src/cpu/x86/vm/c1_LIRAssembler_x86.cpp

Fri, 06 Jan 2012 20:09:20 -0800

author
kvn
date
Fri, 06 Jan 2012 20:09:20 -0800
changeset 3400
22cee0ee8927
parent 3368
52b5d32fbfaf
parent 3391
069ab3f976d3
child 3435
898522ae3c32
permissions
-rw-r--r--

Merge

     1 /*
     2  * Copyright (c) 2000, 2011, Oracle and/or its affiliates. All rights reserved.
     3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
     4  *
     5  * This code is free software; you can redistribute it and/or modify it
     6  * under the terms of the GNU General Public License version 2 only, as
     7  * published by the Free Software Foundation.
     8  *
     9  * This code is distributed in the hope that it will be useful, but WITHOUT
    10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
    11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
    12  * version 2 for more details (a copy is included in the LICENSE file that
    13  * accompanied this code).
    14  *
    15  * You should have received a copy of the GNU General Public License version
    16  * 2 along with this work; if not, write to the Free Software Foundation,
    17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
    18  *
    19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
    20  * or visit www.oracle.com if you need additional information or have any
    21  * questions.
    22  *
    23  */
    25 #include "precompiled.hpp"
    26 #include "asm/assembler.hpp"
    27 #include "c1/c1_Compilation.hpp"
    28 #include "c1/c1_LIRAssembler.hpp"
    29 #include "c1/c1_MacroAssembler.hpp"
    30 #include "c1/c1_Runtime1.hpp"
    31 #include "c1/c1_ValueStack.hpp"
    32 #include "ci/ciArrayKlass.hpp"
    33 #include "ci/ciInstance.hpp"
    34 #include "gc_interface/collectedHeap.hpp"
    35 #include "memory/barrierSet.hpp"
    36 #include "memory/cardTableModRefBS.hpp"
    37 #include "nativeInst_x86.hpp"
    38 #include "oops/objArrayKlass.hpp"
    39 #include "runtime/sharedRuntime.hpp"
    42 // These masks are used to provide 128-bit aligned bitmasks to the XMM
    43 // instructions, to allow sign-masking or sign-bit flipping.  They allow
    44 // fast versions of NegF/NegD and AbsF/AbsD.
    46 // Note: 'double' and 'long long' have 32-bits alignment on x86.
    47 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
    48   // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
    49   // of 128-bits operands for SSE instructions.
    50   jlong *operand = (jlong*)(((intptr_t)adr) & ((intptr_t)(~0xF)));
    51   // Store the value to a 128-bits operand.
    52   operand[0] = lo;
    53   operand[1] = hi;
    54   return operand;
    55 }
    57 // Buffer for 128-bits masks used by SSE instructions.
    58 static jlong fp_signmask_pool[(4+1)*2]; // 4*128bits(data) + 128bits(alignment)
    60 // Static initialization during VM startup.
    61 static jlong *float_signmask_pool  = double_quadword(&fp_signmask_pool[1*2], CONST64(0x7FFFFFFF7FFFFFFF), CONST64(0x7FFFFFFF7FFFFFFF));
    62 static jlong *double_signmask_pool = double_quadword(&fp_signmask_pool[2*2], CONST64(0x7FFFFFFFFFFFFFFF), CONST64(0x7FFFFFFFFFFFFFFF));
    63 static jlong *float_signflip_pool  = double_quadword(&fp_signmask_pool[3*2], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
    64 static jlong *double_signflip_pool = double_quadword(&fp_signmask_pool[4*2], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
    68 NEEDS_CLEANUP // remove this definitions ?
    69 const Register IC_Klass    = rax;   // where the IC klass is cached
    70 const Register SYNC_header = rax;   // synchronization header
    71 const Register SHIFT_count = rcx;   // where count for shift operations must be
    73 #define __ _masm->
    76 static void select_different_registers(Register preserve,
    77                                        Register extra,
    78                                        Register &tmp1,
    79                                        Register &tmp2) {
    80   if (tmp1 == preserve) {
    81     assert_different_registers(tmp1, tmp2, extra);
    82     tmp1 = extra;
    83   } else if (tmp2 == preserve) {
    84     assert_different_registers(tmp1, tmp2, extra);
    85     tmp2 = extra;
    86   }
    87   assert_different_registers(preserve, tmp1, tmp2);
    88 }
    92 static void select_different_registers(Register preserve,
    93                                        Register extra,
    94                                        Register &tmp1,
    95                                        Register &tmp2,
    96                                        Register &tmp3) {
    97   if (tmp1 == preserve) {
    98     assert_different_registers(tmp1, tmp2, tmp3, extra);
    99     tmp1 = extra;
   100   } else if (tmp2 == preserve) {
   101     assert_different_registers(tmp1, tmp2, tmp3, extra);
   102     tmp2 = extra;
   103   } else if (tmp3 == preserve) {
   104     assert_different_registers(tmp1, tmp2, tmp3, extra);
   105     tmp3 = extra;
   106   }
   107   assert_different_registers(preserve, tmp1, tmp2, tmp3);
   108 }
   112 bool LIR_Assembler::is_small_constant(LIR_Opr opr) {
   113   if (opr->is_constant()) {
   114     LIR_Const* constant = opr->as_constant_ptr();
   115     switch (constant->type()) {
   116       case T_INT: {
   117         return true;
   118       }
   120       default:
   121         return false;
   122     }
   123   }
   124   return false;
   125 }
   128 LIR_Opr LIR_Assembler::receiverOpr() {
   129   return FrameMap::receiver_opr;
   130 }
   132 LIR_Opr LIR_Assembler::osrBufferPointer() {
   133   return FrameMap::as_pointer_opr(receiverOpr()->as_register());
   134 }
   136 //--------------fpu register translations-----------------------
   139 address LIR_Assembler::float_constant(float f) {
   140   address const_addr = __ float_constant(f);
   141   if (const_addr == NULL) {
   142     bailout("const section overflow");
   143     return __ code()->consts()->start();
   144   } else {
   145     return const_addr;
   146   }
   147 }
   150 address LIR_Assembler::double_constant(double d) {
   151   address const_addr = __ double_constant(d);
   152   if (const_addr == NULL) {
   153     bailout("const section overflow");
   154     return __ code()->consts()->start();
   155   } else {
   156     return const_addr;
   157   }
   158 }
   161 void LIR_Assembler::set_24bit_FPU() {
   162   __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_24()));
   163 }
   165 void LIR_Assembler::reset_FPU() {
   166   __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
   167 }
   169 void LIR_Assembler::fpop() {
   170   __ fpop();
   171 }
   173 void LIR_Assembler::fxch(int i) {
   174   __ fxch(i);
   175 }
   177 void LIR_Assembler::fld(int i) {
   178   __ fld_s(i);
   179 }
   181 void LIR_Assembler::ffree(int i) {
   182   __ ffree(i);
   183 }
   185 void LIR_Assembler::breakpoint() {
   186   __ int3();
   187 }
   189 void LIR_Assembler::push(LIR_Opr opr) {
   190   if (opr->is_single_cpu()) {
   191     __ push_reg(opr->as_register());
   192   } else if (opr->is_double_cpu()) {
   193     NOT_LP64(__ push_reg(opr->as_register_hi()));
   194     __ push_reg(opr->as_register_lo());
   195   } else if (opr->is_stack()) {
   196     __ push_addr(frame_map()->address_for_slot(opr->single_stack_ix()));
   197   } else if (opr->is_constant()) {
   198     LIR_Const* const_opr = opr->as_constant_ptr();
   199     if (const_opr->type() == T_OBJECT) {
   200       __ push_oop(const_opr->as_jobject());
   201     } else if (const_opr->type() == T_INT) {
   202       __ push_jint(const_opr->as_jint());
   203     } else {
   204       ShouldNotReachHere();
   205     }
   207   } else {
   208     ShouldNotReachHere();
   209   }
   210 }
   212 void LIR_Assembler::pop(LIR_Opr opr) {
   213   if (opr->is_single_cpu()) {
   214     __ pop_reg(opr->as_register());
   215   } else {
   216     ShouldNotReachHere();
   217   }
   218 }
   220 bool LIR_Assembler::is_literal_address(LIR_Address* addr) {
   221   return addr->base()->is_illegal() && addr->index()->is_illegal();
   222 }
   224 //-------------------------------------------
   226 Address LIR_Assembler::as_Address(LIR_Address* addr) {
   227   return as_Address(addr, rscratch1);
   228 }
   230 Address LIR_Assembler::as_Address(LIR_Address* addr, Register tmp) {
   231   if (addr->base()->is_illegal()) {
   232     assert(addr->index()->is_illegal(), "must be illegal too");
   233     AddressLiteral laddr((address)addr->disp(), relocInfo::none);
   234     if (! __ reachable(laddr)) {
   235       __ movptr(tmp, laddr.addr());
   236       Address res(tmp, 0);
   237       return res;
   238     } else {
   239       return __ as_Address(laddr);
   240     }
   241   }
   243   Register base = addr->base()->as_pointer_register();
   245   if (addr->index()->is_illegal()) {
   246     return Address( base, addr->disp());
   247   } else if (addr->index()->is_cpu_register()) {
   248     Register index = addr->index()->as_pointer_register();
   249     return Address(base, index, (Address::ScaleFactor) addr->scale(), addr->disp());
   250   } else if (addr->index()->is_constant()) {
   251     intptr_t addr_offset = (addr->index()->as_constant_ptr()->as_jint() << addr->scale()) + addr->disp();
   252     assert(Assembler::is_simm32(addr_offset), "must be");
   254     return Address(base, addr_offset);
   255   } else {
   256     Unimplemented();
   257     return Address();
   258   }
   259 }
   262 Address LIR_Assembler::as_Address_hi(LIR_Address* addr) {
   263   Address base = as_Address(addr);
   264   return Address(base._base, base._index, base._scale, base._disp + BytesPerWord);
   265 }
   268 Address LIR_Assembler::as_Address_lo(LIR_Address* addr) {
   269   return as_Address(addr);
   270 }
   273 void LIR_Assembler::osr_entry() {
   274   offsets()->set_value(CodeOffsets::OSR_Entry, code_offset());
   275   BlockBegin* osr_entry = compilation()->hir()->osr_entry();
   276   ValueStack* entry_state = osr_entry->state();
   277   int number_of_locks = entry_state->locks_size();
   279   // we jump here if osr happens with the interpreter
   280   // state set up to continue at the beginning of the
   281   // loop that triggered osr - in particular, we have
   282   // the following registers setup:
   283   //
   284   // rcx: osr buffer
   285   //
   287   // build frame
   288   ciMethod* m = compilation()->method();
   289   __ build_frame(initial_frame_size_in_bytes());
   291   // OSR buffer is
   292   //
   293   // locals[nlocals-1..0]
   294   // monitors[0..number_of_locks]
   295   //
   296   // locals is a direct copy of the interpreter frame so in the osr buffer
   297   // so first slot in the local array is the last local from the interpreter
   298   // and last slot is local[0] (receiver) from the interpreter
   299   //
   300   // Similarly with locks. The first lock slot in the osr buffer is the nth lock
   301   // from the interpreter frame, the nth lock slot in the osr buffer is 0th lock
   302   // in the interpreter frame (the method lock if a sync method)
   304   // Initialize monitors in the compiled activation.
   305   //   rcx: pointer to osr buffer
   306   //
   307   // All other registers are dead at this point and the locals will be
   308   // copied into place by code emitted in the IR.
   310   Register OSR_buf = osrBufferPointer()->as_pointer_register();
   311   { assert(frame::interpreter_frame_monitor_size() == BasicObjectLock::size(), "adjust code below");
   312     int monitor_offset = BytesPerWord * method()->max_locals() +
   313       (2 * BytesPerWord) * (number_of_locks - 1);
   314     // SharedRuntime::OSR_migration_begin() packs BasicObjectLocks in
   315     // the OSR buffer using 2 word entries: first the lock and then
   316     // the oop.
   317     for (int i = 0; i < number_of_locks; i++) {
   318       int slot_offset = monitor_offset - ((i * 2) * BytesPerWord);
   319 #ifdef ASSERT
   320       // verify the interpreter's monitor has a non-null object
   321       {
   322         Label L;
   323         __ cmpptr(Address(OSR_buf, slot_offset + 1*BytesPerWord), (int32_t)NULL_WORD);
   324         __ jcc(Assembler::notZero, L);
   325         __ stop("locked object is NULL");
   326         __ bind(L);
   327       }
   328 #endif
   329       __ movptr(rbx, Address(OSR_buf, slot_offset + 0));
   330       __ movptr(frame_map()->address_for_monitor_lock(i), rbx);
   331       __ movptr(rbx, Address(OSR_buf, slot_offset + 1*BytesPerWord));
   332       __ movptr(frame_map()->address_for_monitor_object(i), rbx);
   333     }
   334   }
   335 }
   338 // inline cache check; done before the frame is built.
   339 int LIR_Assembler::check_icache() {
   340   Register receiver = FrameMap::receiver_opr->as_register();
   341   Register ic_klass = IC_Klass;
   342   const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
   343   const bool do_post_padding = VerifyOops || UseCompressedOops;
   344   if (!do_post_padding) {
   345     // insert some nops so that the verified entry point is aligned on CodeEntryAlignment
   346     while ((__ offset() + ic_cmp_size) % CodeEntryAlignment != 0) {
   347       __ nop();
   348     }
   349   }
   350   int offset = __ offset();
   351   __ inline_cache_check(receiver, IC_Klass);
   352   assert(__ offset() % CodeEntryAlignment == 0 || do_post_padding, "alignment must be correct");
   353   if (do_post_padding) {
   354     // force alignment after the cache check.
   355     // It's been verified to be aligned if !VerifyOops
   356     __ align(CodeEntryAlignment);
   357   }
   358   return offset;
   359 }
   362 void LIR_Assembler::jobject2reg_with_patching(Register reg, CodeEmitInfo* info) {
   363   jobject o = NULL;
   364   PatchingStub* patch = new PatchingStub(_masm, PatchingStub::load_klass_id);
   365   __ movoop(reg, o);
   366   patching_epilog(patch, lir_patch_normal, reg, info);
   367 }
   370 // This specifies the rsp decrement needed to build the frame
   371 int LIR_Assembler::initial_frame_size_in_bytes() {
   372   // if rounding, must let FrameMap know!
   374   // The frame_map records size in slots (32bit word)
   376   // subtract two words to account for return address and link
   377   return (frame_map()->framesize() - (2*VMRegImpl::slots_per_word))  * VMRegImpl::stack_slot_size;
   378 }
   381 int LIR_Assembler::emit_exception_handler() {
   382   // if the last instruction is a call (typically to do a throw which
   383   // is coming at the end after block reordering) the return address
   384   // must still point into the code area in order to avoid assertion
   385   // failures when searching for the corresponding bci => add a nop
   386   // (was bug 5/14/1999 - gri)
   387   __ nop();
   389   // generate code for exception handler
   390   address handler_base = __ start_a_stub(exception_handler_size);
   391   if (handler_base == NULL) {
   392     // not enough space left for the handler
   393     bailout("exception handler overflow");
   394     return -1;
   395   }
   397   int offset = code_offset();
   399   // the exception oop and pc are in rax, and rdx
   400   // no other registers need to be preserved, so invalidate them
   401   __ invalidate_registers(false, true, true, false, true, true);
   403   // check that there is really an exception
   404   __ verify_not_null_oop(rax);
   406   // search an exception handler (rax: exception oop, rdx: throwing pc)
   407   __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::handle_exception_from_callee_id)));
   408   __ should_not_reach_here();
   409   assert(code_offset() - offset <= exception_handler_size, "overflow");
   410   __ end_a_stub();
   412   return offset;
   413 }
   416 // Emit the code to remove the frame from the stack in the exception
   417 // unwind path.
   418 int LIR_Assembler::emit_unwind_handler() {
   419 #ifndef PRODUCT
   420   if (CommentedAssembly) {
   421     _masm->block_comment("Unwind handler");
   422   }
   423 #endif
   425   int offset = code_offset();
   427   // Fetch the exception from TLS and clear out exception related thread state
   428   __ get_thread(rsi);
   429   __ movptr(rax, Address(rsi, JavaThread::exception_oop_offset()));
   430   __ movptr(Address(rsi, JavaThread::exception_oop_offset()), (intptr_t)NULL_WORD);
   431   __ movptr(Address(rsi, JavaThread::exception_pc_offset()), (intptr_t)NULL_WORD);
   433   __ bind(_unwind_handler_entry);
   434   __ verify_not_null_oop(rax);
   435   if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
   436     __ mov(rsi, rax);  // Preserve the exception
   437   }
   439   // Preform needed unlocking
   440   MonitorExitStub* stub = NULL;
   441   if (method()->is_synchronized()) {
   442     monitor_address(0, FrameMap::rax_opr);
   443     stub = new MonitorExitStub(FrameMap::rax_opr, true, 0);
   444     __ unlock_object(rdi, rbx, rax, *stub->entry());
   445     __ bind(*stub->continuation());
   446   }
   448   if (compilation()->env()->dtrace_method_probes()) {
   449     __ get_thread(rax);
   450     __ movptr(Address(rsp, 0), rax);
   451     __ movoop(Address(rsp, sizeof(void*)), method()->constant_encoding());
   452     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit)));
   453   }
   455   if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
   456     __ mov(rax, rsi);  // Restore the exception
   457   }
   459   // remove the activation and dispatch to the unwind handler
   460   __ remove_frame(initial_frame_size_in_bytes());
   461   __ jump(RuntimeAddress(Runtime1::entry_for(Runtime1::unwind_exception_id)));
   463   // Emit the slow path assembly
   464   if (stub != NULL) {
   465     stub->emit_code(this);
   466   }
   468   return offset;
   469 }
   472 int LIR_Assembler::emit_deopt_handler() {
   473   // if the last instruction is a call (typically to do a throw which
   474   // is coming at the end after block reordering) the return address
   475   // must still point into the code area in order to avoid assertion
   476   // failures when searching for the corresponding bci => add a nop
   477   // (was bug 5/14/1999 - gri)
   478   __ nop();
   480   // generate code for exception handler
   481   address handler_base = __ start_a_stub(deopt_handler_size);
   482   if (handler_base == NULL) {
   483     // not enough space left for the handler
   484     bailout("deopt handler overflow");
   485     return -1;
   486   }
   488   int offset = code_offset();
   489   InternalAddress here(__ pc());
   491   __ pushptr(here.addr());
   492   __ jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack()));
   494   assert(code_offset() - offset <= deopt_handler_size, "overflow");
   495   __ end_a_stub();
   497   return offset;
   498 }
   501 // This is the fast version of java.lang.String.compare; it has not
   502 // OSR-entry and therefore, we generate a slow version for OSR's
   503 void LIR_Assembler::emit_string_compare(LIR_Opr arg0, LIR_Opr arg1, LIR_Opr dst, CodeEmitInfo* info) {
   504   __ movptr (rbx, rcx); // receiver is in rcx
   505   __ movptr (rax, arg1->as_register());
   507   // Get addresses of first characters from both Strings
   508   __ load_heap_oop(rsi, Address(rax, java_lang_String::value_offset_in_bytes()));
   509   __ movptr       (rcx, Address(rax, java_lang_String::offset_offset_in_bytes()));
   510   __ lea          (rsi, Address(rsi, rcx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
   513   // rbx, may be NULL
   514   add_debug_info_for_null_check_here(info);
   515   __ load_heap_oop(rdi, Address(rbx, java_lang_String::value_offset_in_bytes()));
   516   __ movptr       (rcx, Address(rbx, java_lang_String::offset_offset_in_bytes()));
   517   __ lea          (rdi, Address(rdi, rcx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
   519   // compute minimum length (in rax) and difference of lengths (on top of stack)
   520   __ movl  (rbx, Address(rbx, java_lang_String::count_offset_in_bytes()));
   521   __ movl  (rax, Address(rax, java_lang_String::count_offset_in_bytes()));
   522   __ mov   (rcx, rbx);
   523   __ subptr(rbx, rax); // subtract lengths
   524   __ push  (rbx);      // result
   525   __ cmov  (Assembler::lessEqual, rax, rcx);
   527   // is minimum length 0?
   528   Label noLoop, haveResult;
   529   __ testptr (rax, rax);
   530   __ jcc (Assembler::zero, noLoop);
   532   // compare first characters
   533   __ load_unsigned_short(rcx, Address(rdi, 0));
   534   __ load_unsigned_short(rbx, Address(rsi, 0));
   535   __ subl(rcx, rbx);
   536   __ jcc(Assembler::notZero, haveResult);
   537   // starting loop
   538   __ decrement(rax); // we already tested index: skip one
   539   __ jcc(Assembler::zero, noLoop);
   541   // set rsi.edi to the end of the arrays (arrays have same length)
   542   // negate the index
   544   __ lea(rsi, Address(rsi, rax, Address::times_2, type2aelembytes(T_CHAR)));
   545   __ lea(rdi, Address(rdi, rax, Address::times_2, type2aelembytes(T_CHAR)));
   546   __ negptr(rax);
   548   // compare the strings in a loop
   550   Label loop;
   551   __ align(wordSize);
   552   __ bind(loop);
   553   __ load_unsigned_short(rcx, Address(rdi, rax, Address::times_2, 0));
   554   __ load_unsigned_short(rbx, Address(rsi, rax, Address::times_2, 0));
   555   __ subl(rcx, rbx);
   556   __ jcc(Assembler::notZero, haveResult);
   557   __ increment(rax);
   558   __ jcc(Assembler::notZero, loop);
   560   // strings are equal up to min length
   562   __ bind(noLoop);
   563   __ pop(rax);
   564   return_op(LIR_OprFact::illegalOpr);
   566   __ bind(haveResult);
   567   // leave instruction is going to discard the TOS value
   568   __ mov (rax, rcx); // result of call is in rax,
   569 }
   572 void LIR_Assembler::return_op(LIR_Opr result) {
   573   assert(result->is_illegal() || !result->is_single_cpu() || result->as_register() == rax, "word returns are in rax,");
   574   if (!result->is_illegal() && result->is_float_kind() && !result->is_xmm_register()) {
   575     assert(result->fpu() == 0, "result must already be on TOS");
   576   }
   578   // Pop the stack before the safepoint code
   579   __ remove_frame(initial_frame_size_in_bytes());
   581   bool result_is_oop = result->is_valid() ? result->is_oop() : false;
   583   // Note: we do not need to round double result; float result has the right precision
   584   // the poll sets the condition code, but no data registers
   585   AddressLiteral polling_page(os::get_polling_page() + (SafepointPollOffset % os::vm_page_size()),
   586                               relocInfo::poll_return_type);
   588   if (Assembler::is_polling_page_far()) {
   589     __ lea(rscratch1, polling_page);
   590     __ relocate(relocInfo::poll_return_type);
   591     __ testl(rax, Address(rscratch1, 0));
   592   } else {
   593     __ testl(rax, polling_page);
   594   }
   595   __ ret(0);
   596 }
   599 int LIR_Assembler::safepoint_poll(LIR_Opr tmp, CodeEmitInfo* info) {
   600   AddressLiteral polling_page(os::get_polling_page() + (SafepointPollOffset % os::vm_page_size()),
   601                               relocInfo::poll_type);
   602   guarantee(info != NULL, "Shouldn't be NULL");
   603   int offset = __ offset();
   604   if (Assembler::is_polling_page_far()) {
   605     __ lea(rscratch1, polling_page);
   606     offset = __ offset();
   607     add_debug_info_for_branch(info);
   608     __ testl(rax, Address(rscratch1, 0));
   609   } else {
   610     add_debug_info_for_branch(info);
   611     __ testl(rax, polling_page);
   612   }
   613   return offset;
   614 }
   617 void LIR_Assembler::move_regs(Register from_reg, Register to_reg) {
   618   if (from_reg != to_reg) __ mov(to_reg, from_reg);
   619 }
   621 void LIR_Assembler::swap_reg(Register a, Register b) {
   622   __ xchgptr(a, b);
   623 }
   626 void LIR_Assembler::const2reg(LIR_Opr src, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
   627   assert(src->is_constant(), "should not call otherwise");
   628   assert(dest->is_register(), "should not call otherwise");
   629   LIR_Const* c = src->as_constant_ptr();
   631   switch (c->type()) {
   632     case T_INT: {
   633       assert(patch_code == lir_patch_none, "no patching handled here");
   634       __ movl(dest->as_register(), c->as_jint());
   635       break;
   636     }
   638     case T_ADDRESS: {
   639       assert(patch_code == lir_patch_none, "no patching handled here");
   640       __ movptr(dest->as_register(), c->as_jint());
   641       break;
   642     }
   644     case T_LONG: {
   645       assert(patch_code == lir_patch_none, "no patching handled here");
   646 #ifdef _LP64
   647       __ movptr(dest->as_register_lo(), (intptr_t)c->as_jlong());
   648 #else
   649       __ movptr(dest->as_register_lo(), c->as_jint_lo());
   650       __ movptr(dest->as_register_hi(), c->as_jint_hi());
   651 #endif // _LP64
   652       break;
   653     }
   655     case T_OBJECT: {
   656       if (patch_code != lir_patch_none) {
   657         jobject2reg_with_patching(dest->as_register(), info);
   658       } else {
   659         __ movoop(dest->as_register(), c->as_jobject());
   660       }
   661       break;
   662     }
   664     case T_FLOAT: {
   665       if (dest->is_single_xmm()) {
   666         if (c->is_zero_float()) {
   667           __ xorps(dest->as_xmm_float_reg(), dest->as_xmm_float_reg());
   668         } else {
   669           __ movflt(dest->as_xmm_float_reg(),
   670                    InternalAddress(float_constant(c->as_jfloat())));
   671         }
   672       } else {
   673         assert(dest->is_single_fpu(), "must be");
   674         assert(dest->fpu_regnr() == 0, "dest must be TOS");
   675         if (c->is_zero_float()) {
   676           __ fldz();
   677         } else if (c->is_one_float()) {
   678           __ fld1();
   679         } else {
   680           __ fld_s (InternalAddress(float_constant(c->as_jfloat())));
   681         }
   682       }
   683       break;
   684     }
   686     case T_DOUBLE: {
   687       if (dest->is_double_xmm()) {
   688         if (c->is_zero_double()) {
   689           __ xorpd(dest->as_xmm_double_reg(), dest->as_xmm_double_reg());
   690         } else {
   691           __ movdbl(dest->as_xmm_double_reg(),
   692                     InternalAddress(double_constant(c->as_jdouble())));
   693         }
   694       } else {
   695         assert(dest->is_double_fpu(), "must be");
   696         assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
   697         if (c->is_zero_double()) {
   698           __ fldz();
   699         } else if (c->is_one_double()) {
   700           __ fld1();
   701         } else {
   702           __ fld_d (InternalAddress(double_constant(c->as_jdouble())));
   703         }
   704       }
   705       break;
   706     }
   708     default:
   709       ShouldNotReachHere();
   710   }
   711 }
   713 void LIR_Assembler::const2stack(LIR_Opr src, LIR_Opr dest) {
   714   assert(src->is_constant(), "should not call otherwise");
   715   assert(dest->is_stack(), "should not call otherwise");
   716   LIR_Const* c = src->as_constant_ptr();
   718   switch (c->type()) {
   719     case T_INT:  // fall through
   720     case T_FLOAT:
   721       __ movl(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jint_bits());
   722       break;
   724     case T_ADDRESS:
   725       __ movptr(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jint_bits());
   726       break;
   728     case T_OBJECT:
   729       __ movoop(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jobject());
   730       break;
   732     case T_LONG:  // fall through
   733     case T_DOUBLE:
   734 #ifdef _LP64
   735       __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
   736                                             lo_word_offset_in_bytes), (intptr_t)c->as_jlong_bits());
   737 #else
   738       __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
   739                                               lo_word_offset_in_bytes), c->as_jint_lo_bits());
   740       __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
   741                                               hi_word_offset_in_bytes), c->as_jint_hi_bits());
   742 #endif // _LP64
   743       break;
   745     default:
   746       ShouldNotReachHere();
   747   }
   748 }
   750 void LIR_Assembler::const2mem(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info, bool wide) {
   751   assert(src->is_constant(), "should not call otherwise");
   752   assert(dest->is_address(), "should not call otherwise");
   753   LIR_Const* c = src->as_constant_ptr();
   754   LIR_Address* addr = dest->as_address_ptr();
   756   int null_check_here = code_offset();
   757   switch (type) {
   758     case T_INT:    // fall through
   759     case T_FLOAT:
   760       __ movl(as_Address(addr), c->as_jint_bits());
   761       break;
   763     case T_ADDRESS:
   764       __ movptr(as_Address(addr), c->as_jint_bits());
   765       break;
   767     case T_OBJECT:  // fall through
   768     case T_ARRAY:
   769       if (c->as_jobject() == NULL) {
   770         if (UseCompressedOops && !wide) {
   771           __ movl(as_Address(addr), (int32_t)NULL_WORD);
   772         } else {
   773           __ movptr(as_Address(addr), NULL_WORD);
   774         }
   775       } else {
   776         if (is_literal_address(addr)) {
   777           ShouldNotReachHere();
   778           __ movoop(as_Address(addr, noreg), c->as_jobject());
   779         } else {
   780 #ifdef _LP64
   781           __ movoop(rscratch1, c->as_jobject());
   782           if (UseCompressedOops && !wide) {
   783             __ encode_heap_oop(rscratch1);
   784             null_check_here = code_offset();
   785             __ movl(as_Address_lo(addr), rscratch1);
   786           } else {
   787             null_check_here = code_offset();
   788             __ movptr(as_Address_lo(addr), rscratch1);
   789           }
   790 #else
   791           __ movoop(as_Address(addr), c->as_jobject());
   792 #endif
   793         }
   794       }
   795       break;
   797     case T_LONG:    // fall through
   798     case T_DOUBLE:
   799 #ifdef _LP64
   800       if (is_literal_address(addr)) {
   801         ShouldNotReachHere();
   802         __ movptr(as_Address(addr, r15_thread), (intptr_t)c->as_jlong_bits());
   803       } else {
   804         __ movptr(r10, (intptr_t)c->as_jlong_bits());
   805         null_check_here = code_offset();
   806         __ movptr(as_Address_lo(addr), r10);
   807       }
   808 #else
   809       // Always reachable in 32bit so this doesn't produce useless move literal
   810       __ movptr(as_Address_hi(addr), c->as_jint_hi_bits());
   811       __ movptr(as_Address_lo(addr), c->as_jint_lo_bits());
   812 #endif // _LP64
   813       break;
   815     case T_BOOLEAN: // fall through
   816     case T_BYTE:
   817       __ movb(as_Address(addr), c->as_jint() & 0xFF);
   818       break;
   820     case T_CHAR:    // fall through
   821     case T_SHORT:
   822       __ movw(as_Address(addr), c->as_jint() & 0xFFFF);
   823       break;
   825     default:
   826       ShouldNotReachHere();
   827   };
   829   if (info != NULL) {
   830     add_debug_info_for_null_check(null_check_here, info);
   831   }
   832 }
   835 void LIR_Assembler::reg2reg(LIR_Opr src, LIR_Opr dest) {
   836   assert(src->is_register(), "should not call otherwise");
   837   assert(dest->is_register(), "should not call otherwise");
   839   // move between cpu-registers
   840   if (dest->is_single_cpu()) {
   841 #ifdef _LP64
   842     if (src->type() == T_LONG) {
   843       // Can do LONG -> OBJECT
   844       move_regs(src->as_register_lo(), dest->as_register());
   845       return;
   846     }
   847 #endif
   848     assert(src->is_single_cpu(), "must match");
   849     if (src->type() == T_OBJECT) {
   850       __ verify_oop(src->as_register());
   851     }
   852     move_regs(src->as_register(), dest->as_register());
   854   } else if (dest->is_double_cpu()) {
   855 #ifdef _LP64
   856     if (src->type() == T_OBJECT || src->type() == T_ARRAY) {
   857       // Surprising to me but we can see move of a long to t_object
   858       __ verify_oop(src->as_register());
   859       move_regs(src->as_register(), dest->as_register_lo());
   860       return;
   861     }
   862 #endif
   863     assert(src->is_double_cpu(), "must match");
   864     Register f_lo = src->as_register_lo();
   865     Register f_hi = src->as_register_hi();
   866     Register t_lo = dest->as_register_lo();
   867     Register t_hi = dest->as_register_hi();
   868 #ifdef _LP64
   869     assert(f_hi == f_lo, "must be same");
   870     assert(t_hi == t_lo, "must be same");
   871     move_regs(f_lo, t_lo);
   872 #else
   873     assert(f_lo != f_hi && t_lo != t_hi, "invalid register allocation");
   876     if (f_lo == t_hi && f_hi == t_lo) {
   877       swap_reg(f_lo, f_hi);
   878     } else if (f_hi == t_lo) {
   879       assert(f_lo != t_hi, "overwriting register");
   880       move_regs(f_hi, t_hi);
   881       move_regs(f_lo, t_lo);
   882     } else {
   883       assert(f_hi != t_lo, "overwriting register");
   884       move_regs(f_lo, t_lo);
   885       move_regs(f_hi, t_hi);
   886     }
   887 #endif // LP64
   889     // special moves from fpu-register to xmm-register
   890     // necessary for method results
   891   } else if (src->is_single_xmm() && !dest->is_single_xmm()) {
   892     __ movflt(Address(rsp, 0), src->as_xmm_float_reg());
   893     __ fld_s(Address(rsp, 0));
   894   } else if (src->is_double_xmm() && !dest->is_double_xmm()) {
   895     __ movdbl(Address(rsp, 0), src->as_xmm_double_reg());
   896     __ fld_d(Address(rsp, 0));
   897   } else if (dest->is_single_xmm() && !src->is_single_xmm()) {
   898     __ fstp_s(Address(rsp, 0));
   899     __ movflt(dest->as_xmm_float_reg(), Address(rsp, 0));
   900   } else if (dest->is_double_xmm() && !src->is_double_xmm()) {
   901     __ fstp_d(Address(rsp, 0));
   902     __ movdbl(dest->as_xmm_double_reg(), Address(rsp, 0));
   904     // move between xmm-registers
   905   } else if (dest->is_single_xmm()) {
   906     assert(src->is_single_xmm(), "must match");
   907     __ movflt(dest->as_xmm_float_reg(), src->as_xmm_float_reg());
   908   } else if (dest->is_double_xmm()) {
   909     assert(src->is_double_xmm(), "must match");
   910     __ movdbl(dest->as_xmm_double_reg(), src->as_xmm_double_reg());
   912     // move between fpu-registers (no instruction necessary because of fpu-stack)
   913   } else if (dest->is_single_fpu() || dest->is_double_fpu()) {
   914     assert(src->is_single_fpu() || src->is_double_fpu(), "must match");
   915     assert(src->fpu() == dest->fpu(), "currently should be nothing to do");
   916   } else {
   917     ShouldNotReachHere();
   918   }
   919 }
   921 void LIR_Assembler::reg2stack(LIR_Opr src, LIR_Opr dest, BasicType type, bool pop_fpu_stack) {
   922   assert(src->is_register(), "should not call otherwise");
   923   assert(dest->is_stack(), "should not call otherwise");
   925   if (src->is_single_cpu()) {
   926     Address dst = frame_map()->address_for_slot(dest->single_stack_ix());
   927     if (type == T_OBJECT || type == T_ARRAY) {
   928       __ verify_oop(src->as_register());
   929       __ movptr (dst, src->as_register());
   930     } else {
   931       __ movl (dst, src->as_register());
   932     }
   934   } else if (src->is_double_cpu()) {
   935     Address dstLO = frame_map()->address_for_slot(dest->double_stack_ix(), lo_word_offset_in_bytes);
   936     Address dstHI = frame_map()->address_for_slot(dest->double_stack_ix(), hi_word_offset_in_bytes);
   937     __ movptr (dstLO, src->as_register_lo());
   938     NOT_LP64(__ movptr (dstHI, src->as_register_hi()));
   940   } else if (src->is_single_xmm()) {
   941     Address dst_addr = frame_map()->address_for_slot(dest->single_stack_ix());
   942     __ movflt(dst_addr, src->as_xmm_float_reg());
   944   } else if (src->is_double_xmm()) {
   945     Address dst_addr = frame_map()->address_for_slot(dest->double_stack_ix());
   946     __ movdbl(dst_addr, src->as_xmm_double_reg());
   948   } else if (src->is_single_fpu()) {
   949     assert(src->fpu_regnr() == 0, "argument must be on TOS");
   950     Address dst_addr = frame_map()->address_for_slot(dest->single_stack_ix());
   951     if (pop_fpu_stack)     __ fstp_s (dst_addr);
   952     else                   __ fst_s  (dst_addr);
   954   } else if (src->is_double_fpu()) {
   955     assert(src->fpu_regnrLo() == 0, "argument must be on TOS");
   956     Address dst_addr = frame_map()->address_for_slot(dest->double_stack_ix());
   957     if (pop_fpu_stack)     __ fstp_d (dst_addr);
   958     else                   __ fst_d  (dst_addr);
   960   } else {
   961     ShouldNotReachHere();
   962   }
   963 }
   966 void LIR_Assembler::reg2mem(LIR_Opr src, LIR_Opr dest, BasicType type, LIR_PatchCode patch_code, CodeEmitInfo* info, bool pop_fpu_stack, bool wide, bool /* unaligned */) {
   967   LIR_Address* to_addr = dest->as_address_ptr();
   968   PatchingStub* patch = NULL;
   969   Register compressed_src = rscratch1;
   971   if (type == T_ARRAY || type == T_OBJECT) {
   972     __ verify_oop(src->as_register());
   973 #ifdef _LP64
   974     if (UseCompressedOops && !wide) {
   975       __ movptr(compressed_src, src->as_register());
   976       __ encode_heap_oop(compressed_src);
   977     }
   978 #endif
   979   }
   981   if (patch_code != lir_patch_none) {
   982     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
   983     Address toa = as_Address(to_addr);
   984     assert(toa.disp() != 0, "must have");
   985   }
   987   int null_check_here = code_offset();
   988   switch (type) {
   989     case T_FLOAT: {
   990       if (src->is_single_xmm()) {
   991         __ movflt(as_Address(to_addr), src->as_xmm_float_reg());
   992       } else {
   993         assert(src->is_single_fpu(), "must be");
   994         assert(src->fpu_regnr() == 0, "argument must be on TOS");
   995         if (pop_fpu_stack)      __ fstp_s(as_Address(to_addr));
   996         else                    __ fst_s (as_Address(to_addr));
   997       }
   998       break;
   999     }
  1001     case T_DOUBLE: {
  1002       if (src->is_double_xmm()) {
  1003         __ movdbl(as_Address(to_addr), src->as_xmm_double_reg());
  1004       } else {
  1005         assert(src->is_double_fpu(), "must be");
  1006         assert(src->fpu_regnrLo() == 0, "argument must be on TOS");
  1007         if (pop_fpu_stack)      __ fstp_d(as_Address(to_addr));
  1008         else                    __ fst_d (as_Address(to_addr));
  1010       break;
  1013     case T_ARRAY:   // fall through
  1014     case T_OBJECT:  // fall through
  1015       if (UseCompressedOops && !wide) {
  1016         __ movl(as_Address(to_addr), compressed_src);
  1017       } else {
  1018         __ movptr(as_Address(to_addr), src->as_register());
  1020       break;
  1021     case T_ADDRESS:
  1022       __ movptr(as_Address(to_addr), src->as_register());
  1023       break;
  1024     case T_INT:
  1025       __ movl(as_Address(to_addr), src->as_register());
  1026       break;
  1028     case T_LONG: {
  1029       Register from_lo = src->as_register_lo();
  1030       Register from_hi = src->as_register_hi();
  1031 #ifdef _LP64
  1032       __ movptr(as_Address_lo(to_addr), from_lo);
  1033 #else
  1034       Register base = to_addr->base()->as_register();
  1035       Register index = noreg;
  1036       if (to_addr->index()->is_register()) {
  1037         index = to_addr->index()->as_register();
  1039       if (base == from_lo || index == from_lo) {
  1040         assert(base != from_hi, "can't be");
  1041         assert(index == noreg || (index != base && index != from_hi), "can't handle this");
  1042         __ movl(as_Address_hi(to_addr), from_hi);
  1043         if (patch != NULL) {
  1044           patching_epilog(patch, lir_patch_high, base, info);
  1045           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
  1046           patch_code = lir_patch_low;
  1048         __ movl(as_Address_lo(to_addr), from_lo);
  1049       } else {
  1050         assert(index == noreg || (index != base && index != from_lo), "can't handle this");
  1051         __ movl(as_Address_lo(to_addr), from_lo);
  1052         if (patch != NULL) {
  1053           patching_epilog(patch, lir_patch_low, base, info);
  1054           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
  1055           patch_code = lir_patch_high;
  1057         __ movl(as_Address_hi(to_addr), from_hi);
  1059 #endif // _LP64
  1060       break;
  1063     case T_BYTE:    // fall through
  1064     case T_BOOLEAN: {
  1065       Register src_reg = src->as_register();
  1066       Address dst_addr = as_Address(to_addr);
  1067       assert(VM_Version::is_P6() || src_reg->has_byte_register(), "must use byte registers if not P6");
  1068       __ movb(dst_addr, src_reg);
  1069       break;
  1072     case T_CHAR:    // fall through
  1073     case T_SHORT:
  1074       __ movw(as_Address(to_addr), src->as_register());
  1075       break;
  1077     default:
  1078       ShouldNotReachHere();
  1080   if (info != NULL) {
  1081     add_debug_info_for_null_check(null_check_here, info);
  1084   if (patch_code != lir_patch_none) {
  1085     patching_epilog(patch, patch_code, to_addr->base()->as_register(), info);
  1090 void LIR_Assembler::stack2reg(LIR_Opr src, LIR_Opr dest, BasicType type) {
  1091   assert(src->is_stack(), "should not call otherwise");
  1092   assert(dest->is_register(), "should not call otherwise");
  1094   if (dest->is_single_cpu()) {
  1095     if (type == T_ARRAY || type == T_OBJECT) {
  1096       __ movptr(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()));
  1097       __ verify_oop(dest->as_register());
  1098     } else {
  1099       __ movl(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()));
  1102   } else if (dest->is_double_cpu()) {
  1103     Address src_addr_LO = frame_map()->address_for_slot(src->double_stack_ix(), lo_word_offset_in_bytes);
  1104     Address src_addr_HI = frame_map()->address_for_slot(src->double_stack_ix(), hi_word_offset_in_bytes);
  1105     __ movptr(dest->as_register_lo(), src_addr_LO);
  1106     NOT_LP64(__ movptr(dest->as_register_hi(), src_addr_HI));
  1108   } else if (dest->is_single_xmm()) {
  1109     Address src_addr = frame_map()->address_for_slot(src->single_stack_ix());
  1110     __ movflt(dest->as_xmm_float_reg(), src_addr);
  1112   } else if (dest->is_double_xmm()) {
  1113     Address src_addr = frame_map()->address_for_slot(src->double_stack_ix());
  1114     __ movdbl(dest->as_xmm_double_reg(), src_addr);
  1116   } else if (dest->is_single_fpu()) {
  1117     assert(dest->fpu_regnr() == 0, "dest must be TOS");
  1118     Address src_addr = frame_map()->address_for_slot(src->single_stack_ix());
  1119     __ fld_s(src_addr);
  1121   } else if (dest->is_double_fpu()) {
  1122     assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
  1123     Address src_addr = frame_map()->address_for_slot(src->double_stack_ix());
  1124     __ fld_d(src_addr);
  1126   } else {
  1127     ShouldNotReachHere();
  1132 void LIR_Assembler::stack2stack(LIR_Opr src, LIR_Opr dest, BasicType type) {
  1133   if (src->is_single_stack()) {
  1134     if (type == T_OBJECT || type == T_ARRAY) {
  1135       __ pushptr(frame_map()->address_for_slot(src ->single_stack_ix()));
  1136       __ popptr (frame_map()->address_for_slot(dest->single_stack_ix()));
  1137     } else {
  1138 #ifndef _LP64
  1139       __ pushl(frame_map()->address_for_slot(src ->single_stack_ix()));
  1140       __ popl (frame_map()->address_for_slot(dest->single_stack_ix()));
  1141 #else
  1142       //no pushl on 64bits
  1143       __ movl(rscratch1, frame_map()->address_for_slot(src ->single_stack_ix()));
  1144       __ movl(frame_map()->address_for_slot(dest->single_stack_ix()), rscratch1);
  1145 #endif
  1148   } else if (src->is_double_stack()) {
  1149 #ifdef _LP64
  1150     __ pushptr(frame_map()->address_for_slot(src ->double_stack_ix()));
  1151     __ popptr (frame_map()->address_for_slot(dest->double_stack_ix()));
  1152 #else
  1153     __ pushl(frame_map()->address_for_slot(src ->double_stack_ix(), 0));
  1154     // push and pop the part at src + wordSize, adding wordSize for the previous push
  1155     __ pushl(frame_map()->address_for_slot(src ->double_stack_ix(), 2 * wordSize));
  1156     __ popl (frame_map()->address_for_slot(dest->double_stack_ix(), 2 * wordSize));
  1157     __ popl (frame_map()->address_for_slot(dest->double_stack_ix(), 0));
  1158 #endif // _LP64
  1160   } else {
  1161     ShouldNotReachHere();
  1166 void LIR_Assembler::mem2reg(LIR_Opr src, LIR_Opr dest, BasicType type, LIR_PatchCode patch_code, CodeEmitInfo* info, bool wide, bool /* unaligned */) {
  1167   assert(src->is_address(), "should not call otherwise");
  1168   assert(dest->is_register(), "should not call otherwise");
  1170   LIR_Address* addr = src->as_address_ptr();
  1171   Address from_addr = as_Address(addr);
  1173   switch (type) {
  1174     case T_BOOLEAN: // fall through
  1175     case T_BYTE:    // fall through
  1176     case T_CHAR:    // fall through
  1177     case T_SHORT:
  1178       if (!VM_Version::is_P6() && !from_addr.uses(dest->as_register())) {
  1179         // on pre P6 processors we may get partial register stalls
  1180         // so blow away the value of to_rinfo before loading a
  1181         // partial word into it.  Do it here so that it precedes
  1182         // the potential patch point below.
  1183         __ xorptr(dest->as_register(), dest->as_register());
  1185       break;
  1188   PatchingStub* patch = NULL;
  1189   if (patch_code != lir_patch_none) {
  1190     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
  1191     assert(from_addr.disp() != 0, "must have");
  1193   if (info != NULL) {
  1194     add_debug_info_for_null_check_here(info);
  1197   switch (type) {
  1198     case T_FLOAT: {
  1199       if (dest->is_single_xmm()) {
  1200         __ movflt(dest->as_xmm_float_reg(), from_addr);
  1201       } else {
  1202         assert(dest->is_single_fpu(), "must be");
  1203         assert(dest->fpu_regnr() == 0, "dest must be TOS");
  1204         __ fld_s(from_addr);
  1206       break;
  1209     case T_DOUBLE: {
  1210       if (dest->is_double_xmm()) {
  1211         __ movdbl(dest->as_xmm_double_reg(), from_addr);
  1212       } else {
  1213         assert(dest->is_double_fpu(), "must be");
  1214         assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
  1215         __ fld_d(from_addr);
  1217       break;
  1220     case T_OBJECT:  // fall through
  1221     case T_ARRAY:   // fall through
  1222       if (UseCompressedOops && !wide) {
  1223         __ movl(dest->as_register(), from_addr);
  1224       } else {
  1225         __ movptr(dest->as_register(), from_addr);
  1227       break;
  1229     case T_ADDRESS:
  1230       __ movptr(dest->as_register(), from_addr);
  1231       break;
  1232     case T_INT:
  1233       __ movl(dest->as_register(), from_addr);
  1234       break;
  1236     case T_LONG: {
  1237       Register to_lo = dest->as_register_lo();
  1238       Register to_hi = dest->as_register_hi();
  1239 #ifdef _LP64
  1240       __ movptr(to_lo, as_Address_lo(addr));
  1241 #else
  1242       Register base = addr->base()->as_register();
  1243       Register index = noreg;
  1244       if (addr->index()->is_register()) {
  1245         index = addr->index()->as_register();
  1247       if ((base == to_lo && index == to_hi) ||
  1248           (base == to_hi && index == to_lo)) {
  1249         // addresses with 2 registers are only formed as a result of
  1250         // array access so this code will never have to deal with
  1251         // patches or null checks.
  1252         assert(info == NULL && patch == NULL, "must be");
  1253         __ lea(to_hi, as_Address(addr));
  1254         __ movl(to_lo, Address(to_hi, 0));
  1255         __ movl(to_hi, Address(to_hi, BytesPerWord));
  1256       } else if (base == to_lo || index == to_lo) {
  1257         assert(base != to_hi, "can't be");
  1258         assert(index == noreg || (index != base && index != to_hi), "can't handle this");
  1259         __ movl(to_hi, as_Address_hi(addr));
  1260         if (patch != NULL) {
  1261           patching_epilog(patch, lir_patch_high, base, info);
  1262           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
  1263           patch_code = lir_patch_low;
  1265         __ movl(to_lo, as_Address_lo(addr));
  1266       } else {
  1267         assert(index == noreg || (index != base && index != to_lo), "can't handle this");
  1268         __ movl(to_lo, as_Address_lo(addr));
  1269         if (patch != NULL) {
  1270           patching_epilog(patch, lir_patch_low, base, info);
  1271           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
  1272           patch_code = lir_patch_high;
  1274         __ movl(to_hi, as_Address_hi(addr));
  1276 #endif // _LP64
  1277       break;
  1280     case T_BOOLEAN: // fall through
  1281     case T_BYTE: {
  1282       Register dest_reg = dest->as_register();
  1283       assert(VM_Version::is_P6() || dest_reg->has_byte_register(), "must use byte registers if not P6");
  1284       if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
  1285         __ movsbl(dest_reg, from_addr);
  1286       } else {
  1287         __ movb(dest_reg, from_addr);
  1288         __ shll(dest_reg, 24);
  1289         __ sarl(dest_reg, 24);
  1291       break;
  1294     case T_CHAR: {
  1295       Register dest_reg = dest->as_register();
  1296       assert(VM_Version::is_P6() || dest_reg->has_byte_register(), "must use byte registers if not P6");
  1297       if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
  1298         __ movzwl(dest_reg, from_addr);
  1299       } else {
  1300         __ movw(dest_reg, from_addr);
  1302       break;
  1305     case T_SHORT: {
  1306       Register dest_reg = dest->as_register();
  1307       if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
  1308         __ movswl(dest_reg, from_addr);
  1309       } else {
  1310         __ movw(dest_reg, from_addr);
  1311         __ shll(dest_reg, 16);
  1312         __ sarl(dest_reg, 16);
  1314       break;
  1317     default:
  1318       ShouldNotReachHere();
  1321   if (patch != NULL) {
  1322     patching_epilog(patch, patch_code, addr->base()->as_register(), info);
  1325   if (type == T_ARRAY || type == T_OBJECT) {
  1326 #ifdef _LP64
  1327     if (UseCompressedOops && !wide) {
  1328       __ decode_heap_oop(dest->as_register());
  1330 #endif
  1331     __ verify_oop(dest->as_register());
  1336 void LIR_Assembler::prefetchr(LIR_Opr src) {
  1337   LIR_Address* addr = src->as_address_ptr();
  1338   Address from_addr = as_Address(addr);
  1340   if (VM_Version::supports_sse()) {
  1341     switch (ReadPrefetchInstr) {
  1342       case 0:
  1343         __ prefetchnta(from_addr); break;
  1344       case 1:
  1345         __ prefetcht0(from_addr); break;
  1346       case 2:
  1347         __ prefetcht2(from_addr); break;
  1348       default:
  1349         ShouldNotReachHere(); break;
  1351   } else if (VM_Version::supports_3dnow_prefetch()) {
  1352     __ prefetchr(from_addr);
  1357 void LIR_Assembler::prefetchw(LIR_Opr src) {
  1358   LIR_Address* addr = src->as_address_ptr();
  1359   Address from_addr = as_Address(addr);
  1361   if (VM_Version::supports_sse()) {
  1362     switch (AllocatePrefetchInstr) {
  1363       case 0:
  1364         __ prefetchnta(from_addr); break;
  1365       case 1:
  1366         __ prefetcht0(from_addr); break;
  1367       case 2:
  1368         __ prefetcht2(from_addr); break;
  1369       case 3:
  1370         __ prefetchw(from_addr); break;
  1371       default:
  1372         ShouldNotReachHere(); break;
  1374   } else if (VM_Version::supports_3dnow_prefetch()) {
  1375     __ prefetchw(from_addr);
  1380 NEEDS_CLEANUP; // This could be static?
  1381 Address::ScaleFactor LIR_Assembler::array_element_size(BasicType type) const {
  1382   int elem_size = type2aelembytes(type);
  1383   switch (elem_size) {
  1384     case 1: return Address::times_1;
  1385     case 2: return Address::times_2;
  1386     case 4: return Address::times_4;
  1387     case 8: return Address::times_8;
  1389   ShouldNotReachHere();
  1390   return Address::no_scale;
  1394 void LIR_Assembler::emit_op3(LIR_Op3* op) {
  1395   switch (op->code()) {
  1396     case lir_idiv:
  1397     case lir_irem:
  1398       arithmetic_idiv(op->code(),
  1399                       op->in_opr1(),
  1400                       op->in_opr2(),
  1401                       op->in_opr3(),
  1402                       op->result_opr(),
  1403                       op->info());
  1404       break;
  1405     default:      ShouldNotReachHere(); break;
  1409 void LIR_Assembler::emit_opBranch(LIR_OpBranch* op) {
  1410 #ifdef ASSERT
  1411   assert(op->block() == NULL || op->block()->label() == op->label(), "wrong label");
  1412   if (op->block() != NULL)  _branch_target_blocks.append(op->block());
  1413   if (op->ublock() != NULL) _branch_target_blocks.append(op->ublock());
  1414 #endif
  1416   if (op->cond() == lir_cond_always) {
  1417     if (op->info() != NULL) add_debug_info_for_branch(op->info());
  1418     __ jmp (*(op->label()));
  1419   } else {
  1420     Assembler::Condition acond = Assembler::zero;
  1421     if (op->code() == lir_cond_float_branch) {
  1422       assert(op->ublock() != NULL, "must have unordered successor");
  1423       __ jcc(Assembler::parity, *(op->ublock()->label()));
  1424       switch(op->cond()) {
  1425         case lir_cond_equal:        acond = Assembler::equal;      break;
  1426         case lir_cond_notEqual:     acond = Assembler::notEqual;   break;
  1427         case lir_cond_less:         acond = Assembler::below;      break;
  1428         case lir_cond_lessEqual:    acond = Assembler::belowEqual; break;
  1429         case lir_cond_greaterEqual: acond = Assembler::aboveEqual; break;
  1430         case lir_cond_greater:      acond = Assembler::above;      break;
  1431         default:                         ShouldNotReachHere();
  1433     } else {
  1434       switch (op->cond()) {
  1435         case lir_cond_equal:        acond = Assembler::equal;       break;
  1436         case lir_cond_notEqual:     acond = Assembler::notEqual;    break;
  1437         case lir_cond_less:         acond = Assembler::less;        break;
  1438         case lir_cond_lessEqual:    acond = Assembler::lessEqual;   break;
  1439         case lir_cond_greaterEqual: acond = Assembler::greaterEqual;break;
  1440         case lir_cond_greater:      acond = Assembler::greater;     break;
  1441         case lir_cond_belowEqual:   acond = Assembler::belowEqual;  break;
  1442         case lir_cond_aboveEqual:   acond = Assembler::aboveEqual;  break;
  1443         default:                         ShouldNotReachHere();
  1446     __ jcc(acond,*(op->label()));
  1450 void LIR_Assembler::emit_opConvert(LIR_OpConvert* op) {
  1451   LIR_Opr src  = op->in_opr();
  1452   LIR_Opr dest = op->result_opr();
  1454   switch (op->bytecode()) {
  1455     case Bytecodes::_i2l:
  1456 #ifdef _LP64
  1457       __ movl2ptr(dest->as_register_lo(), src->as_register());
  1458 #else
  1459       move_regs(src->as_register(), dest->as_register_lo());
  1460       move_regs(src->as_register(), dest->as_register_hi());
  1461       __ sarl(dest->as_register_hi(), 31);
  1462 #endif // LP64
  1463       break;
  1465     case Bytecodes::_l2i:
  1466       move_regs(src->as_register_lo(), dest->as_register());
  1467       break;
  1469     case Bytecodes::_i2b:
  1470       move_regs(src->as_register(), dest->as_register());
  1471       __ sign_extend_byte(dest->as_register());
  1472       break;
  1474     case Bytecodes::_i2c:
  1475       move_regs(src->as_register(), dest->as_register());
  1476       __ andl(dest->as_register(), 0xFFFF);
  1477       break;
  1479     case Bytecodes::_i2s:
  1480       move_regs(src->as_register(), dest->as_register());
  1481       __ sign_extend_short(dest->as_register());
  1482       break;
  1485     case Bytecodes::_f2d:
  1486     case Bytecodes::_d2f:
  1487       if (dest->is_single_xmm()) {
  1488         __ cvtsd2ss(dest->as_xmm_float_reg(), src->as_xmm_double_reg());
  1489       } else if (dest->is_double_xmm()) {
  1490         __ cvtss2sd(dest->as_xmm_double_reg(), src->as_xmm_float_reg());
  1491       } else {
  1492         assert(src->fpu() == dest->fpu(), "register must be equal");
  1493         // do nothing (float result is rounded later through spilling)
  1495       break;
  1497     case Bytecodes::_i2f:
  1498     case Bytecodes::_i2d:
  1499       if (dest->is_single_xmm()) {
  1500         __ cvtsi2ssl(dest->as_xmm_float_reg(), src->as_register());
  1501       } else if (dest->is_double_xmm()) {
  1502         __ cvtsi2sdl(dest->as_xmm_double_reg(), src->as_register());
  1503       } else {
  1504         assert(dest->fpu() == 0, "result must be on TOS");
  1505         __ movl(Address(rsp, 0), src->as_register());
  1506         __ fild_s(Address(rsp, 0));
  1508       break;
  1510     case Bytecodes::_f2i:
  1511     case Bytecodes::_d2i:
  1512       if (src->is_single_xmm()) {
  1513         __ cvttss2sil(dest->as_register(), src->as_xmm_float_reg());
  1514       } else if (src->is_double_xmm()) {
  1515         __ cvttsd2sil(dest->as_register(), src->as_xmm_double_reg());
  1516       } else {
  1517         assert(src->fpu() == 0, "input must be on TOS");
  1518         __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_trunc()));
  1519         __ fist_s(Address(rsp, 0));
  1520         __ movl(dest->as_register(), Address(rsp, 0));
  1521         __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
  1524       // IA32 conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
  1525       assert(op->stub() != NULL, "stub required");
  1526       __ cmpl(dest->as_register(), 0x80000000);
  1527       __ jcc(Assembler::equal, *op->stub()->entry());
  1528       __ bind(*op->stub()->continuation());
  1529       break;
  1531     case Bytecodes::_l2f:
  1532     case Bytecodes::_l2d:
  1533       assert(!dest->is_xmm_register(), "result in xmm register not supported (no SSE instruction present)");
  1534       assert(dest->fpu() == 0, "result must be on TOS");
  1536       __ movptr(Address(rsp, 0),            src->as_register_lo());
  1537       NOT_LP64(__ movl(Address(rsp, BytesPerWord), src->as_register_hi()));
  1538       __ fild_d(Address(rsp, 0));
  1539       // float result is rounded later through spilling
  1540       break;
  1542     case Bytecodes::_f2l:
  1543     case Bytecodes::_d2l:
  1544       assert(!src->is_xmm_register(), "input in xmm register not supported (no SSE instruction present)");
  1545       assert(src->fpu() == 0, "input must be on TOS");
  1546       assert(dest == FrameMap::long0_opr, "runtime stub places result in these registers");
  1548       // instruction sequence too long to inline it here
  1550         __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::fpu2long_stub_id)));
  1552       break;
  1554     default: ShouldNotReachHere();
  1558 void LIR_Assembler::emit_alloc_obj(LIR_OpAllocObj* op) {
  1559   if (op->init_check()) {
  1560     __ cmpb(Address(op->klass()->as_register(),
  1561                     instanceKlass::init_state_offset()),
  1562             instanceKlass::fully_initialized);
  1563     add_debug_info_for_null_check_here(op->stub()->info());
  1564     __ jcc(Assembler::notEqual, *op->stub()->entry());
  1566   __ allocate_object(op->obj()->as_register(),
  1567                      op->tmp1()->as_register(),
  1568                      op->tmp2()->as_register(),
  1569                      op->header_size(),
  1570                      op->object_size(),
  1571                      op->klass()->as_register(),
  1572                      *op->stub()->entry());
  1573   __ bind(*op->stub()->continuation());
  1576 void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
  1577   Register len =  op->len()->as_register();
  1578   LP64_ONLY( __ movslq(len, len); )
  1580   if (UseSlowPath ||
  1581       (!UseFastNewObjectArray && (op->type() == T_OBJECT || op->type() == T_ARRAY)) ||
  1582       (!UseFastNewTypeArray   && (op->type() != T_OBJECT && op->type() != T_ARRAY))) {
  1583     __ jmp(*op->stub()->entry());
  1584   } else {
  1585     Register tmp1 = op->tmp1()->as_register();
  1586     Register tmp2 = op->tmp2()->as_register();
  1587     Register tmp3 = op->tmp3()->as_register();
  1588     if (len == tmp1) {
  1589       tmp1 = tmp3;
  1590     } else if (len == tmp2) {
  1591       tmp2 = tmp3;
  1592     } else if (len == tmp3) {
  1593       // everything is ok
  1594     } else {
  1595       __ mov(tmp3, len);
  1597     __ allocate_array(op->obj()->as_register(),
  1598                       len,
  1599                       tmp1,
  1600                       tmp2,
  1601                       arrayOopDesc::header_size(op->type()),
  1602                       array_element_size(op->type()),
  1603                       op->klass()->as_register(),
  1604                       *op->stub()->entry());
  1606   __ bind(*op->stub()->continuation());
  1609 void LIR_Assembler::type_profile_helper(Register mdo,
  1610                                         ciMethodData *md, ciProfileData *data,
  1611                                         Register recv, Label* update_done) {
  1612   for (uint i = 0; i < ReceiverTypeData::row_limit(); i++) {
  1613     Label next_test;
  1614     // See if the receiver is receiver[n].
  1615     __ cmpptr(recv, Address(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i))));
  1616     __ jccb(Assembler::notEqual, next_test);
  1617     Address data_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)));
  1618     __ addptr(data_addr, DataLayout::counter_increment);
  1619     __ jmp(*update_done);
  1620     __ bind(next_test);
  1623   // Didn't find receiver; find next empty slot and fill it in
  1624   for (uint i = 0; i < ReceiverTypeData::row_limit(); i++) {
  1625     Label next_test;
  1626     Address recv_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)));
  1627     __ cmpptr(recv_addr, (intptr_t)NULL_WORD);
  1628     __ jccb(Assembler::notEqual, next_test);
  1629     __ movptr(recv_addr, recv);
  1630     __ movptr(Address(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i))), DataLayout::counter_increment);
  1631     __ jmp(*update_done);
  1632     __ bind(next_test);
  1636 void LIR_Assembler::emit_typecheck_helper(LIR_OpTypeCheck *op, Label* success, Label* failure, Label* obj_is_null) {
  1637   // we always need a stub for the failure case.
  1638   CodeStub* stub = op->stub();
  1639   Register obj = op->object()->as_register();
  1640   Register k_RInfo = op->tmp1()->as_register();
  1641   Register klass_RInfo = op->tmp2()->as_register();
  1642   Register dst = op->result_opr()->as_register();
  1643   ciKlass* k = op->klass();
  1644   Register Rtmp1 = noreg;
  1646   // check if it needs to be profiled
  1647   ciMethodData* md;
  1648   ciProfileData* data;
  1650   if (op->should_profile()) {
  1651     ciMethod* method = op->profiled_method();
  1652     assert(method != NULL, "Should have method");
  1653     int bci = op->profiled_bci();
  1654     md = method->method_data_or_null();
  1655     assert(md != NULL, "Sanity");
  1656     data = md->bci_to_data(bci);
  1657     assert(data != NULL,                "need data for type check");
  1658     assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
  1660   Label profile_cast_success, profile_cast_failure;
  1661   Label *success_target = op->should_profile() ? &profile_cast_success : success;
  1662   Label *failure_target = op->should_profile() ? &profile_cast_failure : failure;
  1664   if (obj == k_RInfo) {
  1665     k_RInfo = dst;
  1666   } else if (obj == klass_RInfo) {
  1667     klass_RInfo = dst;
  1669   if (k->is_loaded() && !UseCompressedOops) {
  1670     select_different_registers(obj, dst, k_RInfo, klass_RInfo);
  1671   } else {
  1672     Rtmp1 = op->tmp3()->as_register();
  1673     select_different_registers(obj, dst, k_RInfo, klass_RInfo, Rtmp1);
  1676   assert_different_registers(obj, k_RInfo, klass_RInfo);
  1677   if (!k->is_loaded()) {
  1678     jobject2reg_with_patching(k_RInfo, op->info_for_patch());
  1679   } else {
  1680 #ifdef _LP64
  1681     __ movoop(k_RInfo, k->constant_encoding());
  1682 #endif // _LP64
  1684   assert(obj != k_RInfo, "must be different");
  1686   __ cmpptr(obj, (int32_t)NULL_WORD);
  1687   if (op->should_profile()) {
  1688     Label not_null;
  1689     __ jccb(Assembler::notEqual, not_null);
  1690     // Object is null; update MDO and exit
  1691     Register mdo  = klass_RInfo;
  1692     __ movoop(mdo, md->constant_encoding());
  1693     Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
  1694     int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
  1695     __ orl(data_addr, header_bits);
  1696     __ jmp(*obj_is_null);
  1697     __ bind(not_null);
  1698   } else {
  1699     __ jcc(Assembler::equal, *obj_is_null);
  1701   __ verify_oop(obj);
  1703   if (op->fast_check()) {
  1704     // get object class
  1705     // not a safepoint as obj null check happens earlier
  1706 #ifdef _LP64
  1707     if (UseCompressedOops) {
  1708       __ load_klass(Rtmp1, obj);
  1709       __ cmpptr(k_RInfo, Rtmp1);
  1710     } else {
  1711       __ cmpptr(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
  1713 #else
  1714     if (k->is_loaded()) {
  1715       __ cmpoop(Address(obj, oopDesc::klass_offset_in_bytes()), k->constant_encoding());
  1716     } else {
  1717       __ cmpptr(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
  1719 #endif
  1720     __ jcc(Assembler::notEqual, *failure_target);
  1721     // successful cast, fall through to profile or jump
  1722   } else {
  1723     // get object class
  1724     // not a safepoint as obj null check happens earlier
  1725     __ load_klass(klass_RInfo, obj);
  1726     if (k->is_loaded()) {
  1727       // See if we get an immediate positive hit
  1728 #ifdef _LP64
  1729       __ cmpptr(k_RInfo, Address(klass_RInfo, k->super_check_offset()));
  1730 #else
  1731       __ cmpoop(Address(klass_RInfo, k->super_check_offset()), k->constant_encoding());
  1732 #endif // _LP64
  1733       if ((juint)in_bytes(Klass::secondary_super_cache_offset()) != k->super_check_offset()) {
  1734         __ jcc(Assembler::notEqual, *failure_target);
  1735         // successful cast, fall through to profile or jump
  1736       } else {
  1737         // See if we get an immediate positive hit
  1738         __ jcc(Assembler::equal, *success_target);
  1739         // check for self
  1740 #ifdef _LP64
  1741         __ cmpptr(klass_RInfo, k_RInfo);
  1742 #else
  1743         __ cmpoop(klass_RInfo, k->constant_encoding());
  1744 #endif // _LP64
  1745         __ jcc(Assembler::equal, *success_target);
  1747         __ push(klass_RInfo);
  1748 #ifdef _LP64
  1749         __ push(k_RInfo);
  1750 #else
  1751         __ pushoop(k->constant_encoding());
  1752 #endif // _LP64
  1753         __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
  1754         __ pop(klass_RInfo);
  1755         __ pop(klass_RInfo);
  1756         // result is a boolean
  1757         __ cmpl(klass_RInfo, 0);
  1758         __ jcc(Assembler::equal, *failure_target);
  1759         // successful cast, fall through to profile or jump
  1761     } else {
  1762       // perform the fast part of the checking logic
  1763       __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, success_target, failure_target, NULL);
  1764       // call out-of-line instance of __ check_klass_subtype_slow_path(...):
  1765       __ push(klass_RInfo);
  1766       __ push(k_RInfo);
  1767       __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
  1768       __ pop(klass_RInfo);
  1769       __ pop(k_RInfo);
  1770       // result is a boolean
  1771       __ cmpl(k_RInfo, 0);
  1772       __ jcc(Assembler::equal, *failure_target);
  1773       // successful cast, fall through to profile or jump
  1776   if (op->should_profile()) {
  1777     Register mdo  = klass_RInfo, recv = k_RInfo;
  1778     __ bind(profile_cast_success);
  1779     __ movoop(mdo, md->constant_encoding());
  1780     __ load_klass(recv, obj);
  1781     Label update_done;
  1782     type_profile_helper(mdo, md, data, recv, success);
  1783     __ jmp(*success);
  1785     __ bind(profile_cast_failure);
  1786     __ movoop(mdo, md->constant_encoding());
  1787     Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
  1788     __ subptr(counter_addr, DataLayout::counter_increment);
  1789     __ jmp(*failure);
  1791   __ jmp(*success);
  1795 void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
  1796   LIR_Code code = op->code();
  1797   if (code == lir_store_check) {
  1798     Register value = op->object()->as_register();
  1799     Register array = op->array()->as_register();
  1800     Register k_RInfo = op->tmp1()->as_register();
  1801     Register klass_RInfo = op->tmp2()->as_register();
  1802     Register Rtmp1 = op->tmp3()->as_register();
  1804     CodeStub* stub = op->stub();
  1806     // check if it needs to be profiled
  1807     ciMethodData* md;
  1808     ciProfileData* data;
  1810     if (op->should_profile()) {
  1811       ciMethod* method = op->profiled_method();
  1812       assert(method != NULL, "Should have method");
  1813       int bci = op->profiled_bci();
  1814       md = method->method_data_or_null();
  1815       assert(md != NULL, "Sanity");
  1816       data = md->bci_to_data(bci);
  1817       assert(data != NULL,                "need data for type check");
  1818       assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
  1820     Label profile_cast_success, profile_cast_failure, done;
  1821     Label *success_target = op->should_profile() ? &profile_cast_success : &done;
  1822     Label *failure_target = op->should_profile() ? &profile_cast_failure : stub->entry();
  1824     __ cmpptr(value, (int32_t)NULL_WORD);
  1825     if (op->should_profile()) {
  1826       Label not_null;
  1827       __ jccb(Assembler::notEqual, not_null);
  1828       // Object is null; update MDO and exit
  1829       Register mdo  = klass_RInfo;
  1830       __ movoop(mdo, md->constant_encoding());
  1831       Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
  1832       int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
  1833       __ orl(data_addr, header_bits);
  1834       __ jmp(done);
  1835       __ bind(not_null);
  1836     } else {
  1837       __ jcc(Assembler::equal, done);
  1840     add_debug_info_for_null_check_here(op->info_for_exception());
  1841     __ load_klass(k_RInfo, array);
  1842     __ load_klass(klass_RInfo, value);
  1844     // get instance klass (it's already uncompressed)
  1845     __ movptr(k_RInfo, Address(k_RInfo, objArrayKlass::element_klass_offset()));
  1846     // perform the fast part of the checking logic
  1847     __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, success_target, failure_target, NULL);
  1848     // call out-of-line instance of __ check_klass_subtype_slow_path(...):
  1849     __ push(klass_RInfo);
  1850     __ push(k_RInfo);
  1851     __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
  1852     __ pop(klass_RInfo);
  1853     __ pop(k_RInfo);
  1854     // result is a boolean
  1855     __ cmpl(k_RInfo, 0);
  1856     __ jcc(Assembler::equal, *failure_target);
  1857     // fall through to the success case
  1859     if (op->should_profile()) {
  1860       Register mdo  = klass_RInfo, recv = k_RInfo;
  1861       __ bind(profile_cast_success);
  1862       __ movoop(mdo, md->constant_encoding());
  1863       __ load_klass(recv, value);
  1864       Label update_done;
  1865       type_profile_helper(mdo, md, data, recv, &done);
  1866       __ jmpb(done);
  1868       __ bind(profile_cast_failure);
  1869       __ movoop(mdo, md->constant_encoding());
  1870       Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
  1871       __ subptr(counter_addr, DataLayout::counter_increment);
  1872       __ jmp(*stub->entry());
  1875     __ bind(done);
  1876   } else
  1877     if (code == lir_checkcast) {
  1878       Register obj = op->object()->as_register();
  1879       Register dst = op->result_opr()->as_register();
  1880       Label success;
  1881       emit_typecheck_helper(op, &success, op->stub()->entry(), &success);
  1882       __ bind(success);
  1883       if (dst != obj) {
  1884         __ mov(dst, obj);
  1886     } else
  1887       if (code == lir_instanceof) {
  1888         Register obj = op->object()->as_register();
  1889         Register dst = op->result_opr()->as_register();
  1890         Label success, failure, done;
  1891         emit_typecheck_helper(op, &success, &failure, &failure);
  1892         __ bind(failure);
  1893         __ xorptr(dst, dst);
  1894         __ jmpb(done);
  1895         __ bind(success);
  1896         __ movptr(dst, 1);
  1897         __ bind(done);
  1898       } else {
  1899         ShouldNotReachHere();
  1905 void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
  1906   if (LP64_ONLY(false &&) op->code() == lir_cas_long && VM_Version::supports_cx8()) {
  1907     assert(op->cmp_value()->as_register_lo() == rax, "wrong register");
  1908     assert(op->cmp_value()->as_register_hi() == rdx, "wrong register");
  1909     assert(op->new_value()->as_register_lo() == rbx, "wrong register");
  1910     assert(op->new_value()->as_register_hi() == rcx, "wrong register");
  1911     Register addr = op->addr()->as_register();
  1912     if (os::is_MP()) {
  1913       __ lock();
  1915     NOT_LP64(__ cmpxchg8(Address(addr, 0)));
  1917   } else if (op->code() == lir_cas_int || op->code() == lir_cas_obj ) {
  1918     NOT_LP64(assert(op->addr()->is_single_cpu(), "must be single");)
  1919     Register addr = (op->addr()->is_single_cpu() ? op->addr()->as_register() : op->addr()->as_register_lo());
  1920     Register newval = op->new_value()->as_register();
  1921     Register cmpval = op->cmp_value()->as_register();
  1922     assert(cmpval == rax, "wrong register");
  1923     assert(newval != NULL, "new val must be register");
  1924     assert(cmpval != newval, "cmp and new values must be in different registers");
  1925     assert(cmpval != addr, "cmp and addr must be in different registers");
  1926     assert(newval != addr, "new value and addr must be in different registers");
  1928     if ( op->code() == lir_cas_obj) {
  1929 #ifdef _LP64
  1930       if (UseCompressedOops) {
  1931         __ encode_heap_oop(cmpval);
  1932         __ mov(rscratch1, newval);
  1933         __ encode_heap_oop(rscratch1);
  1934         if (os::is_MP()) {
  1935           __ lock();
  1937         // cmpval (rax) is implicitly used by this instruction
  1938         __ cmpxchgl(rscratch1, Address(addr, 0));
  1939       } else
  1940 #endif
  1942         if (os::is_MP()) {
  1943           __ lock();
  1945         __ cmpxchgptr(newval, Address(addr, 0));
  1947     } else {
  1948       assert(op->code() == lir_cas_int, "lir_cas_int expected");
  1949       if (os::is_MP()) {
  1950         __ lock();
  1952       __ cmpxchgl(newval, Address(addr, 0));
  1954 #ifdef _LP64
  1955   } else if (op->code() == lir_cas_long) {
  1956     Register addr = (op->addr()->is_single_cpu() ? op->addr()->as_register() : op->addr()->as_register_lo());
  1957     Register newval = op->new_value()->as_register_lo();
  1958     Register cmpval = op->cmp_value()->as_register_lo();
  1959     assert(cmpval == rax, "wrong register");
  1960     assert(newval != NULL, "new val must be register");
  1961     assert(cmpval != newval, "cmp and new values must be in different registers");
  1962     assert(cmpval != addr, "cmp and addr must be in different registers");
  1963     assert(newval != addr, "new value and addr must be in different registers");
  1964     if (os::is_MP()) {
  1965       __ lock();
  1967     __ cmpxchgq(newval, Address(addr, 0));
  1968 #endif // _LP64
  1969   } else {
  1970     Unimplemented();
  1974 void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type) {
  1975   Assembler::Condition acond, ncond;
  1976   switch (condition) {
  1977     case lir_cond_equal:        acond = Assembler::equal;        ncond = Assembler::notEqual;     break;
  1978     case lir_cond_notEqual:     acond = Assembler::notEqual;     ncond = Assembler::equal;        break;
  1979     case lir_cond_less:         acond = Assembler::less;         ncond = Assembler::greaterEqual; break;
  1980     case lir_cond_lessEqual:    acond = Assembler::lessEqual;    ncond = Assembler::greater;      break;
  1981     case lir_cond_greaterEqual: acond = Assembler::greaterEqual; ncond = Assembler::less;         break;
  1982     case lir_cond_greater:      acond = Assembler::greater;      ncond = Assembler::lessEqual;    break;
  1983     case lir_cond_belowEqual:   acond = Assembler::belowEqual;   ncond = Assembler::above;        break;
  1984     case lir_cond_aboveEqual:   acond = Assembler::aboveEqual;   ncond = Assembler::below;        break;
  1985     default:                    ShouldNotReachHere();
  1988   if (opr1->is_cpu_register()) {
  1989     reg2reg(opr1, result);
  1990   } else if (opr1->is_stack()) {
  1991     stack2reg(opr1, result, result->type());
  1992   } else if (opr1->is_constant()) {
  1993     const2reg(opr1, result, lir_patch_none, NULL);
  1994   } else {
  1995     ShouldNotReachHere();
  1998   if (VM_Version::supports_cmov() && !opr2->is_constant()) {
  1999     // optimized version that does not require a branch
  2000     if (opr2->is_single_cpu()) {
  2001       assert(opr2->cpu_regnr() != result->cpu_regnr(), "opr2 already overwritten by previous move");
  2002       __ cmov(ncond, result->as_register(), opr2->as_register());
  2003     } else if (opr2->is_double_cpu()) {
  2004       assert(opr2->cpu_regnrLo() != result->cpu_regnrLo() && opr2->cpu_regnrLo() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
  2005       assert(opr2->cpu_regnrHi() != result->cpu_regnrLo() && opr2->cpu_regnrHi() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
  2006       __ cmovptr(ncond, result->as_register_lo(), opr2->as_register_lo());
  2007       NOT_LP64(__ cmovptr(ncond, result->as_register_hi(), opr2->as_register_hi());)
  2008     } else if (opr2->is_single_stack()) {
  2009       __ cmovl(ncond, result->as_register(), frame_map()->address_for_slot(opr2->single_stack_ix()));
  2010     } else if (opr2->is_double_stack()) {
  2011       __ cmovptr(ncond, result->as_register_lo(), frame_map()->address_for_slot(opr2->double_stack_ix(), lo_word_offset_in_bytes));
  2012       NOT_LP64(__ cmovptr(ncond, result->as_register_hi(), frame_map()->address_for_slot(opr2->double_stack_ix(), hi_word_offset_in_bytes));)
  2013     } else {
  2014       ShouldNotReachHere();
  2017   } else {
  2018     Label skip;
  2019     __ jcc (acond, skip);
  2020     if (opr2->is_cpu_register()) {
  2021       reg2reg(opr2, result);
  2022     } else if (opr2->is_stack()) {
  2023       stack2reg(opr2, result, result->type());
  2024     } else if (opr2->is_constant()) {
  2025       const2reg(opr2, result, lir_patch_none, NULL);
  2026     } else {
  2027       ShouldNotReachHere();
  2029     __ bind(skip);
  2034 void LIR_Assembler::arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest, CodeEmitInfo* info, bool pop_fpu_stack) {
  2035   assert(info == NULL, "should never be used, idiv/irem and ldiv/lrem not handled by this method");
  2037   if (left->is_single_cpu()) {
  2038     assert(left == dest, "left and dest must be equal");
  2039     Register lreg = left->as_register();
  2041     if (right->is_single_cpu()) {
  2042       // cpu register - cpu register
  2043       Register rreg = right->as_register();
  2044       switch (code) {
  2045         case lir_add: __ addl (lreg, rreg); break;
  2046         case lir_sub: __ subl (lreg, rreg); break;
  2047         case lir_mul: __ imull(lreg, rreg); break;
  2048         default:      ShouldNotReachHere();
  2051     } else if (right->is_stack()) {
  2052       // cpu register - stack
  2053       Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
  2054       switch (code) {
  2055         case lir_add: __ addl(lreg, raddr); break;
  2056         case lir_sub: __ subl(lreg, raddr); break;
  2057         default:      ShouldNotReachHere();
  2060     } else if (right->is_constant()) {
  2061       // cpu register - constant
  2062       jint c = right->as_constant_ptr()->as_jint();
  2063       switch (code) {
  2064         case lir_add: {
  2065           __ incrementl(lreg, c);
  2066           break;
  2068         case lir_sub: {
  2069           __ decrementl(lreg, c);
  2070           break;
  2072         default: ShouldNotReachHere();
  2075     } else {
  2076       ShouldNotReachHere();
  2079   } else if (left->is_double_cpu()) {
  2080     assert(left == dest, "left and dest must be equal");
  2081     Register lreg_lo = left->as_register_lo();
  2082     Register lreg_hi = left->as_register_hi();
  2084     if (right->is_double_cpu()) {
  2085       // cpu register - cpu register
  2086       Register rreg_lo = right->as_register_lo();
  2087       Register rreg_hi = right->as_register_hi();
  2088       NOT_LP64(assert_different_registers(lreg_lo, lreg_hi, rreg_lo, rreg_hi));
  2089       LP64_ONLY(assert_different_registers(lreg_lo, rreg_lo));
  2090       switch (code) {
  2091         case lir_add:
  2092           __ addptr(lreg_lo, rreg_lo);
  2093           NOT_LP64(__ adcl(lreg_hi, rreg_hi));
  2094           break;
  2095         case lir_sub:
  2096           __ subptr(lreg_lo, rreg_lo);
  2097           NOT_LP64(__ sbbl(lreg_hi, rreg_hi));
  2098           break;
  2099         case lir_mul:
  2100 #ifdef _LP64
  2101           __ imulq(lreg_lo, rreg_lo);
  2102 #else
  2103           assert(lreg_lo == rax && lreg_hi == rdx, "must be");
  2104           __ imull(lreg_hi, rreg_lo);
  2105           __ imull(rreg_hi, lreg_lo);
  2106           __ addl (rreg_hi, lreg_hi);
  2107           __ mull (rreg_lo);
  2108           __ addl (lreg_hi, rreg_hi);
  2109 #endif // _LP64
  2110           break;
  2111         default:
  2112           ShouldNotReachHere();
  2115     } else if (right->is_constant()) {
  2116       // cpu register - constant
  2117 #ifdef _LP64
  2118       jlong c = right->as_constant_ptr()->as_jlong_bits();
  2119       __ movptr(r10, (intptr_t) c);
  2120       switch (code) {
  2121         case lir_add:
  2122           __ addptr(lreg_lo, r10);
  2123           break;
  2124         case lir_sub:
  2125           __ subptr(lreg_lo, r10);
  2126           break;
  2127         default:
  2128           ShouldNotReachHere();
  2130 #else
  2131       jint c_lo = right->as_constant_ptr()->as_jint_lo();
  2132       jint c_hi = right->as_constant_ptr()->as_jint_hi();
  2133       switch (code) {
  2134         case lir_add:
  2135           __ addptr(lreg_lo, c_lo);
  2136           __ adcl(lreg_hi, c_hi);
  2137           break;
  2138         case lir_sub:
  2139           __ subptr(lreg_lo, c_lo);
  2140           __ sbbl(lreg_hi, c_hi);
  2141           break;
  2142         default:
  2143           ShouldNotReachHere();
  2145 #endif // _LP64
  2147     } else {
  2148       ShouldNotReachHere();
  2151   } else if (left->is_single_xmm()) {
  2152     assert(left == dest, "left and dest must be equal");
  2153     XMMRegister lreg = left->as_xmm_float_reg();
  2155     if (right->is_single_xmm()) {
  2156       XMMRegister rreg = right->as_xmm_float_reg();
  2157       switch (code) {
  2158         case lir_add: __ addss(lreg, rreg);  break;
  2159         case lir_sub: __ subss(lreg, rreg);  break;
  2160         case lir_mul_strictfp: // fall through
  2161         case lir_mul: __ mulss(lreg, rreg);  break;
  2162         case lir_div_strictfp: // fall through
  2163         case lir_div: __ divss(lreg, rreg);  break;
  2164         default: ShouldNotReachHere();
  2166     } else {
  2167       Address raddr;
  2168       if (right->is_single_stack()) {
  2169         raddr = frame_map()->address_for_slot(right->single_stack_ix());
  2170       } else if (right->is_constant()) {
  2171         // hack for now
  2172         raddr = __ as_Address(InternalAddress(float_constant(right->as_jfloat())));
  2173       } else {
  2174         ShouldNotReachHere();
  2176       switch (code) {
  2177         case lir_add: __ addss(lreg, raddr);  break;
  2178         case lir_sub: __ subss(lreg, raddr);  break;
  2179         case lir_mul_strictfp: // fall through
  2180         case lir_mul: __ mulss(lreg, raddr);  break;
  2181         case lir_div_strictfp: // fall through
  2182         case lir_div: __ divss(lreg, raddr);  break;
  2183         default: ShouldNotReachHere();
  2187   } else if (left->is_double_xmm()) {
  2188     assert(left == dest, "left and dest must be equal");
  2190     XMMRegister lreg = left->as_xmm_double_reg();
  2191     if (right->is_double_xmm()) {
  2192       XMMRegister rreg = right->as_xmm_double_reg();
  2193       switch (code) {
  2194         case lir_add: __ addsd(lreg, rreg);  break;
  2195         case lir_sub: __ subsd(lreg, rreg);  break;
  2196         case lir_mul_strictfp: // fall through
  2197         case lir_mul: __ mulsd(lreg, rreg);  break;
  2198         case lir_div_strictfp: // fall through
  2199         case lir_div: __ divsd(lreg, rreg);  break;
  2200         default: ShouldNotReachHere();
  2202     } else {
  2203       Address raddr;
  2204       if (right->is_double_stack()) {
  2205         raddr = frame_map()->address_for_slot(right->double_stack_ix());
  2206       } else if (right->is_constant()) {
  2207         // hack for now
  2208         raddr = __ as_Address(InternalAddress(double_constant(right->as_jdouble())));
  2209       } else {
  2210         ShouldNotReachHere();
  2212       switch (code) {
  2213         case lir_add: __ addsd(lreg, raddr);  break;
  2214         case lir_sub: __ subsd(lreg, raddr);  break;
  2215         case lir_mul_strictfp: // fall through
  2216         case lir_mul: __ mulsd(lreg, raddr);  break;
  2217         case lir_div_strictfp: // fall through
  2218         case lir_div: __ divsd(lreg, raddr);  break;
  2219         default: ShouldNotReachHere();
  2223   } else if (left->is_single_fpu()) {
  2224     assert(dest->is_single_fpu(),  "fpu stack allocation required");
  2226     if (right->is_single_fpu()) {
  2227       arith_fpu_implementation(code, left->fpu_regnr(), right->fpu_regnr(), dest->fpu_regnr(), pop_fpu_stack);
  2229     } else {
  2230       assert(left->fpu_regnr() == 0, "left must be on TOS");
  2231       assert(dest->fpu_regnr() == 0, "dest must be on TOS");
  2233       Address raddr;
  2234       if (right->is_single_stack()) {
  2235         raddr = frame_map()->address_for_slot(right->single_stack_ix());
  2236       } else if (right->is_constant()) {
  2237         address const_addr = float_constant(right->as_jfloat());
  2238         assert(const_addr != NULL, "incorrect float/double constant maintainance");
  2239         // hack for now
  2240         raddr = __ as_Address(InternalAddress(const_addr));
  2241       } else {
  2242         ShouldNotReachHere();
  2245       switch (code) {
  2246         case lir_add: __ fadd_s(raddr); break;
  2247         case lir_sub: __ fsub_s(raddr); break;
  2248         case lir_mul_strictfp: // fall through
  2249         case lir_mul: __ fmul_s(raddr); break;
  2250         case lir_div_strictfp: // fall through
  2251         case lir_div: __ fdiv_s(raddr); break;
  2252         default:      ShouldNotReachHere();
  2256   } else if (left->is_double_fpu()) {
  2257     assert(dest->is_double_fpu(),  "fpu stack allocation required");
  2259     if (code == lir_mul_strictfp || code == lir_div_strictfp) {
  2260       // Double values require special handling for strictfp mul/div on x86
  2261       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
  2262       __ fmulp(left->fpu_regnrLo() + 1);
  2265     if (right->is_double_fpu()) {
  2266       arith_fpu_implementation(code, left->fpu_regnrLo(), right->fpu_regnrLo(), dest->fpu_regnrLo(), pop_fpu_stack);
  2268     } else {
  2269       assert(left->fpu_regnrLo() == 0, "left must be on TOS");
  2270       assert(dest->fpu_regnrLo() == 0, "dest must be on TOS");
  2272       Address raddr;
  2273       if (right->is_double_stack()) {
  2274         raddr = frame_map()->address_for_slot(right->double_stack_ix());
  2275       } else if (right->is_constant()) {
  2276         // hack for now
  2277         raddr = __ as_Address(InternalAddress(double_constant(right->as_jdouble())));
  2278       } else {
  2279         ShouldNotReachHere();
  2282       switch (code) {
  2283         case lir_add: __ fadd_d(raddr); break;
  2284         case lir_sub: __ fsub_d(raddr); break;
  2285         case lir_mul_strictfp: // fall through
  2286         case lir_mul: __ fmul_d(raddr); break;
  2287         case lir_div_strictfp: // fall through
  2288         case lir_div: __ fdiv_d(raddr); break;
  2289         default: ShouldNotReachHere();
  2293     if (code == lir_mul_strictfp || code == lir_div_strictfp) {
  2294       // Double values require special handling for strictfp mul/div on x86
  2295       __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
  2296       __ fmulp(dest->fpu_regnrLo() + 1);
  2299   } else if (left->is_single_stack() || left->is_address()) {
  2300     assert(left == dest, "left and dest must be equal");
  2302     Address laddr;
  2303     if (left->is_single_stack()) {
  2304       laddr = frame_map()->address_for_slot(left->single_stack_ix());
  2305     } else if (left->is_address()) {
  2306       laddr = as_Address(left->as_address_ptr());
  2307     } else {
  2308       ShouldNotReachHere();
  2311     if (right->is_single_cpu()) {
  2312       Register rreg = right->as_register();
  2313       switch (code) {
  2314         case lir_add: __ addl(laddr, rreg); break;
  2315         case lir_sub: __ subl(laddr, rreg); break;
  2316         default:      ShouldNotReachHere();
  2318     } else if (right->is_constant()) {
  2319       jint c = right->as_constant_ptr()->as_jint();
  2320       switch (code) {
  2321         case lir_add: {
  2322           __ incrementl(laddr, c);
  2323           break;
  2325         case lir_sub: {
  2326           __ decrementl(laddr, c);
  2327           break;
  2329         default: ShouldNotReachHere();
  2331     } else {
  2332       ShouldNotReachHere();
  2335   } else {
  2336     ShouldNotReachHere();
  2340 void LIR_Assembler::arith_fpu_implementation(LIR_Code code, int left_index, int right_index, int dest_index, bool pop_fpu_stack) {
  2341   assert(pop_fpu_stack  || (left_index     == dest_index || right_index     == dest_index), "invalid LIR");
  2342   assert(!pop_fpu_stack || (left_index - 1 == dest_index || right_index - 1 == dest_index), "invalid LIR");
  2343   assert(left_index == 0 || right_index == 0, "either must be on top of stack");
  2345   bool left_is_tos = (left_index == 0);
  2346   bool dest_is_tos = (dest_index == 0);
  2347   int non_tos_index = (left_is_tos ? right_index : left_index);
  2349   switch (code) {
  2350     case lir_add:
  2351       if (pop_fpu_stack)       __ faddp(non_tos_index);
  2352       else if (dest_is_tos)    __ fadd (non_tos_index);
  2353       else                     __ fadda(non_tos_index);
  2354       break;
  2356     case lir_sub:
  2357       if (left_is_tos) {
  2358         if (pop_fpu_stack)     __ fsubrp(non_tos_index);
  2359         else if (dest_is_tos)  __ fsub  (non_tos_index);
  2360         else                   __ fsubra(non_tos_index);
  2361       } else {
  2362         if (pop_fpu_stack)     __ fsubp (non_tos_index);
  2363         else if (dest_is_tos)  __ fsubr (non_tos_index);
  2364         else                   __ fsuba (non_tos_index);
  2366       break;
  2368     case lir_mul_strictfp: // fall through
  2369     case lir_mul:
  2370       if (pop_fpu_stack)       __ fmulp(non_tos_index);
  2371       else if (dest_is_tos)    __ fmul (non_tos_index);
  2372       else                     __ fmula(non_tos_index);
  2373       break;
  2375     case lir_div_strictfp: // fall through
  2376     case lir_div:
  2377       if (left_is_tos) {
  2378         if (pop_fpu_stack)     __ fdivrp(non_tos_index);
  2379         else if (dest_is_tos)  __ fdiv  (non_tos_index);
  2380         else                   __ fdivra(non_tos_index);
  2381       } else {
  2382         if (pop_fpu_stack)     __ fdivp (non_tos_index);
  2383         else if (dest_is_tos)  __ fdivr (non_tos_index);
  2384         else                   __ fdiva (non_tos_index);
  2386       break;
  2388     case lir_rem:
  2389       assert(left_is_tos && dest_is_tos && right_index == 1, "must be guaranteed by FPU stack allocation");
  2390       __ fremr(noreg);
  2391       break;
  2393     default:
  2394       ShouldNotReachHere();
  2399 void LIR_Assembler::intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr unused, LIR_Opr dest, LIR_Op* op) {
  2400   if (value->is_double_xmm()) {
  2401     switch(code) {
  2402       case lir_abs :
  2404           if (dest->as_xmm_double_reg() != value->as_xmm_double_reg()) {
  2405             __ movdbl(dest->as_xmm_double_reg(), value->as_xmm_double_reg());
  2407           __ andpd(dest->as_xmm_double_reg(),
  2408                     ExternalAddress((address)double_signmask_pool));
  2410         break;
  2412       case lir_sqrt: __ sqrtsd(dest->as_xmm_double_reg(), value->as_xmm_double_reg()); break;
  2413       // all other intrinsics are not available in the SSE instruction set, so FPU is used
  2414       default      : ShouldNotReachHere();
  2417   } else if (value->is_double_fpu()) {
  2418     assert(value->fpu_regnrLo() == 0 && dest->fpu_regnrLo() == 0, "both must be on TOS");
  2419     switch(code) {
  2420       case lir_log   : __ flog() ; break;
  2421       case lir_log10 : __ flog10() ; break;
  2422       case lir_abs   : __ fabs() ; break;
  2423       case lir_sqrt  : __ fsqrt(); break;
  2424       case lir_sin   :
  2425         // Should consider not saving rbx, if not necessary
  2426         __ trigfunc('s', op->as_Op2()->fpu_stack_size());
  2427         break;
  2428       case lir_cos :
  2429         // Should consider not saving rbx, if not necessary
  2430         assert(op->as_Op2()->fpu_stack_size() <= 6, "sin and cos need two free stack slots");
  2431         __ trigfunc('c', op->as_Op2()->fpu_stack_size());
  2432         break;
  2433       case lir_tan :
  2434         // Should consider not saving rbx, if not necessary
  2435         __ trigfunc('t', op->as_Op2()->fpu_stack_size());
  2436         break;
  2437       default      : ShouldNotReachHere();
  2439   } else {
  2440     Unimplemented();
  2444 void LIR_Assembler::logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst) {
  2445   // assert(left->destroys_register(), "check");
  2446   if (left->is_single_cpu()) {
  2447     Register reg = left->as_register();
  2448     if (right->is_constant()) {
  2449       int val = right->as_constant_ptr()->as_jint();
  2450       switch (code) {
  2451         case lir_logic_and: __ andl (reg, val); break;
  2452         case lir_logic_or:  __ orl  (reg, val); break;
  2453         case lir_logic_xor: __ xorl (reg, val); break;
  2454         default: ShouldNotReachHere();
  2456     } else if (right->is_stack()) {
  2457       // added support for stack operands
  2458       Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
  2459       switch (code) {
  2460         case lir_logic_and: __ andl (reg, raddr); break;
  2461         case lir_logic_or:  __ orl  (reg, raddr); break;
  2462         case lir_logic_xor: __ xorl (reg, raddr); break;
  2463         default: ShouldNotReachHere();
  2465     } else {
  2466       Register rright = right->as_register();
  2467       switch (code) {
  2468         case lir_logic_and: __ andptr (reg, rright); break;
  2469         case lir_logic_or : __ orptr  (reg, rright); break;
  2470         case lir_logic_xor: __ xorptr (reg, rright); break;
  2471         default: ShouldNotReachHere();
  2474     move_regs(reg, dst->as_register());
  2475   } else {
  2476     Register l_lo = left->as_register_lo();
  2477     Register l_hi = left->as_register_hi();
  2478     if (right->is_constant()) {
  2479 #ifdef _LP64
  2480       __ mov64(rscratch1, right->as_constant_ptr()->as_jlong());
  2481       switch (code) {
  2482         case lir_logic_and:
  2483           __ andq(l_lo, rscratch1);
  2484           break;
  2485         case lir_logic_or:
  2486           __ orq(l_lo, rscratch1);
  2487           break;
  2488         case lir_logic_xor:
  2489           __ xorq(l_lo, rscratch1);
  2490           break;
  2491         default: ShouldNotReachHere();
  2493 #else
  2494       int r_lo = right->as_constant_ptr()->as_jint_lo();
  2495       int r_hi = right->as_constant_ptr()->as_jint_hi();
  2496       switch (code) {
  2497         case lir_logic_and:
  2498           __ andl(l_lo, r_lo);
  2499           __ andl(l_hi, r_hi);
  2500           break;
  2501         case lir_logic_or:
  2502           __ orl(l_lo, r_lo);
  2503           __ orl(l_hi, r_hi);
  2504           break;
  2505         case lir_logic_xor:
  2506           __ xorl(l_lo, r_lo);
  2507           __ xorl(l_hi, r_hi);
  2508           break;
  2509         default: ShouldNotReachHere();
  2511 #endif // _LP64
  2512     } else {
  2513 #ifdef _LP64
  2514       Register r_lo;
  2515       if (right->type() == T_OBJECT || right->type() == T_ARRAY) {
  2516         r_lo = right->as_register();
  2517       } else {
  2518         r_lo = right->as_register_lo();
  2520 #else
  2521       Register r_lo = right->as_register_lo();
  2522       Register r_hi = right->as_register_hi();
  2523       assert(l_lo != r_hi, "overwriting registers");
  2524 #endif
  2525       switch (code) {
  2526         case lir_logic_and:
  2527           __ andptr(l_lo, r_lo);
  2528           NOT_LP64(__ andptr(l_hi, r_hi);)
  2529           break;
  2530         case lir_logic_or:
  2531           __ orptr(l_lo, r_lo);
  2532           NOT_LP64(__ orptr(l_hi, r_hi);)
  2533           break;
  2534         case lir_logic_xor:
  2535           __ xorptr(l_lo, r_lo);
  2536           NOT_LP64(__ xorptr(l_hi, r_hi);)
  2537           break;
  2538         default: ShouldNotReachHere();
  2542     Register dst_lo = dst->as_register_lo();
  2543     Register dst_hi = dst->as_register_hi();
  2545 #ifdef _LP64
  2546     move_regs(l_lo, dst_lo);
  2547 #else
  2548     if (dst_lo == l_hi) {
  2549       assert(dst_hi != l_lo, "overwriting registers");
  2550       move_regs(l_hi, dst_hi);
  2551       move_regs(l_lo, dst_lo);
  2552     } else {
  2553       assert(dst_lo != l_hi, "overwriting registers");
  2554       move_regs(l_lo, dst_lo);
  2555       move_regs(l_hi, dst_hi);
  2557 #endif // _LP64
  2562 // we assume that rax, and rdx can be overwritten
  2563 void LIR_Assembler::arithmetic_idiv(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr temp, LIR_Opr result, CodeEmitInfo* info) {
  2565   assert(left->is_single_cpu(),   "left must be register");
  2566   assert(right->is_single_cpu() || right->is_constant(),  "right must be register or constant");
  2567   assert(result->is_single_cpu(), "result must be register");
  2569   //  assert(left->destroys_register(), "check");
  2570   //  assert(right->destroys_register(), "check");
  2572   Register lreg = left->as_register();
  2573   Register dreg = result->as_register();
  2575   if (right->is_constant()) {
  2576     int divisor = right->as_constant_ptr()->as_jint();
  2577     assert(divisor > 0 && is_power_of_2(divisor), "must be");
  2578     if (code == lir_idiv) {
  2579       assert(lreg == rax, "must be rax,");
  2580       assert(temp->as_register() == rdx, "tmp register must be rdx");
  2581       __ cdql(); // sign extend into rdx:rax
  2582       if (divisor == 2) {
  2583         __ subl(lreg, rdx);
  2584       } else {
  2585         __ andl(rdx, divisor - 1);
  2586         __ addl(lreg, rdx);
  2588       __ sarl(lreg, log2_intptr(divisor));
  2589       move_regs(lreg, dreg);
  2590     } else if (code == lir_irem) {
  2591       Label done;
  2592       __ mov(dreg, lreg);
  2593       __ andl(dreg, 0x80000000 | (divisor - 1));
  2594       __ jcc(Assembler::positive, done);
  2595       __ decrement(dreg);
  2596       __ orl(dreg, ~(divisor - 1));
  2597       __ increment(dreg);
  2598       __ bind(done);
  2599     } else {
  2600       ShouldNotReachHere();
  2602   } else {
  2603     Register rreg = right->as_register();
  2604     assert(lreg == rax, "left register must be rax,");
  2605     assert(rreg != rdx, "right register must not be rdx");
  2606     assert(temp->as_register() == rdx, "tmp register must be rdx");
  2608     move_regs(lreg, rax);
  2610     int idivl_offset = __ corrected_idivl(rreg);
  2611     add_debug_info_for_div0(idivl_offset, info);
  2612     if (code == lir_irem) {
  2613       move_regs(rdx, dreg); // result is in rdx
  2614     } else {
  2615       move_regs(rax, dreg);
  2621 void LIR_Assembler::comp_op(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Op2* op) {
  2622   if (opr1->is_single_cpu()) {
  2623     Register reg1 = opr1->as_register();
  2624     if (opr2->is_single_cpu()) {
  2625       // cpu register - cpu register
  2626       if (opr1->type() == T_OBJECT || opr1->type() == T_ARRAY) {
  2627         __ cmpptr(reg1, opr2->as_register());
  2628       } else {
  2629         assert(opr2->type() != T_OBJECT && opr2->type() != T_ARRAY, "cmp int, oop?");
  2630         __ cmpl(reg1, opr2->as_register());
  2632     } else if (opr2->is_stack()) {
  2633       // cpu register - stack
  2634       if (opr1->type() == T_OBJECT || opr1->type() == T_ARRAY) {
  2635         __ cmpptr(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
  2636       } else {
  2637         __ cmpl(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
  2639     } else if (opr2->is_constant()) {
  2640       // cpu register - constant
  2641       LIR_Const* c = opr2->as_constant_ptr();
  2642       if (c->type() == T_INT) {
  2643         __ cmpl(reg1, c->as_jint());
  2644       } else if (c->type() == T_OBJECT || c->type() == T_ARRAY) {
  2645         // In 64bit oops are single register
  2646         jobject o = c->as_jobject();
  2647         if (o == NULL) {
  2648           __ cmpptr(reg1, (int32_t)NULL_WORD);
  2649         } else {
  2650 #ifdef _LP64
  2651           __ movoop(rscratch1, o);
  2652           __ cmpptr(reg1, rscratch1);
  2653 #else
  2654           __ cmpoop(reg1, c->as_jobject());
  2655 #endif // _LP64
  2657       } else {
  2658         ShouldNotReachHere();
  2660       // cpu register - address
  2661     } else if (opr2->is_address()) {
  2662       if (op->info() != NULL) {
  2663         add_debug_info_for_null_check_here(op->info());
  2665       __ cmpl(reg1, as_Address(opr2->as_address_ptr()));
  2666     } else {
  2667       ShouldNotReachHere();
  2670   } else if(opr1->is_double_cpu()) {
  2671     Register xlo = opr1->as_register_lo();
  2672     Register xhi = opr1->as_register_hi();
  2673     if (opr2->is_double_cpu()) {
  2674 #ifdef _LP64
  2675       __ cmpptr(xlo, opr2->as_register_lo());
  2676 #else
  2677       // cpu register - cpu register
  2678       Register ylo = opr2->as_register_lo();
  2679       Register yhi = opr2->as_register_hi();
  2680       __ subl(xlo, ylo);
  2681       __ sbbl(xhi, yhi);
  2682       if (condition == lir_cond_equal || condition == lir_cond_notEqual) {
  2683         __ orl(xhi, xlo);
  2685 #endif // _LP64
  2686     } else if (opr2->is_constant()) {
  2687       // cpu register - constant 0
  2688       assert(opr2->as_jlong() == (jlong)0, "only handles zero");
  2689 #ifdef _LP64
  2690       __ cmpptr(xlo, (int32_t)opr2->as_jlong());
  2691 #else
  2692       assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "only handles equals case");
  2693       __ orl(xhi, xlo);
  2694 #endif // _LP64
  2695     } else {
  2696       ShouldNotReachHere();
  2699   } else if (opr1->is_single_xmm()) {
  2700     XMMRegister reg1 = opr1->as_xmm_float_reg();
  2701     if (opr2->is_single_xmm()) {
  2702       // xmm register - xmm register
  2703       __ ucomiss(reg1, opr2->as_xmm_float_reg());
  2704     } else if (opr2->is_stack()) {
  2705       // xmm register - stack
  2706       __ ucomiss(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
  2707     } else if (opr2->is_constant()) {
  2708       // xmm register - constant
  2709       __ ucomiss(reg1, InternalAddress(float_constant(opr2->as_jfloat())));
  2710     } else if (opr2->is_address()) {
  2711       // xmm register - address
  2712       if (op->info() != NULL) {
  2713         add_debug_info_for_null_check_here(op->info());
  2715       __ ucomiss(reg1, as_Address(opr2->as_address_ptr()));
  2716     } else {
  2717       ShouldNotReachHere();
  2720   } else if (opr1->is_double_xmm()) {
  2721     XMMRegister reg1 = opr1->as_xmm_double_reg();
  2722     if (opr2->is_double_xmm()) {
  2723       // xmm register - xmm register
  2724       __ ucomisd(reg1, opr2->as_xmm_double_reg());
  2725     } else if (opr2->is_stack()) {
  2726       // xmm register - stack
  2727       __ ucomisd(reg1, frame_map()->address_for_slot(opr2->double_stack_ix()));
  2728     } else if (opr2->is_constant()) {
  2729       // xmm register - constant
  2730       __ ucomisd(reg1, InternalAddress(double_constant(opr2->as_jdouble())));
  2731     } else if (opr2->is_address()) {
  2732       // xmm register - address
  2733       if (op->info() != NULL) {
  2734         add_debug_info_for_null_check_here(op->info());
  2736       __ ucomisd(reg1, as_Address(opr2->pointer()->as_address()));
  2737     } else {
  2738       ShouldNotReachHere();
  2741   } else if(opr1->is_single_fpu() || opr1->is_double_fpu()) {
  2742     assert(opr1->is_fpu_register() && opr1->fpu() == 0, "currently left-hand side must be on TOS (relax this restriction)");
  2743     assert(opr2->is_fpu_register(), "both must be registers");
  2744     __ fcmp(noreg, opr2->fpu(), op->fpu_pop_count() > 0, op->fpu_pop_count() > 1);
  2746   } else if (opr1->is_address() && opr2->is_constant()) {
  2747     LIR_Const* c = opr2->as_constant_ptr();
  2748 #ifdef _LP64
  2749     if (c->type() == T_OBJECT || c->type() == T_ARRAY) {
  2750       assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "need to reverse");
  2751       __ movoop(rscratch1, c->as_jobject());
  2753 #endif // LP64
  2754     if (op->info() != NULL) {
  2755       add_debug_info_for_null_check_here(op->info());
  2757     // special case: address - constant
  2758     LIR_Address* addr = opr1->as_address_ptr();
  2759     if (c->type() == T_INT) {
  2760       __ cmpl(as_Address(addr), c->as_jint());
  2761     } else if (c->type() == T_OBJECT || c->type() == T_ARRAY) {
  2762 #ifdef _LP64
  2763       // %%% Make this explode if addr isn't reachable until we figure out a
  2764       // better strategy by giving noreg as the temp for as_Address
  2765       __ cmpptr(rscratch1, as_Address(addr, noreg));
  2766 #else
  2767       __ cmpoop(as_Address(addr), c->as_jobject());
  2768 #endif // _LP64
  2769     } else {
  2770       ShouldNotReachHere();
  2773   } else {
  2774     ShouldNotReachHere();
  2778 void LIR_Assembler::comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst, LIR_Op2* op) {
  2779   if (code == lir_cmp_fd2i || code == lir_ucmp_fd2i) {
  2780     if (left->is_single_xmm()) {
  2781       assert(right->is_single_xmm(), "must match");
  2782       __ cmpss2int(left->as_xmm_float_reg(), right->as_xmm_float_reg(), dst->as_register(), code == lir_ucmp_fd2i);
  2783     } else if (left->is_double_xmm()) {
  2784       assert(right->is_double_xmm(), "must match");
  2785       __ cmpsd2int(left->as_xmm_double_reg(), right->as_xmm_double_reg(), dst->as_register(), code == lir_ucmp_fd2i);
  2787     } else {
  2788       assert(left->is_single_fpu() || left->is_double_fpu(), "must be");
  2789       assert(right->is_single_fpu() || right->is_double_fpu(), "must match");
  2791       assert(left->fpu() == 0, "left must be on TOS");
  2792       __ fcmp2int(dst->as_register(), code == lir_ucmp_fd2i, right->fpu(),
  2793                   op->fpu_pop_count() > 0, op->fpu_pop_count() > 1);
  2795   } else {
  2796     assert(code == lir_cmp_l2i, "check");
  2797 #ifdef _LP64
  2798     Label done;
  2799     Register dest = dst->as_register();
  2800     __ cmpptr(left->as_register_lo(), right->as_register_lo());
  2801     __ movl(dest, -1);
  2802     __ jccb(Assembler::less, done);
  2803     __ set_byte_if_not_zero(dest);
  2804     __ movzbl(dest, dest);
  2805     __ bind(done);
  2806 #else
  2807     __ lcmp2int(left->as_register_hi(),
  2808                 left->as_register_lo(),
  2809                 right->as_register_hi(),
  2810                 right->as_register_lo());
  2811     move_regs(left->as_register_hi(), dst->as_register());
  2812 #endif // _LP64
  2817 void LIR_Assembler::align_call(LIR_Code code) {
  2818   if (os::is_MP()) {
  2819     // make sure that the displacement word of the call ends up word aligned
  2820     int offset = __ offset();
  2821     switch (code) {
  2822       case lir_static_call:
  2823       case lir_optvirtual_call:
  2824       case lir_dynamic_call:
  2825         offset += NativeCall::displacement_offset;
  2826         break;
  2827       case lir_icvirtual_call:
  2828         offset += NativeCall::displacement_offset + NativeMovConstReg::instruction_size;
  2829       break;
  2830       case lir_virtual_call:  // currently, sparc-specific for niagara
  2831       default: ShouldNotReachHere();
  2833     while (offset++ % BytesPerWord != 0) {
  2834       __ nop();
  2840 void LIR_Assembler::call(LIR_OpJavaCall* op, relocInfo::relocType rtype) {
  2841   assert(!os::is_MP() || (__ offset() + NativeCall::displacement_offset) % BytesPerWord == 0,
  2842          "must be aligned");
  2843   __ call(AddressLiteral(op->addr(), rtype));
  2844   add_call_info(code_offset(), op->info());
  2848 void LIR_Assembler::ic_call(LIR_OpJavaCall* op) {
  2849   RelocationHolder rh = virtual_call_Relocation::spec(pc());
  2850   __ movoop(IC_Klass, (jobject)Universe::non_oop_word());
  2851   assert(!os::is_MP() ||
  2852          (__ offset() + NativeCall::displacement_offset) % BytesPerWord == 0,
  2853          "must be aligned");
  2854   __ call(AddressLiteral(op->addr(), rh));
  2855   add_call_info(code_offset(), op->info());
  2859 /* Currently, vtable-dispatch is only enabled for sparc platforms */
  2860 void LIR_Assembler::vtable_call(LIR_OpJavaCall* op) {
  2861   ShouldNotReachHere();
  2865 void LIR_Assembler::emit_static_call_stub() {
  2866   address call_pc = __ pc();
  2867   address stub = __ start_a_stub(call_stub_size);
  2868   if (stub == NULL) {
  2869     bailout("static call stub overflow");
  2870     return;
  2873   int start = __ offset();
  2874   if (os::is_MP()) {
  2875     // make sure that the displacement word of the call ends up word aligned
  2876     int offset = __ offset() + NativeMovConstReg::instruction_size + NativeCall::displacement_offset;
  2877     while (offset++ % BytesPerWord != 0) {
  2878       __ nop();
  2881   __ relocate(static_stub_Relocation::spec(call_pc));
  2882   __ movoop(rbx, (jobject)NULL);
  2883   // must be set to -1 at code generation time
  2884   assert(!os::is_MP() || ((__ offset() + 1) % BytesPerWord) == 0, "must be aligned on MP");
  2885   // On 64bit this will die since it will take a movq & jmp, must be only a jmp
  2886   __ jump(RuntimeAddress(__ pc()));
  2888   assert(__ offset() - start <= call_stub_size, "stub too big");
  2889   __ end_a_stub();
  2893 void LIR_Assembler::throw_op(LIR_Opr exceptionPC, LIR_Opr exceptionOop, CodeEmitInfo* info) {
  2894   assert(exceptionOop->as_register() == rax, "must match");
  2895   assert(exceptionPC->as_register() == rdx, "must match");
  2897   // exception object is not added to oop map by LinearScan
  2898   // (LinearScan assumes that no oops are in fixed registers)
  2899   info->add_register_oop(exceptionOop);
  2900   Runtime1::StubID unwind_id;
  2902   // get current pc information
  2903   // pc is only needed if the method has an exception handler, the unwind code does not need it.
  2904   int pc_for_athrow_offset = __ offset();
  2905   InternalAddress pc_for_athrow(__ pc());
  2906   __ lea(exceptionPC->as_register(), pc_for_athrow);
  2907   add_call_info(pc_for_athrow_offset, info); // for exception handler
  2909   __ verify_not_null_oop(rax);
  2910   // search an exception handler (rax: exception oop, rdx: throwing pc)
  2911   if (compilation()->has_fpu_code()) {
  2912     unwind_id = Runtime1::handle_exception_id;
  2913   } else {
  2914     unwind_id = Runtime1::handle_exception_nofpu_id;
  2916   __ call(RuntimeAddress(Runtime1::entry_for(unwind_id)));
  2918   // enough room for two byte trap
  2919   __ nop();
  2923 void LIR_Assembler::unwind_op(LIR_Opr exceptionOop) {
  2924   assert(exceptionOop->as_register() == rax, "must match");
  2926   __ jmp(_unwind_handler_entry);
  2930 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
  2932   // optimized version for linear scan:
  2933   // * count must be already in ECX (guaranteed by LinearScan)
  2934   // * left and dest must be equal
  2935   // * tmp must be unused
  2936   assert(count->as_register() == SHIFT_count, "count must be in ECX");
  2937   assert(left == dest, "left and dest must be equal");
  2938   assert(tmp->is_illegal(), "wasting a register if tmp is allocated");
  2940   if (left->is_single_cpu()) {
  2941     Register value = left->as_register();
  2942     assert(value != SHIFT_count, "left cannot be ECX");
  2944     switch (code) {
  2945       case lir_shl:  __ shll(value); break;
  2946       case lir_shr:  __ sarl(value); break;
  2947       case lir_ushr: __ shrl(value); break;
  2948       default: ShouldNotReachHere();
  2950   } else if (left->is_double_cpu()) {
  2951     Register lo = left->as_register_lo();
  2952     Register hi = left->as_register_hi();
  2953     assert(lo != SHIFT_count && hi != SHIFT_count, "left cannot be ECX");
  2954 #ifdef _LP64
  2955     switch (code) {
  2956       case lir_shl:  __ shlptr(lo);        break;
  2957       case lir_shr:  __ sarptr(lo);        break;
  2958       case lir_ushr: __ shrptr(lo);        break;
  2959       default: ShouldNotReachHere();
  2961 #else
  2963     switch (code) {
  2964       case lir_shl:  __ lshl(hi, lo);        break;
  2965       case lir_shr:  __ lshr(hi, lo, true);  break;
  2966       case lir_ushr: __ lshr(hi, lo, false); break;
  2967       default: ShouldNotReachHere();
  2969 #endif // LP64
  2970   } else {
  2971     ShouldNotReachHere();
  2976 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, jint count, LIR_Opr dest) {
  2977   if (dest->is_single_cpu()) {
  2978     // first move left into dest so that left is not destroyed by the shift
  2979     Register value = dest->as_register();
  2980     count = count & 0x1F; // Java spec
  2982     move_regs(left->as_register(), value);
  2983     switch (code) {
  2984       case lir_shl:  __ shll(value, count); break;
  2985       case lir_shr:  __ sarl(value, count); break;
  2986       case lir_ushr: __ shrl(value, count); break;
  2987       default: ShouldNotReachHere();
  2989   } else if (dest->is_double_cpu()) {
  2990 #ifndef _LP64
  2991     Unimplemented();
  2992 #else
  2993     // first move left into dest so that left is not destroyed by the shift
  2994     Register value = dest->as_register_lo();
  2995     count = count & 0x1F; // Java spec
  2997     move_regs(left->as_register_lo(), value);
  2998     switch (code) {
  2999       case lir_shl:  __ shlptr(value, count); break;
  3000       case lir_shr:  __ sarptr(value, count); break;
  3001       case lir_ushr: __ shrptr(value, count); break;
  3002       default: ShouldNotReachHere();
  3004 #endif // _LP64
  3005   } else {
  3006     ShouldNotReachHere();
  3011 void LIR_Assembler::store_parameter(Register r, int offset_from_rsp_in_words) {
  3012   assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
  3013   int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
  3014   assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
  3015   __ movptr (Address(rsp, offset_from_rsp_in_bytes), r);
  3019 void LIR_Assembler::store_parameter(jint c,     int offset_from_rsp_in_words) {
  3020   assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
  3021   int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
  3022   assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
  3023   __ movptr (Address(rsp, offset_from_rsp_in_bytes), c);
  3027 void LIR_Assembler::store_parameter(jobject o,  int offset_from_rsp_in_words) {
  3028   assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
  3029   int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
  3030   assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
  3031   __ movoop (Address(rsp, offset_from_rsp_in_bytes), o);
  3035 // This code replaces a call to arraycopy; no exception may
  3036 // be thrown in this code, they must be thrown in the System.arraycopy
  3037 // activation frame; we could save some checks if this would not be the case
  3038 void LIR_Assembler::emit_arraycopy(LIR_OpArrayCopy* op) {
  3039   ciArrayKlass* default_type = op->expected_type();
  3040   Register src = op->src()->as_register();
  3041   Register dst = op->dst()->as_register();
  3042   Register src_pos = op->src_pos()->as_register();
  3043   Register dst_pos = op->dst_pos()->as_register();
  3044   Register length  = op->length()->as_register();
  3045   Register tmp = op->tmp()->as_register();
  3047   CodeStub* stub = op->stub();
  3048   int flags = op->flags();
  3049   BasicType basic_type = default_type != NULL ? default_type->element_type()->basic_type() : T_ILLEGAL;
  3050   if (basic_type == T_ARRAY) basic_type = T_OBJECT;
  3052   // if we don't know anything, just go through the generic arraycopy
  3053   if (default_type == NULL) {
  3054     Label done;
  3055     // save outgoing arguments on stack in case call to System.arraycopy is needed
  3056     // HACK ALERT. This code used to push the parameters in a hardwired fashion
  3057     // for interpreter calling conventions. Now we have to do it in new style conventions.
  3058     // For the moment until C1 gets the new register allocator I just force all the
  3059     // args to the right place (except the register args) and then on the back side
  3060     // reload the register args properly if we go slow path. Yuck
  3062     // These are proper for the calling convention
  3063     store_parameter(length, 2);
  3064     store_parameter(dst_pos, 1);
  3065     store_parameter(dst, 0);
  3067     // these are just temporary placements until we need to reload
  3068     store_parameter(src_pos, 3);
  3069     store_parameter(src, 4);
  3070     NOT_LP64(assert(src == rcx && src_pos == rdx, "mismatch in calling convention");)
  3072     address C_entry = CAST_FROM_FN_PTR(address, Runtime1::arraycopy);
  3074     address copyfunc_addr = StubRoutines::generic_arraycopy();
  3076     // pass arguments: may push as this is not a safepoint; SP must be fix at each safepoint
  3077 #ifdef _LP64
  3078     // The arguments are in java calling convention so we can trivially shift them to C
  3079     // convention
  3080     assert_different_registers(c_rarg0, j_rarg1, j_rarg2, j_rarg3, j_rarg4);
  3081     __ mov(c_rarg0, j_rarg0);
  3082     assert_different_registers(c_rarg1, j_rarg2, j_rarg3, j_rarg4);
  3083     __ mov(c_rarg1, j_rarg1);
  3084     assert_different_registers(c_rarg2, j_rarg3, j_rarg4);
  3085     __ mov(c_rarg2, j_rarg2);
  3086     assert_different_registers(c_rarg3, j_rarg4);
  3087     __ mov(c_rarg3, j_rarg3);
  3088 #ifdef _WIN64
  3089     // Allocate abi space for args but be sure to keep stack aligned
  3090     __ subptr(rsp, 6*wordSize);
  3091     store_parameter(j_rarg4, 4);
  3092     if (copyfunc_addr == NULL) { // Use C version if stub was not generated
  3093       __ call(RuntimeAddress(C_entry));
  3094     } else {
  3095 #ifndef PRODUCT
  3096       if (PrintC1Statistics) {
  3097         __ incrementl(ExternalAddress((address)&Runtime1::_generic_arraycopystub_cnt));
  3099 #endif
  3100       __ call(RuntimeAddress(copyfunc_addr));
  3102     __ addptr(rsp, 6*wordSize);
  3103 #else
  3104     __ mov(c_rarg4, j_rarg4);
  3105     if (copyfunc_addr == NULL) { // Use C version if stub was not generated
  3106       __ call(RuntimeAddress(C_entry));
  3107     } else {
  3108 #ifndef PRODUCT
  3109       if (PrintC1Statistics) {
  3110         __ incrementl(ExternalAddress((address)&Runtime1::_generic_arraycopystub_cnt));
  3112 #endif
  3113       __ call(RuntimeAddress(copyfunc_addr));
  3115 #endif // _WIN64
  3116 #else
  3117     __ push(length);
  3118     __ push(dst_pos);
  3119     __ push(dst);
  3120     __ push(src_pos);
  3121     __ push(src);
  3123     if (copyfunc_addr == NULL) { // Use C version if stub was not generated
  3124       __ call_VM_leaf(C_entry, 5); // removes pushed parameter from the stack
  3125     } else {
  3126 #ifndef PRODUCT
  3127       if (PrintC1Statistics) {
  3128         __ incrementl(ExternalAddress((address)&Runtime1::_generic_arraycopystub_cnt));
  3130 #endif
  3131       __ call_VM_leaf(copyfunc_addr, 5); // removes pushed parameter from the stack
  3134 #endif // _LP64
  3136     __ cmpl(rax, 0);
  3137     __ jcc(Assembler::equal, *stub->continuation());
  3139     if (copyfunc_addr != NULL) {
  3140       __ mov(tmp, rax);
  3141       __ xorl(tmp, -1);
  3144     // Reload values from the stack so they are where the stub
  3145     // expects them.
  3146     __ movptr   (dst,     Address(rsp, 0*BytesPerWord));
  3147     __ movptr   (dst_pos, Address(rsp, 1*BytesPerWord));
  3148     __ movptr   (length,  Address(rsp, 2*BytesPerWord));
  3149     __ movptr   (src_pos, Address(rsp, 3*BytesPerWord));
  3150     __ movptr   (src,     Address(rsp, 4*BytesPerWord));
  3152     if (copyfunc_addr != NULL) {
  3153       __ subl(length, tmp);
  3154       __ addl(src_pos, tmp);
  3155       __ addl(dst_pos, tmp);
  3157     __ jmp(*stub->entry());
  3159     __ bind(*stub->continuation());
  3160     return;
  3163   assert(default_type != NULL && default_type->is_array_klass() && default_type->is_loaded(), "must be true at this point");
  3165   int elem_size = type2aelembytes(basic_type);
  3166   int shift_amount;
  3167   Address::ScaleFactor scale;
  3169   switch (elem_size) {
  3170     case 1 :
  3171       shift_amount = 0;
  3172       scale = Address::times_1;
  3173       break;
  3174     case 2 :
  3175       shift_amount = 1;
  3176       scale = Address::times_2;
  3177       break;
  3178     case 4 :
  3179       shift_amount = 2;
  3180       scale = Address::times_4;
  3181       break;
  3182     case 8 :
  3183       shift_amount = 3;
  3184       scale = Address::times_8;
  3185       break;
  3186     default:
  3187       ShouldNotReachHere();
  3190   Address src_length_addr = Address(src, arrayOopDesc::length_offset_in_bytes());
  3191   Address dst_length_addr = Address(dst, arrayOopDesc::length_offset_in_bytes());
  3192   Address src_klass_addr = Address(src, oopDesc::klass_offset_in_bytes());
  3193   Address dst_klass_addr = Address(dst, oopDesc::klass_offset_in_bytes());
  3195   // length and pos's are all sign extended at this point on 64bit
  3197   // test for NULL
  3198   if (flags & LIR_OpArrayCopy::src_null_check) {
  3199     __ testptr(src, src);
  3200     __ jcc(Assembler::zero, *stub->entry());
  3202   if (flags & LIR_OpArrayCopy::dst_null_check) {
  3203     __ testptr(dst, dst);
  3204     __ jcc(Assembler::zero, *stub->entry());
  3207   // check if negative
  3208   if (flags & LIR_OpArrayCopy::src_pos_positive_check) {
  3209     __ testl(src_pos, src_pos);
  3210     __ jcc(Assembler::less, *stub->entry());
  3212   if (flags & LIR_OpArrayCopy::dst_pos_positive_check) {
  3213     __ testl(dst_pos, dst_pos);
  3214     __ jcc(Assembler::less, *stub->entry());
  3217   if (flags & LIR_OpArrayCopy::src_range_check) {
  3218     __ lea(tmp, Address(src_pos, length, Address::times_1, 0));
  3219     __ cmpl(tmp, src_length_addr);
  3220     __ jcc(Assembler::above, *stub->entry());
  3222   if (flags & LIR_OpArrayCopy::dst_range_check) {
  3223     __ lea(tmp, Address(dst_pos, length, Address::times_1, 0));
  3224     __ cmpl(tmp, dst_length_addr);
  3225     __ jcc(Assembler::above, *stub->entry());
  3228   if (flags & LIR_OpArrayCopy::length_positive_check) {
  3229     __ testl(length, length);
  3230     __ jcc(Assembler::less, *stub->entry());
  3231     __ jcc(Assembler::zero, *stub->continuation());
  3234 #ifdef _LP64
  3235   __ movl2ptr(src_pos, src_pos); //higher 32bits must be null
  3236   __ movl2ptr(dst_pos, dst_pos); //higher 32bits must be null
  3237 #endif
  3239   if (flags & LIR_OpArrayCopy::type_check) {
  3240     // We don't know the array types are compatible
  3241     if (basic_type != T_OBJECT) {
  3242       // Simple test for basic type arrays
  3243       if (UseCompressedOops) {
  3244         __ movl(tmp, src_klass_addr);
  3245         __ cmpl(tmp, dst_klass_addr);
  3246       } else {
  3247         __ movptr(tmp, src_klass_addr);
  3248         __ cmpptr(tmp, dst_klass_addr);
  3250       __ jcc(Assembler::notEqual, *stub->entry());
  3251     } else {
  3252       // For object arrays, if src is a sub class of dst then we can
  3253       // safely do the copy.
  3254       Label cont, slow;
  3256       __ push(src);
  3257       __ push(dst);
  3259       __ load_klass(src, src);
  3260       __ load_klass(dst, dst);
  3262       __ check_klass_subtype_fast_path(src, dst, tmp, &cont, &slow, NULL);
  3264       __ push(src);
  3265       __ push(dst);
  3266       __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
  3267       __ pop(dst);
  3268       __ pop(src);
  3270       __ cmpl(src, 0);
  3271       __ jcc(Assembler::notEqual, cont);
  3273       __ bind(slow);
  3274       __ pop(dst);
  3275       __ pop(src);
  3277       address copyfunc_addr = StubRoutines::checkcast_arraycopy();
  3278       if (copyfunc_addr != NULL) { // use stub if available
  3279         // src is not a sub class of dst so we have to do a
  3280         // per-element check.
  3282         int mask = LIR_OpArrayCopy::src_objarray|LIR_OpArrayCopy::dst_objarray;
  3283         if ((flags & mask) != mask) {
  3284           // Check that at least both of them object arrays.
  3285           assert(flags & mask, "one of the two should be known to be an object array");
  3287           if (!(flags & LIR_OpArrayCopy::src_objarray)) {
  3288             __ load_klass(tmp, src);
  3289           } else if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
  3290             __ load_klass(tmp, dst);
  3292           int lh_offset = in_bytes(Klass::layout_helper_offset());
  3293           Address klass_lh_addr(tmp, lh_offset);
  3294           jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
  3295           __ cmpl(klass_lh_addr, objArray_lh);
  3296           __ jcc(Assembler::notEqual, *stub->entry());
  3299        // Spill because stubs can use any register they like and it's
  3300        // easier to restore just those that we care about.
  3301        store_parameter(dst, 0);
  3302        store_parameter(dst_pos, 1);
  3303        store_parameter(length, 2);
  3304        store_parameter(src_pos, 3);
  3305        store_parameter(src, 4);
  3307 #ifndef _LP64
  3308         __ movptr(tmp, dst_klass_addr);
  3309         __ movptr(tmp, Address(tmp, objArrayKlass::element_klass_offset()));
  3310         __ push(tmp);
  3311         __ movl(tmp, Address(tmp, Klass::super_check_offset_offset()));
  3312         __ push(tmp);
  3313         __ push(length);
  3314         __ lea(tmp, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
  3315         __ push(tmp);
  3316         __ lea(tmp, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
  3317         __ push(tmp);
  3319         __ call_VM_leaf(copyfunc_addr, 5);
  3320 #else
  3321         __ movl2ptr(length, length); //higher 32bits must be null
  3323         __ lea(c_rarg0, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
  3324         assert_different_registers(c_rarg0, dst, dst_pos, length);
  3325         __ lea(c_rarg1, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
  3326         assert_different_registers(c_rarg1, dst, length);
  3328         __ mov(c_rarg2, length);
  3329         assert_different_registers(c_rarg2, dst);
  3331 #ifdef _WIN64
  3332         // Allocate abi space for args but be sure to keep stack aligned
  3333         __ subptr(rsp, 6*wordSize);
  3334         __ load_klass(c_rarg3, dst);
  3335         __ movptr(c_rarg3, Address(c_rarg3, objArrayKlass::element_klass_offset()));
  3336         store_parameter(c_rarg3, 4);
  3337         __ movl(c_rarg3, Address(c_rarg3, Klass::super_check_offset_offset()));
  3338         __ call(RuntimeAddress(copyfunc_addr));
  3339         __ addptr(rsp, 6*wordSize);
  3340 #else
  3341         __ load_klass(c_rarg4, dst);
  3342         __ movptr(c_rarg4, Address(c_rarg4, objArrayKlass::element_klass_offset()));
  3343         __ movl(c_rarg3, Address(c_rarg4, Klass::super_check_offset_offset()));
  3344         __ call(RuntimeAddress(copyfunc_addr));
  3345 #endif
  3347 #endif
  3349 #ifndef PRODUCT
  3350         if (PrintC1Statistics) {
  3351           Label failed;
  3352           __ testl(rax, rax);
  3353           __ jcc(Assembler::notZero, failed);
  3354           __ incrementl(ExternalAddress((address)&Runtime1::_arraycopy_checkcast_cnt));
  3355           __ bind(failed);
  3357 #endif
  3359         __ testl(rax, rax);
  3360         __ jcc(Assembler::zero, *stub->continuation());
  3362 #ifndef PRODUCT
  3363         if (PrintC1Statistics) {
  3364           __ incrementl(ExternalAddress((address)&Runtime1::_arraycopy_checkcast_attempt_cnt));
  3366 #endif
  3368         __ mov(tmp, rax);
  3370         __ xorl(tmp, -1);
  3372         // Restore previously spilled arguments
  3373         __ movptr   (dst,     Address(rsp, 0*BytesPerWord));
  3374         __ movptr   (dst_pos, Address(rsp, 1*BytesPerWord));
  3375         __ movptr   (length,  Address(rsp, 2*BytesPerWord));
  3376         __ movptr   (src_pos, Address(rsp, 3*BytesPerWord));
  3377         __ movptr   (src,     Address(rsp, 4*BytesPerWord));
  3380         __ subl(length, tmp);
  3381         __ addl(src_pos, tmp);
  3382         __ addl(dst_pos, tmp);
  3385       __ jmp(*stub->entry());
  3387       __ bind(cont);
  3388       __ pop(dst);
  3389       __ pop(src);
  3393 #ifdef ASSERT
  3394   if (basic_type != T_OBJECT || !(flags & LIR_OpArrayCopy::type_check)) {
  3395     // Sanity check the known type with the incoming class.  For the
  3396     // primitive case the types must match exactly with src.klass and
  3397     // dst.klass each exactly matching the default type.  For the
  3398     // object array case, if no type check is needed then either the
  3399     // dst type is exactly the expected type and the src type is a
  3400     // subtype which we can't check or src is the same array as dst
  3401     // but not necessarily exactly of type default_type.
  3402     Label known_ok, halt;
  3403     __ movoop(tmp, default_type->constant_encoding());
  3404 #ifdef _LP64
  3405     if (UseCompressedOops) {
  3406       __ encode_heap_oop(tmp);
  3408 #endif
  3410     if (basic_type != T_OBJECT) {
  3412       if (UseCompressedOops) __ cmpl(tmp, dst_klass_addr);
  3413       else                   __ cmpptr(tmp, dst_klass_addr);
  3414       __ jcc(Assembler::notEqual, halt);
  3415       if (UseCompressedOops) __ cmpl(tmp, src_klass_addr);
  3416       else                   __ cmpptr(tmp, src_klass_addr);
  3417       __ jcc(Assembler::equal, known_ok);
  3418     } else {
  3419       if (UseCompressedOops) __ cmpl(tmp, dst_klass_addr);
  3420       else                   __ cmpptr(tmp, dst_klass_addr);
  3421       __ jcc(Assembler::equal, known_ok);
  3422       __ cmpptr(src, dst);
  3423       __ jcc(Assembler::equal, known_ok);
  3425     __ bind(halt);
  3426     __ stop("incorrect type information in arraycopy");
  3427     __ bind(known_ok);
  3429 #endif
  3431 #ifndef PRODUCT
  3432   if (PrintC1Statistics) {
  3433     __ incrementl(ExternalAddress(Runtime1::arraycopy_count_address(basic_type)));
  3435 #endif
  3437 #ifdef _LP64
  3438   assert_different_registers(c_rarg0, dst, dst_pos, length);
  3439   __ lea(c_rarg0, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
  3440   assert_different_registers(c_rarg1, length);
  3441   __ lea(c_rarg1, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
  3442   __ mov(c_rarg2, length);
  3444 #else
  3445   __ lea(tmp, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
  3446   store_parameter(tmp, 0);
  3447   __ lea(tmp, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
  3448   store_parameter(tmp, 1);
  3449   store_parameter(length, 2);
  3450 #endif // _LP64
  3452   bool disjoint = (flags & LIR_OpArrayCopy::overlapping) == 0;
  3453   bool aligned = (flags & LIR_OpArrayCopy::unaligned) == 0;
  3454   const char *name;
  3455   address entry = StubRoutines::select_arraycopy_function(basic_type, aligned, disjoint, name, false);
  3456   __ call_VM_leaf(entry, 0);
  3458   __ bind(*stub->continuation());
  3462 void LIR_Assembler::emit_lock(LIR_OpLock* op) {
  3463   Register obj = op->obj_opr()->as_register();  // may not be an oop
  3464   Register hdr = op->hdr_opr()->as_register();
  3465   Register lock = op->lock_opr()->as_register();
  3466   if (!UseFastLocking) {
  3467     __ jmp(*op->stub()->entry());
  3468   } else if (op->code() == lir_lock) {
  3469     Register scratch = noreg;
  3470     if (UseBiasedLocking) {
  3471       scratch = op->scratch_opr()->as_register();
  3473     assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
  3474     // add debug info for NullPointerException only if one is possible
  3475     int null_check_offset = __ lock_object(hdr, obj, lock, scratch, *op->stub()->entry());
  3476     if (op->info() != NULL) {
  3477       add_debug_info_for_null_check(null_check_offset, op->info());
  3479     // done
  3480   } else if (op->code() == lir_unlock) {
  3481     assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
  3482     __ unlock_object(hdr, obj, lock, *op->stub()->entry());
  3483   } else {
  3484     Unimplemented();
  3486   __ bind(*op->stub()->continuation());
  3490 void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
  3491   ciMethod* method = op->profiled_method();
  3492   int bci          = op->profiled_bci();
  3494   // Update counter for all call types
  3495   ciMethodData* md = method->method_data_or_null();
  3496   assert(md != NULL, "Sanity");
  3497   ciProfileData* data = md->bci_to_data(bci);
  3498   assert(data->is_CounterData(), "need CounterData for calls");
  3499   assert(op->mdo()->is_single_cpu(),  "mdo must be allocated");
  3500   Register mdo  = op->mdo()->as_register();
  3501   __ movoop(mdo, md->constant_encoding());
  3502   Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
  3503   Bytecodes::Code bc = method->java_code_at_bci(bci);
  3504   // Perform additional virtual call profiling for invokevirtual and
  3505   // invokeinterface bytecodes
  3506   if ((bc == Bytecodes::_invokevirtual || bc == Bytecodes::_invokeinterface) &&
  3507       C1ProfileVirtualCalls) {
  3508     assert(op->recv()->is_single_cpu(), "recv must be allocated");
  3509     Register recv = op->recv()->as_register();
  3510     assert_different_registers(mdo, recv);
  3511     assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
  3512     ciKlass* known_klass = op->known_holder();
  3513     if (C1OptimizeVirtualCallProfiling && known_klass != NULL) {
  3514       // We know the type that will be seen at this call site; we can
  3515       // statically update the methodDataOop rather than needing to do
  3516       // dynamic tests on the receiver type
  3518       // NOTE: we should probably put a lock around this search to
  3519       // avoid collisions by concurrent compilations
  3520       ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
  3521       uint i;
  3522       for (i = 0; i < VirtualCallData::row_limit(); i++) {
  3523         ciKlass* receiver = vc_data->receiver(i);
  3524         if (known_klass->equals(receiver)) {
  3525           Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
  3526           __ addptr(data_addr, DataLayout::counter_increment);
  3527           return;
  3531       // Receiver type not found in profile data; select an empty slot
  3533       // Note that this is less efficient than it should be because it
  3534       // always does a write to the receiver part of the
  3535       // VirtualCallData rather than just the first time
  3536       for (i = 0; i < VirtualCallData::row_limit(); i++) {
  3537         ciKlass* receiver = vc_data->receiver(i);
  3538         if (receiver == NULL) {
  3539           Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)));
  3540           __ movoop(recv_addr, known_klass->constant_encoding());
  3541           Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
  3542           __ addptr(data_addr, DataLayout::counter_increment);
  3543           return;
  3546     } else {
  3547       __ load_klass(recv, recv);
  3548       Label update_done;
  3549       type_profile_helper(mdo, md, data, recv, &update_done);
  3550       // Receiver did not match any saved receiver and there is no empty row for it.
  3551       // Increment total counter to indicate polymorphic case.
  3552       __ addptr(counter_addr, DataLayout::counter_increment);
  3554       __ bind(update_done);
  3556   } else {
  3557     // Static call
  3558     __ addptr(counter_addr, DataLayout::counter_increment);
  3562 void LIR_Assembler::emit_delay(LIR_OpDelay*) {
  3563   Unimplemented();
  3567 void LIR_Assembler::monitor_address(int monitor_no, LIR_Opr dst) {
  3568   __ lea(dst->as_register(), frame_map()->address_for_monitor_lock(monitor_no));
  3572 void LIR_Assembler::align_backward_branch_target() {
  3573   __ align(BytesPerWord);
  3577 void LIR_Assembler::negate(LIR_Opr left, LIR_Opr dest) {
  3578   if (left->is_single_cpu()) {
  3579     __ negl(left->as_register());
  3580     move_regs(left->as_register(), dest->as_register());
  3582   } else if (left->is_double_cpu()) {
  3583     Register lo = left->as_register_lo();
  3584 #ifdef _LP64
  3585     Register dst = dest->as_register_lo();
  3586     __ movptr(dst, lo);
  3587     __ negptr(dst);
  3588 #else
  3589     Register hi = left->as_register_hi();
  3590     __ lneg(hi, lo);
  3591     if (dest->as_register_lo() == hi) {
  3592       assert(dest->as_register_hi() != lo, "destroying register");
  3593       move_regs(hi, dest->as_register_hi());
  3594       move_regs(lo, dest->as_register_lo());
  3595     } else {
  3596       move_regs(lo, dest->as_register_lo());
  3597       move_regs(hi, dest->as_register_hi());
  3599 #endif // _LP64
  3601   } else if (dest->is_single_xmm()) {
  3602     if (left->as_xmm_float_reg() != dest->as_xmm_float_reg()) {
  3603       __ movflt(dest->as_xmm_float_reg(), left->as_xmm_float_reg());
  3605     __ xorps(dest->as_xmm_float_reg(),
  3606              ExternalAddress((address)float_signflip_pool));
  3608   } else if (dest->is_double_xmm()) {
  3609     if (left->as_xmm_double_reg() != dest->as_xmm_double_reg()) {
  3610       __ movdbl(dest->as_xmm_double_reg(), left->as_xmm_double_reg());
  3612     __ xorpd(dest->as_xmm_double_reg(),
  3613              ExternalAddress((address)double_signflip_pool));
  3615   } else if (left->is_single_fpu() || left->is_double_fpu()) {
  3616     assert(left->fpu() == 0, "arg must be on TOS");
  3617     assert(dest->fpu() == 0, "dest must be TOS");
  3618     __ fchs();
  3620   } else {
  3621     ShouldNotReachHere();
  3626 void LIR_Assembler::leal(LIR_Opr addr, LIR_Opr dest) {
  3627   assert(addr->is_address() && dest->is_register(), "check");
  3628   Register reg;
  3629   reg = dest->as_pointer_register();
  3630   __ lea(reg, as_Address(addr->as_address_ptr()));
  3635 void LIR_Assembler::rt_call(LIR_Opr result, address dest, const LIR_OprList* args, LIR_Opr tmp, CodeEmitInfo* info) {
  3636   assert(!tmp->is_valid(), "don't need temporary");
  3637   __ call(RuntimeAddress(dest));
  3638   if (info != NULL) {
  3639     add_call_info_here(info);
  3644 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {
  3645   assert(type == T_LONG, "only for volatile long fields");
  3647   if (info != NULL) {
  3648     add_debug_info_for_null_check_here(info);
  3651   if (src->is_double_xmm()) {
  3652     if (dest->is_double_cpu()) {
  3653 #ifdef _LP64
  3654       __ movdq(dest->as_register_lo(), src->as_xmm_double_reg());
  3655 #else
  3656       __ movdl(dest->as_register_lo(), src->as_xmm_double_reg());
  3657       __ psrlq(src->as_xmm_double_reg(), 32);
  3658       __ movdl(dest->as_register_hi(), src->as_xmm_double_reg());
  3659 #endif // _LP64
  3660     } else if (dest->is_double_stack()) {
  3661       __ movdbl(frame_map()->address_for_slot(dest->double_stack_ix()), src->as_xmm_double_reg());
  3662     } else if (dest->is_address()) {
  3663       __ movdbl(as_Address(dest->as_address_ptr()), src->as_xmm_double_reg());
  3664     } else {
  3665       ShouldNotReachHere();
  3668   } else if (dest->is_double_xmm()) {
  3669     if (src->is_double_stack()) {
  3670       __ movdbl(dest->as_xmm_double_reg(), frame_map()->address_for_slot(src->double_stack_ix()));
  3671     } else if (src->is_address()) {
  3672       __ movdbl(dest->as_xmm_double_reg(), as_Address(src->as_address_ptr()));
  3673     } else {
  3674       ShouldNotReachHere();
  3677   } else if (src->is_double_fpu()) {
  3678     assert(src->fpu_regnrLo() == 0, "must be TOS");
  3679     if (dest->is_double_stack()) {
  3680       __ fistp_d(frame_map()->address_for_slot(dest->double_stack_ix()));
  3681     } else if (dest->is_address()) {
  3682       __ fistp_d(as_Address(dest->as_address_ptr()));
  3683     } else {
  3684       ShouldNotReachHere();
  3687   } else if (dest->is_double_fpu()) {
  3688     assert(dest->fpu_regnrLo() == 0, "must be TOS");
  3689     if (src->is_double_stack()) {
  3690       __ fild_d(frame_map()->address_for_slot(src->double_stack_ix()));
  3691     } else if (src->is_address()) {
  3692       __ fild_d(as_Address(src->as_address_ptr()));
  3693     } else {
  3694       ShouldNotReachHere();
  3696   } else {
  3697     ShouldNotReachHere();
  3702 void LIR_Assembler::membar() {
  3703   // QQQ sparc TSO uses this,
  3704   __ membar( Assembler::Membar_mask_bits(Assembler::StoreLoad));
  3707 void LIR_Assembler::membar_acquire() {
  3708   // No x86 machines currently require load fences
  3709   // __ load_fence();
  3712 void LIR_Assembler::membar_release() {
  3713   // No x86 machines currently require store fences
  3714   // __ store_fence();
  3717 void LIR_Assembler::get_thread(LIR_Opr result_reg) {
  3718   assert(result_reg->is_register(), "check");
  3719 #ifdef _LP64
  3720   // __ get_thread(result_reg->as_register_lo());
  3721   __ mov(result_reg->as_register(), r15_thread);
  3722 #else
  3723   __ get_thread(result_reg->as_register());
  3724 #endif // _LP64
  3728 void LIR_Assembler::peephole(LIR_List*) {
  3729   // do nothing for now
  3733 #undef __

mercurial