1 /*
2 * Copyright (c) 2014, Oracle and/or its affiliates. All rights reserved.
3 * Copyright 2013, 2014 SAP AG. 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/templateInterpreter.hpp"
31 #include "interpreter/templateTable.hpp"
32 #include "memory/universe.inline.hpp"
33 #include "oops/objArrayKlass.hpp"
34 #include "oops/oop.inline.hpp"
35 #include "prims/methodHandles.hpp"
36 #include "runtime/sharedRuntime.hpp"
37 #include "runtime/stubRoutines.hpp"
38 #include "runtime/synchronizer.hpp"
39 #include "utilities/macros.hpp"
40
41 #ifndef CC_INTERP
42
43 #undef __
44 #define __ _masm->
45
46 // ============================================================================
47 // Misc helpers
48
49 // Do an oop store like *(base + index) = val OR *(base + offset) = val
50 // (only one of both variants is possible at the same time).
51 // Index can be noreg.
52 // Kills:
53 // Rbase, Rtmp
54 static void do_oop_store(InterpreterMacroAssembler* _masm,
55 Register Rbase,
56 RegisterOrConstant offset,
57 Register Rval, // Noreg means always null.
58 Register Rtmp1,
59 Register Rtmp2,
60 Register Rtmp3,
61 BarrierSet::Name barrier,
62 bool precise,
63 bool check_null) {
64 assert_different_registers(Rtmp1, Rtmp2, Rtmp3, Rval, Rbase);
65
66 switch (barrier) {
67 #if INCLUDE_ALL_GCS
68 case BarrierSet::G1SATBCT:
69 case BarrierSet::G1SATBCTLogging:
70 {
71 // Load and record the previous value.
72 __ g1_write_barrier_pre(Rbase, offset,
73 Rtmp3, /* holder of pre_val ? */
74 Rtmp1, Rtmp2, false /* frame */);
75
76 Label Lnull, Ldone;
77 if (Rval != noreg) {
78 if (check_null) {
79 __ cmpdi(CCR0, Rval, 0);
80 __ beq(CCR0, Lnull);
81 }
82 __ store_heap_oop_not_null(Rval, offset, Rbase, /*Rval must stay uncompressed.*/ Rtmp1);
83 // Mark the card.
84 if (!(offset.is_constant() && offset.as_constant() == 0) && precise) {
85 __ add(Rbase, offset, Rbase);
86 }
87 __ g1_write_barrier_post(Rbase, Rval, Rtmp1, Rtmp2, Rtmp3, /*filtered (fast path)*/ &Ldone);
88 if (check_null) { __ b(Ldone); }
89 }
90
91 if (Rval == noreg || check_null) { // Store null oop.
92 Register Rnull = Rval;
93 __ bind(Lnull);
94 if (Rval == noreg) {
95 Rnull = Rtmp1;
96 __ li(Rnull, 0);
97 }
98 if (UseCompressedOops) {
99 __ stw(Rnull, offset, Rbase);
100 } else {
101 __ std(Rnull, offset, Rbase);
102 }
103 }
104 __ bind(Ldone);
105 }
106 break;
107 #endif // INCLUDE_ALL_GCS
108 case BarrierSet::CardTableModRef:
109 case BarrierSet::CardTableExtension:
110 {
111 Label Lnull, Ldone;
112 if (Rval != noreg) {
113 if (check_null) {
114 __ cmpdi(CCR0, Rval, 0);
115 __ beq(CCR0, Lnull);
116 }
117 __ store_heap_oop_not_null(Rval, offset, Rbase, /*Rval should better stay uncompressed.*/ Rtmp1);
118 // Mark the card.
119 if (!(offset.is_constant() && offset.as_constant() == 0) && precise) {
120 __ add(Rbase, offset, Rbase);
121 }
122 __ card_write_barrier_post(Rbase, Rval, Rtmp1);
123 if (check_null) {
124 __ b(Ldone);
125 }
126 }
127
128 if (Rval == noreg || check_null) { // Store null oop.
129 Register Rnull = Rval;
130 __ bind(Lnull);
131 if (Rval == noreg) {
132 Rnull = Rtmp1;
133 __ li(Rnull, 0);
134 }
135 if (UseCompressedOops) {
136 __ stw(Rnull, offset, Rbase);
137 } else {
138 __ std(Rnull, offset, Rbase);
139 }
140 }
141 __ bind(Ldone);
142 }
143 break;
144 case BarrierSet::Epsilon:
145 Unimplemented(); // Should look like CardTableForRS without card mark?
146 break;
147 case BarrierSet::ModRef:
148 case BarrierSet::Other:
149 ShouldNotReachHere();
150 break;
151 default:
152 ShouldNotReachHere();
153 }
154 }
155
156 // ============================================================================
157 // Platform-dependent initialization
158
159 void TemplateTable::pd_initialize() {
160 // No ppc64 specific initialization.
161 }
162
163 Address TemplateTable::at_bcp(int offset) {
164 // Not used on ppc.
165 ShouldNotReachHere();
166 return Address();
167 }
168
169 // Patches the current bytecode (ptr to it located in bcp)
170 // in the bytecode stream with a new one.
171 void TemplateTable::patch_bytecode(Bytecodes::Code new_bc, Register Rnew_bc, Register Rtemp, bool load_bc_into_bc_reg /*=true*/, int byte_no) {
172 // With sharing on, may need to test method flag.
173 if (!RewriteBytecodes) return;
174 Label L_patch_done;
175
176 switch (new_bc) {
177 case Bytecodes::_fast_aputfield:
178 case Bytecodes::_fast_bputfield:
179 case Bytecodes::_fast_zputfield:
180 case Bytecodes::_fast_cputfield:
181 case Bytecodes::_fast_dputfield:
182 case Bytecodes::_fast_fputfield:
183 case Bytecodes::_fast_iputfield:
184 case Bytecodes::_fast_lputfield:
185 case Bytecodes::_fast_sputfield:
186 {
187 // We skip bytecode quickening for putfield instructions when
188 // the put_code written to the constant pool cache is zero.
189 // This is required so that every execution of this instruction
190 // calls out to InterpreterRuntime::resolve_get_put to do
191 // additional, required work.
192 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
193 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
194 __ get_cache_and_index_at_bcp(Rtemp /* dst = cache */, 1);
195 // ((*(cache+indices))>>((1+byte_no)*8))&0xFF:
196 #if defined(VM_LITTLE_ENDIAN)
197 __ lbz(Rnew_bc, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 1 + byte_no, Rtemp);
198 #else
199 __ lbz(Rnew_bc, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 7 - (1 + byte_no), Rtemp);
200 #endif
201 __ cmpwi(CCR0, Rnew_bc, 0);
202 __ li(Rnew_bc, (unsigned int)(unsigned char)new_bc);
203 __ beq(CCR0, L_patch_done);
204 // __ isync(); // acquire not needed
205 break;
206 }
207
208 default:
209 assert(byte_no == -1, "sanity");
210 if (load_bc_into_bc_reg) {
211 __ li(Rnew_bc, (unsigned int)(unsigned char)new_bc);
212 }
213 }
214
215 if (JvmtiExport::can_post_breakpoint()) {
216 Label L_fast_patch;
217 __ lbz(Rtemp, 0, R14_bcp);
218 __ cmpwi(CCR0, Rtemp, (unsigned int)(unsigned char)Bytecodes::_breakpoint);
219 __ bne(CCR0, L_fast_patch);
220 // Perform the quickening, slowly, in the bowels of the breakpoint table.
221 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), R19_method, R14_bcp, Rnew_bc);
222 __ b(L_patch_done);
223 __ bind(L_fast_patch);
224 }
225
226 // Patch bytecode.
227 __ stb(Rnew_bc, 0, R14_bcp);
228
229 __ bind(L_patch_done);
230 }
231
232 // ============================================================================
233 // Individual instructions
234
235 void TemplateTable::nop() {
236 transition(vtos, vtos);
237 // Nothing to do.
238 }
239
240 void TemplateTable::shouldnotreachhere() {
241 transition(vtos, vtos);
242 __ stop("shouldnotreachhere bytecode");
243 }
244
245 void TemplateTable::aconst_null() {
246 transition(vtos, atos);
247 __ li(R17_tos, 0);
248 }
249
250 void TemplateTable::iconst(int value) {
251 transition(vtos, itos);
252 assert(value >= -1 && value <= 5, "");
253 __ li(R17_tos, value);
254 }
255
256 void TemplateTable::lconst(int value) {
257 transition(vtos, ltos);
258 assert(value >= -1 && value <= 5, "");
259 __ li(R17_tos, value);
260 }
261
262 void TemplateTable::fconst(int value) {
263 transition(vtos, ftos);
264 static float zero = 0.0;
265 static float one = 1.0;
266 static float two = 2.0;
267 switch (value) {
268 default: ShouldNotReachHere();
269 case 0: {
270 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&zero, R0, true);
271 __ lfs(F15_ftos, simm16_offset, R11_scratch1);
272 break;
273 }
274 case 1: {
275 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&one, R0, true);
276 __ lfs(F15_ftos, simm16_offset, R11_scratch1);
277 break;
278 }
279 case 2: {
280 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&two, R0, true);
281 __ lfs(F15_ftos, simm16_offset, R11_scratch1);
282 break;
283 }
284 }
285 }
286
287 void TemplateTable::dconst(int value) {
288 transition(vtos, dtos);
289 static double zero = 0.0;
290 static double one = 1.0;
291 switch (value) {
292 case 0: {
293 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&zero, R0, true);
294 __ lfd(F15_ftos, simm16_offset, R11_scratch1);
295 break;
296 }
297 case 1: {
298 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&one, R0, true);
299 __ lfd(F15_ftos, simm16_offset, R11_scratch1);
300 break;
301 }
302 default: ShouldNotReachHere();
303 }
304 }
305
306 void TemplateTable::bipush() {
307 transition(vtos, itos);
308 __ lbz(R17_tos, 1, R14_bcp);
309 __ extsb(R17_tos, R17_tos);
310 }
311
312 void TemplateTable::sipush() {
313 transition(vtos, itos);
314 __ get_2_byte_integer_at_bcp(1, R17_tos, InterpreterMacroAssembler::Signed);
315 }
316
317 void TemplateTable::ldc(bool wide) {
318 Register Rscratch1 = R11_scratch1,
319 Rscratch2 = R12_scratch2,
320 Rcpool = R3_ARG1;
321
322 transition(vtos, vtos);
323 Label notInt, notClass, exit;
324
325 __ get_cpool_and_tags(Rcpool, Rscratch2); // Set Rscratch2 = &tags.
326 if (wide) { // Read index.
327 __ get_2_byte_integer_at_bcp(1, Rscratch1, InterpreterMacroAssembler::Unsigned);
328 } else {
329 __ lbz(Rscratch1, 1, R14_bcp);
330 }
331
332 const int base_offset = ConstantPool::header_size() * wordSize;
333 const int tags_offset = Array<u1>::base_offset_in_bytes();
334
335 // Get type from tags.
336 __ addi(Rscratch2, Rscratch2, tags_offset);
337 __ lbzx(Rscratch2, Rscratch2, Rscratch1);
338
339 __ cmpwi(CCR0, Rscratch2, JVM_CONSTANT_UnresolvedClass); // Unresolved class?
340 __ cmpwi(CCR1, Rscratch2, JVM_CONSTANT_UnresolvedClassInError); // Unresolved class in error state?
341 __ cror(/*CR0 eq*/2, /*CR1 eq*/4+2, /*CR0 eq*/2);
342
343 // Resolved class - need to call vm to get java mirror of the class.
344 __ cmpwi(CCR1, Rscratch2, JVM_CONSTANT_Class);
345 __ crnor(/*CR0 eq*/2, /*CR1 eq*/4+2, /*CR0 eq*/2); // Neither resolved class nor unresolved case from above?
346 __ beq(CCR0, notClass);
347
348 __ li(R4, wide ? 1 : 0);
349 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), R4);
350 __ push(atos);
351 __ b(exit);
352
353 __ align(32, 12);
354 __ bind(notClass);
355 __ addi(Rcpool, Rcpool, base_offset);
356 __ sldi(Rscratch1, Rscratch1, LogBytesPerWord);
357 __ cmpdi(CCR0, Rscratch2, JVM_CONSTANT_Integer);
358 __ bne(CCR0, notInt);
359 __ lwax(R17_tos, Rcpool, Rscratch1);
360 __ push(itos);
361 __ b(exit);
362
363 __ align(32, 12);
364 __ bind(notInt);
365 #ifdef ASSERT
366 // String and Object are rewritten to fast_aldc
367 __ cmpdi(CCR0, Rscratch2, JVM_CONSTANT_Float);
368 __ asm_assert_eq("unexpected type", 0x8765);
369 #endif
370 __ lfsx(F15_ftos, Rcpool, Rscratch1);
371 __ push(ftos);
372
373 __ align(32, 12);
374 __ bind(exit);
375 }
376
377 // Fast path for caching oop constants.
378 void TemplateTable::fast_aldc(bool wide) {
379 transition(vtos, atos);
380
381 int index_size = wide ? sizeof(u2) : sizeof(u1);
382 const Register Rscratch = R11_scratch1;
383 Label resolved;
384
385 // We are resolved if the resolved reference cache entry contains a
386 // non-null object (CallSite, etc.)
387 __ get_cache_index_at_bcp(Rscratch, 1, index_size); // Load index.
388 __ load_resolved_reference_at_index(R17_tos, Rscratch);
389 __ cmpdi(CCR0, R17_tos, 0);
390 __ bne(CCR0, resolved);
391 __ load_const_optimized(R3_ARG1, (int)bytecode());
392
393 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
394
395 // First time invocation - must resolve first.
396 __ call_VM(R17_tos, entry, R3_ARG1);
397
398 __ align(32, 12);
399 __ bind(resolved);
400 __ verify_oop(R17_tos);
401 }
402
403 void TemplateTable::ldc2_w() {
404 transition(vtos, vtos);
405 Label Llong, Lexit;
406
407 Register Rindex = R11_scratch1,
408 Rcpool = R12_scratch2,
409 Rtag = R3_ARG1;
410 __ get_cpool_and_tags(Rcpool, Rtag);
411 __ get_2_byte_integer_at_bcp(1, Rindex, InterpreterMacroAssembler::Unsigned);
412
413 const int base_offset = ConstantPool::header_size() * wordSize;
414 const int tags_offset = Array<u1>::base_offset_in_bytes();
415 // Get type from tags.
416 __ addi(Rcpool, Rcpool, base_offset);
417 __ addi(Rtag, Rtag, tags_offset);
418
419 __ lbzx(Rtag, Rtag, Rindex);
420
421 __ sldi(Rindex, Rindex, LogBytesPerWord);
422 __ cmpdi(CCR0, Rtag, JVM_CONSTANT_Double);
423 __ bne(CCR0, Llong);
424 // A double can be placed at word-aligned locations in the constant pool.
425 // Check out Conversions.java for an example.
426 // Also ConstantPool::header_size() is 20, which makes it very difficult
427 // to double-align double on the constant pool. SG, 11/7/97
428 __ lfdx(F15_ftos, Rcpool, Rindex);
429 __ push(dtos);
430 __ b(Lexit);
431
432 __ bind(Llong);
433 __ ldx(R17_tos, Rcpool, Rindex);
434 __ push(ltos);
435
436 __ bind(Lexit);
437 }
438
439 // Get the locals index located in the bytecode stream at bcp + offset.
440 void TemplateTable::locals_index(Register Rdst, int offset) {
441 __ lbz(Rdst, offset, R14_bcp);
442 }
443
444 void TemplateTable::iload() {
445 transition(vtos, itos);
446
447 // Get the local value into tos
448 const Register Rindex = R22_tmp2;
449 locals_index(Rindex);
450
451 // Rewrite iload,iload pair into fast_iload2
452 // iload,caload pair into fast_icaload
453 if (RewriteFrequentPairs) {
454 Label Lrewrite, Ldone;
455 Register Rnext_byte = R3_ARG1,
456 Rrewrite_to = R6_ARG4,
457 Rscratch = R11_scratch1;
458
459 // get next byte
460 __ lbz(Rnext_byte, Bytecodes::length_for(Bytecodes::_iload), R14_bcp);
461
462 // if _iload, wait to rewrite to iload2. We only want to rewrite the
463 // last two iloads in a pair. Comparing against fast_iload means that
464 // the next bytecode is neither an iload or a caload, and therefore
465 // an iload pair.
466 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_iload);
467 __ beq(CCR0, Ldone);
468
469 __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_iload);
470 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iload2);
471 __ beq(CCR1, Lrewrite);
472
473 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_caload);
474 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_icaload);
475 __ beq(CCR0, Lrewrite);
476
477 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iload);
478
479 __ bind(Lrewrite);
480 patch_bytecode(Bytecodes::_iload, Rrewrite_to, Rscratch, false);
481 __ bind(Ldone);
482 }
483
484 __ load_local_int(R17_tos, Rindex, Rindex);
485 }
486
487 // Load 2 integers in a row without dispatching
488 void TemplateTable::fast_iload2() {
489 transition(vtos, itos);
490
491 __ lbz(R3_ARG1, 1, R14_bcp);
492 __ lbz(R17_tos, Bytecodes::length_for(Bytecodes::_iload) + 1, R14_bcp);
493
494 __ load_local_int(R3_ARG1, R11_scratch1, R3_ARG1);
495 __ load_local_int(R17_tos, R12_scratch2, R17_tos);
496 __ push_i(R3_ARG1);
497 }
498
499 void TemplateTable::fast_iload() {
500 transition(vtos, itos);
501 // Get the local value into tos
502
503 const Register Rindex = R11_scratch1;
504 locals_index(Rindex);
505 __ load_local_int(R17_tos, Rindex, Rindex);
506 }
507
508 // Load a local variable type long from locals area to TOS cache register.
509 // Local index resides in bytecodestream.
510 void TemplateTable::lload() {
511 transition(vtos, ltos);
512
513 const Register Rindex = R11_scratch1;
514 locals_index(Rindex);
515 __ load_local_long(R17_tos, Rindex, Rindex);
516 }
517
518 void TemplateTable::fload() {
519 transition(vtos, ftos);
520
521 const Register Rindex = R11_scratch1;
522 locals_index(Rindex);
523 __ load_local_float(F15_ftos, Rindex, Rindex);
524 }
525
526 void TemplateTable::dload() {
527 transition(vtos, dtos);
528
529 const Register Rindex = R11_scratch1;
530 locals_index(Rindex);
531 __ load_local_double(F15_ftos, Rindex, Rindex);
532 }
533
534 void TemplateTable::aload() {
535 transition(vtos, atos);
536
537 const Register Rindex = R11_scratch1;
538 locals_index(Rindex);
539 __ load_local_ptr(R17_tos, Rindex, Rindex);
540 }
541
542 void TemplateTable::locals_index_wide(Register Rdst) {
543 // Offset is 2, not 1, because Lbcp points to wide prefix code.
544 __ get_2_byte_integer_at_bcp(2, Rdst, InterpreterMacroAssembler::Unsigned);
545 }
546
547 void TemplateTable::wide_iload() {
548 // Get the local value into tos.
549
550 const Register Rindex = R11_scratch1;
551 locals_index_wide(Rindex);
552 __ load_local_int(R17_tos, Rindex, Rindex);
553 }
554
555 void TemplateTable::wide_lload() {
556 transition(vtos, ltos);
557
558 const Register Rindex = R11_scratch1;
559 locals_index_wide(Rindex);
560 __ load_local_long(R17_tos, Rindex, Rindex);
561 }
562
563 void TemplateTable::wide_fload() {
564 transition(vtos, ftos);
565
566 const Register Rindex = R11_scratch1;
567 locals_index_wide(Rindex);
568 __ load_local_float(F15_ftos, Rindex, Rindex);
569 }
570
571 void TemplateTable::wide_dload() {
572 transition(vtos, dtos);
573
574 const Register Rindex = R11_scratch1;
575 locals_index_wide(Rindex);
576 __ load_local_double(F15_ftos, Rindex, Rindex);
577 }
578
579 void TemplateTable::wide_aload() {
580 transition(vtos, atos);
581
582 const Register Rindex = R11_scratch1;
583 locals_index_wide(Rindex);
584 __ load_local_ptr(R17_tos, Rindex, Rindex);
585 }
586
587 void TemplateTable::iaload() {
588 transition(itos, itos);
589
590 const Register Rload_addr = R3_ARG1,
591 Rarray = R4_ARG2,
592 Rtemp = R5_ARG3;
593 __ index_check(Rarray, R17_tos /* index */, LogBytesPerInt, Rtemp, Rload_addr);
594 __ lwa(R17_tos, arrayOopDesc::base_offset_in_bytes(T_INT), Rload_addr);
595 }
596
597 void TemplateTable::laload() {
598 transition(itos, ltos);
599
600 const Register Rload_addr = R3_ARG1,
601 Rarray = R4_ARG2,
602 Rtemp = R5_ARG3;
603 __ index_check(Rarray, R17_tos /* index */, LogBytesPerLong, Rtemp, Rload_addr);
604 __ ld(R17_tos, arrayOopDesc::base_offset_in_bytes(T_LONG), Rload_addr);
605 }
606
607 void TemplateTable::faload() {
608 transition(itos, ftos);
609
610 const Register Rload_addr = R3_ARG1,
611 Rarray = R4_ARG2,
612 Rtemp = R5_ARG3;
613 __ index_check(Rarray, R17_tos /* index */, LogBytesPerInt, Rtemp, Rload_addr);
614 __ lfs(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Rload_addr);
615 }
616
617 void TemplateTable::daload() {
618 transition(itos, dtos);
619
620 const Register Rload_addr = R3_ARG1,
621 Rarray = R4_ARG2,
622 Rtemp = R5_ARG3;
623 __ index_check(Rarray, R17_tos /* index */, LogBytesPerLong, Rtemp, Rload_addr);
624 __ lfd(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Rload_addr);
625 }
626
627 void TemplateTable::aaload() {
628 transition(itos, atos);
629
630 // tos: index
631 // result tos: array
632 const Register Rload_addr = R3_ARG1,
633 Rarray = R4_ARG2,
634 Rtemp = R5_ARG3;
635 __ index_check(Rarray, R17_tos /* index */, UseCompressedOops ? 2 : LogBytesPerWord, Rtemp, Rload_addr);
636 __ load_heap_oop(R17_tos, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Rload_addr);
637 __ verify_oop(R17_tos);
638 //__ dcbt(R17_tos); // prefetch
639 }
640
641 void TemplateTable::baload() {
642 transition(itos, itos);
643
644 const Register Rload_addr = R3_ARG1,
645 Rarray = R4_ARG2,
646 Rtemp = R5_ARG3;
647 __ index_check(Rarray, R17_tos /* index */, 0, Rtemp, Rload_addr);
648 __ lbz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_BYTE), Rload_addr);
649 __ extsb(R17_tos, R17_tos);
650 }
651
652 void TemplateTable::caload() {
653 transition(itos, itos);
654
655 const Register Rload_addr = R3_ARG1,
656 Rarray = R4_ARG2,
657 Rtemp = R5_ARG3;
658 __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
659 __ lhz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rload_addr);
660 }
661
662 // Iload followed by caload frequent pair.
663 void TemplateTable::fast_icaload() {
664 transition(vtos, itos);
665
666 const Register Rload_addr = R3_ARG1,
667 Rarray = R4_ARG2,
668 Rtemp = R11_scratch1;
669
670 locals_index(R17_tos);
671 __ load_local_int(R17_tos, Rtemp, R17_tos);
672 __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
673 __ lhz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rload_addr);
674 }
675
676 void TemplateTable::saload() {
677 transition(itos, itos);
678
679 const Register Rload_addr = R11_scratch1,
680 Rarray = R12_scratch2,
681 Rtemp = R3_ARG1;
682 __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
683 __ lha(R17_tos, arrayOopDesc::base_offset_in_bytes(T_SHORT), Rload_addr);
684 }
685
686 void TemplateTable::iload(int n) {
687 transition(vtos, itos);
688
689 __ lwz(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
690 }
691
692 void TemplateTable::lload(int n) {
693 transition(vtos, ltos);
694
695 __ ld(R17_tos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
696 }
697
698 void TemplateTable::fload(int n) {
699 transition(vtos, ftos);
700
701 __ lfs(F15_ftos, Interpreter::local_offset_in_bytes(n), R18_locals);
702 }
703
704 void TemplateTable::dload(int n) {
705 transition(vtos, dtos);
706
707 __ lfd(F15_ftos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
708 }
709
710 void TemplateTable::aload(int n) {
711 transition(vtos, atos);
712
713 __ ld(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
714 }
715
716 void TemplateTable::aload_0() {
717 transition(vtos, atos);
718 // According to bytecode histograms, the pairs:
719 //
720 // _aload_0, _fast_igetfield
721 // _aload_0, _fast_agetfield
722 // _aload_0, _fast_fgetfield
723 //
724 // occur frequently. If RewriteFrequentPairs is set, the (slow)
725 // _aload_0 bytecode checks if the next bytecode is either
726 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
727 // rewrites the current bytecode into a pair bytecode; otherwise it
728 // rewrites the current bytecode into _0 that doesn't do
729 // the pair check anymore.
730 //
731 // Note: If the next bytecode is _getfield, the rewrite must be
732 // delayed, otherwise we may miss an opportunity for a pair.
733 //
734 // Also rewrite frequent pairs
735 // aload_0, aload_1
736 // aload_0, iload_1
737 // These bytecodes with a small amount of code are most profitable
738 // to rewrite.
739
740 if (RewriteFrequentPairs) {
741
742 Label Lrewrite, Ldont_rewrite;
743 Register Rnext_byte = R3_ARG1,
744 Rrewrite_to = R6_ARG4,
745 Rscratch = R11_scratch1;
746
747 // Get next byte.
748 __ lbz(Rnext_byte, Bytecodes::length_for(Bytecodes::_aload_0), R14_bcp);
749
750 // If _getfield, wait to rewrite. We only want to rewrite the last two bytecodes in a pair.
751 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_getfield);
752 __ beq(CCR0, Ldont_rewrite);
753
754 __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_igetfield);
755 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iaccess_0);
756 __ beq(CCR1, Lrewrite);
757
758 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_agetfield);
759 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_aaccess_0);
760 __ beq(CCR0, Lrewrite);
761
762 __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_fgetfield);
763 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_faccess_0);
764 __ beq(CCR1, Lrewrite);
765
766 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_aload_0);
767
768 __ bind(Lrewrite);
769 patch_bytecode(Bytecodes::_aload_0, Rrewrite_to, Rscratch, false);
770 __ bind(Ldont_rewrite);
771 }
772
773 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
774 aload(0);
775 }
776
777 void TemplateTable::istore() {
778 transition(itos, vtos);
779
780 const Register Rindex = R11_scratch1;
781 locals_index(Rindex);
782 __ store_local_int(R17_tos, Rindex);
783 }
784
785 void TemplateTable::lstore() {
786 transition(ltos, vtos);
787 const Register Rindex = R11_scratch1;
788 locals_index(Rindex);
789 __ store_local_long(R17_tos, Rindex);
790 }
791
792 void TemplateTable::fstore() {
793 transition(ftos, vtos);
794
795 const Register Rindex = R11_scratch1;
796 locals_index(Rindex);
797 __ store_local_float(F15_ftos, Rindex);
798 }
799
800 void TemplateTable::dstore() {
801 transition(dtos, vtos);
802
803 const Register Rindex = R11_scratch1;
804 locals_index(Rindex);
805 __ store_local_double(F15_ftos, Rindex);
806 }
807
808 void TemplateTable::astore() {
809 transition(vtos, vtos);
810
811 const Register Rindex = R11_scratch1;
812 __ pop_ptr();
813 __ verify_oop_or_return_address(R17_tos, Rindex);
814 locals_index(Rindex);
815 __ store_local_ptr(R17_tos, Rindex);
816 }
817
818 void TemplateTable::wide_istore() {
819 transition(vtos, vtos);
820
821 const Register Rindex = R11_scratch1;
822 __ pop_i();
823 locals_index_wide(Rindex);
824 __ store_local_int(R17_tos, Rindex);
825 }
826
827 void TemplateTable::wide_lstore() {
828 transition(vtos, vtos);
829
830 const Register Rindex = R11_scratch1;
831 __ pop_l();
832 locals_index_wide(Rindex);
833 __ store_local_long(R17_tos, Rindex);
834 }
835
836 void TemplateTable::wide_fstore() {
837 transition(vtos, vtos);
838
839 const Register Rindex = R11_scratch1;
840 __ pop_f();
841 locals_index_wide(Rindex);
842 __ store_local_float(F15_ftos, Rindex);
843 }
844
845 void TemplateTable::wide_dstore() {
846 transition(vtos, vtos);
847
848 const Register Rindex = R11_scratch1;
849 __ pop_d();
850 locals_index_wide(Rindex);
851 __ store_local_double(F15_ftos, Rindex);
852 }
853
854 void TemplateTable::wide_astore() {
855 transition(vtos, vtos);
856
857 const Register Rindex = R11_scratch1;
858 __ pop_ptr();
859 __ verify_oop_or_return_address(R17_tos, Rindex);
860 locals_index_wide(Rindex);
861 __ store_local_ptr(R17_tos, Rindex);
862 }
863
864 void TemplateTable::iastore() {
865 transition(itos, vtos);
866
867 const Register Rindex = R3_ARG1,
868 Rstore_addr = R4_ARG2,
869 Rarray = R5_ARG3,
870 Rtemp = R6_ARG4;
871 __ pop_i(Rindex);
872 __ index_check(Rarray, Rindex, LogBytesPerInt, Rtemp, Rstore_addr);
873 __ stw(R17_tos, arrayOopDesc::base_offset_in_bytes(T_INT), Rstore_addr);
874 }
875
876 void TemplateTable::lastore() {
877 transition(ltos, vtos);
878
879 const Register Rindex = R3_ARG1,
880 Rstore_addr = R4_ARG2,
881 Rarray = R5_ARG3,
882 Rtemp = R6_ARG4;
883 __ pop_i(Rindex);
884 __ index_check(Rarray, Rindex, LogBytesPerLong, Rtemp, Rstore_addr);
885 __ std(R17_tos, arrayOopDesc::base_offset_in_bytes(T_LONG), Rstore_addr);
886 }
887
888 void TemplateTable::fastore() {
889 transition(ftos, vtos);
890
891 const Register Rindex = R3_ARG1,
892 Rstore_addr = R4_ARG2,
893 Rarray = R5_ARG3,
894 Rtemp = R6_ARG4;
895 __ pop_i(Rindex);
896 __ index_check(Rarray, Rindex, LogBytesPerInt, Rtemp, Rstore_addr);
897 __ stfs(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Rstore_addr);
898 }
899
900 void TemplateTable::dastore() {
901 transition(dtos, vtos);
902
903 const Register Rindex = R3_ARG1,
904 Rstore_addr = R4_ARG2,
905 Rarray = R5_ARG3,
906 Rtemp = R6_ARG4;
907 __ pop_i(Rindex);
908 __ index_check(Rarray, Rindex, LogBytesPerLong, Rtemp, Rstore_addr);
909 __ stfd(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Rstore_addr);
910 }
911
912 // Pop 3 values from the stack and...
913 void TemplateTable::aastore() {
914 transition(vtos, vtos);
915
916 Label Lstore_ok, Lis_null, Ldone;
917 const Register Rindex = R3_ARG1,
918 Rarray = R4_ARG2,
919 Rscratch = R11_scratch1,
920 Rscratch2 = R12_scratch2,
921 Rarray_klass = R5_ARG3,
922 Rarray_element_klass = Rarray_klass,
923 Rvalue_klass = R6_ARG4,
924 Rstore_addr = R31; // Use register which survives VM call.
925
926 __ ld(R17_tos, Interpreter::expr_offset_in_bytes(0), R15_esp); // Get value to store.
927 __ lwz(Rindex, Interpreter::expr_offset_in_bytes(1), R15_esp); // Get index.
928 __ ld(Rarray, Interpreter::expr_offset_in_bytes(2), R15_esp); // Get array.
929
930 __ verify_oop(R17_tos);
931 __ index_check_without_pop(Rarray, Rindex, UseCompressedOops ? 2 : LogBytesPerWord, Rscratch, Rstore_addr);
932 // Rindex is dead!
933 Register Rscratch3 = Rindex;
934
935 // Do array store check - check for NULL value first.
936 __ cmpdi(CCR0, R17_tos, 0);
937 __ beq(CCR0, Lis_null);
938
939 __ load_klass(Rarray_klass, Rarray);
940 __ load_klass(Rvalue_klass, R17_tos);
941
942 // Do fast instanceof cache test.
943 __ ld(Rarray_element_klass, in_bytes(ObjArrayKlass::element_klass_offset()), Rarray_klass);
944
945 // Generate a fast subtype check. Branch to store_ok if no failure. Throw if failure.
946 __ gen_subtype_check(Rvalue_klass /*subklass*/, Rarray_element_klass /*superklass*/, Rscratch, Rscratch2, Rscratch3, Lstore_ok);
947
948 // Fell through: subtype check failed => throw an exception.
949 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArrayStoreException_entry);
950 __ mtctr(R11_scratch1);
951 __ bctr();
952
953 __ bind(Lis_null);
954 do_oop_store(_masm, Rstore_addr, arrayOopDesc::base_offset_in_bytes(T_OBJECT), noreg /* 0 */,
955 Rscratch, Rscratch2, Rscratch3, _bs->kind(), true /* precise */, false /* check_null */);
956 __ profile_null_seen(Rscratch, Rscratch2);
957 __ b(Ldone);
958
959 // Store is OK.
960 __ bind(Lstore_ok);
961 do_oop_store(_masm, Rstore_addr, arrayOopDesc::base_offset_in_bytes(T_OBJECT), R17_tos /* value */,
962 Rscratch, Rscratch2, Rscratch3, _bs->kind(), true /* precise */, false /* check_null */);
963
964 __ bind(Ldone);
965 // Adjust sp (pops array, index and value).
966 __ addi(R15_esp, R15_esp, 3 * Interpreter::stackElementSize);
967 }
968
969 void TemplateTable::bastore() {
970 transition(itos, vtos);
971
972 const Register Rindex = R11_scratch1,
973 Rarray = R12_scratch2,
974 Rscratch = R3_ARG1;
975 __ pop_i(Rindex);
976 __ pop_ptr(Rarray);
977 // tos: val
978
979 // Need to check whether array is boolean or byte
980 // since both types share the bastore bytecode.
981 __ load_klass(Rscratch, Rarray);
982 __ lwz(Rscratch, in_bytes(Klass::layout_helper_offset()), Rscratch);
983 int diffbit = exact_log2(Klass::layout_helper_boolean_diffbit());
984 __ testbitdi(CCR0, R0, Rscratch, diffbit);
985 Label L_skip;
986 __ bfalse(CCR0, L_skip);
987 __ andi(R17_tos, R17_tos, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1
988 __ bind(L_skip);
989
990 __ index_check_without_pop(Rarray, Rindex, 0, Rscratch, Rarray);
991 __ stb(R17_tos, arrayOopDesc::base_offset_in_bytes(T_BYTE), Rarray);
992 }
993
994 void TemplateTable::castore() {
995 transition(itos, vtos);
996
997 const Register Rindex = R11_scratch1,
998 Rarray = R12_scratch2,
999 Rscratch = R3_ARG1;
1000 __ pop_i(Rindex);
1001 // tos: val
1002 // Rarray: array ptr (popped by index_check)
1003 __ index_check(Rarray, Rindex, LogBytesPerShort, Rscratch, Rarray);
1004 __ sth(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rarray);
1005 }
1006
1007 void TemplateTable::sastore() {
1008 castore();
1009 }
1010
1011 void TemplateTable::istore(int n) {
1012 transition(itos, vtos);
1013 __ stw(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
1014 }
1015
1016 void TemplateTable::lstore(int n) {
1017 transition(ltos, vtos);
1018 __ std(R17_tos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
1019 }
1020
1021 void TemplateTable::fstore(int n) {
1022 transition(ftos, vtos);
1023 __ stfs(F15_ftos, Interpreter::local_offset_in_bytes(n), R18_locals);
1024 }
1025
1026 void TemplateTable::dstore(int n) {
1027 transition(dtos, vtos);
1028 __ stfd(F15_ftos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
1029 }
1030
1031 void TemplateTable::astore(int n) {
1032 transition(vtos, vtos);
1033
1034 __ pop_ptr();
1035 __ verify_oop_or_return_address(R17_tos, R11_scratch1);
1036 __ std(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
1037 }
1038
1039 void TemplateTable::pop() {
1040 transition(vtos, vtos);
1041
1042 __ addi(R15_esp, R15_esp, Interpreter::stackElementSize);
1043 }
1044
1045 void TemplateTable::pop2() {
1046 transition(vtos, vtos);
1047
1048 __ addi(R15_esp, R15_esp, Interpreter::stackElementSize * 2);
1049 }
1050
1051 void TemplateTable::dup() {
1052 transition(vtos, vtos);
1053
1054 __ ld(R11_scratch1, Interpreter::stackElementSize, R15_esp);
1055 __ push_ptr(R11_scratch1);
1056 }
1057
1058 void TemplateTable::dup_x1() {
1059 transition(vtos, vtos);
1060
1061 Register Ra = R11_scratch1,
1062 Rb = R12_scratch2;
1063 // stack: ..., a, b
1064 __ ld(Rb, Interpreter::stackElementSize, R15_esp);
1065 __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1066 __ std(Rb, Interpreter::stackElementSize * 2, R15_esp);
1067 __ std(Ra, Interpreter::stackElementSize, R15_esp);
1068 __ push_ptr(Rb);
1069 // stack: ..., b, a, b
1070 }
1071
1072 void TemplateTable::dup_x2() {
1073 transition(vtos, vtos);
1074
1075 Register Ra = R11_scratch1,
1076 Rb = R12_scratch2,
1077 Rc = R3_ARG1;
1078
1079 // stack: ..., a, b, c
1080 __ ld(Rc, Interpreter::stackElementSize, R15_esp); // load c
1081 __ ld(Ra, Interpreter::stackElementSize * 3, R15_esp); // load a
1082 __ std(Rc, Interpreter::stackElementSize * 3, R15_esp); // store c in a
1083 __ ld(Rb, Interpreter::stackElementSize * 2, R15_esp); // load b
1084 // stack: ..., c, b, c
1085 __ std(Ra, Interpreter::stackElementSize * 2, R15_esp); // store a in b
1086 // stack: ..., c, a, c
1087 __ std(Rb, Interpreter::stackElementSize, R15_esp); // store b in c
1088 __ push_ptr(Rc); // push c
1089 // stack: ..., c, a, b, c
1090 }
1091
1092 void TemplateTable::dup2() {
1093 transition(vtos, vtos);
1094
1095 Register Ra = R11_scratch1,
1096 Rb = R12_scratch2;
1097 // stack: ..., a, b
1098 __ ld(Rb, Interpreter::stackElementSize, R15_esp);
1099 __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1100 __ push_2ptrs(Ra, Rb);
1101 // stack: ..., a, b, a, b
1102 }
1103
1104 void TemplateTable::dup2_x1() {
1105 transition(vtos, vtos);
1106
1107 Register Ra = R11_scratch1,
1108 Rb = R12_scratch2,
1109 Rc = R3_ARG1;
1110 // stack: ..., a, b, c
1111 __ ld(Rc, Interpreter::stackElementSize, R15_esp);
1112 __ ld(Rb, Interpreter::stackElementSize * 2, R15_esp);
1113 __ std(Rc, Interpreter::stackElementSize * 2, R15_esp);
1114 __ ld(Ra, Interpreter::stackElementSize * 3, R15_esp);
1115 __ std(Ra, Interpreter::stackElementSize, R15_esp);
1116 __ std(Rb, Interpreter::stackElementSize * 3, R15_esp);
1117 // stack: ..., b, c, a
1118 __ push_2ptrs(Rb, Rc);
1119 // stack: ..., b, c, a, b, c
1120 }
1121
1122 void TemplateTable::dup2_x2() {
1123 transition(vtos, vtos);
1124
1125 Register Ra = R11_scratch1,
1126 Rb = R12_scratch2,
1127 Rc = R3_ARG1,
1128 Rd = R4_ARG2;
1129 // stack: ..., a, b, c, d
1130 __ ld(Rb, Interpreter::stackElementSize * 3, R15_esp);
1131 __ ld(Rd, Interpreter::stackElementSize, R15_esp);
1132 __ std(Rb, Interpreter::stackElementSize, R15_esp); // store b in d
1133 __ std(Rd, Interpreter::stackElementSize * 3, R15_esp); // store d in b
1134 __ ld(Ra, Interpreter::stackElementSize * 4, R15_esp);
1135 __ ld(Rc, Interpreter::stackElementSize * 2, R15_esp);
1136 __ std(Ra, Interpreter::stackElementSize * 2, R15_esp); // store a in c
1137 __ std(Rc, Interpreter::stackElementSize * 4, R15_esp); // store c in a
1138 // stack: ..., c, d, a, b
1139 __ push_2ptrs(Rc, Rd);
1140 // stack: ..., c, d, a, b, c, d
1141 }
1142
1143 void TemplateTable::swap() {
1144 transition(vtos, vtos);
1145 // stack: ..., a, b
1146
1147 Register Ra = R11_scratch1,
1148 Rb = R12_scratch2;
1149 // stack: ..., a, b
1150 __ ld(Rb, Interpreter::stackElementSize, R15_esp);
1151 __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1152 __ std(Rb, Interpreter::stackElementSize * 2, R15_esp);
1153 __ std(Ra, Interpreter::stackElementSize, R15_esp);
1154 // stack: ..., b, a
1155 }
1156
1157 void TemplateTable::iop2(Operation op) {
1158 transition(itos, itos);
1159
1160 Register Rscratch = R11_scratch1;
1161
1162 __ pop_i(Rscratch);
1163 // tos = number of bits to shift
1164 // Rscratch = value to shift
1165 switch (op) {
1166 case add: __ add(R17_tos, Rscratch, R17_tos); break;
1167 case sub: __ sub(R17_tos, Rscratch, R17_tos); break;
1168 case mul: __ mullw(R17_tos, Rscratch, R17_tos); break;
1169 case _and: __ andr(R17_tos, Rscratch, R17_tos); break;
1170 case _or: __ orr(R17_tos, Rscratch, R17_tos); break;
1171 case _xor: __ xorr(R17_tos, Rscratch, R17_tos); break;
1172 case shl: __ rldicl(R17_tos, R17_tos, 0, 64-5); __ slw(R17_tos, Rscratch, R17_tos); break;
1173 case shr: __ rldicl(R17_tos, R17_tos, 0, 64-5); __ sraw(R17_tos, Rscratch, R17_tos); break;
1174 case ushr: __ rldicl(R17_tos, R17_tos, 0, 64-5); __ srw(R17_tos, Rscratch, R17_tos); break;
1175 default: ShouldNotReachHere();
1176 }
1177 }
1178
1179 void TemplateTable::lop2(Operation op) {
1180 transition(ltos, ltos);
1181
1182 Register Rscratch = R11_scratch1;
1183 __ pop_l(Rscratch);
1184 switch (op) {
1185 case add: __ add(R17_tos, Rscratch, R17_tos); break;
1186 case sub: __ sub(R17_tos, Rscratch, R17_tos); break;
1187 case _and: __ andr(R17_tos, Rscratch, R17_tos); break;
1188 case _or: __ orr(R17_tos, Rscratch, R17_tos); break;
1189 case _xor: __ xorr(R17_tos, Rscratch, R17_tos); break;
1190 default: ShouldNotReachHere();
1191 }
1192 }
1193
1194 void TemplateTable::idiv() {
1195 transition(itos, itos);
1196
1197 Label Lnormal, Lexception, Ldone;
1198 Register Rdividend = R11_scratch1; // Used by irem.
1199
1200 __ addi(R0, R17_tos, 1);
1201 __ cmplwi(CCR0, R0, 2);
1202 __ bgt(CCR0, Lnormal); // divisor <-1 or >1
1203
1204 __ cmpwi(CCR1, R17_tos, 0);
1205 __ beq(CCR1, Lexception); // divisor == 0
1206
1207 __ pop_i(Rdividend);
1208 __ mullw(R17_tos, Rdividend, R17_tos); // div by +/-1
1209 __ b(Ldone);
1210
1211 __ bind(Lexception);
1212 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArithmeticException_entry);
1213 __ mtctr(R11_scratch1);
1214 __ bctr();
1215
1216 __ align(32, 12);
1217 __ bind(Lnormal);
1218 __ pop_i(Rdividend);
1219 __ divw(R17_tos, Rdividend, R17_tos); // Can't divide minint/-1.
1220 __ bind(Ldone);
1221 }
1222
1223 void TemplateTable::irem() {
1224 transition(itos, itos);
1225
1226 __ mr(R12_scratch2, R17_tos);
1227 idiv();
1228 __ mullw(R17_tos, R17_tos, R12_scratch2);
1229 __ subf(R17_tos, R17_tos, R11_scratch1); // Dividend set by idiv.
1230 }
1231
1232 void TemplateTable::lmul() {
1233 transition(ltos, ltos);
1234
1235 __ pop_l(R11_scratch1);
1236 __ mulld(R17_tos, R11_scratch1, R17_tos);
1237 }
1238
1239 void TemplateTable::ldiv() {
1240 transition(ltos, ltos);
1241
1242 Label Lnormal, Lexception, Ldone;
1243 Register Rdividend = R11_scratch1; // Used by lrem.
1244
1245 __ addi(R0, R17_tos, 1);
1246 __ cmpldi(CCR0, R0, 2);
1247 __ bgt(CCR0, Lnormal); // divisor <-1 or >1
1248
1249 __ cmpdi(CCR1, R17_tos, 0);
1250 __ beq(CCR1, Lexception); // divisor == 0
1251
1252 __ pop_l(Rdividend);
1253 __ mulld(R17_tos, Rdividend, R17_tos); // div by +/-1
1254 __ b(Ldone);
1255
1256 __ bind(Lexception);
1257 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArithmeticException_entry);
1258 __ mtctr(R11_scratch1);
1259 __ bctr();
1260
1261 __ align(32, 12);
1262 __ bind(Lnormal);
1263 __ pop_l(Rdividend);
1264 __ divd(R17_tos, Rdividend, R17_tos); // Can't divide minint/-1.
1265 __ bind(Ldone);
1266 }
1267
1268 void TemplateTable::lrem() {
1269 transition(ltos, ltos);
1270
1271 __ mr(R12_scratch2, R17_tos);
1272 ldiv();
1273 __ mulld(R17_tos, R17_tos, R12_scratch2);
1274 __ subf(R17_tos, R17_tos, R11_scratch1); // Dividend set by ldiv.
1275 }
1276
1277 void TemplateTable::lshl() {
1278 transition(itos, ltos);
1279
1280 __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1281 __ pop_l(R11_scratch1);
1282 __ sld(R17_tos, R11_scratch1, R17_tos);
1283 }
1284
1285 void TemplateTable::lshr() {
1286 transition(itos, ltos);
1287
1288 __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1289 __ pop_l(R11_scratch1);
1290 __ srad(R17_tos, R11_scratch1, R17_tos);
1291 }
1292
1293 void TemplateTable::lushr() {
1294 transition(itos, ltos);
1295
1296 __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1297 __ pop_l(R11_scratch1);
1298 __ srd(R17_tos, R11_scratch1, R17_tos);
1299 }
1300
1301 void TemplateTable::fop2(Operation op) {
1302 transition(ftos, ftos);
1303
1304 switch (op) {
1305 case add: __ pop_f(F0_SCRATCH); __ fadds(F15_ftos, F0_SCRATCH, F15_ftos); break;
1306 case sub: __ pop_f(F0_SCRATCH); __ fsubs(F15_ftos, F0_SCRATCH, F15_ftos); break;
1307 case mul: __ pop_f(F0_SCRATCH); __ fmuls(F15_ftos, F0_SCRATCH, F15_ftos); break;
1308 case div: __ pop_f(F0_SCRATCH); __ fdivs(F15_ftos, F0_SCRATCH, F15_ftos); break;
1309 case rem:
1310 __ pop_f(F1_ARG1);
1311 __ fmr(F2_ARG2, F15_ftos);
1312 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem));
1313 __ fmr(F15_ftos, F1_RET);
1314 break;
1315
1316 default: ShouldNotReachHere();
1317 }
1318 }
1319
1320 void TemplateTable::dop2(Operation op) {
1321 transition(dtos, dtos);
1322
1323 switch (op) {
1324 case add: __ pop_d(F0_SCRATCH); __ fadd(F15_ftos, F0_SCRATCH, F15_ftos); break;
1325 case sub: __ pop_d(F0_SCRATCH); __ fsub(F15_ftos, F0_SCRATCH, F15_ftos); break;
1326 case mul: __ pop_d(F0_SCRATCH); __ fmul(F15_ftos, F0_SCRATCH, F15_ftos); break;
1327 case div: __ pop_d(F0_SCRATCH); __ fdiv(F15_ftos, F0_SCRATCH, F15_ftos); break;
1328 case rem:
1329 __ pop_d(F1_ARG1);
1330 __ fmr(F2_ARG2, F15_ftos);
1331 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem));
1332 __ fmr(F15_ftos, F1_RET);
1333 break;
1334
1335 default: ShouldNotReachHere();
1336 }
1337 }
1338
1339 // Negate the value in the TOS cache.
1340 void TemplateTable::ineg() {
1341 transition(itos, itos);
1342
1343 __ neg(R17_tos, R17_tos);
1344 }
1345
1346 // Negate the value in the TOS cache.
1347 void TemplateTable::lneg() {
1348 transition(ltos, ltos);
1349
1350 __ neg(R17_tos, R17_tos);
1351 }
1352
1353 void TemplateTable::fneg() {
1354 transition(ftos, ftos);
1355
1356 __ fneg(F15_ftos, F15_ftos);
1357 }
1358
1359 void TemplateTable::dneg() {
1360 transition(dtos, dtos);
1361
1362 __ fneg(F15_ftos, F15_ftos);
1363 }
1364
1365 // Increments a local variable in place.
1366 void TemplateTable::iinc() {
1367 transition(vtos, vtos);
1368
1369 const Register Rindex = R11_scratch1,
1370 Rincrement = R0,
1371 Rvalue = R12_scratch2;
1372
1373 locals_index(Rindex); // Load locals index from bytecode stream.
1374 __ lbz(Rincrement, 2, R14_bcp); // Load increment from the bytecode stream.
1375 __ extsb(Rincrement, Rincrement);
1376
1377 __ load_local_int(Rvalue, Rindex, Rindex); // Puts address of local into Rindex.
1378
1379 __ add(Rvalue, Rincrement, Rvalue);
1380 __ stw(Rvalue, 0, Rindex);
1381 }
1382
1383 void TemplateTable::wide_iinc() {
1384 transition(vtos, vtos);
1385
1386 Register Rindex = R11_scratch1,
1387 Rlocals_addr = Rindex,
1388 Rincr = R12_scratch2;
1389 locals_index_wide(Rindex);
1390 __ get_2_byte_integer_at_bcp(4, Rincr, InterpreterMacroAssembler::Signed);
1391 __ load_local_int(R17_tos, Rlocals_addr, Rindex);
1392 __ add(R17_tos, Rincr, R17_tos);
1393 __ stw(R17_tos, 0, Rlocals_addr);
1394 }
1395
1396 void TemplateTable::convert() {
1397 // %%%%% Factor this first part accross platforms
1398 #ifdef ASSERT
1399 TosState tos_in = ilgl;
1400 TosState tos_out = ilgl;
1401 switch (bytecode()) {
1402 case Bytecodes::_i2l: // fall through
1403 case Bytecodes::_i2f: // fall through
1404 case Bytecodes::_i2d: // fall through
1405 case Bytecodes::_i2b: // fall through
1406 case Bytecodes::_i2c: // fall through
1407 case Bytecodes::_i2s: tos_in = itos; break;
1408 case Bytecodes::_l2i: // fall through
1409 case Bytecodes::_l2f: // fall through
1410 case Bytecodes::_l2d: tos_in = ltos; break;
1411 case Bytecodes::_f2i: // fall through
1412 case Bytecodes::_f2l: // fall through
1413 case Bytecodes::_f2d: tos_in = ftos; break;
1414 case Bytecodes::_d2i: // fall through
1415 case Bytecodes::_d2l: // fall through
1416 case Bytecodes::_d2f: tos_in = dtos; break;
1417 default : ShouldNotReachHere();
1418 }
1419 switch (bytecode()) {
1420 case Bytecodes::_l2i: // fall through
1421 case Bytecodes::_f2i: // fall through
1422 case Bytecodes::_d2i: // fall through
1423 case Bytecodes::_i2b: // fall through
1424 case Bytecodes::_i2c: // fall through
1425 case Bytecodes::_i2s: tos_out = itos; break;
1426 case Bytecodes::_i2l: // fall through
1427 case Bytecodes::_f2l: // fall through
1428 case Bytecodes::_d2l: tos_out = ltos; break;
1429 case Bytecodes::_i2f: // fall through
1430 case Bytecodes::_l2f: // fall through
1431 case Bytecodes::_d2f: tos_out = ftos; break;
1432 case Bytecodes::_i2d: // fall through
1433 case Bytecodes::_l2d: // fall through
1434 case Bytecodes::_f2d: tos_out = dtos; break;
1435 default : ShouldNotReachHere();
1436 }
1437 transition(tos_in, tos_out);
1438 #endif
1439
1440 // Conversion
1441 Label done;
1442 switch (bytecode()) {
1443 case Bytecodes::_i2l:
1444 __ extsw(R17_tos, R17_tos);
1445 break;
1446
1447 case Bytecodes::_l2i:
1448 // Nothing to do, we'll continue to work with the lower bits.
1449 break;
1450
1451 case Bytecodes::_i2b:
1452 __ extsb(R17_tos, R17_tos);
1453 break;
1454
1455 case Bytecodes::_i2c:
1456 __ rldicl(R17_tos, R17_tos, 0, 64-2*8);
1457 break;
1458
1459 case Bytecodes::_i2s:
1460 __ extsh(R17_tos, R17_tos);
1461 break;
1462
1463 case Bytecodes::_i2d:
1464 __ extsw(R17_tos, R17_tos);
1465 case Bytecodes::_l2d:
1466 __ push_l_pop_d();
1467 __ fcfid(F15_ftos, F15_ftos);
1468 break;
1469
1470 case Bytecodes::_i2f:
1471 __ extsw(R17_tos, R17_tos);
1472 __ push_l_pop_d();
1473 if (VM_Version::has_fcfids()) { // fcfids is >= Power7 only
1474 // Comment: alternatively, load with sign extend could be done by lfiwax.
1475 __ fcfids(F15_ftos, F15_ftos);
1476 } else {
1477 __ fcfid(F15_ftos, F15_ftos);
1478 __ frsp(F15_ftos, F15_ftos);
1479 }
1480 break;
1481
1482 case Bytecodes::_l2f:
1483 if (VM_Version::has_fcfids()) { // fcfids is >= Power7 only
1484 __ push_l_pop_d();
1485 __ fcfids(F15_ftos, F15_ftos);
1486 } else {
1487 // Avoid rounding problem when result should be 0x3f800001: need fixup code before fcfid+frsp.
1488 __ mr(R3_ARG1, R17_tos);
1489 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::l2f));
1490 __ fmr(F15_ftos, F1_RET);
1491 }
1492 break;
1493
1494 case Bytecodes::_f2d:
1495 // empty
1496 break;
1497
1498 case Bytecodes::_d2f:
1499 __ frsp(F15_ftos, F15_ftos);
1500 break;
1501
1502 case Bytecodes::_d2i:
1503 case Bytecodes::_f2i:
1504 __ fcmpu(CCR0, F15_ftos, F15_ftos);
1505 __ li(R17_tos, 0); // 0 in case of NAN
1506 __ bso(CCR0, done);
1507 __ fctiwz(F15_ftos, F15_ftos);
1508 __ push_d_pop_l();
1509 break;
1510
1511 case Bytecodes::_d2l:
1512 case Bytecodes::_f2l:
1513 __ fcmpu(CCR0, F15_ftos, F15_ftos);
1514 __ li(R17_tos, 0); // 0 in case of NAN
1515 __ bso(CCR0, done);
1516 __ fctidz(F15_ftos, F15_ftos);
1517 __ push_d_pop_l();
1518 break;
1519
1520 default: ShouldNotReachHere();
1521 }
1522 __ bind(done);
1523 }
1524
1525 // Long compare
1526 void TemplateTable::lcmp() {
1527 transition(ltos, itos);
1528
1529 const Register Rscratch = R11_scratch1;
1530 __ pop_l(Rscratch); // first operand, deeper in stack
1531
1532 __ cmpd(CCR0, Rscratch, R17_tos); // compare
1533 __ mfcr(R17_tos); // set bit 32..33 as follows: <: 0b10, =: 0b00, >: 0b01
1534 __ srwi(Rscratch, R17_tos, 30);
1535 __ srawi(R17_tos, R17_tos, 31);
1536 __ orr(R17_tos, Rscratch, R17_tos); // set result as follows: <: -1, =: 0, >: 1
1537 }
1538
1539 // fcmpl/fcmpg and dcmpl/dcmpg bytecodes
1540 // unordered_result == -1 => fcmpl or dcmpl
1541 // unordered_result == 1 => fcmpg or dcmpg
1542 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1543 const FloatRegister Rfirst = F0_SCRATCH,
1544 Rsecond = F15_ftos;
1545 const Register Rscratch = R11_scratch1;
1546
1547 if (is_float) {
1548 __ pop_f(Rfirst);
1549 } else {
1550 __ pop_d(Rfirst);
1551 }
1552
1553 Label Lunordered, Ldone;
1554 __ fcmpu(CCR0, Rfirst, Rsecond); // compare
1555 if (unordered_result) {
1556 __ bso(CCR0, Lunordered);
1557 }
1558 __ mfcr(R17_tos); // set bit 32..33 as follows: <: 0b10, =: 0b00, >: 0b01
1559 __ srwi(Rscratch, R17_tos, 30);
1560 __ srawi(R17_tos, R17_tos, 31);
1561 __ orr(R17_tos, Rscratch, R17_tos); // set result as follows: <: -1, =: 0, >: 1
1562 if (unordered_result) {
1563 __ b(Ldone);
1564 __ bind(Lunordered);
1565 __ load_const_optimized(R17_tos, unordered_result);
1566 }
1567 __ bind(Ldone);
1568 }
1569
1570 // Branch_conditional which takes TemplateTable::Condition.
1571 void TemplateTable::branch_conditional(ConditionRegister crx, TemplateTable::Condition cc, Label& L, bool invert) {
1572 bool positive = false;
1573 Assembler::Condition cond = Assembler::equal;
1574 switch (cc) {
1575 case TemplateTable::equal: positive = true ; cond = Assembler::equal ; break;
1576 case TemplateTable::not_equal: positive = false; cond = Assembler::equal ; break;
1577 case TemplateTable::less: positive = true ; cond = Assembler::less ; break;
1578 case TemplateTable::less_equal: positive = false; cond = Assembler::greater; break;
1579 case TemplateTable::greater: positive = true ; cond = Assembler::greater; break;
1580 case TemplateTable::greater_equal: positive = false; cond = Assembler::less ; break;
1581 default: ShouldNotReachHere();
1582 }
1583 int bo = (positive != invert) ? Assembler::bcondCRbiIs1 : Assembler::bcondCRbiIs0;
1584 int bi = Assembler::bi0(crx, cond);
1585 __ bc(bo, bi, L);
1586 }
1587
1588 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1589
1590 // Note: on SPARC, we use InterpreterMacroAssembler::if_cmp also.
1591 __ verify_thread();
1592
1593 const Register Rscratch1 = R11_scratch1,
1594 Rscratch2 = R12_scratch2,
1595 Rscratch3 = R3_ARG1,
1596 R4_counters = R4_ARG2,
1597 bumped_count = R31,
1598 Rdisp = R22_tmp2;
1599
1600 __ profile_taken_branch(Rscratch1, bumped_count);
1601
1602 // Get (wide) offset.
1603 if (is_wide) {
1604 __ get_4_byte_integer_at_bcp(1, Rdisp, InterpreterMacroAssembler::Signed);
1605 } else {
1606 __ get_2_byte_integer_at_bcp(1, Rdisp, InterpreterMacroAssembler::Signed);
1607 }
1608
1609 // --------------------------------------------------------------------------
1610 // Handle all the JSR stuff here, then exit.
1611 // It's much shorter and cleaner than intermingling with the
1612 // non-JSR normal-branch stuff occurring below.
1613 if (is_jsr) {
1614 // Compute return address as bci in Otos_i.
1615 __ ld(Rscratch1, in_bytes(Method::const_offset()), R19_method);
1616 __ addi(Rscratch2, R14_bcp, -in_bytes(ConstMethod::codes_offset()) + (is_wide ? 5 : 3));
1617 __ subf(R17_tos, Rscratch1, Rscratch2);
1618
1619 // Bump bcp to target of JSR.
1620 __ add(R14_bcp, Rdisp, R14_bcp);
1621 // Push returnAddress for "ret" on stack.
1622 __ push_ptr(R17_tos);
1623 // And away we go!
1624 __ dispatch_next(vtos);
1625 return;
1626 }
1627
1628 // --------------------------------------------------------------------------
1629 // Normal (non-jsr) branch handling
1630
1631 const bool increment_invocation_counter_for_backward_branches = UseCompiler && UseLoopCounter;
1632 if (increment_invocation_counter_for_backward_branches) {
1633 //__ unimplemented("branch invocation counter");
1634
1635 Label Lforward;
1636 __ add(R14_bcp, Rdisp, R14_bcp); // Add to bc addr.
1637
1638 // Check branch direction.
1639 __ cmpdi(CCR0, Rdisp, 0);
1640 __ bgt(CCR0, Lforward);
1641
1642 __ get_method_counters(R19_method, R4_counters, Lforward);
1643
1644 if (TieredCompilation) {
1645 Label Lno_mdo, Loverflow;
1646 const int increment = InvocationCounter::count_increment;
1647 const int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
1648 if (ProfileInterpreter) {
1649 Register Rmdo = Rscratch1;
1650
1651 // If no method data exists, go to profile_continue.
1652 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
1653 __ cmpdi(CCR0, Rmdo, 0);
1654 __ beq(CCR0, Lno_mdo);
1655
1656 // Increment backedge counter in the MDO.
1657 const int mdo_bc_offs = in_bytes(MethodData::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
1658 __ lwz(Rscratch2, mdo_bc_offs, Rmdo);
1659 __ load_const_optimized(Rscratch3, mask, R0);
1660 __ addi(Rscratch2, Rscratch2, increment);
1661 __ stw(Rscratch2, mdo_bc_offs, Rmdo);
1662 __ and_(Rscratch3, Rscratch2, Rscratch3);
1663 __ bne(CCR0, Lforward);
1664 __ b(Loverflow);
1665 }
1666
1667 // If there's no MDO, increment counter in method.
1668 const int mo_bc_offs = in_bytes(MethodCounters::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
1669 __ bind(Lno_mdo);
1670 __ lwz(Rscratch2, mo_bc_offs, R4_counters);
1671 __ load_const_optimized(Rscratch3, mask, R0);
1672 __ addi(Rscratch2, Rscratch2, increment);
1673 __ stw(Rscratch2, mo_bc_offs, R19_method);
1674 __ and_(Rscratch3, Rscratch2, Rscratch3);
1675 __ bne(CCR0, Lforward);
1676
1677 __ bind(Loverflow);
1678
1679 // Notify point for loop, pass branch bytecode.
1680 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R14_bcp, true);
1681
1682 // Was an OSR adapter generated?
1683 // O0 = osr nmethod
1684 __ cmpdi(CCR0, R3_RET, 0);
1685 __ beq(CCR0, Lforward);
1686
1687 // Has the nmethod been invalidated already?
1688 __ lwz(R0, nmethod::entry_bci_offset(), R3_RET);
1689 __ cmpwi(CCR0, R0, InvalidOSREntryBci);
1690 __ beq(CCR0, Lforward);
1691
1692 // Migrate the interpreter frame off of the stack.
1693 // We can use all registers because we will not return to interpreter from this point.
1694
1695 // Save nmethod.
1696 const Register osr_nmethod = R31;
1697 __ mr(osr_nmethod, R3_RET);
1698 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R11_scratch1);
1699 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), R16_thread);
1700 __ reset_last_Java_frame();
1701 // OSR buffer is in ARG1.
1702
1703 // Remove the interpreter frame.
1704 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1705
1706 // Jump to the osr code.
1707 __ ld(R11_scratch1, nmethod::osr_entry_point_offset(), osr_nmethod);
1708 __ mtlr(R0);
1709 __ mtctr(R11_scratch1);
1710 __ bctr();
1711
1712 } else {
1713
1714 const Register invoke_ctr = Rscratch1;
1715 // Update Backedge branch separately from invocations.
1716 __ increment_backedge_counter(R4_counters, invoke_ctr, Rscratch2, Rscratch3);
1717
1718 if (ProfileInterpreter) {
1719 __ test_invocation_counter_for_mdp(invoke_ctr, Rscratch2, Lforward);
1720 if (UseOnStackReplacement) {
1721 __ test_backedge_count_for_osr(bumped_count, R14_bcp, Rscratch2);
1722 }
1723 } else {
1724 if (UseOnStackReplacement) {
1725 __ test_backedge_count_for_osr(invoke_ctr, R14_bcp, Rscratch2);
1726 }
1727 }
1728 }
1729
1730 __ bind(Lforward);
1731
1732 } else {
1733 // Bump bytecode pointer by displacement (take the branch).
1734 __ add(R14_bcp, Rdisp, R14_bcp); // Add to bc addr.
1735 }
1736 // Continue with bytecode @ target.
1737 // %%%%% Like Intel, could speed things up by moving bytecode fetch to code above,
1738 // %%%%% and changing dispatch_next to dispatch_only.
1739 __ dispatch_next(vtos);
1740 }
1741
1742 // Helper function for if_cmp* methods below.
1743 // Factored out common compare and branch code.
1744 void TemplateTable::if_cmp_common(Register Rfirst, Register Rsecond, Register Rscratch1, Register Rscratch2, Condition cc, bool is_jint, bool cmp0) {
1745 Label Lnot_taken;
1746 // Note: The condition code we get is the condition under which we
1747 // *fall through*! So we have to inverse the CC here.
1748
1749 if (is_jint) {
1750 if (cmp0) {
1751 __ cmpwi(CCR0, Rfirst, 0);
1752 } else {
1753 __ cmpw(CCR0, Rfirst, Rsecond);
1754 }
1755 } else {
1756 if (cmp0) {
1757 __ cmpdi(CCR0, Rfirst, 0);
1758 } else {
1759 __ cmpd(CCR0, Rfirst, Rsecond);
1760 }
1761 }
1762 branch_conditional(CCR0, cc, Lnot_taken, /*invert*/ true);
1763
1764 // Conition is false => Jump!
1765 branch(false, false);
1766
1767 // Condition is not true => Continue.
1768 __ align(32, 12);
1769 __ bind(Lnot_taken);
1770 __ profile_not_taken_branch(Rscratch1, Rscratch2);
1771 }
1772
1773 // Compare integer values with zero and fall through if CC holds, branch away otherwise.
1774 void TemplateTable::if_0cmp(Condition cc) {
1775 transition(itos, vtos);
1776
1777 if_cmp_common(R17_tos, noreg, R11_scratch1, R12_scratch2, cc, true, true);
1778 }
1779
1780 // Compare integer values and fall through if CC holds, branch away otherwise.
1781 //
1782 // Interface:
1783 // - Rfirst: First operand (older stack value)
1784 // - tos: Second operand (younger stack value)
1785 void TemplateTable::if_icmp(Condition cc) {
1786 transition(itos, vtos);
1787
1788 const Register Rfirst = R0,
1789 Rsecond = R17_tos;
1790
1791 __ pop_i(Rfirst);
1792 if_cmp_common(Rfirst, Rsecond, R11_scratch1, R12_scratch2, cc, true, false);
1793 }
1794
1795 void TemplateTable::if_nullcmp(Condition cc) {
1796 transition(atos, vtos);
1797
1798 if_cmp_common(R17_tos, noreg, R11_scratch1, R12_scratch2, cc, false, true);
1799 }
1800
1801 void TemplateTable::if_acmp(Condition cc) {
1802 transition(atos, vtos);
1803
1804 const Register Rfirst = R0,
1805 Rsecond = R17_tos;
1806
1807 __ pop_ptr(Rfirst);
1808 if_cmp_common(Rfirst, Rsecond, R11_scratch1, R12_scratch2, cc, false, false);
1809 }
1810
1811 void TemplateTable::ret() {
1812 locals_index(R11_scratch1);
1813 __ load_local_ptr(R17_tos, R11_scratch1, R11_scratch1);
1814
1815 __ profile_ret(vtos, R17_tos, R11_scratch1, R12_scratch2);
1816
1817 __ ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method);
1818 __ add(R11_scratch1, R17_tos, R11_scratch1);
1819 __ addi(R14_bcp, R11_scratch1, in_bytes(ConstMethod::codes_offset()));
1820 __ dispatch_next(vtos);
1821 }
1822
1823 void TemplateTable::wide_ret() {
1824 transition(vtos, vtos);
1825
1826 const Register Rindex = R3_ARG1,
1827 Rscratch1 = R11_scratch1,
1828 Rscratch2 = R12_scratch2;
1829
1830 locals_index_wide(Rindex);
1831 __ load_local_ptr(R17_tos, R17_tos, Rindex);
1832 __ profile_ret(vtos, R17_tos, Rscratch1, R12_scratch2);
1833 // Tos now contains the bci, compute the bcp from that.
1834 __ ld(Rscratch1, in_bytes(Method::const_offset()), R19_method);
1835 __ addi(Rscratch2, R17_tos, in_bytes(ConstMethod::codes_offset()));
1836 __ add(R14_bcp, Rscratch1, Rscratch2);
1837 __ dispatch_next(vtos);
1838 }
1839
1840 void TemplateTable::tableswitch() {
1841 transition(itos, vtos);
1842
1843 Label Ldispatch, Ldefault_case;
1844 Register Rlow_byte = R3_ARG1,
1845 Rindex = Rlow_byte,
1846 Rhigh_byte = R4_ARG2,
1847 Rdef_offset_addr = R5_ARG3, // is going to contain address of default offset
1848 Rscratch1 = R11_scratch1,
1849 Rscratch2 = R12_scratch2,
1850 Roffset = R6_ARG4;
1851
1852 // Align bcp.
1853 __ addi(Rdef_offset_addr, R14_bcp, BytesPerInt);
1854 __ clrrdi(Rdef_offset_addr, Rdef_offset_addr, log2_long((jlong)BytesPerInt));
1855
1856 // Load lo & hi.
1857 __ get_u4(Rlow_byte, Rdef_offset_addr, BytesPerInt, InterpreterMacroAssembler::Unsigned);
1858 __ get_u4(Rhigh_byte, Rdef_offset_addr, 2 *BytesPerInt, InterpreterMacroAssembler::Unsigned);
1859
1860 // Check for default case (=index outside [low,high]).
1861 __ cmpw(CCR0, R17_tos, Rlow_byte);
1862 __ cmpw(CCR1, R17_tos, Rhigh_byte);
1863 __ blt(CCR0, Ldefault_case);
1864 __ bgt(CCR1, Ldefault_case);
1865
1866 // Lookup dispatch offset.
1867 __ sub(Rindex, R17_tos, Rlow_byte);
1868 __ extsw(Rindex, Rindex);
1869 __ profile_switch_case(Rindex, Rhigh_byte /* scratch */, Rscratch1, Rscratch2);
1870 __ sldi(Rindex, Rindex, LogBytesPerInt);
1871 __ addi(Rindex, Rindex, 3 * BytesPerInt);
1872 #if defined(VM_LITTLE_ENDIAN)
1873 __ lwbrx(Roffset, Rdef_offset_addr, Rindex);
1874 __ extsw(Roffset, Roffset);
1875 #else
1876 __ lwax(Roffset, Rdef_offset_addr, Rindex);
1877 #endif
1878 __ b(Ldispatch);
1879
1880 __ bind(Ldefault_case);
1881 __ profile_switch_default(Rhigh_byte, Rscratch1);
1882 __ get_u4(Roffset, Rdef_offset_addr, 0, InterpreterMacroAssembler::Signed);
1883
1884 __ bind(Ldispatch);
1885
1886 __ add(R14_bcp, Roffset, R14_bcp);
1887 __ dispatch_next(vtos);
1888 }
1889
1890 void TemplateTable::lookupswitch() {
1891 transition(itos, itos);
1892 __ stop("lookupswitch bytecode should have been rewritten");
1893 }
1894
1895 // Table switch using linear search through cases.
1896 // Bytecode stream format:
1897 // Bytecode (1) | 4-byte padding | default offset (4) | count (4) | value/offset pair1 (8) | value/offset pair2 (8) | ...
1898 // Note: Everything is big-endian format here.
1899 void TemplateTable::fast_linearswitch() {
1900 transition(itos, vtos);
1901
1902 Label Lloop_entry, Lsearch_loop, Lcontinue_execution, Ldefault_case;
1903 Register Rcount = R3_ARG1,
1904 Rcurrent_pair = R4_ARG2,
1905 Rdef_offset_addr = R5_ARG3, // Is going to contain address of default offset.
1906 Roffset = R31, // Might need to survive C call.
1907 Rvalue = R12_scratch2,
1908 Rscratch = R11_scratch1,
1909 Rcmp_value = R17_tos;
1910
1911 // Align bcp.
1912 __ addi(Rdef_offset_addr, R14_bcp, BytesPerInt);
1913 __ clrrdi(Rdef_offset_addr, Rdef_offset_addr, log2_long((jlong)BytesPerInt));
1914
1915 // Setup loop counter and limit.
1916 __ get_u4(Rcount, Rdef_offset_addr, BytesPerInt, InterpreterMacroAssembler::Unsigned);
1917 __ addi(Rcurrent_pair, Rdef_offset_addr, 2 * BytesPerInt); // Rcurrent_pair now points to first pair.
1918
1919 __ mtctr(Rcount);
1920 __ cmpwi(CCR0, Rcount, 0);
1921 __ bne(CCR0, Lloop_entry);
1922
1923 // Default case
1924 __ bind(Ldefault_case);
1925 __ get_u4(Roffset, Rdef_offset_addr, 0, InterpreterMacroAssembler::Signed);
1926 if (ProfileInterpreter) {
1927 __ profile_switch_default(Rdef_offset_addr, Rcount/* scratch */);
1928 }
1929 __ b(Lcontinue_execution);
1930
1931 // Next iteration
1932 __ bind(Lsearch_loop);
1933 __ bdz(Ldefault_case);
1934 __ addi(Rcurrent_pair, Rcurrent_pair, 2 * BytesPerInt);
1935 __ bind(Lloop_entry);
1936 __ get_u4(Rvalue, Rcurrent_pair, 0, InterpreterMacroAssembler::Unsigned);
1937 __ cmpw(CCR0, Rvalue, Rcmp_value);
1938 __ bne(CCR0, Lsearch_loop);
1939
1940 // Found, load offset.
1941 __ get_u4(Roffset, Rcurrent_pair, BytesPerInt, InterpreterMacroAssembler::Signed);
1942 // Calculate case index and profile
1943 __ mfctr(Rcurrent_pair);
1944 if (ProfileInterpreter) {
1945 __ sub(Rcurrent_pair, Rcount, Rcurrent_pair);
1946 __ profile_switch_case(Rcurrent_pair, Rcount /*scratch*/, Rdef_offset_addr/*scratch*/, Rscratch);
1947 }
1948
1949 __ bind(Lcontinue_execution);
1950 __ add(R14_bcp, Roffset, R14_bcp);
1951 __ dispatch_next(vtos);
1952 }
1953
1954 // Table switch using binary search (value/offset pairs are ordered).
1955 // Bytecode stream format:
1956 // Bytecode (1) | 4-byte padding | default offset (4) | count (4) | value/offset pair1 (8) | value/offset pair2 (8) | ...
1957 // Note: Everything is big-endian format here. So on little endian machines, we have to revers offset and count and cmp value.
1958 void TemplateTable::fast_binaryswitch() {
1959
1960 transition(itos, vtos);
1961 // Implementation using the following core algorithm: (copied from Intel)
1962 //
1963 // int binary_search(int key, LookupswitchPair* array, int n) {
1964 // // Binary search according to "Methodik des Programmierens" by
1965 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1966 // int i = 0;
1967 // int j = n;
1968 // while (i+1 < j) {
1969 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1970 // // with Q: for all i: 0 <= i < n: key < a[i]
1971 // // where a stands for the array and assuming that the (inexisting)
1972 // // element a[n] is infinitely big.
1973 // int h = (i + j) >> 1;
1974 // // i < h < j
1975 // if (key < array[h].fast_match()) {
1976 // j = h;
1977 // } else {
1978 // i = h;
1979 // }
1980 // }
1981 // // R: a[i] <= key < a[i+1] or Q
1982 // // (i.e., if key is within array, i is the correct index)
1983 // return i;
1984 // }
1985
1986 // register allocation
1987 const Register Rkey = R17_tos; // already set (tosca)
1988 const Register Rarray = R3_ARG1;
1989 const Register Ri = R4_ARG2;
1990 const Register Rj = R5_ARG3;
1991 const Register Rh = R6_ARG4;
1992 const Register Rscratch = R11_scratch1;
1993
1994 const int log_entry_size = 3;
1995 const int entry_size = 1 << log_entry_size;
1996
1997 Label found;
1998
1999 // Find Array start,
2000 __ addi(Rarray, R14_bcp, 3 * BytesPerInt);
2001 __ clrrdi(Rarray, Rarray, log2_long((jlong)BytesPerInt));
2002
2003 // initialize i & j
2004 __ li(Ri,0);
2005 __ get_u4(Rj, Rarray, -BytesPerInt, InterpreterMacroAssembler::Unsigned);
2006
2007 // and start.
2008 Label entry;
2009 __ b(entry);
2010
2011 // binary search loop
2012 { Label loop;
2013 __ bind(loop);
2014 // int h = (i + j) >> 1;
2015 __ srdi(Rh, Rh, 1);
2016 // if (key < array[h].fast_match()) {
2017 // j = h;
2018 // } else {
2019 // i = h;
2020 // }
2021 __ sldi(Rscratch, Rh, log_entry_size);
2022 #if defined(VM_LITTLE_ENDIAN)
2023 __ lwbrx(Rscratch, Rscratch, Rarray);
2024 #else
2025 __ lwzx(Rscratch, Rscratch, Rarray);
2026 #endif
2027
2028 // if (key < current value)
2029 // Rh = Rj
2030 // else
2031 // Rh = Ri
2032 Label Lgreater;
2033 __ cmpw(CCR0, Rkey, Rscratch);
2034 __ bge(CCR0, Lgreater);
2035 __ mr(Rj, Rh);
2036 __ b(entry);
2037 __ bind(Lgreater);
2038 __ mr(Ri, Rh);
2039
2040 // while (i+1 < j)
2041 __ bind(entry);
2042 __ addi(Rscratch, Ri, 1);
2043 __ cmpw(CCR0, Rscratch, Rj);
2044 __ add(Rh, Ri, Rj); // start h = i + j >> 1;
2045
2046 __ blt(CCR0, loop);
2047 }
2048
2049 // End of binary search, result index is i (must check again!).
2050 Label default_case;
2051 Label continue_execution;
2052 if (ProfileInterpreter) {
2053 __ mr(Rh, Ri); // Save index in i for profiling.
2054 }
2055 // Ri = value offset
2056 __ sldi(Ri, Ri, log_entry_size);
2057 __ add(Ri, Ri, Rarray);
2058 __ get_u4(Rscratch, Ri, 0, InterpreterMacroAssembler::Unsigned);
2059
2060 Label not_found;
2061 // Ri = offset offset
2062 __ cmpw(CCR0, Rkey, Rscratch);
2063 __ beq(CCR0, not_found);
2064 // entry not found -> j = default offset
2065 __ get_u4(Rj, Rarray, -2 * BytesPerInt, InterpreterMacroAssembler::Unsigned);
2066 __ b(default_case);
2067
2068 __ bind(not_found);
2069 // entry found -> j = offset
2070 __ profile_switch_case(Rh, Rj, Rscratch, Rkey);
2071 __ get_u4(Rj, Ri, BytesPerInt, InterpreterMacroAssembler::Unsigned);
2072
2073 if (ProfileInterpreter) {
2074 __ b(continue_execution);
2075 }
2076
2077 __ bind(default_case); // fall through (if not profiling)
2078 __ profile_switch_default(Ri, Rscratch);
2079
2080 __ bind(continue_execution);
2081
2082 __ extsw(Rj, Rj);
2083 __ add(R14_bcp, Rj, R14_bcp);
2084 __ dispatch_next(vtos);
2085 }
2086
2087 void TemplateTable::_return(TosState state) {
2088 transition(state, state);
2089 assert(_desc->calls_vm(),
2090 "inconsistent calls_vm information"); // call in remove_activation
2091
2092 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2093
2094 Register Rscratch = R11_scratch1,
2095 Rklass = R12_scratch2,
2096 Rklass_flags = Rklass;
2097 Label Lskip_register_finalizer;
2098
2099 // Check if the method has the FINALIZER flag set and call into the VM to finalize in this case.
2100 assert(state == vtos, "only valid state");
2101 __ ld(R17_tos, 0, R18_locals);
2102
2103 // Load klass of this obj.
2104 __ load_klass(Rklass, R17_tos);
2105 __ lwz(Rklass_flags, in_bytes(Klass::access_flags_offset()), Rklass);
2106 __ testbitdi(CCR0, R0, Rklass_flags, exact_log2(JVM_ACC_HAS_FINALIZER));
2107 __ bfalse(CCR0, Lskip_register_finalizer);
2108
2109 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), R17_tos /* obj */);
2110
2111 __ align(32, 12);
2112 __ bind(Lskip_register_finalizer);
2113 }
2114
2115 // Move the result value into the correct register and remove memory stack frame.
2116 __ remove_activation(state, /* throw_monitor_exception */ true);
2117 // Restoration of lr done by remove_activation.
2118 switch (state) {
2119 // Narrow result if state is itos but result type is smaller.
2120 // Need to narrow in the return bytecode rather than in generate_return_entry
2121 // since compiled code callers expect the result to already be narrowed.
2122 case itos: __ narrow(R17_tos); /* fall through */
2123 case ltos:
2124 case btos:
2125 case ztos:
2126 case ctos:
2127 case stos:
2128 case atos: __ mr(R3_RET, R17_tos); break;
2129 case ftos:
2130 case dtos: __ fmr(F1_RET, F15_ftos); break;
2131 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
2132 // to get visible before the reference to the object gets stored anywhere.
2133 __ membar(Assembler::StoreStore); break;
2134 default : ShouldNotReachHere();
2135 }
2136 __ blr();
2137 }
2138
2139 // ============================================================================
2140 // Constant pool cache access
2141 //
2142 // Memory ordering:
2143 //
2144 // Like done in C++ interpreter, we load the fields
2145 // - _indices
2146 // - _f12_oop
2147 // acquired, because these are asked if the cache is already resolved. We don't
2148 // want to float loads above this check.
2149 // See also comments in ConstantPoolCacheEntry::bytecode_1(),
2150 // ConstantPoolCacheEntry::bytecode_2() and ConstantPoolCacheEntry::f1();
2151
2152 // Call into the VM if call site is not yet resolved
2153 //
2154 // Input regs:
2155 // - None, all passed regs are outputs.
2156 //
2157 // Returns:
2158 // - Rcache: The const pool cache entry that contains the resolved result.
2159 // - Rresult: Either noreg or output for f1/f2.
2160 //
2161 // Kills:
2162 // - Rscratch
2163 void TemplateTable::resolve_cache_and_index(int byte_no, Register Rcache, Register Rscratch, size_t index_size) {
2164
2165 __ get_cache_and_index_at_bcp(Rcache, 1, index_size);
2166 Label Lresolved, Ldone;
2167
2168 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2169 // We are resolved if the indices offset contains the current bytecode.
2170 #if defined(VM_LITTLE_ENDIAN)
2171 __ lbz(Rscratch, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + byte_no + 1, Rcache);
2172 #else
2173 __ lbz(Rscratch, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 7 - (byte_no + 1), Rcache);
2174 #endif
2175 // Acquire by cmp-br-isync (see below).
2176 __ cmpdi(CCR0, Rscratch, (int)bytecode());
2177 __ beq(CCR0, Lresolved);
2178
2179 address entry = NULL;
2180 switch (bytecode()) {
2181 case Bytecodes::_getstatic : // fall through
2182 case Bytecodes::_putstatic : // fall through
2183 case Bytecodes::_getfield : // fall through
2184 case Bytecodes::_putfield : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break;
2185 case Bytecodes::_invokevirtual : // fall through
2186 case Bytecodes::_invokespecial : // fall through
2187 case Bytecodes::_invokestatic : // fall through
2188 case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke); break;
2189 case Bytecodes::_invokehandle : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokehandle); break;
2190 case Bytecodes::_invokedynamic : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic); break;
2191 default : ShouldNotReachHere(); break;
2192 }
2193 __ li(R4_ARG2, (int)bytecode());
2194 __ call_VM(noreg, entry, R4_ARG2, true);
2195
2196 // Update registers with resolved info.
2197 __ get_cache_and_index_at_bcp(Rcache, 1, index_size);
2198 __ b(Ldone);
2199
2200 __ bind(Lresolved);
2201 __ isync(); // Order load wrt. succeeding loads.
2202 __ bind(Ldone);
2203 }
2204
2205 // Load the constant pool cache entry at field accesses into registers.
2206 // The Rcache and Rindex registers must be set before call.
2207 // Input:
2208 // - Rcache, Rindex
2209 // Output:
2210 // - Robj, Roffset, Rflags
2211 void TemplateTable::load_field_cp_cache_entry(Register Robj,
2212 Register Rcache,
2213 Register Rindex /* unused on PPC64 */,
2214 Register Roffset,
2215 Register Rflags,
2216 bool is_static = false) {
2217 assert_different_registers(Rcache, Rflags, Roffset);
2218 // assert(Rindex == noreg, "parameter not used on PPC64");
2219
2220 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2221 __ ld(Rflags, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::flags_offset()), Rcache);
2222 __ ld(Roffset, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::f2_offset()), Rcache);
2223 if (is_static) {
2224 __ ld(Robj, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::f1_offset()), Rcache);
2225 __ ld(Robj, in_bytes(Klass::java_mirror_offset()), Robj);
2226 // Acquire not needed here. Following access has an address dependency on this value.
2227 }
2228 }
2229
2230 // Load the constant pool cache entry at invokes into registers.
2231 // Resolve if necessary.
2232
2233 // Input Registers:
2234 // - None, bcp is used, though
2235 //
2236 // Return registers:
2237 // - Rmethod (f1 field or f2 if invokevirtual)
2238 // - Ritable_index (f2 field)
2239 // - Rflags (flags field)
2240 //
2241 // Kills:
2242 // - R21
2243 //
2244 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2245 Register Rmethod,
2246 Register Ritable_index,
2247 Register Rflags,
2248 bool is_invokevirtual,
2249 bool is_invokevfinal,
2250 bool is_invokedynamic) {
2251
2252 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2253 // Determine constant pool cache field offsets.
2254 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2255 const int method_offset = in_bytes(cp_base_offset + (is_invokevirtual ? ConstantPoolCacheEntry::f2_offset() : ConstantPoolCacheEntry::f1_offset()));
2256 const int flags_offset = in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset());
2257 // Access constant pool cache fields.
2258 const int index_offset = in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset());
2259
2260 Register Rcache = R21_tmp1; // Note: same register as R21_sender_SP.
2261
2262 if (is_invokevfinal) {
2263 assert(Ritable_index == noreg, "register not used");
2264 // Already resolved.
2265 __ get_cache_and_index_at_bcp(Rcache, 1);
2266 } else {
2267 resolve_cache_and_index(byte_no, Rcache, R0, is_invokedynamic ? sizeof(u4) : sizeof(u2));
2268 }
2269
2270 __ ld(Rmethod, method_offset, Rcache);
2271 __ ld(Rflags, flags_offset, Rcache);
2272
2273 if (Ritable_index != noreg) {
2274 __ ld(Ritable_index, index_offset, Rcache);
2275 }
2276 }
2277
2278 // ============================================================================
2279 // Field access
2280
2281 // Volatile variables demand their effects be made known to all CPU's
2282 // in order. Store buffers on most chips allow reads & writes to
2283 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2284 // without some kind of memory barrier (i.e., it's not sufficient that
2285 // the interpreter does not reorder volatile references, the hardware
2286 // also must not reorder them).
2287 //
2288 // According to the new Java Memory Model (JMM):
2289 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2290 // writes act as aquire & release, so:
2291 // (2) A read cannot let unrelated NON-volatile memory refs that
2292 // happen after the read float up to before the read. It's OK for
2293 // non-volatile memory refs that happen before the volatile read to
2294 // float down below it.
2295 // (3) Similar a volatile write cannot let unrelated NON-volatile
2296 // memory refs that happen BEFORE the write float down to after the
2297 // write. It's OK for non-volatile memory refs that happen after the
2298 // volatile write to float up before it.
2299 //
2300 // We only put in barriers around volatile refs (they are expensive),
2301 // not _between_ memory refs (that would require us to track the
2302 // flavor of the previous memory refs). Requirements (2) and (3)
2303 // require some barriers before volatile stores and after volatile
2304 // loads. These nearly cover requirement (1) but miss the
2305 // volatile-store-volatile-load case. This final case is placed after
2306 // volatile-stores although it could just as well go before
2307 // volatile-loads.
2308
2309 // The registers cache and index expected to be set before call.
2310 // Correct values of the cache and index registers are preserved.
2311 // Kills:
2312 // Rcache (if has_tos)
2313 // Rscratch
2314 void TemplateTable::jvmti_post_field_access(Register Rcache, Register Rscratch, bool is_static, bool has_tos) {
2315
2316 assert_different_registers(Rcache, Rscratch);
2317
2318 if (JvmtiExport::can_post_field_access()) {
2319 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2320 Label Lno_field_access_post;
2321
2322 // Check if post field access in enabled.
2323 int offs = __ load_const_optimized(Rscratch, JvmtiExport::get_field_access_count_addr(), R0, true);
2324 __ lwz(Rscratch, offs, Rscratch);
2325
2326 __ cmpwi(CCR0, Rscratch, 0);
2327 __ beq(CCR0, Lno_field_access_post);
2328
2329 // Post access enabled - do it!
2330 __ addi(Rcache, Rcache, in_bytes(cp_base_offset));
2331 if (is_static) {
2332 __ li(R17_tos, 0);
2333 } else {
2334 if (has_tos) {
2335 // The fast bytecode versions have obj ptr in register.
2336 // Thus, save object pointer before call_VM() clobbers it
2337 // put object on tos where GC wants it.
2338 __ push_ptr(R17_tos);
2339 } else {
2340 // Load top of stack (do not pop the value off the stack).
2341 __ ld(R17_tos, Interpreter::expr_offset_in_bytes(0), R15_esp);
2342 }
2343 __ verify_oop(R17_tos);
2344 }
2345 // tos: object pointer or NULL if static
2346 // cache: cache entry pointer
2347 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), R17_tos, Rcache);
2348 if (!is_static && has_tos) {
2349 // Restore object pointer.
2350 __ pop_ptr(R17_tos);
2351 __ verify_oop(R17_tos);
2352 } else {
2353 // Cache is still needed to get class or obj.
2354 __ get_cache_and_index_at_bcp(Rcache, 1);
2355 }
2356
2357 __ align(32, 12);
2358 __ bind(Lno_field_access_post);
2359 }
2360 }
2361
2362 // kills R11_scratch1
2363 void TemplateTable::pop_and_check_object(Register Roop) {
2364 Register Rtmp = R11_scratch1;
2365
2366 assert_different_registers(Rtmp, Roop);
2367 __ pop_ptr(Roop);
2368 // For field access must check obj.
2369 __ null_check_throw(Roop, -1, Rtmp);
2370 __ verify_oop(Roop);
2371 }
2372
2373 // PPC64: implement volatile loads as fence-store-acquire.
2374 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2375 transition(vtos, vtos);
2376
2377 Label Lacquire, Lisync;
2378
2379 const Register Rcache = R3_ARG1,
2380 Rclass_or_obj = R22_tmp2,
2381 Roffset = R23_tmp3,
2382 Rflags = R31,
2383 Rbtable = R5_ARG3,
2384 Rbc = R6_ARG4,
2385 Rscratch = R12_scratch2;
2386
2387 static address field_branch_table[number_of_states],
2388 static_branch_table[number_of_states];
2389
2390 address* branch_table = is_static ? static_branch_table : field_branch_table;
2391
2392 // Get field offset.
2393 resolve_cache_and_index(byte_no, Rcache, Rscratch, sizeof(u2));
2394
2395 // JVMTI support
2396 jvmti_post_field_access(Rcache, Rscratch, is_static, false);
2397
2398 // Load after possible GC.
2399 load_field_cp_cache_entry(Rclass_or_obj, Rcache, noreg, Roffset, Rflags, is_static);
2400
2401 // Load pointer to branch table.
2402 __ load_const_optimized(Rbtable, (address)branch_table, Rscratch);
2403
2404 // Get volatile flag.
2405 __ rldicl(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2406 // Note: sync is needed before volatile load on PPC64.
2407
2408 // Check field type.
2409 __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2410
2411 #ifdef ASSERT
2412 Label LFlagInvalid;
2413 __ cmpldi(CCR0, Rflags, number_of_states);
2414 __ bge(CCR0, LFlagInvalid);
2415 #endif
2416
2417 // Load from branch table and dispatch (volatile case: one instruction ahead).
2418 __ sldi(Rflags, Rflags, LogBytesPerWord);
2419 __ cmpwi(CCR6, Rscratch, 1); // Volatile?
2420 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2421 __ sldi(Rscratch, Rscratch, exact_log2(BytesPerInstWord)); // Volatile ? size of 1 instruction : 0.
2422 }
2423 __ ldx(Rbtable, Rbtable, Rflags);
2424
2425 // Get the obj from stack.
2426 if (!is_static) {
2427 pop_and_check_object(Rclass_or_obj); // Kills R11_scratch1.
2428 } else {
2429 __ verify_oop(Rclass_or_obj);
2430 }
2431
2432 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2433 __ subf(Rbtable, Rscratch, Rbtable); // Point to volatile/non-volatile entry point.
2434 }
2435 __ mtctr(Rbtable);
2436 __ bctr();
2437
2438 #ifdef ASSERT
2439 __ bind(LFlagInvalid);
2440 __ stop("got invalid flag", 0x654);
2441
2442 // __ bind(Lvtos);
2443 address pc_before_fence = __ pc();
2444 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2445 assert(__ pc() - pc_before_fence == (ptrdiff_t)BytesPerInstWord, "must be single instruction");
2446 assert(branch_table[vtos] == 0, "can't compute twice");
2447 branch_table[vtos] = __ pc(); // non-volatile_entry point
2448 __ stop("vtos unexpected", 0x655);
2449 #endif
2450
2451 __ align(32, 28, 28); // Align load.
2452 // __ bind(Ldtos);
2453 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2454 assert(branch_table[dtos] == 0, "can't compute twice");
2455 branch_table[dtos] = __ pc(); // non-volatile_entry point
2456 __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
2457 __ push(dtos);
2458 if (!is_static) patch_bytecode(Bytecodes::_fast_dgetfield, Rbc, Rscratch);
2459 {
2460 Label acquire_double;
2461 __ beq(CCR6, acquire_double); // Volatile?
2462 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2463
2464 __ bind(acquire_double);
2465 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
2466 __ beq_predict_taken(CCR0, Lisync);
2467 __ b(Lisync); // In case of NAN.
2468 }
2469
2470 __ align(32, 28, 28); // Align load.
2471 // __ bind(Lftos);
2472 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2473 assert(branch_table[ftos] == 0, "can't compute twice");
2474 branch_table[ftos] = __ pc(); // non-volatile_entry point
2475 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
2476 __ push(ftos);
2477 if (!is_static) { patch_bytecode(Bytecodes::_fast_fgetfield, Rbc, Rscratch); }
2478 {
2479 Label acquire_float;
2480 __ beq(CCR6, acquire_float); // Volatile?
2481 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2482
2483 __ bind(acquire_float);
2484 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
2485 __ beq_predict_taken(CCR0, Lisync);
2486 __ b(Lisync); // In case of NAN.
2487 }
2488
2489 __ align(32, 28, 28); // Align load.
2490 // __ bind(Litos);
2491 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2492 assert(branch_table[itos] == 0, "can't compute twice");
2493 branch_table[itos] = __ pc(); // non-volatile_entry point
2494 __ lwax(R17_tos, Rclass_or_obj, Roffset);
2495 __ push(itos);
2496 if (!is_static) patch_bytecode(Bytecodes::_fast_igetfield, Rbc, Rscratch);
2497 __ beq(CCR6, Lacquire); // Volatile?
2498 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2499
2500 __ align(32, 28, 28); // Align load.
2501 // __ bind(Lltos);
2502 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2503 assert(branch_table[ltos] == 0, "can't compute twice");
2504 branch_table[ltos] = __ pc(); // non-volatile_entry point
2505 __ ldx(R17_tos, Rclass_or_obj, Roffset);
2506 __ push(ltos);
2507 if (!is_static) patch_bytecode(Bytecodes::_fast_lgetfield, Rbc, Rscratch);
2508 __ beq(CCR6, Lacquire); // Volatile?
2509 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2510
2511 __ align(32, 28, 28); // Align load.
2512 // __ bind(Lbtos);
2513 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2514 assert(branch_table[btos] == 0, "can't compute twice");
2515 branch_table[btos] = __ pc(); // non-volatile_entry point
2516 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
2517 __ extsb(R17_tos, R17_tos);
2518 __ push(btos);
2519 if (!is_static) patch_bytecode(Bytecodes::_fast_bgetfield, Rbc, Rscratch);
2520 __ beq(CCR6, Lacquire); // Volatile?
2521 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2522
2523 __ align(32, 28, 28); // Align load.
2524 // __ bind(Lztos); (same code as btos)
2525 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2526 assert(branch_table[ztos] == 0, "can't compute twice");
2527 branch_table[ztos] = __ pc(); // non-volatile_entry point
2528 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
2529 __ extsb(R17_tos, R17_tos);
2530 __ push(ztos);
2531 if (!is_static) {
2532 // use btos rewriting, no truncating to t/f bit is needed for getfield.
2533 patch_bytecode(Bytecodes::_fast_bgetfield, Rbc, Rscratch);
2534 }
2535 __ beq(CCR6, Lacquire); // Volatile?
2536 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2537
2538 __ align(32, 28, 28); // Align load.
2539 // __ bind(Lctos);
2540 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2541 assert(branch_table[ctos] == 0, "can't compute twice");
2542 branch_table[ctos] = __ pc(); // non-volatile_entry point
2543 __ lhzx(R17_tos, Rclass_or_obj, Roffset);
2544 __ push(ctos);
2545 if (!is_static) patch_bytecode(Bytecodes::_fast_cgetfield, Rbc, Rscratch);
2546 __ beq(CCR6, Lacquire); // Volatile?
2547 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2548
2549 __ align(32, 28, 28); // Align load.
2550 // __ bind(Lstos);
2551 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2552 assert(branch_table[stos] == 0, "can't compute twice");
2553 branch_table[stos] = __ pc(); // non-volatile_entry point
2554 __ lhax(R17_tos, Rclass_or_obj, Roffset);
2555 __ push(stos);
2556 if (!is_static) patch_bytecode(Bytecodes::_fast_sgetfield, Rbc, Rscratch);
2557 __ beq(CCR6, Lacquire); // Volatile?
2558 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2559
2560 __ align(32, 28, 28); // Align load.
2561 // __ bind(Latos);
2562 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2563 assert(branch_table[atos] == 0, "can't compute twice");
2564 branch_table[atos] = __ pc(); // non-volatile_entry point
2565 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
2566 __ verify_oop(R17_tos);
2567 __ push(atos);
2568 //__ dcbt(R17_tos); // prefetch
2569 if (!is_static) patch_bytecode(Bytecodes::_fast_agetfield, Rbc, Rscratch);
2570 __ beq(CCR6, Lacquire); // Volatile?
2571 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2572
2573 __ align(32, 12);
2574 __ bind(Lacquire);
2575 __ twi_0(R17_tos);
2576 __ bind(Lisync);
2577 __ isync(); // acquire
2578
2579 #ifdef ASSERT
2580 for (int i = 0; i<number_of_states; ++i) {
2581 assert(branch_table[i], "get initialization");
2582 //tty->print_cr("get: %s_branch_table[%d] = 0x%llx (opcode 0x%llx)",
2583 // is_static ? "static" : "field", i, branch_table[i], *((unsigned int*)branch_table[i]));
2584 }
2585 #endif
2586 }
2587
2588 void TemplateTable::getfield(int byte_no) {
2589 getfield_or_static(byte_no, false);
2590 }
2591
2592 void TemplateTable::getstatic(int byte_no) {
2593 getfield_or_static(byte_no, true);
2594 }
2595
2596 // The registers cache and index expected to be set before call.
2597 // The function may destroy various registers, just not the cache and index registers.
2598 void TemplateTable::jvmti_post_field_mod(Register Rcache, Register Rscratch, bool is_static) {
2599
2600 assert_different_registers(Rcache, Rscratch, R6_ARG4);
2601
2602 if (JvmtiExport::can_post_field_modification()) {
2603 Label Lno_field_mod_post;
2604
2605 // Check if post field access in enabled.
2606 int offs = __ load_const_optimized(Rscratch, JvmtiExport::get_field_modification_count_addr(), R0, true);
2607 __ lwz(Rscratch, offs, Rscratch);
2608
2609 __ cmpwi(CCR0, Rscratch, 0);
2610 __ beq(CCR0, Lno_field_mod_post);
2611
2612 // Do the post
2613 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2614 const Register Robj = Rscratch;
2615
2616 __ addi(Rcache, Rcache, in_bytes(cp_base_offset));
2617 if (is_static) {
2618 // Life is simple. Null out the object pointer.
2619 __ li(Robj, 0);
2620 } else {
2621 // In case of the fast versions, value lives in registers => put it back on tos.
2622 int offs = Interpreter::expr_offset_in_bytes(0);
2623 Register base = R15_esp;
2624 switch(bytecode()) {
2625 case Bytecodes::_fast_aputfield: __ push_ptr(); offs+= Interpreter::stackElementSize; break;
2626 case Bytecodes::_fast_iputfield: // Fall through
2627 case Bytecodes::_fast_bputfield: // Fall through
2628 case Bytecodes::_fast_zputfield: // Fall through
2629 case Bytecodes::_fast_cputfield: // Fall through
2630 case Bytecodes::_fast_sputfield: __ push_i(); offs+= Interpreter::stackElementSize; break;
2631 case Bytecodes::_fast_lputfield: __ push_l(); offs+=2*Interpreter::stackElementSize; break;
2632 case Bytecodes::_fast_fputfield: __ push_f(); offs+= Interpreter::stackElementSize; break;
2633 case Bytecodes::_fast_dputfield: __ push_d(); offs+=2*Interpreter::stackElementSize; break;
2634 default: {
2635 offs = 0;
2636 base = Robj;
2637 const Register Rflags = Robj;
2638 Label is_one_slot;
2639 // Life is harder. The stack holds the value on top, followed by the
2640 // object. We don't know the size of the value, though; it could be
2641 // one or two words depending on its type. As a result, we must find
2642 // the type to determine where the object is.
2643 __ ld(Rflags, in_bytes(ConstantPoolCacheEntry::flags_offset()), Rcache); // Big Endian
2644 __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2645
2646 __ cmpwi(CCR0, Rflags, ltos);
2647 __ cmpwi(CCR1, Rflags, dtos);
2648 __ addi(base, R15_esp, Interpreter::expr_offset_in_bytes(1));
2649 __ crnor(/*CR0 eq*/2, /*CR1 eq*/4+2, /*CR0 eq*/2);
2650 __ beq(CCR0, is_one_slot);
2651 __ addi(base, R15_esp, Interpreter::expr_offset_in_bytes(2));
2652 __ bind(is_one_slot);
2653 break;
2654 }
2655 }
2656 __ ld(Robj, offs, base);
2657 __ verify_oop(Robj);
2658 }
2659
2660 __ addi(R6_ARG4, R15_esp, Interpreter::expr_offset_in_bytes(0));
2661 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), Robj, Rcache, R6_ARG4);
2662 __ get_cache_and_index_at_bcp(Rcache, 1);
2663
2664 // In case of the fast versions, value lives in registers => put it back on tos.
2665 switch(bytecode()) {
2666 case Bytecodes::_fast_aputfield: __ pop_ptr(); break;
2667 case Bytecodes::_fast_iputfield: // Fall through
2668 case Bytecodes::_fast_bputfield: // Fall through
2669 case Bytecodes::_fast_zputfield: // Fall through
2670 case Bytecodes::_fast_cputfield: // Fall through
2671 case Bytecodes::_fast_sputfield: __ pop_i(); break;
2672 case Bytecodes::_fast_lputfield: __ pop_l(); break;
2673 case Bytecodes::_fast_fputfield: __ pop_f(); break;
2674 case Bytecodes::_fast_dputfield: __ pop_d(); break;
2675 default: break; // Nothin' to do.
2676 }
2677
2678 __ align(32, 12);
2679 __ bind(Lno_field_mod_post);
2680 }
2681 }
2682
2683 // PPC64: implement volatile stores as release-store (return bytecode contains an additional release).
2684 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2685 Label Lvolatile;
2686
2687 const Register Rcache = R5_ARG3, // Do not use ARG1/2 (causes trouble in jvmti_post_field_mod).
2688 Rclass_or_obj = R31, // Needs to survive C call.
2689 Roffset = R22_tmp2, // Needs to survive C call.
2690 Rflags = R3_ARG1,
2691 Rbtable = R4_ARG2,
2692 Rscratch = R11_scratch1,
2693 Rscratch2 = R12_scratch2,
2694 Rscratch3 = R6_ARG4,
2695 Rbc = Rscratch3;
2696 const ConditionRegister CR_is_vol = CCR2; // Non-volatile condition register (survives runtime call in do_oop_store).
2697
2698 static address field_branch_table[number_of_states],
2699 static_branch_table[number_of_states];
2700
2701 address* branch_table = is_static ? static_branch_table : field_branch_table;
2702
2703 // Stack (grows up):
2704 // value
2705 // obj
2706
2707 // Load the field offset.
2708 resolve_cache_and_index(byte_no, Rcache, Rscratch, sizeof(u2));
2709 jvmti_post_field_mod(Rcache, Rscratch, is_static);
2710 load_field_cp_cache_entry(Rclass_or_obj, Rcache, noreg, Roffset, Rflags, is_static);
2711
2712 // Load pointer to branch table.
2713 __ load_const_optimized(Rbtable, (address)branch_table, Rscratch);
2714
2715 // Get volatile flag.
2716 __ rldicl(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2717
2718 // Check the field type.
2719 __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2720
2721 #ifdef ASSERT
2722 Label LFlagInvalid;
2723 __ cmpldi(CCR0, Rflags, number_of_states);
2724 __ bge(CCR0, LFlagInvalid);
2725 #endif
2726
2727 // Load from branch table and dispatch (volatile case: one instruction ahead).
2728 __ sldi(Rflags, Rflags, LogBytesPerWord);
2729 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) { __ cmpwi(CR_is_vol, Rscratch, 1); } // Volatile?
2730 __ sldi(Rscratch, Rscratch, exact_log2(BytesPerInstWord)); // Volatile? size of instruction 1 : 0.
2731 __ ldx(Rbtable, Rbtable, Rflags);
2732
2733 __ subf(Rbtable, Rscratch, Rbtable); // Point to volatile/non-volatile entry point.
2734 __ mtctr(Rbtable);
2735 __ bctr();
2736
2737 #ifdef ASSERT
2738 __ bind(LFlagInvalid);
2739 __ stop("got invalid flag", 0x656);
2740
2741 // __ bind(Lvtos);
2742 address pc_before_release = __ pc();
2743 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2744 assert(__ pc() - pc_before_release == (ptrdiff_t)BytesPerInstWord, "must be single instruction");
2745 assert(branch_table[vtos] == 0, "can't compute twice");
2746 branch_table[vtos] = __ pc(); // non-volatile_entry point
2747 __ stop("vtos unexpected", 0x657);
2748 #endif
2749
2750 __ align(32, 28, 28); // Align pop.
2751 // __ bind(Ldtos);
2752 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2753 assert(branch_table[dtos] == 0, "can't compute twice");
2754 branch_table[dtos] = __ pc(); // non-volatile_entry point
2755 __ pop(dtos);
2756 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2757 __ stfdx(F15_ftos, Rclass_or_obj, Roffset);
2758 if (!is_static) { patch_bytecode(Bytecodes::_fast_dputfield, Rbc, Rscratch, true, byte_no); }
2759 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2760 __ beq(CR_is_vol, Lvolatile); // Volatile?
2761 }
2762 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2763
2764 __ align(32, 28, 28); // Align pop.
2765 // __ bind(Lftos);
2766 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2767 assert(branch_table[ftos] == 0, "can't compute twice");
2768 branch_table[ftos] = __ pc(); // non-volatile_entry point
2769 __ pop(ftos);
2770 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2771 __ stfsx(F15_ftos, Rclass_or_obj, Roffset);
2772 if (!is_static) { patch_bytecode(Bytecodes::_fast_fputfield, Rbc, Rscratch, true, byte_no); }
2773 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2774 __ beq(CR_is_vol, Lvolatile); // Volatile?
2775 }
2776 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2777
2778 __ align(32, 28, 28); // Align pop.
2779 // __ bind(Litos);
2780 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2781 assert(branch_table[itos] == 0, "can't compute twice");
2782 branch_table[itos] = __ pc(); // non-volatile_entry point
2783 __ pop(itos);
2784 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2785 __ stwx(R17_tos, Rclass_or_obj, Roffset);
2786 if (!is_static) { patch_bytecode(Bytecodes::_fast_iputfield, Rbc, Rscratch, true, byte_no); }
2787 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2788 __ beq(CR_is_vol, Lvolatile); // Volatile?
2789 }
2790 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2791
2792 __ align(32, 28, 28); // Align pop.
2793 // __ bind(Lltos);
2794 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2795 assert(branch_table[ltos] == 0, "can't compute twice");
2796 branch_table[ltos] = __ pc(); // non-volatile_entry point
2797 __ pop(ltos);
2798 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2799 __ stdx(R17_tos, Rclass_or_obj, Roffset);
2800 if (!is_static) { patch_bytecode(Bytecodes::_fast_lputfield, Rbc, Rscratch, true, byte_no); }
2801 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2802 __ beq(CR_is_vol, Lvolatile); // Volatile?
2803 }
2804 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2805
2806 __ align(32, 28, 28); // Align pop.
2807 // __ bind(Lbtos);
2808 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2809 assert(branch_table[btos] == 0, "can't compute twice");
2810 branch_table[btos] = __ pc(); // non-volatile_entry point
2811 __ pop(btos);
2812 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2813 __ stbx(R17_tos, Rclass_or_obj, Roffset);
2814 if (!is_static) { patch_bytecode(Bytecodes::_fast_bputfield, Rbc, Rscratch, true, byte_no); }
2815 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2816 __ beq(CR_is_vol, Lvolatile); // Volatile?
2817 }
2818 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2819
2820 __ align(32, 28, 28); // Align pop.
2821 // __ bind(Lztos);
2822 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2823 assert(branch_table[ztos] == 0, "can't compute twice");
2824 branch_table[ztos] = __ pc(); // non-volatile_entry point
2825 __ pop(ztos);
2826 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2827 __ andi(R17_tos, R17_tos, 0x1);
2828 __ stbx(R17_tos, Rclass_or_obj, Roffset);
2829 if (!is_static) { patch_bytecode(Bytecodes::_fast_zputfield, Rbc, Rscratch, true, byte_no); }
2830 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2831 __ beq(CR_is_vol, Lvolatile); // Volatile?
2832 }
2833 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2834
2835 __ align(32, 28, 28); // Align pop.
2836 // __ bind(Lctos);
2837 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2838 assert(branch_table[ctos] == 0, "can't compute twice");
2839 branch_table[ctos] = __ pc(); // non-volatile_entry point
2840 __ pop(ctos);
2841 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1..
2842 __ sthx(R17_tos, Rclass_or_obj, Roffset);
2843 if (!is_static) { patch_bytecode(Bytecodes::_fast_cputfield, Rbc, Rscratch, true, byte_no); }
2844 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2845 __ beq(CR_is_vol, Lvolatile); // Volatile?
2846 }
2847 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2848
2849 __ align(32, 28, 28); // Align pop.
2850 // __ bind(Lstos);
2851 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2852 assert(branch_table[stos] == 0, "can't compute twice");
2853 branch_table[stos] = __ pc(); // non-volatile_entry point
2854 __ pop(stos);
2855 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2856 __ sthx(R17_tos, Rclass_or_obj, Roffset);
2857 if (!is_static) { patch_bytecode(Bytecodes::_fast_sputfield, Rbc, Rscratch, true, byte_no); }
2858 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2859 __ beq(CR_is_vol, Lvolatile); // Volatile?
2860 }
2861 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2862
2863 __ align(32, 28, 28); // Align pop.
2864 // __ bind(Latos);
2865 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2866 assert(branch_table[atos] == 0, "can't compute twice");
2867 branch_table[atos] = __ pc(); // non-volatile_entry point
2868 __ pop(atos);
2869 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // kills R11_scratch1
2870 do_oop_store(_masm, Rclass_or_obj, Roffset, R17_tos, Rscratch, Rscratch2, Rscratch3, _bs->kind(), false /* precise */, true /* check null */);
2871 if (!is_static) { patch_bytecode(Bytecodes::_fast_aputfield, Rbc, Rscratch, true, byte_no); }
2872 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2873 __ beq(CR_is_vol, Lvolatile); // Volatile?
2874 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2875
2876 __ align(32, 12);
2877 __ bind(Lvolatile);
2878 __ fence();
2879 }
2880 // fallthru: __ b(Lexit);
2881
2882 #ifdef ASSERT
2883 for (int i = 0; i<number_of_states; ++i) {
2884 assert(branch_table[i], "put initialization");
2885 //tty->print_cr("put: %s_branch_table[%d] = 0x%llx (opcode 0x%llx)",
2886 // is_static ? "static" : "field", i, branch_table[i], *((unsigned int*)branch_table[i]));
2887 }
2888 #endif
2889 }
2890
2891 void TemplateTable::putfield(int byte_no) {
2892 putfield_or_static(byte_no, false);
2893 }
2894
2895 void TemplateTable::putstatic(int byte_no) {
2896 putfield_or_static(byte_no, true);
2897 }
2898
2899 // See SPARC. On PPC64, we have a different jvmti_post_field_mod which does the job.
2900 void TemplateTable::jvmti_post_fast_field_mod() {
2901 __ should_not_reach_here();
2902 }
2903
2904 void TemplateTable::fast_storefield(TosState state) {
2905 transition(state, vtos);
2906
2907 const Register Rcache = R5_ARG3, // Do not use ARG1/2 (causes trouble in jvmti_post_field_mod).
2908 Rclass_or_obj = R31, // Needs to survive C call.
2909 Roffset = R22_tmp2, // Needs to survive C call.
2910 Rflags = R3_ARG1,
2911 Rscratch = R11_scratch1,
2912 Rscratch2 = R12_scratch2,
2913 Rscratch3 = R4_ARG2;
2914 const ConditionRegister CR_is_vol = CCR2; // Non-volatile condition register (survives runtime call in do_oop_store).
2915
2916 // Constant pool already resolved => Load flags and offset of field.
2917 __ get_cache_and_index_at_bcp(Rcache, 1);
2918 jvmti_post_field_mod(Rcache, Rscratch, false /* not static */);
2919 load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
2920
2921 // Get the obj and the final store addr.
2922 pop_and_check_object(Rclass_or_obj); // Kills R11_scratch1.
2923
2924 // Get volatile flag.
2925 __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2926 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) { __ cmpdi(CR_is_vol, Rscratch, 1); }
2927 {
2928 Label LnotVolatile;
2929 __ beq(CCR0, LnotVolatile);
2930 __ release();
2931 __ align(32, 12);
2932 __ bind(LnotVolatile);
2933 }
2934
2935 // Do the store and fencing.
2936 switch(bytecode()) {
2937 case Bytecodes::_fast_aputfield:
2938 // Store into the field.
2939 do_oop_store(_masm, Rclass_or_obj, Roffset, R17_tos, Rscratch, Rscratch2, Rscratch3, _bs->kind(), false /* precise */, true /* check null */);
2940 break;
2941
2942 case Bytecodes::_fast_iputfield:
2943 __ stwx(R17_tos, Rclass_or_obj, Roffset);
2944 break;
2945
2946 case Bytecodes::_fast_lputfield:
2947 __ stdx(R17_tos, Rclass_or_obj, Roffset);
2948 break;
2949
2950 case Bytecodes::_fast_zputfield:
2951 __ andi(R17_tos, R17_tos, 0x1); // boolean is true if LSB is 1
2952 // fall through to bputfield
2953 case Bytecodes::_fast_bputfield:
2954 __ stbx(R17_tos, Rclass_or_obj, Roffset);
2955 break;
2956
2957 case Bytecodes::_fast_cputfield:
2958 case Bytecodes::_fast_sputfield:
2959 __ sthx(R17_tos, Rclass_or_obj, Roffset);
2960 break;
2961
2962 case Bytecodes::_fast_fputfield:
2963 __ stfsx(F15_ftos, Rclass_or_obj, Roffset);
2964 break;
2965
2966 case Bytecodes::_fast_dputfield:
2967 __ stfdx(F15_ftos, Rclass_or_obj, Roffset);
2968 break;
2969
2970 default: ShouldNotReachHere();
2971 }
2972
2973 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2974 Label LVolatile;
2975 __ beq(CR_is_vol, LVolatile);
2976 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2977
2978 __ align(32, 12);
2979 __ bind(LVolatile);
2980 __ fence();
2981 }
2982 }
2983
2984 void TemplateTable::fast_accessfield(TosState state) {
2985 transition(atos, state);
2986
2987 Label LisVolatile;
2988 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2989
2990 const Register Rcache = R3_ARG1,
2991 Rclass_or_obj = R17_tos,
2992 Roffset = R22_tmp2,
2993 Rflags = R23_tmp3,
2994 Rscratch = R12_scratch2;
2995
2996 // Constant pool already resolved. Get the field offset.
2997 __ get_cache_and_index_at_bcp(Rcache, 1);
2998 load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
2999
3000 // JVMTI support
3001 jvmti_post_field_access(Rcache, Rscratch, false, true);
3002
3003 // Get the load address.
3004 __ null_check_throw(Rclass_or_obj, -1, Rscratch);
3005
3006 // Get volatile flag.
3007 __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
3008 __ bne(CCR0, LisVolatile);
3009
3010 switch(bytecode()) {
3011 case Bytecodes::_fast_agetfield:
3012 {
3013 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3014 __ verify_oop(R17_tos);
3015 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3016
3017 __ bind(LisVolatile);
3018 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3019 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3020 __ verify_oop(R17_tos);
3021 __ twi_0(R17_tos);
3022 __ isync();
3023 break;
3024 }
3025 case Bytecodes::_fast_igetfield:
3026 {
3027 __ lwax(R17_tos, Rclass_or_obj, Roffset);
3028 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3029
3030 __ bind(LisVolatile);
3031 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3032 __ lwax(R17_tos, Rclass_or_obj, Roffset);
3033 __ twi_0(R17_tos);
3034 __ isync();
3035 break;
3036 }
3037 case Bytecodes::_fast_lgetfield:
3038 {
3039 __ ldx(R17_tos, Rclass_or_obj, Roffset);
3040 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3041
3042 __ bind(LisVolatile);
3043 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3044 __ ldx(R17_tos, Rclass_or_obj, Roffset);
3045 __ twi_0(R17_tos);
3046 __ isync();
3047 break;
3048 }
3049 case Bytecodes::_fast_bgetfield:
3050 {
3051 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
3052 __ extsb(R17_tos, R17_tos);
3053 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3054
3055 __ bind(LisVolatile);
3056 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3057 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
3058 __ twi_0(R17_tos);
3059 __ extsb(R17_tos, R17_tos);
3060 __ isync();
3061 break;
3062 }
3063 case Bytecodes::_fast_cgetfield:
3064 {
3065 __ lhzx(R17_tos, Rclass_or_obj, Roffset);
3066 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3067
3068 __ bind(LisVolatile);
3069 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3070 __ lhzx(R17_tos, Rclass_or_obj, Roffset);
3071 __ twi_0(R17_tos);
3072 __ isync();
3073 break;
3074 }
3075 case Bytecodes::_fast_sgetfield:
3076 {
3077 __ lhax(R17_tos, Rclass_or_obj, Roffset);
3078 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3079
3080 __ bind(LisVolatile);
3081 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3082 __ lhax(R17_tos, Rclass_or_obj, Roffset);
3083 __ twi_0(R17_tos);
3084 __ isync();
3085 break;
3086 }
3087 case Bytecodes::_fast_fgetfield:
3088 {
3089 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3090 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3091
3092 __ bind(LisVolatile);
3093 Label Ldummy;
3094 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3095 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3096 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3097 __ bne_predict_not_taken(CCR0, Ldummy);
3098 __ bind(Ldummy);
3099 __ isync();
3100 break;
3101 }
3102 case Bytecodes::_fast_dgetfield:
3103 {
3104 __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
3105 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3106
3107 __ bind(LisVolatile);
3108 Label Ldummy;
3109 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3110 __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
3111 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3112 __ bne_predict_not_taken(CCR0, Ldummy);
3113 __ bind(Ldummy);
3114 __ isync();
3115 break;
3116 }
3117 default: ShouldNotReachHere();
3118 }
3119 }
3120
3121 void TemplateTable::fast_xaccess(TosState state) {
3122 transition(vtos, state);
3123
3124 Label LisVolatile;
3125 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
3126 const Register Rcache = R3_ARG1,
3127 Rclass_or_obj = R17_tos,
3128 Roffset = R22_tmp2,
3129 Rflags = R23_tmp3,
3130 Rscratch = R12_scratch2;
3131
3132 __ ld(Rclass_or_obj, 0, R18_locals);
3133
3134 // Constant pool already resolved. Get the field offset.
3135 __ get_cache_and_index_at_bcp(Rcache, 2);
3136 load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
3137
3138 // JVMTI support not needed, since we switch back to single bytecode as soon as debugger attaches.
3139
3140 // Needed to report exception at the correct bcp.
3141 __ addi(R14_bcp, R14_bcp, 1);
3142
3143 // Get the load address.
3144 __ null_check_throw(Rclass_or_obj, -1, Rscratch);
3145
3146 // Get volatile flag.
3147 __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
3148 __ bne(CCR0, LisVolatile);
3149
3150 switch(state) {
3151 case atos:
3152 {
3153 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3154 __ verify_oop(R17_tos);
3155 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3156
3157 __ bind(LisVolatile);
3158 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3159 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3160 __ verify_oop(R17_tos);
3161 __ twi_0(R17_tos);
3162 __ isync();
3163 break;
3164 }
3165 case itos:
3166 {
3167 __ lwax(R17_tos, Rclass_or_obj, Roffset);
3168 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3169
3170 __ bind(LisVolatile);
3171 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3172 __ lwax(R17_tos, Rclass_or_obj, Roffset);
3173 __ twi_0(R17_tos);
3174 __ isync();
3175 break;
3176 }
3177 case ftos:
3178 {
3179 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3180 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3181
3182 __ bind(LisVolatile);
3183 Label Ldummy;
3184 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3185 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3186 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3187 __ bne_predict_not_taken(CCR0, Ldummy);
3188 __ bind(Ldummy);
3189 __ isync();
3190 break;
3191 }
3192 default: ShouldNotReachHere();
3193 }
3194 __ addi(R14_bcp, R14_bcp, -1);
3195 }
3196
3197 // ============================================================================
3198 // Calls
3199
3200 // Common code for invoke
3201 //
3202 // Input:
3203 // - byte_no
3204 //
3205 // Output:
3206 // - Rmethod: The method to invoke next.
3207 // - Rret_addr: The return address to return to.
3208 // - Rindex: MethodType (invokehandle) or CallSite obj (invokedynamic)
3209 // - Rrecv: Cache for "this" pointer, might be noreg if static call.
3210 // - Rflags: Method flags from const pool cache.
3211 //
3212 // Kills:
3213 // - Rscratch1
3214 //
3215 void TemplateTable::prepare_invoke(int byte_no,
3216 Register Rmethod, // linked method (or i-klass)
3217 Register Rret_addr,// return address
3218 Register Rindex, // itable index, MethodType, etc.
3219 Register Rrecv, // If caller wants to see it.
3220 Register Rflags, // If caller wants to test it.
3221 Register Rscratch
3222 ) {
3223 // Determine flags.
3224 const Bytecodes::Code code = bytecode();
3225 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3226 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3227 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3228 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3229 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3230 const bool load_receiver = (Rrecv != noreg);
3231 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3232
3233 assert_different_registers(Rmethod, Rindex, Rflags, Rscratch);
3234 assert_different_registers(Rmethod, Rrecv, Rflags, Rscratch);
3235 assert_different_registers(Rret_addr, Rscratch);
3236
3237 load_invoke_cp_cache_entry(byte_no, Rmethod, Rindex, Rflags, is_invokevirtual, false, is_invokedynamic);
3238
3239 // Saving of SP done in call_from_interpreter.
3240
3241 // Maybe push "appendix" to arguments.
3242 if (is_invokedynamic || is_invokehandle) {
3243 Label Ldone;
3244 __ rldicl_(R0, Rflags, 64-ConstantPoolCacheEntry::has_appendix_shift, 63);
3245 __ beq(CCR0, Ldone);
3246 // Push "appendix" (MethodType, CallSite, etc.).
3247 // This must be done before we get the receiver,
3248 // since the parameter_size includes it.
3249 __ load_resolved_reference_at_index(Rscratch, Rindex);
3250 __ verify_oop(Rscratch);
3251 __ push_ptr(Rscratch);
3252 __ bind(Ldone);
3253 }
3254
3255 // Load receiver if needed (after appendix is pushed so parameter size is correct).
3256 if (load_receiver) {
3257 const Register Rparam_count = Rscratch;
3258 __ andi(Rparam_count, Rflags, ConstantPoolCacheEntry::parameter_size_mask);
3259 __ load_receiver(Rparam_count, Rrecv);
3260 __ verify_oop(Rrecv);
3261 }
3262
3263 // Get return address.
3264 {
3265 Register Rtable_addr = Rscratch;
3266 Register Rret_type = Rret_addr;
3267 address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3268
3269 // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3270 __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3271 __ load_dispatch_table(Rtable_addr, (address*)table_addr);
3272 __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3273 // Get return address.
3274 __ ldx(Rret_addr, Rtable_addr, Rret_type);
3275 }
3276 }
3277
3278 // Helper for virtual calls. Load target out of vtable and jump off!
3279 // Kills all passed registers.
3280 void TemplateTable::generate_vtable_call(Register Rrecv_klass, Register Rindex, Register Rret, Register Rtemp) {
3281
3282 assert_different_registers(Rrecv_klass, Rtemp, Rret);
3283 const Register Rtarget_method = Rindex;
3284
3285 // Get target method & entry point.
3286 const int base = InstanceKlass::vtable_start_offset() * wordSize;
3287 // Calc vtable addr scale the vtable index by 8.
3288 __ sldi(Rindex, Rindex, exact_log2(vtableEntry::size() * wordSize));
3289 // Load target.
3290 __ addi(Rrecv_klass, Rrecv_klass, base + vtableEntry::method_offset_in_bytes());
3291 __ ldx(Rtarget_method, Rindex, Rrecv_klass);
3292 // Argument and return type profiling.
3293 __ profile_arguments_type(Rtarget_method, Rrecv_klass /* scratch1 */, Rtemp /* scratch2 */, true);
3294 __ call_from_interpreter(Rtarget_method, Rret, Rrecv_klass /* scratch1 */, Rtemp /* scratch2 */);
3295 }
3296
3297 // Virtual or final call. Final calls are rewritten on the fly to run through "fast_finalcall" next time.
3298 void TemplateTable::invokevirtual(int byte_no) {
3299 transition(vtos, vtos);
3300
3301 Register Rtable_addr = R11_scratch1,
3302 Rret_type = R12_scratch2,
3303 Rret_addr = R5_ARG3,
3304 Rflags = R22_tmp2, // Should survive C call.
3305 Rrecv = R3_ARG1,
3306 Rrecv_klass = Rrecv,
3307 Rvtableindex_or_method = R31, // Should survive C call.
3308 Rnum_params = R4_ARG2,
3309 Rnew_bc = R6_ARG4;
3310
3311 Label LnotFinal;
3312
3313 load_invoke_cp_cache_entry(byte_no, Rvtableindex_or_method, noreg, Rflags, /*virtual*/ true, false, false);
3314
3315 __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_vfinal_shift);
3316 __ bfalse(CCR0, LnotFinal);
3317
3318 patch_bytecode(Bytecodes::_fast_invokevfinal, Rnew_bc, R12_scratch2);
3319 invokevfinal_helper(Rvtableindex_or_method, Rflags, R11_scratch1, R12_scratch2);
3320
3321 __ align(32, 12);
3322 __ bind(LnotFinal);
3323 // Load "this" pointer (receiver).
3324 __ rldicl(Rnum_params, Rflags, 64, 48);
3325 __ load_receiver(Rnum_params, Rrecv);
3326 __ verify_oop(Rrecv);
3327
3328 // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3329 __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3330 __ load_dispatch_table(Rtable_addr, Interpreter::invoke_return_entry_table());
3331 __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3332 __ ldx(Rret_addr, Rret_type, Rtable_addr);
3333 __ null_check_throw(Rrecv, oopDesc::klass_offset_in_bytes(), R11_scratch1);
3334 __ load_klass(Rrecv_klass, Rrecv);
3335 __ verify_klass_ptr(Rrecv_klass);
3336 __ profile_virtual_call(Rrecv_klass, R11_scratch1, R12_scratch2, false);
3337
3338 generate_vtable_call(Rrecv_klass, Rvtableindex_or_method, Rret_addr, R11_scratch1);
3339 }
3340
3341 void TemplateTable::fast_invokevfinal(int byte_no) {
3342 transition(vtos, vtos);
3343
3344 assert(byte_no == f2_byte, "use this argument");
3345 Register Rflags = R22_tmp2,
3346 Rmethod = R31;
3347 load_invoke_cp_cache_entry(byte_no, Rmethod, noreg, Rflags, /*virtual*/ true, /*is_invokevfinal*/ true, false);
3348 invokevfinal_helper(Rmethod, Rflags, R11_scratch1, R12_scratch2);
3349 }
3350
3351 void TemplateTable::invokevfinal_helper(Register Rmethod, Register Rflags, Register Rscratch1, Register Rscratch2) {
3352
3353 assert_different_registers(Rmethod, Rflags, Rscratch1, Rscratch2);
3354
3355 // Load receiver from stack slot.
3356 Register Rrecv = Rscratch2;
3357 Register Rnum_params = Rrecv;
3358
3359 __ ld(Rnum_params, in_bytes(Method::const_offset()), Rmethod);
3360 __ lhz(Rnum_params /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), Rnum_params);
3361
3362 // Get return address.
3363 Register Rtable_addr = Rscratch1,
3364 Rret_addr = Rflags,
3365 Rret_type = Rret_addr;
3366 // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3367 __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3368 __ load_dispatch_table(Rtable_addr, Interpreter::invoke_return_entry_table());
3369 __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3370 __ ldx(Rret_addr, Rret_type, Rtable_addr);
3371
3372 // Load receiver and receiver NULL check.
3373 __ load_receiver(Rnum_params, Rrecv);
3374 __ null_check_throw(Rrecv, -1, Rscratch1);
3375
3376 __ profile_final_call(Rrecv, Rscratch1);
3377 // Argument and return type profiling.
3378 __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, true);
3379
3380 // Do the call.
3381 __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1, Rscratch2);
3382 }
3383
3384 void TemplateTable::invokespecial(int byte_no) {
3385 assert(byte_no == f1_byte, "use this argument");
3386 transition(vtos, vtos);
3387
3388 Register Rtable_addr = R3_ARG1,
3389 Rret_addr = R4_ARG2,
3390 Rflags = R5_ARG3,
3391 Rreceiver = R6_ARG4,
3392 Rmethod = R31;
3393
3394 prepare_invoke(byte_no, Rmethod, Rret_addr, noreg, Rreceiver, Rflags, R11_scratch1);
3395
3396 // Receiver NULL check.
3397 __ null_check_throw(Rreceiver, -1, R11_scratch1);
3398
3399 __ profile_call(R11_scratch1, R12_scratch2);
3400 // Argument and return type profiling.
3401 __ profile_arguments_type(Rmethod, R11_scratch1, R12_scratch2, false);
3402 __ call_from_interpreter(Rmethod, Rret_addr, R11_scratch1, R12_scratch2);
3403 }
3404
3405 void TemplateTable::invokestatic(int byte_no) {
3406 assert(byte_no == f1_byte, "use this argument");
3407 transition(vtos, vtos);
3408
3409 Register Rtable_addr = R3_ARG1,
3410 Rret_addr = R4_ARG2,
3411 Rflags = R5_ARG3;
3412
3413 prepare_invoke(byte_no, R19_method, Rret_addr, noreg, noreg, Rflags, R11_scratch1);
3414
3415 __ profile_call(R11_scratch1, R12_scratch2);
3416 // Argument and return type profiling.
3417 __ profile_arguments_type(R19_method, R11_scratch1, R12_scratch2, false);
3418 __ call_from_interpreter(R19_method, Rret_addr, R11_scratch1, R12_scratch2);
3419 }
3420
3421 void TemplateTable::invokeinterface_object_method(Register Rrecv_klass,
3422 Register Rret,
3423 Register Rflags,
3424 Register Rindex,
3425 Register Rtemp1,
3426 Register Rtemp2) {
3427
3428 assert_different_registers(Rindex, Rret, Rrecv_klass, Rflags, Rtemp1, Rtemp2);
3429 Label LnotFinal;
3430
3431 // Check for vfinal.
3432 __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_vfinal_shift);
3433 __ bfalse(CCR0, LnotFinal);
3434
3435 Register Rscratch = Rflags; // Rflags is dead now.
3436
3437 // Final call case.
3438 __ profile_final_call(Rtemp1, Rscratch);
3439 // Argument and return type profiling.
3440 __ profile_arguments_type(Rindex, Rscratch, Rrecv_klass /* scratch */, true);
3441 // Do the final call - the index (f2) contains the method.
3442 __ call_from_interpreter(Rindex, Rret, Rscratch, Rrecv_klass /* scratch */);
3443
3444 // Non-final callc case.
3445 __ bind(LnotFinal);
3446 __ profile_virtual_call(Rrecv_klass, Rtemp1, Rscratch, false);
3447 generate_vtable_call(Rrecv_klass, Rindex, Rret, Rscratch);
3448 }
3449
3450 void TemplateTable::invokeinterface(int byte_no) {
3451 assert(byte_no == f1_byte, "use this argument");
3452 transition(vtos, vtos);
3453
3454 const Register Rscratch1 = R11_scratch1,
3455 Rscratch2 = R12_scratch2,
3456 Rscratch3 = R9_ARG7,
3457 Rscratch4 = R10_ARG8,
3458 Rtable_addr = Rscratch2,
3459 Rinterface_klass = R5_ARG3,
3460 Rret_type = R8_ARG6,
3461 Rret_addr = Rret_type,
3462 Rindex = R6_ARG4,
3463 Rreceiver = R4_ARG2,
3464 Rrecv_klass = Rreceiver,
3465 Rflags = R7_ARG5;
3466
3467 prepare_invoke(byte_no, Rinterface_klass, Rret_addr, Rindex, Rreceiver, Rflags, Rscratch1);
3468
3469 // Get receiver klass.
3470 __ null_check_throw(Rreceiver, oopDesc::klass_offset_in_bytes(), Rscratch3);
3471 __ load_klass(Rrecv_klass, Rreceiver);
3472
3473 // Check corner case object method.
3474 Label LobjectMethod;
3475
3476 __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_forced_virtual_shift);
3477 __ btrue(CCR0, LobjectMethod);
3478
3479 // Fallthrough: The normal invokeinterface case.
3480 __ profile_virtual_call(Rrecv_klass, Rscratch1, Rscratch2, false);
3481
3482 // Find entry point to call.
3483 Label Lthrow_icc, Lthrow_ame;
3484 // Result will be returned in Rindex.
3485 __ mr(Rscratch4, Rrecv_klass);
3486 __ mr(Rscratch3, Rindex);
3487 __ lookup_interface_method(Rrecv_klass, Rinterface_klass, Rindex, Rindex, Rscratch1, Rscratch2, Lthrow_icc);
3488
3489 __ cmpdi(CCR0, Rindex, 0);
3490 __ beq(CCR0, Lthrow_ame);
3491 // Found entry. Jump off!
3492 // Argument and return type profiling.
3493 __ profile_arguments_type(Rindex, Rscratch1, Rscratch2, true);
3494 __ call_from_interpreter(Rindex, Rret_addr, Rscratch1, Rscratch2);
3495
3496 // Vtable entry was NULL => Throw abstract method error.
3497 __ bind(Lthrow_ame);
3498 __ mr(Rrecv_klass, Rscratch4);
3499 __ mr(Rindex, Rscratch3);
3500 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3501
3502 // Interface was not found => Throw incompatible class change error.
3503 __ bind(Lthrow_icc);
3504 __ mr(Rrecv_klass, Rscratch4);
3505 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeError));
3506
3507 __ should_not_reach_here();
3508
3509 // Special case of invokeinterface called for virtual method of
3510 // java.lang.Object. See ConstantPoolCacheEntry::set_method() for details:
3511 // The invokeinterface was rewritten to a invokevirtual, hence we have
3512 // to handle this corner case. This code isn't produced by javac, but could
3513 // be produced by another compliant java compiler.
3514 __ bind(LobjectMethod);
3515 invokeinterface_object_method(Rrecv_klass, Rret_addr, Rflags, Rindex, Rscratch1, Rscratch2);
3516 }
3517
3518 void TemplateTable::invokedynamic(int byte_no) {
3519 transition(vtos, vtos);
3520
3521 const Register Rret_addr = R3_ARG1,
3522 Rflags = R4_ARG2,
3523 Rmethod = R22_tmp2,
3524 Rscratch1 = R11_scratch1,
3525 Rscratch2 = R12_scratch2;
3526
3527 if (!EnableInvokeDynamic) {
3528 // We should not encounter this bytecode if !EnableInvokeDynamic.
3529 // The verifier will stop it. However, if we get past the verifier,
3530 // this will stop the thread in a reasonable way, without crashing the JVM.
3531 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeError));
3532 // The call_VM checks for exception, so we should never return here.
3533 __ should_not_reach_here();
3534 return;
3535 }
3536
3537 prepare_invoke(byte_no, Rmethod, Rret_addr, Rscratch1, noreg, Rflags, Rscratch2);
3538
3539 // Profile this call.
3540 __ profile_call(Rscratch1, Rscratch2);
3541
3542 // Off we go. With the new method handles, we don't jump to a method handle
3543 // entry any more. Instead, we pushed an "appendix" in prepare invoke, which happens
3544 // to be the callsite object the bootstrap method returned. This is passed to a
3545 // "link" method which does the dispatch (Most likely just grabs the MH stored
3546 // inside the callsite and does an invokehandle).
3547 // Argument and return type profiling.
3548 __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, false);
3549 __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1 /* scratch1 */, Rscratch2 /* scratch2 */);
3550 }
3551
3552 void TemplateTable::invokehandle(int byte_no) {
3553 transition(vtos, vtos);
3554
3555 const Register Rret_addr = R3_ARG1,
3556 Rflags = R4_ARG2,
3557 Rrecv = R5_ARG3,
3558 Rmethod = R22_tmp2,
3559 Rscratch1 = R11_scratch1,
3560 Rscratch2 = R12_scratch2;
3561
3562 if (!EnableInvokeDynamic) {
3563 // Rewriter does not generate this bytecode.
3564 __ should_not_reach_here();
3565 return;
3566 }
3567
3568 prepare_invoke(byte_no, Rmethod, Rret_addr, Rscratch1, Rrecv, Rflags, Rscratch2);
3569 __ verify_method_ptr(Rmethod);
3570 __ null_check_throw(Rrecv, -1, Rscratch2);
3571
3572 __ profile_final_call(Rrecv, Rscratch1);
3573
3574 // Still no call from handle => We call the method handle interpreter here.
3575 // Argument and return type profiling.
3576 __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, true);
3577 __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1 /* scratch1 */, Rscratch2 /* scratch2 */);
3578 }
3579
3580 // =============================================================================
3581 // Allocation
3582
3583 // Puts allocated obj ref onto the expression stack.
3584 void TemplateTable::_new() {
3585 transition(vtos, atos);
3586
3587 Label Lslow_case,
3588 Ldone,
3589 Linitialize_header,
3590 Lallocate_shared,
3591 Linitialize_object; // Including clearing the fields.
3592
3593 const Register RallocatedObject = R17_tos,
3594 RinstanceKlass = R9_ARG7,
3595 Rscratch = R11_scratch1,
3596 Roffset = R8_ARG6,
3597 Rinstance_size = Roffset,
3598 Rcpool = R4_ARG2,
3599 Rtags = R3_ARG1,
3600 Rindex = R5_ARG3;
3601
3602 const bool allow_shared_alloc = Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
3603
3604 // --------------------------------------------------------------------------
3605 // Check if fast case is possible.
3606
3607 // Load pointers to const pool and const pool's tags array.
3608 __ get_cpool_and_tags(Rcpool, Rtags);
3609 // Load index of constant pool entry.
3610 __ get_2_byte_integer_at_bcp(1, Rindex, InterpreterMacroAssembler::Unsigned);
3611
3612 if (UseTLAB) {
3613 // Make sure the class we're about to instantiate has been resolved
3614 // This is done before loading instanceKlass to be consistent with the order
3615 // how Constant Pool is updated (see ConstantPoolCache::klass_at_put).
3616 __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3617 __ lbzx(Rtags, Rindex, Rtags);
3618
3619 __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3620 __ bne(CCR0, Lslow_case);
3621
3622 // Get instanceKlass (load from Rcpool + sizeof(ConstantPool) + Rindex*BytesPerWord).
3623 __ sldi(Roffset, Rindex, LogBytesPerWord);
3624 __ addi(Rscratch, Rcpool, sizeof(ConstantPool));
3625 __ isync(); // Order load of instance Klass wrt. tags.
3626 __ ldx(RinstanceKlass, Roffset, Rscratch);
3627
3628 // Make sure klass is fully initialized and get instance_size.
3629 __ lbz(Rscratch, in_bytes(InstanceKlass::init_state_offset()), RinstanceKlass);
3630 __ lwz(Rinstance_size, in_bytes(Klass::layout_helper_offset()), RinstanceKlass);
3631
3632 __ cmpdi(CCR1, Rscratch, InstanceKlass::fully_initialized);
3633 // Make sure klass does not have has_finalizer, or is abstract, or interface or java/lang/Class.
3634 __ andi_(R0, Rinstance_size, Klass::_lh_instance_slow_path_bit); // slow path bit equals 0?
3635
3636 __ crnand(/*CR0 eq*/2, /*CR1 eq*/4+2, /*CR0 eq*/2); // slow path bit set or not fully initialized?
3637 __ beq(CCR0, Lslow_case);
3638
3639 // --------------------------------------------------------------------------
3640 // Fast case:
3641 // Allocate the instance.
3642 // 1) Try to allocate in the TLAB.
3643 // 2) If fail, and the TLAB is not full enough to discard, allocate in the shared Eden.
3644 // 3) If the above fails (or is not applicable), go to a slow case (creates a new TLAB, etc.).
3645
3646 Register RoldTopValue = RallocatedObject; // Object will be allocated here if it fits.
3647 Register RnewTopValue = R6_ARG4;
3648 Register RendValue = R7_ARG5;
3649
3650 // Check if we can allocate in the TLAB.
3651 __ ld(RoldTopValue, in_bytes(JavaThread::tlab_top_offset()), R16_thread);
3652 __ ld(RendValue, in_bytes(JavaThread::tlab_end_offset()), R16_thread);
3653
3654 __ add(RnewTopValue, Rinstance_size, RoldTopValue);
3655
3656 // If there is enough space, we do not CAS and do not clear.
3657 __ cmpld(CCR0, RnewTopValue, RendValue);
3658 __ bgt(CCR0, allow_shared_alloc ? Lallocate_shared : Lslow_case);
3659
3660 __ std(RnewTopValue, in_bytes(JavaThread::tlab_top_offset()), R16_thread);
3661
3662 if (ZeroTLAB) {
3663 // The fields have already been cleared.
3664 __ b(Linitialize_header);
3665 } else {
3666 // Initialize both the header and fields.
3667 __ b(Linitialize_object);
3668 }
3669
3670 // Fall through: TLAB was too small.
3671 if (allow_shared_alloc) {
3672 Register RtlabWasteLimitValue = R10_ARG8;
3673 Register RfreeValue = RnewTopValue;
3674
3675 __ bind(Lallocate_shared);
3676 // Check if tlab should be discarded (refill_waste_limit >= free).
3677 __ ld(RtlabWasteLimitValue, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), R16_thread);
3678 __ subf(RfreeValue, RoldTopValue, RendValue);
3679 __ srdi(RfreeValue, RfreeValue, LogHeapWordSize); // in dwords
3680 __ cmpld(CCR0, RtlabWasteLimitValue, RfreeValue);
3681 __ bge(CCR0, Lslow_case);
3682
3683 // Increment waste limit to prevent getting stuck on this slow path.
3684 __ addi(RtlabWasteLimitValue, RtlabWasteLimitValue, (int)ThreadLocalAllocBuffer::refill_waste_limit_increment());
3685 __ std(RtlabWasteLimitValue, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), R16_thread);
3686 }
3687 // else: No allocation in the shared eden. // fallthru: __ b(Lslow_case);
3688 }
3689 // else: Always go the slow path.
3690
3691 // --------------------------------------------------------------------------
3692 // slow case
3693 __ bind(Lslow_case);
3694 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), Rcpool, Rindex);
3695
3696 if (UseTLAB) {
3697 __ b(Ldone);
3698 // --------------------------------------------------------------------------
3699 // Init1: Zero out newly allocated memory.
3700
3701 if (!ZeroTLAB || allow_shared_alloc) {
3702 // Clear object fields.
3703 __ bind(Linitialize_object);
3704
3705 // Initialize remaining object fields.
3706 Register Rbase = Rtags;
3707 __ addi(Rinstance_size, Rinstance_size, 7 - (int)sizeof(oopDesc));
3708 __ addi(Rbase, RallocatedObject, sizeof(oopDesc));
3709 __ srdi(Rinstance_size, Rinstance_size, 3);
3710
3711 // Clear out object skipping header. Takes also care of the zero length case.
3712 __ clear_memory_doubleword(Rbase, Rinstance_size);
3713 // fallthru: __ b(Linitialize_header);
3714 }
3715
3716 // --------------------------------------------------------------------------
3717 // Init2: Initialize the header: mark, klass
3718 __ bind(Linitialize_header);
3719
3720 // Init mark.
3721 if (UseBiasedLocking) {
3722 __ ld(Rscratch, in_bytes(Klass::prototype_header_offset()), RinstanceKlass);
3723 } else {
3724 __ load_const_optimized(Rscratch, markOopDesc::prototype(), R0);
3725 }
3726 __ std(Rscratch, oopDesc::mark_offset_in_bytes(), RallocatedObject);
3727
3728 // Init klass.
3729 __ store_klass_gap(RallocatedObject);
3730 __ store_klass(RallocatedObject, RinstanceKlass, Rscratch); // klass (last for cms)
3731
3732 // Check and trigger dtrace event.
3733 {
3734 SkipIfEqualZero skip_if(_masm, Rscratch, &DTraceAllocProbes);
3735 __ push(atos);
3736 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc));
3737 __ pop(atos);
3738 }
3739 }
3740
3741 // continue
3742 __ bind(Ldone);
3743
3744 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3745 __ membar(Assembler::StoreStore);
3746 }
3747
3748 void TemplateTable::newarray() {
3749 transition(itos, atos);
3750
3751 __ lbz(R4, 1, R14_bcp);
3752 __ extsw(R5, R17_tos);
3753 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), R4, R5 /* size */);
3754
3755 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3756 __ membar(Assembler::StoreStore);
3757 }
3758
3759 void TemplateTable::anewarray() {
3760 transition(itos, atos);
3761
3762 __ get_constant_pool(R4);
3763 __ get_2_byte_integer_at_bcp(1, R5, InterpreterMacroAssembler::Unsigned);
3764 __ extsw(R6, R17_tos); // size
3765 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), R4 /* pool */, R5 /* index */, R6 /* size */);
3766
3767 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3768 __ membar(Assembler::StoreStore);
3769 }
3770
3771 // Allocate a multi dimensional array
3772 void TemplateTable::multianewarray() {
3773 transition(vtos, atos);
3774
3775 Register Rptr = R31; // Needs to survive C call.
3776
3777 // Put ndims * wordSize into frame temp slot
3778 __ lbz(Rptr, 3, R14_bcp);
3779 __ sldi(Rptr, Rptr, Interpreter::logStackElementSize);
3780 // Esp points past last_dim, so set to R4 to first_dim address.
3781 __ add(R4, Rptr, R15_esp);
3782 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), R4 /* first_size_address */);
3783 // Pop all dimensions off the stack.
3784 __ add(R15_esp, Rptr, R15_esp);
3785
3786 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3787 __ membar(Assembler::StoreStore);
3788 }
3789
3790 void TemplateTable::arraylength() {
3791 transition(atos, itos);
3792
3793 Label LnoException;
3794 __ verify_oop(R17_tos);
3795 __ null_check_throw(R17_tos, arrayOopDesc::length_offset_in_bytes(), R11_scratch1);
3796 __ lwa(R17_tos, arrayOopDesc::length_offset_in_bytes(), R17_tos);
3797 }
3798
3799 // ============================================================================
3800 // Typechecks
3801
3802 void TemplateTable::checkcast() {
3803 transition(atos, atos);
3804
3805 Label Ldone, Lis_null, Lquicked, Lresolved;
3806 Register Roffset = R6_ARG4,
3807 RobjKlass = R4_ARG2,
3808 RspecifiedKlass = R5_ARG3, // Generate_ClassCastException_verbose_handler will read value from this register.
3809 Rcpool = R11_scratch1,
3810 Rtags = R12_scratch2;
3811
3812 // Null does not pass.
3813 __ cmpdi(CCR0, R17_tos, 0);
3814 __ beq(CCR0, Lis_null);
3815
3816 // Get constant pool tag to find out if the bytecode has already been "quickened".
3817 __ get_cpool_and_tags(Rcpool, Rtags);
3818
3819 __ get_2_byte_integer_at_bcp(1, Roffset, InterpreterMacroAssembler::Unsigned);
3820
3821 __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3822 __ lbzx(Rtags, Rtags, Roffset);
3823
3824 __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3825 __ beq(CCR0, Lquicked);
3826
3827 // Call into the VM to "quicken" instanceof.
3828 __ push_ptr(); // for GC
3829 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3830 __ get_vm_result_2(RspecifiedKlass);
3831 __ pop_ptr(); // Restore receiver.
3832 __ b(Lresolved);
3833
3834 // Extract target class from constant pool.
3835 __ bind(Lquicked);
3836 __ sldi(Roffset, Roffset, LogBytesPerWord);
3837 __ addi(Rcpool, Rcpool, sizeof(ConstantPool));
3838 __ isync(); // Order load of specified Klass wrt. tags.
3839 __ ldx(RspecifiedKlass, Rcpool, Roffset);
3840
3841 // Do the checkcast.
3842 __ bind(Lresolved);
3843 // Get value klass in RobjKlass.
3844 __ load_klass(RobjKlass, R17_tos);
3845 // Generate a fast subtype check. Branch to cast_ok if no failure. Return 0 if failure.
3846 __ gen_subtype_check(RobjKlass, RspecifiedKlass, /*3 temp regs*/ Roffset, Rcpool, Rtags, /*target if subtype*/ Ldone);
3847
3848 // Not a subtype; so must throw exception
3849 // Target class oop is in register R6_ARG4 == RspecifiedKlass by convention.
3850 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ClassCastException_entry);
3851 __ mtctr(R11_scratch1);
3852 __ bctr();
3853
3854 // Profile the null case.
3855 __ align(32, 12);
3856 __ bind(Lis_null);
3857 __ profile_null_seen(R11_scratch1, Rtags); // Rtags used as scratch.
3858
3859 __ align(32, 12);
3860 __ bind(Ldone);
3861 }
3862
3863 // Output:
3864 // - tos == 0: Obj was null or not an instance of class.
3865 // - tos == 1: Obj was an instance of class.
3866 void TemplateTable::instanceof() {
3867 transition(atos, itos);
3868
3869 Label Ldone, Lis_null, Lquicked, Lresolved;
3870 Register Roffset = R5_ARG3,
3871 RobjKlass = R4_ARG2,
3872 RspecifiedKlass = R6_ARG4, // Generate_ClassCastException_verbose_handler will expect the value in this register.
3873 Rcpool = R11_scratch1,
3874 Rtags = R12_scratch2;
3875
3876 // Null does not pass.
3877 __ cmpdi(CCR0, R17_tos, 0);
3878 __ beq(CCR0, Lis_null);
3879
3880 // Get constant pool tag to find out if the bytecode has already been "quickened".
3881 __ get_cpool_and_tags(Rcpool, Rtags);
3882
3883 __ get_2_byte_integer_at_bcp(1, Roffset, InterpreterMacroAssembler::Unsigned);
3884
3885 __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3886 __ lbzx(Rtags, Rtags, Roffset);
3887
3888 __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3889 __ beq(CCR0, Lquicked);
3890
3891 // Call into the VM to "quicken" instanceof.
3892 __ push_ptr(); // for GC
3893 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3894 __ get_vm_result_2(RspecifiedKlass);
3895 __ pop_ptr(); // Restore receiver.
3896 __ b(Lresolved);
3897
3898 // Extract target class from constant pool.
3899 __ bind(Lquicked);
3900 __ sldi(Roffset, Roffset, LogBytesPerWord);
3901 __ addi(Rcpool, Rcpool, sizeof(ConstantPool));
3902 __ isync(); // Order load of specified Klass wrt. tags.
3903 __ ldx(RspecifiedKlass, Rcpool, Roffset);
3904
3905 // Do the checkcast.
3906 __ bind(Lresolved);
3907 // Get value klass in RobjKlass.
3908 __ load_klass(RobjKlass, R17_tos);
3909 // Generate a fast subtype check. Branch to cast_ok if no failure. Return 0 if failure.
3910 __ li(R17_tos, 1);
3911 __ gen_subtype_check(RobjKlass, RspecifiedKlass, /*3 temp regs*/ Roffset, Rcpool, Rtags, /*target if subtype*/ Ldone);
3912 __ li(R17_tos, 0);
3913
3914 if (ProfileInterpreter) {
3915 __ b(Ldone);
3916 }
3917
3918 // Profile the null case.
3919 __ align(32, 12);
3920 __ bind(Lis_null);
3921 __ profile_null_seen(Rcpool, Rtags); // Rcpool and Rtags used as scratch.
3922
3923 __ align(32, 12);
3924 __ bind(Ldone);
3925 }
3926
3927 // =============================================================================
3928 // Breakpoints
3929
3930 void TemplateTable::_breakpoint() {
3931 transition(vtos, vtos);
3932
3933 // Get the unpatched byte code.
3934 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), R19_method, R14_bcp);
3935 __ mr(R31, R3_RET);
3936
3937 // Post the breakpoint event.
3938 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), R19_method, R14_bcp);
3939
3940 // Complete the execution of original bytecode.
3941 __ dispatch_Lbyte_code(vtos, R31, Interpreter::normal_table(vtos));
3942 }
3943
3944 // =============================================================================
3945 // Exceptions
3946
3947 void TemplateTable::athrow() {
3948 transition(atos, vtos);
3949
3950 // Exception oop is in tos
3951 __ verify_oop(R17_tos);
3952
3953 __ null_check_throw(R17_tos, -1, R11_scratch1);
3954
3955 // Throw exception interpreter entry expects exception oop to be in R3.
3956 __ mr(R3_RET, R17_tos);
3957 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::throw_exception_entry());
3958 __ mtctr(R11_scratch1);
3959 __ bctr();
3960 }
3961
3962 // =============================================================================
3963 // Synchronization
3964 // Searches the basic object lock list on the stack for a free slot
3965 // and uses it to lock the obect in tos.
3966 //
3967 // Recursive locking is enabled by exiting the search if the same
3968 // object is already found in the list. Thus, a new basic lock obj lock
3969 // is allocated "higher up" in the stack and thus is found first
3970 // at next monitor exit.
3971 void TemplateTable::monitorenter() {
3972 transition(atos, vtos);
3973
3974 __ verify_oop(R17_tos);
3975
3976 Register Rcurrent_monitor = R11_scratch1,
3977 Rcurrent_obj = R12_scratch2,
3978 Robj_to_lock = R17_tos,
3979 Rscratch1 = R3_ARG1,
3980 Rscratch2 = R4_ARG2,
3981 Rscratch3 = R5_ARG3,
3982 Rcurrent_obj_addr = R6_ARG4;
3983
3984 // ------------------------------------------------------------------------------
3985 // Null pointer exception.
3986 __ null_check_throw(Robj_to_lock, -1, R11_scratch1);
3987
3988 // Try to acquire a lock on the object.
3989 // Repeat until succeeded (i.e., until monitorenter returns true).
3990
3991 // ------------------------------------------------------------------------------
3992 // Find a free slot in the monitor block.
3993 Label Lfound, Lexit, Lallocate_new;
3994 ConditionRegister found_free_slot = CCR0,
3995 found_same_obj = CCR1,
3996 reached_limit = CCR6;
3997 {
3998 Label Lloop, Lentry;
3999 Register Rlimit = Rcurrent_monitor;
4000
4001 // Set up search loop - start with topmost monitor.
4002 __ add(Rcurrent_obj_addr, BasicObjectLock::obj_offset_in_bytes(), R26_monitor);
4003
4004 __ ld(Rlimit, 0, R1_SP);
4005 __ addi(Rlimit, Rlimit, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes() - BasicObjectLock::obj_offset_in_bytes())); // Monitor base
4006
4007 // Check if any slot is present => short cut to allocation if not.
4008 __ cmpld(reached_limit, Rcurrent_obj_addr, Rlimit);
4009 __ bgt(reached_limit, Lallocate_new);
4010
4011 // Pre-load topmost slot.
4012 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4013 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4014 // The search loop.
4015 __ bind(Lloop);
4016 // Found free slot?
4017 __ cmpdi(found_free_slot, Rcurrent_obj, 0);
4018 // Is this entry for same obj? If so, stop the search and take the found
4019 // free slot or allocate a new one to enable recursive locking.
4020 __ cmpd(found_same_obj, Rcurrent_obj, Robj_to_lock);
4021 __ cmpld(reached_limit, Rcurrent_obj_addr, Rlimit);
4022 __ beq(found_free_slot, Lexit);
4023 __ beq(found_same_obj, Lallocate_new);
4024 __ bgt(reached_limit, Lallocate_new);
4025 // Check if last allocated BasicLockObj reached.
4026 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4027 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4028 // Next iteration if unchecked BasicObjectLocks exist on the stack.
4029 __ b(Lloop);
4030 }
4031
4032 // ------------------------------------------------------------------------------
4033 // Check if we found a free slot.
4034 __ bind(Lexit);
4035
4036 __ addi(Rcurrent_monitor, Rcurrent_obj_addr, -(frame::interpreter_frame_monitor_size() * wordSize) - BasicObjectLock::obj_offset_in_bytes());
4037 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, - frame::interpreter_frame_monitor_size() * wordSize);
4038 __ b(Lfound);
4039
4040 // We didn't find a free BasicObjLock => allocate one.
4041 __ align(32, 12);
4042 __ bind(Lallocate_new);
4043 __ add_monitor_to_stack(false, Rscratch1, Rscratch2);
4044 __ mr(Rcurrent_monitor, R26_monitor);
4045 __ addi(Rcurrent_obj_addr, R26_monitor, BasicObjectLock::obj_offset_in_bytes());
4046
4047 // ------------------------------------------------------------------------------
4048 // We now have a slot to lock.
4049 __ bind(Lfound);
4050
4051 // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
4052 // The object has already been poped from the stack, so the expression stack looks correct.
4053 __ addi(R14_bcp, R14_bcp, 1);
4054
4055 __ std(Robj_to_lock, 0, Rcurrent_obj_addr);
4056 __ lock_object(Rcurrent_monitor, Robj_to_lock);
4057
4058 // Check if there's enough space on the stack for the monitors after locking.
4059 Label Lskip_stack_check;
4060 // Optimization: If the monitors stack section is less then a std page size (4K) don't run
4061 // the stack check. There should be enough shadow pages to fit that in.
4062 __ ld(Rscratch3, 0, R1_SP);
4063 __ sub(Rscratch3, Rscratch3, R26_monitor);
4064 __ cmpdi(CCR0, Rscratch3, 4*K);
4065 __ blt(CCR0, Lskip_stack_check);
4066
4067 DEBUG_ONLY(__ untested("stack overflow check during monitor enter");)
4068 __ li(Rscratch1, 0);
4069 __ generate_stack_overflow_check_with_compare_and_throw(Rscratch1, Rscratch2);
4070
4071 __ align(32, 12);
4072 __ bind(Lskip_stack_check);
4073
4074 // The bcp has already been incremented. Just need to dispatch to next instruction.
4075 __ dispatch_next(vtos);
4076 }
4077
4078 void TemplateTable::monitorexit() {
4079 transition(atos, vtos);
4080 __ verify_oop(R17_tos);
4081
4082 Register Rcurrent_monitor = R11_scratch1,
4083 Rcurrent_obj = R12_scratch2,
4084 Robj_to_lock = R17_tos,
4085 Rcurrent_obj_addr = R3_ARG1,
4086 Rlimit = R4_ARG2;
4087 Label Lfound, Lillegal_monitor_state;
4088
4089 // Check corner case: unbalanced monitorEnter / Exit.
4090 __ ld(Rlimit, 0, R1_SP);
4091 __ addi(Rlimit, Rlimit, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base
4092
4093 // Null pointer check.
4094 __ null_check_throw(Robj_to_lock, -1, R11_scratch1);
4095
4096 __ cmpld(CCR0, R26_monitor, Rlimit);
4097 __ bgt(CCR0, Lillegal_monitor_state);
4098
4099 // Find the corresponding slot in the monitors stack section.
4100 {
4101 Label Lloop;
4102
4103 // Start with topmost monitor.
4104 __ addi(Rcurrent_obj_addr, R26_monitor, BasicObjectLock::obj_offset_in_bytes());
4105 __ addi(Rlimit, Rlimit, BasicObjectLock::obj_offset_in_bytes());
4106 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4107 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4108
4109 __ bind(Lloop);
4110 // Is this entry for same obj?
4111 __ cmpd(CCR0, Rcurrent_obj, Robj_to_lock);
4112 __ beq(CCR0, Lfound);
4113
4114 // Check if last allocated BasicLockObj reached.
4115
4116 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4117 __ cmpld(CCR0, Rcurrent_obj_addr, Rlimit);
4118 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4119
4120 // Next iteration if unchecked BasicObjectLocks exist on the stack.
4121 __ ble(CCR0, Lloop);
4122 }
4123
4124 // Fell through without finding the basic obj lock => throw up!
4125 __ bind(Lillegal_monitor_state);
4126 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
4127 __ should_not_reach_here();
4128
4129 __ align(32, 12);
4130 __ bind(Lfound);
4131 __ addi(Rcurrent_monitor, Rcurrent_obj_addr,
4132 -(frame::interpreter_frame_monitor_size() * wordSize) - BasicObjectLock::obj_offset_in_bytes());
4133 __ unlock_object(Rcurrent_monitor);
4134 }
4135
4136 // ============================================================================
4137 // Wide bytecodes
4138
4139 // Wide instructions. Simply redirects to the wide entry point for that instruction.
4140 void TemplateTable::wide() {
4141 transition(vtos, vtos);
4142
4143 const Register Rtable = R11_scratch1,
4144 Rindex = R12_scratch2,
4145 Rtmp = R0;
4146
4147 __ lbz(Rindex, 1, R14_bcp);
4148
4149 __ load_dispatch_table(Rtable, Interpreter::_wentry_point);
4150
4151 __ slwi(Rindex, Rindex, LogBytesPerWord);
4152 __ ldx(Rtmp, Rtable, Rindex);
4153 __ mtctr(Rtmp);
4154 __ bctr();
4155 // Note: the bcp increment step is part of the individual wide bytecode implementations.
4156 }
4157 #endif // !CC_INTERP