1 /*
2 * Copyright (c) 2014, Oracle and/or its affiliates. All rights reserved.
3 * Copyright 2013, 2014 SAP AG. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "precompiled.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "interpreter/interpreterRuntime.hpp"
30 #include "interpreter/templateInterpreter.hpp"
31 #include "interpreter/templateTable.hpp"
32 #include "memory/universe.inline.hpp"
33 #include "oops/objArrayKlass.hpp"
34 #include "oops/oop.inline.hpp"
35 #include "prims/methodHandles.hpp"
36 #include "runtime/sharedRuntime.hpp"
37 #include "runtime/stubRoutines.hpp"
38 #include "runtime/synchronizer.hpp"
39 #include "utilities/macros.hpp"
40
41 #ifndef CC_INTERP
42
43 #undef __
44 #define __ _masm->
45
46 // ============================================================================
47 // Misc helpers
48
49 // Do an oop store like *(base + index) = val OR *(base + offset) = val
50 // (only one of both variants is possible at the same time).
51 // Index can be noreg.
52 // Kills:
53 // Rbase, Rtmp
54 static void do_oop_store(InterpreterMacroAssembler* _masm,
55 Register Rbase,
56 RegisterOrConstant offset,
57 Register Rval, // Noreg means always null.
58 Register Rtmp1,
59 Register Rtmp2,
60 Register Rtmp3,
61 BarrierSet::Name barrier,
62 bool precise,
63 bool check_null) {
64 assert_different_registers(Rtmp1, Rtmp2, Rtmp3, Rval, Rbase);
65
66 switch (barrier) {
67 #if INCLUDE_ALL_GCS
68 case BarrierSet::G1SATBCT:
69 case BarrierSet::G1SATBCTLogging:
70 {
71 // Load and record the previous value.
72 __ g1_write_barrier_pre(Rbase, offset,
73 Rtmp3, /* holder of pre_val ? */
74 Rtmp1, Rtmp2, false /* frame */);
75
76 Label Lnull, Ldone;
77 if (Rval != noreg) {
78 if (check_null) {
79 __ cmpdi(CCR0, Rval, 0);
80 __ beq(CCR0, Lnull);
81 }
82 __ store_heap_oop_not_null(Rval, offset, Rbase, /*Rval must stay uncompressed.*/ Rtmp1);
83 // Mark the card.
84 if (!(offset.is_constant() && offset.as_constant() == 0) && precise) {
85 __ add(Rbase, offset, Rbase);
86 }
87 __ g1_write_barrier_post(Rbase, Rval, Rtmp1, Rtmp2, Rtmp3, /*filtered (fast path)*/ &Ldone);
88 if (check_null) { __ b(Ldone); }
89 }
90
91 if (Rval == noreg || check_null) { // Store null oop.
92 Register Rnull = Rval;
93 __ bind(Lnull);
94 if (Rval == noreg) {
95 Rnull = Rtmp1;
96 __ li(Rnull, 0);
97 }
98 if (UseCompressedOops) {
99 __ stw(Rnull, offset, Rbase);
100 } else {
101 __ std(Rnull, offset, Rbase);
102 }
103 }
104 __ bind(Ldone);
105 }
106 break;
107 #endif // INCLUDE_ALL_GCS
108 case BarrierSet::CardTableModRef:
109 case BarrierSet::CardTableExtension:
110 {
111 Label Lnull, Ldone;
112 if (Rval != noreg) {
113 if (check_null) {
114 __ cmpdi(CCR0, Rval, 0);
115 __ beq(CCR0, Lnull);
116 }
117 __ store_heap_oop_not_null(Rval, offset, Rbase, /*Rval should better stay uncompressed.*/ Rtmp1);
118 // Mark the card.
119 if (!(offset.is_constant() && offset.as_constant() == 0) && precise) {
120 __ add(Rbase, offset, Rbase);
121 }
122 __ card_write_barrier_post(Rbase, Rval, Rtmp1);
123 if (check_null) {
124 __ b(Ldone);
125 }
126 }
127
128 if (Rval == noreg || check_null) { // Store null oop.
129 Register Rnull = Rval;
130 __ bind(Lnull);
131 if (Rval == noreg) {
132 Rnull = Rtmp1;
133 __ li(Rnull, 0);
134 }
135 if (UseCompressedOops) {
136 __ stw(Rnull, offset, Rbase);
137 } else {
138 __ std(Rnull, offset, Rbase);
139 }
140 }
141 __ bind(Ldone);
142 }
143 break;
144 case BarrierSet::ModRef:
145 case BarrierSet::Other:
146 ShouldNotReachHere();
147 break;
148 default:
149 ShouldNotReachHere();
150 }
151 }
152
153 // ============================================================================
154 // Platform-dependent initialization
155
156 void TemplateTable::pd_initialize() {
157 // No ppc64 specific initialization.
158 }
159
160 Address TemplateTable::at_bcp(int offset) {
161 // Not used on ppc.
162 ShouldNotReachHere();
163 return Address();
164 }
165
166 // Patches the current bytecode (ptr to it located in bcp)
167 // in the bytecode stream with a new one.
168 void TemplateTable::patch_bytecode(Bytecodes::Code new_bc, Register Rnew_bc, Register Rtemp, bool load_bc_into_bc_reg /*=true*/, int byte_no) {
169 // With sharing on, may need to test method flag.
170 if (!RewriteBytecodes) return;
171 Label L_patch_done;
172
173 switch (new_bc) {
174 case Bytecodes::_fast_aputfield:
175 case Bytecodes::_fast_bputfield:
176 case Bytecodes::_fast_cputfield:
177 case Bytecodes::_fast_dputfield:
178 case Bytecodes::_fast_fputfield:
179 case Bytecodes::_fast_iputfield:
180 case Bytecodes::_fast_lputfield:
181 case Bytecodes::_fast_sputfield:
182 {
183 // We skip bytecode quickening for putfield instructions when
184 // the put_code written to the constant pool cache is zero.
185 // This is required so that every execution of this instruction
186 // calls out to InterpreterRuntime::resolve_get_put to do
187 // additional, required work.
188 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
189 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
190 __ get_cache_and_index_at_bcp(Rtemp /* dst = cache */, 1);
191 // Big Endian: ((*(cache+indices))>>((1+byte_no)*8))&0xFF
192 __ lbz(Rnew_bc, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 7 - (1 + byte_no), Rtemp);
193 __ cmpwi(CCR0, Rnew_bc, 0);
194 __ li(Rnew_bc, (unsigned int)(unsigned char)new_bc);
195 __ beq(CCR0, L_patch_done);
196 // __ isync(); // acquire not needed
197 break;
198 }
199
200 default:
201 assert(byte_no == -1, "sanity");
202 if (load_bc_into_bc_reg) {
203 __ li(Rnew_bc, (unsigned int)(unsigned char)new_bc);
204 }
205 }
206
207 if (JvmtiExport::can_post_breakpoint()) {
208 Label L_fast_patch;
209 __ lbz(Rtemp, 0, R14_bcp);
210 __ cmpwi(CCR0, Rtemp, (unsigned int)(unsigned char)Bytecodes::_breakpoint);
211 __ bne(CCR0, L_fast_patch);
212 // Perform the quickening, slowly, in the bowels of the breakpoint table.
213 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), R19_method, R14_bcp, Rnew_bc);
214 __ b(L_patch_done);
215 __ bind(L_fast_patch);
216 }
217
218 // Patch bytecode.
219 __ stb(Rnew_bc, 0, R14_bcp);
220
221 __ bind(L_patch_done);
222 }
223
224 // ============================================================================
225 // Individual instructions
226
227 void TemplateTable::nop() {
228 transition(vtos, vtos);
229 // Nothing to do.
230 }
231
232 void TemplateTable::shouldnotreachhere() {
233 transition(vtos, vtos);
234 __ stop("shouldnotreachhere bytecode");
235 }
236
237 void TemplateTable::aconst_null() {
238 transition(vtos, atos);
239 __ li(R17_tos, 0);
240 }
241
242 void TemplateTable::iconst(int value) {
243 transition(vtos, itos);
244 assert(value >= -1 && value <= 5, "");
245 __ li(R17_tos, value);
246 }
247
248 void TemplateTable::lconst(int value) {
249 transition(vtos, ltos);
250 assert(value >= -1 && value <= 5, "");
251 __ li(R17_tos, value);
252 }
253
254 void TemplateTable::fconst(int value) {
255 transition(vtos, ftos);
256 static float zero = 0.0;
257 static float one = 1.0;
258 static float two = 2.0;
259 switch (value) {
260 default: ShouldNotReachHere();
261 case 0: {
262 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&zero, R0, true);
263 __ lfs(F15_ftos, simm16_offset, R11_scratch1);
264 break;
265 }
266 case 1: {
267 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&one, R0, true);
268 __ lfs(F15_ftos, simm16_offset, R11_scratch1);
269 break;
270 }
271 case 2: {
272 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&two, R0, true);
273 __ lfs(F15_ftos, simm16_offset, R11_scratch1);
274 break;
275 }
276 }
277 }
278
279 void TemplateTable::dconst(int value) {
280 transition(vtos, dtos);
281 static double zero = 0.0;
282 static double one = 1.0;
283 switch (value) {
284 case 0: {
285 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&zero, R0, true);
286 __ lfd(F15_ftos, simm16_offset, R11_scratch1);
287 break;
288 }
289 case 1: {
290 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&one, R0, true);
291 __ lfd(F15_ftos, simm16_offset, R11_scratch1);
292 break;
293 }
294 default: ShouldNotReachHere();
295 }
296 }
297
298 void TemplateTable::bipush() {
299 transition(vtos, itos);
300 __ lbz(R17_tos, 1, R14_bcp);
301 __ extsb(R17_tos, R17_tos);
302 }
303
304 void TemplateTable::sipush() {
305 transition(vtos, itos);
306 __ get_2_byte_integer_at_bcp(1, R17_tos, InterpreterMacroAssembler::Signed);
307 }
308
309 void TemplateTable::ldc(bool wide) {
310 Register Rscratch1 = R11_scratch1,
311 Rscratch2 = R12_scratch2,
312 Rcpool = R3_ARG1;
313
314 transition(vtos, vtos);
315 Label notInt, notClass, exit;
316
317 __ get_cpool_and_tags(Rcpool, Rscratch2); // Set Rscratch2 = &tags.
318 if (wide) { // Read index.
319 __ get_2_byte_integer_at_bcp(1, Rscratch1, InterpreterMacroAssembler::Unsigned);
320 } else {
321 __ lbz(Rscratch1, 1, R14_bcp);
322 }
323
324 const int base_offset = ConstantPool::header_size() * wordSize;
325 const int tags_offset = Array<u1>::base_offset_in_bytes();
326
327 // Get type from tags.
328 __ addi(Rscratch2, Rscratch2, tags_offset);
329 __ lbzx(Rscratch2, Rscratch2, Rscratch1);
330
331 __ cmpwi(CCR0, Rscratch2, JVM_CONSTANT_UnresolvedClass); // Unresolved class?
332 __ cmpwi(CCR1, Rscratch2, JVM_CONSTANT_UnresolvedClassInError); // Unresolved class in error state?
333 __ cror(/*CR0 eq*/2, /*CR1 eq*/4+2, /*CR0 eq*/2);
334
335 // Resolved class - need to call vm to get java mirror of the class.
336 __ cmpwi(CCR1, Rscratch2, JVM_CONSTANT_Class);
337 __ crnor(/*CR0 eq*/2, /*CR1 eq*/4+2, /*CR0 eq*/2); // Neither resolved class nor unresolved case from above?
338 __ beq(CCR0, notClass);
339
340 __ li(R4, wide ? 1 : 0);
341 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), R4);
342 __ push(atos);
343 __ b(exit);
344
345 __ align(32, 12);
346 __ bind(notClass);
347 __ addi(Rcpool, Rcpool, base_offset);
348 __ sldi(Rscratch1, Rscratch1, LogBytesPerWord);
349 __ cmpdi(CCR0, Rscratch2, JVM_CONSTANT_Integer);
350 __ bne(CCR0, notInt);
351 __ isync(); // Order load of constant wrt. tags.
352 __ lwax(R17_tos, Rcpool, Rscratch1);
353 __ push(itos);
354 __ b(exit);
355
356 __ align(32, 12);
357 __ bind(notInt);
358 #ifdef ASSERT
359 // String and Object are rewritten to fast_aldc
360 __ cmpdi(CCR0, Rscratch2, JVM_CONSTANT_Float);
361 __ asm_assert_eq("unexpected type", 0x8765);
362 #endif
363 __ isync(); // Order load of constant wrt. tags.
364 __ lfsx(F15_ftos, Rcpool, Rscratch1);
365 __ push(ftos);
366
367 __ align(32, 12);
368 __ bind(exit);
369 }
370
371 // Fast path for caching oop constants.
372 void TemplateTable::fast_aldc(bool wide) {
373 transition(vtos, atos);
374
375 int index_size = wide ? sizeof(u2) : sizeof(u1);
376 const Register Rscratch = R11_scratch1;
377 Label resolved;
378
379 // We are resolved if the resolved reference cache entry contains a
380 // non-null object (CallSite, etc.)
381 __ get_cache_index_at_bcp(Rscratch, 1, index_size); // Load index.
382 __ load_resolved_reference_at_index(R17_tos, Rscratch);
383 __ cmpdi(CCR0, R17_tos, 0);
384 __ bne(CCR0, resolved);
385 __ load_const_optimized(R3_ARG1, (int)bytecode());
386
387 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
388
389 // First time invocation - must resolve first.
390 __ call_VM(R17_tos, entry, R3_ARG1);
391
392 __ align(32, 12);
393 __ bind(resolved);
394 __ verify_oop(R17_tos);
395 }
396
397 void TemplateTable::ldc2_w() {
398 transition(vtos, vtos);
399 Label Llong, Lexit;
400
401 Register Rindex = R11_scratch1,
402 Rcpool = R12_scratch2,
403 Rtag = R3_ARG1;
404 __ get_cpool_and_tags(Rcpool, Rtag);
405 __ get_2_byte_integer_at_bcp(1, Rindex, InterpreterMacroAssembler::Unsigned);
406
407 const int base_offset = ConstantPool::header_size() * wordSize;
408 const int tags_offset = Array<u1>::base_offset_in_bytes();
409 // Get type from tags.
410 __ addi(Rcpool, Rcpool, base_offset);
411 __ addi(Rtag, Rtag, tags_offset);
412
413 __ lbzx(Rtag, Rtag, Rindex);
414
415 __ sldi(Rindex, Rindex, LogBytesPerWord);
416 __ cmpdi(CCR0, Rtag, JVM_CONSTANT_Double);
417 __ bne(CCR0, Llong);
418 // A double can be placed at word-aligned locations in the constant pool.
419 // Check out Conversions.java for an example.
420 // Also ConstantPool::header_size() is 20, which makes it very difficult
421 // to double-align double on the constant pool. SG, 11/7/97
422 __ isync(); // Order load of constant wrt. tags.
423 __ lfdx(F15_ftos, Rcpool, Rindex);
424 __ push(dtos);
425 __ b(Lexit);
426
427 __ bind(Llong);
428 __ isync(); // Order load of constant wrt. tags.
429 __ ldx(R17_tos, Rcpool, Rindex);
430 __ push(ltos);
431
432 __ bind(Lexit);
433 }
434
435 // Get the locals index located in the bytecode stream at bcp + offset.
436 void TemplateTable::locals_index(Register Rdst, int offset) {
437 __ lbz(Rdst, offset, R14_bcp);
438 }
439
440 void TemplateTable::iload() {
441 transition(vtos, itos);
442
443 // Get the local value into tos
444 const Register Rindex = R22_tmp2;
445 locals_index(Rindex);
446
447 // Rewrite iload,iload pair into fast_iload2
448 // iload,caload pair into fast_icaload
449 if (RewriteFrequentPairs) {
450 Label Lrewrite, Ldone;
451 Register Rnext_byte = R3_ARG1,
452 Rrewrite_to = R6_ARG4,
453 Rscratch = R11_scratch1;
454
455 // get next byte
456 __ lbz(Rnext_byte, Bytecodes::length_for(Bytecodes::_iload), R14_bcp);
457
458 // if _iload, wait to rewrite to iload2. We only want to rewrite the
459 // last two iloads in a pair. Comparing against fast_iload means that
460 // the next bytecode is neither an iload or a caload, and therefore
461 // an iload pair.
462 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_iload);
463 __ beq(CCR0, Ldone);
464
465 __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_iload);
466 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iload2);
467 __ beq(CCR1, Lrewrite);
468
469 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_caload);
470 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_icaload);
471 __ beq(CCR0, Lrewrite);
472
473 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iload);
474
475 __ bind(Lrewrite);
476 patch_bytecode(Bytecodes::_iload, Rrewrite_to, Rscratch, false);
477 __ bind(Ldone);
478 }
479
480 __ load_local_int(R17_tos, Rindex, Rindex);
481 }
482
483 // Load 2 integers in a row without dispatching
484 void TemplateTable::fast_iload2() {
485 transition(vtos, itos);
486
487 __ lbz(R3_ARG1, 1, R14_bcp);
488 __ lbz(R17_tos, Bytecodes::length_for(Bytecodes::_iload) + 1, R14_bcp);
489
490 __ load_local_int(R3_ARG1, R11_scratch1, R3_ARG1);
491 __ load_local_int(R17_tos, R12_scratch2, R17_tos);
492 __ push_i(R3_ARG1);
493 }
494
495 void TemplateTable::fast_iload() {
496 transition(vtos, itos);
497 // Get the local value into tos
498
499 const Register Rindex = R11_scratch1;
500 locals_index(Rindex);
501 __ load_local_int(R17_tos, Rindex, Rindex);
502 }
503
504 // Load a local variable type long from locals area to TOS cache register.
505 // Local index resides in bytecodestream.
506 void TemplateTable::lload() {
507 transition(vtos, ltos);
508
509 const Register Rindex = R11_scratch1;
510 locals_index(Rindex);
511 __ load_local_long(R17_tos, Rindex, Rindex);
512 }
513
514 void TemplateTable::fload() {
515 transition(vtos, ftos);
516
517 const Register Rindex = R11_scratch1;
518 locals_index(Rindex);
519 __ load_local_float(F15_ftos, Rindex, Rindex);
520 }
521
522 void TemplateTable::dload() {
523 transition(vtos, dtos);
524
525 const Register Rindex = R11_scratch1;
526 locals_index(Rindex);
527 __ load_local_double(F15_ftos, Rindex, Rindex);
528 }
529
530 void TemplateTable::aload() {
531 transition(vtos, atos);
532
533 const Register Rindex = R11_scratch1;
534 locals_index(Rindex);
535 __ load_local_ptr(R17_tos, Rindex, Rindex);
536 }
537
538 void TemplateTable::locals_index_wide(Register Rdst) {
539 // Offset is 2, not 1, because Lbcp points to wide prefix code.
540 __ get_2_byte_integer_at_bcp(2, Rdst, InterpreterMacroAssembler::Unsigned);
541 }
542
543 void TemplateTable::wide_iload() {
544 // Get the local value into tos.
545
546 const Register Rindex = R11_scratch1;
547 locals_index_wide(Rindex);
548 __ load_local_int(R17_tos, Rindex, Rindex);
549 }
550
551 void TemplateTable::wide_lload() {
552 transition(vtos, ltos);
553
554 const Register Rindex = R11_scratch1;
555 locals_index_wide(Rindex);
556 __ load_local_long(R17_tos, Rindex, Rindex);
557 }
558
559 void TemplateTable::wide_fload() {
560 transition(vtos, ftos);
561
562 const Register Rindex = R11_scratch1;
563 locals_index_wide(Rindex);
564 __ load_local_float(F15_ftos, Rindex, Rindex);
565 }
566
567 void TemplateTable::wide_dload() {
568 transition(vtos, dtos);
569
570 const Register Rindex = R11_scratch1;
571 locals_index_wide(Rindex);
572 __ load_local_double(F15_ftos, Rindex, Rindex);
573 }
574
575 void TemplateTable::wide_aload() {
576 transition(vtos, atos);
577
578 const Register Rindex = R11_scratch1;
579 locals_index_wide(Rindex);
580 __ load_local_ptr(R17_tos, Rindex, Rindex);
581 }
582
583 void TemplateTable::iaload() {
584 transition(itos, itos);
585
586 const Register Rload_addr = R3_ARG1,
587 Rarray = R4_ARG2,
588 Rtemp = R5_ARG3;
589 __ index_check(Rarray, R17_tos /* index */, LogBytesPerInt, Rtemp, Rload_addr);
590 __ lwa(R17_tos, arrayOopDesc::base_offset_in_bytes(T_INT), Rload_addr);
591 }
592
593 void TemplateTable::laload() {
594 transition(itos, ltos);
595
596 const Register Rload_addr = R3_ARG1,
597 Rarray = R4_ARG2,
598 Rtemp = R5_ARG3;
599 __ index_check(Rarray, R17_tos /* index */, LogBytesPerLong, Rtemp, Rload_addr);
600 __ ld(R17_tos, arrayOopDesc::base_offset_in_bytes(T_LONG), Rload_addr);
601 }
602
603 void TemplateTable::faload() {
604 transition(itos, ftos);
605
606 const Register Rload_addr = R3_ARG1,
607 Rarray = R4_ARG2,
608 Rtemp = R5_ARG3;
609 __ index_check(Rarray, R17_tos /* index */, LogBytesPerInt, Rtemp, Rload_addr);
610 __ lfs(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Rload_addr);
611 }
612
613 void TemplateTable::daload() {
614 transition(itos, dtos);
615
616 const Register Rload_addr = R3_ARG1,
617 Rarray = R4_ARG2,
618 Rtemp = R5_ARG3;
619 __ index_check(Rarray, R17_tos /* index */, LogBytesPerLong, Rtemp, Rload_addr);
620 __ lfd(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Rload_addr);
621 }
622
623 void TemplateTable::aaload() {
624 transition(itos, atos);
625
626 // tos: index
627 // result tos: array
628 const Register Rload_addr = R3_ARG1,
629 Rarray = R4_ARG2,
630 Rtemp = R5_ARG3;
631 __ index_check(Rarray, R17_tos /* index */, UseCompressedOops ? 2 : LogBytesPerWord, Rtemp, Rload_addr);
632 __ load_heap_oop(R17_tos, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Rload_addr);
633 __ verify_oop(R17_tos);
634 //__ dcbt(R17_tos); // prefetch
635 }
636
637 void TemplateTable::baload() {
638 transition(itos, itos);
639
640 const Register Rload_addr = R3_ARG1,
641 Rarray = R4_ARG2,
642 Rtemp = R5_ARG3;
643 __ index_check(Rarray, R17_tos /* index */, 0, Rtemp, Rload_addr);
644 __ lbz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_BYTE), Rload_addr);
645 __ extsb(R17_tos, R17_tos);
646 }
647
648 void TemplateTable::caload() {
649 transition(itos, itos);
650
651 const Register Rload_addr = R3_ARG1,
652 Rarray = R4_ARG2,
653 Rtemp = R5_ARG3;
654 __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
655 __ lhz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rload_addr);
656 }
657
658 // Iload followed by caload frequent pair.
659 void TemplateTable::fast_icaload() {
660 transition(vtos, itos);
661
662 const Register Rload_addr = R3_ARG1,
663 Rarray = R4_ARG2,
664 Rtemp = R11_scratch1;
665
666 locals_index(R17_tos);
667 __ load_local_int(R17_tos, Rtemp, R17_tos);
668 __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
669 __ lhz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rload_addr);
670 }
671
672 void TemplateTable::saload() {
673 transition(itos, itos);
674
675 const Register Rload_addr = R11_scratch1,
676 Rarray = R12_scratch2,
677 Rtemp = R3_ARG1;
678 __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
679 __ lha(R17_tos, arrayOopDesc::base_offset_in_bytes(T_SHORT), Rload_addr);
680 }
681
682 void TemplateTable::iload(int n) {
683 transition(vtos, itos);
684
685 __ lwz(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
686 }
687
688 void TemplateTable::lload(int n) {
689 transition(vtos, ltos);
690
691 __ ld(R17_tos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
692 }
693
694 void TemplateTable::fload(int n) {
695 transition(vtos, ftos);
696
697 __ lfs(F15_ftos, Interpreter::local_offset_in_bytes(n), R18_locals);
698 }
699
700 void TemplateTable::dload(int n) {
701 transition(vtos, dtos);
702
703 __ lfd(F15_ftos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
704 }
705
706 void TemplateTable::aload(int n) {
707 transition(vtos, atos);
708
709 __ ld(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
710 }
711
712 void TemplateTable::aload_0() {
713 transition(vtos, atos);
714 // According to bytecode histograms, the pairs:
715 //
716 // _aload_0, _fast_igetfield
717 // _aload_0, _fast_agetfield
718 // _aload_0, _fast_fgetfield
719 //
720 // occur frequently. If RewriteFrequentPairs is set, the (slow)
721 // _aload_0 bytecode checks if the next bytecode is either
722 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
723 // rewrites the current bytecode into a pair bytecode; otherwise it
724 // rewrites the current bytecode into _0 that doesn't do
725 // the pair check anymore.
726 //
727 // Note: If the next bytecode is _getfield, the rewrite must be
728 // delayed, otherwise we may miss an opportunity for a pair.
729 //
730 // Also rewrite frequent pairs
731 // aload_0, aload_1
732 // aload_0, iload_1
733 // These bytecodes with a small amount of code are most profitable
734 // to rewrite.
735
736 if (RewriteFrequentPairs) {
737
738 Label Lrewrite, Ldont_rewrite;
739 Register Rnext_byte = R3_ARG1,
740 Rrewrite_to = R6_ARG4,
741 Rscratch = R11_scratch1;
742
743 // Get next byte.
744 __ lbz(Rnext_byte, Bytecodes::length_for(Bytecodes::_aload_0), R14_bcp);
745
746 // If _getfield, wait to rewrite. We only want to rewrite the last two bytecodes in a pair.
747 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_getfield);
748 __ beq(CCR0, Ldont_rewrite);
749
750 __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_igetfield);
751 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iaccess_0);
752 __ beq(CCR1, Lrewrite);
753
754 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_agetfield);
755 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_aaccess_0);
756 __ beq(CCR0, Lrewrite);
757
758 __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_fgetfield);
759 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_faccess_0);
760 __ beq(CCR1, Lrewrite);
761
762 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_aload_0);
763
764 __ bind(Lrewrite);
765 patch_bytecode(Bytecodes::_aload_0, Rrewrite_to, Rscratch, false);
766 __ bind(Ldont_rewrite);
767 }
768
769 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
770 aload(0);
771 }
772
773 void TemplateTable::istore() {
774 transition(itos, vtos);
775
776 const Register Rindex = R11_scratch1;
777 locals_index(Rindex);
778 __ store_local_int(R17_tos, Rindex);
779 }
780
781 void TemplateTable::lstore() {
782 transition(ltos, vtos);
783 const Register Rindex = R11_scratch1;
784 locals_index(Rindex);
785 __ store_local_long(R17_tos, Rindex);
786 }
787
788 void TemplateTable::fstore() {
789 transition(ftos, vtos);
790
791 const Register Rindex = R11_scratch1;
792 locals_index(Rindex);
793 __ store_local_float(F15_ftos, Rindex);
794 }
795
796 void TemplateTable::dstore() {
797 transition(dtos, vtos);
798
799 const Register Rindex = R11_scratch1;
800 locals_index(Rindex);
801 __ store_local_double(F15_ftos, Rindex);
802 }
803
804 void TemplateTable::astore() {
805 transition(vtos, vtos);
806
807 const Register Rindex = R11_scratch1;
808 __ pop_ptr();
809 __ verify_oop_or_return_address(R17_tos, Rindex);
810 locals_index(Rindex);
811 __ store_local_ptr(R17_tos, Rindex);
812 }
813
814 void TemplateTable::wide_istore() {
815 transition(vtos, vtos);
816
817 const Register Rindex = R11_scratch1;
818 __ pop_i();
819 locals_index_wide(Rindex);
820 __ store_local_int(R17_tos, Rindex);
821 }
822
823 void TemplateTable::wide_lstore() {
824 transition(vtos, vtos);
825
826 const Register Rindex = R11_scratch1;
827 __ pop_l();
828 locals_index_wide(Rindex);
829 __ store_local_long(R17_tos, Rindex);
830 }
831
832 void TemplateTable::wide_fstore() {
833 transition(vtos, vtos);
834
835 const Register Rindex = R11_scratch1;
836 __ pop_f();
837 locals_index_wide(Rindex);
838 __ store_local_float(F15_ftos, Rindex);
839 }
840
841 void TemplateTable::wide_dstore() {
842 transition(vtos, vtos);
843
844 const Register Rindex = R11_scratch1;
845 __ pop_d();
846 locals_index_wide(Rindex);
847 __ store_local_double(F15_ftos, Rindex);
848 }
849
850 void TemplateTable::wide_astore() {
851 transition(vtos, vtos);
852
853 const Register Rindex = R11_scratch1;
854 __ pop_ptr();
855 __ verify_oop_or_return_address(R17_tos, Rindex);
856 locals_index_wide(Rindex);
857 __ store_local_ptr(R17_tos, Rindex);
858 }
859
860 void TemplateTable::iastore() {
861 transition(itos, vtos);
862
863 const Register Rindex = R3_ARG1,
864 Rstore_addr = R4_ARG2,
865 Rarray = R5_ARG3,
866 Rtemp = R6_ARG4;
867 __ pop_i(Rindex);
868 __ index_check(Rarray, Rindex, LogBytesPerInt, Rtemp, Rstore_addr);
869 __ stw(R17_tos, arrayOopDesc::base_offset_in_bytes(T_INT), Rstore_addr);
870 }
871
872 void TemplateTable::lastore() {
873 transition(ltos, vtos);
874
875 const Register Rindex = R3_ARG1,
876 Rstore_addr = R4_ARG2,
877 Rarray = R5_ARG3,
878 Rtemp = R6_ARG4;
879 __ pop_i(Rindex);
880 __ index_check(Rarray, Rindex, LogBytesPerLong, Rtemp, Rstore_addr);
881 __ std(R17_tos, arrayOopDesc::base_offset_in_bytes(T_LONG), Rstore_addr);
882 }
883
884 void TemplateTable::fastore() {
885 transition(ftos, vtos);
886
887 const Register Rindex = R3_ARG1,
888 Rstore_addr = R4_ARG2,
889 Rarray = R5_ARG3,
890 Rtemp = R6_ARG4;
891 __ pop_i(Rindex);
892 __ index_check(Rarray, Rindex, LogBytesPerInt, Rtemp, Rstore_addr);
893 __ stfs(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Rstore_addr);
894 }
895
896 void TemplateTable::dastore() {
897 transition(dtos, vtos);
898
899 const Register Rindex = R3_ARG1,
900 Rstore_addr = R4_ARG2,
901 Rarray = R5_ARG3,
902 Rtemp = R6_ARG4;
903 __ pop_i(Rindex);
904 __ index_check(Rarray, Rindex, LogBytesPerLong, Rtemp, Rstore_addr);
905 __ stfd(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Rstore_addr);
906 }
907
908 // Pop 3 values from the stack and...
909 void TemplateTable::aastore() {
910 transition(vtos, vtos);
911
912 Label Lstore_ok, Lis_null, Ldone;
913 const Register Rindex = R3_ARG1,
914 Rarray = R4_ARG2,
915 Rscratch = R11_scratch1,
916 Rscratch2 = R12_scratch2,
917 Rarray_klass = R5_ARG3,
918 Rarray_element_klass = Rarray_klass,
919 Rvalue_klass = R6_ARG4,
920 Rstore_addr = R31; // Use register which survives VM call.
921
922 __ ld(R17_tos, Interpreter::expr_offset_in_bytes(0), R15_esp); // Get value to store.
923 __ lwz(Rindex, Interpreter::expr_offset_in_bytes(1), R15_esp); // Get index.
924 __ ld(Rarray, Interpreter::expr_offset_in_bytes(2), R15_esp); // Get array.
925
926 __ verify_oop(R17_tos);
927 __ index_check_without_pop(Rarray, Rindex, UseCompressedOops ? 2 : LogBytesPerWord, Rscratch, Rstore_addr);
928 // Rindex is dead!
929 Register Rscratch3 = Rindex;
930
931 // Do array store check - check for NULL value first.
932 __ cmpdi(CCR0, R17_tos, 0);
933 __ beq(CCR0, Lis_null);
934
935 __ load_klass(Rarray_klass, Rarray);
936 __ load_klass(Rvalue_klass, R17_tos);
937
938 // Do fast instanceof cache test.
939 __ ld(Rarray_element_klass, in_bytes(ObjArrayKlass::element_klass_offset()), Rarray_klass);
940
941 // Generate a fast subtype check. Branch to store_ok if no failure. Throw if failure.
942 __ gen_subtype_check(Rvalue_klass /*subklass*/, Rarray_element_klass /*superklass*/, Rscratch, Rscratch2, Rscratch3, Lstore_ok);
943
944 // Fell through: subtype check failed => throw an exception.
945 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArrayStoreException_entry);
946 __ mtctr(R11_scratch1);
947 __ bctr();
948
949 __ bind(Lis_null);
950 do_oop_store(_masm, Rstore_addr, arrayOopDesc::base_offset_in_bytes(T_OBJECT), noreg /* 0 */,
951 Rscratch, Rscratch2, Rscratch3, _bs->kind(), true /* precise */, false /* check_null */);
952 __ profile_null_seen(Rscratch, Rscratch2);
953 __ b(Ldone);
954
955 // Store is OK.
956 __ bind(Lstore_ok);
957 do_oop_store(_masm, Rstore_addr, arrayOopDesc::base_offset_in_bytes(T_OBJECT), R17_tos /* value */,
958 Rscratch, Rscratch2, Rscratch3, _bs->kind(), true /* precise */, false /* check_null */);
959
960 __ bind(Ldone);
961 // Adjust sp (pops array, index and value).
962 __ addi(R15_esp, R15_esp, 3 * Interpreter::stackElementSize);
963 }
964
965 void TemplateTable::bastore() {
966 transition(itos, vtos);
967
968 const Register Rindex = R11_scratch1,
969 Rarray = R12_scratch2,
970 Rscratch = R3_ARG1;
971 __ pop_i(Rindex);
972 // tos: val
973 // Rarray: array ptr (popped by index_check)
974 __ index_check(Rarray, Rindex, 0, Rscratch, Rarray);
975 __ stb(R17_tos, arrayOopDesc::base_offset_in_bytes(T_BYTE), Rarray);
976 }
977
978 void TemplateTable::castore() {
979 transition(itos, vtos);
980
981 const Register Rindex = R11_scratch1,
982 Rarray = R12_scratch2,
983 Rscratch = R3_ARG1;
984 __ pop_i(Rindex);
985 // tos: val
986 // Rarray: array ptr (popped by index_check)
987 __ index_check(Rarray, Rindex, LogBytesPerShort, Rscratch, Rarray);
988 __ sth(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rarray);
989 }
990
991 void TemplateTable::sastore() {
992 castore();
993 }
994
995 void TemplateTable::istore(int n) {
996 transition(itos, vtos);
997 __ stw(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
998 }
999
1000 void TemplateTable::lstore(int n) {
1001 transition(ltos, vtos);
1002 __ std(R17_tos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
1003 }
1004
1005 void TemplateTable::fstore(int n) {
1006 transition(ftos, vtos);
1007 __ stfs(F15_ftos, Interpreter::local_offset_in_bytes(n), R18_locals);
1008 }
1009
1010 void TemplateTable::dstore(int n) {
1011 transition(dtos, vtos);
1012 __ stfd(F15_ftos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
1013 }
1014
1015 void TemplateTable::astore(int n) {
1016 transition(vtos, vtos);
1017
1018 __ pop_ptr();
1019 __ verify_oop_or_return_address(R17_tos, R11_scratch1);
1020 __ std(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
1021 }
1022
1023 void TemplateTable::pop() {
1024 transition(vtos, vtos);
1025
1026 __ addi(R15_esp, R15_esp, Interpreter::stackElementSize);
1027 }
1028
1029 void TemplateTable::pop2() {
1030 transition(vtos, vtos);
1031
1032 __ addi(R15_esp, R15_esp, Interpreter::stackElementSize * 2);
1033 }
1034
1035 void TemplateTable::dup() {
1036 transition(vtos, vtos);
1037
1038 __ ld(R11_scratch1, Interpreter::stackElementSize, R15_esp);
1039 __ push_ptr(R11_scratch1);
1040 }
1041
1042 void TemplateTable::dup_x1() {
1043 transition(vtos, vtos);
1044
1045 Register Ra = R11_scratch1,
1046 Rb = R12_scratch2;
1047 // stack: ..., a, b
1048 __ ld(Rb, Interpreter::stackElementSize, R15_esp);
1049 __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1050 __ std(Rb, Interpreter::stackElementSize * 2, R15_esp);
1051 __ std(Ra, Interpreter::stackElementSize, R15_esp);
1052 __ push_ptr(Rb);
1053 // stack: ..., b, a, b
1054 }
1055
1056 void TemplateTable::dup_x2() {
1057 transition(vtos, vtos);
1058
1059 Register Ra = R11_scratch1,
1060 Rb = R12_scratch2,
1061 Rc = R3_ARG1;
1062
1063 // stack: ..., a, b, c
1064 __ ld(Rc, Interpreter::stackElementSize, R15_esp); // load c
1065 __ ld(Ra, Interpreter::stackElementSize * 3, R15_esp); // load a
1066 __ std(Rc, Interpreter::stackElementSize * 3, R15_esp); // store c in a
1067 __ ld(Rb, Interpreter::stackElementSize * 2, R15_esp); // load b
1068 // stack: ..., c, b, c
1069 __ std(Ra, Interpreter::stackElementSize * 2, R15_esp); // store a in b
1070 // stack: ..., c, a, c
1071 __ std(Rb, Interpreter::stackElementSize, R15_esp); // store b in c
1072 __ push_ptr(Rc); // push c
1073 // stack: ..., c, a, b, c
1074 }
1075
1076 void TemplateTable::dup2() {
1077 transition(vtos, vtos);
1078
1079 Register Ra = R11_scratch1,
1080 Rb = R12_scratch2;
1081 // stack: ..., a, b
1082 __ ld(Rb, Interpreter::stackElementSize, R15_esp);
1083 __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1084 __ push_2ptrs(Ra, Rb);
1085 // stack: ..., a, b, a, b
1086 }
1087
1088 void TemplateTable::dup2_x1() {
1089 transition(vtos, vtos);
1090
1091 Register Ra = R11_scratch1,
1092 Rb = R12_scratch2,
1093 Rc = R3_ARG1;
1094 // stack: ..., a, b, c
1095 __ ld(Rc, Interpreter::stackElementSize, R15_esp);
1096 __ ld(Rb, Interpreter::stackElementSize * 2, R15_esp);
1097 __ std(Rc, Interpreter::stackElementSize * 2, R15_esp);
1098 __ ld(Ra, Interpreter::stackElementSize * 3, R15_esp);
1099 __ std(Ra, Interpreter::stackElementSize, R15_esp);
1100 __ std(Rb, Interpreter::stackElementSize * 3, R15_esp);
1101 // stack: ..., b, c, a
1102 __ push_2ptrs(Rb, Rc);
1103 // stack: ..., b, c, a, b, c
1104 }
1105
1106 void TemplateTable::dup2_x2() {
1107 transition(vtos, vtos);
1108
1109 Register Ra = R11_scratch1,
1110 Rb = R12_scratch2,
1111 Rc = R3_ARG1,
1112 Rd = R4_ARG2;
1113 // stack: ..., a, b, c, d
1114 __ ld(Rb, Interpreter::stackElementSize * 3, R15_esp);
1115 __ ld(Rd, Interpreter::stackElementSize, R15_esp);
1116 __ std(Rb, Interpreter::stackElementSize, R15_esp); // store b in d
1117 __ std(Rd, Interpreter::stackElementSize * 3, R15_esp); // store d in b
1118 __ ld(Ra, Interpreter::stackElementSize * 4, R15_esp);
1119 __ ld(Rc, Interpreter::stackElementSize * 2, R15_esp);
1120 __ std(Ra, Interpreter::stackElementSize * 2, R15_esp); // store a in c
1121 __ std(Rc, Interpreter::stackElementSize * 4, R15_esp); // store c in a
1122 // stack: ..., c, d, a, b
1123 __ push_2ptrs(Rc, Rd);
1124 // stack: ..., c, d, a, b, c, d
1125 }
1126
1127 void TemplateTable::swap() {
1128 transition(vtos, vtos);
1129 // stack: ..., a, b
1130
1131 Register Ra = R11_scratch1,
1132 Rb = R12_scratch2;
1133 // stack: ..., a, b
1134 __ ld(Rb, Interpreter::stackElementSize, R15_esp);
1135 __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1136 __ std(Rb, Interpreter::stackElementSize * 2, R15_esp);
1137 __ std(Ra, Interpreter::stackElementSize, R15_esp);
1138 // stack: ..., b, a
1139 }
1140
1141 void TemplateTable::iop2(Operation op) {
1142 transition(itos, itos);
1143
1144 Register Rscratch = R11_scratch1;
1145
1146 __ pop_i(Rscratch);
1147 // tos = number of bits to shift
1148 // Rscratch = value to shift
1149 switch (op) {
1150 case add: __ add(R17_tos, Rscratch, R17_tos); break;
1151 case sub: __ sub(R17_tos, Rscratch, R17_tos); break;
1152 case mul: __ mullw(R17_tos, Rscratch, R17_tos); break;
1153 case _and: __ andr(R17_tos, Rscratch, R17_tos); break;
1154 case _or: __ orr(R17_tos, Rscratch, R17_tos); break;
1155 case _xor: __ xorr(R17_tos, Rscratch, R17_tos); break;
1156 case shl: __ rldicl(R17_tos, R17_tos, 0, 64-5); __ slw(R17_tos, Rscratch, R17_tos); break;
1157 case shr: __ rldicl(R17_tos, R17_tos, 0, 64-5); __ sraw(R17_tos, Rscratch, R17_tos); break;
1158 case ushr: __ rldicl(R17_tos, R17_tos, 0, 64-5); __ srw(R17_tos, Rscratch, R17_tos); break;
1159 default: ShouldNotReachHere();
1160 }
1161 }
1162
1163 void TemplateTable::lop2(Operation op) {
1164 transition(ltos, ltos);
1165
1166 Register Rscratch = R11_scratch1;
1167 __ pop_l(Rscratch);
1168 switch (op) {
1169 case add: __ add(R17_tos, Rscratch, R17_tos); break;
1170 case sub: __ sub(R17_tos, Rscratch, R17_tos); break;
1171 case _and: __ andr(R17_tos, Rscratch, R17_tos); break;
1172 case _or: __ orr(R17_tos, Rscratch, R17_tos); break;
1173 case _xor: __ xorr(R17_tos, Rscratch, R17_tos); break;
1174 default: ShouldNotReachHere();
1175 }
1176 }
1177
1178 void TemplateTable::idiv() {
1179 transition(itos, itos);
1180
1181 Label Lnormal, Lexception, Ldone;
1182 Register Rdividend = R11_scratch1; // Used by irem.
1183
1184 __ addi(R0, R17_tos, 1);
1185 __ cmplwi(CCR0, R0, 2);
1186 __ bgt(CCR0, Lnormal); // divisor <-1 or >1
1187
1188 __ cmpwi(CCR1, R17_tos, 0);
1189 __ beq(CCR1, Lexception); // divisor == 0
1190
1191 __ pop_i(Rdividend);
1192 __ mullw(R17_tos, Rdividend, R17_tos); // div by +/-1
1193 __ b(Ldone);
1194
1195 __ bind(Lexception);
1196 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArithmeticException_entry);
1197 __ mtctr(R11_scratch1);
1198 __ bctr();
1199
1200 __ align(32, 12);
1201 __ bind(Lnormal);
1202 __ pop_i(Rdividend);
1203 __ divw(R17_tos, Rdividend, R17_tos); // Can't divide minint/-1.
1204 __ bind(Ldone);
1205 }
1206
1207 void TemplateTable::irem() {
1208 transition(itos, itos);
1209
1210 __ mr(R12_scratch2, R17_tos);
1211 idiv();
1212 __ mullw(R17_tos, R17_tos, R12_scratch2);
1213 __ subf(R17_tos, R17_tos, R11_scratch1); // Dividend set by idiv.
1214 }
1215
1216 void TemplateTable::lmul() {
1217 transition(ltos, ltos);
1218
1219 __ pop_l(R11_scratch1);
1220 __ mulld(R17_tos, R11_scratch1, R17_tos);
1221 }
1222
1223 void TemplateTable::ldiv() {
1224 transition(ltos, ltos);
1225
1226 Label Lnormal, Lexception, Ldone;
1227 Register Rdividend = R11_scratch1; // Used by lrem.
1228
1229 __ addi(R0, R17_tos, 1);
1230 __ cmpldi(CCR0, R0, 2);
1231 __ bgt(CCR0, Lnormal); // divisor <-1 or >1
1232
1233 __ cmpdi(CCR1, R17_tos, 0);
1234 __ beq(CCR1, Lexception); // divisor == 0
1235
1236 __ pop_l(Rdividend);
1237 __ mulld(R17_tos, Rdividend, R17_tos); // div by +/-1
1238 __ b(Ldone);
1239
1240 __ bind(Lexception);
1241 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArithmeticException_entry);
1242 __ mtctr(R11_scratch1);
1243 __ bctr();
1244
1245 __ align(32, 12);
1246 __ bind(Lnormal);
1247 __ pop_l(Rdividend);
1248 __ divd(R17_tos, Rdividend, R17_tos); // Can't divide minint/-1.
1249 __ bind(Ldone);
1250 }
1251
1252 void TemplateTable::lrem() {
1253 transition(ltos, ltos);
1254
1255 __ mr(R12_scratch2, R17_tos);
1256 ldiv();
1257 __ mulld(R17_tos, R17_tos, R12_scratch2);
1258 __ subf(R17_tos, R17_tos, R11_scratch1); // Dividend set by ldiv.
1259 }
1260
1261 void TemplateTable::lshl() {
1262 transition(itos, ltos);
1263
1264 __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1265 __ pop_l(R11_scratch1);
1266 __ sld(R17_tos, R11_scratch1, R17_tos);
1267 }
1268
1269 void TemplateTable::lshr() {
1270 transition(itos, ltos);
1271
1272 __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1273 __ pop_l(R11_scratch1);
1274 __ srad(R17_tos, R11_scratch1, R17_tos);
1275 }
1276
1277 void TemplateTable::lushr() {
1278 transition(itos, ltos);
1279
1280 __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1281 __ pop_l(R11_scratch1);
1282 __ srd(R17_tos, R11_scratch1, R17_tos);
1283 }
1284
1285 void TemplateTable::fop2(Operation op) {
1286 transition(ftos, ftos);
1287
1288 switch (op) {
1289 case add: __ pop_f(F0_SCRATCH); __ fadds(F15_ftos, F0_SCRATCH, F15_ftos); break;
1290 case sub: __ pop_f(F0_SCRATCH); __ fsubs(F15_ftos, F0_SCRATCH, F15_ftos); break;
1291 case mul: __ pop_f(F0_SCRATCH); __ fmuls(F15_ftos, F0_SCRATCH, F15_ftos); break;
1292 case div: __ pop_f(F0_SCRATCH); __ fdivs(F15_ftos, F0_SCRATCH, F15_ftos); break;
1293 case rem:
1294 __ pop_f(F1_ARG1);
1295 __ fmr(F2_ARG2, F15_ftos);
1296 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem));
1297 __ fmr(F15_ftos, F1_RET);
1298 break;
1299
1300 default: ShouldNotReachHere();
1301 }
1302 }
1303
1304 void TemplateTable::dop2(Operation op) {
1305 transition(dtos, dtos);
1306
1307 switch (op) {
1308 case add: __ pop_d(F0_SCRATCH); __ fadd(F15_ftos, F0_SCRATCH, F15_ftos); break;
1309 case sub: __ pop_d(F0_SCRATCH); __ fsub(F15_ftos, F0_SCRATCH, F15_ftos); break;
1310 case mul: __ pop_d(F0_SCRATCH); __ fmul(F15_ftos, F0_SCRATCH, F15_ftos); break;
1311 case div: __ pop_d(F0_SCRATCH); __ fdiv(F15_ftos, F0_SCRATCH, F15_ftos); break;
1312 case rem:
1313 __ pop_d(F1_ARG1);
1314 __ fmr(F2_ARG2, F15_ftos);
1315 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem));
1316 __ fmr(F15_ftos, F1_RET);
1317 break;
1318
1319 default: ShouldNotReachHere();
1320 }
1321 }
1322
1323 // Negate the value in the TOS cache.
1324 void TemplateTable::ineg() {
1325 transition(itos, itos);
1326
1327 __ neg(R17_tos, R17_tos);
1328 }
1329
1330 // Negate the value in the TOS cache.
1331 void TemplateTable::lneg() {
1332 transition(ltos, ltos);
1333
1334 __ neg(R17_tos, R17_tos);
1335 }
1336
1337 void TemplateTable::fneg() {
1338 transition(ftos, ftos);
1339
1340 __ fneg(F15_ftos, F15_ftos);
1341 }
1342
1343 void TemplateTable::dneg() {
1344 transition(dtos, dtos);
1345
1346 __ fneg(F15_ftos, F15_ftos);
1347 }
1348
1349 // Increments a local variable in place.
1350 void TemplateTable::iinc() {
1351 transition(vtos, vtos);
1352
1353 const Register Rindex = R11_scratch1,
1354 Rincrement = R0,
1355 Rvalue = R12_scratch2;
1356
1357 locals_index(Rindex); // Load locals index from bytecode stream.
1358 __ lbz(Rincrement, 2, R14_bcp); // Load increment from the bytecode stream.
1359 __ extsb(Rincrement, Rincrement);
1360
1361 __ load_local_int(Rvalue, Rindex, Rindex); // Puts address of local into Rindex.
1362
1363 __ add(Rvalue, Rincrement, Rvalue);
1364 __ stw(Rvalue, 0, Rindex);
1365 }
1366
1367 void TemplateTable::wide_iinc() {
1368 transition(vtos, vtos);
1369
1370 Register Rindex = R11_scratch1,
1371 Rlocals_addr = Rindex,
1372 Rincr = R12_scratch2;
1373 locals_index_wide(Rindex);
1374 __ get_2_byte_integer_at_bcp(4, Rincr, InterpreterMacroAssembler::Signed);
1375 __ load_local_int(R17_tos, Rlocals_addr, Rindex);
1376 __ add(R17_tos, Rincr, R17_tos);
1377 __ stw(R17_tos, 0, Rlocals_addr);
1378 }
1379
1380 void TemplateTable::convert() {
1381 // %%%%% Factor this first part accross platforms
1382 #ifdef ASSERT
1383 TosState tos_in = ilgl;
1384 TosState tos_out = ilgl;
1385 switch (bytecode()) {
1386 case Bytecodes::_i2l: // fall through
1387 case Bytecodes::_i2f: // fall through
1388 case Bytecodes::_i2d: // fall through
1389 case Bytecodes::_i2b: // fall through
1390 case Bytecodes::_i2c: // fall through
1391 case Bytecodes::_i2s: tos_in = itos; break;
1392 case Bytecodes::_l2i: // fall through
1393 case Bytecodes::_l2f: // fall through
1394 case Bytecodes::_l2d: tos_in = ltos; break;
1395 case Bytecodes::_f2i: // fall through
1396 case Bytecodes::_f2l: // fall through
1397 case Bytecodes::_f2d: tos_in = ftos; break;
1398 case Bytecodes::_d2i: // fall through
1399 case Bytecodes::_d2l: // fall through
1400 case Bytecodes::_d2f: tos_in = dtos; break;
1401 default : ShouldNotReachHere();
1402 }
1403 switch (bytecode()) {
1404 case Bytecodes::_l2i: // fall through
1405 case Bytecodes::_f2i: // fall through
1406 case Bytecodes::_d2i: // fall through
1407 case Bytecodes::_i2b: // fall through
1408 case Bytecodes::_i2c: // fall through
1409 case Bytecodes::_i2s: tos_out = itos; break;
1410 case Bytecodes::_i2l: // fall through
1411 case Bytecodes::_f2l: // fall through
1412 case Bytecodes::_d2l: tos_out = ltos; break;
1413 case Bytecodes::_i2f: // fall through
1414 case Bytecodes::_l2f: // fall through
1415 case Bytecodes::_d2f: tos_out = ftos; break;
1416 case Bytecodes::_i2d: // fall through
1417 case Bytecodes::_l2d: // fall through
1418 case Bytecodes::_f2d: tos_out = dtos; break;
1419 default : ShouldNotReachHere();
1420 }
1421 transition(tos_in, tos_out);
1422 #endif
1423
1424 // Conversion
1425 Label done;
1426 switch (bytecode()) {
1427 case Bytecodes::_i2l:
1428 __ extsw(R17_tos, R17_tos);
1429 break;
1430
1431 case Bytecodes::_l2i:
1432 // Nothing to do, we'll continue to work with the lower bits.
1433 break;
1434
1435 case Bytecodes::_i2b:
1436 __ extsb(R17_tos, R17_tos);
1437 break;
1438
1439 case Bytecodes::_i2c:
1440 __ rldicl(R17_tos, R17_tos, 0, 64-2*8);
1441 break;
1442
1443 case Bytecodes::_i2s:
1444 __ extsh(R17_tos, R17_tos);
1445 break;
1446
1447 case Bytecodes::_i2d:
1448 __ extsw(R17_tos, R17_tos);
1449 case Bytecodes::_l2d:
1450 __ push_l_pop_d();
1451 __ fcfid(F15_ftos, F15_ftos);
1452 break;
1453
1454 case Bytecodes::_i2f:
1455 __ extsw(R17_tos, R17_tos);
1456 __ push_l_pop_d();
1457 if (VM_Version::has_fcfids()) { // fcfids is >= Power7 only
1458 // Comment: alternatively, load with sign extend could be done by lfiwax.
1459 __ fcfids(F15_ftos, F15_ftos);
1460 } else {
1461 __ fcfid(F15_ftos, F15_ftos);
1462 __ frsp(F15_ftos, F15_ftos);
1463 }
1464 break;
1465
1466 case Bytecodes::_l2f:
1467 if (VM_Version::has_fcfids()) { // fcfids is >= Power7 only
1468 __ push_l_pop_d();
1469 __ fcfids(F15_ftos, F15_ftos);
1470 } else {
1471 // Avoid rounding problem when result should be 0x3f800001: need fixup code before fcfid+frsp.
1472 __ mr(R3_ARG1, R17_tos);
1473 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::l2f));
1474 __ fmr(F15_ftos, F1_RET);
1475 }
1476 break;
1477
1478 case Bytecodes::_f2d:
1479 // empty
1480 break;
1481
1482 case Bytecodes::_d2f:
1483 __ frsp(F15_ftos, F15_ftos);
1484 break;
1485
1486 case Bytecodes::_d2i:
1487 case Bytecodes::_f2i:
1488 __ fcmpu(CCR0, F15_ftos, F15_ftos);
1489 __ li(R17_tos, 0); // 0 in case of NAN
1490 __ bso(CCR0, done);
1491 __ fctiwz(F15_ftos, F15_ftos);
1492 __ push_d_pop_l();
1493 break;
1494
1495 case Bytecodes::_d2l:
1496 case Bytecodes::_f2l:
1497 __ fcmpu(CCR0, F15_ftos, F15_ftos);
1498 __ li(R17_tos, 0); // 0 in case of NAN
1499 __ bso(CCR0, done);
1500 __ fctidz(F15_ftos, F15_ftos);
1501 __ push_d_pop_l();
1502 break;
1503
1504 default: ShouldNotReachHere();
1505 }
1506 __ bind(done);
1507 }
1508
1509 // Long compare
1510 void TemplateTable::lcmp() {
1511 transition(ltos, itos);
1512
1513 const Register Rscratch = R11_scratch1;
1514 __ pop_l(Rscratch); // first operand, deeper in stack
1515
1516 __ cmpd(CCR0, Rscratch, R17_tos); // compare
1517 __ mfcr(R17_tos); // set bit 32..33 as follows: <: 0b10, =: 0b00, >: 0b01
1518 __ srwi(Rscratch, R17_tos, 30);
1519 __ srawi(R17_tos, R17_tos, 31);
1520 __ orr(R17_tos, Rscratch, R17_tos); // set result as follows: <: -1, =: 0, >: 1
1521 }
1522
1523 // fcmpl/fcmpg and dcmpl/dcmpg bytecodes
1524 // unordered_result == -1 => fcmpl or dcmpl
1525 // unordered_result == 1 => fcmpg or dcmpg
1526 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1527 const FloatRegister Rfirst = F0_SCRATCH,
1528 Rsecond = F15_ftos;
1529 const Register Rscratch = R11_scratch1;
1530
1531 if (is_float) {
1532 __ pop_f(Rfirst);
1533 } else {
1534 __ pop_d(Rfirst);
1535 }
1536
1537 Label Lunordered, Ldone;
1538 __ fcmpu(CCR0, Rfirst, Rsecond); // compare
1539 if (unordered_result) {
1540 __ bso(CCR0, Lunordered);
1541 }
1542 __ mfcr(R17_tos); // set bit 32..33 as follows: <: 0b10, =: 0b00, >: 0b01
1543 __ srwi(Rscratch, R17_tos, 30);
1544 __ srawi(R17_tos, R17_tos, 31);
1545 __ orr(R17_tos, Rscratch, R17_tos); // set result as follows: <: -1, =: 0, >: 1
1546 if (unordered_result) {
1547 __ b(Ldone);
1548 __ bind(Lunordered);
1549 __ load_const_optimized(R17_tos, unordered_result);
1550 }
1551 __ bind(Ldone);
1552 }
1553
1554 // Branch_conditional which takes TemplateTable::Condition.
1555 void TemplateTable::branch_conditional(ConditionRegister crx, TemplateTable::Condition cc, Label& L, bool invert) {
1556 bool positive = false;
1557 Assembler::Condition cond = Assembler::equal;
1558 switch (cc) {
1559 case TemplateTable::equal: positive = true ; cond = Assembler::equal ; break;
1560 case TemplateTable::not_equal: positive = false; cond = Assembler::equal ; break;
1561 case TemplateTable::less: positive = true ; cond = Assembler::less ; break;
1562 case TemplateTable::less_equal: positive = false; cond = Assembler::greater; break;
1563 case TemplateTable::greater: positive = true ; cond = Assembler::greater; break;
1564 case TemplateTable::greater_equal: positive = false; cond = Assembler::less ; break;
1565 default: ShouldNotReachHere();
1566 }
1567 int bo = (positive != invert) ? Assembler::bcondCRbiIs1 : Assembler::bcondCRbiIs0;
1568 int bi = Assembler::bi0(crx, cond);
1569 __ bc(bo, bi, L);
1570 }
1571
1572 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1573
1574 // Note: on SPARC, we use InterpreterMacroAssembler::if_cmp also.
1575 __ verify_thread();
1576
1577 const Register Rscratch1 = R11_scratch1,
1578 Rscratch2 = R12_scratch2,
1579 Rscratch3 = R3_ARG1,
1580 R4_counters = R4_ARG2,
1581 bumped_count = R31,
1582 Rdisp = R22_tmp2;
1583
1584 __ profile_taken_branch(Rscratch1, bumped_count);
1585
1586 // Get (wide) offset.
1587 if (is_wide) {
1588 __ get_4_byte_integer_at_bcp(1, Rdisp, InterpreterMacroAssembler::Signed);
1589 } else {
1590 __ get_2_byte_integer_at_bcp(1, Rdisp, InterpreterMacroAssembler::Signed);
1591 }
1592
1593 // --------------------------------------------------------------------------
1594 // Handle all the JSR stuff here, then exit.
1595 // It's much shorter and cleaner than intermingling with the
1596 // non-JSR normal-branch stuff occurring below.
1597 if (is_jsr) {
1598 // Compute return address as bci in Otos_i.
1599 __ ld(Rscratch1, in_bytes(Method::const_offset()), R19_method);
1600 __ addi(Rscratch2, R14_bcp, -in_bytes(ConstMethod::codes_offset()) + (is_wide ? 5 : 3));
1601 __ subf(R17_tos, Rscratch1, Rscratch2);
1602
1603 // Bump bcp to target of JSR.
1604 __ add(R14_bcp, Rdisp, R14_bcp);
1605 // Push returnAddress for "ret" on stack.
1606 __ push_ptr(R17_tos);
1607 // And away we go!
1608 __ dispatch_next(vtos);
1609 return;
1610 }
1611
1612 // --------------------------------------------------------------------------
1613 // Normal (non-jsr) branch handling
1614
1615 const bool increment_invocation_counter_for_backward_branches = UseCompiler && UseLoopCounter;
1616 if (increment_invocation_counter_for_backward_branches) {
1617 //__ unimplemented("branch invocation counter");
1618
1619 Label Lforward;
1620 __ add(R14_bcp, Rdisp, R14_bcp); // Add to bc addr.
1621
1622 // Check branch direction.
1623 __ cmpdi(CCR0, Rdisp, 0);
1624 __ bgt(CCR0, Lforward);
1625
1626 __ get_method_counters(R19_method, R4_counters, Lforward);
1627
1628 if (TieredCompilation) {
1629 Label Lno_mdo, Loverflow;
1630 const int increment = InvocationCounter::count_increment;
1631 const int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
1632 if (ProfileInterpreter) {
1633 Register Rmdo = Rscratch1;
1634
1635 // If no method data exists, go to profile_continue.
1636 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
1637 __ cmpdi(CCR0, Rmdo, 0);
1638 __ beq(CCR0, Lno_mdo);
1639
1640 // Increment backedge counter in the MDO.
1641 const int mdo_bc_offs = in_bytes(MethodData::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
1642 __ lwz(Rscratch2, mdo_bc_offs, Rmdo);
1643 __ load_const_optimized(Rscratch3, mask, R0);
1644 __ addi(Rscratch2, Rscratch2, increment);
1645 __ stw(Rscratch2, mdo_bc_offs, Rmdo);
1646 __ and_(Rscratch3, Rscratch2, Rscratch3);
1647 __ bne(CCR0, Lforward);
1648 __ b(Loverflow);
1649 }
1650
1651 // If there's no MDO, increment counter in method.
1652 const int mo_bc_offs = in_bytes(MethodCounters::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
1653 __ bind(Lno_mdo);
1654 __ lwz(Rscratch2, mo_bc_offs, R4_counters);
1655 __ load_const_optimized(Rscratch3, mask, R0);
1656 __ addi(Rscratch2, Rscratch2, increment);
1657 __ stw(Rscratch2, mo_bc_offs, R19_method);
1658 __ and_(Rscratch3, Rscratch2, Rscratch3);
1659 __ bne(CCR0, Lforward);
1660
1661 __ bind(Loverflow);
1662
1663 // Notify point for loop, pass branch bytecode.
1664 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R14_bcp, true);
1665
1666 // Was an OSR adapter generated?
1667 // O0 = osr nmethod
1668 __ cmpdi(CCR0, R3_RET, 0);
1669 __ beq(CCR0, Lforward);
1670
1671 // Has the nmethod been invalidated already?
1672 __ lwz(R0, nmethod::entry_bci_offset(), R3_RET);
1673 __ cmpwi(CCR0, R0, InvalidOSREntryBci);
1674 __ beq(CCR0, Lforward);
1675
1676 // Migrate the interpreter frame off of the stack.
1677 // We can use all registers because we will not return to interpreter from this point.
1678
1679 // Save nmethod.
1680 const Register osr_nmethod = R31;
1681 __ mr(osr_nmethod, R3_RET);
1682 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R11_scratch1);
1683 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), R16_thread);
1684 __ reset_last_Java_frame();
1685 // OSR buffer is in ARG1.
1686
1687 // Remove the interpreter frame.
1688 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1689
1690 // Jump to the osr code.
1691 __ ld(R11_scratch1, nmethod::osr_entry_point_offset(), osr_nmethod);
1692 __ mtlr(R0);
1693 __ mtctr(R11_scratch1);
1694 __ bctr();
1695
1696 } else {
1697
1698 const Register invoke_ctr = Rscratch1;
1699 // Update Backedge branch separately from invocations.
1700 __ increment_backedge_counter(R4_counters, invoke_ctr, Rscratch2, Rscratch3);
1701
1702 if (ProfileInterpreter) {
1703 __ test_invocation_counter_for_mdp(invoke_ctr, Rscratch2, Lforward);
1704 if (UseOnStackReplacement) {
1705 __ test_backedge_count_for_osr(bumped_count, R14_bcp, Rscratch2);
1706 }
1707 } else {
1708 if (UseOnStackReplacement) {
1709 __ test_backedge_count_for_osr(invoke_ctr, R14_bcp, Rscratch2);
1710 }
1711 }
1712 }
1713
1714 __ bind(Lforward);
1715
1716 } else {
1717 // Bump bytecode pointer by displacement (take the branch).
1718 __ add(R14_bcp, Rdisp, R14_bcp); // Add to bc addr.
1719 }
1720 // Continue with bytecode @ target.
1721 // %%%%% Like Intel, could speed things up by moving bytecode fetch to code above,
1722 // %%%%% and changing dispatch_next to dispatch_only.
1723 __ dispatch_next(vtos);
1724 }
1725
1726 // Helper function for if_cmp* methods below.
1727 // Factored out common compare and branch code.
1728 void TemplateTable::if_cmp_common(Register Rfirst, Register Rsecond, Register Rscratch1, Register Rscratch2, Condition cc, bool is_jint, bool cmp0) {
1729 Label Lnot_taken;
1730 // Note: The condition code we get is the condition under which we
1731 // *fall through*! So we have to inverse the CC here.
1732
1733 if (is_jint) {
1734 if (cmp0) {
1735 __ cmpwi(CCR0, Rfirst, 0);
1736 } else {
1737 __ cmpw(CCR0, Rfirst, Rsecond);
1738 }
1739 } else {
1740 if (cmp0) {
1741 __ cmpdi(CCR0, Rfirst, 0);
1742 } else {
1743 __ cmpd(CCR0, Rfirst, Rsecond);
1744 }
1745 }
1746 branch_conditional(CCR0, cc, Lnot_taken, /*invert*/ true);
1747
1748 // Conition is false => Jump!
1749 branch(false, false);
1750
1751 // Condition is not true => Continue.
1752 __ align(32, 12);
1753 __ bind(Lnot_taken);
1754 __ profile_not_taken_branch(Rscratch1, Rscratch2);
1755 }
1756
1757 // Compare integer values with zero and fall through if CC holds, branch away otherwise.
1758 void TemplateTable::if_0cmp(Condition cc) {
1759 transition(itos, vtos);
1760
1761 if_cmp_common(R17_tos, noreg, R11_scratch1, R12_scratch2, cc, true, true);
1762 }
1763
1764 // Compare integer values and fall through if CC holds, branch away otherwise.
1765 //
1766 // Interface:
1767 // - Rfirst: First operand (older stack value)
1768 // - tos: Second operand (younger stack value)
1769 void TemplateTable::if_icmp(Condition cc) {
1770 transition(itos, vtos);
1771
1772 const Register Rfirst = R0,
1773 Rsecond = R17_tos;
1774
1775 __ pop_i(Rfirst);
1776 if_cmp_common(Rfirst, Rsecond, R11_scratch1, R12_scratch2, cc, true, false);
1777 }
1778
1779 void TemplateTable::if_nullcmp(Condition cc) {
1780 transition(atos, vtos);
1781
1782 if_cmp_common(R17_tos, noreg, R11_scratch1, R12_scratch2, cc, false, true);
1783 }
1784
1785 void TemplateTable::if_acmp(Condition cc) {
1786 transition(atos, vtos);
1787
1788 const Register Rfirst = R0,
1789 Rsecond = R17_tos;
1790
1791 __ pop_ptr(Rfirst);
1792 if_cmp_common(Rfirst, Rsecond, R11_scratch1, R12_scratch2, cc, false, false);
1793 }
1794
1795 void TemplateTable::ret() {
1796 locals_index(R11_scratch1);
1797 __ load_local_ptr(R17_tos, R11_scratch1, R11_scratch1);
1798
1799 __ profile_ret(vtos, R17_tos, R11_scratch1, R12_scratch2);
1800
1801 __ ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method);
1802 __ add(R11_scratch1, R17_tos, R11_scratch1);
1803 __ addi(R14_bcp, R11_scratch1, in_bytes(ConstMethod::codes_offset()));
1804 __ dispatch_next(vtos);
1805 }
1806
1807 void TemplateTable::wide_ret() {
1808 transition(vtos, vtos);
1809
1810 const Register Rindex = R3_ARG1,
1811 Rscratch1 = R11_scratch1,
1812 Rscratch2 = R12_scratch2;
1813
1814 locals_index_wide(Rindex);
1815 __ load_local_ptr(R17_tos, R17_tos, Rindex);
1816 __ profile_ret(vtos, R17_tos, Rscratch1, R12_scratch2);
1817 // Tos now contains the bci, compute the bcp from that.
1818 __ ld(Rscratch1, in_bytes(Method::const_offset()), R19_method);
1819 __ addi(Rscratch2, R17_tos, in_bytes(ConstMethod::codes_offset()));
1820 __ add(R14_bcp, Rscratch1, Rscratch2);
1821 __ dispatch_next(vtos);
1822 }
1823
1824 void TemplateTable::tableswitch() {
1825 transition(itos, vtos);
1826
1827 Label Ldispatch, Ldefault_case;
1828 Register Rlow_byte = R3_ARG1,
1829 Rindex = Rlow_byte,
1830 Rhigh_byte = R4_ARG2,
1831 Rdef_offset_addr = R5_ARG3, // is going to contain address of default offset
1832 Rscratch1 = R11_scratch1,
1833 Rscratch2 = R12_scratch2,
1834 Roffset = R6_ARG4;
1835
1836 // Align bcp.
1837 __ addi(Rdef_offset_addr, R14_bcp, BytesPerInt);
1838 __ clrrdi(Rdef_offset_addr, Rdef_offset_addr, log2_long((jlong)BytesPerInt));
1839
1840 // Load lo & hi.
1841 __ lwz(Rlow_byte, BytesPerInt, Rdef_offset_addr);
1842 __ lwz(Rhigh_byte, BytesPerInt * 2, Rdef_offset_addr);
1843
1844 // Check for default case (=index outside [low,high]).
1845 __ cmpw(CCR0, R17_tos, Rlow_byte);
1846 __ cmpw(CCR1, R17_tos, Rhigh_byte);
1847 __ blt(CCR0, Ldefault_case);
1848 __ bgt(CCR1, Ldefault_case);
1849
1850 // Lookup dispatch offset.
1851 __ sub(Rindex, R17_tos, Rlow_byte);
1852 __ extsw(Rindex, Rindex);
1853 __ profile_switch_case(Rindex, Rhigh_byte /* scratch */, Rscratch1, Rscratch2);
1854 __ sldi(Rindex, Rindex, LogBytesPerInt);
1855 __ addi(Rindex, Rindex, 3 * BytesPerInt);
1856 __ lwax(Roffset, Rdef_offset_addr, Rindex);
1857 __ b(Ldispatch);
1858
1859 __ bind(Ldefault_case);
1860 __ profile_switch_default(Rhigh_byte, Rscratch1);
1861 __ lwa(Roffset, 0, Rdef_offset_addr);
1862
1863 __ bind(Ldispatch);
1864
1865 __ add(R14_bcp, Roffset, R14_bcp);
1866 __ dispatch_next(vtos);
1867 }
1868
1869 void TemplateTable::lookupswitch() {
1870 transition(itos, itos);
1871 __ stop("lookupswitch bytecode should have been rewritten");
1872 }
1873
1874 // Table switch using linear search through cases.
1875 // Bytecode stream format:
1876 // Bytecode (1) | 4-byte padding | default offset (4) | count (4) | value/offset pair1 (8) | value/offset pair2 (8) | ...
1877 // Note: Everything is big-endian format here. So on little endian machines, we have to revers offset and count and cmp value.
1878 void TemplateTable::fast_linearswitch() {
1879 transition(itos, vtos);
1880
1881 Label Lloop_entry, Lsearch_loop, Lfound, Lcontinue_execution, Ldefault_case;
1882
1883 Register Rcount = R3_ARG1,
1884 Rcurrent_pair = R4_ARG2,
1885 Rdef_offset_addr = R5_ARG3, // Is going to contain address of default offset.
1886 Roffset = R31, // Might need to survive C call.
1887 Rvalue = R12_scratch2,
1888 Rscratch = R11_scratch1,
1889 Rcmp_value = R17_tos;
1890
1891 // Align bcp.
1892 __ addi(Rdef_offset_addr, R14_bcp, BytesPerInt);
1893 __ clrrdi(Rdef_offset_addr, Rdef_offset_addr, log2_long((jlong)BytesPerInt));
1894
1895 // Setup loop counter and limit.
1896 __ lwz(Rcount, BytesPerInt, Rdef_offset_addr); // Load count.
1897 __ addi(Rcurrent_pair, Rdef_offset_addr, 2 * BytesPerInt); // Rcurrent_pair now points to first pair.
1898
1899 // Set up search loop.
1900 __ cmpwi(CCR0, Rcount, 0);
1901 __ beq(CCR0, Ldefault_case);
1902
1903 __ mtctr(Rcount);
1904
1905 // linear table search
1906 __ bind(Lsearch_loop);
1907
1908 __ lwz(Rvalue, 0, Rcurrent_pair);
1909 __ lwa(Roffset, 1 * BytesPerInt, Rcurrent_pair);
1910
1911 __ cmpw(CCR0, Rvalue, Rcmp_value);
1912 __ beq(CCR0, Lfound);
1913
1914 __ addi(Rcurrent_pair, Rcurrent_pair, 2 * BytesPerInt);
1915 __ bdnz(Lsearch_loop);
1916
1917 // default case
1918 __ bind(Ldefault_case);
1919
1920 __ lwa(Roffset, 0, Rdef_offset_addr);
1921 if (ProfileInterpreter) {
1922 __ profile_switch_default(Rdef_offset_addr, Rcount/* scratch */);
1923 __ b(Lcontinue_execution);
1924 }
1925
1926 // Entry found, skip Roffset bytecodes and continue.
1927 __ bind(Lfound);
1928 if (ProfileInterpreter) {
1929 // Calc the num of the pair we hit. Careful, Rcurrent_pair points 2 ints
1930 // beyond the actual current pair due to the auto update load above!
1931 __ sub(Rcurrent_pair, Rcurrent_pair, Rdef_offset_addr);
1932 __ addi(Rcurrent_pair, Rcurrent_pair, - 2 * BytesPerInt);
1933 __ srdi(Rcurrent_pair, Rcurrent_pair, LogBytesPerInt + 1);
1934 __ profile_switch_case(Rcurrent_pair, Rcount /*scratch*/, Rdef_offset_addr/*scratch*/, Rscratch);
1935 __ bind(Lcontinue_execution);
1936 }
1937 __ add(R14_bcp, Roffset, R14_bcp);
1938 __ dispatch_next(vtos);
1939 }
1940
1941 // Table switch using binary search (value/offset pairs are ordered).
1942 // Bytecode stream format:
1943 // Bytecode (1) | 4-byte padding | default offset (4) | count (4) | value/offset pair1 (8) | value/offset pair2 (8) | ...
1944 // Note: Everything is big-endian format here. So on little endian machines, we have to revers offset and count and cmp value.
1945 void TemplateTable::fast_binaryswitch() {
1946
1947 transition(itos, vtos);
1948 // Implementation using the following core algorithm: (copied from Intel)
1949 //
1950 // int binary_search(int key, LookupswitchPair* array, int n) {
1951 // // Binary search according to "Methodik des Programmierens" by
1952 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1953 // int i = 0;
1954 // int j = n;
1955 // while (i+1 < j) {
1956 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1957 // // with Q: for all i: 0 <= i < n: key < a[i]
1958 // // where a stands for the array and assuming that the (inexisting)
1959 // // element a[n] is infinitely big.
1960 // int h = (i + j) >> 1;
1961 // // i < h < j
1962 // if (key < array[h].fast_match()) {
1963 // j = h;
1964 // } else {
1965 // i = h;
1966 // }
1967 // }
1968 // // R: a[i] <= key < a[i+1] or Q
1969 // // (i.e., if key is within array, i is the correct index)
1970 // return i;
1971 // }
1972
1973 // register allocation
1974 const Register Rkey = R17_tos; // already set (tosca)
1975 const Register Rarray = R3_ARG1;
1976 const Register Ri = R4_ARG2;
1977 const Register Rj = R5_ARG3;
1978 const Register Rh = R6_ARG4;
1979 const Register Rscratch = R11_scratch1;
1980
1981 const int log_entry_size = 3;
1982 const int entry_size = 1 << log_entry_size;
1983
1984 Label found;
1985
1986 // Find Array start,
1987 __ addi(Rarray, R14_bcp, 3 * BytesPerInt);
1988 __ clrrdi(Rarray, Rarray, log2_long((jlong)BytesPerInt));
1989
1990 // initialize i & j
1991 __ li(Ri,0);
1992 __ lwz(Rj, -BytesPerInt, Rarray);
1993
1994 // and start.
1995 Label entry;
1996 __ b(entry);
1997
1998 // binary search loop
1999 { Label loop;
2000 __ bind(loop);
2001 // int h = (i + j) >> 1;
2002 __ srdi(Rh, Rh, 1);
2003 // if (key < array[h].fast_match()) {
2004 // j = h;
2005 // } else {
2006 // i = h;
2007 // }
2008 __ sldi(Rscratch, Rh, log_entry_size);
2009 __ lwzx(Rscratch, Rscratch, Rarray);
2010
2011 // if (key < current value)
2012 // Rh = Rj
2013 // else
2014 // Rh = Ri
2015 Label Lgreater;
2016 __ cmpw(CCR0, Rkey, Rscratch);
2017 __ bge(CCR0, Lgreater);
2018 __ mr(Rj, Rh);
2019 __ b(entry);
2020 __ bind(Lgreater);
2021 __ mr(Ri, Rh);
2022
2023 // while (i+1 < j)
2024 __ bind(entry);
2025 __ addi(Rscratch, Ri, 1);
2026 __ cmpw(CCR0, Rscratch, Rj);
2027 __ add(Rh, Ri, Rj); // start h = i + j >> 1;
2028
2029 __ blt(CCR0, loop);
2030 }
2031
2032 // End of binary search, result index is i (must check again!).
2033 Label default_case;
2034 Label continue_execution;
2035 if (ProfileInterpreter) {
2036 __ mr(Rh, Ri); // Save index in i for profiling.
2037 }
2038 // Ri = value offset
2039 __ sldi(Ri, Ri, log_entry_size);
2040 __ add(Ri, Ri, Rarray);
2041 __ lwz(Rscratch, 0, Ri);
2042
2043 Label not_found;
2044 // Ri = offset offset
2045 __ cmpw(CCR0, Rkey, Rscratch);
2046 __ beq(CCR0, not_found);
2047 // entry not found -> j = default offset
2048 __ lwz(Rj, -2 * BytesPerInt, Rarray);
2049 __ b(default_case);
2050
2051 __ bind(not_found);
2052 // entry found -> j = offset
2053 __ profile_switch_case(Rh, Rj, Rscratch, Rkey);
2054 __ lwz(Rj, BytesPerInt, Ri);
2055
2056 if (ProfileInterpreter) {
2057 __ b(continue_execution);
2058 }
2059
2060 __ bind(default_case); // fall through (if not profiling)
2061 __ profile_switch_default(Ri, Rscratch);
2062
2063 __ bind(continue_execution);
2064
2065 __ extsw(Rj, Rj);
2066 __ add(R14_bcp, Rj, R14_bcp);
2067 __ dispatch_next(vtos);
2068 }
2069
2070 void TemplateTable::_return(TosState state) {
2071 transition(state, state);
2072 assert(_desc->calls_vm(),
2073 "inconsistent calls_vm information"); // call in remove_activation
2074
2075 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2076
2077 Register Rscratch = R11_scratch1,
2078 Rklass = R12_scratch2,
2079 Rklass_flags = Rklass;
2080 Label Lskip_register_finalizer;
2081
2082 // Check if the method has the FINALIZER flag set and call into the VM to finalize in this case.
2083 assert(state == vtos, "only valid state");
2084 __ ld(R17_tos, 0, R18_locals);
2085
2086 // Load klass of this obj.
2087 __ load_klass(Rklass, R17_tos);
2088 __ lwz(Rklass_flags, in_bytes(Klass::access_flags_offset()), Rklass);
2089 __ testbitdi(CCR0, R0, Rklass_flags, exact_log2(JVM_ACC_HAS_FINALIZER));
2090 __ bfalse(CCR0, Lskip_register_finalizer);
2091
2092 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), R17_tos /* obj */);
2093
2094 __ align(32, 12);
2095 __ bind(Lskip_register_finalizer);
2096 }
2097
2098 // Move the result value into the correct register and remove memory stack frame.
2099 __ remove_activation(state, /* throw_monitor_exception */ true);
2100 // Restoration of lr done by remove_activation.
2101 switch (state) {
2102 case ltos:
2103 case btos:
2104 case ctos:
2105 case stos:
2106 case atos:
2107 case itos: __ mr(R3_RET, R17_tos); break;
2108 case ftos:
2109 case dtos: __ fmr(F1_RET, F15_ftos); break;
2110 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
2111 // to get visible before the reference to the object gets stored anywhere.
2112 __ membar(Assembler::StoreStore); break;
2113 default : ShouldNotReachHere();
2114 }
2115 __ blr();
2116 }
2117
2118 // ============================================================================
2119 // Constant pool cache access
2120 //
2121 // Memory ordering:
2122 //
2123 // Like done in C++ interpreter, we load the fields
2124 // - _indices
2125 // - _f12_oop
2126 // acquired, because these are asked if the cache is already resolved. We don't
2127 // want to float loads above this check.
2128 // See also comments in ConstantPoolCacheEntry::bytecode_1(),
2129 // ConstantPoolCacheEntry::bytecode_2() and ConstantPoolCacheEntry::f1();
2130
2131 // Call into the VM if call site is not yet resolved
2132 //
2133 // Input regs:
2134 // - None, all passed regs are outputs.
2135 //
2136 // Returns:
2137 // - Rcache: The const pool cache entry that contains the resolved result.
2138 // - Rresult: Either noreg or output for f1/f2.
2139 //
2140 // Kills:
2141 // - Rscratch
2142 void TemplateTable::resolve_cache_and_index(int byte_no, Register Rcache, Register Rscratch, size_t index_size) {
2143
2144 __ get_cache_and_index_at_bcp(Rcache, 1, index_size);
2145 Label Lresolved, Ldone;
2146
2147 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2148 // We are resolved if the indices offset contains the current bytecode.
2149 // Big Endian:
2150 __ lbz(Rscratch, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 7 - (byte_no + 1), Rcache);
2151 // Acquire by cmp-br-isync (see below).
2152 __ cmpdi(CCR0, Rscratch, (int)bytecode());
2153 __ beq(CCR0, Lresolved);
2154
2155 address entry = NULL;
2156 switch (bytecode()) {
2157 case Bytecodes::_getstatic : // fall through
2158 case Bytecodes::_putstatic : // fall through
2159 case Bytecodes::_getfield : // fall through
2160 case Bytecodes::_putfield : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break;
2161 case Bytecodes::_invokevirtual : // fall through
2162 case Bytecodes::_invokespecial : // fall through
2163 case Bytecodes::_invokestatic : // fall through
2164 case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke); break;
2165 case Bytecodes::_invokehandle : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokehandle); break;
2166 case Bytecodes::_invokedynamic : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic); break;
2167 default : ShouldNotReachHere(); break;
2168 }
2169 __ li(R4_ARG2, (int)bytecode());
2170 __ call_VM(noreg, entry, R4_ARG2, true);
2171
2172 // Update registers with resolved info.
2173 __ get_cache_and_index_at_bcp(Rcache, 1, index_size);
2174 __ b(Ldone);
2175
2176 __ bind(Lresolved);
2177 __ isync(); // Order load wrt. succeeding loads.
2178 __ bind(Ldone);
2179 }
2180
2181 // Load the constant pool cache entry at field accesses into registers.
2182 // The Rcache and Rindex registers must be set before call.
2183 // Input:
2184 // - Rcache, Rindex
2185 // Output:
2186 // - Robj, Roffset, Rflags
2187 void TemplateTable::load_field_cp_cache_entry(Register Robj,
2188 Register Rcache,
2189 Register Rindex /* unused on PPC64 */,
2190 Register Roffset,
2191 Register Rflags,
2192 bool is_static = false) {
2193 assert_different_registers(Rcache, Rflags, Roffset);
2194 // assert(Rindex == noreg, "parameter not used on PPC64");
2195
2196 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2197 __ ld(Rflags, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::flags_offset()), Rcache);
2198 __ ld(Roffset, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::f2_offset()), Rcache);
2199 if (is_static) {
2200 __ ld(Robj, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::f1_offset()), Rcache);
2201 __ ld(Robj, in_bytes(Klass::java_mirror_offset()), Robj);
2202 // Acquire not needed here. Following access has an address dependency on this value.
2203 }
2204 }
2205
2206 // Load the constant pool cache entry at invokes into registers.
2207 // Resolve if necessary.
2208
2209 // Input Registers:
2210 // - None, bcp is used, though
2211 //
2212 // Return registers:
2213 // - Rmethod (f1 field or f2 if invokevirtual)
2214 // - Ritable_index (f2 field)
2215 // - Rflags (flags field)
2216 //
2217 // Kills:
2218 // - R21
2219 //
2220 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2221 Register Rmethod,
2222 Register Ritable_index,
2223 Register Rflags,
2224 bool is_invokevirtual,
2225 bool is_invokevfinal,
2226 bool is_invokedynamic) {
2227
2228 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2229 // Determine constant pool cache field offsets.
2230 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2231 const int method_offset = in_bytes(cp_base_offset + (is_invokevirtual ? ConstantPoolCacheEntry::f2_offset() : ConstantPoolCacheEntry::f1_offset()));
2232 const int flags_offset = in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset());
2233 // Access constant pool cache fields.
2234 const int index_offset = in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset());
2235
2236 Register Rcache = R21_tmp1; // Note: same register as R21_sender_SP.
2237
2238 if (is_invokevfinal) {
2239 assert(Ritable_index == noreg, "register not used");
2240 // Already resolved.
2241 __ get_cache_and_index_at_bcp(Rcache, 1);
2242 } else {
2243 resolve_cache_and_index(byte_no, Rcache, R0, is_invokedynamic ? sizeof(u4) : sizeof(u2));
2244 }
2245
2246 __ ld(Rmethod, method_offset, Rcache);
2247 __ ld(Rflags, flags_offset, Rcache);
2248
2249 if (Ritable_index != noreg) {
2250 __ ld(Ritable_index, index_offset, Rcache);
2251 }
2252 }
2253
2254 // ============================================================================
2255 // Field access
2256
2257 // Volatile variables demand their effects be made known to all CPU's
2258 // in order. Store buffers on most chips allow reads & writes to
2259 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2260 // without some kind of memory barrier (i.e., it's not sufficient that
2261 // the interpreter does not reorder volatile references, the hardware
2262 // also must not reorder them).
2263 //
2264 // According to the new Java Memory Model (JMM):
2265 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2266 // writes act as aquire & release, so:
2267 // (2) A read cannot let unrelated NON-volatile memory refs that
2268 // happen after the read float up to before the read. It's OK for
2269 // non-volatile memory refs that happen before the volatile read to
2270 // float down below it.
2271 // (3) Similar a volatile write cannot let unrelated NON-volatile
2272 // memory refs that happen BEFORE the write float down to after the
2273 // write. It's OK for non-volatile memory refs that happen after the
2274 // volatile write to float up before it.
2275 //
2276 // We only put in barriers around volatile refs (they are expensive),
2277 // not _between_ memory refs (that would require us to track the
2278 // flavor of the previous memory refs). Requirements (2) and (3)
2279 // require some barriers before volatile stores and after volatile
2280 // loads. These nearly cover requirement (1) but miss the
2281 // volatile-store-volatile-load case. This final case is placed after
2282 // volatile-stores although it could just as well go before
2283 // volatile-loads.
2284
2285 // The registers cache and index expected to be set before call.
2286 // Correct values of the cache and index registers are preserved.
2287 // Kills:
2288 // Rcache (if has_tos)
2289 // Rscratch
2290 void TemplateTable::jvmti_post_field_access(Register Rcache, Register Rscratch, bool is_static, bool has_tos) {
2291
2292 assert_different_registers(Rcache, Rscratch);
2293
2294 if (JvmtiExport::can_post_field_access()) {
2295 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2296 Label Lno_field_access_post;
2297
2298 // Check if post field access in enabled.
2299 int offs = __ load_const_optimized(Rscratch, JvmtiExport::get_field_access_count_addr(), R0, true);
2300 __ lwz(Rscratch, offs, Rscratch);
2301
2302 __ cmpwi(CCR0, Rscratch, 0);
2303 __ beq(CCR0, Lno_field_access_post);
2304
2305 // Post access enabled - do it!
2306 __ addi(Rcache, Rcache, in_bytes(cp_base_offset));
2307 if (is_static) {
2308 __ li(R17_tos, 0);
2309 } else {
2310 if (has_tos) {
2311 // The fast bytecode versions have obj ptr in register.
2312 // Thus, save object pointer before call_VM() clobbers it
2313 // put object on tos where GC wants it.
2314 __ push_ptr(R17_tos);
2315 } else {
2316 // Load top of stack (do not pop the value off the stack).
2317 __ ld(R17_tos, Interpreter::expr_offset_in_bytes(0), R15_esp);
2318 }
2319 __ verify_oop(R17_tos);
2320 }
2321 // tos: object pointer or NULL if static
2322 // cache: cache entry pointer
2323 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), R17_tos, Rcache);
2324 if (!is_static && has_tos) {
2325 // Restore object pointer.
2326 __ pop_ptr(R17_tos);
2327 __ verify_oop(R17_tos);
2328 } else {
2329 // Cache is still needed to get class or obj.
2330 __ get_cache_and_index_at_bcp(Rcache, 1);
2331 }
2332
2333 __ align(32, 12);
2334 __ bind(Lno_field_access_post);
2335 }
2336 }
2337
2338 // kills R11_scratch1
2339 void TemplateTable::pop_and_check_object(Register Roop) {
2340 Register Rtmp = R11_scratch1;
2341
2342 assert_different_registers(Rtmp, Roop);
2343 __ pop_ptr(Roop);
2344 // For field access must check obj.
2345 __ null_check_throw(Roop, -1, Rtmp);
2346 __ verify_oop(Roop);
2347 }
2348
2349 // PPC64: implement volatile loads as fence-store-acquire.
2350 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2351 transition(vtos, vtos);
2352
2353 Label Lacquire, Lisync;
2354
2355 const Register Rcache = R3_ARG1,
2356 Rclass_or_obj = R22_tmp2,
2357 Roffset = R23_tmp3,
2358 Rflags = R31,
2359 Rbtable = R5_ARG3,
2360 Rbc = R6_ARG4,
2361 Rscratch = R12_scratch2;
2362
2363 static address field_branch_table[number_of_states],
2364 static_branch_table[number_of_states];
2365
2366 address* branch_table = is_static ? static_branch_table : field_branch_table;
2367
2368 // Get field offset.
2369 resolve_cache_and_index(byte_no, Rcache, Rscratch, sizeof(u2));
2370
2371 // JVMTI support
2372 jvmti_post_field_access(Rcache, Rscratch, is_static, false);
2373
2374 // Load after possible GC.
2375 load_field_cp_cache_entry(Rclass_or_obj, Rcache, noreg, Roffset, Rflags, is_static);
2376
2377 // Load pointer to branch table.
2378 __ load_const_optimized(Rbtable, (address)branch_table, Rscratch);
2379
2380 // Get volatile flag.
2381 __ rldicl(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2382 // Note: sync is needed before volatile load on PPC64.
2383
2384 // Check field type.
2385 __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2386
2387 #ifdef ASSERT
2388 Label LFlagInvalid;
2389 __ cmpldi(CCR0, Rflags, number_of_states);
2390 __ bge(CCR0, LFlagInvalid);
2391 #endif
2392
2393 // Load from branch table and dispatch (volatile case: one instruction ahead).
2394 __ sldi(Rflags, Rflags, LogBytesPerWord);
2395 __ cmpwi(CCR6, Rscratch, 1); // Volatile?
2396 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2397 __ sldi(Rscratch, Rscratch, exact_log2(BytesPerInstWord)); // Volatile ? size of 1 instruction : 0.
2398 }
2399 __ ldx(Rbtable, Rbtable, Rflags);
2400
2401 // Get the obj from stack.
2402 if (!is_static) {
2403 pop_and_check_object(Rclass_or_obj); // Kills R11_scratch1.
2404 } else {
2405 __ verify_oop(Rclass_or_obj);
2406 }
2407
2408 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2409 __ subf(Rbtable, Rscratch, Rbtable); // Point to volatile/non-volatile entry point.
2410 }
2411 __ mtctr(Rbtable);
2412 __ bctr();
2413
2414 #ifdef ASSERT
2415 __ bind(LFlagInvalid);
2416 __ stop("got invalid flag", 0x654);
2417
2418 // __ bind(Lvtos);
2419 address pc_before_fence = __ pc();
2420 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2421 assert(__ pc() - pc_before_fence == (ptrdiff_t)BytesPerInstWord, "must be single instruction");
2422 assert(branch_table[vtos] == 0, "can't compute twice");
2423 branch_table[vtos] = __ pc(); // non-volatile_entry point
2424 __ stop("vtos unexpected", 0x655);
2425 #endif
2426
2427 __ align(32, 28, 28); // Align load.
2428 // __ bind(Ldtos);
2429 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2430 assert(branch_table[dtos] == 0, "can't compute twice");
2431 branch_table[dtos] = __ pc(); // non-volatile_entry point
2432 __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
2433 __ push(dtos);
2434 if (!is_static) patch_bytecode(Bytecodes::_fast_dgetfield, Rbc, Rscratch);
2435 {
2436 Label acquire_double;
2437 __ beq(CCR6, acquire_double); // Volatile?
2438 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2439
2440 __ bind(acquire_double);
2441 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
2442 __ beq_predict_taken(CCR0, Lisync);
2443 __ b(Lisync); // In case of NAN.
2444 }
2445
2446 __ align(32, 28, 28); // Align load.
2447 // __ bind(Lftos);
2448 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2449 assert(branch_table[ftos] == 0, "can't compute twice");
2450 branch_table[ftos] = __ pc(); // non-volatile_entry point
2451 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
2452 __ push(ftos);
2453 if (!is_static) { patch_bytecode(Bytecodes::_fast_fgetfield, Rbc, Rscratch); }
2454 {
2455 Label acquire_float;
2456 __ beq(CCR6, acquire_float); // Volatile?
2457 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2458
2459 __ bind(acquire_float);
2460 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
2461 __ beq_predict_taken(CCR0, Lisync);
2462 __ b(Lisync); // In case of NAN.
2463 }
2464
2465 __ align(32, 28, 28); // Align load.
2466 // __ bind(Litos);
2467 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2468 assert(branch_table[itos] == 0, "can't compute twice");
2469 branch_table[itos] = __ pc(); // non-volatile_entry point
2470 __ lwax(R17_tos, Rclass_or_obj, Roffset);
2471 __ push(itos);
2472 if (!is_static) patch_bytecode(Bytecodes::_fast_igetfield, Rbc, Rscratch);
2473 __ beq(CCR6, Lacquire); // Volatile?
2474 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2475
2476 __ align(32, 28, 28); // Align load.
2477 // __ bind(Lltos);
2478 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2479 assert(branch_table[ltos] == 0, "can't compute twice");
2480 branch_table[ltos] = __ pc(); // non-volatile_entry point
2481 __ ldx(R17_tos, Rclass_or_obj, Roffset);
2482 __ push(ltos);
2483 if (!is_static) patch_bytecode(Bytecodes::_fast_lgetfield, Rbc, Rscratch);
2484 __ beq(CCR6, Lacquire); // Volatile?
2485 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2486
2487 __ align(32, 28, 28); // Align load.
2488 // __ bind(Lbtos);
2489 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2490 assert(branch_table[btos] == 0, "can't compute twice");
2491 branch_table[btos] = __ pc(); // non-volatile_entry point
2492 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
2493 __ extsb(R17_tos, R17_tos);
2494 __ push(btos);
2495 if (!is_static) patch_bytecode(Bytecodes::_fast_bgetfield, Rbc, Rscratch);
2496 __ beq(CCR6, Lacquire); // Volatile?
2497 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2498
2499 __ align(32, 28, 28); // Align load.
2500 // __ bind(Lctos);
2501 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2502 assert(branch_table[ctos] == 0, "can't compute twice");
2503 branch_table[ctos] = __ pc(); // non-volatile_entry point
2504 __ lhzx(R17_tos, Rclass_or_obj, Roffset);
2505 __ push(ctos);
2506 if (!is_static) patch_bytecode(Bytecodes::_fast_cgetfield, Rbc, Rscratch);
2507 __ beq(CCR6, Lacquire); // Volatile?
2508 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2509
2510 __ align(32, 28, 28); // Align load.
2511 // __ bind(Lstos);
2512 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2513 assert(branch_table[stos] == 0, "can't compute twice");
2514 branch_table[stos] = __ pc(); // non-volatile_entry point
2515 __ lhax(R17_tos, Rclass_or_obj, Roffset);
2516 __ push(stos);
2517 if (!is_static) patch_bytecode(Bytecodes::_fast_sgetfield, Rbc, Rscratch);
2518 __ beq(CCR6, Lacquire); // Volatile?
2519 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2520
2521 __ align(32, 28, 28); // Align load.
2522 // __ bind(Latos);
2523 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2524 assert(branch_table[atos] == 0, "can't compute twice");
2525 branch_table[atos] = __ pc(); // non-volatile_entry point
2526 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
2527 __ verify_oop(R17_tos);
2528 __ push(atos);
2529 //__ dcbt(R17_tos); // prefetch
2530 if (!is_static) patch_bytecode(Bytecodes::_fast_agetfield, Rbc, Rscratch);
2531 __ beq(CCR6, Lacquire); // Volatile?
2532 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2533
2534 __ align(32, 12);
2535 __ bind(Lacquire);
2536 __ twi_0(R17_tos);
2537 __ bind(Lisync);
2538 __ isync(); // acquire
2539
2540 #ifdef ASSERT
2541 for (int i = 0; i<number_of_states; ++i) {
2542 assert(branch_table[i], "get initialization");
2543 //tty->print_cr("get: %s_branch_table[%d] = 0x%llx (opcode 0x%llx)",
2544 // is_static ? "static" : "field", i, branch_table[i], *((unsigned int*)branch_table[i]));
2545 }
2546 #endif
2547 }
2548
2549 void TemplateTable::getfield(int byte_no) {
2550 getfield_or_static(byte_no, false);
2551 }
2552
2553 void TemplateTable::getstatic(int byte_no) {
2554 getfield_or_static(byte_no, true);
2555 }
2556
2557 // The registers cache and index expected to be set before call.
2558 // The function may destroy various registers, just not the cache and index registers.
2559 void TemplateTable::jvmti_post_field_mod(Register Rcache, Register Rscratch, bool is_static) {
2560
2561 assert_different_registers(Rcache, Rscratch, R6_ARG4);
2562
2563 if (JvmtiExport::can_post_field_modification()) {
2564 Label Lno_field_mod_post;
2565
2566 // Check if post field access in enabled.
2567 int offs = __ load_const_optimized(Rscratch, JvmtiExport::get_field_modification_count_addr(), R0, true);
2568 __ lwz(Rscratch, offs, Rscratch);
2569
2570 __ cmpwi(CCR0, Rscratch, 0);
2571 __ beq(CCR0, Lno_field_mod_post);
2572
2573 // Do the post
2574 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2575 const Register Robj = Rscratch;
2576
2577 __ addi(Rcache, Rcache, in_bytes(cp_base_offset));
2578 if (is_static) {
2579 // Life is simple. Null out the object pointer.
2580 __ li(Robj, 0);
2581 } else {
2582 // In case of the fast versions, value lives in registers => put it back on tos.
2583 int offs = Interpreter::expr_offset_in_bytes(0);
2584 Register base = R15_esp;
2585 switch(bytecode()) {
2586 case Bytecodes::_fast_aputfield: __ push_ptr(); offs+= Interpreter::stackElementSize; break;
2587 case Bytecodes::_fast_iputfield: // Fall through
2588 case Bytecodes::_fast_bputfield: // Fall through
2589 case Bytecodes::_fast_cputfield: // Fall through
2590 case Bytecodes::_fast_sputfield: __ push_i(); offs+= Interpreter::stackElementSize; break;
2591 case Bytecodes::_fast_lputfield: __ push_l(); offs+=2*Interpreter::stackElementSize; break;
2592 case Bytecodes::_fast_fputfield: __ push_f(); offs+= Interpreter::stackElementSize; break;
2593 case Bytecodes::_fast_dputfield: __ push_d(); offs+=2*Interpreter::stackElementSize; break;
2594 default: {
2595 offs = 0;
2596 base = Robj;
2597 const Register Rflags = Robj;
2598 Label is_one_slot;
2599 // Life is harder. The stack holds the value on top, followed by the
2600 // object. We don't know the size of the value, though; it could be
2601 // one or two words depending on its type. As a result, we must find
2602 // the type to determine where the object is.
2603 __ ld(Rflags, in_bytes(ConstantPoolCacheEntry::flags_offset()), Rcache); // Big Endian
2604 __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2605
2606 __ cmpwi(CCR0, Rflags, ltos);
2607 __ cmpwi(CCR1, Rflags, dtos);
2608 __ addi(base, R15_esp, Interpreter::expr_offset_in_bytes(1));
2609 __ crnor(/*CR0 eq*/2, /*CR1 eq*/4+2, /*CR0 eq*/2);
2610 __ beq(CCR0, is_one_slot);
2611 __ addi(base, R15_esp, Interpreter::expr_offset_in_bytes(2));
2612 __ bind(is_one_slot);
2613 break;
2614 }
2615 }
2616 __ ld(Robj, offs, base);
2617 __ verify_oop(Robj);
2618 }
2619
2620 __ addi(R6_ARG4, R15_esp, Interpreter::expr_offset_in_bytes(0));
2621 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), Robj, Rcache, R6_ARG4);
2622 __ get_cache_and_index_at_bcp(Rcache, 1);
2623
2624 // In case of the fast versions, value lives in registers => put it back on tos.
2625 switch(bytecode()) {
2626 case Bytecodes::_fast_aputfield: __ pop_ptr(); break;
2627 case Bytecodes::_fast_iputfield: // Fall through
2628 case Bytecodes::_fast_bputfield: // Fall through
2629 case Bytecodes::_fast_cputfield: // Fall through
2630 case Bytecodes::_fast_sputfield: __ pop_i(); break;
2631 case Bytecodes::_fast_lputfield: __ pop_l(); break;
2632 case Bytecodes::_fast_fputfield: __ pop_f(); break;
2633 case Bytecodes::_fast_dputfield: __ pop_d(); break;
2634 default: break; // Nothin' to do.
2635 }
2636
2637 __ align(32, 12);
2638 __ bind(Lno_field_mod_post);
2639 }
2640 }
2641
2642 // PPC64: implement volatile stores as release-store (return bytecode contains an additional release).
2643 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2644 Label Lvolatile;
2645
2646 const Register Rcache = R5_ARG3, // Do not use ARG1/2 (causes trouble in jvmti_post_field_mod).
2647 Rclass_or_obj = R31, // Needs to survive C call.
2648 Roffset = R22_tmp2, // Needs to survive C call.
2649 Rflags = R3_ARG1,
2650 Rbtable = R4_ARG2,
2651 Rscratch = R11_scratch1,
2652 Rscratch2 = R12_scratch2,
2653 Rscratch3 = R6_ARG4,
2654 Rbc = Rscratch3;
2655 const ConditionRegister CR_is_vol = CCR2; // Non-volatile condition register (survives runtime call in do_oop_store).
2656
2657 static address field_branch_table[number_of_states],
2658 static_branch_table[number_of_states];
2659
2660 address* branch_table = is_static ? static_branch_table : field_branch_table;
2661
2662 // Stack (grows up):
2663 // value
2664 // obj
2665
2666 // Load the field offset.
2667 resolve_cache_and_index(byte_no, Rcache, Rscratch, sizeof(u2));
2668 jvmti_post_field_mod(Rcache, Rscratch, is_static);
2669 load_field_cp_cache_entry(Rclass_or_obj, Rcache, noreg, Roffset, Rflags, is_static);
2670
2671 // Load pointer to branch table.
2672 __ load_const_optimized(Rbtable, (address)branch_table, Rscratch);
2673
2674 // Get volatile flag.
2675 __ rldicl(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2676
2677 // Check the field type.
2678 __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2679
2680 #ifdef ASSERT
2681 Label LFlagInvalid;
2682 __ cmpldi(CCR0, Rflags, number_of_states);
2683 __ bge(CCR0, LFlagInvalid);
2684 #endif
2685
2686 // Load from branch table and dispatch (volatile case: one instruction ahead).
2687 __ sldi(Rflags, Rflags, LogBytesPerWord);
2688 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) { __ cmpwi(CR_is_vol, Rscratch, 1); } // Volatile?
2689 __ sldi(Rscratch, Rscratch, exact_log2(BytesPerInstWord)); // Volatile? size of instruction 1 : 0.
2690 __ ldx(Rbtable, Rbtable, Rflags);
2691
2692 __ subf(Rbtable, Rscratch, Rbtable); // Point to volatile/non-volatile entry point.
2693 __ mtctr(Rbtable);
2694 __ bctr();
2695
2696 #ifdef ASSERT
2697 __ bind(LFlagInvalid);
2698 __ stop("got invalid flag", 0x656);
2699
2700 // __ bind(Lvtos);
2701 address pc_before_release = __ pc();
2702 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2703 assert(__ pc() - pc_before_release == (ptrdiff_t)BytesPerInstWord, "must be single instruction");
2704 assert(branch_table[vtos] == 0, "can't compute twice");
2705 branch_table[vtos] = __ pc(); // non-volatile_entry point
2706 __ stop("vtos unexpected", 0x657);
2707 #endif
2708
2709 __ align(32, 28, 28); // Align pop.
2710 // __ bind(Ldtos);
2711 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2712 assert(branch_table[dtos] == 0, "can't compute twice");
2713 branch_table[dtos] = __ pc(); // non-volatile_entry point
2714 __ pop(dtos);
2715 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2716 __ stfdx(F15_ftos, Rclass_or_obj, Roffset);
2717 if (!is_static) { patch_bytecode(Bytecodes::_fast_dputfield, Rbc, Rscratch, true, byte_no); }
2718 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2719 __ beq(CR_is_vol, Lvolatile); // Volatile?
2720 }
2721 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2722
2723 __ align(32, 28, 28); // Align pop.
2724 // __ bind(Lftos);
2725 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2726 assert(branch_table[ftos] == 0, "can't compute twice");
2727 branch_table[ftos] = __ pc(); // non-volatile_entry point
2728 __ pop(ftos);
2729 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2730 __ stfsx(F15_ftos, Rclass_or_obj, Roffset);
2731 if (!is_static) { patch_bytecode(Bytecodes::_fast_fputfield, Rbc, Rscratch, true, byte_no); }
2732 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2733 __ beq(CR_is_vol, Lvolatile); // Volatile?
2734 }
2735 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2736
2737 __ align(32, 28, 28); // Align pop.
2738 // __ bind(Litos);
2739 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2740 assert(branch_table[itos] == 0, "can't compute twice");
2741 branch_table[itos] = __ pc(); // non-volatile_entry point
2742 __ pop(itos);
2743 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2744 __ stwx(R17_tos, Rclass_or_obj, Roffset);
2745 if (!is_static) { patch_bytecode(Bytecodes::_fast_iputfield, Rbc, Rscratch, true, byte_no); }
2746 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2747 __ beq(CR_is_vol, Lvolatile); // Volatile?
2748 }
2749 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2750
2751 __ align(32, 28, 28); // Align pop.
2752 // __ bind(Lltos);
2753 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2754 assert(branch_table[ltos] == 0, "can't compute twice");
2755 branch_table[ltos] = __ pc(); // non-volatile_entry point
2756 __ pop(ltos);
2757 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2758 __ stdx(R17_tos, Rclass_or_obj, Roffset);
2759 if (!is_static) { patch_bytecode(Bytecodes::_fast_lputfield, Rbc, Rscratch, true, byte_no); }
2760 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2761 __ beq(CR_is_vol, Lvolatile); // Volatile?
2762 }
2763 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2764
2765 __ align(32, 28, 28); // Align pop.
2766 // __ bind(Lbtos);
2767 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2768 assert(branch_table[btos] == 0, "can't compute twice");
2769 branch_table[btos] = __ pc(); // non-volatile_entry point
2770 __ pop(btos);
2771 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2772 __ stbx(R17_tos, Rclass_or_obj, Roffset);
2773 if (!is_static) { patch_bytecode(Bytecodes::_fast_bputfield, Rbc, Rscratch, true, byte_no); }
2774 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2775 __ beq(CR_is_vol, Lvolatile); // Volatile?
2776 }
2777 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2778
2779 __ align(32, 28, 28); // Align pop.
2780 // __ bind(Lctos);
2781 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2782 assert(branch_table[ctos] == 0, "can't compute twice");
2783 branch_table[ctos] = __ pc(); // non-volatile_entry point
2784 __ pop(ctos);
2785 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1..
2786 __ sthx(R17_tos, Rclass_or_obj, Roffset);
2787 if (!is_static) { patch_bytecode(Bytecodes::_fast_cputfield, Rbc, Rscratch, true, byte_no); }
2788 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2789 __ beq(CR_is_vol, Lvolatile); // Volatile?
2790 }
2791 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2792
2793 __ align(32, 28, 28); // Align pop.
2794 // __ bind(Lstos);
2795 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2796 assert(branch_table[stos] == 0, "can't compute twice");
2797 branch_table[stos] = __ pc(); // non-volatile_entry point
2798 __ pop(stos);
2799 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2800 __ sthx(R17_tos, Rclass_or_obj, Roffset);
2801 if (!is_static) { patch_bytecode(Bytecodes::_fast_sputfield, Rbc, Rscratch, true, byte_no); }
2802 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2803 __ beq(CR_is_vol, Lvolatile); // Volatile?
2804 }
2805 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2806
2807 __ align(32, 28, 28); // Align pop.
2808 // __ bind(Latos);
2809 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2810 assert(branch_table[atos] == 0, "can't compute twice");
2811 branch_table[atos] = __ pc(); // non-volatile_entry point
2812 __ pop(atos);
2813 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // kills R11_scratch1
2814 do_oop_store(_masm, Rclass_or_obj, Roffset, R17_tos, Rscratch, Rscratch2, Rscratch3, _bs->kind(), false /* precise */, true /* check null */);
2815 if (!is_static) { patch_bytecode(Bytecodes::_fast_aputfield, Rbc, Rscratch, true, byte_no); }
2816 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2817 __ beq(CR_is_vol, Lvolatile); // Volatile?
2818 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2819
2820 __ align(32, 12);
2821 __ bind(Lvolatile);
2822 __ fence();
2823 }
2824 // fallthru: __ b(Lexit);
2825
2826 #ifdef ASSERT
2827 for (int i = 0; i<number_of_states; ++i) {
2828 assert(branch_table[i], "put initialization");
2829 //tty->print_cr("put: %s_branch_table[%d] = 0x%llx (opcode 0x%llx)",
2830 // is_static ? "static" : "field", i, branch_table[i], *((unsigned int*)branch_table[i]));
2831 }
2832 #endif
2833 }
2834
2835 void TemplateTable::putfield(int byte_no) {
2836 putfield_or_static(byte_no, false);
2837 }
2838
2839 void TemplateTable::putstatic(int byte_no) {
2840 putfield_or_static(byte_no, true);
2841 }
2842
2843 // See SPARC. On PPC64, we have a different jvmti_post_field_mod which does the job.
2844 void TemplateTable::jvmti_post_fast_field_mod() {
2845 __ should_not_reach_here();
2846 }
2847
2848 void TemplateTable::fast_storefield(TosState state) {
2849 transition(state, vtos);
2850
2851 const Register Rcache = R5_ARG3, // Do not use ARG1/2 (causes trouble in jvmti_post_field_mod).
2852 Rclass_or_obj = R31, // Needs to survive C call.
2853 Roffset = R22_tmp2, // Needs to survive C call.
2854 Rflags = R3_ARG1,
2855 Rscratch = R11_scratch1,
2856 Rscratch2 = R12_scratch2,
2857 Rscratch3 = R4_ARG2;
2858 const ConditionRegister CR_is_vol = CCR2; // Non-volatile condition register (survives runtime call in do_oop_store).
2859
2860 // Constant pool already resolved => Load flags and offset of field.
2861 __ get_cache_and_index_at_bcp(Rcache, 1);
2862 jvmti_post_field_mod(Rcache, Rscratch, false /* not static */);
2863 load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
2864
2865 // Get the obj and the final store addr.
2866 pop_and_check_object(Rclass_or_obj); // Kills R11_scratch1.
2867
2868 // Get volatile flag.
2869 __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2870 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) { __ cmpdi(CR_is_vol, Rscratch, 1); }
2871 {
2872 Label LnotVolatile;
2873 __ beq(CCR0, LnotVolatile);
2874 __ release();
2875 __ align(32, 12);
2876 __ bind(LnotVolatile);
2877 }
2878
2879 // Do the store and fencing.
2880 switch(bytecode()) {
2881 case Bytecodes::_fast_aputfield:
2882 // Store into the field.
2883 do_oop_store(_masm, Rclass_or_obj, Roffset, R17_tos, Rscratch, Rscratch2, Rscratch3, _bs->kind(), false /* precise */, true /* check null */);
2884 break;
2885
2886 case Bytecodes::_fast_iputfield:
2887 __ stwx(R17_tos, Rclass_or_obj, Roffset);
2888 break;
2889
2890 case Bytecodes::_fast_lputfield:
2891 __ stdx(R17_tos, Rclass_or_obj, Roffset);
2892 break;
2893
2894 case Bytecodes::_fast_bputfield:
2895 __ stbx(R17_tos, Rclass_or_obj, Roffset);
2896 break;
2897
2898 case Bytecodes::_fast_cputfield:
2899 case Bytecodes::_fast_sputfield:
2900 __ sthx(R17_tos, Rclass_or_obj, Roffset);
2901 break;
2902
2903 case Bytecodes::_fast_fputfield:
2904 __ stfsx(F15_ftos, Rclass_or_obj, Roffset);
2905 break;
2906
2907 case Bytecodes::_fast_dputfield:
2908 __ stfdx(F15_ftos, Rclass_or_obj, Roffset);
2909 break;
2910
2911 default: ShouldNotReachHere();
2912 }
2913
2914 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2915 Label LVolatile;
2916 __ beq(CR_is_vol, LVolatile);
2917 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2918
2919 __ align(32, 12);
2920 __ bind(LVolatile);
2921 __ fence();
2922 }
2923 }
2924
2925 void TemplateTable::fast_accessfield(TosState state) {
2926 transition(atos, state);
2927
2928 Label LisVolatile;
2929 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2930
2931 const Register Rcache = R3_ARG1,
2932 Rclass_or_obj = R17_tos,
2933 Roffset = R22_tmp2,
2934 Rflags = R23_tmp3,
2935 Rscratch = R12_scratch2;
2936
2937 // Constant pool already resolved. Get the field offset.
2938 __ get_cache_and_index_at_bcp(Rcache, 1);
2939 load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
2940
2941 // JVMTI support
2942 jvmti_post_field_access(Rcache, Rscratch, false, true);
2943
2944 // Get the load address.
2945 __ null_check_throw(Rclass_or_obj, -1, Rscratch);
2946
2947 // Get volatile flag.
2948 __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2949 __ bne(CCR0, LisVolatile);
2950
2951 switch(bytecode()) {
2952 case Bytecodes::_fast_agetfield:
2953 {
2954 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
2955 __ verify_oop(R17_tos);
2956 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
2957
2958 __ bind(LisVolatile);
2959 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
2960 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
2961 __ verify_oop(R17_tos);
2962 __ twi_0(R17_tos);
2963 __ isync();
2964 break;
2965 }
2966 case Bytecodes::_fast_igetfield:
2967 {
2968 __ lwax(R17_tos, Rclass_or_obj, Roffset);
2969 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
2970
2971 __ bind(LisVolatile);
2972 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
2973 __ lwax(R17_tos, Rclass_or_obj, Roffset);
2974 __ twi_0(R17_tos);
2975 __ isync();
2976 break;
2977 }
2978 case Bytecodes::_fast_lgetfield:
2979 {
2980 __ ldx(R17_tos, Rclass_or_obj, Roffset);
2981 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
2982
2983 __ bind(LisVolatile);
2984 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
2985 __ ldx(R17_tos, Rclass_or_obj, Roffset);
2986 __ twi_0(R17_tos);
2987 __ isync();
2988 break;
2989 }
2990 case Bytecodes::_fast_bgetfield:
2991 {
2992 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
2993 __ extsb(R17_tos, R17_tos);
2994 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
2995
2996 __ bind(LisVolatile);
2997 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
2998 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
2999 __ twi_0(R17_tos);
3000 __ extsb(R17_tos, R17_tos);
3001 __ isync();
3002 break;
3003 }
3004 case Bytecodes::_fast_cgetfield:
3005 {
3006 __ lhzx(R17_tos, Rclass_or_obj, Roffset);
3007 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3008
3009 __ bind(LisVolatile);
3010 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3011 __ lhzx(R17_tos, Rclass_or_obj, Roffset);
3012 __ twi_0(R17_tos);
3013 __ isync();
3014 break;
3015 }
3016 case Bytecodes::_fast_sgetfield:
3017 {
3018 __ lhax(R17_tos, Rclass_or_obj, Roffset);
3019 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3020
3021 __ bind(LisVolatile);
3022 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3023 __ lhax(R17_tos, Rclass_or_obj, Roffset);
3024 __ twi_0(R17_tos);
3025 __ isync();
3026 break;
3027 }
3028 case Bytecodes::_fast_fgetfield:
3029 {
3030 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3031 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3032
3033 __ bind(LisVolatile);
3034 Label Ldummy;
3035 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3036 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3037 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3038 __ bne_predict_not_taken(CCR0, Ldummy);
3039 __ bind(Ldummy);
3040 __ isync();
3041 break;
3042 }
3043 case Bytecodes::_fast_dgetfield:
3044 {
3045 __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
3046 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3047
3048 __ bind(LisVolatile);
3049 Label Ldummy;
3050 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3051 __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
3052 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3053 __ bne_predict_not_taken(CCR0, Ldummy);
3054 __ bind(Ldummy);
3055 __ isync();
3056 break;
3057 }
3058 default: ShouldNotReachHere();
3059 }
3060 }
3061
3062 void TemplateTable::fast_xaccess(TosState state) {
3063 transition(vtos, state);
3064
3065 Label LisVolatile;
3066 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
3067 const Register Rcache = R3_ARG1,
3068 Rclass_or_obj = R17_tos,
3069 Roffset = R22_tmp2,
3070 Rflags = R23_tmp3,
3071 Rscratch = R12_scratch2;
3072
3073 __ ld(Rclass_or_obj, 0, R18_locals);
3074
3075 // Constant pool already resolved. Get the field offset.
3076 __ get_cache_and_index_at_bcp(Rcache, 2);
3077 load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
3078
3079 // JVMTI support not needed, since we switch back to single bytecode as soon as debugger attaches.
3080
3081 // Needed to report exception at the correct bcp.
3082 __ addi(R14_bcp, R14_bcp, 1);
3083
3084 // Get the load address.
3085 __ null_check_throw(Rclass_or_obj, -1, Rscratch);
3086
3087 // Get volatile flag.
3088 __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
3089 __ bne(CCR0, LisVolatile);
3090
3091 switch(state) {
3092 case atos:
3093 {
3094 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3095 __ verify_oop(R17_tos);
3096 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3097
3098 __ bind(LisVolatile);
3099 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3100 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3101 __ verify_oop(R17_tos);
3102 __ twi_0(R17_tos);
3103 __ isync();
3104 break;
3105 }
3106 case itos:
3107 {
3108 __ lwax(R17_tos, Rclass_or_obj, Roffset);
3109 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3110
3111 __ bind(LisVolatile);
3112 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3113 __ lwax(R17_tos, Rclass_or_obj, Roffset);
3114 __ twi_0(R17_tos);
3115 __ isync();
3116 break;
3117 }
3118 case ftos:
3119 {
3120 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3121 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3122
3123 __ bind(LisVolatile);
3124 Label Ldummy;
3125 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3126 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3127 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3128 __ bne_predict_not_taken(CCR0, Ldummy);
3129 __ bind(Ldummy);
3130 __ isync();
3131 break;
3132 }
3133 default: ShouldNotReachHere();
3134 }
3135 __ addi(R14_bcp, R14_bcp, -1);
3136 }
3137
3138 // ============================================================================
3139 // Calls
3140
3141 // Common code for invoke
3142 //
3143 // Input:
3144 // - byte_no
3145 //
3146 // Output:
3147 // - Rmethod: The method to invoke next.
3148 // - Rret_addr: The return address to return to.
3149 // - Rindex: MethodType (invokehandle) or CallSite obj (invokedynamic)
3150 // - Rrecv: Cache for "this" pointer, might be noreg if static call.
3151 // - Rflags: Method flags from const pool cache.
3152 //
3153 // Kills:
3154 // - Rscratch1
3155 //
3156 void TemplateTable::prepare_invoke(int byte_no,
3157 Register Rmethod, // linked method (or i-klass)
3158 Register Rret_addr,// return address
3159 Register Rindex, // itable index, MethodType, etc.
3160 Register Rrecv, // If caller wants to see it.
3161 Register Rflags, // If caller wants to test it.
3162 Register Rscratch
3163 ) {
3164 // Determine flags.
3165 const Bytecodes::Code code = bytecode();
3166 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3167 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3168 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3169 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3170 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3171 const bool load_receiver = (Rrecv != noreg);
3172 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3173
3174 assert_different_registers(Rmethod, Rindex, Rflags, Rscratch);
3175 assert_different_registers(Rmethod, Rrecv, Rflags, Rscratch);
3176 assert_different_registers(Rret_addr, Rscratch);
3177
3178 load_invoke_cp_cache_entry(byte_no, Rmethod, Rindex, Rflags, is_invokevirtual, false, is_invokedynamic);
3179
3180 // Saving of SP done in call_from_interpreter.
3181
3182 // Maybe push "appendix" to arguments.
3183 if (is_invokedynamic || is_invokehandle) {
3184 Label Ldone;
3185 __ rldicl_(R0, Rflags, 64-ConstantPoolCacheEntry::has_appendix_shift, 63);
3186 __ beq(CCR0, Ldone);
3187 // Push "appendix" (MethodType, CallSite, etc.).
3188 // This must be done before we get the receiver,
3189 // since the parameter_size includes it.
3190 __ load_resolved_reference_at_index(Rscratch, Rindex);
3191 __ verify_oop(Rscratch);
3192 __ push_ptr(Rscratch);
3193 __ bind(Ldone);
3194 }
3195
3196 // Load receiver if needed (after appendix is pushed so parameter size is correct).
3197 if (load_receiver) {
3198 const Register Rparam_count = Rscratch;
3199 __ andi(Rparam_count, Rflags, ConstantPoolCacheEntry::parameter_size_mask);
3200 __ load_receiver(Rparam_count, Rrecv);
3201 __ verify_oop(Rrecv);
3202 }
3203
3204 // Get return address.
3205 {
3206 Register Rtable_addr = Rscratch;
3207 Register Rret_type = Rret_addr;
3208 address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3209
3210 // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3211 __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3212 __ load_dispatch_table(Rtable_addr, (address*)table_addr);
3213 __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3214 // Get return address.
3215 __ ldx(Rret_addr, Rtable_addr, Rret_type);
3216 }
3217 }
3218
3219 // Helper for virtual calls. Load target out of vtable and jump off!
3220 // Kills all passed registers.
3221 void TemplateTable::generate_vtable_call(Register Rrecv_klass, Register Rindex, Register Rret, Register Rtemp) {
3222
3223 assert_different_registers(Rrecv_klass, Rtemp, Rret);
3224 const Register Rtarget_method = Rindex;
3225
3226 // Get target method & entry point.
3227 const int base = InstanceKlass::vtable_start_offset() * wordSize;
3228 // Calc vtable addr scale the vtable index by 8.
3229 __ sldi(Rindex, Rindex, exact_log2(vtableEntry::size() * wordSize));
3230 // Load target.
3231 __ addi(Rrecv_klass, Rrecv_klass, base + vtableEntry::method_offset_in_bytes());
3232 __ ldx(Rtarget_method, Rindex, Rrecv_klass);
3233 __ call_from_interpreter(Rtarget_method, Rret, Rrecv_klass /* scratch1 */, Rtemp /* scratch2 */);
3234 }
3235
3236 // Virtual or final call. Final calls are rewritten on the fly to run through "fast_finalcall" next time.
3237 void TemplateTable::invokevirtual(int byte_no) {
3238 transition(vtos, vtos);
3239
3240 Register Rtable_addr = R11_scratch1,
3241 Rret_type = R12_scratch2,
3242 Rret_addr = R5_ARG3,
3243 Rflags = R22_tmp2, // Should survive C call.
3244 Rrecv = R3_ARG1,
3245 Rrecv_klass = Rrecv,
3246 Rvtableindex_or_method = R31, // Should survive C call.
3247 Rnum_params = R4_ARG2,
3248 Rnew_bc = R6_ARG4;
3249
3250 Label LnotFinal;
3251
3252 load_invoke_cp_cache_entry(byte_no, Rvtableindex_or_method, noreg, Rflags, /*virtual*/ true, false, false);
3253
3254 __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_vfinal_shift);
3255 __ bfalse(CCR0, LnotFinal);
3256
3257 patch_bytecode(Bytecodes::_fast_invokevfinal, Rnew_bc, R12_scratch2);
3258 invokevfinal_helper(Rvtableindex_or_method, Rflags, R11_scratch1, R12_scratch2);
3259
3260 __ align(32, 12);
3261 __ bind(LnotFinal);
3262 // Load "this" pointer (receiver).
3263 __ rldicl(Rnum_params, Rflags, 64, 48);
3264 __ load_receiver(Rnum_params, Rrecv);
3265 __ verify_oop(Rrecv);
3266
3267 // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3268 __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3269 __ load_dispatch_table(Rtable_addr, Interpreter::invoke_return_entry_table());
3270 __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3271 __ ldx(Rret_addr, Rret_type, Rtable_addr);
3272 __ null_check_throw(Rrecv, oopDesc::klass_offset_in_bytes(), R11_scratch1);
3273 __ load_klass(Rrecv_klass, Rrecv);
3274 __ verify_klass_ptr(Rrecv_klass);
3275 __ profile_virtual_call(Rrecv_klass, R11_scratch1, R12_scratch2, false);
3276
3277 generate_vtable_call(Rrecv_klass, Rvtableindex_or_method, Rret_addr, R11_scratch1);
3278 }
3279
3280 void TemplateTable::fast_invokevfinal(int byte_no) {
3281 transition(vtos, vtos);
3282
3283 assert(byte_no == f2_byte, "use this argument");
3284 Register Rflags = R22_tmp2,
3285 Rmethod = R31;
3286 load_invoke_cp_cache_entry(byte_no, Rmethod, noreg, Rflags, /*virtual*/ true, /*is_invokevfinal*/ true, false);
3287 invokevfinal_helper(Rmethod, Rflags, R11_scratch1, R12_scratch2);
3288 }
3289
3290 void TemplateTable::invokevfinal_helper(Register Rmethod, Register Rflags, Register Rscratch1, Register Rscratch2) {
3291
3292 assert_different_registers(Rmethod, Rflags, Rscratch1, Rscratch2);
3293
3294 // Load receiver from stack slot.
3295 Register Rrecv = Rscratch2;
3296 Register Rnum_params = Rrecv;
3297
3298 __ ld(Rnum_params, in_bytes(Method::const_offset()), Rmethod);
3299 __ lhz(Rnum_params /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), Rnum_params);
3300
3301 // Get return address.
3302 Register Rtable_addr = Rscratch1,
3303 Rret_addr = Rflags,
3304 Rret_type = Rret_addr;
3305 // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3306 __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3307 __ load_dispatch_table(Rtable_addr, Interpreter::invoke_return_entry_table());
3308 __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3309 __ ldx(Rret_addr, Rret_type, Rtable_addr);
3310
3311 // Load receiver and receiver NULL check.
3312 __ load_receiver(Rnum_params, Rrecv);
3313 __ null_check_throw(Rrecv, -1, Rscratch1);
3314
3315 __ profile_final_call(Rrecv, Rscratch1);
3316
3317 // Do the call.
3318 __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1, Rscratch2);
3319 }
3320
3321 void TemplateTable::invokespecial(int byte_no) {
3322 assert(byte_no == f1_byte, "use this argument");
3323 transition(vtos, vtos);
3324
3325 Register Rtable_addr = R3_ARG1,
3326 Rret_addr = R4_ARG2,
3327 Rflags = R5_ARG3,
3328 Rreceiver = R6_ARG4,
3329 Rmethod = R31;
3330
3331 prepare_invoke(byte_no, Rmethod, Rret_addr, noreg, Rreceiver, Rflags, R11_scratch1);
3332
3333 // Receiver NULL check.
3334 __ null_check_throw(Rreceiver, -1, R11_scratch1);
3335
3336 __ profile_call(R11_scratch1, R12_scratch2);
3337 __ call_from_interpreter(Rmethod, Rret_addr, R11_scratch1, R12_scratch2);
3338 }
3339
3340 void TemplateTable::invokestatic(int byte_no) {
3341 assert(byte_no == f1_byte, "use this argument");
3342 transition(vtos, vtos);
3343
3344 Register Rtable_addr = R3_ARG1,
3345 Rret_addr = R4_ARG2,
3346 Rflags = R5_ARG3;
3347
3348 prepare_invoke(byte_no, R19_method, Rret_addr, noreg, noreg, Rflags, R11_scratch1);
3349
3350 __ profile_call(R11_scratch1, R12_scratch2);
3351 __ call_from_interpreter(R19_method, Rret_addr, R11_scratch1, R12_scratch2);
3352 }
3353
3354 void TemplateTable::invokeinterface_object_method(Register Rrecv_klass,
3355 Register Rret,
3356 Register Rflags,
3357 Register Rindex,
3358 Register Rtemp1,
3359 Register Rtemp2) {
3360
3361 assert_different_registers(Rindex, Rret, Rrecv_klass, Rflags, Rtemp1, Rtemp2);
3362 Label LnotFinal;
3363
3364 // Check for vfinal.
3365 __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_vfinal_shift);
3366 __ bfalse(CCR0, LnotFinal);
3367
3368 Register Rscratch = Rflags; // Rflags is dead now.
3369
3370 // Final call case.
3371 __ profile_final_call(Rtemp1, Rscratch);
3372 // Do the final call - the index (f2) contains the method.
3373 __ call_from_interpreter(Rindex, Rret, Rscratch, Rrecv_klass /* scratch */);
3374
3375 // Non-final callc case.
3376 __ bind(LnotFinal);
3377 __ profile_virtual_call(Rrecv_klass, Rtemp1, Rscratch, false);
3378 generate_vtable_call(Rrecv_klass, Rindex, Rret, Rscratch);
3379 }
3380
3381 void TemplateTable::invokeinterface(int byte_no) {
3382 assert(byte_no == f1_byte, "use this argument");
3383 transition(vtos, vtos);
3384
3385 const Register Rscratch1 = R11_scratch1,
3386 Rscratch2 = R12_scratch2,
3387 Rscratch3 = R9_ARG7,
3388 Rscratch4 = R10_ARG8,
3389 Rtable_addr = Rscratch2,
3390 Rinterface_klass = R5_ARG3,
3391 Rret_type = R8_ARG6,
3392 Rret_addr = Rret_type,
3393 Rindex = R6_ARG4,
3394 Rreceiver = R4_ARG2,
3395 Rrecv_klass = Rreceiver,
3396 Rflags = R7_ARG5;
3397
3398 prepare_invoke(byte_no, Rinterface_klass, Rret_addr, Rindex, Rreceiver, Rflags, Rscratch1);
3399
3400 // Get receiver klass.
3401 __ null_check_throw(Rreceiver, oopDesc::klass_offset_in_bytes(), Rscratch3);
3402 __ load_klass(Rrecv_klass, Rreceiver);
3403
3404 // Check corner case object method.
3405 Label LobjectMethod;
3406
3407 __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_forced_virtual_shift);
3408 __ btrue(CCR0, LobjectMethod);
3409
3410 // Fallthrough: The normal invokeinterface case.
3411 __ profile_virtual_call(Rrecv_klass, Rscratch1, Rscratch2, false);
3412
3413 // Find entry point to call.
3414 Label Lthrow_icc, Lthrow_ame;
3415 // Result will be returned in Rindex.
3416 __ mr(Rscratch4, Rrecv_klass);
3417 __ mr(Rscratch3, Rindex);
3418 __ lookup_interface_method(Rrecv_klass, Rinterface_klass, Rindex, Rindex, Rscratch1, Rscratch2, Lthrow_icc);
3419
3420 __ cmpdi(CCR0, Rindex, 0);
3421 __ beq(CCR0, Lthrow_ame);
3422 // Found entry. Jump off!
3423 __ call_from_interpreter(Rindex, Rret_addr, Rscratch1, Rscratch2);
3424
3425 // Vtable entry was NULL => Throw abstract method error.
3426 __ bind(Lthrow_ame);
3427 __ mr(Rrecv_klass, Rscratch4);
3428 __ mr(Rindex, Rscratch3);
3429 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3430
3431 // Interface was not found => Throw incompatible class change error.
3432 __ bind(Lthrow_icc);
3433 __ mr(Rrecv_klass, Rscratch4);
3434 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeError));
3435
3436 __ should_not_reach_here();
3437
3438 // Special case of invokeinterface called for virtual method of
3439 // java.lang.Object. See ConstantPoolCacheEntry::set_method() for details:
3440 // The invokeinterface was rewritten to a invokevirtual, hence we have
3441 // to handle this corner case. This code isn't produced by javac, but could
3442 // be produced by another compliant java compiler.
3443 __ bind(LobjectMethod);
3444 invokeinterface_object_method(Rrecv_klass, Rret_addr, Rflags, Rindex, Rscratch1, Rscratch2);
3445 }
3446
3447 void TemplateTable::invokedynamic(int byte_no) {
3448 transition(vtos, vtos);
3449
3450 const Register Rret_addr = R3_ARG1,
3451 Rflags = R4_ARG2,
3452 Rmethod = R22_tmp2,
3453 Rscratch1 = R11_scratch1,
3454 Rscratch2 = R12_scratch2;
3455
3456 prepare_invoke(byte_no, Rmethod, Rret_addr, Rscratch1, noreg, Rflags, Rscratch2);
3457
3458 // Profile this call.
3459 __ profile_call(Rscratch1, Rscratch2);
3460
3461 // Off we go. With the new method handles, we don't jump to a method handle
3462 // entry any more. Instead, we pushed an "appendix" in prepare invoke, which happens
3463 // to be the callsite object the bootstrap method returned. This is passed to a
3464 // "link" method which does the dispatch (Most likely just grabs the MH stored
3465 // inside the callsite and does an invokehandle).
3466 __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1 /* scratch1 */, Rscratch2 /* scratch2 */);
3467 }
3468
3469 void TemplateTable::invokehandle(int byte_no) {
3470 transition(vtos, vtos);
3471
3472 const Register Rret_addr = R3_ARG1,
3473 Rflags = R4_ARG2,
3474 Rrecv = R5_ARG3,
3475 Rmethod = R22_tmp2,
3476 Rscratch1 = R11_scratch1,
3477 Rscratch2 = R12_scratch2;
3478
3479 prepare_invoke(byte_no, Rmethod, Rret_addr, Rscratch1, Rrecv, Rflags, Rscratch2);
3480 __ verify_method_ptr(Rmethod);
3481 __ null_check_throw(Rrecv, -1, Rscratch2);
3482
3483 __ profile_final_call(Rrecv, Rscratch1);
3484
3485 // Still no call from handle => We call the method handle interpreter here.
3486 __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1 /* scratch1 */, Rscratch2 /* scratch2 */);
3487 }
3488
3489 // =============================================================================
3490 // Allocation
3491
3492 // Puts allocated obj ref onto the expression stack.
3493 void TemplateTable::_new() {
3494 transition(vtos, atos);
3495
3496 Label Lslow_case,
3497 Ldone,
3498 Linitialize_header,
3499 Lallocate_shared,
3500 Linitialize_object; // Including clearing the fields.
3501
3502 const Register RallocatedObject = R17_tos,
3503 RinstanceKlass = R9_ARG7,
3504 Rscratch = R11_scratch1,
3505 Roffset = R8_ARG6,
3506 Rinstance_size = Roffset,
3507 Rcpool = R4_ARG2,
3508 Rtags = R3_ARG1,
3509 Rindex = R5_ARG3;
3510
3511 const bool allow_shared_alloc = Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
3512
3513 // --------------------------------------------------------------------------
3514 // Check if fast case is possible.
3515
3516 // Load pointers to const pool and const pool's tags array.
3517 __ get_cpool_and_tags(Rcpool, Rtags);
3518 // Load index of constant pool entry.
3519 __ get_2_byte_integer_at_bcp(1, Rindex, InterpreterMacroAssembler::Unsigned);
3520
3521 if (UseTLAB) {
3522 // Make sure the class we're about to instantiate has been resolved
3523 // This is done before loading instanceKlass to be consistent with the order
3524 // how Constant Pool is updated (see ConstantPoolCache::klass_at_put).
3525 __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3526 __ lbzx(Rtags, Rindex, Rtags);
3527
3528 __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3529 __ bne(CCR0, Lslow_case);
3530
3531 // Get instanceKlass (load from Rcpool + sizeof(ConstantPool) + Rindex*BytesPerWord).
3532 __ sldi(Roffset, Rindex, LogBytesPerWord);
3533 __ addi(Rscratch, Rcpool, sizeof(ConstantPool));
3534 __ isync(); // Order load of instance Klass wrt. tags.
3535 __ ldx(RinstanceKlass, Roffset, Rscratch);
3536
3537 // Make sure klass is fully initialized and get instance_size.
3538 __ lbz(Rscratch, in_bytes(InstanceKlass::init_state_offset()), RinstanceKlass);
3539 __ lwz(Rinstance_size, in_bytes(Klass::layout_helper_offset()), RinstanceKlass);
3540
3541 __ cmpdi(CCR1, Rscratch, InstanceKlass::fully_initialized);
3542 // Make sure klass does not have has_finalizer, or is abstract, or interface or java/lang/Class.
3543 __ andi_(R0, Rinstance_size, Klass::_lh_instance_slow_path_bit); // slow path bit equals 0?
3544
3545 __ crnand(/*CR0 eq*/2, /*CR1 eq*/4+2, /*CR0 eq*/2); // slow path bit set or not fully initialized?
3546 __ beq(CCR0, Lslow_case);
3547
3548 // --------------------------------------------------------------------------
3549 // Fast case:
3550 // Allocate the instance.
3551 // 1) Try to allocate in the TLAB.
3552 // 2) If fail, and the TLAB is not full enough to discard, allocate in the shared Eden.
3553 // 3) If the above fails (or is not applicable), go to a slow case (creates a new TLAB, etc.).
3554
3555 Register RoldTopValue = RallocatedObject; // Object will be allocated here if it fits.
3556 Register RnewTopValue = R6_ARG4;
3557 Register RendValue = R7_ARG5;
3558
3559 // Check if we can allocate in the TLAB.
3560 __ ld(RoldTopValue, in_bytes(JavaThread::tlab_top_offset()), R16_thread);
3561 __ ld(RendValue, in_bytes(JavaThread::tlab_end_offset()), R16_thread);
3562
3563 __ add(RnewTopValue, Rinstance_size, RoldTopValue);
3564
3565 // If there is enough space, we do not CAS and do not clear.
3566 __ cmpld(CCR0, RnewTopValue, RendValue);
3567 __ bgt(CCR0, allow_shared_alloc ? Lallocate_shared : Lslow_case);
3568
3569 __ std(RnewTopValue, in_bytes(JavaThread::tlab_top_offset()), R16_thread);
3570
3571 if (ZeroTLAB) {
3572 // The fields have already been cleared.
3573 __ b(Linitialize_header);
3574 } else {
3575 // Initialize both the header and fields.
3576 __ b(Linitialize_object);
3577 }
3578
3579 // Fall through: TLAB was too small.
3580 if (allow_shared_alloc) {
3581 Register RtlabWasteLimitValue = R10_ARG8;
3582 Register RfreeValue = RnewTopValue;
3583
3584 __ bind(Lallocate_shared);
3585 // Check if tlab should be discarded (refill_waste_limit >= free).
3586 __ ld(RtlabWasteLimitValue, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), R16_thread);
3587 __ subf(RfreeValue, RoldTopValue, RendValue);
3588 __ srdi(RfreeValue, RfreeValue, LogHeapWordSize); // in dwords
3589 __ cmpld(CCR0, RtlabWasteLimitValue, RfreeValue);
3590 __ bge(CCR0, Lslow_case);
3591
3592 // Increment waste limit to prevent getting stuck on this slow path.
3593 __ addi(RtlabWasteLimitValue, RtlabWasteLimitValue, (int)ThreadLocalAllocBuffer::refill_waste_limit_increment());
3594 __ std(RtlabWasteLimitValue, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), R16_thread);
3595 }
3596 // else: No allocation in the shared eden. // fallthru: __ b(Lslow_case);
3597 }
3598 // else: Always go the slow path.
3599
3600 // --------------------------------------------------------------------------
3601 // slow case
3602 __ bind(Lslow_case);
3603 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), Rcpool, Rindex);
3604
3605 if (UseTLAB) {
3606 __ b(Ldone);
3607 // --------------------------------------------------------------------------
3608 // Init1: Zero out newly allocated memory.
3609
3610 if (!ZeroTLAB || allow_shared_alloc) {
3611 // Clear object fields.
3612 __ bind(Linitialize_object);
3613
3614 // Initialize remaining object fields.
3615 Register Rbase = Rtags;
3616 __ addi(Rinstance_size, Rinstance_size, 7 - (int)sizeof(oopDesc));
3617 __ addi(Rbase, RallocatedObject, sizeof(oopDesc));
3618 __ srdi(Rinstance_size, Rinstance_size, 3);
3619
3620 // Clear out object skipping header. Takes also care of the zero length case.
3621 __ clear_memory_doubleword(Rbase, Rinstance_size);
3622 // fallthru: __ b(Linitialize_header);
3623 }
3624
3625 // --------------------------------------------------------------------------
3626 // Init2: Initialize the header: mark, klass
3627 __ bind(Linitialize_header);
3628
3629 // Init mark.
3630 if (UseBiasedLocking) {
3631 __ ld(Rscratch, in_bytes(Klass::prototype_header_offset()), RinstanceKlass);
3632 } else {
3633 __ load_const_optimized(Rscratch, markOopDesc::prototype(), R0);
3634 }
3635 __ std(Rscratch, oopDesc::mark_offset_in_bytes(), RallocatedObject);
3636
3637 // Init klass.
3638 __ store_klass_gap(RallocatedObject);
3639 __ store_klass(RallocatedObject, RinstanceKlass, Rscratch); // klass (last for cms)
3640
3641 // Check and trigger dtrace event.
3642 {
3643 SkipIfEqualZero skip_if(_masm, Rscratch, &DTraceAllocProbes);
3644 __ push(atos);
3645 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc));
3646 __ pop(atos);
3647 }
3648 }
3649
3650 // continue
3651 __ bind(Ldone);
3652
3653 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3654 __ membar(Assembler::StoreStore);
3655 }
3656
3657 void TemplateTable::newarray() {
3658 transition(itos, atos);
3659
3660 __ lbz(R4, 1, R14_bcp);
3661 __ extsw(R5, R17_tos);
3662 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), R4, R5 /* size */);
3663
3664 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3665 __ membar(Assembler::StoreStore);
3666 }
3667
3668 void TemplateTable::anewarray() {
3669 transition(itos, atos);
3670
3671 __ get_constant_pool(R4);
3672 __ get_2_byte_integer_at_bcp(1, R5, InterpreterMacroAssembler::Unsigned);
3673 __ extsw(R6, R17_tos); // size
3674 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), R4 /* pool */, R5 /* index */, R6 /* size */);
3675
3676 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3677 __ membar(Assembler::StoreStore);
3678 }
3679
3680 // Allocate a multi dimensional array
3681 void TemplateTable::multianewarray() {
3682 transition(vtos, atos);
3683
3684 Register Rptr = R31; // Needs to survive C call.
3685
3686 // Put ndims * wordSize into frame temp slot
3687 __ lbz(Rptr, 3, R14_bcp);
3688 __ sldi(Rptr, Rptr, Interpreter::logStackElementSize);
3689 // Esp points past last_dim, so set to R4 to first_dim address.
3690 __ add(R4, Rptr, R15_esp);
3691 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), R4 /* first_size_address */);
3692 // Pop all dimensions off the stack.
3693 __ add(R15_esp, Rptr, R15_esp);
3694
3695 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3696 __ membar(Assembler::StoreStore);
3697 }
3698
3699 void TemplateTable::arraylength() {
3700 transition(atos, itos);
3701
3702 Label LnoException;
3703 __ verify_oop(R17_tos);
3704 __ null_check_throw(R17_tos, arrayOopDesc::length_offset_in_bytes(), R11_scratch1);
3705 __ lwa(R17_tos, arrayOopDesc::length_offset_in_bytes(), R17_tos);
3706 }
3707
3708 // ============================================================================
3709 // Typechecks
3710
3711 void TemplateTable::checkcast() {
3712 transition(atos, atos);
3713
3714 Label Ldone, Lis_null, Lquicked, Lresolved;
3715 Register Roffset = R6_ARG4,
3716 RobjKlass = R4_ARG2,
3717 RspecifiedKlass = R5_ARG3, // Generate_ClassCastException_verbose_handler will read value from this register.
3718 Rcpool = R11_scratch1,
3719 Rtags = R12_scratch2;
3720
3721 // Null does not pass.
3722 __ cmpdi(CCR0, R17_tos, 0);
3723 __ beq(CCR0, Lis_null);
3724
3725 // Get constant pool tag to find out if the bytecode has already been "quickened".
3726 __ get_cpool_and_tags(Rcpool, Rtags);
3727
3728 __ get_2_byte_integer_at_bcp(1, Roffset, InterpreterMacroAssembler::Unsigned);
3729
3730 __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3731 __ lbzx(Rtags, Rtags, Roffset);
3732
3733 __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3734 __ beq(CCR0, Lquicked);
3735
3736 // Call into the VM to "quicken" instanceof.
3737 __ push_ptr(); // for GC
3738 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3739 __ get_vm_result_2(RspecifiedKlass);
3740 __ pop_ptr(); // Restore receiver.
3741 __ b(Lresolved);
3742
3743 // Extract target class from constant pool.
3744 __ bind(Lquicked);
3745 __ sldi(Roffset, Roffset, LogBytesPerWord);
3746 __ addi(Rcpool, Rcpool, sizeof(ConstantPool));
3747 __ isync(); // Order load of specified Klass wrt. tags.
3748 __ ldx(RspecifiedKlass, Rcpool, Roffset);
3749
3750 // Do the checkcast.
3751 __ bind(Lresolved);
3752 // Get value klass in RobjKlass.
3753 __ load_klass(RobjKlass, R17_tos);
3754 // Generate a fast subtype check. Branch to cast_ok if no failure. Return 0 if failure.
3755 __ gen_subtype_check(RobjKlass, RspecifiedKlass, /*3 temp regs*/ Roffset, Rcpool, Rtags, /*target if subtype*/ Ldone);
3756
3757 // Not a subtype; so must throw exception
3758 // Target class oop is in register R6_ARG4 == RspecifiedKlass by convention.
3759 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ClassCastException_entry);
3760 __ mtctr(R11_scratch1);
3761 __ bctr();
3762
3763 // Profile the null case.
3764 __ align(32, 12);
3765 __ bind(Lis_null);
3766 __ profile_null_seen(R11_scratch1, Rtags); // Rtags used as scratch.
3767
3768 __ align(32, 12);
3769 __ bind(Ldone);
3770 }
3771
3772 // Output:
3773 // - tos == 0: Obj was null or not an instance of class.
3774 // - tos == 1: Obj was an instance of class.
3775 void TemplateTable::instanceof() {
3776 transition(atos, itos);
3777
3778 Label Ldone, Lis_null, Lquicked, Lresolved;
3779 Register Roffset = R5_ARG3,
3780 RobjKlass = R4_ARG2,
3781 RspecifiedKlass = R6_ARG4, // Generate_ClassCastException_verbose_handler will expect the value in this register.
3782 Rcpool = R11_scratch1,
3783 Rtags = R12_scratch2;
3784
3785 // Null does not pass.
3786 __ cmpdi(CCR0, R17_tos, 0);
3787 __ beq(CCR0, Lis_null);
3788
3789 // Get constant pool tag to find out if the bytecode has already been "quickened".
3790 __ get_cpool_and_tags(Rcpool, Rtags);
3791
3792 __ get_2_byte_integer_at_bcp(1, Roffset, InterpreterMacroAssembler::Unsigned);
3793
3794 __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3795 __ lbzx(Rtags, Rtags, Roffset);
3796
3797 __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3798 __ beq(CCR0, Lquicked);
3799
3800 // Call into the VM to "quicken" instanceof.
3801 __ push_ptr(); // for GC
3802 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3803 __ get_vm_result_2(RspecifiedKlass);
3804 __ pop_ptr(); // Restore receiver.
3805 __ b(Lresolved);
3806
3807 // Extract target class from constant pool.
3808 __ bind(Lquicked);
3809 __ sldi(Roffset, Roffset, LogBytesPerWord);
3810 __ addi(Rcpool, Rcpool, sizeof(ConstantPool));
3811 __ isync(); // Order load of specified Klass wrt. tags.
3812 __ ldx(RspecifiedKlass, Rcpool, Roffset);
3813
3814 // Do the checkcast.
3815 __ bind(Lresolved);
3816 // Get value klass in RobjKlass.
3817 __ load_klass(RobjKlass, R17_tos);
3818 // Generate a fast subtype check. Branch to cast_ok if no failure. Return 0 if failure.
3819 __ li(R17_tos, 1);
3820 __ gen_subtype_check(RobjKlass, RspecifiedKlass, /*3 temp regs*/ Roffset, Rcpool, Rtags, /*target if subtype*/ Ldone);
3821 __ li(R17_tos, 0);
3822
3823 if (ProfileInterpreter) {
3824 __ b(Ldone);
3825 }
3826
3827 // Profile the null case.
3828 __ align(32, 12);
3829 __ bind(Lis_null);
3830 __ profile_null_seen(Rcpool, Rtags); // Rcpool and Rtags used as scratch.
3831
3832 __ align(32, 12);
3833 __ bind(Ldone);
3834 }
3835
3836 // =============================================================================
3837 // Breakpoints
3838
3839 void TemplateTable::_breakpoint() {
3840 transition(vtos, vtos);
3841
3842 // Get the unpatched byte code.
3843 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), R19_method, R14_bcp);
3844 __ mr(R31, R3_RET);
3845
3846 // Post the breakpoint event.
3847 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), R19_method, R14_bcp);
3848
3849 // Complete the execution of original bytecode.
3850 __ dispatch_Lbyte_code(vtos, R31, Interpreter::normal_table(vtos));
3851 }
3852
3853 // =============================================================================
3854 // Exceptions
3855
3856 void TemplateTable::athrow() {
3857 transition(atos, vtos);
3858
3859 // Exception oop is in tos
3860 __ verify_oop(R17_tos);
3861
3862 __ null_check_throw(R17_tos, -1, R11_scratch1);
3863
3864 // Throw exception interpreter entry expects exception oop to be in R3.
3865 __ mr(R3_RET, R17_tos);
3866 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::throw_exception_entry());
3867 __ mtctr(R11_scratch1);
3868 __ bctr();
3869 }
3870
3871 // =============================================================================
3872 // Synchronization
3873 // Searches the basic object lock list on the stack for a free slot
3874 // and uses it to lock the obect in tos.
3875 //
3876 // Recursive locking is enabled by exiting the search if the same
3877 // object is already found in the list. Thus, a new basic lock obj lock
3878 // is allocated "higher up" in the stack and thus is found first
3879 // at next monitor exit.
3880 void TemplateTable::monitorenter() {
3881 transition(atos, vtos);
3882
3883 __ verify_oop(R17_tos);
3884
3885 Register Rcurrent_monitor = R11_scratch1,
3886 Rcurrent_obj = R12_scratch2,
3887 Robj_to_lock = R17_tos,
3888 Rscratch1 = R3_ARG1,
3889 Rscratch2 = R4_ARG2,
3890 Rscratch3 = R5_ARG3,
3891 Rcurrent_obj_addr = R6_ARG4;
3892
3893 // ------------------------------------------------------------------------------
3894 // Null pointer exception.
3895 __ null_check_throw(Robj_to_lock, -1, R11_scratch1);
3896
3897 // Try to acquire a lock on the object.
3898 // Repeat until succeeded (i.e., until monitorenter returns true).
3899
3900 // ------------------------------------------------------------------------------
3901 // Find a free slot in the monitor block.
3902 Label Lfound, Lexit, Lallocate_new;
3903 ConditionRegister found_free_slot = CCR0,
3904 found_same_obj = CCR1,
3905 reached_limit = CCR6;
3906 {
3907 Label Lloop, Lentry;
3908 Register Rlimit = Rcurrent_monitor;
3909
3910 // Set up search loop - start with topmost monitor.
3911 __ add(Rcurrent_obj_addr, BasicObjectLock::obj_offset_in_bytes(), R26_monitor);
3912
3913 __ ld(Rlimit, 0, R1_SP);
3914 __ addi(Rlimit, Rlimit, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes() - BasicObjectLock::obj_offset_in_bytes())); // Monitor base
3915
3916 // Check if any slot is present => short cut to allocation if not.
3917 __ cmpld(reached_limit, Rcurrent_obj_addr, Rlimit);
3918 __ bgt(reached_limit, Lallocate_new);
3919
3920 // Pre-load topmost slot.
3921 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
3922 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
3923 // The search loop.
3924 __ bind(Lloop);
3925 // Found free slot?
3926 __ cmpdi(found_free_slot, Rcurrent_obj, 0);
3927 // Is this entry for same obj? If so, stop the search and take the found
3928 // free slot or allocate a new one to enable recursive locking.
3929 __ cmpd(found_same_obj, Rcurrent_obj, Robj_to_lock);
3930 __ cmpld(reached_limit, Rcurrent_obj_addr, Rlimit);
3931 __ beq(found_free_slot, Lexit);
3932 __ beq(found_same_obj, Lallocate_new);
3933 __ bgt(reached_limit, Lallocate_new);
3934 // Check if last allocated BasicLockObj reached.
3935 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
3936 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
3937 // Next iteration if unchecked BasicObjectLocks exist on the stack.
3938 __ b(Lloop);
3939 }
3940
3941 // ------------------------------------------------------------------------------
3942 // Check if we found a free slot.
3943 __ bind(Lexit);
3944
3945 __ addi(Rcurrent_monitor, Rcurrent_obj_addr, -(frame::interpreter_frame_monitor_size() * wordSize) - BasicObjectLock::obj_offset_in_bytes());
3946 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, - frame::interpreter_frame_monitor_size() * wordSize);
3947 __ b(Lfound);
3948
3949 // We didn't find a free BasicObjLock => allocate one.
3950 __ align(32, 12);
3951 __ bind(Lallocate_new);
3952 __ add_monitor_to_stack(false, Rscratch1, Rscratch2);
3953 __ mr(Rcurrent_monitor, R26_monitor);
3954 __ addi(Rcurrent_obj_addr, R26_monitor, BasicObjectLock::obj_offset_in_bytes());
3955
3956 // ------------------------------------------------------------------------------
3957 // We now have a slot to lock.
3958 __ bind(Lfound);
3959
3960 // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
3961 // The object has already been poped from the stack, so the expression stack looks correct.
3962 __ addi(R14_bcp, R14_bcp, 1);
3963
3964 __ std(Robj_to_lock, 0, Rcurrent_obj_addr);
3965 __ lock_object(Rcurrent_monitor, Robj_to_lock);
3966
3967 // Check if there's enough space on the stack for the monitors after locking.
3968 Label Lskip_stack_check;
3969 // Optimization: If the monitors stack section is less then a std page size (4K) don't run
3970 // the stack check. There should be enough shadow pages to fit that in.
3971 __ ld(Rscratch3, 0, R1_SP);
3972 __ sub(Rscratch3, Rscratch3, R26_monitor);
3973 __ cmpdi(CCR0, Rscratch3, 4*K);
3974 __ blt(CCR0, Lskip_stack_check);
3975
3976 DEBUG_ONLY(__ untested("stack overflow check during monitor enter");)
3977 __ li(Rscratch1, 0);
3978 __ generate_stack_overflow_check_with_compare_and_throw(Rscratch1, Rscratch2);
3979
3980 __ align(32, 12);
3981 __ bind(Lskip_stack_check);
3982
3983 // The bcp has already been incremented. Just need to dispatch to next instruction.
3984 __ dispatch_next(vtos);
3985 }
3986
3987 void TemplateTable::monitorexit() {
3988 transition(atos, vtos);
3989 __ verify_oop(R17_tos);
3990
3991 Register Rcurrent_monitor = R11_scratch1,
3992 Rcurrent_obj = R12_scratch2,
3993 Robj_to_lock = R17_tos,
3994 Rcurrent_obj_addr = R3_ARG1,
3995 Rlimit = R4_ARG2;
3996 Label Lfound, Lillegal_monitor_state;
3997
3998 // Check corner case: unbalanced monitorEnter / Exit.
3999 __ ld(Rlimit, 0, R1_SP);
4000 __ addi(Rlimit, Rlimit, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base
4001
4002 // Null pointer check.
4003 __ null_check_throw(Robj_to_lock, -1, R11_scratch1);
4004
4005 __ cmpld(CCR0, R26_monitor, Rlimit);
4006 __ bgt(CCR0, Lillegal_monitor_state);
4007
4008 // Find the corresponding slot in the monitors stack section.
4009 {
4010 Label Lloop;
4011
4012 // Start with topmost monitor.
4013 __ addi(Rcurrent_obj_addr, R26_monitor, BasicObjectLock::obj_offset_in_bytes());
4014 __ addi(Rlimit, Rlimit, BasicObjectLock::obj_offset_in_bytes());
4015 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4016 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4017
4018 __ bind(Lloop);
4019 // Is this entry for same obj?
4020 __ cmpd(CCR0, Rcurrent_obj, Robj_to_lock);
4021 __ beq(CCR0, Lfound);
4022
4023 // Check if last allocated BasicLockObj reached.
4024
4025 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4026 __ cmpld(CCR0, Rcurrent_obj_addr, Rlimit);
4027 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4028
4029 // Next iteration if unchecked BasicObjectLocks exist on the stack.
4030 __ ble(CCR0, Lloop);
4031 }
4032
4033 // Fell through without finding the basic obj lock => throw up!
4034 __ bind(Lillegal_monitor_state);
4035 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
4036 __ should_not_reach_here();
4037
4038 __ align(32, 12);
4039 __ bind(Lfound);
4040 __ addi(Rcurrent_monitor, Rcurrent_obj_addr,
4041 -(frame::interpreter_frame_monitor_size() * wordSize) - BasicObjectLock::obj_offset_in_bytes());
4042 __ unlock_object(Rcurrent_monitor);
4043 }
4044
4045 // ============================================================================
4046 // Wide bytecodes
4047
4048 // Wide instructions. Simply redirects to the wide entry point for that instruction.
4049 void TemplateTable::wide() {
4050 transition(vtos, vtos);
4051
4052 const Register Rtable = R11_scratch1,
4053 Rindex = R12_scratch2,
4054 Rtmp = R0;
4055
4056 __ lbz(Rindex, 1, R14_bcp);
4057
4058 __ load_dispatch_table(Rtable, Interpreter::_wentry_point);
4059
4060 __ slwi(Rindex, Rindex, LogBytesPerWord);
4061 __ ldx(Rtmp, Rtable, Rindex);
4062 __ mtctr(Rtmp);
4063 __ bctr();
4064 // Note: the bcp increment step is part of the individual wide bytecode implementations.
4065 }
4066 #endif // !CC_INTERP