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