1 /*
2 * Copyright (c) 2015, Oracle and/or its affiliates. All rights reserved.
3 * Copyright 2013, 2015 SAP AG. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "precompiled.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "interpreter/interpreterRuntime.hpp"
30 #include "interpreter/interp_masm.hpp"
31 #include "interpreter/templateInterpreter.hpp"
32 #include "interpreter/templateTable.hpp"
33 #include "memory/universe.inline.hpp"
34 #include "oops/objArrayKlass.hpp"
35 #include "oops/oop.inline.hpp"
36 #include "prims/methodHandles.hpp"
37 #include "runtime/sharedRuntime.hpp"
38 #include "runtime/stubRoutines.hpp"
39 #include "runtime/synchronizer.hpp"
40 #include "utilities/macros.hpp"
41
42 #ifndef CC_INTERP
43
44 #undef __
45 #define __ _masm->
46
47 // ============================================================================
48 // Misc helpers
49
50 // Do an oop store like *(base + index) = val OR *(base + offset) = val
51 // (only one of both variants is possible at the same time).
52 // Index can be noreg.
53 // Kills:
54 // Rbase, Rtmp
55 static void do_oop_store(InterpreterMacroAssembler* _masm,
56 Register Rbase,
57 RegisterOrConstant offset,
58 Register Rval, // Noreg means always null.
59 Register Rtmp1,
60 Register Rtmp2,
61 Register Rtmp3,
62 BarrierSet::Name barrier,
63 bool precise,
64 bool check_null) {
65 assert_different_registers(Rtmp1, Rtmp2, Rtmp3, Rval, Rbase);
66
67 switch (barrier) {
68 #if INCLUDE_ALL_GCS
69 case BarrierSet::G1SATBCTLogging:
70 {
71 // Load and record the previous value.
72 __ g1_write_barrier_pre(Rbase, offset,
73 Rtmp3, /* holder of pre_val ? */
74 Rtmp1, Rtmp2, false /* frame */);
75
76 Label Lnull, Ldone;
77 if (Rval != noreg) {
78 if (check_null) {
79 __ cmpdi(CCR0, Rval, 0);
80 __ beq(CCR0, Lnull);
81 }
82 __ store_heap_oop_not_null(Rval, offset, Rbase, /*Rval must stay uncompressed.*/ Rtmp1);
83 // Mark the card.
84 if (!(offset.is_constant() && offset.as_constant() == 0) && precise) {
85 __ add(Rbase, offset, Rbase);
86 }
87 __ g1_write_barrier_post(Rbase, Rval, Rtmp1, Rtmp2, Rtmp3, /*filtered (fast path)*/ &Ldone);
88 if (check_null) { __ b(Ldone); }
89 }
90
91 if (Rval == noreg || check_null) { // Store null oop.
92 Register Rnull = Rval;
93 __ bind(Lnull);
94 if (Rval == noreg) {
95 Rnull = Rtmp1;
96 __ li(Rnull, 0);
97 }
98 if (UseCompressedOops) {
99 __ stw(Rnull, offset, Rbase);
100 } else {
101 __ std(Rnull, offset, Rbase);
102 }
103 }
104 __ bind(Ldone);
105 }
106 break;
107 #endif // INCLUDE_ALL_GCS
108 case BarrierSet::CardTableModRef:
109 case BarrierSet::CardTableExtension:
110 {
111 Label Lnull, Ldone;
112 if (Rval != noreg) {
113 if (check_null) {
114 __ cmpdi(CCR0, Rval, 0);
115 __ beq(CCR0, Lnull);
116 }
117 __ store_heap_oop_not_null(Rval, offset, Rbase, /*Rval should better stay uncompressed.*/ Rtmp1);
118 // Mark the card.
119 if (!(offset.is_constant() && offset.as_constant() == 0) && precise) {
120 __ add(Rbase, offset, Rbase);
121 }
122 __ card_write_barrier_post(Rbase, Rval, Rtmp1);
123 if (check_null) {
124 __ b(Ldone);
125 }
126 }
127
128 if (Rval == noreg || check_null) { // Store null oop.
129 Register Rnull = Rval;
130 __ bind(Lnull);
131 if (Rval == noreg) {
132 Rnull = Rtmp1;
133 __ li(Rnull, 0);
134 }
135 if (UseCompressedOops) {
136 __ stw(Rnull, offset, Rbase);
137 } else {
138 __ std(Rnull, offset, Rbase);
139 }
140 }
141 __ bind(Ldone);
142 }
143 break;
144 case BarrierSet::ModRef:
145 ShouldNotReachHere();
146 break;
147 default:
148 ShouldNotReachHere();
149 }
150 }
151
152 // ============================================================================
153 // Platform-dependent initialization
154
155 void TemplateTable::pd_initialize() {
156 // No ppc64 specific initialization.
157 }
158
159 Address TemplateTable::at_bcp(int offset) {
160 // Not used on ppc.
161 ShouldNotReachHere();
162 return Address();
163 }
164
165 // Patches the current bytecode (ptr to it located in bcp)
166 // in the bytecode stream with a new one.
167 void TemplateTable::patch_bytecode(Bytecodes::Code new_bc, Register Rnew_bc, Register Rtemp, bool load_bc_into_bc_reg /*=true*/, int byte_no) {
168 // With sharing on, may need to test method flag.
169 if (!RewriteBytecodes) return;
170 Label L_patch_done;
171
172 switch (new_bc) {
173 case Bytecodes::_fast_aputfield:
174 case Bytecodes::_fast_bputfield:
175 case Bytecodes::_fast_cputfield:
176 case Bytecodes::_fast_dputfield:
177 case Bytecodes::_fast_fputfield:
178 case Bytecodes::_fast_iputfield:
179 case Bytecodes::_fast_lputfield:
180 case Bytecodes::_fast_sputfield:
181 {
182 // We skip bytecode quickening for putfield instructions when
183 // the put_code written to the constant pool cache is zero.
184 // This is required so that every execution of this instruction
185 // calls out to InterpreterRuntime::resolve_get_put to do
186 // additional, required work.
187 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
188 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
189 __ get_cache_and_index_at_bcp(Rtemp /* dst = cache */, 1);
190 // ((*(cache+indices))>>((1+byte_no)*8))&0xFF:
191 #if defined(VM_LITTLE_ENDIAN)
192 __ lbz(Rnew_bc, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 1 + byte_no, Rtemp);
193 #else
194 __ lbz(Rnew_bc, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 7 - (1 + byte_no), Rtemp);
195 #endif
196 __ cmpwi(CCR0, Rnew_bc, 0);
197 __ li(Rnew_bc, (unsigned int)(unsigned char)new_bc);
198 __ beq(CCR0, L_patch_done);
199 // __ isync(); // acquire not needed
200 break;
201 }
202
203 default:
204 assert(byte_no == -1, "sanity");
205 if (load_bc_into_bc_reg) {
206 __ li(Rnew_bc, (unsigned int)(unsigned char)new_bc);
207 }
208 }
209
210 if (JvmtiExport::can_post_breakpoint()) {
211 Label L_fast_patch;
212 __ lbz(Rtemp, 0, R14_bcp);
213 __ cmpwi(CCR0, Rtemp, (unsigned int)(unsigned char)Bytecodes::_breakpoint);
214 __ bne(CCR0, L_fast_patch);
215 // Perform the quickening, slowly, in the bowels of the breakpoint table.
216 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), R19_method, R14_bcp, Rnew_bc);
217 __ b(L_patch_done);
218 __ bind(L_fast_patch);
219 }
220
221 // Patch bytecode.
222 __ stb(Rnew_bc, 0, R14_bcp);
223
224 __ bind(L_patch_done);
225 }
226
227 // ============================================================================
228 // Individual instructions
229
230 void TemplateTable::nop() {
231 transition(vtos, vtos);
232 // Nothing to do.
233 }
234
235 void TemplateTable::shouldnotreachhere() {
236 transition(vtos, vtos);
237 __ stop("shouldnotreachhere bytecode");
238 }
239
240 void TemplateTable::aconst_null() {
241 transition(vtos, atos);
242 __ li(R17_tos, 0);
243 }
244
245 void TemplateTable::iconst(int value) {
246 transition(vtos, itos);
247 assert(value >= -1 && value <= 5, "");
248 __ li(R17_tos, value);
249 }
250
251 void TemplateTable::lconst(int value) {
252 transition(vtos, ltos);
253 assert(value >= -1 && value <= 5, "");
254 __ li(R17_tos, value);
255 }
256
257 void TemplateTable::fconst(int value) {
258 transition(vtos, ftos);
259 static float zero = 0.0;
260 static float one = 1.0;
261 static float two = 2.0;
262 switch (value) {
263 default: ShouldNotReachHere();
264 case 0: {
265 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&zero, R0, true);
266 __ lfs(F15_ftos, simm16_offset, R11_scratch1);
267 break;
268 }
269 case 1: {
270 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&one, R0, true);
271 __ lfs(F15_ftos, simm16_offset, R11_scratch1);
272 break;
273 }
274 case 2: {
275 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&two, R0, true);
276 __ lfs(F15_ftos, simm16_offset, R11_scratch1);
277 break;
278 }
279 }
280 }
281
282 void TemplateTable::dconst(int value) {
283 transition(vtos, dtos);
284 static double zero = 0.0;
285 static double one = 1.0;
286 switch (value) {
287 case 0: {
288 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&zero, R0, true);
289 __ lfd(F15_ftos, simm16_offset, R11_scratch1);
290 break;
291 }
292 case 1: {
293 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&one, R0, true);
294 __ lfd(F15_ftos, simm16_offset, R11_scratch1);
295 break;
296 }
297 default: ShouldNotReachHere();
298 }
299 }
300
301 void TemplateTable::bipush() {
302 transition(vtos, itos);
303 __ lbz(R17_tos, 1, R14_bcp);
304 __ extsb(R17_tos, R17_tos);
305 }
306
307 void TemplateTable::sipush() {
308 transition(vtos, itos);
309 __ get_2_byte_integer_at_bcp(1, R17_tos, InterpreterMacroAssembler::Signed);
310 }
311
312 void TemplateTable::ldc(bool wide) {
313 Register Rscratch1 = R11_scratch1,
314 Rscratch2 = R12_scratch2,
315 Rcpool = R3_ARG1;
316
317 transition(vtos, vtos);
318 Label notInt, notClass, exit;
319
320 __ get_cpool_and_tags(Rcpool, Rscratch2); // Set Rscratch2 = &tags.
321 if (wide) { // Read index.
322 __ get_2_byte_integer_at_bcp(1, Rscratch1, InterpreterMacroAssembler::Unsigned);
323 } else {
324 __ lbz(Rscratch1, 1, R14_bcp);
325 }
326
327 const int base_offset = ConstantPool::header_size() * wordSize;
328 const int tags_offset = Array<u1>::base_offset_in_bytes();
329
330 // Get type from tags.
331 __ addi(Rscratch2, Rscratch2, tags_offset);
332 __ lbzx(Rscratch2, Rscratch2, Rscratch1);
333
334 __ cmpwi(CCR0, Rscratch2, JVM_CONSTANT_UnresolvedClass); // Unresolved class?
335 __ cmpwi(CCR1, Rscratch2, JVM_CONSTANT_UnresolvedClassInError); // Unresolved class in error state?
336 __ cror(CCR0, Assembler::equal, CCR1, Assembler::equal);
337
338 // Resolved class - need to call vm to get java mirror of the class.
339 __ cmpwi(CCR1, Rscratch2, JVM_CONSTANT_Class);
340 __ crnor(CCR0, Assembler::equal, CCR1, Assembler::equal); // Neither resolved class nor unresolved case from above?
341 __ beq(CCR0, notClass);
342
343 __ li(R4, wide ? 1 : 0);
344 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), R4);
345 __ push(atos);
346 __ b(exit);
347
348 __ align(32, 12);
349 __ bind(notClass);
350 __ addi(Rcpool, Rcpool, base_offset);
351 __ sldi(Rscratch1, Rscratch1, LogBytesPerWord);
352 __ cmpdi(CCR0, Rscratch2, JVM_CONSTANT_Integer);
353 __ bne(CCR0, notInt);
354 __ lwax(R17_tos, Rcpool, Rscratch1);
355 __ push(itos);
356 __ b(exit);
357
358 __ align(32, 12);
359 __ bind(notInt);
360 #ifdef ASSERT
361 // String and Object are rewritten to fast_aldc
362 __ cmpdi(CCR0, Rscratch2, JVM_CONSTANT_Float);
363 __ asm_assert_eq("unexpected type", 0x8765);
364 #endif
365 __ lfsx(F15_ftos, Rcpool, Rscratch1);
366 __ push(ftos);
367
368 __ align(32, 12);
369 __ bind(exit);
370 }
371
372 // Fast path for caching oop constants.
373 void TemplateTable::fast_aldc(bool wide) {
374 transition(vtos, atos);
375
376 int index_size = wide ? sizeof(u2) : sizeof(u1);
377 const Register Rscratch = R11_scratch1;
378 Label resolved;
379
380 // We are resolved if the resolved reference cache entry contains a
381 // non-null object (CallSite, etc.)
382 __ get_cache_index_at_bcp(Rscratch, 1, index_size); // Load index.
383 __ load_resolved_reference_at_index(R17_tos, Rscratch);
384 __ cmpdi(CCR0, R17_tos, 0);
385 __ bne(CCR0, resolved);
386 __ load_const_optimized(R3_ARG1, (int)bytecode());
387
388 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
389
390 // First time invocation - must resolve first.
391 __ call_VM(R17_tos, entry, R3_ARG1);
392
393 __ align(32, 12);
394 __ bind(resolved);
395 __ verify_oop(R17_tos);
396 }
397
398 void TemplateTable::ldc2_w() {
399 transition(vtos, vtos);
400 Label Llong, Lexit;
401
402 Register Rindex = R11_scratch1,
403 Rcpool = R12_scratch2,
404 Rtag = R3_ARG1;
405 __ get_cpool_and_tags(Rcpool, Rtag);
406 __ get_2_byte_integer_at_bcp(1, Rindex, InterpreterMacroAssembler::Unsigned);
407
408 const int base_offset = ConstantPool::header_size() * wordSize;
409 const int tags_offset = Array<u1>::base_offset_in_bytes();
410 // Get type from tags.
411 __ addi(Rcpool, Rcpool, base_offset);
412 __ addi(Rtag, Rtag, tags_offset);
413
414 __ lbzx(Rtag, Rtag, Rindex);
415
416 __ sldi(Rindex, Rindex, LogBytesPerWord);
417 __ cmpdi(CCR0, Rtag, JVM_CONSTANT_Double);
418 __ bne(CCR0, Llong);
419 // A double can be placed at word-aligned locations in the constant pool.
420 // Check out Conversions.java for an example.
421 // Also ConstantPool::header_size() is 20, which makes it very difficult
422 // to double-align double on the constant pool. SG, 11/7/97
423 __ lfdx(F15_ftos, Rcpool, Rindex);
424 __ push(dtos);
425 __ b(Lexit);
426
427 __ bind(Llong);
428 __ ldx(R17_tos, Rcpool, Rindex);
429 __ push(ltos);
430
431 __ bind(Lexit);
432 }
433
434 // Get the locals index located in the bytecode stream at bcp + offset.
435 void TemplateTable::locals_index(Register Rdst, int offset) {
436 __ lbz(Rdst, offset, R14_bcp);
437 }
438
439 void TemplateTable::iload() {
440 transition(vtos, itos);
441
442 // Get the local value into tos
443 const Register Rindex = R22_tmp2;
444 locals_index(Rindex);
445
446 // Rewrite iload,iload pair into fast_iload2
447 // iload,caload pair into fast_icaload
448 if (RewriteFrequentPairs) {
449 Label Lrewrite, Ldone;
450 Register Rnext_byte = R3_ARG1,
451 Rrewrite_to = R6_ARG4,
452 Rscratch = R11_scratch1;
453
454 // get next byte
455 __ lbz(Rnext_byte, Bytecodes::length_for(Bytecodes::_iload), R14_bcp);
456
457 // if _iload, wait to rewrite to iload2. We only want to rewrite the
458 // last two iloads in a pair. Comparing against fast_iload means that
459 // the next bytecode is neither an iload or a caload, and therefore
460 // an iload pair.
461 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_iload);
462 __ beq(CCR0, Ldone);
463
464 __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_iload);
465 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iload2);
466 __ beq(CCR1, Lrewrite);
467
468 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_caload);
469 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_icaload);
470 __ beq(CCR0, Lrewrite);
471
472 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iload);
473
474 __ bind(Lrewrite);
475 patch_bytecode(Bytecodes::_iload, Rrewrite_to, Rscratch, false);
476 __ bind(Ldone);
477 }
478
479 __ load_local_int(R17_tos, Rindex, Rindex);
480 }
481
482 // Load 2 integers in a row without dispatching
483 void TemplateTable::fast_iload2() {
484 transition(vtos, itos);
485
486 __ lbz(R3_ARG1, 1, R14_bcp);
487 __ lbz(R17_tos, Bytecodes::length_for(Bytecodes::_iload) + 1, R14_bcp);
488
489 __ load_local_int(R3_ARG1, R11_scratch1, R3_ARG1);
490 __ load_local_int(R17_tos, R12_scratch2, R17_tos);
491 __ push_i(R3_ARG1);
492 }
493
494 void TemplateTable::fast_iload() {
495 transition(vtos, itos);
496 // Get the local value into tos
497
498 const Register Rindex = R11_scratch1;
499 locals_index(Rindex);
500 __ load_local_int(R17_tos, Rindex, Rindex);
501 }
502
503 // Load a local variable type long from locals area to TOS cache register.
504 // Local index resides in bytecodestream.
505 void TemplateTable::lload() {
506 transition(vtos, ltos);
507
508 const Register Rindex = R11_scratch1;
509 locals_index(Rindex);
510 __ load_local_long(R17_tos, Rindex, Rindex);
511 }
512
513 void TemplateTable::fload() {
514 transition(vtos, ftos);
515
516 const Register Rindex = R11_scratch1;
517 locals_index(Rindex);
518 __ load_local_float(F15_ftos, Rindex, Rindex);
519 }
520
521 void TemplateTable::dload() {
522 transition(vtos, dtos);
523
524 const Register Rindex = R11_scratch1;
525 locals_index(Rindex);
526 __ load_local_double(F15_ftos, Rindex, Rindex);
527 }
528
529 void TemplateTable::aload() {
530 transition(vtos, atos);
531
532 const Register Rindex = R11_scratch1;
533 locals_index(Rindex);
534 __ load_local_ptr(R17_tos, Rindex, Rindex);
535 }
536
537 void TemplateTable::locals_index_wide(Register Rdst) {
538 // Offset is 2, not 1, because Lbcp points to wide prefix code.
539 __ get_2_byte_integer_at_bcp(2, Rdst, InterpreterMacroAssembler::Unsigned);
540 }
541
542 void TemplateTable::wide_iload() {
543 // Get the local value into tos.
544
545 const Register Rindex = R11_scratch1;
546 locals_index_wide(Rindex);
547 __ load_local_int(R17_tos, Rindex, Rindex);
548 }
549
550 void TemplateTable::wide_lload() {
551 transition(vtos, ltos);
552
553 const Register Rindex = R11_scratch1;
554 locals_index_wide(Rindex);
555 __ load_local_long(R17_tos, Rindex, Rindex);
556 }
557
558 void TemplateTable::wide_fload() {
559 transition(vtos, ftos);
560
561 const Register Rindex = R11_scratch1;
562 locals_index_wide(Rindex);
563 __ load_local_float(F15_ftos, Rindex, Rindex);
564 }
565
566 void TemplateTable::wide_dload() {
567 transition(vtos, dtos);
568
569 const Register Rindex = R11_scratch1;
570 locals_index_wide(Rindex);
571 __ load_local_double(F15_ftos, Rindex, Rindex);
572 }
573
574 void TemplateTable::wide_aload() {
575 transition(vtos, atos);
576
577 const Register Rindex = R11_scratch1;
578 locals_index_wide(Rindex);
579 __ load_local_ptr(R17_tos, Rindex, Rindex);
580 }
581
582 void TemplateTable::iaload() {
583 transition(itos, itos);
584
585 const Register Rload_addr = R3_ARG1,
586 Rarray = R4_ARG2,
587 Rtemp = R5_ARG3;
588 __ index_check(Rarray, R17_tos /* index */, LogBytesPerInt, Rtemp, Rload_addr);
589 __ lwa(R17_tos, arrayOopDesc::base_offset_in_bytes(T_INT), Rload_addr);
590 }
591
592 void TemplateTable::laload() {
593 transition(itos, ltos);
594
595 const Register Rload_addr = R3_ARG1,
596 Rarray = R4_ARG2,
597 Rtemp = R5_ARG3;
598 __ index_check(Rarray, R17_tos /* index */, LogBytesPerLong, Rtemp, Rload_addr);
599 __ ld(R17_tos, arrayOopDesc::base_offset_in_bytes(T_LONG), Rload_addr);
600 }
601
602 void TemplateTable::faload() {
603 transition(itos, ftos);
604
605 const Register Rload_addr = R3_ARG1,
606 Rarray = R4_ARG2,
607 Rtemp = R5_ARG3;
608 __ index_check(Rarray, R17_tos /* index */, LogBytesPerInt, Rtemp, Rload_addr);
609 __ lfs(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Rload_addr);
610 }
611
612 void TemplateTable::daload() {
613 transition(itos, dtos);
614
615 const Register Rload_addr = R3_ARG1,
616 Rarray = R4_ARG2,
617 Rtemp = R5_ARG3;
618 __ index_check(Rarray, R17_tos /* index */, LogBytesPerLong, Rtemp, Rload_addr);
619 __ lfd(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Rload_addr);
620 }
621
622 void TemplateTable::aaload() {
623 transition(itos, atos);
624
625 // tos: index
626 // result tos: array
627 const Register Rload_addr = R3_ARG1,
628 Rarray = R4_ARG2,
629 Rtemp = R5_ARG3;
630 __ index_check(Rarray, R17_tos /* index */, UseCompressedOops ? 2 : LogBytesPerWord, Rtemp, Rload_addr);
631 __ load_heap_oop(R17_tos, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Rload_addr);
632 __ verify_oop(R17_tos);
633 //__ dcbt(R17_tos); // prefetch
634 }
635
636 void TemplateTable::baload() {
637 transition(itos, itos);
638
639 const Register Rload_addr = R3_ARG1,
640 Rarray = R4_ARG2,
641 Rtemp = R5_ARG3;
642 __ index_check(Rarray, R17_tos /* index */, 0, Rtemp, Rload_addr);
643 __ lbz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_BYTE), Rload_addr);
644 __ extsb(R17_tos, R17_tos);
645 }
646
647 void TemplateTable::caload() {
648 transition(itos, itos);
649
650 const Register Rload_addr = R3_ARG1,
651 Rarray = R4_ARG2,
652 Rtemp = R5_ARG3;
653 __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
654 __ lhz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rload_addr);
655 }
656
657 // Iload followed by caload frequent pair.
658 void TemplateTable::fast_icaload() {
659 transition(vtos, itos);
660
661 const Register Rload_addr = R3_ARG1,
662 Rarray = R4_ARG2,
663 Rtemp = R11_scratch1;
664
665 locals_index(R17_tos);
666 __ load_local_int(R17_tos, Rtemp, R17_tos);
667 __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
668 __ lhz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rload_addr);
669 }
670
671 void TemplateTable::saload() {
672 transition(itos, itos);
673
674 const Register Rload_addr = R11_scratch1,
675 Rarray = R12_scratch2,
676 Rtemp = R3_ARG1;
677 __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
678 __ lha(R17_tos, arrayOopDesc::base_offset_in_bytes(T_SHORT), Rload_addr);
679 }
680
681 void TemplateTable::iload(int n) {
682 transition(vtos, itos);
683
684 __ lwz(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
685 }
686
687 void TemplateTable::lload(int n) {
688 transition(vtos, ltos);
689
690 __ ld(R17_tos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
691 }
692
693 void TemplateTable::fload(int n) {
694 transition(vtos, ftos);
695
696 __ lfs(F15_ftos, Interpreter::local_offset_in_bytes(n), R18_locals);
697 }
698
699 void TemplateTable::dload(int n) {
700 transition(vtos, dtos);
701
702 __ lfd(F15_ftos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
703 }
704
705 void TemplateTable::aload(int n) {
706 transition(vtos, atos);
707
708 __ ld(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
709 }
710
711 void TemplateTable::aload_0() {
712 transition(vtos, atos);
713 // According to bytecode histograms, the pairs:
714 //
715 // _aload_0, _fast_igetfield
716 // _aload_0, _fast_agetfield
717 // _aload_0, _fast_fgetfield
718 //
719 // occur frequently. If RewriteFrequentPairs is set, the (slow)
720 // _aload_0 bytecode checks if the next bytecode is either
721 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
722 // rewrites the current bytecode into a pair bytecode; otherwise it
723 // rewrites the current bytecode into _0 that doesn't do
724 // the pair check anymore.
725 //
726 // Note: If the next bytecode is _getfield, the rewrite must be
727 // delayed, otherwise we may miss an opportunity for a pair.
728 //
729 // Also rewrite frequent pairs
730 // aload_0, aload_1
731 // aload_0, iload_1
732 // These bytecodes with a small amount of code are most profitable
733 // to rewrite.
734
735 if (RewriteFrequentPairs) {
736
737 Label Lrewrite, Ldont_rewrite;
738 Register Rnext_byte = R3_ARG1,
739 Rrewrite_to = R6_ARG4,
740 Rscratch = R11_scratch1;
741
742 // Get next byte.
743 __ lbz(Rnext_byte, Bytecodes::length_for(Bytecodes::_aload_0), R14_bcp);
744
745 // If _getfield, wait to rewrite. We only want to rewrite the last two bytecodes in a pair.
746 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_getfield);
747 __ beq(CCR0, Ldont_rewrite);
748
749 __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_igetfield);
750 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iaccess_0);
751 __ beq(CCR1, Lrewrite);
752
753 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_agetfield);
754 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_aaccess_0);
755 __ beq(CCR0, Lrewrite);
756
757 __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_fgetfield);
758 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_faccess_0);
759 __ beq(CCR1, Lrewrite);
760
761 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_aload_0);
762
763 __ bind(Lrewrite);
764 patch_bytecode(Bytecodes::_aload_0, Rrewrite_to, Rscratch, false);
765 __ bind(Ldont_rewrite);
766 }
767
768 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
769 aload(0);
770 }
771
772 void TemplateTable::istore() {
773 transition(itos, vtos);
774
775 const Register Rindex = R11_scratch1;
776 locals_index(Rindex);
777 __ store_local_int(R17_tos, Rindex);
778 }
779
780 void TemplateTable::lstore() {
781 transition(ltos, vtos);
782 const Register Rindex = R11_scratch1;
783 locals_index(Rindex);
784 __ store_local_long(R17_tos, Rindex);
785 }
786
787 void TemplateTable::fstore() {
788 transition(ftos, vtos);
789
790 const Register Rindex = R11_scratch1;
791 locals_index(Rindex);
792 __ store_local_float(F15_ftos, Rindex);
793 }
794
795 void TemplateTable::dstore() {
796 transition(dtos, vtos);
797
798 const Register Rindex = R11_scratch1;
799 locals_index(Rindex);
800 __ store_local_double(F15_ftos, Rindex);
801 }
802
803 void TemplateTable::astore() {
804 transition(vtos, vtos);
805
806 const Register Rindex = R11_scratch1;
807 __ pop_ptr();
808 __ verify_oop_or_return_address(R17_tos, Rindex);
809 locals_index(Rindex);
810 __ store_local_ptr(R17_tos, Rindex);
811 }
812
813 void TemplateTable::wide_istore() {
814 transition(vtos, vtos);
815
816 const Register Rindex = R11_scratch1;
817 __ pop_i();
818 locals_index_wide(Rindex);
819 __ store_local_int(R17_tos, Rindex);
820 }
821
822 void TemplateTable::wide_lstore() {
823 transition(vtos, vtos);
824
825 const Register Rindex = R11_scratch1;
826 __ pop_l();
827 locals_index_wide(Rindex);
828 __ store_local_long(R17_tos, Rindex);
829 }
830
831 void TemplateTable::wide_fstore() {
832 transition(vtos, vtos);
833
834 const Register Rindex = R11_scratch1;
835 __ pop_f();
836 locals_index_wide(Rindex);
837 __ store_local_float(F15_ftos, Rindex);
838 }
839
840 void TemplateTable::wide_dstore() {
841 transition(vtos, vtos);
842
843 const Register Rindex = R11_scratch1;
844 __ pop_d();
845 locals_index_wide(Rindex);
846 __ store_local_double(F15_ftos, Rindex);
847 }
848
849 void TemplateTable::wide_astore() {
850 transition(vtos, vtos);
851
852 const Register Rindex = R11_scratch1;
853 __ pop_ptr();
854 __ verify_oop_or_return_address(R17_tos, Rindex);
855 locals_index_wide(Rindex);
856 __ store_local_ptr(R17_tos, Rindex);
857 }
858
859 void TemplateTable::iastore() {
860 transition(itos, vtos);
861
862 const Register Rindex = R3_ARG1,
863 Rstore_addr = R4_ARG2,
864 Rarray = R5_ARG3,
865 Rtemp = R6_ARG4;
866 __ pop_i(Rindex);
867 __ index_check(Rarray, Rindex, LogBytesPerInt, Rtemp, Rstore_addr);
868 __ stw(R17_tos, arrayOopDesc::base_offset_in_bytes(T_INT), Rstore_addr);
869 }
870
871 void TemplateTable::lastore() {
872 transition(ltos, vtos);
873
874 const Register Rindex = R3_ARG1,
875 Rstore_addr = R4_ARG2,
876 Rarray = R5_ARG3,
877 Rtemp = R6_ARG4;
878 __ pop_i(Rindex);
879 __ index_check(Rarray, Rindex, LogBytesPerLong, Rtemp, Rstore_addr);
880 __ std(R17_tos, arrayOopDesc::base_offset_in_bytes(T_LONG), Rstore_addr);
881 }
882
883 void TemplateTable::fastore() {
884 transition(ftos, vtos);
885
886 const Register Rindex = R3_ARG1,
887 Rstore_addr = R4_ARG2,
888 Rarray = R5_ARG3,
889 Rtemp = R6_ARG4;
890 __ pop_i(Rindex);
891 __ index_check(Rarray, Rindex, LogBytesPerInt, Rtemp, Rstore_addr);
892 __ stfs(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Rstore_addr);
893 }
894
895 void TemplateTable::dastore() {
896 transition(dtos, vtos);
897
898 const Register Rindex = R3_ARG1,
899 Rstore_addr = R4_ARG2,
900 Rarray = R5_ARG3,
901 Rtemp = R6_ARG4;
902 __ pop_i(Rindex);
903 __ index_check(Rarray, Rindex, LogBytesPerLong, Rtemp, Rstore_addr);
904 __ stfd(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Rstore_addr);
905 }
906
907 // Pop 3 values from the stack and...
908 void TemplateTable::aastore() {
909 transition(vtos, vtos);
910
911 Label Lstore_ok, Lis_null, Ldone;
912 const Register Rindex = R3_ARG1,
913 Rarray = R4_ARG2,
914 Rscratch = R11_scratch1,
915 Rscratch2 = R12_scratch2,
916 Rarray_klass = R5_ARG3,
917 Rarray_element_klass = Rarray_klass,
918 Rvalue_klass = R6_ARG4,
919 Rstore_addr = R31; // Use register which survives VM call.
920
921 __ ld(R17_tos, Interpreter::expr_offset_in_bytes(0), R15_esp); // Get value to store.
922 __ lwz(Rindex, Interpreter::expr_offset_in_bytes(1), R15_esp); // Get index.
923 __ ld(Rarray, Interpreter::expr_offset_in_bytes(2), R15_esp); // Get array.
924
925 __ verify_oop(R17_tos);
926 __ index_check_without_pop(Rarray, Rindex, UseCompressedOops ? 2 : LogBytesPerWord, Rscratch, Rstore_addr);
927 // Rindex is dead!
928 Register Rscratch3 = Rindex;
929
930 // Do array store check - check for NULL value first.
931 __ cmpdi(CCR0, R17_tos, 0);
932 __ beq(CCR0, Lis_null);
933
934 __ load_klass(Rarray_klass, Rarray);
935 __ load_klass(Rvalue_klass, R17_tos);
936
937 // Do fast instanceof cache test.
938 __ ld(Rarray_element_klass, in_bytes(ObjArrayKlass::element_klass_offset()), Rarray_klass);
939
940 // Generate a fast subtype check. Branch to store_ok if no failure. Throw if failure.
941 __ gen_subtype_check(Rvalue_klass /*subklass*/, Rarray_element_klass /*superklass*/, Rscratch, Rscratch2, Rscratch3, Lstore_ok);
942
943 // Fell through: subtype check failed => throw an exception.
944 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArrayStoreException_entry);
945 __ mtctr(R11_scratch1);
946 __ bctr();
947
948 __ bind(Lis_null);
949 do_oop_store(_masm, Rstore_addr, arrayOopDesc::base_offset_in_bytes(T_OBJECT), noreg /* 0 */,
950 Rscratch, Rscratch2, Rscratch3, _bs->kind(), true /* precise */, false /* check_null */);
951 __ profile_null_seen(Rscratch, Rscratch2);
952 __ b(Ldone);
953
954 // Store is OK.
955 __ bind(Lstore_ok);
956 do_oop_store(_masm, Rstore_addr, arrayOopDesc::base_offset_in_bytes(T_OBJECT), R17_tos /* value */,
957 Rscratch, Rscratch2, Rscratch3, _bs->kind(), true /* precise */, false /* check_null */);
958
959 __ bind(Ldone);
960 // Adjust sp (pops array, index and value).
961 __ addi(R15_esp, R15_esp, 3 * Interpreter::stackElementSize);
962 }
963
964 void TemplateTable::bastore() {
965 transition(itos, vtos);
966
967 const Register Rindex = R11_scratch1,
968 Rarray = R12_scratch2,
969 Rscratch = R3_ARG1;
970 __ pop_i(Rindex);
971 // tos: val
972 // Rarray: array ptr (popped by index_check)
973 __ index_check(Rarray, Rindex, 0, Rscratch, Rarray);
974 __ stb(R17_tos, arrayOopDesc::base_offset_in_bytes(T_BYTE), Rarray);
975 }
976
977 void TemplateTable::castore() {
978 transition(itos, vtos);
979
980 const Register Rindex = R11_scratch1,
981 Rarray = R12_scratch2,
982 Rscratch = R3_ARG1;
983 __ pop_i(Rindex);
984 // tos: val
985 // Rarray: array ptr (popped by index_check)
986 __ index_check(Rarray, Rindex, LogBytesPerShort, Rscratch, Rarray);
987 __ sth(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rarray);
988 }
989
990 void TemplateTable::sastore() {
991 castore();
992 }
993
994 void TemplateTable::istore(int n) {
995 transition(itos, vtos);
996 __ stw(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
997 }
998
999 void TemplateTable::lstore(int n) {
1000 transition(ltos, vtos);
1001 __ std(R17_tos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
1002 }
1003
1004 void TemplateTable::fstore(int n) {
1005 transition(ftos, vtos);
1006 __ stfs(F15_ftos, Interpreter::local_offset_in_bytes(n), R18_locals);
1007 }
1008
1009 void TemplateTable::dstore(int n) {
1010 transition(dtos, vtos);
1011 __ stfd(F15_ftos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
1012 }
1013
1014 void TemplateTable::astore(int n) {
1015 transition(vtos, vtos);
1016
1017 __ pop_ptr();
1018 __ verify_oop_or_return_address(R17_tos, R11_scratch1);
1019 __ std(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
1020 }
1021
1022 void TemplateTable::pop() {
1023 transition(vtos, vtos);
1024
1025 __ addi(R15_esp, R15_esp, Interpreter::stackElementSize);
1026 }
1027
1028 void TemplateTable::pop2() {
1029 transition(vtos, vtos);
1030
1031 __ addi(R15_esp, R15_esp, Interpreter::stackElementSize * 2);
1032 }
1033
1034 void TemplateTable::dup() {
1035 transition(vtos, vtos);
1036
1037 __ ld(R11_scratch1, Interpreter::stackElementSize, R15_esp);
1038 __ push_ptr(R11_scratch1);
1039 }
1040
1041 void TemplateTable::dup_x1() {
1042 transition(vtos, vtos);
1043
1044 Register Ra = R11_scratch1,
1045 Rb = R12_scratch2;
1046 // stack: ..., a, b
1047 __ ld(Rb, Interpreter::stackElementSize, R15_esp);
1048 __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1049 __ std(Rb, Interpreter::stackElementSize * 2, R15_esp);
1050 __ std(Ra, Interpreter::stackElementSize, R15_esp);
1051 __ push_ptr(Rb);
1052 // stack: ..., b, a, b
1053 }
1054
1055 void TemplateTable::dup_x2() {
1056 transition(vtos, vtos);
1057
1058 Register Ra = R11_scratch1,
1059 Rb = R12_scratch2,
1060 Rc = R3_ARG1;
1061
1062 // stack: ..., a, b, c
1063 __ ld(Rc, Interpreter::stackElementSize, R15_esp); // load c
1064 __ ld(Ra, Interpreter::stackElementSize * 3, R15_esp); // load a
1065 __ std(Rc, Interpreter::stackElementSize * 3, R15_esp); // store c in a
1066 __ ld(Rb, Interpreter::stackElementSize * 2, R15_esp); // load b
1067 // stack: ..., c, b, c
1068 __ std(Ra, Interpreter::stackElementSize * 2, R15_esp); // store a in b
1069 // stack: ..., c, a, c
1070 __ std(Rb, Interpreter::stackElementSize, R15_esp); // store b in c
1071 __ push_ptr(Rc); // push c
1072 // stack: ..., c, a, b, c
1073 }
1074
1075 void TemplateTable::dup2() {
1076 transition(vtos, vtos);
1077
1078 Register Ra = R11_scratch1,
1079 Rb = R12_scratch2;
1080 // stack: ..., a, b
1081 __ ld(Rb, Interpreter::stackElementSize, R15_esp);
1082 __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1083 __ push_2ptrs(Ra, Rb);
1084 // stack: ..., a, b, a, b
1085 }
1086
1087 void TemplateTable::dup2_x1() {
1088 transition(vtos, vtos);
1089
1090 Register Ra = R11_scratch1,
1091 Rb = R12_scratch2,
1092 Rc = R3_ARG1;
1093 // stack: ..., a, b, c
1094 __ ld(Rc, Interpreter::stackElementSize, R15_esp);
1095 __ ld(Rb, Interpreter::stackElementSize * 2, R15_esp);
1096 __ std(Rc, Interpreter::stackElementSize * 2, R15_esp);
1097 __ ld(Ra, Interpreter::stackElementSize * 3, R15_esp);
1098 __ std(Ra, Interpreter::stackElementSize, R15_esp);
1099 __ std(Rb, Interpreter::stackElementSize * 3, R15_esp);
1100 // stack: ..., b, c, a
1101 __ push_2ptrs(Rb, Rc);
1102 // stack: ..., b, c, a, b, c
1103 }
1104
1105 void TemplateTable::dup2_x2() {
1106 transition(vtos, vtos);
1107
1108 Register Ra = R11_scratch1,
1109 Rb = R12_scratch2,
1110 Rc = R3_ARG1,
1111 Rd = R4_ARG2;
1112 // stack: ..., a, b, c, d
1113 __ ld(Rb, Interpreter::stackElementSize * 3, R15_esp);
1114 __ ld(Rd, Interpreter::stackElementSize, R15_esp);
1115 __ std(Rb, Interpreter::stackElementSize, R15_esp); // store b in d
1116 __ std(Rd, Interpreter::stackElementSize * 3, R15_esp); // store d in b
1117 __ ld(Ra, Interpreter::stackElementSize * 4, R15_esp);
1118 __ ld(Rc, Interpreter::stackElementSize * 2, R15_esp);
1119 __ std(Ra, Interpreter::stackElementSize * 2, R15_esp); // store a in c
1120 __ std(Rc, Interpreter::stackElementSize * 4, R15_esp); // store c in a
1121 // stack: ..., c, d, a, b
1122 __ push_2ptrs(Rc, Rd);
1123 // stack: ..., c, d, a, b, c, d
1124 }
1125
1126 void TemplateTable::swap() {
1127 transition(vtos, vtos);
1128 // stack: ..., a, b
1129
1130 Register Ra = R11_scratch1,
1131 Rb = R12_scratch2;
1132 // stack: ..., a, b
1133 __ ld(Rb, Interpreter::stackElementSize, R15_esp);
1134 __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1135 __ std(Rb, Interpreter::stackElementSize * 2, R15_esp);
1136 __ std(Ra, Interpreter::stackElementSize, R15_esp);
1137 // stack: ..., b, a
1138 }
1139
1140 void TemplateTable::iop2(Operation op) {
1141 transition(itos, itos);
1142
1143 Register Rscratch = R11_scratch1;
1144
1145 __ pop_i(Rscratch);
1146 // tos = number of bits to shift
1147 // Rscratch = value to shift
1148 switch (op) {
1149 case add: __ add(R17_tos, Rscratch, R17_tos); break;
1150 case sub: __ sub(R17_tos, Rscratch, R17_tos); break;
1151 case mul: __ mullw(R17_tos, Rscratch, R17_tos); break;
1152 case _and: __ andr(R17_tos, Rscratch, R17_tos); break;
1153 case _or: __ orr(R17_tos, Rscratch, R17_tos); break;
1154 case _xor: __ xorr(R17_tos, Rscratch, R17_tos); break;
1155 case shl: __ rldicl(R17_tos, R17_tos, 0, 64-5); __ slw(R17_tos, Rscratch, R17_tos); break;
1156 case shr: __ rldicl(R17_tos, R17_tos, 0, 64-5); __ sraw(R17_tos, Rscratch, R17_tos); break;
1157 case ushr: __ rldicl(R17_tos, R17_tos, 0, 64-5); __ srw(R17_tos, Rscratch, R17_tos); break;
1158 default: ShouldNotReachHere();
1159 }
1160 }
1161
1162 void TemplateTable::lop2(Operation op) {
1163 transition(ltos, ltos);
1164
1165 Register Rscratch = R11_scratch1;
1166 __ pop_l(Rscratch);
1167 switch (op) {
1168 case add: __ add(R17_tos, Rscratch, R17_tos); break;
1169 case sub: __ sub(R17_tos, Rscratch, R17_tos); break;
1170 case _and: __ andr(R17_tos, Rscratch, R17_tos); break;
1171 case _or: __ orr(R17_tos, Rscratch, R17_tos); break;
1172 case _xor: __ xorr(R17_tos, Rscratch, R17_tos); break;
1173 default: ShouldNotReachHere();
1174 }
1175 }
1176
1177 void TemplateTable::idiv() {
1178 transition(itos, itos);
1179
1180 Label Lnormal, Lexception, Ldone;
1181 Register Rdividend = R11_scratch1; // Used by irem.
1182
1183 __ addi(R0, R17_tos, 1);
1184 __ cmplwi(CCR0, R0, 2);
1185 __ bgt(CCR0, Lnormal); // divisor <-1 or >1
1186
1187 __ cmpwi(CCR1, R17_tos, 0);
1188 __ beq(CCR1, Lexception); // divisor == 0
1189
1190 __ pop_i(Rdividend);
1191 __ mullw(R17_tos, Rdividend, R17_tos); // div by +/-1
1192 __ b(Ldone);
1193
1194 __ bind(Lexception);
1195 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArithmeticException_entry);
1196 __ mtctr(R11_scratch1);
1197 __ bctr();
1198
1199 __ align(32, 12);
1200 __ bind(Lnormal);
1201 __ pop_i(Rdividend);
1202 __ divw(R17_tos, Rdividend, R17_tos); // Can't divide minint/-1.
1203 __ bind(Ldone);
1204 }
1205
1206 void TemplateTable::irem() {
1207 transition(itos, itos);
1208
1209 __ mr(R12_scratch2, R17_tos);
1210 idiv();
1211 __ mullw(R17_tos, R17_tos, R12_scratch2);
1212 __ subf(R17_tos, R17_tos, R11_scratch1); // Dividend set by idiv.
1213 }
1214
1215 void TemplateTable::lmul() {
1216 transition(ltos, ltos);
1217
1218 __ pop_l(R11_scratch1);
1219 __ mulld(R17_tos, R11_scratch1, R17_tos);
1220 }
1221
1222 void TemplateTable::ldiv() {
1223 transition(ltos, ltos);
1224
1225 Label Lnormal, Lexception, Ldone;
1226 Register Rdividend = R11_scratch1; // Used by lrem.
1227
1228 __ addi(R0, R17_tos, 1);
1229 __ cmpldi(CCR0, R0, 2);
1230 __ bgt(CCR0, Lnormal); // divisor <-1 or >1
1231
1232 __ cmpdi(CCR1, R17_tos, 0);
1233 __ beq(CCR1, Lexception); // divisor == 0
1234
1235 __ pop_l(Rdividend);
1236 __ mulld(R17_tos, Rdividend, R17_tos); // div by +/-1
1237 __ b(Ldone);
1238
1239 __ bind(Lexception);
1240 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArithmeticException_entry);
1241 __ mtctr(R11_scratch1);
1242 __ bctr();
1243
1244 __ align(32, 12);
1245 __ bind(Lnormal);
1246 __ pop_l(Rdividend);
1247 __ divd(R17_tos, Rdividend, R17_tos); // Can't divide minint/-1.
1248 __ bind(Ldone);
1249 }
1250
1251 void TemplateTable::lrem() {
1252 transition(ltos, ltos);
1253
1254 __ mr(R12_scratch2, R17_tos);
1255 ldiv();
1256 __ mulld(R17_tos, R17_tos, R12_scratch2);
1257 __ subf(R17_tos, R17_tos, R11_scratch1); // Dividend set by ldiv.
1258 }
1259
1260 void TemplateTable::lshl() {
1261 transition(itos, ltos);
1262
1263 __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1264 __ pop_l(R11_scratch1);
1265 __ sld(R17_tos, R11_scratch1, R17_tos);
1266 }
1267
1268 void TemplateTable::lshr() {
1269 transition(itos, ltos);
1270
1271 __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1272 __ pop_l(R11_scratch1);
1273 __ srad(R17_tos, R11_scratch1, R17_tos);
1274 }
1275
1276 void TemplateTable::lushr() {
1277 transition(itos, ltos);
1278
1279 __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1280 __ pop_l(R11_scratch1);
1281 __ srd(R17_tos, R11_scratch1, R17_tos);
1282 }
1283
1284 void TemplateTable::fop2(Operation op) {
1285 transition(ftos, ftos);
1286
1287 switch (op) {
1288 case add: __ pop_f(F0_SCRATCH); __ fadds(F15_ftos, F0_SCRATCH, F15_ftos); break;
1289 case sub: __ pop_f(F0_SCRATCH); __ fsubs(F15_ftos, F0_SCRATCH, F15_ftos); break;
1290 case mul: __ pop_f(F0_SCRATCH); __ fmuls(F15_ftos, F0_SCRATCH, F15_ftos); break;
1291 case div: __ pop_f(F0_SCRATCH); __ fdivs(F15_ftos, F0_SCRATCH, F15_ftos); break;
1292 case rem:
1293 __ pop_f(F1_ARG1);
1294 __ fmr(F2_ARG2, F15_ftos);
1295 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem));
1296 __ fmr(F15_ftos, F1_RET);
1297 break;
1298
1299 default: ShouldNotReachHere();
1300 }
1301 }
1302
1303 void TemplateTable::dop2(Operation op) {
1304 transition(dtos, dtos);
1305
1306 switch (op) {
1307 case add: __ pop_d(F0_SCRATCH); __ fadd(F15_ftos, F0_SCRATCH, F15_ftos); break;
1308 case sub: __ pop_d(F0_SCRATCH); __ fsub(F15_ftos, F0_SCRATCH, F15_ftos); break;
1309 case mul: __ pop_d(F0_SCRATCH); __ fmul(F15_ftos, F0_SCRATCH, F15_ftos); break;
1310 case div: __ pop_d(F0_SCRATCH); __ fdiv(F15_ftos, F0_SCRATCH, F15_ftos); break;
1311 case rem:
1312 __ pop_d(F1_ARG1);
1313 __ fmr(F2_ARG2, F15_ftos);
1314 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem));
1315 __ fmr(F15_ftos, F1_RET);
1316 break;
1317
1318 default: ShouldNotReachHere();
1319 }
1320 }
1321
1322 // Negate the value in the TOS cache.
1323 void TemplateTable::ineg() {
1324 transition(itos, itos);
1325
1326 __ neg(R17_tos, R17_tos);
1327 }
1328
1329 // Negate the value in the TOS cache.
1330 void TemplateTable::lneg() {
1331 transition(ltos, ltos);
1332
1333 __ neg(R17_tos, R17_tos);
1334 }
1335
1336 void TemplateTable::fneg() {
1337 transition(ftos, ftos);
1338
1339 __ fneg(F15_ftos, F15_ftos);
1340 }
1341
1342 void TemplateTable::dneg() {
1343 transition(dtos, dtos);
1344
1345 __ fneg(F15_ftos, F15_ftos);
1346 }
1347
1348 // Increments a local variable in place.
1349 void TemplateTable::iinc() {
1350 transition(vtos, vtos);
1351
1352 const Register Rindex = R11_scratch1,
1353 Rincrement = R0,
1354 Rvalue = R12_scratch2;
1355
1356 locals_index(Rindex); // Load locals index from bytecode stream.
1357 __ lbz(Rincrement, 2, R14_bcp); // Load increment from the bytecode stream.
1358 __ extsb(Rincrement, Rincrement);
1359
1360 __ load_local_int(Rvalue, Rindex, Rindex); // Puts address of local into Rindex.
1361
1362 __ add(Rvalue, Rincrement, Rvalue);
1363 __ stw(Rvalue, 0, Rindex);
1364 }
1365
1366 void TemplateTable::wide_iinc() {
1367 transition(vtos, vtos);
1368
1369 Register Rindex = R11_scratch1,
1370 Rlocals_addr = Rindex,
1371 Rincr = R12_scratch2;
1372 locals_index_wide(Rindex);
1373 __ get_2_byte_integer_at_bcp(4, Rincr, InterpreterMacroAssembler::Signed);
1374 __ load_local_int(R17_tos, Rlocals_addr, Rindex);
1375 __ add(R17_tos, Rincr, R17_tos);
1376 __ stw(R17_tos, 0, Rlocals_addr);
1377 }
1378
1379 void TemplateTable::convert() {
1380 // %%%%% Factor this first part accross platforms
1381 #ifdef ASSERT
1382 TosState tos_in = ilgl;
1383 TosState tos_out = ilgl;
1384 switch (bytecode()) {
1385 case Bytecodes::_i2l: // fall through
1386 case Bytecodes::_i2f: // fall through
1387 case Bytecodes::_i2d: // fall through
1388 case Bytecodes::_i2b: // fall through
1389 case Bytecodes::_i2c: // fall through
1390 case Bytecodes::_i2s: tos_in = itos; break;
1391 case Bytecodes::_l2i: // fall through
1392 case Bytecodes::_l2f: // fall through
1393 case Bytecodes::_l2d: tos_in = ltos; break;
1394 case Bytecodes::_f2i: // fall through
1395 case Bytecodes::_f2l: // fall through
1396 case Bytecodes::_f2d: tos_in = ftos; break;
1397 case Bytecodes::_d2i: // fall through
1398 case Bytecodes::_d2l: // fall through
1399 case Bytecodes::_d2f: tos_in = dtos; break;
1400 default : ShouldNotReachHere();
1401 }
1402 switch (bytecode()) {
1403 case Bytecodes::_l2i: // fall through
1404 case Bytecodes::_f2i: // fall through
1405 case Bytecodes::_d2i: // fall through
1406 case Bytecodes::_i2b: // fall through
1407 case Bytecodes::_i2c: // fall through
1408 case Bytecodes::_i2s: tos_out = itos; break;
1409 case Bytecodes::_i2l: // fall through
1410 case Bytecodes::_f2l: // fall through
1411 case Bytecodes::_d2l: tos_out = ltos; break;
1412 case Bytecodes::_i2f: // fall through
1413 case Bytecodes::_l2f: // fall through
1414 case Bytecodes::_d2f: tos_out = ftos; break;
1415 case Bytecodes::_i2d: // fall through
1416 case Bytecodes::_l2d: // fall through
1417 case Bytecodes::_f2d: tos_out = dtos; break;
1418 default : ShouldNotReachHere();
1419 }
1420 transition(tos_in, tos_out);
1421 #endif
1422
1423 // Conversion
1424 Label done;
1425 switch (bytecode()) {
1426 case Bytecodes::_i2l:
1427 __ extsw(R17_tos, R17_tos);
1428 break;
1429
1430 case Bytecodes::_l2i:
1431 // Nothing to do, we'll continue to work with the lower bits.
1432 break;
1433
1434 case Bytecodes::_i2b:
1435 __ extsb(R17_tos, R17_tos);
1436 break;
1437
1438 case Bytecodes::_i2c:
1439 __ rldicl(R17_tos, R17_tos, 0, 64-2*8);
1440 break;
1441
1442 case Bytecodes::_i2s:
1443 __ extsh(R17_tos, R17_tos);
1444 break;
1445
1446 case Bytecodes::_i2d:
1447 __ extsw(R17_tos, R17_tos);
1448 case Bytecodes::_l2d:
1449 __ push_l_pop_d();
1450 __ fcfid(F15_ftos, F15_ftos);
1451 break;
1452
1453 case Bytecodes::_i2f:
1454 __ extsw(R17_tos, R17_tos);
1455 __ push_l_pop_d();
1456 if (VM_Version::has_fcfids()) { // fcfids is >= Power7 only
1457 // Comment: alternatively, load with sign extend could be done by lfiwax.
1458 __ fcfids(F15_ftos, F15_ftos);
1459 } else {
1460 __ fcfid(F15_ftos, F15_ftos);
1461 __ frsp(F15_ftos, F15_ftos);
1462 }
1463 break;
1464
1465 case Bytecodes::_l2f:
1466 if (VM_Version::has_fcfids()) { // fcfids is >= Power7 only
1467 __ push_l_pop_d();
1468 __ fcfids(F15_ftos, F15_ftos);
1469 } else {
1470 // Avoid rounding problem when result should be 0x3f800001: need fixup code before fcfid+frsp.
1471 __ mr(R3_ARG1, R17_tos);
1472 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::l2f));
1473 __ fmr(F15_ftos, F1_RET);
1474 }
1475 break;
1476
1477 case Bytecodes::_f2d:
1478 // empty
1479 break;
1480
1481 case Bytecodes::_d2f:
1482 __ frsp(F15_ftos, F15_ftos);
1483 break;
1484
1485 case Bytecodes::_d2i:
1486 case Bytecodes::_f2i:
1487 __ fcmpu(CCR0, F15_ftos, F15_ftos);
1488 __ li(R17_tos, 0); // 0 in case of NAN
1489 __ bso(CCR0, done);
1490 __ fctiwz(F15_ftos, F15_ftos);
1491 __ push_d_pop_l();
1492 break;
1493
1494 case Bytecodes::_d2l:
1495 case Bytecodes::_f2l:
1496 __ fcmpu(CCR0, F15_ftos, F15_ftos);
1497 __ li(R17_tos, 0); // 0 in case of NAN
1498 __ bso(CCR0, done);
1499 __ fctidz(F15_ftos, F15_ftos);
1500 __ push_d_pop_l();
1501 break;
1502
1503 default: ShouldNotReachHere();
1504 }
1505 __ bind(done);
1506 }
1507
1508 // Long compare
1509 void TemplateTable::lcmp() {
1510 transition(ltos, itos);
1511
1512 const Register Rscratch = R11_scratch1;
1513 __ pop_l(Rscratch); // first operand, deeper in stack
1514
1515 __ cmpd(CCR0, Rscratch, R17_tos); // compare
1516 __ mfcr(R17_tos); // set bit 32..33 as follows: <: 0b10, =: 0b00, >: 0b01
1517 __ srwi(Rscratch, R17_tos, 30);
1518 __ srawi(R17_tos, R17_tos, 31);
1519 __ orr(R17_tos, Rscratch, R17_tos); // set result as follows: <: -1, =: 0, >: 1
1520 }
1521
1522 // fcmpl/fcmpg and dcmpl/dcmpg bytecodes
1523 // unordered_result == -1 => fcmpl or dcmpl
1524 // unordered_result == 1 => fcmpg or dcmpg
1525 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1526 const FloatRegister Rfirst = F0_SCRATCH,
1527 Rsecond = F15_ftos;
1528 const Register Rscratch = R11_scratch1;
1529
1530 if (is_float) {
1531 __ pop_f(Rfirst);
1532 } else {
1533 __ pop_d(Rfirst);
1534 }
1535
1536 Label Lunordered, Ldone;
1537 __ fcmpu(CCR0, Rfirst, Rsecond); // compare
1538 if (unordered_result) {
1539 __ bso(CCR0, Lunordered);
1540 }
1541 __ mfcr(R17_tos); // set bit 32..33 as follows: <: 0b10, =: 0b00, >: 0b01
1542 __ srwi(Rscratch, R17_tos, 30);
1543 __ srawi(R17_tos, R17_tos, 31);
1544 __ orr(R17_tos, Rscratch, R17_tos); // set result as follows: <: -1, =: 0, >: 1
1545 if (unordered_result) {
1546 __ b(Ldone);
1547 __ bind(Lunordered);
1548 __ load_const_optimized(R17_tos, unordered_result);
1549 }
1550 __ bind(Ldone);
1551 }
1552
1553 // Branch_conditional which takes TemplateTable::Condition.
1554 void TemplateTable::branch_conditional(ConditionRegister crx, TemplateTable::Condition cc, Label& L, bool invert) {
1555 bool positive = false;
1556 Assembler::Condition cond = Assembler::equal;
1557 switch (cc) {
1558 case TemplateTable::equal: positive = true ; cond = Assembler::equal ; break;
1559 case TemplateTable::not_equal: positive = false; cond = Assembler::equal ; break;
1560 case TemplateTable::less: positive = true ; cond = Assembler::less ; break;
1561 case TemplateTable::less_equal: positive = false; cond = Assembler::greater; break;
1562 case TemplateTable::greater: positive = true ; cond = Assembler::greater; break;
1563 case TemplateTable::greater_equal: positive = false; cond = Assembler::less ; break;
1564 default: ShouldNotReachHere();
1565 }
1566 int bo = (positive != invert) ? Assembler::bcondCRbiIs1 : Assembler::bcondCRbiIs0;
1567 int bi = Assembler::bi0(crx, cond);
1568 __ bc(bo, bi, L);
1569 }
1570
1571 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1572
1573 // Note: on SPARC, we use InterpreterMacroAssembler::if_cmp also.
1574 __ verify_thread();
1575
1576 const Register Rscratch1 = R11_scratch1,
1577 Rscratch2 = R12_scratch2,
1578 Rscratch3 = R3_ARG1,
1579 R4_counters = R4_ARG2,
1580 bumped_count = R31,
1581 Rdisp = R22_tmp2;
1582
1583 __ profile_taken_branch(Rscratch1, bumped_count);
1584
1585 // Get (wide) offset.
1586 if (is_wide) {
1587 __ get_4_byte_integer_at_bcp(1, Rdisp, InterpreterMacroAssembler::Signed);
1588 } else {
1589 __ get_2_byte_integer_at_bcp(1, Rdisp, InterpreterMacroAssembler::Signed);
1590 }
1591
1592 // --------------------------------------------------------------------------
1593 // Handle all the JSR stuff here, then exit.
1594 // It's much shorter and cleaner than intermingling with the
1595 // non-JSR normal-branch stuff occurring below.
1596 if (is_jsr) {
1597 // Compute return address as bci in Otos_i.
1598 __ ld(Rscratch1, in_bytes(Method::const_offset()), R19_method);
1599 __ addi(Rscratch2, R14_bcp, -in_bytes(ConstMethod::codes_offset()) + (is_wide ? 5 : 3));
1600 __ subf(R17_tos, Rscratch1, Rscratch2);
1601
1602 // Bump bcp to target of JSR.
1603 __ add(R14_bcp, Rdisp, R14_bcp);
1604 // Push returnAddress for "ret" on stack.
1605 __ push_ptr(R17_tos);
1606 // And away we go!
1607 __ dispatch_next(vtos);
1608 return;
1609 }
1610
1611 // --------------------------------------------------------------------------
1612 // Normal (non-jsr) branch handling
1613
1614 const bool increment_invocation_counter_for_backward_branches = UseCompiler && UseLoopCounter;
1615 if (increment_invocation_counter_for_backward_branches) {
1616 //__ unimplemented("branch invocation counter");
1617
1618 Label Lforward;
1619 __ add(R14_bcp, Rdisp, R14_bcp); // Add to bc addr.
1620
1621 // Check branch direction.
1622 __ cmpdi(CCR0, Rdisp, 0);
1623 __ bgt(CCR0, Lforward);
1624
1625 __ get_method_counters(R19_method, R4_counters, Lforward);
1626
1627 if (TieredCompilation) {
1628 Label Lno_mdo, Loverflow;
1629 const int increment = InvocationCounter::count_increment;
1630 const int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
1631 if (ProfileInterpreter) {
1632 Register Rmdo = Rscratch1;
1633
1634 // If no method data exists, go to profile_continue.
1635 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
1636 __ cmpdi(CCR0, Rmdo, 0);
1637 __ beq(CCR0, Lno_mdo);
1638
1639 // Increment backedge counter in the MDO.
1640 const int mdo_bc_offs = in_bytes(MethodData::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
1641 __ lwz(Rscratch2, mdo_bc_offs, Rmdo);
1642 __ load_const_optimized(Rscratch3, mask, R0);
1643 __ addi(Rscratch2, Rscratch2, increment);
1644 __ stw(Rscratch2, mdo_bc_offs, Rmdo);
1645 __ and_(Rscratch3, Rscratch2, Rscratch3);
1646 __ bne(CCR0, Lforward);
1647 __ b(Loverflow);
1648 }
1649
1650 // If there's no MDO, increment counter in method.
1651 const int mo_bc_offs = in_bytes(MethodCounters::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
1652 __ bind(Lno_mdo);
1653 __ lwz(Rscratch2, mo_bc_offs, R4_counters);
1654 __ load_const_optimized(Rscratch3, mask, R0);
1655 __ addi(Rscratch2, Rscratch2, increment);
1656 __ stw(Rscratch2, mo_bc_offs, R19_method);
1657 __ and_(Rscratch3, Rscratch2, Rscratch3);
1658 __ bne(CCR0, Lforward);
1659
1660 __ bind(Loverflow);
1661
1662 // Notify point for loop, pass branch bytecode.
1663 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R14_bcp, true);
1664
1665 // Was an OSR adapter generated?
1666 // O0 = osr nmethod
1667 __ cmpdi(CCR0, R3_RET, 0);
1668 __ beq(CCR0, Lforward);
1669
1670 // Has the nmethod been invalidated already?
1671 __ lbz(R0, nmethod::state_offset(), R3_RET);
1672 __ cmpwi(CCR0, R0, nmethod::in_use);
1673 __ bne(CCR0, Lforward);
1674
1675 // Migrate the interpreter frame off of the stack.
1676 // We can use all registers because we will not return to interpreter from this point.
1677
1678 // Save nmethod.
1679 const Register osr_nmethod = R31;
1680 __ mr(osr_nmethod, R3_RET);
1681 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R11_scratch1);
1682 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), R16_thread);
1683 __ reset_last_Java_frame();
1684 // OSR buffer is in ARG1.
1685
1686 // Remove the interpreter frame.
1687 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1688
1689 // Jump to the osr code.
1690 __ ld(R11_scratch1, nmethod::osr_entry_point_offset(), osr_nmethod);
1691 __ mtlr(R0);
1692 __ mtctr(R11_scratch1);
1693 __ bctr();
1694
1695 } else {
1696
1697 const Register invoke_ctr = Rscratch1;
1698 // Update Backedge branch separately from invocations.
1699 __ increment_backedge_counter(R4_counters, invoke_ctr, Rscratch2, Rscratch3);
1700
1701 if (ProfileInterpreter) {
1702 __ test_invocation_counter_for_mdp(invoke_ctr, Rscratch2, Lforward);
1703 if (UseOnStackReplacement) {
1704 __ test_backedge_count_for_osr(bumped_count, R14_bcp, Rscratch2);
1705 }
1706 } else {
1707 if (UseOnStackReplacement) {
1708 __ test_backedge_count_for_osr(invoke_ctr, R14_bcp, Rscratch2);
1709 }
1710 }
1711 }
1712
1713 __ bind(Lforward);
1714
1715 } else {
1716 // Bump bytecode pointer by displacement (take the branch).
1717 __ add(R14_bcp, Rdisp, R14_bcp); // Add to bc addr.
1718 }
1719 // Continue with bytecode @ target.
1720 // %%%%% Like Intel, could speed things up by moving bytecode fetch to code above,
1721 // %%%%% and changing dispatch_next to dispatch_only.
1722 __ dispatch_next(vtos);
1723 }
1724
1725 // Helper function for if_cmp* methods below.
1726 // Factored out common compare and branch code.
1727 void TemplateTable::if_cmp_common(Register Rfirst, Register Rsecond, Register Rscratch1, Register Rscratch2, Condition cc, bool is_jint, bool cmp0) {
1728 Label Lnot_taken;
1729 // Note: The condition code we get is the condition under which we
1730 // *fall through*! So we have to inverse the CC here.
1731
1732 if (is_jint) {
1733 if (cmp0) {
1734 __ cmpwi(CCR0, Rfirst, 0);
1735 } else {
1736 __ cmpw(CCR0, Rfirst, Rsecond);
1737 }
1738 } else {
1739 if (cmp0) {
1740 __ cmpdi(CCR0, Rfirst, 0);
1741 } else {
1742 __ cmpd(CCR0, Rfirst, Rsecond);
1743 }
1744 }
1745 branch_conditional(CCR0, cc, Lnot_taken, /*invert*/ true);
1746
1747 // Conition is false => Jump!
1748 branch(false, false);
1749
1750 // Condition is not true => Continue.
1751 __ align(32, 12);
1752 __ bind(Lnot_taken);
1753 __ profile_not_taken_branch(Rscratch1, Rscratch2);
1754 }
1755
1756 // Compare integer values with zero and fall through if CC holds, branch away otherwise.
1757 void TemplateTable::if_0cmp(Condition cc) {
1758 transition(itos, vtos);
1759
1760 if_cmp_common(R17_tos, noreg, R11_scratch1, R12_scratch2, cc, true, true);
1761 }
1762
1763 // Compare integer values and fall through if CC holds, branch away otherwise.
1764 //
1765 // Interface:
1766 // - Rfirst: First operand (older stack value)
1767 // - tos: Second operand (younger stack value)
1768 void TemplateTable::if_icmp(Condition cc) {
1769 transition(itos, vtos);
1770
1771 const Register Rfirst = R0,
1772 Rsecond = R17_tos;
1773
1774 __ pop_i(Rfirst);
1775 if_cmp_common(Rfirst, Rsecond, R11_scratch1, R12_scratch2, cc, true, false);
1776 }
1777
1778 void TemplateTable::if_nullcmp(Condition cc) {
1779 transition(atos, vtos);
1780
1781 if_cmp_common(R17_tos, noreg, R11_scratch1, R12_scratch2, cc, false, true);
1782 }
1783
1784 void TemplateTable::if_acmp(Condition cc) {
1785 transition(atos, vtos);
1786
1787 const Register Rfirst = R0,
1788 Rsecond = R17_tos;
1789
1790 __ pop_ptr(Rfirst);
1791 if_cmp_common(Rfirst, Rsecond, R11_scratch1, R12_scratch2, cc, false, false);
1792 }
1793
1794 void TemplateTable::ret() {
1795 locals_index(R11_scratch1);
1796 __ load_local_ptr(R17_tos, R11_scratch1, R11_scratch1);
1797
1798 __ profile_ret(vtos, R17_tos, R11_scratch1, R12_scratch2);
1799
1800 __ ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method);
1801 __ add(R11_scratch1, R17_tos, R11_scratch1);
1802 __ addi(R14_bcp, R11_scratch1, in_bytes(ConstMethod::codes_offset()));
1803 __ dispatch_next(vtos);
1804 }
1805
1806 void TemplateTable::wide_ret() {
1807 transition(vtos, vtos);
1808
1809 const Register Rindex = R3_ARG1,
1810 Rscratch1 = R11_scratch1,
1811 Rscratch2 = R12_scratch2;
1812
1813 locals_index_wide(Rindex);
1814 __ load_local_ptr(R17_tos, R17_tos, Rindex);
1815 __ profile_ret(vtos, R17_tos, Rscratch1, R12_scratch2);
1816 // Tos now contains the bci, compute the bcp from that.
1817 __ ld(Rscratch1, in_bytes(Method::const_offset()), R19_method);
1818 __ addi(Rscratch2, R17_tos, in_bytes(ConstMethod::codes_offset()));
1819 __ add(R14_bcp, Rscratch1, Rscratch2);
1820 __ dispatch_next(vtos);
1821 }
1822
1823 void TemplateTable::tableswitch() {
1824 transition(itos, vtos);
1825
1826 Label Ldispatch, Ldefault_case;
1827 Register Rlow_byte = R3_ARG1,
1828 Rindex = Rlow_byte,
1829 Rhigh_byte = R4_ARG2,
1830 Rdef_offset_addr = R5_ARG3, // is going to contain address of default offset
1831 Rscratch1 = R11_scratch1,
1832 Rscratch2 = R12_scratch2,
1833 Roffset = R6_ARG4;
1834
1835 // Align bcp.
1836 __ addi(Rdef_offset_addr, R14_bcp, BytesPerInt);
1837 __ clrrdi(Rdef_offset_addr, Rdef_offset_addr, log2_long((jlong)BytesPerInt));
1838
1839 // Load lo & hi.
1840 __ get_u4(Rlow_byte, Rdef_offset_addr, BytesPerInt, InterpreterMacroAssembler::Unsigned);
1841 __ get_u4(Rhigh_byte, Rdef_offset_addr, 2 *BytesPerInt, InterpreterMacroAssembler::Unsigned);
1842
1843 // Check for default case (=index outside [low,high]).
1844 __ cmpw(CCR0, R17_tos, Rlow_byte);
1845 __ cmpw(CCR1, R17_tos, Rhigh_byte);
1846 __ blt(CCR0, Ldefault_case);
1847 __ bgt(CCR1, Ldefault_case);
1848
1849 // Lookup dispatch offset.
1850 __ sub(Rindex, R17_tos, Rlow_byte);
1851 __ extsw(Rindex, Rindex);
1852 __ profile_switch_case(Rindex, Rhigh_byte /* scratch */, Rscratch1, Rscratch2);
1853 __ sldi(Rindex, Rindex, LogBytesPerInt);
1854 __ addi(Rindex, Rindex, 3 * BytesPerInt);
1855 #if defined(VM_LITTLE_ENDIAN)
1856 __ lwbrx(Roffset, Rdef_offset_addr, Rindex);
1857 __ extsw(Roffset, Roffset);
1858 #else
1859 __ lwax(Roffset, Rdef_offset_addr, Rindex);
1860 #endif
1861 __ b(Ldispatch);
1862
1863 __ bind(Ldefault_case);
1864 __ profile_switch_default(Rhigh_byte, Rscratch1);
1865 __ get_u4(Roffset, Rdef_offset_addr, 0, InterpreterMacroAssembler::Signed);
1866
1867 __ bind(Ldispatch);
1868
1869 __ add(R14_bcp, Roffset, R14_bcp);
1870 __ dispatch_next(vtos);
1871 }
1872
1873 void TemplateTable::lookupswitch() {
1874 transition(itos, itos);
1875 __ stop("lookupswitch bytecode should have been rewritten");
1876 }
1877
1878 // Table switch using linear search through cases.
1879 // Bytecode stream format:
1880 // Bytecode (1) | 4-byte padding | default offset (4) | count (4) | value/offset pair1 (8) | value/offset pair2 (8) | ...
1881 // Note: Everything is big-endian format here.
1882 void TemplateTable::fast_linearswitch() {
1883 transition(itos, vtos);
1884
1885 Label Lloop_entry, Lsearch_loop, Lcontinue_execution, Ldefault_case;
1886 Register Rcount = R3_ARG1,
1887 Rcurrent_pair = R4_ARG2,
1888 Rdef_offset_addr = R5_ARG3, // Is going to contain address of default offset.
1889 Roffset = R31, // Might need to survive C call.
1890 Rvalue = R12_scratch2,
1891 Rscratch = R11_scratch1,
1892 Rcmp_value = R17_tos;
1893
1894 // Align bcp.
1895 __ addi(Rdef_offset_addr, R14_bcp, BytesPerInt);
1896 __ clrrdi(Rdef_offset_addr, Rdef_offset_addr, log2_long((jlong)BytesPerInt));
1897
1898 // Setup loop counter and limit.
1899 __ get_u4(Rcount, Rdef_offset_addr, BytesPerInt, InterpreterMacroAssembler::Unsigned);
1900 __ addi(Rcurrent_pair, Rdef_offset_addr, 2 * BytesPerInt); // Rcurrent_pair now points to first pair.
1901
1902 __ mtctr(Rcount);
1903 __ cmpwi(CCR0, Rcount, 0);
1904 __ bne(CCR0, Lloop_entry);
1905
1906 // Default case
1907 __ bind(Ldefault_case);
1908 __ get_u4(Roffset, Rdef_offset_addr, 0, InterpreterMacroAssembler::Signed);
1909 if (ProfileInterpreter) {
1910 __ profile_switch_default(Rdef_offset_addr, Rcount/* scratch */);
1911 }
1912 __ b(Lcontinue_execution);
1913
1914 // Next iteration
1915 __ bind(Lsearch_loop);
1916 __ bdz(Ldefault_case);
1917 __ addi(Rcurrent_pair, Rcurrent_pair, 2 * BytesPerInt);
1918 __ bind(Lloop_entry);
1919 __ get_u4(Rvalue, Rcurrent_pair, 0, InterpreterMacroAssembler::Unsigned);
1920 __ cmpw(CCR0, Rvalue, Rcmp_value);
1921 __ bne(CCR0, Lsearch_loop);
1922
1923 // Found, load offset.
1924 __ get_u4(Roffset, Rcurrent_pair, BytesPerInt, InterpreterMacroAssembler::Signed);
1925 // Calculate case index and profile
1926 __ mfctr(Rcurrent_pair);
1927 if (ProfileInterpreter) {
1928 __ sub(Rcurrent_pair, Rcount, Rcurrent_pair);
1929 __ profile_switch_case(Rcurrent_pair, Rcount /*scratch*/, Rdef_offset_addr/*scratch*/, Rscratch);
1930 }
1931
1932 __ bind(Lcontinue_execution);
1933 __ add(R14_bcp, Roffset, R14_bcp);
1934 __ dispatch_next(vtos);
1935 }
1936
1937 // Table switch using binary search (value/offset pairs are ordered).
1938 // Bytecode stream format:
1939 // Bytecode (1) | 4-byte padding | default offset (4) | count (4) | value/offset pair1 (8) | value/offset pair2 (8) | ...
1940 // Note: Everything is big-endian format here. So on little endian machines, we have to revers offset and count and cmp value.
1941 void TemplateTable::fast_binaryswitch() {
1942
1943 transition(itos, vtos);
1944 // Implementation using the following core algorithm: (copied from Intel)
1945 //
1946 // int binary_search(int key, LookupswitchPair* array, int n) {
1947 // // Binary search according to "Methodik des Programmierens" by
1948 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1949 // int i = 0;
1950 // int j = n;
1951 // while (i+1 < j) {
1952 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1953 // // with Q: for all i: 0 <= i < n: key < a[i]
1954 // // where a stands for the array and assuming that the (inexisting)
1955 // // element a[n] is infinitely big.
1956 // int h = (i + j) >> 1;
1957 // // i < h < j
1958 // if (key < array[h].fast_match()) {
1959 // j = h;
1960 // } else {
1961 // i = h;
1962 // }
1963 // }
1964 // // R: a[i] <= key < a[i+1] or Q
1965 // // (i.e., if key is within array, i is the correct index)
1966 // return i;
1967 // }
1968
1969 // register allocation
1970 const Register Rkey = R17_tos; // already set (tosca)
1971 const Register Rarray = R3_ARG1;
1972 const Register Ri = R4_ARG2;
1973 const Register Rj = R5_ARG3;
1974 const Register Rh = R6_ARG4;
1975 const Register Rscratch = R11_scratch1;
1976
1977 const int log_entry_size = 3;
1978 const int entry_size = 1 << log_entry_size;
1979
1980 Label found;
1981
1982 // Find Array start,
1983 __ addi(Rarray, R14_bcp, 3 * BytesPerInt);
1984 __ clrrdi(Rarray, Rarray, log2_long((jlong)BytesPerInt));
1985
1986 // initialize i & j
1987 __ li(Ri,0);
1988 __ get_u4(Rj, Rarray, -BytesPerInt, InterpreterMacroAssembler::Unsigned);
1989
1990 // and start.
1991 Label entry;
1992 __ b(entry);
1993
1994 // binary search loop
1995 { Label loop;
1996 __ bind(loop);
1997 // int h = (i + j) >> 1;
1998 __ srdi(Rh, Rh, 1);
1999 // if (key < array[h].fast_match()) {
2000 // j = h;
2001 // } else {
2002 // i = h;
2003 // }
2004 __ sldi(Rscratch, Rh, log_entry_size);
2005 #if defined(VM_LITTLE_ENDIAN)
2006 __ lwbrx(Rscratch, Rscratch, Rarray);
2007 #else
2008 __ lwzx(Rscratch, Rscratch, Rarray);
2009 #endif
2010
2011 // if (key < current value)
2012 // Rh = Rj
2013 // else
2014 // Rh = Ri
2015 Label Lgreater;
2016 __ cmpw(CCR0, Rkey, Rscratch);
2017 __ bge(CCR0, Lgreater);
2018 __ mr(Rj, Rh);
2019 __ b(entry);
2020 __ bind(Lgreater);
2021 __ mr(Ri, Rh);
2022
2023 // while (i+1 < j)
2024 __ bind(entry);
2025 __ addi(Rscratch, Ri, 1);
2026 __ cmpw(CCR0, Rscratch, Rj);
2027 __ add(Rh, Ri, Rj); // start h = i + j >> 1;
2028
2029 __ blt(CCR0, loop);
2030 }
2031
2032 // End of binary search, result index is i (must check again!).
2033 Label default_case;
2034 Label continue_execution;
2035 if (ProfileInterpreter) {
2036 __ mr(Rh, Ri); // Save index in i for profiling.
2037 }
2038 // Ri = value offset
2039 __ sldi(Ri, Ri, log_entry_size);
2040 __ add(Ri, Ri, Rarray);
2041 __ get_u4(Rscratch, Ri, 0, InterpreterMacroAssembler::Unsigned);
2042
2043 Label not_found;
2044 // Ri = offset offset
2045 __ cmpw(CCR0, Rkey, Rscratch);
2046 __ beq(CCR0, not_found);
2047 // entry not found -> j = default offset
2048 __ get_u4(Rj, Rarray, -2 * BytesPerInt, InterpreterMacroAssembler::Unsigned);
2049 __ b(default_case);
2050
2051 __ bind(not_found);
2052 // entry found -> j = offset
2053 __ profile_switch_case(Rh, Rj, Rscratch, Rkey);
2054 __ get_u4(Rj, Ri, BytesPerInt, InterpreterMacroAssembler::Unsigned);
2055
2056 if (ProfileInterpreter) {
2057 __ b(continue_execution);
2058 }
2059
2060 __ bind(default_case); // fall through (if not profiling)
2061 __ profile_switch_default(Ri, Rscratch);
2062
2063 __ bind(continue_execution);
2064
2065 __ extsw(Rj, Rj);
2066 __ add(R14_bcp, Rj, R14_bcp);
2067 __ dispatch_next(vtos);
2068 }
2069
2070 void TemplateTable::_return(TosState state) {
2071 transition(state, state);
2072 assert(_desc->calls_vm(),
2073 "inconsistent calls_vm information"); // call in remove_activation
2074
2075 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2076
2077 Register Rscratch = R11_scratch1,
2078 Rklass = R12_scratch2,
2079 Rklass_flags = Rklass;
2080 Label Lskip_register_finalizer;
2081
2082 // Check if the method has the FINALIZER flag set and call into the VM to finalize in this case.
2083 assert(state == vtos, "only valid state");
2084 __ ld(R17_tos, 0, R18_locals);
2085
2086 // Load klass of this obj.
2087 __ load_klass(Rklass, R17_tos);
2088 __ lwz(Rklass_flags, in_bytes(Klass::access_flags_offset()), Rklass);
2089 __ testbitdi(CCR0, R0, Rklass_flags, exact_log2(JVM_ACC_HAS_FINALIZER));
2090 __ bfalse(CCR0, Lskip_register_finalizer);
2091
2092 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), R17_tos /* obj */);
2093
2094 __ align(32, 12);
2095 __ bind(Lskip_register_finalizer);
2096 }
2097
2098 // Move the result value into the correct register and remove memory stack frame.
2099 __ remove_activation(state, /* throw_monitor_exception */ true);
2100 // Restoration of lr done by remove_activation.
2101 switch (state) {
2102 case ltos:
2103 case btos:
2104 case ctos:
2105 case stos:
2106 case atos:
2107 case itos: __ mr(R3_RET, R17_tos); break;
2108 case ftos:
2109 case dtos: __ fmr(F1_RET, F15_ftos); break;
2110 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
2111 // to get visible before the reference to the object gets stored anywhere.
2112 __ membar(Assembler::StoreStore); break;
2113 default : ShouldNotReachHere();
2114 }
2115 __ blr();
2116 }
2117
2118 // ============================================================================
2119 // Constant pool cache access
2120 //
2121 // Memory ordering:
2122 //
2123 // Like done in C++ interpreter, we load the fields
2124 // - _indices
2125 // - _f12_oop
2126 // acquired, because these are asked if the cache is already resolved. We don't
2127 // want to float loads above this check.
2128 // See also comments in ConstantPoolCacheEntry::bytecode_1(),
2129 // ConstantPoolCacheEntry::bytecode_2() and ConstantPoolCacheEntry::f1();
2130
2131 // Call into the VM if call site is not yet resolved
2132 //
2133 // Input regs:
2134 // - None, all passed regs are outputs.
2135 //
2136 // Returns:
2137 // - Rcache: The const pool cache entry that contains the resolved result.
2138 // - Rresult: Either noreg or output for f1/f2.
2139 //
2140 // Kills:
2141 // - Rscratch
2142 void TemplateTable::resolve_cache_and_index(int byte_no, Register Rcache, Register Rscratch, size_t index_size) {
2143
2144 __ get_cache_and_index_at_bcp(Rcache, 1, index_size);
2145 Label Lresolved, Ldone;
2146
2147 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2148 // We are resolved if the indices offset contains the current bytecode.
2149 #if defined(VM_LITTLE_ENDIAN)
2150 __ lbz(Rscratch, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + byte_no + 1, Rcache);
2151 #else
2152 __ lbz(Rscratch, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 7 - (byte_no + 1), Rcache);
2153 #endif
2154 // Acquire by cmp-br-isync (see below).
2155 __ cmpdi(CCR0, Rscratch, (int)bytecode());
2156 __ beq(CCR0, Lresolved);
2157
2158 address entry = NULL;
2159 switch (bytecode()) {
2160 case Bytecodes::_getstatic : // fall through
2161 case Bytecodes::_putstatic : // fall through
2162 case Bytecodes::_getfield : // fall through
2163 case Bytecodes::_putfield : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break;
2164 case Bytecodes::_invokevirtual : // fall through
2165 case Bytecodes::_invokespecial : // fall through
2166 case Bytecodes::_invokestatic : // fall through
2167 case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke); break;
2168 case Bytecodes::_invokehandle : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokehandle); break;
2169 case Bytecodes::_invokedynamic : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic); break;
2170 default : ShouldNotReachHere(); break;
2171 }
2172 __ li(R4_ARG2, (int)bytecode());
2173 __ call_VM(noreg, entry, R4_ARG2, true);
2174
2175 // Update registers with resolved info.
2176 __ get_cache_and_index_at_bcp(Rcache, 1, index_size);
2177 __ b(Ldone);
2178
2179 __ bind(Lresolved);
2180 __ isync(); // Order load wrt. succeeding loads.
2181 __ bind(Ldone);
2182 }
2183
2184 // Load the constant pool cache entry at field accesses into registers.
2185 // The Rcache and Rindex registers must be set before call.
2186 // Input:
2187 // - Rcache, Rindex
2188 // Output:
2189 // - Robj, Roffset, Rflags
2190 void TemplateTable::load_field_cp_cache_entry(Register Robj,
2191 Register Rcache,
2192 Register Rindex /* unused on PPC64 */,
2193 Register Roffset,
2194 Register Rflags,
2195 bool is_static = false) {
2196 assert_different_registers(Rcache, Rflags, Roffset);
2197 // assert(Rindex == noreg, "parameter not used on PPC64");
2198
2199 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2200 __ ld(Rflags, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::flags_offset()), Rcache);
2201 __ ld(Roffset, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::f2_offset()), Rcache);
2202 if (is_static) {
2203 __ ld(Robj, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::f1_offset()), Rcache);
2204 __ ld(Robj, in_bytes(Klass::java_mirror_offset()), Robj);
2205 // Acquire not needed here. Following access has an address dependency on this value.
2206 }
2207 }
2208
2209 // Load the constant pool cache entry at invokes into registers.
2210 // Resolve if necessary.
2211
2212 // Input Registers:
2213 // - None, bcp is used, though
2214 //
2215 // Return registers:
2216 // - Rmethod (f1 field or f2 if invokevirtual)
2217 // - Ritable_index (f2 field)
2218 // - Rflags (flags field)
2219 //
2220 // Kills:
2221 // - R21
2222 //
2223 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2224 Register Rmethod,
2225 Register Ritable_index,
2226 Register Rflags,
2227 bool is_invokevirtual,
2228 bool is_invokevfinal,
2229 bool is_invokedynamic) {
2230
2231 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2232 // Determine constant pool cache field offsets.
2233 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2234 const int method_offset = in_bytes(cp_base_offset + (is_invokevirtual ? ConstantPoolCacheEntry::f2_offset() : ConstantPoolCacheEntry::f1_offset()));
2235 const int flags_offset = in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset());
2236 // Access constant pool cache fields.
2237 const int index_offset = in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset());
2238
2239 Register Rcache = R21_tmp1; // Note: same register as R21_sender_SP.
2240
2241 if (is_invokevfinal) {
2242 assert(Ritable_index == noreg, "register not used");
2243 // Already resolved.
2244 __ get_cache_and_index_at_bcp(Rcache, 1);
2245 } else {
2246 resolve_cache_and_index(byte_no, Rcache, R0, is_invokedynamic ? sizeof(u4) : sizeof(u2));
2247 }
2248
2249 __ ld(Rmethod, method_offset, Rcache);
2250 __ ld(Rflags, flags_offset, Rcache);
2251
2252 if (Ritable_index != noreg) {
2253 __ ld(Ritable_index, index_offset, Rcache);
2254 }
2255 }
2256
2257 // ============================================================================
2258 // Field access
2259
2260 // Volatile variables demand their effects be made known to all CPU's
2261 // in order. Store buffers on most chips allow reads & writes to
2262 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2263 // without some kind of memory barrier (i.e., it's not sufficient that
2264 // the interpreter does not reorder volatile references, the hardware
2265 // also must not reorder them).
2266 //
2267 // According to the new Java Memory Model (JMM):
2268 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2269 // writes act as aquire & release, so:
2270 // (2) A read cannot let unrelated NON-volatile memory refs that
2271 // happen after the read float up to before the read. It's OK for
2272 // non-volatile memory refs that happen before the volatile read to
2273 // float down below it.
2274 // (3) Similar a volatile write cannot let unrelated NON-volatile
2275 // memory refs that happen BEFORE the write float down to after the
2276 // write. It's OK for non-volatile memory refs that happen after the
2277 // volatile write to float up before it.
2278 //
2279 // We only put in barriers around volatile refs (they are expensive),
2280 // not _between_ memory refs (that would require us to track the
2281 // flavor of the previous memory refs). Requirements (2) and (3)
2282 // require some barriers before volatile stores and after volatile
2283 // loads. These nearly cover requirement (1) but miss the
2284 // volatile-store-volatile-load case. This final case is placed after
2285 // volatile-stores although it could just as well go before
2286 // volatile-loads.
2287
2288 // The registers cache and index expected to be set before call.
2289 // Correct values of the cache and index registers are preserved.
2290 // Kills:
2291 // Rcache (if has_tos)
2292 // Rscratch
2293 void TemplateTable::jvmti_post_field_access(Register Rcache, Register Rscratch, bool is_static, bool has_tos) {
2294
2295 assert_different_registers(Rcache, Rscratch);
2296
2297 if (JvmtiExport::can_post_field_access()) {
2298 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2299 Label Lno_field_access_post;
2300
2301 // Check if post field access in enabled.
2302 int offs = __ load_const_optimized(Rscratch, JvmtiExport::get_field_access_count_addr(), R0, true);
2303 __ lwz(Rscratch, offs, Rscratch);
2304
2305 __ cmpwi(CCR0, Rscratch, 0);
2306 __ beq(CCR0, Lno_field_access_post);
2307
2308 // Post access enabled - do it!
2309 __ addi(Rcache, Rcache, in_bytes(cp_base_offset));
2310 if (is_static) {
2311 __ li(R17_tos, 0);
2312 } else {
2313 if (has_tos) {
2314 // The fast bytecode versions have obj ptr in register.
2315 // Thus, save object pointer before call_VM() clobbers it
2316 // put object on tos where GC wants it.
2317 __ push_ptr(R17_tos);
2318 } else {
2319 // Load top of stack (do not pop the value off the stack).
2320 __ ld(R17_tos, Interpreter::expr_offset_in_bytes(0), R15_esp);
2321 }
2322 __ verify_oop(R17_tos);
2323 }
2324 // tos: object pointer or NULL if static
2325 // cache: cache entry pointer
2326 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), R17_tos, Rcache);
2327 if (!is_static && has_tos) {
2328 // Restore object pointer.
2329 __ pop_ptr(R17_tos);
2330 __ verify_oop(R17_tos);
2331 } else {
2332 // Cache is still needed to get class or obj.
2333 __ get_cache_and_index_at_bcp(Rcache, 1);
2334 }
2335
2336 __ align(32, 12);
2337 __ bind(Lno_field_access_post);
2338 }
2339 }
2340
2341 // kills R11_scratch1
2342 void TemplateTable::pop_and_check_object(Register Roop) {
2343 Register Rtmp = R11_scratch1;
2344
2345 assert_different_registers(Rtmp, Roop);
2346 __ pop_ptr(Roop);
2347 // For field access must check obj.
2348 __ null_check_throw(Roop, -1, Rtmp);
2349 __ verify_oop(Roop);
2350 }
2351
2352 // PPC64: implement volatile loads as fence-store-acquire.
2353 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2354 transition(vtos, vtos);
2355
2356 Label Lacquire, Lisync;
2357
2358 const Register Rcache = R3_ARG1,
2359 Rclass_or_obj = R22_tmp2,
2360 Roffset = R23_tmp3,
2361 Rflags = R31,
2362 Rbtable = R5_ARG3,
2363 Rbc = R6_ARG4,
2364 Rscratch = R12_scratch2;
2365
2366 static address field_branch_table[number_of_states],
2367 static_branch_table[number_of_states];
2368
2369 address* branch_table = is_static ? static_branch_table : field_branch_table;
2370
2371 // Get field offset.
2372 resolve_cache_and_index(byte_no, Rcache, Rscratch, sizeof(u2));
2373
2374 // JVMTI support
2375 jvmti_post_field_access(Rcache, Rscratch, is_static, false);
2376
2377 // Load after possible GC.
2378 load_field_cp_cache_entry(Rclass_or_obj, Rcache, noreg, Roffset, Rflags, is_static);
2379
2380 // Load pointer to branch table.
2381 __ load_const_optimized(Rbtable, (address)branch_table, Rscratch);
2382
2383 // Get volatile flag.
2384 __ rldicl(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2385 // Note: sync is needed before volatile load on PPC64.
2386
2387 // Check field type.
2388 __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2389
2390 #ifdef ASSERT
2391 Label LFlagInvalid;
2392 __ cmpldi(CCR0, Rflags, number_of_states);
2393 __ bge(CCR0, LFlagInvalid);
2394 #endif
2395
2396 // Load from branch table and dispatch (volatile case: one instruction ahead).
2397 __ sldi(Rflags, Rflags, LogBytesPerWord);
2398 __ cmpwi(CCR6, Rscratch, 1); // Volatile?
2399 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2400 __ sldi(Rscratch, Rscratch, exact_log2(BytesPerInstWord)); // Volatile ? size of 1 instruction : 0.
2401 }
2402 __ ldx(Rbtable, Rbtable, Rflags);
2403
2404 // Get the obj from stack.
2405 if (!is_static) {
2406 pop_and_check_object(Rclass_or_obj); // Kills R11_scratch1.
2407 } else {
2408 __ verify_oop(Rclass_or_obj);
2409 }
2410
2411 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2412 __ subf(Rbtable, Rscratch, Rbtable); // Point to volatile/non-volatile entry point.
2413 }
2414 __ mtctr(Rbtable);
2415 __ bctr();
2416
2417 #ifdef ASSERT
2418 __ bind(LFlagInvalid);
2419 __ stop("got invalid flag", 0x654);
2420
2421 // __ bind(Lvtos);
2422 address pc_before_fence = __ pc();
2423 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2424 assert(__ pc() - pc_before_fence == (ptrdiff_t)BytesPerInstWord, "must be single instruction");
2425 assert(branch_table[vtos] == 0, "can't compute twice");
2426 branch_table[vtos] = __ pc(); // non-volatile_entry point
2427 __ stop("vtos unexpected", 0x655);
2428 #endif
2429
2430 __ align(32, 28, 28); // Align load.
2431 // __ bind(Ldtos);
2432 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2433 assert(branch_table[dtos] == 0, "can't compute twice");
2434 branch_table[dtos] = __ pc(); // non-volatile_entry point
2435 __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
2436 __ push(dtos);
2437 if (!is_static) patch_bytecode(Bytecodes::_fast_dgetfield, Rbc, Rscratch);
2438 {
2439 Label acquire_double;
2440 __ beq(CCR6, acquire_double); // Volatile?
2441 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2442
2443 __ bind(acquire_double);
2444 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
2445 __ beq_predict_taken(CCR0, Lisync);
2446 __ b(Lisync); // In case of NAN.
2447 }
2448
2449 __ align(32, 28, 28); // Align load.
2450 // __ bind(Lftos);
2451 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2452 assert(branch_table[ftos] == 0, "can't compute twice");
2453 branch_table[ftos] = __ pc(); // non-volatile_entry point
2454 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
2455 __ push(ftos);
2456 if (!is_static) { patch_bytecode(Bytecodes::_fast_fgetfield, Rbc, Rscratch); }
2457 {
2458 Label acquire_float;
2459 __ beq(CCR6, acquire_float); // Volatile?
2460 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2461
2462 __ bind(acquire_float);
2463 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
2464 __ beq_predict_taken(CCR0, Lisync);
2465 __ b(Lisync); // In case of NAN.
2466 }
2467
2468 __ align(32, 28, 28); // Align load.
2469 // __ bind(Litos);
2470 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2471 assert(branch_table[itos] == 0, "can't compute twice");
2472 branch_table[itos] = __ pc(); // non-volatile_entry point
2473 __ lwax(R17_tos, Rclass_or_obj, Roffset);
2474 __ push(itos);
2475 if (!is_static) patch_bytecode(Bytecodes::_fast_igetfield, Rbc, Rscratch);
2476 __ beq(CCR6, Lacquire); // Volatile?
2477 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2478
2479 __ align(32, 28, 28); // Align load.
2480 // __ bind(Lltos);
2481 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2482 assert(branch_table[ltos] == 0, "can't compute twice");
2483 branch_table[ltos] = __ pc(); // non-volatile_entry point
2484 __ ldx(R17_tos, Rclass_or_obj, Roffset);
2485 __ push(ltos);
2486 if (!is_static) patch_bytecode(Bytecodes::_fast_lgetfield, Rbc, Rscratch);
2487 __ beq(CCR6, Lacquire); // Volatile?
2488 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2489
2490 __ align(32, 28, 28); // Align load.
2491 // __ bind(Lbtos);
2492 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2493 assert(branch_table[btos] == 0, "can't compute twice");
2494 branch_table[btos] = __ pc(); // non-volatile_entry point
2495 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
2496 __ extsb(R17_tos, R17_tos);
2497 __ push(btos);
2498 if (!is_static) patch_bytecode(Bytecodes::_fast_bgetfield, Rbc, Rscratch);
2499 __ beq(CCR6, Lacquire); // Volatile?
2500 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2501
2502 __ align(32, 28, 28); // Align load.
2503 // __ bind(Lctos);
2504 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2505 assert(branch_table[ctos] == 0, "can't compute twice");
2506 branch_table[ctos] = __ pc(); // non-volatile_entry point
2507 __ lhzx(R17_tos, Rclass_or_obj, Roffset);
2508 __ push(ctos);
2509 if (!is_static) patch_bytecode(Bytecodes::_fast_cgetfield, Rbc, Rscratch);
2510 __ beq(CCR6, Lacquire); // Volatile?
2511 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2512
2513 __ align(32, 28, 28); // Align load.
2514 // __ bind(Lstos);
2515 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2516 assert(branch_table[stos] == 0, "can't compute twice");
2517 branch_table[stos] = __ pc(); // non-volatile_entry point
2518 __ lhax(R17_tos, Rclass_or_obj, Roffset);
2519 __ push(stos);
2520 if (!is_static) patch_bytecode(Bytecodes::_fast_sgetfield, Rbc, Rscratch);
2521 __ beq(CCR6, Lacquire); // Volatile?
2522 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2523
2524 __ align(32, 28, 28); // Align load.
2525 // __ bind(Latos);
2526 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2527 assert(branch_table[atos] == 0, "can't compute twice");
2528 branch_table[atos] = __ pc(); // non-volatile_entry point
2529 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
2530 __ verify_oop(R17_tos);
2531 __ push(atos);
2532 //__ dcbt(R17_tos); // prefetch
2533 if (!is_static) patch_bytecode(Bytecodes::_fast_agetfield, Rbc, Rscratch);
2534 __ beq(CCR6, Lacquire); // Volatile?
2535 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2536
2537 __ align(32, 12);
2538 __ bind(Lacquire);
2539 __ twi_0(R17_tos);
2540 __ bind(Lisync);
2541 __ isync(); // acquire
2542
2543 #ifdef ASSERT
2544 for (int i = 0; i<number_of_states; ++i) {
2545 assert(branch_table[i], "get initialization");
2546 //tty->print_cr("get: %s_branch_table[%d] = 0x%llx (opcode 0x%llx)",
2547 // is_static ? "static" : "field", i, branch_table[i], *((unsigned int*)branch_table[i]));
2548 }
2549 #endif
2550 }
2551
2552 void TemplateTable::getfield(int byte_no) {
2553 getfield_or_static(byte_no, false);
2554 }
2555
2556 void TemplateTable::getstatic(int byte_no) {
2557 getfield_or_static(byte_no, true);
2558 }
2559
2560 // The registers cache and index expected to be set before call.
2561 // The function may destroy various registers, just not the cache and index registers.
2562 void TemplateTable::jvmti_post_field_mod(Register Rcache, Register Rscratch, bool is_static) {
2563
2564 assert_different_registers(Rcache, Rscratch, R6_ARG4);
2565
2566 if (JvmtiExport::can_post_field_modification()) {
2567 Label Lno_field_mod_post;
2568
2569 // Check if post field access in enabled.
2570 int offs = __ load_const_optimized(Rscratch, JvmtiExport::get_field_modification_count_addr(), R0, true);
2571 __ lwz(Rscratch, offs, Rscratch);
2572
2573 __ cmpwi(CCR0, Rscratch, 0);
2574 __ beq(CCR0, Lno_field_mod_post);
2575
2576 // Do the post
2577 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2578 const Register Robj = Rscratch;
2579
2580 __ addi(Rcache, Rcache, in_bytes(cp_base_offset));
2581 if (is_static) {
2582 // Life is simple. Null out the object pointer.
2583 __ li(Robj, 0);
2584 } else {
2585 // In case of the fast versions, value lives in registers => put it back on tos.
2586 int offs = Interpreter::expr_offset_in_bytes(0);
2587 Register base = R15_esp;
2588 switch(bytecode()) {
2589 case Bytecodes::_fast_aputfield: __ push_ptr(); offs+= Interpreter::stackElementSize; break;
2590 case Bytecodes::_fast_iputfield: // Fall through
2591 case Bytecodes::_fast_bputfield: // Fall through
2592 case Bytecodes::_fast_cputfield: // Fall through
2593 case Bytecodes::_fast_sputfield: __ push_i(); offs+= Interpreter::stackElementSize; break;
2594 case Bytecodes::_fast_lputfield: __ push_l(); offs+=2*Interpreter::stackElementSize; break;
2595 case Bytecodes::_fast_fputfield: __ push_f(); offs+= Interpreter::stackElementSize; break;
2596 case Bytecodes::_fast_dputfield: __ push_d(); offs+=2*Interpreter::stackElementSize; break;
2597 default: {
2598 offs = 0;
2599 base = Robj;
2600 const Register Rflags = Robj;
2601 Label is_one_slot;
2602 // Life is harder. The stack holds the value on top, followed by the
2603 // object. We don't know the size of the value, though; it could be
2604 // one or two words depending on its type. As a result, we must find
2605 // the type to determine where the object is.
2606 __ ld(Rflags, in_bytes(ConstantPoolCacheEntry::flags_offset()), Rcache); // Big Endian
2607 __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2608
2609 __ cmpwi(CCR0, Rflags, ltos);
2610 __ cmpwi(CCR1, Rflags, dtos);
2611 __ addi(base, R15_esp, Interpreter::expr_offset_in_bytes(1));
2612 __ crnor(CCR0, Assembler::equal, CCR1, Assembler::equal);
2613 __ beq(CCR0, is_one_slot);
2614 __ addi(base, R15_esp, Interpreter::expr_offset_in_bytes(2));
2615 __ bind(is_one_slot);
2616 break;
2617 }
2618 }
2619 __ ld(Robj, offs, base);
2620 __ verify_oop(Robj);
2621 }
2622
2623 __ addi(R6_ARG4, R15_esp, Interpreter::expr_offset_in_bytes(0));
2624 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), Robj, Rcache, R6_ARG4);
2625 __ get_cache_and_index_at_bcp(Rcache, 1);
2626
2627 // In case of the fast versions, value lives in registers => put it back on tos.
2628 switch(bytecode()) {
2629 case Bytecodes::_fast_aputfield: __ pop_ptr(); break;
2630 case Bytecodes::_fast_iputfield: // Fall through
2631 case Bytecodes::_fast_bputfield: // Fall through
2632 case Bytecodes::_fast_cputfield: // Fall through
2633 case Bytecodes::_fast_sputfield: __ pop_i(); break;
2634 case Bytecodes::_fast_lputfield: __ pop_l(); break;
2635 case Bytecodes::_fast_fputfield: __ pop_f(); break;
2636 case Bytecodes::_fast_dputfield: __ pop_d(); break;
2637 default: break; // Nothin' to do.
2638 }
2639
2640 __ align(32, 12);
2641 __ bind(Lno_field_mod_post);
2642 }
2643 }
2644
2645 // PPC64: implement volatile stores as release-store (return bytecode contains an additional release).
2646 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2647 Label Lvolatile;
2648
2649 const Register Rcache = R5_ARG3, // Do not use ARG1/2 (causes trouble in jvmti_post_field_mod).
2650 Rclass_or_obj = R31, // Needs to survive C call.
2651 Roffset = R22_tmp2, // Needs to survive C call.
2652 Rflags = R3_ARG1,
2653 Rbtable = R4_ARG2,
2654 Rscratch = R11_scratch1,
2655 Rscratch2 = R12_scratch2,
2656 Rscratch3 = R6_ARG4,
2657 Rbc = Rscratch3;
2658 const ConditionRegister CR_is_vol = CCR2; // Non-volatile condition register (survives runtime call in do_oop_store).
2659
2660 static address field_branch_table[number_of_states],
2661 static_branch_table[number_of_states];
2662
2663 address* branch_table = is_static ? static_branch_table : field_branch_table;
2664
2665 // Stack (grows up):
2666 // value
2667 // obj
2668
2669 // Load the field offset.
2670 resolve_cache_and_index(byte_no, Rcache, Rscratch, sizeof(u2));
2671 jvmti_post_field_mod(Rcache, Rscratch, is_static);
2672 load_field_cp_cache_entry(Rclass_or_obj, Rcache, noreg, Roffset, Rflags, is_static);
2673
2674 // Load pointer to branch table.
2675 __ load_const_optimized(Rbtable, (address)branch_table, Rscratch);
2676
2677 // Get volatile flag.
2678 __ rldicl(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2679
2680 // Check the field type.
2681 __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2682
2683 #ifdef ASSERT
2684 Label LFlagInvalid;
2685 __ cmpldi(CCR0, Rflags, number_of_states);
2686 __ bge(CCR0, LFlagInvalid);
2687 #endif
2688
2689 // Load from branch table and dispatch (volatile case: one instruction ahead).
2690 __ sldi(Rflags, Rflags, LogBytesPerWord);
2691 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) { __ cmpwi(CR_is_vol, Rscratch, 1); } // Volatile?
2692 __ sldi(Rscratch, Rscratch, exact_log2(BytesPerInstWord)); // Volatile? size of instruction 1 : 0.
2693 __ ldx(Rbtable, Rbtable, Rflags);
2694
2695 __ subf(Rbtable, Rscratch, Rbtable); // Point to volatile/non-volatile entry point.
2696 __ mtctr(Rbtable);
2697 __ bctr();
2698
2699 #ifdef ASSERT
2700 __ bind(LFlagInvalid);
2701 __ stop("got invalid flag", 0x656);
2702
2703 // __ bind(Lvtos);
2704 address pc_before_release = __ pc();
2705 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2706 assert(__ pc() - pc_before_release == (ptrdiff_t)BytesPerInstWord, "must be single instruction");
2707 assert(branch_table[vtos] == 0, "can't compute twice");
2708 branch_table[vtos] = __ pc(); // non-volatile_entry point
2709 __ stop("vtos unexpected", 0x657);
2710 #endif
2711
2712 __ align(32, 28, 28); // Align pop.
2713 // __ bind(Ldtos);
2714 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2715 assert(branch_table[dtos] == 0, "can't compute twice");
2716 branch_table[dtos] = __ pc(); // non-volatile_entry point
2717 __ pop(dtos);
2718 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2719 __ stfdx(F15_ftos, Rclass_or_obj, Roffset);
2720 if (!is_static) { patch_bytecode(Bytecodes::_fast_dputfield, Rbc, Rscratch, true, byte_no); }
2721 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2722 __ beq(CR_is_vol, Lvolatile); // Volatile?
2723 }
2724 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2725
2726 __ align(32, 28, 28); // Align pop.
2727 // __ bind(Lftos);
2728 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2729 assert(branch_table[ftos] == 0, "can't compute twice");
2730 branch_table[ftos] = __ pc(); // non-volatile_entry point
2731 __ pop(ftos);
2732 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2733 __ stfsx(F15_ftos, Rclass_or_obj, Roffset);
2734 if (!is_static) { patch_bytecode(Bytecodes::_fast_fputfield, Rbc, Rscratch, true, byte_no); }
2735 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2736 __ beq(CR_is_vol, Lvolatile); // Volatile?
2737 }
2738 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2739
2740 __ align(32, 28, 28); // Align pop.
2741 // __ bind(Litos);
2742 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2743 assert(branch_table[itos] == 0, "can't compute twice");
2744 branch_table[itos] = __ pc(); // non-volatile_entry point
2745 __ pop(itos);
2746 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2747 __ stwx(R17_tos, Rclass_or_obj, Roffset);
2748 if (!is_static) { patch_bytecode(Bytecodes::_fast_iputfield, Rbc, Rscratch, true, byte_no); }
2749 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2750 __ beq(CR_is_vol, Lvolatile); // Volatile?
2751 }
2752 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2753
2754 __ align(32, 28, 28); // Align pop.
2755 // __ bind(Lltos);
2756 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2757 assert(branch_table[ltos] == 0, "can't compute twice");
2758 branch_table[ltos] = __ pc(); // non-volatile_entry point
2759 __ pop(ltos);
2760 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2761 __ stdx(R17_tos, Rclass_or_obj, Roffset);
2762 if (!is_static) { patch_bytecode(Bytecodes::_fast_lputfield, Rbc, Rscratch, true, byte_no); }
2763 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2764 __ beq(CR_is_vol, Lvolatile); // Volatile?
2765 }
2766 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2767
2768 __ align(32, 28, 28); // Align pop.
2769 // __ bind(Lbtos);
2770 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2771 assert(branch_table[btos] == 0, "can't compute twice");
2772 branch_table[btos] = __ pc(); // non-volatile_entry point
2773 __ pop(btos);
2774 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2775 __ stbx(R17_tos, Rclass_or_obj, Roffset);
2776 if (!is_static) { patch_bytecode(Bytecodes::_fast_bputfield, Rbc, Rscratch, true, byte_no); }
2777 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2778 __ beq(CR_is_vol, Lvolatile); // Volatile?
2779 }
2780 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2781
2782 __ align(32, 28, 28); // Align pop.
2783 // __ bind(Lctos);
2784 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2785 assert(branch_table[ctos] == 0, "can't compute twice");
2786 branch_table[ctos] = __ pc(); // non-volatile_entry point
2787 __ pop(ctos);
2788 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1..
2789 __ sthx(R17_tos, Rclass_or_obj, Roffset);
2790 if (!is_static) { patch_bytecode(Bytecodes::_fast_cputfield, Rbc, Rscratch, true, byte_no); }
2791 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2792 __ beq(CR_is_vol, Lvolatile); // Volatile?
2793 }
2794 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2795
2796 __ align(32, 28, 28); // Align pop.
2797 // __ bind(Lstos);
2798 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2799 assert(branch_table[stos] == 0, "can't compute twice");
2800 branch_table[stos] = __ pc(); // non-volatile_entry point
2801 __ pop(stos);
2802 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2803 __ sthx(R17_tos, Rclass_or_obj, Roffset);
2804 if (!is_static) { patch_bytecode(Bytecodes::_fast_sputfield, Rbc, Rscratch, true, byte_no); }
2805 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2806 __ beq(CR_is_vol, Lvolatile); // Volatile?
2807 }
2808 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2809
2810 __ align(32, 28, 28); // Align pop.
2811 // __ bind(Latos);
2812 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2813 assert(branch_table[atos] == 0, "can't compute twice");
2814 branch_table[atos] = __ pc(); // non-volatile_entry point
2815 __ pop(atos);
2816 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // kills R11_scratch1
2817 do_oop_store(_masm, Rclass_or_obj, Roffset, R17_tos, Rscratch, Rscratch2, Rscratch3, _bs->kind(), false /* precise */, true /* check null */);
2818 if (!is_static) { patch_bytecode(Bytecodes::_fast_aputfield, Rbc, Rscratch, true, byte_no); }
2819 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2820 __ beq(CR_is_vol, Lvolatile); // Volatile?
2821 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2822
2823 __ align(32, 12);
2824 __ bind(Lvolatile);
2825 __ fence();
2826 }
2827 // fallthru: __ b(Lexit);
2828
2829 #ifdef ASSERT
2830 for (int i = 0; i<number_of_states; ++i) {
2831 assert(branch_table[i], "put initialization");
2832 //tty->print_cr("put: %s_branch_table[%d] = 0x%llx (opcode 0x%llx)",
2833 // is_static ? "static" : "field", i, branch_table[i], *((unsigned int*)branch_table[i]));
2834 }
2835 #endif
2836 }
2837
2838 void TemplateTable::putfield(int byte_no) {
2839 putfield_or_static(byte_no, false);
2840 }
2841
2842 void TemplateTable::putstatic(int byte_no) {
2843 putfield_or_static(byte_no, true);
2844 }
2845
2846 // See SPARC. On PPC64, we have a different jvmti_post_field_mod which does the job.
2847 void TemplateTable::jvmti_post_fast_field_mod() {
2848 __ should_not_reach_here();
2849 }
2850
2851 void TemplateTable::fast_storefield(TosState state) {
2852 transition(state, vtos);
2853
2854 const Register Rcache = R5_ARG3, // Do not use ARG1/2 (causes trouble in jvmti_post_field_mod).
2855 Rclass_or_obj = R31, // Needs to survive C call.
2856 Roffset = R22_tmp2, // Needs to survive C call.
2857 Rflags = R3_ARG1,
2858 Rscratch = R11_scratch1,
2859 Rscratch2 = R12_scratch2,
2860 Rscratch3 = R4_ARG2;
2861 const ConditionRegister CR_is_vol = CCR2; // Non-volatile condition register (survives runtime call in do_oop_store).
2862
2863 // Constant pool already resolved => Load flags and offset of field.
2864 __ get_cache_and_index_at_bcp(Rcache, 1);
2865 jvmti_post_field_mod(Rcache, Rscratch, false /* not static */);
2866 load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
2867
2868 // Get the obj and the final store addr.
2869 pop_and_check_object(Rclass_or_obj); // Kills R11_scratch1.
2870
2871 // Get volatile flag.
2872 __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2873 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) { __ cmpdi(CR_is_vol, Rscratch, 1); }
2874 {
2875 Label LnotVolatile;
2876 __ beq(CCR0, LnotVolatile);
2877 __ release();
2878 __ align(32, 12);
2879 __ bind(LnotVolatile);
2880 }
2881
2882 // Do the store and fencing.
2883 switch(bytecode()) {
2884 case Bytecodes::_fast_aputfield:
2885 // Store into the field.
2886 do_oop_store(_masm, Rclass_or_obj, Roffset, R17_tos, Rscratch, Rscratch2, Rscratch3, _bs->kind(), false /* precise */, true /* check null */);
2887 break;
2888
2889 case Bytecodes::_fast_iputfield:
2890 __ stwx(R17_tos, Rclass_or_obj, Roffset);
2891 break;
2892
2893 case Bytecodes::_fast_lputfield:
2894 __ stdx(R17_tos, Rclass_or_obj, Roffset);
2895 break;
2896
2897 case Bytecodes::_fast_bputfield:
2898 __ stbx(R17_tos, Rclass_or_obj, Roffset);
2899 break;
2900
2901 case Bytecodes::_fast_cputfield:
2902 case Bytecodes::_fast_sputfield:
2903 __ sthx(R17_tos, Rclass_or_obj, Roffset);
2904 break;
2905
2906 case Bytecodes::_fast_fputfield:
2907 __ stfsx(F15_ftos, Rclass_or_obj, Roffset);
2908 break;
2909
2910 case Bytecodes::_fast_dputfield:
2911 __ stfdx(F15_ftos, Rclass_or_obj, Roffset);
2912 break;
2913
2914 default: ShouldNotReachHere();
2915 }
2916
2917 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2918 Label LVolatile;
2919 __ beq(CR_is_vol, LVolatile);
2920 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2921
2922 __ align(32, 12);
2923 __ bind(LVolatile);
2924 __ fence();
2925 }
2926 }
2927
2928 void TemplateTable::fast_accessfield(TosState state) {
2929 transition(atos, state);
2930
2931 Label LisVolatile;
2932 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2933
2934 const Register Rcache = R3_ARG1,
2935 Rclass_or_obj = R17_tos,
2936 Roffset = R22_tmp2,
2937 Rflags = R23_tmp3,
2938 Rscratch = R12_scratch2;
2939
2940 // Constant pool already resolved. Get the field offset.
2941 __ get_cache_and_index_at_bcp(Rcache, 1);
2942 load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
2943
2944 // JVMTI support
2945 jvmti_post_field_access(Rcache, Rscratch, false, true);
2946
2947 // Get the load address.
2948 __ null_check_throw(Rclass_or_obj, -1, Rscratch);
2949
2950 // Get volatile flag.
2951 __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2952 __ bne(CCR0, LisVolatile);
2953
2954 switch(bytecode()) {
2955 case Bytecodes::_fast_agetfield:
2956 {
2957 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
2958 __ verify_oop(R17_tos);
2959 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
2960
2961 __ bind(LisVolatile);
2962 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
2963 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
2964 __ verify_oop(R17_tos);
2965 __ twi_0(R17_tos);
2966 __ isync();
2967 break;
2968 }
2969 case Bytecodes::_fast_igetfield:
2970 {
2971 __ lwax(R17_tos, Rclass_or_obj, Roffset);
2972 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
2973
2974 __ bind(LisVolatile);
2975 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
2976 __ lwax(R17_tos, Rclass_or_obj, Roffset);
2977 __ twi_0(R17_tos);
2978 __ isync();
2979 break;
2980 }
2981 case Bytecodes::_fast_lgetfield:
2982 {
2983 __ ldx(R17_tos, Rclass_or_obj, Roffset);
2984 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
2985
2986 __ bind(LisVolatile);
2987 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
2988 __ ldx(R17_tos, Rclass_or_obj, Roffset);
2989 __ twi_0(R17_tos);
2990 __ isync();
2991 break;
2992 }
2993 case Bytecodes::_fast_bgetfield:
2994 {
2995 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
2996 __ extsb(R17_tos, R17_tos);
2997 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
2998
2999 __ bind(LisVolatile);
3000 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3001 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
3002 __ twi_0(R17_tos);
3003 __ extsb(R17_tos, R17_tos);
3004 __ isync();
3005 break;
3006 }
3007 case Bytecodes::_fast_cgetfield:
3008 {
3009 __ lhzx(R17_tos, Rclass_or_obj, Roffset);
3010 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3011
3012 __ bind(LisVolatile);
3013 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3014 __ lhzx(R17_tos, Rclass_or_obj, Roffset);
3015 __ twi_0(R17_tos);
3016 __ isync();
3017 break;
3018 }
3019 case Bytecodes::_fast_sgetfield:
3020 {
3021 __ lhax(R17_tos, Rclass_or_obj, Roffset);
3022 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3023
3024 __ bind(LisVolatile);
3025 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3026 __ lhax(R17_tos, Rclass_or_obj, Roffset);
3027 __ twi_0(R17_tos);
3028 __ isync();
3029 break;
3030 }
3031 case Bytecodes::_fast_fgetfield:
3032 {
3033 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3034 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3035
3036 __ bind(LisVolatile);
3037 Label Ldummy;
3038 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3039 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3040 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3041 __ bne_predict_not_taken(CCR0, Ldummy);
3042 __ bind(Ldummy);
3043 __ isync();
3044 break;
3045 }
3046 case Bytecodes::_fast_dgetfield:
3047 {
3048 __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
3049 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3050
3051 __ bind(LisVolatile);
3052 Label Ldummy;
3053 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3054 __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
3055 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3056 __ bne_predict_not_taken(CCR0, Ldummy);
3057 __ bind(Ldummy);
3058 __ isync();
3059 break;
3060 }
3061 default: ShouldNotReachHere();
3062 }
3063 }
3064
3065 void TemplateTable::fast_xaccess(TosState state) {
3066 transition(vtos, state);
3067
3068 Label LisVolatile;
3069 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
3070 const Register Rcache = R3_ARG1,
3071 Rclass_or_obj = R17_tos,
3072 Roffset = R22_tmp2,
3073 Rflags = R23_tmp3,
3074 Rscratch = R12_scratch2;
3075
3076 __ ld(Rclass_or_obj, 0, R18_locals);
3077
3078 // Constant pool already resolved. Get the field offset.
3079 __ get_cache_and_index_at_bcp(Rcache, 2);
3080 load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
3081
3082 // JVMTI support not needed, since we switch back to single bytecode as soon as debugger attaches.
3083
3084 // Needed to report exception at the correct bcp.
3085 __ addi(R14_bcp, R14_bcp, 1);
3086
3087 // Get the load address.
3088 __ null_check_throw(Rclass_or_obj, -1, Rscratch);
3089
3090 // Get volatile flag.
3091 __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
3092 __ bne(CCR0, LisVolatile);
3093
3094 switch(state) {
3095 case atos:
3096 {
3097 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3098 __ verify_oop(R17_tos);
3099 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3100
3101 __ bind(LisVolatile);
3102 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3103 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3104 __ verify_oop(R17_tos);
3105 __ twi_0(R17_tos);
3106 __ isync();
3107 break;
3108 }
3109 case itos:
3110 {
3111 __ lwax(R17_tos, Rclass_or_obj, Roffset);
3112 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3113
3114 __ bind(LisVolatile);
3115 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3116 __ lwax(R17_tos, Rclass_or_obj, Roffset);
3117 __ twi_0(R17_tos);
3118 __ isync();
3119 break;
3120 }
3121 case ftos:
3122 {
3123 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3124 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3125
3126 __ bind(LisVolatile);
3127 Label Ldummy;
3128 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3129 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3130 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3131 __ bne_predict_not_taken(CCR0, Ldummy);
3132 __ bind(Ldummy);
3133 __ isync();
3134 break;
3135 }
3136 default: ShouldNotReachHere();
3137 }
3138 __ addi(R14_bcp, R14_bcp, -1);
3139 }
3140
3141 // ============================================================================
3142 // Calls
3143
3144 // Common code for invoke
3145 //
3146 // Input:
3147 // - byte_no
3148 //
3149 // Output:
3150 // - Rmethod: The method to invoke next.
3151 // - Rret_addr: The return address to return to.
3152 // - Rindex: MethodType (invokehandle) or CallSite obj (invokedynamic)
3153 // - Rrecv: Cache for "this" pointer, might be noreg if static call.
3154 // - Rflags: Method flags from const pool cache.
3155 //
3156 // Kills:
3157 // - Rscratch1
3158 //
3159 void TemplateTable::prepare_invoke(int byte_no,
3160 Register Rmethod, // linked method (or i-klass)
3161 Register Rret_addr,// return address
3162 Register Rindex, // itable index, MethodType, etc.
3163 Register Rrecv, // If caller wants to see it.
3164 Register Rflags, // If caller wants to test it.
3165 Register Rscratch
3166 ) {
3167 // Determine flags.
3168 const Bytecodes::Code code = bytecode();
3169 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3170 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3171 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3172 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3173 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3174 const bool load_receiver = (Rrecv != noreg);
3175 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3176
3177 assert_different_registers(Rmethod, Rindex, Rflags, Rscratch);
3178 assert_different_registers(Rmethod, Rrecv, Rflags, Rscratch);
3179 assert_different_registers(Rret_addr, Rscratch);
3180
3181 load_invoke_cp_cache_entry(byte_no, Rmethod, Rindex, Rflags, is_invokevirtual, false, is_invokedynamic);
3182
3183 // Saving of SP done in call_from_interpreter.
3184
3185 // Maybe push "appendix" to arguments.
3186 if (is_invokedynamic || is_invokehandle) {
3187 Label Ldone;
3188 __ rldicl_(R0, Rflags, 64-ConstantPoolCacheEntry::has_appendix_shift, 63);
3189 __ beq(CCR0, Ldone);
3190 // Push "appendix" (MethodType, CallSite, etc.).
3191 // This must be done before we get the receiver,
3192 // since the parameter_size includes it.
3193 __ load_resolved_reference_at_index(Rscratch, Rindex);
3194 __ verify_oop(Rscratch);
3195 __ push_ptr(Rscratch);
3196 __ bind(Ldone);
3197 }
3198
3199 // Load receiver if needed (after appendix is pushed so parameter size is correct).
3200 if (load_receiver) {
3201 const Register Rparam_count = Rscratch;
3202 __ andi(Rparam_count, Rflags, ConstantPoolCacheEntry::parameter_size_mask);
3203 __ load_receiver(Rparam_count, Rrecv);
3204 __ verify_oop(Rrecv);
3205 }
3206
3207 // Get return address.
3208 {
3209 Register Rtable_addr = Rscratch;
3210 Register Rret_type = Rret_addr;
3211 address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3212
3213 // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3214 __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3215 __ load_dispatch_table(Rtable_addr, (address*)table_addr);
3216 __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3217 // Get return address.
3218 __ ldx(Rret_addr, Rtable_addr, Rret_type);
3219 }
3220 }
3221
3222 // Helper for virtual calls. Load target out of vtable and jump off!
3223 // Kills all passed registers.
3224 void TemplateTable::generate_vtable_call(Register Rrecv_klass, Register Rindex, Register Rret, Register Rtemp) {
3225
3226 assert_different_registers(Rrecv_klass, Rtemp, Rret);
3227 const Register Rtarget_method = Rindex;
3228
3229 // Get target method & entry point.
3230 const int base = InstanceKlass::vtable_start_offset() * wordSize;
3231 // Calc vtable addr scale the vtable index by 8.
3232 __ sldi(Rindex, Rindex, exact_log2(vtableEntry::size() * wordSize));
3233 // Load target.
3234 __ addi(Rrecv_klass, Rrecv_klass, base + vtableEntry::method_offset_in_bytes());
3235 __ ldx(Rtarget_method, Rindex, Rrecv_klass);
3236 // Argument and return type profiling.
3237 __ profile_arguments_type(Rtarget_method, Rrecv_klass /* scratch1 */, Rtemp /* scratch2 */, true);
3238 __ call_from_interpreter(Rtarget_method, Rret, Rrecv_klass /* scratch1 */, Rtemp /* scratch2 */);
3239 }
3240
3241 // Virtual or final call. Final calls are rewritten on the fly to run through "fast_finalcall" next time.
3242 void TemplateTable::invokevirtual(int byte_no) {
3243 transition(vtos, vtos);
3244
3245 Register Rtable_addr = R11_scratch1,
3246 Rret_type = R12_scratch2,
3247 Rret_addr = R5_ARG3,
3248 Rflags = R22_tmp2, // Should survive C call.
3249 Rrecv = R3_ARG1,
3250 Rrecv_klass = Rrecv,
3251 Rvtableindex_or_method = R31, // Should survive C call.
3252 Rnum_params = R4_ARG2,
3253 Rnew_bc = R6_ARG4;
3254
3255 Label LnotFinal;
3256
3257 load_invoke_cp_cache_entry(byte_no, Rvtableindex_or_method, noreg, Rflags, /*virtual*/ true, false, false);
3258
3259 __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_vfinal_shift);
3260 __ bfalse(CCR0, LnotFinal);
3261
3262 patch_bytecode(Bytecodes::_fast_invokevfinal, Rnew_bc, R12_scratch2);
3263 invokevfinal_helper(Rvtableindex_or_method, Rflags, R11_scratch1, R12_scratch2);
3264
3265 __ align(32, 12);
3266 __ bind(LnotFinal);
3267 // Load "this" pointer (receiver).
3268 __ rldicl(Rnum_params, Rflags, 64, 48);
3269 __ load_receiver(Rnum_params, Rrecv);
3270 __ verify_oop(Rrecv);
3271
3272 // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3273 __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3274 __ load_dispatch_table(Rtable_addr, Interpreter::invoke_return_entry_table());
3275 __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3276 __ ldx(Rret_addr, Rret_type, Rtable_addr);
3277 __ null_check_throw(Rrecv, oopDesc::klass_offset_in_bytes(), R11_scratch1);
3278 __ load_klass(Rrecv_klass, Rrecv);
3279 __ verify_klass_ptr(Rrecv_klass);
3280 __ profile_virtual_call(Rrecv_klass, R11_scratch1, R12_scratch2, false);
3281
3282 generate_vtable_call(Rrecv_klass, Rvtableindex_or_method, Rret_addr, R11_scratch1);
3283 }
3284
3285 void TemplateTable::fast_invokevfinal(int byte_no) {
3286 transition(vtos, vtos);
3287
3288 assert(byte_no == f2_byte, "use this argument");
3289 Register Rflags = R22_tmp2,
3290 Rmethod = R31;
3291 load_invoke_cp_cache_entry(byte_no, Rmethod, noreg, Rflags, /*virtual*/ true, /*is_invokevfinal*/ true, false);
3292 invokevfinal_helper(Rmethod, Rflags, R11_scratch1, R12_scratch2);
3293 }
3294
3295 void TemplateTable::invokevfinal_helper(Register Rmethod, Register Rflags, Register Rscratch1, Register Rscratch2) {
3296
3297 assert_different_registers(Rmethod, Rflags, Rscratch1, Rscratch2);
3298
3299 // Load receiver from stack slot.
3300 Register Rrecv = Rscratch2;
3301 Register Rnum_params = Rrecv;
3302
3303 __ ld(Rnum_params, in_bytes(Method::const_offset()), Rmethod);
3304 __ lhz(Rnum_params /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), Rnum_params);
3305
3306 // Get return address.
3307 Register Rtable_addr = Rscratch1,
3308 Rret_addr = Rflags,
3309 Rret_type = Rret_addr;
3310 // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3311 __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3312 __ load_dispatch_table(Rtable_addr, Interpreter::invoke_return_entry_table());
3313 __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3314 __ ldx(Rret_addr, Rret_type, Rtable_addr);
3315
3316 // Load receiver and receiver NULL check.
3317 __ load_receiver(Rnum_params, Rrecv);
3318 __ null_check_throw(Rrecv, -1, Rscratch1);
3319
3320 __ profile_final_call(Rrecv, Rscratch1);
3321 // Argument and return type profiling.
3322 __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, true);
3323
3324 // Do the call.
3325 __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1, Rscratch2);
3326 }
3327
3328 void TemplateTable::invokespecial(int byte_no) {
3329 assert(byte_no == f1_byte, "use this argument");
3330 transition(vtos, vtos);
3331
3332 Register Rtable_addr = R3_ARG1,
3333 Rret_addr = R4_ARG2,
3334 Rflags = R5_ARG3,
3335 Rreceiver = R6_ARG4,
3336 Rmethod = R31;
3337
3338 prepare_invoke(byte_no, Rmethod, Rret_addr, noreg, Rreceiver, Rflags, R11_scratch1);
3339
3340 // Receiver NULL check.
3341 __ null_check_throw(Rreceiver, -1, R11_scratch1);
3342
3343 __ profile_call(R11_scratch1, R12_scratch2);
3344 // Argument and return type profiling.
3345 __ profile_arguments_type(Rmethod, R11_scratch1, R12_scratch2, false);
3346 __ call_from_interpreter(Rmethod, Rret_addr, R11_scratch1, R12_scratch2);
3347 }
3348
3349 void TemplateTable::invokestatic(int byte_no) {
3350 assert(byte_no == f1_byte, "use this argument");
3351 transition(vtos, vtos);
3352
3353 Register Rtable_addr = R3_ARG1,
3354 Rret_addr = R4_ARG2,
3355 Rflags = R5_ARG3;
3356
3357 prepare_invoke(byte_no, R19_method, Rret_addr, noreg, noreg, Rflags, R11_scratch1);
3358
3359 __ profile_call(R11_scratch1, R12_scratch2);
3360 // Argument and return type profiling.
3361 __ profile_arguments_type(R19_method, R11_scratch1, R12_scratch2, false);
3362 __ call_from_interpreter(R19_method, Rret_addr, R11_scratch1, R12_scratch2);
3363 }
3364
3365 void TemplateTable::invokeinterface_object_method(Register Rrecv_klass,
3366 Register Rret,
3367 Register Rflags,
3368 Register Rindex,
3369 Register Rtemp1,
3370 Register Rtemp2) {
3371
3372 assert_different_registers(Rindex, Rret, Rrecv_klass, Rflags, Rtemp1, Rtemp2);
3373 Label LnotFinal;
3374
3375 // Check for vfinal.
3376 __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_vfinal_shift);
3377 __ bfalse(CCR0, LnotFinal);
3378
3379 Register Rscratch = Rflags; // Rflags is dead now.
3380
3381 // Final call case.
3382 __ profile_final_call(Rtemp1, Rscratch);
3383 // Argument and return type profiling.
3384 __ profile_arguments_type(Rindex, Rscratch, Rrecv_klass /* scratch */, true);
3385 // Do the final call - the index (f2) contains the method.
3386 __ call_from_interpreter(Rindex, Rret, Rscratch, Rrecv_klass /* scratch */);
3387
3388 // Non-final callc case.
3389 __ bind(LnotFinal);
3390 __ profile_virtual_call(Rrecv_klass, Rtemp1, Rscratch, false);
3391 generate_vtable_call(Rrecv_klass, Rindex, Rret, Rscratch);
3392 }
3393
3394 void TemplateTable::invokeinterface(int byte_no) {
3395 assert(byte_no == f1_byte, "use this argument");
3396 transition(vtos, vtos);
3397
3398 const Register Rscratch1 = R11_scratch1,
3399 Rscratch2 = R12_scratch2,
3400 Rscratch3 = R9_ARG7,
3401 Rscratch4 = R10_ARG8,
3402 Rtable_addr = Rscratch2,
3403 Rinterface_klass = R5_ARG3,
3404 Rret_type = R8_ARG6,
3405 Rret_addr = Rret_type,
3406 Rindex = R6_ARG4,
3407 Rreceiver = R4_ARG2,
3408 Rrecv_klass = Rreceiver,
3409 Rflags = R7_ARG5;
3410
3411 prepare_invoke(byte_no, Rinterface_klass, Rret_addr, Rindex, Rreceiver, Rflags, Rscratch1);
3412
3413 // Get receiver klass.
3414 __ null_check_throw(Rreceiver, oopDesc::klass_offset_in_bytes(), Rscratch3);
3415 __ load_klass(Rrecv_klass, Rreceiver);
3416
3417 // Check corner case object method.
3418 Label LobjectMethod;
3419
3420 __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_forced_virtual_shift);
3421 __ btrue(CCR0, LobjectMethod);
3422
3423 // Fallthrough: The normal invokeinterface case.
3424 __ profile_virtual_call(Rrecv_klass, Rscratch1, Rscratch2, false);
3425
3426 // Find entry point to call.
3427 Label Lthrow_icc, Lthrow_ame;
3428 // Result will be returned in Rindex.
3429 __ mr(Rscratch4, Rrecv_klass);
3430 __ mr(Rscratch3, Rindex);
3431 __ lookup_interface_method(Rrecv_klass, Rinterface_klass, Rindex, Rindex, Rscratch1, Rscratch2, Lthrow_icc);
3432
3433 __ cmpdi(CCR0, Rindex, 0);
3434 __ beq(CCR0, Lthrow_ame);
3435 // Found entry. Jump off!
3436 // Argument and return type profiling.
3437 __ profile_arguments_type(Rindex, Rscratch1, Rscratch2, true);
3438 __ call_from_interpreter(Rindex, Rret_addr, Rscratch1, Rscratch2);
3439
3440 // Vtable entry was NULL => Throw abstract method error.
3441 __ bind(Lthrow_ame);
3442 __ mr(Rrecv_klass, Rscratch4);
3443 __ mr(Rindex, Rscratch3);
3444 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3445
3446 // Interface was not found => Throw incompatible class change error.
3447 __ bind(Lthrow_icc);
3448 __ mr(Rrecv_klass, Rscratch4);
3449 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeError));
3450
3451 __ should_not_reach_here();
3452
3453 // Special case of invokeinterface called for virtual method of
3454 // java.lang.Object. See ConstantPoolCacheEntry::set_method() for details:
3455 // The invokeinterface was rewritten to a invokevirtual, hence we have
3456 // to handle this corner case. This code isn't produced by javac, but could
3457 // be produced by another compliant java compiler.
3458 __ bind(LobjectMethod);
3459 invokeinterface_object_method(Rrecv_klass, Rret_addr, Rflags, Rindex, Rscratch1, Rscratch2);
3460 }
3461
3462 void TemplateTable::invokedynamic(int byte_no) {
3463 transition(vtos, vtos);
3464
3465 const Register Rret_addr = R3_ARG1,
3466 Rflags = R4_ARG2,
3467 Rmethod = R22_tmp2,
3468 Rscratch1 = R11_scratch1,
3469 Rscratch2 = R12_scratch2;
3470
3471 prepare_invoke(byte_no, Rmethod, Rret_addr, Rscratch1, noreg, Rflags, Rscratch2);
3472
3473 // Profile this call.
3474 __ profile_call(Rscratch1, Rscratch2);
3475
3476 // Off we go. With the new method handles, we don't jump to a method handle
3477 // entry any more. Instead, we pushed an "appendix" in prepare invoke, which happens
3478 // to be the callsite object the bootstrap method returned. This is passed to a
3479 // "link" method which does the dispatch (Most likely just grabs the MH stored
3480 // inside the callsite and does an invokehandle).
3481 // Argument and return type profiling.
3482 __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, false);
3483 __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1 /* scratch1 */, Rscratch2 /* scratch2 */);
3484 }
3485
3486 void TemplateTable::invokehandle(int byte_no) {
3487 transition(vtos, vtos);
3488
3489 const Register Rret_addr = R3_ARG1,
3490 Rflags = R4_ARG2,
3491 Rrecv = R5_ARG3,
3492 Rmethod = R22_tmp2,
3493 Rscratch1 = R11_scratch1,
3494 Rscratch2 = R12_scratch2;
3495
3496 prepare_invoke(byte_no, Rmethod, Rret_addr, Rscratch1, Rrecv, Rflags, Rscratch2);
3497 __ verify_method_ptr(Rmethod);
3498 __ null_check_throw(Rrecv, -1, Rscratch2);
3499
3500 __ profile_final_call(Rrecv, Rscratch1);
3501
3502 // Still no call from handle => We call the method handle interpreter here.
3503 // Argument and return type profiling.
3504 __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, true);
3505 __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1 /* scratch1 */, Rscratch2 /* scratch2 */);
3506 }
3507
3508 // =============================================================================
3509 // Allocation
3510
3511 // Puts allocated obj ref onto the expression stack.
3512 void TemplateTable::_new() {
3513 transition(vtos, atos);
3514
3515 Label Lslow_case,
3516 Ldone,
3517 Linitialize_header,
3518 Lallocate_shared,
3519 Linitialize_object; // Including clearing the fields.
3520
3521 const Register RallocatedObject = R17_tos,
3522 RinstanceKlass = R9_ARG7,
3523 Rscratch = R11_scratch1,
3524 Roffset = R8_ARG6,
3525 Rinstance_size = Roffset,
3526 Rcpool = R4_ARG2,
3527 Rtags = R3_ARG1,
3528 Rindex = R5_ARG3;
3529
3530 const bool allow_shared_alloc = Universe::heap()->supports_inline_contig_alloc();
3531
3532 // --------------------------------------------------------------------------
3533 // Check if fast case is possible.
3534
3535 // Load pointers to const pool and const pool's tags array.
3536 __ get_cpool_and_tags(Rcpool, Rtags);
3537 // Load index of constant pool entry.
3538 __ get_2_byte_integer_at_bcp(1, Rindex, InterpreterMacroAssembler::Unsigned);
3539
3540 if (UseTLAB) {
3541 // Make sure the class we're about to instantiate has been resolved
3542 // This is done before loading instanceKlass to be consistent with the order
3543 // how Constant Pool is updated (see ConstantPoolCache::klass_at_put).
3544 __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3545 __ lbzx(Rtags, Rindex, Rtags);
3546
3547 __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3548 __ bne(CCR0, Lslow_case);
3549
3550 // Get instanceKlass (load from Rcpool + sizeof(ConstantPool) + Rindex*BytesPerWord).
3551 __ sldi(Roffset, Rindex, LogBytesPerWord);
3552 __ addi(Rscratch, Rcpool, sizeof(ConstantPool));
3553 __ isync(); // Order load of instance Klass wrt. tags.
3554 __ ldx(RinstanceKlass, Roffset, Rscratch);
3555
3556 // Make sure klass is fully initialized and get instance_size.
3557 __ lbz(Rscratch, in_bytes(InstanceKlass::init_state_offset()), RinstanceKlass);
3558 __ lwz(Rinstance_size, in_bytes(Klass::layout_helper_offset()), RinstanceKlass);
3559
3560 __ cmpdi(CCR1, Rscratch, InstanceKlass::fully_initialized);
3561 // Make sure klass does not have has_finalizer, or is abstract, or interface or java/lang/Class.
3562 __ andi_(R0, Rinstance_size, Klass::_lh_instance_slow_path_bit); // slow path bit equals 0?
3563
3564 __ crnand(CCR0, Assembler::equal, CCR1, Assembler::equal); // slow path bit set or not fully initialized?
3565 __ beq(CCR0, Lslow_case);
3566
3567 // --------------------------------------------------------------------------
3568 // Fast case:
3569 // Allocate the instance.
3570 // 1) Try to allocate in the TLAB.
3571 // 2) If fail, and the TLAB is not full enough to discard, allocate in the shared Eden.
3572 // 3) If the above fails (or is not applicable), go to a slow case (creates a new TLAB, etc.).
3573
3574 Register RoldTopValue = RallocatedObject; // Object will be allocated here if it fits.
3575 Register RnewTopValue = R6_ARG4;
3576 Register RendValue = R7_ARG5;
3577
3578 // Check if we can allocate in the TLAB.
3579 __ ld(RoldTopValue, in_bytes(JavaThread::tlab_top_offset()), R16_thread);
3580 __ ld(RendValue, in_bytes(JavaThread::tlab_end_offset()), R16_thread);
3581
3582 __ add(RnewTopValue, Rinstance_size, RoldTopValue);
3583
3584 // If there is enough space, we do not CAS and do not clear.
3585 __ cmpld(CCR0, RnewTopValue, RendValue);
3586 __ bgt(CCR0, allow_shared_alloc ? Lallocate_shared : Lslow_case);
3587
3588 __ std(RnewTopValue, in_bytes(JavaThread::tlab_top_offset()), R16_thread);
3589
3590 if (ZeroTLAB) {
3591 // The fields have already been cleared.
3592 __ b(Linitialize_header);
3593 } else {
3594 // Initialize both the header and fields.
3595 __ b(Linitialize_object);
3596 }
3597
3598 // Fall through: TLAB was too small.
3599 if (allow_shared_alloc) {
3600 Register RtlabWasteLimitValue = R10_ARG8;
3601 Register RfreeValue = RnewTopValue;
3602
3603 __ bind(Lallocate_shared);
3604 // Check if tlab should be discarded (refill_waste_limit >= free).
3605 __ ld(RtlabWasteLimitValue, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), R16_thread);
3606 __ subf(RfreeValue, RoldTopValue, RendValue);
3607 __ srdi(RfreeValue, RfreeValue, LogHeapWordSize); // in dwords
3608 __ cmpld(CCR0, RtlabWasteLimitValue, RfreeValue);
3609 __ bge(CCR0, Lslow_case);
3610
3611 // Increment waste limit to prevent getting stuck on this slow path.
3612 __ addi(RtlabWasteLimitValue, RtlabWasteLimitValue, (int)ThreadLocalAllocBuffer::refill_waste_limit_increment());
3613 __ std(RtlabWasteLimitValue, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), R16_thread);
3614 }
3615 // else: No allocation in the shared eden. // fallthru: __ b(Lslow_case);
3616 }
3617 // else: Always go the slow path.
3618
3619 // --------------------------------------------------------------------------
3620 // slow case
3621 __ bind(Lslow_case);
3622 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), Rcpool, Rindex);
3623
3624 if (UseTLAB) {
3625 __ b(Ldone);
3626 // --------------------------------------------------------------------------
3627 // Init1: Zero out newly allocated memory.
3628
3629 if (!ZeroTLAB || allow_shared_alloc) {
3630 // Clear object fields.
3631 __ bind(Linitialize_object);
3632
3633 // Initialize remaining object fields.
3634 Register Rbase = Rtags;
3635 __ addi(Rinstance_size, Rinstance_size, 7 - (int)sizeof(oopDesc));
3636 __ addi(Rbase, RallocatedObject, sizeof(oopDesc));
3637 __ srdi(Rinstance_size, Rinstance_size, 3);
3638
3639 // Clear out object skipping header. Takes also care of the zero length case.
3640 __ clear_memory_doubleword(Rbase, Rinstance_size);
3641 // fallthru: __ b(Linitialize_header);
3642 }
3643
3644 // --------------------------------------------------------------------------
3645 // Init2: Initialize the header: mark, klass
3646 __ bind(Linitialize_header);
3647
3648 // Init mark.
3649 if (UseBiasedLocking) {
3650 __ ld(Rscratch, in_bytes(Klass::prototype_header_offset()), RinstanceKlass);
3651 } else {
3652 __ load_const_optimized(Rscratch, markOopDesc::prototype(), R0);
3653 }
3654 __ std(Rscratch, oopDesc::mark_offset_in_bytes(), RallocatedObject);
3655
3656 // Init klass.
3657 __ store_klass_gap(RallocatedObject);
3658 __ store_klass(RallocatedObject, RinstanceKlass, Rscratch); // klass (last for cms)
3659
3660 // Check and trigger dtrace event.
3661 {
3662 SkipIfEqualZero skip_if(_masm, Rscratch, &DTraceAllocProbes);
3663 __ push(atos);
3664 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc));
3665 __ pop(atos);
3666 }
3667 }
3668
3669 // continue
3670 __ bind(Ldone);
3671
3672 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3673 __ membar(Assembler::StoreStore);
3674 }
3675
3676 void TemplateTable::newarray() {
3677 transition(itos, atos);
3678
3679 __ lbz(R4, 1, R14_bcp);
3680 __ extsw(R5, R17_tos);
3681 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), R4, R5 /* size */);
3682
3683 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3684 __ membar(Assembler::StoreStore);
3685 }
3686
3687 void TemplateTable::anewarray() {
3688 transition(itos, atos);
3689
3690 __ get_constant_pool(R4);
3691 __ get_2_byte_integer_at_bcp(1, R5, InterpreterMacroAssembler::Unsigned);
3692 __ extsw(R6, R17_tos); // size
3693 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), R4 /* pool */, R5 /* index */, R6 /* size */);
3694
3695 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3696 __ membar(Assembler::StoreStore);
3697 }
3698
3699 // Allocate a multi dimensional array
3700 void TemplateTable::multianewarray() {
3701 transition(vtos, atos);
3702
3703 Register Rptr = R31; // Needs to survive C call.
3704
3705 // Put ndims * wordSize into frame temp slot
3706 __ lbz(Rptr, 3, R14_bcp);
3707 __ sldi(Rptr, Rptr, Interpreter::logStackElementSize);
3708 // Esp points past last_dim, so set to R4 to first_dim address.
3709 __ add(R4, Rptr, R15_esp);
3710 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), R4 /* first_size_address */);
3711 // Pop all dimensions off the stack.
3712 __ add(R15_esp, Rptr, R15_esp);
3713
3714 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3715 __ membar(Assembler::StoreStore);
3716 }
3717
3718 void TemplateTable::arraylength() {
3719 transition(atos, itos);
3720
3721 Label LnoException;
3722 __ verify_oop(R17_tos);
3723 __ null_check_throw(R17_tos, arrayOopDesc::length_offset_in_bytes(), R11_scratch1);
3724 __ lwa(R17_tos, arrayOopDesc::length_offset_in_bytes(), R17_tos);
3725 }
3726
3727 // ============================================================================
3728 // Typechecks
3729
3730 void TemplateTable::checkcast() {
3731 transition(atos, atos);
3732
3733 Label Ldone, Lis_null, Lquicked, Lresolved;
3734 Register Roffset = R6_ARG4,
3735 RobjKlass = R4_ARG2,
3736 RspecifiedKlass = R5_ARG3, // Generate_ClassCastException_verbose_handler will read value from this register.
3737 Rcpool = R11_scratch1,
3738 Rtags = R12_scratch2;
3739
3740 // Null does not pass.
3741 __ cmpdi(CCR0, R17_tos, 0);
3742 __ beq(CCR0, Lis_null);
3743
3744 // Get constant pool tag to find out if the bytecode has already been "quickened".
3745 __ get_cpool_and_tags(Rcpool, Rtags);
3746
3747 __ get_2_byte_integer_at_bcp(1, Roffset, InterpreterMacroAssembler::Unsigned);
3748
3749 __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3750 __ lbzx(Rtags, Rtags, Roffset);
3751
3752 __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3753 __ beq(CCR0, Lquicked);
3754
3755 // Call into the VM to "quicken" instanceof.
3756 __ push_ptr(); // for GC
3757 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3758 __ get_vm_result_2(RspecifiedKlass);
3759 __ pop_ptr(); // Restore receiver.
3760 __ b(Lresolved);
3761
3762 // Extract target class from constant pool.
3763 __ bind(Lquicked);
3764 __ sldi(Roffset, Roffset, LogBytesPerWord);
3765 __ addi(Rcpool, Rcpool, sizeof(ConstantPool));
3766 __ isync(); // Order load of specified Klass wrt. tags.
3767 __ ldx(RspecifiedKlass, Rcpool, Roffset);
3768
3769 // Do the checkcast.
3770 __ bind(Lresolved);
3771 // Get value klass in RobjKlass.
3772 __ load_klass(RobjKlass, R17_tos);
3773 // Generate a fast subtype check. Branch to cast_ok if no failure. Return 0 if failure.
3774 __ gen_subtype_check(RobjKlass, RspecifiedKlass, /*3 temp regs*/ Roffset, Rcpool, Rtags, /*target if subtype*/ Ldone);
3775
3776 // Not a subtype; so must throw exception
3777 // Target class oop is in register R6_ARG4 == RspecifiedKlass by convention.
3778 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ClassCastException_entry);
3779 __ mtctr(R11_scratch1);
3780 __ bctr();
3781
3782 // Profile the null case.
3783 __ align(32, 12);
3784 __ bind(Lis_null);
3785 __ profile_null_seen(R11_scratch1, Rtags); // Rtags used as scratch.
3786
3787 __ align(32, 12);
3788 __ bind(Ldone);
3789 }
3790
3791 // Output:
3792 // - tos == 0: Obj was null or not an instance of class.
3793 // - tos == 1: Obj was an instance of class.
3794 void TemplateTable::instanceof() {
3795 transition(atos, itos);
3796
3797 Label Ldone, Lis_null, Lquicked, Lresolved;
3798 Register Roffset = R5_ARG3,
3799 RobjKlass = R4_ARG2,
3800 RspecifiedKlass = R6_ARG4, // Generate_ClassCastException_verbose_handler will expect the value in this register.
3801 Rcpool = R11_scratch1,
3802 Rtags = R12_scratch2;
3803
3804 // Null does not pass.
3805 __ cmpdi(CCR0, R17_tos, 0);
3806 __ beq(CCR0, Lis_null);
3807
3808 // Get constant pool tag to find out if the bytecode has already been "quickened".
3809 __ get_cpool_and_tags(Rcpool, Rtags);
3810
3811 __ get_2_byte_integer_at_bcp(1, Roffset, InterpreterMacroAssembler::Unsigned);
3812
3813 __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3814 __ lbzx(Rtags, Rtags, Roffset);
3815
3816 __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3817 __ beq(CCR0, Lquicked);
3818
3819 // Call into the VM to "quicken" instanceof.
3820 __ push_ptr(); // for GC
3821 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3822 __ get_vm_result_2(RspecifiedKlass);
3823 __ pop_ptr(); // Restore receiver.
3824 __ b(Lresolved);
3825
3826 // Extract target class from constant pool.
3827 __ bind(Lquicked);
3828 __ sldi(Roffset, Roffset, LogBytesPerWord);
3829 __ addi(Rcpool, Rcpool, sizeof(ConstantPool));
3830 __ isync(); // Order load of specified Klass wrt. tags.
3831 __ ldx(RspecifiedKlass, Rcpool, Roffset);
3832
3833 // Do the checkcast.
3834 __ bind(Lresolved);
3835 // Get value klass in RobjKlass.
3836 __ load_klass(RobjKlass, R17_tos);
3837 // Generate a fast subtype check. Branch to cast_ok if no failure. Return 0 if failure.
3838 __ li(R17_tos, 1);
3839 __ gen_subtype_check(RobjKlass, RspecifiedKlass, /*3 temp regs*/ Roffset, Rcpool, Rtags, /*target if subtype*/ Ldone);
3840 __ li(R17_tos, 0);
3841
3842 if (ProfileInterpreter) {
3843 __ b(Ldone);
3844 }
3845
3846 // Profile the null case.
3847 __ align(32, 12);
3848 __ bind(Lis_null);
3849 __ profile_null_seen(Rcpool, Rtags); // Rcpool and Rtags used as scratch.
3850
3851 __ align(32, 12);
3852 __ bind(Ldone);
3853 }
3854
3855 // =============================================================================
3856 // Breakpoints
3857
3858 void TemplateTable::_breakpoint() {
3859 transition(vtos, vtos);
3860
3861 // Get the unpatched byte code.
3862 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), R19_method, R14_bcp);
3863 __ mr(R31, R3_RET);
3864
3865 // Post the breakpoint event.
3866 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), R19_method, R14_bcp);
3867
3868 // Complete the execution of original bytecode.
3869 __ dispatch_Lbyte_code(vtos, R31, Interpreter::normal_table(vtos));
3870 }
3871
3872 // =============================================================================
3873 // Exceptions
3874
3875 void TemplateTable::athrow() {
3876 transition(atos, vtos);
3877
3878 // Exception oop is in tos
3879 __ verify_oop(R17_tos);
3880
3881 __ null_check_throw(R17_tos, -1, R11_scratch1);
3882
3883 // Throw exception interpreter entry expects exception oop to be in R3.
3884 __ mr(R3_RET, R17_tos);
3885 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::throw_exception_entry());
3886 __ mtctr(R11_scratch1);
3887 __ bctr();
3888 }
3889
3890 // =============================================================================
3891 // Synchronization
3892 // Searches the basic object lock list on the stack for a free slot
3893 // and uses it to lock the obect in tos.
3894 //
3895 // Recursive locking is enabled by exiting the search if the same
3896 // object is already found in the list. Thus, a new basic lock obj lock
3897 // is allocated "higher up" in the stack and thus is found first
3898 // at next monitor exit.
3899 void TemplateTable::monitorenter() {
3900 transition(atos, vtos);
3901
3902 __ verify_oop(R17_tos);
3903
3904 Register Rcurrent_monitor = R11_scratch1,
3905 Rcurrent_obj = R12_scratch2,
3906 Robj_to_lock = R17_tos,
3907 Rscratch1 = R3_ARG1,
3908 Rscratch2 = R4_ARG2,
3909 Rscratch3 = R5_ARG3,
3910 Rcurrent_obj_addr = R6_ARG4;
3911
3912 // ------------------------------------------------------------------------------
3913 // Null pointer exception.
3914 __ null_check_throw(Robj_to_lock, -1, R11_scratch1);
3915
3916 // Try to acquire a lock on the object.
3917 // Repeat until succeeded (i.e., until monitorenter returns true).
3918
3919 // ------------------------------------------------------------------------------
3920 // Find a free slot in the monitor block.
3921 Label Lfound, Lexit, Lallocate_new;
3922 ConditionRegister found_free_slot = CCR0,
3923 found_same_obj = CCR1,
3924 reached_limit = CCR6;
3925 {
3926 Label Lloop, Lentry;
3927 Register Rlimit = Rcurrent_monitor;
3928
3929 // Set up search loop - start with topmost monitor.
3930 __ add(Rcurrent_obj_addr, BasicObjectLock::obj_offset_in_bytes(), R26_monitor);
3931
3932 __ ld(Rlimit, 0, R1_SP);
3933 __ addi(Rlimit, Rlimit, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes() - BasicObjectLock::obj_offset_in_bytes())); // Monitor base
3934
3935 // Check if any slot is present => short cut to allocation if not.
3936 __ cmpld(reached_limit, Rcurrent_obj_addr, Rlimit);
3937 __ bgt(reached_limit, Lallocate_new);
3938
3939 // Pre-load topmost slot.
3940 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
3941 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
3942 // The search loop.
3943 __ bind(Lloop);
3944 // Found free slot?
3945 __ cmpdi(found_free_slot, Rcurrent_obj, 0);
3946 // Is this entry for same obj? If so, stop the search and take the found
3947 // free slot or allocate a new one to enable recursive locking.
3948 __ cmpd(found_same_obj, Rcurrent_obj, Robj_to_lock);
3949 __ cmpld(reached_limit, Rcurrent_obj_addr, Rlimit);
3950 __ beq(found_free_slot, Lexit);
3951 __ beq(found_same_obj, Lallocate_new);
3952 __ bgt(reached_limit, Lallocate_new);
3953 // Check if last allocated BasicLockObj reached.
3954 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
3955 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
3956 // Next iteration if unchecked BasicObjectLocks exist on the stack.
3957 __ b(Lloop);
3958 }
3959
3960 // ------------------------------------------------------------------------------
3961 // Check if we found a free slot.
3962 __ bind(Lexit);
3963
3964 __ addi(Rcurrent_monitor, Rcurrent_obj_addr, -(frame::interpreter_frame_monitor_size() * wordSize) - BasicObjectLock::obj_offset_in_bytes());
3965 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, - frame::interpreter_frame_monitor_size() * wordSize);
3966 __ b(Lfound);
3967
3968 // We didn't find a free BasicObjLock => allocate one.
3969 __ align(32, 12);
3970 __ bind(Lallocate_new);
3971 __ add_monitor_to_stack(false, Rscratch1, Rscratch2);
3972 __ mr(Rcurrent_monitor, R26_monitor);
3973 __ addi(Rcurrent_obj_addr, R26_monitor, BasicObjectLock::obj_offset_in_bytes());
3974
3975 // ------------------------------------------------------------------------------
3976 // We now have a slot to lock.
3977 __ bind(Lfound);
3978
3979 // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
3980 // The object has already been poped from the stack, so the expression stack looks correct.
3981 __ addi(R14_bcp, R14_bcp, 1);
3982
3983 __ std(Robj_to_lock, 0, Rcurrent_obj_addr);
3984 __ lock_object(Rcurrent_monitor, Robj_to_lock);
3985
3986 // Check if there's enough space on the stack for the monitors after locking.
3987 Label Lskip_stack_check;
3988 // Optimization: If the monitors stack section is less then a std page size (4K) don't run
3989 // the stack check. There should be enough shadow pages to fit that in.
3990 __ ld(Rscratch3, 0, R1_SP);
3991 __ sub(Rscratch3, Rscratch3, R26_monitor);
3992 __ cmpdi(CCR0, Rscratch3, 4*K);
3993 __ blt(CCR0, Lskip_stack_check);
3994
3995 DEBUG_ONLY(__ untested("stack overflow check during monitor enter");)
3996 __ li(Rscratch1, 0);
3997 __ generate_stack_overflow_check_with_compare_and_throw(Rscratch1, Rscratch2);
3998
3999 __ align(32, 12);
4000 __ bind(Lskip_stack_check);
4001
4002 // The bcp has already been incremented. Just need to dispatch to next instruction.
4003 __ dispatch_next(vtos);
4004 }
4005
4006 void TemplateTable::monitorexit() {
4007 transition(atos, vtos);
4008 __ verify_oop(R17_tos);
4009
4010 Register Rcurrent_monitor = R11_scratch1,
4011 Rcurrent_obj = R12_scratch2,
4012 Robj_to_lock = R17_tos,
4013 Rcurrent_obj_addr = R3_ARG1,
4014 Rlimit = R4_ARG2;
4015 Label Lfound, Lillegal_monitor_state;
4016
4017 // Check corner case: unbalanced monitorEnter / Exit.
4018 __ ld(Rlimit, 0, R1_SP);
4019 __ addi(Rlimit, Rlimit, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base
4020
4021 // Null pointer check.
4022 __ null_check_throw(Robj_to_lock, -1, R11_scratch1);
4023
4024 __ cmpld(CCR0, R26_monitor, Rlimit);
4025 __ bgt(CCR0, Lillegal_monitor_state);
4026
4027 // Find the corresponding slot in the monitors stack section.
4028 {
4029 Label Lloop;
4030
4031 // Start with topmost monitor.
4032 __ addi(Rcurrent_obj_addr, R26_monitor, BasicObjectLock::obj_offset_in_bytes());
4033 __ addi(Rlimit, Rlimit, BasicObjectLock::obj_offset_in_bytes());
4034 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4035 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4036
4037 __ bind(Lloop);
4038 // Is this entry for same obj?
4039 __ cmpd(CCR0, Rcurrent_obj, Robj_to_lock);
4040 __ beq(CCR0, Lfound);
4041
4042 // Check if last allocated BasicLockObj reached.
4043
4044 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4045 __ cmpld(CCR0, Rcurrent_obj_addr, Rlimit);
4046 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4047
4048 // Next iteration if unchecked BasicObjectLocks exist on the stack.
4049 __ ble(CCR0, Lloop);
4050 }
4051
4052 // Fell through without finding the basic obj lock => throw up!
4053 __ bind(Lillegal_monitor_state);
4054 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
4055 __ should_not_reach_here();
4056
4057 __ align(32, 12);
4058 __ bind(Lfound);
4059 __ addi(Rcurrent_monitor, Rcurrent_obj_addr,
4060 -(frame::interpreter_frame_monitor_size() * wordSize) - BasicObjectLock::obj_offset_in_bytes());
4061 __ unlock_object(Rcurrent_monitor);
4062 }
4063
4064 // ============================================================================
4065 // Wide bytecodes
4066
4067 // Wide instructions. Simply redirects to the wide entry point for that instruction.
4068 void TemplateTable::wide() {
4069 transition(vtos, vtos);
4070
4071 const Register Rtable = R11_scratch1,
4072 Rindex = R12_scratch2,
4073 Rtmp = R0;
4074
4075 __ lbz(Rindex, 1, R14_bcp);
4076
4077 __ load_dispatch_table(Rtable, Interpreter::_wentry_point);
4078
4079 __ slwi(Rindex, Rindex, LogBytesPerWord);
4080 __ ldx(Rtmp, Rtable, Rindex);
4081 __ mtctr(Rtmp);
4082 __ bctr();
4083 // Note: the bcp increment step is part of the individual wide bytecode implementations.
4084 }
4085 #endif // !CC_INTERP