src/cpu/x86/vm/c1_LIRAssembler_x86.cpp

Wed, 21 May 2008 16:31:35 -0700

author
kvn
date
Wed, 21 May 2008 16:31:35 -0700
changeset 600
437d03ea40b1
parent 464
d5fc211aea19
child 631
d1605aabd0a1
permissions
-rw-r--r--

6703888: Compressed Oops: use the 32-bits gap after klass in a object
Summary: Use the gap also for a narrow oop field and a boxing object value.
Reviewed-by: coleenp, never

     1 /*
     2  * Copyright 2000-2007 Sun Microsystems, Inc.  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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
    20  * CA 95054 USA or visit www.sun.com if you need additional information or
    21  * have any questions.
    22  *
    23  */
    25 # include "incls/_precompiled.incl"
    26 # include "incls/_c1_LIRAssembler_x86.cpp.incl"
    29 // These masks are used to provide 128-bit aligned bitmasks to the XMM
    30 // instructions, to allow sign-masking or sign-bit flipping.  They allow
    31 // fast versions of NegF/NegD and AbsF/AbsD.
    33 // Note: 'double' and 'long long' have 32-bits alignment on x86.
    34 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
    35   // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
    36   // of 128-bits operands for SSE instructions.
    37   jlong *operand = (jlong*)(((long)adr)&((long)(~0xF)));
    38   // Store the value to a 128-bits operand.
    39   operand[0] = lo;
    40   operand[1] = hi;
    41   return operand;
    42 }
    44 // Buffer for 128-bits masks used by SSE instructions.
    45 static jlong fp_signmask_pool[(4+1)*2]; // 4*128bits(data) + 128bits(alignment)
    47 // Static initialization during VM startup.
    48 static jlong *float_signmask_pool  = double_quadword(&fp_signmask_pool[1*2], CONST64(0x7FFFFFFF7FFFFFFF), CONST64(0x7FFFFFFF7FFFFFFF));
    49 static jlong *double_signmask_pool = double_quadword(&fp_signmask_pool[2*2], CONST64(0x7FFFFFFFFFFFFFFF), CONST64(0x7FFFFFFFFFFFFFFF));
    50 static jlong *float_signflip_pool  = double_quadword(&fp_signmask_pool[3*2], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
    51 static jlong *double_signflip_pool = double_quadword(&fp_signmask_pool[4*2], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
    55 NEEDS_CLEANUP // remove this definitions ?
    56 const Register IC_Klass    = rax;   // where the IC klass is cached
    57 const Register SYNC_header = rax;   // synchronization header
    58 const Register SHIFT_count = rcx;   // where count for shift operations must be
    60 #define __ _masm->
    63 static void select_different_registers(Register preserve,
    64                                        Register extra,
    65                                        Register &tmp1,
    66                                        Register &tmp2) {
    67   if (tmp1 == preserve) {
    68     assert_different_registers(tmp1, tmp2, extra);
    69     tmp1 = extra;
    70   } else if (tmp2 == preserve) {
    71     assert_different_registers(tmp1, tmp2, extra);
    72     tmp2 = extra;
    73   }
    74   assert_different_registers(preserve, tmp1, tmp2);
    75 }
    79 static void select_different_registers(Register preserve,
    80                                        Register extra,
    81                                        Register &tmp1,
    82                                        Register &tmp2,
    83                                        Register &tmp3) {
    84   if (tmp1 == preserve) {
    85     assert_different_registers(tmp1, tmp2, tmp3, extra);
    86     tmp1 = extra;
    87   } else if (tmp2 == preserve) {
    88     assert_different_registers(tmp1, tmp2, tmp3, extra);
    89     tmp2 = extra;
    90   } else if (tmp3 == preserve) {
    91     assert_different_registers(tmp1, tmp2, tmp3, extra);
    92     tmp3 = extra;
    93   }
    94   assert_different_registers(preserve, tmp1, tmp2, tmp3);
    95 }
    99 bool LIR_Assembler::is_small_constant(LIR_Opr opr) {
   100   if (opr->is_constant()) {
   101     LIR_Const* constant = opr->as_constant_ptr();
   102     switch (constant->type()) {
   103       case T_INT: {
   104         return true;
   105       }
   107       default:
   108         return false;
   109     }
   110   }
   111   return false;
   112 }
   115 LIR_Opr LIR_Assembler::receiverOpr() {
   116   return FrameMap::rcx_oop_opr;
   117 }
   119 LIR_Opr LIR_Assembler::incomingReceiverOpr() {
   120   return receiverOpr();
   121 }
   123 LIR_Opr LIR_Assembler::osrBufferPointer() {
   124   return FrameMap::rcx_opr;
   125 }
   127 //--------------fpu register translations-----------------------
   130 address LIR_Assembler::float_constant(float f) {
   131   address const_addr = __ float_constant(f);
   132   if (const_addr == NULL) {
   133     bailout("const section overflow");
   134     return __ code()->consts()->start();
   135   } else {
   136     return const_addr;
   137   }
   138 }
   141 address LIR_Assembler::double_constant(double d) {
   142   address const_addr = __ double_constant(d);
   143   if (const_addr == NULL) {
   144     bailout("const section overflow");
   145     return __ code()->consts()->start();
   146   } else {
   147     return const_addr;
   148   }
   149 }
   152 void LIR_Assembler::set_24bit_FPU() {
   153   __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_24()));
   154 }
   156 void LIR_Assembler::reset_FPU() {
   157   __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
   158 }
   160 void LIR_Assembler::fpop() {
   161   __ fpop();
   162 }
   164 void LIR_Assembler::fxch(int i) {
   165   __ fxch(i);
   166 }
   168 void LIR_Assembler::fld(int i) {
   169   __ fld_s(i);
   170 }
   172 void LIR_Assembler::ffree(int i) {
   173   __ ffree(i);
   174 }
   176 void LIR_Assembler::breakpoint() {
   177   __ int3();
   178 }
   180 void LIR_Assembler::push(LIR_Opr opr) {
   181   if (opr->is_single_cpu()) {
   182     __ push_reg(opr->as_register());
   183   } else if (opr->is_double_cpu()) {
   184     __ push_reg(opr->as_register_hi());
   185     __ push_reg(opr->as_register_lo());
   186   } else if (opr->is_stack()) {
   187     __ push_addr(frame_map()->address_for_slot(opr->single_stack_ix()));
   188   } else if (opr->is_constant()) {
   189     LIR_Const* const_opr = opr->as_constant_ptr();
   190     if (const_opr->type() == T_OBJECT) {
   191       __ push_oop(const_opr->as_jobject());
   192     } else if (const_opr->type() == T_INT) {
   193       __ push_jint(const_opr->as_jint());
   194     } else {
   195       ShouldNotReachHere();
   196     }
   198   } else {
   199     ShouldNotReachHere();
   200   }
   201 }
   203 void LIR_Assembler::pop(LIR_Opr opr) {
   204   if (opr->is_single_cpu()) {
   205     __ pop(opr->as_register());
   206   } else {
   207     ShouldNotReachHere();
   208   }
   209 }
   211 //-------------------------------------------
   212 Address LIR_Assembler::as_Address(LIR_Address* addr) {
   213   if (addr->base()->is_illegal()) {
   214     assert(addr->index()->is_illegal(), "must be illegal too");
   215     //return Address(addr->disp(), relocInfo::none);
   216     // hack for now since this should really return an AddressLiteral
   217     // which will have to await 64bit c1 changes.
   218     return Address(noreg, addr->disp());
   219   }
   221   Register base = addr->base()->as_register();
   223   if (addr->index()->is_illegal()) {
   224     return Address( base, addr->disp());
   225   } else if (addr->index()->is_single_cpu()) {
   226     Register index = addr->index()->as_register();
   227     return Address(base, index, (Address::ScaleFactor) addr->scale(), addr->disp());
   228   } else if (addr->index()->is_constant()) {
   229     int addr_offset = (addr->index()->as_constant_ptr()->as_jint() << addr->scale()) + addr->disp();
   231     return Address(base, addr_offset);
   232   } else {
   233     Unimplemented();
   234     return Address();
   235   }
   236 }
   239 Address LIR_Assembler::as_Address_hi(LIR_Address* addr) {
   240   Address base = as_Address(addr);
   241   return Address(base._base, base._index, base._scale, base._disp + BytesPerWord);
   242 }
   245 Address LIR_Assembler::as_Address_lo(LIR_Address* addr) {
   246   return as_Address(addr);
   247 }
   250 void LIR_Assembler::osr_entry() {
   251   offsets()->set_value(CodeOffsets::OSR_Entry, code_offset());
   252   BlockBegin* osr_entry = compilation()->hir()->osr_entry();
   253   ValueStack* entry_state = osr_entry->state();
   254   int number_of_locks = entry_state->locks_size();
   256   // we jump here if osr happens with the interpreter
   257   // state set up to continue at the beginning of the
   258   // loop that triggered osr - in particular, we have
   259   // the following registers setup:
   260   //
   261   // rcx: osr buffer
   262   //
   264   // build frame
   265   ciMethod* m = compilation()->method();
   266   __ build_frame(initial_frame_size_in_bytes());
   268   // OSR buffer is
   269   //
   270   // locals[nlocals-1..0]
   271   // monitors[0..number_of_locks]
   272   //
   273   // locals is a direct copy of the interpreter frame so in the osr buffer
   274   // so first slot in the local array is the last local from the interpreter
   275   // and last slot is local[0] (receiver) from the interpreter
   276   //
   277   // Similarly with locks. The first lock slot in the osr buffer is the nth lock
   278   // from the interpreter frame, the nth lock slot in the osr buffer is 0th lock
   279   // in the interpreter frame (the method lock if a sync method)
   281   // Initialize monitors in the compiled activation.
   282   //   rcx: pointer to osr buffer
   283   //
   284   // All other registers are dead at this point and the locals will be
   285   // copied into place by code emitted in the IR.
   287   Register OSR_buf = osrBufferPointer()->as_register();
   288   { assert(frame::interpreter_frame_monitor_size() == BasicObjectLock::size(), "adjust code below");
   289     int monitor_offset = BytesPerWord * method()->max_locals() +
   290       (BasicObjectLock::size() * BytesPerWord) * (number_of_locks - 1);
   291     for (int i = 0; i < number_of_locks; i++) {
   292       int slot_offset = monitor_offset - ((i * BasicObjectLock::size()) * BytesPerWord);
   293 #ifdef ASSERT
   294       // verify the interpreter's monitor has a non-null object
   295       {
   296         Label L;
   297         __ cmpl(Address(OSR_buf, slot_offset + BasicObjectLock::obj_offset_in_bytes()), NULL_WORD);
   298         __ jcc(Assembler::notZero, L);
   299         __ stop("locked object is NULL");
   300         __ bind(L);
   301       }
   302 #endif
   303       __ movl(rbx, Address(OSR_buf, slot_offset + BasicObjectLock::lock_offset_in_bytes()));
   304       __ movl(frame_map()->address_for_monitor_lock(i), rbx);
   305       __ movl(rbx, Address(OSR_buf, slot_offset + BasicObjectLock::obj_offset_in_bytes()));
   306       __ movl(frame_map()->address_for_monitor_object(i), rbx);
   307     }
   308   }
   309 }
   312 // inline cache check; done before the frame is built.
   313 int LIR_Assembler::check_icache() {
   314   Register receiver = FrameMap::receiver_opr->as_register();
   315   Register ic_klass = IC_Klass;
   317   if (!VerifyOops) {
   318     // insert some nops so that the verified entry point is aligned on CodeEntryAlignment
   319     while ((__ offset() + 9) % CodeEntryAlignment != 0) {
   320       __ nop();
   321     }
   322   }
   323   int offset = __ offset();
   324   __ inline_cache_check(receiver, IC_Klass);
   325   assert(__ offset() % CodeEntryAlignment == 0 || VerifyOops, "alignment must be correct");
   326   if (VerifyOops) {
   327     // force alignment after the cache check.
   328     // It's been verified to be aligned if !VerifyOops
   329     __ align(CodeEntryAlignment);
   330   }
   331   return offset;
   332 }
   335 void LIR_Assembler::jobject2reg_with_patching(Register reg, CodeEmitInfo* info) {
   336   jobject o = NULL;
   337   PatchingStub* patch = new PatchingStub(_masm, PatchingStub::load_klass_id);
   338   __ movoop(reg, o);
   339   patching_epilog(patch, lir_patch_normal, reg, info);
   340 }
   343 void LIR_Assembler::monitorexit(LIR_Opr obj_opr, LIR_Opr lock_opr, Register new_hdr, int monitor_no, Register exception) {
   344   if (exception->is_valid()) {
   345     // preserve exception
   346     // note: the monitor_exit runtime call is a leaf routine
   347     //       and cannot block => no GC can happen
   348     // The slow case (MonitorAccessStub) uses the first two stack slots
   349     // ([esp+0] and [esp+4]), therefore we store the exception at [esp+8]
   350     __ movl (Address(rsp, 2*wordSize), exception);
   351   }
   353   Register obj_reg  = obj_opr->as_register();
   354   Register lock_reg = lock_opr->as_register();
   356   // setup registers (lock_reg must be rax, for lock_object)
   357   assert(obj_reg != SYNC_header && lock_reg != SYNC_header, "rax, must be available here");
   358   Register hdr = lock_reg;
   359   assert(new_hdr == SYNC_header, "wrong register");
   360   lock_reg = new_hdr;
   361   // compute pointer to BasicLock
   362   Address lock_addr = frame_map()->address_for_monitor_lock(monitor_no);
   363   __ leal(lock_reg, lock_addr);
   364   // unlock object
   365   MonitorAccessStub* slow_case = new MonitorExitStub(lock_opr, true, monitor_no);
   366   // _slow_case_stubs->append(slow_case);
   367   // temporary fix: must be created after exceptionhandler, therefore as call stub
   368   _slow_case_stubs->append(slow_case);
   369   if (UseFastLocking) {
   370     // try inlined fast unlocking first, revert to slow locking if it fails
   371     // note: lock_reg points to the displaced header since the displaced header offset is 0!
   372     assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
   373     __ unlock_object(hdr, obj_reg, lock_reg, *slow_case->entry());
   374   } else {
   375     // always do slow unlocking
   376     // note: the slow unlocking code could be inlined here, however if we use
   377     //       slow unlocking, speed doesn't matter anyway and this solution is
   378     //       simpler and requires less duplicated code - additionally, the
   379     //       slow unlocking code is the same in either case which simplifies
   380     //       debugging
   381     __ jmp(*slow_case->entry());
   382   }
   383   // done
   384   __ bind(*slow_case->continuation());
   386   if (exception->is_valid()) {
   387     // restore exception
   388     __ movl (exception, Address(rsp, 2 * wordSize));
   389   }
   390 }
   392 // This specifies the rsp decrement needed to build the frame
   393 int LIR_Assembler::initial_frame_size_in_bytes() {
   394   // if rounding, must let FrameMap know!
   395   return (frame_map()->framesize() - 2)  * BytesPerWord; // subtract two words to account for return address and link
   396 }
   399 void LIR_Assembler::emit_exception_handler() {
   400   // if the last instruction is a call (typically to do a throw which
   401   // is coming at the end after block reordering) the return address
   402   // must still point into the code area in order to avoid assertion
   403   // failures when searching for the corresponding bci => add a nop
   404   // (was bug 5/14/1999 - gri)
   406   __ nop();
   408   // generate code for exception handler
   409   address handler_base = __ start_a_stub(exception_handler_size);
   410   if (handler_base == NULL) {
   411     // not enough space left for the handler
   412     bailout("exception handler overflow");
   413     return;
   414   }
   415 #ifdef ASSERT
   416   int offset = code_offset();
   417 #endif // ASSERT
   419   compilation()->offsets()->set_value(CodeOffsets::Exceptions, code_offset());
   421   // if the method does not have an exception handler, then there is
   422   // no reason to search for one
   423   if (compilation()->has_exception_handlers() || JvmtiExport::can_post_exceptions()) {
   424     // the exception oop and pc are in rax, and rdx
   425     // no other registers need to be preserved, so invalidate them
   426     __ invalidate_registers(false, true, true, false, true, true);
   428     // check that there is really an exception
   429     __ verify_not_null_oop(rax);
   431     // search an exception handler (rax: exception oop, rdx: throwing pc)
   432     __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::handle_exception_nofpu_id)));
   434     // if the call returns here, then the exception handler for particular
   435     // exception doesn't exist -> unwind activation and forward exception to caller
   436   }
   438   // the exception oop is in rax,
   439   // no other registers need to be preserved, so invalidate them
   440   __ invalidate_registers(false, true, true, true, true, true);
   442   // check that there is really an exception
   443   __ verify_not_null_oop(rax);
   445   // unlock the receiver/klass if necessary
   446   // rax,: exception
   447   ciMethod* method = compilation()->method();
   448   if (method->is_synchronized() && GenerateSynchronizationCode) {
   449     monitorexit(FrameMap::rbx_oop_opr, FrameMap::rcx_opr, SYNC_header, 0, rax);
   450   }
   452   // unwind activation and forward exception to caller
   453   // rax,: exception
   454   __ jump(RuntimeAddress(Runtime1::entry_for(Runtime1::unwind_exception_id)));
   456   assert(code_offset() - offset <= exception_handler_size, "overflow");
   458   __ end_a_stub();
   459 }
   461 void LIR_Assembler::emit_deopt_handler() {
   462   // if the last instruction is a call (typically to do a throw which
   463   // is coming at the end after block reordering) the return address
   464   // must still point into the code area in order to avoid assertion
   465   // failures when searching for the corresponding bci => add a nop
   466   // (was bug 5/14/1999 - gri)
   468   __ nop();
   470   // generate code for exception handler
   471   address handler_base = __ start_a_stub(deopt_handler_size);
   472   if (handler_base == NULL) {
   473     // not enough space left for the handler
   474     bailout("deopt handler overflow");
   475     return;
   476   }
   477 #ifdef ASSERT
   478   int offset = code_offset();
   479 #endif // ASSERT
   481   compilation()->offsets()->set_value(CodeOffsets::Deopt, code_offset());
   483   InternalAddress here(__ pc());
   484   __ pushptr(here.addr());
   486   __ jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack()));
   488   assert(code_offset() - offset <= deopt_handler_size, "overflow");
   490   __ end_a_stub();
   492 }
   495 // This is the fast version of java.lang.String.compare; it has not
   496 // OSR-entry and therefore, we generate a slow version for OSR's
   497 void LIR_Assembler::emit_string_compare(LIR_Opr arg0, LIR_Opr arg1, LIR_Opr dst, CodeEmitInfo* info) {
   498   __ movl (rbx, rcx); // receiver is in rcx
   499   __ movl (rax, arg1->as_register());
   501   // Get addresses of first characters from both Strings
   502   __ movl (rsi, Address(rax, java_lang_String::value_offset_in_bytes()));
   503   __ movl (rcx, Address(rax, java_lang_String::offset_offset_in_bytes()));
   504   __ leal (rsi, Address(rsi, rcx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
   507   // rbx, may be NULL
   508   add_debug_info_for_null_check_here(info);
   509   __ movl (rdi, Address(rbx, java_lang_String::value_offset_in_bytes()));
   510   __ movl (rcx, Address(rbx, java_lang_String::offset_offset_in_bytes()));
   511   __ leal (rdi, Address(rdi, rcx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
   513   // compute minimum length (in rax) and difference of lengths (on top of stack)
   514   if (VM_Version::supports_cmov()) {
   515     __ movl (rbx, Address(rbx, java_lang_String::count_offset_in_bytes()));
   516     __ movl (rax, Address(rax, java_lang_String::count_offset_in_bytes()));
   517     __ movl (rcx, rbx);
   518     __ subl (rbx, rax); // subtract lengths
   519     __ pushl(rbx);      // result
   520     __ cmovl(Assembler::lessEqual, rax, rcx);
   521   } else {
   522     Label L;
   523     __ movl (rbx, Address(rbx, java_lang_String::count_offset_in_bytes()));
   524     __ movl (rcx, Address(rax, java_lang_String::count_offset_in_bytes()));
   525     __ movl (rax, rbx);
   526     __ subl (rbx, rcx);
   527     __ pushl(rbx);
   528     __ jcc  (Assembler::lessEqual, L);
   529     __ movl (rax, rcx);
   530     __ bind (L);
   531   }
   532   // is minimum length 0?
   533   Label noLoop, haveResult;
   534   __ testl (rax, rax);
   535   __ jcc (Assembler::zero, noLoop);
   537   // compare first characters
   538   __ load_unsigned_word(rcx, Address(rdi, 0));
   539   __ load_unsigned_word(rbx, Address(rsi, 0));
   540   __ subl(rcx, rbx);
   541   __ jcc(Assembler::notZero, haveResult);
   542   // starting loop
   543   __ decrement(rax); // we already tested index: skip one
   544   __ jcc(Assembler::zero, noLoop);
   546   // set rsi.edi to the end of the arrays (arrays have same length)
   547   // negate the index
   549   __ leal(rsi, Address(rsi, rax, Address::times_2, type2aelembytes(T_CHAR)));
   550   __ leal(rdi, Address(rdi, rax, Address::times_2, type2aelembytes(T_CHAR)));
   551   __ negl(rax);
   553   // compare the strings in a loop
   555   Label loop;
   556   __ align(wordSize);
   557   __ bind(loop);
   558   __ load_unsigned_word(rcx, Address(rdi, rax, Address::times_2, 0));
   559   __ load_unsigned_word(rbx, Address(rsi, rax, Address::times_2, 0));
   560   __ subl(rcx, rbx);
   561   __ jcc(Assembler::notZero, haveResult);
   562   __ increment(rax);
   563   __ jcc(Assembler::notZero, loop);
   565   // strings are equal up to min length
   567   __ bind(noLoop);
   568   __ popl(rax);
   569   return_op(LIR_OprFact::illegalOpr);
   571   __ bind(haveResult);
   572   // leave instruction is going to discard the TOS value
   573   __ movl (rax, rcx); // result of call is in rax,
   574 }
   577 void LIR_Assembler::return_op(LIR_Opr result) {
   578   assert(result->is_illegal() || !result->is_single_cpu() || result->as_register() == rax, "word returns are in rax,");
   579   if (!result->is_illegal() && result->is_float_kind() && !result->is_xmm_register()) {
   580     assert(result->fpu() == 0, "result must already be on TOS");
   581   }
   583   // Pop the stack before the safepoint code
   584   __ leave();
   586   bool result_is_oop = result->is_valid() ? result->is_oop() : false;
   588   // Note: we do not need to round double result; float result has the right precision
   589   // the poll sets the condition code, but no data registers
   590   AddressLiteral polling_page(os::get_polling_page() + (SafepointPollOffset % os::vm_page_size()),
   591                               relocInfo::poll_return_type);
   592   __ test32(rax, polling_page);
   594   __ ret(0);
   595 }
   598 int LIR_Assembler::safepoint_poll(LIR_Opr tmp, CodeEmitInfo* info) {
   599   AddressLiteral polling_page(os::get_polling_page() + (SafepointPollOffset % os::vm_page_size()),
   600                               relocInfo::poll_type);
   602   if (info != NULL) {
   603     add_debug_info_for_branch(info);
   604   } else {
   605     ShouldNotReachHere();
   606   }
   608   int offset = __ offset();
   609   __ test32(rax, polling_page);
   610   return offset;
   611 }
   614 void LIR_Assembler::move_regs(Register from_reg, Register to_reg) {
   615   if (from_reg != to_reg) __ movl(to_reg, from_reg);
   616 }
   618 void LIR_Assembler::swap_reg(Register a, Register b) {
   619   __ xchgl(a, b);
   620 }
   623 void LIR_Assembler::const2reg(LIR_Opr src, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
   624   assert(src->is_constant(), "should not call otherwise");
   625   assert(dest->is_register(), "should not call otherwise");
   626   LIR_Const* c = src->as_constant_ptr();
   628   switch (c->type()) {
   629     case T_INT: {
   630       assert(patch_code == lir_patch_none, "no patching handled here");
   631       __ movl(dest->as_register(), c->as_jint());
   632       break;
   633     }
   635     case T_LONG: {
   636       assert(patch_code == lir_patch_none, "no patching handled here");
   637       __ movl(dest->as_register_lo(), c->as_jint_lo());
   638       __ movl(dest->as_register_hi(), c->as_jint_hi());
   639       break;
   640     }
   642     case T_OBJECT: {
   643       if (patch_code != lir_patch_none) {
   644         jobject2reg_with_patching(dest->as_register(), info);
   645       } else {
   646         __ movoop(dest->as_register(), c->as_jobject());
   647       }
   648       break;
   649     }
   651     case T_FLOAT: {
   652       if (dest->is_single_xmm()) {
   653         if (c->is_zero_float()) {
   654           __ xorps(dest->as_xmm_float_reg(), dest->as_xmm_float_reg());
   655         } else {
   656           __ movflt(dest->as_xmm_float_reg(),
   657                    InternalAddress(float_constant(c->as_jfloat())));
   658         }
   659       } else {
   660         assert(dest->is_single_fpu(), "must be");
   661         assert(dest->fpu_regnr() == 0, "dest must be TOS");
   662         if (c->is_zero_float()) {
   663           __ fldz();
   664         } else if (c->is_one_float()) {
   665           __ fld1();
   666         } else {
   667           __ fld_s (InternalAddress(float_constant(c->as_jfloat())));
   668         }
   669       }
   670       break;
   671     }
   673     case T_DOUBLE: {
   674       if (dest->is_double_xmm()) {
   675         if (c->is_zero_double()) {
   676           __ xorpd(dest->as_xmm_double_reg(), dest->as_xmm_double_reg());
   677         } else {
   678           __ movdbl(dest->as_xmm_double_reg(),
   679                     InternalAddress(double_constant(c->as_jdouble())));
   680         }
   681       } else {
   682         assert(dest->is_double_fpu(), "must be");
   683         assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
   684         if (c->is_zero_double()) {
   685           __ fldz();
   686         } else if (c->is_one_double()) {
   687           __ fld1();
   688         } else {
   689           __ fld_d (InternalAddress(double_constant(c->as_jdouble())));
   690         }
   691       }
   692       break;
   693     }
   695     default:
   696       ShouldNotReachHere();
   697   }
   698 }
   700 void LIR_Assembler::const2stack(LIR_Opr src, LIR_Opr dest) {
   701   assert(src->is_constant(), "should not call otherwise");
   702   assert(dest->is_stack(), "should not call otherwise");
   703   LIR_Const* c = src->as_constant_ptr();
   705   switch (c->type()) {
   706     case T_INT:  // fall through
   707     case T_FLOAT:
   708       __ movl(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jint_bits());
   709       break;
   711     case T_OBJECT:
   712       __ movoop(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jobject());
   713       break;
   715     case T_LONG:  // fall through
   716     case T_DOUBLE:
   717       __ movl(frame_map()->address_for_slot(dest->double_stack_ix(),
   718                                             lo_word_offset_in_bytes), c->as_jint_lo_bits());
   719       __ movl(frame_map()->address_for_slot(dest->double_stack_ix(),
   720                                             hi_word_offset_in_bytes), c->as_jint_hi_bits());
   721       break;
   723     default:
   724       ShouldNotReachHere();
   725   }
   726 }
   728 void LIR_Assembler::const2mem(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info ) {
   729   assert(src->is_constant(), "should not call otherwise");
   730   assert(dest->is_address(), "should not call otherwise");
   731   LIR_Const* c = src->as_constant_ptr();
   732   LIR_Address* addr = dest->as_address_ptr();
   734   if (info != NULL) add_debug_info_for_null_check_here(info);
   735   switch (type) {
   736     case T_INT:    // fall through
   737     case T_FLOAT:
   738       __ movl(as_Address(addr), c->as_jint_bits());
   739       break;
   741     case T_OBJECT:  // fall through
   742     case T_ARRAY:
   743       if (c->as_jobject() == NULL) {
   744         __ movl(as_Address(addr), NULL_WORD);
   745       } else {
   746         __ movoop(as_Address(addr), c->as_jobject());
   747       }
   748       break;
   750     case T_LONG:    // fall through
   751     case T_DOUBLE:
   752       __ movl(as_Address_hi(addr), c->as_jint_hi_bits());
   753       __ movl(as_Address_lo(addr), c->as_jint_lo_bits());
   754       break;
   756     case T_BOOLEAN: // fall through
   757     case T_BYTE:
   758       __ movb(as_Address(addr), c->as_jint() & 0xFF);
   759       break;
   761     case T_CHAR:    // fall through
   762     case T_SHORT:
   763       __ movw(as_Address(addr), c->as_jint() & 0xFFFF);
   764       break;
   766     default:
   767       ShouldNotReachHere();
   768   };
   769 }
   772 void LIR_Assembler::reg2reg(LIR_Opr src, LIR_Opr dest) {
   773   assert(src->is_register(), "should not call otherwise");
   774   assert(dest->is_register(), "should not call otherwise");
   776   // move between cpu-registers
   777   if (dest->is_single_cpu()) {
   778     assert(src->is_single_cpu(), "must match");
   779     if (src->type() == T_OBJECT) {
   780       __ verify_oop(src->as_register());
   781     }
   782     move_regs(src->as_register(), dest->as_register());
   784   } else if (dest->is_double_cpu()) {
   785     assert(src->is_double_cpu(), "must match");
   786     Register f_lo = src->as_register_lo();
   787     Register f_hi = src->as_register_hi();
   788     Register t_lo = dest->as_register_lo();
   789     Register t_hi = dest->as_register_hi();
   790     assert(f_lo != f_hi && t_lo != t_hi, "invalid register allocation");
   792     if (f_lo == t_hi && f_hi == t_lo) {
   793       swap_reg(f_lo, f_hi);
   794     } else if (f_hi == t_lo) {
   795       assert(f_lo != t_hi, "overwriting register");
   796       move_regs(f_hi, t_hi);
   797       move_regs(f_lo, t_lo);
   798     } else {
   799       assert(f_hi != t_lo, "overwriting register");
   800       move_regs(f_lo, t_lo);
   801       move_regs(f_hi, t_hi);
   802     }
   804     // special moves from fpu-register to xmm-register
   805     // necessary for method results
   806   } else if (src->is_single_xmm() && !dest->is_single_xmm()) {
   807     __ movflt(Address(rsp, 0), src->as_xmm_float_reg());
   808     __ fld_s(Address(rsp, 0));
   809   } else if (src->is_double_xmm() && !dest->is_double_xmm()) {
   810     __ movdbl(Address(rsp, 0), src->as_xmm_double_reg());
   811     __ fld_d(Address(rsp, 0));
   812   } else if (dest->is_single_xmm() && !src->is_single_xmm()) {
   813     __ fstp_s(Address(rsp, 0));
   814     __ movflt(dest->as_xmm_float_reg(), Address(rsp, 0));
   815   } else if (dest->is_double_xmm() && !src->is_double_xmm()) {
   816     __ fstp_d(Address(rsp, 0));
   817     __ movdbl(dest->as_xmm_double_reg(), Address(rsp, 0));
   819     // move between xmm-registers
   820   } else if (dest->is_single_xmm()) {
   821     assert(src->is_single_xmm(), "must match");
   822     __ movflt(dest->as_xmm_float_reg(), src->as_xmm_float_reg());
   823   } else if (dest->is_double_xmm()) {
   824     assert(src->is_double_xmm(), "must match");
   825     __ movdbl(dest->as_xmm_double_reg(), src->as_xmm_double_reg());
   827     // move between fpu-registers (no instruction necessary because of fpu-stack)
   828   } else if (dest->is_single_fpu() || dest->is_double_fpu()) {
   829     assert(src->is_single_fpu() || src->is_double_fpu(), "must match");
   830     assert(src->fpu() == dest->fpu(), "currently should be nothing to do");
   831   } else {
   832     ShouldNotReachHere();
   833   }
   834 }
   836 void LIR_Assembler::reg2stack(LIR_Opr src, LIR_Opr dest, BasicType type, bool pop_fpu_stack) {
   837   assert(src->is_register(), "should not call otherwise");
   838   assert(dest->is_stack(), "should not call otherwise");
   840   if (src->is_single_cpu()) {
   841     Address dst = frame_map()->address_for_slot(dest->single_stack_ix());
   842     if (type == T_OBJECT || type == T_ARRAY) {
   843       __ verify_oop(src->as_register());
   844     }
   845     __ movl (dst, src->as_register());
   847   } else if (src->is_double_cpu()) {
   848     Address dstLO = frame_map()->address_for_slot(dest->double_stack_ix(), lo_word_offset_in_bytes);
   849     Address dstHI = frame_map()->address_for_slot(dest->double_stack_ix(), hi_word_offset_in_bytes);
   850     __ movl (dstLO, src->as_register_lo());
   851     __ movl (dstHI, src->as_register_hi());
   853   } else if (src->is_single_xmm()) {
   854     Address dst_addr = frame_map()->address_for_slot(dest->single_stack_ix());
   855     __ movflt(dst_addr, src->as_xmm_float_reg());
   857   } else if (src->is_double_xmm()) {
   858     Address dst_addr = frame_map()->address_for_slot(dest->double_stack_ix());
   859     __ movdbl(dst_addr, src->as_xmm_double_reg());
   861   } else if (src->is_single_fpu()) {
   862     assert(src->fpu_regnr() == 0, "argument must be on TOS");
   863     Address dst_addr = frame_map()->address_for_slot(dest->single_stack_ix());
   864     if (pop_fpu_stack)     __ fstp_s (dst_addr);
   865     else                   __ fst_s  (dst_addr);
   867   } else if (src->is_double_fpu()) {
   868     assert(src->fpu_regnrLo() == 0, "argument must be on TOS");
   869     Address dst_addr = frame_map()->address_for_slot(dest->double_stack_ix());
   870     if (pop_fpu_stack)     __ fstp_d (dst_addr);
   871     else                   __ fst_d  (dst_addr);
   873   } else {
   874     ShouldNotReachHere();
   875   }
   876 }
   879 void LIR_Assembler::reg2mem(LIR_Opr src, LIR_Opr dest, BasicType type, LIR_PatchCode patch_code, CodeEmitInfo* info, bool pop_fpu_stack, bool /* unaligned */) {
   880   LIR_Address* to_addr = dest->as_address_ptr();
   881   PatchingStub* patch = NULL;
   883   if (type == T_ARRAY || type == T_OBJECT) {
   884     __ verify_oop(src->as_register());
   885   }
   886   if (patch_code != lir_patch_none) {
   887     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
   888   }
   889   if (info != NULL) {
   890     add_debug_info_for_null_check_here(info);
   891   }
   893   switch (type) {
   894     case T_FLOAT: {
   895       if (src->is_single_xmm()) {
   896         __ movflt(as_Address(to_addr), src->as_xmm_float_reg());
   897       } else {
   898         assert(src->is_single_fpu(), "must be");
   899         assert(src->fpu_regnr() == 0, "argument must be on TOS");
   900         if (pop_fpu_stack)      __ fstp_s(as_Address(to_addr));
   901         else                    __ fst_s (as_Address(to_addr));
   902       }
   903       break;
   904     }
   906     case T_DOUBLE: {
   907       if (src->is_double_xmm()) {
   908         __ movdbl(as_Address(to_addr), src->as_xmm_double_reg());
   909       } else {
   910         assert(src->is_double_fpu(), "must be");
   911         assert(src->fpu_regnrLo() == 0, "argument must be on TOS");
   912         if (pop_fpu_stack)      __ fstp_d(as_Address(to_addr));
   913         else                    __ fst_d (as_Address(to_addr));
   914       }
   915       break;
   916     }
   918     case T_ADDRESS: // fall through
   919     case T_ARRAY:   // fall through
   920     case T_OBJECT:  // fall through
   921     case T_INT:
   922       __ movl(as_Address(to_addr), src->as_register());
   923       break;
   925     case T_LONG: {
   926       Register from_lo = src->as_register_lo();
   927       Register from_hi = src->as_register_hi();
   928       Register base = to_addr->base()->as_register();
   929       Register index = noreg;
   930       if (to_addr->index()->is_register()) {
   931         index = to_addr->index()->as_register();
   932       }
   933       if (base == from_lo || index == from_lo) {
   934         assert(base != from_hi, "can't be");
   935         assert(index == noreg || (index != base && index != from_hi), "can't handle this");
   936         __ movl(as_Address_hi(to_addr), from_hi);
   937         if (patch != NULL) {
   938           patching_epilog(patch, lir_patch_high, base, info);
   939           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
   940           patch_code = lir_patch_low;
   941         }
   942         __ movl(as_Address_lo(to_addr), from_lo);
   943       } else {
   944         assert(index == noreg || (index != base && index != from_lo), "can't handle this");
   945         __ movl(as_Address_lo(to_addr), from_lo);
   946         if (patch != NULL) {
   947           patching_epilog(patch, lir_patch_low, base, info);
   948           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
   949           patch_code = lir_patch_high;
   950         }
   951         __ movl(as_Address_hi(to_addr), from_hi);
   952       }
   953       break;
   954     }
   956     case T_BYTE:    // fall through
   957     case T_BOOLEAN: {
   958       Register src_reg = src->as_register();
   959       Address dst_addr = as_Address(to_addr);
   960       assert(VM_Version::is_P6() || src_reg->has_byte_register(), "must use byte registers if not P6");
   961       __ movb(dst_addr, src_reg);
   962       break;
   963     }
   965     case T_CHAR:    // fall through
   966     case T_SHORT:
   967       __ movw(as_Address(to_addr), src->as_register());
   968       break;
   970     default:
   971       ShouldNotReachHere();
   972   }
   974   if (patch_code != lir_patch_none) {
   975     patching_epilog(patch, patch_code, to_addr->base()->as_register(), info);
   976   }
   977 }
   980 void LIR_Assembler::stack2reg(LIR_Opr src, LIR_Opr dest, BasicType type) {
   981   assert(src->is_stack(), "should not call otherwise");
   982   assert(dest->is_register(), "should not call otherwise");
   984   if (dest->is_single_cpu()) {
   985     __ movl(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()));
   986     if (type == T_ARRAY || type == T_OBJECT) {
   987       __ verify_oop(dest->as_register());
   988     }
   990   } else if (dest->is_double_cpu()) {
   991     Address src_addr_LO = frame_map()->address_for_slot(src->double_stack_ix(), lo_word_offset_in_bytes);
   992     Address src_addr_HI = frame_map()->address_for_slot(src->double_stack_ix(), hi_word_offset_in_bytes);
   993     __ movl(dest->as_register_hi(), src_addr_HI);
   994     __ movl(dest->as_register_lo(), src_addr_LO);
   996   } else if (dest->is_single_xmm()) {
   997     Address src_addr = frame_map()->address_for_slot(src->single_stack_ix());
   998     __ movflt(dest->as_xmm_float_reg(), src_addr);
  1000   } else if (dest->is_double_xmm()) {
  1001     Address src_addr = frame_map()->address_for_slot(src->double_stack_ix());
  1002     __ movdbl(dest->as_xmm_double_reg(), src_addr);
  1004   } else if (dest->is_single_fpu()) {
  1005     assert(dest->fpu_regnr() == 0, "dest must be TOS");
  1006     Address src_addr = frame_map()->address_for_slot(src->single_stack_ix());
  1007     __ fld_s(src_addr);
  1009   } else if (dest->is_double_fpu()) {
  1010     assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
  1011     Address src_addr = frame_map()->address_for_slot(src->double_stack_ix());
  1012     __ fld_d(src_addr);
  1014   } else {
  1015     ShouldNotReachHere();
  1020 void LIR_Assembler::stack2stack(LIR_Opr src, LIR_Opr dest, BasicType type) {
  1021   if (src->is_single_stack()) {
  1022     __ pushl(frame_map()->address_for_slot(src ->single_stack_ix()));
  1023     __ popl (frame_map()->address_for_slot(dest->single_stack_ix()));
  1025   } else if (src->is_double_stack()) {
  1026     __ pushl(frame_map()->address_for_slot(src ->double_stack_ix(), 0));
  1027     // push and pop the part at src + 4, adding 4 for the previous push
  1028     __ pushl(frame_map()->address_for_slot(src ->double_stack_ix(), 4 + 4));
  1029     __ popl (frame_map()->address_for_slot(dest->double_stack_ix(), 4 + 4));
  1030     __ popl (frame_map()->address_for_slot(dest->double_stack_ix(), 0));
  1032   } else {
  1033     ShouldNotReachHere();
  1038 void LIR_Assembler::mem2reg(LIR_Opr src, LIR_Opr dest, BasicType type, LIR_PatchCode patch_code, CodeEmitInfo* info, bool /* unaligned */) {
  1039   assert(src->is_address(), "should not call otherwise");
  1040   assert(dest->is_register(), "should not call otherwise");
  1042   LIR_Address* addr = src->as_address_ptr();
  1043   Address from_addr = as_Address(addr);
  1045   switch (type) {
  1046     case T_BOOLEAN: // fall through
  1047     case T_BYTE:    // fall through
  1048     case T_CHAR:    // fall through
  1049     case T_SHORT:
  1050       if (!VM_Version::is_P6() && !from_addr.uses(dest->as_register())) {
  1051         // on pre P6 processors we may get partial register stalls
  1052         // so blow away the value of to_rinfo before loading a
  1053         // partial word into it.  Do it here so that it precedes
  1054         // the potential patch point below.
  1055         __ xorl(dest->as_register(), dest->as_register());
  1057       break;
  1060   PatchingStub* patch = NULL;
  1061   if (patch_code != lir_patch_none) {
  1062     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
  1064   if (info != NULL) {
  1065     add_debug_info_for_null_check_here(info);
  1068   switch (type) {
  1069     case T_FLOAT: {
  1070       if (dest->is_single_xmm()) {
  1071         __ movflt(dest->as_xmm_float_reg(), from_addr);
  1072       } else {
  1073         assert(dest->is_single_fpu(), "must be");
  1074         assert(dest->fpu_regnr() == 0, "dest must be TOS");
  1075         __ fld_s(from_addr);
  1077       break;
  1080     case T_DOUBLE: {
  1081       if (dest->is_double_xmm()) {
  1082         __ movdbl(dest->as_xmm_double_reg(), from_addr);
  1083       } else {
  1084         assert(dest->is_double_fpu(), "must be");
  1085         assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
  1086         __ fld_d(from_addr);
  1088       break;
  1091     case T_ADDRESS: // fall through
  1092     case T_OBJECT:  // fall through
  1093     case T_ARRAY:   // fall through
  1094     case T_INT:
  1095       __ movl(dest->as_register(), from_addr);
  1096       break;
  1098     case T_LONG: {
  1099       Register to_lo = dest->as_register_lo();
  1100       Register to_hi = dest->as_register_hi();
  1101       Register base = addr->base()->as_register();
  1102       Register index = noreg;
  1103       if (addr->index()->is_register()) {
  1104         index = addr->index()->as_register();
  1106       if ((base == to_lo && index == to_hi) ||
  1107           (base == to_hi && index == to_lo)) {
  1108         // addresses with 2 registers are only formed as a result of
  1109         // array access so this code will never have to deal with
  1110         // patches or null checks.
  1111         assert(info == NULL && patch == NULL, "must be");
  1112         __ leal(to_hi, as_Address(addr));
  1113         __ movl(to_lo, Address(to_hi, 0));
  1114         __ movl(to_hi, Address(to_hi, BytesPerWord));
  1115       } else if (base == to_lo || index == to_lo) {
  1116         assert(base != to_hi, "can't be");
  1117         assert(index == noreg || (index != base && index != to_hi), "can't handle this");
  1118         __ movl(to_hi, as_Address_hi(addr));
  1119         if (patch != NULL) {
  1120           patching_epilog(patch, lir_patch_high, base, info);
  1121           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
  1122           patch_code = lir_patch_low;
  1124         __ movl(to_lo, as_Address_lo(addr));
  1125       } else {
  1126         assert(index == noreg || (index != base && index != to_lo), "can't handle this");
  1127         __ movl(to_lo, as_Address_lo(addr));
  1128         if (patch != NULL) {
  1129           patching_epilog(patch, lir_patch_low, base, info);
  1130           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
  1131           patch_code = lir_patch_high;
  1133         __ movl(to_hi, as_Address_hi(addr));
  1135       break;
  1138     case T_BOOLEAN: // fall through
  1139     case T_BYTE: {
  1140       Register dest_reg = dest->as_register();
  1141       assert(VM_Version::is_P6() || dest_reg->has_byte_register(), "must use byte registers if not P6");
  1142       if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
  1143         __ movsxb(dest_reg, from_addr);
  1144       } else {
  1145         __ movb(dest_reg, from_addr);
  1146         __ shll(dest_reg, 24);
  1147         __ sarl(dest_reg, 24);
  1149       break;
  1152     case T_CHAR: {
  1153       Register dest_reg = dest->as_register();
  1154       assert(VM_Version::is_P6() || dest_reg->has_byte_register(), "must use byte registers if not P6");
  1155       if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
  1156         __ movzxw(dest_reg, from_addr);
  1157       } else {
  1158         __ movw(dest_reg, from_addr);
  1160       break;
  1163     case T_SHORT: {
  1164       Register dest_reg = dest->as_register();
  1165       if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
  1166         __ movsxw(dest_reg, from_addr);
  1167       } else {
  1168         __ movw(dest_reg, from_addr);
  1169         __ shll(dest_reg, 16);
  1170         __ sarl(dest_reg, 16);
  1172       break;
  1175     default:
  1176       ShouldNotReachHere();
  1179   if (patch != NULL) {
  1180     patching_epilog(patch, patch_code, addr->base()->as_register(), info);
  1183   if (type == T_ARRAY || type == T_OBJECT) {
  1184     __ verify_oop(dest->as_register());
  1189 void LIR_Assembler::prefetchr(LIR_Opr src) {
  1190   LIR_Address* addr = src->as_address_ptr();
  1191   Address from_addr = as_Address(addr);
  1193   if (VM_Version::supports_sse()) {
  1194     switch (ReadPrefetchInstr) {
  1195       case 0:
  1196         __ prefetchnta(from_addr); break;
  1197       case 1:
  1198         __ prefetcht0(from_addr); break;
  1199       case 2:
  1200         __ prefetcht2(from_addr); break;
  1201       default:
  1202         ShouldNotReachHere(); break;
  1204   } else if (VM_Version::supports_3dnow()) {
  1205     __ prefetchr(from_addr);
  1210 void LIR_Assembler::prefetchw(LIR_Opr src) {
  1211   LIR_Address* addr = src->as_address_ptr();
  1212   Address from_addr = as_Address(addr);
  1214   if (VM_Version::supports_sse()) {
  1215     switch (AllocatePrefetchInstr) {
  1216       case 0:
  1217         __ prefetchnta(from_addr); break;
  1218       case 1:
  1219         __ prefetcht0(from_addr); break;
  1220       case 2:
  1221         __ prefetcht2(from_addr); break;
  1222       case 3:
  1223         __ prefetchw(from_addr); break;
  1224       default:
  1225         ShouldNotReachHere(); break;
  1227   } else if (VM_Version::supports_3dnow()) {
  1228     __ prefetchw(from_addr);
  1233 NEEDS_CLEANUP; // This could be static?
  1234 Address::ScaleFactor LIR_Assembler::array_element_size(BasicType type) const {
  1235   int elem_size = type2aelembytes(type);
  1236   switch (elem_size) {
  1237     case 1: return Address::times_1;
  1238     case 2: return Address::times_2;
  1239     case 4: return Address::times_4;
  1240     case 8: return Address::times_8;
  1242   ShouldNotReachHere();
  1243   return Address::no_scale;
  1247 void LIR_Assembler::emit_op3(LIR_Op3* op) {
  1248   switch (op->code()) {
  1249     case lir_idiv:
  1250     case lir_irem:
  1251       arithmetic_idiv(op->code(),
  1252                       op->in_opr1(),
  1253                       op->in_opr2(),
  1254                       op->in_opr3(),
  1255                       op->result_opr(),
  1256                       op->info());
  1257       break;
  1258     default:      ShouldNotReachHere(); break;
  1262 void LIR_Assembler::emit_opBranch(LIR_OpBranch* op) {
  1263 #ifdef ASSERT
  1264   assert(op->block() == NULL || op->block()->label() == op->label(), "wrong label");
  1265   if (op->block() != NULL)  _branch_target_blocks.append(op->block());
  1266   if (op->ublock() != NULL) _branch_target_blocks.append(op->ublock());
  1267 #endif
  1269   if (op->cond() == lir_cond_always) {
  1270     if (op->info() != NULL) add_debug_info_for_branch(op->info());
  1271     __ jmp (*(op->label()));
  1272   } else {
  1273     Assembler::Condition acond = Assembler::zero;
  1274     if (op->code() == lir_cond_float_branch) {
  1275       assert(op->ublock() != NULL, "must have unordered successor");
  1276       __ jcc(Assembler::parity, *(op->ublock()->label()));
  1277       switch(op->cond()) {
  1278         case lir_cond_equal:        acond = Assembler::equal;      break;
  1279         case lir_cond_notEqual:     acond = Assembler::notEqual;   break;
  1280         case lir_cond_less:         acond = Assembler::below;      break;
  1281         case lir_cond_lessEqual:    acond = Assembler::belowEqual; break;
  1282         case lir_cond_greaterEqual: acond = Assembler::aboveEqual; break;
  1283         case lir_cond_greater:      acond = Assembler::above;      break;
  1284         default:                         ShouldNotReachHere();
  1286     } else {
  1287       switch (op->cond()) {
  1288         case lir_cond_equal:        acond = Assembler::equal;       break;
  1289         case lir_cond_notEqual:     acond = Assembler::notEqual;    break;
  1290         case lir_cond_less:         acond = Assembler::less;        break;
  1291         case lir_cond_lessEqual:    acond = Assembler::lessEqual;   break;
  1292         case lir_cond_greaterEqual: acond = Assembler::greaterEqual;break;
  1293         case lir_cond_greater:      acond = Assembler::greater;     break;
  1294         case lir_cond_belowEqual:   acond = Assembler::belowEqual;  break;
  1295         case lir_cond_aboveEqual:   acond = Assembler::aboveEqual;  break;
  1296         default:                         ShouldNotReachHere();
  1299     __ jcc(acond,*(op->label()));
  1303 void LIR_Assembler::emit_opConvert(LIR_OpConvert* op) {
  1304   LIR_Opr src  = op->in_opr();
  1305   LIR_Opr dest = op->result_opr();
  1307   switch (op->bytecode()) {
  1308     case Bytecodes::_i2l:
  1309       move_regs(src->as_register(), dest->as_register_lo());
  1310       move_regs(src->as_register(), dest->as_register_hi());
  1311       __ sarl(dest->as_register_hi(), 31);
  1312       break;
  1314     case Bytecodes::_l2i:
  1315       move_regs(src->as_register_lo(), dest->as_register());
  1316       break;
  1318     case Bytecodes::_i2b:
  1319       move_regs(src->as_register(), dest->as_register());
  1320       __ sign_extend_byte(dest->as_register());
  1321       break;
  1323     case Bytecodes::_i2c:
  1324       move_regs(src->as_register(), dest->as_register());
  1325       __ andl(dest->as_register(), 0xFFFF);
  1326       break;
  1328     case Bytecodes::_i2s:
  1329       move_regs(src->as_register(), dest->as_register());
  1330       __ sign_extend_short(dest->as_register());
  1331       break;
  1334     case Bytecodes::_f2d:
  1335     case Bytecodes::_d2f:
  1336       if (dest->is_single_xmm()) {
  1337         __ cvtsd2ss(dest->as_xmm_float_reg(), src->as_xmm_double_reg());
  1338       } else if (dest->is_double_xmm()) {
  1339         __ cvtss2sd(dest->as_xmm_double_reg(), src->as_xmm_float_reg());
  1340       } else {
  1341         assert(src->fpu() == dest->fpu(), "register must be equal");
  1342         // do nothing (float result is rounded later through spilling)
  1344       break;
  1346     case Bytecodes::_i2f:
  1347     case Bytecodes::_i2d:
  1348       if (dest->is_single_xmm()) {
  1349         __ cvtsi2ss(dest->as_xmm_float_reg(), src->as_register());
  1350       } else if (dest->is_double_xmm()) {
  1351         __ cvtsi2sd(dest->as_xmm_double_reg(), src->as_register());
  1352       } else {
  1353         assert(dest->fpu() == 0, "result must be on TOS");
  1354         __ movl(Address(rsp, 0), src->as_register());
  1355         __ fild_s(Address(rsp, 0));
  1357       break;
  1359     case Bytecodes::_f2i:
  1360     case Bytecodes::_d2i:
  1361       if (src->is_single_xmm()) {
  1362         __ cvttss2si(dest->as_register(), src->as_xmm_float_reg());
  1363       } else if (src->is_double_xmm()) {
  1364         __ cvttsd2si(dest->as_register(), src->as_xmm_double_reg());
  1365       } else {
  1366         assert(src->fpu() == 0, "input must be on TOS");
  1367         __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_trunc()));
  1368         __ fist_s(Address(rsp, 0));
  1369         __ movl(dest->as_register(), Address(rsp, 0));
  1370         __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
  1373       // IA32 conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
  1374       assert(op->stub() != NULL, "stub required");
  1375       __ cmpl(dest->as_register(), 0x80000000);
  1376       __ jcc(Assembler::equal, *op->stub()->entry());
  1377       __ bind(*op->stub()->continuation());
  1378       break;
  1380     case Bytecodes::_l2f:
  1381     case Bytecodes::_l2d:
  1382       assert(!dest->is_xmm_register(), "result in xmm register not supported (no SSE instruction present)");
  1383       assert(dest->fpu() == 0, "result must be on TOS");
  1385       __ movl(Address(rsp, 0),            src->as_register_lo());
  1386       __ movl(Address(rsp, BytesPerWord), src->as_register_hi());
  1387       __ fild_d(Address(rsp, 0));
  1388       // float result is rounded later through spilling
  1389       break;
  1391     case Bytecodes::_f2l:
  1392     case Bytecodes::_d2l:
  1393       assert(!src->is_xmm_register(), "input in xmm register not supported (no SSE instruction present)");
  1394       assert(src->fpu() == 0, "input must be on TOS");
  1395       assert(dest == FrameMap::rax_rdx_long_opr, "runtime stub places result in these registers");
  1397       // instruction sequence too long to inline it here
  1399         __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::fpu2long_stub_id)));
  1401       break;
  1403     default: ShouldNotReachHere();
  1407 void LIR_Assembler::emit_alloc_obj(LIR_OpAllocObj* op) {
  1408   if (op->init_check()) {
  1409     __ cmpl(Address(op->klass()->as_register(),
  1410                     instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc)),
  1411             instanceKlass::fully_initialized);
  1412     add_debug_info_for_null_check_here(op->stub()->info());
  1413     __ jcc(Assembler::notEqual, *op->stub()->entry());
  1415   __ allocate_object(op->obj()->as_register(),
  1416                      op->tmp1()->as_register(),
  1417                      op->tmp2()->as_register(),
  1418                      op->header_size(),
  1419                      op->object_size(),
  1420                      op->klass()->as_register(),
  1421                      *op->stub()->entry());
  1422   __ bind(*op->stub()->continuation());
  1425 void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
  1426   if (UseSlowPath ||
  1427       (!UseFastNewObjectArray && (op->type() == T_OBJECT || op->type() == T_ARRAY)) ||
  1428       (!UseFastNewTypeArray   && (op->type() != T_OBJECT && op->type() != T_ARRAY))) {
  1429     __ jmp(*op->stub()->entry());
  1430   } else {
  1431     Register len =  op->len()->as_register();
  1432     Register tmp1 = op->tmp1()->as_register();
  1433     Register tmp2 = op->tmp2()->as_register();
  1434     Register tmp3 = op->tmp3()->as_register();
  1435     if (len == tmp1) {
  1436       tmp1 = tmp3;
  1437     } else if (len == tmp2) {
  1438       tmp2 = tmp3;
  1439     } else if (len == tmp3) {
  1440       // everything is ok
  1441     } else {
  1442       __ movl(tmp3, len);
  1444     __ allocate_array(op->obj()->as_register(),
  1445                       len,
  1446                       tmp1,
  1447                       tmp2,
  1448                       arrayOopDesc::header_size(op->type()),
  1449                       array_element_size(op->type()),
  1450                       op->klass()->as_register(),
  1451                       *op->stub()->entry());
  1453   __ bind(*op->stub()->continuation());
  1458 void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
  1459   LIR_Code code = op->code();
  1460   if (code == lir_store_check) {
  1461     Register value = op->object()->as_register();
  1462     Register array = op->array()->as_register();
  1463     Register k_RInfo = op->tmp1()->as_register();
  1464     Register klass_RInfo = op->tmp2()->as_register();
  1465     Register Rtmp1 = op->tmp3()->as_register();
  1467     CodeStub* stub = op->stub();
  1468     Label done;
  1469     __ cmpl(value, 0);
  1470     __ jcc(Assembler::equal, done);
  1471     add_debug_info_for_null_check_here(op->info_for_exception());
  1472     __ movl(k_RInfo, Address(array, oopDesc::klass_offset_in_bytes()));
  1473     __ movl(klass_RInfo, Address(value, oopDesc::klass_offset_in_bytes()));
  1475     // get instance klass
  1476     __ movl(k_RInfo, Address(k_RInfo, objArrayKlass::element_klass_offset_in_bytes() + sizeof(oopDesc)));
  1477     // get super_check_offset
  1478     __ movl(Rtmp1, Address(k_RInfo, sizeof(oopDesc) + Klass::super_check_offset_offset_in_bytes()));
  1479     // See if we get an immediate positive hit
  1480     __ cmpl(k_RInfo, Address(klass_RInfo, Rtmp1, Address::times_1));
  1481     __ jcc(Assembler::equal, done);
  1482     // check for immediate negative hit
  1483     __ cmpl(Rtmp1, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes());
  1484     __ jcc(Assembler::notEqual, *stub->entry());
  1485     // check for self
  1486     __ cmpl(klass_RInfo, k_RInfo);
  1487     __ jcc(Assembler::equal, done);
  1489     __ pushl(klass_RInfo);
  1490     __ pushl(k_RInfo);
  1491     __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
  1492     __ popl(klass_RInfo);
  1493     __ popl(k_RInfo);
  1494     __ cmpl(k_RInfo, 0);
  1495     __ jcc(Assembler::equal, *stub->entry());
  1496     __ bind(done);
  1497   } else if (op->code() == lir_checkcast) {
  1498     // we always need a stub for the failure case.
  1499     CodeStub* stub = op->stub();
  1500     Register obj = op->object()->as_register();
  1501     Register k_RInfo = op->tmp1()->as_register();
  1502     Register klass_RInfo = op->tmp2()->as_register();
  1503     Register dst = op->result_opr()->as_register();
  1504     ciKlass* k = op->klass();
  1505     Register Rtmp1 = noreg;
  1507     Label done;
  1508     if (obj == k_RInfo) {
  1509       k_RInfo = dst;
  1510     } else if (obj == klass_RInfo) {
  1511       klass_RInfo = dst;
  1513     if (k->is_loaded()) {
  1514       select_different_registers(obj, dst, k_RInfo, klass_RInfo);
  1515     } else {
  1516       Rtmp1 = op->tmp3()->as_register();
  1517       select_different_registers(obj, dst, k_RInfo, klass_RInfo, Rtmp1);
  1520     assert_different_registers(obj, k_RInfo, klass_RInfo);
  1521     if (!k->is_loaded()) {
  1522       jobject2reg_with_patching(k_RInfo, op->info_for_patch());
  1523     } else {
  1524       k_RInfo = noreg;
  1526     assert(obj != k_RInfo, "must be different");
  1527     __ cmpl(obj, 0);
  1528     if (op->profiled_method() != NULL) {
  1529       ciMethod* method = op->profiled_method();
  1530       int bci          = op->profiled_bci();
  1532       Label profile_done;
  1533       __ jcc(Assembler::notEqual, profile_done);
  1534       // Object is null; update methodDataOop
  1535       ciMethodData* md = method->method_data();
  1536       if (md == NULL) {
  1537         bailout("out of memory building methodDataOop");
  1538         return;
  1540       ciProfileData* data = md->bci_to_data(bci);
  1541       assert(data != NULL,       "need data for checkcast");
  1542       assert(data->is_BitData(), "need BitData for checkcast");
  1543       Register mdo  = klass_RInfo;
  1544       __ movoop(mdo, md->encoding());
  1545       Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
  1546       int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
  1547       __ orl(data_addr, header_bits);
  1548       __ jmp(done);
  1549       __ bind(profile_done);
  1550     } else {
  1551       __ jcc(Assembler::equal, done);
  1553     __ verify_oop(obj);
  1555     if (op->fast_check()) {
  1556       // get object classo
  1557       // not a safepoint as obj null check happens earlier
  1558       if (k->is_loaded()) {
  1559         __ cmpoop(Address(obj, oopDesc::klass_offset_in_bytes()), k->encoding());
  1560       } else {
  1561         __ cmpl(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
  1564       __ jcc(Assembler::notEqual, *stub->entry());
  1565       __ bind(done);
  1566     } else {
  1567       // get object class
  1568       // not a safepoint as obj null check happens earlier
  1569       __ movl(klass_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
  1570       if (k->is_loaded()) {
  1571         // See if we get an immediate positive hit
  1572         __ cmpoop(Address(klass_RInfo, k->super_check_offset()), k->encoding());
  1573         if (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes() != k->super_check_offset()) {
  1574           __ jcc(Assembler::notEqual, *stub->entry());
  1575         } else {
  1576           // See if we get an immediate positive hit
  1577           __ jcc(Assembler::equal, done);
  1578           // check for self
  1579           __ cmpoop(klass_RInfo, k->encoding());
  1580           __ jcc(Assembler::equal, done);
  1582           __ pushl(klass_RInfo);
  1583           __ pushoop(k->encoding());
  1584           __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
  1585           __ popl(klass_RInfo);
  1586           __ popl(klass_RInfo);
  1587           __ cmpl(klass_RInfo, 0);
  1588           __ jcc(Assembler::equal, *stub->entry());
  1590         __ bind(done);
  1591       } else {
  1592         __ movl(Rtmp1, Address(k_RInfo, sizeof(oopDesc) + Klass::super_check_offset_offset_in_bytes()));
  1593         // See if we get an immediate positive hit
  1594         __ cmpl(k_RInfo, Address(klass_RInfo, Rtmp1, Address::times_1));
  1595         __ jcc(Assembler::equal, done);
  1596         // check for immediate negative hit
  1597         __ cmpl(Rtmp1, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes());
  1598         __ jcc(Assembler::notEqual, *stub->entry());
  1599         // check for self
  1600         __ cmpl(klass_RInfo, k_RInfo);
  1601         __ jcc(Assembler::equal, done);
  1603         __ pushl(klass_RInfo);
  1604         __ pushl(k_RInfo);
  1605         __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
  1606         __ popl(klass_RInfo);
  1607         __ popl(k_RInfo);
  1608         __ cmpl(k_RInfo, 0);
  1609         __ jcc(Assembler::equal, *stub->entry());
  1610         __ bind(done);
  1614     if (dst != obj) {
  1615       __ movl(dst, obj);
  1617   } else if (code == lir_instanceof) {
  1618     Register obj = op->object()->as_register();
  1619     Register k_RInfo = op->tmp1()->as_register();
  1620     Register klass_RInfo = op->tmp2()->as_register();
  1621     Register dst = op->result_opr()->as_register();
  1622     ciKlass* k = op->klass();
  1624     Label done;
  1625     Label zero;
  1626     Label one;
  1627     if (obj == k_RInfo) {
  1628       k_RInfo = klass_RInfo;
  1629       klass_RInfo = obj;
  1631     // patching may screw with our temporaries on sparc,
  1632     // so let's do it before loading the class
  1633     if (!k->is_loaded()) {
  1634       jobject2reg_with_patching(k_RInfo, op->info_for_patch());
  1636     assert(obj != k_RInfo, "must be different");
  1638     __ verify_oop(obj);
  1639     if (op->fast_check()) {
  1640       __ cmpl(obj, 0);
  1641       __ jcc(Assembler::equal, zero);
  1642       // get object class
  1643       // not a safepoint as obj null check happens earlier
  1644       if (k->is_loaded()) {
  1645         __ cmpoop(Address(obj, oopDesc::klass_offset_in_bytes()), k->encoding());
  1646         k_RInfo = noreg;
  1647       } else {
  1648         __ cmpl(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
  1651       __ jcc(Assembler::equal, one);
  1652     } else {
  1653       // get object class
  1654       // not a safepoint as obj null check happens earlier
  1655       __ cmpl(obj, 0);
  1656       __ jcc(Assembler::equal, zero);
  1657       __ movl(klass_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
  1658       if (k->is_loaded()) {
  1659         // See if we get an immediate positive hit
  1660         __ cmpoop(Address(klass_RInfo, k->super_check_offset()), k->encoding());
  1661         __ jcc(Assembler::equal, one);
  1662         if (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes() == k->super_check_offset()) {
  1663           // check for self
  1664           __ cmpoop(klass_RInfo, k->encoding());
  1665           __ jcc(Assembler::equal, one);
  1666           __ pushl(klass_RInfo);
  1667           __ pushoop(k->encoding());
  1668           __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
  1669           __ popl(klass_RInfo);
  1670           __ popl(dst);
  1671           __ jmp(done);
  1673       } else {
  1674         assert(dst != klass_RInfo && dst != k_RInfo, "need 3 registers");
  1676         __ movl(dst, Address(k_RInfo, sizeof(oopDesc) + Klass::super_check_offset_offset_in_bytes()));
  1677         // See if we get an immediate positive hit
  1678         __ cmpl(k_RInfo, Address(klass_RInfo, dst, Address::times_1));
  1679         __ jcc(Assembler::equal, one);
  1680         // check for immediate negative hit
  1681         __ cmpl(dst, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes());
  1682         __ jcc(Assembler::notEqual, zero);
  1683         // check for self
  1684         __ cmpl(klass_RInfo, k_RInfo);
  1685         __ jcc(Assembler::equal, one);
  1687         __ pushl(klass_RInfo);
  1688         __ pushl(k_RInfo);
  1689         __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
  1690         __ popl(klass_RInfo);
  1691         __ popl(dst);
  1692         __ jmp(done);
  1695     __ bind(zero);
  1696     __ xorl(dst, dst);
  1697     __ jmp(done);
  1698     __ bind(one);
  1699     __ movl(dst, 1);
  1700     __ bind(done);
  1701   } else {
  1702     ShouldNotReachHere();
  1708 void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
  1709   if (op->code() == lir_cas_long) {
  1710     assert(VM_Version::supports_cx8(), "wrong machine");
  1711     assert(op->cmp_value()->as_register_lo() == rax, "wrong register");
  1712     assert(op->cmp_value()->as_register_hi() == rdx, "wrong register");
  1713     assert(op->new_value()->as_register_lo() == rbx, "wrong register");
  1714     assert(op->new_value()->as_register_hi() == rcx, "wrong register");
  1715     Register addr = op->addr()->as_register();
  1716     if (os::is_MP()) {
  1717       __ lock();
  1719     __ cmpxchg8(Address(addr, 0));
  1721   } else if (op->code() == lir_cas_int || op->code() == lir_cas_obj) {
  1722     Register addr = op->addr()->as_register();
  1723     Register newval = op->new_value()->as_register();
  1724     Register cmpval = op->cmp_value()->as_register();
  1725     assert(cmpval == rax, "wrong register");
  1726     assert(newval != NULL, "new val must be register");
  1727     assert(cmpval != newval, "cmp and new values must be in different registers");
  1728     assert(cmpval != addr, "cmp and addr must be in different registers");
  1729     assert(newval != addr, "new value and addr must be in different registers");
  1730     if (os::is_MP()) {
  1731       __ lock();
  1733     __ cmpxchg(newval, Address(addr, 0));
  1734   } else {
  1735     Unimplemented();
  1740 void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result) {
  1741   Assembler::Condition acond, ncond;
  1742   switch (condition) {
  1743     case lir_cond_equal:        acond = Assembler::equal;        ncond = Assembler::notEqual;     break;
  1744     case lir_cond_notEqual:     acond = Assembler::notEqual;     ncond = Assembler::equal;        break;
  1745     case lir_cond_less:         acond = Assembler::less;         ncond = Assembler::greaterEqual; break;
  1746     case lir_cond_lessEqual:    acond = Assembler::lessEqual;    ncond = Assembler::greater;      break;
  1747     case lir_cond_greaterEqual: acond = Assembler::greaterEqual; ncond = Assembler::less;         break;
  1748     case lir_cond_greater:      acond = Assembler::greater;      ncond = Assembler::lessEqual;    break;
  1749     case lir_cond_belowEqual:   acond = Assembler::belowEqual;   ncond = Assembler::above;        break;
  1750     case lir_cond_aboveEqual:   acond = Assembler::aboveEqual;   ncond = Assembler::below;        break;
  1751     default:                    ShouldNotReachHere();
  1754   if (opr1->is_cpu_register()) {
  1755     reg2reg(opr1, result);
  1756   } else if (opr1->is_stack()) {
  1757     stack2reg(opr1, result, result->type());
  1758   } else if (opr1->is_constant()) {
  1759     const2reg(opr1, result, lir_patch_none, NULL);
  1760   } else {
  1761     ShouldNotReachHere();
  1764   if (VM_Version::supports_cmov() && !opr2->is_constant()) {
  1765     // optimized version that does not require a branch
  1766     if (opr2->is_single_cpu()) {
  1767       assert(opr2->cpu_regnr() != result->cpu_regnr(), "opr2 already overwritten by previous move");
  1768       __ cmovl(ncond, result->as_register(), opr2->as_register());
  1769     } else if (opr2->is_double_cpu()) {
  1770       assert(opr2->cpu_regnrLo() != result->cpu_regnrLo() && opr2->cpu_regnrLo() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
  1771       assert(opr2->cpu_regnrHi() != result->cpu_regnrLo() && opr2->cpu_regnrHi() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
  1772       __ cmovl(ncond, result->as_register_lo(), opr2->as_register_lo());
  1773       __ cmovl(ncond, result->as_register_hi(), opr2->as_register_hi());
  1774     } else if (opr2->is_single_stack()) {
  1775       __ cmovl(ncond, result->as_register(), frame_map()->address_for_slot(opr2->single_stack_ix()));
  1776     } else if (opr2->is_double_stack()) {
  1777       __ cmovl(ncond, result->as_register_lo(), frame_map()->address_for_slot(opr2->double_stack_ix(), lo_word_offset_in_bytes));
  1778       __ cmovl(ncond, result->as_register_hi(), frame_map()->address_for_slot(opr2->double_stack_ix(), hi_word_offset_in_bytes));
  1779     } else {
  1780       ShouldNotReachHere();
  1783   } else {
  1784     Label skip;
  1785     __ jcc (acond, skip);
  1786     if (opr2->is_cpu_register()) {
  1787       reg2reg(opr2, result);
  1788     } else if (opr2->is_stack()) {
  1789       stack2reg(opr2, result, result->type());
  1790     } else if (opr2->is_constant()) {
  1791       const2reg(opr2, result, lir_patch_none, NULL);
  1792     } else {
  1793       ShouldNotReachHere();
  1795     __ bind(skip);
  1800 void LIR_Assembler::arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest, CodeEmitInfo* info, bool pop_fpu_stack) {
  1801   assert(info == NULL, "should never be used, idiv/irem and ldiv/lrem not handled by this method");
  1803   if (left->is_single_cpu()) {
  1804     assert(left == dest, "left and dest must be equal");
  1805     Register lreg = left->as_register();
  1807     if (right->is_single_cpu()) {
  1808       // cpu register - cpu register
  1809       Register rreg = right->as_register();
  1810       switch (code) {
  1811         case lir_add: __ addl (lreg, rreg); break;
  1812         case lir_sub: __ subl (lreg, rreg); break;
  1813         case lir_mul: __ imull(lreg, rreg); break;
  1814         default:      ShouldNotReachHere();
  1817     } else if (right->is_stack()) {
  1818       // cpu register - stack
  1819       Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
  1820       switch (code) {
  1821         case lir_add: __ addl(lreg, raddr); break;
  1822         case lir_sub: __ subl(lreg, raddr); break;
  1823         default:      ShouldNotReachHere();
  1826     } else if (right->is_constant()) {
  1827       // cpu register - constant
  1828       jint c = right->as_constant_ptr()->as_jint();
  1829       switch (code) {
  1830         case lir_add: {
  1831           __ increment(lreg, c);
  1832           break;
  1834         case lir_sub: {
  1835           __ decrement(lreg, c);
  1836           break;
  1838         default: ShouldNotReachHere();
  1841     } else {
  1842       ShouldNotReachHere();
  1845   } else if (left->is_double_cpu()) {
  1846     assert(left == dest, "left and dest must be equal");
  1847     Register lreg_lo = left->as_register_lo();
  1848     Register lreg_hi = left->as_register_hi();
  1850     if (right->is_double_cpu()) {
  1851       // cpu register - cpu register
  1852       Register rreg_lo = right->as_register_lo();
  1853       Register rreg_hi = right->as_register_hi();
  1854       assert_different_registers(lreg_lo, lreg_hi, rreg_lo, rreg_hi);
  1855       switch (code) {
  1856         case lir_add:
  1857           __ addl(lreg_lo, rreg_lo);
  1858           __ adcl(lreg_hi, rreg_hi);
  1859           break;
  1860         case lir_sub:
  1861           __ subl(lreg_lo, rreg_lo);
  1862           __ sbbl(lreg_hi, rreg_hi);
  1863           break;
  1864         case lir_mul:
  1865           assert(lreg_lo == rax && lreg_hi == rdx, "must be");
  1866           __ imull(lreg_hi, rreg_lo);
  1867           __ imull(rreg_hi, lreg_lo);
  1868           __ addl (rreg_hi, lreg_hi);
  1869           __ mull (rreg_lo);
  1870           __ addl (lreg_hi, rreg_hi);
  1871           break;
  1872         default:
  1873           ShouldNotReachHere();
  1876     } else if (right->is_constant()) {
  1877       // cpu register - constant
  1878       jint c_lo = right->as_constant_ptr()->as_jint_lo();
  1879       jint c_hi = right->as_constant_ptr()->as_jint_hi();
  1880       switch (code) {
  1881         case lir_add:
  1882           __ addl(lreg_lo, c_lo);
  1883           __ adcl(lreg_hi, c_hi);
  1884           break;
  1885         case lir_sub:
  1886           __ subl(lreg_lo, c_lo);
  1887           __ sbbl(lreg_hi, c_hi);
  1888           break;
  1889         default:
  1890           ShouldNotReachHere();
  1893     } else {
  1894       ShouldNotReachHere();
  1897   } else if (left->is_single_xmm()) {
  1898     assert(left == dest, "left and dest must be equal");
  1899     XMMRegister lreg = left->as_xmm_float_reg();
  1901     if (right->is_single_xmm()) {
  1902       XMMRegister rreg = right->as_xmm_float_reg();
  1903       switch (code) {
  1904         case lir_add: __ addss(lreg, rreg);  break;
  1905         case lir_sub: __ subss(lreg, rreg);  break;
  1906         case lir_mul_strictfp: // fall through
  1907         case lir_mul: __ mulss(lreg, rreg);  break;
  1908         case lir_div_strictfp: // fall through
  1909         case lir_div: __ divss(lreg, rreg);  break;
  1910         default: ShouldNotReachHere();
  1912     } else {
  1913       Address raddr;
  1914       if (right->is_single_stack()) {
  1915         raddr = frame_map()->address_for_slot(right->single_stack_ix());
  1916       } else if (right->is_constant()) {
  1917         // hack for now
  1918         raddr = __ as_Address(InternalAddress(float_constant(right->as_jfloat())));
  1919       } else {
  1920         ShouldNotReachHere();
  1922       switch (code) {
  1923         case lir_add: __ addss(lreg, raddr);  break;
  1924         case lir_sub: __ subss(lreg, raddr);  break;
  1925         case lir_mul_strictfp: // fall through
  1926         case lir_mul: __ mulss(lreg, raddr);  break;
  1927         case lir_div_strictfp: // fall through
  1928         case lir_div: __ divss(lreg, raddr);  break;
  1929         default: ShouldNotReachHere();
  1933   } else if (left->is_double_xmm()) {
  1934     assert(left == dest, "left and dest must be equal");
  1936     XMMRegister lreg = left->as_xmm_double_reg();
  1937     if (right->is_double_xmm()) {
  1938       XMMRegister rreg = right->as_xmm_double_reg();
  1939       switch (code) {
  1940         case lir_add: __ addsd(lreg, rreg);  break;
  1941         case lir_sub: __ subsd(lreg, rreg);  break;
  1942         case lir_mul_strictfp: // fall through
  1943         case lir_mul: __ mulsd(lreg, rreg);  break;
  1944         case lir_div_strictfp: // fall through
  1945         case lir_div: __ divsd(lreg, rreg);  break;
  1946         default: ShouldNotReachHere();
  1948     } else {
  1949       Address raddr;
  1950       if (right->is_double_stack()) {
  1951         raddr = frame_map()->address_for_slot(right->double_stack_ix());
  1952       } else if (right->is_constant()) {
  1953         // hack for now
  1954         raddr = __ as_Address(InternalAddress(double_constant(right->as_jdouble())));
  1955       } else {
  1956         ShouldNotReachHere();
  1958       switch (code) {
  1959         case lir_add: __ addsd(lreg, raddr);  break;
  1960         case lir_sub: __ subsd(lreg, raddr);  break;
  1961         case lir_mul_strictfp: // fall through
  1962         case lir_mul: __ mulsd(lreg, raddr);  break;
  1963         case lir_div_strictfp: // fall through
  1964         case lir_div: __ divsd(lreg, raddr);  break;
  1965         default: ShouldNotReachHere();
  1969   } else if (left->is_single_fpu()) {
  1970     assert(dest->is_single_fpu(),  "fpu stack allocation required");
  1972     if (right->is_single_fpu()) {
  1973       arith_fpu_implementation(code, left->fpu_regnr(), right->fpu_regnr(), dest->fpu_regnr(), pop_fpu_stack);
  1975     } else {
  1976       assert(left->fpu_regnr() == 0, "left must be on TOS");
  1977       assert(dest->fpu_regnr() == 0, "dest must be on TOS");
  1979       Address raddr;
  1980       if (right->is_single_stack()) {
  1981         raddr = frame_map()->address_for_slot(right->single_stack_ix());
  1982       } else if (right->is_constant()) {
  1983         address const_addr = float_constant(right->as_jfloat());
  1984         assert(const_addr != NULL, "incorrect float/double constant maintainance");
  1985         // hack for now
  1986         raddr = __ as_Address(InternalAddress(const_addr));
  1987       } else {
  1988         ShouldNotReachHere();
  1991       switch (code) {
  1992         case lir_add: __ fadd_s(raddr); break;
  1993         case lir_sub: __ fsub_s(raddr); break;
  1994         case lir_mul_strictfp: // fall through
  1995         case lir_mul: __ fmul_s(raddr); break;
  1996         case lir_div_strictfp: // fall through
  1997         case lir_div: __ fdiv_s(raddr); break;
  1998         default:      ShouldNotReachHere();
  2002   } else if (left->is_double_fpu()) {
  2003     assert(dest->is_double_fpu(),  "fpu stack allocation required");
  2005     if (code == lir_mul_strictfp || code == lir_div_strictfp) {
  2006       // Double values require special handling for strictfp mul/div on x86
  2007       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
  2008       __ fmulp(left->fpu_regnrLo() + 1);
  2011     if (right->is_double_fpu()) {
  2012       arith_fpu_implementation(code, left->fpu_regnrLo(), right->fpu_regnrLo(), dest->fpu_regnrLo(), pop_fpu_stack);
  2014     } else {
  2015       assert(left->fpu_regnrLo() == 0, "left must be on TOS");
  2016       assert(dest->fpu_regnrLo() == 0, "dest must be on TOS");
  2018       Address raddr;
  2019       if (right->is_double_stack()) {
  2020         raddr = frame_map()->address_for_slot(right->double_stack_ix());
  2021       } else if (right->is_constant()) {
  2022         // hack for now
  2023         raddr = __ as_Address(InternalAddress(double_constant(right->as_jdouble())));
  2024       } else {
  2025         ShouldNotReachHere();
  2028       switch (code) {
  2029         case lir_add: __ fadd_d(raddr); break;
  2030         case lir_sub: __ fsub_d(raddr); break;
  2031         case lir_mul_strictfp: // fall through
  2032         case lir_mul: __ fmul_d(raddr); break;
  2033         case lir_div_strictfp: // fall through
  2034         case lir_div: __ fdiv_d(raddr); break;
  2035         default: ShouldNotReachHere();
  2039     if (code == lir_mul_strictfp || code == lir_div_strictfp) {
  2040       // Double values require special handling for strictfp mul/div on x86
  2041       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
  2042       __ fmulp(dest->fpu_regnrLo() + 1);
  2045   } else if (left->is_single_stack() || left->is_address()) {
  2046     assert(left == dest, "left and dest must be equal");
  2048     Address laddr;
  2049     if (left->is_single_stack()) {
  2050       laddr = frame_map()->address_for_slot(left->single_stack_ix());
  2051     } else if (left->is_address()) {
  2052       laddr = as_Address(left->as_address_ptr());
  2053     } else {
  2054       ShouldNotReachHere();
  2057     if (right->is_single_cpu()) {
  2058       Register rreg = right->as_register();
  2059       switch (code) {
  2060         case lir_add: __ addl(laddr, rreg); break;
  2061         case lir_sub: __ subl(laddr, rreg); break;
  2062         default:      ShouldNotReachHere();
  2064     } else if (right->is_constant()) {
  2065       jint c = right->as_constant_ptr()->as_jint();
  2066       switch (code) {
  2067         case lir_add: {
  2068           __ increment(laddr, c);
  2069           break;
  2071         case lir_sub: {
  2072           __ decrement(laddr, c);
  2073           break;
  2075         default: ShouldNotReachHere();
  2077     } else {
  2078       ShouldNotReachHere();
  2081   } else {
  2082     ShouldNotReachHere();
  2086 void LIR_Assembler::arith_fpu_implementation(LIR_Code code, int left_index, int right_index, int dest_index, bool pop_fpu_stack) {
  2087   assert(pop_fpu_stack  || (left_index     == dest_index || right_index     == dest_index), "invalid LIR");
  2088   assert(!pop_fpu_stack || (left_index - 1 == dest_index || right_index - 1 == dest_index), "invalid LIR");
  2089   assert(left_index == 0 || right_index == 0, "either must be on top of stack");
  2091   bool left_is_tos = (left_index == 0);
  2092   bool dest_is_tos = (dest_index == 0);
  2093   int non_tos_index = (left_is_tos ? right_index : left_index);
  2095   switch (code) {
  2096     case lir_add:
  2097       if (pop_fpu_stack)       __ faddp(non_tos_index);
  2098       else if (dest_is_tos)    __ fadd (non_tos_index);
  2099       else                     __ fadda(non_tos_index);
  2100       break;
  2102     case lir_sub:
  2103       if (left_is_tos) {
  2104         if (pop_fpu_stack)     __ fsubrp(non_tos_index);
  2105         else if (dest_is_tos)  __ fsub  (non_tos_index);
  2106         else                   __ fsubra(non_tos_index);
  2107       } else {
  2108         if (pop_fpu_stack)     __ fsubp (non_tos_index);
  2109         else if (dest_is_tos)  __ fsubr (non_tos_index);
  2110         else                   __ fsuba (non_tos_index);
  2112       break;
  2114     case lir_mul_strictfp: // fall through
  2115     case lir_mul:
  2116       if (pop_fpu_stack)       __ fmulp(non_tos_index);
  2117       else if (dest_is_tos)    __ fmul (non_tos_index);
  2118       else                     __ fmula(non_tos_index);
  2119       break;
  2121     case lir_div_strictfp: // fall through
  2122     case lir_div:
  2123       if (left_is_tos) {
  2124         if (pop_fpu_stack)     __ fdivrp(non_tos_index);
  2125         else if (dest_is_tos)  __ fdiv  (non_tos_index);
  2126         else                   __ fdivra(non_tos_index);
  2127       } else {
  2128         if (pop_fpu_stack)     __ fdivp (non_tos_index);
  2129         else if (dest_is_tos)  __ fdivr (non_tos_index);
  2130         else                   __ fdiva (non_tos_index);
  2132       break;
  2134     case lir_rem:
  2135       assert(left_is_tos && dest_is_tos && right_index == 1, "must be guaranteed by FPU stack allocation");
  2136       __ fremr(noreg);
  2137       break;
  2139     default:
  2140       ShouldNotReachHere();
  2145 void LIR_Assembler::intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr unused, LIR_Opr dest, LIR_Op* op) {
  2146   if (value->is_double_xmm()) {
  2147     switch(code) {
  2148       case lir_abs :
  2150           if (dest->as_xmm_double_reg() != value->as_xmm_double_reg()) {
  2151             __ movdbl(dest->as_xmm_double_reg(), value->as_xmm_double_reg());
  2153           __ andpd(dest->as_xmm_double_reg(),
  2154                     ExternalAddress((address)double_signmask_pool));
  2156         break;
  2158       case lir_sqrt: __ sqrtsd(dest->as_xmm_double_reg(), value->as_xmm_double_reg()); break;
  2159       // all other intrinsics are not available in the SSE instruction set, so FPU is used
  2160       default      : ShouldNotReachHere();
  2163   } else if (value->is_double_fpu()) {
  2164     assert(value->fpu_regnrLo() == 0 && dest->fpu_regnrLo() == 0, "both must be on TOS");
  2165     switch(code) {
  2166       case lir_log   : __ flog() ; break;
  2167       case lir_log10 : __ flog10() ; break;
  2168       case lir_abs   : __ fabs() ; break;
  2169       case lir_sqrt  : __ fsqrt(); break;
  2170       case lir_sin   :
  2171         // Should consider not saving rbx, if not necessary
  2172         __ trigfunc('s', op->as_Op2()->fpu_stack_size());
  2173         break;
  2174       case lir_cos :
  2175         // Should consider not saving rbx, if not necessary
  2176         assert(op->as_Op2()->fpu_stack_size() <= 6, "sin and cos need two free stack slots");
  2177         __ trigfunc('c', op->as_Op2()->fpu_stack_size());
  2178         break;
  2179       case lir_tan :
  2180         // Should consider not saving rbx, if not necessary
  2181         __ trigfunc('t', op->as_Op2()->fpu_stack_size());
  2182         break;
  2183       default      : ShouldNotReachHere();
  2185   } else {
  2186     Unimplemented();
  2190 void LIR_Assembler::logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst) {
  2191   // assert(left->destroys_register(), "check");
  2192   if (left->is_single_cpu()) {
  2193     Register reg = left->as_register();
  2194     if (right->is_constant()) {
  2195       int val = right->as_constant_ptr()->as_jint();
  2196       switch (code) {
  2197         case lir_logic_and: __ andl (reg, val); break;
  2198         case lir_logic_or:  __ orl  (reg, val); break;
  2199         case lir_logic_xor: __ xorl (reg, val); break;
  2200         default: ShouldNotReachHere();
  2202     } else if (right->is_stack()) {
  2203       // added support for stack operands
  2204       Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
  2205       switch (code) {
  2206         case lir_logic_and: __ andl (reg, raddr); break;
  2207         case lir_logic_or:  __ orl  (reg, raddr); break;
  2208         case lir_logic_xor: __ xorl (reg, raddr); break;
  2209         default: ShouldNotReachHere();
  2211     } else {
  2212       Register rright = right->as_register();
  2213       switch (code) {
  2214         case lir_logic_and: __ andl (reg, rright); break;
  2215         case lir_logic_or : __ orl  (reg, rright); break;
  2216         case lir_logic_xor: __ xorl (reg, rright); break;
  2217         default: ShouldNotReachHere();
  2220     move_regs(reg, dst->as_register());
  2221   } else {
  2222     Register l_lo = left->as_register_lo();
  2223     Register l_hi = left->as_register_hi();
  2224     if (right->is_constant()) {
  2225       int r_lo = right->as_constant_ptr()->as_jint_lo();
  2226       int r_hi = right->as_constant_ptr()->as_jint_hi();
  2227       switch (code) {
  2228         case lir_logic_and:
  2229           __ andl(l_lo, r_lo);
  2230           __ andl(l_hi, r_hi);
  2231           break;
  2232         case lir_logic_or:
  2233           __ orl(l_lo, r_lo);
  2234           __ orl(l_hi, r_hi);
  2235           break;
  2236         case lir_logic_xor:
  2237           __ xorl(l_lo, r_lo);
  2238           __ xorl(l_hi, r_hi);
  2239           break;
  2240         default: ShouldNotReachHere();
  2242     } else {
  2243       Register r_lo = right->as_register_lo();
  2244       Register r_hi = right->as_register_hi();
  2245       assert(l_lo != r_hi, "overwriting registers");
  2246       switch (code) {
  2247         case lir_logic_and:
  2248           __ andl(l_lo, r_lo);
  2249           __ andl(l_hi, r_hi);
  2250           break;
  2251         case lir_logic_or:
  2252           __ orl(l_lo, r_lo);
  2253           __ orl(l_hi, r_hi);
  2254           break;
  2255         case lir_logic_xor:
  2256           __ xorl(l_lo, r_lo);
  2257           __ xorl(l_hi, r_hi);
  2258           break;
  2259         default: ShouldNotReachHere();
  2263     Register dst_lo = dst->as_register_lo();
  2264     Register dst_hi = dst->as_register_hi();
  2266     if (dst_lo == l_hi) {
  2267       assert(dst_hi != l_lo, "overwriting registers");
  2268       move_regs(l_hi, dst_hi);
  2269       move_regs(l_lo, dst_lo);
  2270     } else {
  2271       assert(dst_lo != l_hi, "overwriting registers");
  2272       move_regs(l_lo, dst_lo);
  2273       move_regs(l_hi, dst_hi);
  2279 // we assume that rax, and rdx can be overwritten
  2280 void LIR_Assembler::arithmetic_idiv(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr temp, LIR_Opr result, CodeEmitInfo* info) {
  2282   assert(left->is_single_cpu(),   "left must be register");
  2283   assert(right->is_single_cpu() || right->is_constant(),  "right must be register or constant");
  2284   assert(result->is_single_cpu(), "result must be register");
  2286   //  assert(left->destroys_register(), "check");
  2287   //  assert(right->destroys_register(), "check");
  2289   Register lreg = left->as_register();
  2290   Register dreg = result->as_register();
  2292   if (right->is_constant()) {
  2293     int divisor = right->as_constant_ptr()->as_jint();
  2294     assert(divisor > 0 && is_power_of_2(divisor), "must be");
  2295     if (code == lir_idiv) {
  2296       assert(lreg == rax, "must be rax,");
  2297       assert(temp->as_register() == rdx, "tmp register must be rdx");
  2298       __ cdql(); // sign extend into rdx:rax
  2299       if (divisor == 2) {
  2300         __ subl(lreg, rdx);
  2301       } else {
  2302         __ andl(rdx, divisor - 1);
  2303         __ addl(lreg, rdx);
  2305       __ sarl(lreg, log2_intptr(divisor));
  2306       move_regs(lreg, dreg);
  2307     } else if (code == lir_irem) {
  2308       Label done;
  2309       __ movl(dreg, lreg);
  2310       __ andl(dreg, 0x80000000 | (divisor - 1));
  2311       __ jcc(Assembler::positive, done);
  2312       __ decrement(dreg);
  2313       __ orl(dreg, ~(divisor - 1));
  2314       __ increment(dreg);
  2315       __ bind(done);
  2316     } else {
  2317       ShouldNotReachHere();
  2319   } else {
  2320     Register rreg = right->as_register();
  2321     assert(lreg == rax, "left register must be rax,");
  2322     assert(rreg != rdx, "right register must not be rdx");
  2323     assert(temp->as_register() == rdx, "tmp register must be rdx");
  2325     move_regs(lreg, rax);
  2327     int idivl_offset = __ corrected_idivl(rreg);
  2328     add_debug_info_for_div0(idivl_offset, info);
  2329     if (code == lir_irem) {
  2330       move_regs(rdx, dreg); // result is in rdx
  2331     } else {
  2332       move_regs(rax, dreg);
  2338 void LIR_Assembler::comp_op(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Op2* op) {
  2339   if (opr1->is_single_cpu()) {
  2340     Register reg1 = opr1->as_register();
  2341     if (opr2->is_single_cpu()) {
  2342       // cpu register - cpu register
  2343       __ cmpl(reg1, opr2->as_register());
  2344     } else if (opr2->is_stack()) {
  2345       // cpu register - stack
  2346       __ cmpl(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
  2347     } else if (opr2->is_constant()) {
  2348       // cpu register - constant
  2349       LIR_Const* c = opr2->as_constant_ptr();
  2350       if (c->type() == T_INT) {
  2351         __ cmpl(reg1, c->as_jint());
  2352       } else if (c->type() == T_OBJECT) {
  2353         jobject o = c->as_jobject();
  2354         if (o == NULL) {
  2355           __ cmpl(reg1, NULL_WORD);
  2356         } else {
  2357           __ cmpoop(reg1, c->as_jobject());
  2359       } else {
  2360         ShouldNotReachHere();
  2362       // cpu register - address
  2363     } else if (opr2->is_address()) {
  2364       if (op->info() != NULL) {
  2365         add_debug_info_for_null_check_here(op->info());
  2367       __ cmpl(reg1, as_Address(opr2->as_address_ptr()));
  2368     } else {
  2369       ShouldNotReachHere();
  2372   } else if(opr1->is_double_cpu()) {
  2373     Register xlo = opr1->as_register_lo();
  2374     Register xhi = opr1->as_register_hi();
  2375     if (opr2->is_double_cpu()) {
  2376       // cpu register - cpu register
  2377       Register ylo = opr2->as_register_lo();
  2378       Register yhi = opr2->as_register_hi();
  2379       __ subl(xlo, ylo);
  2380       __ sbbl(xhi, yhi);
  2381       if (condition == lir_cond_equal || condition == lir_cond_notEqual) {
  2382         __ orl(xhi, xlo);
  2384     } else if (opr2->is_constant()) {
  2385       // cpu register - constant 0
  2386       assert(opr2->as_jlong() == (jlong)0, "only handles zero");
  2387       assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "only handles equals case");
  2388       __ orl(xhi, xlo);
  2389     } else {
  2390       ShouldNotReachHere();
  2393   } else if (opr1->is_single_xmm()) {
  2394     XMMRegister reg1 = opr1->as_xmm_float_reg();
  2395     if (opr2->is_single_xmm()) {
  2396       // xmm register - xmm register
  2397       __ ucomiss(reg1, opr2->as_xmm_float_reg());
  2398     } else if (opr2->is_stack()) {
  2399       // xmm register - stack
  2400       __ ucomiss(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
  2401     } else if (opr2->is_constant()) {
  2402       // xmm register - constant
  2403       __ ucomiss(reg1, InternalAddress(float_constant(opr2->as_jfloat())));
  2404     } else if (opr2->is_address()) {
  2405       // xmm register - address
  2406       if (op->info() != NULL) {
  2407         add_debug_info_for_null_check_here(op->info());
  2409       __ ucomiss(reg1, as_Address(opr2->as_address_ptr()));
  2410     } else {
  2411       ShouldNotReachHere();
  2414   } else if (opr1->is_double_xmm()) {
  2415     XMMRegister reg1 = opr1->as_xmm_double_reg();
  2416     if (opr2->is_double_xmm()) {
  2417       // xmm register - xmm register
  2418       __ ucomisd(reg1, opr2->as_xmm_double_reg());
  2419     } else if (opr2->is_stack()) {
  2420       // xmm register - stack
  2421       __ ucomisd(reg1, frame_map()->address_for_slot(opr2->double_stack_ix()));
  2422     } else if (opr2->is_constant()) {
  2423       // xmm register - constant
  2424       __ ucomisd(reg1, InternalAddress(double_constant(opr2->as_jdouble())));
  2425     } else if (opr2->is_address()) {
  2426       // xmm register - address
  2427       if (op->info() != NULL) {
  2428         add_debug_info_for_null_check_here(op->info());
  2430       __ ucomisd(reg1, as_Address(opr2->pointer()->as_address()));
  2431     } else {
  2432       ShouldNotReachHere();
  2435   } else if(opr1->is_single_fpu() || opr1->is_double_fpu()) {
  2436     assert(opr1->is_fpu_register() && opr1->fpu() == 0, "currently left-hand side must be on TOS (relax this restriction)");
  2437     assert(opr2->is_fpu_register(), "both must be registers");
  2438     __ fcmp(noreg, opr2->fpu(), op->fpu_pop_count() > 0, op->fpu_pop_count() > 1);
  2440   } else if (opr1->is_address() && opr2->is_constant()) {
  2441     if (op->info() != NULL) {
  2442       add_debug_info_for_null_check_here(op->info());
  2444     // special case: address - constant
  2445     LIR_Address* addr = opr1->as_address_ptr();
  2446     LIR_Const* c = opr2->as_constant_ptr();
  2447     if (c->type() == T_INT) {
  2448       __ cmpl(as_Address(addr), c->as_jint());
  2449     } else if (c->type() == T_OBJECT) {
  2450       __ cmpoop(as_Address(addr), c->as_jobject());
  2451     } else {
  2452       ShouldNotReachHere();
  2455   } else {
  2456     ShouldNotReachHere();
  2460 void LIR_Assembler::comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst, LIR_Op2* op) {
  2461   if (code == lir_cmp_fd2i || code == lir_ucmp_fd2i) {
  2462     if (left->is_single_xmm()) {
  2463       assert(right->is_single_xmm(), "must match");
  2464       __ cmpss2int(left->as_xmm_float_reg(), right->as_xmm_float_reg(), dst->as_register(), code == lir_ucmp_fd2i);
  2465     } else if (left->is_double_xmm()) {
  2466       assert(right->is_double_xmm(), "must match");
  2467       __ cmpsd2int(left->as_xmm_double_reg(), right->as_xmm_double_reg(), dst->as_register(), code == lir_ucmp_fd2i);
  2469     } else {
  2470       assert(left->is_single_fpu() || left->is_double_fpu(), "must be");
  2471       assert(right->is_single_fpu() || right->is_double_fpu(), "must match");
  2473       assert(left->fpu() == 0, "left must be on TOS");
  2474       __ fcmp2int(dst->as_register(), code == lir_ucmp_fd2i, right->fpu(),
  2475                   op->fpu_pop_count() > 0, op->fpu_pop_count() > 1);
  2477   } else {
  2478     assert(code == lir_cmp_l2i, "check");
  2479     __ lcmp2int(left->as_register_hi(),
  2480                 left->as_register_lo(),
  2481                 right->as_register_hi(),
  2482                 right->as_register_lo());
  2483     move_regs(left->as_register_hi(), dst->as_register());
  2488 void LIR_Assembler::align_call(LIR_Code code) {
  2489   if (os::is_MP()) {
  2490     // make sure that the displacement word of the call ends up word aligned
  2491     int offset = __ offset();
  2492     switch (code) {
  2493       case lir_static_call:
  2494       case lir_optvirtual_call:
  2495         offset += NativeCall::displacement_offset;
  2496         break;
  2497       case lir_icvirtual_call:
  2498         offset += NativeCall::displacement_offset + NativeMovConstReg::instruction_size;
  2499       break;
  2500       case lir_virtual_call:  // currently, sparc-specific for niagara
  2501       default: ShouldNotReachHere();
  2503     while (offset++ % BytesPerWord != 0) {
  2504       __ nop();
  2510 void LIR_Assembler::call(address entry, relocInfo::relocType rtype, CodeEmitInfo* info) {
  2511   assert(!os::is_MP() || (__ offset() + NativeCall::displacement_offset) % BytesPerWord == 0,
  2512          "must be aligned");
  2513   __ call(AddressLiteral(entry, rtype));
  2514   add_call_info(code_offset(), info);
  2518 void LIR_Assembler::ic_call(address entry, CodeEmitInfo* info) {
  2519   RelocationHolder rh = virtual_call_Relocation::spec(pc());
  2520   __ movoop(IC_Klass, (jobject)Universe::non_oop_word());
  2521   assert(!os::is_MP() ||
  2522          (__ offset() + NativeCall::displacement_offset) % BytesPerWord == 0,
  2523          "must be aligned");
  2524   __ call(AddressLiteral(entry, rh));
  2525   add_call_info(code_offset(), info);
  2529 /* Currently, vtable-dispatch is only enabled for sparc platforms */
  2530 void LIR_Assembler::vtable_call(int vtable_offset, CodeEmitInfo* info) {
  2531   ShouldNotReachHere();
  2534 void LIR_Assembler::emit_static_call_stub() {
  2535   address call_pc = __ pc();
  2536   address stub = __ start_a_stub(call_stub_size);
  2537   if (stub == NULL) {
  2538     bailout("static call stub overflow");
  2539     return;
  2542   int start = __ offset();
  2543   if (os::is_MP()) {
  2544     // make sure that the displacement word of the call ends up word aligned
  2545     int offset = __ offset() + NativeMovConstReg::instruction_size + NativeCall::displacement_offset;
  2546     while (offset++ % BytesPerWord != 0) {
  2547       __ nop();
  2550   __ relocate(static_stub_Relocation::spec(call_pc));
  2551   __ movoop(rbx, (jobject)NULL);
  2552   // must be set to -1 at code generation time
  2553   assert(!os::is_MP() || ((__ offset() + 1) % BytesPerWord) == 0, "must be aligned on MP");
  2554   __ jump(RuntimeAddress((address)-1));
  2556   assert(__ offset() - start <= call_stub_size, "stub too big")
  2557   __ end_a_stub();
  2561 void LIR_Assembler::throw_op(LIR_Opr exceptionPC, LIR_Opr exceptionOop, CodeEmitInfo* info, bool unwind) {
  2562   assert(exceptionOop->as_register() == rax, "must match");
  2563   assert(unwind || exceptionPC->as_register() == rdx, "must match");
  2565   // exception object is not added to oop map by LinearScan
  2566   // (LinearScan assumes that no oops are in fixed registers)
  2567   info->add_register_oop(exceptionOop);
  2568   Runtime1::StubID unwind_id;
  2570   if (!unwind) {
  2571     // get current pc information
  2572     // pc is only needed if the method has an exception handler, the unwind code does not need it.
  2573     int pc_for_athrow_offset = __ offset();
  2574     InternalAddress pc_for_athrow(__ pc());
  2575     __ lea(exceptionPC->as_register(), pc_for_athrow);
  2576     add_call_info(pc_for_athrow_offset, info); // for exception handler
  2578     __ verify_not_null_oop(rax);
  2579     // search an exception handler (rax: exception oop, rdx: throwing pc)
  2580     if (compilation()->has_fpu_code()) {
  2581       unwind_id = Runtime1::handle_exception_id;
  2582     } else {
  2583       unwind_id = Runtime1::handle_exception_nofpu_id;
  2585   } else {
  2586     unwind_id = Runtime1::unwind_exception_id;
  2588   __ call(RuntimeAddress(Runtime1::entry_for(unwind_id)));
  2590   // enough room for two byte trap
  2591   __ nop();
  2595 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
  2597   // optimized version for linear scan:
  2598   // * count must be already in ECX (guaranteed by LinearScan)
  2599   // * left and dest must be equal
  2600   // * tmp must be unused
  2601   assert(count->as_register() == SHIFT_count, "count must be in ECX");
  2602   assert(left == dest, "left and dest must be equal");
  2603   assert(tmp->is_illegal(), "wasting a register if tmp is allocated");
  2605   if (left->is_single_cpu()) {
  2606     Register value = left->as_register();
  2607     assert(value != SHIFT_count, "left cannot be ECX");
  2609     switch (code) {
  2610       case lir_shl:  __ shll(value); break;
  2611       case lir_shr:  __ sarl(value); break;
  2612       case lir_ushr: __ shrl(value); break;
  2613       default: ShouldNotReachHere();
  2615   } else if (left->is_double_cpu()) {
  2616     Register lo = left->as_register_lo();
  2617     Register hi = left->as_register_hi();
  2618     assert(lo != SHIFT_count && hi != SHIFT_count, "left cannot be ECX");
  2620     switch (code) {
  2621       case lir_shl:  __ lshl(hi, lo);        break;
  2622       case lir_shr:  __ lshr(hi, lo, true);  break;
  2623       case lir_ushr: __ lshr(hi, lo, false); break;
  2624       default: ShouldNotReachHere();
  2626   } else {
  2627     ShouldNotReachHere();
  2632 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, jint count, LIR_Opr dest) {
  2633   if (dest->is_single_cpu()) {
  2634     // first move left into dest so that left is not destroyed by the shift
  2635     Register value = dest->as_register();
  2636     count = count & 0x1F; // Java spec
  2638     move_regs(left->as_register(), value);
  2639     switch (code) {
  2640       case lir_shl:  __ shll(value, count); break;
  2641       case lir_shr:  __ sarl(value, count); break;
  2642       case lir_ushr: __ shrl(value, count); break;
  2643       default: ShouldNotReachHere();
  2645   } else if (dest->is_double_cpu()) {
  2646     Unimplemented();
  2647   } else {
  2648     ShouldNotReachHere();
  2653 void LIR_Assembler::store_parameter(Register r, int offset_from_rsp_in_words) {
  2654   assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
  2655   int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
  2656   assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
  2657   __ movl (Address(rsp, offset_from_rsp_in_bytes), r);
  2661 void LIR_Assembler::store_parameter(jint c,     int offset_from_rsp_in_words) {
  2662   assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
  2663   int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
  2664   assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
  2665   __ movl (Address(rsp, offset_from_rsp_in_bytes), c);
  2669 void LIR_Assembler::store_parameter(jobject o,  int offset_from_rsp_in_words) {
  2670   assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
  2671   int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
  2672   assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
  2673   __ movoop (Address(rsp, offset_from_rsp_in_bytes), o);
  2677 // This code replaces a call to arraycopy; no exception may
  2678 // be thrown in this code, they must be thrown in the System.arraycopy
  2679 // activation frame; we could save some checks if this would not be the case
  2680 void LIR_Assembler::emit_arraycopy(LIR_OpArrayCopy* op) {
  2681   ciArrayKlass* default_type = op->expected_type();
  2682   Register src = op->src()->as_register();
  2683   Register dst = op->dst()->as_register();
  2684   Register src_pos = op->src_pos()->as_register();
  2685   Register dst_pos = op->dst_pos()->as_register();
  2686   Register length  = op->length()->as_register();
  2687   Register tmp = op->tmp()->as_register();
  2689   CodeStub* stub = op->stub();
  2690   int flags = op->flags();
  2691   BasicType basic_type = default_type != NULL ? default_type->element_type()->basic_type() : T_ILLEGAL;
  2692   if (basic_type == T_ARRAY) basic_type = T_OBJECT;
  2694   // if we don't know anything or it's an object array, just go through the generic arraycopy
  2695   if (default_type == NULL) {
  2696     Label done;
  2697     // save outgoing arguments on stack in case call to System.arraycopy is needed
  2698     // HACK ALERT. This code used to push the parameters in a hardwired fashion
  2699     // for interpreter calling conventions. Now we have to do it in new style conventions.
  2700     // For the moment until C1 gets the new register allocator I just force all the
  2701     // args to the right place (except the register args) and then on the back side
  2702     // reload the register args properly if we go slow path. Yuck
  2704     // These are proper for the calling convention
  2706     store_parameter(length, 2);
  2707     store_parameter(dst_pos, 1);
  2708     store_parameter(dst, 0);
  2710     // these are just temporary placements until we need to reload
  2711     store_parameter(src_pos, 3);
  2712     store_parameter(src, 4);
  2713     assert(src == rcx && src_pos == rdx, "mismatch in calling convention");
  2715     // pass arguments: may push as this is not a safepoint; SP must be fix at each safepoint
  2716     __ pushl(length);
  2717     __ pushl(dst_pos);
  2718     __ pushl(dst);
  2719     __ pushl(src_pos);
  2720     __ pushl(src);
  2721     address entry = CAST_FROM_FN_PTR(address, Runtime1::arraycopy);
  2722     __ call_VM_leaf(entry, 5); // removes pushed parameter from the stack
  2724     __ cmpl(rax, 0);
  2725     __ jcc(Assembler::equal, *stub->continuation());
  2727     // Reload values from the stack so they are where the stub
  2728     // expects them.
  2729     __ movl (dst,     Address(rsp, 0*BytesPerWord));
  2730     __ movl (dst_pos, Address(rsp, 1*BytesPerWord));
  2731     __ movl (length,  Address(rsp, 2*BytesPerWord));
  2732     __ movl (src_pos, Address(rsp, 3*BytesPerWord));
  2733     __ movl (src,     Address(rsp, 4*BytesPerWord));
  2734     __ jmp(*stub->entry());
  2736     __ bind(*stub->continuation());
  2737     return;
  2740   assert(default_type != NULL && default_type->is_array_klass() && default_type->is_loaded(), "must be true at this point");
  2742   int elem_size = type2aelembytes(basic_type);
  2743   int shift_amount;
  2744   Address::ScaleFactor scale;
  2746   switch (elem_size) {
  2747     case 1 :
  2748       shift_amount = 0;
  2749       scale = Address::times_1;
  2750       break;
  2751     case 2 :
  2752       shift_amount = 1;
  2753       scale = Address::times_2;
  2754       break;
  2755     case 4 :
  2756       shift_amount = 2;
  2757       scale = Address::times_4;
  2758       break;
  2759     case 8 :
  2760       shift_amount = 3;
  2761       scale = Address::times_8;
  2762       break;
  2763     default:
  2764       ShouldNotReachHere();
  2767   Address src_length_addr = Address(src, arrayOopDesc::length_offset_in_bytes());
  2768   Address dst_length_addr = Address(dst, arrayOopDesc::length_offset_in_bytes());
  2769   Address src_klass_addr = Address(src, oopDesc::klass_offset_in_bytes());
  2770   Address dst_klass_addr = Address(dst, oopDesc::klass_offset_in_bytes());
  2772   // test for NULL
  2773   if (flags & LIR_OpArrayCopy::src_null_check) {
  2774     __ testl(src, src);
  2775     __ jcc(Assembler::zero, *stub->entry());
  2777   if (flags & LIR_OpArrayCopy::dst_null_check) {
  2778     __ testl(dst, dst);
  2779     __ jcc(Assembler::zero, *stub->entry());
  2782   // check if negative
  2783   if (flags & LIR_OpArrayCopy::src_pos_positive_check) {
  2784     __ testl(src_pos, src_pos);
  2785     __ jcc(Assembler::less, *stub->entry());
  2787   if (flags & LIR_OpArrayCopy::dst_pos_positive_check) {
  2788     __ testl(dst_pos, dst_pos);
  2789     __ jcc(Assembler::less, *stub->entry());
  2791   if (flags & LIR_OpArrayCopy::length_positive_check) {
  2792     __ testl(length, length);
  2793     __ jcc(Assembler::less, *stub->entry());
  2796   if (flags & LIR_OpArrayCopy::src_range_check) {
  2797     __ leal(tmp, Address(src_pos, length, Address::times_1, 0));
  2798     __ cmpl(tmp, src_length_addr);
  2799     __ jcc(Assembler::above, *stub->entry());
  2801   if (flags & LIR_OpArrayCopy::dst_range_check) {
  2802     __ leal(tmp, Address(dst_pos, length, Address::times_1, 0));
  2803     __ cmpl(tmp, dst_length_addr);
  2804     __ jcc(Assembler::above, *stub->entry());
  2807   if (flags & LIR_OpArrayCopy::type_check) {
  2808     __ movl(tmp, src_klass_addr);
  2809     __ cmpl(tmp, dst_klass_addr);
  2810     __ jcc(Assembler::notEqual, *stub->entry());
  2813 #ifdef ASSERT
  2814   if (basic_type != T_OBJECT || !(flags & LIR_OpArrayCopy::type_check)) {
  2815     // Sanity check the known type with the incoming class.  For the
  2816     // primitive case the types must match exactly with src.klass and
  2817     // dst.klass each exactly matching the default type.  For the
  2818     // object array case, if no type check is needed then either the
  2819     // dst type is exactly the expected type and the src type is a
  2820     // subtype which we can't check or src is the same array as dst
  2821     // but not necessarily exactly of type default_type.
  2822     Label known_ok, halt;
  2823     __ movoop(tmp, default_type->encoding());
  2824     if (basic_type != T_OBJECT) {
  2825       __ cmpl(tmp, dst_klass_addr);
  2826       __ jcc(Assembler::notEqual, halt);
  2827       __ cmpl(tmp, src_klass_addr);
  2828       __ jcc(Assembler::equal, known_ok);
  2829     } else {
  2830       __ cmpl(tmp, dst_klass_addr);
  2831       __ jcc(Assembler::equal, known_ok);
  2832       __ cmpl(src, dst);
  2833       __ jcc(Assembler::equal, known_ok);
  2835     __ bind(halt);
  2836     __ stop("incorrect type information in arraycopy");
  2837     __ bind(known_ok);
  2839 #endif
  2841   __ leal(tmp, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
  2842   store_parameter(tmp, 0);
  2843   __ leal(tmp, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
  2844   store_parameter(tmp, 1);
  2845   if (shift_amount > 0 && basic_type != T_OBJECT) {
  2846     __ shll(length, shift_amount);
  2848   store_parameter(length, 2);
  2849   if (basic_type == T_OBJECT) {
  2850     __ call_VM_leaf(CAST_FROM_FN_PTR(address, Runtime1::oop_arraycopy), 0);
  2851   } else {
  2852     __ call_VM_leaf(CAST_FROM_FN_PTR(address, Runtime1::primitive_arraycopy), 0);
  2855   __ bind(*stub->continuation());
  2859 void LIR_Assembler::emit_lock(LIR_OpLock* op) {
  2860   Register obj = op->obj_opr()->as_register();  // may not be an oop
  2861   Register hdr = op->hdr_opr()->as_register();
  2862   Register lock = op->lock_opr()->as_register();
  2863   if (!UseFastLocking) {
  2864     __ jmp(*op->stub()->entry());
  2865   } else if (op->code() == lir_lock) {
  2866     Register scratch = noreg;
  2867     if (UseBiasedLocking) {
  2868       scratch = op->scratch_opr()->as_register();
  2870     assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
  2871     // add debug info for NullPointerException only if one is possible
  2872     int null_check_offset = __ lock_object(hdr, obj, lock, scratch, *op->stub()->entry());
  2873     if (op->info() != NULL) {
  2874       add_debug_info_for_null_check(null_check_offset, op->info());
  2876     // done
  2877   } else if (op->code() == lir_unlock) {
  2878     assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
  2879     __ unlock_object(hdr, obj, lock, *op->stub()->entry());
  2880   } else {
  2881     Unimplemented();
  2883   __ bind(*op->stub()->continuation());
  2887 void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
  2888   ciMethod* method = op->profiled_method();
  2889   int bci          = op->profiled_bci();
  2891   // Update counter for all call types
  2892   ciMethodData* md = method->method_data();
  2893   if (md == NULL) {
  2894     bailout("out of memory building methodDataOop");
  2895     return;
  2897   ciProfileData* data = md->bci_to_data(bci);
  2898   assert(data->is_CounterData(), "need CounterData for calls");
  2899   assert(op->mdo()->is_single_cpu(),  "mdo must be allocated");
  2900   Register mdo  = op->mdo()->as_register();
  2901   __ movoop(mdo, md->encoding());
  2902   Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
  2903   __ addl(counter_addr, DataLayout::counter_increment);
  2904   Bytecodes::Code bc = method->java_code_at_bci(bci);
  2905   // Perform additional virtual call profiling for invokevirtual and
  2906   // invokeinterface bytecodes
  2907   if ((bc == Bytecodes::_invokevirtual || bc == Bytecodes::_invokeinterface) &&
  2908       Tier1ProfileVirtualCalls) {
  2909     assert(op->recv()->is_single_cpu(), "recv must be allocated");
  2910     Register recv = op->recv()->as_register();
  2911     assert_different_registers(mdo, recv);
  2912     assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
  2913     ciKlass* known_klass = op->known_holder();
  2914     if (Tier1OptimizeVirtualCallProfiling && known_klass != NULL) {
  2915       // We know the type that will be seen at this call site; we can
  2916       // statically update the methodDataOop rather than needing to do
  2917       // dynamic tests on the receiver type
  2919       // NOTE: we should probably put a lock around this search to
  2920       // avoid collisions by concurrent compilations
  2921       ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
  2922       uint i;
  2923       for (i = 0; i < VirtualCallData::row_limit(); i++) {
  2924         ciKlass* receiver = vc_data->receiver(i);
  2925         if (known_klass->equals(receiver)) {
  2926           Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
  2927           __ addl(data_addr, DataLayout::counter_increment);
  2928           return;
  2932       // Receiver type not found in profile data; select an empty slot
  2934       // Note that this is less efficient than it should be because it
  2935       // always does a write to the receiver part of the
  2936       // VirtualCallData rather than just the first time
  2937       for (i = 0; i < VirtualCallData::row_limit(); i++) {
  2938         ciKlass* receiver = vc_data->receiver(i);
  2939         if (receiver == NULL) {
  2940           Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)));
  2941           __ movoop(recv_addr, known_klass->encoding());
  2942           Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
  2943           __ addl(data_addr, DataLayout::counter_increment);
  2944           return;
  2947     } else {
  2948       __ movl(recv, Address(recv, oopDesc::klass_offset_in_bytes()));
  2949       Label update_done;
  2950       uint i;
  2951       for (i = 0; i < VirtualCallData::row_limit(); i++) {
  2952         Label next_test;
  2953         // See if the receiver is receiver[n].
  2954         __ cmpl(recv, Address(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i))));
  2955         __ jcc(Assembler::notEqual, next_test);
  2956         Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
  2957         __ addl(data_addr, DataLayout::counter_increment);
  2958         __ jmp(update_done);
  2959         __ bind(next_test);
  2962       // Didn't find receiver; find next empty slot and fill it in
  2963       for (i = 0; i < VirtualCallData::row_limit(); i++) {
  2964         Label next_test;
  2965         Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)));
  2966         __ cmpl(recv_addr, NULL_WORD);
  2967         __ jcc(Assembler::notEqual, next_test);
  2968         __ movl(recv_addr, recv);
  2969         __ movl(Address(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i))), DataLayout::counter_increment);
  2970         if (i < (VirtualCallData::row_limit() - 1)) {
  2971           __ jmp(update_done);
  2973         __ bind(next_test);
  2976       __ bind(update_done);
  2982 void LIR_Assembler::emit_delay(LIR_OpDelay*) {
  2983   Unimplemented();
  2987 void LIR_Assembler::monitor_address(int monitor_no, LIR_Opr dst) {
  2988   __ leal(dst->as_register(), frame_map()->address_for_monitor_lock(monitor_no));
  2992 void LIR_Assembler::align_backward_branch_target() {
  2993   __ align(BytesPerWord);
  2997 void LIR_Assembler::negate(LIR_Opr left, LIR_Opr dest) {
  2998   if (left->is_single_cpu()) {
  2999     __ negl(left->as_register());
  3000     move_regs(left->as_register(), dest->as_register());
  3002   } else if (left->is_double_cpu()) {
  3003     Register lo = left->as_register_lo();
  3004     Register hi = left->as_register_hi();
  3005     __ lneg(hi, lo);
  3006     if (dest->as_register_lo() == hi) {
  3007       assert(dest->as_register_hi() != lo, "destroying register");
  3008       move_regs(hi, dest->as_register_hi());
  3009       move_regs(lo, dest->as_register_lo());
  3010     } else {
  3011       move_regs(lo, dest->as_register_lo());
  3012       move_regs(hi, dest->as_register_hi());
  3015   } else if (dest->is_single_xmm()) {
  3016     if (left->as_xmm_float_reg() != dest->as_xmm_float_reg()) {
  3017       __ movflt(dest->as_xmm_float_reg(), left->as_xmm_float_reg());
  3019     __ xorps(dest->as_xmm_float_reg(),
  3020              ExternalAddress((address)float_signflip_pool));
  3022   } else if (dest->is_double_xmm()) {
  3023     if (left->as_xmm_double_reg() != dest->as_xmm_double_reg()) {
  3024       __ movdbl(dest->as_xmm_double_reg(), left->as_xmm_double_reg());
  3026     __ xorpd(dest->as_xmm_double_reg(),
  3027              ExternalAddress((address)double_signflip_pool));
  3029   } else if (left->is_single_fpu() || left->is_double_fpu()) {
  3030     assert(left->fpu() == 0, "arg must be on TOS");
  3031     assert(dest->fpu() == 0, "dest must be TOS");
  3032     __ fchs();
  3034   } else {
  3035     ShouldNotReachHere();
  3040 void LIR_Assembler::leal(LIR_Opr addr, LIR_Opr dest) {
  3041   assert(addr->is_address() && dest->is_register(), "check");
  3042   Register reg = dest->as_register();
  3043   __ leal(dest->as_register(), as_Address(addr->as_address_ptr()));
  3048 void LIR_Assembler::rt_call(LIR_Opr result, address dest, const LIR_OprList* args, LIR_Opr tmp, CodeEmitInfo* info) {
  3049   assert(!tmp->is_valid(), "don't need temporary");
  3050   __ call(RuntimeAddress(dest));
  3051   if (info != NULL) {
  3052     add_call_info_here(info);
  3057 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {
  3058   assert(type == T_LONG, "only for volatile long fields");
  3060   if (info != NULL) {
  3061     add_debug_info_for_null_check_here(info);
  3064   if (src->is_double_xmm()) {
  3065     if (dest->is_double_cpu()) {
  3066       __ movd(dest->as_register_lo(), src->as_xmm_double_reg());
  3067       __ psrlq(src->as_xmm_double_reg(), 32);
  3068       __ movd(dest->as_register_hi(), src->as_xmm_double_reg());
  3069     } else if (dest->is_double_stack()) {
  3070       __ movdbl(frame_map()->address_for_slot(dest->double_stack_ix()), src->as_xmm_double_reg());
  3071     } else if (dest->is_address()) {
  3072       __ movdbl(as_Address(dest->as_address_ptr()), src->as_xmm_double_reg());
  3073     } else {
  3074       ShouldNotReachHere();
  3077   } else if (dest->is_double_xmm()) {
  3078     if (src->is_double_stack()) {
  3079       __ movdbl(dest->as_xmm_double_reg(), frame_map()->address_for_slot(src->double_stack_ix()));
  3080     } else if (src->is_address()) {
  3081       __ movdbl(dest->as_xmm_double_reg(), as_Address(src->as_address_ptr()));
  3082     } else {
  3083       ShouldNotReachHere();
  3086   } else if (src->is_double_fpu()) {
  3087     assert(src->fpu_regnrLo() == 0, "must be TOS");
  3088     if (dest->is_double_stack()) {
  3089       __ fistp_d(frame_map()->address_for_slot(dest->double_stack_ix()));
  3090     } else if (dest->is_address()) {
  3091       __ fistp_d(as_Address(dest->as_address_ptr()));
  3092     } else {
  3093       ShouldNotReachHere();
  3096   } else if (dest->is_double_fpu()) {
  3097     assert(dest->fpu_regnrLo() == 0, "must be TOS");
  3098     if (src->is_double_stack()) {
  3099       __ fild_d(frame_map()->address_for_slot(src->double_stack_ix()));
  3100     } else if (src->is_address()) {
  3101       __ fild_d(as_Address(src->as_address_ptr()));
  3102     } else {
  3103       ShouldNotReachHere();
  3105   } else {
  3106     ShouldNotReachHere();
  3111 void LIR_Assembler::membar() {
  3112   __ membar();
  3115 void LIR_Assembler::membar_acquire() {
  3116   // No x86 machines currently require load fences
  3117   // __ load_fence();
  3120 void LIR_Assembler::membar_release() {
  3121   // No x86 machines currently require store fences
  3122   // __ store_fence();
  3125 void LIR_Assembler::get_thread(LIR_Opr result_reg) {
  3126   assert(result_reg->is_register(), "check");
  3127   __ get_thread(result_reg->as_register());
  3131 void LIR_Assembler::peephole(LIR_List*) {
  3132   // do nothing for now
  3136 #undef __

mercurial