Mon, 12 Mar 2012 15:28:07 -0700
7152957: VM crashes with assert(false) failed: bad AD file
Reviewed-by: kvn, never
Contributed-by: nils.eliasson@oracle.com
1 /*
2 * Copyright (c) 1997, 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 "interp_masm_sparc.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "interpreter/interpreterRuntime.hpp"
29 #include "oops/arrayOop.hpp"
30 #include "oops/markOop.hpp"
31 #include "oops/methodDataOop.hpp"
32 #include "oops/methodOop.hpp"
33 #include "prims/jvmtiExport.hpp"
34 #include "prims/jvmtiRedefineClassesTrace.hpp"
35 #include "prims/jvmtiThreadState.hpp"
36 #include "runtime/basicLock.hpp"
37 #include "runtime/biasedLocking.hpp"
38 #include "runtime/sharedRuntime.hpp"
39 #ifdef TARGET_OS_FAMILY_linux
40 # include "thread_linux.inline.hpp"
41 #endif
42 #ifdef TARGET_OS_FAMILY_solaris
43 # include "thread_solaris.inline.hpp"
44 #endif
46 #ifndef CC_INTERP
47 #ifndef FAST_DISPATCH
48 #define FAST_DISPATCH 1
49 #endif
50 #undef FAST_DISPATCH
52 // Implementation of InterpreterMacroAssembler
54 // This file specializes the assember with interpreter-specific macros
56 const Address InterpreterMacroAssembler::l_tmp(FP, (frame::interpreter_frame_l_scratch_fp_offset * wordSize) + STACK_BIAS);
57 const Address InterpreterMacroAssembler::d_tmp(FP, (frame::interpreter_frame_d_scratch_fp_offset * wordSize) + STACK_BIAS);
59 #else // CC_INTERP
60 #ifndef STATE
61 #define STATE(field_name) Lstate, in_bytes(byte_offset_of(BytecodeInterpreter, field_name))
62 #endif // STATE
64 #endif // CC_INTERP
66 void InterpreterMacroAssembler::compute_extra_locals_size_in_bytes(Register args_size, Register locals_size, Register delta) {
67 // Note: this algorithm is also used by C1's OSR entry sequence.
68 // Any changes should also be applied to CodeEmitter::emit_osr_entry().
69 assert_different_registers(args_size, locals_size);
70 // max_locals*2 for TAGS. Assumes that args_size has already been adjusted.
71 subcc(locals_size, args_size, delta);// extra space for non-arguments locals in words
72 // Use br/mov combination because it works on both V8 and V9 and is
73 // faster.
74 Label skip_move;
75 br(Assembler::negative, true, Assembler::pt, skip_move);
76 delayed()->mov(G0, delta);
77 bind(skip_move);
78 round_to(delta, WordsPerLong); // make multiple of 2 (SP must be 2-word aligned)
79 sll(delta, LogBytesPerWord, delta); // extra space for locals in bytes
80 }
82 #ifndef CC_INTERP
84 // Dispatch code executed in the prolog of a bytecode which does not do it's
85 // own dispatch. The dispatch address is computed and placed in IdispatchAddress
86 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) {
87 assert_not_delayed();
88 #ifdef FAST_DISPATCH
89 // FAST_DISPATCH and ProfileInterpreter are mutually exclusive since
90 // they both use I2.
91 assert(!ProfileInterpreter, "FAST_DISPATCH and +ProfileInterpreter are mutually exclusive");
92 ldub(Lbcp, bcp_incr, Lbyte_code); // load next bytecode
93 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code);
94 // add offset to correct dispatch table
95 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize
96 ld_ptr(IdispatchTables, Lbyte_code, IdispatchAddress);// get entry addr
97 #else
98 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode
99 // dispatch table to use
100 AddressLiteral tbl(Interpreter::dispatch_table(state));
101 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize
102 set(tbl, G3_scratch); // compute addr of table
103 ld_ptr(G3_scratch, Lbyte_code, IdispatchAddress); // get entry addr
104 #endif
105 }
108 // Dispatch code executed in the epilog of a bytecode which does not do it's
109 // own dispatch. The dispatch address in IdispatchAddress is used for the
110 // dispatch.
111 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) {
112 assert_not_delayed();
113 verify_FPU(1, state);
114 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
115 jmp( IdispatchAddress, 0 );
116 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr);
117 else delayed()->nop();
118 }
121 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr) {
122 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr)
123 assert_not_delayed();
124 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode
125 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr);
126 }
129 void InterpreterMacroAssembler::dispatch_next_noverify_oop(TosState state, int bcp_incr) {
130 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr)
131 assert_not_delayed();
132 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode
133 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr, false);
134 }
137 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
138 // load current bytecode
139 assert_not_delayed();
140 ldub( Lbcp, 0, Lbyte_code); // load next bytecode
141 dispatch_base(state, table);
142 }
145 void InterpreterMacroAssembler::call_VM_leaf_base(
146 Register java_thread,
147 address entry_point,
148 int number_of_arguments
149 ) {
150 if (!java_thread->is_valid())
151 java_thread = L7_thread_cache;
152 // super call
153 MacroAssembler::call_VM_leaf_base(java_thread, entry_point, number_of_arguments);
154 }
157 void InterpreterMacroAssembler::call_VM_base(
158 Register oop_result,
159 Register java_thread,
160 Register last_java_sp,
161 address entry_point,
162 int number_of_arguments,
163 bool check_exception
164 ) {
165 if (!java_thread->is_valid())
166 java_thread = L7_thread_cache;
167 // See class ThreadInVMfromInterpreter, which assumes that the interpreter
168 // takes responsibility for setting its own thread-state on call-out.
169 // However, ThreadInVMfromInterpreter resets the state to "in_Java".
171 //save_bcp(); // save bcp
172 MacroAssembler::call_VM_base(oop_result, java_thread, last_java_sp, entry_point, number_of_arguments, check_exception);
173 //restore_bcp(); // restore bcp
174 //restore_locals(); // restore locals pointer
175 }
178 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) {
179 if (JvmtiExport::can_pop_frame()) {
180 Label L;
182 // Check the "pending popframe condition" flag in the current thread
183 ld(G2_thread, JavaThread::popframe_condition_offset(), scratch_reg);
185 // Initiate popframe handling only if it is not already being processed. If the flag
186 // has the popframe_processing bit set, it means that this code is called *during* popframe
187 // handling - we don't want to reenter.
188 btst(JavaThread::popframe_pending_bit, scratch_reg);
189 br(zero, false, pt, L);
190 delayed()->nop();
191 btst(JavaThread::popframe_processing_bit, scratch_reg);
192 br(notZero, false, pt, L);
193 delayed()->nop();
195 // Call Interpreter::remove_activation_preserving_args_entry() to get the
196 // address of the same-named entrypoint in the generated interpreter code.
197 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
199 // Jump to Interpreter::_remove_activation_preserving_args_entry
200 jmpl(O0, G0, G0);
201 delayed()->nop();
202 bind(L);
203 }
204 }
207 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
208 Register thr_state = G4_scratch;
209 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state);
210 const Address tos_addr(thr_state, JvmtiThreadState::earlyret_tos_offset());
211 const Address oop_addr(thr_state, JvmtiThreadState::earlyret_oop_offset());
212 const Address val_addr(thr_state, JvmtiThreadState::earlyret_value_offset());
213 switch (state) {
214 case ltos: ld_long(val_addr, Otos_l); break;
215 case atos: ld_ptr(oop_addr, Otos_l);
216 st_ptr(G0, oop_addr); break;
217 case btos: // fall through
218 case ctos: // fall through
219 case stos: // fall through
220 case itos: ld(val_addr, Otos_l1); break;
221 case ftos: ldf(FloatRegisterImpl::S, val_addr, Ftos_f); break;
222 case dtos: ldf(FloatRegisterImpl::D, val_addr, Ftos_d); break;
223 case vtos: /* nothing to do */ break;
224 default : ShouldNotReachHere();
225 }
226 // Clean up tos value in the jvmti thread state
227 or3(G0, ilgl, G3_scratch);
228 stw(G3_scratch, tos_addr);
229 st_long(G0, val_addr);
230 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
231 }
234 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) {
235 if (JvmtiExport::can_force_early_return()) {
236 Label L;
237 Register thr_state = G3_scratch;
238 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state);
239 br_null_short(thr_state, pt, L); // if (thread->jvmti_thread_state() == NULL) exit;
241 // Initiate earlyret handling only if it is not already being processed.
242 // If the flag has the earlyret_processing bit set, it means that this code
243 // is called *during* earlyret handling - we don't want to reenter.
244 ld(thr_state, JvmtiThreadState::earlyret_state_offset(), G4_scratch);
245 cmp_and_br_short(G4_scratch, JvmtiThreadState::earlyret_pending, Assembler::notEqual, pt, L);
247 // Call Interpreter::remove_activation_early_entry() to get the address of the
248 // same-named entrypoint in the generated interpreter code
249 ld(thr_state, JvmtiThreadState::earlyret_tos_offset(), Otos_l1);
250 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), Otos_l1);
252 // Jump to Interpreter::_remove_activation_early_entry
253 jmpl(O0, G0, G0);
254 delayed()->nop();
255 bind(L);
256 }
257 }
260 void InterpreterMacroAssembler::super_call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2) {
261 mov(arg_1, O0);
262 mov(arg_2, O1);
263 MacroAssembler::call_VM_leaf_base(thread_cache, entry_point, 2);
264 }
265 #endif /* CC_INTERP */
268 #ifndef CC_INTERP
270 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table) {
271 assert_not_delayed();
272 dispatch_Lbyte_code(state, table);
273 }
276 void InterpreterMacroAssembler::dispatch_normal(TosState state) {
277 dispatch_base(state, Interpreter::normal_table(state));
278 }
281 void InterpreterMacroAssembler::dispatch_only(TosState state) {
282 dispatch_base(state, Interpreter::dispatch_table(state));
283 }
286 // common code to dispatch and dispatch_only
287 // dispatch value in Lbyte_code and increment Lbcp
289 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, address* table, int bcp_incr, bool verify) {
290 verify_FPU(1, state);
291 // %%%%% maybe implement +VerifyActivationFrameSize here
292 //verify_thread(); //too slow; we will just verify on method entry & exit
293 if (verify) interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
294 #ifdef FAST_DISPATCH
295 if (table == Interpreter::dispatch_table(state)) {
296 // use IdispatchTables
297 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code);
298 // add offset to correct dispatch table
299 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize
300 ld_ptr(IdispatchTables, Lbyte_code, G3_scratch); // get entry addr
301 } else {
302 #endif
303 // dispatch table to use
304 AddressLiteral tbl(table);
305 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize
306 set(tbl, G3_scratch); // compute addr of table
307 ld_ptr(G3_scratch, Lbyte_code, G3_scratch); // get entry addr
308 #ifdef FAST_DISPATCH
309 }
310 #endif
311 jmp( G3_scratch, 0 );
312 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr);
313 else delayed()->nop();
314 }
317 // Helpers for expression stack
319 // Longs and doubles are Category 2 computational types in the
320 // JVM specification (section 3.11.1) and take 2 expression stack or
321 // local slots.
322 // Aligning them on 32 bit with tagged stacks is hard because the code generated
323 // for the dup* bytecodes depends on what types are already on the stack.
324 // If the types are split into the two stack/local slots, that is much easier
325 // (and we can use 0 for non-reference tags).
327 // Known good alignment in _LP64 but unknown otherwise
328 void InterpreterMacroAssembler::load_unaligned_double(Register r1, int offset, FloatRegister d) {
329 assert_not_delayed();
331 #ifdef _LP64
332 ldf(FloatRegisterImpl::D, r1, offset, d);
333 #else
334 ldf(FloatRegisterImpl::S, r1, offset, d);
335 ldf(FloatRegisterImpl::S, r1, offset + Interpreter::stackElementSize, d->successor());
336 #endif
337 }
339 // Known good alignment in _LP64 but unknown otherwise
340 void InterpreterMacroAssembler::store_unaligned_double(FloatRegister d, Register r1, int offset) {
341 assert_not_delayed();
343 #ifdef _LP64
344 stf(FloatRegisterImpl::D, d, r1, offset);
345 // store something more useful here
346 debug_only(stx(G0, r1, offset+Interpreter::stackElementSize);)
347 #else
348 stf(FloatRegisterImpl::S, d, r1, offset);
349 stf(FloatRegisterImpl::S, d->successor(), r1, offset + Interpreter::stackElementSize);
350 #endif
351 }
354 // Known good alignment in _LP64 but unknown otherwise
355 void InterpreterMacroAssembler::load_unaligned_long(Register r1, int offset, Register rd) {
356 assert_not_delayed();
357 #ifdef _LP64
358 ldx(r1, offset, rd);
359 #else
360 ld(r1, offset, rd);
361 ld(r1, offset + Interpreter::stackElementSize, rd->successor());
362 #endif
363 }
365 // Known good alignment in _LP64 but unknown otherwise
366 void InterpreterMacroAssembler::store_unaligned_long(Register l, Register r1, int offset) {
367 assert_not_delayed();
369 #ifdef _LP64
370 stx(l, r1, offset);
371 // store something more useful here
372 debug_only(stx(G0, r1, offset+Interpreter::stackElementSize);)
373 #else
374 st(l, r1, offset);
375 st(l->successor(), r1, offset + Interpreter::stackElementSize);
376 #endif
377 }
379 void InterpreterMacroAssembler::pop_i(Register r) {
380 assert_not_delayed();
381 ld(Lesp, Interpreter::expr_offset_in_bytes(0), r);
382 inc(Lesp, Interpreter::stackElementSize);
383 debug_only(verify_esp(Lesp));
384 }
386 void InterpreterMacroAssembler::pop_ptr(Register r, Register scratch) {
387 assert_not_delayed();
388 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), r);
389 inc(Lesp, Interpreter::stackElementSize);
390 debug_only(verify_esp(Lesp));
391 }
393 void InterpreterMacroAssembler::pop_l(Register r) {
394 assert_not_delayed();
395 load_unaligned_long(Lesp, Interpreter::expr_offset_in_bytes(0), r);
396 inc(Lesp, 2*Interpreter::stackElementSize);
397 debug_only(verify_esp(Lesp));
398 }
401 void InterpreterMacroAssembler::pop_f(FloatRegister f, Register scratch) {
402 assert_not_delayed();
403 ldf(FloatRegisterImpl::S, Lesp, Interpreter::expr_offset_in_bytes(0), f);
404 inc(Lesp, Interpreter::stackElementSize);
405 debug_only(verify_esp(Lesp));
406 }
409 void InterpreterMacroAssembler::pop_d(FloatRegister f, Register scratch) {
410 assert_not_delayed();
411 load_unaligned_double(Lesp, Interpreter::expr_offset_in_bytes(0), f);
412 inc(Lesp, 2*Interpreter::stackElementSize);
413 debug_only(verify_esp(Lesp));
414 }
417 void InterpreterMacroAssembler::push_i(Register r) {
418 assert_not_delayed();
419 debug_only(verify_esp(Lesp));
420 st(r, Lesp, 0);
421 dec(Lesp, Interpreter::stackElementSize);
422 }
424 void InterpreterMacroAssembler::push_ptr(Register r) {
425 assert_not_delayed();
426 st_ptr(r, Lesp, 0);
427 dec(Lesp, Interpreter::stackElementSize);
428 }
430 // remember: our convention for longs in SPARC is:
431 // O0 (Otos_l1) has high-order part in first word,
432 // O1 (Otos_l2) has low-order part in second word
434 void InterpreterMacroAssembler::push_l(Register r) {
435 assert_not_delayed();
436 debug_only(verify_esp(Lesp));
437 // Longs are stored in memory-correct order, even if unaligned.
438 int offset = -Interpreter::stackElementSize;
439 store_unaligned_long(r, Lesp, offset);
440 dec(Lesp, 2 * Interpreter::stackElementSize);
441 }
444 void InterpreterMacroAssembler::push_f(FloatRegister f) {
445 assert_not_delayed();
446 debug_only(verify_esp(Lesp));
447 stf(FloatRegisterImpl::S, f, Lesp, 0);
448 dec(Lesp, Interpreter::stackElementSize);
449 }
452 void InterpreterMacroAssembler::push_d(FloatRegister d) {
453 assert_not_delayed();
454 debug_only(verify_esp(Lesp));
455 // Longs are stored in memory-correct order, even if unaligned.
456 int offset = -Interpreter::stackElementSize;
457 store_unaligned_double(d, Lesp, offset);
458 dec(Lesp, 2 * Interpreter::stackElementSize);
459 }
462 void InterpreterMacroAssembler::push(TosState state) {
463 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
464 switch (state) {
465 case atos: push_ptr(); break;
466 case btos: push_i(); break;
467 case ctos:
468 case stos: push_i(); break;
469 case itos: push_i(); break;
470 case ltos: push_l(); break;
471 case ftos: push_f(); break;
472 case dtos: push_d(); break;
473 case vtos: /* nothing to do */ break;
474 default : ShouldNotReachHere();
475 }
476 }
479 void InterpreterMacroAssembler::pop(TosState state) {
480 switch (state) {
481 case atos: pop_ptr(); break;
482 case btos: pop_i(); break;
483 case ctos:
484 case stos: pop_i(); break;
485 case itos: pop_i(); break;
486 case ltos: pop_l(); break;
487 case ftos: pop_f(); break;
488 case dtos: pop_d(); break;
489 case vtos: /* nothing to do */ break;
490 default : ShouldNotReachHere();
491 }
492 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
493 }
496 // Helpers for swap and dup
497 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
498 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(n), val);
499 }
500 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
501 st_ptr(val, Lesp, Interpreter::expr_offset_in_bytes(n));
502 }
505 void InterpreterMacroAssembler::load_receiver(Register param_count,
506 Register recv) {
507 sll(param_count, Interpreter::logStackElementSize, param_count);
508 ld_ptr(Lesp, param_count, recv); // gets receiver Oop
509 }
511 void InterpreterMacroAssembler::empty_expression_stack() {
512 // Reset Lesp.
513 sub( Lmonitors, wordSize, Lesp );
515 // Reset SP by subtracting more space from Lesp.
516 Label done;
517 verify_oop(Lmethod);
518 assert(G4_scratch != Gframe_size, "Only you can prevent register aliasing!");
520 // A native does not need to do this, since its callee does not change SP.
521 ld(Lmethod, methodOopDesc::access_flags_offset(), Gframe_size); // Load access flags.
522 btst(JVM_ACC_NATIVE, Gframe_size);
523 br(Assembler::notZero, false, Assembler::pt, done);
524 delayed()->nop();
526 // Compute max expression stack+register save area
527 lduh(Lmethod, in_bytes(methodOopDesc::max_stack_offset()), Gframe_size); // Load max stack.
528 add( Gframe_size, frame::memory_parameter_word_sp_offset, Gframe_size );
530 //
531 // now set up a stack frame with the size computed above
532 //
533 //round_to( Gframe_size, WordsPerLong ); // -- moved down to the "and" below
534 sll( Gframe_size, LogBytesPerWord, Gframe_size );
535 sub( Lesp, Gframe_size, Gframe_size );
536 and3( Gframe_size, -(2 * wordSize), Gframe_size ); // align SP (downwards) to an 8/16-byte boundary
537 debug_only(verify_sp(Gframe_size, G4_scratch));
538 #ifdef _LP64
539 sub(Gframe_size, STACK_BIAS, Gframe_size );
540 #endif
541 mov(Gframe_size, SP);
543 bind(done);
544 }
547 #ifdef ASSERT
548 void InterpreterMacroAssembler::verify_sp(Register Rsp, Register Rtemp) {
549 Label Bad, OK;
551 // Saved SP must be aligned.
552 #ifdef _LP64
553 btst(2*BytesPerWord-1, Rsp);
554 #else
555 btst(LongAlignmentMask, Rsp);
556 #endif
557 br(Assembler::notZero, false, Assembler::pn, Bad);
558 delayed()->nop();
560 // Saved SP, plus register window size, must not be above FP.
561 add(Rsp, frame::register_save_words * wordSize, Rtemp);
562 #ifdef _LP64
563 sub(Rtemp, STACK_BIAS, Rtemp); // Bias Rtemp before cmp to FP
564 #endif
565 cmp_and_brx_short(Rtemp, FP, Assembler::greaterUnsigned, Assembler::pn, Bad);
567 // Saved SP must not be ridiculously below current SP.
568 size_t maxstack = MAX2(JavaThread::stack_size_at_create(), (size_t) 4*K*K);
569 set(maxstack, Rtemp);
570 sub(SP, Rtemp, Rtemp);
571 #ifdef _LP64
572 add(Rtemp, STACK_BIAS, Rtemp); // Unbias Rtemp before cmp to Rsp
573 #endif
574 cmp_and_brx_short(Rsp, Rtemp, Assembler::lessUnsigned, Assembler::pn, Bad);
576 ba_short(OK);
578 bind(Bad);
579 stop("on return to interpreted call, restored SP is corrupted");
581 bind(OK);
582 }
585 void InterpreterMacroAssembler::verify_esp(Register Resp) {
586 // about to read or write Resp[0]
587 // make sure it is not in the monitors or the register save area
588 Label OK1, OK2;
590 cmp(Resp, Lmonitors);
591 brx(Assembler::lessUnsigned, true, Assembler::pt, OK1);
592 delayed()->sub(Resp, frame::memory_parameter_word_sp_offset * wordSize, Resp);
593 stop("too many pops: Lesp points into monitor area");
594 bind(OK1);
595 #ifdef _LP64
596 sub(Resp, STACK_BIAS, Resp);
597 #endif
598 cmp(Resp, SP);
599 brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, OK2);
600 delayed()->add(Resp, STACK_BIAS + frame::memory_parameter_word_sp_offset * wordSize, Resp);
601 stop("too many pushes: Lesp points into register window");
602 bind(OK2);
603 }
604 #endif // ASSERT
606 // Load compiled (i2c) or interpreter entry when calling from interpreted and
607 // do the call. Centralized so that all interpreter calls will do the same actions.
608 // If jvmti single stepping is on for a thread we must not call compiled code.
609 void InterpreterMacroAssembler::call_from_interpreter(Register target, Register scratch, Register Rret) {
611 // Assume we want to go compiled if available
613 ld_ptr(G5_method, in_bytes(methodOopDesc::from_interpreted_offset()), target);
615 if (JvmtiExport::can_post_interpreter_events()) {
616 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
617 // compiled code in threads for which the event is enabled. Check here for
618 // interp_only_mode if these events CAN be enabled.
619 verify_thread();
620 Label skip_compiled_code;
622 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset());
623 ld(interp_only, scratch);
624 cmp_zero_and_br(Assembler::notZero, scratch, skip_compiled_code, true, Assembler::pn);
625 delayed()->ld_ptr(G5_method, in_bytes(methodOopDesc::interpreter_entry_offset()), target);
626 bind(skip_compiled_code);
627 }
629 // the i2c_adapters need methodOop in G5_method (right? %%%)
630 // do the call
631 #ifdef ASSERT
632 {
633 Label ok;
634 br_notnull_short(target, Assembler::pt, ok);
635 stop("null entry point");
636 bind(ok);
637 }
638 #endif // ASSERT
640 // Adjust Rret first so Llast_SP can be same as Rret
641 add(Rret, -frame::pc_return_offset, O7);
642 add(Lesp, BytesPerWord, Gargs); // setup parameter pointer
643 // Record SP so we can remove any stack space allocated by adapter transition
644 jmp(target, 0);
645 delayed()->mov(SP, Llast_SP);
646 }
648 void InterpreterMacroAssembler::if_cmp(Condition cc, bool ptr_compare) {
649 assert_not_delayed();
651 Label not_taken;
652 if (ptr_compare) brx(cc, false, Assembler::pn, not_taken);
653 else br (cc, false, Assembler::pn, not_taken);
654 delayed()->nop();
656 TemplateTable::branch(false,false);
658 bind(not_taken);
660 profile_not_taken_branch(G3_scratch);
661 }
664 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(
665 int bcp_offset,
666 Register Rtmp,
667 Register Rdst,
668 signedOrNot is_signed,
669 setCCOrNot should_set_CC ) {
670 assert(Rtmp != Rdst, "need separate temp register");
671 assert_not_delayed();
672 switch (is_signed) {
673 default: ShouldNotReachHere();
675 case Signed: ldsb( Lbcp, bcp_offset, Rdst ); break; // high byte
676 case Unsigned: ldub( Lbcp, bcp_offset, Rdst ); break; // high byte
677 }
678 ldub( Lbcp, bcp_offset + 1, Rtmp ); // low byte
679 sll( Rdst, BitsPerByte, Rdst);
680 switch (should_set_CC ) {
681 default: ShouldNotReachHere();
683 case set_CC: orcc( Rdst, Rtmp, Rdst ); break;
684 case dont_set_CC: or3( Rdst, Rtmp, Rdst ); break;
685 }
686 }
689 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(
690 int bcp_offset,
691 Register Rtmp,
692 Register Rdst,
693 setCCOrNot should_set_CC ) {
694 assert(Rtmp != Rdst, "need separate temp register");
695 assert_not_delayed();
696 add( Lbcp, bcp_offset, Rtmp);
697 andcc( Rtmp, 3, G0);
698 Label aligned;
699 switch (should_set_CC ) {
700 default: ShouldNotReachHere();
702 case set_CC: break;
703 case dont_set_CC: break;
704 }
706 br(Assembler::zero, true, Assembler::pn, aligned);
707 #ifdef _LP64
708 delayed()->ldsw(Rtmp, 0, Rdst);
709 #else
710 delayed()->ld(Rtmp, 0, Rdst);
711 #endif
713 ldub(Lbcp, bcp_offset + 3, Rdst);
714 ldub(Lbcp, bcp_offset + 2, Rtmp); sll(Rtmp, 8, Rtmp); or3(Rtmp, Rdst, Rdst);
715 ldub(Lbcp, bcp_offset + 1, Rtmp); sll(Rtmp, 16, Rtmp); or3(Rtmp, Rdst, Rdst);
716 #ifdef _LP64
717 ldsb(Lbcp, bcp_offset + 0, Rtmp); sll(Rtmp, 24, Rtmp);
718 #else
719 // Unsigned load is faster than signed on some implementations
720 ldub(Lbcp, bcp_offset + 0, Rtmp); sll(Rtmp, 24, Rtmp);
721 #endif
722 or3(Rtmp, Rdst, Rdst );
724 bind(aligned);
725 if (should_set_CC == set_CC) tst(Rdst);
726 }
729 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register cache, Register tmp,
730 int bcp_offset, size_t index_size) {
731 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
732 if (index_size == sizeof(u2)) {
733 get_2_byte_integer_at_bcp(bcp_offset, cache, tmp, Unsigned);
734 } else if (index_size == sizeof(u4)) {
735 assert(EnableInvokeDynamic, "giant index used only for JSR 292");
736 get_4_byte_integer_at_bcp(bcp_offset, cache, tmp);
737 assert(constantPoolCacheOopDesc::decode_secondary_index(~123) == 123, "else change next line");
738 xor3(tmp, -1, tmp); // convert to plain index
739 } else if (index_size == sizeof(u1)) {
740 assert(EnableInvokeDynamic, "tiny index used only for JSR 292");
741 ldub(Lbcp, bcp_offset, tmp);
742 } else {
743 ShouldNotReachHere();
744 }
745 }
748 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, Register tmp,
749 int bcp_offset, size_t index_size) {
750 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
751 assert_different_registers(cache, tmp);
752 assert_not_delayed();
753 get_cache_index_at_bcp(cache, tmp, bcp_offset, index_size);
754 // convert from field index to ConstantPoolCacheEntry index and from
755 // word index to byte offset
756 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp);
757 add(LcpoolCache, tmp, cache);
758 }
761 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
762 Register temp,
763 Register bytecode,
764 int byte_no,
765 int bcp_offset,
766 size_t index_size) {
767 get_cache_and_index_at_bcp(cache, temp, bcp_offset, index_size);
768 ld_ptr(cache, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset(), bytecode);
769 const int shift_count = (1 + byte_no) * BitsPerByte;
770 srl( bytecode, shift_count, bytecode);
771 and3(bytecode, 0xFF, bytecode);
772 }
775 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, Register tmp,
776 int bcp_offset, size_t index_size) {
777 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
778 assert_different_registers(cache, tmp);
779 assert_not_delayed();
780 if (index_size == sizeof(u2)) {
781 get_2_byte_integer_at_bcp(bcp_offset, cache, tmp, Unsigned);
782 } else {
783 ShouldNotReachHere(); // other sizes not supported here
784 }
785 // convert from field index to ConstantPoolCacheEntry index
786 // and from word index to byte offset
787 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp);
788 // skip past the header
789 add(tmp, in_bytes(constantPoolCacheOopDesc::base_offset()), tmp);
790 // construct pointer to cache entry
791 add(LcpoolCache, tmp, cache);
792 }
795 // Generate a subtype check: branch to ok_is_subtype if sub_klass is
796 // a subtype of super_klass. Blows registers Rsuper_klass, Rsub_klass, tmp1, tmp2.
797 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
798 Register Rsuper_klass,
799 Register Rtmp1,
800 Register Rtmp2,
801 Register Rtmp3,
802 Label &ok_is_subtype ) {
803 Label not_subtype;
805 // Profile the not-null value's klass.
806 profile_typecheck(Rsub_klass, Rtmp1);
808 check_klass_subtype_fast_path(Rsub_klass, Rsuper_klass,
809 Rtmp1, Rtmp2,
810 &ok_is_subtype, ¬_subtype, NULL);
812 check_klass_subtype_slow_path(Rsub_klass, Rsuper_klass,
813 Rtmp1, Rtmp2, Rtmp3, /*hack:*/ noreg,
814 &ok_is_subtype, NULL);
816 bind(not_subtype);
817 profile_typecheck_failed(Rtmp1);
818 }
820 // Separate these two to allow for delay slot in middle
821 // These are used to do a test and full jump to exception-throwing code.
823 // %%%%% Could possibly reoptimize this by testing to see if could use
824 // a single conditional branch (i.e. if span is small enough.
825 // If you go that route, than get rid of the split and give up
826 // on the delay-slot hack.
828 void InterpreterMacroAssembler::throw_if_not_1_icc( Condition ok_condition,
829 Label& ok ) {
830 assert_not_delayed();
831 br(ok_condition, true, pt, ok);
832 // DELAY SLOT
833 }
835 void InterpreterMacroAssembler::throw_if_not_1_xcc( Condition ok_condition,
836 Label& ok ) {
837 assert_not_delayed();
838 bp( ok_condition, true, Assembler::xcc, pt, ok);
839 // DELAY SLOT
840 }
842 void InterpreterMacroAssembler::throw_if_not_1_x( Condition ok_condition,
843 Label& ok ) {
844 assert_not_delayed();
845 brx(ok_condition, true, pt, ok);
846 // DELAY SLOT
847 }
849 void InterpreterMacroAssembler::throw_if_not_2( address throw_entry_point,
850 Register Rscratch,
851 Label& ok ) {
852 assert(throw_entry_point != NULL, "entry point must be generated by now");
853 AddressLiteral dest(throw_entry_point);
854 jump_to(dest, Rscratch);
855 delayed()->nop();
856 bind(ok);
857 }
860 // And if you cannot use the delay slot, here is a shorthand:
862 void InterpreterMacroAssembler::throw_if_not_icc( Condition ok_condition,
863 address throw_entry_point,
864 Register Rscratch ) {
865 Label ok;
866 if (ok_condition != never) {
867 throw_if_not_1_icc( ok_condition, ok);
868 delayed()->nop();
869 }
870 throw_if_not_2( throw_entry_point, Rscratch, ok);
871 }
872 void InterpreterMacroAssembler::throw_if_not_xcc( Condition ok_condition,
873 address throw_entry_point,
874 Register Rscratch ) {
875 Label ok;
876 if (ok_condition != never) {
877 throw_if_not_1_xcc( ok_condition, ok);
878 delayed()->nop();
879 }
880 throw_if_not_2( throw_entry_point, Rscratch, ok);
881 }
882 void InterpreterMacroAssembler::throw_if_not_x( Condition ok_condition,
883 address throw_entry_point,
884 Register Rscratch ) {
885 Label ok;
886 if (ok_condition != never) {
887 throw_if_not_1_x( ok_condition, ok);
888 delayed()->nop();
889 }
890 throw_if_not_2( throw_entry_point, Rscratch, ok);
891 }
893 // Check that index is in range for array, then shift index by index_shift, and put arrayOop + shifted_index into res
894 // Note: res is still shy of address by array offset into object.
896 void InterpreterMacroAssembler::index_check_without_pop(Register array, Register index, int index_shift, Register tmp, Register res) {
897 assert_not_delayed();
899 verify_oop(array);
900 #ifdef _LP64
901 // sign extend since tos (index) can be a 32bit value
902 sra(index, G0, index);
903 #endif // _LP64
905 // check array
906 Label ptr_ok;
907 tst(array);
908 throw_if_not_1_x( notZero, ptr_ok );
909 delayed()->ld( array, arrayOopDesc::length_offset_in_bytes(), tmp ); // check index
910 throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ptr_ok);
912 Label index_ok;
913 cmp(index, tmp);
914 throw_if_not_1_icc( lessUnsigned, index_ok );
915 if (index_shift > 0) delayed()->sll(index, index_shift, index);
916 else delayed()->add(array, index, res); // addr - const offset in index
917 // convention: move aberrant index into G3_scratch for exception message
918 mov(index, G3_scratch);
919 throw_if_not_2( Interpreter::_throw_ArrayIndexOutOfBoundsException_entry, G4_scratch, index_ok);
921 // add offset if didn't do it in delay slot
922 if (index_shift > 0) add(array, index, res); // addr - const offset in index
923 }
926 void InterpreterMacroAssembler::index_check(Register array, Register index, int index_shift, Register tmp, Register res) {
927 assert_not_delayed();
929 // pop array
930 pop_ptr(array);
932 // check array
933 index_check_without_pop(array, index, index_shift, tmp, res);
934 }
937 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) {
938 ld_ptr(Lmethod, in_bytes(methodOopDesc::constants_offset()), Rdst);
939 }
942 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) {
943 get_constant_pool(Rdst);
944 ld_ptr(Rdst, constantPoolOopDesc::cache_offset_in_bytes(), Rdst);
945 }
948 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) {
949 get_constant_pool(Rcpool);
950 ld_ptr(Rcpool, constantPoolOopDesc::tags_offset_in_bytes(), Rtags);
951 }
954 // unlock if synchronized method
955 //
956 // Unlock the receiver if this is a synchronized method.
957 // Unlock any Java monitors from syncronized blocks.
958 //
959 // If there are locked Java monitors
960 // If throw_monitor_exception
961 // throws IllegalMonitorStateException
962 // Else if install_monitor_exception
963 // installs IllegalMonitorStateException
964 // Else
965 // no error processing
966 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state,
967 bool throw_monitor_exception,
968 bool install_monitor_exception) {
969 Label unlocked, unlock, no_unlock;
971 // get the value of _do_not_unlock_if_synchronized into G1_scratch
972 const Address do_not_unlock_if_synchronized(G2_thread,
973 JavaThread::do_not_unlock_if_synchronized_offset());
974 ldbool(do_not_unlock_if_synchronized, G1_scratch);
975 stbool(G0, do_not_unlock_if_synchronized); // reset the flag
977 // check if synchronized method
978 const Address access_flags(Lmethod, methodOopDesc::access_flags_offset());
979 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
980 push(state); // save tos
981 ld(access_flags, G3_scratch); // Load access flags.
982 btst(JVM_ACC_SYNCHRONIZED, G3_scratch);
983 br(zero, false, pt, unlocked);
984 delayed()->nop();
986 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
987 // is set.
988 cmp_zero_and_br(Assembler::notZero, G1_scratch, no_unlock);
989 delayed()->nop();
991 // BasicObjectLock will be first in list, since this is a synchronized method. However, need
992 // to check that the object has not been unlocked by an explicit monitorexit bytecode.
994 //Intel: if (throw_monitor_exception) ... else ...
995 // Entry already unlocked, need to throw exception
996 //...
998 // pass top-most monitor elem
999 add( top_most_monitor(), O1 );
1001 ld_ptr(O1, BasicObjectLock::obj_offset_in_bytes(), G3_scratch);
1002 br_notnull_short(G3_scratch, pt, unlock);
1004 if (throw_monitor_exception) {
1005 // Entry already unlocked need to throw an exception
1006 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1007 should_not_reach_here();
1008 } else {
1009 // Monitor already unlocked during a stack unroll.
1010 // If requested, install an illegal_monitor_state_exception.
1011 // Continue with stack unrolling.
1012 if (install_monitor_exception) {
1013 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
1014 }
1015 ba_short(unlocked);
1016 }
1018 bind(unlock);
1020 unlock_object(O1);
1022 bind(unlocked);
1024 // I0, I1: Might contain return value
1026 // Check that all monitors are unlocked
1027 { Label loop, exception, entry, restart;
1029 Register Rmptr = O0;
1030 Register Rtemp = O1;
1031 Register Rlimit = Lmonitors;
1032 const jint delta = frame::interpreter_frame_monitor_size() * wordSize;
1033 assert( (delta & LongAlignmentMask) == 0,
1034 "sizeof BasicObjectLock must be even number of doublewords");
1036 #ifdef ASSERT
1037 add(top_most_monitor(), Rmptr, delta);
1038 { Label L;
1039 // ensure that Rmptr starts out above (or at) Rlimit
1040 cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L);
1041 stop("monitor stack has negative size");
1042 bind(L);
1043 }
1044 #endif
1045 bind(restart);
1046 ba(entry);
1047 delayed()->
1048 add(top_most_monitor(), Rmptr, delta); // points to current entry, starting with bottom-most entry
1050 // Entry is still locked, need to throw exception
1051 bind(exception);
1052 if (throw_monitor_exception) {
1053 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1054 should_not_reach_here();
1055 } else {
1056 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception.
1057 // Unlock does not block, so don't have to worry about the frame
1058 unlock_object(Rmptr);
1059 if (install_monitor_exception) {
1060 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
1061 }
1062 ba_short(restart);
1063 }
1065 bind(loop);
1066 cmp(Rtemp, G0); // check if current entry is used
1067 brx(Assembler::notEqual, false, pn, exception);
1068 delayed()->
1069 dec(Rmptr, delta); // otherwise advance to next entry
1070 #ifdef ASSERT
1071 { Label L;
1072 // ensure that Rmptr has not somehow stepped below Rlimit
1073 cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L);
1074 stop("ran off the end of the monitor stack");
1075 bind(L);
1076 }
1077 #endif
1078 bind(entry);
1079 cmp(Rmptr, Rlimit); // check if bottom reached
1080 brx(Assembler::notEqual, true, pn, loop); // if not at bottom then check this entry
1081 delayed()->
1082 ld_ptr(Rmptr, BasicObjectLock::obj_offset_in_bytes() - delta, Rtemp);
1083 }
1085 bind(no_unlock);
1086 pop(state);
1087 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
1088 }
1091 // remove activation
1092 //
1093 // Unlock the receiver if this is a synchronized method.
1094 // Unlock any Java monitors from syncronized blocks.
1095 // Remove the activation from the stack.
1096 //
1097 // If there are locked Java monitors
1098 // If throw_monitor_exception
1099 // throws IllegalMonitorStateException
1100 // Else if install_monitor_exception
1101 // installs IllegalMonitorStateException
1102 // Else
1103 // no error processing
1104 void InterpreterMacroAssembler::remove_activation(TosState state,
1105 bool throw_monitor_exception,
1106 bool install_monitor_exception) {
1108 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception);
1110 // save result (push state before jvmti call and pop it afterwards) and notify jvmti
1111 notify_method_exit(false, state, NotifyJVMTI);
1113 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
1114 verify_oop(Lmethod);
1115 verify_thread();
1117 // return tos
1118 assert(Otos_l1 == Otos_i, "adjust code below");
1119 switch (state) {
1120 #ifdef _LP64
1121 case ltos: mov(Otos_l, Otos_l->after_save()); break; // O0 -> I0
1122 #else
1123 case ltos: mov(Otos_l2, Otos_l2->after_save()); // fall through // O1 -> I1
1124 #endif
1125 case btos: // fall through
1126 case ctos:
1127 case stos: // fall through
1128 case atos: // fall through
1129 case itos: mov(Otos_l1, Otos_l1->after_save()); break; // O0 -> I0
1130 case ftos: // fall through
1131 case dtos: // fall through
1132 case vtos: /* nothing to do */ break;
1133 default : ShouldNotReachHere();
1134 }
1136 #if defined(COMPILER2) && !defined(_LP64)
1137 if (state == ltos) {
1138 // C2 expects long results in G1 we can't tell if we're returning to interpreted
1139 // or compiled so just be safe use G1 and O0/O1
1141 // Shift bits into high (msb) of G1
1142 sllx(Otos_l1->after_save(), 32, G1);
1143 // Zero extend low bits
1144 srl (Otos_l2->after_save(), 0, Otos_l2->after_save());
1145 or3 (Otos_l2->after_save(), G1, G1);
1146 }
1147 #endif /* COMPILER2 */
1149 }
1150 #endif /* CC_INTERP */
1153 // Lock object
1154 //
1155 // Argument - lock_reg points to the BasicObjectLock to be used for locking,
1156 // it must be initialized with the object to lock
1157 void InterpreterMacroAssembler::lock_object(Register lock_reg, Register Object) {
1158 if (UseHeavyMonitors) {
1159 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg);
1160 }
1161 else {
1162 Register obj_reg = Object;
1163 Register mark_reg = G4_scratch;
1164 Register temp_reg = G1_scratch;
1165 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes());
1166 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes());
1167 Label done;
1169 Label slow_case;
1171 assert_different_registers(lock_reg, obj_reg, mark_reg, temp_reg);
1173 // load markOop from object into mark_reg
1174 ld_ptr(mark_addr, mark_reg);
1176 if (UseBiasedLocking) {
1177 biased_locking_enter(obj_reg, mark_reg, temp_reg, done, &slow_case);
1178 }
1180 // get the address of basicLock on stack that will be stored in the object
1181 // we need a temporary register here as we do not want to clobber lock_reg
1182 // (cas clobbers the destination register)
1183 mov(lock_reg, temp_reg);
1184 // set mark reg to be (markOop of object | UNLOCK_VALUE)
1185 or3(mark_reg, markOopDesc::unlocked_value, mark_reg);
1186 // initialize the box (Must happen before we update the object mark!)
1187 st_ptr(mark_reg, lock_addr, BasicLock::displaced_header_offset_in_bytes());
1188 // compare and exchange object_addr, markOop | 1, stack address of basicLock
1189 assert(mark_addr.disp() == 0, "cas must take a zero displacement");
1190 casx_under_lock(mark_addr.base(), mark_reg, temp_reg,
1191 (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr());
1193 // if the compare and exchange succeeded we are done (we saw an unlocked object)
1194 cmp_and_brx_short(mark_reg, temp_reg, Assembler::equal, Assembler::pt, done);
1196 // We did not see an unlocked object so try the fast recursive case
1198 // Check if owner is self by comparing the value in the markOop of object
1199 // with the stack pointer
1200 sub(temp_reg, SP, temp_reg);
1201 #ifdef _LP64
1202 sub(temp_reg, STACK_BIAS, temp_reg);
1203 #endif
1204 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
1206 // Composite "andcc" test:
1207 // (a) %sp -vs- markword proximity check, and,
1208 // (b) verify mark word LSBs == 0 (Stack-locked).
1209 //
1210 // FFFFF003/FFFFFFFFFFFF003 is (markOopDesc::lock_mask_in_place | -os::vm_page_size())
1211 // Note that the page size used for %sp proximity testing is arbitrary and is
1212 // unrelated to the actual MMU page size. We use a 'logical' page size of
1213 // 4096 bytes. F..FFF003 is designed to fit conveniently in the SIMM13 immediate
1214 // field of the andcc instruction.
1215 andcc (temp_reg, 0xFFFFF003, G0) ;
1217 // if condition is true we are done and hence we can store 0 in the displaced
1218 // header indicating it is a recursive lock and be done
1219 brx(Assembler::zero, true, Assembler::pt, done);
1220 delayed()->st_ptr(G0, lock_addr, BasicLock::displaced_header_offset_in_bytes());
1222 // none of the above fast optimizations worked so we have to get into the
1223 // slow case of monitor enter
1224 bind(slow_case);
1225 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg);
1227 bind(done);
1228 }
1229 }
1231 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
1232 //
1233 // Argument - lock_reg points to the BasicObjectLock for lock
1234 // Throw IllegalMonitorException if object is not locked by current thread
1235 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1236 if (UseHeavyMonitors) {
1237 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1238 } else {
1239 Register obj_reg = G3_scratch;
1240 Register mark_reg = G4_scratch;
1241 Register displaced_header_reg = G1_scratch;
1242 Address lockobj_addr(lock_reg, BasicObjectLock::obj_offset_in_bytes());
1243 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes());
1244 Label done;
1246 if (UseBiasedLocking) {
1247 // load the object out of the BasicObjectLock
1248 ld_ptr(lockobj_addr, obj_reg);
1249 biased_locking_exit(mark_addr, mark_reg, done, true);
1250 st_ptr(G0, lockobj_addr); // free entry
1251 }
1253 // Test first if we are in the fast recursive case
1254 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes() + BasicLock::displaced_header_offset_in_bytes());
1255 ld_ptr(lock_addr, displaced_header_reg);
1256 br_null(displaced_header_reg, true, Assembler::pn, done);
1257 delayed()->st_ptr(G0, lockobj_addr); // free entry
1259 // See if it is still a light weight lock, if so we just unlock
1260 // the object and we are done
1262 if (!UseBiasedLocking) {
1263 // load the object out of the BasicObjectLock
1264 ld_ptr(lockobj_addr, obj_reg);
1265 }
1267 // we have the displaced header in displaced_header_reg
1268 // we expect to see the stack address of the basicLock in case the
1269 // lock is still a light weight lock (lock_reg)
1270 assert(mark_addr.disp() == 0, "cas must take a zero displacement");
1271 casx_under_lock(mark_addr.base(), lock_reg, displaced_header_reg,
1272 (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr());
1273 cmp(lock_reg, displaced_header_reg);
1274 brx(Assembler::equal, true, Assembler::pn, done);
1275 delayed()->st_ptr(G0, lockobj_addr); // free entry
1277 // The lock has been converted into a heavy lock and hence
1278 // we need to get into the slow case
1280 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1282 bind(done);
1283 }
1284 }
1286 #ifndef CC_INTERP
1288 // Get the method data pointer from the methodOop and set the
1289 // specified register to its value.
1291 void InterpreterMacroAssembler::set_method_data_pointer() {
1292 assert(ProfileInterpreter, "must be profiling interpreter");
1293 Label get_continue;
1295 ld_ptr(Lmethod, in_bytes(methodOopDesc::method_data_offset()), ImethodDataPtr);
1296 test_method_data_pointer(get_continue);
1297 add(ImethodDataPtr, in_bytes(methodDataOopDesc::data_offset()), ImethodDataPtr);
1298 bind(get_continue);
1299 }
1301 // Set the method data pointer for the current bcp.
1303 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1304 assert(ProfileInterpreter, "must be profiling interpreter");
1305 Label zero_continue;
1307 // Test MDO to avoid the call if it is NULL.
1308 ld_ptr(Lmethod, in_bytes(methodOopDesc::method_data_offset()), ImethodDataPtr);
1309 test_method_data_pointer(zero_continue);
1310 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), Lmethod, Lbcp);
1311 add(ImethodDataPtr, in_bytes(methodDataOopDesc::data_offset()), ImethodDataPtr);
1312 add(ImethodDataPtr, O0, ImethodDataPtr);
1313 bind(zero_continue);
1314 }
1316 // Test ImethodDataPtr. If it is null, continue at the specified label
1318 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) {
1319 assert(ProfileInterpreter, "must be profiling interpreter");
1320 br_null_short(ImethodDataPtr, Assembler::pn, zero_continue);
1321 }
1323 void InterpreterMacroAssembler::verify_method_data_pointer() {
1324 assert(ProfileInterpreter, "must be profiling interpreter");
1325 #ifdef ASSERT
1326 Label verify_continue;
1327 test_method_data_pointer(verify_continue);
1329 // If the mdp is valid, it will point to a DataLayout header which is
1330 // consistent with the bcp. The converse is highly probable also.
1331 lduh(ImethodDataPtr, in_bytes(DataLayout::bci_offset()), G3_scratch);
1332 ld_ptr(Lmethod, methodOopDesc::const_offset(), O5);
1333 add(G3_scratch, in_bytes(constMethodOopDesc::codes_offset()), G3_scratch);
1334 add(G3_scratch, O5, G3_scratch);
1335 cmp(Lbcp, G3_scratch);
1336 brx(Assembler::equal, false, Assembler::pt, verify_continue);
1338 Register temp_reg = O5;
1339 delayed()->mov(ImethodDataPtr, temp_reg);
1340 // %%% should use call_VM_leaf here?
1341 //call_VM_leaf(noreg, ..., Lmethod, Lbcp, ImethodDataPtr);
1342 save_frame_and_mov(sizeof(jdouble) / wordSize, Lmethod, O0, Lbcp, O1);
1343 Address d_save(FP, -sizeof(jdouble) + STACK_BIAS);
1344 stf(FloatRegisterImpl::D, Ftos_d, d_save);
1345 mov(temp_reg->after_save(), O2);
1346 save_thread(L7_thread_cache);
1347 call(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), relocInfo::none);
1348 delayed()->nop();
1349 restore_thread(L7_thread_cache);
1350 ldf(FloatRegisterImpl::D, d_save, Ftos_d);
1351 restore();
1352 bind(verify_continue);
1353 #endif // ASSERT
1354 }
1356 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count,
1357 Register Rtmp,
1358 Label &profile_continue) {
1359 assert(ProfileInterpreter, "must be profiling interpreter");
1360 // Control will flow to "profile_continue" if the counter is less than the
1361 // limit or if we call profile_method()
1363 Label done;
1365 // if no method data exists, and the counter is high enough, make one
1366 br_notnull_short(ImethodDataPtr, Assembler::pn, done);
1368 // Test to see if we should create a method data oop
1369 AddressLiteral profile_limit((address) &InvocationCounter::InterpreterProfileLimit);
1370 sethi(profile_limit, Rtmp);
1371 ld(Rtmp, profile_limit.low10(), Rtmp);
1372 cmp_and_br_short(invocation_count, Rtmp, Assembler::lessUnsigned, Assembler::pn, profile_continue);
1374 // Build it now.
1375 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1376 set_method_data_pointer_for_bcp();
1377 ba_short(profile_continue);
1378 bind(done);
1379 }
1381 // Store a value at some constant offset from the method data pointer.
1383 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) {
1384 assert(ProfileInterpreter, "must be profiling interpreter");
1385 st_ptr(value, ImethodDataPtr, constant);
1386 }
1388 void InterpreterMacroAssembler::increment_mdp_data_at(Address counter,
1389 Register bumped_count,
1390 bool decrement) {
1391 assert(ProfileInterpreter, "must be profiling interpreter");
1393 // Load the counter.
1394 ld_ptr(counter, bumped_count);
1396 if (decrement) {
1397 // Decrement the register. Set condition codes.
1398 subcc(bumped_count, DataLayout::counter_increment, bumped_count);
1400 // If the decrement causes the counter to overflow, stay negative
1401 Label L;
1402 brx(Assembler::negative, true, Assembler::pn, L);
1404 // Store the decremented counter, if it is still negative.
1405 delayed()->st_ptr(bumped_count, counter);
1406 bind(L);
1407 } else {
1408 // Increment the register. Set carry flag.
1409 addcc(bumped_count, DataLayout::counter_increment, bumped_count);
1411 // If the increment causes the counter to overflow, pull back by 1.
1412 assert(DataLayout::counter_increment == 1, "subc works");
1413 subc(bumped_count, G0, bumped_count);
1415 // Store the incremented counter.
1416 st_ptr(bumped_count, counter);
1417 }
1418 }
1420 // Increment the value at some constant offset from the method data pointer.
1422 void InterpreterMacroAssembler::increment_mdp_data_at(int constant,
1423 Register bumped_count,
1424 bool decrement) {
1425 // Locate the counter at a fixed offset from the mdp:
1426 Address counter(ImethodDataPtr, constant);
1427 increment_mdp_data_at(counter, bumped_count, decrement);
1428 }
1430 // Increment the value at some non-fixed (reg + constant) offset from
1431 // the method data pointer.
1433 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg,
1434 int constant,
1435 Register bumped_count,
1436 Register scratch2,
1437 bool decrement) {
1438 // Add the constant to reg to get the offset.
1439 add(ImethodDataPtr, reg, scratch2);
1440 Address counter(scratch2, constant);
1441 increment_mdp_data_at(counter, bumped_count, decrement);
1442 }
1444 // Set a flag value at the current method data pointer position.
1445 // Updates a single byte of the header, to avoid races with other header bits.
1447 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant,
1448 Register scratch) {
1449 assert(ProfileInterpreter, "must be profiling interpreter");
1450 // Load the data header
1451 ldub(ImethodDataPtr, in_bytes(DataLayout::flags_offset()), scratch);
1453 // Set the flag
1454 or3(scratch, flag_constant, scratch);
1456 // Store the modified header.
1457 stb(scratch, ImethodDataPtr, in_bytes(DataLayout::flags_offset()));
1458 }
1460 // Test the location at some offset from the method data pointer.
1461 // If it is not equal to value, branch to the not_equal_continue Label.
1462 // Set condition codes to match the nullness of the loaded value.
1464 void InterpreterMacroAssembler::test_mdp_data_at(int offset,
1465 Register value,
1466 Label& not_equal_continue,
1467 Register scratch) {
1468 assert(ProfileInterpreter, "must be profiling interpreter");
1469 ld_ptr(ImethodDataPtr, offset, scratch);
1470 cmp(value, scratch);
1471 brx(Assembler::notEqual, false, Assembler::pn, not_equal_continue);
1472 delayed()->tst(scratch);
1473 }
1475 // Update the method data pointer by the displacement located at some fixed
1476 // offset from the method data pointer.
1478 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp,
1479 Register scratch) {
1480 assert(ProfileInterpreter, "must be profiling interpreter");
1481 ld_ptr(ImethodDataPtr, offset_of_disp, scratch);
1482 add(ImethodDataPtr, scratch, ImethodDataPtr);
1483 }
1485 // Update the method data pointer by the displacement located at the
1486 // offset (reg + offset_of_disp).
1488 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg,
1489 int offset_of_disp,
1490 Register scratch) {
1491 assert(ProfileInterpreter, "must be profiling interpreter");
1492 add(reg, offset_of_disp, scratch);
1493 ld_ptr(ImethodDataPtr, scratch, scratch);
1494 add(ImethodDataPtr, scratch, ImethodDataPtr);
1495 }
1497 // Update the method data pointer by a simple constant displacement.
1499 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) {
1500 assert(ProfileInterpreter, "must be profiling interpreter");
1501 add(ImethodDataPtr, constant, ImethodDataPtr);
1502 }
1504 // Update the method data pointer for a _ret bytecode whose target
1505 // was not among our cached targets.
1507 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state,
1508 Register return_bci) {
1509 assert(ProfileInterpreter, "must be profiling interpreter");
1510 push(state);
1511 st_ptr(return_bci, l_tmp); // protect return_bci, in case it is volatile
1512 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci);
1513 ld_ptr(l_tmp, return_bci);
1514 pop(state);
1515 }
1517 // Count a taken branch in the bytecodes.
1519 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) {
1520 if (ProfileInterpreter) {
1521 Label profile_continue;
1523 // If no method data exists, go to profile_continue.
1524 test_method_data_pointer(profile_continue);
1526 // We are taking a branch. Increment the taken count.
1527 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), bumped_count);
1529 // The method data pointer needs to be updated to reflect the new target.
1530 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch);
1531 bind (profile_continue);
1532 }
1533 }
1536 // Count a not-taken branch in the bytecodes.
1538 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch) {
1539 if (ProfileInterpreter) {
1540 Label profile_continue;
1542 // If no method data exists, go to profile_continue.
1543 test_method_data_pointer(profile_continue);
1545 // We are taking a branch. Increment the not taken count.
1546 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch);
1548 // The method data pointer needs to be updated to correspond to the
1549 // next bytecode.
1550 update_mdp_by_constant(in_bytes(BranchData::branch_data_size()));
1551 bind (profile_continue);
1552 }
1553 }
1556 // Count a non-virtual call in the bytecodes.
1558 void InterpreterMacroAssembler::profile_call(Register scratch) {
1559 if (ProfileInterpreter) {
1560 Label profile_continue;
1562 // If no method data exists, go to profile_continue.
1563 test_method_data_pointer(profile_continue);
1565 // We are making a call. Increment the count.
1566 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1568 // The method data pointer needs to be updated to reflect the new target.
1569 update_mdp_by_constant(in_bytes(CounterData::counter_data_size()));
1570 bind (profile_continue);
1571 }
1572 }
1575 // Count a final call in the bytecodes.
1577 void InterpreterMacroAssembler::profile_final_call(Register scratch) {
1578 if (ProfileInterpreter) {
1579 Label profile_continue;
1581 // If no method data exists, go to profile_continue.
1582 test_method_data_pointer(profile_continue);
1584 // We are making a call. Increment the count.
1585 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1587 // The method data pointer needs to be updated to reflect the new target.
1588 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1589 bind (profile_continue);
1590 }
1591 }
1594 // Count a virtual call in the bytecodes.
1596 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1597 Register scratch,
1598 bool receiver_can_be_null) {
1599 if (ProfileInterpreter) {
1600 Label profile_continue;
1602 // If no method data exists, go to profile_continue.
1603 test_method_data_pointer(profile_continue);
1606 Label skip_receiver_profile;
1607 if (receiver_can_be_null) {
1608 Label not_null;
1609 br_notnull_short(receiver, Assembler::pt, not_null);
1610 // We are making a call. Increment the count for null receiver.
1611 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1612 ba_short(skip_receiver_profile);
1613 bind(not_null);
1614 }
1616 // Record the receiver type.
1617 record_klass_in_profile(receiver, scratch, true);
1618 bind(skip_receiver_profile);
1620 // The method data pointer needs to be updated to reflect the new target.
1621 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1622 bind (profile_continue);
1623 }
1624 }
1626 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1627 Register receiver, Register scratch,
1628 int start_row, Label& done, bool is_virtual_call) {
1629 if (TypeProfileWidth == 0) {
1630 if (is_virtual_call) {
1631 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1632 }
1633 return;
1634 }
1636 int last_row = VirtualCallData::row_limit() - 1;
1637 assert(start_row <= last_row, "must be work left to do");
1638 // Test this row for both the receiver and for null.
1639 // Take any of three different outcomes:
1640 // 1. found receiver => increment count and goto done
1641 // 2. found null => keep looking for case 1, maybe allocate this cell
1642 // 3. found something else => keep looking for cases 1 and 2
1643 // Case 3 is handled by a recursive call.
1644 for (int row = start_row; row <= last_row; row++) {
1645 Label next_test;
1646 bool test_for_null_also = (row == start_row);
1648 // See if the receiver is receiver[n].
1649 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1650 test_mdp_data_at(recvr_offset, receiver, next_test, scratch);
1651 // delayed()->tst(scratch);
1653 // The receiver is receiver[n]. Increment count[n].
1654 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1655 increment_mdp_data_at(count_offset, scratch);
1656 ba_short(done);
1657 bind(next_test);
1659 if (test_for_null_also) {
1660 Label found_null;
1661 // Failed the equality check on receiver[n]... Test for null.
1662 if (start_row == last_row) {
1663 // The only thing left to do is handle the null case.
1664 if (is_virtual_call) {
1665 brx(Assembler::zero, false, Assembler::pn, found_null);
1666 delayed()->nop();
1667 // Receiver did not match any saved receiver and there is no empty row for it.
1668 // Increment total counter to indicate polymorphic case.
1669 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1670 ba_short(done);
1671 bind(found_null);
1672 } else {
1673 brx(Assembler::notZero, false, Assembler::pt, done);
1674 delayed()->nop();
1675 }
1676 break;
1677 }
1678 // Since null is rare, make it be the branch-taken case.
1679 brx(Assembler::zero, false, Assembler::pn, found_null);
1680 delayed()->nop();
1682 // Put all the "Case 3" tests here.
1683 record_klass_in_profile_helper(receiver, scratch, start_row + 1, done, is_virtual_call);
1685 // Found a null. Keep searching for a matching receiver,
1686 // but remember that this is an empty (unused) slot.
1687 bind(found_null);
1688 }
1689 }
1691 // In the fall-through case, we found no matching receiver, but we
1692 // observed the receiver[start_row] is NULL.
1694 // Fill in the receiver field and increment the count.
1695 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1696 set_mdp_data_at(recvr_offset, receiver);
1697 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1698 mov(DataLayout::counter_increment, scratch);
1699 set_mdp_data_at(count_offset, scratch);
1700 if (start_row > 0) {
1701 ba_short(done);
1702 }
1703 }
1705 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1706 Register scratch, bool is_virtual_call) {
1707 assert(ProfileInterpreter, "must be profiling");
1708 Label done;
1710 record_klass_in_profile_helper(receiver, scratch, 0, done, is_virtual_call);
1712 bind (done);
1713 }
1716 // Count a ret in the bytecodes.
1718 void InterpreterMacroAssembler::profile_ret(TosState state,
1719 Register return_bci,
1720 Register scratch) {
1721 if (ProfileInterpreter) {
1722 Label profile_continue;
1723 uint row;
1725 // If no method data exists, go to profile_continue.
1726 test_method_data_pointer(profile_continue);
1728 // Update the total ret count.
1729 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1731 for (row = 0; row < RetData::row_limit(); row++) {
1732 Label next_test;
1734 // See if return_bci is equal to bci[n]:
1735 test_mdp_data_at(in_bytes(RetData::bci_offset(row)),
1736 return_bci, next_test, scratch);
1738 // return_bci is equal to bci[n]. Increment the count.
1739 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch);
1741 // The method data pointer needs to be updated to reflect the new target.
1742 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch);
1743 ba_short(profile_continue);
1744 bind(next_test);
1745 }
1747 update_mdp_for_ret(state, return_bci);
1749 bind (profile_continue);
1750 }
1751 }
1753 // Profile an unexpected null in the bytecodes.
1754 void InterpreterMacroAssembler::profile_null_seen(Register scratch) {
1755 if (ProfileInterpreter) {
1756 Label profile_continue;
1758 // If no method data exists, go to profile_continue.
1759 test_method_data_pointer(profile_continue);
1761 set_mdp_flag_at(BitData::null_seen_byte_constant(), scratch);
1763 // The method data pointer needs to be updated.
1764 int mdp_delta = in_bytes(BitData::bit_data_size());
1765 if (TypeProfileCasts) {
1766 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1767 }
1768 update_mdp_by_constant(mdp_delta);
1770 bind (profile_continue);
1771 }
1772 }
1774 void InterpreterMacroAssembler::profile_typecheck(Register klass,
1775 Register scratch) {
1776 if (ProfileInterpreter) {
1777 Label profile_continue;
1779 // If no method data exists, go to profile_continue.
1780 test_method_data_pointer(profile_continue);
1782 int mdp_delta = in_bytes(BitData::bit_data_size());
1783 if (TypeProfileCasts) {
1784 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1786 // Record the object type.
1787 record_klass_in_profile(klass, scratch, false);
1788 }
1790 // The method data pointer needs to be updated.
1791 update_mdp_by_constant(mdp_delta);
1793 bind (profile_continue);
1794 }
1795 }
1797 void InterpreterMacroAssembler::profile_typecheck_failed(Register scratch) {
1798 if (ProfileInterpreter && TypeProfileCasts) {
1799 Label profile_continue;
1801 // If no method data exists, go to profile_continue.
1802 test_method_data_pointer(profile_continue);
1804 int count_offset = in_bytes(CounterData::count_offset());
1805 // Back up the address, since we have already bumped the mdp.
1806 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1808 // *Decrement* the counter. We expect to see zero or small negatives.
1809 increment_mdp_data_at(count_offset, scratch, true);
1811 bind (profile_continue);
1812 }
1813 }
1815 // Count the default case of a switch construct.
1817 void InterpreterMacroAssembler::profile_switch_default(Register scratch) {
1818 if (ProfileInterpreter) {
1819 Label profile_continue;
1821 // If no method data exists, go to profile_continue.
1822 test_method_data_pointer(profile_continue);
1824 // Update the default case count
1825 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()),
1826 scratch);
1828 // The method data pointer needs to be updated.
1829 update_mdp_by_offset(
1830 in_bytes(MultiBranchData::default_displacement_offset()),
1831 scratch);
1833 bind (profile_continue);
1834 }
1835 }
1837 // Count the index'th case of a switch construct.
1839 void InterpreterMacroAssembler::profile_switch_case(Register index,
1840 Register scratch,
1841 Register scratch2,
1842 Register scratch3) {
1843 if (ProfileInterpreter) {
1844 Label profile_continue;
1846 // If no method data exists, go to profile_continue.
1847 test_method_data_pointer(profile_continue);
1849 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes()
1850 set(in_bytes(MultiBranchData::per_case_size()), scratch);
1851 smul(index, scratch, scratch);
1852 add(scratch, in_bytes(MultiBranchData::case_array_offset()), scratch);
1854 // Update the case count
1855 increment_mdp_data_at(scratch,
1856 in_bytes(MultiBranchData::relative_count_offset()),
1857 scratch2,
1858 scratch3);
1860 // The method data pointer needs to be updated.
1861 update_mdp_by_offset(scratch,
1862 in_bytes(MultiBranchData::relative_displacement_offset()),
1863 scratch2);
1865 bind (profile_continue);
1866 }
1867 }
1869 // add a InterpMonitorElem to stack (see frame_sparc.hpp)
1871 void InterpreterMacroAssembler::add_monitor_to_stack( bool stack_is_empty,
1872 Register Rtemp,
1873 Register Rtemp2 ) {
1875 Register Rlimit = Lmonitors;
1876 const jint delta = frame::interpreter_frame_monitor_size() * wordSize;
1877 assert( (delta & LongAlignmentMask) == 0,
1878 "sizeof BasicObjectLock must be even number of doublewords");
1880 sub( SP, delta, SP);
1881 sub( Lesp, delta, Lesp);
1882 sub( Lmonitors, delta, Lmonitors);
1884 if (!stack_is_empty) {
1886 // must copy stack contents down
1888 Label start_copying, next;
1890 // untested("monitor stack expansion");
1891 compute_stack_base(Rtemp);
1892 ba(start_copying);
1893 delayed()->cmp(Rtemp, Rlimit); // done? duplicated below
1895 // note: must copy from low memory upwards
1896 // On entry to loop,
1897 // Rtemp points to new base of stack, Lesp points to new end of stack (1 past TOS)
1898 // Loop mutates Rtemp
1900 bind( next);
1902 st_ptr(Rtemp2, Rtemp, 0);
1903 inc(Rtemp, wordSize);
1904 cmp(Rtemp, Rlimit); // are we done? (duplicated above)
1906 bind( start_copying );
1908 brx( notEqual, true, pn, next );
1909 delayed()->ld_ptr( Rtemp, delta, Rtemp2 );
1911 // done copying stack
1912 }
1913 }
1915 // Locals
1916 void InterpreterMacroAssembler::access_local_ptr( Register index, Register dst ) {
1917 assert_not_delayed();
1918 sll(index, Interpreter::logStackElementSize, index);
1919 sub(Llocals, index, index);
1920 ld_ptr(index, 0, dst);
1921 // Note: index must hold the effective address--the iinc template uses it
1922 }
1924 // Just like access_local_ptr but the tag is a returnAddress
1925 void InterpreterMacroAssembler::access_local_returnAddress(Register index,
1926 Register dst ) {
1927 assert_not_delayed();
1928 sll(index, Interpreter::logStackElementSize, index);
1929 sub(Llocals, index, index);
1930 ld_ptr(index, 0, dst);
1931 }
1933 void InterpreterMacroAssembler::access_local_int( Register index, Register dst ) {
1934 assert_not_delayed();
1935 sll(index, Interpreter::logStackElementSize, index);
1936 sub(Llocals, index, index);
1937 ld(index, 0, dst);
1938 // Note: index must hold the effective address--the iinc template uses it
1939 }
1942 void InterpreterMacroAssembler::access_local_long( Register index, Register dst ) {
1943 assert_not_delayed();
1944 sll(index, Interpreter::logStackElementSize, index);
1945 sub(Llocals, index, index);
1946 // First half stored at index n+1 (which grows down from Llocals[n])
1947 load_unaligned_long(index, Interpreter::local_offset_in_bytes(1), dst);
1948 }
1951 void InterpreterMacroAssembler::access_local_float( Register index, FloatRegister dst ) {
1952 assert_not_delayed();
1953 sll(index, Interpreter::logStackElementSize, index);
1954 sub(Llocals, index, index);
1955 ldf(FloatRegisterImpl::S, index, 0, dst);
1956 }
1959 void InterpreterMacroAssembler::access_local_double( Register index, FloatRegister dst ) {
1960 assert_not_delayed();
1961 sll(index, Interpreter::logStackElementSize, index);
1962 sub(Llocals, index, index);
1963 load_unaligned_double(index, Interpreter::local_offset_in_bytes(1), dst);
1964 }
1967 #ifdef ASSERT
1968 void InterpreterMacroAssembler::check_for_regarea_stomp(Register Rindex, int offset, Register Rlimit, Register Rscratch, Register Rscratch1) {
1969 Label L;
1971 assert(Rindex != Rscratch, "Registers cannot be same");
1972 assert(Rindex != Rscratch1, "Registers cannot be same");
1973 assert(Rlimit != Rscratch, "Registers cannot be same");
1974 assert(Rlimit != Rscratch1, "Registers cannot be same");
1975 assert(Rscratch1 != Rscratch, "Registers cannot be same");
1977 // untested("reg area corruption");
1978 add(Rindex, offset, Rscratch);
1979 add(Rlimit, 64 + STACK_BIAS, Rscratch1);
1980 cmp_and_brx_short(Rscratch, Rscratch1, Assembler::greaterEqualUnsigned, pn, L);
1981 stop("regsave area is being clobbered");
1982 bind(L);
1983 }
1984 #endif // ASSERT
1987 void InterpreterMacroAssembler::store_local_int( Register index, Register src ) {
1988 assert_not_delayed();
1989 sll(index, Interpreter::logStackElementSize, index);
1990 sub(Llocals, index, index);
1991 debug_only(check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);)
1992 st(src, index, 0);
1993 }
1995 void InterpreterMacroAssembler::store_local_ptr( Register index, Register src ) {
1996 assert_not_delayed();
1997 sll(index, Interpreter::logStackElementSize, index);
1998 sub(Llocals, index, index);
1999 #ifdef ASSERT
2000 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);
2001 #endif
2002 st_ptr(src, index, 0);
2003 }
2007 void InterpreterMacroAssembler::store_local_ptr( int n, Register src ) {
2008 st_ptr(src, Llocals, Interpreter::local_offset_in_bytes(n));
2009 }
2011 void InterpreterMacroAssembler::store_local_long( Register index, Register src ) {
2012 assert_not_delayed();
2013 sll(index, Interpreter::logStackElementSize, index);
2014 sub(Llocals, index, index);
2015 #ifdef ASSERT
2016 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch);
2017 #endif
2018 store_unaligned_long(src, index, Interpreter::local_offset_in_bytes(1)); // which is n+1
2019 }
2022 void InterpreterMacroAssembler::store_local_float( Register index, FloatRegister src ) {
2023 assert_not_delayed();
2024 sll(index, Interpreter::logStackElementSize, index);
2025 sub(Llocals, index, index);
2026 #ifdef ASSERT
2027 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);
2028 #endif
2029 stf(FloatRegisterImpl::S, src, index, 0);
2030 }
2033 void InterpreterMacroAssembler::store_local_double( Register index, FloatRegister src ) {
2034 assert_not_delayed();
2035 sll(index, Interpreter::logStackElementSize, index);
2036 sub(Llocals, index, index);
2037 #ifdef ASSERT
2038 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch);
2039 #endif
2040 store_unaligned_double(src, index, Interpreter::local_offset_in_bytes(1));
2041 }
2044 int InterpreterMacroAssembler::top_most_monitor_byte_offset() {
2045 const jint delta = frame::interpreter_frame_monitor_size() * wordSize;
2046 int rounded_vm_local_words = ::round_to(frame::interpreter_frame_vm_local_words, WordsPerLong);
2047 return ((-rounded_vm_local_words * wordSize) - delta ) + STACK_BIAS;
2048 }
2051 Address InterpreterMacroAssembler::top_most_monitor() {
2052 return Address(FP, top_most_monitor_byte_offset());
2053 }
2056 void InterpreterMacroAssembler::compute_stack_base( Register Rdest ) {
2057 add( Lesp, wordSize, Rdest );
2058 }
2060 #endif /* CC_INTERP */
2062 void InterpreterMacroAssembler::increment_invocation_counter( Register Rtmp, Register Rtmp2 ) {
2063 assert(UseCompiler, "incrementing must be useful");
2064 #ifdef CC_INTERP
2065 Address inv_counter(G5_method, methodOopDesc::invocation_counter_offset() +
2066 InvocationCounter::counter_offset());
2067 Address be_counter (G5_method, methodOopDesc::backedge_counter_offset() +
2068 InvocationCounter::counter_offset());
2069 #else
2070 Address inv_counter(Lmethod, methodOopDesc::invocation_counter_offset() +
2071 InvocationCounter::counter_offset());
2072 Address be_counter (Lmethod, methodOopDesc::backedge_counter_offset() +
2073 InvocationCounter::counter_offset());
2074 #endif /* CC_INTERP */
2075 int delta = InvocationCounter::count_increment;
2077 // Load each counter in a register
2078 ld( inv_counter, Rtmp );
2079 ld( be_counter, Rtmp2 );
2081 assert( is_simm13( delta ), " delta too large.");
2083 // Add the delta to the invocation counter and store the result
2084 add( Rtmp, delta, Rtmp );
2086 // Mask the backedge counter
2087 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 );
2089 // Store value
2090 st( Rtmp, inv_counter);
2092 // Add invocation counter + backedge counter
2093 add( Rtmp, Rtmp2, Rtmp);
2095 // Note that this macro must leave the backedge_count + invocation_count in Rtmp!
2096 }
2098 void InterpreterMacroAssembler::increment_backedge_counter( Register Rtmp, Register Rtmp2 ) {
2099 assert(UseCompiler, "incrementing must be useful");
2100 #ifdef CC_INTERP
2101 Address be_counter (G5_method, methodOopDesc::backedge_counter_offset() +
2102 InvocationCounter::counter_offset());
2103 Address inv_counter(G5_method, methodOopDesc::invocation_counter_offset() +
2104 InvocationCounter::counter_offset());
2105 #else
2106 Address be_counter (Lmethod, methodOopDesc::backedge_counter_offset() +
2107 InvocationCounter::counter_offset());
2108 Address inv_counter(Lmethod, methodOopDesc::invocation_counter_offset() +
2109 InvocationCounter::counter_offset());
2110 #endif /* CC_INTERP */
2111 int delta = InvocationCounter::count_increment;
2112 // Load each counter in a register
2113 ld( be_counter, Rtmp );
2114 ld( inv_counter, Rtmp2 );
2116 // Add the delta to the backedge counter
2117 add( Rtmp, delta, Rtmp );
2119 // Mask the invocation counter, add to backedge counter
2120 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 );
2122 // and store the result to memory
2123 st( Rtmp, be_counter );
2125 // Add backedge + invocation counter
2126 add( Rtmp, Rtmp2, Rtmp );
2128 // Note that this macro must leave backedge_count + invocation_count in Rtmp!
2129 }
2131 #ifndef CC_INTERP
2132 void InterpreterMacroAssembler::test_backedge_count_for_osr( Register backedge_count,
2133 Register branch_bcp,
2134 Register Rtmp ) {
2135 Label did_not_overflow;
2136 Label overflow_with_error;
2137 assert_different_registers(backedge_count, Rtmp, branch_bcp);
2138 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr");
2140 AddressLiteral limit(&InvocationCounter::InterpreterBackwardBranchLimit);
2141 load_contents(limit, Rtmp);
2142 cmp_and_br_short(backedge_count, Rtmp, Assembler::lessUnsigned, Assembler::pt, did_not_overflow);
2144 // When ProfileInterpreter is on, the backedge_count comes from the
2145 // methodDataOop, which value does not get reset on the call to
2146 // frequency_counter_overflow(). To avoid excessive calls to the overflow
2147 // routine while the method is being compiled, add a second test to make sure
2148 // the overflow function is called only once every overflow_frequency.
2149 if (ProfileInterpreter) {
2150 const int overflow_frequency = 1024;
2151 andcc(backedge_count, overflow_frequency-1, Rtmp);
2152 brx(Assembler::notZero, false, Assembler::pt, did_not_overflow);
2153 delayed()->nop();
2154 }
2156 // overflow in loop, pass branch bytecode
2157 set(6,Rtmp);
2158 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, Rtmp);
2160 // Was an OSR adapter generated?
2161 // O0 = osr nmethod
2162 br_null_short(O0, Assembler::pn, overflow_with_error);
2164 // Has the nmethod been invalidated already?
2165 ld(O0, nmethod::entry_bci_offset(), O2);
2166 cmp_and_br_short(O2, InvalidOSREntryBci, Assembler::equal, Assembler::pn, overflow_with_error);
2168 // migrate the interpreter frame off of the stack
2170 mov(G2_thread, L7);
2171 // save nmethod
2172 mov(O0, L6);
2173 set_last_Java_frame(SP, noreg);
2174 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7);
2175 reset_last_Java_frame();
2176 mov(L7, G2_thread);
2178 // move OSR nmethod to I1
2179 mov(L6, I1);
2181 // OSR buffer to I0
2182 mov(O0, I0);
2184 // remove the interpreter frame
2185 restore(I5_savedSP, 0, SP);
2187 // Jump to the osr code.
2188 ld_ptr(O1, nmethod::osr_entry_point_offset(), O2);
2189 jmp(O2, G0);
2190 delayed()->nop();
2192 bind(overflow_with_error);
2194 bind(did_not_overflow);
2195 }
2199 void InterpreterMacroAssembler::interp_verify_oop(Register reg, TosState state, const char * file, int line) {
2200 if (state == atos) { MacroAssembler::_verify_oop(reg, "broken oop ", file, line); }
2201 }
2204 // local helper function for the verify_oop_or_return_address macro
2205 static bool verify_return_address(methodOopDesc* m, int bci) {
2206 #ifndef PRODUCT
2207 address pc = (address)(m->constMethod())
2208 + in_bytes(constMethodOopDesc::codes_offset()) + bci;
2209 // assume it is a valid return address if it is inside m and is preceded by a jsr
2210 if (!m->contains(pc)) return false;
2211 address jsr_pc;
2212 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr);
2213 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true;
2214 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w);
2215 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true;
2216 #endif // PRODUCT
2217 return false;
2218 }
2221 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) {
2222 if (!VerifyOops) return;
2223 // the VM documentation for the astore[_wide] bytecode allows
2224 // the TOS to be not only an oop but also a return address
2225 Label test;
2226 Label skip;
2227 // See if it is an address (in the current method):
2229 mov(reg, Rtmp);
2230 const int log2_bytecode_size_limit = 16;
2231 srl(Rtmp, log2_bytecode_size_limit, Rtmp);
2232 br_notnull_short( Rtmp, pt, test );
2234 // %%% should use call_VM_leaf here?
2235 save_frame_and_mov(0, Lmethod, O0, reg, O1);
2236 save_thread(L7_thread_cache);
2237 call(CAST_FROM_FN_PTR(address,verify_return_address), relocInfo::none);
2238 delayed()->nop();
2239 restore_thread(L7_thread_cache);
2240 br_notnull( O0, false, pt, skip );
2241 delayed()->restore();
2243 // Perform a more elaborate out-of-line call
2244 // Not an address; verify it:
2245 bind(test);
2246 verify_oop(reg);
2247 bind(skip);
2248 }
2251 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2252 if (state == ftos || state == dtos) MacroAssembler::verify_FPU(stack_depth);
2253 }
2254 #endif /* CC_INTERP */
2256 // Inline assembly for:
2257 //
2258 // if (thread is in interp_only_mode) {
2259 // InterpreterRuntime::post_method_entry();
2260 // }
2261 // if (DTraceMethodProbes) {
2262 // SharedRuntime::dtrace_method_entry(method, receiver);
2263 // }
2264 // if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
2265 // SharedRuntime::rc_trace_method_entry(method, receiver);
2266 // }
2268 void InterpreterMacroAssembler::notify_method_entry() {
2270 // C++ interpreter only uses this for native methods.
2272 // Whenever JVMTI puts a thread in interp_only_mode, method
2273 // entry/exit events are sent for that thread to track stack
2274 // depth. If it is possible to enter interp_only_mode we add
2275 // the code to check if the event should be sent.
2276 if (JvmtiExport::can_post_interpreter_events()) {
2277 Label L;
2278 Register temp_reg = O5;
2279 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset());
2280 ld(interp_only, temp_reg);
2281 cmp_and_br_short(temp_reg, 0, equal, pt, L);
2282 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry));
2283 bind(L);
2284 }
2286 {
2287 Register temp_reg = O5;
2288 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero);
2289 call_VM_leaf(noreg,
2290 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
2291 G2_thread, Lmethod);
2292 }
2294 // RedefineClasses() tracing support for obsolete method entry
2295 if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
2296 call_VM_leaf(noreg,
2297 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
2298 G2_thread, Lmethod);
2299 }
2300 }
2303 // Inline assembly for:
2304 //
2305 // if (thread is in interp_only_mode) {
2306 // // save result
2307 // InterpreterRuntime::post_method_exit();
2308 // // restore result
2309 // }
2310 // if (DTraceMethodProbes) {
2311 // SharedRuntime::dtrace_method_exit(thread, method);
2312 // }
2313 //
2314 // Native methods have their result stored in d_tmp and l_tmp
2315 // Java methods have their result stored in the expression stack
2317 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method,
2318 TosState state,
2319 NotifyMethodExitMode mode) {
2320 // C++ interpreter only uses this for native methods.
2322 // Whenever JVMTI puts a thread in interp_only_mode, method
2323 // entry/exit events are sent for that thread to track stack
2324 // depth. If it is possible to enter interp_only_mode we add
2325 // the code to check if the event should be sent.
2326 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2327 Label L;
2328 Register temp_reg = O5;
2329 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset());
2330 ld(interp_only, temp_reg);
2331 cmp_and_br_short(temp_reg, 0, equal, pt, L);
2333 // Note: frame::interpreter_frame_result has a dependency on how the
2334 // method result is saved across the call to post_method_exit. For
2335 // native methods it assumes the result registers are saved to
2336 // l_scratch and d_scratch. If this changes then the interpreter_frame_result
2337 // implementation will need to be updated too.
2339 save_return_value(state, is_native_method);
2340 call_VM(noreg,
2341 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2342 restore_return_value(state, is_native_method);
2343 bind(L);
2344 }
2346 {
2347 Register temp_reg = O5;
2348 // Dtrace notification
2349 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero);
2350 save_return_value(state, is_native_method);
2351 call_VM_leaf(
2352 noreg,
2353 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2354 G2_thread, Lmethod);
2355 restore_return_value(state, is_native_method);
2356 }
2357 }
2359 void InterpreterMacroAssembler::save_return_value(TosState state, bool is_native_call) {
2360 #ifdef CC_INTERP
2361 // result potentially in O0/O1: save it across calls
2362 stf(FloatRegisterImpl::D, F0, STATE(_native_fresult));
2363 #ifdef _LP64
2364 stx(O0, STATE(_native_lresult));
2365 #else
2366 std(O0, STATE(_native_lresult));
2367 #endif
2368 #else // CC_INTERP
2369 if (is_native_call) {
2370 stf(FloatRegisterImpl::D, F0, d_tmp);
2371 #ifdef _LP64
2372 stx(O0, l_tmp);
2373 #else
2374 std(O0, l_tmp);
2375 #endif
2376 } else {
2377 push(state);
2378 }
2379 #endif // CC_INTERP
2380 }
2382 void InterpreterMacroAssembler::restore_return_value( TosState state, bool is_native_call) {
2383 #ifdef CC_INTERP
2384 ldf(FloatRegisterImpl::D, STATE(_native_fresult), F0);
2385 #ifdef _LP64
2386 ldx(STATE(_native_lresult), O0);
2387 #else
2388 ldd(STATE(_native_lresult), O0);
2389 #endif
2390 #else // CC_INTERP
2391 if (is_native_call) {
2392 ldf(FloatRegisterImpl::D, d_tmp, F0);
2393 #ifdef _LP64
2394 ldx(l_tmp, O0);
2395 #else
2396 ldd(l_tmp, O0);
2397 #endif
2398 } else {
2399 pop(state);
2400 }
2401 #endif // CC_INTERP
2402 }
2404 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
2405 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
2406 int increment, int mask,
2407 Register scratch1, Register scratch2,
2408 Condition cond, Label *where) {
2409 ld(counter_addr, scratch1);
2410 add(scratch1, increment, scratch1);
2411 if (is_simm13(mask)) {
2412 andcc(scratch1, mask, G0);
2413 } else {
2414 set(mask, scratch2);
2415 andcc(scratch1, scratch2, G0);
2416 }
2417 br(cond, false, Assembler::pn, *where);
2418 delayed()->st(scratch1, counter_addr);
2419 }