
2521   ciMethodData* md;
2522   ciProfileData* data;
2523   int mdo_offset_bias = 0;
2524   if (op->should_profile()) {
2525     ciMethod* method = op->profiled_method();
2526     assert(method != NULL, "Should have method");
2527     setup_md_access(method, op->profiled_bci(), md, data, mdo_offset_bias);
2528 
2529     Label not_null;
2530     __ br_notnull(obj, false, Assembler::pn, not_null);
2531     __ delayed()->nop();
2532     Register mdo      = k_RInfo;
2533     Register data_val = Rtmp1;
2534     jobject2reg(md->constant_encoding(), mdo);
2535     if (mdo_offset_bias > 0) {
2536       __ set(mdo_offset_bias, data_val);
2537       __ add(mdo, data_val, mdo);
2538     }
2539     Address flags_addr(mdo, md->byte_offset_of_slot(data, DataLayout::flags_offset()) - mdo_offset_bias);
2540     __ ldub(flags_addr, data_val);
2541     __ or3(data_val, BitData::null_seen_byte_constant(), data_val);
2542     __ stb(data_val, flags_addr);
2543     __ ba(false, *obj_is_null);
2544     __ delayed()->nop();
2545     __ bind(not_null);
2546   } else {
2547     __ br_null(obj, false, Assembler::pn, *obj_is_null);
2548     __ delayed()->nop();
2549   }
2550 
2551   Label profile_cast_failure, profile_cast_success;
2552   Label *failure_target = op->should_profile() ? &profile_cast_failure : failure;
2553   Label *success_target = op->should_profile() ? &profile_cast_success : success;
2554 
2555   // patching may screw with our temporaries on sparc,
2556   // so let's do it before loading the class
2557   if (k->is_loaded()) {
2558     jobject2reg(k->constant_encoding(), k_RInfo);
2559   } else {
2560     jobject2reg_with_patching(k_RInfo, op->info_for_patch());
2561   }
2562   assert(obj != k_RInfo, "must be different");
2563 
2564   // get object class
2565   // not a safepoint as obj null check happens earlier
2566   load(obj, oopDesc::klass_offset_in_bytes(), klass_RInfo, T_OBJECT, NULL);
2567   if (op->fast_check()) {
2568     assert_different_registers(klass_RInfo, k_RInfo);
2569     __ cmp(k_RInfo, klass_RInfo);
2570     __ brx(Assembler::notEqual, false, Assembler::pt, *failure_target);
2571     __ delayed()->nop();
2572   } else {
2573     bool need_slow_path = true;
2574     if (k->is_loaded()) {
2575       if (k->super_check_offset() != sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes())
2576         need_slow_path = false;
2577       // perform the fast part of the checking logic
2578       __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, noreg,
2579                                        (need_slow_path ? success_target : NULL),
2580                                        failure_target, NULL,
2581                                        RegisterOrConstant(k->super_check_offset()));
2582     } else {
2583       // perform the fast part of the checking logic
2584       __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, O7, success_target,
2585                                        failure_target, NULL);
2586     }
2587     if (need_slow_path) {
2588       // call out-of-line instance of __ check_klass_subtype_slow_path(...):
2589       assert(klass_RInfo == G3 && k_RInfo == G1, "incorrect call setup");
2590       __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type);
2591       __ delayed()->nop();
2592       __ cmp(G3, 0);
2593       __ br(Assembler::equal, false, Assembler::pn, *failure_target);
2594       __ delayed()->nop();
2595       // Fall through to success case
2596     }
2597   }
2598 
2599   if (op->should_profile()) {
2600     Register mdo  = klass_RInfo, recv = k_RInfo, tmp1 = Rtmp1;
2601     assert_different_registers(obj, mdo, recv, tmp1);
2602     __ bind(profile_cast_success);
2603     jobject2reg(md->constant_encoding(), mdo);
2604     if (mdo_offset_bias > 0) {
2605       __ set(mdo_offset_bias, tmp1);
2606       __ add(mdo, tmp1, mdo);
2607     }
2608     load(Address(obj, oopDesc::klass_offset_in_bytes()), recv, T_OBJECT);
2609     type_profile_helper(mdo, mdo_offset_bias, md, data, recv, tmp1, success);
2610     // Jump over the failure case
2611     __ ba(false, *success);
2612     __ delayed()->nop();
2613     // Cast failure case
2614     __ bind(profile_cast_failure);
2615     jobject2reg(md->constant_encoding(), mdo);
2616     if (mdo_offset_bias > 0) {
2617       __ set(mdo_offset_bias, tmp1);
2618       __ add(mdo, tmp1, mdo);
2619     }
2620     Address data_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()) - mdo_offset_bias);
2621     __ ld_ptr(data_addr, tmp1);
2622     __ sub(tmp1, DataLayout::counter_increment, tmp1);
2623     __ st_ptr(tmp1, data_addr);
2624     __ ba(false, *failure);
2625     __ delayed()->nop();
2626   }
2627   __ ba(false, *success);
2628   __ delayed()->nop();
2629 }
2630 
2631 void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
2632   LIR_Code code = op->code();
2633   if (code == lir_store_check) {
2634     Register value = op->object()->as_register();
2635     Register array = op->array()->as_register();
2636     Register k_RInfo = op->tmp1()->as_register();
2637     Register klass_RInfo = op->tmp2()->as_register();
2638     Register Rtmp1 = op->tmp3()->as_register();
2639 
2640     __ verify_oop(value);
2641     CodeStub* stub = op->stub();
2642     // check if it needs to be profiled
2643     ciMethodData* md;
2644     ciProfileData* data;
2645     int mdo_offset_bias = 0;
2646     if (op->should_profile()) {
2647       ciMethod* method = op->profiled_method();
2648       assert(method != NULL, "Should have method");
2649       setup_md_access(method, op->profiled_bci(), md, data, mdo_offset_bias);
2650     }
2651     Label profile_cast_success, profile_cast_failure, done;
2652     Label *success_target = op->should_profile() ? &profile_cast_success : &done;
2653     Label *failure_target = op->should_profile() ? &profile_cast_failure : stub->entry();
2654 
2655     if (op->should_profile()) {
2656       Label not_null;
2657       __ br_notnull(value, false, Assembler::pn, not_null);
2658       __ delayed()->nop();
2659       Register mdo      = k_RInfo;
2660       Register data_val = Rtmp1;
2661       jobject2reg(md->constant_encoding(), mdo);
2662       if (mdo_offset_bias > 0) {
2663         __ set(mdo_offset_bias, data_val);
2664         __ add(mdo, data_val, mdo);
2665       }
2666       Address flags_addr(mdo, md->byte_offset_of_slot(data, DataLayout::flags_offset()) - mdo_offset_bias);
2667       __ ldub(flags_addr, data_val);
2668       __ or3(data_val, BitData::null_seen_byte_constant(), data_val);
2669       __ stb(data_val, flags_addr);
2670       __ ba(false, done);
2671       __ delayed()->nop();
2672       __ bind(not_null);
2673     } else {
2674       __ br_null(value, false, Assembler::pn, done);
2675       __ delayed()->nop();
2676     }
2677     load(array, oopDesc::klass_offset_in_bytes(), k_RInfo, T_OBJECT, op->info_for_exception());
2678     load(value, oopDesc::klass_offset_in_bytes(), klass_RInfo, T_OBJECT, NULL);
2679 
2680     // get instance klass
2681     load(k_RInfo, objArrayKlass::element_klass_offset_in_bytes() + sizeof(oopDesc), k_RInfo, T_OBJECT, NULL);
2682     // perform the fast part of the checking logic
2683     __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, O7, success_target, failure_target, NULL);
2684 
2685     // call out-of-line instance of __ check_klass_subtype_slow_path(...):
2686     assert(klass_RInfo == G3 && k_RInfo == G1, "incorrect call setup");
2687     __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type);
2688     __ delayed()->nop();
2689     __ cmp(G3, 0);
2690     __ br(Assembler::equal, false, Assembler::pn, *failure_target);
2691     __ delayed()->nop();
2692     // fall through to the success case
2693 
2694     if (op->should_profile()) {
2695       Register mdo  = klass_RInfo, recv = k_RInfo, tmp1 = Rtmp1;
2696       assert_different_registers(value, mdo, recv, tmp1);
2697       __ bind(profile_cast_success);
2698       jobject2reg(md->constant_encoding(), mdo);
2699       if (mdo_offset_bias > 0) {
2700         __ set(mdo_offset_bias, tmp1);
2701         __ add(mdo, tmp1, mdo);
2702       }
2703       load(Address(value, oopDesc::klass_offset_in_bytes()), recv, T_OBJECT);
2704       type_profile_helper(mdo, mdo_offset_bias, md, data, recv, tmp1, &done);
2705       __ ba(false, done);
2706       __ delayed()->nop();
2707       // Cast failure case
2708       __ bind(profile_cast_failure);
2709       jobject2reg(md->constant_encoding(), mdo);
2710       if (mdo_offset_bias > 0) {
2711         __ set(mdo_offset_bias, tmp1);
2712         __ add(mdo, tmp1, mdo);
2713       }
2714       Address data_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()) - mdo_offset_bias);
2715       __ ld_ptr(data_addr, tmp1);
2716       __ sub(tmp1, DataLayout::counter_increment, tmp1);
2717       __ st_ptr(tmp1, data_addr);
2718       __ ba(false, *stub->entry());
2719       __ delayed()->nop();
2720     }
2721     __ bind(done);
2722   } else if (code == lir_checkcast) {
2723     Register obj = op->object()->as_register();
2724     Register dst = op->result_opr()->as_register();
2725     Label success;
2726     emit_typecheck_helper(op, &success, op->stub()->entry(), &success);
2727     __ bind(success);
2728     __ mov(obj, dst);
2729   } else if (code == lir_instanceof) {
2730     Register obj = op->object()->as_register();
2731     Register dst = op->result_opr()->as_register();
2732     Label success, failure, done;
2733     emit_typecheck_helper(op, &success, &failure, &failure);
2734     __ bind(failure);
2735     __ set(0, dst);
2736     __ ba(false, done);
2737     __ delayed()->nop();
2738     __ bind(success);
2739     __ set(1, dst);
2740     __ bind(done);
2741   } else {
2742     ShouldNotReachHere();
2743   }
2744 
2745 }
2746 
2747 
2748 void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
2749   if (op->code() == lir_cas_long) {
2750     assert(VM_Version::supports_cx8(), "wrong machine");
2751     Register addr = op->addr()->as_pointer_register();
2752     Register cmp_value_lo = op->cmp_value()->as_register_lo();
2753     Register cmp_value_hi = op->cmp_value()->as_register_hi();
2754     Register new_value_lo = op->new_value()->as_register_lo();
2755     Register new_value_hi = op->new_value()->as_register_hi();
2756     Register t1 = op->tmp1()->as_register();
2757     Register t2 = op->tmp2()->as_register();
2758 #ifdef _LP64
2759     __ mov(cmp_value_lo, t1);
2760     __ mov(new_value_lo, t2);
2761 #else
2762     // move high and low halves of long values into single registers
2763     __ sllx(cmp_value_hi, 32, t1);         // shift high half into temp reg
2764     __ srl(cmp_value_lo, 0, cmp_value_lo); // clear upper 32 bits of low half
2765     __ or3(t1, cmp_value_lo, t1);          // t1 holds 64-bit compare value
2766     __ sllx(new_value_hi, 32, t2);
2767     __ srl(new_value_lo, 0, new_value_lo);
2768     __ or3(t2, new_value_lo, t2);          // t2 holds 64-bit value to swap
2769 #endif
2770     // perform the compare and swap operation
2771     __ casx(addr, t1, t2);
2772     // generate condition code - if the swap succeeded, t2 ("new value" reg) was
2773     // overwritten with the original value in "addr" and will be equal to t1.
2774     __ cmp(t1, t2);
2775 
2776   } else if (op->code() == lir_cas_int || op->code() == lir_cas_obj) {
2777     Register addr = op->addr()->as_pointer_register();
2778     Register cmp_value = op->cmp_value()->as_register();
2779     Register new_value = op->new_value()->as_register();
2780     Register t1 = op->tmp1()->as_register();
2781     Register t2 = op->tmp2()->as_register();
2782     __ mov(cmp_value, t1);
2783     __ mov(new_value, t2);
2784 #ifdef _LP64
2785     if (op->code() == lir_cas_obj) {
2786       __ casx(addr, t1, t2);
2787     } else
2788 #endif
2789       {
2790         __ cas(addr, t1, t2);
2791       }
2792     __ cmp(t1, t2);
2793   } else {
2794     Unimplemented();
2795   }
2796 }
2797 
2798 void LIR_Assembler::set_24bit_FPU() {
2799   Unimplemented();
2800 }
2801 
2802 
2803 void LIR_Assembler::reset_FPU() {
2804   Unimplemented();
2805 }
2806 
2807 
2808 void LIR_Assembler::breakpoint() {
2809   __ breakpoint_trap();
2810 }
2811 
2812 
2813 void LIR_Assembler::push(LIR_Opr opr) {
2814   Unimplemented();
2815 }
2816 
2817 
2818 void LIR_Assembler::pop(LIR_Opr opr) {
2819   Unimplemented();
2820 }
2821 
2822 
2823 void LIR_Assembler::monitor_address(int monitor_no, LIR_Opr dst_opr) {
2824   Address mon_addr = frame_map()->address_for_monitor_lock(monitor_no);
2825   Register dst = dst_opr->as_register();
2826   Register reg = mon_addr.base();
2827   int offset = mon_addr.disp();
2828   // compute pointer to BasicLock
2829   if (mon_addr.is_simm13()) {
2830     __ add(reg, offset, dst);
2831   } else {
2832     __ set(offset, dst);
2833     __ add(dst, reg, dst);
2834   }
2835 }
2836 
2837 
2838 void LIR_Assembler::emit_lock(LIR_OpLock* op) {
2839   Register obj = op->obj_opr()->as_register();
2840   Register hdr = op->hdr_opr()->as_register();
2841   Register lock = op->lock_opr()->as_register();
2842 
2843   // obj may not be an oop
2844   if (op->code() == lir_lock) {
2845     MonitorEnterStub* stub = (MonitorEnterStub*)op->stub();
2846     if (UseFastLocking) {
2847       assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
2848       // add debug info for NullPointerException only if one is possible
2849       if (op->info() != NULL) {
2850         add_debug_info_for_null_check_here(op->info());
2851       }
2852       __ lock_object(hdr, obj, lock, op->scratch_opr()->as_register(), *op->stub()->entry());
2853     } else {
2854       // always do slow locking
2855       // note: the slow locking code could be inlined here, however if we use
2856       //       slow locking, speed doesn't matter anyway and this solution is
2857       //       simpler and requires less duplicated code - additionally, the
2858       //       slow locking code is the same in either case which simplifies
2859       //       debugging
2860       __ br(Assembler::always, false, Assembler::pt, *op->stub()->entry());
2861       __ delayed()->nop();
2862     }
2863   } else {
2864     assert (op->code() == lir_unlock, "Invalid code, expected lir_unlock");
2865     if (UseFastLocking) {
2866       assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
2867       __ unlock_object(hdr, obj, lock, *op->stub()->entry());
2868     } else {
2869       // always do slow unlocking
2870       // note: the slow unlocking code could be inlined here, however if we use
2871       //       slow unlocking, speed doesn't matter anyway and this solution is
2872       //       simpler and requires less duplicated code - additionally, the
2873       //       slow unlocking code is the same in either case which simplifies
2874       //       debugging
2875       __ br(Assembler::always, false, Assembler::pt, *op->stub()->entry());
2876       __ delayed()->nop();
2877     }
2878   }
2879   __ bind(*op->stub()->continuation());
2880 }
2881 
2882 
2883 void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
2884   ciMethod* method = op->profiled_method();
2885   int bci          = op->profiled_bci();
2886 
2887   // Update counter for all call types
2888   ciMethodData* md = method->method_data();
2889   if (md == NULL) {
2890     bailout("out of memory building methodDataOop");
2891     return;
2892   }
2893   ciProfileData* data = md->bci_to_data(bci);
2894   assert(data->is_CounterData(), "need CounterData for calls");
2895   assert(op->mdo()->is_single_cpu(),  "mdo must be allocated");
2896   Register mdo  = op->mdo()->as_register();
2897 #ifdef _LP64
2898   assert(op->tmp1()->is_double_cpu(), "tmp1 must be allocated");
2899   Register tmp1 = op->tmp1()->as_register_lo();
2900 #else
2901   assert(op->tmp1()->is_single_cpu(), "tmp1 must be allocated");
2902   Register tmp1 = op->tmp1()->as_register();
2903 #endif
2904   jobject2reg(md->constant_encoding(), mdo);
2905   int mdo_offset_bias = 0;
2906   if (!Assembler::is_simm13(md->byte_offset_of_slot(data, CounterData::count_offset()) +
2907                             data->size_in_bytes())) {
2908     // The offset is large so bias the mdo by the base of the slot so
2909     // that the ld can use simm13s to reference the slots of the data
2910     mdo_offset_bias = md->byte_offset_of_slot(data, CounterData::count_offset());
2911     __ set(mdo_offset_bias, O7);
2912     __ add(mdo, O7, mdo);
2913   }
2914 
2915   Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()) - mdo_offset_bias);
2916   Bytecodes::Code bc = method->java_code_at_bci(bci);
2917   // Perform additional virtual call profiling for invokevirtual and
2918   // invokeinterface bytecodes
2919   if ((bc == Bytecodes::_invokevirtual || bc == Bytecodes::_invokeinterface) &&
2920       C1ProfileVirtualCalls) {
2921     assert(op->recv()->is_single_cpu(), "recv must be allocated");
2922     Register recv = op->recv()->as_register();
2923     assert_different_registers(mdo, tmp1, recv);
2924     assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
2925     ciKlass* known_klass = op->known_holder();
2926     if (C1OptimizeVirtualCallProfiling && known_klass != NULL) {
2927       // We know the type that will be seen at this call site; we can
2928       // statically update the methodDataOop rather than needing to do
2929       // dynamic tests on the receiver type
2930 
2931       // NOTE: we should probably put a lock around this search to
2932       // avoid collisions by concurrent compilations
2933       ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
2934       uint i;
2935       for (i = 0; i < VirtualCallData::row_limit(); i++) {
2936         ciKlass* receiver = vc_data->receiver(i);
2937         if (known_klass->equals(receiver)) {
2938           Address data_addr(mdo, md->byte_offset_of_slot(data,
2939                                                          VirtualCallData::receiver_count_offset(i)) -
2940                             mdo_offset_bias);
2941           __ ld_ptr(data_addr, tmp1);
2942           __ add(tmp1, DataLayout::counter_increment, tmp1);
2943           __ st_ptr(tmp1, data_addr);
2944           return;
2945         }
2946       }
2947 
2948       // Receiver type not found in profile data; select an empty slot
2949 
2950       // Note that this is less efficient than it should be because it
2951       // always does a write to the receiver part of the
2952       // VirtualCallData rather than just the first time
2953       for (i = 0; i < VirtualCallData::row_limit(); i++) {
2954         ciKlass* receiver = vc_data->receiver(i);
2955         if (receiver == NULL) {
2956           Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)) -
2957                             mdo_offset_bias);
2958           jobject2reg(known_klass->constant_encoding(), tmp1);
2959           __ st_ptr(tmp1, recv_addr);
2960           Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)) -
2961                             mdo_offset_bias);
2962           __ ld_ptr(data_addr, tmp1);
2963           __ add(tmp1, DataLayout::counter_increment, tmp1);
2964           __ st_ptr(tmp1, data_addr);
2965           return;
2966         }
2967       }
2968     } else {
2969       load(Address(recv, oopDesc::klass_offset_in_bytes()), recv, T_OBJECT);
2970       Label update_done;
2971       type_profile_helper(mdo, mdo_offset_bias, md, data, recv, tmp1, &update_done);
2972       // Receiver did not match any saved receiver and there is no empty row for it.
2973       // Increment total counter to indicate polymorphic case.
2974       __ ld_ptr(counter_addr, tmp1);
2975       __ add(tmp1, DataLayout::counter_increment, tmp1);
2976       __ st_ptr(tmp1, counter_addr);
2977 
2978       __ bind(update_done);
2979     }
2980   } else {
2981     // Static call
2982     __ ld_ptr(counter_addr, tmp1);
2983     __ add(tmp1, DataLayout::counter_increment, tmp1);
2984     __ st_ptr(tmp1, counter_addr);
2985   }
2986 }
2987 
2988 void LIR_Assembler::align_backward_branch_target() {
2989   __ align(OptoLoopAlignment);
2990 }
2991 
2992 
2993 void LIR_Assembler::emit_delay(LIR_OpDelay* op) {
2994   // make sure we are expecting a delay
2995   // this has the side effect of clearing the delay state
2996   // so we can use _masm instead of _masm->delayed() to do the
2997   // code generation.
2998   __ delayed();
2999 
3000   // make sure we only emit one instruction
3001   int offset = code_offset();
3002   op->delay_op()->emit_code(this);
3003 #ifdef ASSERT
3004   if (code_offset() - offset != NativeInstruction::nop_instruction_size) {
3005     op->delay_op()->print();
3006   }
3007   assert(code_offset() - offset == NativeInstruction::nop_instruction_size,
3008          "only one instruction can go in a delay slot");
3009 #endif
3010 
3011   // we may also be emitting the call info for the instruction
3012   // which we are the delay slot of.
3013   CodeEmitInfo* call_info = op->call_info();
3014   if (call_info) {
3015     add_call_info(code_offset(), call_info);
3016   }
3017 
3018   if (VerifyStackAtCalls) {
3019     _masm->sub(FP, SP, O7);
3020     _masm->cmp(O7, initial_frame_size_in_bytes());
3021     _masm->trap(Assembler::notEqual, Assembler::ptr_cc, G0, ST_RESERVED_FOR_USER_0+2 );
3022   }
3023 }
3024 
3025 
3026 void LIR_Assembler::negate(LIR_Opr left, LIR_Opr dest) {
3027   assert(left->is_register(), "can only handle registers");
3028 
3029   if (left->is_single_cpu()) {
3030     __ neg(left->as_register(), dest->as_register());
3031   } else if (left->is_single_fpu()) {
3032     __ fneg(FloatRegisterImpl::S, left->as_float_reg(), dest->as_float_reg());
3033   } else if (left->is_double_fpu()) {
3034     __ fneg(FloatRegisterImpl::D, left->as_double_reg(), dest->as_double_reg());
3035   } else {
3036     assert (left->is_double_cpu(), "Must be a long");
3037     Register Rlow = left->as_register_lo();
3038     Register Rhi = left->as_register_hi();
3039 #ifdef _LP64
3040     __ sub(G0, Rlow, dest->as_register_lo());
3041 #else
3042     __ subcc(G0, Rlow, dest->as_register_lo());
3043     __ subc (G0, Rhi,  dest->as_register_hi());
3044 #endif
3045   }
3046 }
3047 
3048 
3049 void LIR_Assembler::fxch(int i) {
3050   Unimplemented();
3051 }
3052 
3053 void LIR_Assembler::fld(int i) {
3054   Unimplemented();
3055 }
3056 
3057 void LIR_Assembler::ffree(int i) {
3058   Unimplemented();
3059 }
3060 
3061 void LIR_Assembler::rt_call(LIR_Opr result, address dest,
3062                             const LIR_OprList* args, LIR_Opr tmp, CodeEmitInfo* info) {
3063 
3064   // if tmp is invalid, then the function being called doesn't destroy the thread
3065   if (tmp->is_valid()) {
3066     __ save_thread(tmp->as_register());
3067   }
3068   __ call(dest, relocInfo::runtime_call_type);
3069   __ delayed()->nop();
3070   if (info != NULL) {
3071     add_call_info_here(info);
3072   }
3073   if (tmp->is_valid()) {
3074     __ restore_thread(tmp->as_register());
3075   }
3076 
3077 #ifdef ASSERT
3078   __ verify_thread();
3079 #endif // ASSERT
3080 }
3081 
3082 
3083 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {
3084 #ifdef _LP64
3085   ShouldNotReachHere();
3086 #endif
3087 
3088   NEEDS_CLEANUP;
3089   if (type == T_LONG) {
3090     LIR_Address* mem_addr = dest->is_address() ? dest->as_address_ptr() : src->as_address_ptr();
3091 
3092     // (extended to allow indexed as well as constant displaced for JSR-166)
3093     Register idx = noreg; // contains either constant offset or index
3094 
3095     int disp = mem_addr->disp();
3096     if (mem_addr->index() == LIR_OprFact::illegalOpr) {
3097       if (!Assembler::is_simm13(disp)) {
3098         idx = O7;
3099         __ set(disp, idx);
3100       }
3101     } else {
3102       assert(disp == 0, "not both indexed and disp");
3103       idx = mem_addr->index()->as_register();
3104     }
3105 
3106     int null_check_offset = -1;
3107 
3108     Register base = mem_addr->base()->as_register();
3109     if (src->is_register() && dest->is_address()) {
3110       // G4 is high half, G5 is low half
3111       if (VM_Version::v9_instructions_work()) {
3112         // clear the top bits of G5, and scale up G4
3113         __ srl (src->as_register_lo(),  0, G5);
3114         __ sllx(src->as_register_hi(), 32, G4);
3115         // combine the two halves into the 64 bits of G4
3116         __ or3(G4, G5, G4);
3117         null_check_offset = __ offset();
3118         if (idx == noreg) {
3119           __ stx(G4, base, disp);
3120         } else {
3121           __ stx(G4, base, idx);
3122         }
3123       } else {
3124         __ mov (src->as_register_hi(), G4);
3125         __ mov (src->as_register_lo(), G5);
3126         null_check_offset = __ offset();
3127         if (idx == noreg) {
3128           __ std(G4, base, disp);
3129         } else {
3130           __ std(G4, base, idx);
3131         }
3132       }
3133     } else if (src->is_address() && dest->is_register()) {
3134       null_check_offset = __ offset();
3135       if (VM_Version::v9_instructions_work()) {
3136         if (idx == noreg) {
3137           __ ldx(base, disp, G5);
3138         } else {
3139           __ ldx(base, idx, G5);
3140         }
3141         __ srax(G5, 32, dest->as_register_hi()); // fetch the high half into hi
3142         __ mov (G5, dest->as_register_lo());     // copy low half into lo
3143       } else {
3144         if (idx == noreg) {
3145           __ ldd(base, disp, G4);
3146         } else {
3147           __ ldd(base, idx, G4);
3148         }
3149         // G4 is high half, G5 is low half
3150         __ mov (G4, dest->as_register_hi());
3151         __ mov (G5, dest->as_register_lo());
3152       }
3153     } else {
3154       Unimplemented();
3155     }
3156     if (info != NULL) {
3157       add_debug_info_for_null_check(null_check_offset, info);
3158     }
3159 
3160   } else {
3161     // use normal move for all other volatiles since they don't need
3162     // special handling to remain atomic.
3163     move_op(src, dest, type, lir_patch_none, info, false, false);
3164   }
3165 }
3166 
3167 void LIR_Assembler::membar() {
3168   // only StoreLoad membars are ever explicitly needed on sparcs in TSO mode
3169   __ membar( Assembler::Membar_mask_bits(Assembler::StoreLoad) );
3170 }
3171 
3172 void LIR_Assembler::membar_acquire() {
3173   // no-op on TSO
3174 }
3175 
3176 void LIR_Assembler::membar_release() {
3177   // no-op on TSO
3178 }
3179 
3180 // Pack two sequential registers containing 32 bit values
3181 // into a single 64 bit register.
3182 // src and src->successor() are packed into dst
3183 // src and dst may be the same register.
3184 // Note: src is destroyed
3185 void LIR_Assembler::pack64(LIR_Opr src, LIR_Opr dst) {
3186   Register rs = src->as_register();
3187   Register rd = dst->as_register_lo();
3188   __ sllx(rs, 32, rs);
3189   __ srl(rs->successor(), 0, rs->successor());
3190   __ or3(rs, rs->successor(), rd);
3191 }
3192 
3193 // Unpack a 64 bit value in a register into
3194 // two sequential registers.
3195 // src is unpacked into dst and dst->successor()
3196 void LIR_Assembler::unpack64(LIR_Opr src, LIR_Opr dst) {
3197   Register rs = src->as_register_lo();
3198   Register rd = dst->as_register_hi();
3199   assert_different_registers(rs, rd, rd->successor());
3200   __ srlx(rs, 32, rd);
3201   __ srl (rs,  0, rd->successor());
3202 }
3203 
3204 
3205 void LIR_Assembler::leal(LIR_Opr addr_opr, LIR_Opr dest) {
3206   LIR_Address* addr = addr_opr->as_address_ptr();
3207   assert(addr->index()->is_illegal() && addr->scale() == LIR_Address::times_1 && Assembler::is_simm13(addr->disp()), "can't handle complex addresses yet");
3208 
3209   __ add(addr->base()->as_pointer_register(), addr->disp(), dest->as_pointer_register());
3210 }
3211 
3212 
3213 void LIR_Assembler::get_thread(LIR_Opr result_reg) {
3214   assert(result_reg->is_register(), "check");
3215   __ mov(G2_thread, result_reg->as_register());
3216 }
3217 
3218 
3219 void LIR_Assembler::peephole(LIR_List* lir) {
3220   LIR_OpList* inst = lir->instructions_list();
3221   for (int i = 0; i < inst->length(); i++) {
3222     LIR_Op* op = inst->at(i);
3223     switch (op->code()) {
3224       case lir_cond_float_branch:
3225       case lir_branch: {
3226         LIR_OpBranch* branch = op->as_OpBranch();
3227         assert(branch->info() == NULL, "shouldn't be state on branches anymore");
3228         LIR_Op* delay_op = NULL;
3229         // we'd like to be able to pull following instructions into
3230         // this slot but we don't know enough to do it safely yet so
3231         // only optimize block to block control flow.
3232         if (LIRFillDelaySlots && branch->block()) {
3233           LIR_Op* prev = inst->at(i - 1);
3234           if (prev && LIR_Assembler::is_single_instruction(prev) && prev->info() == NULL) {
3235             // swap previous instruction into delay slot
3236             inst->at_put(i - 1, op);
3237             inst->at_put(i, new LIR_OpDelay(prev, op->info()));
3238 #ifndef PRODUCT
3239             if (LIRTracePeephole) {
3240               tty->print_cr("delayed");
3241               inst->at(i - 1)->print();
3242               inst->at(i)->print();
3243               tty->cr();
3244             }
3245 #endif
3246             continue;
3247           }
3248         }
3249 
3250         if (!delay_op) {
3251           delay_op = new LIR_OpDelay(new LIR_Op0(lir_nop), NULL);
3252         }
3253         inst->insert_before(i + 1, delay_op);
3254         break;
3255       }
3256       case lir_static_call:
3257       case lir_virtual_call:
3258       case lir_icvirtual_call:
3259       case lir_optvirtual_call:
3260       case lir_dynamic_call: {
3261         LIR_Op* prev = inst->at(i - 1);
3262         if (LIRFillDelaySlots && prev && prev->code() == lir_move && prev->info() == NULL &&
3263             (op->code() != lir_virtual_call ||
3264              !prev->result_opr()->is_single_cpu() ||
3265              prev->result_opr()->as_register() != O0) &&
3266             LIR_Assembler::is_single_instruction(prev)) {
3267           // Only moves without info can be put into the delay slot.
3268           // Also don't allow the setup of the receiver in the delay
3269           // slot for vtable calls.
3270           inst->at_put(i - 1, op);
3271           inst->at_put(i, new LIR_OpDelay(prev, op->info()));
3272 #ifndef PRODUCT
3273           if (LIRTracePeephole) {
3274             tty->print_cr("delayed");
3275             inst->at(i - 1)->print();
3276             inst->at(i)->print();
3277             tty->cr();
3278           }
3279 #endif
3280         } else {
3281           LIR_Op* delay_op = new LIR_OpDelay(new LIR_Op0(lir_nop), op->as_OpJavaCall()->info());
3282           inst->insert_before(i + 1, delay_op);
3283           i++;
3284         }
3285 
3286 #if defined(TIERED) && !defined(_LP64)
3287         // fixup the return value from G1 to O0/O1 for long returns.
3288         // It's done here instead of in LIRGenerator because there's
3289         // such a mismatch between the single reg and double reg
3290         // calling convention.
3291         LIR_OpJavaCall* callop = op->as_OpJavaCall();
3292         if (callop->result_opr() == FrameMap::out_long_opr) {
3293           LIR_OpJavaCall* call;
3294           LIR_OprList* arguments = new LIR_OprList(callop->arguments()->length());
3295           for (int a = 0; a < arguments->length(); a++) {
3296             arguments[a] = callop->arguments()[a];
3297           }
3298           if (op->code() == lir_virtual_call) {
3299             call = new LIR_OpJavaCall(op->code(), callop->method(), callop->receiver(), FrameMap::g1_long_single_opr,
3300                                       callop->vtable_offset(), arguments, callop->info());
3301           } else {
3302             call = new LIR_OpJavaCall(op->code(), callop->method(), callop->receiver(), FrameMap::g1_long_single_opr,
3303                                       callop->addr(), arguments, callop->info());
3304           }
3305           inst->at_put(i - 1, call);
3306           inst->insert_before(i + 1, new LIR_Op1(lir_unpack64, FrameMap::g1_long_single_opr, callop->result_opr(),
3307                                                  T_LONG, lir_patch_none, NULL));
3308         }
3309 #endif
3310         break;
3311       }
3312     }
3313   }
3314 }
3315 
3316 
3317 
3318 
3319 #undef __