1 /*
2 * Copyright (c) 2014, 2018, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2013, 2017 SAP SE. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "precompiled.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "interpreter/interpreterRuntime.hpp"
30 #include "interpreter/interp_masm.hpp"
31 #include "interpreter/templateInterpreter.hpp"
32 #include "interpreter/templateTable.hpp"
33 #include "memory/universe.hpp"
34 #include "oops/objArrayKlass.hpp"
35 #include "oops/oop.inline.hpp"
36 #include "prims/methodHandles.hpp"
37 #include "runtime/frame.inline.hpp"
38 #include "runtime/safepointMechanism.hpp"
39 #include "runtime/sharedRuntime.hpp"
40 #include "runtime/stubRoutines.hpp"
41 #include "runtime/synchronizer.hpp"
42 #include "utilities/macros.hpp"
43
44 #undef __
45 #define __ _masm->
46
47 // ============================================================================
48 // Misc helpers
49
50 // Do an oop store like *(base + index) = val OR *(base + offset) = val
51 // (only one of both variants is possible at the same time).
52 // Index can be noreg.
53 // Kills:
54 // Rbase, Rtmp
55 static void do_oop_store(InterpreterMacroAssembler* _masm,
56 Register Rbase,
57 RegisterOrConstant offset,
58 Register Rval, // Noreg means always null.
59 Register Rtmp1,
60 Register Rtmp2,
61 Register Rtmp3,
62 BarrierSet::Name barrier,
63 bool precise,
64 bool check_null) {
65 assert_different_registers(Rtmp1, Rtmp2, Rtmp3, Rval, Rbase);
66
67 switch (barrier) {
68 #if INCLUDE_ALL_GCS
69 case BarrierSet::G1BarrierSet:
70 {
71 // Load and record the previous value.
72 __ g1_write_barrier_pre(Rbase, offset,
73 Rtmp3, /* holder of pre_val ? */
74 Rtmp1, Rtmp2, false /* frame */);
75
76 Label Lnull, Ldone;
77 if (Rval != noreg) {
78 if (check_null) {
79 __ cmpdi(CCR0, Rval, 0);
80 __ beq(CCR0, Lnull);
81 }
82 __ store_heap_oop_not_null(Rval, offset, Rbase, /*Rval must stay uncompressed.*/ Rtmp1);
83 // Mark the card.
84 if (!(offset.is_constant() && offset.as_constant() == 0) && precise) {
85 __ add(Rbase, offset, Rbase);
86 }
87 __ g1_write_barrier_post(Rbase, Rval, Rtmp1, Rtmp2, Rtmp3, /*filtered (fast path)*/ &Ldone);
88 if (check_null) { __ b(Ldone); }
89 }
90
91 if (Rval == noreg || check_null) { // Store null oop.
92 Register Rnull = Rval;
93 __ bind(Lnull);
94 if (Rval == noreg) {
95 Rnull = Rtmp1;
96 __ li(Rnull, 0);
97 }
98 if (UseCompressedOops) {
99 __ stw(Rnull, offset, Rbase);
100 } else {
101 __ std(Rnull, offset, Rbase);
102 }
103 }
104 __ bind(Ldone);
105 }
106 break;
107 #endif // INCLUDE_ALL_GCS
108 case BarrierSet::CardTableBarrierSet:
109 {
110 Label Lnull, Ldone;
111 if (Rval != noreg) {
112 if (check_null) {
113 __ cmpdi(CCR0, Rval, 0);
114 __ beq(CCR0, Lnull);
115 }
116 __ store_heap_oop_not_null(Rval, offset, Rbase, /*Rval should better stay uncompressed.*/ Rtmp1);
117 // Mark the card.
118 if (!(offset.is_constant() && offset.as_constant() == 0) && precise) {
119 __ add(Rbase, offset, Rbase);
120 }
121 __ card_write_barrier_post(Rbase, Rval, Rtmp1);
122 if (check_null) {
123 __ b(Ldone);
124 }
125 }
126
127 if (Rval == noreg || check_null) { // Store null oop.
128 Register Rnull = Rval;
129 __ bind(Lnull);
130 if (Rval == noreg) {
131 Rnull = Rtmp1;
132 __ li(Rnull, 0);
133 }
134 if (UseCompressedOops) {
135 __ stw(Rnull, offset, Rbase);
136 } else {
137 __ std(Rnull, offset, Rbase);
138 }
139 }
140 __ bind(Ldone);
141 }
142 break;
143 case BarrierSet::ModRef:
144 ShouldNotReachHere();
145 break;
146 default:
147 ShouldNotReachHere();
148 }
149 }
150
151 // ============================================================================
152 // Platform-dependent initialization
153
154 void TemplateTable::pd_initialize() {
155 // No ppc64 specific initialization.
156 }
157
158 Address TemplateTable::at_bcp(int offset) {
159 // Not used on ppc.
160 ShouldNotReachHere();
161 return Address();
162 }
163
164 // Patches the current bytecode (ptr to it located in bcp)
165 // in the bytecode stream with a new one.
166 void TemplateTable::patch_bytecode(Bytecodes::Code new_bc, Register Rnew_bc, Register Rtemp, bool load_bc_into_bc_reg /*=true*/, int byte_no) {
167 // With sharing on, may need to test method flag.
168 if (!RewriteBytecodes) return;
169 Label L_patch_done;
170
171 switch (new_bc) {
172 case Bytecodes::_fast_aputfield:
173 case Bytecodes::_fast_bputfield:
174 case Bytecodes::_fast_zputfield:
175 case Bytecodes::_fast_cputfield:
176 case Bytecodes::_fast_dputfield:
177 case Bytecodes::_fast_fputfield:
178 case Bytecodes::_fast_iputfield:
179 case Bytecodes::_fast_lputfield:
180 case Bytecodes::_fast_sputfield:
181 {
182 // We skip bytecode quickening for putfield instructions when
183 // the put_code written to the constant pool cache is zero.
184 // This is required so that every execution of this instruction
185 // calls out to InterpreterRuntime::resolve_get_put to do
186 // additional, required work.
187 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
188 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
189 __ get_cache_and_index_at_bcp(Rtemp /* dst = cache */, 1);
190 // ((*(cache+indices))>>((1+byte_no)*8))&0xFF:
191 #if defined(VM_LITTLE_ENDIAN)
192 __ lbz(Rnew_bc, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 1 + byte_no, Rtemp);
193 #else
194 __ lbz(Rnew_bc, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 7 - (1 + byte_no), Rtemp);
195 #endif
196 __ cmpwi(CCR0, Rnew_bc, 0);
197 __ li(Rnew_bc, (unsigned int)(unsigned char)new_bc);
198 __ beq(CCR0, L_patch_done);
199 // __ isync(); // acquire not needed
200 break;
201 }
202
203 default:
204 assert(byte_no == -1, "sanity");
205 if (load_bc_into_bc_reg) {
206 __ li(Rnew_bc, (unsigned int)(unsigned char)new_bc);
207 }
208 }
209
210 if (JvmtiExport::can_post_breakpoint()) {
211 Label L_fast_patch;
212 __ lbz(Rtemp, 0, R14_bcp);
213 __ cmpwi(CCR0, Rtemp, (unsigned int)(unsigned char)Bytecodes::_breakpoint);
214 __ bne(CCR0, L_fast_patch);
215 // Perform the quickening, slowly, in the bowels of the breakpoint table.
216 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), R19_method, R14_bcp, Rnew_bc);
217 __ b(L_patch_done);
218 __ bind(L_fast_patch);
219 }
220
221 // Patch bytecode.
222 __ stb(Rnew_bc, 0, R14_bcp);
223
224 __ bind(L_patch_done);
225 }
226
227 // ============================================================================
228 // Individual instructions
229
230 void TemplateTable::nop() {
231 transition(vtos, vtos);
232 // Nothing to do.
233 }
234
235 void TemplateTable::shouldnotreachhere() {
236 transition(vtos, vtos);
237 __ stop("shouldnotreachhere bytecode");
238 }
239
240 void TemplateTable::aconst_null() {
241 transition(vtos, atos);
242 __ li(R17_tos, 0);
243 }
244
245 void TemplateTable::iconst(int value) {
246 transition(vtos, itos);
247 assert(value >= -1 && value <= 5, "");
248 __ li(R17_tos, value);
249 }
250
251 void TemplateTable::lconst(int value) {
252 transition(vtos, ltos);
253 assert(value >= -1 && value <= 5, "");
254 __ li(R17_tos, value);
255 }
256
257 void TemplateTable::fconst(int value) {
258 transition(vtos, ftos);
259 static float zero = 0.0;
260 static float one = 1.0;
261 static float two = 2.0;
262 switch (value) {
263 default: ShouldNotReachHere();
264 case 0: {
265 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&zero, R0, true);
266 __ lfs(F15_ftos, simm16_offset, R11_scratch1);
267 break;
268 }
269 case 1: {
270 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&one, R0, true);
271 __ lfs(F15_ftos, simm16_offset, R11_scratch1);
272 break;
273 }
274 case 2: {
275 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&two, R0, true);
276 __ lfs(F15_ftos, simm16_offset, R11_scratch1);
277 break;
278 }
279 }
280 }
281
282 void TemplateTable::dconst(int value) {
283 transition(vtos, dtos);
284 static double zero = 0.0;
285 static double one = 1.0;
286 switch (value) {
287 case 0: {
288 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&zero, R0, true);
289 __ lfd(F15_ftos, simm16_offset, R11_scratch1);
290 break;
291 }
292 case 1: {
293 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&one, R0, true);
294 __ lfd(F15_ftos, simm16_offset, R11_scratch1);
295 break;
296 }
297 default: ShouldNotReachHere();
298 }
299 }
300
301 void TemplateTable::bipush() {
302 transition(vtos, itos);
303 __ lbz(R17_tos, 1, R14_bcp);
304 __ extsb(R17_tos, R17_tos);
305 }
306
307 void TemplateTable::sipush() {
308 transition(vtos, itos);
309 __ get_2_byte_integer_at_bcp(1, R17_tos, InterpreterMacroAssembler::Signed);
310 }
311
312 void TemplateTable::ldc(bool wide) {
313 Register Rscratch1 = R11_scratch1,
314 Rscratch2 = R12_scratch2,
315 Rcpool = R3_ARG1;
316
317 transition(vtos, vtos);
318 Label notInt, notFloat, notClass, exit;
319
320 __ get_cpool_and_tags(Rcpool, Rscratch2); // Set Rscratch2 = &tags.
321 if (wide) { // Read index.
322 __ get_2_byte_integer_at_bcp(1, Rscratch1, InterpreterMacroAssembler::Unsigned);
323 } else {
324 __ lbz(Rscratch1, 1, R14_bcp);
325 }
326
327 const int base_offset = ConstantPool::header_size() * wordSize;
328 const int tags_offset = Array<u1>::base_offset_in_bytes();
329
330 // Get type from tags.
331 __ addi(Rscratch2, Rscratch2, tags_offset);
332 __ lbzx(Rscratch2, Rscratch2, Rscratch1);
333
334 __ cmpwi(CCR0, Rscratch2, JVM_CONSTANT_UnresolvedClass); // Unresolved class?
335 __ cmpwi(CCR1, Rscratch2, JVM_CONSTANT_UnresolvedClassInError); // Unresolved class in error state?
336 __ cror(CCR0, Assembler::equal, CCR1, Assembler::equal);
337
338 // Resolved class - need to call vm to get java mirror of the class.
339 __ cmpwi(CCR1, Rscratch2, JVM_CONSTANT_Class);
340 __ crnor(CCR0, Assembler::equal, CCR1, Assembler::equal); // Neither resolved class nor unresolved case from above?
341 __ beq(CCR0, notClass);
342
343 __ li(R4, wide ? 1 : 0);
344 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), R4);
345 __ push(atos);
346 __ b(exit);
347
348 __ align(32, 12);
349 __ bind(notClass);
350 __ addi(Rcpool, Rcpool, base_offset);
351 __ sldi(Rscratch1, Rscratch1, LogBytesPerWord);
352 __ cmpdi(CCR0, Rscratch2, JVM_CONSTANT_Integer);
353 __ bne(CCR0, notInt);
354 __ lwax(R17_tos, Rcpool, Rscratch1);
355 __ push(itos);
356 __ b(exit);
357
358 __ align(32, 12);
359 __ bind(notInt);
360 __ cmpdi(CCR0, Rscratch2, JVM_CONSTANT_Float);
361 __ bne(CCR0, notFloat);
362 __ lfsx(F15_ftos, Rcpool, Rscratch1);
363 __ push(ftos);
364 __ b(exit);
365
366 __ align(32, 12);
367 // assume the tag is for condy; if not, the VM runtime will tell us
368 __ bind(notFloat);
369 condy_helper(exit);
370
371 __ align(32, 12);
372 __ bind(exit);
373 }
374
375 // Fast path for caching oop constants.
376 void TemplateTable::fast_aldc(bool wide) {
377 transition(vtos, atos);
378
379 int index_size = wide ? sizeof(u2) : sizeof(u1);
380 const Register Rscratch = R11_scratch1;
381 Label is_null;
382
383 // We are resolved if the resolved reference cache entry contains a
384 // non-null object (CallSite, etc.)
385 __ get_cache_index_at_bcp(Rscratch, 1, index_size); // Load index.
386 __ load_resolved_reference_at_index(R17_tos, Rscratch, &is_null);
387
388 // Convert null sentinel to NULL.
389 int simm16_rest = __ load_const_optimized(Rscratch, Universe::the_null_sentinel_addr(), R0, true);
390 __ ld(Rscratch, simm16_rest, Rscratch);
391 __ cmpld(CCR0, R17_tos, Rscratch);
392 if (VM_Version::has_isel()) {
393 __ isel_0(R17_tos, CCR0, Assembler::equal);
394 } else {
395 Label not_sentinel;
396 __ bne(CCR0, not_sentinel);
397 __ li(R17_tos, 0);
398 __ bind(not_sentinel);
399 }
400 __ verify_oop(R17_tos);
401 __ dispatch_epilog(atos, Bytecodes::length_for(bytecode()));
402
403 __ bind(is_null);
404 __ load_const_optimized(R3_ARG1, (int)bytecode());
405
406 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
407
408 // First time invocation - must resolve first.
409 __ call_VM(R17_tos, entry, R3_ARG1);
410 __ verify_oop(R17_tos);
411 }
412
413 void TemplateTable::ldc2_w() {
414 transition(vtos, vtos);
415 Label not_double, not_long, exit;
416
417 Register Rindex = R11_scratch1,
418 Rcpool = R12_scratch2,
419 Rtag = R3_ARG1;
420 __ get_cpool_and_tags(Rcpool, Rtag);
421 __ get_2_byte_integer_at_bcp(1, Rindex, InterpreterMacroAssembler::Unsigned);
422
423 const int base_offset = ConstantPool::header_size() * wordSize;
424 const int tags_offset = Array<u1>::base_offset_in_bytes();
425 // Get type from tags.
426 __ addi(Rcpool, Rcpool, base_offset);
427 __ addi(Rtag, Rtag, tags_offset);
428
429 __ lbzx(Rtag, Rtag, Rindex);
430 __ sldi(Rindex, Rindex, LogBytesPerWord);
431
432 __ cmpdi(CCR0, Rtag, JVM_CONSTANT_Double);
433 __ bne(CCR0, not_double);
434 __ lfdx(F15_ftos, Rcpool, Rindex);
435 __ push(dtos);
436 __ b(exit);
437
438 __ bind(not_double);
439 __ cmpdi(CCR0, Rtag, JVM_CONSTANT_Long);
440 __ bne(CCR0, not_long);
441 __ ldx(R17_tos, Rcpool, Rindex);
442 __ push(ltos);
443 __ b(exit);
444
445 __ bind(not_long);
446 condy_helper(exit);
447
448 __ align(32, 12);
449 __ bind(exit);
450 }
451
452 void TemplateTable::condy_helper(Label& Done) {
453 const Register obj = R31;
454 const Register off = R11_scratch1;
455 const Register flags = R12_scratch2;
456 const Register rarg = R4_ARG2;
457 __ li(rarg, (int)bytecode());
458 call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc), rarg);
459 __ get_vm_result_2(flags);
460
461 // VMr = obj = base address to find primitive value to push
462 // VMr2 = flags = (tos, off) using format of CPCE::_flags
463 __ andi(off, flags, ConstantPoolCacheEntry::field_index_mask);
464
465 // What sort of thing are we loading?
466 __ rldicl(flags, flags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
467
468 switch (bytecode()) {
469 case Bytecodes::_ldc:
470 case Bytecodes::_ldc_w:
471 {
472 // tos in (itos, ftos, stos, btos, ctos, ztos)
473 Label notInt, notFloat, notShort, notByte, notChar, notBool;
474 __ cmplwi(CCR0, flags, itos);
475 __ bne(CCR0, notInt);
476 // itos
477 __ lwax(R17_tos, obj, off);
478 __ push(itos);
479 __ b(Done);
480
481 __ bind(notInt);
482 __ cmplwi(CCR0, flags, ftos);
483 __ bne(CCR0, notFloat);
484 // ftos
485 __ lfsx(F15_ftos, obj, off);
486 __ push(ftos);
487 __ b(Done);
488
489 __ bind(notFloat);
490 __ cmplwi(CCR0, flags, stos);
491 __ bne(CCR0, notShort);
492 // stos
493 __ lhax(R17_tos, obj, off);
494 __ push(stos);
495 __ b(Done);
496
497 __ bind(notShort);
498 __ cmplwi(CCR0, flags, btos);
499 __ bne(CCR0, notByte);
500 // btos
501 __ lbzx(R17_tos, obj, off);
502 __ extsb(R17_tos, R17_tos);
503 __ push(btos);
504 __ b(Done);
505
506 __ bind(notByte);
507 __ cmplwi(CCR0, flags, ctos);
508 __ bne(CCR0, notChar);
509 // ctos
510 __ lhzx(R17_tos, obj, off);
511 __ push(ctos);
512 __ b(Done);
513
514 __ bind(notChar);
515 __ cmplwi(CCR0, flags, ztos);
516 __ bne(CCR0, notBool);
517 // ztos
518 __ lbzx(R17_tos, obj, off);
519 __ push(ztos);
520 __ b(Done);
521
522 __ bind(notBool);
523 break;
524 }
525
526 case Bytecodes::_ldc2_w:
527 {
528 Label notLong, notDouble;
529 __ cmplwi(CCR0, flags, ltos);
530 __ bne(CCR0, notLong);
531 // ltos
532 __ ldx(R17_tos, obj, off);
533 __ push(ltos);
534 __ b(Done);
535
536 __ bind(notLong);
537 __ cmplwi(CCR0, flags, dtos);
538 __ bne(CCR0, notDouble);
539 // dtos
540 __ lfdx(F15_ftos, obj, off);
541 __ push(dtos);
542 __ b(Done);
543
544 __ bind(notDouble);
545 break;
546 }
547
548 default:
549 ShouldNotReachHere();
550 }
551
552 __ stop("bad ldc/condy");
553 }
554
555 // Get the locals index located in the bytecode stream at bcp + offset.
556 void TemplateTable::locals_index(Register Rdst, int offset) {
557 __ lbz(Rdst, offset, R14_bcp);
558 }
559
560 void TemplateTable::iload() {
561 iload_internal();
562 }
563
564 void TemplateTable::nofast_iload() {
565 iload_internal(may_not_rewrite);
566 }
567
568 void TemplateTable::iload_internal(RewriteControl rc) {
569 transition(vtos, itos);
570
571 // Get the local value into tos
572 const Register Rindex = R22_tmp2;
573 locals_index(Rindex);
574
575 // Rewrite iload,iload pair into fast_iload2
576 // iload,caload pair into fast_icaload
577 if (RewriteFrequentPairs && rc == may_rewrite) {
578 Label Lrewrite, Ldone;
579 Register Rnext_byte = R3_ARG1,
580 Rrewrite_to = R6_ARG4,
581 Rscratch = R11_scratch1;
582
583 // get next byte
584 __ lbz(Rnext_byte, Bytecodes::length_for(Bytecodes::_iload), R14_bcp);
585
586 // if _iload, wait to rewrite to iload2. We only want to rewrite the
587 // last two iloads in a pair. Comparing against fast_iload means that
588 // the next bytecode is neither an iload or a caload, and therefore
589 // an iload pair.
590 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_iload);
591 __ beq(CCR0, Ldone);
592
593 __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_iload);
594 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iload2);
595 __ beq(CCR1, Lrewrite);
596
597 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_caload);
598 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_icaload);
599 __ beq(CCR0, Lrewrite);
600
601 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iload);
602
603 __ bind(Lrewrite);
604 patch_bytecode(Bytecodes::_iload, Rrewrite_to, Rscratch, false);
605 __ bind(Ldone);
606 }
607
608 __ load_local_int(R17_tos, Rindex, Rindex);
609 }
610
611 // Load 2 integers in a row without dispatching
612 void TemplateTable::fast_iload2() {
613 transition(vtos, itos);
614
615 __ lbz(R3_ARG1, 1, R14_bcp);
616 __ lbz(R17_tos, Bytecodes::length_for(Bytecodes::_iload) + 1, R14_bcp);
617
618 __ load_local_int(R3_ARG1, R11_scratch1, R3_ARG1);
619 __ load_local_int(R17_tos, R12_scratch2, R17_tos);
620 __ push_i(R3_ARG1);
621 }
622
623 void TemplateTable::fast_iload() {
624 transition(vtos, itos);
625 // Get the local value into tos
626
627 const Register Rindex = R11_scratch1;
628 locals_index(Rindex);
629 __ load_local_int(R17_tos, Rindex, Rindex);
630 }
631
632 // Load a local variable type long from locals area to TOS cache register.
633 // Local index resides in bytecodestream.
634 void TemplateTable::lload() {
635 transition(vtos, ltos);
636
637 const Register Rindex = R11_scratch1;
638 locals_index(Rindex);
639 __ load_local_long(R17_tos, Rindex, Rindex);
640 }
641
642 void TemplateTable::fload() {
643 transition(vtos, ftos);
644
645 const Register Rindex = R11_scratch1;
646 locals_index(Rindex);
647 __ load_local_float(F15_ftos, Rindex, Rindex);
648 }
649
650 void TemplateTable::dload() {
651 transition(vtos, dtos);
652
653 const Register Rindex = R11_scratch1;
654 locals_index(Rindex);
655 __ load_local_double(F15_ftos, Rindex, Rindex);
656 }
657
658 void TemplateTable::aload() {
659 transition(vtos, atos);
660
661 const Register Rindex = R11_scratch1;
662 locals_index(Rindex);
663 __ load_local_ptr(R17_tos, Rindex, Rindex);
664 }
665
666 void TemplateTable::locals_index_wide(Register Rdst) {
667 // Offset is 2, not 1, because Lbcp points to wide prefix code.
668 __ get_2_byte_integer_at_bcp(2, Rdst, InterpreterMacroAssembler::Unsigned);
669 }
670
671 void TemplateTable::wide_iload() {
672 // Get the local value into tos.
673
674 const Register Rindex = R11_scratch1;
675 locals_index_wide(Rindex);
676 __ load_local_int(R17_tos, Rindex, Rindex);
677 }
678
679 void TemplateTable::wide_lload() {
680 transition(vtos, ltos);
681
682 const Register Rindex = R11_scratch1;
683 locals_index_wide(Rindex);
684 __ load_local_long(R17_tos, Rindex, Rindex);
685 }
686
687 void TemplateTable::wide_fload() {
688 transition(vtos, ftos);
689
690 const Register Rindex = R11_scratch1;
691 locals_index_wide(Rindex);
692 __ load_local_float(F15_ftos, Rindex, Rindex);
693 }
694
695 void TemplateTable::wide_dload() {
696 transition(vtos, dtos);
697
698 const Register Rindex = R11_scratch1;
699 locals_index_wide(Rindex);
700 __ load_local_double(F15_ftos, Rindex, Rindex);
701 }
702
703 void TemplateTable::wide_aload() {
704 transition(vtos, atos);
705
706 const Register Rindex = R11_scratch1;
707 locals_index_wide(Rindex);
708 __ load_local_ptr(R17_tos, Rindex, Rindex);
709 }
710
711 void TemplateTable::iaload() {
712 transition(itos, itos);
713
714 const Register Rload_addr = R3_ARG1,
715 Rarray = R4_ARG2,
716 Rtemp = R5_ARG3;
717 __ index_check(Rarray, R17_tos /* index */, LogBytesPerInt, Rtemp, Rload_addr);
718 __ lwa(R17_tos, arrayOopDesc::base_offset_in_bytes(T_INT), Rload_addr);
719 }
720
721 void TemplateTable::laload() {
722 transition(itos, ltos);
723
724 const Register Rload_addr = R3_ARG1,
725 Rarray = R4_ARG2,
726 Rtemp = R5_ARG3;
727 __ index_check(Rarray, R17_tos /* index */, LogBytesPerLong, Rtemp, Rload_addr);
728 __ ld(R17_tos, arrayOopDesc::base_offset_in_bytes(T_LONG), Rload_addr);
729 }
730
731 void TemplateTable::faload() {
732 transition(itos, ftos);
733
734 const Register Rload_addr = R3_ARG1,
735 Rarray = R4_ARG2,
736 Rtemp = R5_ARG3;
737 __ index_check(Rarray, R17_tos /* index */, LogBytesPerInt, Rtemp, Rload_addr);
738 __ lfs(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Rload_addr);
739 }
740
741 void TemplateTable::daload() {
742 transition(itos, dtos);
743
744 const Register Rload_addr = R3_ARG1,
745 Rarray = R4_ARG2,
746 Rtemp = R5_ARG3;
747 __ index_check(Rarray, R17_tos /* index */, LogBytesPerLong, Rtemp, Rload_addr);
748 __ lfd(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Rload_addr);
749 }
750
751 void TemplateTable::aaload() {
752 transition(itos, atos);
753
754 // tos: index
755 // result tos: array
756 const Register Rload_addr = R3_ARG1,
757 Rarray = R4_ARG2,
758 Rtemp = R5_ARG3;
759 __ index_check(Rarray, R17_tos /* index */, UseCompressedOops ? 2 : LogBytesPerWord, Rtemp, Rload_addr);
760 __ load_heap_oop(R17_tos, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Rload_addr);
761 __ verify_oop(R17_tos);
762 //__ dcbt(R17_tos); // prefetch
763 }
764
765 void TemplateTable::baload() {
766 transition(itos, itos);
767
768 const Register Rload_addr = R3_ARG1,
769 Rarray = R4_ARG2,
770 Rtemp = R5_ARG3;
771 __ index_check(Rarray, R17_tos /* index */, 0, Rtemp, Rload_addr);
772 __ lbz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_BYTE), Rload_addr);
773 __ extsb(R17_tos, R17_tos);
774 }
775
776 void TemplateTable::caload() {
777 transition(itos, itos);
778
779 const Register Rload_addr = R3_ARG1,
780 Rarray = R4_ARG2,
781 Rtemp = R5_ARG3;
782 __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
783 __ lhz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rload_addr);
784 }
785
786 // Iload followed by caload frequent pair.
787 void TemplateTable::fast_icaload() {
788 transition(vtos, itos);
789
790 const Register Rload_addr = R3_ARG1,
791 Rarray = R4_ARG2,
792 Rtemp = R11_scratch1;
793
794 locals_index(R17_tos);
795 __ load_local_int(R17_tos, Rtemp, R17_tos);
796 __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
797 __ lhz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rload_addr);
798 }
799
800 void TemplateTable::saload() {
801 transition(itos, itos);
802
803 const Register Rload_addr = R11_scratch1,
804 Rarray = R12_scratch2,
805 Rtemp = R3_ARG1;
806 __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
807 __ lha(R17_tos, arrayOopDesc::base_offset_in_bytes(T_SHORT), Rload_addr);
808 }
809
810 void TemplateTable::iload(int n) {
811 transition(vtos, itos);
812
813 __ lwz(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
814 }
815
816 void TemplateTable::lload(int n) {
817 transition(vtos, ltos);
818
819 __ ld(R17_tos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
820 }
821
822 void TemplateTable::fload(int n) {
823 transition(vtos, ftos);
824
825 __ lfs(F15_ftos, Interpreter::local_offset_in_bytes(n), R18_locals);
826 }
827
828 void TemplateTable::dload(int n) {
829 transition(vtos, dtos);
830
831 __ lfd(F15_ftos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
832 }
833
834 void TemplateTable::aload(int n) {
835 transition(vtos, atos);
836
837 __ ld(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
838 }
839
840 void TemplateTable::aload_0() {
841 aload_0_internal();
842 }
843
844 void TemplateTable::nofast_aload_0() {
845 aload_0_internal(may_not_rewrite);
846 }
847
848 void TemplateTable::aload_0_internal(RewriteControl rc) {
849 transition(vtos, atos);
850 // According to bytecode histograms, the pairs:
851 //
852 // _aload_0, _fast_igetfield
853 // _aload_0, _fast_agetfield
854 // _aload_0, _fast_fgetfield
855 //
856 // occur frequently. If RewriteFrequentPairs is set, the (slow)
857 // _aload_0 bytecode checks if the next bytecode is either
858 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
859 // rewrites the current bytecode into a pair bytecode; otherwise it
860 // rewrites the current bytecode into _0 that doesn't do
861 // the pair check anymore.
862 //
863 // Note: If the next bytecode is _getfield, the rewrite must be
864 // delayed, otherwise we may miss an opportunity for a pair.
865 //
866 // Also rewrite frequent pairs
867 // aload_0, aload_1
868 // aload_0, iload_1
869 // These bytecodes with a small amount of code are most profitable
870 // to rewrite.
871
872 if (RewriteFrequentPairs && rc == may_rewrite) {
873
874 Label Lrewrite, Ldont_rewrite;
875 Register Rnext_byte = R3_ARG1,
876 Rrewrite_to = R6_ARG4,
877 Rscratch = R11_scratch1;
878
879 // Get next byte.
880 __ lbz(Rnext_byte, Bytecodes::length_for(Bytecodes::_aload_0), R14_bcp);
881
882 // If _getfield, wait to rewrite. We only want to rewrite the last two bytecodes in a pair.
883 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_getfield);
884 __ beq(CCR0, Ldont_rewrite);
885
886 __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_igetfield);
887 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iaccess_0);
888 __ beq(CCR1, Lrewrite);
889
890 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_agetfield);
891 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_aaccess_0);
892 __ beq(CCR0, Lrewrite);
893
894 __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_fgetfield);
895 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_faccess_0);
896 __ beq(CCR1, Lrewrite);
897
898 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_aload_0);
899
900 __ bind(Lrewrite);
901 patch_bytecode(Bytecodes::_aload_0, Rrewrite_to, Rscratch, false);
902 __ bind(Ldont_rewrite);
903 }
904
905 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
906 aload(0);
907 }
908
909 void TemplateTable::istore() {
910 transition(itos, vtos);
911
912 const Register Rindex = R11_scratch1;
913 locals_index(Rindex);
914 __ store_local_int(R17_tos, Rindex);
915 }
916
917 void TemplateTable::lstore() {
918 transition(ltos, vtos);
919 const Register Rindex = R11_scratch1;
920 locals_index(Rindex);
921 __ store_local_long(R17_tos, Rindex);
922 }
923
924 void TemplateTable::fstore() {
925 transition(ftos, vtos);
926
927 const Register Rindex = R11_scratch1;
928 locals_index(Rindex);
929 __ store_local_float(F15_ftos, Rindex);
930 }
931
932 void TemplateTable::dstore() {
933 transition(dtos, vtos);
934
935 const Register Rindex = R11_scratch1;
936 locals_index(Rindex);
937 __ store_local_double(F15_ftos, Rindex);
938 }
939
940 void TemplateTable::astore() {
941 transition(vtos, vtos);
942
943 const Register Rindex = R11_scratch1;
944 __ pop_ptr();
945 __ verify_oop_or_return_address(R17_tos, Rindex);
946 locals_index(Rindex);
947 __ store_local_ptr(R17_tos, Rindex);
948 }
949
950 void TemplateTable::wide_istore() {
951 transition(vtos, vtos);
952
953 const Register Rindex = R11_scratch1;
954 __ pop_i();
955 locals_index_wide(Rindex);
956 __ store_local_int(R17_tos, Rindex);
957 }
958
959 void TemplateTable::wide_lstore() {
960 transition(vtos, vtos);
961
962 const Register Rindex = R11_scratch1;
963 __ pop_l();
964 locals_index_wide(Rindex);
965 __ store_local_long(R17_tos, Rindex);
966 }
967
968 void TemplateTable::wide_fstore() {
969 transition(vtos, vtos);
970
971 const Register Rindex = R11_scratch1;
972 __ pop_f();
973 locals_index_wide(Rindex);
974 __ store_local_float(F15_ftos, Rindex);
975 }
976
977 void TemplateTable::wide_dstore() {
978 transition(vtos, vtos);
979
980 const Register Rindex = R11_scratch1;
981 __ pop_d();
982 locals_index_wide(Rindex);
983 __ store_local_double(F15_ftos, Rindex);
984 }
985
986 void TemplateTable::wide_astore() {
987 transition(vtos, vtos);
988
989 const Register Rindex = R11_scratch1;
990 __ pop_ptr();
991 __ verify_oop_or_return_address(R17_tos, Rindex);
992 locals_index_wide(Rindex);
993 __ store_local_ptr(R17_tos, Rindex);
994 }
995
996 void TemplateTable::iastore() {
997 transition(itos, vtos);
998
999 const Register Rindex = R3_ARG1,
1000 Rstore_addr = R4_ARG2,
1001 Rarray = R5_ARG3,
1002 Rtemp = R6_ARG4;
1003 __ pop_i(Rindex);
1004 __ index_check(Rarray, Rindex, LogBytesPerInt, Rtemp, Rstore_addr);
1005 __ stw(R17_tos, arrayOopDesc::base_offset_in_bytes(T_INT), Rstore_addr);
1006 }
1007
1008 void TemplateTable::lastore() {
1009 transition(ltos, vtos);
1010
1011 const Register Rindex = R3_ARG1,
1012 Rstore_addr = R4_ARG2,
1013 Rarray = R5_ARG3,
1014 Rtemp = R6_ARG4;
1015 __ pop_i(Rindex);
1016 __ index_check(Rarray, Rindex, LogBytesPerLong, Rtemp, Rstore_addr);
1017 __ std(R17_tos, arrayOopDesc::base_offset_in_bytes(T_LONG), Rstore_addr);
1018 }
1019
1020 void TemplateTable::fastore() {
1021 transition(ftos, vtos);
1022
1023 const Register Rindex = R3_ARG1,
1024 Rstore_addr = R4_ARG2,
1025 Rarray = R5_ARG3,
1026 Rtemp = R6_ARG4;
1027 __ pop_i(Rindex);
1028 __ index_check(Rarray, Rindex, LogBytesPerInt, Rtemp, Rstore_addr);
1029 __ stfs(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Rstore_addr);
1030 }
1031
1032 void TemplateTable::dastore() {
1033 transition(dtos, vtos);
1034
1035 const Register Rindex = R3_ARG1,
1036 Rstore_addr = R4_ARG2,
1037 Rarray = R5_ARG3,
1038 Rtemp = R6_ARG4;
1039 __ pop_i(Rindex);
1040 __ index_check(Rarray, Rindex, LogBytesPerLong, Rtemp, Rstore_addr);
1041 __ stfd(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Rstore_addr);
1042 }
1043
1044 // Pop 3 values from the stack and...
1045 void TemplateTable::aastore() {
1046 transition(vtos, vtos);
1047
1048 Label Lstore_ok, Lis_null, Ldone;
1049 const Register Rindex = R3_ARG1,
1050 Rarray = R4_ARG2,
1051 Rscratch = R11_scratch1,
1052 Rscratch2 = R12_scratch2,
1053 Rarray_klass = R5_ARG3,
1054 Rarray_element_klass = Rarray_klass,
1055 Rvalue_klass = R6_ARG4,
1056 Rstore_addr = R31; // Use register which survives VM call.
1057
1058 __ ld(R17_tos, Interpreter::expr_offset_in_bytes(0), R15_esp); // Get value to store.
1059 __ lwz(Rindex, Interpreter::expr_offset_in_bytes(1), R15_esp); // Get index.
1060 __ ld(Rarray, Interpreter::expr_offset_in_bytes(2), R15_esp); // Get array.
1061
1062 __ verify_oop(R17_tos);
1063 __ index_check_without_pop(Rarray, Rindex, UseCompressedOops ? 2 : LogBytesPerWord, Rscratch, Rstore_addr);
1064 // Rindex is dead!
1065 Register Rscratch3 = Rindex;
1066
1067 // Do array store check - check for NULL value first.
1068 __ cmpdi(CCR0, R17_tos, 0);
1069 __ beq(CCR0, Lis_null);
1070
1071 __ load_klass(Rarray_klass, Rarray);
1072 __ load_klass(Rvalue_klass, R17_tos);
1073
1074 // Do fast instanceof cache test.
1075 __ ld(Rarray_element_klass, in_bytes(ObjArrayKlass::element_klass_offset()), Rarray_klass);
1076
1077 // Generate a fast subtype check. Branch to store_ok if no failure. Throw if failure.
1078 __ gen_subtype_check(Rvalue_klass /*subklass*/, Rarray_element_klass /*superklass*/, Rscratch, Rscratch2, Rscratch3, Lstore_ok);
1079
1080 // Fell through: subtype check failed => throw an exception.
1081 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArrayStoreException_entry);
1082 __ mtctr(R11_scratch1);
1083 __ bctr();
1084
1085 __ bind(Lis_null);
1086 do_oop_store(_masm, Rstore_addr, arrayOopDesc::base_offset_in_bytes(T_OBJECT), noreg /* 0 */,
1087 Rscratch, Rscratch2, Rscratch3, _bs->kind(), true /* precise */, false /* check_null */);
1088 __ profile_null_seen(Rscratch, Rscratch2);
1089 __ b(Ldone);
1090
1091 // Store is OK.
1092 __ bind(Lstore_ok);
1093 do_oop_store(_masm, Rstore_addr, arrayOopDesc::base_offset_in_bytes(T_OBJECT), R17_tos /* value */,
1094 Rscratch, Rscratch2, Rscratch3, _bs->kind(), true /* precise */, false /* check_null */);
1095
1096 __ bind(Ldone);
1097 // Adjust sp (pops array, index and value).
1098 __ addi(R15_esp, R15_esp, 3 * Interpreter::stackElementSize);
1099 }
1100
1101 void TemplateTable::bastore() {
1102 transition(itos, vtos);
1103
1104 const Register Rindex = R11_scratch1,
1105 Rarray = R12_scratch2,
1106 Rscratch = R3_ARG1;
1107 __ pop_i(Rindex);
1108 __ pop_ptr(Rarray);
1109 // tos: val
1110
1111 // Need to check whether array is boolean or byte
1112 // since both types share the bastore bytecode.
1113 __ load_klass(Rscratch, Rarray);
1114 __ lwz(Rscratch, in_bytes(Klass::layout_helper_offset()), Rscratch);
1115 int diffbit = exact_log2(Klass::layout_helper_boolean_diffbit());
1116 __ testbitdi(CCR0, R0, Rscratch, diffbit);
1117 Label L_skip;
1118 __ bfalse(CCR0, L_skip);
1119 __ andi(R17_tos, R17_tos, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1
1120 __ bind(L_skip);
1121
1122 __ index_check_without_pop(Rarray, Rindex, 0, Rscratch, Rarray);
1123 __ stb(R17_tos, arrayOopDesc::base_offset_in_bytes(T_BYTE), Rarray);
1124 }
1125
1126 void TemplateTable::castore() {
1127 transition(itos, vtos);
1128
1129 const Register Rindex = R11_scratch1,
1130 Rarray = R12_scratch2,
1131 Rscratch = R3_ARG1;
1132 __ pop_i(Rindex);
1133 // tos: val
1134 // Rarray: array ptr (popped by index_check)
1135 __ index_check(Rarray, Rindex, LogBytesPerShort, Rscratch, Rarray);
1136 __ sth(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rarray);
1137 }
1138
1139 void TemplateTable::sastore() {
1140 castore();
1141 }
1142
1143 void TemplateTable::istore(int n) {
1144 transition(itos, vtos);
1145 __ stw(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
1146 }
1147
1148 void TemplateTable::lstore(int n) {
1149 transition(ltos, vtos);
1150 __ std(R17_tos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
1151 }
1152
1153 void TemplateTable::fstore(int n) {
1154 transition(ftos, vtos);
1155 __ stfs(F15_ftos, Interpreter::local_offset_in_bytes(n), R18_locals);
1156 }
1157
1158 void TemplateTable::dstore(int n) {
1159 transition(dtos, vtos);
1160 __ stfd(F15_ftos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
1161 }
1162
1163 void TemplateTable::astore(int n) {
1164 transition(vtos, vtos);
1165
1166 __ pop_ptr();
1167 __ verify_oop_or_return_address(R17_tos, R11_scratch1);
1168 __ std(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
1169 }
1170
1171 void TemplateTable::pop() {
1172 transition(vtos, vtos);
1173
1174 __ addi(R15_esp, R15_esp, Interpreter::stackElementSize);
1175 }
1176
1177 void TemplateTable::pop2() {
1178 transition(vtos, vtos);
1179
1180 __ addi(R15_esp, R15_esp, Interpreter::stackElementSize * 2);
1181 }
1182
1183 void TemplateTable::dup() {
1184 transition(vtos, vtos);
1185
1186 __ ld(R11_scratch1, Interpreter::stackElementSize, R15_esp);
1187 __ push_ptr(R11_scratch1);
1188 }
1189
1190 void TemplateTable::dup_x1() {
1191 transition(vtos, vtos);
1192
1193 Register Ra = R11_scratch1,
1194 Rb = R12_scratch2;
1195 // stack: ..., a, b
1196 __ ld(Rb, Interpreter::stackElementSize, R15_esp);
1197 __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1198 __ std(Rb, Interpreter::stackElementSize * 2, R15_esp);
1199 __ std(Ra, Interpreter::stackElementSize, R15_esp);
1200 __ push_ptr(Rb);
1201 // stack: ..., b, a, b
1202 }
1203
1204 void TemplateTable::dup_x2() {
1205 transition(vtos, vtos);
1206
1207 Register Ra = R11_scratch1,
1208 Rb = R12_scratch2,
1209 Rc = R3_ARG1;
1210
1211 // stack: ..., a, b, c
1212 __ ld(Rc, Interpreter::stackElementSize, R15_esp); // load c
1213 __ ld(Ra, Interpreter::stackElementSize * 3, R15_esp); // load a
1214 __ std(Rc, Interpreter::stackElementSize * 3, R15_esp); // store c in a
1215 __ ld(Rb, Interpreter::stackElementSize * 2, R15_esp); // load b
1216 // stack: ..., c, b, c
1217 __ std(Ra, Interpreter::stackElementSize * 2, R15_esp); // store a in b
1218 // stack: ..., c, a, c
1219 __ std(Rb, Interpreter::stackElementSize, R15_esp); // store b in c
1220 __ push_ptr(Rc); // push c
1221 // stack: ..., c, a, b, c
1222 }
1223
1224 void TemplateTable::dup2() {
1225 transition(vtos, vtos);
1226
1227 Register Ra = R11_scratch1,
1228 Rb = R12_scratch2;
1229 // stack: ..., a, b
1230 __ ld(Rb, Interpreter::stackElementSize, R15_esp);
1231 __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1232 __ push_2ptrs(Ra, Rb);
1233 // stack: ..., a, b, a, b
1234 }
1235
1236 void TemplateTable::dup2_x1() {
1237 transition(vtos, vtos);
1238
1239 Register Ra = R11_scratch1,
1240 Rb = R12_scratch2,
1241 Rc = R3_ARG1;
1242 // stack: ..., a, b, c
1243 __ ld(Rc, Interpreter::stackElementSize, R15_esp);
1244 __ ld(Rb, Interpreter::stackElementSize * 2, R15_esp);
1245 __ std(Rc, Interpreter::stackElementSize * 2, R15_esp);
1246 __ ld(Ra, Interpreter::stackElementSize * 3, R15_esp);
1247 __ std(Ra, Interpreter::stackElementSize, R15_esp);
1248 __ std(Rb, Interpreter::stackElementSize * 3, R15_esp);
1249 // stack: ..., b, c, a
1250 __ push_2ptrs(Rb, Rc);
1251 // stack: ..., b, c, a, b, c
1252 }
1253
1254 void TemplateTable::dup2_x2() {
1255 transition(vtos, vtos);
1256
1257 Register Ra = R11_scratch1,
1258 Rb = R12_scratch2,
1259 Rc = R3_ARG1,
1260 Rd = R4_ARG2;
1261 // stack: ..., a, b, c, d
1262 __ ld(Rb, Interpreter::stackElementSize * 3, R15_esp);
1263 __ ld(Rd, Interpreter::stackElementSize, R15_esp);
1264 __ std(Rb, Interpreter::stackElementSize, R15_esp); // store b in d
1265 __ std(Rd, Interpreter::stackElementSize * 3, R15_esp); // store d in b
1266 __ ld(Ra, Interpreter::stackElementSize * 4, R15_esp);
1267 __ ld(Rc, Interpreter::stackElementSize * 2, R15_esp);
1268 __ std(Ra, Interpreter::stackElementSize * 2, R15_esp); // store a in c
1269 __ std(Rc, Interpreter::stackElementSize * 4, R15_esp); // store c in a
1270 // stack: ..., c, d, a, b
1271 __ push_2ptrs(Rc, Rd);
1272 // stack: ..., c, d, a, b, c, d
1273 }
1274
1275 void TemplateTable::swap() {
1276 transition(vtos, vtos);
1277 // stack: ..., a, b
1278
1279 Register Ra = R11_scratch1,
1280 Rb = R12_scratch2;
1281 // stack: ..., a, b
1282 __ ld(Rb, Interpreter::stackElementSize, R15_esp);
1283 __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1284 __ std(Rb, Interpreter::stackElementSize * 2, R15_esp);
1285 __ std(Ra, Interpreter::stackElementSize, R15_esp);
1286 // stack: ..., b, a
1287 }
1288
1289 void TemplateTable::iop2(Operation op) {
1290 transition(itos, itos);
1291
1292 Register Rscratch = R11_scratch1;
1293
1294 __ pop_i(Rscratch);
1295 // tos = number of bits to shift
1296 // Rscratch = value to shift
1297 switch (op) {
1298 case add: __ add(R17_tos, Rscratch, R17_tos); break;
1299 case sub: __ sub(R17_tos, Rscratch, R17_tos); break;
1300 case mul: __ mullw(R17_tos, Rscratch, R17_tos); break;
1301 case _and: __ andr(R17_tos, Rscratch, R17_tos); break;
1302 case _or: __ orr(R17_tos, Rscratch, R17_tos); break;
1303 case _xor: __ xorr(R17_tos, Rscratch, R17_tos); break;
1304 case shl: __ rldicl(R17_tos, R17_tos, 0, 64-5); __ slw(R17_tos, Rscratch, R17_tos); break;
1305 case shr: __ rldicl(R17_tos, R17_tos, 0, 64-5); __ sraw(R17_tos, Rscratch, R17_tos); break;
1306 case ushr: __ rldicl(R17_tos, R17_tos, 0, 64-5); __ srw(R17_tos, Rscratch, R17_tos); break;
1307 default: ShouldNotReachHere();
1308 }
1309 }
1310
1311 void TemplateTable::lop2(Operation op) {
1312 transition(ltos, ltos);
1313
1314 Register Rscratch = R11_scratch1;
1315 __ pop_l(Rscratch);
1316 switch (op) {
1317 case add: __ add(R17_tos, Rscratch, R17_tos); break;
1318 case sub: __ sub(R17_tos, Rscratch, R17_tos); break;
1319 case _and: __ andr(R17_tos, Rscratch, R17_tos); break;
1320 case _or: __ orr(R17_tos, Rscratch, R17_tos); break;
1321 case _xor: __ xorr(R17_tos, Rscratch, R17_tos); break;
1322 default: ShouldNotReachHere();
1323 }
1324 }
1325
1326 void TemplateTable::idiv() {
1327 transition(itos, itos);
1328
1329 Label Lnormal, Lexception, Ldone;
1330 Register Rdividend = R11_scratch1; // Used by irem.
1331
1332 __ addi(R0, R17_tos, 1);
1333 __ cmplwi(CCR0, R0, 2);
1334 __ bgt(CCR0, Lnormal); // divisor <-1 or >1
1335
1336 __ cmpwi(CCR1, R17_tos, 0);
1337 __ beq(CCR1, Lexception); // divisor == 0
1338
1339 __ pop_i(Rdividend);
1340 __ mullw(R17_tos, Rdividend, R17_tos); // div by +/-1
1341 __ b(Ldone);
1342
1343 __ bind(Lexception);
1344 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArithmeticException_entry);
1345 __ mtctr(R11_scratch1);
1346 __ bctr();
1347
1348 __ align(32, 12);
1349 __ bind(Lnormal);
1350 __ pop_i(Rdividend);
1351 __ divw(R17_tos, Rdividend, R17_tos); // Can't divide minint/-1.
1352 __ bind(Ldone);
1353 }
1354
1355 void TemplateTable::irem() {
1356 transition(itos, itos);
1357
1358 __ mr(R12_scratch2, R17_tos);
1359 idiv();
1360 __ mullw(R17_tos, R17_tos, R12_scratch2);
1361 __ subf(R17_tos, R17_tos, R11_scratch1); // Dividend set by idiv.
1362 }
1363
1364 void TemplateTable::lmul() {
1365 transition(ltos, ltos);
1366
1367 __ pop_l(R11_scratch1);
1368 __ mulld(R17_tos, R11_scratch1, R17_tos);
1369 }
1370
1371 void TemplateTable::ldiv() {
1372 transition(ltos, ltos);
1373
1374 Label Lnormal, Lexception, Ldone;
1375 Register Rdividend = R11_scratch1; // Used by lrem.
1376
1377 __ addi(R0, R17_tos, 1);
1378 __ cmpldi(CCR0, R0, 2);
1379 __ bgt(CCR0, Lnormal); // divisor <-1 or >1
1380
1381 __ cmpdi(CCR1, R17_tos, 0);
1382 __ beq(CCR1, Lexception); // divisor == 0
1383
1384 __ pop_l(Rdividend);
1385 __ mulld(R17_tos, Rdividend, R17_tos); // div by +/-1
1386 __ b(Ldone);
1387
1388 __ bind(Lexception);
1389 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArithmeticException_entry);
1390 __ mtctr(R11_scratch1);
1391 __ bctr();
1392
1393 __ align(32, 12);
1394 __ bind(Lnormal);
1395 __ pop_l(Rdividend);
1396 __ divd(R17_tos, Rdividend, R17_tos); // Can't divide minint/-1.
1397 __ bind(Ldone);
1398 }
1399
1400 void TemplateTable::lrem() {
1401 transition(ltos, ltos);
1402
1403 __ mr(R12_scratch2, R17_tos);
1404 ldiv();
1405 __ mulld(R17_tos, R17_tos, R12_scratch2);
1406 __ subf(R17_tos, R17_tos, R11_scratch1); // Dividend set by ldiv.
1407 }
1408
1409 void TemplateTable::lshl() {
1410 transition(itos, ltos);
1411
1412 __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1413 __ pop_l(R11_scratch1);
1414 __ sld(R17_tos, R11_scratch1, R17_tos);
1415 }
1416
1417 void TemplateTable::lshr() {
1418 transition(itos, ltos);
1419
1420 __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1421 __ pop_l(R11_scratch1);
1422 __ srad(R17_tos, R11_scratch1, R17_tos);
1423 }
1424
1425 void TemplateTable::lushr() {
1426 transition(itos, ltos);
1427
1428 __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1429 __ pop_l(R11_scratch1);
1430 __ srd(R17_tos, R11_scratch1, R17_tos);
1431 }
1432
1433 void TemplateTable::fop2(Operation op) {
1434 transition(ftos, ftos);
1435
1436 switch (op) {
1437 case add: __ pop_f(F0_SCRATCH); __ fadds(F15_ftos, F0_SCRATCH, F15_ftos); break;
1438 case sub: __ pop_f(F0_SCRATCH); __ fsubs(F15_ftos, F0_SCRATCH, F15_ftos); break;
1439 case mul: __ pop_f(F0_SCRATCH); __ fmuls(F15_ftos, F0_SCRATCH, F15_ftos); break;
1440 case div: __ pop_f(F0_SCRATCH); __ fdivs(F15_ftos, F0_SCRATCH, F15_ftos); break;
1441 case rem:
1442 __ pop_f(F1_ARG1);
1443 __ fmr(F2_ARG2, F15_ftos);
1444 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem));
1445 __ fmr(F15_ftos, F1_RET);
1446 break;
1447
1448 default: ShouldNotReachHere();
1449 }
1450 }
1451
1452 void TemplateTable::dop2(Operation op) {
1453 transition(dtos, dtos);
1454
1455 switch (op) {
1456 case add: __ pop_d(F0_SCRATCH); __ fadd(F15_ftos, F0_SCRATCH, F15_ftos); break;
1457 case sub: __ pop_d(F0_SCRATCH); __ fsub(F15_ftos, F0_SCRATCH, F15_ftos); break;
1458 case mul: __ pop_d(F0_SCRATCH); __ fmul(F15_ftos, F0_SCRATCH, F15_ftos); break;
1459 case div: __ pop_d(F0_SCRATCH); __ fdiv(F15_ftos, F0_SCRATCH, F15_ftos); break;
1460 case rem:
1461 __ pop_d(F1_ARG1);
1462 __ fmr(F2_ARG2, F15_ftos);
1463 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem));
1464 __ fmr(F15_ftos, F1_RET);
1465 break;
1466
1467 default: ShouldNotReachHere();
1468 }
1469 }
1470
1471 // Negate the value in the TOS cache.
1472 void TemplateTable::ineg() {
1473 transition(itos, itos);
1474
1475 __ neg(R17_tos, R17_tos);
1476 }
1477
1478 // Negate the value in the TOS cache.
1479 void TemplateTable::lneg() {
1480 transition(ltos, ltos);
1481
1482 __ neg(R17_tos, R17_tos);
1483 }
1484
1485 void TemplateTable::fneg() {
1486 transition(ftos, ftos);
1487
1488 __ fneg(F15_ftos, F15_ftos);
1489 }
1490
1491 void TemplateTable::dneg() {
1492 transition(dtos, dtos);
1493
1494 __ fneg(F15_ftos, F15_ftos);
1495 }
1496
1497 // Increments a local variable in place.
1498 void TemplateTable::iinc() {
1499 transition(vtos, vtos);
1500
1501 const Register Rindex = R11_scratch1,
1502 Rincrement = R0,
1503 Rvalue = R12_scratch2;
1504
1505 locals_index(Rindex); // Load locals index from bytecode stream.
1506 __ lbz(Rincrement, 2, R14_bcp); // Load increment from the bytecode stream.
1507 __ extsb(Rincrement, Rincrement);
1508
1509 __ load_local_int(Rvalue, Rindex, Rindex); // Puts address of local into Rindex.
1510
1511 __ add(Rvalue, Rincrement, Rvalue);
1512 __ stw(Rvalue, 0, Rindex);
1513 }
1514
1515 void TemplateTable::wide_iinc() {
1516 transition(vtos, vtos);
1517
1518 Register Rindex = R11_scratch1,
1519 Rlocals_addr = Rindex,
1520 Rincr = R12_scratch2;
1521 locals_index_wide(Rindex);
1522 __ get_2_byte_integer_at_bcp(4, Rincr, InterpreterMacroAssembler::Signed);
1523 __ load_local_int(R17_tos, Rlocals_addr, Rindex);
1524 __ add(R17_tos, Rincr, R17_tos);
1525 __ stw(R17_tos, 0, Rlocals_addr);
1526 }
1527
1528 void TemplateTable::convert() {
1529 // %%%%% Factor this first part accross platforms
1530 #ifdef ASSERT
1531 TosState tos_in = ilgl;
1532 TosState tos_out = ilgl;
1533 switch (bytecode()) {
1534 case Bytecodes::_i2l: // fall through
1535 case Bytecodes::_i2f: // fall through
1536 case Bytecodes::_i2d: // fall through
1537 case Bytecodes::_i2b: // fall through
1538 case Bytecodes::_i2c: // fall through
1539 case Bytecodes::_i2s: tos_in = itos; break;
1540 case Bytecodes::_l2i: // fall through
1541 case Bytecodes::_l2f: // fall through
1542 case Bytecodes::_l2d: tos_in = ltos; break;
1543 case Bytecodes::_f2i: // fall through
1544 case Bytecodes::_f2l: // fall through
1545 case Bytecodes::_f2d: tos_in = ftos; break;
1546 case Bytecodes::_d2i: // fall through
1547 case Bytecodes::_d2l: // fall through
1548 case Bytecodes::_d2f: tos_in = dtos; break;
1549 default : ShouldNotReachHere();
1550 }
1551 switch (bytecode()) {
1552 case Bytecodes::_l2i: // fall through
1553 case Bytecodes::_f2i: // fall through
1554 case Bytecodes::_d2i: // fall through
1555 case Bytecodes::_i2b: // fall through
1556 case Bytecodes::_i2c: // fall through
1557 case Bytecodes::_i2s: tos_out = itos; break;
1558 case Bytecodes::_i2l: // fall through
1559 case Bytecodes::_f2l: // fall through
1560 case Bytecodes::_d2l: tos_out = ltos; break;
1561 case Bytecodes::_i2f: // fall through
1562 case Bytecodes::_l2f: // fall through
1563 case Bytecodes::_d2f: tos_out = ftos; break;
1564 case Bytecodes::_i2d: // fall through
1565 case Bytecodes::_l2d: // fall through
1566 case Bytecodes::_f2d: tos_out = dtos; break;
1567 default : ShouldNotReachHere();
1568 }
1569 transition(tos_in, tos_out);
1570 #endif
1571
1572 // Conversion
1573 Label done;
1574 switch (bytecode()) {
1575 case Bytecodes::_i2l:
1576 __ extsw(R17_tos, R17_tos);
1577 break;
1578
1579 case Bytecodes::_l2i:
1580 // Nothing to do, we'll continue to work with the lower bits.
1581 break;
1582
1583 case Bytecodes::_i2b:
1584 __ extsb(R17_tos, R17_tos);
1585 break;
1586
1587 case Bytecodes::_i2c:
1588 __ rldicl(R17_tos, R17_tos, 0, 64-2*8);
1589 break;
1590
1591 case Bytecodes::_i2s:
1592 __ extsh(R17_tos, R17_tos);
1593 break;
1594
1595 case Bytecodes::_i2d:
1596 __ extsw(R17_tos, R17_tos);
1597 case Bytecodes::_l2d:
1598 __ move_l_to_d();
1599 __ fcfid(F15_ftos, F15_ftos);
1600 break;
1601
1602 case Bytecodes::_i2f:
1603 __ extsw(R17_tos, R17_tos);
1604 __ move_l_to_d();
1605 if (VM_Version::has_fcfids()) { // fcfids is >= Power7 only
1606 // Comment: alternatively, load with sign extend could be done by lfiwax.
1607 __ fcfids(F15_ftos, F15_ftos);
1608 } else {
1609 __ fcfid(F15_ftos, F15_ftos);
1610 __ frsp(F15_ftos, F15_ftos);
1611 }
1612 break;
1613
1614 case Bytecodes::_l2f:
1615 if (VM_Version::has_fcfids()) { // fcfids is >= Power7 only
1616 __ move_l_to_d();
1617 __ fcfids(F15_ftos, F15_ftos);
1618 } else {
1619 // Avoid rounding problem when result should be 0x3f800001: need fixup code before fcfid+frsp.
1620 __ mr(R3_ARG1, R17_tos);
1621 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::l2f));
1622 __ fmr(F15_ftos, F1_RET);
1623 }
1624 break;
1625
1626 case Bytecodes::_f2d:
1627 // empty
1628 break;
1629
1630 case Bytecodes::_d2f:
1631 __ frsp(F15_ftos, F15_ftos);
1632 break;
1633
1634 case Bytecodes::_d2i:
1635 case Bytecodes::_f2i:
1636 __ fcmpu(CCR0, F15_ftos, F15_ftos);
1637 __ li(R17_tos, 0); // 0 in case of NAN
1638 __ bso(CCR0, done);
1639 __ fctiwz(F15_ftos, F15_ftos);
1640 __ move_d_to_l();
1641 break;
1642
1643 case Bytecodes::_d2l:
1644 case Bytecodes::_f2l:
1645 __ fcmpu(CCR0, F15_ftos, F15_ftos);
1646 __ li(R17_tos, 0); // 0 in case of NAN
1647 __ bso(CCR0, done);
1648 __ fctidz(F15_ftos, F15_ftos);
1649 __ move_d_to_l();
1650 break;
1651
1652 default: ShouldNotReachHere();
1653 }
1654 __ bind(done);
1655 }
1656
1657 // Long compare
1658 void TemplateTable::lcmp() {
1659 transition(ltos, itos);
1660
1661 const Register Rscratch = R11_scratch1;
1662 __ pop_l(Rscratch); // first operand, deeper in stack
1663
1664 __ cmpd(CCR0, Rscratch, R17_tos); // compare
1665 __ mfcr(R17_tos); // set bit 32..33 as follows: <: 0b10, =: 0b00, >: 0b01
1666 __ srwi(Rscratch, R17_tos, 30);
1667 __ srawi(R17_tos, R17_tos, 31);
1668 __ orr(R17_tos, Rscratch, R17_tos); // set result as follows: <: -1, =: 0, >: 1
1669 }
1670
1671 // fcmpl/fcmpg and dcmpl/dcmpg bytecodes
1672 // unordered_result == -1 => fcmpl or dcmpl
1673 // unordered_result == 1 => fcmpg or dcmpg
1674 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1675 const FloatRegister Rfirst = F0_SCRATCH,
1676 Rsecond = F15_ftos;
1677 const Register Rscratch = R11_scratch1;
1678
1679 if (is_float) {
1680 __ pop_f(Rfirst);
1681 } else {
1682 __ pop_d(Rfirst);
1683 }
1684
1685 Label Lunordered, Ldone;
1686 __ fcmpu(CCR0, Rfirst, Rsecond); // compare
1687 if (unordered_result) {
1688 __ bso(CCR0, Lunordered);
1689 }
1690 __ mfcr(R17_tos); // set bit 32..33 as follows: <: 0b10, =: 0b00, >: 0b01
1691 __ srwi(Rscratch, R17_tos, 30);
1692 __ srawi(R17_tos, R17_tos, 31);
1693 __ orr(R17_tos, Rscratch, R17_tos); // set result as follows: <: -1, =: 0, >: 1
1694 if (unordered_result) {
1695 __ b(Ldone);
1696 __ bind(Lunordered);
1697 __ load_const_optimized(R17_tos, unordered_result);
1698 }
1699 __ bind(Ldone);
1700 }
1701
1702 // Branch_conditional which takes TemplateTable::Condition.
1703 void TemplateTable::branch_conditional(ConditionRegister crx, TemplateTable::Condition cc, Label& L, bool invert) {
1704 bool positive = false;
1705 Assembler::Condition cond = Assembler::equal;
1706 switch (cc) {
1707 case TemplateTable::equal: positive = true ; cond = Assembler::equal ; break;
1708 case TemplateTable::not_equal: positive = false; cond = Assembler::equal ; break;
1709 case TemplateTable::less: positive = true ; cond = Assembler::less ; break;
1710 case TemplateTable::less_equal: positive = false; cond = Assembler::greater; break;
1711 case TemplateTable::greater: positive = true ; cond = Assembler::greater; break;
1712 case TemplateTable::greater_equal: positive = false; cond = Assembler::less ; break;
1713 default: ShouldNotReachHere();
1714 }
1715 int bo = (positive != invert) ? Assembler::bcondCRbiIs1 : Assembler::bcondCRbiIs0;
1716 int bi = Assembler::bi0(crx, cond);
1717 __ bc(bo, bi, L);
1718 }
1719
1720 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1721
1722 // Note: on SPARC, we use InterpreterMacroAssembler::if_cmp also.
1723 __ verify_thread();
1724
1725 const Register Rscratch1 = R11_scratch1,
1726 Rscratch2 = R12_scratch2,
1727 Rscratch3 = R3_ARG1,
1728 R4_counters = R4_ARG2,
1729 bumped_count = R31,
1730 Rdisp = R22_tmp2;
1731
1732 __ profile_taken_branch(Rscratch1, bumped_count);
1733
1734 // Get (wide) offset.
1735 if (is_wide) {
1736 __ get_4_byte_integer_at_bcp(1, Rdisp, InterpreterMacroAssembler::Signed);
1737 } else {
1738 __ get_2_byte_integer_at_bcp(1, Rdisp, InterpreterMacroAssembler::Signed);
1739 }
1740
1741 // --------------------------------------------------------------------------
1742 // Handle all the JSR stuff here, then exit.
1743 // It's much shorter and cleaner than intermingling with the
1744 // non-JSR normal-branch stuff occurring below.
1745 if (is_jsr) {
1746 // Compute return address as bci in Otos_i.
1747 __ ld(Rscratch1, in_bytes(Method::const_offset()), R19_method);
1748 __ addi(Rscratch2, R14_bcp, -in_bytes(ConstMethod::codes_offset()) + (is_wide ? 5 : 3));
1749 __ subf(R17_tos, Rscratch1, Rscratch2);
1750
1751 // Bump bcp to target of JSR.
1752 __ add(R14_bcp, Rdisp, R14_bcp);
1753 // Push returnAddress for "ret" on stack.
1754 __ push_ptr(R17_tos);
1755 // And away we go!
1756 __ dispatch_next(vtos, 0 ,true);
1757 return;
1758 }
1759
1760 // --------------------------------------------------------------------------
1761 // Normal (non-jsr) branch handling
1762
1763 // Bump bytecode pointer by displacement (take the branch).
1764 __ add(R14_bcp, Rdisp, R14_bcp); // Add to bc addr.
1765
1766 const bool increment_invocation_counter_for_backward_branches = UseCompiler && UseLoopCounter;
1767 if (increment_invocation_counter_for_backward_branches) {
1768 Label Lforward;
1769
1770 // Check branch direction.
1771 __ cmpdi(CCR0, Rdisp, 0);
1772 __ bgt(CCR0, Lforward);
1773
1774 __ get_method_counters(R19_method, R4_counters, Lforward);
1775
1776 if (TieredCompilation) {
1777 Label Lno_mdo, Loverflow;
1778 const int increment = InvocationCounter::count_increment;
1779 if (ProfileInterpreter) {
1780 Register Rmdo = Rscratch1;
1781
1782 // If no method data exists, go to profile_continue.
1783 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
1784 __ cmpdi(CCR0, Rmdo, 0);
1785 __ beq(CCR0, Lno_mdo);
1786
1787 // Increment backedge counter in the MDO.
1788 const int mdo_bc_offs = in_bytes(MethodData::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
1789 __ lwz(Rscratch2, mdo_bc_offs, Rmdo);
1790 __ lwz(Rscratch3, in_bytes(MethodData::backedge_mask_offset()), Rmdo);
1791 __ addi(Rscratch2, Rscratch2, increment);
1792 __ stw(Rscratch2, mdo_bc_offs, Rmdo);
1793 if (UseOnStackReplacement) {
1794 __ and_(Rscratch3, Rscratch2, Rscratch3);
1795 __ bne(CCR0, Lforward);
1796 __ b(Loverflow);
1797 } else {
1798 __ b(Lforward);
1799 }
1800 }
1801
1802 // If there's no MDO, increment counter in method.
1803 const int mo_bc_offs = in_bytes(MethodCounters::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
1804 __ bind(Lno_mdo);
1805 __ lwz(Rscratch2, mo_bc_offs, R4_counters);
1806 __ lwz(Rscratch3, in_bytes(MethodCounters::backedge_mask_offset()), R4_counters);
1807 __ addi(Rscratch2, Rscratch2, increment);
1808 __ stw(Rscratch2, mo_bc_offs, R4_counters);
1809 if (UseOnStackReplacement) {
1810 __ and_(Rscratch3, Rscratch2, Rscratch3);
1811 __ bne(CCR0, Lforward);
1812 } else {
1813 __ b(Lforward);
1814 }
1815 __ bind(Loverflow);
1816
1817 // Notify point for loop, pass branch bytecode.
1818 __ subf(R4_ARG2, Rdisp, R14_bcp); // Compute branch bytecode (previous bcp).
1819 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true);
1820
1821 // Was an OSR adapter generated?
1822 __ cmpdi(CCR0, R3_RET, 0);
1823 __ beq(CCR0, Lforward);
1824
1825 // Has the nmethod been invalidated already?
1826 __ lbz(R0, nmethod::state_offset(), R3_RET);
1827 __ cmpwi(CCR0, R0, nmethod::in_use);
1828 __ bne(CCR0, Lforward);
1829
1830 // Migrate the interpreter frame off of the stack.
1831 // We can use all registers because we will not return to interpreter from this point.
1832
1833 // Save nmethod.
1834 const Register osr_nmethod = R31;
1835 __ mr(osr_nmethod, R3_RET);
1836 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R11_scratch1);
1837 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), R16_thread);
1838 __ reset_last_Java_frame();
1839 // OSR buffer is in ARG1.
1840
1841 // Remove the interpreter frame.
1842 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1843
1844 // Jump to the osr code.
1845 __ ld(R11_scratch1, nmethod::osr_entry_point_offset(), osr_nmethod);
1846 __ mtlr(R0);
1847 __ mtctr(R11_scratch1);
1848 __ bctr();
1849
1850 } else {
1851
1852 const Register invoke_ctr = Rscratch1;
1853 // Update Backedge branch separately from invocations.
1854 __ increment_backedge_counter(R4_counters, invoke_ctr, Rscratch2, Rscratch3);
1855
1856 if (ProfileInterpreter) {
1857 __ test_invocation_counter_for_mdp(invoke_ctr, R4_counters, Rscratch2, Lforward);
1858 if (UseOnStackReplacement) {
1859 __ test_backedge_count_for_osr(bumped_count, R4_counters, R14_bcp, Rdisp, Rscratch2);
1860 }
1861 } else {
1862 if (UseOnStackReplacement) {
1863 __ test_backedge_count_for_osr(invoke_ctr, R4_counters, R14_bcp, Rdisp, Rscratch2);
1864 }
1865 }
1866 }
1867
1868 __ bind(Lforward);
1869 }
1870 __ dispatch_next(vtos, 0, true);
1871 }
1872
1873 // Helper function for if_cmp* methods below.
1874 // Factored out common compare and branch code.
1875 void TemplateTable::if_cmp_common(Register Rfirst, Register Rsecond, Register Rscratch1, Register Rscratch2, Condition cc, bool is_jint, bool cmp0) {
1876 Label Lnot_taken;
1877 // Note: The condition code we get is the condition under which we
1878 // *fall through*! So we have to inverse the CC here.
1879
1880 if (is_jint) {
1881 if (cmp0) {
1882 __ cmpwi(CCR0, Rfirst, 0);
1883 } else {
1884 __ cmpw(CCR0, Rfirst, Rsecond);
1885 }
1886 } else {
1887 if (cmp0) {
1888 __ cmpdi(CCR0, Rfirst, 0);
1889 } else {
1890 __ cmpd(CCR0, Rfirst, Rsecond);
1891 }
1892 }
1893 branch_conditional(CCR0, cc, Lnot_taken, /*invert*/ true);
1894
1895 // Conition is false => Jump!
1896 branch(false, false);
1897
1898 // Condition is not true => Continue.
1899 __ align(32, 12);
1900 __ bind(Lnot_taken);
1901 __ profile_not_taken_branch(Rscratch1, Rscratch2);
1902 }
1903
1904 // Compare integer values with zero and fall through if CC holds, branch away otherwise.
1905 void TemplateTable::if_0cmp(Condition cc) {
1906 transition(itos, vtos);
1907
1908 if_cmp_common(R17_tos, noreg, R11_scratch1, R12_scratch2, cc, true, true);
1909 }
1910
1911 // Compare integer values and fall through if CC holds, branch away otherwise.
1912 //
1913 // Interface:
1914 // - Rfirst: First operand (older stack value)
1915 // - tos: Second operand (younger stack value)
1916 void TemplateTable::if_icmp(Condition cc) {
1917 transition(itos, vtos);
1918
1919 const Register Rfirst = R0,
1920 Rsecond = R17_tos;
1921
1922 __ pop_i(Rfirst);
1923 if_cmp_common(Rfirst, Rsecond, R11_scratch1, R12_scratch2, cc, true, false);
1924 }
1925
1926 void TemplateTable::if_nullcmp(Condition cc) {
1927 transition(atos, vtos);
1928
1929 if_cmp_common(R17_tos, noreg, R11_scratch1, R12_scratch2, cc, false, true);
1930 }
1931
1932 void TemplateTable::if_acmp(Condition cc) {
1933 transition(atos, vtos);
1934
1935 const Register Rfirst = R0,
1936 Rsecond = R17_tos;
1937
1938 __ pop_ptr(Rfirst);
1939 if_cmp_common(Rfirst, Rsecond, R11_scratch1, R12_scratch2, cc, false, false);
1940 }
1941
1942 void TemplateTable::ret() {
1943 locals_index(R11_scratch1);
1944 __ load_local_ptr(R17_tos, R11_scratch1, R11_scratch1);
1945
1946 __ profile_ret(vtos, R17_tos, R11_scratch1, R12_scratch2);
1947
1948 __ ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method);
1949 __ add(R11_scratch1, R17_tos, R11_scratch1);
1950 __ addi(R14_bcp, R11_scratch1, in_bytes(ConstMethod::codes_offset()));
1951 __ dispatch_next(vtos, 0, true);
1952 }
1953
1954 void TemplateTable::wide_ret() {
1955 transition(vtos, vtos);
1956
1957 const Register Rindex = R3_ARG1,
1958 Rscratch1 = R11_scratch1,
1959 Rscratch2 = R12_scratch2;
1960
1961 locals_index_wide(Rindex);
1962 __ load_local_ptr(R17_tos, R17_tos, Rindex);
1963 __ profile_ret(vtos, R17_tos, Rscratch1, R12_scratch2);
1964 // Tos now contains the bci, compute the bcp from that.
1965 __ ld(Rscratch1, in_bytes(Method::const_offset()), R19_method);
1966 __ addi(Rscratch2, R17_tos, in_bytes(ConstMethod::codes_offset()));
1967 __ add(R14_bcp, Rscratch1, Rscratch2);
1968 __ dispatch_next(vtos, 0, true);
1969 }
1970
1971 void TemplateTable::tableswitch() {
1972 transition(itos, vtos);
1973
1974 Label Ldispatch, Ldefault_case;
1975 Register Rlow_byte = R3_ARG1,
1976 Rindex = Rlow_byte,
1977 Rhigh_byte = R4_ARG2,
1978 Rdef_offset_addr = R5_ARG3, // is going to contain address of default offset
1979 Rscratch1 = R11_scratch1,
1980 Rscratch2 = R12_scratch2,
1981 Roffset = R6_ARG4;
1982
1983 // Align bcp.
1984 __ addi(Rdef_offset_addr, R14_bcp, BytesPerInt);
1985 __ clrrdi(Rdef_offset_addr, Rdef_offset_addr, log2_long((jlong)BytesPerInt));
1986
1987 // Load lo & hi.
1988 __ get_u4(Rlow_byte, Rdef_offset_addr, BytesPerInt, InterpreterMacroAssembler::Unsigned);
1989 __ get_u4(Rhigh_byte, Rdef_offset_addr, 2 *BytesPerInt, InterpreterMacroAssembler::Unsigned);
1990
1991 // Check for default case (=index outside [low,high]).
1992 __ cmpw(CCR0, R17_tos, Rlow_byte);
1993 __ cmpw(CCR1, R17_tos, Rhigh_byte);
1994 __ blt(CCR0, Ldefault_case);
1995 __ bgt(CCR1, Ldefault_case);
1996
1997 // Lookup dispatch offset.
1998 __ sub(Rindex, R17_tos, Rlow_byte);
1999 __ extsw(Rindex, Rindex);
2000 __ profile_switch_case(Rindex, Rhigh_byte /* scratch */, Rscratch1, Rscratch2);
2001 __ sldi(Rindex, Rindex, LogBytesPerInt);
2002 __ addi(Rindex, Rindex, 3 * BytesPerInt);
2003 #if defined(VM_LITTLE_ENDIAN)
2004 __ lwbrx(Roffset, Rdef_offset_addr, Rindex);
2005 __ extsw(Roffset, Roffset);
2006 #else
2007 __ lwax(Roffset, Rdef_offset_addr, Rindex);
2008 #endif
2009 __ b(Ldispatch);
2010
2011 __ bind(Ldefault_case);
2012 __ profile_switch_default(Rhigh_byte, Rscratch1);
2013 __ get_u4(Roffset, Rdef_offset_addr, 0, InterpreterMacroAssembler::Signed);
2014
2015 __ bind(Ldispatch);
2016
2017 __ add(R14_bcp, Roffset, R14_bcp);
2018 __ dispatch_next(vtos, 0, true);
2019 }
2020
2021 void TemplateTable::lookupswitch() {
2022 transition(itos, itos);
2023 __ stop("lookupswitch bytecode should have been rewritten");
2024 }
2025
2026 // Table switch using linear search through cases.
2027 // Bytecode stream format:
2028 // Bytecode (1) | 4-byte padding | default offset (4) | count (4) | value/offset pair1 (8) | value/offset pair2 (8) | ...
2029 // Note: Everything is big-endian format here.
2030 void TemplateTable::fast_linearswitch() {
2031 transition(itos, vtos);
2032
2033 Label Lloop_entry, Lsearch_loop, Lcontinue_execution, Ldefault_case;
2034 Register Rcount = R3_ARG1,
2035 Rcurrent_pair = R4_ARG2,
2036 Rdef_offset_addr = R5_ARG3, // Is going to contain address of default offset.
2037 Roffset = R31, // Might need to survive C call.
2038 Rvalue = R12_scratch2,
2039 Rscratch = R11_scratch1,
2040 Rcmp_value = R17_tos;
2041
2042 // Align bcp.
2043 __ addi(Rdef_offset_addr, R14_bcp, BytesPerInt);
2044 __ clrrdi(Rdef_offset_addr, Rdef_offset_addr, log2_long((jlong)BytesPerInt));
2045
2046 // Setup loop counter and limit.
2047 __ get_u4(Rcount, Rdef_offset_addr, BytesPerInt, InterpreterMacroAssembler::Unsigned);
2048 __ addi(Rcurrent_pair, Rdef_offset_addr, 2 * BytesPerInt); // Rcurrent_pair now points to first pair.
2049
2050 __ mtctr(Rcount);
2051 __ cmpwi(CCR0, Rcount, 0);
2052 __ bne(CCR0, Lloop_entry);
2053
2054 // Default case
2055 __ bind(Ldefault_case);
2056 __ get_u4(Roffset, Rdef_offset_addr, 0, InterpreterMacroAssembler::Signed);
2057 if (ProfileInterpreter) {
2058 __ profile_switch_default(Rdef_offset_addr, Rcount/* scratch */);
2059 }
2060 __ b(Lcontinue_execution);
2061
2062 // Next iteration
2063 __ bind(Lsearch_loop);
2064 __ bdz(Ldefault_case);
2065 __ addi(Rcurrent_pair, Rcurrent_pair, 2 * BytesPerInt);
2066 __ bind(Lloop_entry);
2067 __ get_u4(Rvalue, Rcurrent_pair, 0, InterpreterMacroAssembler::Unsigned);
2068 __ cmpw(CCR0, Rvalue, Rcmp_value);
2069 __ bne(CCR0, Lsearch_loop);
2070
2071 // Found, load offset.
2072 __ get_u4(Roffset, Rcurrent_pair, BytesPerInt, InterpreterMacroAssembler::Signed);
2073 // Calculate case index and profile
2074 __ mfctr(Rcurrent_pair);
2075 if (ProfileInterpreter) {
2076 __ sub(Rcurrent_pair, Rcount, Rcurrent_pair);
2077 __ profile_switch_case(Rcurrent_pair, Rcount /*scratch*/, Rdef_offset_addr/*scratch*/, Rscratch);
2078 }
2079
2080 __ bind(Lcontinue_execution);
2081 __ add(R14_bcp, Roffset, R14_bcp);
2082 __ dispatch_next(vtos, 0, true);
2083 }
2084
2085 // Table switch using binary search (value/offset pairs are ordered).
2086 // Bytecode stream format:
2087 // Bytecode (1) | 4-byte padding | default offset (4) | count (4) | value/offset pair1 (8) | value/offset pair2 (8) | ...
2088 // Note: Everything is big-endian format here. So on little endian machines, we have to revers offset and count and cmp value.
2089 void TemplateTable::fast_binaryswitch() {
2090
2091 transition(itos, vtos);
2092 // Implementation using the following core algorithm: (copied from Intel)
2093 //
2094 // int binary_search(int key, LookupswitchPair* array, int n) {
2095 // // Binary search according to "Methodik des Programmierens" by
2096 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2097 // int i = 0;
2098 // int j = n;
2099 // while (i+1 < j) {
2100 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2101 // // with Q: for all i: 0 <= i < n: key < a[i]
2102 // // where a stands for the array and assuming that the (inexisting)
2103 // // element a[n] is infinitely big.
2104 // int h = (i + j) >> 1;
2105 // // i < h < j
2106 // if (key < array[h].fast_match()) {
2107 // j = h;
2108 // } else {
2109 // i = h;
2110 // }
2111 // }
2112 // // R: a[i] <= key < a[i+1] or Q
2113 // // (i.e., if key is within array, i is the correct index)
2114 // return i;
2115 // }
2116
2117 // register allocation
2118 const Register Rkey = R17_tos; // already set (tosca)
2119 const Register Rarray = R3_ARG1;
2120 const Register Ri = R4_ARG2;
2121 const Register Rj = R5_ARG3;
2122 const Register Rh = R6_ARG4;
2123 const Register Rscratch = R11_scratch1;
2124
2125 const int log_entry_size = 3;
2126 const int entry_size = 1 << log_entry_size;
2127
2128 Label found;
2129
2130 // Find Array start,
2131 __ addi(Rarray, R14_bcp, 3 * BytesPerInt);
2132 __ clrrdi(Rarray, Rarray, log2_long((jlong)BytesPerInt));
2133
2134 // initialize i & j
2135 __ li(Ri,0);
2136 __ get_u4(Rj, Rarray, -BytesPerInt, InterpreterMacroAssembler::Unsigned);
2137
2138 // and start.
2139 Label entry;
2140 __ b(entry);
2141
2142 // binary search loop
2143 { Label loop;
2144 __ bind(loop);
2145 // int h = (i + j) >> 1;
2146 __ srdi(Rh, Rh, 1);
2147 // if (key < array[h].fast_match()) {
2148 // j = h;
2149 // } else {
2150 // i = h;
2151 // }
2152 __ sldi(Rscratch, Rh, log_entry_size);
2153 #if defined(VM_LITTLE_ENDIAN)
2154 __ lwbrx(Rscratch, Rscratch, Rarray);
2155 #else
2156 __ lwzx(Rscratch, Rscratch, Rarray);
2157 #endif
2158
2159 // if (key < current value)
2160 // Rh = Rj
2161 // else
2162 // Rh = Ri
2163 Label Lgreater;
2164 __ cmpw(CCR0, Rkey, Rscratch);
2165 __ bge(CCR0, Lgreater);
2166 __ mr(Rj, Rh);
2167 __ b(entry);
2168 __ bind(Lgreater);
2169 __ mr(Ri, Rh);
2170
2171 // while (i+1 < j)
2172 __ bind(entry);
2173 __ addi(Rscratch, Ri, 1);
2174 __ cmpw(CCR0, Rscratch, Rj);
2175 __ add(Rh, Ri, Rj); // start h = i + j >> 1;
2176
2177 __ blt(CCR0, loop);
2178 }
2179
2180 // End of binary search, result index is i (must check again!).
2181 Label default_case;
2182 Label continue_execution;
2183 if (ProfileInterpreter) {
2184 __ mr(Rh, Ri); // Save index in i for profiling.
2185 }
2186 // Ri = value offset
2187 __ sldi(Ri, Ri, log_entry_size);
2188 __ add(Ri, Ri, Rarray);
2189 __ get_u4(Rscratch, Ri, 0, InterpreterMacroAssembler::Unsigned);
2190
2191 Label not_found;
2192 // Ri = offset offset
2193 __ cmpw(CCR0, Rkey, Rscratch);
2194 __ beq(CCR0, not_found);
2195 // entry not found -> j = default offset
2196 __ get_u4(Rj, Rarray, -2 * BytesPerInt, InterpreterMacroAssembler::Unsigned);
2197 __ b(default_case);
2198
2199 __ bind(not_found);
2200 // entry found -> j = offset
2201 __ profile_switch_case(Rh, Rj, Rscratch, Rkey);
2202 __ get_u4(Rj, Ri, BytesPerInt, InterpreterMacroAssembler::Unsigned);
2203
2204 if (ProfileInterpreter) {
2205 __ b(continue_execution);
2206 }
2207
2208 __ bind(default_case); // fall through (if not profiling)
2209 __ profile_switch_default(Ri, Rscratch);
2210
2211 __ bind(continue_execution);
2212
2213 __ extsw(Rj, Rj);
2214 __ add(R14_bcp, Rj, R14_bcp);
2215 __ dispatch_next(vtos, 0 , true);
2216 }
2217
2218 void TemplateTable::_return(TosState state) {
2219 transition(state, state);
2220 assert(_desc->calls_vm(),
2221 "inconsistent calls_vm information"); // call in remove_activation
2222
2223 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2224
2225 Register Rscratch = R11_scratch1,
2226 Rklass = R12_scratch2,
2227 Rklass_flags = Rklass;
2228 Label Lskip_register_finalizer;
2229
2230 // Check if the method has the FINALIZER flag set and call into the VM to finalize in this case.
2231 assert(state == vtos, "only valid state");
2232 __ ld(R17_tos, 0, R18_locals);
2233
2234 // Load klass of this obj.
2235 __ load_klass(Rklass, R17_tos);
2236 __ lwz(Rklass_flags, in_bytes(Klass::access_flags_offset()), Rklass);
2237 __ testbitdi(CCR0, R0, Rklass_flags, exact_log2(JVM_ACC_HAS_FINALIZER));
2238 __ bfalse(CCR0, Lskip_register_finalizer);
2239
2240 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), R17_tos /* obj */);
2241
2242 __ align(32, 12);
2243 __ bind(Lskip_register_finalizer);
2244 }
2245
2246 if (SafepointMechanism::uses_thread_local_poll() && _desc->bytecode() != Bytecodes::_return_register_finalizer) {
2247 Label no_safepoint;
2248 __ ld(R11_scratch1, in_bytes(Thread::polling_page_offset()), R16_thread);
2249 __ andi_(R11_scratch1, R11_scratch1, SafepointMechanism::poll_bit());
2250 __ beq(CCR0, no_safepoint);
2251 __ push(state);
2252 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::at_safepoint));
2253 __ pop(state);
2254 __ bind(no_safepoint);
2255 }
2256
2257 // Move the result value into the correct register and remove memory stack frame.
2258 __ remove_activation(state, /* throw_monitor_exception */ true);
2259 // Restoration of lr done by remove_activation.
2260 switch (state) {
2261 // Narrow result if state is itos but result type is smaller.
2262 // Need to narrow in the return bytecode rather than in generate_return_entry
2263 // since compiled code callers expect the result to already be narrowed.
2264 case itos: __ narrow(R17_tos); /* fall through */
2265 case ltos:
2266 case atos: __ mr(R3_RET, R17_tos); break;
2267 case ftos:
2268 case dtos: __ fmr(F1_RET, F15_ftos); break;
2269 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
2270 // to get visible before the reference to the object gets stored anywhere.
2271 __ membar(Assembler::StoreStore); break;
2272 default : ShouldNotReachHere();
2273 }
2274 __ blr();
2275 }
2276
2277 // ============================================================================
2278 // Constant pool cache access
2279 //
2280 // Memory ordering:
2281 //
2282 // Like done in C++ interpreter, we load the fields
2283 // - _indices
2284 // - _f12_oop
2285 // acquired, because these are asked if the cache is already resolved. We don't
2286 // want to float loads above this check.
2287 // See also comments in ConstantPoolCacheEntry::bytecode_1(),
2288 // ConstantPoolCacheEntry::bytecode_2() and ConstantPoolCacheEntry::f1();
2289
2290 // Call into the VM if call site is not yet resolved
2291 //
2292 // Input regs:
2293 // - None, all passed regs are outputs.
2294 //
2295 // Returns:
2296 // - Rcache: The const pool cache entry that contains the resolved result.
2297 // - Rresult: Either noreg or output for f1/f2.
2298 //
2299 // Kills:
2300 // - Rscratch
2301 void TemplateTable::resolve_cache_and_index(int byte_no, Register Rcache, Register Rscratch, size_t index_size) {
2302
2303 __ get_cache_and_index_at_bcp(Rcache, 1, index_size);
2304 Label Lresolved, Ldone;
2305
2306 Bytecodes::Code code = bytecode();
2307 switch (code) {
2308 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2309 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2310 }
2311
2312 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2313 // We are resolved if the indices offset contains the current bytecode.
2314 #if defined(VM_LITTLE_ENDIAN)
2315 __ lbz(Rscratch, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + byte_no + 1, Rcache);
2316 #else
2317 __ lbz(Rscratch, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 7 - (byte_no + 1), Rcache);
2318 #endif
2319 // Acquire by cmp-br-isync (see below).
2320 __ cmpdi(CCR0, Rscratch, (int)code);
2321 __ beq(CCR0, Lresolved);
2322
2323 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2324 __ li(R4_ARG2, code);
2325 __ call_VM(noreg, entry, R4_ARG2, true);
2326
2327 // Update registers with resolved info.
2328 __ get_cache_and_index_at_bcp(Rcache, 1, index_size);
2329 __ b(Ldone);
2330
2331 __ bind(Lresolved);
2332 __ isync(); // Order load wrt. succeeding loads.
2333 __ bind(Ldone);
2334 }
2335
2336 // Load the constant pool cache entry at field accesses into registers.
2337 // The Rcache and Rindex registers must be set before call.
2338 // Input:
2339 // - Rcache, Rindex
2340 // Output:
2341 // - Robj, Roffset, Rflags
2342 void TemplateTable::load_field_cp_cache_entry(Register Robj,
2343 Register Rcache,
2344 Register Rindex /* unused on PPC64 */,
2345 Register Roffset,
2346 Register Rflags,
2347 bool is_static = false) {
2348 assert_different_registers(Rcache, Rflags, Roffset);
2349 // assert(Rindex == noreg, "parameter not used on PPC64");
2350
2351 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2352 __ ld(Rflags, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::flags_offset()), Rcache);
2353 __ ld(Roffset, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::f2_offset()), Rcache);
2354 if (is_static) {
2355 __ ld(Robj, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::f1_offset()), Rcache);
2356 __ ld(Robj, in_bytes(Klass::java_mirror_offset()), Robj);
2357 __ resolve_oop_handle(Robj);
2358 // Acquire not needed here. Following access has an address dependency on this value.
2359 }
2360 }
2361
2362 // Load the constant pool cache entry at invokes into registers.
2363 // Resolve if necessary.
2364
2365 // Input Registers:
2366 // - None, bcp is used, though
2367 //
2368 // Return registers:
2369 // - Rmethod (f1 field or f2 if invokevirtual)
2370 // - Ritable_index (f2 field)
2371 // - Rflags (flags field)
2372 //
2373 // Kills:
2374 // - R21
2375 //
2376 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2377 Register Rmethod,
2378 Register Ritable_index,
2379 Register Rflags,
2380 bool is_invokevirtual,
2381 bool is_invokevfinal,
2382 bool is_invokedynamic) {
2383
2384 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2385 // Determine constant pool cache field offsets.
2386 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2387 const int method_offset = in_bytes(cp_base_offset + (is_invokevirtual ? ConstantPoolCacheEntry::f2_offset() : ConstantPoolCacheEntry::f1_offset()));
2388 const int flags_offset = in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset());
2389 // Access constant pool cache fields.
2390 const int index_offset = in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset());
2391
2392 Register Rcache = R21_tmp1; // Note: same register as R21_sender_SP.
2393
2394 if (is_invokevfinal) {
2395 assert(Ritable_index == noreg, "register not used");
2396 // Already resolved.
2397 __ get_cache_and_index_at_bcp(Rcache, 1);
2398 } else {
2399 resolve_cache_and_index(byte_no, Rcache, R0, is_invokedynamic ? sizeof(u4) : sizeof(u2));
2400 }
2401
2402 __ ld(Rmethod, method_offset, Rcache);
2403 __ ld(Rflags, flags_offset, Rcache);
2404
2405 if (Ritable_index != noreg) {
2406 __ ld(Ritable_index, index_offset, Rcache);
2407 }
2408 }
2409
2410 // ============================================================================
2411 // Field access
2412
2413 // Volatile variables demand their effects be made known to all CPU's
2414 // in order. Store buffers on most chips allow reads & writes to
2415 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2416 // without some kind of memory barrier (i.e., it's not sufficient that
2417 // the interpreter does not reorder volatile references, the hardware
2418 // also must not reorder them).
2419 //
2420 // According to the new Java Memory Model (JMM):
2421 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2422 // writes act as aquire & release, so:
2423 // (2) A read cannot let unrelated NON-volatile memory refs that
2424 // happen after the read float up to before the read. It's OK for
2425 // non-volatile memory refs that happen before the volatile read to
2426 // float down below it.
2427 // (3) Similar a volatile write cannot let unrelated NON-volatile
2428 // memory refs that happen BEFORE the write float down to after the
2429 // write. It's OK for non-volatile memory refs that happen after the
2430 // volatile write to float up before it.
2431 //
2432 // We only put in barriers around volatile refs (they are expensive),
2433 // not _between_ memory refs (that would require us to track the
2434 // flavor of the previous memory refs). Requirements (2) and (3)
2435 // require some barriers before volatile stores and after volatile
2436 // loads. These nearly cover requirement (1) but miss the
2437 // volatile-store-volatile-load case. This final case is placed after
2438 // volatile-stores although it could just as well go before
2439 // volatile-loads.
2440
2441 // The registers cache and index expected to be set before call.
2442 // Correct values of the cache and index registers are preserved.
2443 // Kills:
2444 // Rcache (if has_tos)
2445 // Rscratch
2446 void TemplateTable::jvmti_post_field_access(Register Rcache, Register Rscratch, bool is_static, bool has_tos) {
2447
2448 assert_different_registers(Rcache, Rscratch);
2449
2450 if (JvmtiExport::can_post_field_access()) {
2451 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2452 Label Lno_field_access_post;
2453
2454 // Check if post field access in enabled.
2455 int offs = __ load_const_optimized(Rscratch, JvmtiExport::get_field_access_count_addr(), R0, true);
2456 __ lwz(Rscratch, offs, Rscratch);
2457
2458 __ cmpwi(CCR0, Rscratch, 0);
2459 __ beq(CCR0, Lno_field_access_post);
2460
2461 // Post access enabled - do it!
2462 __ addi(Rcache, Rcache, in_bytes(cp_base_offset));
2463 if (is_static) {
2464 __ li(R17_tos, 0);
2465 } else {
2466 if (has_tos) {
2467 // The fast bytecode versions have obj ptr in register.
2468 // Thus, save object pointer before call_VM() clobbers it
2469 // put object on tos where GC wants it.
2470 __ push_ptr(R17_tos);
2471 } else {
2472 // Load top of stack (do not pop the value off the stack).
2473 __ ld(R17_tos, Interpreter::expr_offset_in_bytes(0), R15_esp);
2474 }
2475 __ verify_oop(R17_tos);
2476 }
2477 // tos: object pointer or NULL if static
2478 // cache: cache entry pointer
2479 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), R17_tos, Rcache);
2480 if (!is_static && has_tos) {
2481 // Restore object pointer.
2482 __ pop_ptr(R17_tos);
2483 __ verify_oop(R17_tos);
2484 } else {
2485 // Cache is still needed to get class or obj.
2486 __ get_cache_and_index_at_bcp(Rcache, 1);
2487 }
2488
2489 __ align(32, 12);
2490 __ bind(Lno_field_access_post);
2491 }
2492 }
2493
2494 // kills R11_scratch1
2495 void TemplateTable::pop_and_check_object(Register Roop) {
2496 Register Rtmp = R11_scratch1;
2497
2498 assert_different_registers(Rtmp, Roop);
2499 __ pop_ptr(Roop);
2500 // For field access must check obj.
2501 __ null_check_throw(Roop, -1, Rtmp);
2502 __ verify_oop(Roop);
2503 }
2504
2505 // PPC64: implement volatile loads as fence-store-acquire.
2506 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2507 transition(vtos, vtos);
2508
2509 Label Lacquire, Lisync;
2510
2511 const Register Rcache = R3_ARG1,
2512 Rclass_or_obj = R22_tmp2,
2513 Roffset = R23_tmp3,
2514 Rflags = R31,
2515 Rbtable = R5_ARG3,
2516 Rbc = R6_ARG4,
2517 Rscratch = R12_scratch2;
2518
2519 static address field_branch_table[number_of_states],
2520 static_branch_table[number_of_states];
2521
2522 address* branch_table = (is_static || rc == may_not_rewrite) ? static_branch_table : field_branch_table;
2523
2524 // Get field offset.
2525 resolve_cache_and_index(byte_no, Rcache, Rscratch, sizeof(u2));
2526
2527 // JVMTI support
2528 jvmti_post_field_access(Rcache, Rscratch, is_static, false);
2529
2530 // Load after possible GC.
2531 load_field_cp_cache_entry(Rclass_or_obj, Rcache, noreg, Roffset, Rflags, is_static);
2532
2533 // Load pointer to branch table.
2534 __ load_const_optimized(Rbtable, (address)branch_table, Rscratch);
2535
2536 // Get volatile flag.
2537 __ rldicl(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2538 // Note: sync is needed before volatile load on PPC64.
2539
2540 // Check field type.
2541 __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2542
2543 #ifdef ASSERT
2544 Label LFlagInvalid;
2545 __ cmpldi(CCR0, Rflags, number_of_states);
2546 __ bge(CCR0, LFlagInvalid);
2547 #endif
2548
2549 // Load from branch table and dispatch (volatile case: one instruction ahead).
2550 __ sldi(Rflags, Rflags, LogBytesPerWord);
2551 __ cmpwi(CCR6, Rscratch, 1); // Volatile?
2552 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2553 __ sldi(Rscratch, Rscratch, exact_log2(BytesPerInstWord)); // Volatile ? size of 1 instruction : 0.
2554 }
2555 __ ldx(Rbtable, Rbtable, Rflags);
2556
2557 // Get the obj from stack.
2558 if (!is_static) {
2559 pop_and_check_object(Rclass_or_obj); // Kills R11_scratch1.
2560 } else {
2561 __ verify_oop(Rclass_or_obj);
2562 }
2563
2564 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2565 __ subf(Rbtable, Rscratch, Rbtable); // Point to volatile/non-volatile entry point.
2566 }
2567 __ mtctr(Rbtable);
2568 __ bctr();
2569
2570 #ifdef ASSERT
2571 __ bind(LFlagInvalid);
2572 __ stop("got invalid flag", 0x654);
2573 #endif
2574
2575 if (!is_static && rc == may_not_rewrite) {
2576 // We reuse the code from is_static. It's jumped to via the table above.
2577 return;
2578 }
2579
2580 #ifdef ASSERT
2581 // __ bind(Lvtos);
2582 address pc_before_fence = __ pc();
2583 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2584 assert(__ pc() - pc_before_fence == (ptrdiff_t)BytesPerInstWord, "must be single instruction");
2585 assert(branch_table[vtos] == 0, "can't compute twice");
2586 branch_table[vtos] = __ pc(); // non-volatile_entry point
2587 __ stop("vtos unexpected", 0x655);
2588 #endif
2589
2590 __ align(32, 28, 28); // Align load.
2591 // __ bind(Ldtos);
2592 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2593 assert(branch_table[dtos] == 0, "can't compute twice");
2594 branch_table[dtos] = __ pc(); // non-volatile_entry point
2595 __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
2596 __ push(dtos);
2597 if (!is_static && rc == may_rewrite) {
2598 patch_bytecode(Bytecodes::_fast_dgetfield, Rbc, Rscratch);
2599 }
2600 {
2601 Label acquire_double;
2602 __ beq(CCR6, acquire_double); // Volatile?
2603 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2604
2605 __ bind(acquire_double);
2606 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
2607 __ beq_predict_taken(CCR0, Lisync);
2608 __ b(Lisync); // In case of NAN.
2609 }
2610
2611 __ align(32, 28, 28); // Align load.
2612 // __ bind(Lftos);
2613 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2614 assert(branch_table[ftos] == 0, "can't compute twice");
2615 branch_table[ftos] = __ pc(); // non-volatile_entry point
2616 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
2617 __ push(ftos);
2618 if (!is_static && rc == may_rewrite) {
2619 patch_bytecode(Bytecodes::_fast_fgetfield, Rbc, Rscratch);
2620 }
2621 {
2622 Label acquire_float;
2623 __ beq(CCR6, acquire_float); // Volatile?
2624 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2625
2626 __ bind(acquire_float);
2627 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
2628 __ beq_predict_taken(CCR0, Lisync);
2629 __ b(Lisync); // In case of NAN.
2630 }
2631
2632 __ align(32, 28, 28); // Align load.
2633 // __ bind(Litos);
2634 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2635 assert(branch_table[itos] == 0, "can't compute twice");
2636 branch_table[itos] = __ pc(); // non-volatile_entry point
2637 __ lwax(R17_tos, Rclass_or_obj, Roffset);
2638 __ push(itos);
2639 if (!is_static && rc == may_rewrite) {
2640 patch_bytecode(Bytecodes::_fast_igetfield, Rbc, Rscratch);
2641 }
2642 __ beq(CCR6, Lacquire); // Volatile?
2643 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2644
2645 __ align(32, 28, 28); // Align load.
2646 // __ bind(Lltos);
2647 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2648 assert(branch_table[ltos] == 0, "can't compute twice");
2649 branch_table[ltos] = __ pc(); // non-volatile_entry point
2650 __ ldx(R17_tos, Rclass_or_obj, Roffset);
2651 __ push(ltos);
2652 if (!is_static && rc == may_rewrite) {
2653 patch_bytecode(Bytecodes::_fast_lgetfield, Rbc, Rscratch);
2654 }
2655 __ beq(CCR6, Lacquire); // Volatile?
2656 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2657
2658 __ align(32, 28, 28); // Align load.
2659 // __ bind(Lbtos);
2660 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2661 assert(branch_table[btos] == 0, "can't compute twice");
2662 branch_table[btos] = __ pc(); // non-volatile_entry point
2663 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
2664 __ extsb(R17_tos, R17_tos);
2665 __ push(btos);
2666 if (!is_static && rc == may_rewrite) {
2667 patch_bytecode(Bytecodes::_fast_bgetfield, Rbc, Rscratch);
2668 }
2669 __ beq(CCR6, Lacquire); // Volatile?
2670 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2671
2672 __ align(32, 28, 28); // Align load.
2673 // __ bind(Lztos); (same code as btos)
2674 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2675 assert(branch_table[ztos] == 0, "can't compute twice");
2676 branch_table[ztos] = __ pc(); // non-volatile_entry point
2677 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
2678 __ push(ztos);
2679 if (!is_static && rc == may_rewrite) {
2680 // use btos rewriting, no truncating to t/f bit is needed for getfield.
2681 patch_bytecode(Bytecodes::_fast_bgetfield, Rbc, Rscratch);
2682 }
2683 __ beq(CCR6, Lacquire); // Volatile?
2684 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2685
2686 __ align(32, 28, 28); // Align load.
2687 // __ bind(Lctos);
2688 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2689 assert(branch_table[ctos] == 0, "can't compute twice");
2690 branch_table[ctos] = __ pc(); // non-volatile_entry point
2691 __ lhzx(R17_tos, Rclass_or_obj, Roffset);
2692 __ push(ctos);
2693 if (!is_static && rc == may_rewrite) {
2694 patch_bytecode(Bytecodes::_fast_cgetfield, Rbc, Rscratch);
2695 }
2696 __ beq(CCR6, Lacquire); // Volatile?
2697 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2698
2699 __ align(32, 28, 28); // Align load.
2700 // __ bind(Lstos);
2701 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2702 assert(branch_table[stos] == 0, "can't compute twice");
2703 branch_table[stos] = __ pc(); // non-volatile_entry point
2704 __ lhax(R17_tos, Rclass_or_obj, Roffset);
2705 __ push(stos);
2706 if (!is_static && rc == may_rewrite) {
2707 patch_bytecode(Bytecodes::_fast_sgetfield, Rbc, Rscratch);
2708 }
2709 __ beq(CCR6, Lacquire); // Volatile?
2710 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2711
2712 __ align(32, 28, 28); // Align load.
2713 // __ bind(Latos);
2714 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2715 assert(branch_table[atos] == 0, "can't compute twice");
2716 branch_table[atos] = __ pc(); // non-volatile_entry point
2717 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
2718 __ verify_oop(R17_tos);
2719 __ push(atos);
2720 //__ dcbt(R17_tos); // prefetch
2721 if (!is_static && rc == may_rewrite) {
2722 patch_bytecode(Bytecodes::_fast_agetfield, Rbc, Rscratch);
2723 }
2724 __ beq(CCR6, Lacquire); // Volatile?
2725 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2726
2727 __ align(32, 12);
2728 __ bind(Lacquire);
2729 __ twi_0(R17_tos);
2730 __ bind(Lisync);
2731 __ isync(); // acquire
2732
2733 #ifdef ASSERT
2734 for (int i = 0; i<number_of_states; ++i) {
2735 assert(branch_table[i], "get initialization");
2736 //tty->print_cr("get: %s_branch_table[%d] = 0x%llx (opcode 0x%llx)",
2737 // is_static ? "static" : "field", i, branch_table[i], *((unsigned int*)branch_table[i]));
2738 }
2739 #endif
2740 }
2741
2742 void TemplateTable::getfield(int byte_no) {
2743 getfield_or_static(byte_no, false);
2744 }
2745
2746 void TemplateTable::nofast_getfield(int byte_no) {
2747 getfield_or_static(byte_no, false, may_not_rewrite);
2748 }
2749
2750 void TemplateTable::getstatic(int byte_no) {
2751 getfield_or_static(byte_no, true);
2752 }
2753
2754 // The registers cache and index expected to be set before call.
2755 // The function may destroy various registers, just not the cache and index registers.
2756 void TemplateTable::jvmti_post_field_mod(Register Rcache, Register Rscratch, bool is_static) {
2757
2758 assert_different_registers(Rcache, Rscratch, R6_ARG4);
2759
2760 if (JvmtiExport::can_post_field_modification()) {
2761 Label Lno_field_mod_post;
2762
2763 // Check if post field access in enabled.
2764 int offs = __ load_const_optimized(Rscratch, JvmtiExport::get_field_modification_count_addr(), R0, true);
2765 __ lwz(Rscratch, offs, Rscratch);
2766
2767 __ cmpwi(CCR0, Rscratch, 0);
2768 __ beq(CCR0, Lno_field_mod_post);
2769
2770 // Do the post
2771 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2772 const Register Robj = Rscratch;
2773
2774 __ addi(Rcache, Rcache, in_bytes(cp_base_offset));
2775 if (is_static) {
2776 // Life is simple. Null out the object pointer.
2777 __ li(Robj, 0);
2778 } else {
2779 // In case of the fast versions, value lives in registers => put it back on tos.
2780 int offs = Interpreter::expr_offset_in_bytes(0);
2781 Register base = R15_esp;
2782 switch(bytecode()) {
2783 case Bytecodes::_fast_aputfield: __ push_ptr(); offs+= Interpreter::stackElementSize; break;
2784 case Bytecodes::_fast_iputfield: // Fall through
2785 case Bytecodes::_fast_bputfield: // Fall through
2786 case Bytecodes::_fast_zputfield: // Fall through
2787 case Bytecodes::_fast_cputfield: // Fall through
2788 case Bytecodes::_fast_sputfield: __ push_i(); offs+= Interpreter::stackElementSize; break;
2789 case Bytecodes::_fast_lputfield: __ push_l(); offs+=2*Interpreter::stackElementSize; break;
2790 case Bytecodes::_fast_fputfield: __ push_f(); offs+= Interpreter::stackElementSize; break;
2791 case Bytecodes::_fast_dputfield: __ push_d(); offs+=2*Interpreter::stackElementSize; break;
2792 default: {
2793 offs = 0;
2794 base = Robj;
2795 const Register Rflags = Robj;
2796 Label is_one_slot;
2797 // Life is harder. The stack holds the value on top, followed by the
2798 // object. We don't know the size of the value, though; it could be
2799 // one or two words depending on its type. As a result, we must find
2800 // the type to determine where the object is.
2801 __ ld(Rflags, in_bytes(ConstantPoolCacheEntry::flags_offset()), Rcache); // Big Endian
2802 __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2803
2804 __ cmpwi(CCR0, Rflags, ltos);
2805 __ cmpwi(CCR1, Rflags, dtos);
2806 __ addi(base, R15_esp, Interpreter::expr_offset_in_bytes(1));
2807 __ crnor(CCR0, Assembler::equal, CCR1, Assembler::equal);
2808 __ beq(CCR0, is_one_slot);
2809 __ addi(base, R15_esp, Interpreter::expr_offset_in_bytes(2));
2810 __ bind(is_one_slot);
2811 break;
2812 }
2813 }
2814 __ ld(Robj, offs, base);
2815 __ verify_oop(Robj);
2816 }
2817
2818 __ addi(R6_ARG4, R15_esp, Interpreter::expr_offset_in_bytes(0));
2819 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), Robj, Rcache, R6_ARG4);
2820 __ get_cache_and_index_at_bcp(Rcache, 1);
2821
2822 // In case of the fast versions, value lives in registers => put it back on tos.
2823 switch(bytecode()) {
2824 case Bytecodes::_fast_aputfield: __ pop_ptr(); break;
2825 case Bytecodes::_fast_iputfield: // Fall through
2826 case Bytecodes::_fast_bputfield: // Fall through
2827 case Bytecodes::_fast_zputfield: // Fall through
2828 case Bytecodes::_fast_cputfield: // Fall through
2829 case Bytecodes::_fast_sputfield: __ pop_i(); break;
2830 case Bytecodes::_fast_lputfield: __ pop_l(); break;
2831 case Bytecodes::_fast_fputfield: __ pop_f(); break;
2832 case Bytecodes::_fast_dputfield: __ pop_d(); break;
2833 default: break; // Nothin' to do.
2834 }
2835
2836 __ align(32, 12);
2837 __ bind(Lno_field_mod_post);
2838 }
2839 }
2840
2841 // PPC64: implement volatile stores as release-store (return bytecode contains an additional release).
2842 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2843 Label Lvolatile;
2844
2845 const Register Rcache = R5_ARG3, // Do not use ARG1/2 (causes trouble in jvmti_post_field_mod).
2846 Rclass_or_obj = R31, // Needs to survive C call.
2847 Roffset = R22_tmp2, // Needs to survive C call.
2848 Rflags = R3_ARG1,
2849 Rbtable = R4_ARG2,
2850 Rscratch = R11_scratch1,
2851 Rscratch2 = R12_scratch2,
2852 Rscratch3 = R6_ARG4,
2853 Rbc = Rscratch3;
2854 const ConditionRegister CR_is_vol = CCR2; // Non-volatile condition register (survives runtime call in do_oop_store).
2855
2856 static address field_rw_branch_table[number_of_states],
2857 field_norw_branch_table[number_of_states],
2858 static_branch_table[number_of_states];
2859
2860 address* branch_table = is_static ? static_branch_table :
2861 (rc == may_rewrite ? field_rw_branch_table : field_norw_branch_table);
2862
2863 // Stack (grows up):
2864 // value
2865 // obj
2866
2867 // Load the field offset.
2868 resolve_cache_and_index(byte_no, Rcache, Rscratch, sizeof(u2));
2869 jvmti_post_field_mod(Rcache, Rscratch, is_static);
2870 load_field_cp_cache_entry(Rclass_or_obj, Rcache, noreg, Roffset, Rflags, is_static);
2871
2872 // Load pointer to branch table.
2873 __ load_const_optimized(Rbtable, (address)branch_table, Rscratch);
2874
2875 // Get volatile flag.
2876 __ rldicl(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2877
2878 // Check the field type.
2879 __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2880
2881 #ifdef ASSERT
2882 Label LFlagInvalid;
2883 __ cmpldi(CCR0, Rflags, number_of_states);
2884 __ bge(CCR0, LFlagInvalid);
2885 #endif
2886
2887 // Load from branch table and dispatch (volatile case: one instruction ahead).
2888 __ sldi(Rflags, Rflags, LogBytesPerWord);
2889 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2890 __ cmpwi(CR_is_vol, Rscratch, 1); // Volatile?
2891 }
2892 __ sldi(Rscratch, Rscratch, exact_log2(BytesPerInstWord)); // Volatile? size of instruction 1 : 0.
2893 __ ldx(Rbtable, Rbtable, Rflags);
2894
2895 __ subf(Rbtable, Rscratch, Rbtable); // Point to volatile/non-volatile entry point.
2896 __ mtctr(Rbtable);
2897 __ bctr();
2898
2899 #ifdef ASSERT
2900 __ bind(LFlagInvalid);
2901 __ stop("got invalid flag", 0x656);
2902
2903 // __ bind(Lvtos);
2904 address pc_before_release = __ pc();
2905 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2906 assert(__ pc() - pc_before_release == (ptrdiff_t)BytesPerInstWord, "must be single instruction");
2907 assert(branch_table[vtos] == 0, "can't compute twice");
2908 branch_table[vtos] = __ pc(); // non-volatile_entry point
2909 __ stop("vtos unexpected", 0x657);
2910 #endif
2911
2912 __ align(32, 28, 28); // Align pop.
2913 // __ bind(Ldtos);
2914 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2915 assert(branch_table[dtos] == 0, "can't compute twice");
2916 branch_table[dtos] = __ pc(); // non-volatile_entry point
2917 __ pop(dtos);
2918 if (!is_static) {
2919 pop_and_check_object(Rclass_or_obj); // Kills R11_scratch1.
2920 }
2921 __ stfdx(F15_ftos, Rclass_or_obj, Roffset);
2922 if (!is_static && rc == may_rewrite) {
2923 patch_bytecode(Bytecodes::_fast_dputfield, Rbc, Rscratch, true, byte_no);
2924 }
2925 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2926 __ beq(CR_is_vol, Lvolatile); // Volatile?
2927 }
2928 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2929
2930 __ align(32, 28, 28); // Align pop.
2931 // __ bind(Lftos);
2932 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2933 assert(branch_table[ftos] == 0, "can't compute twice");
2934 branch_table[ftos] = __ pc(); // non-volatile_entry point
2935 __ pop(ftos);
2936 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2937 __ stfsx(F15_ftos, Rclass_or_obj, Roffset);
2938 if (!is_static && rc == may_rewrite) {
2939 patch_bytecode(Bytecodes::_fast_fputfield, Rbc, Rscratch, true, byte_no);
2940 }
2941 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2942 __ beq(CR_is_vol, Lvolatile); // Volatile?
2943 }
2944 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2945
2946 __ align(32, 28, 28); // Align pop.
2947 // __ bind(Litos);
2948 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2949 assert(branch_table[itos] == 0, "can't compute twice");
2950 branch_table[itos] = __ pc(); // non-volatile_entry point
2951 __ pop(itos);
2952 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2953 __ stwx(R17_tos, Rclass_or_obj, Roffset);
2954 if (!is_static && rc == may_rewrite) {
2955 patch_bytecode(Bytecodes::_fast_iputfield, Rbc, Rscratch, true, byte_no);
2956 }
2957 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2958 __ beq(CR_is_vol, Lvolatile); // Volatile?
2959 }
2960 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2961
2962 __ align(32, 28, 28); // Align pop.
2963 // __ bind(Lltos);
2964 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2965 assert(branch_table[ltos] == 0, "can't compute twice");
2966 branch_table[ltos] = __ pc(); // non-volatile_entry point
2967 __ pop(ltos);
2968 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2969 __ stdx(R17_tos, Rclass_or_obj, Roffset);
2970 if (!is_static && rc == may_rewrite) {
2971 patch_bytecode(Bytecodes::_fast_lputfield, Rbc, Rscratch, true, byte_no);
2972 }
2973 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2974 __ beq(CR_is_vol, Lvolatile); // Volatile?
2975 }
2976 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2977
2978 __ align(32, 28, 28); // Align pop.
2979 // __ bind(Lbtos);
2980 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2981 assert(branch_table[btos] == 0, "can't compute twice");
2982 branch_table[btos] = __ pc(); // non-volatile_entry point
2983 __ pop(btos);
2984 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2985 __ stbx(R17_tos, Rclass_or_obj, Roffset);
2986 if (!is_static && rc == may_rewrite) {
2987 patch_bytecode(Bytecodes::_fast_bputfield, Rbc, Rscratch, true, byte_no);
2988 }
2989 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2990 __ beq(CR_is_vol, Lvolatile); // Volatile?
2991 }
2992 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2993
2994 __ align(32, 28, 28); // Align pop.
2995 // __ bind(Lztos);
2996 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2997 assert(branch_table[ztos] == 0, "can't compute twice");
2998 branch_table[ztos] = __ pc(); // non-volatile_entry point
2999 __ pop(ztos);
3000 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
3001 __ andi(R17_tos, R17_tos, 0x1);
3002 __ stbx(R17_tos, Rclass_or_obj, Roffset);
3003 if (!is_static && rc == may_rewrite) {
3004 patch_bytecode(Bytecodes::_fast_zputfield, Rbc, Rscratch, true, byte_no);
3005 }
3006 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
3007 __ beq(CR_is_vol, Lvolatile); // Volatile?
3008 }
3009 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
3010
3011 __ align(32, 28, 28); // Align pop.
3012 // __ bind(Lctos);
3013 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
3014 assert(branch_table[ctos] == 0, "can't compute twice");
3015 branch_table[ctos] = __ pc(); // non-volatile_entry point
3016 __ pop(ctos);
3017 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1..
3018 __ sthx(R17_tos, Rclass_or_obj, Roffset);
3019 if (!is_static && rc == may_rewrite) {
3020 patch_bytecode(Bytecodes::_fast_cputfield, Rbc, Rscratch, true, byte_no);
3021 }
3022 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
3023 __ beq(CR_is_vol, Lvolatile); // Volatile?
3024 }
3025 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
3026
3027 __ align(32, 28, 28); // Align pop.
3028 // __ bind(Lstos);
3029 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
3030 assert(branch_table[stos] == 0, "can't compute twice");
3031 branch_table[stos] = __ pc(); // non-volatile_entry point
3032 __ pop(stos);
3033 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
3034 __ sthx(R17_tos, Rclass_or_obj, Roffset);
3035 if (!is_static && rc == may_rewrite) {
3036 patch_bytecode(Bytecodes::_fast_sputfield, Rbc, Rscratch, true, byte_no);
3037 }
3038 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
3039 __ beq(CR_is_vol, Lvolatile); // Volatile?
3040 }
3041 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
3042
3043 __ align(32, 28, 28); // Align pop.
3044 // __ bind(Latos);
3045 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
3046 assert(branch_table[atos] == 0, "can't compute twice");
3047 branch_table[atos] = __ pc(); // non-volatile_entry point
3048 __ pop(atos);
3049 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // kills R11_scratch1
3050 do_oop_store(_masm, Rclass_or_obj, Roffset, R17_tos, Rscratch, Rscratch2, Rscratch3, _bs->kind(), false /* precise */, true /* check null */);
3051 if (!is_static && rc == may_rewrite) {
3052 patch_bytecode(Bytecodes::_fast_aputfield, Rbc, Rscratch, true, byte_no);
3053 }
3054 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
3055 __ beq(CR_is_vol, Lvolatile); // Volatile?
3056 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
3057
3058 __ align(32, 12);
3059 __ bind(Lvolatile);
3060 __ fence();
3061 }
3062 // fallthru: __ b(Lexit);
3063
3064 #ifdef ASSERT
3065 for (int i = 0; i<number_of_states; ++i) {
3066 assert(branch_table[i], "put initialization");
3067 //tty->print_cr("put: %s_branch_table[%d] = 0x%llx (opcode 0x%llx)",
3068 // is_static ? "static" : "field", i, branch_table[i], *((unsigned int*)branch_table[i]));
3069 }
3070 #endif
3071 }
3072
3073 void TemplateTable::putfield(int byte_no) {
3074 putfield_or_static(byte_no, false);
3075 }
3076
3077 void TemplateTable::nofast_putfield(int byte_no) {
3078 putfield_or_static(byte_no, false, may_not_rewrite);
3079 }
3080
3081 void TemplateTable::putstatic(int byte_no) {
3082 putfield_or_static(byte_no, true);
3083 }
3084
3085 // See SPARC. On PPC64, we have a different jvmti_post_field_mod which does the job.
3086 void TemplateTable::jvmti_post_fast_field_mod() {
3087 __ should_not_reach_here();
3088 }
3089
3090 void TemplateTable::fast_storefield(TosState state) {
3091 transition(state, vtos);
3092
3093 const Register Rcache = R5_ARG3, // Do not use ARG1/2 (causes trouble in jvmti_post_field_mod).
3094 Rclass_or_obj = R31, // Needs to survive C call.
3095 Roffset = R22_tmp2, // Needs to survive C call.
3096 Rflags = R3_ARG1,
3097 Rscratch = R11_scratch1,
3098 Rscratch2 = R12_scratch2,
3099 Rscratch3 = R4_ARG2;
3100 const ConditionRegister CR_is_vol = CCR2; // Non-volatile condition register (survives runtime call in do_oop_store).
3101
3102 // Constant pool already resolved => Load flags and offset of field.
3103 __ get_cache_and_index_at_bcp(Rcache, 1);
3104 jvmti_post_field_mod(Rcache, Rscratch, false /* not static */);
3105 load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
3106
3107 // Get the obj and the final store addr.
3108 pop_and_check_object(Rclass_or_obj); // Kills R11_scratch1.
3109
3110 // Get volatile flag.
3111 __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
3112 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) { __ cmpdi(CR_is_vol, Rscratch, 1); }
3113 {
3114 Label LnotVolatile;
3115 __ beq(CCR0, LnotVolatile);
3116 __ release();
3117 __ align(32, 12);
3118 __ bind(LnotVolatile);
3119 }
3120
3121 // Do the store and fencing.
3122 switch(bytecode()) {
3123 case Bytecodes::_fast_aputfield:
3124 // Store into the field.
3125 do_oop_store(_masm, Rclass_or_obj, Roffset, R17_tos, Rscratch, Rscratch2, Rscratch3, _bs->kind(), false /* precise */, true /* check null */);
3126 break;
3127
3128 case Bytecodes::_fast_iputfield:
3129 __ stwx(R17_tos, Rclass_or_obj, Roffset);
3130 break;
3131
3132 case Bytecodes::_fast_lputfield:
3133 __ stdx(R17_tos, Rclass_or_obj, Roffset);
3134 break;
3135
3136 case Bytecodes::_fast_zputfield:
3137 __ andi(R17_tos, R17_tos, 0x1); // boolean is true if LSB is 1
3138 // fall through to bputfield
3139 case Bytecodes::_fast_bputfield:
3140 __ stbx(R17_tos, Rclass_or_obj, Roffset);
3141 break;
3142
3143 case Bytecodes::_fast_cputfield:
3144 case Bytecodes::_fast_sputfield:
3145 __ sthx(R17_tos, Rclass_or_obj, Roffset);
3146 break;
3147
3148 case Bytecodes::_fast_fputfield:
3149 __ stfsx(F15_ftos, Rclass_or_obj, Roffset);
3150 break;
3151
3152 case Bytecodes::_fast_dputfield:
3153 __ stfdx(F15_ftos, Rclass_or_obj, Roffset);
3154 break;
3155
3156 default: ShouldNotReachHere();
3157 }
3158
3159 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
3160 Label LVolatile;
3161 __ beq(CR_is_vol, LVolatile);
3162 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
3163
3164 __ align(32, 12);
3165 __ bind(LVolatile);
3166 __ fence();
3167 }
3168 }
3169
3170 void TemplateTable::fast_accessfield(TosState state) {
3171 transition(atos, state);
3172
3173 Label LisVolatile;
3174 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
3175
3176 const Register Rcache = R3_ARG1,
3177 Rclass_or_obj = R17_tos,
3178 Roffset = R22_tmp2,
3179 Rflags = R23_tmp3,
3180 Rscratch = R12_scratch2;
3181
3182 // Constant pool already resolved. Get the field offset.
3183 __ get_cache_and_index_at_bcp(Rcache, 1);
3184 load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
3185
3186 // JVMTI support
3187 jvmti_post_field_access(Rcache, Rscratch, false, true);
3188
3189 // Get the load address.
3190 __ null_check_throw(Rclass_or_obj, -1, Rscratch);
3191
3192 // Get volatile flag.
3193 __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
3194 __ bne(CCR0, LisVolatile);
3195
3196 switch(bytecode()) {
3197 case Bytecodes::_fast_agetfield:
3198 {
3199 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3200 __ verify_oop(R17_tos);
3201 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3202
3203 __ bind(LisVolatile);
3204 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3205 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3206 __ verify_oop(R17_tos);
3207 __ twi_0(R17_tos);
3208 __ isync();
3209 break;
3210 }
3211 case Bytecodes::_fast_igetfield:
3212 {
3213 __ lwax(R17_tos, Rclass_or_obj, Roffset);
3214 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3215
3216 __ bind(LisVolatile);
3217 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3218 __ lwax(R17_tos, Rclass_or_obj, Roffset);
3219 __ twi_0(R17_tos);
3220 __ isync();
3221 break;
3222 }
3223 case Bytecodes::_fast_lgetfield:
3224 {
3225 __ ldx(R17_tos, Rclass_or_obj, Roffset);
3226 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3227
3228 __ bind(LisVolatile);
3229 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3230 __ ldx(R17_tos, Rclass_or_obj, Roffset);
3231 __ twi_0(R17_tos);
3232 __ isync();
3233 break;
3234 }
3235 case Bytecodes::_fast_bgetfield:
3236 {
3237 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
3238 __ extsb(R17_tos, R17_tos);
3239 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3240
3241 __ bind(LisVolatile);
3242 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3243 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
3244 __ twi_0(R17_tos);
3245 __ extsb(R17_tos, R17_tos);
3246 __ isync();
3247 break;
3248 }
3249 case Bytecodes::_fast_cgetfield:
3250 {
3251 __ lhzx(R17_tos, Rclass_or_obj, Roffset);
3252 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3253
3254 __ bind(LisVolatile);
3255 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3256 __ lhzx(R17_tos, Rclass_or_obj, Roffset);
3257 __ twi_0(R17_tos);
3258 __ isync();
3259 break;
3260 }
3261 case Bytecodes::_fast_sgetfield:
3262 {
3263 __ lhax(R17_tos, Rclass_or_obj, Roffset);
3264 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3265
3266 __ bind(LisVolatile);
3267 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3268 __ lhax(R17_tos, Rclass_or_obj, Roffset);
3269 __ twi_0(R17_tos);
3270 __ isync();
3271 break;
3272 }
3273 case Bytecodes::_fast_fgetfield:
3274 {
3275 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3276 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3277
3278 __ bind(LisVolatile);
3279 Label Ldummy;
3280 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3281 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3282 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3283 __ bne_predict_not_taken(CCR0, Ldummy);
3284 __ bind(Ldummy);
3285 __ isync();
3286 break;
3287 }
3288 case Bytecodes::_fast_dgetfield:
3289 {
3290 __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
3291 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3292
3293 __ bind(LisVolatile);
3294 Label Ldummy;
3295 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3296 __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
3297 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3298 __ bne_predict_not_taken(CCR0, Ldummy);
3299 __ bind(Ldummy);
3300 __ isync();
3301 break;
3302 }
3303 default: ShouldNotReachHere();
3304 }
3305 }
3306
3307 void TemplateTable::fast_xaccess(TosState state) {
3308 transition(vtos, state);
3309
3310 Label LisVolatile;
3311 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
3312 const Register Rcache = R3_ARG1,
3313 Rclass_or_obj = R17_tos,
3314 Roffset = R22_tmp2,
3315 Rflags = R23_tmp3,
3316 Rscratch = R12_scratch2;
3317
3318 __ ld(Rclass_or_obj, 0, R18_locals);
3319
3320 // Constant pool already resolved. Get the field offset.
3321 __ get_cache_and_index_at_bcp(Rcache, 2);
3322 load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
3323
3324 // JVMTI support not needed, since we switch back to single bytecode as soon as debugger attaches.
3325
3326 // Needed to report exception at the correct bcp.
3327 __ addi(R14_bcp, R14_bcp, 1);
3328
3329 // Get the load address.
3330 __ null_check_throw(Rclass_or_obj, -1, Rscratch);
3331
3332 // Get volatile flag.
3333 __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
3334 __ bne(CCR0, LisVolatile);
3335
3336 switch(state) {
3337 case atos:
3338 {
3339 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3340 __ verify_oop(R17_tos);
3341 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3342
3343 __ bind(LisVolatile);
3344 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3345 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3346 __ verify_oop(R17_tos);
3347 __ twi_0(R17_tos);
3348 __ isync();
3349 break;
3350 }
3351 case itos:
3352 {
3353 __ lwax(R17_tos, Rclass_or_obj, Roffset);
3354 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3355
3356 __ bind(LisVolatile);
3357 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3358 __ lwax(R17_tos, Rclass_or_obj, Roffset);
3359 __ twi_0(R17_tos);
3360 __ isync();
3361 break;
3362 }
3363 case ftos:
3364 {
3365 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3366 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3367
3368 __ bind(LisVolatile);
3369 Label Ldummy;
3370 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3371 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3372 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3373 __ bne_predict_not_taken(CCR0, Ldummy);
3374 __ bind(Ldummy);
3375 __ isync();
3376 break;
3377 }
3378 default: ShouldNotReachHere();
3379 }
3380 __ addi(R14_bcp, R14_bcp, -1);
3381 }
3382
3383 // ============================================================================
3384 // Calls
3385
3386 // Common code for invoke
3387 //
3388 // Input:
3389 // - byte_no
3390 //
3391 // Output:
3392 // - Rmethod: The method to invoke next.
3393 // - Rret_addr: The return address to return to.
3394 // - Rindex: MethodType (invokehandle) or CallSite obj (invokedynamic)
3395 // - Rrecv: Cache for "this" pointer, might be noreg if static call.
3396 // - Rflags: Method flags from const pool cache.
3397 //
3398 // Kills:
3399 // - Rscratch1
3400 //
3401 void TemplateTable::prepare_invoke(int byte_no,
3402 Register Rmethod, // linked method (or i-klass)
3403 Register Rret_addr,// return address
3404 Register Rindex, // itable index, MethodType, etc.
3405 Register Rrecv, // If caller wants to see it.
3406 Register Rflags, // If caller wants to test it.
3407 Register Rscratch
3408 ) {
3409 // Determine flags.
3410 const Bytecodes::Code code = bytecode();
3411 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3412 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3413 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3414 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3415 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3416 const bool load_receiver = (Rrecv != noreg);
3417 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3418
3419 assert_different_registers(Rmethod, Rindex, Rflags, Rscratch);
3420 assert_different_registers(Rmethod, Rrecv, Rflags, Rscratch);
3421 assert_different_registers(Rret_addr, Rscratch);
3422
3423 load_invoke_cp_cache_entry(byte_no, Rmethod, Rindex, Rflags, is_invokevirtual, false, is_invokedynamic);
3424
3425 // Saving of SP done in call_from_interpreter.
3426
3427 // Maybe push "appendix" to arguments.
3428 if (is_invokedynamic || is_invokehandle) {
3429 Label Ldone;
3430 __ rldicl_(R0, Rflags, 64-ConstantPoolCacheEntry::has_appendix_shift, 63);
3431 __ beq(CCR0, Ldone);
3432 // Push "appendix" (MethodType, CallSite, etc.).
3433 // This must be done before we get the receiver,
3434 // since the parameter_size includes it.
3435 __ load_resolved_reference_at_index(Rscratch, Rindex);
3436 __ verify_oop(Rscratch);
3437 __ push_ptr(Rscratch);
3438 __ bind(Ldone);
3439 }
3440
3441 // Load receiver if needed (after appendix is pushed so parameter size is correct).
3442 if (load_receiver) {
3443 const Register Rparam_count = Rscratch;
3444 __ andi(Rparam_count, Rflags, ConstantPoolCacheEntry::parameter_size_mask);
3445 __ load_receiver(Rparam_count, Rrecv);
3446 __ verify_oop(Rrecv);
3447 }
3448
3449 // Get return address.
3450 {
3451 Register Rtable_addr = Rscratch;
3452 Register Rret_type = Rret_addr;
3453 address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3454
3455 // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3456 __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3457 __ load_dispatch_table(Rtable_addr, (address*)table_addr);
3458 __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3459 // Get return address.
3460 __ ldx(Rret_addr, Rtable_addr, Rret_type);
3461 }
3462 }
3463
3464 // Helper for virtual calls. Load target out of vtable and jump off!
3465 // Kills all passed registers.
3466 void TemplateTable::generate_vtable_call(Register Rrecv_klass, Register Rindex, Register Rret, Register Rtemp) {
3467
3468 assert_different_registers(Rrecv_klass, Rtemp, Rret);
3469 const Register Rtarget_method = Rindex;
3470
3471 // Get target method & entry point.
3472 const int base = in_bytes(Klass::vtable_start_offset());
3473 // Calc vtable addr scale the vtable index by 8.
3474 __ sldi(Rindex, Rindex, exact_log2(vtableEntry::size_in_bytes()));
3475 // Load target.
3476 __ addi(Rrecv_klass, Rrecv_klass, base + vtableEntry::method_offset_in_bytes());
3477 __ ldx(Rtarget_method, Rindex, Rrecv_klass);
3478 // Argument and return type profiling.
3479 __ profile_arguments_type(Rtarget_method, Rrecv_klass /* scratch1 */, Rtemp /* scratch2 */, true);
3480 __ call_from_interpreter(Rtarget_method, Rret, Rrecv_klass /* scratch1 */, Rtemp /* scratch2 */);
3481 }
3482
3483 // Virtual or final call. Final calls are rewritten on the fly to run through "fast_finalcall" next time.
3484 void TemplateTable::invokevirtual(int byte_no) {
3485 transition(vtos, vtos);
3486
3487 Register Rtable_addr = R11_scratch1,
3488 Rret_type = R12_scratch2,
3489 Rret_addr = R5_ARG3,
3490 Rflags = R22_tmp2, // Should survive C call.
3491 Rrecv = R3_ARG1,
3492 Rrecv_klass = Rrecv,
3493 Rvtableindex_or_method = R31, // Should survive C call.
3494 Rnum_params = R4_ARG2,
3495 Rnew_bc = R6_ARG4;
3496
3497 Label LnotFinal;
3498
3499 load_invoke_cp_cache_entry(byte_no, Rvtableindex_or_method, noreg, Rflags, /*virtual*/ true, false, false);
3500
3501 __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_vfinal_shift);
3502 __ bfalse(CCR0, LnotFinal);
3503
3504 if (RewriteBytecodes && !UseSharedSpaces && !DumpSharedSpaces) {
3505 patch_bytecode(Bytecodes::_fast_invokevfinal, Rnew_bc, R12_scratch2);
3506 }
3507 invokevfinal_helper(Rvtableindex_or_method, Rflags, R11_scratch1, R12_scratch2);
3508
3509 __ align(32, 12);
3510 __ bind(LnotFinal);
3511 // Load "this" pointer (receiver).
3512 __ rldicl(Rnum_params, Rflags, 64, 48);
3513 __ load_receiver(Rnum_params, Rrecv);
3514 __ verify_oop(Rrecv);
3515
3516 // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3517 __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3518 __ load_dispatch_table(Rtable_addr, Interpreter::invoke_return_entry_table());
3519 __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3520 __ ldx(Rret_addr, Rret_type, Rtable_addr);
3521 __ null_check_throw(Rrecv, oopDesc::klass_offset_in_bytes(), R11_scratch1);
3522 __ load_klass(Rrecv_klass, Rrecv);
3523 __ verify_klass_ptr(Rrecv_klass);
3524 __ profile_virtual_call(Rrecv_klass, R11_scratch1, R12_scratch2, false);
3525
3526 generate_vtable_call(Rrecv_klass, Rvtableindex_or_method, Rret_addr, R11_scratch1);
3527 }
3528
3529 void TemplateTable::fast_invokevfinal(int byte_no) {
3530 transition(vtos, vtos);
3531
3532 assert(byte_no == f2_byte, "use this argument");
3533 Register Rflags = R22_tmp2,
3534 Rmethod = R31;
3535 load_invoke_cp_cache_entry(byte_no, Rmethod, noreg, Rflags, /*virtual*/ true, /*is_invokevfinal*/ true, false);
3536 invokevfinal_helper(Rmethod, Rflags, R11_scratch1, R12_scratch2);
3537 }
3538
3539 void TemplateTable::invokevfinal_helper(Register Rmethod, Register Rflags, Register Rscratch1, Register Rscratch2) {
3540
3541 assert_different_registers(Rmethod, Rflags, Rscratch1, Rscratch2);
3542
3543 // Load receiver from stack slot.
3544 Register Rrecv = Rscratch2;
3545 Register Rnum_params = Rrecv;
3546
3547 __ ld(Rnum_params, in_bytes(Method::const_offset()), Rmethod);
3548 __ lhz(Rnum_params /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), Rnum_params);
3549
3550 // Get return address.
3551 Register Rtable_addr = Rscratch1,
3552 Rret_addr = Rflags,
3553 Rret_type = Rret_addr;
3554 // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3555 __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3556 __ load_dispatch_table(Rtable_addr, Interpreter::invoke_return_entry_table());
3557 __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3558 __ ldx(Rret_addr, Rret_type, Rtable_addr);
3559
3560 // Load receiver and receiver NULL check.
3561 __ load_receiver(Rnum_params, Rrecv);
3562 __ null_check_throw(Rrecv, -1, Rscratch1);
3563
3564 __ profile_final_call(Rrecv, Rscratch1);
3565 // Argument and return type profiling.
3566 __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, true);
3567
3568 // Do the call.
3569 __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1, Rscratch2);
3570 }
3571
3572 void TemplateTable::invokespecial(int byte_no) {
3573 assert(byte_no == f1_byte, "use this argument");
3574 transition(vtos, vtos);
3575
3576 Register Rtable_addr = R3_ARG1,
3577 Rret_addr = R4_ARG2,
3578 Rflags = R5_ARG3,
3579 Rreceiver = R6_ARG4,
3580 Rmethod = R31;
3581
3582 prepare_invoke(byte_no, Rmethod, Rret_addr, noreg, Rreceiver, Rflags, R11_scratch1);
3583
3584 // Receiver NULL check.
3585 __ null_check_throw(Rreceiver, -1, R11_scratch1);
3586
3587 __ profile_call(R11_scratch1, R12_scratch2);
3588 // Argument and return type profiling.
3589 __ profile_arguments_type(Rmethod, R11_scratch1, R12_scratch2, false);
3590 __ call_from_interpreter(Rmethod, Rret_addr, R11_scratch1, R12_scratch2);
3591 }
3592
3593 void TemplateTable::invokestatic(int byte_no) {
3594 assert(byte_no == f1_byte, "use this argument");
3595 transition(vtos, vtos);
3596
3597 Register Rtable_addr = R3_ARG1,
3598 Rret_addr = R4_ARG2,
3599 Rflags = R5_ARG3;
3600
3601 prepare_invoke(byte_no, R19_method, Rret_addr, noreg, noreg, Rflags, R11_scratch1);
3602
3603 __ profile_call(R11_scratch1, R12_scratch2);
3604 // Argument and return type profiling.
3605 __ profile_arguments_type(R19_method, R11_scratch1, R12_scratch2, false);
3606 __ call_from_interpreter(R19_method, Rret_addr, R11_scratch1, R12_scratch2);
3607 }
3608
3609 void TemplateTable::invokeinterface_object_method(Register Rrecv_klass,
3610 Register Rret,
3611 Register Rflags,
3612 Register Rmethod,
3613 Register Rtemp1,
3614 Register Rtemp2) {
3615
3616 assert_different_registers(Rmethod, Rret, Rrecv_klass, Rflags, Rtemp1, Rtemp2);
3617 Label LnotFinal;
3618
3619 // Check for vfinal.
3620 __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_vfinal_shift);
3621 __ bfalse(CCR0, LnotFinal);
3622
3623 Register Rscratch = Rflags; // Rflags is dead now.
3624
3625 // Final call case.
3626 __ profile_final_call(Rtemp1, Rscratch);
3627 // Argument and return type profiling.
3628 __ profile_arguments_type(Rmethod, Rscratch, Rrecv_klass /* scratch */, true);
3629 // Do the final call - the index (f2) contains the method.
3630 __ call_from_interpreter(Rmethod, Rret, Rscratch, Rrecv_klass /* scratch */);
3631
3632 // Non-final callc case.
3633 __ bind(LnotFinal);
3634 __ profile_virtual_call(Rrecv_klass, Rtemp1, Rscratch, false);
3635 generate_vtable_call(Rrecv_klass, Rmethod, Rret, Rscratch);
3636 }
3637
3638 void TemplateTable::invokeinterface(int byte_no) {
3639 assert(byte_no == f1_byte, "use this argument");
3640 transition(vtos, vtos);
3641
3642 const Register Rscratch1 = R11_scratch1,
3643 Rscratch2 = R12_scratch2,
3644 Rmethod = R6_ARG4,
3645 Rmethod2 = R9_ARG7,
3646 Rinterface_klass = R5_ARG3,
3647 Rret_addr = R8_ARG6,
3648 Rindex = R10_ARG8,
3649 Rreceiver = R3_ARG1,
3650 Rrecv_klass = R4_ARG2,
3651 Rflags = R7_ARG5;
3652
3653 prepare_invoke(byte_no, Rinterface_klass, Rret_addr, Rmethod, Rreceiver, Rflags, Rscratch1);
3654
3655 // Get receiver klass.
3656 __ null_check_throw(Rreceiver, oopDesc::klass_offset_in_bytes(), Rscratch2);
3657 __ load_klass(Rrecv_klass, Rreceiver);
3658
3659 // Check corner case object method.
3660 Label LobjectMethod, L_no_such_interface, Lthrow_ame;
3661 __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_forced_virtual_shift);
3662 __ btrue(CCR0, LobjectMethod);
3663
3664 __ lookup_interface_method(Rrecv_klass, Rinterface_klass, noreg, noreg, Rscratch1, Rscratch2,
3665 L_no_such_interface, /*return_method=*/false);
3666
3667 __ profile_virtual_call(Rrecv_klass, Rscratch1, Rscratch2, false);
3668
3669 // Find entry point to call.
3670
3671 // Get declaring interface class from method
3672 __ ld(Rinterface_klass, in_bytes(Method::const_offset()), Rmethod);
3673 __ ld(Rinterface_klass, in_bytes(ConstMethod::constants_offset()), Rinterface_klass);
3674 __ ld(Rinterface_klass, ConstantPool::pool_holder_offset_in_bytes(), Rinterface_klass);
3675
3676 // Get itable index from method
3677 __ lwa(Rindex, in_bytes(Method::itable_index_offset()), Rmethod);
3678 __ subfic(Rindex, Rindex, Method::itable_index_max);
3679
3680 __ lookup_interface_method(Rrecv_klass, Rinterface_klass, Rindex, Rmethod2, Rscratch1, Rscratch2,
3681 L_no_such_interface);
3682
3683 __ cmpdi(CCR0, Rmethod2, 0);
3684 __ beq(CCR0, Lthrow_ame);
3685 // Found entry. Jump off!
3686 // Argument and return type profiling.
3687 __ profile_arguments_type(Rmethod2, Rscratch1, Rscratch2, true);
3688 //__ profile_called_method(Rindex, Rscratch1);
3689 __ call_from_interpreter(Rmethod2, Rret_addr, Rscratch1, Rscratch2);
3690
3691 // Vtable entry was NULL => Throw abstract method error.
3692 __ bind(Lthrow_ame);
3693 // Pass arguments for generating a verbose error message.
3694 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorVerbose),
3695 Rrecv_klass, Rmethod);
3696
3697 // Interface was not found => Throw incompatible class change error.
3698 __ bind(L_no_such_interface);
3699 // Pass arguments for generating a verbose error message.
3700 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose),
3701 Rrecv_klass, Rinterface_klass);
3702 DEBUG_ONLY( __ should_not_reach_here(); )
3703
3704 // Special case of invokeinterface called for virtual method of
3705 // java.lang.Object. See ConstantPoolCacheEntry::set_method() for details:
3706 // The invokeinterface was rewritten to a invokevirtual, hence we have
3707 // to handle this corner case. This code isn't produced by javac, but could
3708 // be produced by another compliant java compiler.
3709 __ bind(LobjectMethod);
3710 invokeinterface_object_method(Rrecv_klass, Rret_addr, Rflags, Rmethod, Rscratch1, Rscratch2);
3711 }
3712
3713 void TemplateTable::invokedynamic(int byte_no) {
3714 transition(vtos, vtos);
3715
3716 const Register Rret_addr = R3_ARG1,
3717 Rflags = R4_ARG2,
3718 Rmethod = R22_tmp2,
3719 Rscratch1 = R11_scratch1,
3720 Rscratch2 = R12_scratch2;
3721
3722 prepare_invoke(byte_no, Rmethod, Rret_addr, Rscratch1, noreg, Rflags, Rscratch2);
3723
3724 // Profile this call.
3725 __ profile_call(Rscratch1, Rscratch2);
3726
3727 // Off we go. With the new method handles, we don't jump to a method handle
3728 // entry any more. Instead, we pushed an "appendix" in prepare invoke, which happens
3729 // to be the callsite object the bootstrap method returned. This is passed to a
3730 // "link" method which does the dispatch (Most likely just grabs the MH stored
3731 // inside the callsite and does an invokehandle).
3732 // Argument and return type profiling.
3733 __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, false);
3734 __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1 /* scratch1 */, Rscratch2 /* scratch2 */);
3735 }
3736
3737 void TemplateTable::invokehandle(int byte_no) {
3738 transition(vtos, vtos);
3739
3740 const Register Rret_addr = R3_ARG1,
3741 Rflags = R4_ARG2,
3742 Rrecv = R5_ARG3,
3743 Rmethod = R22_tmp2,
3744 Rscratch1 = R11_scratch1,
3745 Rscratch2 = R12_scratch2;
3746
3747 prepare_invoke(byte_no, Rmethod, Rret_addr, Rscratch1, Rrecv, Rflags, Rscratch2);
3748 __ verify_method_ptr(Rmethod);
3749 __ null_check_throw(Rrecv, -1, Rscratch2);
3750
3751 __ profile_final_call(Rrecv, Rscratch1);
3752
3753 // Still no call from handle => We call the method handle interpreter here.
3754 // Argument and return type profiling.
3755 __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, true);
3756 __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1 /* scratch1 */, Rscratch2 /* scratch2 */);
3757 }
3758
3759 // =============================================================================
3760 // Allocation
3761
3762 // Puts allocated obj ref onto the expression stack.
3763 void TemplateTable::_new() {
3764 transition(vtos, atos);
3765
3766 Label Lslow_case,
3767 Ldone;
3768
3769 const Register RallocatedObject = R17_tos,
3770 RinstanceKlass = R9_ARG7,
3771 Rscratch = R11_scratch1,
3772 Roffset = R8_ARG6,
3773 Rinstance_size = Roffset,
3774 Rcpool = R4_ARG2,
3775 Rtags = R3_ARG1,
3776 Rindex = R5_ARG3;
3777
3778 // --------------------------------------------------------------------------
3779 // Check if fast case is possible.
3780
3781 // Load pointers to const pool and const pool's tags array.
3782 __ get_cpool_and_tags(Rcpool, Rtags);
3783 // Load index of constant pool entry.
3784 __ get_2_byte_integer_at_bcp(1, Rindex, InterpreterMacroAssembler::Unsigned);
3785
3786 // Note: compared to other architectures, PPC's implementation always goes
3787 // to the slow path if TLAB is used and fails.
3788 if (UseTLAB) {
3789 // Make sure the class we're about to instantiate has been resolved
3790 // This is done before loading instanceKlass to be consistent with the order
3791 // how Constant Pool is updated (see ConstantPoolCache::klass_at_put).
3792 __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3793 __ lbzx(Rtags, Rindex, Rtags);
3794
3795 __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3796 __ bne(CCR0, Lslow_case);
3797
3798 // Get instanceKlass
3799 __ sldi(Roffset, Rindex, LogBytesPerWord);
3800 __ load_resolved_klass_at_offset(Rcpool, Roffset, RinstanceKlass);
3801
3802 // Make sure klass is fully initialized and get instance_size.
3803 __ lbz(Rscratch, in_bytes(InstanceKlass::init_state_offset()), RinstanceKlass);
3804 __ lwz(Rinstance_size, in_bytes(Klass::layout_helper_offset()), RinstanceKlass);
3805
3806 __ cmpdi(CCR1, Rscratch, InstanceKlass::fully_initialized);
3807 // Make sure klass does not have has_finalizer, or is abstract, or interface or java/lang/Class.
3808 __ andi_(R0, Rinstance_size, Klass::_lh_instance_slow_path_bit); // slow path bit equals 0?
3809
3810 __ crnand(CCR0, Assembler::equal, CCR1, Assembler::equal); // slow path bit set or not fully initialized?
3811 __ beq(CCR0, Lslow_case);
3812
3813 // --------------------------------------------------------------------------
3814 // Fast case:
3815 // Allocate the instance.
3816 // 1) Try to allocate in the TLAB.
3817 // 2) If the above fails (or is not applicable), go to a slow case (creates a new TLAB, etc.).
3818
3819 Register RoldTopValue = RallocatedObject; // Object will be allocated here if it fits.
3820 Register RnewTopValue = R6_ARG4;
3821 Register RendValue = R7_ARG5;
3822
3823 // Check if we can allocate in the TLAB.
3824 __ ld(RoldTopValue, in_bytes(JavaThread::tlab_top_offset()), R16_thread);
3825 __ ld(RendValue, in_bytes(JavaThread::tlab_current_end_offset()), R16_thread);
3826
3827 __ add(RnewTopValue, Rinstance_size, RoldTopValue);
3828
3829 // If there is enough space, we do not CAS and do not clear.
3830 __ cmpld(CCR0, RnewTopValue, RendValue);
3831 __ bgt(CCR0, Lslow_case);
3832
3833 __ std(RnewTopValue, in_bytes(JavaThread::tlab_top_offset()), R16_thread);
3834
3835 if (!ZeroTLAB) {
3836 // --------------------------------------------------------------------------
3837 // Init1: Zero out newly allocated memory.
3838 // Initialize remaining object fields.
3839 Register Rbase = Rtags;
3840 __ addi(Rinstance_size, Rinstance_size, 7 - (int)sizeof(oopDesc));
3841 __ addi(Rbase, RallocatedObject, sizeof(oopDesc));
3842 __ srdi(Rinstance_size, Rinstance_size, 3);
3843
3844 // Clear out object skipping header. Takes also care of the zero length case.
3845 __ clear_memory_doubleword(Rbase, Rinstance_size);
3846 }
3847
3848 // --------------------------------------------------------------------------
3849 // Init2: Initialize the header: mark, klass
3850 // Init mark.
3851 if (UseBiasedLocking) {
3852 __ ld(Rscratch, in_bytes(Klass::prototype_header_offset()), RinstanceKlass);
3853 } else {
3854 __ load_const_optimized(Rscratch, markOopDesc::prototype(), R0);
3855 }
3856 __ std(Rscratch, oopDesc::mark_offset_in_bytes(), RallocatedObject);
3857
3858 // Init klass.
3859 __ store_klass_gap(RallocatedObject);
3860 __ store_klass(RallocatedObject, RinstanceKlass, Rscratch); // klass (last for cms)
3861
3862 // Check and trigger dtrace event.
3863 SkipIfEqualZero::skip_to_label_if_equal_zero(_masm, Rscratch, &DTraceAllocProbes, Ldone);
3864 __ push(atos);
3865 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc));
3866 __ pop(atos);
3867
3868 __ b(Ldone);
3869 }
3870
3871 // --------------------------------------------------------------------------
3872 // slow case
3873 __ bind(Lslow_case);
3874 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), Rcpool, Rindex);
3875
3876 // continue
3877 __ bind(Ldone);
3878
3879 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3880 __ membar(Assembler::StoreStore);
3881 }
3882
3883 void TemplateTable::newarray() {
3884 transition(itos, atos);
3885
3886 __ lbz(R4, 1, R14_bcp);
3887 __ extsw(R5, R17_tos);
3888 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), R4, R5 /* size */);
3889
3890 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3891 __ membar(Assembler::StoreStore);
3892 }
3893
3894 void TemplateTable::anewarray() {
3895 transition(itos, atos);
3896
3897 __ get_constant_pool(R4);
3898 __ get_2_byte_integer_at_bcp(1, R5, InterpreterMacroAssembler::Unsigned);
3899 __ extsw(R6, R17_tos); // size
3900 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), R4 /* pool */, R5 /* index */, R6 /* size */);
3901
3902 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3903 __ membar(Assembler::StoreStore);
3904 }
3905
3906 // Allocate a multi dimensional array
3907 void TemplateTable::multianewarray() {
3908 transition(vtos, atos);
3909
3910 Register Rptr = R31; // Needs to survive C call.
3911
3912 // Put ndims * wordSize into frame temp slot
3913 __ lbz(Rptr, 3, R14_bcp);
3914 __ sldi(Rptr, Rptr, Interpreter::logStackElementSize);
3915 // Esp points past last_dim, so set to R4 to first_dim address.
3916 __ add(R4, Rptr, R15_esp);
3917 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), R4 /* first_size_address */);
3918 // Pop all dimensions off the stack.
3919 __ add(R15_esp, Rptr, R15_esp);
3920
3921 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3922 __ membar(Assembler::StoreStore);
3923 }
3924
3925 void TemplateTable::arraylength() {
3926 transition(atos, itos);
3927
3928 Label LnoException;
3929 __ verify_oop(R17_tos);
3930 __ null_check_throw(R17_tos, arrayOopDesc::length_offset_in_bytes(), R11_scratch1);
3931 __ lwa(R17_tos, arrayOopDesc::length_offset_in_bytes(), R17_tos);
3932 }
3933
3934 // ============================================================================
3935 // Typechecks
3936
3937 void TemplateTable::checkcast() {
3938 transition(atos, atos);
3939
3940 Label Ldone, Lis_null, Lquicked, Lresolved;
3941 Register Roffset = R6_ARG4,
3942 RobjKlass = R4_ARG2,
3943 RspecifiedKlass = R5_ARG3, // Generate_ClassCastException_verbose_handler will read value from this register.
3944 Rcpool = R11_scratch1,
3945 Rtags = R12_scratch2;
3946
3947 // Null does not pass.
3948 __ cmpdi(CCR0, R17_tos, 0);
3949 __ beq(CCR0, Lis_null);
3950
3951 // Get constant pool tag to find out if the bytecode has already been "quickened".
3952 __ get_cpool_and_tags(Rcpool, Rtags);
3953
3954 __ get_2_byte_integer_at_bcp(1, Roffset, InterpreterMacroAssembler::Unsigned);
3955
3956 __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3957 __ lbzx(Rtags, Rtags, Roffset);
3958
3959 __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3960 __ beq(CCR0, Lquicked);
3961
3962 // Call into the VM to "quicken" instanceof.
3963 __ push_ptr(); // for GC
3964 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3965 __ get_vm_result_2(RspecifiedKlass);
3966 __ pop_ptr(); // Restore receiver.
3967 __ b(Lresolved);
3968
3969 // Extract target class from constant pool.
3970 __ bind(Lquicked);
3971 __ sldi(Roffset, Roffset, LogBytesPerWord);
3972 __ load_resolved_klass_at_offset(Rcpool, Roffset, RspecifiedKlass);
3973
3974 // Do the checkcast.
3975 __ bind(Lresolved);
3976 // Get value klass in RobjKlass.
3977 __ load_klass(RobjKlass, R17_tos);
3978 // Generate a fast subtype check. Branch to cast_ok if no failure. Return 0 if failure.
3979 __ gen_subtype_check(RobjKlass, RspecifiedKlass, /*3 temp regs*/ Roffset, Rcpool, Rtags, /*target if subtype*/ Ldone);
3980
3981 // Not a subtype; so must throw exception
3982 // Target class oop is in register R6_ARG4 == RspecifiedKlass by convention.
3983 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ClassCastException_entry);
3984 __ mtctr(R11_scratch1);
3985 __ bctr();
3986
3987 // Profile the null case.
3988 __ align(32, 12);
3989 __ bind(Lis_null);
3990 __ profile_null_seen(R11_scratch1, Rtags); // Rtags used as scratch.
3991
3992 __ align(32, 12);
3993 __ bind(Ldone);
3994 }
3995
3996 // Output:
3997 // - tos == 0: Obj was null or not an instance of class.
3998 // - tos == 1: Obj was an instance of class.
3999 void TemplateTable::instanceof() {
4000 transition(atos, itos);
4001
4002 Label Ldone, Lis_null, Lquicked, Lresolved;
4003 Register Roffset = R6_ARG4,
4004 RobjKlass = R4_ARG2,
4005 RspecifiedKlass = R5_ARG3,
4006 Rcpool = R11_scratch1,
4007 Rtags = R12_scratch2;
4008
4009 // Null does not pass.
4010 __ cmpdi(CCR0, R17_tos, 0);
4011 __ beq(CCR0, Lis_null);
4012
4013 // Get constant pool tag to find out if the bytecode has already been "quickened".
4014 __ get_cpool_and_tags(Rcpool, Rtags);
4015
4016 __ get_2_byte_integer_at_bcp(1, Roffset, InterpreterMacroAssembler::Unsigned);
4017
4018 __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
4019 __ lbzx(Rtags, Rtags, Roffset);
4020
4021 __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
4022 __ beq(CCR0, Lquicked);
4023
4024 // Call into the VM to "quicken" instanceof.
4025 __ push_ptr(); // for GC
4026 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4027 __ get_vm_result_2(RspecifiedKlass);
4028 __ pop_ptr(); // Restore receiver.
4029 __ b(Lresolved);
4030
4031 // Extract target class from constant pool.
4032 __ bind(Lquicked);
4033 __ sldi(Roffset, Roffset, LogBytesPerWord);
4034 __ load_resolved_klass_at_offset(Rcpool, Roffset, RspecifiedKlass);
4035
4036 // Do the checkcast.
4037 __ bind(Lresolved);
4038 // Get value klass in RobjKlass.
4039 __ load_klass(RobjKlass, R17_tos);
4040 // Generate a fast subtype check. Branch to cast_ok if no failure. Return 0 if failure.
4041 __ li(R17_tos, 1);
4042 __ gen_subtype_check(RobjKlass, RspecifiedKlass, /*3 temp regs*/ Roffset, Rcpool, Rtags, /*target if subtype*/ Ldone);
4043 __ li(R17_tos, 0);
4044
4045 if (ProfileInterpreter) {
4046 __ b(Ldone);
4047 }
4048
4049 // Profile the null case.
4050 __ align(32, 12);
4051 __ bind(Lis_null);
4052 __ profile_null_seen(Rcpool, Rtags); // Rcpool and Rtags used as scratch.
4053
4054 __ align(32, 12);
4055 __ bind(Ldone);
4056 }
4057
4058 // =============================================================================
4059 // Breakpoints
4060
4061 void TemplateTable::_breakpoint() {
4062 transition(vtos, vtos);
4063
4064 // Get the unpatched byte code.
4065 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), R19_method, R14_bcp);
4066 __ mr(R31, R3_RET);
4067
4068 // Post the breakpoint event.
4069 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), R19_method, R14_bcp);
4070
4071 // Complete the execution of original bytecode.
4072 __ dispatch_Lbyte_code(vtos, R31, Interpreter::normal_table(vtos));
4073 }
4074
4075 // =============================================================================
4076 // Exceptions
4077
4078 void TemplateTable::athrow() {
4079 transition(atos, vtos);
4080
4081 // Exception oop is in tos
4082 __ verify_oop(R17_tos);
4083
4084 __ null_check_throw(R17_tos, -1, R11_scratch1);
4085
4086 // Throw exception interpreter entry expects exception oop to be in R3.
4087 __ mr(R3_RET, R17_tos);
4088 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::throw_exception_entry());
4089 __ mtctr(R11_scratch1);
4090 __ bctr();
4091 }
4092
4093 // =============================================================================
4094 // Synchronization
4095 // Searches the basic object lock list on the stack for a free slot
4096 // and uses it to lock the obect in tos.
4097 //
4098 // Recursive locking is enabled by exiting the search if the same
4099 // object is already found in the list. Thus, a new basic lock obj lock
4100 // is allocated "higher up" in the stack and thus is found first
4101 // at next monitor exit.
4102 void TemplateTable::monitorenter() {
4103 transition(atos, vtos);
4104
4105 __ verify_oop(R17_tos);
4106
4107 Register Rcurrent_monitor = R11_scratch1,
4108 Rcurrent_obj = R12_scratch2,
4109 Robj_to_lock = R17_tos,
4110 Rscratch1 = R3_ARG1,
4111 Rscratch2 = R4_ARG2,
4112 Rscratch3 = R5_ARG3,
4113 Rcurrent_obj_addr = R6_ARG4;
4114
4115 // ------------------------------------------------------------------------------
4116 // Null pointer exception.
4117 __ null_check_throw(Robj_to_lock, -1, R11_scratch1);
4118
4119 // Try to acquire a lock on the object.
4120 // Repeat until succeeded (i.e., until monitorenter returns true).
4121
4122 // ------------------------------------------------------------------------------
4123 // Find a free slot in the monitor block.
4124 Label Lfound, Lexit, Lallocate_new;
4125 ConditionRegister found_free_slot = CCR0,
4126 found_same_obj = CCR1,
4127 reached_limit = CCR6;
4128 {
4129 Label Lloop, Lentry;
4130 Register Rlimit = Rcurrent_monitor;
4131
4132 // Set up search loop - start with topmost monitor.
4133 __ add(Rcurrent_obj_addr, BasicObjectLock::obj_offset_in_bytes(), R26_monitor);
4134
4135 __ ld(Rlimit, 0, R1_SP);
4136 __ addi(Rlimit, Rlimit, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes() - BasicObjectLock::obj_offset_in_bytes())); // Monitor base
4137
4138 // Check if any slot is present => short cut to allocation if not.
4139 __ cmpld(reached_limit, Rcurrent_obj_addr, Rlimit);
4140 __ bgt(reached_limit, Lallocate_new);
4141
4142 // Pre-load topmost slot.
4143 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4144 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4145 // The search loop.
4146 __ bind(Lloop);
4147 // Found free slot?
4148 __ cmpdi(found_free_slot, Rcurrent_obj, 0);
4149 // Is this entry for same obj? If so, stop the search and take the found
4150 // free slot or allocate a new one to enable recursive locking.
4151 __ cmpd(found_same_obj, Rcurrent_obj, Robj_to_lock);
4152 __ cmpld(reached_limit, Rcurrent_obj_addr, Rlimit);
4153 __ beq(found_free_slot, Lexit);
4154 __ beq(found_same_obj, Lallocate_new);
4155 __ bgt(reached_limit, Lallocate_new);
4156 // Check if last allocated BasicLockObj reached.
4157 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4158 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4159 // Next iteration if unchecked BasicObjectLocks exist on the stack.
4160 __ b(Lloop);
4161 }
4162
4163 // ------------------------------------------------------------------------------
4164 // Check if we found a free slot.
4165 __ bind(Lexit);
4166
4167 __ addi(Rcurrent_monitor, Rcurrent_obj_addr, -(frame::interpreter_frame_monitor_size() * wordSize) - BasicObjectLock::obj_offset_in_bytes());
4168 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, - frame::interpreter_frame_monitor_size() * wordSize);
4169 __ b(Lfound);
4170
4171 // We didn't find a free BasicObjLock => allocate one.
4172 __ align(32, 12);
4173 __ bind(Lallocate_new);
4174 __ add_monitor_to_stack(false, Rscratch1, Rscratch2);
4175 __ mr(Rcurrent_monitor, R26_monitor);
4176 __ addi(Rcurrent_obj_addr, R26_monitor, BasicObjectLock::obj_offset_in_bytes());
4177
4178 // ------------------------------------------------------------------------------
4179 // We now have a slot to lock.
4180 __ bind(Lfound);
4181
4182 // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
4183 // The object has already been poped from the stack, so the expression stack looks correct.
4184 __ addi(R14_bcp, R14_bcp, 1);
4185
4186 __ std(Robj_to_lock, 0, Rcurrent_obj_addr);
4187 __ lock_object(Rcurrent_monitor, Robj_to_lock);
4188
4189 // Check if there's enough space on the stack for the monitors after locking.
4190 // This emits a single store.
4191 __ generate_stack_overflow_check(0);
4192
4193 // The bcp has already been incremented. Just need to dispatch to next instruction.
4194 __ dispatch_next(vtos);
4195 }
4196
4197 void TemplateTable::monitorexit() {
4198 transition(atos, vtos);
4199 __ verify_oop(R17_tos);
4200
4201 Register Rcurrent_monitor = R11_scratch1,
4202 Rcurrent_obj = R12_scratch2,
4203 Robj_to_lock = R17_tos,
4204 Rcurrent_obj_addr = R3_ARG1,
4205 Rlimit = R4_ARG2;
4206 Label Lfound, Lillegal_monitor_state;
4207
4208 // Check corner case: unbalanced monitorEnter / Exit.
4209 __ ld(Rlimit, 0, R1_SP);
4210 __ addi(Rlimit, Rlimit, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base
4211
4212 // Null pointer check.
4213 __ null_check_throw(Robj_to_lock, -1, R11_scratch1);
4214
4215 __ cmpld(CCR0, R26_monitor, Rlimit);
4216 __ bgt(CCR0, Lillegal_monitor_state);
4217
4218 // Find the corresponding slot in the monitors stack section.
4219 {
4220 Label Lloop;
4221
4222 // Start with topmost monitor.
4223 __ addi(Rcurrent_obj_addr, R26_monitor, BasicObjectLock::obj_offset_in_bytes());
4224 __ addi(Rlimit, Rlimit, BasicObjectLock::obj_offset_in_bytes());
4225 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4226 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4227
4228 __ bind(Lloop);
4229 // Is this entry for same obj?
4230 __ cmpd(CCR0, Rcurrent_obj, Robj_to_lock);
4231 __ beq(CCR0, Lfound);
4232
4233 // Check if last allocated BasicLockObj reached.
4234
4235 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4236 __ cmpld(CCR0, Rcurrent_obj_addr, Rlimit);
4237 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4238
4239 // Next iteration if unchecked BasicObjectLocks exist on the stack.
4240 __ ble(CCR0, Lloop);
4241 }
4242
4243 // Fell through without finding the basic obj lock => throw up!
4244 __ bind(Lillegal_monitor_state);
4245 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
4246 __ should_not_reach_here();
4247
4248 __ align(32, 12);
4249 __ bind(Lfound);
4250 __ addi(Rcurrent_monitor, Rcurrent_obj_addr,
4251 -(frame::interpreter_frame_monitor_size() * wordSize) - BasicObjectLock::obj_offset_in_bytes());
4252 __ unlock_object(Rcurrent_monitor);
4253 }
4254
4255 // ============================================================================
4256 // Wide bytecodes
4257
4258 // Wide instructions. Simply redirects to the wide entry point for that instruction.
4259 void TemplateTable::wide() {
4260 transition(vtos, vtos);
4261
4262 const Register Rtable = R11_scratch1,
4263 Rindex = R12_scratch2,
4264 Rtmp = R0;
4265
4266 __ lbz(Rindex, 1, R14_bcp);
4267
4268 __ load_dispatch_table(Rtable, Interpreter::_wentry_point);
4269
4270 __ slwi(Rindex, Rindex, LogBytesPerWord);
4271 __ ldx(Rtmp, Rtable, Rindex);
4272 __ mtctr(Rtmp);
4273 __ bctr();
4274 // Note: the bcp increment step is part of the individual wide bytecode implementations.
4275 }