1 /*
  2  * Copyright (c) 2018, Oracle and/or its affiliates. All rights reserved.
  3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  4  *
  5  * This code is free software; you can redistribute it and/or modify it
  6  * under the terms of the GNU General Public License version 2 only, as
  7  * published by the Free Software Foundation.
  8  *
  9  * This code is distributed in the hope that it will be useful, but WITHOUT
 10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 12  * version 2 for more details (a copy is included in the LICENSE file that
 13  * accompanied this code).
 14  *
 15  * You should have received a copy of the GNU General Public License version
 16  * 2 along with this work; if not, write to the Free Software Foundation,
 17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 18  *
 19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
 20  * or visit www.oracle.com if you need additional information or have any
 21  * questions.
 22  *
 23  */
 24 
 25 #include "precompiled.hpp"
 26 #include "asm/macroAssembler.inline.hpp"
 27 #include "gc/g1/g1BarrierSet.hpp"
 28 #include "gc/g1/g1BarrierSetAssembler.hpp"
 29 #include "gc/g1/g1BarrierSetRuntime.hpp"
 30 #include "gc/g1/g1CardTable.hpp"
 31 #include "gc/g1/g1ThreadLocalData.hpp"
 32 #include "gc/g1/heapRegion.hpp"
 33 #include "interpreter/interp_masm.hpp"
 34 #include "runtime/sharedRuntime.hpp"
 35 #include "utilities/macros.hpp"
 36 #ifdef COMPILER1
 37 #include "c1/c1_LIRAssembler.hpp"
 38 #include "c1/c1_MacroAssembler.hpp"
 39 #include "gc/g1/c1/g1BarrierSetC1.hpp"
 40 #endif
 41 
 42 #define __ masm->
 43 
 44 void G1BarrierSetAssembler::gen_write_ref_array_pre_barrier(MacroAssembler* masm, DecoratorSet decorators,
 45                                                             Register addr, Register count) {
 46   bool dest_uninitialized = (decorators & AS_DEST_NOT_INITIALIZED) != 0;
 47 
 48   if (!dest_uninitialized) {
 49     Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
 50 #ifndef _LP64
 51     __ push(thread);
 52     __ get_thread(thread);
 53 #endif
 54 
 55     Label filtered;
 56     Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
 57     // Is marking active?
 58     if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
 59       __ cmpl(in_progress, 0);
 60     } else {
 61       assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
 62       __ cmpb(in_progress, 0);
 63     }
 64 
 65     NOT_LP64(__ pop(thread);)
 66 
 67     __ jcc(Assembler::equal, filtered);
 68 
 69     __ pusha();                      // push registers
 70 #ifdef _LP64
 71     if (count == c_rarg0) {
 72       if (addr == c_rarg1) {
 73         // exactly backwards!!
 74         __ xchgptr(c_rarg1, c_rarg0);
 75       } else {
 76         __ movptr(c_rarg1, count);
 77         __ movptr(c_rarg0, addr);
 78       }
 79     } else {
 80       __ movptr(c_rarg0, addr);
 81       __ movptr(c_rarg1, count);
 82     }
 83     if (UseCompressedOops) {
 84       __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_narrow_oop_entry), 2);
 85     } else {
 86       __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_oop_entry), 2);
 87     }
 88 #else
 89     __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_pre_oop_entry),
 90                     addr, count);
 91 #endif
 92     __ popa();
 93 
 94     __ bind(filtered);
 95   }
 96 }
 97 
 98 void G1BarrierSetAssembler::gen_write_ref_array_post_barrier(MacroAssembler* masm, DecoratorSet decorators,
 99                                                              Register addr, Register count, Register tmp) {
100   __ pusha();             // push registers (overkill)
101 #ifdef _LP64
102   if (c_rarg0 == count) { // On win64 c_rarg0 == rcx
103     assert_different_registers(c_rarg1, addr);
104     __ mov(c_rarg1, count);
105     __ mov(c_rarg0, addr);
106   } else {
107     assert_different_registers(c_rarg0, count);
108     __ mov(c_rarg0, addr);
109     __ mov(c_rarg1, count);
110   }
111   __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_post_entry), 2);
112 #else
113   __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_array_post_entry),
114                   addr, count);
115 #endif
116   __ popa();
117 }
118 
119 void G1BarrierSetAssembler::load_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
120                                     Register dst, Address src, Register tmp1, Register tmp_thread) {
121   bool on_oop = type == T_OBJECT || type == T_ARRAY;
122   bool on_weak = (decorators & ON_WEAK_OOP_REF) != 0;
123   bool on_phantom = (decorators & ON_PHANTOM_OOP_REF) != 0;
124   bool on_reference = on_weak || on_phantom;
125   ModRefBarrierSetAssembler::load_at(masm, decorators, type, dst, src, tmp1, tmp_thread);
126   if (on_oop && on_reference) {
127     const Register thread = NOT_LP64(tmp_thread) LP64_ONLY(r15_thread);
128     NOT_LP64(__ get_thread(thread));
129 
130     // Generate the G1 pre-barrier code to log the value of
131     // the referent field in an SATB buffer.
132     g1_write_barrier_pre(masm /* masm */,
133                          noreg /* obj */,
134                          dst /* pre_val */,
135                          thread /* thread */,
136                          tmp1 /* tmp */,
137                          true /* tosca_live */,
138                          true /* expand_call */);
139   }
140 }
141 
142 void G1BarrierSetAssembler::g1_write_barrier_pre(MacroAssembler* masm,
143                                                  Register obj,
144                                                  Register pre_val,
145                                                  Register thread,
146                                                  Register tmp,
147                                                  bool tosca_live,
148                                                  bool expand_call) {
149   // If expand_call is true then we expand the call_VM_leaf macro
150   // directly to skip generating the check by
151   // InterpreterMacroAssembler::call_VM_leaf_base that checks _last_sp.
152 
153 #ifdef _LP64
154   assert(thread == r15_thread, "must be");
155 #endif // _LP64
156 
157   Label done;
158   Label runtime;
159 
160   assert(pre_val != noreg, "check this code");
161 
162   if (obj != noreg) {
163     assert_different_registers(obj, pre_val, tmp);
164     assert(pre_val != rax, "check this code");
165   }
166 
167   Address in_progress(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
168   Address index(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_index_offset()));
169   Address buffer(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_buffer_offset()));
170 
171   // Is marking active?
172   if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
173     __ cmpl(in_progress, 0);
174   } else {
175     assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
176     __ cmpb(in_progress, 0);
177   }
178   __ jcc(Assembler::equal, done);
179 
180   // Do we need to load the previous value?
181   if (obj != noreg) {
182     __ load_heap_oop(pre_val, Address(obj, 0), noreg, noreg, AS_RAW);
183   }
184 
185   // Is the previous value null?
186   __ cmpptr(pre_val, (int32_t) NULL_WORD);
187   __ jcc(Assembler::equal, done);
188 
189   // Can we store original value in the thread's buffer?
190   // Is index == 0?
191   // (The index field is typed as size_t.)
192 
193   __ movptr(tmp, index);                   // tmp := *index_adr
194   __ cmpptr(tmp, 0);                       // tmp == 0?
195   __ jcc(Assembler::equal, runtime);       // If yes, goto runtime
196 
197   __ subptr(tmp, wordSize);                // tmp := tmp - wordSize
198   __ movptr(index, tmp);                   // *index_adr := tmp
199   __ addptr(tmp, buffer);                  // tmp := tmp + *buffer_adr
200 
201   // Record the previous value
202   __ movptr(Address(tmp, 0), pre_val);
203   __ jmp(done);
204 
205   __ bind(runtime);
206   // save the live input values
207   if(tosca_live) __ push(rax);
208 
209   if (obj != noreg && obj != rax)
210     __ push(obj);
211 
212   if (pre_val != rax)
213     __ push(pre_val);
214 
215   // Calling the runtime using the regular call_VM_leaf mechanism generates
216   // code (generated by InterpreterMacroAssember::call_VM_leaf_base)
217   // that checks that the *(ebp+frame::interpreter_frame_last_sp) == NULL.
218   //
219   // If we care generating the pre-barrier without a frame (e.g. in the
220   // intrinsified Reference.get() routine) then ebp might be pointing to
221   // the caller frame and so this check will most likely fail at runtime.
222   //
223   // Expanding the call directly bypasses the generation of the check.
224   // So when we do not have have a full interpreter frame on the stack
225   // expand_call should be passed true.
226 
227   NOT_LP64( __ push(thread); )
228 
229   if (expand_call) {
230     LP64_ONLY( assert(pre_val != c_rarg1, "smashed arg"); )
231 #ifdef _LP64
232     if (c_rarg1 != thread) {
233       __ mov(c_rarg1, thread);
234     }
235     if (c_rarg0 != pre_val) {
236       __ mov(c_rarg0, pre_val);
237     }
238 #else
239     __ push(thread);
240     __ push(pre_val);
241 #endif
242     __ MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), 2);
243   } else {
244     __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), pre_val, thread);
245   }
246 
247   NOT_LP64( __ pop(thread); )
248 
249   // save the live input values
250   if (pre_val != rax)
251     __ pop(pre_val);
252 
253   if (obj != noreg && obj != rax)
254     __ pop(obj);
255 
256   if(tosca_live) __ pop(rax);
257 
258   __ bind(done);
259 }
260 
261 void G1BarrierSetAssembler::g1_write_barrier_post(MacroAssembler* masm,
262                                                   Register store_addr,
263                                                   Register new_val,
264                                                   Register thread,
265                                                   Register tmp,
266                                                   Register tmp2) {
267 #ifdef _LP64
268   assert(thread == r15_thread, "must be");
269 #endif // _LP64
270 
271   Address queue_index(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_index_offset()));
272   Address buffer(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_buffer_offset()));
273 
274   CardTableBarrierSet* ct =
275     barrier_set_cast<CardTableBarrierSet>(BarrierSet::barrier_set());
276   assert(sizeof(*ct->card_table()->byte_map_base()) == sizeof(jbyte), "adjust this code");
277 
278   Label done;
279   Label runtime;
280 
281   // Does store cross heap regions?
282 
283   __ movptr(tmp, store_addr);
284   __ xorptr(tmp, new_val);
285   __ shrptr(tmp, HeapRegion::LogOfHRGrainBytes);
286   __ jcc(Assembler::equal, done);
287 
288   // crosses regions, storing NULL?
289 
290   __ cmpptr(new_val, (int32_t) NULL_WORD);
291   __ jcc(Assembler::equal, done);
292 
293   // storing region crossing non-NULL, is card already dirty?
294 
295   const Register card_addr = tmp;
296   const Register cardtable = tmp2;
297 
298   __ movptr(card_addr, store_addr);
299   __ shrptr(card_addr, CardTable::card_shift);
300   // Do not use ExternalAddress to load 'byte_map_base', since 'byte_map_base' is NOT
301   // a valid address and therefore is not properly handled by the relocation code.
302   __ movptr(cardtable, (intptr_t)ct->card_table()->byte_map_base());
303   __ addptr(card_addr, cardtable);
304 
305   __ cmpb(Address(card_addr, 0), (int)G1CardTable::g1_young_card_val());
306   __ jcc(Assembler::equal, done);
307 
308   __ membar(Assembler::Membar_mask_bits(Assembler::StoreLoad));
309   __ cmpb(Address(card_addr, 0), (int)G1CardTable::dirty_card_val());
310   __ jcc(Assembler::equal, done);
311 
312 
313   // storing a region crossing, non-NULL oop, card is clean.
314   // dirty card and log.
315 
316   __ movb(Address(card_addr, 0), (int)G1CardTable::dirty_card_val());
317 
318   __ cmpl(queue_index, 0);
319   __ jcc(Assembler::equal, runtime);
320   __ subl(queue_index, wordSize);
321   __ movptr(tmp2, buffer);
322 #ifdef _LP64
323   __ movslq(rscratch1, queue_index);
324   __ addq(tmp2, rscratch1);
325   __ movq(Address(tmp2, 0), card_addr);
326 #else
327   __ addl(tmp2, queue_index);
328   __ movl(Address(tmp2, 0), card_addr);
329 #endif
330   __ jmp(done);
331 
332   __ bind(runtime);
333   // save the live input values
334   __ push(store_addr);
335   __ push(new_val);
336 #ifdef _LP64
337   __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_post_entry), card_addr, r15_thread);
338 #else
339   __ push(thread);
340   __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_post_entry), card_addr, thread);
341   __ pop(thread);
342 #endif
343   __ pop(new_val);
344   __ pop(store_addr);
345 
346   __ bind(done);
347 }
348 
349 void G1BarrierSetAssembler::oop_store_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
350                                          Address dst, Register val, Register tmp1, Register tmp2) {
351   bool in_heap = (decorators & IN_HEAP) != 0;
352   bool in_concurrent_root = (decorators & IN_CONCURRENT_ROOT) != 0;
353 
354   bool needs_pre_barrier = in_heap || in_concurrent_root;
355   bool needs_post_barrier = val != noreg && in_heap;
356 
357   Register tmp3 = LP64_ONLY(r8) NOT_LP64(rsi);
358   Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
359   // flatten object address if needed
360   // We do it regardless of precise because we need the registers
361   if (dst.index() == noreg && dst.disp() == 0) {
362     if (dst.base() != tmp1) {
363       __ movptr(tmp1, dst.base());
364     }
365   } else {
366     __ lea(tmp1, dst);
367   }
368 
369 #ifndef _LP64
370   InterpreterMacroAssembler *imasm = static_cast<InterpreterMacroAssembler*>(masm);
371 #endif
372 
373   NOT_LP64(__ get_thread(rcx));
374   NOT_LP64(imasm->save_bcp());
375 
376   if (needs_pre_barrier) {
377     g1_write_barrier_pre(masm /*masm*/,
378                          tmp1 /* obj */,
379                          tmp2 /* pre_val */,
380                          rthread /* thread */,
381                          tmp3  /* tmp */,
382                          val != noreg /* tosca_live */,
383                          false /* expand_call */);
384   }
385   if (val == noreg) {
386     BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp1, 0), val, noreg, noreg);
387   } else {
388     Register new_val = val;
389     if (needs_post_barrier) {
390       // G1 barrier needs uncompressed oop for region cross check.
391       if (UseCompressedOops) {
392         new_val = tmp2;
393         __ movptr(new_val, val);
394       }
395     }
396     BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp1, 0), val, noreg, noreg);
397     if (needs_post_barrier) {
398       g1_write_barrier_post(masm /*masm*/,
399                             tmp1 /* store_adr */,
400                             new_val /* new_val */,
401                             rthread /* thread */,
402                             tmp3 /* tmp */,
403                             tmp2 /* tmp2 */);
404     }
405   }
406   NOT_LP64(imasm->restore_bcp());
407 }
408 
409 #ifdef COMPILER1
410 
411 #undef __
412 #define __ ce->masm()->
413 
414 void G1BarrierSetAssembler::gen_pre_barrier_stub(LIR_Assembler* ce, G1PreBarrierStub* stub) {
415   G1BarrierSetC1* bs = (G1BarrierSetC1*)BarrierSet::barrier_set()->barrier_set_c1();
416   // At this point we know that marking is in progress.
417   // If do_load() is true then we have to emit the
418   // load of the previous value; otherwise it has already
419   // been loaded into _pre_val.
420 
421   __ bind(*stub->entry());
422   assert(stub->pre_val()->is_register(), "Precondition.");
423 
424   Register pre_val_reg = stub->pre_val()->as_register();
425 
426   if (stub->do_load()) {
427     ce->mem2reg(stub->addr(), stub->pre_val(), T_OBJECT, stub->patch_code(), stub->info(), false /*wide*/, false /*unaligned*/);
428   }
429 
430   __ cmpptr(pre_val_reg, (int32_t)NULL_WORD);
431   __ jcc(Assembler::equal, *stub->continuation());
432   ce->store_parameter(stub->pre_val()->as_register(), 0);
433   __ call(RuntimeAddress(bs->pre_barrier_c1_runtime_code_blob()->code_begin()));
434   __ jmp(*stub->continuation());
435 
436 }
437 
438 void G1BarrierSetAssembler::gen_post_barrier_stub(LIR_Assembler* ce, G1PostBarrierStub* stub) {
439   G1BarrierSetC1* bs = (G1BarrierSetC1*)BarrierSet::barrier_set()->barrier_set_c1();
440   __ bind(*stub->entry());
441   assert(stub->addr()->is_register(), "Precondition.");
442   assert(stub->new_val()->is_register(), "Precondition.");
443   Register new_val_reg = stub->new_val()->as_register();
444   __ cmpptr(new_val_reg, (int32_t) NULL_WORD);
445   __ jcc(Assembler::equal, *stub->continuation());
446   ce->store_parameter(stub->addr()->as_pointer_register(), 0);
447   __ call(RuntimeAddress(bs->post_barrier_c1_runtime_code_blob()->code_begin()));
448   __ jmp(*stub->continuation());
449 }
450 
451 #undef __
452 
453 #define __ sasm->
454 
455 void G1BarrierSetAssembler::generate_c1_pre_barrier_runtime_stub(StubAssembler* sasm) {
456   __ prologue("g1_pre_barrier", false);
457   // arg0 : previous value of memory
458 
459   __ push(rax);
460   __ push(rdx);
461 
462   const Register pre_val = rax;
463   const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
464   const Register tmp = rdx;
465 
466   NOT_LP64(__ get_thread(thread);)
467 
468   Address queue_active(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset()));
469   Address queue_index(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_index_offset()));
470   Address buffer(thread, in_bytes(G1ThreadLocalData::satb_mark_queue_buffer_offset()));
471 
472   Label done;
473   Label runtime;
474 
475   // Is marking still active?
476   if (in_bytes(SATBMarkQueue::byte_width_of_active()) == 4) {
477     __ cmpl(queue_active, 0);
478   } else {
479     assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "Assumption");
480     __ cmpb(queue_active, 0);
481   }
482   __ jcc(Assembler::equal, done);
483 
484   // Can we store original value in the thread's buffer?
485 
486   __ movptr(tmp, queue_index);
487   __ testptr(tmp, tmp);
488   __ jcc(Assembler::zero, runtime);
489   __ subptr(tmp, wordSize);
490   __ movptr(queue_index, tmp);
491   __ addptr(tmp, buffer);
492 
493   // prev_val (rax)
494   __ load_parameter(0, pre_val);
495   __ movptr(Address(tmp, 0), pre_val);
496   __ jmp(done);
497 
498   __ bind(runtime);
499 
500   __ save_live_registers_no_oop_map(true);
501 
502   // load the pre-value
503   __ load_parameter(0, rcx);
504   __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), rcx, thread);
505 
506   __ restore_live_registers(true);
507 
508   __ bind(done);
509 
510   __ pop(rdx);
511   __ pop(rax);
512 
513   __ epilogue();
514 }
515 
516 void G1BarrierSetAssembler::generate_c1_post_barrier_runtime_stub(StubAssembler* sasm) {
517   __ prologue("g1_post_barrier", false);
518 
519   // arg0: store_address
520   Address store_addr(rbp, 2*BytesPerWord);
521 
522   CardTableBarrierSet* ct =
523     barrier_set_cast<CardTableBarrierSet>(BarrierSet::barrier_set());
524   assert(sizeof(*ct->card_table()->byte_map_base()) == sizeof(jbyte), "adjust this code");
525 
526   Label done;
527   Label enqueued;
528   Label runtime;
529 
530   // At this point we know new_value is non-NULL and the new_value crosses regions.
531   // Must check to see if card is already dirty
532 
533   const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
534 
535   Address queue_index(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_index_offset()));
536   Address buffer(thread, in_bytes(G1ThreadLocalData::dirty_card_queue_buffer_offset()));
537 
538   __ push(rax);
539   __ push(rcx);
540 
541   const Register cardtable = rax;
542   const Register card_addr = rcx;
543 
544   __ load_parameter(0, card_addr);
545   __ shrptr(card_addr, CardTable::card_shift);
546   // Do not use ExternalAddress to load 'byte_map_base', since 'byte_map_base' is NOT
547   // a valid address and therefore is not properly handled by the relocation code.
548   __ movptr(cardtable, (intptr_t)ct->card_table()->byte_map_base());
549   __ addptr(card_addr, cardtable);
550 
551   NOT_LP64(__ get_thread(thread);)
552 
553   __ cmpb(Address(card_addr, 0), (int)G1CardTable::g1_young_card_val());
554   __ jcc(Assembler::equal, done);
555 
556   __ membar(Assembler::Membar_mask_bits(Assembler::StoreLoad));
557   __ cmpb(Address(card_addr, 0), (int)CardTable::dirty_card_val());
558   __ jcc(Assembler::equal, done);
559 
560   // storing region crossing non-NULL, card is clean.
561   // dirty card and log.
562 
563   __ movb(Address(card_addr, 0), (int)CardTable::dirty_card_val());
564 
565   const Register tmp = rdx;
566   __ push(rdx);
567 
568   __ movptr(tmp, queue_index);
569   __ testptr(tmp, tmp);
570   __ jcc(Assembler::zero, runtime);
571   __ subptr(tmp, wordSize);
572   __ movptr(queue_index, tmp);
573   __ addptr(tmp, buffer);
574   __ movptr(Address(tmp, 0), card_addr);
575   __ jmp(enqueued);
576 
577   __ bind(runtime);
578 
579   __ save_live_registers_no_oop_map(true);
580 
581   __ call_VM_leaf(CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_post_entry), card_addr, thread);
582 
583   __ restore_live_registers(true);
584 
585   __ bind(enqueued);
586   __ pop(rdx);
587 
588   __ bind(done);
589   __ pop(rcx);
590   __ pop(rax);
591 
592   __ epilogue();
593 }
594 
595 #undef __
596 
597 #endif // COMPILER1