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