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