1 /*
2 * Copyright (c) 1997, 2017, 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 "interp_masm_sparc.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "interpreter/interpreterRuntime.hpp"
29 #include "logging/log.hpp"
30 #include "oops/arrayOop.hpp"
31 #include "oops/markOop.hpp"
32 #include "oops/methodData.hpp"
33 #include "oops/method.hpp"
34 #include "oops/methodCounters.hpp"
35 #include "prims/jvmtiExport.hpp"
36 #include "prims/jvmtiThreadState.hpp"
37 #include "runtime/basicLock.hpp"
38 #include "runtime/biasedLocking.hpp"
39 #include "runtime/sharedRuntime.hpp"
40 #include "runtime/thread.inline.hpp"
41
42 #ifndef FAST_DISPATCH
43 #define FAST_DISPATCH 1
44 #endif
45 #undef FAST_DISPATCH
46
47 // Implementation of InterpreterMacroAssembler
48
49 // This file specializes the assember with interpreter-specific macros
50
51 const Address InterpreterMacroAssembler::l_tmp(FP, (frame::interpreter_frame_l_scratch_fp_offset * wordSize) + STACK_BIAS);
52 const Address InterpreterMacroAssembler::d_tmp(FP, (frame::interpreter_frame_d_scratch_fp_offset * wordSize) + STACK_BIAS);
53
54 void InterpreterMacroAssembler::jump_to_entry(address entry) {
55 assert(entry, "Entry must have been generated by now");
56 AddressLiteral al(entry);
57 jump_to(al, G3_scratch);
58 delayed()->nop();
59 }
60
61 void InterpreterMacroAssembler::compute_extra_locals_size_in_bytes(Register args_size, Register locals_size, Register delta) {
62 // Note: this algorithm is also used by C1's OSR entry sequence.
63 // Any changes should also be applied to CodeEmitter::emit_osr_entry().
64 assert_different_registers(args_size, locals_size);
65 // max_locals*2 for TAGS. Assumes that args_size has already been adjusted.
66 subcc(locals_size, args_size, delta);// extra space for non-arguments locals in words
67 // Use br/mov combination because it works on both V8 and V9 and is
68 // faster.
69 Label skip_move;
70 br(Assembler::negative, true, Assembler::pt, skip_move);
71 delayed()->mov(G0, delta);
72 bind(skip_move);
73 round_to(delta, WordsPerLong); // make multiple of 2 (SP must be 2-word aligned)
74 sll(delta, LogBytesPerWord, delta); // extra space for locals in bytes
75 }
76
77 // Dispatch code executed in the prolog of a bytecode which does not do it's
78 // own dispatch. The dispatch address is computed and placed in IdispatchAddress
79 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) {
80 assert_not_delayed();
81 #ifdef FAST_DISPATCH
82 // FAST_DISPATCH and ProfileInterpreter are mutually exclusive since
83 // they both use I2.
84 assert(!ProfileInterpreter, "FAST_DISPATCH and +ProfileInterpreter are mutually exclusive");
85 ldub(Lbcp, bcp_incr, Lbyte_code); // load next bytecode
86 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code);
87 // add offset to correct dispatch table
88 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize
89 ld_ptr(IdispatchTables, Lbyte_code, IdispatchAddress);// get entry addr
90 #else
91 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode
92 // dispatch table to use
93 AddressLiteral tbl(Interpreter::dispatch_table(state));
94 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize
95 set(tbl, G3_scratch); // compute addr of table
96 ld_ptr(G3_scratch, Lbyte_code, IdispatchAddress); // get entry addr
97 #endif
98 }
99
100
101 // Dispatch code executed in the epilog of a bytecode which does not do it's
102 // own dispatch. The dispatch address in IdispatchAddress is used for the
103 // dispatch.
104 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) {
105 assert_not_delayed();
106 verify_FPU(1, state);
107 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
108 jmp( IdispatchAddress, 0 );
109 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr);
110 else delayed()->nop();
111 }
112
113
114 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr) {
115 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr)
116 assert_not_delayed();
117 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode
118 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr);
119 }
120
121
122 void InterpreterMacroAssembler::dispatch_next_noverify_oop(TosState state, int bcp_incr) {
123 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr)
124 assert_not_delayed();
125 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode
126 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr, false);
127 }
128
129
130 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
131 // load current bytecode
132 assert_not_delayed();
133 ldub( Lbcp, 0, Lbyte_code); // load next bytecode
134 dispatch_base(state, table);
135 }
136
137
138 void InterpreterMacroAssembler::call_VM_leaf_base(
139 Register java_thread,
140 address entry_point,
141 int number_of_arguments
142 ) {
143 if (!java_thread->is_valid())
144 java_thread = L7_thread_cache;
145 // super call
146 MacroAssembler::call_VM_leaf_base(java_thread, entry_point, number_of_arguments);
147 }
148
149
150 void InterpreterMacroAssembler::call_VM_base(
151 Register oop_result,
152 Register java_thread,
153 Register last_java_sp,
154 address entry_point,
155 int number_of_arguments,
156 bool check_exception
157 ) {
158 if (!java_thread->is_valid())
159 java_thread = L7_thread_cache;
160 // See class ThreadInVMfromInterpreter, which assumes that the interpreter
161 // takes responsibility for setting its own thread-state on call-out.
162 // However, ThreadInVMfromInterpreter resets the state to "in_Java".
163
164 //save_bcp(); // save bcp
165 MacroAssembler::call_VM_base(oop_result, java_thread, last_java_sp, entry_point, number_of_arguments, check_exception);
166 //restore_bcp(); // restore bcp
167 //restore_locals(); // restore locals pointer
168 }
169
170
171 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) {
172 if (JvmtiExport::can_pop_frame()) {
173 Label L;
174
175 // Check the "pending popframe condition" flag in the current thread
176 ld(G2_thread, JavaThread::popframe_condition_offset(), scratch_reg);
177
178 // Initiate popframe handling only if it is not already being processed. If the flag
179 // has the popframe_processing bit set, it means that this code is called *during* popframe
180 // handling - we don't want to reenter.
181 btst(JavaThread::popframe_pending_bit, scratch_reg);
182 br(zero, false, pt, L);
183 delayed()->nop();
184 btst(JavaThread::popframe_processing_bit, scratch_reg);
185 br(notZero, false, pt, L);
186 delayed()->nop();
187
188 // Call Interpreter::remove_activation_preserving_args_entry() to get the
189 // address of the same-named entrypoint in the generated interpreter code.
190 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
191
192 // Jump to Interpreter::_remove_activation_preserving_args_entry
193 jmpl(O0, G0, G0);
194 delayed()->nop();
195 bind(L);
196 }
197 }
198
199
200 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
201 Register thr_state = G4_scratch;
202 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state);
203 const Address tos_addr(thr_state, JvmtiThreadState::earlyret_tos_offset());
204 const Address oop_addr(thr_state, JvmtiThreadState::earlyret_oop_offset());
205 const Address val_addr(thr_state, JvmtiThreadState::earlyret_value_offset());
206 switch (state) {
207 case ltos: ld_long(val_addr, Otos_l); break;
208 case atos: ld_ptr(oop_addr, Otos_l);
209 st_ptr(G0, oop_addr); break;
210 case btos: // fall through
211 case ztos: // fall through
212 case ctos: // fall through
213 case stos: // fall through
214 case itos: ld(val_addr, Otos_l1); break;
215 case ftos: ldf(FloatRegisterImpl::S, val_addr, Ftos_f); break;
216 case dtos: ldf(FloatRegisterImpl::D, val_addr, Ftos_d); break;
217 case vtos: /* nothing to do */ break;
218 default : ShouldNotReachHere();
219 }
220 // Clean up tos value in the jvmti thread state
221 or3(G0, ilgl, G3_scratch);
222 stw(G3_scratch, tos_addr);
223 st_long(G0, val_addr);
224 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
225 }
226
227
228 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) {
229 if (JvmtiExport::can_force_early_return()) {
230 Label L;
231 Register thr_state = G3_scratch;
232 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state);
233 br_null_short(thr_state, pt, L); // if (thread->jvmti_thread_state() == NULL) exit;
234
235 // Initiate earlyret handling only if it is not already being processed.
236 // If the flag has the earlyret_processing bit set, it means that this code
237 // is called *during* earlyret handling - we don't want to reenter.
238 ld(thr_state, JvmtiThreadState::earlyret_state_offset(), G4_scratch);
239 cmp_and_br_short(G4_scratch, JvmtiThreadState::earlyret_pending, Assembler::notEqual, pt, L);
240
241 // Call Interpreter::remove_activation_early_entry() to get the address of the
242 // same-named entrypoint in the generated interpreter code
243 ld(thr_state, JvmtiThreadState::earlyret_tos_offset(), Otos_l1);
244 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), Otos_l1);
245
246 // Jump to Interpreter::_remove_activation_early_entry
247 jmpl(O0, G0, G0);
248 delayed()->nop();
249 bind(L);
250 }
251 }
252
253
254 void InterpreterMacroAssembler::super_call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2) {
255 mov(arg_1, O0);
256 mov(arg_2, O1);
257 MacroAssembler::call_VM_leaf_base(thread_cache, entry_point, 2);
258 }
259
260 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table) {
261 assert_not_delayed();
262 dispatch_Lbyte_code(state, table);
263 }
264
265
266 void InterpreterMacroAssembler::dispatch_normal(TosState state) {
267 dispatch_base(state, Interpreter::normal_table(state));
268 }
269
270
271 void InterpreterMacroAssembler::dispatch_only(TosState state) {
272 dispatch_base(state, Interpreter::dispatch_table(state));
273 }
274
275
276 // common code to dispatch and dispatch_only
277 // dispatch value in Lbyte_code and increment Lbcp
278
279 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, address* table, int bcp_incr, bool verify) {
280 verify_FPU(1, state);
281 // %%%%% maybe implement +VerifyActivationFrameSize here
282 //verify_thread(); //too slow; we will just verify on method entry & exit
283 if (verify) interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
284 #ifdef FAST_DISPATCH
285 if (table == Interpreter::dispatch_table(state)) {
286 // use IdispatchTables
287 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code);
288 // add offset to correct dispatch table
289 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize
290 ld_ptr(IdispatchTables, Lbyte_code, G3_scratch); // get entry addr
291 } else {
292 #endif
293 // dispatch table to use
294 AddressLiteral tbl(table);
295 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize
296 set(tbl, G3_scratch); // compute addr of table
297 ld_ptr(G3_scratch, Lbyte_code, G3_scratch); // get entry addr
298 #ifdef FAST_DISPATCH
299 }
300 #endif
301 jmp( G3_scratch, 0 );
302 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr);
303 else delayed()->nop();
304 }
305
306
307 // Helpers for expression stack
308
309 // Longs and doubles are Category 2 computational types in the
310 // JVM specification (section 3.11.1) and take 2 expression stack or
311 // local slots.
312 // Aligning them on 32 bit with tagged stacks is hard because the code generated
313 // for the dup* bytecodes depends on what types are already on the stack.
314 // If the types are split into the two stack/local slots, that is much easier
315 // (and we can use 0 for non-reference tags).
316
317 // Known good alignment in _LP64 but unknown otherwise
318 void InterpreterMacroAssembler::load_unaligned_double(Register r1, int offset, FloatRegister d) {
319 assert_not_delayed();
320
321 #ifdef _LP64
322 ldf(FloatRegisterImpl::D, r1, offset, d);
323 #else
324 ldf(FloatRegisterImpl::S, r1, offset, d);
325 ldf(FloatRegisterImpl::S, r1, offset + Interpreter::stackElementSize, d->successor());
326 #endif
327 }
328
329 // Known good alignment in _LP64 but unknown otherwise
330 void InterpreterMacroAssembler::store_unaligned_double(FloatRegister d, Register r1, int offset) {
331 assert_not_delayed();
332
333 #ifdef _LP64
334 stf(FloatRegisterImpl::D, d, r1, offset);
335 // store something more useful here
336 debug_only(stx(G0, r1, offset+Interpreter::stackElementSize);)
337 #else
338 stf(FloatRegisterImpl::S, d, r1, offset);
339 stf(FloatRegisterImpl::S, d->successor(), r1, offset + Interpreter::stackElementSize);
340 #endif
341 }
342
343
344 // Known good alignment in _LP64 but unknown otherwise
345 void InterpreterMacroAssembler::load_unaligned_long(Register r1, int offset, Register rd) {
346 assert_not_delayed();
347 #ifdef _LP64
348 ldx(r1, offset, rd);
349 #else
350 ld(r1, offset, rd);
351 ld(r1, offset + Interpreter::stackElementSize, rd->successor());
352 #endif
353 }
354
355 // Known good alignment in _LP64 but unknown otherwise
356 void InterpreterMacroAssembler::store_unaligned_long(Register l, Register r1, int offset) {
357 assert_not_delayed();
358
359 #ifdef _LP64
360 stx(l, r1, offset);
361 // store something more useful here
362 stx(G0, r1, offset+Interpreter::stackElementSize);
363 #else
364 st(l, r1, offset);
365 st(l->successor(), r1, offset + Interpreter::stackElementSize);
366 #endif
367 }
368
369 void InterpreterMacroAssembler::pop_i(Register r) {
370 assert_not_delayed();
371 ld(Lesp, Interpreter::expr_offset_in_bytes(0), r);
372 inc(Lesp, Interpreter::stackElementSize);
373 debug_only(verify_esp(Lesp));
374 }
375
376 void InterpreterMacroAssembler::pop_ptr(Register r, Register scratch) {
377 assert_not_delayed();
378 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), r);
379 inc(Lesp, Interpreter::stackElementSize);
380 debug_only(verify_esp(Lesp));
381 }
382
383 void InterpreterMacroAssembler::pop_l(Register r) {
384 assert_not_delayed();
385 load_unaligned_long(Lesp, Interpreter::expr_offset_in_bytes(0), r);
386 inc(Lesp, 2*Interpreter::stackElementSize);
387 debug_only(verify_esp(Lesp));
388 }
389
390
391 void InterpreterMacroAssembler::pop_f(FloatRegister f, Register scratch) {
392 assert_not_delayed();
393 ldf(FloatRegisterImpl::S, Lesp, Interpreter::expr_offset_in_bytes(0), f);
394 inc(Lesp, Interpreter::stackElementSize);
395 debug_only(verify_esp(Lesp));
396 }
397
398
399 void InterpreterMacroAssembler::pop_d(FloatRegister f, Register scratch) {
400 assert_not_delayed();
401 load_unaligned_double(Lesp, Interpreter::expr_offset_in_bytes(0), f);
402 inc(Lesp, 2*Interpreter::stackElementSize);
403 debug_only(verify_esp(Lesp));
404 }
405
406
407 void InterpreterMacroAssembler::push_i(Register r) {
408 assert_not_delayed();
409 debug_only(verify_esp(Lesp));
410 st(r, Lesp, 0);
411 dec(Lesp, Interpreter::stackElementSize);
412 }
413
414 void InterpreterMacroAssembler::push_ptr(Register r) {
415 assert_not_delayed();
416 st_ptr(r, Lesp, 0);
417 dec(Lesp, Interpreter::stackElementSize);
418 }
419
420 // remember: our convention for longs in SPARC is:
421 // O0 (Otos_l1) has high-order part in first word,
422 // O1 (Otos_l2) has low-order part in second word
423
424 void InterpreterMacroAssembler::push_l(Register r) {
425 assert_not_delayed();
426 debug_only(verify_esp(Lesp));
427 // Longs are stored in memory-correct order, even if unaligned.
428 int offset = -Interpreter::stackElementSize;
429 store_unaligned_long(r, Lesp, offset);
430 dec(Lesp, 2 * Interpreter::stackElementSize);
431 }
432
433
434 void InterpreterMacroAssembler::push_f(FloatRegister f) {
435 assert_not_delayed();
436 debug_only(verify_esp(Lesp));
437 stf(FloatRegisterImpl::S, f, Lesp, 0);
438 dec(Lesp, Interpreter::stackElementSize);
439 }
440
441
442 void InterpreterMacroAssembler::push_d(FloatRegister d) {
443 assert_not_delayed();
444 debug_only(verify_esp(Lesp));
445 // Longs are stored in memory-correct order, even if unaligned.
446 int offset = -Interpreter::stackElementSize;
447 store_unaligned_double(d, Lesp, offset);
448 dec(Lesp, 2 * Interpreter::stackElementSize);
449 }
450
451
452 void InterpreterMacroAssembler::push(TosState state) {
453 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
454 switch (state) {
455 case atos: push_ptr(); break;
456 case btos: // fall through
457 case ztos: // fall through
458 case ctos: // fall through
459 case stos: // fall through
460 case itos: push_i(); break;
461 case ltos: push_l(); break;
462 case ftos: push_f(); break;
463 case dtos: push_d(); break;
464 case vtos: /* nothing to do */ break;
465 default : ShouldNotReachHere();
466 }
467 }
468
469
470 void InterpreterMacroAssembler::pop(TosState state) {
471 switch (state) {
472 case atos: pop_ptr(); break;
473 case btos: // fall through
474 case ztos: // fall through
475 case ctos: // fall through
476 case stos: // fall through
477 case itos: pop_i(); break;
478 case ltos: pop_l(); break;
479 case ftos: pop_f(); break;
480 case dtos: pop_d(); break;
481 case vtos: /* nothing to do */ break;
482 default : ShouldNotReachHere();
483 }
484 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
485 }
486
487
488 // Helpers for swap and dup
489 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
490 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(n), val);
491 }
492 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
493 st_ptr(val, Lesp, Interpreter::expr_offset_in_bytes(n));
494 }
495
496
497 void InterpreterMacroAssembler::load_receiver(Register param_count,
498 Register recv) {
499 sll(param_count, Interpreter::logStackElementSize, param_count);
500 ld_ptr(Lesp, param_count, recv); // gets receiver oop
501 }
502
503 void InterpreterMacroAssembler::empty_expression_stack() {
504 // Reset Lesp.
505 sub( Lmonitors, wordSize, Lesp );
506
507 // Reset SP by subtracting more space from Lesp.
508 Label done;
509 assert(G4_scratch != Gframe_size, "Only you can prevent register aliasing!");
510
511 // A native does not need to do this, since its callee does not change SP.
512 ld(Lmethod, Method::access_flags_offset(), Gframe_size); // Load access flags.
513 btst(JVM_ACC_NATIVE, Gframe_size);
514 br(Assembler::notZero, false, Assembler::pt, done);
515 delayed()->nop();
516
517 // Compute max expression stack+register save area
518 ld_ptr(Lmethod, in_bytes(Method::const_offset()), Gframe_size);
519 lduh(Gframe_size, in_bytes(ConstMethod::max_stack_offset()), Gframe_size); // Load max stack.
520 add(Gframe_size, frame::memory_parameter_word_sp_offset+Method::extra_stack_entries(), Gframe_size );
521
522 //
523 // now set up a stack frame with the size computed above
524 //
525 //round_to( Gframe_size, WordsPerLong ); // -- moved down to the "and" below
526 sll( Gframe_size, LogBytesPerWord, Gframe_size );
527 sub( Lesp, Gframe_size, Gframe_size );
528 and3( Gframe_size, -(2 * wordSize), Gframe_size ); // align SP (downwards) to an 8/16-byte boundary
529 debug_only(verify_sp(Gframe_size, G4_scratch));
530 #ifdef _LP64
531 sub(Gframe_size, STACK_BIAS, Gframe_size );
532 #endif
533 mov(Gframe_size, SP);
534
535 bind(done);
536 }
537
538
539 #ifdef ASSERT
540 void InterpreterMacroAssembler::verify_sp(Register Rsp, Register Rtemp) {
541 Label Bad, OK;
542
543 // Saved SP must be aligned.
544 #ifdef _LP64
545 btst(2*BytesPerWord-1, Rsp);
546 #else
547 btst(LongAlignmentMask, Rsp);
548 #endif
549 br(Assembler::notZero, false, Assembler::pn, Bad);
550 delayed()->nop();
551
552 // Saved SP, plus register window size, must not be above FP.
553 add(Rsp, frame::register_save_words * wordSize, Rtemp);
554 #ifdef _LP64
555 sub(Rtemp, STACK_BIAS, Rtemp); // Bias Rtemp before cmp to FP
556 #endif
557 cmp_and_brx_short(Rtemp, FP, Assembler::greaterUnsigned, Assembler::pn, Bad);
558
559 // Saved SP must not be ridiculously below current SP.
560 size_t maxstack = MAX2(JavaThread::stack_size_at_create(), (size_t) 4*K*K);
561 set(maxstack, Rtemp);
562 sub(SP, Rtemp, Rtemp);
563 #ifdef _LP64
564 add(Rtemp, STACK_BIAS, Rtemp); // Unbias Rtemp before cmp to Rsp
565 #endif
566 cmp_and_brx_short(Rsp, Rtemp, Assembler::lessUnsigned, Assembler::pn, Bad);
567
568 ba_short(OK);
569
570 bind(Bad);
571 stop("on return to interpreted call, restored SP is corrupted");
572
573 bind(OK);
574 }
575
576
577 void InterpreterMacroAssembler::verify_esp(Register Resp) {
578 // about to read or write Resp[0]
579 // make sure it is not in the monitors or the register save area
580 Label OK1, OK2;
581
582 cmp(Resp, Lmonitors);
583 brx(Assembler::lessUnsigned, true, Assembler::pt, OK1);
584 delayed()->sub(Resp, frame::memory_parameter_word_sp_offset * wordSize, Resp);
585 stop("too many pops: Lesp points into monitor area");
586 bind(OK1);
587 #ifdef _LP64
588 sub(Resp, STACK_BIAS, Resp);
589 #endif
590 cmp(Resp, SP);
591 brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, OK2);
592 delayed()->add(Resp, STACK_BIAS + frame::memory_parameter_word_sp_offset * wordSize, Resp);
593 stop("too many pushes: Lesp points into register window");
594 bind(OK2);
595 }
596 #endif // ASSERT
597
598 // Load compiled (i2c) or interpreter entry when calling from interpreted and
599 // do the call. Centralized so that all interpreter calls will do the same actions.
600 // If jvmti single stepping is on for a thread we must not call compiled code.
601 void InterpreterMacroAssembler::call_from_interpreter(Register target, Register scratch, Register Rret) {
602
603 // Assume we want to go compiled if available
604
605 ld_ptr(G5_method, in_bytes(Method::from_interpreted_offset()), target);
606
607 if (JvmtiExport::can_post_interpreter_events()) {
608 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
609 // compiled code in threads for which the event is enabled. Check here for
610 // interp_only_mode if these events CAN be enabled.
611 verify_thread();
612 Label skip_compiled_code;
613
614 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset());
615 ld(interp_only, scratch);
616 cmp_zero_and_br(Assembler::notZero, scratch, skip_compiled_code, true, Assembler::pn);
617 delayed()->ld_ptr(G5_method, in_bytes(Method::interpreter_entry_offset()), target);
618 bind(skip_compiled_code);
619 }
620
621 // the i2c_adapters need Method* in G5_method (right? %%%)
622 // do the call
623 #ifdef ASSERT
624 {
625 Label ok;
626 br_notnull_short(target, Assembler::pt, ok);
627 stop("null entry point");
628 bind(ok);
629 }
630 #endif // ASSERT
631
632 // Adjust Rret first so Llast_SP can be same as Rret
633 add(Rret, -frame::pc_return_offset, O7);
634 add(Lesp, BytesPerWord, Gargs); // setup parameter pointer
635 // Record SP so we can remove any stack space allocated by adapter transition
636 jmp(target, 0);
637 delayed()->mov(SP, Llast_SP);
638 }
639
640 void InterpreterMacroAssembler::if_cmp(Condition cc, bool ptr_compare) {
641 assert_not_delayed();
642
643 Label not_taken;
644 if (ptr_compare) brx(cc, false, Assembler::pn, not_taken);
645 else br (cc, false, Assembler::pn, not_taken);
646 delayed()->nop();
647
648 TemplateTable::branch(false,false);
649
650 bind(not_taken);
651
652 profile_not_taken_branch(G3_scratch);
653 }
654
655
656 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(
657 int bcp_offset,
658 Register Rtmp,
659 Register Rdst,
660 signedOrNot is_signed,
661 setCCOrNot should_set_CC ) {
662 assert(Rtmp != Rdst, "need separate temp register");
663 assert_not_delayed();
664 switch (is_signed) {
665 default: ShouldNotReachHere();
666
667 case Signed: ldsb( Lbcp, bcp_offset, Rdst ); break; // high byte
668 case Unsigned: ldub( Lbcp, bcp_offset, Rdst ); break; // high byte
669 }
670 ldub( Lbcp, bcp_offset + 1, Rtmp ); // low byte
671 sll( Rdst, BitsPerByte, Rdst);
672 switch (should_set_CC ) {
673 default: ShouldNotReachHere();
674
675 case set_CC: orcc( Rdst, Rtmp, Rdst ); break;
676 case dont_set_CC: or3( Rdst, Rtmp, Rdst ); break;
677 }
678 }
679
680
681 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(
682 int bcp_offset,
683 Register Rtmp,
684 Register Rdst,
685 setCCOrNot should_set_CC ) {
686 assert(Rtmp != Rdst, "need separate temp register");
687 assert_not_delayed();
688 add( Lbcp, bcp_offset, Rtmp);
689 andcc( Rtmp, 3, G0);
690 Label aligned;
691 switch (should_set_CC ) {
692 default: ShouldNotReachHere();
693
694 case set_CC: break;
695 case dont_set_CC: break;
696 }
697
698 br(Assembler::zero, true, Assembler::pn, aligned);
699 #ifdef _LP64
700 delayed()->ldsw(Rtmp, 0, Rdst);
701 #else
702 delayed()->ld(Rtmp, 0, Rdst);
703 #endif
704
705 ldub(Lbcp, bcp_offset + 3, Rdst);
706 ldub(Lbcp, bcp_offset + 2, Rtmp); sll(Rtmp, 8, Rtmp); or3(Rtmp, Rdst, Rdst);
707 ldub(Lbcp, bcp_offset + 1, Rtmp); sll(Rtmp, 16, Rtmp); or3(Rtmp, Rdst, Rdst);
708 #ifdef _LP64
709 ldsb(Lbcp, bcp_offset + 0, Rtmp); sll(Rtmp, 24, Rtmp);
710 #else
711 // Unsigned load is faster than signed on some implementations
712 ldub(Lbcp, bcp_offset + 0, Rtmp); sll(Rtmp, 24, Rtmp);
713 #endif
714 or3(Rtmp, Rdst, Rdst );
715
716 bind(aligned);
717 if (should_set_CC == set_CC) tst(Rdst);
718 }
719
720 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register temp, Register index,
721 int bcp_offset, size_t index_size) {
722 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
723 if (index_size == sizeof(u2)) {
724 get_2_byte_integer_at_bcp(bcp_offset, temp, index, Unsigned);
725 } else if (index_size == sizeof(u4)) {
726 get_4_byte_integer_at_bcp(bcp_offset, temp, index);
727 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
728 xor3(index, -1, index); // convert to plain index
729 } else if (index_size == sizeof(u1)) {
730 ldub(Lbcp, bcp_offset, index);
731 } else {
732 ShouldNotReachHere();
733 }
734 }
735
736
737 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, Register tmp,
738 int bcp_offset, size_t index_size) {
739 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
740 assert_different_registers(cache, tmp);
741 assert_not_delayed();
742 get_cache_index_at_bcp(cache, tmp, bcp_offset, index_size);
743 // convert from field index to ConstantPoolCacheEntry index and from
744 // word index to byte offset
745 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp);
746 add(LcpoolCache, tmp, cache);
747 }
748
749
750 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
751 Register temp,
752 Register bytecode,
753 int byte_no,
754 int bcp_offset,
755 size_t index_size) {
756 get_cache_and_index_at_bcp(cache, temp, bcp_offset, index_size);
757 ld_ptr(cache, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset(), bytecode);
758 const int shift_count = (1 + byte_no) * BitsPerByte;
759 assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
760 (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
761 "correct shift count");
762 srl(bytecode, shift_count, bytecode);
763 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
764 and3(bytecode, ConstantPoolCacheEntry::bytecode_1_mask, bytecode);
765 }
766
767
768 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, Register tmp,
769 int bcp_offset, size_t index_size) {
770 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
771 assert_different_registers(cache, tmp);
772 assert_not_delayed();
773 if (index_size == sizeof(u2)) {
774 get_2_byte_integer_at_bcp(bcp_offset, cache, tmp, Unsigned);
775 } else {
776 ShouldNotReachHere(); // other sizes not supported here
777 }
778 // convert from field index to ConstantPoolCacheEntry index
779 // and from word index to byte offset
780 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp);
781 // skip past the header
782 add(tmp, in_bytes(ConstantPoolCache::base_offset()), tmp);
783 // construct pointer to cache entry
784 add(LcpoolCache, tmp, cache);
785 }
786
787
788 // Load object from cpool->resolved_references(index)
789 void InterpreterMacroAssembler::load_resolved_reference_at_index(
790 Register result, Register index) {
791 assert_different_registers(result, index);
792 assert_not_delayed();
793 // convert from field index to resolved_references() index and from
794 // word index to byte offset. Since this is a java object, it can be compressed
795 Register tmp = index; // reuse
796 sll(index, LogBytesPerHeapOop, tmp);
797 get_constant_pool(result);
798 // load pointer for resolved_references[] objArray
799 ld_ptr(result, ConstantPool::cache_offset_in_bytes(), result);
800 ld_ptr(result, ConstantPoolCache::resolved_references_offset_in_bytes(), result);
801 // JNIHandles::resolve(result)
802 ld_ptr(result, 0, result);
803 // Add in the index
804 add(result, tmp, result);
805 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result);
806 }
807
808
809 // load cpool->resolved_klass_at(index)
810 void InterpreterMacroAssembler::load_resolved_klass_at_offset(Register Rcpool,
811 Register Roffset, Register Rklass) {
812 // int value = *this_cp->int_at_addr(which);
813 // int resolved_klass_index = extract_low_short_from_int(value);
814 //
815 // Because SPARC is big-endian, the low_short is at (cpool->int_at_addr(which) + 2 bytes)
816 add(Roffset, Rcpool, Roffset);
817 lduh(Roffset, sizeof(ConstantPool) + 2, Roffset); // Roffset = resolved_klass_index
818
819 Register Rresolved_klasses = Rklass;
820 ld_ptr(Rcpool, ConstantPool::resolved_klasses_offset_in_bytes(), Rresolved_klasses);
821 sll(Roffset, LogBytesPerWord, Roffset);
822 add(Roffset, Array<Klass*>::base_offset_in_bytes(), Roffset);
823 ld_ptr(Rresolved_klasses, Roffset, Rklass);
824 }
825
826
827 // Generate a subtype check: branch to ok_is_subtype if sub_klass is
828 // a subtype of super_klass. Blows registers Rsuper_klass, Rsub_klass, tmp1, tmp2.
829 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
830 Register Rsuper_klass,
831 Register Rtmp1,
832 Register Rtmp2,
833 Register Rtmp3,
834 Label &ok_is_subtype ) {
835 Label not_subtype;
836
837 // Profile the not-null value's klass.
838 profile_typecheck(Rsub_klass, Rtmp1);
839
840 check_klass_subtype_fast_path(Rsub_klass, Rsuper_klass,
841 Rtmp1, Rtmp2,
842 &ok_is_subtype, ¬_subtype, NULL);
843
844 check_klass_subtype_slow_path(Rsub_klass, Rsuper_klass,
845 Rtmp1, Rtmp2, Rtmp3, /*hack:*/ noreg,
846 &ok_is_subtype, NULL);
847
848 bind(not_subtype);
849 profile_typecheck_failed(Rtmp1);
850 }
851
852 // Separate these two to allow for delay slot in middle
853 // These are used to do a test and full jump to exception-throwing code.
854
855 // %%%%% Could possibly reoptimize this by testing to see if could use
856 // a single conditional branch (i.e. if span is small enough.
857 // If you go that route, than get rid of the split and give up
858 // on the delay-slot hack.
859
860 void InterpreterMacroAssembler::throw_if_not_1_icc( Condition ok_condition,
861 Label& ok ) {
862 assert_not_delayed();
863 br(ok_condition, true, pt, ok);
864 // DELAY SLOT
865 }
866
867 void InterpreterMacroAssembler::throw_if_not_1_xcc( Condition ok_condition,
868 Label& ok ) {
869 assert_not_delayed();
870 bp( ok_condition, true, Assembler::xcc, pt, ok);
871 // DELAY SLOT
872 }
873
874 void InterpreterMacroAssembler::throw_if_not_1_x( Condition ok_condition,
875 Label& ok ) {
876 assert_not_delayed();
877 brx(ok_condition, true, pt, ok);
878 // DELAY SLOT
879 }
880
881 void InterpreterMacroAssembler::throw_if_not_2( address throw_entry_point,
882 Register Rscratch,
883 Label& ok ) {
884 assert(throw_entry_point != NULL, "entry point must be generated by now");
885 AddressLiteral dest(throw_entry_point);
886 jump_to(dest, Rscratch);
887 delayed()->nop();
888 bind(ok);
889 }
890
891
892 // And if you cannot use the delay slot, here is a shorthand:
893
894 void InterpreterMacroAssembler::throw_if_not_icc( Condition ok_condition,
895 address throw_entry_point,
896 Register Rscratch ) {
897 Label ok;
898 if (ok_condition != never) {
899 throw_if_not_1_icc( ok_condition, ok);
900 delayed()->nop();
901 }
902 throw_if_not_2( throw_entry_point, Rscratch, ok);
903 }
904 void InterpreterMacroAssembler::throw_if_not_xcc( Condition ok_condition,
905 address throw_entry_point,
906 Register Rscratch ) {
907 Label ok;
908 if (ok_condition != never) {
909 throw_if_not_1_xcc( ok_condition, ok);
910 delayed()->nop();
911 }
912 throw_if_not_2( throw_entry_point, Rscratch, ok);
913 }
914 void InterpreterMacroAssembler::throw_if_not_x( Condition ok_condition,
915 address throw_entry_point,
916 Register Rscratch ) {
917 Label ok;
918 if (ok_condition != never) {
919 throw_if_not_1_x( ok_condition, ok);
920 delayed()->nop();
921 }
922 throw_if_not_2( throw_entry_point, Rscratch, ok);
923 }
924
925 // Check that index is in range for array, then shift index by index_shift, and put arrayOop + shifted_index into res
926 // Note: res is still shy of address by array offset into object.
927
928 void InterpreterMacroAssembler::index_check_without_pop(Register array, Register index, int index_shift, Register tmp, Register res) {
929 assert_not_delayed();
930
931 verify_oop(array);
932 #ifdef _LP64
933 // sign extend since tos (index) can be a 32bit value
934 sra(index, G0, index);
935 #endif // _LP64
936
937 // check array
938 Label ptr_ok;
939 tst(array);
940 throw_if_not_1_x( notZero, ptr_ok );
941 delayed()->ld( array, arrayOopDesc::length_offset_in_bytes(), tmp ); // check index
942 throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ptr_ok);
943
944 Label index_ok;
945 cmp(index, tmp);
946 throw_if_not_1_icc( lessUnsigned, index_ok );
947 if (index_shift > 0) delayed()->sll(index, index_shift, index);
948 else delayed()->add(array, index, res); // addr - const offset in index
949 // convention: move aberrant index into G3_scratch for exception message
950 mov(index, G3_scratch);
951 throw_if_not_2( Interpreter::_throw_ArrayIndexOutOfBoundsException_entry, G4_scratch, index_ok);
952
953 // add offset if didn't do it in delay slot
954 if (index_shift > 0) add(array, index, res); // addr - const offset in index
955 }
956
957
958 void InterpreterMacroAssembler::index_check(Register array, Register index, int index_shift, Register tmp, Register res) {
959 assert_not_delayed();
960
961 // pop array
962 pop_ptr(array);
963
964 // check array
965 index_check_without_pop(array, index, index_shift, tmp, res);
966 }
967
968
969 void InterpreterMacroAssembler::get_const(Register Rdst) {
970 ld_ptr(Lmethod, in_bytes(Method::const_offset()), Rdst);
971 }
972
973
974 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) {
975 get_const(Rdst);
976 ld_ptr(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst);
977 }
978
979
980 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) {
981 get_constant_pool(Rdst);
982 ld_ptr(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst);
983 }
984
985
986 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) {
987 get_constant_pool(Rcpool);
988 ld_ptr(Rcpool, ConstantPool::tags_offset_in_bytes(), Rtags);
989 }
990
991
992 // unlock if synchronized method
993 //
994 // Unlock the receiver if this is a synchronized method.
995 // Unlock any Java monitors from syncronized blocks.
996 //
997 // If there are locked Java monitors
998 // If throw_monitor_exception
999 // throws IllegalMonitorStateException
1000 // Else if install_monitor_exception
1001 // installs IllegalMonitorStateException
1002 // Else
1003 // no error processing
1004 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state,
1005 bool throw_monitor_exception,
1006 bool install_monitor_exception) {
1007 Label unlocked, unlock, no_unlock;
1008
1009 // get the value of _do_not_unlock_if_synchronized into G1_scratch
1010 const Address do_not_unlock_if_synchronized(G2_thread,
1011 JavaThread::do_not_unlock_if_synchronized_offset());
1012 ldbool(do_not_unlock_if_synchronized, G1_scratch);
1013 stbool(G0, do_not_unlock_if_synchronized); // reset the flag
1014
1015 // check if synchronized method
1016 const Address access_flags(Lmethod, Method::access_flags_offset());
1017 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
1018 push(state); // save tos
1019 ld(access_flags, G3_scratch); // Load access flags.
1020 btst(JVM_ACC_SYNCHRONIZED, G3_scratch);
1021 br(zero, false, pt, unlocked);
1022 delayed()->nop();
1023
1024 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
1025 // is set.
1026 cmp_zero_and_br(Assembler::notZero, G1_scratch, no_unlock);
1027 delayed()->nop();
1028
1029 // BasicObjectLock will be first in list, since this is a synchronized method. However, need
1030 // to check that the object has not been unlocked by an explicit monitorexit bytecode.
1031
1032 //Intel: if (throw_monitor_exception) ... else ...
1033 // Entry already unlocked, need to throw exception
1034 //...
1035
1036 // pass top-most monitor elem
1037 add( top_most_monitor(), O1 );
1038
1039 ld_ptr(O1, BasicObjectLock::obj_offset_in_bytes(), G3_scratch);
1040 br_notnull_short(G3_scratch, pt, unlock);
1041
1042 if (throw_monitor_exception) {
1043 // Entry already unlocked need to throw an exception
1044 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1045 should_not_reach_here();
1046 } else {
1047 // Monitor already unlocked during a stack unroll.
1048 // If requested, install an illegal_monitor_state_exception.
1049 // Continue with stack unrolling.
1050 if (install_monitor_exception) {
1051 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
1052 }
1053 ba_short(unlocked);
1054 }
1055
1056 bind(unlock);
1057
1058 unlock_object(O1);
1059
1060 bind(unlocked);
1061
1062 // I0, I1: Might contain return value
1063
1064 // Check that all monitors are unlocked
1065 { Label loop, exception, entry, restart;
1066
1067 Register Rmptr = O0;
1068 Register Rtemp = O1;
1069 Register Rlimit = Lmonitors;
1070 const jint delta = frame::interpreter_frame_monitor_size() * wordSize;
1071 assert( (delta & LongAlignmentMask) == 0,
1072 "sizeof BasicObjectLock must be even number of doublewords");
1073
1074 #ifdef ASSERT
1075 add(top_most_monitor(), Rmptr, delta);
1076 { Label L;
1077 // ensure that Rmptr starts out above (or at) Rlimit
1078 cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L);
1079 stop("monitor stack has negative size");
1080 bind(L);
1081 }
1082 #endif
1083 bind(restart);
1084 ba(entry);
1085 delayed()->
1086 add(top_most_monitor(), Rmptr, delta); // points to current entry, starting with bottom-most entry
1087
1088 // Entry is still locked, need to throw exception
1089 bind(exception);
1090 if (throw_monitor_exception) {
1091 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1092 should_not_reach_here();
1093 } else {
1094 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception.
1095 // Unlock does not block, so don't have to worry about the frame
1096 unlock_object(Rmptr);
1097 if (install_monitor_exception) {
1098 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
1099 }
1100 ba_short(restart);
1101 }
1102
1103 bind(loop);
1104 cmp(Rtemp, G0); // check if current entry is used
1105 brx(Assembler::notEqual, false, pn, exception);
1106 delayed()->
1107 dec(Rmptr, delta); // otherwise advance to next entry
1108 #ifdef ASSERT
1109 { Label L;
1110 // ensure that Rmptr has not somehow stepped below Rlimit
1111 cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L);
1112 stop("ran off the end of the monitor stack");
1113 bind(L);
1114 }
1115 #endif
1116 bind(entry);
1117 cmp(Rmptr, Rlimit); // check if bottom reached
1118 brx(Assembler::notEqual, true, pn, loop); // if not at bottom then check this entry
1119 delayed()->
1120 ld_ptr(Rmptr, BasicObjectLock::obj_offset_in_bytes() - delta, Rtemp);
1121 }
1122
1123 bind(no_unlock);
1124 pop(state);
1125 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
1126 }
1127
1128 void InterpreterMacroAssembler::narrow(Register result) {
1129
1130 ld_ptr(Address(Lmethod, Method::const_offset()), G3_scratch);
1131 ldub(G3_scratch, in_bytes(ConstMethod::result_type_offset()), G3_scratch);
1132
1133 Label notBool, notByte, notChar, done;
1134
1135 // common case first
1136 cmp(G3_scratch, T_INT);
1137 br(Assembler::equal, true, pn, done);
1138 delayed()->nop();
1139
1140 cmp(G3_scratch, T_BOOLEAN);
1141 br(Assembler::notEqual, true, pn, notBool);
1142 delayed()->cmp(G3_scratch, T_BYTE);
1143 and3(result, 1, result);
1144 ba(done);
1145 delayed()->nop();
1146
1147 bind(notBool);
1148 // cmp(G3_scratch, T_BYTE);
1149 br(Assembler::notEqual, true, pn, notByte);
1150 delayed()->cmp(G3_scratch, T_CHAR);
1151 sll(result, 24, result);
1152 sra(result, 24, result);
1153 ba(done);
1154 delayed()->nop();
1155
1156 bind(notByte);
1157 // cmp(G3_scratch, T_CHAR);
1158 sll(result, 16, result);
1159 br(Assembler::notEqual, true, pn, done);
1160 delayed()->sra(result, 16, result);
1161 // sll(result, 16, result);
1162 srl(result, 16, result);
1163
1164 // bind(notChar);
1165 // must be short, instructions already executed in delay slot
1166 // sll(result, 16, result);
1167 // sra(result, 16, result);
1168
1169 bind(done);
1170 }
1171
1172 // remove activation
1173 //
1174 // Unlock the receiver if this is a synchronized method.
1175 // Unlock any Java monitors from syncronized blocks.
1176 // Remove the activation from the stack.
1177 //
1178 // If there are locked Java monitors
1179 // If throw_monitor_exception
1180 // throws IllegalMonitorStateException
1181 // Else if install_monitor_exception
1182 // installs IllegalMonitorStateException
1183 // Else
1184 // no error processing
1185 void InterpreterMacroAssembler::remove_activation(TosState state,
1186 bool throw_monitor_exception,
1187 bool install_monitor_exception) {
1188
1189 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception);
1190
1191 // save result (push state before jvmti call and pop it afterwards) and notify jvmti
1192 notify_method_exit(false, state, NotifyJVMTI);
1193
1194 if (StackReservedPages > 0) {
1195 // testing if Stack Reserved Area needs to be re-enabled
1196 Label no_reserved_zone_enabling;
1197 ld_ptr(G2_thread, JavaThread::reserved_stack_activation_offset(), G3_scratch);
1198 cmp_and_brx_short(SP, G3_scratch, Assembler::lessUnsigned, Assembler::pt, no_reserved_zone_enabling);
1199
1200 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), G2_thread);
1201 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError), G2_thread);
1202 should_not_reach_here();
1203
1204 bind(no_reserved_zone_enabling);
1205 }
1206
1207 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
1208 verify_thread();
1209
1210 // return tos
1211 assert(Otos_l1 == Otos_i, "adjust code below");
1212 switch (state) {
1213 #ifdef _LP64
1214 case ltos: mov(Otos_l, Otos_l->after_save()); break; // O0 -> I0
1215 #else
1216 case ltos: mov(Otos_l2, Otos_l2->after_save()); // fall through // O1 -> I1
1217 #endif
1218 case btos: // fall through
1219 case ztos: // fall through
1220 case ctos:
1221 case stos: // fall through
1222 case atos: // fall through
1223 case itos: mov(Otos_l1, Otos_l1->after_save()); break; // O0 -> I0
1224 case ftos: // fall through
1225 case dtos: // fall through
1226 case vtos: /* nothing to do */ break;
1227 default : ShouldNotReachHere();
1228 }
1229
1230 #if defined(COMPILER2) && !defined(_LP64)
1231 if (state == ltos) {
1232 // C2 expects long results in G1 we can't tell if we're returning to interpreted
1233 // or compiled so just be safe use G1 and O0/O1
1234
1235 // Shift bits into high (msb) of G1
1236 sllx(Otos_l1->after_save(), 32, G1);
1237 // Zero extend low bits
1238 srl (Otos_l2->after_save(), 0, Otos_l2->after_save());
1239 or3 (Otos_l2->after_save(), G1, G1);
1240 }
1241 #endif /* COMPILER2 */
1242
1243 }
1244
1245 // Lock object
1246 //
1247 // Argument - lock_reg points to the BasicObjectLock to be used for locking,
1248 // it must be initialized with the object to lock
1249 void InterpreterMacroAssembler::lock_object(Register lock_reg, Register Object) {
1250 if (UseHeavyMonitors) {
1251 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg);
1252 }
1253 else {
1254 Register obj_reg = Object;
1255 Register mark_reg = G4_scratch;
1256 Register temp_reg = G1_scratch;
1257 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes());
1258 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes());
1259 Label done;
1260
1261 Label slow_case;
1262
1263 assert_different_registers(lock_reg, obj_reg, mark_reg, temp_reg);
1264
1265 // load markOop from object into mark_reg
1266 ld_ptr(mark_addr, mark_reg);
1267
1268 if (UseBiasedLocking) {
1269 biased_locking_enter(obj_reg, mark_reg, temp_reg, done, &slow_case);
1270 }
1271
1272 // get the address of basicLock on stack that will be stored in the object
1273 // we need a temporary register here as we do not want to clobber lock_reg
1274 // (cas clobbers the destination register)
1275 mov(lock_reg, temp_reg);
1276 // set mark reg to be (markOop of object | UNLOCK_VALUE)
1277 or3(mark_reg, markOopDesc::unlocked_value, mark_reg);
1278 // initialize the box (Must happen before we update the object mark!)
1279 st_ptr(mark_reg, lock_addr, BasicLock::displaced_header_offset_in_bytes());
1280 // compare and exchange object_addr, markOop | 1, stack address of basicLock
1281 assert(mark_addr.disp() == 0, "cas must take a zero displacement");
1282 cas_ptr(mark_addr.base(), mark_reg, temp_reg);
1283
1284 // if the compare and exchange succeeded we are done (we saw an unlocked object)
1285 cmp_and_brx_short(mark_reg, temp_reg, Assembler::equal, Assembler::pt, done);
1286
1287 // We did not see an unlocked object so try the fast recursive case
1288
1289 // Check if owner is self by comparing the value in the markOop of object
1290 // with the stack pointer
1291 sub(temp_reg, SP, temp_reg);
1292 #ifdef _LP64
1293 sub(temp_reg, STACK_BIAS, temp_reg);
1294 #endif
1295 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
1296
1297 // Composite "andcc" test:
1298 // (a) %sp -vs- markword proximity check, and,
1299 // (b) verify mark word LSBs == 0 (Stack-locked).
1300 //
1301 // FFFFF003/FFFFFFFFFFFF003 is (markOopDesc::lock_mask_in_place | -os::vm_page_size())
1302 // Note that the page size used for %sp proximity testing is arbitrary and is
1303 // unrelated to the actual MMU page size. We use a 'logical' page size of
1304 // 4096 bytes. F..FFF003 is designed to fit conveniently in the SIMM13 immediate
1305 // field of the andcc instruction.
1306 andcc (temp_reg, 0xFFFFF003, G0) ;
1307
1308 // if condition is true we are done and hence we can store 0 in the displaced
1309 // header indicating it is a recursive lock and be done
1310 brx(Assembler::zero, true, Assembler::pt, done);
1311 delayed()->st_ptr(G0, lock_addr, BasicLock::displaced_header_offset_in_bytes());
1312
1313 // none of the above fast optimizations worked so we have to get into the
1314 // slow case of monitor enter
1315 bind(slow_case);
1316 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg);
1317
1318 bind(done);
1319 }
1320 }
1321
1322 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
1323 //
1324 // Argument - lock_reg points to the BasicObjectLock for lock
1325 // Throw IllegalMonitorException if object is not locked by current thread
1326 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1327 if (UseHeavyMonitors) {
1328 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1329 } else {
1330 Register obj_reg = G3_scratch;
1331 Register mark_reg = G4_scratch;
1332 Register displaced_header_reg = G1_scratch;
1333 Address lockobj_addr(lock_reg, BasicObjectLock::obj_offset_in_bytes());
1334 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes());
1335 Label done;
1336
1337 if (UseBiasedLocking) {
1338 // load the object out of the BasicObjectLock
1339 ld_ptr(lockobj_addr, obj_reg);
1340 biased_locking_exit(mark_addr, mark_reg, done, true);
1341 st_ptr(G0, lockobj_addr); // free entry
1342 }
1343
1344 // Test first if we are in the fast recursive case
1345 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes() + BasicLock::displaced_header_offset_in_bytes());
1346 ld_ptr(lock_addr, displaced_header_reg);
1347 br_null(displaced_header_reg, true, Assembler::pn, done);
1348 delayed()->st_ptr(G0, lockobj_addr); // free entry
1349
1350 // See if it is still a light weight lock, if so we just unlock
1351 // the object and we are done
1352
1353 if (!UseBiasedLocking) {
1354 // load the object out of the BasicObjectLock
1355 ld_ptr(lockobj_addr, obj_reg);
1356 }
1357
1358 // we have the displaced header in displaced_header_reg
1359 // we expect to see the stack address of the basicLock in case the
1360 // lock is still a light weight lock (lock_reg)
1361 assert(mark_addr.disp() == 0, "cas must take a zero displacement");
1362 cas_ptr(mark_addr.base(), lock_reg, displaced_header_reg);
1363 cmp(lock_reg, displaced_header_reg);
1364 brx(Assembler::equal, true, Assembler::pn, done);
1365 delayed()->st_ptr(G0, lockobj_addr); // free entry
1366
1367 // The lock has been converted into a heavy lock and hence
1368 // we need to get into the slow case
1369
1370 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1371
1372 bind(done);
1373 }
1374 }
1375
1376 // Get the method data pointer from the Method* and set the
1377 // specified register to its value.
1378
1379 void InterpreterMacroAssembler::set_method_data_pointer() {
1380 assert(ProfileInterpreter, "must be profiling interpreter");
1381 Label get_continue;
1382
1383 ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr);
1384 test_method_data_pointer(get_continue);
1385 add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr);
1386 bind(get_continue);
1387 }
1388
1389 // Set the method data pointer for the current bcp.
1390
1391 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1392 assert(ProfileInterpreter, "must be profiling interpreter");
1393 Label zero_continue;
1394
1395 // Test MDO to avoid the call if it is NULL.
1396 ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr);
1397 test_method_data_pointer(zero_continue);
1398 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), Lmethod, Lbcp);
1399 add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr);
1400 add(ImethodDataPtr, O0, ImethodDataPtr);
1401 bind(zero_continue);
1402 }
1403
1404 // Test ImethodDataPtr. If it is null, continue at the specified label
1405
1406 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) {
1407 assert(ProfileInterpreter, "must be profiling interpreter");
1408 br_null_short(ImethodDataPtr, Assembler::pn, zero_continue);
1409 }
1410
1411 void InterpreterMacroAssembler::verify_method_data_pointer() {
1412 assert(ProfileInterpreter, "must be profiling interpreter");
1413 #ifdef ASSERT
1414 Label verify_continue;
1415 test_method_data_pointer(verify_continue);
1416
1417 // If the mdp is valid, it will point to a DataLayout header which is
1418 // consistent with the bcp. The converse is highly probable also.
1419 lduh(ImethodDataPtr, in_bytes(DataLayout::bci_offset()), G3_scratch);
1420 ld_ptr(Lmethod, Method::const_offset(), O5);
1421 add(G3_scratch, in_bytes(ConstMethod::codes_offset()), G3_scratch);
1422 add(G3_scratch, O5, G3_scratch);
1423 cmp(Lbcp, G3_scratch);
1424 brx(Assembler::equal, false, Assembler::pt, verify_continue);
1425
1426 Register temp_reg = O5;
1427 delayed()->mov(ImethodDataPtr, temp_reg);
1428 // %%% should use call_VM_leaf here?
1429 //call_VM_leaf(noreg, ..., Lmethod, Lbcp, ImethodDataPtr);
1430 save_frame_and_mov(sizeof(jdouble) / wordSize, Lmethod, O0, Lbcp, O1);
1431 Address d_save(FP, -sizeof(jdouble) + STACK_BIAS);
1432 stf(FloatRegisterImpl::D, Ftos_d, d_save);
1433 mov(temp_reg->after_save(), O2);
1434 save_thread(L7_thread_cache);
1435 call(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), relocInfo::none);
1436 delayed()->nop();
1437 restore_thread(L7_thread_cache);
1438 ldf(FloatRegisterImpl::D, d_save, Ftos_d);
1439 restore();
1440 bind(verify_continue);
1441 #endif // ASSERT
1442 }
1443
1444 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count,
1445 Register method_counters,
1446 Register Rtmp,
1447 Label &profile_continue) {
1448 assert(ProfileInterpreter, "must be profiling interpreter");
1449 // Control will flow to "profile_continue" if the counter is less than the
1450 // limit or if we call profile_method()
1451
1452 Label done;
1453
1454 // if no method data exists, and the counter is high enough, make one
1455 br_notnull_short(ImethodDataPtr, Assembler::pn, done);
1456
1457 // Test to see if we should create a method data oop
1458 Address profile_limit(method_counters, MethodCounters::interpreter_profile_limit_offset());
1459 ld(profile_limit, Rtmp);
1460 cmp(invocation_count, Rtmp);
1461 // Use long branches because call_VM() code and following code generated by
1462 // test_backedge_count_for_osr() is large in debug VM.
1463 br(Assembler::lessUnsigned, false, Assembler::pn, profile_continue);
1464 delayed()->nop();
1465
1466 // Build it now.
1467 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1468 set_method_data_pointer_for_bcp();
1469 ba(profile_continue);
1470 delayed()->nop();
1471 bind(done);
1472 }
1473
1474 // Store a value at some constant offset from the method data pointer.
1475
1476 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) {
1477 assert(ProfileInterpreter, "must be profiling interpreter");
1478 st_ptr(value, ImethodDataPtr, constant);
1479 }
1480
1481 void InterpreterMacroAssembler::increment_mdp_data_at(Address counter,
1482 Register bumped_count,
1483 bool decrement) {
1484 assert(ProfileInterpreter, "must be profiling interpreter");
1485
1486 // Load the counter.
1487 ld_ptr(counter, bumped_count);
1488
1489 if (decrement) {
1490 // Decrement the register. Set condition codes.
1491 subcc(bumped_count, DataLayout::counter_increment, bumped_count);
1492
1493 // If the decrement causes the counter to overflow, stay negative
1494 Label L;
1495 brx(Assembler::negative, true, Assembler::pn, L);
1496
1497 // Store the decremented counter, if it is still negative.
1498 delayed()->st_ptr(bumped_count, counter);
1499 bind(L);
1500 } else {
1501 // Increment the register. Set carry flag.
1502 addcc(bumped_count, DataLayout::counter_increment, bumped_count);
1503
1504 // If the increment causes the counter to overflow, pull back by 1.
1505 assert(DataLayout::counter_increment == 1, "subc works");
1506 subc(bumped_count, G0, bumped_count);
1507
1508 // Store the incremented counter.
1509 st_ptr(bumped_count, counter);
1510 }
1511 }
1512
1513 // Increment the value at some constant offset from the method data pointer.
1514
1515 void InterpreterMacroAssembler::increment_mdp_data_at(int constant,
1516 Register bumped_count,
1517 bool decrement) {
1518 // Locate the counter at a fixed offset from the mdp:
1519 Address counter(ImethodDataPtr, constant);
1520 increment_mdp_data_at(counter, bumped_count, decrement);
1521 }
1522
1523 // Increment the value at some non-fixed (reg + constant) offset from
1524 // the method data pointer.
1525
1526 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg,
1527 int constant,
1528 Register bumped_count,
1529 Register scratch2,
1530 bool decrement) {
1531 // Add the constant to reg to get the offset.
1532 add(ImethodDataPtr, reg, scratch2);
1533 Address counter(scratch2, constant);
1534 increment_mdp_data_at(counter, bumped_count, decrement);
1535 }
1536
1537 // Set a flag value at the current method data pointer position.
1538 // Updates a single byte of the header, to avoid races with other header bits.
1539
1540 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant,
1541 Register scratch) {
1542 assert(ProfileInterpreter, "must be profiling interpreter");
1543 // Load the data header
1544 ldub(ImethodDataPtr, in_bytes(DataLayout::flags_offset()), scratch);
1545
1546 // Set the flag
1547 or3(scratch, flag_constant, scratch);
1548
1549 // Store the modified header.
1550 stb(scratch, ImethodDataPtr, in_bytes(DataLayout::flags_offset()));
1551 }
1552
1553 // Test the location at some offset from the method data pointer.
1554 // If it is not equal to value, branch to the not_equal_continue Label.
1555 // Set condition codes to match the nullness of the loaded value.
1556
1557 void InterpreterMacroAssembler::test_mdp_data_at(int offset,
1558 Register value,
1559 Label& not_equal_continue,
1560 Register scratch) {
1561 assert(ProfileInterpreter, "must be profiling interpreter");
1562 ld_ptr(ImethodDataPtr, offset, scratch);
1563 cmp(value, scratch);
1564 brx(Assembler::notEqual, false, Assembler::pn, not_equal_continue);
1565 delayed()->tst(scratch);
1566 }
1567
1568 // Update the method data pointer by the displacement located at some fixed
1569 // offset from the method data pointer.
1570
1571 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp,
1572 Register scratch) {
1573 assert(ProfileInterpreter, "must be profiling interpreter");
1574 ld_ptr(ImethodDataPtr, offset_of_disp, scratch);
1575 add(ImethodDataPtr, scratch, ImethodDataPtr);
1576 }
1577
1578 // Update the method data pointer by the displacement located at the
1579 // offset (reg + offset_of_disp).
1580
1581 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg,
1582 int offset_of_disp,
1583 Register scratch) {
1584 assert(ProfileInterpreter, "must be profiling interpreter");
1585 add(reg, offset_of_disp, scratch);
1586 ld_ptr(ImethodDataPtr, scratch, scratch);
1587 add(ImethodDataPtr, scratch, ImethodDataPtr);
1588 }
1589
1590 // Update the method data pointer by a simple constant displacement.
1591
1592 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) {
1593 assert(ProfileInterpreter, "must be profiling interpreter");
1594 add(ImethodDataPtr, constant, ImethodDataPtr);
1595 }
1596
1597 // Update the method data pointer for a _ret bytecode whose target
1598 // was not among our cached targets.
1599
1600 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state,
1601 Register return_bci) {
1602 assert(ProfileInterpreter, "must be profiling interpreter");
1603 push(state);
1604 st_ptr(return_bci, l_tmp); // protect return_bci, in case it is volatile
1605 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci);
1606 ld_ptr(l_tmp, return_bci);
1607 pop(state);
1608 }
1609
1610 // Count a taken branch in the bytecodes.
1611
1612 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) {
1613 if (ProfileInterpreter) {
1614 Label profile_continue;
1615
1616 // If no method data exists, go to profile_continue.
1617 test_method_data_pointer(profile_continue);
1618
1619 // We are taking a branch. Increment the taken count.
1620 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), bumped_count);
1621
1622 // The method data pointer needs to be updated to reflect the new target.
1623 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch);
1624 bind (profile_continue);
1625 }
1626 }
1627
1628
1629 // Count a not-taken branch in the bytecodes.
1630
1631 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch) {
1632 if (ProfileInterpreter) {
1633 Label profile_continue;
1634
1635 // If no method data exists, go to profile_continue.
1636 test_method_data_pointer(profile_continue);
1637
1638 // We are taking a branch. Increment the not taken count.
1639 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch);
1640
1641 // The method data pointer needs to be updated to correspond to the
1642 // next bytecode.
1643 update_mdp_by_constant(in_bytes(BranchData::branch_data_size()));
1644 bind (profile_continue);
1645 }
1646 }
1647
1648
1649 // Count a non-virtual call in the bytecodes.
1650
1651 void InterpreterMacroAssembler::profile_call(Register scratch) {
1652 if (ProfileInterpreter) {
1653 Label profile_continue;
1654
1655 // If no method data exists, go to profile_continue.
1656 test_method_data_pointer(profile_continue);
1657
1658 // We are making a call. Increment the count.
1659 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1660
1661 // The method data pointer needs to be updated to reflect the new target.
1662 update_mdp_by_constant(in_bytes(CounterData::counter_data_size()));
1663 bind (profile_continue);
1664 }
1665 }
1666
1667
1668 // Count a final call in the bytecodes.
1669
1670 void InterpreterMacroAssembler::profile_final_call(Register scratch) {
1671 if (ProfileInterpreter) {
1672 Label profile_continue;
1673
1674 // If no method data exists, go to profile_continue.
1675 test_method_data_pointer(profile_continue);
1676
1677 // We are making a call. Increment the count.
1678 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1679
1680 // The method data pointer needs to be updated to reflect the new target.
1681 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1682 bind (profile_continue);
1683 }
1684 }
1685
1686
1687 // Count a virtual call in the bytecodes.
1688
1689 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1690 Register scratch,
1691 bool receiver_can_be_null) {
1692 if (ProfileInterpreter) {
1693 Label profile_continue;
1694
1695 // If no method data exists, go to profile_continue.
1696 test_method_data_pointer(profile_continue);
1697
1698
1699 Label skip_receiver_profile;
1700 if (receiver_can_be_null) {
1701 Label not_null;
1702 br_notnull_short(receiver, Assembler::pt, not_null);
1703 // We are making a call. Increment the count for null receiver.
1704 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1705 ba_short(skip_receiver_profile);
1706 bind(not_null);
1707 }
1708
1709 // Record the receiver type.
1710 record_klass_in_profile(receiver, scratch, true);
1711 bind(skip_receiver_profile);
1712
1713 // The method data pointer needs to be updated to reflect the new target.
1714 #if INCLUDE_JVMCI
1715 if (MethodProfileWidth == 0) {
1716 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1717 }
1718 #else
1719 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1720 #endif
1721 bind(profile_continue);
1722 }
1723 }
1724
1725 #if INCLUDE_JVMCI
1726 void InterpreterMacroAssembler::profile_called_method(Register method, Register scratch) {
1727 assert_different_registers(method, scratch);
1728 if (ProfileInterpreter && MethodProfileWidth > 0) {
1729 Label profile_continue;
1730
1731 // If no method data exists, go to profile_continue.
1732 test_method_data_pointer(profile_continue);
1733
1734 Label done;
1735 record_item_in_profile_helper(method, scratch, 0, done, MethodProfileWidth,
1736 &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset()));
1737 bind(done);
1738
1739 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1740 bind(profile_continue);
1741 }
1742 }
1743 #endif // INCLUDE_JVMCI
1744
1745 void InterpreterMacroAssembler::record_klass_in_profile_helper(Register receiver, Register scratch,
1746 Label& done, bool is_virtual_call) {
1747 if (TypeProfileWidth == 0) {
1748 if (is_virtual_call) {
1749 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1750 }
1751 #if INCLUDE_JVMCI
1752 else if (EnableJVMCI) {
1753 increment_mdp_data_at(in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()), scratch);
1754 }
1755 #endif
1756 } else {
1757 int non_profiled_offset = -1;
1758 if (is_virtual_call) {
1759 non_profiled_offset = in_bytes(CounterData::count_offset());
1760 }
1761 #if INCLUDE_JVMCI
1762 else if (EnableJVMCI) {
1763 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1764 }
1765 #endif
1766
1767 record_item_in_profile_helper(receiver, scratch, 0, done, TypeProfileWidth,
1768 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1769 }
1770 }
1771
1772 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item,
1773 Register scratch, int start_row, Label& done, int total_rows,
1774 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1775 int non_profiled_offset) {
1776 int last_row = total_rows - 1;
1777 assert(start_row <= last_row, "must be work left to do");
1778 // Test this row for both the item and for null.
1779 // Take any of three different outcomes:
1780 // 1. found item => increment count and goto done
1781 // 2. found null => keep looking for case 1, maybe allocate this cell
1782 // 3. found something else => keep looking for cases 1 and 2
1783 // Case 3 is handled by a recursive call.
1784 for (int row = start_row; row <= last_row; row++) {
1785 Label next_test;
1786 bool test_for_null_also = (row == start_row);
1787
1788 // See if the item is item[n].
1789 int item_offset = in_bytes(item_offset_fn(row));
1790 test_mdp_data_at(item_offset, item, next_test, scratch);
1791 // delayed()->tst(scratch);
1792
1793 // The receiver is item[n]. Increment count[n].
1794 int count_offset = in_bytes(item_count_offset_fn(row));
1795 increment_mdp_data_at(count_offset, scratch);
1796 ba_short(done);
1797 bind(next_test);
1798
1799 if (test_for_null_also) {
1800 Label found_null;
1801 // Failed the equality check on item[n]... Test for null.
1802 if (start_row == last_row) {
1803 // The only thing left to do is handle the null case.
1804 if (non_profiled_offset >= 0) {
1805 brx(Assembler::zero, false, Assembler::pn, found_null);
1806 delayed()->nop();
1807 // Item did not match any saved item and there is no empty row for it.
1808 // Increment total counter to indicate polymorphic case.
1809 increment_mdp_data_at(non_profiled_offset, scratch);
1810 ba_short(done);
1811 bind(found_null);
1812 } else {
1813 brx(Assembler::notZero, false, Assembler::pt, done);
1814 delayed()->nop();
1815 }
1816 break;
1817 }
1818 // Since null is rare, make it be the branch-taken case.
1819 brx(Assembler::zero, false, Assembler::pn, found_null);
1820 delayed()->nop();
1821
1822 // Put all the "Case 3" tests here.
1823 record_item_in_profile_helper(item, scratch, start_row + 1, done, total_rows,
1824 item_offset_fn, item_count_offset_fn, non_profiled_offset);
1825
1826 // Found a null. Keep searching for a matching item,
1827 // but remember that this is an empty (unused) slot.
1828 bind(found_null);
1829 }
1830 }
1831
1832 // In the fall-through case, we found no matching item, but we
1833 // observed the item[start_row] is NULL.
1834
1835 // Fill in the item field and increment the count.
1836 int item_offset = in_bytes(item_offset_fn(start_row));
1837 set_mdp_data_at(item_offset, item);
1838 int count_offset = in_bytes(item_count_offset_fn(start_row));
1839 mov(DataLayout::counter_increment, scratch);
1840 set_mdp_data_at(count_offset, scratch);
1841 if (start_row > 0) {
1842 ba_short(done);
1843 }
1844 }
1845
1846 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1847 Register scratch, bool is_virtual_call) {
1848 assert(ProfileInterpreter, "must be profiling");
1849 Label done;
1850
1851 record_klass_in_profile_helper(receiver, scratch, done, is_virtual_call);
1852
1853 bind (done);
1854 }
1855
1856
1857 // Count a ret in the bytecodes.
1858
1859 void InterpreterMacroAssembler::profile_ret(TosState state,
1860 Register return_bci,
1861 Register scratch) {
1862 if (ProfileInterpreter) {
1863 Label profile_continue;
1864 uint row;
1865
1866 // If no method data exists, go to profile_continue.
1867 test_method_data_pointer(profile_continue);
1868
1869 // Update the total ret count.
1870 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1871
1872 for (row = 0; row < RetData::row_limit(); row++) {
1873 Label next_test;
1874
1875 // See if return_bci is equal to bci[n]:
1876 test_mdp_data_at(in_bytes(RetData::bci_offset(row)),
1877 return_bci, next_test, scratch);
1878
1879 // return_bci is equal to bci[n]. Increment the count.
1880 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch);
1881
1882 // The method data pointer needs to be updated to reflect the new target.
1883 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch);
1884 ba_short(profile_continue);
1885 bind(next_test);
1886 }
1887
1888 update_mdp_for_ret(state, return_bci);
1889
1890 bind (profile_continue);
1891 }
1892 }
1893
1894 // Profile an unexpected null in the bytecodes.
1895 void InterpreterMacroAssembler::profile_null_seen(Register scratch) {
1896 if (ProfileInterpreter) {
1897 Label profile_continue;
1898
1899 // If no method data exists, go to profile_continue.
1900 test_method_data_pointer(profile_continue);
1901
1902 set_mdp_flag_at(BitData::null_seen_byte_constant(), scratch);
1903
1904 // The method data pointer needs to be updated.
1905 int mdp_delta = in_bytes(BitData::bit_data_size());
1906 if (TypeProfileCasts) {
1907 mdp_delta = in_bytes(ReceiverTypeData::receiver_type_data_size());
1908 }
1909 update_mdp_by_constant(mdp_delta);
1910
1911 bind (profile_continue);
1912 }
1913 }
1914
1915 void InterpreterMacroAssembler::profile_typecheck(Register klass,
1916 Register scratch) {
1917 if (ProfileInterpreter) {
1918 Label profile_continue;
1919
1920 // If no method data exists, go to profile_continue.
1921 test_method_data_pointer(profile_continue);
1922
1923 int mdp_delta = in_bytes(BitData::bit_data_size());
1924 if (TypeProfileCasts) {
1925 mdp_delta = in_bytes(ReceiverTypeData::receiver_type_data_size());
1926
1927 // Record the object type.
1928 record_klass_in_profile(klass, scratch, false);
1929 }
1930
1931 // The method data pointer needs to be updated.
1932 update_mdp_by_constant(mdp_delta);
1933
1934 bind (profile_continue);
1935 }
1936 }
1937
1938 void InterpreterMacroAssembler::profile_typecheck_failed(Register scratch) {
1939 if (ProfileInterpreter && TypeProfileCasts) {
1940 Label profile_continue;
1941
1942 // If no method data exists, go to profile_continue.
1943 test_method_data_pointer(profile_continue);
1944
1945 int count_offset = in_bytes(CounterData::count_offset());
1946 // Back up the address, since we have already bumped the mdp.
1947 count_offset -= in_bytes(ReceiverTypeData::receiver_type_data_size());
1948
1949 // *Decrement* the counter. We expect to see zero or small negatives.
1950 increment_mdp_data_at(count_offset, scratch, true);
1951
1952 bind (profile_continue);
1953 }
1954 }
1955
1956 // Count the default case of a switch construct.
1957
1958 void InterpreterMacroAssembler::profile_switch_default(Register scratch) {
1959 if (ProfileInterpreter) {
1960 Label profile_continue;
1961
1962 // If no method data exists, go to profile_continue.
1963 test_method_data_pointer(profile_continue);
1964
1965 // Update the default case count
1966 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()),
1967 scratch);
1968
1969 // The method data pointer needs to be updated.
1970 update_mdp_by_offset(
1971 in_bytes(MultiBranchData::default_displacement_offset()),
1972 scratch);
1973
1974 bind (profile_continue);
1975 }
1976 }
1977
1978 // Count the index'th case of a switch construct.
1979
1980 void InterpreterMacroAssembler::profile_switch_case(Register index,
1981 Register scratch,
1982 Register scratch2,
1983 Register scratch3) {
1984 if (ProfileInterpreter) {
1985 Label profile_continue;
1986
1987 // If no method data exists, go to profile_continue.
1988 test_method_data_pointer(profile_continue);
1989
1990 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes()
1991 set(in_bytes(MultiBranchData::per_case_size()), scratch);
1992 smul(index, scratch, scratch);
1993 add(scratch, in_bytes(MultiBranchData::case_array_offset()), scratch);
1994
1995 // Update the case count
1996 increment_mdp_data_at(scratch,
1997 in_bytes(MultiBranchData::relative_count_offset()),
1998 scratch2,
1999 scratch3);
2000
2001 // The method data pointer needs to be updated.
2002 update_mdp_by_offset(scratch,
2003 in_bytes(MultiBranchData::relative_displacement_offset()),
2004 scratch2);
2005
2006 bind (profile_continue);
2007 }
2008 }
2009
2010 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr, Register tmp) {
2011 Label not_null, do_nothing, do_update;
2012
2013 assert_different_registers(obj, mdo_addr.base(), tmp);
2014
2015 verify_oop(obj);
2016
2017 ld_ptr(mdo_addr, tmp);
2018
2019 br_notnull_short(obj, pt, not_null);
2020 or3(tmp, TypeEntries::null_seen, tmp);
2021 ba_short(do_update);
2022
2023 bind(not_null);
2024 load_klass(obj, obj);
2025
2026 xor3(obj, tmp, obj);
2027 btst(TypeEntries::type_klass_mask, obj);
2028 // klass seen before, nothing to do. The unknown bit may have been
2029 // set already but no need to check.
2030 brx(zero, false, pt, do_nothing);
2031 delayed()->
2032
2033 btst(TypeEntries::type_unknown, obj);
2034 // already unknown. Nothing to do anymore.
2035 brx(notZero, false, pt, do_nothing);
2036 delayed()->
2037
2038 btst(TypeEntries::type_mask, tmp);
2039 brx(zero, true, pt, do_update);
2040 // first time here. Set profile type.
2041 delayed()->or3(tmp, obj, tmp);
2042
2043 // different than before. Cannot keep accurate profile.
2044 or3(tmp, TypeEntries::type_unknown, tmp);
2045
2046 bind(do_update);
2047 // update profile
2048 st_ptr(tmp, mdo_addr);
2049
2050 bind(do_nothing);
2051 }
2052
2053 void InterpreterMacroAssembler::profile_arguments_type(Register callee, Register tmp1, Register tmp2, bool is_virtual) {
2054 if (!ProfileInterpreter) {
2055 return;
2056 }
2057
2058 assert_different_registers(callee, tmp1, tmp2, ImethodDataPtr);
2059
2060 if (MethodData::profile_arguments() || MethodData::profile_return()) {
2061 Label profile_continue;
2062
2063 test_method_data_pointer(profile_continue);
2064
2065 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
2066
2067 ldub(ImethodDataPtr, in_bytes(DataLayout::tag_offset()) - off_to_start, tmp1);
2068 cmp_and_br_short(tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag, notEqual, pn, profile_continue);
2069
2070 if (MethodData::profile_arguments()) {
2071 Label done;
2072 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
2073 add(ImethodDataPtr, off_to_args, ImethodDataPtr);
2074
2075 for (int i = 0; i < TypeProfileArgsLimit; i++) {
2076 if (i > 0 || MethodData::profile_return()) {
2077 // If return value type is profiled we may have no argument to profile
2078 ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, tmp1);
2079 sub(tmp1, i*TypeStackSlotEntries::per_arg_count(), tmp1);
2080 cmp_and_br_short(tmp1, TypeStackSlotEntries::per_arg_count(), less, pn, done);
2081 }
2082 ld_ptr(Address(callee, Method::const_offset()), tmp1);
2083 lduh(Address(tmp1, ConstMethod::size_of_parameters_offset()), tmp1);
2084 // stack offset o (zero based) from the start of the argument
2085 // list, for n arguments translates into offset n - o - 1 from
2086 // the end of the argument list. But there's an extra slot at
2087 // the stop of the stack. So the offset is n - o from Lesp.
2088 ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, tmp2);
2089 sub(tmp1, tmp2, tmp1);
2090
2091 // Can't use MacroAssembler::argument_address() which needs Gargs to be set up
2092 sll(tmp1, Interpreter::logStackElementSize, tmp1);
2093 ld_ptr(Lesp, tmp1, tmp1);
2094
2095 Address mdo_arg_addr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
2096 profile_obj_type(tmp1, mdo_arg_addr, tmp2);
2097
2098 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
2099 add(ImethodDataPtr, to_add, ImethodDataPtr);
2100 off_to_args += to_add;
2101 }
2102
2103 if (MethodData::profile_return()) {
2104 ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, tmp1);
2105 sub(tmp1, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count(), tmp1);
2106 }
2107
2108 bind(done);
2109
2110 if (MethodData::profile_return()) {
2111 // We're right after the type profile for the last
2112 // argument. tmp1 is the number of cells left in the
2113 // CallTypeData/VirtualCallTypeData to reach its end. Non null
2114 // if there's a return to profile.
2115 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
2116 sll(tmp1, exact_log2(DataLayout::cell_size), tmp1);
2117 add(ImethodDataPtr, tmp1, ImethodDataPtr);
2118 }
2119 } else {
2120 assert(MethodData::profile_return(), "either profile call args or call ret");
2121 update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size()));
2122 }
2123
2124 // mdp points right after the end of the
2125 // CallTypeData/VirtualCallTypeData, right after the cells for the
2126 // return value type if there's one.
2127
2128 bind(profile_continue);
2129 }
2130 }
2131
2132 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) {
2133 assert_different_registers(ret, tmp1, tmp2);
2134 if (ProfileInterpreter && MethodData::profile_return()) {
2135 Label profile_continue, done;
2136
2137 test_method_data_pointer(profile_continue);
2138
2139 if (MethodData::profile_return_jsr292_only()) {
2140 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
2141
2142 // If we don't profile all invoke bytecodes we must make sure
2143 // it's a bytecode we indeed profile. We can't go back to the
2144 // begining of the ProfileData we intend to update to check its
2145 // type because we're right after it and we don't known its
2146 // length.
2147 Label do_profile;
2148 ldub(Lbcp, 0, tmp1);
2149 cmp_and_br_short(tmp1, Bytecodes::_invokedynamic, equal, pn, do_profile);
2150 cmp(tmp1, Bytecodes::_invokehandle);
2151 br(equal, false, pn, do_profile);
2152 delayed()->lduh(Lmethod, Method::intrinsic_id_offset_in_bytes(), tmp1);
2153 cmp_and_br_short(tmp1, vmIntrinsics::_compiledLambdaForm, notEqual, pt, profile_continue);
2154
2155 bind(do_profile);
2156 }
2157
2158 Address mdo_ret_addr(ImethodDataPtr, -in_bytes(ReturnTypeEntry::size()));
2159 mov(ret, tmp1);
2160 profile_obj_type(tmp1, mdo_ret_addr, tmp2);
2161
2162 bind(profile_continue);
2163 }
2164 }
2165
2166 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
2167 if (ProfileInterpreter && MethodData::profile_parameters()) {
2168 Label profile_continue, done;
2169
2170 test_method_data_pointer(profile_continue);
2171
2172 // Load the offset of the area within the MDO used for
2173 // parameters. If it's negative we're not profiling any parameters.
2174 lduw(ImethodDataPtr, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), tmp1);
2175 cmp_and_br_short(tmp1, 0, less, pn, profile_continue);
2176
2177 // Compute a pointer to the area for parameters from the offset
2178 // and move the pointer to the slot for the last
2179 // parameters. Collect profiling from last parameter down.
2180 // mdo start + parameters offset + array length - 1
2181
2182 // Pointer to the parameter area in the MDO
2183 Register mdp = tmp1;
2184 add(ImethodDataPtr, tmp1, mdp);
2185
2186 // offset of the current profile entry to update
2187 Register entry_offset = tmp2;
2188 // entry_offset = array len in number of cells
2189 ld_ptr(mdp, ArrayData::array_len_offset(), entry_offset);
2190
2191 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
2192 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells");
2193
2194 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field
2195 sub(entry_offset, TypeStackSlotEntries::per_arg_count() - (off_base / DataLayout::cell_size), entry_offset);
2196 // entry_offset in bytes
2197 sll(entry_offset, exact_log2(DataLayout::cell_size), entry_offset);
2198
2199 Label loop;
2200 bind(loop);
2201
2202 // load offset on the stack from the slot for this parameter
2203 ld_ptr(mdp, entry_offset, tmp3);
2204 sll(tmp3,Interpreter::logStackElementSize, tmp3);
2205 neg(tmp3);
2206 // read the parameter from the local area
2207 ld_ptr(Llocals, tmp3, tmp3);
2208
2209 // make entry_offset now point to the type field for this parameter
2210 int type_base = in_bytes(ParametersTypeData::type_offset(0));
2211 assert(type_base > off_base, "unexpected");
2212 add(entry_offset, type_base - off_base, entry_offset);
2213
2214 // profile the parameter
2215 Address arg_type(mdp, entry_offset);
2216 profile_obj_type(tmp3, arg_type, tmp4);
2217
2218 // go to next parameter
2219 sub(entry_offset, TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base), entry_offset);
2220 cmp_and_br_short(entry_offset, off_base, greaterEqual, pt, loop);
2221
2222 bind(profile_continue);
2223 }
2224 }
2225
2226 // add a InterpMonitorElem to stack (see frame_sparc.hpp)
2227
2228 void InterpreterMacroAssembler::add_monitor_to_stack( bool stack_is_empty,
2229 Register Rtemp,
2230 Register Rtemp2 ) {
2231
2232 Register Rlimit = Lmonitors;
2233 const jint delta = frame::interpreter_frame_monitor_size() * wordSize;
2234 assert( (delta & LongAlignmentMask) == 0,
2235 "sizeof BasicObjectLock must be even number of doublewords");
2236
2237 sub( SP, delta, SP);
2238 sub( Lesp, delta, Lesp);
2239 sub( Lmonitors, delta, Lmonitors);
2240
2241 if (!stack_is_empty) {
2242
2243 // must copy stack contents down
2244
2245 Label start_copying, next;
2246
2247 // untested("monitor stack expansion");
2248 compute_stack_base(Rtemp);
2249 ba(start_copying);
2250 delayed()->cmp(Rtemp, Rlimit); // done? duplicated below
2251
2252 // note: must copy from low memory upwards
2253 // On entry to loop,
2254 // Rtemp points to new base of stack, Lesp points to new end of stack (1 past TOS)
2255 // Loop mutates Rtemp
2256
2257 bind( next);
2258
2259 st_ptr(Rtemp2, Rtemp, 0);
2260 inc(Rtemp, wordSize);
2261 cmp(Rtemp, Rlimit); // are we done? (duplicated above)
2262
2263 bind( start_copying );
2264
2265 brx( notEqual, true, pn, next );
2266 delayed()->ld_ptr( Rtemp, delta, Rtemp2 );
2267
2268 // done copying stack
2269 }
2270 }
2271
2272 // Locals
2273 void InterpreterMacroAssembler::access_local_ptr( Register index, Register dst ) {
2274 assert_not_delayed();
2275 sll(index, Interpreter::logStackElementSize, index);
2276 sub(Llocals, index, index);
2277 ld_ptr(index, 0, dst);
2278 // Note: index must hold the effective address--the iinc template uses it
2279 }
2280
2281 // Just like access_local_ptr but the tag is a returnAddress
2282 void InterpreterMacroAssembler::access_local_returnAddress(Register index,
2283 Register dst ) {
2284 assert_not_delayed();
2285 sll(index, Interpreter::logStackElementSize, index);
2286 sub(Llocals, index, index);
2287 ld_ptr(index, 0, dst);
2288 }
2289
2290 void InterpreterMacroAssembler::access_local_int( Register index, Register dst ) {
2291 assert_not_delayed();
2292 sll(index, Interpreter::logStackElementSize, index);
2293 sub(Llocals, index, index);
2294 ld(index, 0, dst);
2295 // Note: index must hold the effective address--the iinc template uses it
2296 }
2297
2298
2299 void InterpreterMacroAssembler::access_local_long( Register index, Register dst ) {
2300 assert_not_delayed();
2301 sll(index, Interpreter::logStackElementSize, index);
2302 sub(Llocals, index, index);
2303 // First half stored at index n+1 (which grows down from Llocals[n])
2304 load_unaligned_long(index, Interpreter::local_offset_in_bytes(1), dst);
2305 }
2306
2307
2308 void InterpreterMacroAssembler::access_local_float( Register index, FloatRegister dst ) {
2309 assert_not_delayed();
2310 sll(index, Interpreter::logStackElementSize, index);
2311 sub(Llocals, index, index);
2312 ldf(FloatRegisterImpl::S, index, 0, dst);
2313 }
2314
2315
2316 void InterpreterMacroAssembler::access_local_double( Register index, FloatRegister dst ) {
2317 assert_not_delayed();
2318 sll(index, Interpreter::logStackElementSize, index);
2319 sub(Llocals, index, index);
2320 load_unaligned_double(index, Interpreter::local_offset_in_bytes(1), dst);
2321 }
2322
2323
2324 #ifdef ASSERT
2325 void InterpreterMacroAssembler::check_for_regarea_stomp(Register Rindex, int offset, Register Rlimit, Register Rscratch, Register Rscratch1) {
2326 Label L;
2327
2328 assert(Rindex != Rscratch, "Registers cannot be same");
2329 assert(Rindex != Rscratch1, "Registers cannot be same");
2330 assert(Rlimit != Rscratch, "Registers cannot be same");
2331 assert(Rlimit != Rscratch1, "Registers cannot be same");
2332 assert(Rscratch1 != Rscratch, "Registers cannot be same");
2333
2334 // untested("reg area corruption");
2335 add(Rindex, offset, Rscratch);
2336 add(Rlimit, 64 + STACK_BIAS, Rscratch1);
2337 cmp_and_brx_short(Rscratch, Rscratch1, Assembler::greaterEqualUnsigned, pn, L);
2338 stop("regsave area is being clobbered");
2339 bind(L);
2340 }
2341 #endif // ASSERT
2342
2343
2344 void InterpreterMacroAssembler::store_local_int( Register index, Register src ) {
2345 assert_not_delayed();
2346 sll(index, Interpreter::logStackElementSize, index);
2347 sub(Llocals, index, index);
2348 debug_only(check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);)
2349 st(src, index, 0);
2350 }
2351
2352 void InterpreterMacroAssembler::store_local_ptr( Register index, Register src ) {
2353 assert_not_delayed();
2354 sll(index, Interpreter::logStackElementSize, index);
2355 sub(Llocals, index, index);
2356 #ifdef ASSERT
2357 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);
2358 #endif
2359 st_ptr(src, index, 0);
2360 }
2361
2362
2363
2364 void InterpreterMacroAssembler::store_local_ptr( int n, Register src ) {
2365 st_ptr(src, Llocals, Interpreter::local_offset_in_bytes(n));
2366 }
2367
2368 void InterpreterMacroAssembler::store_local_long( Register index, Register src ) {
2369 assert_not_delayed();
2370 sll(index, Interpreter::logStackElementSize, index);
2371 sub(Llocals, index, index);
2372 #ifdef ASSERT
2373 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch);
2374 #endif
2375 store_unaligned_long(src, index, Interpreter::local_offset_in_bytes(1)); // which is n+1
2376 }
2377
2378
2379 void InterpreterMacroAssembler::store_local_float( Register index, FloatRegister src ) {
2380 assert_not_delayed();
2381 sll(index, Interpreter::logStackElementSize, index);
2382 sub(Llocals, index, index);
2383 #ifdef ASSERT
2384 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);
2385 #endif
2386 stf(FloatRegisterImpl::S, src, index, 0);
2387 }
2388
2389
2390 void InterpreterMacroAssembler::store_local_double( Register index, FloatRegister src ) {
2391 assert_not_delayed();
2392 sll(index, Interpreter::logStackElementSize, index);
2393 sub(Llocals, index, index);
2394 #ifdef ASSERT
2395 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch);
2396 #endif
2397 store_unaligned_double(src, index, Interpreter::local_offset_in_bytes(1));
2398 }
2399
2400
2401 int InterpreterMacroAssembler::top_most_monitor_byte_offset() {
2402 const jint delta = frame::interpreter_frame_monitor_size() * wordSize;
2403 int rounded_vm_local_words = ::round_to(frame::interpreter_frame_vm_local_words, WordsPerLong);
2404 return ((-rounded_vm_local_words * wordSize) - delta ) + STACK_BIAS;
2405 }
2406
2407
2408 Address InterpreterMacroAssembler::top_most_monitor() {
2409 return Address(FP, top_most_monitor_byte_offset());
2410 }
2411
2412
2413 void InterpreterMacroAssembler::compute_stack_base( Register Rdest ) {
2414 add( Lesp, wordSize, Rdest );
2415 }
2416
2417 void InterpreterMacroAssembler::get_method_counters(Register method,
2418 Register Rcounters,
2419 Label& skip) {
2420 Label has_counters;
2421 Address method_counters(method, in_bytes(Method::method_counters_offset()));
2422 ld_ptr(method_counters, Rcounters);
2423 br_notnull_short(Rcounters, Assembler::pt, has_counters);
2424 call_VM(noreg, CAST_FROM_FN_PTR(address,
2425 InterpreterRuntime::build_method_counters), method);
2426 ld_ptr(method_counters, Rcounters);
2427 br_null(Rcounters, false, Assembler::pn, skip); // No MethodCounters, OutOfMemory
2428 delayed()->nop();
2429 bind(has_counters);
2430 }
2431
2432 void InterpreterMacroAssembler::increment_invocation_counter( Register Rcounters, Register Rtmp, Register Rtmp2 ) {
2433 assert(UseCompiler || LogTouchedMethods, "incrementing must be useful");
2434 assert_different_registers(Rcounters, Rtmp, Rtmp2);
2435
2436 Address inv_counter(Rcounters, MethodCounters::invocation_counter_offset() +
2437 InvocationCounter::counter_offset());
2438 Address be_counter (Rcounters, MethodCounters::backedge_counter_offset() +
2439 InvocationCounter::counter_offset());
2440 int delta = InvocationCounter::count_increment;
2441
2442 // Load each counter in a register
2443 ld( inv_counter, Rtmp );
2444 ld( be_counter, Rtmp2 );
2445
2446 assert( is_simm13( delta ), " delta too large.");
2447
2448 // Add the delta to the invocation counter and store the result
2449 add( Rtmp, delta, Rtmp );
2450
2451 // Mask the backedge counter
2452 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 );
2453
2454 // Store value
2455 st( Rtmp, inv_counter);
2456
2457 // Add invocation counter + backedge counter
2458 add( Rtmp, Rtmp2, Rtmp);
2459
2460 // Note that this macro must leave the backedge_count + invocation_count in Rtmp!
2461 }
2462
2463 void InterpreterMacroAssembler::increment_backedge_counter( Register Rcounters, Register Rtmp, Register Rtmp2 ) {
2464 assert(UseCompiler, "incrementing must be useful");
2465 assert_different_registers(Rcounters, Rtmp, Rtmp2);
2466
2467 Address be_counter (Rcounters, MethodCounters::backedge_counter_offset() +
2468 InvocationCounter::counter_offset());
2469 Address inv_counter(Rcounters, MethodCounters::invocation_counter_offset() +
2470 InvocationCounter::counter_offset());
2471
2472 int delta = InvocationCounter::count_increment;
2473 // Load each counter in a register
2474 ld( be_counter, Rtmp );
2475 ld( inv_counter, Rtmp2 );
2476
2477 // Add the delta to the backedge counter
2478 add( Rtmp, delta, Rtmp );
2479
2480 // Mask the invocation counter, add to backedge counter
2481 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 );
2482
2483 // and store the result to memory
2484 st( Rtmp, be_counter );
2485
2486 // Add backedge + invocation counter
2487 add( Rtmp, Rtmp2, Rtmp );
2488
2489 // Note that this macro must leave backedge_count + invocation_count in Rtmp!
2490 }
2491
2492 void InterpreterMacroAssembler::test_backedge_count_for_osr( Register backedge_count,
2493 Register method_counters,
2494 Register branch_bcp,
2495 Register Rtmp ) {
2496 Label did_not_overflow;
2497 Label overflow_with_error;
2498 assert_different_registers(backedge_count, Rtmp, branch_bcp);
2499 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr");
2500
2501 Address limit(method_counters, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset()));
2502 ld(limit, Rtmp);
2503 cmp_and_br_short(backedge_count, Rtmp, Assembler::lessUnsigned, Assembler::pt, did_not_overflow);
2504
2505 // When ProfileInterpreter is on, the backedge_count comes from the
2506 // MethodData*, which value does not get reset on the call to
2507 // frequency_counter_overflow(). To avoid excessive calls to the overflow
2508 // routine while the method is being compiled, add a second test to make sure
2509 // the overflow function is called only once every overflow_frequency.
2510 if (ProfileInterpreter) {
2511 const int overflow_frequency = 1024;
2512 andcc(backedge_count, overflow_frequency-1, Rtmp);
2513 brx(Assembler::notZero, false, Assembler::pt, did_not_overflow);
2514 delayed()->nop();
2515 }
2516
2517 // overflow in loop, pass branch bytecode
2518 set(6,Rtmp);
2519 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, Rtmp);
2520
2521 // Was an OSR adapter generated?
2522 // O0 = osr nmethod
2523 br_null_short(O0, Assembler::pn, overflow_with_error);
2524
2525 // Has the nmethod been invalidated already?
2526 ldub(O0, nmethod::state_offset(), O2);
2527 cmp_and_br_short(O2, nmethod::in_use, Assembler::notEqual, Assembler::pn, overflow_with_error);
2528
2529 // migrate the interpreter frame off of the stack
2530
2531 mov(G2_thread, L7);
2532 // save nmethod
2533 mov(O0, L6);
2534 set_last_Java_frame(SP, noreg);
2535 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7);
2536 reset_last_Java_frame();
2537 mov(L7, G2_thread);
2538
2539 // move OSR nmethod to I1
2540 mov(L6, I1);
2541
2542 // OSR buffer to I0
2543 mov(O0, I0);
2544
2545 // remove the interpreter frame
2546 restore(I5_savedSP, 0, SP);
2547
2548 // Jump to the osr code.
2549 ld_ptr(O1, nmethod::osr_entry_point_offset(), O2);
2550 jmp(O2, G0);
2551 delayed()->nop();
2552
2553 bind(overflow_with_error);
2554
2555 bind(did_not_overflow);
2556 }
2557
2558
2559
2560 void InterpreterMacroAssembler::interp_verify_oop(Register reg, TosState state, const char * file, int line) {
2561 if (state == atos) { MacroAssembler::_verify_oop(reg, "broken oop ", file, line); }
2562 }
2563
2564
2565 // local helper function for the verify_oop_or_return_address macro
2566 static bool verify_return_address(Method* m, int bci) {
2567 #ifndef PRODUCT
2568 address pc = (address)(m->constMethod())
2569 + in_bytes(ConstMethod::codes_offset()) + bci;
2570 // assume it is a valid return address if it is inside m and is preceded by a jsr
2571 if (!m->contains(pc)) return false;
2572 address jsr_pc;
2573 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr);
2574 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true;
2575 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w);
2576 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true;
2577 #endif // PRODUCT
2578 return false;
2579 }
2580
2581
2582 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) {
2583 if (!VerifyOops) return;
2584 // the VM documentation for the astore[_wide] bytecode allows
2585 // the TOS to be not only an oop but also a return address
2586 Label test;
2587 Label skip;
2588 // See if it is an address (in the current method):
2589
2590 mov(reg, Rtmp);
2591 const int log2_bytecode_size_limit = 16;
2592 srl(Rtmp, log2_bytecode_size_limit, Rtmp);
2593 br_notnull_short( Rtmp, pt, test );
2594
2595 // %%% should use call_VM_leaf here?
2596 save_frame_and_mov(0, Lmethod, O0, reg, O1);
2597 save_thread(L7_thread_cache);
2598 call(CAST_FROM_FN_PTR(address,verify_return_address), relocInfo::none);
2599 delayed()->nop();
2600 restore_thread(L7_thread_cache);
2601 br_notnull( O0, false, pt, skip );
2602 delayed()->restore();
2603
2604 // Perform a more elaborate out-of-line call
2605 // Not an address; verify it:
2606 bind(test);
2607 verify_oop(reg);
2608 bind(skip);
2609 }
2610
2611
2612 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2613 if (state == ftos || state == dtos) MacroAssembler::verify_FPU(stack_depth);
2614 }
2615
2616
2617 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
2618 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
2619 int increment, Address mask_addr,
2620 Register scratch1, Register scratch2,
2621 Condition cond, Label *where) {
2622 ld(counter_addr, scratch1);
2623 add(scratch1, increment, scratch1);
2624 ld(mask_addr, scratch2);
2625 andcc(scratch1, scratch2, G0);
2626 br(cond, false, Assembler::pn, *where);
2627 delayed()->st(scratch1, counter_addr);
2628 }
2629
2630 // Inline assembly for:
2631 //
2632 // if (thread is in interp_only_mode) {
2633 // InterpreterRuntime::post_method_entry();
2634 // }
2635 // if (DTraceMethodProbes) {
2636 // SharedRuntime::dtrace_method_entry(method, receiver);
2637 // }
2638 // if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
2639 // SharedRuntime::rc_trace_method_entry(method, receiver);
2640 // }
2641
2642 void InterpreterMacroAssembler::notify_method_entry() {
2643
2644 // Whenever JVMTI puts a thread in interp_only_mode, method
2645 // entry/exit events are sent for that thread to track stack
2646 // depth. If it is possible to enter interp_only_mode we add
2647 // the code to check if the event should be sent.
2648 if (JvmtiExport::can_post_interpreter_events()) {
2649 Label L;
2650 Register temp_reg = O5;
2651 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset());
2652 ld(interp_only, temp_reg);
2653 cmp_and_br_short(temp_reg, 0, equal, pt, L);
2654 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry));
2655 bind(L);
2656 }
2657
2658 {
2659 Register temp_reg = O5;
2660 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero);
2661 call_VM_leaf(noreg,
2662 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
2663 G2_thread, Lmethod);
2664 }
2665
2666 // RedefineClasses() tracing support for obsolete method entry
2667 if (log_is_enabled(Trace, redefine, class, obsolete)) {
2668 call_VM_leaf(noreg,
2669 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
2670 G2_thread, Lmethod);
2671 }
2672 }
2673
2674
2675 // Inline assembly for:
2676 //
2677 // if (thread is in interp_only_mode) {
2678 // // save result
2679 // InterpreterRuntime::post_method_exit();
2680 // // restore result
2681 // }
2682 // if (DTraceMethodProbes) {
2683 // SharedRuntime::dtrace_method_exit(thread, method);
2684 // }
2685 //
2686 // Native methods have their result stored in d_tmp and l_tmp
2687 // Java methods have their result stored in the expression stack
2688
2689 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method,
2690 TosState state,
2691 NotifyMethodExitMode mode) {
2692
2693 // Whenever JVMTI puts a thread in interp_only_mode, method
2694 // entry/exit events are sent for that thread to track stack
2695 // depth. If it is possible to enter interp_only_mode we add
2696 // the code to check if the event should be sent.
2697 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2698 Label L;
2699 Register temp_reg = O5;
2700 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset());
2701 ld(interp_only, temp_reg);
2702 cmp_and_br_short(temp_reg, 0, equal, pt, L);
2703
2704 // Note: frame::interpreter_frame_result has a dependency on how the
2705 // method result is saved across the call to post_method_exit. For
2706 // native methods it assumes the result registers are saved to
2707 // l_scratch and d_scratch. If this changes then the interpreter_frame_result
2708 // implementation will need to be updated too.
2709
2710 save_return_value(state, is_native_method);
2711 call_VM(noreg,
2712 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2713 restore_return_value(state, is_native_method);
2714 bind(L);
2715 }
2716
2717 {
2718 Register temp_reg = O5;
2719 // Dtrace notification
2720 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero);
2721 save_return_value(state, is_native_method);
2722 call_VM_leaf(
2723 noreg,
2724 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2725 G2_thread, Lmethod);
2726 restore_return_value(state, is_native_method);
2727 }
2728 }
2729
2730 void InterpreterMacroAssembler::save_return_value(TosState state, bool is_native_call) {
2731 if (is_native_call) {
2732 stf(FloatRegisterImpl::D, F0, d_tmp);
2733 #ifdef _LP64
2734 stx(O0, l_tmp);
2735 #else
2736 std(O0, l_tmp);
2737 #endif
2738 } else {
2739 push(state);
2740 }
2741 }
2742
2743 void InterpreterMacroAssembler::restore_return_value( TosState state, bool is_native_call) {
2744 if (is_native_call) {
2745 ldf(FloatRegisterImpl::D, d_tmp, F0);
2746 #ifdef _LP64
2747 ldx(l_tmp, O0);
2748 #else
2749 ldd(l_tmp, O0);
2750 #endif
2751 } else {
2752 pop(state);
2753 }
2754 }