1 /*
2 * Copyright (c) 2014, 2015, Oracle and/or its affiliates. All rights reserved.
3 * Copyright 2013, 2015 SAP AG. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "precompiled.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "interpreter/interpreterRuntime.hpp"
30 #include "interpreter/interp_masm.hpp"
31 #include "interpreter/templateInterpreter.hpp"
32 #include "interpreter/templateTable.hpp"
33 #include "memory/universe.inline.hpp"
34 #include "oops/objArrayKlass.hpp"
35 #include "oops/oop.inline.hpp"
36 #include "prims/methodHandles.hpp"
37 #include "runtime/sharedRuntime.hpp"
38 #include "runtime/stubRoutines.hpp"
39 #include "runtime/synchronizer.hpp"
40 #include "utilities/macros.hpp"
41
42 #ifndef CC_INTERP
43
44 #undef __
45 #define __ _masm->
46
47 // ============================================================================
48 // Misc helpers
49
50 // Do an oop store like *(base + index) = val OR *(base + offset) = val
51 // (only one of both variants is possible at the same time).
52 // Index can be noreg.
53 // Kills:
54 // Rbase, Rtmp
55 static void do_oop_store(InterpreterMacroAssembler* _masm,
56 Register Rbase,
57 RegisterOrConstant offset,
58 Register Rval, // Noreg means always null.
59 Register Rtmp1,
60 Register Rtmp2,
61 Register Rtmp3,
62 BarrierSet::Name barrier,
63 bool precise,
64 bool check_null) {
65 assert_different_registers(Rtmp1, Rtmp2, Rtmp3, Rval, Rbase);
66
67 switch (barrier) {
68 #if INCLUDE_ALL_GCS
69 case BarrierSet::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 ShouldNotReachHere();
146 break;
147 default:
148 ShouldNotReachHere();
149 }
150 }
151
152 // ============================================================================
153 // Platform-dependent initialization
154
155 void TemplateTable::pd_initialize() {
156 // No ppc64 specific initialization.
157 }
158
159 Address TemplateTable::at_bcp(int offset) {
160 // Not used on ppc.
161 ShouldNotReachHere();
162 return Address();
163 }
164
165 // Patches the current bytecode (ptr to it located in bcp)
166 // in the bytecode stream with a new one.
167 void TemplateTable::patch_bytecode(Bytecodes::Code new_bc, Register Rnew_bc, Register Rtemp, bool load_bc_into_bc_reg /*=true*/, int byte_no) {
168 // With sharing on, may need to test method flag.
169 if (!RewriteBytecodes) return;
170 Label L_patch_done;
171
172 switch (new_bc) {
173 case Bytecodes::_fast_aputfield:
174 case Bytecodes::_fast_bputfield:
175 case Bytecodes::_fast_cputfield:
176 case Bytecodes::_fast_dputfield:
177 case Bytecodes::_fast_fputfield:
178 case Bytecodes::_fast_iputfield:
179 case Bytecodes::_fast_lputfield:
180 case Bytecodes::_fast_sputfield:
181 {
182 // We skip bytecode quickening for putfield instructions when
183 // the put_code written to the constant pool cache is zero.
184 // This is required so that every execution of this instruction
185 // calls out to InterpreterRuntime::resolve_get_put to do
186 // additional, required work.
187 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
188 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
189 __ get_cache_and_index_at_bcp(Rtemp /* dst = cache */, 1);
190 // ((*(cache+indices))>>((1+byte_no)*8))&0xFF:
191 #if defined(VM_LITTLE_ENDIAN)
192 __ lbz(Rnew_bc, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 1 + byte_no, Rtemp);
193 #else
194 __ lbz(Rnew_bc, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 7 - (1 + byte_no), Rtemp);
195 #endif
196 __ cmpwi(CCR0, Rnew_bc, 0);
197 __ li(Rnew_bc, (unsigned int)(unsigned char)new_bc);
198 __ beq(CCR0, L_patch_done);
199 // __ isync(); // acquire not needed
200 break;
201 }
202
203 default:
204 assert(byte_no == -1, "sanity");
205 if (load_bc_into_bc_reg) {
206 __ li(Rnew_bc, (unsigned int)(unsigned char)new_bc);
207 }
208 }
209
210 if (JvmtiExport::can_post_breakpoint()) {
211 Label L_fast_patch;
212 __ lbz(Rtemp, 0, R14_bcp);
213 __ cmpwi(CCR0, Rtemp, (unsigned int)(unsigned char)Bytecodes::_breakpoint);
214 __ bne(CCR0, L_fast_patch);
215 // Perform the quickening, slowly, in the bowels of the breakpoint table.
216 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), R19_method, R14_bcp, Rnew_bc);
217 __ b(L_patch_done);
218 __ bind(L_fast_patch);
219 }
220
221 // Patch bytecode.
222 __ stb(Rnew_bc, 0, R14_bcp);
223
224 __ bind(L_patch_done);
225 }
226
227 // ============================================================================
228 // Individual instructions
229
230 void TemplateTable::nop() {
231 transition(vtos, vtos);
232 // Nothing to do.
233 }
234
235 void TemplateTable::shouldnotreachhere() {
236 transition(vtos, vtos);
237 __ stop("shouldnotreachhere bytecode");
238 }
239
240 void TemplateTable::aconst_null() {
241 transition(vtos, atos);
242 __ li(R17_tos, 0);
243 }
244
245 void TemplateTable::iconst(int value) {
246 transition(vtos, itos);
247 assert(value >= -1 && value <= 5, "");
248 __ li(R17_tos, value);
249 }
250
251 void TemplateTable::lconst(int value) {
252 transition(vtos, ltos);
253 assert(value >= -1 && value <= 5, "");
254 __ li(R17_tos, value);
255 }
256
257 void TemplateTable::fconst(int value) {
258 transition(vtos, ftos);
259 static float zero = 0.0;
260 static float one = 1.0;
261 static float two = 2.0;
262 switch (value) {
263 default: ShouldNotReachHere();
264 case 0: {
265 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&zero, R0, true);
266 __ lfs(F15_ftos, simm16_offset, R11_scratch1);
267 break;
268 }
269 case 1: {
270 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&one, R0, true);
271 __ lfs(F15_ftos, simm16_offset, R11_scratch1);
272 break;
273 }
274 case 2: {
275 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&two, R0, true);
276 __ lfs(F15_ftos, simm16_offset, R11_scratch1);
277 break;
278 }
279 }
280 }
281
282 void TemplateTable::dconst(int value) {
283 transition(vtos, dtos);
284 static double zero = 0.0;
285 static double one = 1.0;
286 switch (value) {
287 case 0: {
288 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&zero, R0, true);
289 __ lfd(F15_ftos, simm16_offset, R11_scratch1);
290 break;
291 }
292 case 1: {
293 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&one, R0, true);
294 __ lfd(F15_ftos, simm16_offset, R11_scratch1);
295 break;
296 }
297 default: ShouldNotReachHere();
298 }
299 }
300
301 void TemplateTable::bipush() {
302 transition(vtos, itos);
303 __ lbz(R17_tos, 1, R14_bcp);
304 __ extsb(R17_tos, R17_tos);
305 }
306
307 void TemplateTable::sipush() {
308 transition(vtos, itos);
309 __ get_2_byte_integer_at_bcp(1, R17_tos, InterpreterMacroAssembler::Signed);
310 }
311
312 void TemplateTable::ldc(bool wide) {
313 Register Rscratch1 = R11_scratch1,
314 Rscratch2 = R12_scratch2,
315 Rcpool = R3_ARG1;
316
317 transition(vtos, vtos);
318 Label notInt, notClass, exit;
319
320 __ get_cpool_and_tags(Rcpool, Rscratch2); // Set Rscratch2 = &tags.
321 if (wide) { // Read index.
322 __ get_2_byte_integer_at_bcp(1, Rscratch1, InterpreterMacroAssembler::Unsigned);
323 } else {
324 __ lbz(Rscratch1, 1, R14_bcp);
325 }
326
327 const int base_offset = ConstantPool::header_size() * wordSize;
328 const int tags_offset = Array<u1>::base_offset_in_bytes();
329
330 // Get type from tags.
331 __ addi(Rscratch2, Rscratch2, tags_offset);
332 __ lbzx(Rscratch2, Rscratch2, Rscratch1);
333
334 __ cmpwi(CCR0, Rscratch2, JVM_CONSTANT_UnresolvedClass); // Unresolved class?
335 __ cmpwi(CCR1, Rscratch2, JVM_CONSTANT_UnresolvedClassInError); // Unresolved class in error state?
336 __ cror(CCR0, Assembler::equal, CCR1, Assembler::equal);
337
338 // Resolved class - need to call vm to get java mirror of the class.
339 __ cmpwi(CCR1, Rscratch2, JVM_CONSTANT_Class);
340 __ crnor(CCR0, Assembler::equal, CCR1, Assembler::equal); // Neither resolved class nor unresolved case from above?
341 __ beq(CCR0, notClass);
342
343 __ li(R4, wide ? 1 : 0);
344 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), R4);
345 __ push(atos);
346 __ b(exit);
347
348 __ align(32, 12);
349 __ bind(notClass);
350 __ addi(Rcpool, Rcpool, base_offset);
351 __ sldi(Rscratch1, Rscratch1, LogBytesPerWord);
352 __ cmpdi(CCR0, Rscratch2, JVM_CONSTANT_Integer);
353 __ bne(CCR0, notInt);
354 __ lwax(R17_tos, Rcpool, Rscratch1);
355 __ push(itos);
356 __ b(exit);
357
358 __ align(32, 12);
359 __ bind(notInt);
360 #ifdef ASSERT
361 // String and Object are rewritten to fast_aldc
362 __ cmpdi(CCR0, Rscratch2, JVM_CONSTANT_Float);
363 __ asm_assert_eq("unexpected type", 0x8765);
364 #endif
365 __ lfsx(F15_ftos, Rcpool, Rscratch1);
366 __ push(ftos);
367
368 __ align(32, 12);
369 __ bind(exit);
370 }
371
372 // Fast path for caching oop constants.
373 void TemplateTable::fast_aldc(bool wide) {
374 transition(vtos, atos);
375
376 int index_size = wide ? sizeof(u2) : sizeof(u1);
377 const Register Rscratch = R11_scratch1;
378 Label is_null;
379
380 // We are resolved if the resolved reference cache entry contains a
381 // non-null object (CallSite, etc.)
382 __ get_cache_index_at_bcp(Rscratch, 1, index_size); // Load index.
383 __ load_resolved_reference_at_index(R17_tos, Rscratch, &is_null);
384 __ verify_oop(R17_tos);
385 __ dispatch_epilog(atos, Bytecodes::length_for(bytecode()));
386
387 __ bind(is_null);
388 __ load_const_optimized(R3_ARG1, (int)bytecode());
389
390 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
391
392 // First time invocation - must resolve first.
393 __ call_VM(R17_tos, entry, R3_ARG1);
394 __ 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 __ lfdx(F15_ftos, Rcpool, Rindex);
423 __ push(dtos);
424 __ b(Lexit);
425
426 __ bind(Llong);
427 __ ldx(R17_tos, Rcpool, Rindex);
428 __ push(ltos);
429
430 __ bind(Lexit);
431 }
432
433 // Get the locals index located in the bytecode stream at bcp + offset.
434 void TemplateTable::locals_index(Register Rdst, int offset) {
435 __ lbz(Rdst, offset, R14_bcp);
436 }
437
438 void TemplateTable::iload() {
439 transition(vtos, itos);
440
441 // Get the local value into tos
442 const Register Rindex = R22_tmp2;
443 locals_index(Rindex);
444
445 // Rewrite iload,iload pair into fast_iload2
446 // iload,caload pair into fast_icaload
447 if (RewriteFrequentPairs) {
448 Label Lrewrite, Ldone;
449 Register Rnext_byte = R3_ARG1,
450 Rrewrite_to = R6_ARG4,
451 Rscratch = R11_scratch1;
452
453 // get next byte
454 __ lbz(Rnext_byte, Bytecodes::length_for(Bytecodes::_iload), R14_bcp);
455
456 // if _iload, wait to rewrite to iload2. We only want to rewrite the
457 // last two iloads in a pair. Comparing against fast_iload means that
458 // the next bytecode is neither an iload or a caload, and therefore
459 // an iload pair.
460 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_iload);
461 __ beq(CCR0, Ldone);
462
463 __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_iload);
464 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iload2);
465 __ beq(CCR1, Lrewrite);
466
467 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_caload);
468 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_icaload);
469 __ beq(CCR0, Lrewrite);
470
471 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iload);
472
473 __ bind(Lrewrite);
474 patch_bytecode(Bytecodes::_iload, Rrewrite_to, Rscratch, false);
475 __ bind(Ldone);
476 }
477
478 __ load_local_int(R17_tos, Rindex, Rindex);
479 }
480
481 // Load 2 integers in a row without dispatching
482 void TemplateTable::fast_iload2() {
483 transition(vtos, itos);
484
485 __ lbz(R3_ARG1, 1, R14_bcp);
486 __ lbz(R17_tos, Bytecodes::length_for(Bytecodes::_iload) + 1, R14_bcp);
487
488 __ load_local_int(R3_ARG1, R11_scratch1, R3_ARG1);
489 __ load_local_int(R17_tos, R12_scratch2, R17_tos);
490 __ push_i(R3_ARG1);
491 }
492
493 void TemplateTable::fast_iload() {
494 transition(vtos, itos);
495 // Get the local value into tos
496
497 const Register Rindex = R11_scratch1;
498 locals_index(Rindex);
499 __ load_local_int(R17_tos, Rindex, Rindex);
500 }
501
502 // Load a local variable type long from locals area to TOS cache register.
503 // Local index resides in bytecodestream.
504 void TemplateTable::lload() {
505 transition(vtos, ltos);
506
507 const Register Rindex = R11_scratch1;
508 locals_index(Rindex);
509 __ load_local_long(R17_tos, Rindex, Rindex);
510 }
511
512 void TemplateTable::fload() {
513 transition(vtos, ftos);
514
515 const Register Rindex = R11_scratch1;
516 locals_index(Rindex);
517 __ load_local_float(F15_ftos, Rindex, Rindex);
518 }
519
520 void TemplateTable::dload() {
521 transition(vtos, dtos);
522
523 const Register Rindex = R11_scratch1;
524 locals_index(Rindex);
525 __ load_local_double(F15_ftos, Rindex, Rindex);
526 }
527
528 void TemplateTable::aload() {
529 transition(vtos, atos);
530
531 const Register Rindex = R11_scratch1;
532 locals_index(Rindex);
533 __ load_local_ptr(R17_tos, Rindex, Rindex);
534 }
535
536 void TemplateTable::locals_index_wide(Register Rdst) {
537 // Offset is 2, not 1, because Lbcp points to wide prefix code.
538 __ get_2_byte_integer_at_bcp(2, Rdst, InterpreterMacroAssembler::Unsigned);
539 }
540
541 void TemplateTable::wide_iload() {
542 // Get the local value into tos.
543
544 const Register Rindex = R11_scratch1;
545 locals_index_wide(Rindex);
546 __ load_local_int(R17_tos, Rindex, Rindex);
547 }
548
549 void TemplateTable::wide_lload() {
550 transition(vtos, ltos);
551
552 const Register Rindex = R11_scratch1;
553 locals_index_wide(Rindex);
554 __ load_local_long(R17_tos, Rindex, Rindex);
555 }
556
557 void TemplateTable::wide_fload() {
558 transition(vtos, ftos);
559
560 const Register Rindex = R11_scratch1;
561 locals_index_wide(Rindex);
562 __ load_local_float(F15_ftos, Rindex, Rindex);
563 }
564
565 void TemplateTable::wide_dload() {
566 transition(vtos, dtos);
567
568 const Register Rindex = R11_scratch1;
569 locals_index_wide(Rindex);
570 __ load_local_double(F15_ftos, Rindex, Rindex);
571 }
572
573 void TemplateTable::wide_aload() {
574 transition(vtos, atos);
575
576 const Register Rindex = R11_scratch1;
577 locals_index_wide(Rindex);
578 __ load_local_ptr(R17_tos, Rindex, Rindex);
579 }
580
581 void TemplateTable::iaload() {
582 transition(itos, itos);
583
584 const Register Rload_addr = R3_ARG1,
585 Rarray = R4_ARG2,
586 Rtemp = R5_ARG3;
587 __ index_check(Rarray, R17_tos /* index */, LogBytesPerInt, Rtemp, Rload_addr);
588 __ lwa(R17_tos, arrayOopDesc::base_offset_in_bytes(T_INT), Rload_addr);
589 }
590
591 void TemplateTable::laload() {
592 transition(itos, ltos);
593
594 const Register Rload_addr = R3_ARG1,
595 Rarray = R4_ARG2,
596 Rtemp = R5_ARG3;
597 __ index_check(Rarray, R17_tos /* index */, LogBytesPerLong, Rtemp, Rload_addr);
598 __ ld(R17_tos, arrayOopDesc::base_offset_in_bytes(T_LONG), Rload_addr);
599 }
600
601 void TemplateTable::faload() {
602 transition(itos, ftos);
603
604 const Register Rload_addr = R3_ARG1,
605 Rarray = R4_ARG2,
606 Rtemp = R5_ARG3;
607 __ index_check(Rarray, R17_tos /* index */, LogBytesPerInt, Rtemp, Rload_addr);
608 __ lfs(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Rload_addr);
609 }
610
611 void TemplateTable::daload() {
612 transition(itos, dtos);
613
614 const Register Rload_addr = R3_ARG1,
615 Rarray = R4_ARG2,
616 Rtemp = R5_ARG3;
617 __ index_check(Rarray, R17_tos /* index */, LogBytesPerLong, Rtemp, Rload_addr);
618 __ lfd(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Rload_addr);
619 }
620
621 void TemplateTable::aaload() {
622 transition(itos, atos);
623
624 // tos: index
625 // result tos: array
626 const Register Rload_addr = R3_ARG1,
627 Rarray = R4_ARG2,
628 Rtemp = R5_ARG3;
629 __ index_check(Rarray, R17_tos /* index */, UseCompressedOops ? 2 : LogBytesPerWord, Rtemp, Rload_addr);
630 __ load_heap_oop(R17_tos, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Rload_addr);
631 __ verify_oop(R17_tos);
632 //__ dcbt(R17_tos); // prefetch
633 }
634
635 void TemplateTable::baload() {
636 transition(itos, itos);
637
638 const Register Rload_addr = R3_ARG1,
639 Rarray = R4_ARG2,
640 Rtemp = R5_ARG3;
641 __ index_check(Rarray, R17_tos /* index */, 0, Rtemp, Rload_addr);
642 __ lbz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_BYTE), Rload_addr);
643 __ extsb(R17_tos, R17_tos);
644 }
645
646 void TemplateTable::caload() {
647 transition(itos, itos);
648
649 const Register Rload_addr = R3_ARG1,
650 Rarray = R4_ARG2,
651 Rtemp = R5_ARG3;
652 __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
653 __ lhz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rload_addr);
654 }
655
656 // Iload followed by caload frequent pair.
657 void TemplateTable::fast_icaload() {
658 transition(vtos, itos);
659
660 const Register Rload_addr = R3_ARG1,
661 Rarray = R4_ARG2,
662 Rtemp = R11_scratch1;
663
664 locals_index(R17_tos);
665 __ load_local_int(R17_tos, Rtemp, R17_tos);
666 __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
667 __ lhz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rload_addr);
668 }
669
670 void TemplateTable::saload() {
671 transition(itos, itos);
672
673 const Register Rload_addr = R11_scratch1,
674 Rarray = R12_scratch2,
675 Rtemp = R3_ARG1;
676 __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
677 __ lha(R17_tos, arrayOopDesc::base_offset_in_bytes(T_SHORT), Rload_addr);
678 }
679
680 void TemplateTable::iload(int n) {
681 transition(vtos, itos);
682
683 __ lwz(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
684 }
685
686 void TemplateTable::lload(int n) {
687 transition(vtos, ltos);
688
689 __ ld(R17_tos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
690 }
691
692 void TemplateTable::fload(int n) {
693 transition(vtos, ftos);
694
695 __ lfs(F15_ftos, Interpreter::local_offset_in_bytes(n), R18_locals);
696 }
697
698 void TemplateTable::dload(int n) {
699 transition(vtos, dtos);
700
701 __ lfd(F15_ftos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
702 }
703
704 void TemplateTable::aload(int n) {
705 transition(vtos, atos);
706
707 __ ld(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
708 }
709
710 void TemplateTable::aload_0() {
711 transition(vtos, atos);
712 // According to bytecode histograms, the pairs:
713 //
714 // _aload_0, _fast_igetfield
715 // _aload_0, _fast_agetfield
716 // _aload_0, _fast_fgetfield
717 //
718 // occur frequently. If RewriteFrequentPairs is set, the (slow)
719 // _aload_0 bytecode checks if the next bytecode is either
720 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
721 // rewrites the current bytecode into a pair bytecode; otherwise it
722 // rewrites the current bytecode into _0 that doesn't do
723 // the pair check anymore.
724 //
725 // Note: If the next bytecode is _getfield, the rewrite must be
726 // delayed, otherwise we may miss an opportunity for a pair.
727 //
728 // Also rewrite frequent pairs
729 // aload_0, aload_1
730 // aload_0, iload_1
731 // These bytecodes with a small amount of code are most profitable
732 // to rewrite.
733
734 if (RewriteFrequentPairs) {
735
736 Label Lrewrite, Ldont_rewrite;
737 Register Rnext_byte = R3_ARG1,
738 Rrewrite_to = R6_ARG4,
739 Rscratch = R11_scratch1;
740
741 // Get next byte.
742 __ lbz(Rnext_byte, Bytecodes::length_for(Bytecodes::_aload_0), R14_bcp);
743
744 // If _getfield, wait to rewrite. We only want to rewrite the last two bytecodes in a pair.
745 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_getfield);
746 __ beq(CCR0, Ldont_rewrite);
747
748 __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_igetfield);
749 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iaccess_0);
750 __ beq(CCR1, Lrewrite);
751
752 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_agetfield);
753 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_aaccess_0);
754 __ beq(CCR0, Lrewrite);
755
756 __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_fgetfield);
757 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_faccess_0);
758 __ beq(CCR1, Lrewrite);
759
760 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_aload_0);
761
762 __ bind(Lrewrite);
763 patch_bytecode(Bytecodes::_aload_0, Rrewrite_to, Rscratch, false);
764 __ bind(Ldont_rewrite);
765 }
766
767 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
768 aload(0);
769 }
770
771 void TemplateTable::istore() {
772 transition(itos, vtos);
773
774 const Register Rindex = R11_scratch1;
775 locals_index(Rindex);
776 __ store_local_int(R17_tos, Rindex);
777 }
778
779 void TemplateTable::lstore() {
780 transition(ltos, vtos);
781 const Register Rindex = R11_scratch1;
782 locals_index(Rindex);
783 __ store_local_long(R17_tos, Rindex);
784 }
785
786 void TemplateTable::fstore() {
787 transition(ftos, vtos);
788
789 const Register Rindex = R11_scratch1;
790 locals_index(Rindex);
791 __ store_local_float(F15_ftos, Rindex);
792 }
793
794 void TemplateTable::dstore() {
795 transition(dtos, vtos);
796
797 const Register Rindex = R11_scratch1;
798 locals_index(Rindex);
799 __ store_local_double(F15_ftos, Rindex);
800 }
801
802 void TemplateTable::astore() {
803 transition(vtos, vtos);
804
805 const Register Rindex = R11_scratch1;
806 __ pop_ptr();
807 __ verify_oop_or_return_address(R17_tos, Rindex);
808 locals_index(Rindex);
809 __ store_local_ptr(R17_tos, Rindex);
810 }
811
812 void TemplateTable::wide_istore() {
813 transition(vtos, vtos);
814
815 const Register Rindex = R11_scratch1;
816 __ pop_i();
817 locals_index_wide(Rindex);
818 __ store_local_int(R17_tos, Rindex);
819 }
820
821 void TemplateTable::wide_lstore() {
822 transition(vtos, vtos);
823
824 const Register Rindex = R11_scratch1;
825 __ pop_l();
826 locals_index_wide(Rindex);
827 __ store_local_long(R17_tos, Rindex);
828 }
829
830 void TemplateTable::wide_fstore() {
831 transition(vtos, vtos);
832
833 const Register Rindex = R11_scratch1;
834 __ pop_f();
835 locals_index_wide(Rindex);
836 __ store_local_float(F15_ftos, Rindex);
837 }
838
839 void TemplateTable::wide_dstore() {
840 transition(vtos, vtos);
841
842 const Register Rindex = R11_scratch1;
843 __ pop_d();
844 locals_index_wide(Rindex);
845 __ store_local_double(F15_ftos, Rindex);
846 }
847
848 void TemplateTable::wide_astore() {
849 transition(vtos, vtos);
850
851 const Register Rindex = R11_scratch1;
852 __ pop_ptr();
853 __ verify_oop_or_return_address(R17_tos, Rindex);
854 locals_index_wide(Rindex);
855 __ store_local_ptr(R17_tos, Rindex);
856 }
857
858 void TemplateTable::iastore() {
859 transition(itos, vtos);
860
861 const Register Rindex = R3_ARG1,
862 Rstore_addr = R4_ARG2,
863 Rarray = R5_ARG3,
864 Rtemp = R6_ARG4;
865 __ pop_i(Rindex);
866 __ index_check(Rarray, Rindex, LogBytesPerInt, Rtemp, Rstore_addr);
867 __ stw(R17_tos, arrayOopDesc::base_offset_in_bytes(T_INT), Rstore_addr);
868 }
869
870 void TemplateTable::lastore() {
871 transition(ltos, vtos);
872
873 const Register Rindex = R3_ARG1,
874 Rstore_addr = R4_ARG2,
875 Rarray = R5_ARG3,
876 Rtemp = R6_ARG4;
877 __ pop_i(Rindex);
878 __ index_check(Rarray, Rindex, LogBytesPerLong, Rtemp, Rstore_addr);
879 __ std(R17_tos, arrayOopDesc::base_offset_in_bytes(T_LONG), Rstore_addr);
880 }
881
882 void TemplateTable::fastore() {
883 transition(ftos, vtos);
884
885 const Register Rindex = R3_ARG1,
886 Rstore_addr = R4_ARG2,
887 Rarray = R5_ARG3,
888 Rtemp = R6_ARG4;
889 __ pop_i(Rindex);
890 __ index_check(Rarray, Rindex, LogBytesPerInt, Rtemp, Rstore_addr);
891 __ stfs(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Rstore_addr);
892 }
893
894 void TemplateTable::dastore() {
895 transition(dtos, vtos);
896
897 const Register Rindex = R3_ARG1,
898 Rstore_addr = R4_ARG2,
899 Rarray = R5_ARG3,
900 Rtemp = R6_ARG4;
901 __ pop_i(Rindex);
902 __ index_check(Rarray, Rindex, LogBytesPerLong, Rtemp, Rstore_addr);
903 __ stfd(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Rstore_addr);
904 }
905
906 // Pop 3 values from the stack and...
907 void TemplateTable::aastore() {
908 transition(vtos, vtos);
909
910 Label Lstore_ok, Lis_null, Ldone;
911 const Register Rindex = R3_ARG1,
912 Rarray = R4_ARG2,
913 Rscratch = R11_scratch1,
914 Rscratch2 = R12_scratch2,
915 Rarray_klass = R5_ARG3,
916 Rarray_element_klass = Rarray_klass,
917 Rvalue_klass = R6_ARG4,
918 Rstore_addr = R31; // Use register which survives VM call.
919
920 __ ld(R17_tos, Interpreter::expr_offset_in_bytes(0), R15_esp); // Get value to store.
921 __ lwz(Rindex, Interpreter::expr_offset_in_bytes(1), R15_esp); // Get index.
922 __ ld(Rarray, Interpreter::expr_offset_in_bytes(2), R15_esp); // Get array.
923
924 __ verify_oop(R17_tos);
925 __ index_check_without_pop(Rarray, Rindex, UseCompressedOops ? 2 : LogBytesPerWord, Rscratch, Rstore_addr);
926 // Rindex is dead!
927 Register Rscratch3 = Rindex;
928
929 // Do array store check - check for NULL value first.
930 __ cmpdi(CCR0, R17_tos, 0);
931 __ beq(CCR0, Lis_null);
932
933 __ load_klass(Rarray_klass, Rarray);
934 __ load_klass(Rvalue_klass, R17_tos);
935
936 // Do fast instanceof cache test.
937 __ ld(Rarray_element_klass, in_bytes(ObjArrayKlass::element_klass_offset()), Rarray_klass);
938
939 // Generate a fast subtype check. Branch to store_ok if no failure. Throw if failure.
940 __ gen_subtype_check(Rvalue_klass /*subklass*/, Rarray_element_klass /*superklass*/, Rscratch, Rscratch2, Rscratch3, Lstore_ok);
941
942 // Fell through: subtype check failed => throw an exception.
943 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArrayStoreException_entry);
944 __ mtctr(R11_scratch1);
945 __ bctr();
946
947 __ bind(Lis_null);
948 do_oop_store(_masm, Rstore_addr, arrayOopDesc::base_offset_in_bytes(T_OBJECT), noreg /* 0 */,
949 Rscratch, Rscratch2, Rscratch3, _bs->kind(), true /* precise */, false /* check_null */);
950 __ profile_null_seen(Rscratch, Rscratch2);
951 __ b(Ldone);
952
953 // Store is OK.
954 __ bind(Lstore_ok);
955 do_oop_store(_masm, Rstore_addr, arrayOopDesc::base_offset_in_bytes(T_OBJECT), R17_tos /* value */,
956 Rscratch, Rscratch2, Rscratch3, _bs->kind(), true /* precise */, false /* check_null */);
957
958 __ bind(Ldone);
959 // Adjust sp (pops array, index and value).
960 __ addi(R15_esp, R15_esp, 3 * Interpreter::stackElementSize);
961 }
962
963 void TemplateTable::bastore() {
964 transition(itos, vtos);
965
966 const Register Rindex = R11_scratch1,
967 Rarray = R12_scratch2,
968 Rscratch = R3_ARG1;
969 __ pop_i(Rindex);
970 // tos: val
971 // Rarray: array ptr (popped by index_check)
972 __ index_check(Rarray, Rindex, 0, Rscratch, Rarray);
973 __ stb(R17_tos, arrayOopDesc::base_offset_in_bytes(T_BYTE), Rarray);
974 }
975
976 void TemplateTable::castore() {
977 transition(itos, vtos);
978
979 const Register Rindex = R11_scratch1,
980 Rarray = R12_scratch2,
981 Rscratch = R3_ARG1;
982 __ pop_i(Rindex);
983 // tos: val
984 // Rarray: array ptr (popped by index_check)
985 __ index_check(Rarray, Rindex, LogBytesPerShort, Rscratch, Rarray);
986 __ sth(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rarray);
987 }
988
989 void TemplateTable::sastore() {
990 castore();
991 }
992
993 void TemplateTable::istore(int n) {
994 transition(itos, vtos);
995 __ stw(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
996 }
997
998 void TemplateTable::lstore(int n) {
999 transition(ltos, vtos);
1000 __ std(R17_tos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
1001 }
1002
1003 void TemplateTable::fstore(int n) {
1004 transition(ftos, vtos);
1005 __ stfs(F15_ftos, Interpreter::local_offset_in_bytes(n), R18_locals);
1006 }
1007
1008 void TemplateTable::dstore(int n) {
1009 transition(dtos, vtos);
1010 __ stfd(F15_ftos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
1011 }
1012
1013 void TemplateTable::astore(int n) {
1014 transition(vtos, vtos);
1015
1016 __ pop_ptr();
1017 __ verify_oop_or_return_address(R17_tos, R11_scratch1);
1018 __ std(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
1019 }
1020
1021 void TemplateTable::pop() {
1022 transition(vtos, vtos);
1023
1024 __ addi(R15_esp, R15_esp, Interpreter::stackElementSize);
1025 }
1026
1027 void TemplateTable::pop2() {
1028 transition(vtos, vtos);
1029
1030 __ addi(R15_esp, R15_esp, Interpreter::stackElementSize * 2);
1031 }
1032
1033 void TemplateTable::dup() {
1034 transition(vtos, vtos);
1035
1036 __ ld(R11_scratch1, Interpreter::stackElementSize, R15_esp);
1037 __ push_ptr(R11_scratch1);
1038 }
1039
1040 void TemplateTable::dup_x1() {
1041 transition(vtos, vtos);
1042
1043 Register Ra = R11_scratch1,
1044 Rb = R12_scratch2;
1045 // stack: ..., a, b
1046 __ ld(Rb, Interpreter::stackElementSize, R15_esp);
1047 __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1048 __ std(Rb, Interpreter::stackElementSize * 2, R15_esp);
1049 __ std(Ra, Interpreter::stackElementSize, R15_esp);
1050 __ push_ptr(Rb);
1051 // stack: ..., b, a, b
1052 }
1053
1054 void TemplateTable::dup_x2() {
1055 transition(vtos, vtos);
1056
1057 Register Ra = R11_scratch1,
1058 Rb = R12_scratch2,
1059 Rc = R3_ARG1;
1060
1061 // stack: ..., a, b, c
1062 __ ld(Rc, Interpreter::stackElementSize, R15_esp); // load c
1063 __ ld(Ra, Interpreter::stackElementSize * 3, R15_esp); // load a
1064 __ std(Rc, Interpreter::stackElementSize * 3, R15_esp); // store c in a
1065 __ ld(Rb, Interpreter::stackElementSize * 2, R15_esp); // load b
1066 // stack: ..., c, b, c
1067 __ std(Ra, Interpreter::stackElementSize * 2, R15_esp); // store a in b
1068 // stack: ..., c, a, c
1069 __ std(Rb, Interpreter::stackElementSize, R15_esp); // store b in c
1070 __ push_ptr(Rc); // push c
1071 // stack: ..., c, a, b, c
1072 }
1073
1074 void TemplateTable::dup2() {
1075 transition(vtos, vtos);
1076
1077 Register Ra = R11_scratch1,
1078 Rb = R12_scratch2;
1079 // stack: ..., a, b
1080 __ ld(Rb, Interpreter::stackElementSize, R15_esp);
1081 __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1082 __ push_2ptrs(Ra, Rb);
1083 // stack: ..., a, b, a, b
1084 }
1085
1086 void TemplateTable::dup2_x1() {
1087 transition(vtos, vtos);
1088
1089 Register Ra = R11_scratch1,
1090 Rb = R12_scratch2,
1091 Rc = R3_ARG1;
1092 // stack: ..., a, b, c
1093 __ ld(Rc, Interpreter::stackElementSize, R15_esp);
1094 __ ld(Rb, Interpreter::stackElementSize * 2, R15_esp);
1095 __ std(Rc, Interpreter::stackElementSize * 2, R15_esp);
1096 __ ld(Ra, Interpreter::stackElementSize * 3, R15_esp);
1097 __ std(Ra, Interpreter::stackElementSize, R15_esp);
1098 __ std(Rb, Interpreter::stackElementSize * 3, R15_esp);
1099 // stack: ..., b, c, a
1100 __ push_2ptrs(Rb, Rc);
1101 // stack: ..., b, c, a, b, c
1102 }
1103
1104 void TemplateTable::dup2_x2() {
1105 transition(vtos, vtos);
1106
1107 Register Ra = R11_scratch1,
1108 Rb = R12_scratch2,
1109 Rc = R3_ARG1,
1110 Rd = R4_ARG2;
1111 // stack: ..., a, b, c, d
1112 __ ld(Rb, Interpreter::stackElementSize * 3, R15_esp);
1113 __ ld(Rd, Interpreter::stackElementSize, R15_esp);
1114 __ std(Rb, Interpreter::stackElementSize, R15_esp); // store b in d
1115 __ std(Rd, Interpreter::stackElementSize * 3, R15_esp); // store d in b
1116 __ ld(Ra, Interpreter::stackElementSize * 4, R15_esp);
1117 __ ld(Rc, Interpreter::stackElementSize * 2, R15_esp);
1118 __ std(Ra, Interpreter::stackElementSize * 2, R15_esp); // store a in c
1119 __ std(Rc, Interpreter::stackElementSize * 4, R15_esp); // store c in a
1120 // stack: ..., c, d, a, b
1121 __ push_2ptrs(Rc, Rd);
1122 // stack: ..., c, d, a, b, c, d
1123 }
1124
1125 void TemplateTable::swap() {
1126 transition(vtos, vtos);
1127 // stack: ..., a, b
1128
1129 Register Ra = R11_scratch1,
1130 Rb = R12_scratch2;
1131 // stack: ..., a, b
1132 __ ld(Rb, Interpreter::stackElementSize, R15_esp);
1133 __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1134 __ std(Rb, Interpreter::stackElementSize * 2, R15_esp);
1135 __ std(Ra, Interpreter::stackElementSize, R15_esp);
1136 // stack: ..., b, a
1137 }
1138
1139 void TemplateTable::iop2(Operation op) {
1140 transition(itos, itos);
1141
1142 Register Rscratch = R11_scratch1;
1143
1144 __ pop_i(Rscratch);
1145 // tos = number of bits to shift
1146 // Rscratch = value to shift
1147 switch (op) {
1148 case add: __ add(R17_tos, Rscratch, R17_tos); break;
1149 case sub: __ sub(R17_tos, Rscratch, R17_tos); break;
1150 case mul: __ mullw(R17_tos, Rscratch, R17_tos); break;
1151 case _and: __ andr(R17_tos, Rscratch, R17_tos); break;
1152 case _or: __ orr(R17_tos, Rscratch, R17_tos); break;
1153 case _xor: __ xorr(R17_tos, Rscratch, R17_tos); break;
1154 case shl: __ rldicl(R17_tos, R17_tos, 0, 64-5); __ slw(R17_tos, Rscratch, R17_tos); break;
1155 case shr: __ rldicl(R17_tos, R17_tos, 0, 64-5); __ sraw(R17_tos, Rscratch, R17_tos); break;
1156 case ushr: __ rldicl(R17_tos, R17_tos, 0, 64-5); __ srw(R17_tos, Rscratch, R17_tos); break;
1157 default: ShouldNotReachHere();
1158 }
1159 }
1160
1161 void TemplateTable::lop2(Operation op) {
1162 transition(ltos, ltos);
1163
1164 Register Rscratch = R11_scratch1;
1165 __ pop_l(Rscratch);
1166 switch (op) {
1167 case add: __ add(R17_tos, Rscratch, R17_tos); break;
1168 case sub: __ sub(R17_tos, Rscratch, R17_tos); break;
1169 case _and: __ andr(R17_tos, Rscratch, R17_tos); break;
1170 case _or: __ orr(R17_tos, Rscratch, R17_tos); break;
1171 case _xor: __ xorr(R17_tos, Rscratch, R17_tos); break;
1172 default: ShouldNotReachHere();
1173 }
1174 }
1175
1176 void TemplateTable::idiv() {
1177 transition(itos, itos);
1178
1179 Label Lnormal, Lexception, Ldone;
1180 Register Rdividend = R11_scratch1; // Used by irem.
1181
1182 __ addi(R0, R17_tos, 1);
1183 __ cmplwi(CCR0, R0, 2);
1184 __ bgt(CCR0, Lnormal); // divisor <-1 or >1
1185
1186 __ cmpwi(CCR1, R17_tos, 0);
1187 __ beq(CCR1, Lexception); // divisor == 0
1188
1189 __ pop_i(Rdividend);
1190 __ mullw(R17_tos, Rdividend, R17_tos); // div by +/-1
1191 __ b(Ldone);
1192
1193 __ bind(Lexception);
1194 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArithmeticException_entry);
1195 __ mtctr(R11_scratch1);
1196 __ bctr();
1197
1198 __ align(32, 12);
1199 __ bind(Lnormal);
1200 __ pop_i(Rdividend);
1201 __ divw(R17_tos, Rdividend, R17_tos); // Can't divide minint/-1.
1202 __ bind(Ldone);
1203 }
1204
1205 void TemplateTable::irem() {
1206 transition(itos, itos);
1207
1208 __ mr(R12_scratch2, R17_tos);
1209 idiv();
1210 __ mullw(R17_tos, R17_tos, R12_scratch2);
1211 __ subf(R17_tos, R17_tos, R11_scratch1); // Dividend set by idiv.
1212 }
1213
1214 void TemplateTable::lmul() {
1215 transition(ltos, ltos);
1216
1217 __ pop_l(R11_scratch1);
1218 __ mulld(R17_tos, R11_scratch1, R17_tos);
1219 }
1220
1221 void TemplateTable::ldiv() {
1222 transition(ltos, ltos);
1223
1224 Label Lnormal, Lexception, Ldone;
1225 Register Rdividend = R11_scratch1; // Used by lrem.
1226
1227 __ addi(R0, R17_tos, 1);
1228 __ cmpldi(CCR0, R0, 2);
1229 __ bgt(CCR0, Lnormal); // divisor <-1 or >1
1230
1231 __ cmpdi(CCR1, R17_tos, 0);
1232 __ beq(CCR1, Lexception); // divisor == 0
1233
1234 __ pop_l(Rdividend);
1235 __ mulld(R17_tos, Rdividend, R17_tos); // div by +/-1
1236 __ b(Ldone);
1237
1238 __ bind(Lexception);
1239 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArithmeticException_entry);
1240 __ mtctr(R11_scratch1);
1241 __ bctr();
1242
1243 __ align(32, 12);
1244 __ bind(Lnormal);
1245 __ pop_l(Rdividend);
1246 __ divd(R17_tos, Rdividend, R17_tos); // Can't divide minint/-1.
1247 __ bind(Ldone);
1248 }
1249
1250 void TemplateTable::lrem() {
1251 transition(ltos, ltos);
1252
1253 __ mr(R12_scratch2, R17_tos);
1254 ldiv();
1255 __ mulld(R17_tos, R17_tos, R12_scratch2);
1256 __ subf(R17_tos, R17_tos, R11_scratch1); // Dividend set by ldiv.
1257 }
1258
1259 void TemplateTable::lshl() {
1260 transition(itos, ltos);
1261
1262 __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1263 __ pop_l(R11_scratch1);
1264 __ sld(R17_tos, R11_scratch1, R17_tos);
1265 }
1266
1267 void TemplateTable::lshr() {
1268 transition(itos, ltos);
1269
1270 __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1271 __ pop_l(R11_scratch1);
1272 __ srad(R17_tos, R11_scratch1, R17_tos);
1273 }
1274
1275 void TemplateTable::lushr() {
1276 transition(itos, ltos);
1277
1278 __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1279 __ pop_l(R11_scratch1);
1280 __ srd(R17_tos, R11_scratch1, R17_tos);
1281 }
1282
1283 void TemplateTable::fop2(Operation op) {
1284 transition(ftos, ftos);
1285
1286 switch (op) {
1287 case add: __ pop_f(F0_SCRATCH); __ fadds(F15_ftos, F0_SCRATCH, F15_ftos); break;
1288 case sub: __ pop_f(F0_SCRATCH); __ fsubs(F15_ftos, F0_SCRATCH, F15_ftos); break;
1289 case mul: __ pop_f(F0_SCRATCH); __ fmuls(F15_ftos, F0_SCRATCH, F15_ftos); break;
1290 case div: __ pop_f(F0_SCRATCH); __ fdivs(F15_ftos, F0_SCRATCH, F15_ftos); break;
1291 case rem:
1292 __ pop_f(F1_ARG1);
1293 __ fmr(F2_ARG2, F15_ftos);
1294 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem));
1295 __ fmr(F15_ftos, F1_RET);
1296 break;
1297
1298 default: ShouldNotReachHere();
1299 }
1300 }
1301
1302 void TemplateTable::dop2(Operation op) {
1303 transition(dtos, dtos);
1304
1305 switch (op) {
1306 case add: __ pop_d(F0_SCRATCH); __ fadd(F15_ftos, F0_SCRATCH, F15_ftos); break;
1307 case sub: __ pop_d(F0_SCRATCH); __ fsub(F15_ftos, F0_SCRATCH, F15_ftos); break;
1308 case mul: __ pop_d(F0_SCRATCH); __ fmul(F15_ftos, F0_SCRATCH, F15_ftos); break;
1309 case div: __ pop_d(F0_SCRATCH); __ fdiv(F15_ftos, F0_SCRATCH, F15_ftos); break;
1310 case rem:
1311 __ pop_d(F1_ARG1);
1312 __ fmr(F2_ARG2, F15_ftos);
1313 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem));
1314 __ fmr(F15_ftos, F1_RET);
1315 break;
1316
1317 default: ShouldNotReachHere();
1318 }
1319 }
1320
1321 // Negate the value in the TOS cache.
1322 void TemplateTable::ineg() {
1323 transition(itos, itos);
1324
1325 __ neg(R17_tos, R17_tos);
1326 }
1327
1328 // Negate the value in the TOS cache.
1329 void TemplateTable::lneg() {
1330 transition(ltos, ltos);
1331
1332 __ neg(R17_tos, R17_tos);
1333 }
1334
1335 void TemplateTable::fneg() {
1336 transition(ftos, ftos);
1337
1338 __ fneg(F15_ftos, F15_ftos);
1339 }
1340
1341 void TemplateTable::dneg() {
1342 transition(dtos, dtos);
1343
1344 __ fneg(F15_ftos, F15_ftos);
1345 }
1346
1347 // Increments a local variable in place.
1348 void TemplateTable::iinc() {
1349 transition(vtos, vtos);
1350
1351 const Register Rindex = R11_scratch1,
1352 Rincrement = R0,
1353 Rvalue = R12_scratch2;
1354
1355 locals_index(Rindex); // Load locals index from bytecode stream.
1356 __ lbz(Rincrement, 2, R14_bcp); // Load increment from the bytecode stream.
1357 __ extsb(Rincrement, Rincrement);
1358
1359 __ load_local_int(Rvalue, Rindex, Rindex); // Puts address of local into Rindex.
1360
1361 __ add(Rvalue, Rincrement, Rvalue);
1362 __ stw(Rvalue, 0, Rindex);
1363 }
1364
1365 void TemplateTable::wide_iinc() {
1366 transition(vtos, vtos);
1367
1368 Register Rindex = R11_scratch1,
1369 Rlocals_addr = Rindex,
1370 Rincr = R12_scratch2;
1371 locals_index_wide(Rindex);
1372 __ get_2_byte_integer_at_bcp(4, Rincr, InterpreterMacroAssembler::Signed);
1373 __ load_local_int(R17_tos, Rlocals_addr, Rindex);
1374 __ add(R17_tos, Rincr, R17_tos);
1375 __ stw(R17_tos, 0, Rlocals_addr);
1376 }
1377
1378 void TemplateTable::convert() {
1379 // %%%%% Factor this first part accross platforms
1380 #ifdef ASSERT
1381 TosState tos_in = ilgl;
1382 TosState tos_out = ilgl;
1383 switch (bytecode()) {
1384 case Bytecodes::_i2l: // fall through
1385 case Bytecodes::_i2f: // fall through
1386 case Bytecodes::_i2d: // fall through
1387 case Bytecodes::_i2b: // fall through
1388 case Bytecodes::_i2c: // fall through
1389 case Bytecodes::_i2s: tos_in = itos; break;
1390 case Bytecodes::_l2i: // fall through
1391 case Bytecodes::_l2f: // fall through
1392 case Bytecodes::_l2d: tos_in = ltos; break;
1393 case Bytecodes::_f2i: // fall through
1394 case Bytecodes::_f2l: // fall through
1395 case Bytecodes::_f2d: tos_in = ftos; break;
1396 case Bytecodes::_d2i: // fall through
1397 case Bytecodes::_d2l: // fall through
1398 case Bytecodes::_d2f: tos_in = dtos; break;
1399 default : ShouldNotReachHere();
1400 }
1401 switch (bytecode()) {
1402 case Bytecodes::_l2i: // fall through
1403 case Bytecodes::_f2i: // fall through
1404 case Bytecodes::_d2i: // fall through
1405 case Bytecodes::_i2b: // fall through
1406 case Bytecodes::_i2c: // fall through
1407 case Bytecodes::_i2s: tos_out = itos; break;
1408 case Bytecodes::_i2l: // fall through
1409 case Bytecodes::_f2l: // fall through
1410 case Bytecodes::_d2l: tos_out = ltos; break;
1411 case Bytecodes::_i2f: // fall through
1412 case Bytecodes::_l2f: // fall through
1413 case Bytecodes::_d2f: tos_out = ftos; break;
1414 case Bytecodes::_i2d: // fall through
1415 case Bytecodes::_l2d: // fall through
1416 case Bytecodes::_f2d: tos_out = dtos; break;
1417 default : ShouldNotReachHere();
1418 }
1419 transition(tos_in, tos_out);
1420 #endif
1421
1422 // Conversion
1423 Label done;
1424 switch (bytecode()) {
1425 case Bytecodes::_i2l:
1426 __ extsw(R17_tos, R17_tos);
1427 break;
1428
1429 case Bytecodes::_l2i:
1430 // Nothing to do, we'll continue to work with the lower bits.
1431 break;
1432
1433 case Bytecodes::_i2b:
1434 __ extsb(R17_tos, R17_tos);
1435 break;
1436
1437 case Bytecodes::_i2c:
1438 __ rldicl(R17_tos, R17_tos, 0, 64-2*8);
1439 break;
1440
1441 case Bytecodes::_i2s:
1442 __ extsh(R17_tos, R17_tos);
1443 break;
1444
1445 case Bytecodes::_i2d:
1446 __ extsw(R17_tos, R17_tos);
1447 case Bytecodes::_l2d:
1448 __ push_l_pop_d();
1449 __ fcfid(F15_ftos, F15_ftos);
1450 break;
1451
1452 case Bytecodes::_i2f:
1453 __ extsw(R17_tos, R17_tos);
1454 __ push_l_pop_d();
1455 if (VM_Version::has_fcfids()) { // fcfids is >= Power7 only
1456 // Comment: alternatively, load with sign extend could be done by lfiwax.
1457 __ fcfids(F15_ftos, F15_ftos);
1458 } else {
1459 __ fcfid(F15_ftos, F15_ftos);
1460 __ frsp(F15_ftos, F15_ftos);
1461 }
1462 break;
1463
1464 case Bytecodes::_l2f:
1465 if (VM_Version::has_fcfids()) { // fcfids is >= Power7 only
1466 __ push_l_pop_d();
1467 __ fcfids(F15_ftos, F15_ftos);
1468 } else {
1469 // Avoid rounding problem when result should be 0x3f800001: need fixup code before fcfid+frsp.
1470 __ mr(R3_ARG1, R17_tos);
1471 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::l2f));
1472 __ fmr(F15_ftos, F1_RET);
1473 }
1474 break;
1475
1476 case Bytecodes::_f2d:
1477 // empty
1478 break;
1479
1480 case Bytecodes::_d2f:
1481 __ frsp(F15_ftos, F15_ftos);
1482 break;
1483
1484 case Bytecodes::_d2i:
1485 case Bytecodes::_f2i:
1486 __ fcmpu(CCR0, F15_ftos, F15_ftos);
1487 __ li(R17_tos, 0); // 0 in case of NAN
1488 __ bso(CCR0, done);
1489 __ fctiwz(F15_ftos, F15_ftos);
1490 __ push_d_pop_l();
1491 break;
1492
1493 case Bytecodes::_d2l:
1494 case Bytecodes::_f2l:
1495 __ fcmpu(CCR0, F15_ftos, F15_ftos);
1496 __ li(R17_tos, 0); // 0 in case of NAN
1497 __ bso(CCR0, done);
1498 __ fctidz(F15_ftos, F15_ftos);
1499 __ push_d_pop_l();
1500 break;
1501
1502 default: ShouldNotReachHere();
1503 }
1504 __ bind(done);
1505 }
1506
1507 // Long compare
1508 void TemplateTable::lcmp() {
1509 transition(ltos, itos);
1510
1511 const Register Rscratch = R11_scratch1;
1512 __ pop_l(Rscratch); // first operand, deeper in stack
1513
1514 __ cmpd(CCR0, Rscratch, R17_tos); // compare
1515 __ mfcr(R17_tos); // set bit 32..33 as follows: <: 0b10, =: 0b00, >: 0b01
1516 __ srwi(Rscratch, R17_tos, 30);
1517 __ srawi(R17_tos, R17_tos, 31);
1518 __ orr(R17_tos, Rscratch, R17_tos); // set result as follows: <: -1, =: 0, >: 1
1519 }
1520
1521 // fcmpl/fcmpg and dcmpl/dcmpg bytecodes
1522 // unordered_result == -1 => fcmpl or dcmpl
1523 // unordered_result == 1 => fcmpg or dcmpg
1524 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1525 const FloatRegister Rfirst = F0_SCRATCH,
1526 Rsecond = F15_ftos;
1527 const Register Rscratch = R11_scratch1;
1528
1529 if (is_float) {
1530 __ pop_f(Rfirst);
1531 } else {
1532 __ pop_d(Rfirst);
1533 }
1534
1535 Label Lunordered, Ldone;
1536 __ fcmpu(CCR0, Rfirst, Rsecond); // compare
1537 if (unordered_result) {
1538 __ bso(CCR0, Lunordered);
1539 }
1540 __ mfcr(R17_tos); // set bit 32..33 as follows: <: 0b10, =: 0b00, >: 0b01
1541 __ srwi(Rscratch, R17_tos, 30);
1542 __ srawi(R17_tos, R17_tos, 31);
1543 __ orr(R17_tos, Rscratch, R17_tos); // set result as follows: <: -1, =: 0, >: 1
1544 if (unordered_result) {
1545 __ b(Ldone);
1546 __ bind(Lunordered);
1547 __ load_const_optimized(R17_tos, unordered_result);
1548 }
1549 __ bind(Ldone);
1550 }
1551
1552 // Branch_conditional which takes TemplateTable::Condition.
1553 void TemplateTable::branch_conditional(ConditionRegister crx, TemplateTable::Condition cc, Label& L, bool invert) {
1554 bool positive = false;
1555 Assembler::Condition cond = Assembler::equal;
1556 switch (cc) {
1557 case TemplateTable::equal: positive = true ; cond = Assembler::equal ; break;
1558 case TemplateTable::not_equal: positive = false; cond = Assembler::equal ; break;
1559 case TemplateTable::less: positive = true ; cond = Assembler::less ; break;
1560 case TemplateTable::less_equal: positive = false; cond = Assembler::greater; break;
1561 case TemplateTable::greater: positive = true ; cond = Assembler::greater; break;
1562 case TemplateTable::greater_equal: positive = false; cond = Assembler::less ; break;
1563 default: ShouldNotReachHere();
1564 }
1565 int bo = (positive != invert) ? Assembler::bcondCRbiIs1 : Assembler::bcondCRbiIs0;
1566 int bi = Assembler::bi0(crx, cond);
1567 __ bc(bo, bi, L);
1568 }
1569
1570 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1571
1572 // Note: on SPARC, we use InterpreterMacroAssembler::if_cmp also.
1573 __ verify_thread();
1574
1575 const Register Rscratch1 = R11_scratch1,
1576 Rscratch2 = R12_scratch2,
1577 Rscratch3 = R3_ARG1,
1578 R4_counters = R4_ARG2,
1579 bumped_count = R31,
1580 Rdisp = R22_tmp2;
1581
1582 __ profile_taken_branch(Rscratch1, bumped_count);
1583
1584 // Get (wide) offset.
1585 if (is_wide) {
1586 __ get_4_byte_integer_at_bcp(1, Rdisp, InterpreterMacroAssembler::Signed);
1587 } else {
1588 __ get_2_byte_integer_at_bcp(1, Rdisp, InterpreterMacroAssembler::Signed);
1589 }
1590
1591 // --------------------------------------------------------------------------
1592 // Handle all the JSR stuff here, then exit.
1593 // It's much shorter and cleaner than intermingling with the
1594 // non-JSR normal-branch stuff occurring below.
1595 if (is_jsr) {
1596 // Compute return address as bci in Otos_i.
1597 __ ld(Rscratch1, in_bytes(Method::const_offset()), R19_method);
1598 __ addi(Rscratch2, R14_bcp, -in_bytes(ConstMethod::codes_offset()) + (is_wide ? 5 : 3));
1599 __ subf(R17_tos, Rscratch1, Rscratch2);
1600
1601 // Bump bcp to target of JSR.
1602 __ add(R14_bcp, Rdisp, R14_bcp);
1603 // Push returnAddress for "ret" on stack.
1604 __ push_ptr(R17_tos);
1605 // And away we go!
1606 __ dispatch_next(vtos);
1607 return;
1608 }
1609
1610 // --------------------------------------------------------------------------
1611 // Normal (non-jsr) branch handling
1612
1613 const bool increment_invocation_counter_for_backward_branches = UseCompiler && UseLoopCounter;
1614 if (increment_invocation_counter_for_backward_branches) {
1615 //__ unimplemented("branch invocation counter");
1616
1617 Label Lforward;
1618 __ add(R14_bcp, Rdisp, R14_bcp); // Add to bc addr.
1619
1620 // Check branch direction.
1621 __ cmpdi(CCR0, Rdisp, 0);
1622 __ bgt(CCR0, Lforward);
1623
1624 __ get_method_counters(R19_method, R4_counters, Lforward);
1625
1626 if (TieredCompilation) {
1627 Label Lno_mdo, Loverflow;
1628 const int increment = InvocationCounter::count_increment;
1629 const int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
1630 if (ProfileInterpreter) {
1631 Register Rmdo = Rscratch1;
1632
1633 // If no method data exists, go to profile_continue.
1634 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
1635 __ cmpdi(CCR0, Rmdo, 0);
1636 __ beq(CCR0, Lno_mdo);
1637
1638 // Increment backedge counter in the MDO.
1639 const int mdo_bc_offs = in_bytes(MethodData::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
1640 __ lwz(Rscratch2, mdo_bc_offs, Rmdo);
1641 __ load_const_optimized(Rscratch3, mask, R0);
1642 __ addi(Rscratch2, Rscratch2, increment);
1643 __ stw(Rscratch2, mdo_bc_offs, Rmdo);
1644 __ and_(Rscratch3, Rscratch2, Rscratch3);
1645 __ bne(CCR0, Lforward);
1646 __ b(Loverflow);
1647 }
1648
1649 // If there's no MDO, increment counter in method.
1650 const int mo_bc_offs = in_bytes(MethodCounters::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
1651 __ bind(Lno_mdo);
1652 __ lwz(Rscratch2, mo_bc_offs, R4_counters);
1653 __ load_const_optimized(Rscratch3, mask, R0);
1654 __ addi(Rscratch2, Rscratch2, increment);
1655 __ stw(Rscratch2, mo_bc_offs, R19_method);
1656 __ and_(Rscratch3, Rscratch2, Rscratch3);
1657 __ bne(CCR0, Lforward);
1658
1659 __ bind(Loverflow);
1660
1661 // Notify point for loop, pass branch bytecode.
1662 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R14_bcp, true);
1663
1664 // Was an OSR adapter generated?
1665 // O0 = osr nmethod
1666 __ cmpdi(CCR0, R3_RET, 0);
1667 __ beq(CCR0, Lforward);
1668
1669 // Has the nmethod been invalidated already?
1670 __ lbz(R0, nmethod::state_offset(), R3_RET);
1671 __ cmpwi(CCR0, R0, nmethod::in_use);
1672 __ bne(CCR0, Lforward);
1673
1674 // Migrate the interpreter frame off of the stack.
1675 // We can use all registers because we will not return to interpreter from this point.
1676
1677 // Save nmethod.
1678 const Register osr_nmethod = R31;
1679 __ mr(osr_nmethod, R3_RET);
1680 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R11_scratch1);
1681 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), R16_thread);
1682 __ reset_last_Java_frame();
1683 // OSR buffer is in ARG1.
1684
1685 // Remove the interpreter frame.
1686 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1687
1688 // Jump to the osr code.
1689 __ ld(R11_scratch1, nmethod::osr_entry_point_offset(), osr_nmethod);
1690 __ mtlr(R0);
1691 __ mtctr(R11_scratch1);
1692 __ bctr();
1693
1694 } else {
1695
1696 const Register invoke_ctr = Rscratch1;
1697 // Update Backedge branch separately from invocations.
1698 __ increment_backedge_counter(R4_counters, invoke_ctr, Rscratch2, Rscratch3);
1699
1700 if (ProfileInterpreter) {
1701 __ test_invocation_counter_for_mdp(invoke_ctr, Rscratch2, Lforward);
1702 if (UseOnStackReplacement) {
1703 __ test_backedge_count_for_osr(bumped_count, R14_bcp, Rscratch2);
1704 }
1705 } else {
1706 if (UseOnStackReplacement) {
1707 __ test_backedge_count_for_osr(invoke_ctr, R14_bcp, Rscratch2);
1708 }
1709 }
1710 }
1711
1712 __ bind(Lforward);
1713
1714 } else {
1715 // Bump bytecode pointer by displacement (take the branch).
1716 __ add(R14_bcp, Rdisp, R14_bcp); // Add to bc addr.
1717 }
1718 // Continue with bytecode @ target.
1719 // %%%%% Like Intel, could speed things up by moving bytecode fetch to code above,
1720 // %%%%% and changing dispatch_next to dispatch_only.
1721 __ dispatch_next(vtos);
1722 }
1723
1724 // Helper function for if_cmp* methods below.
1725 // Factored out common compare and branch code.
1726 void TemplateTable::if_cmp_common(Register Rfirst, Register Rsecond, Register Rscratch1, Register Rscratch2, Condition cc, bool is_jint, bool cmp0) {
1727 Label Lnot_taken;
1728 // Note: The condition code we get is the condition under which we
1729 // *fall through*! So we have to inverse the CC here.
1730
1731 if (is_jint) {
1732 if (cmp0) {
1733 __ cmpwi(CCR0, Rfirst, 0);
1734 } else {
1735 __ cmpw(CCR0, Rfirst, Rsecond);
1736 }
1737 } else {
1738 if (cmp0) {
1739 __ cmpdi(CCR0, Rfirst, 0);
1740 } else {
1741 __ cmpd(CCR0, Rfirst, Rsecond);
1742 }
1743 }
1744 branch_conditional(CCR0, cc, Lnot_taken, /*invert*/ true);
1745
1746 // Conition is false => Jump!
1747 branch(false, false);
1748
1749 // Condition is not true => Continue.
1750 __ align(32, 12);
1751 __ bind(Lnot_taken);
1752 __ profile_not_taken_branch(Rscratch1, Rscratch2);
1753 }
1754
1755 // Compare integer values with zero and fall through if CC holds, branch away otherwise.
1756 void TemplateTable::if_0cmp(Condition cc) {
1757 transition(itos, vtos);
1758
1759 if_cmp_common(R17_tos, noreg, R11_scratch1, R12_scratch2, cc, true, true);
1760 }
1761
1762 // Compare integer values and fall through if CC holds, branch away otherwise.
1763 //
1764 // Interface:
1765 // - Rfirst: First operand (older stack value)
1766 // - tos: Second operand (younger stack value)
1767 void TemplateTable::if_icmp(Condition cc) {
1768 transition(itos, vtos);
1769
1770 const Register Rfirst = R0,
1771 Rsecond = R17_tos;
1772
1773 __ pop_i(Rfirst);
1774 if_cmp_common(Rfirst, Rsecond, R11_scratch1, R12_scratch2, cc, true, false);
1775 }
1776
1777 void TemplateTable::if_nullcmp(Condition cc) {
1778 transition(atos, vtos);
1779
1780 if_cmp_common(R17_tos, noreg, R11_scratch1, R12_scratch2, cc, false, true);
1781 }
1782
1783 void TemplateTable::if_acmp(Condition cc) {
1784 transition(atos, vtos);
1785
1786 const Register Rfirst = R0,
1787 Rsecond = R17_tos;
1788
1789 __ pop_ptr(Rfirst);
1790 if_cmp_common(Rfirst, Rsecond, R11_scratch1, R12_scratch2, cc, false, false);
1791 }
1792
1793 void TemplateTable::ret() {
1794 locals_index(R11_scratch1);
1795 __ load_local_ptr(R17_tos, R11_scratch1, R11_scratch1);
1796
1797 __ profile_ret(vtos, R17_tos, R11_scratch1, R12_scratch2);
1798
1799 __ ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method);
1800 __ add(R11_scratch1, R17_tos, R11_scratch1);
1801 __ addi(R14_bcp, R11_scratch1, in_bytes(ConstMethod::codes_offset()));
1802 __ dispatch_next(vtos);
1803 }
1804
1805 void TemplateTable::wide_ret() {
1806 transition(vtos, vtos);
1807
1808 const Register Rindex = R3_ARG1,
1809 Rscratch1 = R11_scratch1,
1810 Rscratch2 = R12_scratch2;
1811
1812 locals_index_wide(Rindex);
1813 __ load_local_ptr(R17_tos, R17_tos, Rindex);
1814 __ profile_ret(vtos, R17_tos, Rscratch1, R12_scratch2);
1815 // Tos now contains the bci, compute the bcp from that.
1816 __ ld(Rscratch1, in_bytes(Method::const_offset()), R19_method);
1817 __ addi(Rscratch2, R17_tos, in_bytes(ConstMethod::codes_offset()));
1818 __ add(R14_bcp, Rscratch1, Rscratch2);
1819 __ dispatch_next(vtos);
1820 }
1821
1822 void TemplateTable::tableswitch() {
1823 transition(itos, vtos);
1824
1825 Label Ldispatch, Ldefault_case;
1826 Register Rlow_byte = R3_ARG1,
1827 Rindex = Rlow_byte,
1828 Rhigh_byte = R4_ARG2,
1829 Rdef_offset_addr = R5_ARG3, // is going to contain address of default offset
1830 Rscratch1 = R11_scratch1,
1831 Rscratch2 = R12_scratch2,
1832 Roffset = R6_ARG4;
1833
1834 // Align bcp.
1835 __ addi(Rdef_offset_addr, R14_bcp, BytesPerInt);
1836 __ clrrdi(Rdef_offset_addr, Rdef_offset_addr, log2_long((jlong)BytesPerInt));
1837
1838 // Load lo & hi.
1839 __ get_u4(Rlow_byte, Rdef_offset_addr, BytesPerInt, InterpreterMacroAssembler::Unsigned);
1840 __ get_u4(Rhigh_byte, Rdef_offset_addr, 2 *BytesPerInt, InterpreterMacroAssembler::Unsigned);
1841
1842 // Check for default case (=index outside [low,high]).
1843 __ cmpw(CCR0, R17_tos, Rlow_byte);
1844 __ cmpw(CCR1, R17_tos, Rhigh_byte);
1845 __ blt(CCR0, Ldefault_case);
1846 __ bgt(CCR1, Ldefault_case);
1847
1848 // Lookup dispatch offset.
1849 __ sub(Rindex, R17_tos, Rlow_byte);
1850 __ extsw(Rindex, Rindex);
1851 __ profile_switch_case(Rindex, Rhigh_byte /* scratch */, Rscratch1, Rscratch2);
1852 __ sldi(Rindex, Rindex, LogBytesPerInt);
1853 __ addi(Rindex, Rindex, 3 * BytesPerInt);
1854 #if defined(VM_LITTLE_ENDIAN)
1855 __ lwbrx(Roffset, Rdef_offset_addr, Rindex);
1856 __ extsw(Roffset, Roffset);
1857 #else
1858 __ lwax(Roffset, Rdef_offset_addr, Rindex);
1859 #endif
1860 __ b(Ldispatch);
1861
1862 __ bind(Ldefault_case);
1863 __ profile_switch_default(Rhigh_byte, Rscratch1);
1864 __ get_u4(Roffset, Rdef_offset_addr, 0, InterpreterMacroAssembler::Signed);
1865
1866 __ bind(Ldispatch);
1867
1868 __ add(R14_bcp, Roffset, R14_bcp);
1869 __ dispatch_next(vtos);
1870 }
1871
1872 void TemplateTable::lookupswitch() {
1873 transition(itos, itos);
1874 __ stop("lookupswitch bytecode should have been rewritten");
1875 }
1876
1877 // Table switch using linear search through cases.
1878 // Bytecode stream format:
1879 // Bytecode (1) | 4-byte padding | default offset (4) | count (4) | value/offset pair1 (8) | value/offset pair2 (8) | ...
1880 // Note: Everything is big-endian format here.
1881 void TemplateTable::fast_linearswitch() {
1882 transition(itos, vtos);
1883
1884 Label Lloop_entry, Lsearch_loop, Lcontinue_execution, Ldefault_case;
1885 Register Rcount = R3_ARG1,
1886 Rcurrent_pair = R4_ARG2,
1887 Rdef_offset_addr = R5_ARG3, // Is going to contain address of default offset.
1888 Roffset = R31, // Might need to survive C call.
1889 Rvalue = R12_scratch2,
1890 Rscratch = R11_scratch1,
1891 Rcmp_value = R17_tos;
1892
1893 // Align bcp.
1894 __ addi(Rdef_offset_addr, R14_bcp, BytesPerInt);
1895 __ clrrdi(Rdef_offset_addr, Rdef_offset_addr, log2_long((jlong)BytesPerInt));
1896
1897 // Setup loop counter and limit.
1898 __ get_u4(Rcount, Rdef_offset_addr, BytesPerInt, InterpreterMacroAssembler::Unsigned);
1899 __ addi(Rcurrent_pair, Rdef_offset_addr, 2 * BytesPerInt); // Rcurrent_pair now points to first pair.
1900
1901 __ mtctr(Rcount);
1902 __ cmpwi(CCR0, Rcount, 0);
1903 __ bne(CCR0, Lloop_entry);
1904
1905 // Default case
1906 __ bind(Ldefault_case);
1907 __ get_u4(Roffset, Rdef_offset_addr, 0, InterpreterMacroAssembler::Signed);
1908 if (ProfileInterpreter) {
1909 __ profile_switch_default(Rdef_offset_addr, Rcount/* scratch */);
1910 }
1911 __ b(Lcontinue_execution);
1912
1913 // Next iteration
1914 __ bind(Lsearch_loop);
1915 __ bdz(Ldefault_case);
1916 __ addi(Rcurrent_pair, Rcurrent_pair, 2 * BytesPerInt);
1917 __ bind(Lloop_entry);
1918 __ get_u4(Rvalue, Rcurrent_pair, 0, InterpreterMacroAssembler::Unsigned);
1919 __ cmpw(CCR0, Rvalue, Rcmp_value);
1920 __ bne(CCR0, Lsearch_loop);
1921
1922 // Found, load offset.
1923 __ get_u4(Roffset, Rcurrent_pair, BytesPerInt, InterpreterMacroAssembler::Signed);
1924 // Calculate case index and profile
1925 __ mfctr(Rcurrent_pair);
1926 if (ProfileInterpreter) {
1927 __ sub(Rcurrent_pair, Rcount, Rcurrent_pair);
1928 __ profile_switch_case(Rcurrent_pair, Rcount /*scratch*/, Rdef_offset_addr/*scratch*/, Rscratch);
1929 }
1930
1931 __ bind(Lcontinue_execution);
1932 __ add(R14_bcp, Roffset, R14_bcp);
1933 __ dispatch_next(vtos);
1934 }
1935
1936 // Table switch using binary search (value/offset pairs are ordered).
1937 // Bytecode stream format:
1938 // Bytecode (1) | 4-byte padding | default offset (4) | count (4) | value/offset pair1 (8) | value/offset pair2 (8) | ...
1939 // Note: Everything is big-endian format here. So on little endian machines, we have to revers offset and count and cmp value.
1940 void TemplateTable::fast_binaryswitch() {
1941
1942 transition(itos, vtos);
1943 // Implementation using the following core algorithm: (copied from Intel)
1944 //
1945 // int binary_search(int key, LookupswitchPair* array, int n) {
1946 // // Binary search according to "Methodik des Programmierens" by
1947 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1948 // int i = 0;
1949 // int j = n;
1950 // while (i+1 < j) {
1951 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1952 // // with Q: for all i: 0 <= i < n: key < a[i]
1953 // // where a stands for the array and assuming that the (inexisting)
1954 // // element a[n] is infinitely big.
1955 // int h = (i + j) >> 1;
1956 // // i < h < j
1957 // if (key < array[h].fast_match()) {
1958 // j = h;
1959 // } else {
1960 // i = h;
1961 // }
1962 // }
1963 // // R: a[i] <= key < a[i+1] or Q
1964 // // (i.e., if key is within array, i is the correct index)
1965 // return i;
1966 // }
1967
1968 // register allocation
1969 const Register Rkey = R17_tos; // already set (tosca)
1970 const Register Rarray = R3_ARG1;
1971 const Register Ri = R4_ARG2;
1972 const Register Rj = R5_ARG3;
1973 const Register Rh = R6_ARG4;
1974 const Register Rscratch = R11_scratch1;
1975
1976 const int log_entry_size = 3;
1977 const int entry_size = 1 << log_entry_size;
1978
1979 Label found;
1980
1981 // Find Array start,
1982 __ addi(Rarray, R14_bcp, 3 * BytesPerInt);
1983 __ clrrdi(Rarray, Rarray, log2_long((jlong)BytesPerInt));
1984
1985 // initialize i & j
1986 __ li(Ri,0);
1987 __ get_u4(Rj, Rarray, -BytesPerInt, InterpreterMacroAssembler::Unsigned);
1988
1989 // and start.
1990 Label entry;
1991 __ b(entry);
1992
1993 // binary search loop
1994 { Label loop;
1995 __ bind(loop);
1996 // int h = (i + j) >> 1;
1997 __ srdi(Rh, Rh, 1);
1998 // if (key < array[h].fast_match()) {
1999 // j = h;
2000 // } else {
2001 // i = h;
2002 // }
2003 __ sldi(Rscratch, Rh, log_entry_size);
2004 #if defined(VM_LITTLE_ENDIAN)
2005 __ lwbrx(Rscratch, Rscratch, Rarray);
2006 #else
2007 __ lwzx(Rscratch, Rscratch, Rarray);
2008 #endif
2009
2010 // if (key < current value)
2011 // Rh = Rj
2012 // else
2013 // Rh = Ri
2014 Label Lgreater;
2015 __ cmpw(CCR0, Rkey, Rscratch);
2016 __ bge(CCR0, Lgreater);
2017 __ mr(Rj, Rh);
2018 __ b(entry);
2019 __ bind(Lgreater);
2020 __ mr(Ri, Rh);
2021
2022 // while (i+1 < j)
2023 __ bind(entry);
2024 __ addi(Rscratch, Ri, 1);
2025 __ cmpw(CCR0, Rscratch, Rj);
2026 __ add(Rh, Ri, Rj); // start h = i + j >> 1;
2027
2028 __ blt(CCR0, loop);
2029 }
2030
2031 // End of binary search, result index is i (must check again!).
2032 Label default_case;
2033 Label continue_execution;
2034 if (ProfileInterpreter) {
2035 __ mr(Rh, Ri); // Save index in i for profiling.
2036 }
2037 // Ri = value offset
2038 __ sldi(Ri, Ri, log_entry_size);
2039 __ add(Ri, Ri, Rarray);
2040 __ get_u4(Rscratch, Ri, 0, InterpreterMacroAssembler::Unsigned);
2041
2042 Label not_found;
2043 // Ri = offset offset
2044 __ cmpw(CCR0, Rkey, Rscratch);
2045 __ beq(CCR0, not_found);
2046 // entry not found -> j = default offset
2047 __ get_u4(Rj, Rarray, -2 * BytesPerInt, InterpreterMacroAssembler::Unsigned);
2048 __ b(default_case);
2049
2050 __ bind(not_found);
2051 // entry found -> j = offset
2052 __ profile_switch_case(Rh, Rj, Rscratch, Rkey);
2053 __ get_u4(Rj, Ri, BytesPerInt, InterpreterMacroAssembler::Unsigned);
2054
2055 if (ProfileInterpreter) {
2056 __ b(continue_execution);
2057 }
2058
2059 __ bind(default_case); // fall through (if not profiling)
2060 __ profile_switch_default(Ri, Rscratch);
2061
2062 __ bind(continue_execution);
2063
2064 __ extsw(Rj, Rj);
2065 __ add(R14_bcp, Rj, R14_bcp);
2066 __ dispatch_next(vtos);
2067 }
2068
2069 void TemplateTable::_return(TosState state) {
2070 transition(state, state);
2071 assert(_desc->calls_vm(),
2072 "inconsistent calls_vm information"); // call in remove_activation
2073
2074 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2075
2076 Register Rscratch = R11_scratch1,
2077 Rklass = R12_scratch2,
2078 Rklass_flags = Rklass;
2079 Label Lskip_register_finalizer;
2080
2081 // Check if the method has the FINALIZER flag set and call into the VM to finalize in this case.
2082 assert(state == vtos, "only valid state");
2083 __ ld(R17_tos, 0, R18_locals);
2084
2085 // Load klass of this obj.
2086 __ load_klass(Rklass, R17_tos);
2087 __ lwz(Rklass_flags, in_bytes(Klass::access_flags_offset()), Rklass);
2088 __ testbitdi(CCR0, R0, Rklass_flags, exact_log2(JVM_ACC_HAS_FINALIZER));
2089 __ bfalse(CCR0, Lskip_register_finalizer);
2090
2091 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), R17_tos /* obj */);
2092
2093 __ align(32, 12);
2094 __ bind(Lskip_register_finalizer);
2095 }
2096
2097 // Move the result value into the correct register and remove memory stack frame.
2098 __ remove_activation(state, /* throw_monitor_exception */ true);
2099 // Restoration of lr done by remove_activation.
2100 switch (state) {
2101 case ltos:
2102 case btos:
2103 case ctos:
2104 case stos:
2105 case atos:
2106 case itos: __ mr(R3_RET, R17_tos); break;
2107 case ftos:
2108 case dtos: __ fmr(F1_RET, F15_ftos); break;
2109 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
2110 // to get visible before the reference to the object gets stored anywhere.
2111 __ membar(Assembler::StoreStore); break;
2112 default : ShouldNotReachHere();
2113 }
2114 __ blr();
2115 }
2116
2117 // ============================================================================
2118 // Constant pool cache access
2119 //
2120 // Memory ordering:
2121 //
2122 // Like done in C++ interpreter, we load the fields
2123 // - _indices
2124 // - _f12_oop
2125 // acquired, because these are asked if the cache is already resolved. We don't
2126 // want to float loads above this check.
2127 // See also comments in ConstantPoolCacheEntry::bytecode_1(),
2128 // ConstantPoolCacheEntry::bytecode_2() and ConstantPoolCacheEntry::f1();
2129
2130 // Call into the VM if call site is not yet resolved
2131 //
2132 // Input regs:
2133 // - None, all passed regs are outputs.
2134 //
2135 // Returns:
2136 // - Rcache: The const pool cache entry that contains the resolved result.
2137 // - Rresult: Either noreg or output for f1/f2.
2138 //
2139 // Kills:
2140 // - Rscratch
2141 void TemplateTable::resolve_cache_and_index(int byte_no, Register Rcache, Register Rscratch, size_t index_size) {
2142
2143 __ get_cache_and_index_at_bcp(Rcache, 1, index_size);
2144 Label Lresolved, Ldone;
2145
2146 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2147 // We are resolved if the indices offset contains the current bytecode.
2148 #if defined(VM_LITTLE_ENDIAN)
2149 __ lbz(Rscratch, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + byte_no + 1, Rcache);
2150 #else
2151 __ lbz(Rscratch, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 7 - (byte_no + 1), Rcache);
2152 #endif
2153 // Acquire by cmp-br-isync (see below).
2154 __ cmpdi(CCR0, Rscratch, (int)bytecode());
2155 __ beq(CCR0, Lresolved);
2156
2157 address entry = NULL;
2158 switch (bytecode()) {
2159 case Bytecodes::_getstatic : // fall through
2160 case Bytecodes::_putstatic : // fall through
2161 case Bytecodes::_getfield : // fall through
2162 case Bytecodes::_putfield : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break;
2163 case Bytecodes::_invokevirtual : // fall through
2164 case Bytecodes::_invokespecial : // fall through
2165 case Bytecodes::_invokestatic : // fall through
2166 case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke); break;
2167 case Bytecodes::_invokehandle : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokehandle); break;
2168 case Bytecodes::_invokedynamic : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic); break;
2169 default : ShouldNotReachHere(); break;
2170 }
2171 __ li(R4_ARG2, (int)bytecode());
2172 __ call_VM(noreg, entry, R4_ARG2, true);
2173
2174 // Update registers with resolved info.
2175 __ get_cache_and_index_at_bcp(Rcache, 1, index_size);
2176 __ b(Ldone);
2177
2178 __ bind(Lresolved);
2179 __ isync(); // Order load wrt. succeeding loads.
2180 __ bind(Ldone);
2181 }
2182
2183 // Load the constant pool cache entry at field accesses into registers.
2184 // The Rcache and Rindex registers must be set before call.
2185 // Input:
2186 // - Rcache, Rindex
2187 // Output:
2188 // - Robj, Roffset, Rflags
2189 void TemplateTable::load_field_cp_cache_entry(Register Robj,
2190 Register Rcache,
2191 Register Rindex /* unused on PPC64 */,
2192 Register Roffset,
2193 Register Rflags,
2194 bool is_static = false) {
2195 assert_different_registers(Rcache, Rflags, Roffset);
2196 // assert(Rindex == noreg, "parameter not used on PPC64");
2197
2198 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2199 __ ld(Rflags, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::flags_offset()), Rcache);
2200 __ ld(Roffset, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::f2_offset()), Rcache);
2201 if (is_static) {
2202 __ ld(Robj, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::f1_offset()), Rcache);
2203 __ ld(Robj, in_bytes(Klass::java_mirror_offset()), Robj);
2204 // Acquire not needed here. Following access has an address dependency on this value.
2205 }
2206 }
2207
2208 // Load the constant pool cache entry at invokes into registers.
2209 // Resolve if necessary.
2210
2211 // Input Registers:
2212 // - None, bcp is used, though
2213 //
2214 // Return registers:
2215 // - Rmethod (f1 field or f2 if invokevirtual)
2216 // - Ritable_index (f2 field)
2217 // - Rflags (flags field)
2218 //
2219 // Kills:
2220 // - R21
2221 //
2222 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2223 Register Rmethod,
2224 Register Ritable_index,
2225 Register Rflags,
2226 bool is_invokevirtual,
2227 bool is_invokevfinal,
2228 bool is_invokedynamic) {
2229
2230 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2231 // Determine constant pool cache field offsets.
2232 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2233 const int method_offset = in_bytes(cp_base_offset + (is_invokevirtual ? ConstantPoolCacheEntry::f2_offset() : ConstantPoolCacheEntry::f1_offset()));
2234 const int flags_offset = in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset());
2235 // Access constant pool cache fields.
2236 const int index_offset = in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset());
2237
2238 Register Rcache = R21_tmp1; // Note: same register as R21_sender_SP.
2239
2240 if (is_invokevfinal) {
2241 assert(Ritable_index == noreg, "register not used");
2242 // Already resolved.
2243 __ get_cache_and_index_at_bcp(Rcache, 1);
2244 } else {
2245 resolve_cache_and_index(byte_no, Rcache, R0, is_invokedynamic ? sizeof(u4) : sizeof(u2));
2246 }
2247
2248 __ ld(Rmethod, method_offset, Rcache);
2249 __ ld(Rflags, flags_offset, Rcache);
2250
2251 if (Ritable_index != noreg) {
2252 __ ld(Ritable_index, index_offset, Rcache);
2253 }
2254 }
2255
2256 // ============================================================================
2257 // Field access
2258
2259 // Volatile variables demand their effects be made known to all CPU's
2260 // in order. Store buffers on most chips allow reads & writes to
2261 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2262 // without some kind of memory barrier (i.e., it's not sufficient that
2263 // the interpreter does not reorder volatile references, the hardware
2264 // also must not reorder them).
2265 //
2266 // According to the new Java Memory Model (JMM):
2267 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2268 // writes act as aquire & release, so:
2269 // (2) A read cannot let unrelated NON-volatile memory refs that
2270 // happen after the read float up to before the read. It's OK for
2271 // non-volatile memory refs that happen before the volatile read to
2272 // float down below it.
2273 // (3) Similar a volatile write cannot let unrelated NON-volatile
2274 // memory refs that happen BEFORE the write float down to after the
2275 // write. It's OK for non-volatile memory refs that happen after the
2276 // volatile write to float up before it.
2277 //
2278 // We only put in barriers around volatile refs (they are expensive),
2279 // not _between_ memory refs (that would require us to track the
2280 // flavor of the previous memory refs). Requirements (2) and (3)
2281 // require some barriers before volatile stores and after volatile
2282 // loads. These nearly cover requirement (1) but miss the
2283 // volatile-store-volatile-load case. This final case is placed after
2284 // volatile-stores although it could just as well go before
2285 // volatile-loads.
2286
2287 // The registers cache and index expected to be set before call.
2288 // Correct values of the cache and index registers are preserved.
2289 // Kills:
2290 // Rcache (if has_tos)
2291 // Rscratch
2292 void TemplateTable::jvmti_post_field_access(Register Rcache, Register Rscratch, bool is_static, bool has_tos) {
2293
2294 assert_different_registers(Rcache, Rscratch);
2295
2296 if (JvmtiExport::can_post_field_access()) {
2297 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2298 Label Lno_field_access_post;
2299
2300 // Check if post field access in enabled.
2301 int offs = __ load_const_optimized(Rscratch, JvmtiExport::get_field_access_count_addr(), R0, true);
2302 __ lwz(Rscratch, offs, Rscratch);
2303
2304 __ cmpwi(CCR0, Rscratch, 0);
2305 __ beq(CCR0, Lno_field_access_post);
2306
2307 // Post access enabled - do it!
2308 __ addi(Rcache, Rcache, in_bytes(cp_base_offset));
2309 if (is_static) {
2310 __ li(R17_tos, 0);
2311 } else {
2312 if (has_tos) {
2313 // The fast bytecode versions have obj ptr in register.
2314 // Thus, save object pointer before call_VM() clobbers it
2315 // put object on tos where GC wants it.
2316 __ push_ptr(R17_tos);
2317 } else {
2318 // Load top of stack (do not pop the value off the stack).
2319 __ ld(R17_tos, Interpreter::expr_offset_in_bytes(0), R15_esp);
2320 }
2321 __ verify_oop(R17_tos);
2322 }
2323 // tos: object pointer or NULL if static
2324 // cache: cache entry pointer
2325 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), R17_tos, Rcache);
2326 if (!is_static && has_tos) {
2327 // Restore object pointer.
2328 __ pop_ptr(R17_tos);
2329 __ verify_oop(R17_tos);
2330 } else {
2331 // Cache is still needed to get class or obj.
2332 __ get_cache_and_index_at_bcp(Rcache, 1);
2333 }
2334
2335 __ align(32, 12);
2336 __ bind(Lno_field_access_post);
2337 }
2338 }
2339
2340 // kills R11_scratch1
2341 void TemplateTable::pop_and_check_object(Register Roop) {
2342 Register Rtmp = R11_scratch1;
2343
2344 assert_different_registers(Rtmp, Roop);
2345 __ pop_ptr(Roop);
2346 // For field access must check obj.
2347 __ null_check_throw(Roop, -1, Rtmp);
2348 __ verify_oop(Roop);
2349 }
2350
2351 // PPC64: implement volatile loads as fence-store-acquire.
2352 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2353 transition(vtos, vtos);
2354
2355 Label Lacquire, Lisync;
2356
2357 const Register Rcache = R3_ARG1,
2358 Rclass_or_obj = R22_tmp2,
2359 Roffset = R23_tmp3,
2360 Rflags = R31,
2361 Rbtable = R5_ARG3,
2362 Rbc = R6_ARG4,
2363 Rscratch = R12_scratch2;
2364
2365 static address field_branch_table[number_of_states],
2366 static_branch_table[number_of_states];
2367
2368 address* branch_table = is_static ? static_branch_table : field_branch_table;
2369
2370 // Get field offset.
2371 resolve_cache_and_index(byte_no, Rcache, Rscratch, sizeof(u2));
2372
2373 // JVMTI support
2374 jvmti_post_field_access(Rcache, Rscratch, is_static, false);
2375
2376 // Load after possible GC.
2377 load_field_cp_cache_entry(Rclass_or_obj, Rcache, noreg, Roffset, Rflags, is_static);
2378
2379 // Load pointer to branch table.
2380 __ load_const_optimized(Rbtable, (address)branch_table, Rscratch);
2381
2382 // Get volatile flag.
2383 __ rldicl(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2384 // Note: sync is needed before volatile load on PPC64.
2385
2386 // Check field type.
2387 __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2388
2389 #ifdef ASSERT
2390 Label LFlagInvalid;
2391 __ cmpldi(CCR0, Rflags, number_of_states);
2392 __ bge(CCR0, LFlagInvalid);
2393 #endif
2394
2395 // Load from branch table and dispatch (volatile case: one instruction ahead).
2396 __ sldi(Rflags, Rflags, LogBytesPerWord);
2397 __ cmpwi(CCR6, Rscratch, 1); // Volatile?
2398 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2399 __ sldi(Rscratch, Rscratch, exact_log2(BytesPerInstWord)); // Volatile ? size of 1 instruction : 0.
2400 }
2401 __ ldx(Rbtable, Rbtable, Rflags);
2402
2403 // Get the obj from stack.
2404 if (!is_static) {
2405 pop_and_check_object(Rclass_or_obj); // Kills R11_scratch1.
2406 } else {
2407 __ verify_oop(Rclass_or_obj);
2408 }
2409
2410 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2411 __ subf(Rbtable, Rscratch, Rbtable); // Point to volatile/non-volatile entry point.
2412 }
2413 __ mtctr(Rbtable);
2414 __ bctr();
2415
2416 #ifdef ASSERT
2417 __ bind(LFlagInvalid);
2418 __ stop("got invalid flag", 0x654);
2419
2420 // __ bind(Lvtos);
2421 address pc_before_fence = __ pc();
2422 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2423 assert(__ pc() - pc_before_fence == (ptrdiff_t)BytesPerInstWord, "must be single instruction");
2424 assert(branch_table[vtos] == 0, "can't compute twice");
2425 branch_table[vtos] = __ pc(); // non-volatile_entry point
2426 __ stop("vtos unexpected", 0x655);
2427 #endif
2428
2429 __ align(32, 28, 28); // Align load.
2430 // __ bind(Ldtos);
2431 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2432 assert(branch_table[dtos] == 0, "can't compute twice");
2433 branch_table[dtos] = __ pc(); // non-volatile_entry point
2434 __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
2435 __ push(dtos);
2436 if (!is_static) patch_bytecode(Bytecodes::_fast_dgetfield, Rbc, Rscratch);
2437 {
2438 Label acquire_double;
2439 __ beq(CCR6, acquire_double); // Volatile?
2440 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2441
2442 __ bind(acquire_double);
2443 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
2444 __ beq_predict_taken(CCR0, Lisync);
2445 __ b(Lisync); // In case of NAN.
2446 }
2447
2448 __ align(32, 28, 28); // Align load.
2449 // __ bind(Lftos);
2450 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2451 assert(branch_table[ftos] == 0, "can't compute twice");
2452 branch_table[ftos] = __ pc(); // non-volatile_entry point
2453 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
2454 __ push(ftos);
2455 if (!is_static) { patch_bytecode(Bytecodes::_fast_fgetfield, Rbc, Rscratch); }
2456 {
2457 Label acquire_float;
2458 __ beq(CCR6, acquire_float); // Volatile?
2459 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2460
2461 __ bind(acquire_float);
2462 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
2463 __ beq_predict_taken(CCR0, Lisync);
2464 __ b(Lisync); // In case of NAN.
2465 }
2466
2467 __ align(32, 28, 28); // Align load.
2468 // __ bind(Litos);
2469 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2470 assert(branch_table[itos] == 0, "can't compute twice");
2471 branch_table[itos] = __ pc(); // non-volatile_entry point
2472 __ lwax(R17_tos, Rclass_or_obj, Roffset);
2473 __ push(itos);
2474 if (!is_static) patch_bytecode(Bytecodes::_fast_igetfield, Rbc, Rscratch);
2475 __ beq(CCR6, Lacquire); // Volatile?
2476 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2477
2478 __ align(32, 28, 28); // Align load.
2479 // __ bind(Lltos);
2480 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2481 assert(branch_table[ltos] == 0, "can't compute twice");
2482 branch_table[ltos] = __ pc(); // non-volatile_entry point
2483 __ ldx(R17_tos, Rclass_or_obj, Roffset);
2484 __ push(ltos);
2485 if (!is_static) patch_bytecode(Bytecodes::_fast_lgetfield, Rbc, Rscratch);
2486 __ beq(CCR6, Lacquire); // Volatile?
2487 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2488
2489 __ align(32, 28, 28); // Align load.
2490 // __ bind(Lbtos);
2491 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2492 assert(branch_table[btos] == 0, "can't compute twice");
2493 branch_table[btos] = __ pc(); // non-volatile_entry point
2494 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
2495 __ extsb(R17_tos, R17_tos);
2496 __ push(btos);
2497 if (!is_static) patch_bytecode(Bytecodes::_fast_bgetfield, Rbc, Rscratch);
2498 __ beq(CCR6, Lacquire); // Volatile?
2499 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2500
2501 __ align(32, 28, 28); // Align load.
2502 // __ bind(Lctos);
2503 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2504 assert(branch_table[ctos] == 0, "can't compute twice");
2505 branch_table[ctos] = __ pc(); // non-volatile_entry point
2506 __ lhzx(R17_tos, Rclass_or_obj, Roffset);
2507 __ push(ctos);
2508 if (!is_static) patch_bytecode(Bytecodes::_fast_cgetfield, Rbc, Rscratch);
2509 __ beq(CCR6, Lacquire); // Volatile?
2510 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2511
2512 __ align(32, 28, 28); // Align load.
2513 // __ bind(Lstos);
2514 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2515 assert(branch_table[stos] == 0, "can't compute twice");
2516 branch_table[stos] = __ pc(); // non-volatile_entry point
2517 __ lhax(R17_tos, Rclass_or_obj, Roffset);
2518 __ push(stos);
2519 if (!is_static) patch_bytecode(Bytecodes::_fast_sgetfield, Rbc, Rscratch);
2520 __ beq(CCR6, Lacquire); // Volatile?
2521 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2522
2523 __ align(32, 28, 28); // Align load.
2524 // __ bind(Latos);
2525 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2526 assert(branch_table[atos] == 0, "can't compute twice");
2527 branch_table[atos] = __ pc(); // non-volatile_entry point
2528 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
2529 __ verify_oop(R17_tos);
2530 __ push(atos);
2531 //__ dcbt(R17_tos); // prefetch
2532 if (!is_static) patch_bytecode(Bytecodes::_fast_agetfield, Rbc, Rscratch);
2533 __ beq(CCR6, Lacquire); // Volatile?
2534 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2535
2536 __ align(32, 12);
2537 __ bind(Lacquire);
2538 __ twi_0(R17_tos);
2539 __ bind(Lisync);
2540 __ isync(); // acquire
2541
2542 #ifdef ASSERT
2543 for (int i = 0; i<number_of_states; ++i) {
2544 assert(branch_table[i], "get initialization");
2545 //tty->print_cr("get: %s_branch_table[%d] = 0x%llx (opcode 0x%llx)",
2546 // is_static ? "static" : "field", i, branch_table[i], *((unsigned int*)branch_table[i]));
2547 }
2548 #endif
2549 }
2550
2551 void TemplateTable::getfield(int byte_no) {
2552 getfield_or_static(byte_no, false);
2553 }
2554
2555 void TemplateTable::getstatic(int byte_no) {
2556 getfield_or_static(byte_no, true);
2557 }
2558
2559 // The registers cache and index expected to be set before call.
2560 // The function may destroy various registers, just not the cache and index registers.
2561 void TemplateTable::jvmti_post_field_mod(Register Rcache, Register Rscratch, bool is_static) {
2562
2563 assert_different_registers(Rcache, Rscratch, R6_ARG4);
2564
2565 if (JvmtiExport::can_post_field_modification()) {
2566 Label Lno_field_mod_post;
2567
2568 // Check if post field access in enabled.
2569 int offs = __ load_const_optimized(Rscratch, JvmtiExport::get_field_modification_count_addr(), R0, true);
2570 __ lwz(Rscratch, offs, Rscratch);
2571
2572 __ cmpwi(CCR0, Rscratch, 0);
2573 __ beq(CCR0, Lno_field_mod_post);
2574
2575 // Do the post
2576 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2577 const Register Robj = Rscratch;
2578
2579 __ addi(Rcache, Rcache, in_bytes(cp_base_offset));
2580 if (is_static) {
2581 // Life is simple. Null out the object pointer.
2582 __ li(Robj, 0);
2583 } else {
2584 // In case of the fast versions, value lives in registers => put it back on tos.
2585 int offs = Interpreter::expr_offset_in_bytes(0);
2586 Register base = R15_esp;
2587 switch(bytecode()) {
2588 case Bytecodes::_fast_aputfield: __ push_ptr(); offs+= Interpreter::stackElementSize; break;
2589 case Bytecodes::_fast_iputfield: // Fall through
2590 case Bytecodes::_fast_bputfield: // Fall through
2591 case Bytecodes::_fast_cputfield: // Fall through
2592 case Bytecodes::_fast_sputfield: __ push_i(); offs+= Interpreter::stackElementSize; break;
2593 case Bytecodes::_fast_lputfield: __ push_l(); offs+=2*Interpreter::stackElementSize; break;
2594 case Bytecodes::_fast_fputfield: __ push_f(); offs+= Interpreter::stackElementSize; break;
2595 case Bytecodes::_fast_dputfield: __ push_d(); offs+=2*Interpreter::stackElementSize; break;
2596 default: {
2597 offs = 0;
2598 base = Robj;
2599 const Register Rflags = Robj;
2600 Label is_one_slot;
2601 // Life is harder. The stack holds the value on top, followed by the
2602 // object. We don't know the size of the value, though; it could be
2603 // one or two words depending on its type. As a result, we must find
2604 // the type to determine where the object is.
2605 __ ld(Rflags, in_bytes(ConstantPoolCacheEntry::flags_offset()), Rcache); // Big Endian
2606 __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2607
2608 __ cmpwi(CCR0, Rflags, ltos);
2609 __ cmpwi(CCR1, Rflags, dtos);
2610 __ addi(base, R15_esp, Interpreter::expr_offset_in_bytes(1));
2611 __ crnor(CCR0, Assembler::equal, CCR1, Assembler::equal);
2612 __ beq(CCR0, is_one_slot);
2613 __ addi(base, R15_esp, Interpreter::expr_offset_in_bytes(2));
2614 __ bind(is_one_slot);
2615 break;
2616 }
2617 }
2618 __ ld(Robj, offs, base);
2619 __ verify_oop(Robj);
2620 }
2621
2622 __ addi(R6_ARG4, R15_esp, Interpreter::expr_offset_in_bytes(0));
2623 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), Robj, Rcache, R6_ARG4);
2624 __ get_cache_and_index_at_bcp(Rcache, 1);
2625
2626 // In case of the fast versions, value lives in registers => put it back on tos.
2627 switch(bytecode()) {
2628 case Bytecodes::_fast_aputfield: __ pop_ptr(); break;
2629 case Bytecodes::_fast_iputfield: // Fall through
2630 case Bytecodes::_fast_bputfield: // Fall through
2631 case Bytecodes::_fast_cputfield: // Fall through
2632 case Bytecodes::_fast_sputfield: __ pop_i(); break;
2633 case Bytecodes::_fast_lputfield: __ pop_l(); break;
2634 case Bytecodes::_fast_fputfield: __ pop_f(); break;
2635 case Bytecodes::_fast_dputfield: __ pop_d(); break;
2636 default: break; // Nothin' to do.
2637 }
2638
2639 __ align(32, 12);
2640 __ bind(Lno_field_mod_post);
2641 }
2642 }
2643
2644 // PPC64: implement volatile stores as release-store (return bytecode contains an additional release).
2645 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2646 Label Lvolatile;
2647
2648 const Register Rcache = R5_ARG3, // Do not use ARG1/2 (causes trouble in jvmti_post_field_mod).
2649 Rclass_or_obj = R31, // Needs to survive C call.
2650 Roffset = R22_tmp2, // Needs to survive C call.
2651 Rflags = R3_ARG1,
2652 Rbtable = R4_ARG2,
2653 Rscratch = R11_scratch1,
2654 Rscratch2 = R12_scratch2,
2655 Rscratch3 = R6_ARG4,
2656 Rbc = Rscratch3;
2657 const ConditionRegister CR_is_vol = CCR2; // Non-volatile condition register (survives runtime call in do_oop_store).
2658
2659 static address field_branch_table[number_of_states],
2660 static_branch_table[number_of_states];
2661
2662 address* branch_table = is_static ? static_branch_table : field_branch_table;
2663
2664 // Stack (grows up):
2665 // value
2666 // obj
2667
2668 // Load the field offset.
2669 resolve_cache_and_index(byte_no, Rcache, Rscratch, sizeof(u2));
2670 jvmti_post_field_mod(Rcache, Rscratch, is_static);
2671 load_field_cp_cache_entry(Rclass_or_obj, Rcache, noreg, Roffset, Rflags, is_static);
2672
2673 // Load pointer to branch table.
2674 __ load_const_optimized(Rbtable, (address)branch_table, Rscratch);
2675
2676 // Get volatile flag.
2677 __ rldicl(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2678
2679 // Check the field type.
2680 __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2681
2682 #ifdef ASSERT
2683 Label LFlagInvalid;
2684 __ cmpldi(CCR0, Rflags, number_of_states);
2685 __ bge(CCR0, LFlagInvalid);
2686 #endif
2687
2688 // Load from branch table and dispatch (volatile case: one instruction ahead).
2689 __ sldi(Rflags, Rflags, LogBytesPerWord);
2690 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) { __ cmpwi(CR_is_vol, Rscratch, 1); } // Volatile?
2691 __ sldi(Rscratch, Rscratch, exact_log2(BytesPerInstWord)); // Volatile? size of instruction 1 : 0.
2692 __ ldx(Rbtable, Rbtable, Rflags);
2693
2694 __ subf(Rbtable, Rscratch, Rbtable); // Point to volatile/non-volatile entry point.
2695 __ mtctr(Rbtable);
2696 __ bctr();
2697
2698 #ifdef ASSERT
2699 __ bind(LFlagInvalid);
2700 __ stop("got invalid flag", 0x656);
2701
2702 // __ bind(Lvtos);
2703 address pc_before_release = __ pc();
2704 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2705 assert(__ pc() - pc_before_release == (ptrdiff_t)BytesPerInstWord, "must be single instruction");
2706 assert(branch_table[vtos] == 0, "can't compute twice");
2707 branch_table[vtos] = __ pc(); // non-volatile_entry point
2708 __ stop("vtos unexpected", 0x657);
2709 #endif
2710
2711 __ align(32, 28, 28); // Align pop.
2712 // __ bind(Ldtos);
2713 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2714 assert(branch_table[dtos] == 0, "can't compute twice");
2715 branch_table[dtos] = __ pc(); // non-volatile_entry point
2716 __ pop(dtos);
2717 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2718 __ stfdx(F15_ftos, Rclass_or_obj, Roffset);
2719 if (!is_static) { patch_bytecode(Bytecodes::_fast_dputfield, Rbc, Rscratch, true, byte_no); }
2720 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2721 __ beq(CR_is_vol, Lvolatile); // Volatile?
2722 }
2723 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2724
2725 __ align(32, 28, 28); // Align pop.
2726 // __ bind(Lftos);
2727 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2728 assert(branch_table[ftos] == 0, "can't compute twice");
2729 branch_table[ftos] = __ pc(); // non-volatile_entry point
2730 __ pop(ftos);
2731 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2732 __ stfsx(F15_ftos, Rclass_or_obj, Roffset);
2733 if (!is_static) { patch_bytecode(Bytecodes::_fast_fputfield, Rbc, Rscratch, true, byte_no); }
2734 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2735 __ beq(CR_is_vol, Lvolatile); // Volatile?
2736 }
2737 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2738
2739 __ align(32, 28, 28); // Align pop.
2740 // __ bind(Litos);
2741 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2742 assert(branch_table[itos] == 0, "can't compute twice");
2743 branch_table[itos] = __ pc(); // non-volatile_entry point
2744 __ pop(itos);
2745 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2746 __ stwx(R17_tos, Rclass_or_obj, Roffset);
2747 if (!is_static) { patch_bytecode(Bytecodes::_fast_iputfield, Rbc, Rscratch, true, byte_no); }
2748 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2749 __ beq(CR_is_vol, Lvolatile); // Volatile?
2750 }
2751 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2752
2753 __ align(32, 28, 28); // Align pop.
2754 // __ bind(Lltos);
2755 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2756 assert(branch_table[ltos] == 0, "can't compute twice");
2757 branch_table[ltos] = __ pc(); // non-volatile_entry point
2758 __ pop(ltos);
2759 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2760 __ stdx(R17_tos, Rclass_or_obj, Roffset);
2761 if (!is_static) { patch_bytecode(Bytecodes::_fast_lputfield, Rbc, Rscratch, true, byte_no); }
2762 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2763 __ beq(CR_is_vol, Lvolatile); // Volatile?
2764 }
2765 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2766
2767 __ align(32, 28, 28); // Align pop.
2768 // __ bind(Lbtos);
2769 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2770 assert(branch_table[btos] == 0, "can't compute twice");
2771 branch_table[btos] = __ pc(); // non-volatile_entry point
2772 __ pop(btos);
2773 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2774 __ stbx(R17_tos, Rclass_or_obj, Roffset);
2775 if (!is_static) { patch_bytecode(Bytecodes::_fast_bputfield, Rbc, Rscratch, true, byte_no); }
2776 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2777 __ beq(CR_is_vol, Lvolatile); // Volatile?
2778 }
2779 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2780
2781 __ align(32, 28, 28); // Align pop.
2782 // __ bind(Lctos);
2783 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2784 assert(branch_table[ctos] == 0, "can't compute twice");
2785 branch_table[ctos] = __ pc(); // non-volatile_entry point
2786 __ pop(ctos);
2787 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1..
2788 __ sthx(R17_tos, Rclass_or_obj, Roffset);
2789 if (!is_static) { patch_bytecode(Bytecodes::_fast_cputfield, Rbc, Rscratch, true, byte_no); }
2790 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2791 __ beq(CR_is_vol, Lvolatile); // Volatile?
2792 }
2793 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2794
2795 __ align(32, 28, 28); // Align pop.
2796 // __ bind(Lstos);
2797 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2798 assert(branch_table[stos] == 0, "can't compute twice");
2799 branch_table[stos] = __ pc(); // non-volatile_entry point
2800 __ pop(stos);
2801 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2802 __ sthx(R17_tos, Rclass_or_obj, Roffset);
2803 if (!is_static) { patch_bytecode(Bytecodes::_fast_sputfield, Rbc, Rscratch, true, byte_no); }
2804 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2805 __ beq(CR_is_vol, Lvolatile); // Volatile?
2806 }
2807 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2808
2809 __ align(32, 28, 28); // Align pop.
2810 // __ bind(Latos);
2811 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2812 assert(branch_table[atos] == 0, "can't compute twice");
2813 branch_table[atos] = __ pc(); // non-volatile_entry point
2814 __ pop(atos);
2815 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // kills R11_scratch1
2816 do_oop_store(_masm, Rclass_or_obj, Roffset, R17_tos, Rscratch, Rscratch2, Rscratch3, _bs->kind(), false /* precise */, true /* check null */);
2817 if (!is_static) { patch_bytecode(Bytecodes::_fast_aputfield, Rbc, Rscratch, true, byte_no); }
2818 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2819 __ beq(CR_is_vol, Lvolatile); // Volatile?
2820 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2821
2822 __ align(32, 12);
2823 __ bind(Lvolatile);
2824 __ fence();
2825 }
2826 // fallthru: __ b(Lexit);
2827
2828 #ifdef ASSERT
2829 for (int i = 0; i<number_of_states; ++i) {
2830 assert(branch_table[i], "put initialization");
2831 //tty->print_cr("put: %s_branch_table[%d] = 0x%llx (opcode 0x%llx)",
2832 // is_static ? "static" : "field", i, branch_table[i], *((unsigned int*)branch_table[i]));
2833 }
2834 #endif
2835 }
2836
2837 void TemplateTable::putfield(int byte_no) {
2838 putfield_or_static(byte_no, false);
2839 }
2840
2841 void TemplateTable::putstatic(int byte_no) {
2842 putfield_or_static(byte_no, true);
2843 }
2844
2845 // See SPARC. On PPC64, we have a different jvmti_post_field_mod which does the job.
2846 void TemplateTable::jvmti_post_fast_field_mod() {
2847 __ should_not_reach_here();
2848 }
2849
2850 void TemplateTable::fast_storefield(TosState state) {
2851 transition(state, vtos);
2852
2853 const Register Rcache = R5_ARG3, // Do not use ARG1/2 (causes trouble in jvmti_post_field_mod).
2854 Rclass_or_obj = R31, // Needs to survive C call.
2855 Roffset = R22_tmp2, // Needs to survive C call.
2856 Rflags = R3_ARG1,
2857 Rscratch = R11_scratch1,
2858 Rscratch2 = R12_scratch2,
2859 Rscratch3 = R4_ARG2;
2860 const ConditionRegister CR_is_vol = CCR2; // Non-volatile condition register (survives runtime call in do_oop_store).
2861
2862 // Constant pool already resolved => Load flags and offset of field.
2863 __ get_cache_and_index_at_bcp(Rcache, 1);
2864 jvmti_post_field_mod(Rcache, Rscratch, false /* not static */);
2865 load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
2866
2867 // Get the obj and the final store addr.
2868 pop_and_check_object(Rclass_or_obj); // Kills R11_scratch1.
2869
2870 // Get volatile flag.
2871 __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2872 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) { __ cmpdi(CR_is_vol, Rscratch, 1); }
2873 {
2874 Label LnotVolatile;
2875 __ beq(CCR0, LnotVolatile);
2876 __ release();
2877 __ align(32, 12);
2878 __ bind(LnotVolatile);
2879 }
2880
2881 // Do the store and fencing.
2882 switch(bytecode()) {
2883 case Bytecodes::_fast_aputfield:
2884 // Store into the field.
2885 do_oop_store(_masm, Rclass_or_obj, Roffset, R17_tos, Rscratch, Rscratch2, Rscratch3, _bs->kind(), false /* precise */, true /* check null */);
2886 break;
2887
2888 case Bytecodes::_fast_iputfield:
2889 __ stwx(R17_tos, Rclass_or_obj, Roffset);
2890 break;
2891
2892 case Bytecodes::_fast_lputfield:
2893 __ stdx(R17_tos, Rclass_or_obj, Roffset);
2894 break;
2895
2896 case Bytecodes::_fast_bputfield:
2897 __ stbx(R17_tos, Rclass_or_obj, Roffset);
2898 break;
2899
2900 case Bytecodes::_fast_cputfield:
2901 case Bytecodes::_fast_sputfield:
2902 __ sthx(R17_tos, Rclass_or_obj, Roffset);
2903 break;
2904
2905 case Bytecodes::_fast_fputfield:
2906 __ stfsx(F15_ftos, Rclass_or_obj, Roffset);
2907 break;
2908
2909 case Bytecodes::_fast_dputfield:
2910 __ stfdx(F15_ftos, Rclass_or_obj, Roffset);
2911 break;
2912
2913 default: ShouldNotReachHere();
2914 }
2915
2916 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2917 Label LVolatile;
2918 __ beq(CR_is_vol, LVolatile);
2919 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2920
2921 __ align(32, 12);
2922 __ bind(LVolatile);
2923 __ fence();
2924 }
2925 }
2926
2927 void TemplateTable::fast_accessfield(TosState state) {
2928 transition(atos, state);
2929
2930 Label LisVolatile;
2931 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2932
2933 const Register Rcache = R3_ARG1,
2934 Rclass_or_obj = R17_tos,
2935 Roffset = R22_tmp2,
2936 Rflags = R23_tmp3,
2937 Rscratch = R12_scratch2;
2938
2939 // Constant pool already resolved. Get the field offset.
2940 __ get_cache_and_index_at_bcp(Rcache, 1);
2941 load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
2942
2943 // JVMTI support
2944 jvmti_post_field_access(Rcache, Rscratch, false, true);
2945
2946 // Get the load address.
2947 __ null_check_throw(Rclass_or_obj, -1, Rscratch);
2948
2949 // Get volatile flag.
2950 __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2951 __ bne(CCR0, LisVolatile);
2952
2953 switch(bytecode()) {
2954 case Bytecodes::_fast_agetfield:
2955 {
2956 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
2957 __ verify_oop(R17_tos);
2958 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
2959
2960 __ bind(LisVolatile);
2961 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
2962 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
2963 __ verify_oop(R17_tos);
2964 __ twi_0(R17_tos);
2965 __ isync();
2966 break;
2967 }
2968 case Bytecodes::_fast_igetfield:
2969 {
2970 __ lwax(R17_tos, Rclass_or_obj, Roffset);
2971 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
2972
2973 __ bind(LisVolatile);
2974 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
2975 __ lwax(R17_tos, Rclass_or_obj, Roffset);
2976 __ twi_0(R17_tos);
2977 __ isync();
2978 break;
2979 }
2980 case Bytecodes::_fast_lgetfield:
2981 {
2982 __ ldx(R17_tos, Rclass_or_obj, Roffset);
2983 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
2984
2985 __ bind(LisVolatile);
2986 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
2987 __ ldx(R17_tos, Rclass_or_obj, Roffset);
2988 __ twi_0(R17_tos);
2989 __ isync();
2990 break;
2991 }
2992 case Bytecodes::_fast_bgetfield:
2993 {
2994 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
2995 __ extsb(R17_tos, R17_tos);
2996 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
2997
2998 __ bind(LisVolatile);
2999 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3000 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
3001 __ twi_0(R17_tos);
3002 __ extsb(R17_tos, R17_tos);
3003 __ isync();
3004 break;
3005 }
3006 case Bytecodes::_fast_cgetfield:
3007 {
3008 __ lhzx(R17_tos, Rclass_or_obj, Roffset);
3009 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3010
3011 __ bind(LisVolatile);
3012 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3013 __ lhzx(R17_tos, Rclass_or_obj, Roffset);
3014 __ twi_0(R17_tos);
3015 __ isync();
3016 break;
3017 }
3018 case Bytecodes::_fast_sgetfield:
3019 {
3020 __ lhax(R17_tos, Rclass_or_obj, Roffset);
3021 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3022
3023 __ bind(LisVolatile);
3024 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3025 __ lhax(R17_tos, Rclass_or_obj, Roffset);
3026 __ twi_0(R17_tos);
3027 __ isync();
3028 break;
3029 }
3030 case Bytecodes::_fast_fgetfield:
3031 {
3032 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3033 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3034
3035 __ bind(LisVolatile);
3036 Label Ldummy;
3037 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3038 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3039 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3040 __ bne_predict_not_taken(CCR0, Ldummy);
3041 __ bind(Ldummy);
3042 __ isync();
3043 break;
3044 }
3045 case Bytecodes::_fast_dgetfield:
3046 {
3047 __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
3048 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3049
3050 __ bind(LisVolatile);
3051 Label Ldummy;
3052 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3053 __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
3054 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3055 __ bne_predict_not_taken(CCR0, Ldummy);
3056 __ bind(Ldummy);
3057 __ isync();
3058 break;
3059 }
3060 default: ShouldNotReachHere();
3061 }
3062 }
3063
3064 void TemplateTable::fast_xaccess(TosState state) {
3065 transition(vtos, state);
3066
3067 Label LisVolatile;
3068 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
3069 const Register Rcache = R3_ARG1,
3070 Rclass_or_obj = R17_tos,
3071 Roffset = R22_tmp2,
3072 Rflags = R23_tmp3,
3073 Rscratch = R12_scratch2;
3074
3075 __ ld(Rclass_or_obj, 0, R18_locals);
3076
3077 // Constant pool already resolved. Get the field offset.
3078 __ get_cache_and_index_at_bcp(Rcache, 2);
3079 load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
3080
3081 // JVMTI support not needed, since we switch back to single bytecode as soon as debugger attaches.
3082
3083 // Needed to report exception at the correct bcp.
3084 __ addi(R14_bcp, R14_bcp, 1);
3085
3086 // Get the load address.
3087 __ null_check_throw(Rclass_or_obj, -1, Rscratch);
3088
3089 // Get volatile flag.
3090 __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
3091 __ bne(CCR0, LisVolatile);
3092
3093 switch(state) {
3094 case atos:
3095 {
3096 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3097 __ verify_oop(R17_tos);
3098 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3099
3100 __ bind(LisVolatile);
3101 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3102 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3103 __ verify_oop(R17_tos);
3104 __ twi_0(R17_tos);
3105 __ isync();
3106 break;
3107 }
3108 case itos:
3109 {
3110 __ lwax(R17_tos, Rclass_or_obj, Roffset);
3111 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3112
3113 __ bind(LisVolatile);
3114 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3115 __ lwax(R17_tos, Rclass_or_obj, Roffset);
3116 __ twi_0(R17_tos);
3117 __ isync();
3118 break;
3119 }
3120 case ftos:
3121 {
3122 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3123 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3124
3125 __ bind(LisVolatile);
3126 Label Ldummy;
3127 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3128 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3129 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3130 __ bne_predict_not_taken(CCR0, Ldummy);
3131 __ bind(Ldummy);
3132 __ isync();
3133 break;
3134 }
3135 default: ShouldNotReachHere();
3136 }
3137 __ addi(R14_bcp, R14_bcp, -1);
3138 }
3139
3140 // ============================================================================
3141 // Calls
3142
3143 // Common code for invoke
3144 //
3145 // Input:
3146 // - byte_no
3147 //
3148 // Output:
3149 // - Rmethod: The method to invoke next.
3150 // - Rret_addr: The return address to return to.
3151 // - Rindex: MethodType (invokehandle) or CallSite obj (invokedynamic)
3152 // - Rrecv: Cache for "this" pointer, might be noreg if static call.
3153 // - Rflags: Method flags from const pool cache.
3154 //
3155 // Kills:
3156 // - Rscratch1
3157 //
3158 void TemplateTable::prepare_invoke(int byte_no,
3159 Register Rmethod, // linked method (or i-klass)
3160 Register Rret_addr,// return address
3161 Register Rindex, // itable index, MethodType, etc.
3162 Register Rrecv, // If caller wants to see it.
3163 Register Rflags, // If caller wants to test it.
3164 Register Rscratch
3165 ) {
3166 // Determine flags.
3167 const Bytecodes::Code code = bytecode();
3168 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3169 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3170 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3171 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3172 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3173 const bool load_receiver = (Rrecv != noreg);
3174 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3175
3176 assert_different_registers(Rmethod, Rindex, Rflags, Rscratch);
3177 assert_different_registers(Rmethod, Rrecv, Rflags, Rscratch);
3178 assert_different_registers(Rret_addr, Rscratch);
3179
3180 load_invoke_cp_cache_entry(byte_no, Rmethod, Rindex, Rflags, is_invokevirtual, false, is_invokedynamic);
3181
3182 // Saving of SP done in call_from_interpreter.
3183
3184 // Maybe push "appendix" to arguments.
3185 if (is_invokedynamic || is_invokehandle) {
3186 Label Ldone;
3187 __ rldicl_(R0, Rflags, 64-ConstantPoolCacheEntry::has_appendix_shift, 63);
3188 __ beq(CCR0, Ldone);
3189 // Push "appendix" (MethodType, CallSite, etc.).
3190 // This must be done before we get the receiver,
3191 // since the parameter_size includes it.
3192 __ load_resolved_reference_at_index(Rscratch, Rindex);
3193 __ verify_oop(Rscratch);
3194 __ push_ptr(Rscratch);
3195 __ bind(Ldone);
3196 }
3197
3198 // Load receiver if needed (after appendix is pushed so parameter size is correct).
3199 if (load_receiver) {
3200 const Register Rparam_count = Rscratch;
3201 __ andi(Rparam_count, Rflags, ConstantPoolCacheEntry::parameter_size_mask);
3202 __ load_receiver(Rparam_count, Rrecv);
3203 __ verify_oop(Rrecv);
3204 }
3205
3206 // Get return address.
3207 {
3208 Register Rtable_addr = Rscratch;
3209 Register Rret_type = Rret_addr;
3210 address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3211
3212 // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3213 __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3214 __ load_dispatch_table(Rtable_addr, (address*)table_addr);
3215 __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3216 // Get return address.
3217 __ ldx(Rret_addr, Rtable_addr, Rret_type);
3218 }
3219 }
3220
3221 // Helper for virtual calls. Load target out of vtable and jump off!
3222 // Kills all passed registers.
3223 void TemplateTable::generate_vtable_call(Register Rrecv_klass, Register Rindex, Register Rret, Register Rtemp) {
3224
3225 assert_different_registers(Rrecv_klass, Rtemp, Rret);
3226 const Register Rtarget_method = Rindex;
3227
3228 // Get target method & entry point.
3229 const int base = InstanceKlass::vtable_start_offset() * wordSize;
3230 // Calc vtable addr scale the vtable index by 8.
3231 __ sldi(Rindex, Rindex, exact_log2(vtableEntry::size() * wordSize));
3232 // Load target.
3233 __ addi(Rrecv_klass, Rrecv_klass, base + vtableEntry::method_offset_in_bytes());
3234 __ ldx(Rtarget_method, Rindex, Rrecv_klass);
3235 // Argument and return type profiling.
3236 __ profile_arguments_type(Rtarget_method, Rrecv_klass /* scratch1 */, Rtemp /* scratch2 */, true);
3237 __ call_from_interpreter(Rtarget_method, Rret, Rrecv_klass /* scratch1 */, Rtemp /* scratch2 */);
3238 }
3239
3240 // Virtual or final call. Final calls are rewritten on the fly to run through "fast_finalcall" next time.
3241 void TemplateTable::invokevirtual(int byte_no) {
3242 transition(vtos, vtos);
3243
3244 Register Rtable_addr = R11_scratch1,
3245 Rret_type = R12_scratch2,
3246 Rret_addr = R5_ARG3,
3247 Rflags = R22_tmp2, // Should survive C call.
3248 Rrecv = R3_ARG1,
3249 Rrecv_klass = Rrecv,
3250 Rvtableindex_or_method = R31, // Should survive C call.
3251 Rnum_params = R4_ARG2,
3252 Rnew_bc = R6_ARG4;
3253
3254 Label LnotFinal;
3255
3256 load_invoke_cp_cache_entry(byte_no, Rvtableindex_or_method, noreg, Rflags, /*virtual*/ true, false, false);
3257
3258 __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_vfinal_shift);
3259 __ bfalse(CCR0, LnotFinal);
3260
3261 patch_bytecode(Bytecodes::_fast_invokevfinal, Rnew_bc, R12_scratch2);
3262 invokevfinal_helper(Rvtableindex_or_method, Rflags, R11_scratch1, R12_scratch2);
3263
3264 __ align(32, 12);
3265 __ bind(LnotFinal);
3266 // Load "this" pointer (receiver).
3267 __ rldicl(Rnum_params, Rflags, 64, 48);
3268 __ load_receiver(Rnum_params, Rrecv);
3269 __ verify_oop(Rrecv);
3270
3271 // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3272 __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3273 __ load_dispatch_table(Rtable_addr, Interpreter::invoke_return_entry_table());
3274 __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3275 __ ldx(Rret_addr, Rret_type, Rtable_addr);
3276 __ null_check_throw(Rrecv, oopDesc::klass_offset_in_bytes(), R11_scratch1);
3277 __ load_klass(Rrecv_klass, Rrecv);
3278 __ verify_klass_ptr(Rrecv_klass);
3279 __ profile_virtual_call(Rrecv_klass, R11_scratch1, R12_scratch2, false);
3280
3281 generate_vtable_call(Rrecv_klass, Rvtableindex_or_method, Rret_addr, R11_scratch1);
3282 }
3283
3284 void TemplateTable::fast_invokevfinal(int byte_no) {
3285 transition(vtos, vtos);
3286
3287 assert(byte_no == f2_byte, "use this argument");
3288 Register Rflags = R22_tmp2,
3289 Rmethod = R31;
3290 load_invoke_cp_cache_entry(byte_no, Rmethod, noreg, Rflags, /*virtual*/ true, /*is_invokevfinal*/ true, false);
3291 invokevfinal_helper(Rmethod, Rflags, R11_scratch1, R12_scratch2);
3292 }
3293
3294 void TemplateTable::invokevfinal_helper(Register Rmethod, Register Rflags, Register Rscratch1, Register Rscratch2) {
3295
3296 assert_different_registers(Rmethod, Rflags, Rscratch1, Rscratch2);
3297
3298 // Load receiver from stack slot.
3299 Register Rrecv = Rscratch2;
3300 Register Rnum_params = Rrecv;
3301
3302 __ ld(Rnum_params, in_bytes(Method::const_offset()), Rmethod);
3303 __ lhz(Rnum_params /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), Rnum_params);
3304
3305 // Get return address.
3306 Register Rtable_addr = Rscratch1,
3307 Rret_addr = Rflags,
3308 Rret_type = Rret_addr;
3309 // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3310 __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3311 __ load_dispatch_table(Rtable_addr, Interpreter::invoke_return_entry_table());
3312 __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3313 __ ldx(Rret_addr, Rret_type, Rtable_addr);
3314
3315 // Load receiver and receiver NULL check.
3316 __ load_receiver(Rnum_params, Rrecv);
3317 __ null_check_throw(Rrecv, -1, Rscratch1);
3318
3319 __ profile_final_call(Rrecv, Rscratch1);
3320 // Argument and return type profiling.
3321 __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, true);
3322
3323 // Do the call.
3324 __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1, Rscratch2);
3325 }
3326
3327 void TemplateTable::invokespecial(int byte_no) {
3328 assert(byte_no == f1_byte, "use this argument");
3329 transition(vtos, vtos);
3330
3331 Register Rtable_addr = R3_ARG1,
3332 Rret_addr = R4_ARG2,
3333 Rflags = R5_ARG3,
3334 Rreceiver = R6_ARG4,
3335 Rmethod = R31;
3336
3337 prepare_invoke(byte_no, Rmethod, Rret_addr, noreg, Rreceiver, Rflags, R11_scratch1);
3338
3339 // Receiver NULL check.
3340 __ null_check_throw(Rreceiver, -1, R11_scratch1);
3341
3342 __ profile_call(R11_scratch1, R12_scratch2);
3343 // Argument and return type profiling.
3344 __ profile_arguments_type(Rmethod, R11_scratch1, R12_scratch2, false);
3345 __ call_from_interpreter(Rmethod, Rret_addr, R11_scratch1, R12_scratch2);
3346 }
3347
3348 void TemplateTable::invokestatic(int byte_no) {
3349 assert(byte_no == f1_byte, "use this argument");
3350 transition(vtos, vtos);
3351
3352 Register Rtable_addr = R3_ARG1,
3353 Rret_addr = R4_ARG2,
3354 Rflags = R5_ARG3;
3355
3356 prepare_invoke(byte_no, R19_method, Rret_addr, noreg, noreg, Rflags, R11_scratch1);
3357
3358 __ profile_call(R11_scratch1, R12_scratch2);
3359 // Argument and return type profiling.
3360 __ profile_arguments_type(R19_method, R11_scratch1, R12_scratch2, false);
3361 __ call_from_interpreter(R19_method, Rret_addr, R11_scratch1, R12_scratch2);
3362 }
3363
3364 void TemplateTable::invokeinterface_object_method(Register Rrecv_klass,
3365 Register Rret,
3366 Register Rflags,
3367 Register Rindex,
3368 Register Rtemp1,
3369 Register Rtemp2) {
3370
3371 assert_different_registers(Rindex, Rret, Rrecv_klass, Rflags, Rtemp1, Rtemp2);
3372 Label LnotFinal;
3373
3374 // Check for vfinal.
3375 __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_vfinal_shift);
3376 __ bfalse(CCR0, LnotFinal);
3377
3378 Register Rscratch = Rflags; // Rflags is dead now.
3379
3380 // Final call case.
3381 __ profile_final_call(Rtemp1, Rscratch);
3382 // Argument and return type profiling.
3383 __ profile_arguments_type(Rindex, Rscratch, Rrecv_klass /* scratch */, true);
3384 // Do the final call - the index (f2) contains the method.
3385 __ call_from_interpreter(Rindex, Rret, Rscratch, Rrecv_klass /* scratch */);
3386
3387 // Non-final callc case.
3388 __ bind(LnotFinal);
3389 __ profile_virtual_call(Rrecv_klass, Rtemp1, Rscratch, false);
3390 generate_vtable_call(Rrecv_klass, Rindex, Rret, Rscratch);
3391 }
3392
3393 void TemplateTable::invokeinterface(int byte_no) {
3394 assert(byte_no == f1_byte, "use this argument");
3395 transition(vtos, vtos);
3396
3397 const Register Rscratch1 = R11_scratch1,
3398 Rscratch2 = R12_scratch2,
3399 Rscratch3 = R9_ARG7,
3400 Rscratch4 = R10_ARG8,
3401 Rtable_addr = Rscratch2,
3402 Rinterface_klass = R5_ARG3,
3403 Rret_type = R8_ARG6,
3404 Rret_addr = Rret_type,
3405 Rindex = R6_ARG4,
3406 Rreceiver = R4_ARG2,
3407 Rrecv_klass = Rreceiver,
3408 Rflags = R7_ARG5;
3409
3410 prepare_invoke(byte_no, Rinterface_klass, Rret_addr, Rindex, Rreceiver, Rflags, Rscratch1);
3411
3412 // Get receiver klass.
3413 __ null_check_throw(Rreceiver, oopDesc::klass_offset_in_bytes(), Rscratch3);
3414 __ load_klass(Rrecv_klass, Rreceiver);
3415
3416 // Check corner case object method.
3417 Label LobjectMethod;
3418
3419 __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_forced_virtual_shift);
3420 __ btrue(CCR0, LobjectMethod);
3421
3422 // Fallthrough: The normal invokeinterface case.
3423 __ profile_virtual_call(Rrecv_klass, Rscratch1, Rscratch2, false);
3424
3425 // Find entry point to call.
3426 Label Lthrow_icc, Lthrow_ame;
3427 // Result will be returned in Rindex.
3428 __ mr(Rscratch4, Rrecv_klass);
3429 __ mr(Rscratch3, Rindex);
3430 __ lookup_interface_method(Rrecv_klass, Rinterface_klass, Rindex, Rindex, Rscratch1, Rscratch2, Lthrow_icc);
3431
3432 __ cmpdi(CCR0, Rindex, 0);
3433 __ beq(CCR0, Lthrow_ame);
3434 // Found entry. Jump off!
3435 // Argument and return type profiling.
3436 __ profile_arguments_type(Rindex, Rscratch1, Rscratch2, true);
3437 __ call_from_interpreter(Rindex, Rret_addr, Rscratch1, Rscratch2);
3438
3439 // Vtable entry was NULL => Throw abstract method error.
3440 __ bind(Lthrow_ame);
3441 __ mr(Rrecv_klass, Rscratch4);
3442 __ mr(Rindex, Rscratch3);
3443 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3444
3445 // Interface was not found => Throw incompatible class change error.
3446 __ bind(Lthrow_icc);
3447 __ mr(Rrecv_klass, Rscratch4);
3448 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeError));
3449
3450 __ should_not_reach_here();
3451
3452 // Special case of invokeinterface called for virtual method of
3453 // java.lang.Object. See ConstantPoolCacheEntry::set_method() for details:
3454 // The invokeinterface was rewritten to a invokevirtual, hence we have
3455 // to handle this corner case. This code isn't produced by javac, but could
3456 // be produced by another compliant java compiler.
3457 __ bind(LobjectMethod);
3458 invokeinterface_object_method(Rrecv_klass, Rret_addr, Rflags, Rindex, Rscratch1, Rscratch2);
3459 }
3460
3461 void TemplateTable::invokedynamic(int byte_no) {
3462 transition(vtos, vtos);
3463
3464 const Register Rret_addr = R3_ARG1,
3465 Rflags = R4_ARG2,
3466 Rmethod = R22_tmp2,
3467 Rscratch1 = R11_scratch1,
3468 Rscratch2 = R12_scratch2;
3469
3470 prepare_invoke(byte_no, Rmethod, Rret_addr, Rscratch1, noreg, Rflags, Rscratch2);
3471
3472 // Profile this call.
3473 __ profile_call(Rscratch1, Rscratch2);
3474
3475 // Off we go. With the new method handles, we don't jump to a method handle
3476 // entry any more. Instead, we pushed an "appendix" in prepare invoke, which happens
3477 // to be the callsite object the bootstrap method returned. This is passed to a
3478 // "link" method which does the dispatch (Most likely just grabs the MH stored
3479 // inside the callsite and does an invokehandle).
3480 // Argument and return type profiling.
3481 __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, false);
3482 __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1 /* scratch1 */, Rscratch2 /* scratch2 */);
3483 }
3484
3485 void TemplateTable::invokehandle(int byte_no) {
3486 transition(vtos, vtos);
3487
3488 const Register Rret_addr = R3_ARG1,
3489 Rflags = R4_ARG2,
3490 Rrecv = R5_ARG3,
3491 Rmethod = R22_tmp2,
3492 Rscratch1 = R11_scratch1,
3493 Rscratch2 = R12_scratch2;
3494
3495 prepare_invoke(byte_no, Rmethod, Rret_addr, Rscratch1, Rrecv, Rflags, Rscratch2);
3496 __ verify_method_ptr(Rmethod);
3497 __ null_check_throw(Rrecv, -1, Rscratch2);
3498
3499 __ profile_final_call(Rrecv, Rscratch1);
3500
3501 // Still no call from handle => We call the method handle interpreter here.
3502 // Argument and return type profiling.
3503 __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, true);
3504 __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1 /* scratch1 */, Rscratch2 /* scratch2 */);
3505 }
3506
3507 // =============================================================================
3508 // Allocation
3509
3510 // Puts allocated obj ref onto the expression stack.
3511 void TemplateTable::_new() {
3512 transition(vtos, atos);
3513
3514 Label Lslow_case,
3515 Ldone,
3516 Linitialize_header,
3517 Lallocate_shared,
3518 Linitialize_object; // Including clearing the fields.
3519
3520 const Register RallocatedObject = R17_tos,
3521 RinstanceKlass = R9_ARG7,
3522 Rscratch = R11_scratch1,
3523 Roffset = R8_ARG6,
3524 Rinstance_size = Roffset,
3525 Rcpool = R4_ARG2,
3526 Rtags = R3_ARG1,
3527 Rindex = R5_ARG3;
3528
3529 const bool allow_shared_alloc = Universe::heap()->supports_inline_contig_alloc();
3530
3531 // --------------------------------------------------------------------------
3532 // Check if fast case is possible.
3533
3534 // Load pointers to const pool and const pool's tags array.
3535 __ get_cpool_and_tags(Rcpool, Rtags);
3536 // Load index of constant pool entry.
3537 __ get_2_byte_integer_at_bcp(1, Rindex, InterpreterMacroAssembler::Unsigned);
3538
3539 if (UseTLAB) {
3540 // Make sure the class we're about to instantiate has been resolved
3541 // This is done before loading instanceKlass to be consistent with the order
3542 // how Constant Pool is updated (see ConstantPoolCache::klass_at_put).
3543 __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3544 __ lbzx(Rtags, Rindex, Rtags);
3545
3546 __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3547 __ bne(CCR0, Lslow_case);
3548
3549 // Get instanceKlass (load from Rcpool + sizeof(ConstantPool) + Rindex*BytesPerWord).
3550 __ sldi(Roffset, Rindex, LogBytesPerWord);
3551 __ addi(Rscratch, Rcpool, sizeof(ConstantPool));
3552 __ isync(); // Order load of instance Klass wrt. tags.
3553 __ ldx(RinstanceKlass, Roffset, Rscratch);
3554
3555 // Make sure klass is fully initialized and get instance_size.
3556 __ lbz(Rscratch, in_bytes(InstanceKlass::init_state_offset()), RinstanceKlass);
3557 __ lwz(Rinstance_size, in_bytes(Klass::layout_helper_offset()), RinstanceKlass);
3558
3559 __ cmpdi(CCR1, Rscratch, InstanceKlass::fully_initialized);
3560 // Make sure klass does not have has_finalizer, or is abstract, or interface or java/lang/Class.
3561 __ andi_(R0, Rinstance_size, Klass::_lh_instance_slow_path_bit); // slow path bit equals 0?
3562
3563 __ crnand(CCR0, Assembler::equal, CCR1, Assembler::equal); // slow path bit set or not fully initialized?
3564 __ beq(CCR0, Lslow_case);
3565
3566 // --------------------------------------------------------------------------
3567 // Fast case:
3568 // Allocate the instance.
3569 // 1) Try to allocate in the TLAB.
3570 // 2) If fail, and the TLAB is not full enough to discard, allocate in the shared Eden.
3571 // 3) If the above fails (or is not applicable), go to a slow case (creates a new TLAB, etc.).
3572
3573 Register RoldTopValue = RallocatedObject; // Object will be allocated here if it fits.
3574 Register RnewTopValue = R6_ARG4;
3575 Register RendValue = R7_ARG5;
3576
3577 // Check if we can allocate in the TLAB.
3578 __ ld(RoldTopValue, in_bytes(JavaThread::tlab_top_offset()), R16_thread);
3579 __ ld(RendValue, in_bytes(JavaThread::tlab_end_offset()), R16_thread);
3580
3581 __ add(RnewTopValue, Rinstance_size, RoldTopValue);
3582
3583 // If there is enough space, we do not CAS and do not clear.
3584 __ cmpld(CCR0, RnewTopValue, RendValue);
3585 __ bgt(CCR0, allow_shared_alloc ? Lallocate_shared : Lslow_case);
3586
3587 __ std(RnewTopValue, in_bytes(JavaThread::tlab_top_offset()), R16_thread);
3588
3589 if (ZeroTLAB) {
3590 // The fields have already been cleared.
3591 __ b(Linitialize_header);
3592 } else {
3593 // Initialize both the header and fields.
3594 __ b(Linitialize_object);
3595 }
3596
3597 // Fall through: TLAB was too small.
3598 if (allow_shared_alloc) {
3599 Register RtlabWasteLimitValue = R10_ARG8;
3600 Register RfreeValue = RnewTopValue;
3601
3602 __ bind(Lallocate_shared);
3603 // Check if tlab should be discarded (refill_waste_limit >= free).
3604 __ ld(RtlabWasteLimitValue, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), R16_thread);
3605 __ subf(RfreeValue, RoldTopValue, RendValue);
3606 __ srdi(RfreeValue, RfreeValue, LogHeapWordSize); // in dwords
3607 __ cmpld(CCR0, RtlabWasteLimitValue, RfreeValue);
3608 __ bge(CCR0, Lslow_case);
3609
3610 // Increment waste limit to prevent getting stuck on this slow path.
3611 __ addi(RtlabWasteLimitValue, RtlabWasteLimitValue, (int)ThreadLocalAllocBuffer::refill_waste_limit_increment());
3612 __ std(RtlabWasteLimitValue, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), R16_thread);
3613 }
3614 // else: No allocation in the shared eden. // fallthru: __ b(Lslow_case);
3615 }
3616 // else: Always go the slow path.
3617
3618 // --------------------------------------------------------------------------
3619 // slow case
3620 __ bind(Lslow_case);
3621 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), Rcpool, Rindex);
3622
3623 if (UseTLAB) {
3624 __ b(Ldone);
3625 // --------------------------------------------------------------------------
3626 // Init1: Zero out newly allocated memory.
3627
3628 if (!ZeroTLAB || allow_shared_alloc) {
3629 // Clear object fields.
3630 __ bind(Linitialize_object);
3631
3632 // Initialize remaining object fields.
3633 Register Rbase = Rtags;
3634 __ addi(Rinstance_size, Rinstance_size, 7 - (int)sizeof(oopDesc));
3635 __ addi(Rbase, RallocatedObject, sizeof(oopDesc));
3636 __ srdi(Rinstance_size, Rinstance_size, 3);
3637
3638 // Clear out object skipping header. Takes also care of the zero length case.
3639 __ clear_memory_doubleword(Rbase, Rinstance_size);
3640 // fallthru: __ b(Linitialize_header);
3641 }
3642
3643 // --------------------------------------------------------------------------
3644 // Init2: Initialize the header: mark, klass
3645 __ bind(Linitialize_header);
3646
3647 // Init mark.
3648 if (UseBiasedLocking) {
3649 __ ld(Rscratch, in_bytes(Klass::prototype_header_offset()), RinstanceKlass);
3650 } else {
3651 __ load_const_optimized(Rscratch, markOopDesc::prototype(), R0);
3652 }
3653 __ std(Rscratch, oopDesc::mark_offset_in_bytes(), RallocatedObject);
3654
3655 // Init klass.
3656 __ store_klass_gap(RallocatedObject);
3657 __ store_klass(RallocatedObject, RinstanceKlass, Rscratch); // klass (last for cms)
3658
3659 // Check and trigger dtrace event.
3660 {
3661 SkipIfEqualZero skip_if(_masm, Rscratch, &DTraceAllocProbes);
3662 __ push(atos);
3663 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc));
3664 __ pop(atos);
3665 }
3666 }
3667
3668 // continue
3669 __ bind(Ldone);
3670
3671 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3672 __ membar(Assembler::StoreStore);
3673 }
3674
3675 void TemplateTable::newarray() {
3676 transition(itos, atos);
3677
3678 __ lbz(R4, 1, R14_bcp);
3679 __ extsw(R5, R17_tos);
3680 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), R4, R5 /* size */);
3681
3682 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3683 __ membar(Assembler::StoreStore);
3684 }
3685
3686 void TemplateTable::anewarray() {
3687 transition(itos, atos);
3688
3689 __ get_constant_pool(R4);
3690 __ get_2_byte_integer_at_bcp(1, R5, InterpreterMacroAssembler::Unsigned);
3691 __ extsw(R6, R17_tos); // size
3692 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), R4 /* pool */, R5 /* index */, R6 /* size */);
3693
3694 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3695 __ membar(Assembler::StoreStore);
3696 }
3697
3698 // Allocate a multi dimensional array
3699 void TemplateTable::multianewarray() {
3700 transition(vtos, atos);
3701
3702 Register Rptr = R31; // Needs to survive C call.
3703
3704 // Put ndims * wordSize into frame temp slot
3705 __ lbz(Rptr, 3, R14_bcp);
3706 __ sldi(Rptr, Rptr, Interpreter::logStackElementSize);
3707 // Esp points past last_dim, so set to R4 to first_dim address.
3708 __ add(R4, Rptr, R15_esp);
3709 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), R4 /* first_size_address */);
3710 // Pop all dimensions off the stack.
3711 __ add(R15_esp, Rptr, R15_esp);
3712
3713 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3714 __ membar(Assembler::StoreStore);
3715 }
3716
3717 void TemplateTable::arraylength() {
3718 transition(atos, itos);
3719
3720 Label LnoException;
3721 __ verify_oop(R17_tos);
3722 __ null_check_throw(R17_tos, arrayOopDesc::length_offset_in_bytes(), R11_scratch1);
3723 __ lwa(R17_tos, arrayOopDesc::length_offset_in_bytes(), R17_tos);
3724 }
3725
3726 // ============================================================================
3727 // Typechecks
3728
3729 void TemplateTable::checkcast() {
3730 transition(atos, atos);
3731
3732 Label Ldone, Lis_null, Lquicked, Lresolved;
3733 Register Roffset = R6_ARG4,
3734 RobjKlass = R4_ARG2,
3735 RspecifiedKlass = R5_ARG3, // Generate_ClassCastException_verbose_handler will read value from this register.
3736 Rcpool = R11_scratch1,
3737 Rtags = R12_scratch2;
3738
3739 // Null does not pass.
3740 __ cmpdi(CCR0, R17_tos, 0);
3741 __ beq(CCR0, Lis_null);
3742
3743 // Get constant pool tag to find out if the bytecode has already been "quickened".
3744 __ get_cpool_and_tags(Rcpool, Rtags);
3745
3746 __ get_2_byte_integer_at_bcp(1, Roffset, InterpreterMacroAssembler::Unsigned);
3747
3748 __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3749 __ lbzx(Rtags, Rtags, Roffset);
3750
3751 __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3752 __ beq(CCR0, Lquicked);
3753
3754 // Call into the VM to "quicken" instanceof.
3755 __ push_ptr(); // for GC
3756 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3757 __ get_vm_result_2(RspecifiedKlass);
3758 __ pop_ptr(); // Restore receiver.
3759 __ b(Lresolved);
3760
3761 // Extract target class from constant pool.
3762 __ bind(Lquicked);
3763 __ sldi(Roffset, Roffset, LogBytesPerWord);
3764 __ addi(Rcpool, Rcpool, sizeof(ConstantPool));
3765 __ isync(); // Order load of specified Klass wrt. tags.
3766 __ ldx(RspecifiedKlass, Rcpool, Roffset);
3767
3768 // Do the checkcast.
3769 __ bind(Lresolved);
3770 // Get value klass in RobjKlass.
3771 __ load_klass(RobjKlass, R17_tos);
3772 // Generate a fast subtype check. Branch to cast_ok if no failure. Return 0 if failure.
3773 __ gen_subtype_check(RobjKlass, RspecifiedKlass, /*3 temp regs*/ Roffset, Rcpool, Rtags, /*target if subtype*/ Ldone);
3774
3775 // Not a subtype; so must throw exception
3776 // Target class oop is in register R6_ARG4 == RspecifiedKlass by convention.
3777 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ClassCastException_entry);
3778 __ mtctr(R11_scratch1);
3779 __ bctr();
3780
3781 // Profile the null case.
3782 __ align(32, 12);
3783 __ bind(Lis_null);
3784 __ profile_null_seen(R11_scratch1, Rtags); // Rtags used as scratch.
3785
3786 __ align(32, 12);
3787 __ bind(Ldone);
3788 }
3789
3790 // Output:
3791 // - tos == 0: Obj was null or not an instance of class.
3792 // - tos == 1: Obj was an instance of class.
3793 void TemplateTable::instanceof() {
3794 transition(atos, itos);
3795
3796 Label Ldone, Lis_null, Lquicked, Lresolved;
3797 Register Roffset = R6_ARG4,
3798 RobjKlass = R4_ARG2,
3799 RspecifiedKlass = R5_ARG3,
3800 Rcpool = R11_scratch1,
3801 Rtags = R12_scratch2;
3802
3803 // Null does not pass.
3804 __ cmpdi(CCR0, R17_tos, 0);
3805 __ beq(CCR0, Lis_null);
3806
3807 // Get constant pool tag to find out if the bytecode has already been "quickened".
3808 __ get_cpool_and_tags(Rcpool, Rtags);
3809
3810 __ get_2_byte_integer_at_bcp(1, Roffset, InterpreterMacroAssembler::Unsigned);
3811
3812 __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3813 __ lbzx(Rtags, Rtags, Roffset);
3814
3815 __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3816 __ beq(CCR0, Lquicked);
3817
3818 // Call into the VM to "quicken" instanceof.
3819 __ push_ptr(); // for GC
3820 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3821 __ get_vm_result_2(RspecifiedKlass);
3822 __ pop_ptr(); // Restore receiver.
3823 __ b(Lresolved);
3824
3825 // Extract target class from constant pool.
3826 __ bind(Lquicked);
3827 __ sldi(Roffset, Roffset, LogBytesPerWord);
3828 __ addi(Rcpool, Rcpool, sizeof(ConstantPool));
3829 __ isync(); // Order load of specified Klass wrt. tags.
3830 __ ldx(RspecifiedKlass, Rcpool, Roffset);
3831
3832 // Do the checkcast.
3833 __ bind(Lresolved);
3834 // Get value klass in RobjKlass.
3835 __ load_klass(RobjKlass, R17_tos);
3836 // Generate a fast subtype check. Branch to cast_ok if no failure. Return 0 if failure.
3837 __ li(R17_tos, 1);
3838 __ gen_subtype_check(RobjKlass, RspecifiedKlass, /*3 temp regs*/ Roffset, Rcpool, Rtags, /*target if subtype*/ Ldone);
3839 __ li(R17_tos, 0);
3840
3841 if (ProfileInterpreter) {
3842 __ b(Ldone);
3843 }
3844
3845 // Profile the null case.
3846 __ align(32, 12);
3847 __ bind(Lis_null);
3848 __ profile_null_seen(Rcpool, Rtags); // Rcpool and Rtags used as scratch.
3849
3850 __ align(32, 12);
3851 __ bind(Ldone);
3852 }
3853
3854 // =============================================================================
3855 // Breakpoints
3856
3857 void TemplateTable::_breakpoint() {
3858 transition(vtos, vtos);
3859
3860 // Get the unpatched byte code.
3861 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), R19_method, R14_bcp);
3862 __ mr(R31, R3_RET);
3863
3864 // Post the breakpoint event.
3865 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), R19_method, R14_bcp);
3866
3867 // Complete the execution of original bytecode.
3868 __ dispatch_Lbyte_code(vtos, R31, Interpreter::normal_table(vtos));
3869 }
3870
3871 // =============================================================================
3872 // Exceptions
3873
3874 void TemplateTable::athrow() {
3875 transition(atos, vtos);
3876
3877 // Exception oop is in tos
3878 __ verify_oop(R17_tos);
3879
3880 __ null_check_throw(R17_tos, -1, R11_scratch1);
3881
3882 // Throw exception interpreter entry expects exception oop to be in R3.
3883 __ mr(R3_RET, R17_tos);
3884 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::throw_exception_entry());
3885 __ mtctr(R11_scratch1);
3886 __ bctr();
3887 }
3888
3889 // =============================================================================
3890 // Synchronization
3891 // Searches the basic object lock list on the stack for a free slot
3892 // and uses it to lock the obect in tos.
3893 //
3894 // Recursive locking is enabled by exiting the search if the same
3895 // object is already found in the list. Thus, a new basic lock obj lock
3896 // is allocated "higher up" in the stack and thus is found first
3897 // at next monitor exit.
3898 void TemplateTable::monitorenter() {
3899 transition(atos, vtos);
3900
3901 __ verify_oop(R17_tos);
3902
3903 Register Rcurrent_monitor = R11_scratch1,
3904 Rcurrent_obj = R12_scratch2,
3905 Robj_to_lock = R17_tos,
3906 Rscratch1 = R3_ARG1,
3907 Rscratch2 = R4_ARG2,
3908 Rscratch3 = R5_ARG3,
3909 Rcurrent_obj_addr = R6_ARG4;
3910
3911 // ------------------------------------------------------------------------------
3912 // Null pointer exception.
3913 __ null_check_throw(Robj_to_lock, -1, R11_scratch1);
3914
3915 // Try to acquire a lock on the object.
3916 // Repeat until succeeded (i.e., until monitorenter returns true).
3917
3918 // ------------------------------------------------------------------------------
3919 // Find a free slot in the monitor block.
3920 Label Lfound, Lexit, Lallocate_new;
3921 ConditionRegister found_free_slot = CCR0,
3922 found_same_obj = CCR1,
3923 reached_limit = CCR6;
3924 {
3925 Label Lloop, Lentry;
3926 Register Rlimit = Rcurrent_monitor;
3927
3928 // Set up search loop - start with topmost monitor.
3929 __ add(Rcurrent_obj_addr, BasicObjectLock::obj_offset_in_bytes(), R26_monitor);
3930
3931 __ ld(Rlimit, 0, R1_SP);
3932 __ addi(Rlimit, Rlimit, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes() - BasicObjectLock::obj_offset_in_bytes())); // Monitor base
3933
3934 // Check if any slot is present => short cut to allocation if not.
3935 __ cmpld(reached_limit, Rcurrent_obj_addr, Rlimit);
3936 __ bgt(reached_limit, Lallocate_new);
3937
3938 // Pre-load topmost slot.
3939 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
3940 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
3941 // The search loop.
3942 __ bind(Lloop);
3943 // Found free slot?
3944 __ cmpdi(found_free_slot, Rcurrent_obj, 0);
3945 // Is this entry for same obj? If so, stop the search and take the found
3946 // free slot or allocate a new one to enable recursive locking.
3947 __ cmpd(found_same_obj, Rcurrent_obj, Robj_to_lock);
3948 __ cmpld(reached_limit, Rcurrent_obj_addr, Rlimit);
3949 __ beq(found_free_slot, Lexit);
3950 __ beq(found_same_obj, Lallocate_new);
3951 __ bgt(reached_limit, Lallocate_new);
3952 // Check if last allocated BasicLockObj reached.
3953 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
3954 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
3955 // Next iteration if unchecked BasicObjectLocks exist on the stack.
3956 __ b(Lloop);
3957 }
3958
3959 // ------------------------------------------------------------------------------
3960 // Check if we found a free slot.
3961 __ bind(Lexit);
3962
3963 __ addi(Rcurrent_monitor, Rcurrent_obj_addr, -(frame::interpreter_frame_monitor_size() * wordSize) - BasicObjectLock::obj_offset_in_bytes());
3964 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, - frame::interpreter_frame_monitor_size() * wordSize);
3965 __ b(Lfound);
3966
3967 // We didn't find a free BasicObjLock => allocate one.
3968 __ align(32, 12);
3969 __ bind(Lallocate_new);
3970 __ add_monitor_to_stack(false, Rscratch1, Rscratch2);
3971 __ mr(Rcurrent_monitor, R26_monitor);
3972 __ addi(Rcurrent_obj_addr, R26_monitor, BasicObjectLock::obj_offset_in_bytes());
3973
3974 // ------------------------------------------------------------------------------
3975 // We now have a slot to lock.
3976 __ bind(Lfound);
3977
3978 // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
3979 // The object has already been poped from the stack, so the expression stack looks correct.
3980 __ addi(R14_bcp, R14_bcp, 1);
3981
3982 __ std(Robj_to_lock, 0, Rcurrent_obj_addr);
3983 __ lock_object(Rcurrent_monitor, Robj_to_lock);
3984
3985 // Check if there's enough space on the stack for the monitors after locking.
3986 Label Lskip_stack_check;
3987 // Optimization: If the monitors stack section is less then a std page size (4K) don't run
3988 // the stack check. There should be enough shadow pages to fit that in.
3989 __ ld(Rscratch3, 0, R1_SP);
3990 __ sub(Rscratch3, Rscratch3, R26_monitor);
3991 __ cmpdi(CCR0, Rscratch3, 4*K);
3992 __ blt(CCR0, Lskip_stack_check);
3993
3994 DEBUG_ONLY(__ untested("stack overflow check during monitor enter");)
3995 __ li(Rscratch1, 0);
3996 __ generate_stack_overflow_check_with_compare_and_throw(Rscratch1, Rscratch2);
3997
3998 __ align(32, 12);
3999 __ bind(Lskip_stack_check);
4000
4001 // The bcp has already been incremented. Just need to dispatch to next instruction.
4002 __ dispatch_next(vtos);
4003 }
4004
4005 void TemplateTable::monitorexit() {
4006 transition(atos, vtos);
4007 __ verify_oop(R17_tos);
4008
4009 Register Rcurrent_monitor = R11_scratch1,
4010 Rcurrent_obj = R12_scratch2,
4011 Robj_to_lock = R17_tos,
4012 Rcurrent_obj_addr = R3_ARG1,
4013 Rlimit = R4_ARG2;
4014 Label Lfound, Lillegal_monitor_state;
4015
4016 // Check corner case: unbalanced monitorEnter / Exit.
4017 __ ld(Rlimit, 0, R1_SP);
4018 __ addi(Rlimit, Rlimit, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base
4019
4020 // Null pointer check.
4021 __ null_check_throw(Robj_to_lock, -1, R11_scratch1);
4022
4023 __ cmpld(CCR0, R26_monitor, Rlimit);
4024 __ bgt(CCR0, Lillegal_monitor_state);
4025
4026 // Find the corresponding slot in the monitors stack section.
4027 {
4028 Label Lloop;
4029
4030 // Start with topmost monitor.
4031 __ addi(Rcurrent_obj_addr, R26_monitor, BasicObjectLock::obj_offset_in_bytes());
4032 __ addi(Rlimit, Rlimit, BasicObjectLock::obj_offset_in_bytes());
4033 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4034 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4035
4036 __ bind(Lloop);
4037 // Is this entry for same obj?
4038 __ cmpd(CCR0, Rcurrent_obj, Robj_to_lock);
4039 __ beq(CCR0, Lfound);
4040
4041 // Check if last allocated BasicLockObj reached.
4042
4043 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4044 __ cmpld(CCR0, Rcurrent_obj_addr, Rlimit);
4045 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4046
4047 // Next iteration if unchecked BasicObjectLocks exist on the stack.
4048 __ ble(CCR0, Lloop);
4049 }
4050
4051 // Fell through without finding the basic obj lock => throw up!
4052 __ bind(Lillegal_monitor_state);
4053 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
4054 __ should_not_reach_here();
4055
4056 __ align(32, 12);
4057 __ bind(Lfound);
4058 __ addi(Rcurrent_monitor, Rcurrent_obj_addr,
4059 -(frame::interpreter_frame_monitor_size() * wordSize) - BasicObjectLock::obj_offset_in_bytes());
4060 __ unlock_object(Rcurrent_monitor);
4061 }
4062
4063 // ============================================================================
4064 // Wide bytecodes
4065
4066 // Wide instructions. Simply redirects to the wide entry point for that instruction.
4067 void TemplateTable::wide() {
4068 transition(vtos, vtos);
4069
4070 const Register Rtable = R11_scratch1,
4071 Rindex = R12_scratch2,
4072 Rtmp = R0;
4073
4074 __ lbz(Rindex, 1, R14_bcp);
4075
4076 __ load_dispatch_table(Rtable, Interpreter::_wentry_point);
4077
4078 __ slwi(Rindex, Rindex, LogBytesPerWord);
4079 __ ldx(Rtmp, Rtable, Rindex);
4080 __ mtctr(Rtmp);
4081 __ bctr();
4082 // Note: the bcp increment step is part of the individual wide bytecode implementations.
4083 }
4084 #endif // !CC_INTERP