1 /*
2 * Copyright (c) 2014, 2018, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2013, 2017 SAP SE. 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 "gc/shared/barrierSetAssembler.hpp"
29 #include "interpreter/interpreter.hpp"
30 #include "interpreter/interpreterRuntime.hpp"
31 #include "interpreter/interp_masm.hpp"
32 #include "interpreter/templateInterpreter.hpp"
33 #include "interpreter/templateTable.hpp"
34 #include "memory/universe.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "oops/oop.inline.hpp"
37 #include "prims/methodHandles.hpp"
38 #include "runtime/frame.inline.hpp"
39 #include "runtime/safepointMechanism.hpp"
40 #include "runtime/sharedRuntime.hpp"
41 #include "runtime/stubRoutines.hpp"
42 #include "runtime/synchronizer.hpp"
43 #include "utilities/macros.hpp"
44
45 #undef __
46 #define __ _masm->
47
48 // ============================================================================
49 // Misc helpers
50
51 // Do an oop store like *(base + index) = val OR *(base + offset) = val
52 // (only one of both variants is possible at the same time).
53 // Index can be noreg.
54 // Kills:
55 // Rbase, Rtmp
56 static void do_oop_store(InterpreterMacroAssembler* _masm,
57 Register base,
58 RegisterOrConstant offset,
59 Register val, // Noreg means always null.
60 Register tmp1,
61 Register tmp2,
62 Register tmp3,
63 DecoratorSet decorators) {
64 assert_different_registers(tmp1, tmp2, tmp3, val, base);
65 __ store_heap_oop(val, offset, base, tmp1, tmp2, tmp3, false, decorators);
66 }
67
68 static void do_oop_load(InterpreterMacroAssembler* _masm,
69 Register base,
70 RegisterOrConstant offset,
71 Register dst,
72 Register tmp1,
73 Register tmp2,
74 DecoratorSet decorators) {
75 assert_different_registers(base, tmp1, tmp2);
76 assert_different_registers(dst, tmp1, tmp2);
77 __ load_heap_oop(dst, offset, base, tmp1, tmp2, false, decorators);
78 }
79
80 // ============================================================================
81 // Platform-dependent initialization
82
83 void TemplateTable::pd_initialize() {
84 // No ppc64 specific initialization.
85 }
86
87 Address TemplateTable::at_bcp(int offset) {
88 // Not used on ppc.
89 ShouldNotReachHere();
90 return Address();
91 }
92
93 // Patches the current bytecode (ptr to it located in bcp)
94 // in the bytecode stream with a new one.
95 void TemplateTable::patch_bytecode(Bytecodes::Code new_bc, Register Rnew_bc, Register Rtemp, bool load_bc_into_bc_reg /*=true*/, int byte_no) {
96 // With sharing on, may need to test method flag.
97 if (!RewriteBytecodes) return;
98 Label L_patch_done;
99
100 switch (new_bc) {
101 case Bytecodes::_fast_aputfield:
102 case Bytecodes::_fast_bputfield:
103 case Bytecodes::_fast_zputfield:
104 case Bytecodes::_fast_cputfield:
105 case Bytecodes::_fast_dputfield:
106 case Bytecodes::_fast_fputfield:
107 case Bytecodes::_fast_iputfield:
108 case Bytecodes::_fast_lputfield:
109 case Bytecodes::_fast_sputfield:
110 {
111 // We skip bytecode quickening for putfield instructions when
112 // the put_code written to the constant pool cache is zero.
113 // This is required so that every execution of this instruction
114 // calls out to InterpreterRuntime::resolve_get_put to do
115 // additional, required work.
116 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
117 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
118 __ get_cache_and_index_at_bcp(Rtemp /* dst = cache */, 1);
119 // ((*(cache+indices))>>((1+byte_no)*8))&0xFF:
120 #if defined(VM_LITTLE_ENDIAN)
121 __ lbz(Rnew_bc, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 1 + byte_no, Rtemp);
122 #else
123 __ lbz(Rnew_bc, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 7 - (1 + byte_no), Rtemp);
124 #endif
125 __ cmpwi(CCR0, Rnew_bc, 0);
126 __ li(Rnew_bc, (unsigned int)(unsigned char)new_bc);
127 __ beq(CCR0, L_patch_done);
128 // __ isync(); // acquire not needed
129 break;
130 }
131
132 default:
133 assert(byte_no == -1, "sanity");
134 if (load_bc_into_bc_reg) {
135 __ li(Rnew_bc, (unsigned int)(unsigned char)new_bc);
136 }
137 }
138
139 if (JvmtiExport::can_post_breakpoint()) {
140 Label L_fast_patch;
141 __ lbz(Rtemp, 0, R14_bcp);
142 __ cmpwi(CCR0, Rtemp, (unsigned int)(unsigned char)Bytecodes::_breakpoint);
143 __ bne(CCR0, L_fast_patch);
144 // Perform the quickening, slowly, in the bowels of the breakpoint table.
145 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), R19_method, R14_bcp, Rnew_bc);
146 __ b(L_patch_done);
147 __ bind(L_fast_patch);
148 }
149
150 // Patch bytecode.
151 __ stb(Rnew_bc, 0, R14_bcp);
152
153 __ bind(L_patch_done);
154 }
155
156 // ============================================================================
157 // Individual instructions
158
159 void TemplateTable::nop() {
160 transition(vtos, vtos);
161 // Nothing to do.
162 }
163
164 void TemplateTable::shouldnotreachhere() {
165 transition(vtos, vtos);
166 __ stop("shouldnotreachhere bytecode");
167 }
168
169 void TemplateTable::aconst_null() {
170 transition(vtos, atos);
171 __ li(R17_tos, 0);
172 }
173
174 void TemplateTable::iconst(int value) {
175 transition(vtos, itos);
176 assert(value >= -1 && value <= 5, "");
177 __ li(R17_tos, value);
178 }
179
180 void TemplateTable::lconst(int value) {
181 transition(vtos, ltos);
182 assert(value >= -1 && value <= 5, "");
183 __ li(R17_tos, value);
184 }
185
186 void TemplateTable::fconst(int value) {
187 transition(vtos, ftos);
188 static float zero = 0.0;
189 static float one = 1.0;
190 static float two = 2.0;
191 switch (value) {
192 default: ShouldNotReachHere();
193 case 0: {
194 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&zero, R0, true);
195 __ lfs(F15_ftos, simm16_offset, R11_scratch1);
196 break;
197 }
198 case 1: {
199 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&one, R0, true);
200 __ lfs(F15_ftos, simm16_offset, R11_scratch1);
201 break;
202 }
203 case 2: {
204 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&two, R0, true);
205 __ lfs(F15_ftos, simm16_offset, R11_scratch1);
206 break;
207 }
208 }
209 }
210
211 void TemplateTable::dconst(int value) {
212 transition(vtos, dtos);
213 static double zero = 0.0;
214 static double one = 1.0;
215 switch (value) {
216 case 0: {
217 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&zero, R0, true);
218 __ lfd(F15_ftos, simm16_offset, R11_scratch1);
219 break;
220 }
221 case 1: {
222 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&one, R0, true);
223 __ lfd(F15_ftos, simm16_offset, R11_scratch1);
224 break;
225 }
226 default: ShouldNotReachHere();
227 }
228 }
229
230 void TemplateTable::bipush() {
231 transition(vtos, itos);
232 __ lbz(R17_tos, 1, R14_bcp);
233 __ extsb(R17_tos, R17_tos);
234 }
235
236 void TemplateTable::sipush() {
237 transition(vtos, itos);
238 __ get_2_byte_integer_at_bcp(1, R17_tos, InterpreterMacroAssembler::Signed);
239 }
240
241 void TemplateTable::ldc(bool wide) {
242 Register Rscratch1 = R11_scratch1,
243 Rscratch2 = R12_scratch2,
244 Rcpool = R3_ARG1;
245
246 transition(vtos, vtos);
247 Label notInt, notFloat, notClass, exit;
248
249 __ get_cpool_and_tags(Rcpool, Rscratch2); // Set Rscratch2 = &tags.
250 if (wide) { // Read index.
251 __ get_2_byte_integer_at_bcp(1, Rscratch1, InterpreterMacroAssembler::Unsigned);
252 } else {
253 __ lbz(Rscratch1, 1, R14_bcp);
254 }
255
256 const int base_offset = ConstantPool::header_size() * wordSize;
257 const int tags_offset = Array<u1>::base_offset_in_bytes();
258
259 // Get type from tags.
260 __ addi(Rscratch2, Rscratch2, tags_offset);
261 __ lbzx(Rscratch2, Rscratch2, Rscratch1);
262
263 __ cmpwi(CCR0, Rscratch2, JVM_CONSTANT_UnresolvedClass); // Unresolved class?
264 __ cmpwi(CCR1, Rscratch2, JVM_CONSTANT_UnresolvedClassInError); // Unresolved class in error state?
265 __ cror(CCR0, Assembler::equal, CCR1, Assembler::equal);
266
267 // Resolved class - need to call vm to get java mirror of the class.
268 __ cmpwi(CCR1, Rscratch2, JVM_CONSTANT_Class);
269 __ crnor(CCR0, Assembler::equal, CCR1, Assembler::equal); // Neither resolved class nor unresolved case from above?
270 __ beq(CCR0, notClass);
271
272 __ li(R4, wide ? 1 : 0);
273 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), R4);
274 __ push(atos);
275 __ b(exit);
276
277 __ align(32, 12);
278 __ bind(notClass);
279 __ addi(Rcpool, Rcpool, base_offset);
280 __ sldi(Rscratch1, Rscratch1, LogBytesPerWord);
281 __ cmpdi(CCR0, Rscratch2, JVM_CONSTANT_Integer);
282 __ bne(CCR0, notInt);
283 __ lwax(R17_tos, Rcpool, Rscratch1);
284 __ push(itos);
285 __ b(exit);
286
287 __ align(32, 12);
288 __ bind(notInt);
289 __ cmpdi(CCR0, Rscratch2, JVM_CONSTANT_Float);
290 __ bne(CCR0, notFloat);
291 __ lfsx(F15_ftos, Rcpool, Rscratch1);
292 __ push(ftos);
293 __ b(exit);
294
295 __ align(32, 12);
296 // assume the tag is for condy; if not, the VM runtime will tell us
297 __ bind(notFloat);
298 condy_helper(exit);
299
300 __ align(32, 12);
301 __ bind(exit);
302 }
303
304 // Fast path for caching oop constants.
305 void TemplateTable::fast_aldc(bool wide) {
306 transition(vtos, atos);
307
308 int index_size = wide ? sizeof(u2) : sizeof(u1);
309 const Register Rscratch = R11_scratch1;
310 Label is_null;
311
312 // We are resolved if the resolved reference cache entry contains a
313 // non-null object (CallSite, etc.)
314 __ get_cache_index_at_bcp(Rscratch, 1, index_size); // Load index.
315 __ load_resolved_reference_at_index(R17_tos, Rscratch, &is_null);
316
317 // Convert null sentinel to NULL.
318 int simm16_rest = __ load_const_optimized(Rscratch, Universe::the_null_sentinel_addr(), R0, true);
319 __ ld(Rscratch, simm16_rest, Rscratch);
320 __ cmpld(CCR0, R17_tos, Rscratch);
321 if (VM_Version::has_isel()) {
322 __ isel_0(R17_tos, CCR0, Assembler::equal);
323 } else {
324 Label not_sentinel;
325 __ bne(CCR0, not_sentinel);
326 __ li(R17_tos, 0);
327 __ bind(not_sentinel);
328 }
329 __ verify_oop(R17_tos);
330 __ dispatch_epilog(atos, Bytecodes::length_for(bytecode()));
331
332 __ bind(is_null);
333 __ load_const_optimized(R3_ARG1, (int)bytecode());
334
335 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
336
337 // First time invocation - must resolve first.
338 __ call_VM(R17_tos, entry, R3_ARG1);
339 __ verify_oop(R17_tos);
340 }
341
342 void TemplateTable::ldc2_w() {
343 transition(vtos, vtos);
344 Label not_double, not_long, exit;
345
346 Register Rindex = R11_scratch1,
347 Rcpool = R12_scratch2,
348 Rtag = R3_ARG1;
349 __ get_cpool_and_tags(Rcpool, Rtag);
350 __ get_2_byte_integer_at_bcp(1, Rindex, InterpreterMacroAssembler::Unsigned);
351
352 const int base_offset = ConstantPool::header_size() * wordSize;
353 const int tags_offset = Array<u1>::base_offset_in_bytes();
354 // Get type from tags.
355 __ addi(Rcpool, Rcpool, base_offset);
356 __ addi(Rtag, Rtag, tags_offset);
357
358 __ lbzx(Rtag, Rtag, Rindex);
359 __ sldi(Rindex, Rindex, LogBytesPerWord);
360
361 __ cmpdi(CCR0, Rtag, JVM_CONSTANT_Double);
362 __ bne(CCR0, not_double);
363 __ lfdx(F15_ftos, Rcpool, Rindex);
364 __ push(dtos);
365 __ b(exit);
366
367 __ bind(not_double);
368 __ cmpdi(CCR0, Rtag, JVM_CONSTANT_Long);
369 __ bne(CCR0, not_long);
370 __ ldx(R17_tos, Rcpool, Rindex);
371 __ push(ltos);
372 __ b(exit);
373
374 __ bind(not_long);
375 condy_helper(exit);
376
377 __ align(32, 12);
378 __ bind(exit);
379 }
380
381 void TemplateTable::condy_helper(Label& Done) {
382 const Register obj = R31;
383 const Register off = R11_scratch1;
384 const Register flags = R12_scratch2;
385 const Register rarg = R4_ARG2;
386 __ li(rarg, (int)bytecode());
387 call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc), rarg);
388 __ get_vm_result_2(flags);
389
390 // VMr = obj = base address to find primitive value to push
391 // VMr2 = flags = (tos, off) using format of CPCE::_flags
392 __ andi(off, flags, ConstantPoolCacheEntry::field_index_mask);
393
394 // What sort of thing are we loading?
395 __ rldicl(flags, flags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
396
397 switch (bytecode()) {
398 case Bytecodes::_ldc:
399 case Bytecodes::_ldc_w:
400 {
401 // tos in (itos, ftos, stos, btos, ctos, ztos)
402 Label notInt, notFloat, notShort, notByte, notChar, notBool;
403 __ cmplwi(CCR0, flags, itos);
404 __ bne(CCR0, notInt);
405 // itos
406 __ lwax(R17_tos, obj, off);
407 __ push(itos);
408 __ b(Done);
409
410 __ bind(notInt);
411 __ cmplwi(CCR0, flags, ftos);
412 __ bne(CCR0, notFloat);
413 // ftos
414 __ lfsx(F15_ftos, obj, off);
415 __ push(ftos);
416 __ b(Done);
417
418 __ bind(notFloat);
419 __ cmplwi(CCR0, flags, stos);
420 __ bne(CCR0, notShort);
421 // stos
422 __ lhax(R17_tos, obj, off);
423 __ push(stos);
424 __ b(Done);
425
426 __ bind(notShort);
427 __ cmplwi(CCR0, flags, btos);
428 __ bne(CCR0, notByte);
429 // btos
430 __ lbzx(R17_tos, obj, off);
431 __ extsb(R17_tos, R17_tos);
432 __ push(btos);
433 __ b(Done);
434
435 __ bind(notByte);
436 __ cmplwi(CCR0, flags, ctos);
437 __ bne(CCR0, notChar);
438 // ctos
439 __ lhzx(R17_tos, obj, off);
440 __ push(ctos);
441 __ b(Done);
442
443 __ bind(notChar);
444 __ cmplwi(CCR0, flags, ztos);
445 __ bne(CCR0, notBool);
446 // ztos
447 __ lbzx(R17_tos, obj, off);
448 __ push(ztos);
449 __ b(Done);
450
451 __ bind(notBool);
452 break;
453 }
454
455 case Bytecodes::_ldc2_w:
456 {
457 Label notLong, notDouble;
458 __ cmplwi(CCR0, flags, ltos);
459 __ bne(CCR0, notLong);
460 // ltos
461 __ ldx(R17_tos, obj, off);
462 __ push(ltos);
463 __ b(Done);
464
465 __ bind(notLong);
466 __ cmplwi(CCR0, flags, dtos);
467 __ bne(CCR0, notDouble);
468 // dtos
469 __ lfdx(F15_ftos, obj, off);
470 __ push(dtos);
471 __ b(Done);
472
473 __ bind(notDouble);
474 break;
475 }
476
477 default:
478 ShouldNotReachHere();
479 }
480
481 __ stop("bad ldc/condy");
482 }
483
484 // Get the locals index located in the bytecode stream at bcp + offset.
485 void TemplateTable::locals_index(Register Rdst, int offset) {
486 __ lbz(Rdst, offset, R14_bcp);
487 }
488
489 void TemplateTable::iload() {
490 iload_internal();
491 }
492
493 void TemplateTable::nofast_iload() {
494 iload_internal(may_not_rewrite);
495 }
496
497 void TemplateTable::iload_internal(RewriteControl rc) {
498 transition(vtos, itos);
499
500 // Get the local value into tos
501 const Register Rindex = R22_tmp2;
502 locals_index(Rindex);
503
504 // Rewrite iload,iload pair into fast_iload2
505 // iload,caload pair into fast_icaload
506 if (RewriteFrequentPairs && rc == may_rewrite) {
507 Label Lrewrite, Ldone;
508 Register Rnext_byte = R3_ARG1,
509 Rrewrite_to = R6_ARG4,
510 Rscratch = R11_scratch1;
511
512 // get next byte
513 __ lbz(Rnext_byte, Bytecodes::length_for(Bytecodes::_iload), R14_bcp);
514
515 // if _iload, wait to rewrite to iload2. We only want to rewrite the
516 // last two iloads in a pair. Comparing against fast_iload means that
517 // the next bytecode is neither an iload or a caload, and therefore
518 // an iload pair.
519 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_iload);
520 __ beq(CCR0, Ldone);
521
522 __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_iload);
523 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iload2);
524 __ beq(CCR1, Lrewrite);
525
526 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_caload);
527 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_icaload);
528 __ beq(CCR0, Lrewrite);
529
530 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iload);
531
532 __ bind(Lrewrite);
533 patch_bytecode(Bytecodes::_iload, Rrewrite_to, Rscratch, false);
534 __ bind(Ldone);
535 }
536
537 __ load_local_int(R17_tos, Rindex, Rindex);
538 }
539
540 // Load 2 integers in a row without dispatching
541 void TemplateTable::fast_iload2() {
542 transition(vtos, itos);
543
544 __ lbz(R3_ARG1, 1, R14_bcp);
545 __ lbz(R17_tos, Bytecodes::length_for(Bytecodes::_iload) + 1, R14_bcp);
546
547 __ load_local_int(R3_ARG1, R11_scratch1, R3_ARG1);
548 __ load_local_int(R17_tos, R12_scratch2, R17_tos);
549 __ push_i(R3_ARG1);
550 }
551
552 void TemplateTable::fast_iload() {
553 transition(vtos, itos);
554 // Get the local value into tos
555
556 const Register Rindex = R11_scratch1;
557 locals_index(Rindex);
558 __ load_local_int(R17_tos, Rindex, Rindex);
559 }
560
561 // Load a local variable type long from locals area to TOS cache register.
562 // Local index resides in bytecodestream.
563 void TemplateTable::lload() {
564 transition(vtos, ltos);
565
566 const Register Rindex = R11_scratch1;
567 locals_index(Rindex);
568 __ load_local_long(R17_tos, Rindex, Rindex);
569 }
570
571 void TemplateTable::fload() {
572 transition(vtos, ftos);
573
574 const Register Rindex = R11_scratch1;
575 locals_index(Rindex);
576 __ load_local_float(F15_ftos, Rindex, Rindex);
577 }
578
579 void TemplateTable::dload() {
580 transition(vtos, dtos);
581
582 const Register Rindex = R11_scratch1;
583 locals_index(Rindex);
584 __ load_local_double(F15_ftos, Rindex, Rindex);
585 }
586
587 void TemplateTable::aload() {
588 transition(vtos, atos);
589
590 const Register Rindex = R11_scratch1;
591 locals_index(Rindex);
592 __ load_local_ptr(R17_tos, Rindex, Rindex);
593 }
594
595 void TemplateTable::locals_index_wide(Register Rdst) {
596 // Offset is 2, not 1, because Lbcp points to wide prefix code.
597 __ get_2_byte_integer_at_bcp(2, Rdst, InterpreterMacroAssembler::Unsigned);
598 }
599
600 void TemplateTable::wide_iload() {
601 // Get the local value into tos.
602
603 const Register Rindex = R11_scratch1;
604 locals_index_wide(Rindex);
605 __ load_local_int(R17_tos, Rindex, Rindex);
606 }
607
608 void TemplateTable::wide_lload() {
609 transition(vtos, ltos);
610
611 const Register Rindex = R11_scratch1;
612 locals_index_wide(Rindex);
613 __ load_local_long(R17_tos, Rindex, Rindex);
614 }
615
616 void TemplateTable::wide_fload() {
617 transition(vtos, ftos);
618
619 const Register Rindex = R11_scratch1;
620 locals_index_wide(Rindex);
621 __ load_local_float(F15_ftos, Rindex, Rindex);
622 }
623
624 void TemplateTable::wide_dload() {
625 transition(vtos, dtos);
626
627 const Register Rindex = R11_scratch1;
628 locals_index_wide(Rindex);
629 __ load_local_double(F15_ftos, Rindex, Rindex);
630 }
631
632 void TemplateTable::wide_aload() {
633 transition(vtos, atos);
634
635 const Register Rindex = R11_scratch1;
636 locals_index_wide(Rindex);
637 __ load_local_ptr(R17_tos, Rindex, Rindex);
638 }
639
640 void TemplateTable::iaload() {
641 transition(itos, itos);
642
643 const Register Rload_addr = R3_ARG1,
644 Rarray = R4_ARG2,
645 Rtemp = R5_ARG3;
646 __ index_check(Rarray, R17_tos /* index */, LogBytesPerInt, Rtemp, Rload_addr);
647 __ lwa(R17_tos, arrayOopDesc::base_offset_in_bytes(T_INT), Rload_addr);
648 }
649
650 void TemplateTable::laload() {
651 transition(itos, ltos);
652
653 const Register Rload_addr = R3_ARG1,
654 Rarray = R4_ARG2,
655 Rtemp = R5_ARG3;
656 __ index_check(Rarray, R17_tos /* index */, LogBytesPerLong, Rtemp, Rload_addr);
657 __ ld(R17_tos, arrayOopDesc::base_offset_in_bytes(T_LONG), Rload_addr);
658 }
659
660 void TemplateTable::faload() {
661 transition(itos, ftos);
662
663 const Register Rload_addr = R3_ARG1,
664 Rarray = R4_ARG2,
665 Rtemp = R5_ARG3;
666 __ index_check(Rarray, R17_tos /* index */, LogBytesPerInt, Rtemp, Rload_addr);
667 __ lfs(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Rload_addr);
668 }
669
670 void TemplateTable::daload() {
671 transition(itos, dtos);
672
673 const Register Rload_addr = R3_ARG1,
674 Rarray = R4_ARG2,
675 Rtemp = R5_ARG3;
676 __ index_check(Rarray, R17_tos /* index */, LogBytesPerLong, Rtemp, Rload_addr);
677 __ lfd(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Rload_addr);
678 }
679
680 void TemplateTable::aaload() {
681 transition(itos, atos);
682
683 // tos: index
684 // result tos: array
685 const Register Rload_addr = R3_ARG1,
686 Rarray = R4_ARG2,
687 Rtemp = R5_ARG3,
688 Rtemp2 = R31;
689 __ index_check(Rarray, R17_tos /* index */, UseCompressedOops ? 2 : LogBytesPerWord, Rtemp, Rload_addr);
690 do_oop_load(_masm, Rload_addr, arrayOopDesc::base_offset_in_bytes(T_OBJECT), R17_tos, Rtemp, Rtemp2,
691 IS_ARRAY);
692 __ verify_oop(R17_tos);
693 //__ dcbt(R17_tos); // prefetch
694 }
695
696 void TemplateTable::baload() {
697 transition(itos, itos);
698
699 const Register Rload_addr = R3_ARG1,
700 Rarray = R4_ARG2,
701 Rtemp = R5_ARG3;
702 __ index_check(Rarray, R17_tos /* index */, 0, Rtemp, Rload_addr);
703 __ lbz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_BYTE), Rload_addr);
704 __ extsb(R17_tos, R17_tos);
705 }
706
707 void TemplateTable::caload() {
708 transition(itos, itos);
709
710 const Register Rload_addr = R3_ARG1,
711 Rarray = R4_ARG2,
712 Rtemp = R5_ARG3;
713 __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
714 __ lhz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rload_addr);
715 }
716
717 // Iload followed by caload frequent pair.
718 void TemplateTable::fast_icaload() {
719 transition(vtos, itos);
720
721 const Register Rload_addr = R3_ARG1,
722 Rarray = R4_ARG2,
723 Rtemp = R11_scratch1;
724
725 locals_index(R17_tos);
726 __ load_local_int(R17_tos, Rtemp, R17_tos);
727 __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
728 __ lhz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rload_addr);
729 }
730
731 void TemplateTable::saload() {
732 transition(itos, itos);
733
734 const Register Rload_addr = R11_scratch1,
735 Rarray = R12_scratch2,
736 Rtemp = R3_ARG1;
737 __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
738 __ lha(R17_tos, arrayOopDesc::base_offset_in_bytes(T_SHORT), Rload_addr);
739 }
740
741 void TemplateTable::iload(int n) {
742 transition(vtos, itos);
743
744 __ lwz(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
745 }
746
747 void TemplateTable::lload(int n) {
748 transition(vtos, ltos);
749
750 __ ld(R17_tos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
751 }
752
753 void TemplateTable::fload(int n) {
754 transition(vtos, ftos);
755
756 __ lfs(F15_ftos, Interpreter::local_offset_in_bytes(n), R18_locals);
757 }
758
759 void TemplateTable::dload(int n) {
760 transition(vtos, dtos);
761
762 __ lfd(F15_ftos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
763 }
764
765 void TemplateTable::aload(int n) {
766 transition(vtos, atos);
767
768 __ ld(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
769 }
770
771 void TemplateTable::aload_0() {
772 aload_0_internal();
773 }
774
775 void TemplateTable::nofast_aload_0() {
776 aload_0_internal(may_not_rewrite);
777 }
778
779 void TemplateTable::aload_0_internal(RewriteControl rc) {
780 transition(vtos, atos);
781 // According to bytecode histograms, the pairs:
782 //
783 // _aload_0, _fast_igetfield
784 // _aload_0, _fast_agetfield
785 // _aload_0, _fast_fgetfield
786 //
787 // occur frequently. If RewriteFrequentPairs is set, the (slow)
788 // _aload_0 bytecode checks if the next bytecode is either
789 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
790 // rewrites the current bytecode into a pair bytecode; otherwise it
791 // rewrites the current bytecode into _0 that doesn't do
792 // the pair check anymore.
793 //
794 // Note: If the next bytecode is _getfield, the rewrite must be
795 // delayed, otherwise we may miss an opportunity for a pair.
796 //
797 // Also rewrite frequent pairs
798 // aload_0, aload_1
799 // aload_0, iload_1
800 // These bytecodes with a small amount of code are most profitable
801 // to rewrite.
802
803 if (RewriteFrequentPairs && rc == may_rewrite) {
804
805 Label Lrewrite, Ldont_rewrite;
806 Register Rnext_byte = R3_ARG1,
807 Rrewrite_to = R6_ARG4,
808 Rscratch = R11_scratch1;
809
810 // Get next byte.
811 __ lbz(Rnext_byte, Bytecodes::length_for(Bytecodes::_aload_0), R14_bcp);
812
813 // If _getfield, wait to rewrite. We only want to rewrite the last two bytecodes in a pair.
814 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_getfield);
815 __ beq(CCR0, Ldont_rewrite);
816
817 __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_igetfield);
818 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iaccess_0);
819 __ beq(CCR1, Lrewrite);
820
821 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_agetfield);
822 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_aaccess_0);
823 __ beq(CCR0, Lrewrite);
824
825 __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_fgetfield);
826 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_faccess_0);
827 __ beq(CCR1, Lrewrite);
828
829 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_aload_0);
830
831 __ bind(Lrewrite);
832 patch_bytecode(Bytecodes::_aload_0, Rrewrite_to, Rscratch, false);
833 __ bind(Ldont_rewrite);
834 }
835
836 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
837 aload(0);
838 }
839
840 void TemplateTable::istore() {
841 transition(itos, vtos);
842
843 const Register Rindex = R11_scratch1;
844 locals_index(Rindex);
845 __ store_local_int(R17_tos, Rindex);
846 }
847
848 void TemplateTable::lstore() {
849 transition(ltos, vtos);
850 const Register Rindex = R11_scratch1;
851 locals_index(Rindex);
852 __ store_local_long(R17_tos, Rindex);
853 }
854
855 void TemplateTable::fstore() {
856 transition(ftos, vtos);
857
858 const Register Rindex = R11_scratch1;
859 locals_index(Rindex);
860 __ store_local_float(F15_ftos, Rindex);
861 }
862
863 void TemplateTable::dstore() {
864 transition(dtos, vtos);
865
866 const Register Rindex = R11_scratch1;
867 locals_index(Rindex);
868 __ store_local_double(F15_ftos, Rindex);
869 }
870
871 void TemplateTable::astore() {
872 transition(vtos, vtos);
873
874 const Register Rindex = R11_scratch1;
875 __ pop_ptr();
876 __ verify_oop_or_return_address(R17_tos, Rindex);
877 locals_index(Rindex);
878 __ store_local_ptr(R17_tos, Rindex);
879 }
880
881 void TemplateTable::wide_istore() {
882 transition(vtos, vtos);
883
884 const Register Rindex = R11_scratch1;
885 __ pop_i();
886 locals_index_wide(Rindex);
887 __ store_local_int(R17_tos, Rindex);
888 }
889
890 void TemplateTable::wide_lstore() {
891 transition(vtos, vtos);
892
893 const Register Rindex = R11_scratch1;
894 __ pop_l();
895 locals_index_wide(Rindex);
896 __ store_local_long(R17_tos, Rindex);
897 }
898
899 void TemplateTable::wide_fstore() {
900 transition(vtos, vtos);
901
902 const Register Rindex = R11_scratch1;
903 __ pop_f();
904 locals_index_wide(Rindex);
905 __ store_local_float(F15_ftos, Rindex);
906 }
907
908 void TemplateTable::wide_dstore() {
909 transition(vtos, vtos);
910
911 const Register Rindex = R11_scratch1;
912 __ pop_d();
913 locals_index_wide(Rindex);
914 __ store_local_double(F15_ftos, Rindex);
915 }
916
917 void TemplateTable::wide_astore() {
918 transition(vtos, vtos);
919
920 const Register Rindex = R11_scratch1;
921 __ pop_ptr();
922 __ verify_oop_or_return_address(R17_tos, Rindex);
923 locals_index_wide(Rindex);
924 __ store_local_ptr(R17_tos, Rindex);
925 }
926
927 void TemplateTable::iastore() {
928 transition(itos, vtos);
929
930 const Register Rindex = R3_ARG1,
931 Rstore_addr = R4_ARG2,
932 Rarray = R5_ARG3,
933 Rtemp = R6_ARG4;
934 __ pop_i(Rindex);
935 __ index_check(Rarray, Rindex, LogBytesPerInt, Rtemp, Rstore_addr);
936 __ stw(R17_tos, arrayOopDesc::base_offset_in_bytes(T_INT), Rstore_addr);
937 }
938
939 void TemplateTable::lastore() {
940 transition(ltos, vtos);
941
942 const Register Rindex = R3_ARG1,
943 Rstore_addr = R4_ARG2,
944 Rarray = R5_ARG3,
945 Rtemp = R6_ARG4;
946 __ pop_i(Rindex);
947 __ index_check(Rarray, Rindex, LogBytesPerLong, Rtemp, Rstore_addr);
948 __ std(R17_tos, arrayOopDesc::base_offset_in_bytes(T_LONG), Rstore_addr);
949 }
950
951 void TemplateTable::fastore() {
952 transition(ftos, vtos);
953
954 const Register Rindex = R3_ARG1,
955 Rstore_addr = R4_ARG2,
956 Rarray = R5_ARG3,
957 Rtemp = R6_ARG4;
958 __ pop_i(Rindex);
959 __ index_check(Rarray, Rindex, LogBytesPerInt, Rtemp, Rstore_addr);
960 __ stfs(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Rstore_addr);
961 }
962
963 void TemplateTable::dastore() {
964 transition(dtos, vtos);
965
966 const Register Rindex = R3_ARG1,
967 Rstore_addr = R4_ARG2,
968 Rarray = R5_ARG3,
969 Rtemp = R6_ARG4;
970 __ pop_i(Rindex);
971 __ index_check(Rarray, Rindex, LogBytesPerLong, Rtemp, Rstore_addr);
972 __ stfd(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Rstore_addr);
973 }
974
975 // Pop 3 values from the stack and...
976 void TemplateTable::aastore() {
977 transition(vtos, vtos);
978
979 Label Lstore_ok, Lis_null, Ldone;
980 const Register Rindex = R3_ARG1,
981 Rarray = R4_ARG2,
982 Rscratch = R11_scratch1,
983 Rscratch2 = R12_scratch2,
984 Rarray_klass = R5_ARG3,
985 Rarray_element_klass = Rarray_klass,
986 Rvalue_klass = R6_ARG4,
987 Rstore_addr = R31; // Use register which survives VM call.
988
989 __ ld(R17_tos, Interpreter::expr_offset_in_bytes(0), R15_esp); // Get value to store.
990 __ lwz(Rindex, Interpreter::expr_offset_in_bytes(1), R15_esp); // Get index.
991 __ ld(Rarray, Interpreter::expr_offset_in_bytes(2), R15_esp); // Get array.
992
993 __ verify_oop(R17_tos);
994 __ index_check_without_pop(Rarray, Rindex, UseCompressedOops ? 2 : LogBytesPerWord, Rscratch, Rstore_addr);
995 // Rindex is dead!
996 Register Rscratch3 = Rindex;
997
998 // Do array store check - check for NULL value first.
999 __ cmpdi(CCR0, R17_tos, 0);
1000 __ beq(CCR0, Lis_null);
1001
1002 __ load_klass(Rarray_klass, Rarray);
1003 __ load_klass(Rvalue_klass, R17_tos);
1004
1005 // Do fast instanceof cache test.
1006 __ ld(Rarray_element_klass, in_bytes(ObjArrayKlass::element_klass_offset()), Rarray_klass);
1007
1008 // Generate a fast subtype check. Branch to store_ok if no failure. Throw if failure.
1009 __ gen_subtype_check(Rvalue_klass /*subklass*/, Rarray_element_klass /*superklass*/, Rscratch, Rscratch2, Rscratch3, Lstore_ok);
1010
1011 // Fell through: subtype check failed => throw an exception.
1012 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArrayStoreException_entry);
1013 __ mtctr(R11_scratch1);
1014 __ bctr();
1015
1016 __ bind(Lis_null);
1017 do_oop_store(_masm, Rstore_addr, arrayOopDesc::base_offset_in_bytes(T_OBJECT), noreg /* 0 */,
1018 Rscratch, Rscratch2, Rscratch3, IS_ARRAY);
1019 __ profile_null_seen(Rscratch, Rscratch2);
1020 __ b(Ldone);
1021
1022 // Store is OK.
1023 __ bind(Lstore_ok);
1024 do_oop_store(_masm, Rstore_addr, arrayOopDesc::base_offset_in_bytes(T_OBJECT), R17_tos /* value */,
1025 Rscratch, Rscratch2, Rscratch3, IS_ARRAY | IS_NOT_NULL);
1026
1027 __ bind(Ldone);
1028 // Adjust sp (pops array, index and value).
1029 __ addi(R15_esp, R15_esp, 3 * Interpreter::stackElementSize);
1030 }
1031
1032 void TemplateTable::bastore() {
1033 transition(itos, vtos);
1034
1035 const Register Rindex = R11_scratch1,
1036 Rarray = R12_scratch2,
1037 Rscratch = R3_ARG1;
1038 __ pop_i(Rindex);
1039 __ pop_ptr(Rarray);
1040 // tos: val
1041
1042 // Need to check whether array is boolean or byte
1043 // since both types share the bastore bytecode.
1044 __ load_klass(Rscratch, Rarray);
1045 __ lwz(Rscratch, in_bytes(Klass::layout_helper_offset()), Rscratch);
1046 int diffbit = exact_log2(Klass::layout_helper_boolean_diffbit());
1047 __ testbitdi(CCR0, R0, Rscratch, diffbit);
1048 Label L_skip;
1049 __ bfalse(CCR0, L_skip);
1050 __ andi(R17_tos, R17_tos, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1
1051 __ bind(L_skip);
1052
1053 __ index_check_without_pop(Rarray, Rindex, 0, Rscratch, Rarray);
1054 __ stb(R17_tos, arrayOopDesc::base_offset_in_bytes(T_BYTE), Rarray);
1055 }
1056
1057 void TemplateTable::castore() {
1058 transition(itos, vtos);
1059
1060 const Register Rindex = R11_scratch1,
1061 Rarray = R12_scratch2,
1062 Rscratch = R3_ARG1;
1063 __ pop_i(Rindex);
1064 // tos: val
1065 // Rarray: array ptr (popped by index_check)
1066 __ index_check(Rarray, Rindex, LogBytesPerShort, Rscratch, Rarray);
1067 __ sth(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rarray);
1068 }
1069
1070 void TemplateTable::sastore() {
1071 castore();
1072 }
1073
1074 void TemplateTable::istore(int n) {
1075 transition(itos, vtos);
1076 __ stw(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
1077 }
1078
1079 void TemplateTable::lstore(int n) {
1080 transition(ltos, vtos);
1081 __ std(R17_tos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
1082 }
1083
1084 void TemplateTable::fstore(int n) {
1085 transition(ftos, vtos);
1086 __ stfs(F15_ftos, Interpreter::local_offset_in_bytes(n), R18_locals);
1087 }
1088
1089 void TemplateTable::dstore(int n) {
1090 transition(dtos, vtos);
1091 __ stfd(F15_ftos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
1092 }
1093
1094 void TemplateTable::astore(int n) {
1095 transition(vtos, vtos);
1096
1097 __ pop_ptr();
1098 __ verify_oop_or_return_address(R17_tos, R11_scratch1);
1099 __ std(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
1100 }
1101
1102 void TemplateTable::pop() {
1103 transition(vtos, vtos);
1104
1105 __ addi(R15_esp, R15_esp, Interpreter::stackElementSize);
1106 }
1107
1108 void TemplateTable::pop2() {
1109 transition(vtos, vtos);
1110
1111 __ addi(R15_esp, R15_esp, Interpreter::stackElementSize * 2);
1112 }
1113
1114 void TemplateTable::dup() {
1115 transition(vtos, vtos);
1116
1117 __ ld(R11_scratch1, Interpreter::stackElementSize, R15_esp);
1118 __ push_ptr(R11_scratch1);
1119 }
1120
1121 void TemplateTable::dup_x1() {
1122 transition(vtos, vtos);
1123
1124 Register Ra = R11_scratch1,
1125 Rb = R12_scratch2;
1126 // stack: ..., a, b
1127 __ ld(Rb, Interpreter::stackElementSize, R15_esp);
1128 __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1129 __ std(Rb, Interpreter::stackElementSize * 2, R15_esp);
1130 __ std(Ra, Interpreter::stackElementSize, R15_esp);
1131 __ push_ptr(Rb);
1132 // stack: ..., b, a, b
1133 }
1134
1135 void TemplateTable::dup_x2() {
1136 transition(vtos, vtos);
1137
1138 Register Ra = R11_scratch1,
1139 Rb = R12_scratch2,
1140 Rc = R3_ARG1;
1141
1142 // stack: ..., a, b, c
1143 __ ld(Rc, Interpreter::stackElementSize, R15_esp); // load c
1144 __ ld(Ra, Interpreter::stackElementSize * 3, R15_esp); // load a
1145 __ std(Rc, Interpreter::stackElementSize * 3, R15_esp); // store c in a
1146 __ ld(Rb, Interpreter::stackElementSize * 2, R15_esp); // load b
1147 // stack: ..., c, b, c
1148 __ std(Ra, Interpreter::stackElementSize * 2, R15_esp); // store a in b
1149 // stack: ..., c, a, c
1150 __ std(Rb, Interpreter::stackElementSize, R15_esp); // store b in c
1151 __ push_ptr(Rc); // push c
1152 // stack: ..., c, a, b, c
1153 }
1154
1155 void TemplateTable::dup2() {
1156 transition(vtos, vtos);
1157
1158 Register Ra = R11_scratch1,
1159 Rb = R12_scratch2;
1160 // stack: ..., a, b
1161 __ ld(Rb, Interpreter::stackElementSize, R15_esp);
1162 __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1163 __ push_2ptrs(Ra, Rb);
1164 // stack: ..., a, b, a, b
1165 }
1166
1167 void TemplateTable::dup2_x1() {
1168 transition(vtos, vtos);
1169
1170 Register Ra = R11_scratch1,
1171 Rb = R12_scratch2,
1172 Rc = R3_ARG1;
1173 // stack: ..., a, b, c
1174 __ ld(Rc, Interpreter::stackElementSize, R15_esp);
1175 __ ld(Rb, Interpreter::stackElementSize * 2, R15_esp);
1176 __ std(Rc, Interpreter::stackElementSize * 2, R15_esp);
1177 __ ld(Ra, Interpreter::stackElementSize * 3, R15_esp);
1178 __ std(Ra, Interpreter::stackElementSize, R15_esp);
1179 __ std(Rb, Interpreter::stackElementSize * 3, R15_esp);
1180 // stack: ..., b, c, a
1181 __ push_2ptrs(Rb, Rc);
1182 // stack: ..., b, c, a, b, c
1183 }
1184
1185 void TemplateTable::dup2_x2() {
1186 transition(vtos, vtos);
1187
1188 Register Ra = R11_scratch1,
1189 Rb = R12_scratch2,
1190 Rc = R3_ARG1,
1191 Rd = R4_ARG2;
1192 // stack: ..., a, b, c, d
1193 __ ld(Rb, Interpreter::stackElementSize * 3, R15_esp);
1194 __ ld(Rd, Interpreter::stackElementSize, R15_esp);
1195 __ std(Rb, Interpreter::stackElementSize, R15_esp); // store b in d
1196 __ std(Rd, Interpreter::stackElementSize * 3, R15_esp); // store d in b
1197 __ ld(Ra, Interpreter::stackElementSize * 4, R15_esp);
1198 __ ld(Rc, Interpreter::stackElementSize * 2, R15_esp);
1199 __ std(Ra, Interpreter::stackElementSize * 2, R15_esp); // store a in c
1200 __ std(Rc, Interpreter::stackElementSize * 4, R15_esp); // store c in a
1201 // stack: ..., c, d, a, b
1202 __ push_2ptrs(Rc, Rd);
1203 // stack: ..., c, d, a, b, c, d
1204 }
1205
1206 void TemplateTable::swap() {
1207 transition(vtos, vtos);
1208 // stack: ..., a, b
1209
1210 Register Ra = R11_scratch1,
1211 Rb = R12_scratch2;
1212 // stack: ..., a, b
1213 __ ld(Rb, Interpreter::stackElementSize, R15_esp);
1214 __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1215 __ std(Rb, Interpreter::stackElementSize * 2, R15_esp);
1216 __ std(Ra, Interpreter::stackElementSize, R15_esp);
1217 // stack: ..., b, a
1218 }
1219
1220 void TemplateTable::iop2(Operation op) {
1221 transition(itos, itos);
1222
1223 Register Rscratch = R11_scratch1;
1224
1225 __ pop_i(Rscratch);
1226 // tos = number of bits to shift
1227 // Rscratch = value to shift
1228 switch (op) {
1229 case add: __ add(R17_tos, Rscratch, R17_tos); break;
1230 case sub: __ sub(R17_tos, Rscratch, R17_tos); break;
1231 case mul: __ mullw(R17_tos, Rscratch, R17_tos); break;
1232 case _and: __ andr(R17_tos, Rscratch, R17_tos); break;
1233 case _or: __ orr(R17_tos, Rscratch, R17_tos); break;
1234 case _xor: __ xorr(R17_tos, Rscratch, R17_tos); break;
1235 case shl: __ rldicl(R17_tos, R17_tos, 0, 64-5); __ slw(R17_tos, Rscratch, R17_tos); break;
1236 case shr: __ rldicl(R17_tos, R17_tos, 0, 64-5); __ sraw(R17_tos, Rscratch, R17_tos); break;
1237 case ushr: __ rldicl(R17_tos, R17_tos, 0, 64-5); __ srw(R17_tos, Rscratch, R17_tos); break;
1238 default: ShouldNotReachHere();
1239 }
1240 }
1241
1242 void TemplateTable::lop2(Operation op) {
1243 transition(ltos, ltos);
1244
1245 Register Rscratch = R11_scratch1;
1246 __ pop_l(Rscratch);
1247 switch (op) {
1248 case add: __ add(R17_tos, Rscratch, R17_tos); break;
1249 case sub: __ sub(R17_tos, Rscratch, R17_tos); break;
1250 case _and: __ andr(R17_tos, Rscratch, R17_tos); break;
1251 case _or: __ orr(R17_tos, Rscratch, R17_tos); break;
1252 case _xor: __ xorr(R17_tos, Rscratch, R17_tos); break;
1253 default: ShouldNotReachHere();
1254 }
1255 }
1256
1257 void TemplateTable::idiv() {
1258 transition(itos, itos);
1259
1260 Label Lnormal, Lexception, Ldone;
1261 Register Rdividend = R11_scratch1; // Used by irem.
1262
1263 __ addi(R0, R17_tos, 1);
1264 __ cmplwi(CCR0, R0, 2);
1265 __ bgt(CCR0, Lnormal); // divisor <-1 or >1
1266
1267 __ cmpwi(CCR1, R17_tos, 0);
1268 __ beq(CCR1, Lexception); // divisor == 0
1269
1270 __ pop_i(Rdividend);
1271 __ mullw(R17_tos, Rdividend, R17_tos); // div by +/-1
1272 __ b(Ldone);
1273
1274 __ bind(Lexception);
1275 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArithmeticException_entry);
1276 __ mtctr(R11_scratch1);
1277 __ bctr();
1278
1279 __ align(32, 12);
1280 __ bind(Lnormal);
1281 __ pop_i(Rdividend);
1282 __ divw(R17_tos, Rdividend, R17_tos); // Can't divide minint/-1.
1283 __ bind(Ldone);
1284 }
1285
1286 void TemplateTable::irem() {
1287 transition(itos, itos);
1288
1289 __ mr(R12_scratch2, R17_tos);
1290 idiv();
1291 __ mullw(R17_tos, R17_tos, R12_scratch2);
1292 __ subf(R17_tos, R17_tos, R11_scratch1); // Dividend set by idiv.
1293 }
1294
1295 void TemplateTable::lmul() {
1296 transition(ltos, ltos);
1297
1298 __ pop_l(R11_scratch1);
1299 __ mulld(R17_tos, R11_scratch1, R17_tos);
1300 }
1301
1302 void TemplateTable::ldiv() {
1303 transition(ltos, ltos);
1304
1305 Label Lnormal, Lexception, Ldone;
1306 Register Rdividend = R11_scratch1; // Used by lrem.
1307
1308 __ addi(R0, R17_tos, 1);
1309 __ cmpldi(CCR0, R0, 2);
1310 __ bgt(CCR0, Lnormal); // divisor <-1 or >1
1311
1312 __ cmpdi(CCR1, R17_tos, 0);
1313 __ beq(CCR1, Lexception); // divisor == 0
1314
1315 __ pop_l(Rdividend);
1316 __ mulld(R17_tos, Rdividend, R17_tos); // div by +/-1
1317 __ b(Ldone);
1318
1319 __ bind(Lexception);
1320 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArithmeticException_entry);
1321 __ mtctr(R11_scratch1);
1322 __ bctr();
1323
1324 __ align(32, 12);
1325 __ bind(Lnormal);
1326 __ pop_l(Rdividend);
1327 __ divd(R17_tos, Rdividend, R17_tos); // Can't divide minint/-1.
1328 __ bind(Ldone);
1329 }
1330
1331 void TemplateTable::lrem() {
1332 transition(ltos, ltos);
1333
1334 __ mr(R12_scratch2, R17_tos);
1335 ldiv();
1336 __ mulld(R17_tos, R17_tos, R12_scratch2);
1337 __ subf(R17_tos, R17_tos, R11_scratch1); // Dividend set by ldiv.
1338 }
1339
1340 void TemplateTable::lshl() {
1341 transition(itos, ltos);
1342
1343 __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1344 __ pop_l(R11_scratch1);
1345 __ sld(R17_tos, R11_scratch1, R17_tos);
1346 }
1347
1348 void TemplateTable::lshr() {
1349 transition(itos, ltos);
1350
1351 __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1352 __ pop_l(R11_scratch1);
1353 __ srad(R17_tos, R11_scratch1, R17_tos);
1354 }
1355
1356 void TemplateTable::lushr() {
1357 transition(itos, ltos);
1358
1359 __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1360 __ pop_l(R11_scratch1);
1361 __ srd(R17_tos, R11_scratch1, R17_tos);
1362 }
1363
1364 void TemplateTable::fop2(Operation op) {
1365 transition(ftos, ftos);
1366
1367 switch (op) {
1368 case add: __ pop_f(F0_SCRATCH); __ fadds(F15_ftos, F0_SCRATCH, F15_ftos); break;
1369 case sub: __ pop_f(F0_SCRATCH); __ fsubs(F15_ftos, F0_SCRATCH, F15_ftos); break;
1370 case mul: __ pop_f(F0_SCRATCH); __ fmuls(F15_ftos, F0_SCRATCH, F15_ftos); break;
1371 case div: __ pop_f(F0_SCRATCH); __ fdivs(F15_ftos, F0_SCRATCH, F15_ftos); break;
1372 case rem:
1373 __ pop_f(F1_ARG1);
1374 __ fmr(F2_ARG2, F15_ftos);
1375 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem));
1376 __ fmr(F15_ftos, F1_RET);
1377 break;
1378
1379 default: ShouldNotReachHere();
1380 }
1381 }
1382
1383 void TemplateTable::dop2(Operation op) {
1384 transition(dtos, dtos);
1385
1386 switch (op) {
1387 case add: __ pop_d(F0_SCRATCH); __ fadd(F15_ftos, F0_SCRATCH, F15_ftos); break;
1388 case sub: __ pop_d(F0_SCRATCH); __ fsub(F15_ftos, F0_SCRATCH, F15_ftos); break;
1389 case mul: __ pop_d(F0_SCRATCH); __ fmul(F15_ftos, F0_SCRATCH, F15_ftos); break;
1390 case div: __ pop_d(F0_SCRATCH); __ fdiv(F15_ftos, F0_SCRATCH, F15_ftos); break;
1391 case rem:
1392 __ pop_d(F1_ARG1);
1393 __ fmr(F2_ARG2, F15_ftos);
1394 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem));
1395 __ fmr(F15_ftos, F1_RET);
1396 break;
1397
1398 default: ShouldNotReachHere();
1399 }
1400 }
1401
1402 // Negate the value in the TOS cache.
1403 void TemplateTable::ineg() {
1404 transition(itos, itos);
1405
1406 __ neg(R17_tos, R17_tos);
1407 }
1408
1409 // Negate the value in the TOS cache.
1410 void TemplateTable::lneg() {
1411 transition(ltos, ltos);
1412
1413 __ neg(R17_tos, R17_tos);
1414 }
1415
1416 void TemplateTable::fneg() {
1417 transition(ftos, ftos);
1418
1419 __ fneg(F15_ftos, F15_ftos);
1420 }
1421
1422 void TemplateTable::dneg() {
1423 transition(dtos, dtos);
1424
1425 __ fneg(F15_ftos, F15_ftos);
1426 }
1427
1428 // Increments a local variable in place.
1429 void TemplateTable::iinc() {
1430 transition(vtos, vtos);
1431
1432 const Register Rindex = R11_scratch1,
1433 Rincrement = R0,
1434 Rvalue = R12_scratch2;
1435
1436 locals_index(Rindex); // Load locals index from bytecode stream.
1437 __ lbz(Rincrement, 2, R14_bcp); // Load increment from the bytecode stream.
1438 __ extsb(Rincrement, Rincrement);
1439
1440 __ load_local_int(Rvalue, Rindex, Rindex); // Puts address of local into Rindex.
1441
1442 __ add(Rvalue, Rincrement, Rvalue);
1443 __ stw(Rvalue, 0, Rindex);
1444 }
1445
1446 void TemplateTable::wide_iinc() {
1447 transition(vtos, vtos);
1448
1449 Register Rindex = R11_scratch1,
1450 Rlocals_addr = Rindex,
1451 Rincr = R12_scratch2;
1452 locals_index_wide(Rindex);
1453 __ get_2_byte_integer_at_bcp(4, Rincr, InterpreterMacroAssembler::Signed);
1454 __ load_local_int(R17_tos, Rlocals_addr, Rindex);
1455 __ add(R17_tos, Rincr, R17_tos);
1456 __ stw(R17_tos, 0, Rlocals_addr);
1457 }
1458
1459 void TemplateTable::convert() {
1460 // %%%%% Factor this first part accross platforms
1461 #ifdef ASSERT
1462 TosState tos_in = ilgl;
1463 TosState tos_out = ilgl;
1464 switch (bytecode()) {
1465 case Bytecodes::_i2l: // fall through
1466 case Bytecodes::_i2f: // fall through
1467 case Bytecodes::_i2d: // fall through
1468 case Bytecodes::_i2b: // fall through
1469 case Bytecodes::_i2c: // fall through
1470 case Bytecodes::_i2s: tos_in = itos; break;
1471 case Bytecodes::_l2i: // fall through
1472 case Bytecodes::_l2f: // fall through
1473 case Bytecodes::_l2d: tos_in = ltos; break;
1474 case Bytecodes::_f2i: // fall through
1475 case Bytecodes::_f2l: // fall through
1476 case Bytecodes::_f2d: tos_in = ftos; break;
1477 case Bytecodes::_d2i: // fall through
1478 case Bytecodes::_d2l: // fall through
1479 case Bytecodes::_d2f: tos_in = dtos; break;
1480 default : ShouldNotReachHere();
1481 }
1482 switch (bytecode()) {
1483 case Bytecodes::_l2i: // fall through
1484 case Bytecodes::_f2i: // fall through
1485 case Bytecodes::_d2i: // fall through
1486 case Bytecodes::_i2b: // fall through
1487 case Bytecodes::_i2c: // fall through
1488 case Bytecodes::_i2s: tos_out = itos; break;
1489 case Bytecodes::_i2l: // fall through
1490 case Bytecodes::_f2l: // fall through
1491 case Bytecodes::_d2l: tos_out = ltos; break;
1492 case Bytecodes::_i2f: // fall through
1493 case Bytecodes::_l2f: // fall through
1494 case Bytecodes::_d2f: tos_out = ftos; break;
1495 case Bytecodes::_i2d: // fall through
1496 case Bytecodes::_l2d: // fall through
1497 case Bytecodes::_f2d: tos_out = dtos; break;
1498 default : ShouldNotReachHere();
1499 }
1500 transition(tos_in, tos_out);
1501 #endif
1502
1503 // Conversion
1504 Label done;
1505 switch (bytecode()) {
1506 case Bytecodes::_i2l:
1507 __ extsw(R17_tos, R17_tos);
1508 break;
1509
1510 case Bytecodes::_l2i:
1511 // Nothing to do, we'll continue to work with the lower bits.
1512 break;
1513
1514 case Bytecodes::_i2b:
1515 __ extsb(R17_tos, R17_tos);
1516 break;
1517
1518 case Bytecodes::_i2c:
1519 __ rldicl(R17_tos, R17_tos, 0, 64-2*8);
1520 break;
1521
1522 case Bytecodes::_i2s:
1523 __ extsh(R17_tos, R17_tos);
1524 break;
1525
1526 case Bytecodes::_i2d:
1527 __ extsw(R17_tos, R17_tos);
1528 case Bytecodes::_l2d:
1529 __ move_l_to_d();
1530 __ fcfid(F15_ftos, F15_ftos);
1531 break;
1532
1533 case Bytecodes::_i2f:
1534 __ extsw(R17_tos, R17_tos);
1535 __ move_l_to_d();
1536 if (VM_Version::has_fcfids()) { // fcfids is >= Power7 only
1537 // Comment: alternatively, load with sign extend could be done by lfiwax.
1538 __ fcfids(F15_ftos, F15_ftos);
1539 } else {
1540 __ fcfid(F15_ftos, F15_ftos);
1541 __ frsp(F15_ftos, F15_ftos);
1542 }
1543 break;
1544
1545 case Bytecodes::_l2f:
1546 if (VM_Version::has_fcfids()) { // fcfids is >= Power7 only
1547 __ move_l_to_d();
1548 __ fcfids(F15_ftos, F15_ftos);
1549 } else {
1550 // Avoid rounding problem when result should be 0x3f800001: need fixup code before fcfid+frsp.
1551 __ mr(R3_ARG1, R17_tos);
1552 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::l2f));
1553 __ fmr(F15_ftos, F1_RET);
1554 }
1555 break;
1556
1557 case Bytecodes::_f2d:
1558 // empty
1559 break;
1560
1561 case Bytecodes::_d2f:
1562 __ frsp(F15_ftos, F15_ftos);
1563 break;
1564
1565 case Bytecodes::_d2i:
1566 case Bytecodes::_f2i:
1567 __ fcmpu(CCR0, F15_ftos, F15_ftos);
1568 __ li(R17_tos, 0); // 0 in case of NAN
1569 __ bso(CCR0, done);
1570 __ fctiwz(F15_ftos, F15_ftos);
1571 __ move_d_to_l();
1572 break;
1573
1574 case Bytecodes::_d2l:
1575 case Bytecodes::_f2l:
1576 __ fcmpu(CCR0, F15_ftos, F15_ftos);
1577 __ li(R17_tos, 0); // 0 in case of NAN
1578 __ bso(CCR0, done);
1579 __ fctidz(F15_ftos, F15_ftos);
1580 __ move_d_to_l();
1581 break;
1582
1583 default: ShouldNotReachHere();
1584 }
1585 __ bind(done);
1586 }
1587
1588 // Long compare
1589 void TemplateTable::lcmp() {
1590 transition(ltos, itos);
1591
1592 const Register Rscratch = R11_scratch1;
1593 __ pop_l(Rscratch); // first operand, deeper in stack
1594
1595 __ cmpd(CCR0, Rscratch, R17_tos); // compare
1596 __ mfcr(R17_tos); // set bit 32..33 as follows: <: 0b10, =: 0b00, >: 0b01
1597 __ srwi(Rscratch, R17_tos, 30);
1598 __ srawi(R17_tos, R17_tos, 31);
1599 __ orr(R17_tos, Rscratch, R17_tos); // set result as follows: <: -1, =: 0, >: 1
1600 }
1601
1602 // fcmpl/fcmpg and dcmpl/dcmpg bytecodes
1603 // unordered_result == -1 => fcmpl or dcmpl
1604 // unordered_result == 1 => fcmpg or dcmpg
1605 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1606 const FloatRegister Rfirst = F0_SCRATCH,
1607 Rsecond = F15_ftos;
1608 const Register Rscratch = R11_scratch1;
1609
1610 if (is_float) {
1611 __ pop_f(Rfirst);
1612 } else {
1613 __ pop_d(Rfirst);
1614 }
1615
1616 Label Lunordered, Ldone;
1617 __ fcmpu(CCR0, Rfirst, Rsecond); // compare
1618 if (unordered_result) {
1619 __ bso(CCR0, Lunordered);
1620 }
1621 __ mfcr(R17_tos); // set bit 32..33 as follows: <: 0b10, =: 0b00, >: 0b01
1622 __ srwi(Rscratch, R17_tos, 30);
1623 __ srawi(R17_tos, R17_tos, 31);
1624 __ orr(R17_tos, Rscratch, R17_tos); // set result as follows: <: -1, =: 0, >: 1
1625 if (unordered_result) {
1626 __ b(Ldone);
1627 __ bind(Lunordered);
1628 __ load_const_optimized(R17_tos, unordered_result);
1629 }
1630 __ bind(Ldone);
1631 }
1632
1633 // Branch_conditional which takes TemplateTable::Condition.
1634 void TemplateTable::branch_conditional(ConditionRegister crx, TemplateTable::Condition cc, Label& L, bool invert) {
1635 bool positive = false;
1636 Assembler::Condition cond = Assembler::equal;
1637 switch (cc) {
1638 case TemplateTable::equal: positive = true ; cond = Assembler::equal ; break;
1639 case TemplateTable::not_equal: positive = false; cond = Assembler::equal ; break;
1640 case TemplateTable::less: positive = true ; cond = Assembler::less ; break;
1641 case TemplateTable::less_equal: positive = false; cond = Assembler::greater; break;
1642 case TemplateTable::greater: positive = true ; cond = Assembler::greater; break;
1643 case TemplateTable::greater_equal: positive = false; cond = Assembler::less ; break;
1644 default: ShouldNotReachHere();
1645 }
1646 int bo = (positive != invert) ? Assembler::bcondCRbiIs1 : Assembler::bcondCRbiIs0;
1647 int bi = Assembler::bi0(crx, cond);
1648 __ bc(bo, bi, L);
1649 }
1650
1651 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1652
1653 // Note: on SPARC, we use InterpreterMacroAssembler::if_cmp also.
1654 __ verify_thread();
1655
1656 const Register Rscratch1 = R11_scratch1,
1657 Rscratch2 = R12_scratch2,
1658 Rscratch3 = R3_ARG1,
1659 R4_counters = R4_ARG2,
1660 bumped_count = R31,
1661 Rdisp = R22_tmp2;
1662
1663 __ profile_taken_branch(Rscratch1, bumped_count);
1664
1665 // Get (wide) offset.
1666 if (is_wide) {
1667 __ get_4_byte_integer_at_bcp(1, Rdisp, InterpreterMacroAssembler::Signed);
1668 } else {
1669 __ get_2_byte_integer_at_bcp(1, Rdisp, InterpreterMacroAssembler::Signed);
1670 }
1671
1672 // --------------------------------------------------------------------------
1673 // Handle all the JSR stuff here, then exit.
1674 // It's much shorter and cleaner than intermingling with the
1675 // non-JSR normal-branch stuff occurring below.
1676 if (is_jsr) {
1677 // Compute return address as bci in Otos_i.
1678 __ ld(Rscratch1, in_bytes(Method::const_offset()), R19_method);
1679 __ addi(Rscratch2, R14_bcp, -in_bytes(ConstMethod::codes_offset()) + (is_wide ? 5 : 3));
1680 __ subf(R17_tos, Rscratch1, Rscratch2);
1681
1682 // Bump bcp to target of JSR.
1683 __ add(R14_bcp, Rdisp, R14_bcp);
1684 // Push returnAddress for "ret" on stack.
1685 __ push_ptr(R17_tos);
1686 // And away we go!
1687 __ dispatch_next(vtos, 0 ,true);
1688 return;
1689 }
1690
1691 // --------------------------------------------------------------------------
1692 // Normal (non-jsr) branch handling
1693
1694 // Bump bytecode pointer by displacement (take the branch).
1695 __ add(R14_bcp, Rdisp, R14_bcp); // Add to bc addr.
1696
1697 const bool increment_invocation_counter_for_backward_branches = UseCompiler && UseLoopCounter;
1698 if (increment_invocation_counter_for_backward_branches) {
1699 Label Lforward;
1700
1701 // Check branch direction.
1702 __ cmpdi(CCR0, Rdisp, 0);
1703 __ bgt(CCR0, Lforward);
1704
1705 __ get_method_counters(R19_method, R4_counters, Lforward);
1706
1707 if (TieredCompilation) {
1708 Label Lno_mdo, Loverflow;
1709 const int increment = InvocationCounter::count_increment;
1710 if (ProfileInterpreter) {
1711 Register Rmdo = Rscratch1;
1712
1713 // If no method data exists, go to profile_continue.
1714 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
1715 __ cmpdi(CCR0, Rmdo, 0);
1716 __ beq(CCR0, Lno_mdo);
1717
1718 // Increment backedge counter in the MDO.
1719 const int mdo_bc_offs = in_bytes(MethodData::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
1720 __ lwz(Rscratch2, mdo_bc_offs, Rmdo);
1721 __ lwz(Rscratch3, in_bytes(MethodData::backedge_mask_offset()), Rmdo);
1722 __ addi(Rscratch2, Rscratch2, increment);
1723 __ stw(Rscratch2, mdo_bc_offs, Rmdo);
1724 if (UseOnStackReplacement) {
1725 __ and_(Rscratch3, Rscratch2, Rscratch3);
1726 __ bne(CCR0, Lforward);
1727 __ b(Loverflow);
1728 } else {
1729 __ b(Lforward);
1730 }
1731 }
1732
1733 // If there's no MDO, increment counter in method.
1734 const int mo_bc_offs = in_bytes(MethodCounters::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
1735 __ bind(Lno_mdo);
1736 __ lwz(Rscratch2, mo_bc_offs, R4_counters);
1737 __ lwz(Rscratch3, in_bytes(MethodCounters::backedge_mask_offset()), R4_counters);
1738 __ addi(Rscratch2, Rscratch2, increment);
1739 __ stw(Rscratch2, mo_bc_offs, R4_counters);
1740 if (UseOnStackReplacement) {
1741 __ and_(Rscratch3, Rscratch2, Rscratch3);
1742 __ bne(CCR0, Lforward);
1743 } else {
1744 __ b(Lforward);
1745 }
1746 __ bind(Loverflow);
1747
1748 // Notify point for loop, pass branch bytecode.
1749 __ subf(R4_ARG2, Rdisp, R14_bcp); // Compute branch bytecode (previous bcp).
1750 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true);
1751
1752 // Was an OSR adapter generated?
1753 __ cmpdi(CCR0, R3_RET, 0);
1754 __ beq(CCR0, Lforward);
1755
1756 // Has the nmethod been invalidated already?
1757 __ lbz(R0, nmethod::state_offset(), R3_RET);
1758 __ cmpwi(CCR0, R0, nmethod::in_use);
1759 __ bne(CCR0, Lforward);
1760
1761 // Migrate the interpreter frame off of the stack.
1762 // We can use all registers because we will not return to interpreter from this point.
1763
1764 // Save nmethod.
1765 const Register osr_nmethod = R31;
1766 __ mr(osr_nmethod, R3_RET);
1767 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R11_scratch1);
1768 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), R16_thread);
1769 __ reset_last_Java_frame();
1770 // OSR buffer is in ARG1.
1771
1772 // Remove the interpreter frame.
1773 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1774
1775 // Jump to the osr code.
1776 __ ld(R11_scratch1, nmethod::osr_entry_point_offset(), osr_nmethod);
1777 __ mtlr(R0);
1778 __ mtctr(R11_scratch1);
1779 __ bctr();
1780
1781 } else {
1782
1783 const Register invoke_ctr = Rscratch1;
1784 // Update Backedge branch separately from invocations.
1785 __ increment_backedge_counter(R4_counters, invoke_ctr, Rscratch2, Rscratch3);
1786
1787 if (ProfileInterpreter) {
1788 __ test_invocation_counter_for_mdp(invoke_ctr, R4_counters, Rscratch2, Lforward);
1789 if (UseOnStackReplacement) {
1790 __ test_backedge_count_for_osr(bumped_count, R4_counters, R14_bcp, Rdisp, Rscratch2);
1791 }
1792 } else {
1793 if (UseOnStackReplacement) {
1794 __ test_backedge_count_for_osr(invoke_ctr, R4_counters, R14_bcp, Rdisp, Rscratch2);
1795 }
1796 }
1797 }
1798
1799 __ bind(Lforward);
1800 }
1801 __ dispatch_next(vtos, 0, true);
1802 }
1803
1804 // Helper function for if_cmp* methods below.
1805 // Factored out common compare and branch code.
1806 void TemplateTable::if_cmp_common(Register Rfirst, Register Rsecond, Register Rscratch1, Register Rscratch2, Condition cc, bool is_jint, bool cmp0) {
1807 Label Lnot_taken;
1808 // Note: The condition code we get is the condition under which we
1809 // *fall through*! So we have to inverse the CC here.
1810
1811 if (is_jint) {
1812 if (cmp0) {
1813 __ cmpwi(CCR0, Rfirst, 0);
1814 } else {
1815 __ cmpw(CCR0, Rfirst, Rsecond);
1816 }
1817 } else {
1818 if (cmp0) {
1819 __ cmpdi(CCR0, Rfirst, 0);
1820 } else {
1821 __ cmpd(CCR0, Rfirst, Rsecond);
1822 }
1823 }
1824 branch_conditional(CCR0, cc, Lnot_taken, /*invert*/ true);
1825
1826 // Conition is false => Jump!
1827 branch(false, false);
1828
1829 // Condition is not true => Continue.
1830 __ align(32, 12);
1831 __ bind(Lnot_taken);
1832 __ profile_not_taken_branch(Rscratch1, Rscratch2);
1833 }
1834
1835 // Compare integer values with zero and fall through if CC holds, branch away otherwise.
1836 void TemplateTable::if_0cmp(Condition cc) {
1837 transition(itos, vtos);
1838
1839 if_cmp_common(R17_tos, noreg, R11_scratch1, R12_scratch2, cc, true, true);
1840 }
1841
1842 // Compare integer values and fall through if CC holds, branch away otherwise.
1843 //
1844 // Interface:
1845 // - Rfirst: First operand (older stack value)
1846 // - tos: Second operand (younger stack value)
1847 void TemplateTable::if_icmp(Condition cc) {
1848 transition(itos, vtos);
1849
1850 const Register Rfirst = R0,
1851 Rsecond = R17_tos;
1852
1853 __ pop_i(Rfirst);
1854 if_cmp_common(Rfirst, Rsecond, R11_scratch1, R12_scratch2, cc, true, false);
1855 }
1856
1857 void TemplateTable::if_nullcmp(Condition cc) {
1858 transition(atos, vtos);
1859
1860 if_cmp_common(R17_tos, noreg, R11_scratch1, R12_scratch2, cc, false, true);
1861 }
1862
1863 void TemplateTable::if_acmp(Condition cc) {
1864 transition(atos, vtos);
1865
1866 const Register Rfirst = R0,
1867 Rsecond = R17_tos;
1868
1869 __ pop_ptr(Rfirst);
1870 if_cmp_common(Rfirst, Rsecond, R11_scratch1, R12_scratch2, cc, false, false);
1871 }
1872
1873 void TemplateTable::ret() {
1874 locals_index(R11_scratch1);
1875 __ load_local_ptr(R17_tos, R11_scratch1, R11_scratch1);
1876
1877 __ profile_ret(vtos, R17_tos, R11_scratch1, R12_scratch2);
1878
1879 __ ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method);
1880 __ add(R11_scratch1, R17_tos, R11_scratch1);
1881 __ addi(R14_bcp, R11_scratch1, in_bytes(ConstMethod::codes_offset()));
1882 __ dispatch_next(vtos, 0, true);
1883 }
1884
1885 void TemplateTable::wide_ret() {
1886 transition(vtos, vtos);
1887
1888 const Register Rindex = R3_ARG1,
1889 Rscratch1 = R11_scratch1,
1890 Rscratch2 = R12_scratch2;
1891
1892 locals_index_wide(Rindex);
1893 __ load_local_ptr(R17_tos, R17_tos, Rindex);
1894 __ profile_ret(vtos, R17_tos, Rscratch1, R12_scratch2);
1895 // Tos now contains the bci, compute the bcp from that.
1896 __ ld(Rscratch1, in_bytes(Method::const_offset()), R19_method);
1897 __ addi(Rscratch2, R17_tos, in_bytes(ConstMethod::codes_offset()));
1898 __ add(R14_bcp, Rscratch1, Rscratch2);
1899 __ dispatch_next(vtos, 0, true);
1900 }
1901
1902 void TemplateTable::tableswitch() {
1903 transition(itos, vtos);
1904
1905 Label Ldispatch, Ldefault_case;
1906 Register Rlow_byte = R3_ARG1,
1907 Rindex = Rlow_byte,
1908 Rhigh_byte = R4_ARG2,
1909 Rdef_offset_addr = R5_ARG3, // is going to contain address of default offset
1910 Rscratch1 = R11_scratch1,
1911 Rscratch2 = R12_scratch2,
1912 Roffset = R6_ARG4;
1913
1914 // Align bcp.
1915 __ addi(Rdef_offset_addr, R14_bcp, BytesPerInt);
1916 __ clrrdi(Rdef_offset_addr, Rdef_offset_addr, log2_long((jlong)BytesPerInt));
1917
1918 // Load lo & hi.
1919 __ get_u4(Rlow_byte, Rdef_offset_addr, BytesPerInt, InterpreterMacroAssembler::Unsigned);
1920 __ get_u4(Rhigh_byte, Rdef_offset_addr, 2 *BytesPerInt, InterpreterMacroAssembler::Unsigned);
1921
1922 // Check for default case (=index outside [low,high]).
1923 __ cmpw(CCR0, R17_tos, Rlow_byte);
1924 __ cmpw(CCR1, R17_tos, Rhigh_byte);
1925 __ blt(CCR0, Ldefault_case);
1926 __ bgt(CCR1, Ldefault_case);
1927
1928 // Lookup dispatch offset.
1929 __ sub(Rindex, R17_tos, Rlow_byte);
1930 __ extsw(Rindex, Rindex);
1931 __ profile_switch_case(Rindex, Rhigh_byte /* scratch */, Rscratch1, Rscratch2);
1932 __ sldi(Rindex, Rindex, LogBytesPerInt);
1933 __ addi(Rindex, Rindex, 3 * BytesPerInt);
1934 #if defined(VM_LITTLE_ENDIAN)
1935 __ lwbrx(Roffset, Rdef_offset_addr, Rindex);
1936 __ extsw(Roffset, Roffset);
1937 #else
1938 __ lwax(Roffset, Rdef_offset_addr, Rindex);
1939 #endif
1940 __ b(Ldispatch);
1941
1942 __ bind(Ldefault_case);
1943 __ profile_switch_default(Rhigh_byte, Rscratch1);
1944 __ get_u4(Roffset, Rdef_offset_addr, 0, InterpreterMacroAssembler::Signed);
1945
1946 __ bind(Ldispatch);
1947
1948 __ add(R14_bcp, Roffset, R14_bcp);
1949 __ dispatch_next(vtos, 0, true);
1950 }
1951
1952 void TemplateTable::lookupswitch() {
1953 transition(itos, itos);
1954 __ stop("lookupswitch bytecode should have been rewritten");
1955 }
1956
1957 // Table switch using linear search through cases.
1958 // Bytecode stream format:
1959 // Bytecode (1) | 4-byte padding | default offset (4) | count (4) | value/offset pair1 (8) | value/offset pair2 (8) | ...
1960 // Note: Everything is big-endian format here.
1961 void TemplateTable::fast_linearswitch() {
1962 transition(itos, vtos);
1963
1964 Label Lloop_entry, Lsearch_loop, Lcontinue_execution, Ldefault_case;
1965 Register Rcount = R3_ARG1,
1966 Rcurrent_pair = R4_ARG2,
1967 Rdef_offset_addr = R5_ARG3, // Is going to contain address of default offset.
1968 Roffset = R31, // Might need to survive C call.
1969 Rvalue = R12_scratch2,
1970 Rscratch = R11_scratch1,
1971 Rcmp_value = R17_tos;
1972
1973 // Align bcp.
1974 __ addi(Rdef_offset_addr, R14_bcp, BytesPerInt);
1975 __ clrrdi(Rdef_offset_addr, Rdef_offset_addr, log2_long((jlong)BytesPerInt));
1976
1977 // Setup loop counter and limit.
1978 __ get_u4(Rcount, Rdef_offset_addr, BytesPerInt, InterpreterMacroAssembler::Unsigned);
1979 __ addi(Rcurrent_pair, Rdef_offset_addr, 2 * BytesPerInt); // Rcurrent_pair now points to first pair.
1980
1981 __ mtctr(Rcount);
1982 __ cmpwi(CCR0, Rcount, 0);
1983 __ bne(CCR0, Lloop_entry);
1984
1985 // Default case
1986 __ bind(Ldefault_case);
1987 __ get_u4(Roffset, Rdef_offset_addr, 0, InterpreterMacroAssembler::Signed);
1988 if (ProfileInterpreter) {
1989 __ profile_switch_default(Rdef_offset_addr, Rcount/* scratch */);
1990 }
1991 __ b(Lcontinue_execution);
1992
1993 // Next iteration
1994 __ bind(Lsearch_loop);
1995 __ bdz(Ldefault_case);
1996 __ addi(Rcurrent_pair, Rcurrent_pair, 2 * BytesPerInt);
1997 __ bind(Lloop_entry);
1998 __ get_u4(Rvalue, Rcurrent_pair, 0, InterpreterMacroAssembler::Unsigned);
1999 __ cmpw(CCR0, Rvalue, Rcmp_value);
2000 __ bne(CCR0, Lsearch_loop);
2001
2002 // Found, load offset.
2003 __ get_u4(Roffset, Rcurrent_pair, BytesPerInt, InterpreterMacroAssembler::Signed);
2004 // Calculate case index and profile
2005 __ mfctr(Rcurrent_pair);
2006 if (ProfileInterpreter) {
2007 __ sub(Rcurrent_pair, Rcount, Rcurrent_pair);
2008 __ profile_switch_case(Rcurrent_pair, Rcount /*scratch*/, Rdef_offset_addr/*scratch*/, Rscratch);
2009 }
2010
2011 __ bind(Lcontinue_execution);
2012 __ add(R14_bcp, Roffset, R14_bcp);
2013 __ dispatch_next(vtos, 0, true);
2014 }
2015
2016 // Table switch using binary search (value/offset pairs are ordered).
2017 // Bytecode stream format:
2018 // Bytecode (1) | 4-byte padding | default offset (4) | count (4) | value/offset pair1 (8) | value/offset pair2 (8) | ...
2019 // Note: Everything is big-endian format here. So on little endian machines, we have to revers offset and count and cmp value.
2020 void TemplateTable::fast_binaryswitch() {
2021
2022 transition(itos, vtos);
2023 // Implementation using the following core algorithm: (copied from Intel)
2024 //
2025 // int binary_search(int key, LookupswitchPair* array, int n) {
2026 // // Binary search according to "Methodik des Programmierens" by
2027 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2028 // int i = 0;
2029 // int j = n;
2030 // while (i+1 < j) {
2031 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2032 // // with Q: for all i: 0 <= i < n: key < a[i]
2033 // // where a stands for the array and assuming that the (inexisting)
2034 // // element a[n] is infinitely big.
2035 // int h = (i + j) >> 1;
2036 // // i < h < j
2037 // if (key < array[h].fast_match()) {
2038 // j = h;
2039 // } else {
2040 // i = h;
2041 // }
2042 // }
2043 // // R: a[i] <= key < a[i+1] or Q
2044 // // (i.e., if key is within array, i is the correct index)
2045 // return i;
2046 // }
2047
2048 // register allocation
2049 const Register Rkey = R17_tos; // already set (tosca)
2050 const Register Rarray = R3_ARG1;
2051 const Register Ri = R4_ARG2;
2052 const Register Rj = R5_ARG3;
2053 const Register Rh = R6_ARG4;
2054 const Register Rscratch = R11_scratch1;
2055
2056 const int log_entry_size = 3;
2057 const int entry_size = 1 << log_entry_size;
2058
2059 Label found;
2060
2061 // Find Array start,
2062 __ addi(Rarray, R14_bcp, 3 * BytesPerInt);
2063 __ clrrdi(Rarray, Rarray, log2_long((jlong)BytesPerInt));
2064
2065 // initialize i & j
2066 __ li(Ri,0);
2067 __ get_u4(Rj, Rarray, -BytesPerInt, InterpreterMacroAssembler::Unsigned);
2068
2069 // and start.
2070 Label entry;
2071 __ b(entry);
2072
2073 // binary search loop
2074 { Label loop;
2075 __ bind(loop);
2076 // int h = (i + j) >> 1;
2077 __ srdi(Rh, Rh, 1);
2078 // if (key < array[h].fast_match()) {
2079 // j = h;
2080 // } else {
2081 // i = h;
2082 // }
2083 __ sldi(Rscratch, Rh, log_entry_size);
2084 #if defined(VM_LITTLE_ENDIAN)
2085 __ lwbrx(Rscratch, Rscratch, Rarray);
2086 #else
2087 __ lwzx(Rscratch, Rscratch, Rarray);
2088 #endif
2089
2090 // if (key < current value)
2091 // Rh = Rj
2092 // else
2093 // Rh = Ri
2094 Label Lgreater;
2095 __ cmpw(CCR0, Rkey, Rscratch);
2096 __ bge(CCR0, Lgreater);
2097 __ mr(Rj, Rh);
2098 __ b(entry);
2099 __ bind(Lgreater);
2100 __ mr(Ri, Rh);
2101
2102 // while (i+1 < j)
2103 __ bind(entry);
2104 __ addi(Rscratch, Ri, 1);
2105 __ cmpw(CCR0, Rscratch, Rj);
2106 __ add(Rh, Ri, Rj); // start h = i + j >> 1;
2107
2108 __ blt(CCR0, loop);
2109 }
2110
2111 // End of binary search, result index is i (must check again!).
2112 Label default_case;
2113 Label continue_execution;
2114 if (ProfileInterpreter) {
2115 __ mr(Rh, Ri); // Save index in i for profiling.
2116 }
2117 // Ri = value offset
2118 __ sldi(Ri, Ri, log_entry_size);
2119 __ add(Ri, Ri, Rarray);
2120 __ get_u4(Rscratch, Ri, 0, InterpreterMacroAssembler::Unsigned);
2121
2122 Label not_found;
2123 // Ri = offset offset
2124 __ cmpw(CCR0, Rkey, Rscratch);
2125 __ beq(CCR0, not_found);
2126 // entry not found -> j = default offset
2127 __ get_u4(Rj, Rarray, -2 * BytesPerInt, InterpreterMacroAssembler::Unsigned);
2128 __ b(default_case);
2129
2130 __ bind(not_found);
2131 // entry found -> j = offset
2132 __ profile_switch_case(Rh, Rj, Rscratch, Rkey);
2133 __ get_u4(Rj, Ri, BytesPerInt, InterpreterMacroAssembler::Unsigned);
2134
2135 if (ProfileInterpreter) {
2136 __ b(continue_execution);
2137 }
2138
2139 __ bind(default_case); // fall through (if not profiling)
2140 __ profile_switch_default(Ri, Rscratch);
2141
2142 __ bind(continue_execution);
2143
2144 __ extsw(Rj, Rj);
2145 __ add(R14_bcp, Rj, R14_bcp);
2146 __ dispatch_next(vtos, 0 , true);
2147 }
2148
2149 void TemplateTable::_return(TosState state) {
2150 transition(state, state);
2151 assert(_desc->calls_vm(),
2152 "inconsistent calls_vm information"); // call in remove_activation
2153
2154 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2155
2156 Register Rscratch = R11_scratch1,
2157 Rklass = R12_scratch2,
2158 Rklass_flags = Rklass;
2159 Label Lskip_register_finalizer;
2160
2161 // Check if the method has the FINALIZER flag set and call into the VM to finalize in this case.
2162 assert(state == vtos, "only valid state");
2163 __ ld(R17_tos, 0, R18_locals);
2164
2165 // Load klass of this obj.
2166 __ load_klass(Rklass, R17_tos);
2167 __ lwz(Rklass_flags, in_bytes(Klass::access_flags_offset()), Rklass);
2168 __ testbitdi(CCR0, R0, Rklass_flags, exact_log2(JVM_ACC_HAS_FINALIZER));
2169 __ bfalse(CCR0, Lskip_register_finalizer);
2170
2171 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), R17_tos /* obj */);
2172
2173 __ align(32, 12);
2174 __ bind(Lskip_register_finalizer);
2175 }
2176
2177 if (SafepointMechanism::uses_thread_local_poll() && _desc->bytecode() != Bytecodes::_return_register_finalizer) {
2178 Label no_safepoint;
2179 __ ld(R11_scratch1, in_bytes(Thread::polling_page_offset()), R16_thread);
2180 __ andi_(R11_scratch1, R11_scratch1, SafepointMechanism::poll_bit());
2181 __ beq(CCR0, no_safepoint);
2182 __ push(state);
2183 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::at_safepoint));
2184 __ pop(state);
2185 __ bind(no_safepoint);
2186 }
2187
2188 // Move the result value into the correct register and remove memory stack frame.
2189 __ remove_activation(state, /* throw_monitor_exception */ true);
2190 // Restoration of lr done by remove_activation.
2191 switch (state) {
2192 // Narrow result if state is itos but result type is smaller.
2193 // Need to narrow in the return bytecode rather than in generate_return_entry
2194 // since compiled code callers expect the result to already be narrowed.
2195 case itos: __ narrow(R17_tos); /* fall through */
2196 case ltos:
2197 case atos: __ mr(R3_RET, R17_tos); break;
2198 case ftos:
2199 case dtos: __ fmr(F1_RET, F15_ftos); break;
2200 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
2201 // to get visible before the reference to the object gets stored anywhere.
2202 __ membar(Assembler::StoreStore); break;
2203 default : ShouldNotReachHere();
2204 }
2205 __ blr();
2206 }
2207
2208 // ============================================================================
2209 // Constant pool cache access
2210 //
2211 // Memory ordering:
2212 //
2213 // Like done in C++ interpreter, we load the fields
2214 // - _indices
2215 // - _f12_oop
2216 // acquired, because these are asked if the cache is already resolved. We don't
2217 // want to float loads above this check.
2218 // See also comments in ConstantPoolCacheEntry::bytecode_1(),
2219 // ConstantPoolCacheEntry::bytecode_2() and ConstantPoolCacheEntry::f1();
2220
2221 // Call into the VM if call site is not yet resolved
2222 //
2223 // Input regs:
2224 // - None, all passed regs are outputs.
2225 //
2226 // Returns:
2227 // - Rcache: The const pool cache entry that contains the resolved result.
2228 // - Rresult: Either noreg or output for f1/f2.
2229 //
2230 // Kills:
2231 // - Rscratch
2232 void TemplateTable::resolve_cache_and_index(int byte_no, Register Rcache, Register Rscratch, size_t index_size) {
2233
2234 __ get_cache_and_index_at_bcp(Rcache, 1, index_size);
2235 Label Lresolved, Ldone, L_clinit_barrier_slow;
2236
2237 Bytecodes::Code code = bytecode();
2238 switch (code) {
2239 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2240 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2241 default:
2242 break;
2243 }
2244
2245 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2246 // We are resolved if the indices offset contains the current bytecode.
2247 #if defined(VM_LITTLE_ENDIAN)
2248 __ lbz(Rscratch, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + byte_no + 1, Rcache);
2249 #else
2250 __ lbz(Rscratch, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 7 - (byte_no + 1), Rcache);
2251 #endif
2252 // Acquire by cmp-br-isync (see below).
2253 __ cmpdi(CCR0, Rscratch, (int)code);
2254 __ beq(CCR0, Lresolved);
2255
2256 // Class initialization barrier slow path lands here as well.
2257 __ bind(L_clinit_barrier_slow);
2258
2259 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2260 __ li(R4_ARG2, code);
2261 __ call_VM(noreg, entry, R4_ARG2, true);
2262
2263 // Update registers with resolved info.
2264 __ get_cache_and_index_at_bcp(Rcache, 1, index_size);
2265 __ b(Ldone);
2266
2267 __ bind(Lresolved);
2268 __ isync(); // Order load wrt. succeeding loads.
2269
2270 // Class initialization barrier for static methods
2271 if (VM_Version::supports_fast_class_init_checks() && bytecode() == Bytecodes::_invokestatic) {
2272 const Register method = Rscratch;
2273 const Register klass = Rscratch;
2274
2275 __ load_resolved_method_at_index(byte_no, Rcache, method);
2276 __ load_method_holder(klass, method);
2277 __ clinit_barrier(klass, R16_thread, NULL /*L_fast_path*/, &L_clinit_barrier_slow);
2278 }
2279
2280 __ bind(Ldone);
2281 }
2282
2283 // Load the constant pool cache entry at field accesses into registers.
2284 // The Rcache and Rindex registers must be set before call.
2285 // Input:
2286 // - Rcache, Rindex
2287 // Output:
2288 // - Robj, Roffset, Rflags
2289 void TemplateTable::load_field_cp_cache_entry(Register Robj,
2290 Register Rcache,
2291 Register Rindex /* unused on PPC64 */,
2292 Register Roffset,
2293 Register Rflags,
2294 bool is_static = false) {
2295 assert_different_registers(Rcache, Rflags, Roffset);
2296 // assert(Rindex == noreg, "parameter not used on PPC64");
2297
2298 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2299 __ ld(Rflags, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::flags_offset()), Rcache);
2300 __ ld(Roffset, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::f2_offset()), Rcache);
2301 if (is_static) {
2302 __ ld(Robj, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::f1_offset()), Rcache);
2303 __ ld(Robj, in_bytes(Klass::java_mirror_offset()), Robj);
2304 __ resolve_oop_handle(Robj);
2305 // Acquire not needed here. Following access has an address dependency on this value.
2306 }
2307 }
2308
2309 // Load the constant pool cache entry at invokes into registers.
2310 // Resolve if necessary.
2311
2312 // Input Registers:
2313 // - None, bcp is used, though
2314 //
2315 // Return registers:
2316 // - Rmethod (f1 field or f2 if invokevirtual)
2317 // - Ritable_index (f2 field)
2318 // - Rflags (flags field)
2319 //
2320 // Kills:
2321 // - R21
2322 //
2323 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2324 Register Rmethod,
2325 Register Ritable_index,
2326 Register Rflags,
2327 bool is_invokevirtual,
2328 bool is_invokevfinal,
2329 bool is_invokedynamic) {
2330
2331 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2332 // Determine constant pool cache field offsets.
2333 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2334 const int method_offset = in_bytes(cp_base_offset + (is_invokevirtual ? ConstantPoolCacheEntry::f2_offset() : ConstantPoolCacheEntry::f1_offset()));
2335 const int flags_offset = in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset());
2336 // Access constant pool cache fields.
2337 const int index_offset = in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset());
2338
2339 Register Rcache = R21_tmp1; // Note: same register as R21_sender_SP.
2340
2341 if (is_invokevfinal) {
2342 assert(Ritable_index == noreg, "register not used");
2343 // Already resolved.
2344 __ get_cache_and_index_at_bcp(Rcache, 1);
2345 } else {
2346 resolve_cache_and_index(byte_no, Rcache, /* temp */ Rmethod, is_invokedynamic ? sizeof(u4) : sizeof(u2));
2347 }
2348
2349 __ ld(Rmethod, method_offset, Rcache);
2350 __ ld(Rflags, flags_offset, Rcache);
2351
2352 if (Ritable_index != noreg) {
2353 __ ld(Ritable_index, index_offset, Rcache);
2354 }
2355 }
2356
2357 // ============================================================================
2358 // Field access
2359
2360 // Volatile variables demand their effects be made known to all CPU's
2361 // in order. Store buffers on most chips allow reads & writes to
2362 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2363 // without some kind of memory barrier (i.e., it's not sufficient that
2364 // the interpreter does not reorder volatile references, the hardware
2365 // also must not reorder them).
2366 //
2367 // According to the new Java Memory Model (JMM):
2368 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2369 // writes act as aquire & release, so:
2370 // (2) A read cannot let unrelated NON-volatile memory refs that
2371 // happen after the read float up to before the read. It's OK for
2372 // non-volatile memory refs that happen before the volatile read to
2373 // float down below it.
2374 // (3) Similar a volatile write cannot let unrelated NON-volatile
2375 // memory refs that happen BEFORE the write float down to after the
2376 // write. It's OK for non-volatile memory refs that happen after the
2377 // volatile write to float up before it.
2378 //
2379 // We only put in barriers around volatile refs (they are expensive),
2380 // not _between_ memory refs (that would require us to track the
2381 // flavor of the previous memory refs). Requirements (2) and (3)
2382 // require some barriers before volatile stores and after volatile
2383 // loads. These nearly cover requirement (1) but miss the
2384 // volatile-store-volatile-load case. This final case is placed after
2385 // volatile-stores although it could just as well go before
2386 // volatile-loads.
2387
2388 // The registers cache and index expected to be set before call.
2389 // Correct values of the cache and index registers are preserved.
2390 // Kills:
2391 // Rcache (if has_tos)
2392 // Rscratch
2393 void TemplateTable::jvmti_post_field_access(Register Rcache, Register Rscratch, bool is_static, bool has_tos) {
2394
2395 assert_different_registers(Rcache, Rscratch);
2396
2397 if (JvmtiExport::can_post_field_access()) {
2398 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2399 Label Lno_field_access_post;
2400
2401 // Check if post field access in enabled.
2402 int offs = __ load_const_optimized(Rscratch, JvmtiExport::get_field_access_count_addr(), R0, true);
2403 __ lwz(Rscratch, offs, Rscratch);
2404
2405 __ cmpwi(CCR0, Rscratch, 0);
2406 __ beq(CCR0, Lno_field_access_post);
2407
2408 // Post access enabled - do it!
2409 __ addi(Rcache, Rcache, in_bytes(cp_base_offset));
2410 if (is_static) {
2411 __ li(R17_tos, 0);
2412 } else {
2413 if (has_tos) {
2414 // The fast bytecode versions have obj ptr in register.
2415 // Thus, save object pointer before call_VM() clobbers it
2416 // put object on tos where GC wants it.
2417 __ push_ptr(R17_tos);
2418 } else {
2419 // Load top of stack (do not pop the value off the stack).
2420 __ ld(R17_tos, Interpreter::expr_offset_in_bytes(0), R15_esp);
2421 }
2422 __ verify_oop(R17_tos);
2423 }
2424 // tos: object pointer or NULL if static
2425 // cache: cache entry pointer
2426 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), R17_tos, Rcache);
2427 if (!is_static && has_tos) {
2428 // Restore object pointer.
2429 __ pop_ptr(R17_tos);
2430 __ verify_oop(R17_tos);
2431 } else {
2432 // Cache is still needed to get class or obj.
2433 __ get_cache_and_index_at_bcp(Rcache, 1);
2434 }
2435
2436 __ align(32, 12);
2437 __ bind(Lno_field_access_post);
2438 }
2439 }
2440
2441 // kills R11_scratch1
2442 void TemplateTable::pop_and_check_object(Register Roop) {
2443 Register Rtmp = R11_scratch1;
2444
2445 assert_different_registers(Rtmp, Roop);
2446 __ pop_ptr(Roop);
2447 // For field access must check obj.
2448 __ null_check_throw(Roop, -1, Rtmp);
2449 __ verify_oop(Roop);
2450 }
2451
2452 // PPC64: implement volatile loads as fence-store-acquire.
2453 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2454 transition(vtos, vtos);
2455
2456 Label Lacquire, Lisync;
2457
2458 const Register Rcache = R3_ARG1,
2459 Rclass_or_obj = R22_tmp2,
2460 Roffset = R23_tmp3,
2461 Rflags = R31,
2462 Rbtable = R5_ARG3,
2463 Rbc = R6_ARG4,
2464 Rscratch = R12_scratch2;
2465
2466 static address field_branch_table[number_of_states],
2467 static_branch_table[number_of_states];
2468
2469 address* branch_table = (is_static || rc == may_not_rewrite) ? static_branch_table : field_branch_table;
2470
2471 // Get field offset.
2472 resolve_cache_and_index(byte_no, Rcache, Rscratch, sizeof(u2));
2473
2474 // JVMTI support
2475 jvmti_post_field_access(Rcache, Rscratch, is_static, false);
2476
2477 // Load after possible GC.
2478 load_field_cp_cache_entry(Rclass_or_obj, Rcache, noreg, Roffset, Rflags, is_static);
2479
2480 // Load pointer to branch table.
2481 __ load_const_optimized(Rbtable, (address)branch_table, Rscratch);
2482
2483 // Get volatile flag.
2484 __ rldicl(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2485 // Note: sync is needed before volatile load on PPC64.
2486
2487 // Check field type.
2488 __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2489
2490 #ifdef ASSERT
2491 Label LFlagInvalid;
2492 __ cmpldi(CCR0, Rflags, number_of_states);
2493 __ bge(CCR0, LFlagInvalid);
2494 #endif
2495
2496 // Load from branch table and dispatch (volatile case: one instruction ahead).
2497 __ sldi(Rflags, Rflags, LogBytesPerWord);
2498 __ cmpwi(CCR6, Rscratch, 1); // Volatile?
2499 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2500 __ sldi(Rscratch, Rscratch, exact_log2(BytesPerInstWord)); // Volatile ? size of 1 instruction : 0.
2501 }
2502 __ ldx(Rbtable, Rbtable, Rflags);
2503
2504 // Get the obj from stack.
2505 if (!is_static) {
2506 pop_and_check_object(Rclass_or_obj); // Kills R11_scratch1.
2507 } else {
2508 __ verify_oop(Rclass_or_obj);
2509 }
2510
2511 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2512 __ subf(Rbtable, Rscratch, Rbtable); // Point to volatile/non-volatile entry point.
2513 }
2514 __ mtctr(Rbtable);
2515 __ bctr();
2516
2517 #ifdef ASSERT
2518 __ bind(LFlagInvalid);
2519 __ stop("got invalid flag", 0x654);
2520 #endif
2521
2522 if (!is_static && rc == may_not_rewrite) {
2523 // We reuse the code from is_static. It's jumped to via the table above.
2524 return;
2525 }
2526
2527 #ifdef ASSERT
2528 // __ bind(Lvtos);
2529 address pc_before_fence = __ pc();
2530 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2531 assert(__ pc() - pc_before_fence == (ptrdiff_t)BytesPerInstWord, "must be single instruction");
2532 assert(branch_table[vtos] == 0, "can't compute twice");
2533 branch_table[vtos] = __ pc(); // non-volatile_entry point
2534 __ stop("vtos unexpected", 0x655);
2535 #endif
2536
2537 __ align(32, 28, 28); // Align load.
2538 // __ bind(Ldtos);
2539 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2540 assert(branch_table[dtos] == 0, "can't compute twice");
2541 branch_table[dtos] = __ pc(); // non-volatile_entry point
2542 __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
2543 __ push(dtos);
2544 if (!is_static && rc == may_rewrite) {
2545 patch_bytecode(Bytecodes::_fast_dgetfield, Rbc, Rscratch);
2546 }
2547 {
2548 Label acquire_double;
2549 __ beq(CCR6, acquire_double); // Volatile?
2550 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2551
2552 __ bind(acquire_double);
2553 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
2554 __ beq_predict_taken(CCR0, Lisync);
2555 __ b(Lisync); // In case of NAN.
2556 }
2557
2558 __ align(32, 28, 28); // Align load.
2559 // __ bind(Lftos);
2560 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2561 assert(branch_table[ftos] == 0, "can't compute twice");
2562 branch_table[ftos] = __ pc(); // non-volatile_entry point
2563 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
2564 __ push(ftos);
2565 if (!is_static && rc == may_rewrite) {
2566 patch_bytecode(Bytecodes::_fast_fgetfield, Rbc, Rscratch);
2567 }
2568 {
2569 Label acquire_float;
2570 __ beq(CCR6, acquire_float); // Volatile?
2571 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2572
2573 __ bind(acquire_float);
2574 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
2575 __ beq_predict_taken(CCR0, Lisync);
2576 __ b(Lisync); // In case of NAN.
2577 }
2578
2579 __ align(32, 28, 28); // Align load.
2580 // __ bind(Litos);
2581 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2582 assert(branch_table[itos] == 0, "can't compute twice");
2583 branch_table[itos] = __ pc(); // non-volatile_entry point
2584 __ lwax(R17_tos, Rclass_or_obj, Roffset);
2585 __ push(itos);
2586 if (!is_static && rc == may_rewrite) {
2587 patch_bytecode(Bytecodes::_fast_igetfield, Rbc, Rscratch);
2588 }
2589 __ beq(CCR6, Lacquire); // Volatile?
2590 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2591
2592 __ align(32, 28, 28); // Align load.
2593 // __ bind(Lltos);
2594 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2595 assert(branch_table[ltos] == 0, "can't compute twice");
2596 branch_table[ltos] = __ pc(); // non-volatile_entry point
2597 __ ldx(R17_tos, Rclass_or_obj, Roffset);
2598 __ push(ltos);
2599 if (!is_static && rc == may_rewrite) {
2600 patch_bytecode(Bytecodes::_fast_lgetfield, Rbc, Rscratch);
2601 }
2602 __ beq(CCR6, Lacquire); // Volatile?
2603 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2604
2605 __ align(32, 28, 28); // Align load.
2606 // __ bind(Lbtos);
2607 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2608 assert(branch_table[btos] == 0, "can't compute twice");
2609 branch_table[btos] = __ pc(); // non-volatile_entry point
2610 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
2611 __ extsb(R17_tos, R17_tos);
2612 __ push(btos);
2613 if (!is_static && rc == may_rewrite) {
2614 patch_bytecode(Bytecodes::_fast_bgetfield, Rbc, Rscratch);
2615 }
2616 __ beq(CCR6, Lacquire); // Volatile?
2617 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2618
2619 __ align(32, 28, 28); // Align load.
2620 // __ bind(Lztos); (same code as btos)
2621 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2622 assert(branch_table[ztos] == 0, "can't compute twice");
2623 branch_table[ztos] = __ pc(); // non-volatile_entry point
2624 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
2625 __ push(ztos);
2626 if (!is_static && rc == may_rewrite) {
2627 // use btos rewriting, no truncating to t/f bit is needed for getfield.
2628 patch_bytecode(Bytecodes::_fast_bgetfield, Rbc, Rscratch);
2629 }
2630 __ beq(CCR6, Lacquire); // Volatile?
2631 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2632
2633 __ align(32, 28, 28); // Align load.
2634 // __ bind(Lctos);
2635 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2636 assert(branch_table[ctos] == 0, "can't compute twice");
2637 branch_table[ctos] = __ pc(); // non-volatile_entry point
2638 __ lhzx(R17_tos, Rclass_or_obj, Roffset);
2639 __ push(ctos);
2640 if (!is_static && rc == may_rewrite) {
2641 patch_bytecode(Bytecodes::_fast_cgetfield, Rbc, Rscratch);
2642 }
2643 __ beq(CCR6, Lacquire); // Volatile?
2644 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2645
2646 __ align(32, 28, 28); // Align load.
2647 // __ bind(Lstos);
2648 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2649 assert(branch_table[stos] == 0, "can't compute twice");
2650 branch_table[stos] = __ pc(); // non-volatile_entry point
2651 __ lhax(R17_tos, Rclass_or_obj, Roffset);
2652 __ push(stos);
2653 if (!is_static && rc == may_rewrite) {
2654 patch_bytecode(Bytecodes::_fast_sgetfield, Rbc, Rscratch);
2655 }
2656 __ beq(CCR6, Lacquire); // Volatile?
2657 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2658
2659 __ align(32, 28, 28); // Align load.
2660 // __ bind(Latos);
2661 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2662 assert(branch_table[atos] == 0, "can't compute twice");
2663 branch_table[atos] = __ pc(); // non-volatile_entry point
2664 do_oop_load(_masm, Rclass_or_obj, Roffset, R17_tos, Rscratch, /* nv temp */ Rflags, IN_HEAP);
2665 __ verify_oop(R17_tos);
2666 __ push(atos);
2667 //__ dcbt(R17_tos); // prefetch
2668 if (!is_static && rc == may_rewrite) {
2669 patch_bytecode(Bytecodes::_fast_agetfield, Rbc, Rscratch);
2670 }
2671 __ beq(CCR6, Lacquire); // Volatile?
2672 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2673
2674 __ align(32, 12);
2675 __ bind(Lacquire);
2676 __ twi_0(R17_tos);
2677 __ bind(Lisync);
2678 __ isync(); // acquire
2679
2680 #ifdef ASSERT
2681 for (int i = 0; i<number_of_states; ++i) {
2682 assert(branch_table[i], "get initialization");
2683 //tty->print_cr("get: %s_branch_table[%d] = 0x%llx (opcode 0x%llx)",
2684 // is_static ? "static" : "field", i, branch_table[i], *((unsigned int*)branch_table[i]));
2685 }
2686 #endif
2687 }
2688
2689 void TemplateTable::getfield(int byte_no) {
2690 getfield_or_static(byte_no, false);
2691 }
2692
2693 void TemplateTable::nofast_getfield(int byte_no) {
2694 getfield_or_static(byte_no, false, may_not_rewrite);
2695 }
2696
2697 void TemplateTable::getstatic(int byte_no) {
2698 getfield_or_static(byte_no, true);
2699 }
2700
2701 // The registers cache and index expected to be set before call.
2702 // The function may destroy various registers, just not the cache and index registers.
2703 void TemplateTable::jvmti_post_field_mod(Register Rcache, Register Rscratch, bool is_static) {
2704
2705 assert_different_registers(Rcache, Rscratch, R6_ARG4);
2706
2707 if (JvmtiExport::can_post_field_modification()) {
2708 Label Lno_field_mod_post;
2709
2710 // Check if post field access in enabled.
2711 int offs = __ load_const_optimized(Rscratch, JvmtiExport::get_field_modification_count_addr(), R0, true);
2712 __ lwz(Rscratch, offs, Rscratch);
2713
2714 __ cmpwi(CCR0, Rscratch, 0);
2715 __ beq(CCR0, Lno_field_mod_post);
2716
2717 // Do the post
2718 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2719 const Register Robj = Rscratch;
2720
2721 __ addi(Rcache, Rcache, in_bytes(cp_base_offset));
2722 if (is_static) {
2723 // Life is simple. Null out the object pointer.
2724 __ li(Robj, 0);
2725 } else {
2726 // In case of the fast versions, value lives in registers => put it back on tos.
2727 int offs = Interpreter::expr_offset_in_bytes(0);
2728 Register base = R15_esp;
2729 switch(bytecode()) {
2730 case Bytecodes::_fast_aputfield: __ push_ptr(); offs+= Interpreter::stackElementSize; break;
2731 case Bytecodes::_fast_iputfield: // Fall through
2732 case Bytecodes::_fast_bputfield: // Fall through
2733 case Bytecodes::_fast_zputfield: // Fall through
2734 case Bytecodes::_fast_cputfield: // Fall through
2735 case Bytecodes::_fast_sputfield: __ push_i(); offs+= Interpreter::stackElementSize; break;
2736 case Bytecodes::_fast_lputfield: __ push_l(); offs+=2*Interpreter::stackElementSize; break;
2737 case Bytecodes::_fast_fputfield: __ push_f(); offs+= Interpreter::stackElementSize; break;
2738 case Bytecodes::_fast_dputfield: __ push_d(); offs+=2*Interpreter::stackElementSize; break;
2739 default: {
2740 offs = 0;
2741 base = Robj;
2742 const Register Rflags = Robj;
2743 Label is_one_slot;
2744 // Life is harder. The stack holds the value on top, followed by the
2745 // object. We don't know the size of the value, though; it could be
2746 // one or two words depending on its type. As a result, we must find
2747 // the type to determine where the object is.
2748 __ ld(Rflags, in_bytes(ConstantPoolCacheEntry::flags_offset()), Rcache); // Big Endian
2749 __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2750
2751 __ cmpwi(CCR0, Rflags, ltos);
2752 __ cmpwi(CCR1, Rflags, dtos);
2753 __ addi(base, R15_esp, Interpreter::expr_offset_in_bytes(1));
2754 __ crnor(CCR0, Assembler::equal, CCR1, Assembler::equal);
2755 __ beq(CCR0, is_one_slot);
2756 __ addi(base, R15_esp, Interpreter::expr_offset_in_bytes(2));
2757 __ bind(is_one_slot);
2758 break;
2759 }
2760 }
2761 __ ld(Robj, offs, base);
2762 __ verify_oop(Robj);
2763 }
2764
2765 __ addi(R6_ARG4, R15_esp, Interpreter::expr_offset_in_bytes(0));
2766 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), Robj, Rcache, R6_ARG4);
2767 __ get_cache_and_index_at_bcp(Rcache, 1);
2768
2769 // In case of the fast versions, value lives in registers => put it back on tos.
2770 switch(bytecode()) {
2771 case Bytecodes::_fast_aputfield: __ pop_ptr(); break;
2772 case Bytecodes::_fast_iputfield: // Fall through
2773 case Bytecodes::_fast_bputfield: // Fall through
2774 case Bytecodes::_fast_zputfield: // Fall through
2775 case Bytecodes::_fast_cputfield: // Fall through
2776 case Bytecodes::_fast_sputfield: __ pop_i(); break;
2777 case Bytecodes::_fast_lputfield: __ pop_l(); break;
2778 case Bytecodes::_fast_fputfield: __ pop_f(); break;
2779 case Bytecodes::_fast_dputfield: __ pop_d(); break;
2780 default: break; // Nothin' to do.
2781 }
2782
2783 __ align(32, 12);
2784 __ bind(Lno_field_mod_post);
2785 }
2786 }
2787
2788 // PPC64: implement volatile stores as release-store (return bytecode contains an additional release).
2789 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2790 Label Lvolatile;
2791
2792 const Register Rcache = R5_ARG3, // Do not use ARG1/2 (causes trouble in jvmti_post_field_mod).
2793 Rclass_or_obj = R31, // Needs to survive C call.
2794 Roffset = R22_tmp2, // Needs to survive C call.
2795 Rflags = R3_ARG1,
2796 Rbtable = R4_ARG2,
2797 Rscratch = R11_scratch1,
2798 Rscratch2 = R12_scratch2,
2799 Rscratch3 = R6_ARG4,
2800 Rbc = Rscratch3;
2801 const ConditionRegister CR_is_vol = CCR2; // Non-volatile condition register (survives runtime call in do_oop_store).
2802
2803 static address field_rw_branch_table[number_of_states],
2804 field_norw_branch_table[number_of_states],
2805 static_branch_table[number_of_states];
2806
2807 address* branch_table = is_static ? static_branch_table :
2808 (rc == may_rewrite ? field_rw_branch_table : field_norw_branch_table);
2809
2810 // Stack (grows up):
2811 // value
2812 // obj
2813
2814 // Load the field offset.
2815 resolve_cache_and_index(byte_no, Rcache, Rscratch, sizeof(u2));
2816 jvmti_post_field_mod(Rcache, Rscratch, is_static);
2817 load_field_cp_cache_entry(Rclass_or_obj, Rcache, noreg, Roffset, Rflags, is_static);
2818
2819 // Load pointer to branch table.
2820 __ load_const_optimized(Rbtable, (address)branch_table, Rscratch);
2821
2822 // Get volatile flag.
2823 __ rldicl(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2824
2825 // Check the field type.
2826 __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2827
2828 #ifdef ASSERT
2829 Label LFlagInvalid;
2830 __ cmpldi(CCR0, Rflags, number_of_states);
2831 __ bge(CCR0, LFlagInvalid);
2832 #endif
2833
2834 // Load from branch table and dispatch (volatile case: one instruction ahead).
2835 __ sldi(Rflags, Rflags, LogBytesPerWord);
2836 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2837 __ cmpwi(CR_is_vol, Rscratch, 1); // Volatile?
2838 }
2839 __ sldi(Rscratch, Rscratch, exact_log2(BytesPerInstWord)); // Volatile? size of instruction 1 : 0.
2840 __ ldx(Rbtable, Rbtable, Rflags);
2841
2842 __ subf(Rbtable, Rscratch, Rbtable); // Point to volatile/non-volatile entry point.
2843 __ mtctr(Rbtable);
2844 __ bctr();
2845
2846 #ifdef ASSERT
2847 __ bind(LFlagInvalid);
2848 __ stop("got invalid flag", 0x656);
2849
2850 // __ bind(Lvtos);
2851 address pc_before_release = __ pc();
2852 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2853 assert(__ pc() - pc_before_release == (ptrdiff_t)BytesPerInstWord, "must be single instruction");
2854 assert(branch_table[vtos] == 0, "can't compute twice");
2855 branch_table[vtos] = __ pc(); // non-volatile_entry point
2856 __ stop("vtos unexpected", 0x657);
2857 #endif
2858
2859 __ align(32, 28, 28); // Align pop.
2860 // __ bind(Ldtos);
2861 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2862 assert(branch_table[dtos] == 0, "can't compute twice");
2863 branch_table[dtos] = __ pc(); // non-volatile_entry point
2864 __ pop(dtos);
2865 if (!is_static) {
2866 pop_and_check_object(Rclass_or_obj); // Kills R11_scratch1.
2867 }
2868 __ stfdx(F15_ftos, Rclass_or_obj, Roffset);
2869 if (!is_static && rc == may_rewrite) {
2870 patch_bytecode(Bytecodes::_fast_dputfield, Rbc, Rscratch, true, byte_no);
2871 }
2872 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2873 __ beq(CR_is_vol, Lvolatile); // Volatile?
2874 }
2875 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2876
2877 __ align(32, 28, 28); // Align pop.
2878 // __ bind(Lftos);
2879 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2880 assert(branch_table[ftos] == 0, "can't compute twice");
2881 branch_table[ftos] = __ pc(); // non-volatile_entry point
2882 __ pop(ftos);
2883 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2884 __ stfsx(F15_ftos, Rclass_or_obj, Roffset);
2885 if (!is_static && rc == may_rewrite) {
2886 patch_bytecode(Bytecodes::_fast_fputfield, Rbc, Rscratch, true, byte_no);
2887 }
2888 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2889 __ beq(CR_is_vol, Lvolatile); // Volatile?
2890 }
2891 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2892
2893 __ align(32, 28, 28); // Align pop.
2894 // __ bind(Litos);
2895 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2896 assert(branch_table[itos] == 0, "can't compute twice");
2897 branch_table[itos] = __ pc(); // non-volatile_entry point
2898 __ pop(itos);
2899 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2900 __ stwx(R17_tos, Rclass_or_obj, Roffset);
2901 if (!is_static && rc == may_rewrite) {
2902 patch_bytecode(Bytecodes::_fast_iputfield, Rbc, Rscratch, true, byte_no);
2903 }
2904 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2905 __ beq(CR_is_vol, Lvolatile); // Volatile?
2906 }
2907 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2908
2909 __ align(32, 28, 28); // Align pop.
2910 // __ bind(Lltos);
2911 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2912 assert(branch_table[ltos] == 0, "can't compute twice");
2913 branch_table[ltos] = __ pc(); // non-volatile_entry point
2914 __ pop(ltos);
2915 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2916 __ stdx(R17_tos, Rclass_or_obj, Roffset);
2917 if (!is_static && rc == may_rewrite) {
2918 patch_bytecode(Bytecodes::_fast_lputfield, Rbc, Rscratch, true, byte_no);
2919 }
2920 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2921 __ beq(CR_is_vol, Lvolatile); // Volatile?
2922 }
2923 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2924
2925 __ align(32, 28, 28); // Align pop.
2926 // __ bind(Lbtos);
2927 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2928 assert(branch_table[btos] == 0, "can't compute twice");
2929 branch_table[btos] = __ pc(); // non-volatile_entry point
2930 __ pop(btos);
2931 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2932 __ stbx(R17_tos, Rclass_or_obj, Roffset);
2933 if (!is_static && rc == may_rewrite) {
2934 patch_bytecode(Bytecodes::_fast_bputfield, Rbc, Rscratch, true, byte_no);
2935 }
2936 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2937 __ beq(CR_is_vol, Lvolatile); // Volatile?
2938 }
2939 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2940
2941 __ align(32, 28, 28); // Align pop.
2942 // __ bind(Lztos);
2943 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2944 assert(branch_table[ztos] == 0, "can't compute twice");
2945 branch_table[ztos] = __ pc(); // non-volatile_entry point
2946 __ pop(ztos);
2947 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2948 __ andi(R17_tos, R17_tos, 0x1);
2949 __ stbx(R17_tos, Rclass_or_obj, Roffset);
2950 if (!is_static && rc == may_rewrite) {
2951 patch_bytecode(Bytecodes::_fast_zputfield, Rbc, Rscratch, true, byte_no);
2952 }
2953 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2954 __ beq(CR_is_vol, Lvolatile); // Volatile?
2955 }
2956 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2957
2958 __ align(32, 28, 28); // Align pop.
2959 // __ bind(Lctos);
2960 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2961 assert(branch_table[ctos] == 0, "can't compute twice");
2962 branch_table[ctos] = __ pc(); // non-volatile_entry point
2963 __ pop(ctos);
2964 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1..
2965 __ sthx(R17_tos, Rclass_or_obj, Roffset);
2966 if (!is_static && rc == may_rewrite) {
2967 patch_bytecode(Bytecodes::_fast_cputfield, Rbc, Rscratch, true, byte_no);
2968 }
2969 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2970 __ beq(CR_is_vol, Lvolatile); // Volatile?
2971 }
2972 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2973
2974 __ align(32, 28, 28); // Align pop.
2975 // __ bind(Lstos);
2976 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2977 assert(branch_table[stos] == 0, "can't compute twice");
2978 branch_table[stos] = __ pc(); // non-volatile_entry point
2979 __ pop(stos);
2980 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2981 __ sthx(R17_tos, Rclass_or_obj, Roffset);
2982 if (!is_static && rc == may_rewrite) {
2983 patch_bytecode(Bytecodes::_fast_sputfield, Rbc, Rscratch, true, byte_no);
2984 }
2985 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2986 __ beq(CR_is_vol, Lvolatile); // Volatile?
2987 }
2988 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2989
2990 __ align(32, 28, 28); // Align pop.
2991 // __ bind(Latos);
2992 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2993 assert(branch_table[atos] == 0, "can't compute twice");
2994 branch_table[atos] = __ pc(); // non-volatile_entry point
2995 __ pop(atos);
2996 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // kills R11_scratch1
2997 do_oop_store(_masm, Rclass_or_obj, Roffset, R17_tos, Rscratch, Rscratch2, Rscratch3, IN_HEAP);
2998 if (!is_static && rc == may_rewrite) {
2999 patch_bytecode(Bytecodes::_fast_aputfield, Rbc, Rscratch, true, byte_no);
3000 }
3001 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
3002 __ beq(CR_is_vol, Lvolatile); // Volatile?
3003 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
3004
3005 __ align(32, 12);
3006 __ bind(Lvolatile);
3007 __ fence();
3008 }
3009 // fallthru: __ b(Lexit);
3010
3011 #ifdef ASSERT
3012 for (int i = 0; i<number_of_states; ++i) {
3013 assert(branch_table[i], "put initialization");
3014 //tty->print_cr("put: %s_branch_table[%d] = 0x%llx (opcode 0x%llx)",
3015 // is_static ? "static" : "field", i, branch_table[i], *((unsigned int*)branch_table[i]));
3016 }
3017 #endif
3018 }
3019
3020 void TemplateTable::putfield(int byte_no) {
3021 putfield_or_static(byte_no, false);
3022 }
3023
3024 void TemplateTable::nofast_putfield(int byte_no) {
3025 putfield_or_static(byte_no, false, may_not_rewrite);
3026 }
3027
3028 void TemplateTable::putstatic(int byte_no) {
3029 putfield_or_static(byte_no, true);
3030 }
3031
3032 // See SPARC. On PPC64, we have a different jvmti_post_field_mod which does the job.
3033 void TemplateTable::jvmti_post_fast_field_mod() {
3034 __ should_not_reach_here();
3035 }
3036
3037 void TemplateTable::fast_storefield(TosState state) {
3038 transition(state, vtos);
3039
3040 const Register Rcache = R5_ARG3, // Do not use ARG1/2 (causes trouble in jvmti_post_field_mod).
3041 Rclass_or_obj = R31, // Needs to survive C call.
3042 Roffset = R22_tmp2, // Needs to survive C call.
3043 Rflags = R3_ARG1,
3044 Rscratch = R11_scratch1,
3045 Rscratch2 = R12_scratch2,
3046 Rscratch3 = R4_ARG2;
3047 const ConditionRegister CR_is_vol = CCR2; // Non-volatile condition register (survives runtime call in do_oop_store).
3048
3049 // Constant pool already resolved => Load flags and offset of field.
3050 __ get_cache_and_index_at_bcp(Rcache, 1);
3051 jvmti_post_field_mod(Rcache, Rscratch, false /* not static */);
3052 load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
3053
3054 // Get the obj and the final store addr.
3055 pop_and_check_object(Rclass_or_obj); // Kills R11_scratch1.
3056
3057 // Get volatile flag.
3058 __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
3059 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) { __ cmpdi(CR_is_vol, Rscratch, 1); }
3060 {
3061 Label LnotVolatile;
3062 __ beq(CCR0, LnotVolatile);
3063 __ release();
3064 __ align(32, 12);
3065 __ bind(LnotVolatile);
3066 }
3067
3068 // Do the store and fencing.
3069 switch(bytecode()) {
3070 case Bytecodes::_fast_aputfield:
3071 // Store into the field.
3072 do_oop_store(_masm, Rclass_or_obj, Roffset, R17_tos, Rscratch, Rscratch2, Rscratch3, IN_HEAP);
3073 break;
3074
3075 case Bytecodes::_fast_iputfield:
3076 __ stwx(R17_tos, Rclass_or_obj, Roffset);
3077 break;
3078
3079 case Bytecodes::_fast_lputfield:
3080 __ stdx(R17_tos, Rclass_or_obj, Roffset);
3081 break;
3082
3083 case Bytecodes::_fast_zputfield:
3084 __ andi(R17_tos, R17_tos, 0x1); // boolean is true if LSB is 1
3085 // fall through to bputfield
3086 case Bytecodes::_fast_bputfield:
3087 __ stbx(R17_tos, Rclass_or_obj, Roffset);
3088 break;
3089
3090 case Bytecodes::_fast_cputfield:
3091 case Bytecodes::_fast_sputfield:
3092 __ sthx(R17_tos, Rclass_or_obj, Roffset);
3093 break;
3094
3095 case Bytecodes::_fast_fputfield:
3096 __ stfsx(F15_ftos, Rclass_or_obj, Roffset);
3097 break;
3098
3099 case Bytecodes::_fast_dputfield:
3100 __ stfdx(F15_ftos, Rclass_or_obj, Roffset);
3101 break;
3102
3103 default: ShouldNotReachHere();
3104 }
3105
3106 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
3107 Label LVolatile;
3108 __ beq(CR_is_vol, LVolatile);
3109 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
3110
3111 __ align(32, 12);
3112 __ bind(LVolatile);
3113 __ fence();
3114 }
3115 }
3116
3117 void TemplateTable::fast_accessfield(TosState state) {
3118 transition(atos, state);
3119
3120 Label LisVolatile;
3121 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
3122
3123 const Register Rcache = R3_ARG1,
3124 Rclass_or_obj = R17_tos,
3125 Roffset = R22_tmp2,
3126 Rflags = R23_tmp3,
3127 Rscratch = R12_scratch2;
3128
3129 // Constant pool already resolved. Get the field offset.
3130 __ get_cache_and_index_at_bcp(Rcache, 1);
3131 load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
3132
3133 // JVMTI support
3134 jvmti_post_field_access(Rcache, Rscratch, false, true);
3135
3136 // Get the load address.
3137 __ null_check_throw(Rclass_or_obj, -1, Rscratch);
3138
3139 // Get volatile flag.
3140 __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
3141 __ bne(CCR0, LisVolatile);
3142
3143 switch(bytecode()) {
3144 case Bytecodes::_fast_agetfield:
3145 {
3146 do_oop_load(_masm, Rclass_or_obj, Roffset, R17_tos, Rscratch, /* nv temp */ Rflags, IN_HEAP);
3147 __ verify_oop(R17_tos);
3148 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3149
3150 __ bind(LisVolatile);
3151 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3152 do_oop_load(_masm, Rclass_or_obj, Roffset, R17_tos, Rscratch, /* nv temp */ Rflags, IN_HEAP);
3153 __ verify_oop(R17_tos);
3154 __ twi_0(R17_tos);
3155 __ isync();
3156 break;
3157 }
3158 case Bytecodes::_fast_igetfield:
3159 {
3160 __ lwax(R17_tos, Rclass_or_obj, Roffset);
3161 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3162
3163 __ bind(LisVolatile);
3164 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3165 __ lwax(R17_tos, Rclass_or_obj, Roffset);
3166 __ twi_0(R17_tos);
3167 __ isync();
3168 break;
3169 }
3170 case Bytecodes::_fast_lgetfield:
3171 {
3172 __ ldx(R17_tos, Rclass_or_obj, Roffset);
3173 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3174
3175 __ bind(LisVolatile);
3176 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3177 __ ldx(R17_tos, Rclass_or_obj, Roffset);
3178 __ twi_0(R17_tos);
3179 __ isync();
3180 break;
3181 }
3182 case Bytecodes::_fast_bgetfield:
3183 {
3184 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
3185 __ extsb(R17_tos, R17_tos);
3186 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3187
3188 __ bind(LisVolatile);
3189 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3190 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
3191 __ twi_0(R17_tos);
3192 __ extsb(R17_tos, R17_tos);
3193 __ isync();
3194 break;
3195 }
3196 case Bytecodes::_fast_cgetfield:
3197 {
3198 __ lhzx(R17_tos, Rclass_or_obj, Roffset);
3199 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3200
3201 __ bind(LisVolatile);
3202 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3203 __ lhzx(R17_tos, Rclass_or_obj, Roffset);
3204 __ twi_0(R17_tos);
3205 __ isync();
3206 break;
3207 }
3208 case Bytecodes::_fast_sgetfield:
3209 {
3210 __ lhax(R17_tos, Rclass_or_obj, Roffset);
3211 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3212
3213 __ bind(LisVolatile);
3214 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3215 __ lhax(R17_tos, Rclass_or_obj, Roffset);
3216 __ twi_0(R17_tos);
3217 __ isync();
3218 break;
3219 }
3220 case Bytecodes::_fast_fgetfield:
3221 {
3222 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3223 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3224
3225 __ bind(LisVolatile);
3226 Label Ldummy;
3227 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3228 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3229 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3230 __ bne_predict_not_taken(CCR0, Ldummy);
3231 __ bind(Ldummy);
3232 __ isync();
3233 break;
3234 }
3235 case Bytecodes::_fast_dgetfield:
3236 {
3237 __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
3238 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3239
3240 __ bind(LisVolatile);
3241 Label Ldummy;
3242 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3243 __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
3244 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3245 __ bne_predict_not_taken(CCR0, Ldummy);
3246 __ bind(Ldummy);
3247 __ isync();
3248 break;
3249 }
3250 default: ShouldNotReachHere();
3251 }
3252 }
3253
3254 void TemplateTable::fast_xaccess(TosState state) {
3255 transition(vtos, state);
3256
3257 Label LisVolatile;
3258 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
3259 const Register Rcache = R3_ARG1,
3260 Rclass_or_obj = R17_tos,
3261 Roffset = R22_tmp2,
3262 Rflags = R23_tmp3,
3263 Rscratch = R12_scratch2;
3264
3265 __ ld(Rclass_or_obj, 0, R18_locals);
3266
3267 // Constant pool already resolved. Get the field offset.
3268 __ get_cache_and_index_at_bcp(Rcache, 2);
3269 load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
3270
3271 // JVMTI support not needed, since we switch back to single bytecode as soon as debugger attaches.
3272
3273 // Needed to report exception at the correct bcp.
3274 __ addi(R14_bcp, R14_bcp, 1);
3275
3276 // Get the load address.
3277 __ null_check_throw(Rclass_or_obj, -1, Rscratch);
3278
3279 // Get volatile flag.
3280 __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
3281 __ bne(CCR0, LisVolatile);
3282
3283 switch(state) {
3284 case atos:
3285 {
3286 do_oop_load(_masm, Rclass_or_obj, Roffset, R17_tos, Rscratch, /* nv temp */ Rflags, IN_HEAP);
3287 __ verify_oop(R17_tos);
3288 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3289
3290 __ bind(LisVolatile);
3291 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3292 do_oop_load(_masm, Rclass_or_obj, Roffset, R17_tos, Rscratch, /* nv temp */ Rflags, IN_HEAP);
3293 __ verify_oop(R17_tos);
3294 __ twi_0(R17_tos);
3295 __ isync();
3296 break;
3297 }
3298 case itos:
3299 {
3300 __ lwax(R17_tos, Rclass_or_obj, Roffset);
3301 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3302
3303 __ bind(LisVolatile);
3304 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3305 __ lwax(R17_tos, Rclass_or_obj, Roffset);
3306 __ twi_0(R17_tos);
3307 __ isync();
3308 break;
3309 }
3310 case ftos:
3311 {
3312 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3313 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3314
3315 __ bind(LisVolatile);
3316 Label Ldummy;
3317 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3318 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3319 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3320 __ bne_predict_not_taken(CCR0, Ldummy);
3321 __ bind(Ldummy);
3322 __ isync();
3323 break;
3324 }
3325 default: ShouldNotReachHere();
3326 }
3327 __ addi(R14_bcp, R14_bcp, -1);
3328 }
3329
3330 // ============================================================================
3331 // Calls
3332
3333 // Common code for invoke
3334 //
3335 // Input:
3336 // - byte_no
3337 //
3338 // Output:
3339 // - Rmethod: The method to invoke next or i-klass (invokeinterface).
3340 // - Rret_addr: The return address to return to.
3341 // - Rindex: MethodType (invokehandle), CallSite obj (invokedynamic) or Method (invokeinterface)
3342 // - Rrecv: Cache for "this" pointer, might be noreg if static call.
3343 // - Rflags: Method flags from const pool cache.
3344 //
3345 // Kills:
3346 // - Rscratch1
3347 //
3348 void TemplateTable::prepare_invoke(int byte_no,
3349 Register Rmethod, // linked method (or i-klass)
3350 Register Rret_addr,// return address
3351 Register Rindex, // itable index, MethodType, Method, etc.
3352 Register Rrecv, // If caller wants to see it.
3353 Register Rflags, // If caller wants to test it.
3354 Register Rscratch
3355 ) {
3356 // Determine flags.
3357 const Bytecodes::Code code = bytecode();
3358 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3359 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3360 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3361 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3362 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3363 const bool load_receiver = (Rrecv != noreg);
3364 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3365
3366 assert_different_registers(Rmethod, Rindex, Rflags, Rscratch);
3367 assert_different_registers(Rmethod, Rrecv, Rflags, Rscratch);
3368 assert_different_registers(Rret_addr, Rscratch);
3369
3370 load_invoke_cp_cache_entry(byte_no, Rmethod, Rindex, Rflags, is_invokevirtual, false, is_invokedynamic);
3371
3372 // Saving of SP done in call_from_interpreter.
3373
3374 // Maybe push "appendix" to arguments.
3375 if (is_invokedynamic || is_invokehandle) {
3376 Label Ldone;
3377 __ rldicl_(R0, Rflags, 64-ConstantPoolCacheEntry::has_appendix_shift, 63);
3378 __ beq(CCR0, Ldone);
3379 // Push "appendix" (MethodType, CallSite, etc.).
3380 // This must be done before we get the receiver,
3381 // since the parameter_size includes it.
3382 __ load_resolved_reference_at_index(Rscratch, Rindex);
3383 __ verify_oop(Rscratch);
3384 __ push_ptr(Rscratch);
3385 __ bind(Ldone);
3386 }
3387
3388 // Load receiver if needed (after appendix is pushed so parameter size is correct).
3389 if (load_receiver) {
3390 const Register Rparam_count = Rscratch;
3391 __ andi(Rparam_count, Rflags, ConstantPoolCacheEntry::parameter_size_mask);
3392 __ load_receiver(Rparam_count, Rrecv);
3393 __ verify_oop(Rrecv);
3394 }
3395
3396 // Get return address.
3397 {
3398 Register Rtable_addr = Rscratch;
3399 Register Rret_type = Rret_addr;
3400 address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3401
3402 // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3403 __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3404 __ load_dispatch_table(Rtable_addr, (address*)table_addr);
3405 __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3406 // Get return address.
3407 __ ldx(Rret_addr, Rtable_addr, Rret_type);
3408 }
3409 }
3410
3411 // Helper for virtual calls. Load target out of vtable and jump off!
3412 // Kills all passed registers.
3413 void TemplateTable::generate_vtable_call(Register Rrecv_klass, Register Rindex, Register Rret, Register Rtemp) {
3414
3415 assert_different_registers(Rrecv_klass, Rtemp, Rret);
3416 const Register Rtarget_method = Rindex;
3417
3418 // Get target method & entry point.
3419 const int base = in_bytes(Klass::vtable_start_offset());
3420 // Calc vtable addr scale the vtable index by 8.
3421 __ sldi(Rindex, Rindex, exact_log2(vtableEntry::size_in_bytes()));
3422 // Load target.
3423 __ addi(Rrecv_klass, Rrecv_klass, base + vtableEntry::method_offset_in_bytes());
3424 __ ldx(Rtarget_method, Rindex, Rrecv_klass);
3425 // Argument and return type profiling.
3426 __ profile_arguments_type(Rtarget_method, Rrecv_klass /* scratch1 */, Rtemp /* scratch2 */, true);
3427 __ call_from_interpreter(Rtarget_method, Rret, Rrecv_klass /* scratch1 */, Rtemp /* scratch2 */);
3428 }
3429
3430 // Virtual or final call. Final calls are rewritten on the fly to run through "fast_finalcall" next time.
3431 void TemplateTable::invokevirtual(int byte_no) {
3432 transition(vtos, vtos);
3433
3434 Register Rtable_addr = R11_scratch1,
3435 Rret_type = R12_scratch2,
3436 Rret_addr = R5_ARG3,
3437 Rflags = R22_tmp2, // Should survive C call.
3438 Rrecv = R3_ARG1,
3439 Rrecv_klass = Rrecv,
3440 Rvtableindex_or_method = R31, // Should survive C call.
3441 Rnum_params = R4_ARG2,
3442 Rnew_bc = R6_ARG4;
3443
3444 Label LnotFinal;
3445
3446 load_invoke_cp_cache_entry(byte_no, Rvtableindex_or_method, noreg, Rflags, /*virtual*/ true, false, false);
3447
3448 __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_vfinal_shift);
3449 __ bfalse(CCR0, LnotFinal);
3450
3451 if (RewriteBytecodes && !UseSharedSpaces && !DumpSharedSpaces) {
3452 patch_bytecode(Bytecodes::_fast_invokevfinal, Rnew_bc, R12_scratch2);
3453 }
3454 invokevfinal_helper(Rvtableindex_or_method, Rflags, R11_scratch1, R12_scratch2);
3455
3456 __ align(32, 12);
3457 __ bind(LnotFinal);
3458 // Load "this" pointer (receiver).
3459 __ rldicl(Rnum_params, Rflags, 64, 48);
3460 __ load_receiver(Rnum_params, Rrecv);
3461 __ verify_oop(Rrecv);
3462
3463 // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3464 __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3465 __ load_dispatch_table(Rtable_addr, Interpreter::invoke_return_entry_table());
3466 __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3467 __ ldx(Rret_addr, Rret_type, Rtable_addr);
3468 __ null_check_throw(Rrecv, oopDesc::klass_offset_in_bytes(), R11_scratch1);
3469 __ load_klass(Rrecv_klass, Rrecv);
3470 __ verify_klass_ptr(Rrecv_klass);
3471 __ profile_virtual_call(Rrecv_klass, R11_scratch1, R12_scratch2, false);
3472
3473 generate_vtable_call(Rrecv_klass, Rvtableindex_or_method, Rret_addr, R11_scratch1);
3474 }
3475
3476 void TemplateTable::fast_invokevfinal(int byte_no) {
3477 transition(vtos, vtos);
3478
3479 assert(byte_no == f2_byte, "use this argument");
3480 Register Rflags = R22_tmp2,
3481 Rmethod = R31;
3482 load_invoke_cp_cache_entry(byte_no, Rmethod, noreg, Rflags, /*virtual*/ true, /*is_invokevfinal*/ true, false);
3483 invokevfinal_helper(Rmethod, Rflags, R11_scratch1, R12_scratch2);
3484 }
3485
3486 void TemplateTable::invokevfinal_helper(Register Rmethod, Register Rflags, Register Rscratch1, Register Rscratch2) {
3487
3488 assert_different_registers(Rmethod, Rflags, Rscratch1, Rscratch2);
3489
3490 // Load receiver from stack slot.
3491 Register Rrecv = Rscratch2;
3492 Register Rnum_params = Rrecv;
3493
3494 __ ld(Rnum_params, in_bytes(Method::const_offset()), Rmethod);
3495 __ lhz(Rnum_params /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), Rnum_params);
3496
3497 // Get return address.
3498 Register Rtable_addr = Rscratch1,
3499 Rret_addr = Rflags,
3500 Rret_type = Rret_addr;
3501 // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3502 __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3503 __ load_dispatch_table(Rtable_addr, Interpreter::invoke_return_entry_table());
3504 __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3505 __ ldx(Rret_addr, Rret_type, Rtable_addr);
3506
3507 // Load receiver and receiver NULL check.
3508 __ load_receiver(Rnum_params, Rrecv);
3509 __ null_check_throw(Rrecv, -1, Rscratch1);
3510
3511 __ profile_final_call(Rrecv, Rscratch1);
3512 // Argument and return type profiling.
3513 __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, true);
3514
3515 // Do the call.
3516 __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1, Rscratch2);
3517 }
3518
3519 void TemplateTable::invokespecial(int byte_no) {
3520 assert(byte_no == f1_byte, "use this argument");
3521 transition(vtos, vtos);
3522
3523 Register Rtable_addr = R3_ARG1,
3524 Rret_addr = R4_ARG2,
3525 Rflags = R5_ARG3,
3526 Rreceiver = R6_ARG4,
3527 Rmethod = R31;
3528
3529 prepare_invoke(byte_no, Rmethod, Rret_addr, noreg, Rreceiver, Rflags, R11_scratch1);
3530
3531 // Receiver NULL check.
3532 __ null_check_throw(Rreceiver, -1, R11_scratch1);
3533
3534 __ profile_call(R11_scratch1, R12_scratch2);
3535 // Argument and return type profiling.
3536 __ profile_arguments_type(Rmethod, R11_scratch1, R12_scratch2, false);
3537 __ call_from_interpreter(Rmethod, Rret_addr, R11_scratch1, R12_scratch2);
3538 }
3539
3540 void TemplateTable::invokestatic(int byte_no) {
3541 assert(byte_no == f1_byte, "use this argument");
3542 transition(vtos, vtos);
3543
3544 Register Rtable_addr = R3_ARG1,
3545 Rret_addr = R4_ARG2,
3546 Rflags = R5_ARG3;
3547
3548 prepare_invoke(byte_no, R19_method, Rret_addr, noreg, noreg, Rflags, R11_scratch1);
3549
3550 __ profile_call(R11_scratch1, R12_scratch2);
3551 // Argument and return type profiling.
3552 __ profile_arguments_type(R19_method, R11_scratch1, R12_scratch2, false);
3553 __ call_from_interpreter(R19_method, Rret_addr, R11_scratch1, R12_scratch2);
3554 }
3555
3556 void TemplateTable::invokeinterface_object_method(Register Rrecv_klass,
3557 Register Rret,
3558 Register Rflags,
3559 Register Rmethod,
3560 Register Rtemp1,
3561 Register Rtemp2) {
3562
3563 assert_different_registers(Rmethod, Rret, Rrecv_klass, Rflags, Rtemp1, Rtemp2);
3564 Label LnotFinal;
3565
3566 // Check for vfinal.
3567 __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_vfinal_shift);
3568 __ bfalse(CCR0, LnotFinal);
3569
3570 Register Rscratch = Rflags; // Rflags is dead now.
3571
3572 // Final call case.
3573 __ profile_final_call(Rtemp1, Rscratch);
3574 // Argument and return type profiling.
3575 __ profile_arguments_type(Rmethod, Rscratch, Rrecv_klass /* scratch */, true);
3576 // Do the final call - the index (f2) contains the method.
3577 __ call_from_interpreter(Rmethod, Rret, Rscratch, Rrecv_klass /* scratch */);
3578
3579 // Non-final callc case.
3580 __ bind(LnotFinal);
3581 __ profile_virtual_call(Rrecv_klass, Rtemp1, Rscratch, false);
3582 generate_vtable_call(Rrecv_klass, Rmethod, Rret, Rscratch);
3583 }
3584
3585 void TemplateTable::invokeinterface(int byte_no) {
3586 assert(byte_no == f1_byte, "use this argument");
3587 transition(vtos, vtos);
3588
3589 const Register Rscratch1 = R11_scratch1,
3590 Rscratch2 = R12_scratch2,
3591 Rmethod = R6_ARG4,
3592 Rmethod2 = R9_ARG7,
3593 Rinterface_klass = R5_ARG3,
3594 Rret_addr = R8_ARG6,
3595 Rindex = R10_ARG8,
3596 Rreceiver = R3_ARG1,
3597 Rrecv_klass = R4_ARG2,
3598 Rflags = R7_ARG5;
3599
3600 prepare_invoke(byte_no, Rinterface_klass, Rret_addr, Rmethod, Rreceiver, Rflags, Rscratch1);
3601
3602 // First check for Object case, then private interface method,
3603 // then regular interface method.
3604
3605 // Get receiver klass - this is also a null check
3606 __ null_check_throw(Rreceiver, oopDesc::klass_offset_in_bytes(), Rscratch2);
3607 __ load_klass(Rrecv_klass, Rreceiver);
3608
3609 // Check corner case object method.
3610 // Special case of invokeinterface called for virtual method of
3611 // java.lang.Object. See ConstantPoolCacheEntry::set_method() for details:
3612 // The invokeinterface was rewritten to a invokevirtual, hence we have
3613 // to handle this corner case.
3614
3615 Label LnotObjectMethod, Lthrow_ame;
3616 __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_forced_virtual_shift);
3617 __ bfalse(CCR0, LnotObjectMethod);
3618 invokeinterface_object_method(Rrecv_klass, Rret_addr, Rflags, Rmethod, Rscratch1, Rscratch2);
3619 __ bind(LnotObjectMethod);
3620
3621 // Check for private method invocation - indicated by vfinal
3622 Label LnotVFinal, L_no_such_interface, L_subtype;
3623
3624 __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_vfinal_shift);
3625 __ bfalse(CCR0, LnotVFinal);
3626
3627 __ check_klass_subtype(Rrecv_klass, Rinterface_klass, Rscratch1, Rscratch2, L_subtype);
3628 // If we get here the typecheck failed
3629 __ b(L_no_such_interface);
3630 __ bind(L_subtype);
3631
3632 // do the call
3633
3634 Register Rscratch = Rflags; // Rflags is dead now.
3635
3636 __ profile_final_call(Rscratch1, Rscratch);
3637 __ profile_arguments_type(Rmethod, Rscratch, Rrecv_klass /* scratch */, true);
3638
3639 __ call_from_interpreter(Rmethod, Rret_addr, Rscratch, Rrecv_klass /* scratch */);
3640
3641 __ bind(LnotVFinal);
3642
3643 __ lookup_interface_method(Rrecv_klass, Rinterface_klass, noreg, noreg, Rscratch1, Rscratch2,
3644 L_no_such_interface, /*return_method=*/false);
3645
3646 __ profile_virtual_call(Rrecv_klass, Rscratch1, Rscratch2, false);
3647
3648 // Find entry point to call.
3649
3650 // Get declaring interface class from method
3651 __ load_method_holder(Rinterface_klass, Rmethod);
3652
3653 // Get itable index from method
3654 __ lwa(Rindex, in_bytes(Method::itable_index_offset()), Rmethod);
3655 __ subfic(Rindex, Rindex, Method::itable_index_max);
3656
3657 __ lookup_interface_method(Rrecv_klass, Rinterface_klass, Rindex, Rmethod2, Rscratch1, Rscratch2,
3658 L_no_such_interface);
3659
3660 __ cmpdi(CCR0, Rmethod2, 0);
3661 __ beq(CCR0, Lthrow_ame);
3662 // Found entry. Jump off!
3663 // Argument and return type profiling.
3664 __ profile_arguments_type(Rmethod2, Rscratch1, Rscratch2, true);
3665 //__ profile_called_method(Rindex, Rscratch1);
3666 __ call_from_interpreter(Rmethod2, Rret_addr, Rscratch1, Rscratch2);
3667
3668 // Vtable entry was NULL => Throw abstract method error.
3669 __ bind(Lthrow_ame);
3670 // Pass arguments for generating a verbose error message.
3671 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorVerbose),
3672 Rrecv_klass, Rmethod);
3673
3674 // Interface was not found => Throw incompatible class change error.
3675 __ bind(L_no_such_interface);
3676 // Pass arguments for generating a verbose error message.
3677 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose),
3678 Rrecv_klass, Rinterface_klass);
3679 DEBUG_ONLY( __ should_not_reach_here(); )
3680 }
3681
3682 void TemplateTable::invokedynamic(int byte_no) {
3683 transition(vtos, vtos);
3684
3685 const Register Rret_addr = R3_ARG1,
3686 Rflags = R4_ARG2,
3687 Rmethod = R22_tmp2,
3688 Rscratch1 = R11_scratch1,
3689 Rscratch2 = R12_scratch2;
3690
3691 prepare_invoke(byte_no, Rmethod, Rret_addr, Rscratch1, noreg, Rflags, Rscratch2);
3692
3693 // Profile this call.
3694 __ profile_call(Rscratch1, Rscratch2);
3695
3696 // Off we go. With the new method handles, we don't jump to a method handle
3697 // entry any more. Instead, we pushed an "appendix" in prepare invoke, which happens
3698 // to be the callsite object the bootstrap method returned. This is passed to a
3699 // "link" method which does the dispatch (Most likely just grabs the MH stored
3700 // inside the callsite and does an invokehandle).
3701 // Argument and return type profiling.
3702 __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, false);
3703 __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1 /* scratch1 */, Rscratch2 /* scratch2 */);
3704 }
3705
3706 void TemplateTable::invokehandle(int byte_no) {
3707 transition(vtos, vtos);
3708
3709 const Register Rret_addr = R3_ARG1,
3710 Rflags = R4_ARG2,
3711 Rrecv = R5_ARG3,
3712 Rmethod = R22_tmp2,
3713 Rscratch1 = R11_scratch1,
3714 Rscratch2 = R12_scratch2;
3715
3716 prepare_invoke(byte_no, Rmethod, Rret_addr, Rscratch1, Rrecv, Rflags, Rscratch2);
3717 __ verify_method_ptr(Rmethod);
3718 __ null_check_throw(Rrecv, -1, Rscratch2);
3719
3720 __ profile_final_call(Rrecv, Rscratch1);
3721
3722 // Still no call from handle => We call the method handle interpreter here.
3723 // Argument and return type profiling.
3724 __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, true);
3725 __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1 /* scratch1 */, Rscratch2 /* scratch2 */);
3726 }
3727
3728 // =============================================================================
3729 // Allocation
3730
3731 // Puts allocated obj ref onto the expression stack.
3732 void TemplateTable::_new() {
3733 transition(vtos, atos);
3734
3735 Label Lslow_case,
3736 Ldone;
3737
3738 const Register RallocatedObject = R17_tos,
3739 RinstanceKlass = R9_ARG7,
3740 Rscratch = R11_scratch1,
3741 Roffset = R8_ARG6,
3742 Rinstance_size = Roffset,
3743 Rcpool = R4_ARG2,
3744 Rtags = R3_ARG1,
3745 Rindex = R5_ARG3;
3746
3747 // --------------------------------------------------------------------------
3748 // Check if fast case is possible.
3749
3750 // Load pointers to const pool and const pool's tags array.
3751 __ get_cpool_and_tags(Rcpool, Rtags);
3752 // Load index of constant pool entry.
3753 __ get_2_byte_integer_at_bcp(1, Rindex, InterpreterMacroAssembler::Unsigned);
3754
3755 // Note: compared to other architectures, PPC's implementation always goes
3756 // to the slow path if TLAB is used and fails.
3757 if (UseTLAB) {
3758 // Make sure the class we're about to instantiate has been resolved
3759 // This is done before loading instanceKlass to be consistent with the order
3760 // how Constant Pool is updated (see ConstantPoolCache::klass_at_put).
3761 __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3762 __ lbzx(Rtags, Rindex, Rtags);
3763
3764 __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3765 __ bne(CCR0, Lslow_case);
3766
3767 // Get instanceKlass
3768 __ sldi(Roffset, Rindex, LogBytesPerWord);
3769 __ load_resolved_klass_at_offset(Rcpool, Roffset, RinstanceKlass);
3770
3771 // Make sure klass is fully initialized and get instance_size.
3772 __ lbz(Rscratch, in_bytes(InstanceKlass::init_state_offset()), RinstanceKlass);
3773 __ lwz(Rinstance_size, in_bytes(Klass::layout_helper_offset()), RinstanceKlass);
3774
3775 __ cmpdi(CCR1, Rscratch, InstanceKlass::fully_initialized);
3776 // Make sure klass does not have has_finalizer, or is abstract, or interface or java/lang/Class.
3777 __ andi_(R0, Rinstance_size, Klass::_lh_instance_slow_path_bit); // slow path bit equals 0?
3778
3779 __ crnand(CCR0, Assembler::equal, CCR1, Assembler::equal); // slow path bit set or not fully initialized?
3780 __ beq(CCR0, Lslow_case);
3781
3782 // --------------------------------------------------------------------------
3783 // Fast case:
3784 // Allocate the instance.
3785 // 1) Try to allocate in the TLAB.
3786 // 2) If the above fails (or is not applicable), go to a slow case (creates a new TLAB, etc.).
3787
3788 Register RoldTopValue = RallocatedObject; // Object will be allocated here if it fits.
3789 Register RnewTopValue = R6_ARG4;
3790 Register RendValue = R7_ARG5;
3791
3792 // Check if we can allocate in the TLAB.
3793 __ ld(RoldTopValue, in_bytes(JavaThread::tlab_top_offset()), R16_thread);
3794 __ ld(RendValue, in_bytes(JavaThread::tlab_end_offset()), R16_thread);
3795
3796 __ add(RnewTopValue, Rinstance_size, RoldTopValue);
3797
3798 // If there is enough space, we do not CAS and do not clear.
3799 __ cmpld(CCR0, RnewTopValue, RendValue);
3800 __ bgt(CCR0, Lslow_case);
3801
3802 __ std(RnewTopValue, in_bytes(JavaThread::tlab_top_offset()), R16_thread);
3803
3804 if (!ZeroTLAB) {
3805 // --------------------------------------------------------------------------
3806 // Init1: Zero out newly allocated memory.
3807 // Initialize remaining object fields.
3808 Register Rbase = Rtags;
3809 __ addi(Rinstance_size, Rinstance_size, 7 - (int)sizeof(oopDesc));
3810 __ addi(Rbase, RallocatedObject, sizeof(oopDesc));
3811 __ srdi(Rinstance_size, Rinstance_size, 3);
3812
3813 // Clear out object skipping header. Takes also care of the zero length case.
3814 __ clear_memory_doubleword(Rbase, Rinstance_size);
3815 }
3816
3817 // --------------------------------------------------------------------------
3818 // Init2: Initialize the header: mark, klass
3819 // Init mark.
3820 if (UseBiasedLocking) {
3821 __ ld(Rscratch, in_bytes(Klass::prototype_header_offset()), RinstanceKlass);
3822 } else {
3823 __ load_const_optimized(Rscratch, markWord::prototype().value(), R0);
3824 }
3825 __ std(Rscratch, oopDesc::mark_offset_in_bytes(), RallocatedObject);
3826
3827 // Init klass.
3828 __ store_klass_gap(RallocatedObject);
3829 __ store_klass(RallocatedObject, RinstanceKlass, Rscratch); // klass (last for cms)
3830
3831 // Check and trigger dtrace event.
3832 SkipIfEqualZero::skip_to_label_if_equal_zero(_masm, Rscratch, &DTraceAllocProbes, Ldone);
3833 __ push(atos);
3834 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc));
3835 __ pop(atos);
3836
3837 __ b(Ldone);
3838 }
3839
3840 // --------------------------------------------------------------------------
3841 // slow case
3842 __ bind(Lslow_case);
3843 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), Rcpool, Rindex);
3844
3845 // continue
3846 __ bind(Ldone);
3847
3848 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3849 __ membar(Assembler::StoreStore);
3850 }
3851
3852 void TemplateTable::newarray() {
3853 transition(itos, atos);
3854
3855 __ lbz(R4, 1, R14_bcp);
3856 __ extsw(R5, R17_tos);
3857 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), R4, R5 /* size */);
3858
3859 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3860 __ membar(Assembler::StoreStore);
3861 }
3862
3863 void TemplateTable::anewarray() {
3864 transition(itos, atos);
3865
3866 __ get_constant_pool(R4);
3867 __ get_2_byte_integer_at_bcp(1, R5, InterpreterMacroAssembler::Unsigned);
3868 __ extsw(R6, R17_tos); // size
3869 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), R4 /* pool */, R5 /* index */, R6 /* size */);
3870
3871 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3872 __ membar(Assembler::StoreStore);
3873 }
3874
3875 // Allocate a multi dimensional array
3876 void TemplateTable::multianewarray() {
3877 transition(vtos, atos);
3878
3879 Register Rptr = R31; // Needs to survive C call.
3880
3881 // Put ndims * wordSize into frame temp slot
3882 __ lbz(Rptr, 3, R14_bcp);
3883 __ sldi(Rptr, Rptr, Interpreter::logStackElementSize);
3884 // Esp points past last_dim, so set to R4 to first_dim address.
3885 __ add(R4, Rptr, R15_esp);
3886 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), R4 /* first_size_address */);
3887 // Pop all dimensions off the stack.
3888 __ add(R15_esp, Rptr, R15_esp);
3889
3890 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3891 __ membar(Assembler::StoreStore);
3892 }
3893
3894 void TemplateTable::arraylength() {
3895 transition(atos, itos);
3896
3897 __ verify_oop(R17_tos);
3898 __ null_check_throw(R17_tos, arrayOopDesc::length_offset_in_bytes(), R11_scratch1);
3899 __ lwa(R17_tos, arrayOopDesc::length_offset_in_bytes(), R17_tos);
3900 }
3901
3902 // ============================================================================
3903 // Typechecks
3904
3905 void TemplateTable::checkcast() {
3906 transition(atos, atos);
3907
3908 Label Ldone, Lis_null, Lquicked, Lresolved;
3909 Register Roffset = R6_ARG4,
3910 RobjKlass = R4_ARG2,
3911 RspecifiedKlass = R5_ARG3, // Generate_ClassCastException_verbose_handler will read value from this register.
3912 Rcpool = R11_scratch1,
3913 Rtags = R12_scratch2;
3914
3915 // Null does not pass.
3916 __ cmpdi(CCR0, R17_tos, 0);
3917 __ beq(CCR0, Lis_null);
3918
3919 // Get constant pool tag to find out if the bytecode has already been "quickened".
3920 __ get_cpool_and_tags(Rcpool, Rtags);
3921
3922 __ get_2_byte_integer_at_bcp(1, Roffset, InterpreterMacroAssembler::Unsigned);
3923
3924 __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3925 __ lbzx(Rtags, Rtags, Roffset);
3926
3927 __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3928 __ beq(CCR0, Lquicked);
3929
3930 // Call into the VM to "quicken" instanceof.
3931 __ push_ptr(); // for GC
3932 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3933 __ get_vm_result_2(RspecifiedKlass);
3934 __ pop_ptr(); // Restore receiver.
3935 __ b(Lresolved);
3936
3937 // Extract target class from constant pool.
3938 __ bind(Lquicked);
3939 __ sldi(Roffset, Roffset, LogBytesPerWord);
3940 __ load_resolved_klass_at_offset(Rcpool, Roffset, RspecifiedKlass);
3941
3942 // Do the checkcast.
3943 __ bind(Lresolved);
3944 // Get value klass in RobjKlass.
3945 __ load_klass(RobjKlass, R17_tos);
3946 // Generate a fast subtype check. Branch to cast_ok if no failure. Return 0 if failure.
3947 __ gen_subtype_check(RobjKlass, RspecifiedKlass, /*3 temp regs*/ Roffset, Rcpool, Rtags, /*target if subtype*/ Ldone);
3948
3949 // Not a subtype; so must throw exception
3950 // Target class oop is in register R6_ARG4 == RspecifiedKlass by convention.
3951 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ClassCastException_entry);
3952 __ mtctr(R11_scratch1);
3953 __ bctr();
3954
3955 // Profile the null case.
3956 __ align(32, 12);
3957 __ bind(Lis_null);
3958 __ profile_null_seen(R11_scratch1, Rtags); // Rtags used as scratch.
3959
3960 __ align(32, 12);
3961 __ bind(Ldone);
3962 }
3963
3964 // Output:
3965 // - tos == 0: Obj was null or not an instance of class.
3966 // - tos == 1: Obj was an instance of class.
3967 void TemplateTable::instanceof() {
3968 transition(atos, itos);
3969
3970 Label Ldone, Lis_null, Lquicked, Lresolved;
3971 Register Roffset = R6_ARG4,
3972 RobjKlass = R4_ARG2,
3973 RspecifiedKlass = R5_ARG3,
3974 Rcpool = R11_scratch1,
3975 Rtags = R12_scratch2;
3976
3977 // Null does not pass.
3978 __ cmpdi(CCR0, R17_tos, 0);
3979 __ beq(CCR0, Lis_null);
3980
3981 // Get constant pool tag to find out if the bytecode has already been "quickened".
3982 __ get_cpool_and_tags(Rcpool, Rtags);
3983
3984 __ get_2_byte_integer_at_bcp(1, Roffset, InterpreterMacroAssembler::Unsigned);
3985
3986 __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3987 __ lbzx(Rtags, Rtags, Roffset);
3988
3989 __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3990 __ beq(CCR0, Lquicked);
3991
3992 // Call into the VM to "quicken" instanceof.
3993 __ push_ptr(); // for GC
3994 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3995 __ get_vm_result_2(RspecifiedKlass);
3996 __ pop_ptr(); // Restore receiver.
3997 __ b(Lresolved);
3998
3999 // Extract target class from constant pool.
4000 __ bind(Lquicked);
4001 __ sldi(Roffset, Roffset, LogBytesPerWord);
4002 __ load_resolved_klass_at_offset(Rcpool, Roffset, RspecifiedKlass);
4003
4004 // Do the checkcast.
4005 __ bind(Lresolved);
4006 // Get value klass in RobjKlass.
4007 __ load_klass(RobjKlass, R17_tos);
4008 // Generate a fast subtype check. Branch to cast_ok if no failure. Return 0 if failure.
4009 __ li(R17_tos, 1);
4010 __ gen_subtype_check(RobjKlass, RspecifiedKlass, /*3 temp regs*/ Roffset, Rcpool, Rtags, /*target if subtype*/ Ldone);
4011 __ li(R17_tos, 0);
4012
4013 if (ProfileInterpreter) {
4014 __ b(Ldone);
4015 }
4016
4017 // Profile the null case.
4018 __ align(32, 12);
4019 __ bind(Lis_null);
4020 __ profile_null_seen(Rcpool, Rtags); // Rcpool and Rtags used as scratch.
4021
4022 __ align(32, 12);
4023 __ bind(Ldone);
4024 }
4025
4026 // =============================================================================
4027 // Breakpoints
4028
4029 void TemplateTable::_breakpoint() {
4030 transition(vtos, vtos);
4031
4032 // Get the unpatched byte code.
4033 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), R19_method, R14_bcp);
4034 __ mr(R31, R3_RET);
4035
4036 // Post the breakpoint event.
4037 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), R19_method, R14_bcp);
4038
4039 // Complete the execution of original bytecode.
4040 __ dispatch_Lbyte_code(vtos, R31, Interpreter::normal_table(vtos));
4041 }
4042
4043 // =============================================================================
4044 // Exceptions
4045
4046 void TemplateTable::athrow() {
4047 transition(atos, vtos);
4048
4049 // Exception oop is in tos
4050 __ verify_oop(R17_tos);
4051
4052 __ null_check_throw(R17_tos, -1, R11_scratch1);
4053
4054 // Throw exception interpreter entry expects exception oop to be in R3.
4055 __ mr(R3_RET, R17_tos);
4056 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::throw_exception_entry());
4057 __ mtctr(R11_scratch1);
4058 __ bctr();
4059 }
4060
4061 // =============================================================================
4062 // Synchronization
4063 // Searches the basic object lock list on the stack for a free slot
4064 // and uses it to lock the obect in tos.
4065 //
4066 // Recursive locking is enabled by exiting the search if the same
4067 // object is already found in the list. Thus, a new basic lock obj lock
4068 // is allocated "higher up" in the stack and thus is found first
4069 // at next monitor exit.
4070 void TemplateTable::monitorenter() {
4071 transition(atos, vtos);
4072
4073 __ verify_oop(R17_tos);
4074
4075 Register Rcurrent_monitor = R11_scratch1,
4076 Rcurrent_obj = R12_scratch2,
4077 Robj_to_lock = R17_tos,
4078 Rscratch1 = R3_ARG1,
4079 Rscratch2 = R4_ARG2,
4080 Rscratch3 = R5_ARG3,
4081 Rcurrent_obj_addr = R6_ARG4;
4082
4083 // ------------------------------------------------------------------------------
4084 // Null pointer exception.
4085 __ null_check_throw(Robj_to_lock, -1, R11_scratch1);
4086
4087 // Try to acquire a lock on the object.
4088 // Repeat until succeeded (i.e., until monitorenter returns true).
4089
4090 // ------------------------------------------------------------------------------
4091 // Find a free slot in the monitor block.
4092 Label Lfound, Lexit, Lallocate_new;
4093 ConditionRegister found_free_slot = CCR0,
4094 found_same_obj = CCR1,
4095 reached_limit = CCR6;
4096 {
4097 Label Lloop;
4098 Register Rlimit = Rcurrent_monitor;
4099
4100 // Set up search loop - start with topmost monitor.
4101 __ add(Rcurrent_obj_addr, BasicObjectLock::obj_offset_in_bytes(), R26_monitor);
4102
4103 __ ld(Rlimit, 0, R1_SP);
4104 __ addi(Rlimit, Rlimit, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes() - BasicObjectLock::obj_offset_in_bytes())); // Monitor base
4105
4106 // Check if any slot is present => short cut to allocation if not.
4107 __ cmpld(reached_limit, Rcurrent_obj_addr, Rlimit);
4108 __ bgt(reached_limit, Lallocate_new);
4109
4110 // Pre-load topmost slot.
4111 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4112 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4113 // The search loop.
4114 __ bind(Lloop);
4115 // Found free slot?
4116 __ cmpdi(found_free_slot, Rcurrent_obj, 0);
4117 // Is this entry for same obj? If so, stop the search and take the found
4118 // free slot or allocate a new one to enable recursive locking.
4119 __ cmpd(found_same_obj, Rcurrent_obj, Robj_to_lock);
4120 __ cmpld(reached_limit, Rcurrent_obj_addr, Rlimit);
4121 __ beq(found_free_slot, Lexit);
4122 __ beq(found_same_obj, Lallocate_new);
4123 __ bgt(reached_limit, Lallocate_new);
4124 // Check if last allocated BasicLockObj reached.
4125 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4126 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4127 // Next iteration if unchecked BasicObjectLocks exist on the stack.
4128 __ b(Lloop);
4129 }
4130
4131 // ------------------------------------------------------------------------------
4132 // Check if we found a free slot.
4133 __ bind(Lexit);
4134
4135 __ addi(Rcurrent_monitor, Rcurrent_obj_addr, -(frame::interpreter_frame_monitor_size() * wordSize) - BasicObjectLock::obj_offset_in_bytes());
4136 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, - frame::interpreter_frame_monitor_size() * wordSize);
4137 __ b(Lfound);
4138
4139 // We didn't find a free BasicObjLock => allocate one.
4140 __ align(32, 12);
4141 __ bind(Lallocate_new);
4142 __ add_monitor_to_stack(false, Rscratch1, Rscratch2);
4143 __ mr(Rcurrent_monitor, R26_monitor);
4144 __ addi(Rcurrent_obj_addr, R26_monitor, BasicObjectLock::obj_offset_in_bytes());
4145
4146 // ------------------------------------------------------------------------------
4147 // We now have a slot to lock.
4148 __ bind(Lfound);
4149
4150 // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
4151 // The object has already been poped from the stack, so the expression stack looks correct.
4152 __ addi(R14_bcp, R14_bcp, 1);
4153
4154 __ std(Robj_to_lock, 0, Rcurrent_obj_addr);
4155 __ lock_object(Rcurrent_monitor, Robj_to_lock);
4156
4157 // Check if there's enough space on the stack for the monitors after locking.
4158 // This emits a single store.
4159 __ generate_stack_overflow_check(0);
4160
4161 // The bcp has already been incremented. Just need to dispatch to next instruction.
4162 __ dispatch_next(vtos);
4163 }
4164
4165 void TemplateTable::monitorexit() {
4166 transition(atos, vtos);
4167 __ verify_oop(R17_tos);
4168
4169 Register Rcurrent_monitor = R11_scratch1,
4170 Rcurrent_obj = R12_scratch2,
4171 Robj_to_lock = R17_tos,
4172 Rcurrent_obj_addr = R3_ARG1,
4173 Rlimit = R4_ARG2;
4174 Label Lfound, Lillegal_monitor_state;
4175
4176 // Check corner case: unbalanced monitorEnter / Exit.
4177 __ ld(Rlimit, 0, R1_SP);
4178 __ addi(Rlimit, Rlimit, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base
4179
4180 // Null pointer check.
4181 __ null_check_throw(Robj_to_lock, -1, R11_scratch1);
4182
4183 __ cmpld(CCR0, R26_monitor, Rlimit);
4184 __ bgt(CCR0, Lillegal_monitor_state);
4185
4186 // Find the corresponding slot in the monitors stack section.
4187 {
4188 Label Lloop;
4189
4190 // Start with topmost monitor.
4191 __ addi(Rcurrent_obj_addr, R26_monitor, BasicObjectLock::obj_offset_in_bytes());
4192 __ addi(Rlimit, Rlimit, BasicObjectLock::obj_offset_in_bytes());
4193 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4194 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4195
4196 __ bind(Lloop);
4197 // Is this entry for same obj?
4198 __ cmpd(CCR0, Rcurrent_obj, Robj_to_lock);
4199 __ beq(CCR0, Lfound);
4200
4201 // Check if last allocated BasicLockObj reached.
4202
4203 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4204 __ cmpld(CCR0, Rcurrent_obj_addr, Rlimit);
4205 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4206
4207 // Next iteration if unchecked BasicObjectLocks exist on the stack.
4208 __ ble(CCR0, Lloop);
4209 }
4210
4211 // Fell through without finding the basic obj lock => throw up!
4212 __ bind(Lillegal_monitor_state);
4213 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
4214 __ should_not_reach_here();
4215
4216 __ align(32, 12);
4217 __ bind(Lfound);
4218 __ addi(Rcurrent_monitor, Rcurrent_obj_addr,
4219 -(frame::interpreter_frame_monitor_size() * wordSize) - BasicObjectLock::obj_offset_in_bytes());
4220 __ unlock_object(Rcurrent_monitor);
4221 }
4222
4223 // ============================================================================
4224 // Wide bytecodes
4225
4226 // Wide instructions. Simply redirects to the wide entry point for that instruction.
4227 void TemplateTable::wide() {
4228 transition(vtos, vtos);
4229
4230 const Register Rtable = R11_scratch1,
4231 Rindex = R12_scratch2,
4232 Rtmp = R0;
4233
4234 __ lbz(Rindex, 1, R14_bcp);
4235
4236 __ load_dispatch_table(Rtable, Interpreter::_wentry_point);
4237
4238 __ slwi(Rindex, Rindex, LogBytesPerWord);
4239 __ ldx(Rtmp, Rtable, Rindex);
4240 __ mtctr(Rtmp);
4241 __ bctr();
4242 // Note: the bcp increment step is part of the individual wide bytecode implementations.
4243 }