hotspot/src/cpu/sparc/vm/cppInterpreter_sparc.cpp

Print this page
rev 611 : Merge
   1 #ifdef USE_PRAGMA_IDENT_HDR
   2 #pragma ident "@(#)cppInterpreter_sparc.cpp     1.1 07/08/29 13:42:16 JVM"
   3 #endif
   4 /*
   5  * Copyright 2007 Sun Microsystems, Inc.  All Rights Reserved.
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This code is free software; you can redistribute it and/or modify it
   9  * under the terms of the GNU General Public License version 2 only, as
  10  * published by the Free Software Foundation.
  11  *
  12  * This code is distributed in the hope that it will be useful, but WITHOUT
  13  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  14  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  15  * version 2 for more details (a copy is included in the LICENSE file that
  16  * accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License version
  19  * 2 along with this work; if not, write to the Free Software Foundation,
  20  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *  


 145       __ srl(O0, 16, O0);   
 146       __ st(O0, L1_scratch, 0); 
 147       __ sub(L1_scratch, wordSize, L1_scratch);
 148       break; 
 149 
 150     case T_BYTE   : 
 151       __ sll(O0, 24, O0); 
 152       __ sra(O0, 24, O0);   
 153       __ st(O0, L1_scratch, 0); 
 154       __ sub(L1_scratch, wordSize, L1_scratch);
 155       break;
 156 
 157     case T_SHORT  : 
 158       __ sll(O0, 16, O0); 
 159       __ sra(O0, 16, O0);   
 160       __ st(O0, L1_scratch, 0); 
 161       __ sub(L1_scratch, wordSize, L1_scratch);
 162       break;
 163     case T_LONG   : 
 164 #ifndef _LP64
 165 #if !defined(_LP64) && defined(COMPILER2) 
 166   // All return values are where we want them, except for Longs.  C2 returns
 167   // longs in G1 in the 32-bit build whereas the interpreter wants them in O0/O1.
 168   // Since the interpreter will return longs in G1 and O0/O1 in the 32bit
 169   // build even if we are returning from interpreted we just do a little
 170   // stupid shuffing. 
 171   // Note: I tried to make c2 return longs in O0/O1 and G1 so we wouldn't have to
 172   // do this here. Unfortunately if we did a rethrow we'd see an machepilog node
 173   // first which would move g1 -> O0/O1 and destroy the exception we were throwing.
 174       __ stx(G1, L1_scratch, -wordSize);
 175 #else
 176       // native result is in O0, O1
 177       __ st(O1, L1_scratch, 0);                      // Low order
 178       __ st(O0, L1_scratch, -wordSize);              // High order
 179 #endif /* !_LP64 && COMPILER2 */
 180 #else
 181       __ stx(O0, L1_scratch, 0);
 182 __ breakpoint_trap();
 183 #endif
 184       __ sub(L1_scratch, 2*wordSize, L1_scratch);
 185       break;
 186 
 187     case T_INT    : 
 188       __ st(O0, L1_scratch, 0); 
 189       __ sub(L1_scratch, wordSize, L1_scratch);
 190       break;
 191 
 192     case T_VOID   : /* nothing to do */
 193       break;
 194 
 195     case T_FLOAT  : 
 196       __ stf(FloatRegisterImpl::S, F0, L1_scratch, 0);      
 197       __ sub(L1_scratch, wordSize, L1_scratch);
 198       break;
 199 
 200     case T_DOUBLE : 
 201       // Every stack slot is aligned on 64 bit, However is this
 202       // the correct stack slot on 64bit?? QQQ


 223   // The current interpreter activation in Lstate is for the method just returning its
 224   // result. So we know that the result of this method is on the top of the current
 225   // execution stack (which is pre-pushed) and will be return to the top of the caller
 226   // stack. The top of the callers stack is the bottom of the locals of the current
 227   // activation.
 228   // Because of the way activation are managed by the frame manager the value of esp is
 229   // below both the stack top of the current activation and naturally the stack top
 230   // of the calling activation. This enable this routine to leave the return address
 231   // to the frame manager on the stack and do a vanilla return.
 232   //
 233   // On entry: O0 - points to source (callee stack top)
 234   //           O1 - points to destination (caller stack top [i.e. free location])
 235   // destroys O2, O3
 236   //
 237 
 238   address entry = __ pc();
 239   switch (type) {
 240     case T_VOID:  break;  
 241       break;
 242     case T_FLOAT  :
 243       __ breakpoint_trap(Assembler::zero);
 244     case T_BOOLEAN: 
 245     case T_CHAR   : 
 246     case T_BYTE   : 
 247     case T_SHORT  : 
 248     case T_INT    : 
 249       // 1 word result
 250       __ ld(O0, 0, O2);
 251       __ st(O2, O1, 0);
 252       __ sub(O1, wordSize, O1);
 253       break;
 254     case T_DOUBLE  :
 255     case T_LONG    : 
 256       // return top two words on current expression stack to caller's expression stack
 257       // The caller's expression stack is adjacent to the current frame manager's intepretState
 258       // except we allocated one extra word for this intepretState so we won't overwrite it
 259       // when we return a two word result.
 260 #ifdef _LP64
 261 __ breakpoint_trap();
 262       // Hmm now that longs are in one entry should "_ptr" really be "x"?
 263       __ ld_ptr(O0, 0, O2);
 264       __ ld_ptr(O0, wordSize, O3);
 265       __ st_ptr(O3, O1, 0);
 266       __ st_ptr(O2, O1, -wordSize);
 267 #else
 268       __ ld(O0, 0, O2);
 269       __ ld(O0, wordSize, O3);
 270       __ st(O3, O1, 0);
 271       __ st(O2, O1, -wordSize);
 272 #endif
 273       __ sub(O1, 2*wordSize, O1);
 274       break;
 275     case T_OBJECT :
 276       __ ld_ptr(O0, 0, O2);
 277       __ verify_oop(O2);                                               // verify it
 278       __ st_ptr(O2, O1, 0);
 279       __ sub(O1, wordSize, O1);
 280       break;
 281     default       : ShouldNotReachHere();
 282   }
 283   __ retl();
 284   __ delayed()->nop(); // QQ schedule this better
 285   return entry;


 305     case T_FLOAT  :
 306       __ ldf(FloatRegisterImpl::S, O0, 0, F0);
 307       break;
 308     case T_BOOLEAN: 
 309     case T_CHAR   : 
 310     case T_BYTE   : 
 311     case T_SHORT  : 
 312     case T_INT    : 
 313       // 1 word result
 314       __ ld(O0, 0, O0->after_save());
 315       break;
 316     case T_DOUBLE  :
 317       __ ldf(FloatRegisterImpl::D, O0, 0, F0);
 318       break;
 319     case T_LONG    : 
 320       // return top two words on current expression stack to caller's expression stack
 321       // The caller's expression stack is adjacent to the current frame manager's interpretState
 322       // except we allocated one extra word for this intepretState so we won't overwrite it
 323       // when we return a two word result.
 324 #ifdef _LP64
 325 __ breakpoint_trap();
 326       // Hmm now that longs are in one entry should "_ptr" really be "x"?
 327       __ ld_ptr(O0, 0, O0->after_save());
 328       __ ld_ptr(O0, wordSize, O1->after_save());
 329 #else
 330       __ ld(O0, wordSize, O1->after_save());
 331       __ ld(O0, 0, O0->after_save());
 332 #endif
 333 #if defined(COMPILER2) && !defined(_LP64)
 334       // C2 expects long results in G1 we can't tell if we're returning to interpreted
 335       // or compiled so just be safe use G1 and O0/O1
 336     
 337       // Shift bits into high (msb) of G1
 338       __ sllx(Otos_l1->after_save(), 32, G1);
 339       // Zero extend low bits
 340       __ srl (Otos_l2->after_save(), 0, Otos_l2->after_save());
 341       __ or3 (Otos_l2->after_save(), G1, G1);           
 342 #endif /* COMPILER2 */
 343       break;
 344     case T_OBJECT :
 345       __ ld_ptr(O0, 0, O0->after_save());
 346       __ verify_oop(O0->after_save());                                               // verify it
 347       break;
 348     default       : ShouldNotReachHere();


1359   __ sub(L2_scratch, entry_size, L2_scratch);
1360   __ st_ptr(L2_scratch, STATE(_stack_limit));
1361 
1362   __ ld_ptr(STATE(_stack), L1_scratch);                // Get current stack top
1363   __ sub(L1_scratch, entry_size, L1_scratch);
1364   __ st_ptr(L1_scratch, STATE(_stack));
1365   __ ba(false, entry);
1366   __ delayed()->add(L1_scratch, wordSize, L1_scratch);        // first real entry (undo prepush)
1367  
1368   // 2. move expression stack
1369 
1370   __ bind(loop);
1371   __ st_ptr(L3_scratch, Address(L1_scratch, 0));
1372   __ add(L1_scratch, wordSize, L1_scratch);
1373   __ bind(entry);
1374   __ cmp(L1_scratch, L4_scratch);
1375   __ br(Assembler::notEqual, false, Assembler::pt, loop);
1376   __ delayed()->ld_ptr(L1_scratch, entry_size, L3_scratch);
1377 
1378   // now zero the slot so we can find it.
1379   __ st(G0, L4_scratch, BasicObjectLock::obj_offset_in_bytes());
1380 
1381 }
1382 
1383 // Initial entry to C++ interpreter from the call_stub.
1384 // This entry point is called the frame manager since it handles the generation
1385 // of interpreter activation frames via requests directly from the vm (via call_stub)
1386 // and via requests from the interpreter. The requests from the call_stub happen
1387 // directly thru the entry point. Requests from the interpreter happen via returning
1388 // from the interpreter and examining the message the interpreter has returned to
1389 // the frame manager. The frame manager can take the following requests:
1390 
1391 // NO_REQUEST - error, should never happen.
1392 // MORE_MONITORS - need a new monitor. Shuffle the expression stack on down and
1393 //                 allocate a new monitor.
1394 // CALL_METHOD - setup a new activation to call a new method. Very similar to what
1395 //               happens during entry during the entry via the call stub.
1396 // RETURN_FROM_METHOD - remove an activation. Return to interpreter or call stub.
1397 //
1398 // Arguments:
1399 //


1699   // stack is in the state that the  calling convention left it.
1700   // Copy the result from native abi result and place it on java expression stack.
1701 
1702   // Current interpreter state is present in Lstate
1703 
1704   // Exception pending?
1705 
1706   __ ld_ptr(STATE(_frame_bottom), SP);                             // restore to full stack frame
1707   __ ld_ptr(exception_addr, Lscratch);                                         // get any pending exception
1708   __ tst(Lscratch);                                                            // exception pending?
1709   __ brx(Assembler::notZero, false, Assembler::pt, return_with_exception);
1710   __ delayed()->nop();
1711   
1712   // Process the native abi result to java expression stack
1713 
1714   __ ld_ptr(STATE(_result._to_call._callee), L4_scratch);                        // called method
1715   __ ld_ptr(STATE(_stack), L1_scratch);                                          // get top of java expr stack
1716   __ lduh(L4_scratch, in_bytes(methodOopDesc::size_of_parameters_offset()), L2_scratch); // get parameter size
1717   __ sll(L2_scratch, LogBytesPerWord, L2_scratch     );                           // parameter size in bytes
1718   __ add(L1_scratch, L2_scratch, L1_scratch);                                      // stack destination for result
1719   __ ld_ptr(L4_scratch, in_bytes(methodOopDesc::result_index_offset()), L3_scratch); // called method result type index
1720 
1721   // tosca is really just native abi
1722   __ set((intptr_t)CppInterpreter::_tosca_to_stack, L4_scratch);
1723   __ sll(L3_scratch, LogBytesPerWord, L3_scratch);
1724   __ ld_ptr(L4_scratch, L3_scratch, Lscratch);                                       // get typed result converter address
1725   __ jmpl(Lscratch, G0, O7);                                                   // and convert it
1726   __ delayed()->nop();
1727 
1728   // L1_scratch points to top of stack (prepushed)
1729 
1730   __ ba(false, resume_interpreter);
1731   __ delayed()->mov(L1_scratch, O1);
1732 
1733   // An exception is being caught on return to a vanilla interpreter frame.
1734   // Empty the stack and resume interpreter
1735 
1736   __ bind(return_with_exception);
1737 
1738   __ ld_ptr(STATE(_frame_bottom), SP);                             // restore to full stack frame
1739   __ ld_ptr(STATE(_stack_base), O1);                               // empty java expression stack


1743   // Return from interpreted method we return result appropriate to the caller (i.e. "recursive"
1744   // interpreter call, or native) and unwind this interpreter activation.
1745   // All monitors should be unlocked.
1746 
1747   __ bind(return_from_interpreted_method); 
1748 
1749   VALIDATE_STATE(G3_scratch, 7);
1750 
1751   Label return_to_initial_caller;
1752 
1753   // Interpreted result is on the top of the completed activation expression stack.
1754   // We must return it to the top of the callers stack if caller was interpreted
1755   // otherwise we convert to native abi result and return to call_stub/c1/c2 
1756   // The caller's expression stack was truncated by the call however the current activation
1757   // has enough stuff on the stack that we have usable space there no matter what. The
1758   // other thing that makes it easy is that the top of the caller's stack is stored in STATE(_locals)
1759   // for the current activation
1760 
1761   __ ld_ptr(STATE(_prev_link), L1_scratch);
1762   __ ld_ptr(STATE(_method), L2_scratch);                               // get method just executed
1763   __ ld_ptr(L2_scratch, in_bytes(methodOopDesc::result_index_offset()), L2_scratch);
1764   __ tst(L1_scratch);
1765   __ brx(Assembler::zero, false, Assembler::pt, return_to_initial_caller);
1766   __ delayed()->sll(L2_scratch, LogBytesPerWord, L2_scratch);
1767 
1768   // Copy result to callers java stack
1769 
1770   __ set((intptr_t)CppInterpreter::_stack_to_stack, L4_scratch);
1771   __ ld_ptr(L4_scratch, L2_scratch, Lscratch);                          // get typed result converter address
1772   __ ld_ptr(STATE(_stack), O0);                                       // current top (prepushed)
1773   __ ld_ptr(STATE(_locals), O1);                                      // stack destination
1774 
1775   // O0 - will be source, O1 - will be destination (preserved)
1776   __ jmpl(Lscratch, G0, O7);                                          // and convert it
1777   __ delayed()->add(O0, wordSize, O0);                                // get source (top of current expr stack)
1778 
1779   // O1 == &locals[0]
1780 
1781   // Result is now on caller's stack. Just unwind current activation and resume
1782 
1783   Label unwind_recursive_activation;


1909 
1910   // stack points to next free location and not top element on expression stack
1911   // method expects sp to be pointing to topmost element
1912 
1913   __ ld_ptr(STATE(_thread), G2_thread);
1914   __ ld_ptr(STATE(_result._to_call._callee), G5_method);
1915 
1916 
1917   // SP already takes in to account the 2 extra words we use for slop
1918   // when we call a "static long no_params()" method. So if
1919   // we trim back sp by the amount of unused java expression stack
1920   // there will be automagically the 2 extra words we need.
1921   // We also have to worry about keeping SP aligned.
1922 
1923   __ ld_ptr(STATE(_stack), Gargs);
1924   __ ld_ptr(STATE(_stack_limit), L1_scratch);
1925 
1926   // compute the unused java stack size
1927   __ sub(Gargs, L1_scratch, L2_scratch);                       // compute unused space
1928 
1929   // Round down the unused space to that stack is always aligned
1930   // by making the unused space a multiple of the size of a long.
1931 
1932   __ and3(L2_scratch, -BytesPerLong, L2_scratch);
1933 
1934   // Now trim the stack
1935   __ add(SP, L2_scratch, SP);
1936 
1937 
1938   // Now point to the final argument (account for prepush)
1939   __ add(Gargs, wordSize, Gargs);                         
1940 #ifdef ASSERT
1941   // Make sure we have space for the window
1942   __ sub(Gargs, SP, L1_scratch);
1943   __ cmp(L1_scratch, 16*wordSize);
1944   {
1945     Label skip;
1946     __ brx(Assembler::greaterEqual, false, Assembler::pt, skip);
1947     __ delayed()->nop();
1948     __ stop("killed stack");
1949     __ bind(skip);
1950   }
1951 #endif // ASSERT
1952 


2162   // calculates the extra locals based on itself. Not what the callee does
2163   // to it. So it ignores last_frame_adjust value. Seems suspicious as far
2164   // as getting sender_sp correct.
2165 
2166   int extra_locals_size = callee_locals_size - callee_param_size;
2167   int monitor_size = (sizeof(BasicObjectLock) * moncount) / wordSize;
2168   int full_frame_words = size_activation_helper(extra_locals_size, method->max_stack(), monitor_size);
2169   int short_frame_words = size_activation_helper(extra_locals_size, method->max_stack(), monitor_size);
2170   int frame_words = is_top_frame ? full_frame_words : short_frame_words;
2171                                                 
2172                                                 
2173   /*
2174     if we actually have a frame to layout we must now fill in all the pieces. This means both
2175     the interpreterState and the registers.
2176   */
2177   if (interpreter_frame != NULL) {
2178 
2179     // MUCHO HACK
2180 
2181     intptr_t* frame_bottom = interpreter_frame->sp() - (full_frame_words - frame_words);



2182 
2183     /* Now fillin the interpreterState object */
2184 
2185     interpreterState cur_state = (interpreterState) ((intptr_t)interpreter_frame->fp() -  sizeof(BytecodeInterpreter));
2186 
2187 
2188     intptr_t* locals;
2189 
2190     // Calculate the postion of locals[0]. This is painful because of
2191     // stack alignment (same as ia64). The problem is that we can
2192     // not compute the location of locals from fp(). fp() will account
2193     // for the extra locals but it also accounts for aligning the stack
2194     // and we can't determine if the locals[0] was misaligned but max_locals
2195     // was enough to have the 
2196     // calculate postion of locals. fp already accounts for extra locals.
2197     // +2 for the static long no_params() issue.
2198 
2199     if (caller->is_interpreted_frame()) {
2200       // locals must agree with the caller because it will be used to set the 
2201       // caller's tos when we return. 


2227                              
2228 
2229     BytecodeInterpreter::layout_interpreterState(cur_state, 
2230                                           caller,
2231                                           interpreter_frame,
2232                                           method, 
2233                                           locals, 
2234                                           stack, 
2235                                           stack_base, 
2236                                           monitor_base, 
2237                                           frame_bottom,
2238                                           is_top_frame);
2239 
2240     BytecodeInterpreter::pd_layout_interpreterState(cur_state, interpreter_return_address, interpreter_frame->fp());
2241 
2242   }
2243   return frame_words;
2244 }
2245 
2246 #endif // CC_INTERP 
2247 



   1 /*
   2  * Copyright 2007-2008 Sun Microsystems, Inc.  All Rights Reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  20  * CA 95054 USA or visit www.sun.com if you need additional information or
  21  * have any questions.
  22  *


 142       __ srl(O0, 16, O0);
 143       __ st(O0, L1_scratch, 0);
 144       __ sub(L1_scratch, wordSize, L1_scratch);
 145       break;
 146 
 147     case T_BYTE   :
 148       __ sll(O0, 24, O0);
 149       __ sra(O0, 24, O0);
 150       __ st(O0, L1_scratch, 0);
 151       __ sub(L1_scratch, wordSize, L1_scratch);
 152       break;
 153 
 154     case T_SHORT  :
 155       __ sll(O0, 16, O0);
 156       __ sra(O0, 16, O0);
 157       __ st(O0, L1_scratch, 0);
 158       __ sub(L1_scratch, wordSize, L1_scratch);
 159       break;
 160     case T_LONG   :
 161 #ifndef _LP64
 162 #if defined(COMPILER2)
 163   // All return values are where we want them, except for Longs.  C2 returns
 164   // longs in G1 in the 32-bit build whereas the interpreter wants them in O0/O1.
 165   // Since the interpreter will return longs in G1 and O0/O1 in the 32bit
 166   // build even if we are returning from interpreted we just do a little
 167   // stupid shuffing.
 168   // Note: I tried to make c2 return longs in O0/O1 and G1 so we wouldn't have to
 169   // do this here. Unfortunately if we did a rethrow we'd see an machepilog node
 170   // first which would move g1 -> O0/O1 and destroy the exception we were throwing.
 171       __ stx(G1, L1_scratch, -wordSize);
 172 #else
 173       // native result is in O0, O1
 174       __ st(O1, L1_scratch, 0);                      // Low order
 175       __ st(O0, L1_scratch, -wordSize);              // High order
 176 #endif /* COMPILER2 */
 177 #else
 178       __ stx(O0, L1_scratch, -wordSize);

 179 #endif
 180       __ sub(L1_scratch, 2*wordSize, L1_scratch);
 181       break;
 182 
 183     case T_INT    :
 184       __ st(O0, L1_scratch, 0);
 185       __ sub(L1_scratch, wordSize, L1_scratch);
 186       break;
 187 
 188     case T_VOID   : /* nothing to do */
 189       break;
 190 
 191     case T_FLOAT  :
 192       __ stf(FloatRegisterImpl::S, F0, L1_scratch, 0);
 193       __ sub(L1_scratch, wordSize, L1_scratch);
 194       break;
 195 
 196     case T_DOUBLE :
 197       // Every stack slot is aligned on 64 bit, However is this
 198       // the correct stack slot on 64bit?? QQQ


 219   // The current interpreter activation in Lstate is for the method just returning its
 220   // result. So we know that the result of this method is on the top of the current
 221   // execution stack (which is pre-pushed) and will be return to the top of the caller
 222   // stack. The top of the callers stack is the bottom of the locals of the current
 223   // activation.
 224   // Because of the way activation are managed by the frame manager the value of esp is
 225   // below both the stack top of the current activation and naturally the stack top
 226   // of the calling activation. This enable this routine to leave the return address
 227   // to the frame manager on the stack and do a vanilla return.
 228   //
 229   // On entry: O0 - points to source (callee stack top)
 230   //           O1 - points to destination (caller stack top [i.e. free location])
 231   // destroys O2, O3
 232   //
 233 
 234   address entry = __ pc();
 235   switch (type) {
 236     case T_VOID:  break;
 237       break;
 238     case T_FLOAT  :

 239     case T_BOOLEAN:
 240     case T_CHAR   :
 241     case T_BYTE   :
 242     case T_SHORT  :
 243     case T_INT    :
 244       // 1 word result
 245       __ ld(O0, 0, O2);
 246       __ st(O2, O1, 0);
 247       __ sub(O1, wordSize, O1);
 248       break;
 249     case T_DOUBLE  :
 250     case T_LONG    :
 251       // return top two words on current expression stack to caller's expression stack
 252       // The caller's expression stack is adjacent to the current frame manager's intepretState
 253       // except we allocated one extra word for this intepretState so we won't overwrite it
 254       // when we return a two word result.
 255 #ifdef _LP64


 256       __ ld_ptr(O0, 0, O2);


 257       __ st_ptr(O2, O1, -wordSize);
 258 #else
 259       __ ld(O0, 0, O2);
 260       __ ld(O0, wordSize, O3);
 261       __ st(O3, O1, 0);
 262       __ st(O2, O1, -wordSize);
 263 #endif
 264       __ sub(O1, 2*wordSize, O1);
 265       break;
 266     case T_OBJECT :
 267       __ ld_ptr(O0, 0, O2);
 268       __ verify_oop(O2);                                               // verify it
 269       __ st_ptr(O2, O1, 0);
 270       __ sub(O1, wordSize, O1);
 271       break;
 272     default       : ShouldNotReachHere();
 273   }
 274   __ retl();
 275   __ delayed()->nop(); // QQ schedule this better
 276   return entry;


 296     case T_FLOAT  :
 297       __ ldf(FloatRegisterImpl::S, O0, 0, F0);
 298       break;
 299     case T_BOOLEAN:
 300     case T_CHAR   :
 301     case T_BYTE   :
 302     case T_SHORT  :
 303     case T_INT    :
 304       // 1 word result
 305       __ ld(O0, 0, O0->after_save());
 306       break;
 307     case T_DOUBLE  :
 308       __ ldf(FloatRegisterImpl::D, O0, 0, F0);
 309       break;
 310     case T_LONG    :
 311       // return top two words on current expression stack to caller's expression stack
 312       // The caller's expression stack is adjacent to the current frame manager's interpretState
 313       // except we allocated one extra word for this intepretState so we won't overwrite it
 314       // when we return a two word result.
 315 #ifdef _LP64


 316       __ ld_ptr(O0, 0, O0->after_save());

 317 #else
 318       __ ld(O0, wordSize, O1->after_save());
 319       __ ld(O0, 0, O0->after_save());
 320 #endif
 321 #if defined(COMPILER2) && !defined(_LP64)
 322       // C2 expects long results in G1 we can't tell if we're returning to interpreted
 323       // or compiled so just be safe use G1 and O0/O1
 324 
 325       // Shift bits into high (msb) of G1
 326       __ sllx(Otos_l1->after_save(), 32, G1);
 327       // Zero extend low bits
 328       __ srl (Otos_l2->after_save(), 0, Otos_l2->after_save());
 329       __ or3 (Otos_l2->after_save(), G1, G1);
 330 #endif /* COMPILER2 */
 331       break;
 332     case T_OBJECT :
 333       __ ld_ptr(O0, 0, O0->after_save());
 334       __ verify_oop(O0->after_save());                                               // verify it
 335       break;
 336     default       : ShouldNotReachHere();


1347   __ sub(L2_scratch, entry_size, L2_scratch);
1348   __ st_ptr(L2_scratch, STATE(_stack_limit));
1349 
1350   __ ld_ptr(STATE(_stack), L1_scratch);                // Get current stack top
1351   __ sub(L1_scratch, entry_size, L1_scratch);
1352   __ st_ptr(L1_scratch, STATE(_stack));
1353   __ ba(false, entry);
1354   __ delayed()->add(L1_scratch, wordSize, L1_scratch);        // first real entry (undo prepush)
1355 
1356   // 2. move expression stack
1357 
1358   __ bind(loop);
1359   __ st_ptr(L3_scratch, Address(L1_scratch, 0));
1360   __ add(L1_scratch, wordSize, L1_scratch);
1361   __ bind(entry);
1362   __ cmp(L1_scratch, L4_scratch);
1363   __ br(Assembler::notEqual, false, Assembler::pt, loop);
1364   __ delayed()->ld_ptr(L1_scratch, entry_size, L3_scratch);
1365 
1366   // now zero the slot so we can find it.
1367   __ st_ptr(G0, L4_scratch, BasicObjectLock::obj_offset_in_bytes());
1368 
1369 }
1370 
1371 // Initial entry to C++ interpreter from the call_stub.
1372 // This entry point is called the frame manager since it handles the generation
1373 // of interpreter activation frames via requests directly from the vm (via call_stub)
1374 // and via requests from the interpreter. The requests from the call_stub happen
1375 // directly thru the entry point. Requests from the interpreter happen via returning
1376 // from the interpreter and examining the message the interpreter has returned to
1377 // the frame manager. The frame manager can take the following requests:
1378 
1379 // NO_REQUEST - error, should never happen.
1380 // MORE_MONITORS - need a new monitor. Shuffle the expression stack on down and
1381 //                 allocate a new monitor.
1382 // CALL_METHOD - setup a new activation to call a new method. Very similar to what
1383 //               happens during entry during the entry via the call stub.
1384 // RETURN_FROM_METHOD - remove an activation. Return to interpreter or call stub.
1385 //
1386 // Arguments:
1387 //


1687   // stack is in the state that the  calling convention left it.
1688   // Copy the result from native abi result and place it on java expression stack.
1689 
1690   // Current interpreter state is present in Lstate
1691 
1692   // Exception pending?
1693 
1694   __ ld_ptr(STATE(_frame_bottom), SP);                             // restore to full stack frame
1695   __ ld_ptr(exception_addr, Lscratch);                                         // get any pending exception
1696   __ tst(Lscratch);                                                            // exception pending?
1697   __ brx(Assembler::notZero, false, Assembler::pt, return_with_exception);
1698   __ delayed()->nop();
1699 
1700   // Process the native abi result to java expression stack
1701 
1702   __ ld_ptr(STATE(_result._to_call._callee), L4_scratch);                        // called method
1703   __ ld_ptr(STATE(_stack), L1_scratch);                                          // get top of java expr stack
1704   __ lduh(L4_scratch, in_bytes(methodOopDesc::size_of_parameters_offset()), L2_scratch); // get parameter size
1705   __ sll(L2_scratch, LogBytesPerWord, L2_scratch     );                           // parameter size in bytes
1706   __ add(L1_scratch, L2_scratch, L1_scratch);                                      // stack destination for result
1707   __ ld(L4_scratch, in_bytes(methodOopDesc::result_index_offset()), L3_scratch); // called method result type index
1708 
1709   // tosca is really just native abi
1710   __ set((intptr_t)CppInterpreter::_tosca_to_stack, L4_scratch);
1711   __ sll(L3_scratch, LogBytesPerWord, L3_scratch);
1712   __ ld_ptr(L4_scratch, L3_scratch, Lscratch);                                       // get typed result converter address
1713   __ jmpl(Lscratch, G0, O7);                                                   // and convert it
1714   __ delayed()->nop();
1715 
1716   // L1_scratch points to top of stack (prepushed)
1717 
1718   __ ba(false, resume_interpreter);
1719   __ delayed()->mov(L1_scratch, O1);
1720 
1721   // An exception is being caught on return to a vanilla interpreter frame.
1722   // Empty the stack and resume interpreter
1723 
1724   __ bind(return_with_exception);
1725 
1726   __ ld_ptr(STATE(_frame_bottom), SP);                             // restore to full stack frame
1727   __ ld_ptr(STATE(_stack_base), O1);                               // empty java expression stack


1731   // Return from interpreted method we return result appropriate to the caller (i.e. "recursive"
1732   // interpreter call, or native) and unwind this interpreter activation.
1733   // All monitors should be unlocked.
1734 
1735   __ bind(return_from_interpreted_method);
1736 
1737   VALIDATE_STATE(G3_scratch, 7);
1738 
1739   Label return_to_initial_caller;
1740 
1741   // Interpreted result is on the top of the completed activation expression stack.
1742   // We must return it to the top of the callers stack if caller was interpreted
1743   // otherwise we convert to native abi result and return to call_stub/c1/c2
1744   // The caller's expression stack was truncated by the call however the current activation
1745   // has enough stuff on the stack that we have usable space there no matter what. The
1746   // other thing that makes it easy is that the top of the caller's stack is stored in STATE(_locals)
1747   // for the current activation
1748 
1749   __ ld_ptr(STATE(_prev_link), L1_scratch);
1750   __ ld_ptr(STATE(_method), L2_scratch);                               // get method just executed
1751   __ ld(L2_scratch, in_bytes(methodOopDesc::result_index_offset()), L2_scratch);
1752   __ tst(L1_scratch);
1753   __ brx(Assembler::zero, false, Assembler::pt, return_to_initial_caller);
1754   __ delayed()->sll(L2_scratch, LogBytesPerWord, L2_scratch);
1755 
1756   // Copy result to callers java stack
1757 
1758   __ set((intptr_t)CppInterpreter::_stack_to_stack, L4_scratch);
1759   __ ld_ptr(L4_scratch, L2_scratch, Lscratch);                          // get typed result converter address
1760   __ ld_ptr(STATE(_stack), O0);                                       // current top (prepushed)
1761   __ ld_ptr(STATE(_locals), O1);                                      // stack destination
1762 
1763   // O0 - will be source, O1 - will be destination (preserved)
1764   __ jmpl(Lscratch, G0, O7);                                          // and convert it
1765   __ delayed()->add(O0, wordSize, O0);                                // get source (top of current expr stack)
1766 
1767   // O1 == &locals[0]
1768 
1769   // Result is now on caller's stack. Just unwind current activation and resume
1770 
1771   Label unwind_recursive_activation;


1897 
1898   // stack points to next free location and not top element on expression stack
1899   // method expects sp to be pointing to topmost element
1900 
1901   __ ld_ptr(STATE(_thread), G2_thread);
1902   __ ld_ptr(STATE(_result._to_call._callee), G5_method);
1903 
1904 
1905   // SP already takes in to account the 2 extra words we use for slop
1906   // when we call a "static long no_params()" method. So if
1907   // we trim back sp by the amount of unused java expression stack
1908   // there will be automagically the 2 extra words we need.
1909   // We also have to worry about keeping SP aligned.
1910 
1911   __ ld_ptr(STATE(_stack), Gargs);
1912   __ ld_ptr(STATE(_stack_limit), L1_scratch);
1913 
1914   // compute the unused java stack size
1915   __ sub(Gargs, L1_scratch, L2_scratch);                       // compute unused space
1916 
1917   // Round down the unused space to that stack is always 16-byte aligned
1918   // by making the unused space a multiple of the size of two longs.
1919 
1920   __ and3(L2_scratch, -2*BytesPerLong, L2_scratch);
1921 
1922   // Now trim the stack
1923   __ add(SP, L2_scratch, SP);
1924 
1925 
1926   // Now point to the final argument (account for prepush)
1927   __ add(Gargs, wordSize, Gargs);
1928 #ifdef ASSERT
1929   // Make sure we have space for the window
1930   __ sub(Gargs, SP, L1_scratch);
1931   __ cmp(L1_scratch, 16*wordSize);
1932   {
1933     Label skip;
1934     __ brx(Assembler::greaterEqual, false, Assembler::pt, skip);
1935     __ delayed()->nop();
1936     __ stop("killed stack");
1937     __ bind(skip);
1938   }
1939 #endif // ASSERT
1940 


2150   // calculates the extra locals based on itself. Not what the callee does
2151   // to it. So it ignores last_frame_adjust value. Seems suspicious as far
2152   // as getting sender_sp correct.
2153 
2154   int extra_locals_size = callee_locals_size - callee_param_size;
2155   int monitor_size = (sizeof(BasicObjectLock) * moncount) / wordSize;
2156   int full_frame_words = size_activation_helper(extra_locals_size, method->max_stack(), monitor_size);
2157   int short_frame_words = size_activation_helper(extra_locals_size, method->max_stack(), monitor_size);
2158   int frame_words = is_top_frame ? full_frame_words : short_frame_words;
2159 
2160 
2161   /*
2162     if we actually have a frame to layout we must now fill in all the pieces. This means both
2163     the interpreterState and the registers.
2164   */
2165   if (interpreter_frame != NULL) {
2166 
2167     // MUCHO HACK
2168 
2169     intptr_t* frame_bottom = interpreter_frame->sp() - (full_frame_words - frame_words);
2170     // 'interpreter_frame->sp()' is unbiased while 'frame_bottom' must be a biased value in 64bit mode.
2171     assert(((intptr_t)frame_bottom & 0xf) == 0, "SP biased in layout_activation");
2172     frame_bottom = (intptr_t*)((intptr_t)frame_bottom - STACK_BIAS);
2173 
2174     /* Now fillin the interpreterState object */
2175 
2176     interpreterState cur_state = (interpreterState) ((intptr_t)interpreter_frame->fp() -  sizeof(BytecodeInterpreter));
2177 
2178 
2179     intptr_t* locals;
2180 
2181     // Calculate the postion of locals[0]. This is painful because of
2182     // stack alignment (same as ia64). The problem is that we can
2183     // not compute the location of locals from fp(). fp() will account
2184     // for the extra locals but it also accounts for aligning the stack
2185     // and we can't determine if the locals[0] was misaligned but max_locals
2186     // was enough to have the
2187     // calculate postion of locals. fp already accounts for extra locals.
2188     // +2 for the static long no_params() issue.
2189 
2190     if (caller->is_interpreted_frame()) {
2191       // locals must agree with the caller because it will be used to set the
2192       // caller's tos when we return.


2218 
2219 
2220     BytecodeInterpreter::layout_interpreterState(cur_state,
2221                                           caller,
2222                                           interpreter_frame,
2223                                           method,
2224                                           locals,
2225                                           stack,
2226                                           stack_base,
2227                                           monitor_base,
2228                                           frame_bottom,
2229                                           is_top_frame);
2230 
2231     BytecodeInterpreter::pd_layout_interpreterState(cur_state, interpreter_return_address, interpreter_frame->fp());
2232 
2233   }
2234   return frame_words;
2235 }
2236 
2237 #endif // CC_INTERP