Wed, 15 Apr 2020 11:49:55 +0800
Merge
1 /*
2 * Copyright (c) 2014, 2017 Oracle and/or its affiliates. All rights reserved.
3 * Copyright 2013, 2017 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 */
26 #include "precompiled.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "interpreter/interpreterRuntime.hpp"
30 #include "interpreter/templateInterpreter.hpp"
31 #include "interpreter/templateTable.hpp"
32 #include "memory/universe.inline.hpp"
33 #include "oops/objArrayKlass.hpp"
34 #include "oops/oop.inline.hpp"
35 #include "prims/methodHandles.hpp"
36 #include "runtime/sharedRuntime.hpp"
37 #include "runtime/stubRoutines.hpp"
38 #include "runtime/synchronizer.hpp"
39 #include "utilities/macros.hpp"
41 #ifndef CC_INTERP
43 #undef __
44 #define __ _masm->
46 // ============================================================================
47 // Misc helpers
49 // Do an oop store like *(base + index) = val OR *(base + offset) = val
50 // (only one of both variants is possible at the same time).
51 // Index can be noreg.
52 // Kills:
53 // Rbase, Rtmp
54 static void do_oop_store(InterpreterMacroAssembler* _masm,
55 Register Rbase,
56 RegisterOrConstant offset,
57 Register Rval, // Noreg means always null.
58 Register Rtmp1,
59 Register Rtmp2,
60 Register Rtmp3,
61 BarrierSet::Name barrier,
62 bool precise,
63 bool check_null) {
64 assert_different_registers(Rtmp1, Rtmp2, Rtmp3, Rval, Rbase);
66 switch (barrier) {
67 #if INCLUDE_ALL_GCS
68 case BarrierSet::G1SATBCT:
69 case BarrierSet::G1SATBCTLogging:
70 {
71 // Load and record the previous value.
72 __ g1_write_barrier_pre(Rbase, offset,
73 Rtmp3, /* holder of pre_val ? */
74 Rtmp1, Rtmp2, false /* frame */);
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 }
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 }
128 if (Rval == noreg || check_null) { // Store null oop.
129 Register Rnull = Rval;
130 __ bind(Lnull);
131 if (Rval == noreg) {
132 Rnull = Rtmp1;
133 __ li(Rnull, 0);
134 }
135 if (UseCompressedOops) {
136 __ stw(Rnull, offset, Rbase);
137 } else {
138 __ std(Rnull, offset, Rbase);
139 }
140 }
141 __ bind(Ldone);
142 }
143 break;
144 case BarrierSet::ModRef:
145 case BarrierSet::Other:
146 ShouldNotReachHere();
147 break;
148 default:
149 ShouldNotReachHere();
150 }
151 }
153 // ============================================================================
154 // Platform-dependent initialization
156 void TemplateTable::pd_initialize() {
157 // No ppc64 specific initialization.
158 }
160 Address TemplateTable::at_bcp(int offset) {
161 // Not used on ppc.
162 ShouldNotReachHere();
163 return Address();
164 }
166 // Patches the current bytecode (ptr to it located in bcp)
167 // in the bytecode stream with a new one.
168 void TemplateTable::patch_bytecode(Bytecodes::Code new_bc, Register Rnew_bc, Register Rtemp, bool load_bc_into_bc_reg /*=true*/, int byte_no) {
169 // With sharing on, may need to test method flag.
170 if (!RewriteBytecodes) return;
171 Label L_patch_done;
173 switch (new_bc) {
174 case Bytecodes::_fast_aputfield:
175 case Bytecodes::_fast_bputfield:
176 case Bytecodes::_fast_zputfield:
177 case Bytecodes::_fast_cputfield:
178 case Bytecodes::_fast_dputfield:
179 case Bytecodes::_fast_fputfield:
180 case Bytecodes::_fast_iputfield:
181 case Bytecodes::_fast_lputfield:
182 case Bytecodes::_fast_sputfield:
183 {
184 // We skip bytecode quickening for putfield instructions when
185 // the put_code written to the constant pool cache is zero.
186 // This is required so that every execution of this instruction
187 // calls out to InterpreterRuntime::resolve_get_put to do
188 // additional, required work.
189 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
190 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
191 __ get_cache_and_index_at_bcp(Rtemp /* dst = cache */, 1);
192 // ((*(cache+indices))>>((1+byte_no)*8))&0xFF:
193 #if defined(VM_LITTLE_ENDIAN)
194 __ lbz(Rnew_bc, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 1 + byte_no, Rtemp);
195 #else
196 __ lbz(Rnew_bc, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 7 - (1 + byte_no), Rtemp);
197 #endif
198 __ cmpwi(CCR0, Rnew_bc, 0);
199 __ li(Rnew_bc, (unsigned int)(unsigned char)new_bc);
200 __ beq(CCR0, L_patch_done);
201 // __ isync(); // acquire not needed
202 break;
203 }
205 default:
206 assert(byte_no == -1, "sanity");
207 if (load_bc_into_bc_reg) {
208 __ li(Rnew_bc, (unsigned int)(unsigned char)new_bc);
209 }
210 }
212 if (JvmtiExport::can_post_breakpoint()) {
213 Label L_fast_patch;
214 __ lbz(Rtemp, 0, R14_bcp);
215 __ cmpwi(CCR0, Rtemp, (unsigned int)(unsigned char)Bytecodes::_breakpoint);
216 __ bne(CCR0, L_fast_patch);
217 // Perform the quickening, slowly, in the bowels of the breakpoint table.
218 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), R19_method, R14_bcp, Rnew_bc);
219 __ b(L_patch_done);
220 __ bind(L_fast_patch);
221 }
223 // Patch bytecode.
224 __ stb(Rnew_bc, 0, R14_bcp);
226 __ bind(L_patch_done);
227 }
229 // ============================================================================
230 // Individual instructions
232 void TemplateTable::nop() {
233 transition(vtos, vtos);
234 // Nothing to do.
235 }
237 void TemplateTable::shouldnotreachhere() {
238 transition(vtos, vtos);
239 __ stop("shouldnotreachhere bytecode");
240 }
242 void TemplateTable::aconst_null() {
243 transition(vtos, atos);
244 __ li(R17_tos, 0);
245 }
247 void TemplateTable::iconst(int value) {
248 transition(vtos, itos);
249 assert(value >= -1 && value <= 5, "");
250 __ li(R17_tos, value);
251 }
253 void TemplateTable::lconst(int value) {
254 transition(vtos, ltos);
255 assert(value >= -1 && value <= 5, "");
256 __ li(R17_tos, value);
257 }
259 void TemplateTable::fconst(int value) {
260 transition(vtos, ftos);
261 static float zero = 0.0;
262 static float one = 1.0;
263 static float two = 2.0;
264 switch (value) {
265 default: ShouldNotReachHere();
266 case 0: {
267 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&zero, R0, true);
268 __ lfs(F15_ftos, simm16_offset, R11_scratch1);
269 break;
270 }
271 case 1: {
272 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&one, R0, true);
273 __ lfs(F15_ftos, simm16_offset, R11_scratch1);
274 break;
275 }
276 case 2: {
277 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&two, R0, true);
278 __ lfs(F15_ftos, simm16_offset, R11_scratch1);
279 break;
280 }
281 }
282 }
284 void TemplateTable::dconst(int value) {
285 transition(vtos, dtos);
286 static double zero = 0.0;
287 static double one = 1.0;
288 switch (value) {
289 case 0: {
290 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&zero, R0, true);
291 __ lfd(F15_ftos, simm16_offset, R11_scratch1);
292 break;
293 }
294 case 1: {
295 int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&one, R0, true);
296 __ lfd(F15_ftos, simm16_offset, R11_scratch1);
297 break;
298 }
299 default: ShouldNotReachHere();
300 }
301 }
303 void TemplateTable::bipush() {
304 transition(vtos, itos);
305 __ lbz(R17_tos, 1, R14_bcp);
306 __ extsb(R17_tos, R17_tos);
307 }
309 void TemplateTable::sipush() {
310 transition(vtos, itos);
311 __ get_2_byte_integer_at_bcp(1, R17_tos, InterpreterMacroAssembler::Signed);
312 }
314 void TemplateTable::ldc(bool wide) {
315 Register Rscratch1 = R11_scratch1,
316 Rscratch2 = R12_scratch2,
317 Rcpool = R3_ARG1;
319 transition(vtos, vtos);
320 Label notInt, notClass, exit;
322 __ get_cpool_and_tags(Rcpool, Rscratch2); // Set Rscratch2 = &tags.
323 if (wide) { // Read index.
324 __ get_2_byte_integer_at_bcp(1, Rscratch1, InterpreterMacroAssembler::Unsigned);
325 } else {
326 __ lbz(Rscratch1, 1, R14_bcp);
327 }
329 const int base_offset = ConstantPool::header_size() * wordSize;
330 const int tags_offset = Array<u1>::base_offset_in_bytes();
332 // Get type from tags.
333 __ addi(Rscratch2, Rscratch2, tags_offset);
334 __ lbzx(Rscratch2, Rscratch2, Rscratch1);
336 __ cmpwi(CCR0, Rscratch2, JVM_CONSTANT_UnresolvedClass); // Unresolved class?
337 __ cmpwi(CCR1, Rscratch2, JVM_CONSTANT_UnresolvedClassInError); // Unresolved class in error state?
338 __ cror(/*CR0 eq*/2, /*CR1 eq*/4+2, /*CR0 eq*/2);
340 // Resolved class - need to call vm to get java mirror of the class.
341 __ cmpwi(CCR1, Rscratch2, JVM_CONSTANT_Class);
342 __ crnor(/*CR0 eq*/2, /*CR1 eq*/4+2, /*CR0 eq*/2); // Neither resolved class nor unresolved case from above?
343 __ beq(CCR0, notClass);
345 __ li(R4, wide ? 1 : 0);
346 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), R4);
347 __ push(atos);
348 __ b(exit);
350 __ align(32, 12);
351 __ bind(notClass);
352 __ addi(Rcpool, Rcpool, base_offset);
353 __ sldi(Rscratch1, Rscratch1, LogBytesPerWord);
354 __ cmpdi(CCR0, Rscratch2, JVM_CONSTANT_Integer);
355 __ bne(CCR0, notInt);
356 __ lwax(R17_tos, Rcpool, Rscratch1);
357 __ push(itos);
358 __ b(exit);
360 __ align(32, 12);
361 __ bind(notInt);
362 #ifdef ASSERT
363 // String and Object are rewritten to fast_aldc
364 __ cmpdi(CCR0, Rscratch2, JVM_CONSTANT_Float);
365 __ asm_assert_eq("unexpected type", 0x8765);
366 #endif
367 __ lfsx(F15_ftos, Rcpool, Rscratch1);
368 __ push(ftos);
370 __ align(32, 12);
371 __ bind(exit);
372 }
374 // Fast path for caching oop constants.
375 void TemplateTable::fast_aldc(bool wide) {
376 transition(vtos, atos);
378 int index_size = wide ? sizeof(u2) : sizeof(u1);
379 const Register Rscratch = R11_scratch1;
380 Label resolved;
382 // We are resolved if the resolved reference cache entry contains a
383 // non-null object (CallSite, etc.)
384 __ get_cache_index_at_bcp(Rscratch, 1, index_size); // Load index.
385 __ load_resolved_reference_at_index(R17_tos, Rscratch);
386 __ cmpdi(CCR0, R17_tos, 0);
387 __ bne(CCR0, resolved);
388 __ load_const_optimized(R3_ARG1, (int)bytecode());
390 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
392 // First time invocation - must resolve first.
393 __ call_VM(R17_tos, entry, R3_ARG1);
395 __ align(32, 12);
396 __ bind(resolved);
397 __ verify_oop(R17_tos);
398 }
400 void TemplateTable::ldc2_w() {
401 transition(vtos, vtos);
402 Label Llong, Lexit;
404 Register Rindex = R11_scratch1,
405 Rcpool = R12_scratch2,
406 Rtag = R3_ARG1;
407 __ get_cpool_and_tags(Rcpool, Rtag);
408 __ get_2_byte_integer_at_bcp(1, Rindex, InterpreterMacroAssembler::Unsigned);
410 const int base_offset = ConstantPool::header_size() * wordSize;
411 const int tags_offset = Array<u1>::base_offset_in_bytes();
412 // Get type from tags.
413 __ addi(Rcpool, Rcpool, base_offset);
414 __ addi(Rtag, Rtag, tags_offset);
416 __ lbzx(Rtag, Rtag, Rindex);
418 __ sldi(Rindex, Rindex, LogBytesPerWord);
419 __ cmpdi(CCR0, Rtag, JVM_CONSTANT_Double);
420 __ bne(CCR0, Llong);
421 // A double can be placed at word-aligned locations in the constant pool.
422 // Check out Conversions.java for an example.
423 // Also ConstantPool::header_size() is 20, which makes it very difficult
424 // to double-align double on the constant pool. SG, 11/7/97
425 __ lfdx(F15_ftos, Rcpool, Rindex);
426 __ push(dtos);
427 __ b(Lexit);
429 __ bind(Llong);
430 __ ldx(R17_tos, Rcpool, Rindex);
431 __ push(ltos);
433 __ bind(Lexit);
434 }
436 // Get the locals index located in the bytecode stream at bcp + offset.
437 void TemplateTable::locals_index(Register Rdst, int offset) {
438 __ lbz(Rdst, offset, R14_bcp);
439 }
441 void TemplateTable::iload() {
442 transition(vtos, itos);
444 // Get the local value into tos
445 const Register Rindex = R22_tmp2;
446 locals_index(Rindex);
448 // Rewrite iload,iload pair into fast_iload2
449 // iload,caload pair into fast_icaload
450 if (RewriteFrequentPairs) {
451 Label Lrewrite, Ldone;
452 Register Rnext_byte = R3_ARG1,
453 Rrewrite_to = R6_ARG4,
454 Rscratch = R11_scratch1;
456 // get next byte
457 __ lbz(Rnext_byte, Bytecodes::length_for(Bytecodes::_iload), R14_bcp);
459 // if _iload, wait to rewrite to iload2. We only want to rewrite the
460 // last two iloads in a pair. Comparing against fast_iload means that
461 // the next bytecode is neither an iload or a caload, and therefore
462 // an iload pair.
463 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_iload);
464 __ beq(CCR0, Ldone);
466 __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_iload);
467 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iload2);
468 __ beq(CCR1, Lrewrite);
470 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_caload);
471 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_icaload);
472 __ beq(CCR0, Lrewrite);
474 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iload);
476 __ bind(Lrewrite);
477 patch_bytecode(Bytecodes::_iload, Rrewrite_to, Rscratch, false);
478 __ bind(Ldone);
479 }
481 __ load_local_int(R17_tos, Rindex, Rindex);
482 }
484 // Load 2 integers in a row without dispatching
485 void TemplateTable::fast_iload2() {
486 transition(vtos, itos);
488 __ lbz(R3_ARG1, 1, R14_bcp);
489 __ lbz(R17_tos, Bytecodes::length_for(Bytecodes::_iload) + 1, R14_bcp);
491 __ load_local_int(R3_ARG1, R11_scratch1, R3_ARG1);
492 __ load_local_int(R17_tos, R12_scratch2, R17_tos);
493 __ push_i(R3_ARG1);
494 }
496 void TemplateTable::fast_iload() {
497 transition(vtos, itos);
498 // Get the local value into tos
500 const Register Rindex = R11_scratch1;
501 locals_index(Rindex);
502 __ load_local_int(R17_tos, Rindex, Rindex);
503 }
505 // Load a local variable type long from locals area to TOS cache register.
506 // Local index resides in bytecodestream.
507 void TemplateTable::lload() {
508 transition(vtos, ltos);
510 const Register Rindex = R11_scratch1;
511 locals_index(Rindex);
512 __ load_local_long(R17_tos, Rindex, Rindex);
513 }
515 void TemplateTable::fload() {
516 transition(vtos, ftos);
518 const Register Rindex = R11_scratch1;
519 locals_index(Rindex);
520 __ load_local_float(F15_ftos, Rindex, Rindex);
521 }
523 void TemplateTable::dload() {
524 transition(vtos, dtos);
526 const Register Rindex = R11_scratch1;
527 locals_index(Rindex);
528 __ load_local_double(F15_ftos, Rindex, Rindex);
529 }
531 void TemplateTable::aload() {
532 transition(vtos, atos);
534 const Register Rindex = R11_scratch1;
535 locals_index(Rindex);
536 __ load_local_ptr(R17_tos, Rindex, Rindex);
537 }
539 void TemplateTable::locals_index_wide(Register Rdst) {
540 // Offset is 2, not 1, because Lbcp points to wide prefix code.
541 __ get_2_byte_integer_at_bcp(2, Rdst, InterpreterMacroAssembler::Unsigned);
542 }
544 void TemplateTable::wide_iload() {
545 // Get the local value into tos.
547 const Register Rindex = R11_scratch1;
548 locals_index_wide(Rindex);
549 __ load_local_int(R17_tos, Rindex, Rindex);
550 }
552 void TemplateTable::wide_lload() {
553 transition(vtos, ltos);
555 const Register Rindex = R11_scratch1;
556 locals_index_wide(Rindex);
557 __ load_local_long(R17_tos, Rindex, Rindex);
558 }
560 void TemplateTable::wide_fload() {
561 transition(vtos, ftos);
563 const Register Rindex = R11_scratch1;
564 locals_index_wide(Rindex);
565 __ load_local_float(F15_ftos, Rindex, Rindex);
566 }
568 void TemplateTable::wide_dload() {
569 transition(vtos, dtos);
571 const Register Rindex = R11_scratch1;
572 locals_index_wide(Rindex);
573 __ load_local_double(F15_ftos, Rindex, Rindex);
574 }
576 void TemplateTable::wide_aload() {
577 transition(vtos, atos);
579 const Register Rindex = R11_scratch1;
580 locals_index_wide(Rindex);
581 __ load_local_ptr(R17_tos, Rindex, Rindex);
582 }
584 void TemplateTable::iaload() {
585 transition(itos, itos);
587 const Register Rload_addr = R3_ARG1,
588 Rarray = R4_ARG2,
589 Rtemp = R5_ARG3;
590 __ index_check(Rarray, R17_tos /* index */, LogBytesPerInt, Rtemp, Rload_addr);
591 __ lwa(R17_tos, arrayOopDesc::base_offset_in_bytes(T_INT), Rload_addr);
592 }
594 void TemplateTable::laload() {
595 transition(itos, ltos);
597 const Register Rload_addr = R3_ARG1,
598 Rarray = R4_ARG2,
599 Rtemp = R5_ARG3;
600 __ index_check(Rarray, R17_tos /* index */, LogBytesPerLong, Rtemp, Rload_addr);
601 __ ld(R17_tos, arrayOopDesc::base_offset_in_bytes(T_LONG), Rload_addr);
602 }
604 void TemplateTable::faload() {
605 transition(itos, ftos);
607 const Register Rload_addr = R3_ARG1,
608 Rarray = R4_ARG2,
609 Rtemp = R5_ARG3;
610 __ index_check(Rarray, R17_tos /* index */, LogBytesPerInt, Rtemp, Rload_addr);
611 __ lfs(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Rload_addr);
612 }
614 void TemplateTable::daload() {
615 transition(itos, dtos);
617 const Register Rload_addr = R3_ARG1,
618 Rarray = R4_ARG2,
619 Rtemp = R5_ARG3;
620 __ index_check(Rarray, R17_tos /* index */, LogBytesPerLong, Rtemp, Rload_addr);
621 __ lfd(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Rload_addr);
622 }
624 void TemplateTable::aaload() {
625 transition(itos, atos);
627 // tos: index
628 // result tos: array
629 const Register Rload_addr = R3_ARG1,
630 Rarray = R4_ARG2,
631 Rtemp = R5_ARG3;
632 __ index_check(Rarray, R17_tos /* index */, UseCompressedOops ? 2 : LogBytesPerWord, Rtemp, Rload_addr);
633 __ load_heap_oop(R17_tos, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Rload_addr);
634 __ verify_oop(R17_tos);
635 //__ dcbt(R17_tos); // prefetch
636 }
638 void TemplateTable::baload() {
639 transition(itos, itos);
641 const Register Rload_addr = R3_ARG1,
642 Rarray = R4_ARG2,
643 Rtemp = R5_ARG3;
644 __ index_check(Rarray, R17_tos /* index */, 0, Rtemp, Rload_addr);
645 __ lbz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_BYTE), Rload_addr);
646 __ extsb(R17_tos, R17_tos);
647 }
649 void TemplateTable::caload() {
650 transition(itos, itos);
652 const Register Rload_addr = R3_ARG1,
653 Rarray = R4_ARG2,
654 Rtemp = R5_ARG3;
655 __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
656 __ lhz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rload_addr);
657 }
659 // Iload followed by caload frequent pair.
660 void TemplateTable::fast_icaload() {
661 transition(vtos, itos);
663 const Register Rload_addr = R3_ARG1,
664 Rarray = R4_ARG2,
665 Rtemp = R11_scratch1;
667 locals_index(R17_tos);
668 __ load_local_int(R17_tos, Rtemp, R17_tos);
669 __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
670 __ lhz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rload_addr);
671 }
673 void TemplateTable::saload() {
674 transition(itos, itos);
676 const Register Rload_addr = R11_scratch1,
677 Rarray = R12_scratch2,
678 Rtemp = R3_ARG1;
679 __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
680 __ lha(R17_tos, arrayOopDesc::base_offset_in_bytes(T_SHORT), Rload_addr);
681 }
683 void TemplateTable::iload(int n) {
684 transition(vtos, itos);
686 __ lwz(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
687 }
689 void TemplateTable::lload(int n) {
690 transition(vtos, ltos);
692 __ ld(R17_tos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
693 }
695 void TemplateTable::fload(int n) {
696 transition(vtos, ftos);
698 __ lfs(F15_ftos, Interpreter::local_offset_in_bytes(n), R18_locals);
699 }
701 void TemplateTable::dload(int n) {
702 transition(vtos, dtos);
704 __ lfd(F15_ftos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
705 }
707 void TemplateTable::aload(int n) {
708 transition(vtos, atos);
710 __ ld(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
711 }
713 void TemplateTable::aload_0() {
714 transition(vtos, atos);
715 // According to bytecode histograms, the pairs:
716 //
717 // _aload_0, _fast_igetfield
718 // _aload_0, _fast_agetfield
719 // _aload_0, _fast_fgetfield
720 //
721 // occur frequently. If RewriteFrequentPairs is set, the (slow)
722 // _aload_0 bytecode checks if the next bytecode is either
723 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
724 // rewrites the current bytecode into a pair bytecode; otherwise it
725 // rewrites the current bytecode into _0 that doesn't do
726 // the pair check anymore.
727 //
728 // Note: If the next bytecode is _getfield, the rewrite must be
729 // delayed, otherwise we may miss an opportunity for a pair.
730 //
731 // Also rewrite frequent pairs
732 // aload_0, aload_1
733 // aload_0, iload_1
734 // These bytecodes with a small amount of code are most profitable
735 // to rewrite.
737 if (RewriteFrequentPairs) {
739 Label Lrewrite, Ldont_rewrite;
740 Register Rnext_byte = R3_ARG1,
741 Rrewrite_to = R6_ARG4,
742 Rscratch = R11_scratch1;
744 // Get next byte.
745 __ lbz(Rnext_byte, Bytecodes::length_for(Bytecodes::_aload_0), R14_bcp);
747 // If _getfield, wait to rewrite. We only want to rewrite the last two bytecodes in a pair.
748 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_getfield);
749 __ beq(CCR0, Ldont_rewrite);
751 __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_igetfield);
752 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iaccess_0);
753 __ beq(CCR1, Lrewrite);
755 __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_agetfield);
756 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_aaccess_0);
757 __ beq(CCR0, Lrewrite);
759 __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_fgetfield);
760 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_faccess_0);
761 __ beq(CCR1, Lrewrite);
763 __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_aload_0);
765 __ bind(Lrewrite);
766 patch_bytecode(Bytecodes::_aload_0, Rrewrite_to, Rscratch, false);
767 __ bind(Ldont_rewrite);
768 }
770 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
771 aload(0);
772 }
774 void TemplateTable::istore() {
775 transition(itos, vtos);
777 const Register Rindex = R11_scratch1;
778 locals_index(Rindex);
779 __ store_local_int(R17_tos, Rindex);
780 }
782 void TemplateTable::lstore() {
783 transition(ltos, vtos);
784 const Register Rindex = R11_scratch1;
785 locals_index(Rindex);
786 __ store_local_long(R17_tos, Rindex);
787 }
789 void TemplateTable::fstore() {
790 transition(ftos, vtos);
792 const Register Rindex = R11_scratch1;
793 locals_index(Rindex);
794 __ store_local_float(F15_ftos, Rindex);
795 }
797 void TemplateTable::dstore() {
798 transition(dtos, vtos);
800 const Register Rindex = R11_scratch1;
801 locals_index(Rindex);
802 __ store_local_double(F15_ftos, Rindex);
803 }
805 void TemplateTable::astore() {
806 transition(vtos, vtos);
808 const Register Rindex = R11_scratch1;
809 __ pop_ptr();
810 __ verify_oop_or_return_address(R17_tos, Rindex);
811 locals_index(Rindex);
812 __ store_local_ptr(R17_tos, Rindex);
813 }
815 void TemplateTable::wide_istore() {
816 transition(vtos, vtos);
818 const Register Rindex = R11_scratch1;
819 __ pop_i();
820 locals_index_wide(Rindex);
821 __ store_local_int(R17_tos, Rindex);
822 }
824 void TemplateTable::wide_lstore() {
825 transition(vtos, vtos);
827 const Register Rindex = R11_scratch1;
828 __ pop_l();
829 locals_index_wide(Rindex);
830 __ store_local_long(R17_tos, Rindex);
831 }
833 void TemplateTable::wide_fstore() {
834 transition(vtos, vtos);
836 const Register Rindex = R11_scratch1;
837 __ pop_f();
838 locals_index_wide(Rindex);
839 __ store_local_float(F15_ftos, Rindex);
840 }
842 void TemplateTable::wide_dstore() {
843 transition(vtos, vtos);
845 const Register Rindex = R11_scratch1;
846 __ pop_d();
847 locals_index_wide(Rindex);
848 __ store_local_double(F15_ftos, Rindex);
849 }
851 void TemplateTable::wide_astore() {
852 transition(vtos, vtos);
854 const Register Rindex = R11_scratch1;
855 __ pop_ptr();
856 __ verify_oop_or_return_address(R17_tos, Rindex);
857 locals_index_wide(Rindex);
858 __ store_local_ptr(R17_tos, Rindex);
859 }
861 void TemplateTable::iastore() {
862 transition(itos, vtos);
864 const Register Rindex = R3_ARG1,
865 Rstore_addr = R4_ARG2,
866 Rarray = R5_ARG3,
867 Rtemp = R6_ARG4;
868 __ pop_i(Rindex);
869 __ index_check(Rarray, Rindex, LogBytesPerInt, Rtemp, Rstore_addr);
870 __ stw(R17_tos, arrayOopDesc::base_offset_in_bytes(T_INT), Rstore_addr);
871 }
873 void TemplateTable::lastore() {
874 transition(ltos, vtos);
876 const Register Rindex = R3_ARG1,
877 Rstore_addr = R4_ARG2,
878 Rarray = R5_ARG3,
879 Rtemp = R6_ARG4;
880 __ pop_i(Rindex);
881 __ index_check(Rarray, Rindex, LogBytesPerLong, Rtemp, Rstore_addr);
882 __ std(R17_tos, arrayOopDesc::base_offset_in_bytes(T_LONG), Rstore_addr);
883 }
885 void TemplateTable::fastore() {
886 transition(ftos, vtos);
888 const Register Rindex = R3_ARG1,
889 Rstore_addr = R4_ARG2,
890 Rarray = R5_ARG3,
891 Rtemp = R6_ARG4;
892 __ pop_i(Rindex);
893 __ index_check(Rarray, Rindex, LogBytesPerInt, Rtemp, Rstore_addr);
894 __ stfs(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Rstore_addr);
895 }
897 void TemplateTable::dastore() {
898 transition(dtos, vtos);
900 const Register Rindex = R3_ARG1,
901 Rstore_addr = R4_ARG2,
902 Rarray = R5_ARG3,
903 Rtemp = R6_ARG4;
904 __ pop_i(Rindex);
905 __ index_check(Rarray, Rindex, LogBytesPerLong, Rtemp, Rstore_addr);
906 __ stfd(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Rstore_addr);
907 }
909 // Pop 3 values from the stack and...
910 void TemplateTable::aastore() {
911 transition(vtos, vtos);
913 Label Lstore_ok, Lis_null, Ldone;
914 const Register Rindex = R3_ARG1,
915 Rarray = R4_ARG2,
916 Rscratch = R11_scratch1,
917 Rscratch2 = R12_scratch2,
918 Rarray_klass = R5_ARG3,
919 Rarray_element_klass = Rarray_klass,
920 Rvalue_klass = R6_ARG4,
921 Rstore_addr = R31; // Use register which survives VM call.
923 __ ld(R17_tos, Interpreter::expr_offset_in_bytes(0), R15_esp); // Get value to store.
924 __ lwz(Rindex, Interpreter::expr_offset_in_bytes(1), R15_esp); // Get index.
925 __ ld(Rarray, Interpreter::expr_offset_in_bytes(2), R15_esp); // Get array.
927 __ verify_oop(R17_tos);
928 __ index_check_without_pop(Rarray, Rindex, UseCompressedOops ? 2 : LogBytesPerWord, Rscratch, Rstore_addr);
929 // Rindex is dead!
930 Register Rscratch3 = Rindex;
932 // Do array store check - check for NULL value first.
933 __ cmpdi(CCR0, R17_tos, 0);
934 __ beq(CCR0, Lis_null);
936 __ load_klass(Rarray_klass, Rarray);
937 __ load_klass(Rvalue_klass, R17_tos);
939 // Do fast instanceof cache test.
940 __ ld(Rarray_element_klass, in_bytes(ObjArrayKlass::element_klass_offset()), Rarray_klass);
942 // Generate a fast subtype check. Branch to store_ok if no failure. Throw if failure.
943 __ gen_subtype_check(Rvalue_klass /*subklass*/, Rarray_element_klass /*superklass*/, Rscratch, Rscratch2, Rscratch3, Lstore_ok);
945 // Fell through: subtype check failed => throw an exception.
946 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArrayStoreException_entry);
947 __ mtctr(R11_scratch1);
948 __ bctr();
950 __ bind(Lis_null);
951 do_oop_store(_masm, Rstore_addr, arrayOopDesc::base_offset_in_bytes(T_OBJECT), noreg /* 0 */,
952 Rscratch, Rscratch2, Rscratch3, _bs->kind(), true /* precise */, false /* check_null */);
953 __ profile_null_seen(Rscratch, Rscratch2);
954 __ b(Ldone);
956 // Store is OK.
957 __ bind(Lstore_ok);
958 do_oop_store(_masm, Rstore_addr, arrayOopDesc::base_offset_in_bytes(T_OBJECT), R17_tos /* value */,
959 Rscratch, Rscratch2, Rscratch3, _bs->kind(), true /* precise */, false /* check_null */);
961 __ bind(Ldone);
962 // Adjust sp (pops array, index and value).
963 __ addi(R15_esp, R15_esp, 3 * Interpreter::stackElementSize);
964 }
966 void TemplateTable::bastore() {
967 transition(itos, vtos);
969 const Register Rindex = R11_scratch1,
970 Rarray = R12_scratch2,
971 Rscratch = R3_ARG1;
972 __ pop_i(Rindex);
973 __ pop_ptr(Rarray);
974 // tos: val
976 // Need to check whether array is boolean or byte
977 // since both types share the bastore bytecode.
978 __ load_klass(Rscratch, Rarray);
979 __ lwz(Rscratch, in_bytes(Klass::layout_helper_offset()), Rscratch);
980 int diffbit = exact_log2(Klass::layout_helper_boolean_diffbit());
981 __ testbitdi(CCR0, R0, Rscratch, diffbit);
982 Label L_skip;
983 __ bfalse(CCR0, L_skip);
984 __ andi(R17_tos, R17_tos, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1
985 __ bind(L_skip);
987 __ index_check_without_pop(Rarray, Rindex, 0, Rscratch, Rarray);
988 __ stb(R17_tos, arrayOopDesc::base_offset_in_bytes(T_BYTE), Rarray);
989 }
991 void TemplateTable::castore() {
992 transition(itos, vtos);
994 const Register Rindex = R11_scratch1,
995 Rarray = R12_scratch2,
996 Rscratch = R3_ARG1;
997 __ pop_i(Rindex);
998 // tos: val
999 // Rarray: array ptr (popped by index_check)
1000 __ index_check(Rarray, Rindex, LogBytesPerShort, Rscratch, Rarray);
1001 __ sth(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rarray);
1002 }
1004 void TemplateTable::sastore() {
1005 castore();
1006 }
1008 void TemplateTable::istore(int n) {
1009 transition(itos, vtos);
1010 __ stw(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
1011 }
1013 void TemplateTable::lstore(int n) {
1014 transition(ltos, vtos);
1015 __ std(R17_tos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
1016 }
1018 void TemplateTable::fstore(int n) {
1019 transition(ftos, vtos);
1020 __ stfs(F15_ftos, Interpreter::local_offset_in_bytes(n), R18_locals);
1021 }
1023 void TemplateTable::dstore(int n) {
1024 transition(dtos, vtos);
1025 __ stfd(F15_ftos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
1026 }
1028 void TemplateTable::astore(int n) {
1029 transition(vtos, vtos);
1031 __ pop_ptr();
1032 __ verify_oop_or_return_address(R17_tos, R11_scratch1);
1033 __ std(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
1034 }
1036 void TemplateTable::pop() {
1037 transition(vtos, vtos);
1039 __ addi(R15_esp, R15_esp, Interpreter::stackElementSize);
1040 }
1042 void TemplateTable::pop2() {
1043 transition(vtos, vtos);
1045 __ addi(R15_esp, R15_esp, Interpreter::stackElementSize * 2);
1046 }
1048 void TemplateTable::dup() {
1049 transition(vtos, vtos);
1051 __ ld(R11_scratch1, Interpreter::stackElementSize, R15_esp);
1052 __ push_ptr(R11_scratch1);
1053 }
1055 void TemplateTable::dup_x1() {
1056 transition(vtos, vtos);
1058 Register Ra = R11_scratch1,
1059 Rb = R12_scratch2;
1060 // stack: ..., a, b
1061 __ ld(Rb, Interpreter::stackElementSize, R15_esp);
1062 __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1063 __ std(Rb, Interpreter::stackElementSize * 2, R15_esp);
1064 __ std(Ra, Interpreter::stackElementSize, R15_esp);
1065 __ push_ptr(Rb);
1066 // stack: ..., b, a, b
1067 }
1069 void TemplateTable::dup_x2() {
1070 transition(vtos, vtos);
1072 Register Ra = R11_scratch1,
1073 Rb = R12_scratch2,
1074 Rc = R3_ARG1;
1076 // stack: ..., a, b, c
1077 __ ld(Rc, Interpreter::stackElementSize, R15_esp); // load c
1078 __ ld(Ra, Interpreter::stackElementSize * 3, R15_esp); // load a
1079 __ std(Rc, Interpreter::stackElementSize * 3, R15_esp); // store c in a
1080 __ ld(Rb, Interpreter::stackElementSize * 2, R15_esp); // load b
1081 // stack: ..., c, b, c
1082 __ std(Ra, Interpreter::stackElementSize * 2, R15_esp); // store a in b
1083 // stack: ..., c, a, c
1084 __ std(Rb, Interpreter::stackElementSize, R15_esp); // store b in c
1085 __ push_ptr(Rc); // push c
1086 // stack: ..., c, a, b, c
1087 }
1089 void TemplateTable::dup2() {
1090 transition(vtos, vtos);
1092 Register Ra = R11_scratch1,
1093 Rb = R12_scratch2;
1094 // stack: ..., a, b
1095 __ ld(Rb, Interpreter::stackElementSize, R15_esp);
1096 __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1097 __ push_2ptrs(Ra, Rb);
1098 // stack: ..., a, b, a, b
1099 }
1101 void TemplateTable::dup2_x1() {
1102 transition(vtos, vtos);
1104 Register Ra = R11_scratch1,
1105 Rb = R12_scratch2,
1106 Rc = R3_ARG1;
1107 // stack: ..., a, b, c
1108 __ ld(Rc, Interpreter::stackElementSize, R15_esp);
1109 __ ld(Rb, Interpreter::stackElementSize * 2, R15_esp);
1110 __ std(Rc, Interpreter::stackElementSize * 2, R15_esp);
1111 __ ld(Ra, Interpreter::stackElementSize * 3, R15_esp);
1112 __ std(Ra, Interpreter::stackElementSize, R15_esp);
1113 __ std(Rb, Interpreter::stackElementSize * 3, R15_esp);
1114 // stack: ..., b, c, a
1115 __ push_2ptrs(Rb, Rc);
1116 // stack: ..., b, c, a, b, c
1117 }
1119 void TemplateTable::dup2_x2() {
1120 transition(vtos, vtos);
1122 Register Ra = R11_scratch1,
1123 Rb = R12_scratch2,
1124 Rc = R3_ARG1,
1125 Rd = R4_ARG2;
1126 // stack: ..., a, b, c, d
1127 __ ld(Rb, Interpreter::stackElementSize * 3, R15_esp);
1128 __ ld(Rd, Interpreter::stackElementSize, R15_esp);
1129 __ std(Rb, Interpreter::stackElementSize, R15_esp); // store b in d
1130 __ std(Rd, Interpreter::stackElementSize * 3, R15_esp); // store d in b
1131 __ ld(Ra, Interpreter::stackElementSize * 4, R15_esp);
1132 __ ld(Rc, Interpreter::stackElementSize * 2, R15_esp);
1133 __ std(Ra, Interpreter::stackElementSize * 2, R15_esp); // store a in c
1134 __ std(Rc, Interpreter::stackElementSize * 4, R15_esp); // store c in a
1135 // stack: ..., c, d, a, b
1136 __ push_2ptrs(Rc, Rd);
1137 // stack: ..., c, d, a, b, c, d
1138 }
1140 void TemplateTable::swap() {
1141 transition(vtos, vtos);
1142 // stack: ..., a, b
1144 Register Ra = R11_scratch1,
1145 Rb = R12_scratch2;
1146 // stack: ..., a, b
1147 __ ld(Rb, Interpreter::stackElementSize, R15_esp);
1148 __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1149 __ std(Rb, Interpreter::stackElementSize * 2, R15_esp);
1150 __ std(Ra, Interpreter::stackElementSize, R15_esp);
1151 // stack: ..., b, a
1152 }
1154 void TemplateTable::iop2(Operation op) {
1155 transition(itos, itos);
1157 Register Rscratch = R11_scratch1;
1159 __ pop_i(Rscratch);
1160 // tos = number of bits to shift
1161 // Rscratch = value to shift
1162 switch (op) {
1163 case add: __ add(R17_tos, Rscratch, R17_tos); break;
1164 case sub: __ sub(R17_tos, Rscratch, R17_tos); break;
1165 case mul: __ mullw(R17_tos, Rscratch, R17_tos); break;
1166 case _and: __ andr(R17_tos, Rscratch, R17_tos); break;
1167 case _or: __ orr(R17_tos, Rscratch, R17_tos); break;
1168 case _xor: __ xorr(R17_tos, Rscratch, R17_tos); break;
1169 case shl: __ rldicl(R17_tos, R17_tos, 0, 64-5); __ slw(R17_tos, Rscratch, R17_tos); break;
1170 case shr: __ rldicl(R17_tos, R17_tos, 0, 64-5); __ sraw(R17_tos, Rscratch, R17_tos); break;
1171 case ushr: __ rldicl(R17_tos, R17_tos, 0, 64-5); __ srw(R17_tos, Rscratch, R17_tos); break;
1172 default: ShouldNotReachHere();
1173 }
1174 }
1176 void TemplateTable::lop2(Operation op) {
1177 transition(ltos, ltos);
1179 Register Rscratch = R11_scratch1;
1180 __ pop_l(Rscratch);
1181 switch (op) {
1182 case add: __ add(R17_tos, Rscratch, R17_tos); break;
1183 case sub: __ sub(R17_tos, Rscratch, R17_tos); break;
1184 case _and: __ andr(R17_tos, Rscratch, R17_tos); break;
1185 case _or: __ orr(R17_tos, Rscratch, R17_tos); break;
1186 case _xor: __ xorr(R17_tos, Rscratch, R17_tos); break;
1187 default: ShouldNotReachHere();
1188 }
1189 }
1191 void TemplateTable::idiv() {
1192 transition(itos, itos);
1194 Label Lnormal, Lexception, Ldone;
1195 Register Rdividend = R11_scratch1; // Used by irem.
1197 __ addi(R0, R17_tos, 1);
1198 __ cmplwi(CCR0, R0, 2);
1199 __ bgt(CCR0, Lnormal); // divisor <-1 or >1
1201 __ cmpwi(CCR1, R17_tos, 0);
1202 __ beq(CCR1, Lexception); // divisor == 0
1204 __ pop_i(Rdividend);
1205 __ mullw(R17_tos, Rdividend, R17_tos); // div by +/-1
1206 __ b(Ldone);
1208 __ bind(Lexception);
1209 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArithmeticException_entry);
1210 __ mtctr(R11_scratch1);
1211 __ bctr();
1213 __ align(32, 12);
1214 __ bind(Lnormal);
1215 __ pop_i(Rdividend);
1216 __ divw(R17_tos, Rdividend, R17_tos); // Can't divide minint/-1.
1217 __ bind(Ldone);
1218 }
1220 void TemplateTable::irem() {
1221 transition(itos, itos);
1223 __ mr(R12_scratch2, R17_tos);
1224 idiv();
1225 __ mullw(R17_tos, R17_tos, R12_scratch2);
1226 __ subf(R17_tos, R17_tos, R11_scratch1); // Dividend set by idiv.
1227 }
1229 void TemplateTable::lmul() {
1230 transition(ltos, ltos);
1232 __ pop_l(R11_scratch1);
1233 __ mulld(R17_tos, R11_scratch1, R17_tos);
1234 }
1236 void TemplateTable::ldiv() {
1237 transition(ltos, ltos);
1239 Label Lnormal, Lexception, Ldone;
1240 Register Rdividend = R11_scratch1; // Used by lrem.
1242 __ addi(R0, R17_tos, 1);
1243 __ cmpldi(CCR0, R0, 2);
1244 __ bgt(CCR0, Lnormal); // divisor <-1 or >1
1246 __ cmpdi(CCR1, R17_tos, 0);
1247 __ beq(CCR1, Lexception); // divisor == 0
1249 __ pop_l(Rdividend);
1250 __ mulld(R17_tos, Rdividend, R17_tos); // div by +/-1
1251 __ b(Ldone);
1253 __ bind(Lexception);
1254 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArithmeticException_entry);
1255 __ mtctr(R11_scratch1);
1256 __ bctr();
1258 __ align(32, 12);
1259 __ bind(Lnormal);
1260 __ pop_l(Rdividend);
1261 __ divd(R17_tos, Rdividend, R17_tos); // Can't divide minint/-1.
1262 __ bind(Ldone);
1263 }
1265 void TemplateTable::lrem() {
1266 transition(ltos, ltos);
1268 __ mr(R12_scratch2, R17_tos);
1269 ldiv();
1270 __ mulld(R17_tos, R17_tos, R12_scratch2);
1271 __ subf(R17_tos, R17_tos, R11_scratch1); // Dividend set by ldiv.
1272 }
1274 void TemplateTable::lshl() {
1275 transition(itos, ltos);
1277 __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1278 __ pop_l(R11_scratch1);
1279 __ sld(R17_tos, R11_scratch1, R17_tos);
1280 }
1282 void TemplateTable::lshr() {
1283 transition(itos, ltos);
1285 __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1286 __ pop_l(R11_scratch1);
1287 __ srad(R17_tos, R11_scratch1, R17_tos);
1288 }
1290 void TemplateTable::lushr() {
1291 transition(itos, ltos);
1293 __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1294 __ pop_l(R11_scratch1);
1295 __ srd(R17_tos, R11_scratch1, R17_tos);
1296 }
1298 void TemplateTable::fop2(Operation op) {
1299 transition(ftos, ftos);
1301 switch (op) {
1302 case add: __ pop_f(F0_SCRATCH); __ fadds(F15_ftos, F0_SCRATCH, F15_ftos); break;
1303 case sub: __ pop_f(F0_SCRATCH); __ fsubs(F15_ftos, F0_SCRATCH, F15_ftos); break;
1304 case mul: __ pop_f(F0_SCRATCH); __ fmuls(F15_ftos, F0_SCRATCH, F15_ftos); break;
1305 case div: __ pop_f(F0_SCRATCH); __ fdivs(F15_ftos, F0_SCRATCH, F15_ftos); break;
1306 case rem:
1307 __ pop_f(F1_ARG1);
1308 __ fmr(F2_ARG2, F15_ftos);
1309 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem));
1310 __ fmr(F15_ftos, F1_RET);
1311 break;
1313 default: ShouldNotReachHere();
1314 }
1315 }
1317 void TemplateTable::dop2(Operation op) {
1318 transition(dtos, dtos);
1320 switch (op) {
1321 case add: __ pop_d(F0_SCRATCH); __ fadd(F15_ftos, F0_SCRATCH, F15_ftos); break;
1322 case sub: __ pop_d(F0_SCRATCH); __ fsub(F15_ftos, F0_SCRATCH, F15_ftos); break;
1323 case mul: __ pop_d(F0_SCRATCH); __ fmul(F15_ftos, F0_SCRATCH, F15_ftos); break;
1324 case div: __ pop_d(F0_SCRATCH); __ fdiv(F15_ftos, F0_SCRATCH, F15_ftos); break;
1325 case rem:
1326 __ pop_d(F1_ARG1);
1327 __ fmr(F2_ARG2, F15_ftos);
1328 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem));
1329 __ fmr(F15_ftos, F1_RET);
1330 break;
1332 default: ShouldNotReachHere();
1333 }
1334 }
1336 // Negate the value in the TOS cache.
1337 void TemplateTable::ineg() {
1338 transition(itos, itos);
1340 __ neg(R17_tos, R17_tos);
1341 }
1343 // Negate the value in the TOS cache.
1344 void TemplateTable::lneg() {
1345 transition(ltos, ltos);
1347 __ neg(R17_tos, R17_tos);
1348 }
1350 void TemplateTable::fneg() {
1351 transition(ftos, ftos);
1353 __ fneg(F15_ftos, F15_ftos);
1354 }
1356 void TemplateTable::dneg() {
1357 transition(dtos, dtos);
1359 __ fneg(F15_ftos, F15_ftos);
1360 }
1362 // Increments a local variable in place.
1363 void TemplateTable::iinc() {
1364 transition(vtos, vtos);
1366 const Register Rindex = R11_scratch1,
1367 Rincrement = R0,
1368 Rvalue = R12_scratch2;
1370 locals_index(Rindex); // Load locals index from bytecode stream.
1371 __ lbz(Rincrement, 2, R14_bcp); // Load increment from the bytecode stream.
1372 __ extsb(Rincrement, Rincrement);
1374 __ load_local_int(Rvalue, Rindex, Rindex); // Puts address of local into Rindex.
1376 __ add(Rvalue, Rincrement, Rvalue);
1377 __ stw(Rvalue, 0, Rindex);
1378 }
1380 void TemplateTable::wide_iinc() {
1381 transition(vtos, vtos);
1383 Register Rindex = R11_scratch1,
1384 Rlocals_addr = Rindex,
1385 Rincr = R12_scratch2;
1386 locals_index_wide(Rindex);
1387 __ get_2_byte_integer_at_bcp(4, Rincr, InterpreterMacroAssembler::Signed);
1388 __ load_local_int(R17_tos, Rlocals_addr, Rindex);
1389 __ add(R17_tos, Rincr, R17_tos);
1390 __ stw(R17_tos, 0, Rlocals_addr);
1391 }
1393 void TemplateTable::convert() {
1394 // %%%%% Factor this first part accross platforms
1395 #ifdef ASSERT
1396 TosState tos_in = ilgl;
1397 TosState tos_out = ilgl;
1398 switch (bytecode()) {
1399 case Bytecodes::_i2l: // fall through
1400 case Bytecodes::_i2f: // fall through
1401 case Bytecodes::_i2d: // fall through
1402 case Bytecodes::_i2b: // fall through
1403 case Bytecodes::_i2c: // fall through
1404 case Bytecodes::_i2s: tos_in = itos; break;
1405 case Bytecodes::_l2i: // fall through
1406 case Bytecodes::_l2f: // fall through
1407 case Bytecodes::_l2d: tos_in = ltos; break;
1408 case Bytecodes::_f2i: // fall through
1409 case Bytecodes::_f2l: // fall through
1410 case Bytecodes::_f2d: tos_in = ftos; break;
1411 case Bytecodes::_d2i: // fall through
1412 case Bytecodes::_d2l: // fall through
1413 case Bytecodes::_d2f: tos_in = dtos; break;
1414 default : ShouldNotReachHere();
1415 }
1416 switch (bytecode()) {
1417 case Bytecodes::_l2i: // fall through
1418 case Bytecodes::_f2i: // fall through
1419 case Bytecodes::_d2i: // fall through
1420 case Bytecodes::_i2b: // fall through
1421 case Bytecodes::_i2c: // fall through
1422 case Bytecodes::_i2s: tos_out = itos; break;
1423 case Bytecodes::_i2l: // fall through
1424 case Bytecodes::_f2l: // fall through
1425 case Bytecodes::_d2l: tos_out = ltos; break;
1426 case Bytecodes::_i2f: // fall through
1427 case Bytecodes::_l2f: // fall through
1428 case Bytecodes::_d2f: tos_out = ftos; break;
1429 case Bytecodes::_i2d: // fall through
1430 case Bytecodes::_l2d: // fall through
1431 case Bytecodes::_f2d: tos_out = dtos; break;
1432 default : ShouldNotReachHere();
1433 }
1434 transition(tos_in, tos_out);
1435 #endif
1437 // Conversion
1438 Label done;
1439 switch (bytecode()) {
1440 case Bytecodes::_i2l:
1441 __ extsw(R17_tos, R17_tos);
1442 break;
1444 case Bytecodes::_l2i:
1445 // Nothing to do, we'll continue to work with the lower bits.
1446 break;
1448 case Bytecodes::_i2b:
1449 __ extsb(R17_tos, R17_tos);
1450 break;
1452 case Bytecodes::_i2c:
1453 __ rldicl(R17_tos, R17_tos, 0, 64-2*8);
1454 break;
1456 case Bytecodes::_i2s:
1457 __ extsh(R17_tos, R17_tos);
1458 break;
1460 case Bytecodes::_i2d:
1461 __ extsw(R17_tos, R17_tos);
1462 case Bytecodes::_l2d:
1463 __ push_l_pop_d();
1464 __ fcfid(F15_ftos, F15_ftos);
1465 break;
1467 case Bytecodes::_i2f:
1468 __ extsw(R17_tos, R17_tos);
1469 __ push_l_pop_d();
1470 if (VM_Version::has_fcfids()) { // fcfids is >= Power7 only
1471 // Comment: alternatively, load with sign extend could be done by lfiwax.
1472 __ fcfids(F15_ftos, F15_ftos);
1473 } else {
1474 __ fcfid(F15_ftos, F15_ftos);
1475 __ frsp(F15_ftos, F15_ftos);
1476 }
1477 break;
1479 case Bytecodes::_l2f:
1480 if (VM_Version::has_fcfids()) { // fcfids is >= Power7 only
1481 __ push_l_pop_d();
1482 __ fcfids(F15_ftos, F15_ftos);
1483 } else {
1484 // Avoid rounding problem when result should be 0x3f800001: need fixup code before fcfid+frsp.
1485 __ mr(R3_ARG1, R17_tos);
1486 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::l2f));
1487 __ fmr(F15_ftos, F1_RET);
1488 }
1489 break;
1491 case Bytecodes::_f2d:
1492 // empty
1493 break;
1495 case Bytecodes::_d2f:
1496 __ frsp(F15_ftos, F15_ftos);
1497 break;
1499 case Bytecodes::_d2i:
1500 case Bytecodes::_f2i:
1501 __ fcmpu(CCR0, F15_ftos, F15_ftos);
1502 __ li(R17_tos, 0); // 0 in case of NAN
1503 __ bso(CCR0, done);
1504 __ fctiwz(F15_ftos, F15_ftos);
1505 __ push_d_pop_l();
1506 break;
1508 case Bytecodes::_d2l:
1509 case Bytecodes::_f2l:
1510 __ fcmpu(CCR0, F15_ftos, F15_ftos);
1511 __ li(R17_tos, 0); // 0 in case of NAN
1512 __ bso(CCR0, done);
1513 __ fctidz(F15_ftos, F15_ftos);
1514 __ push_d_pop_l();
1515 break;
1517 default: ShouldNotReachHere();
1518 }
1519 __ bind(done);
1520 }
1522 // Long compare
1523 void TemplateTable::lcmp() {
1524 transition(ltos, itos);
1526 const Register Rscratch = R11_scratch1;
1527 __ pop_l(Rscratch); // first operand, deeper in stack
1529 __ cmpd(CCR0, Rscratch, R17_tos); // compare
1530 __ mfcr(R17_tos); // set bit 32..33 as follows: <: 0b10, =: 0b00, >: 0b01
1531 __ srwi(Rscratch, R17_tos, 30);
1532 __ srawi(R17_tos, R17_tos, 31);
1533 __ orr(R17_tos, Rscratch, R17_tos); // set result as follows: <: -1, =: 0, >: 1
1534 }
1536 // fcmpl/fcmpg and dcmpl/dcmpg bytecodes
1537 // unordered_result == -1 => fcmpl or dcmpl
1538 // unordered_result == 1 => fcmpg or dcmpg
1539 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1540 const FloatRegister Rfirst = F0_SCRATCH,
1541 Rsecond = F15_ftos;
1542 const Register Rscratch = R11_scratch1;
1544 if (is_float) {
1545 __ pop_f(Rfirst);
1546 } else {
1547 __ pop_d(Rfirst);
1548 }
1550 Label Lunordered, Ldone;
1551 __ fcmpu(CCR0, Rfirst, Rsecond); // compare
1552 if (unordered_result) {
1553 __ bso(CCR0, Lunordered);
1554 }
1555 __ mfcr(R17_tos); // set bit 32..33 as follows: <: 0b10, =: 0b00, >: 0b01
1556 __ srwi(Rscratch, R17_tos, 30);
1557 __ srawi(R17_tos, R17_tos, 31);
1558 __ orr(R17_tos, Rscratch, R17_tos); // set result as follows: <: -1, =: 0, >: 1
1559 if (unordered_result) {
1560 __ b(Ldone);
1561 __ bind(Lunordered);
1562 __ load_const_optimized(R17_tos, unordered_result);
1563 }
1564 __ bind(Ldone);
1565 }
1567 // Branch_conditional which takes TemplateTable::Condition.
1568 void TemplateTable::branch_conditional(ConditionRegister crx, TemplateTable::Condition cc, Label& L, bool invert) {
1569 bool positive = false;
1570 Assembler::Condition cond = Assembler::equal;
1571 switch (cc) {
1572 case TemplateTable::equal: positive = true ; cond = Assembler::equal ; break;
1573 case TemplateTable::not_equal: positive = false; cond = Assembler::equal ; break;
1574 case TemplateTable::less: positive = true ; cond = Assembler::less ; break;
1575 case TemplateTable::less_equal: positive = false; cond = Assembler::greater; break;
1576 case TemplateTable::greater: positive = true ; cond = Assembler::greater; break;
1577 case TemplateTable::greater_equal: positive = false; cond = Assembler::less ; break;
1578 default: ShouldNotReachHere();
1579 }
1580 int bo = (positive != invert) ? Assembler::bcondCRbiIs1 : Assembler::bcondCRbiIs0;
1581 int bi = Assembler::bi0(crx, cond);
1582 __ bc(bo, bi, L);
1583 }
1585 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1587 // Note: on SPARC, we use InterpreterMacroAssembler::if_cmp also.
1588 __ verify_thread();
1590 const Register Rscratch1 = R11_scratch1,
1591 Rscratch2 = R12_scratch2,
1592 Rscratch3 = R3_ARG1,
1593 R4_counters = R4_ARG2,
1594 bumped_count = R31,
1595 Rdisp = R22_tmp2;
1597 __ profile_taken_branch(Rscratch1, bumped_count);
1599 // Get (wide) offset.
1600 if (is_wide) {
1601 __ get_4_byte_integer_at_bcp(1, Rdisp, InterpreterMacroAssembler::Signed);
1602 } else {
1603 __ get_2_byte_integer_at_bcp(1, Rdisp, InterpreterMacroAssembler::Signed);
1604 }
1606 // --------------------------------------------------------------------------
1607 // Handle all the JSR stuff here, then exit.
1608 // It's much shorter and cleaner than intermingling with the
1609 // non-JSR normal-branch stuff occurring below.
1610 if (is_jsr) {
1611 // Compute return address as bci in Otos_i.
1612 __ ld(Rscratch1, in_bytes(Method::const_offset()), R19_method);
1613 __ addi(Rscratch2, R14_bcp, -in_bytes(ConstMethod::codes_offset()) + (is_wide ? 5 : 3));
1614 __ subf(R17_tos, Rscratch1, Rscratch2);
1616 // Bump bcp to target of JSR.
1617 __ add(R14_bcp, Rdisp, R14_bcp);
1618 // Push returnAddress for "ret" on stack.
1619 __ push_ptr(R17_tos);
1620 // And away we go!
1621 __ dispatch_next(vtos);
1622 return;
1623 }
1625 // --------------------------------------------------------------------------
1626 // Normal (non-jsr) branch handling
1628 const bool increment_invocation_counter_for_backward_branches = UseCompiler && UseLoopCounter;
1629 if (increment_invocation_counter_for_backward_branches) {
1630 //__ unimplemented("branch invocation counter");
1632 Label Lforward;
1633 __ add(R14_bcp, Rdisp, R14_bcp); // Add to bc addr.
1635 // Check branch direction.
1636 __ cmpdi(CCR0, Rdisp, 0);
1637 __ bgt(CCR0, Lforward);
1639 __ get_method_counters(R19_method, R4_counters, Lforward);
1641 if (TieredCompilation) {
1642 Label Lno_mdo, Loverflow;
1643 const int increment = InvocationCounter::count_increment;
1644 const int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
1645 if (ProfileInterpreter) {
1646 Register Rmdo = Rscratch1;
1648 // If no method data exists, go to profile_continue.
1649 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
1650 __ cmpdi(CCR0, Rmdo, 0);
1651 __ beq(CCR0, Lno_mdo);
1653 // Increment backedge counter in the MDO.
1654 const int mdo_bc_offs = in_bytes(MethodData::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
1655 __ lwz(Rscratch2, mdo_bc_offs, Rmdo);
1656 __ load_const_optimized(Rscratch3, mask, R0);
1657 __ addi(Rscratch2, Rscratch2, increment);
1658 __ stw(Rscratch2, mdo_bc_offs, Rmdo);
1659 __ and_(Rscratch3, Rscratch2, Rscratch3);
1660 __ bne(CCR0, Lforward);
1661 __ b(Loverflow);
1662 }
1664 // If there's no MDO, increment counter in method.
1665 const int mo_bc_offs = in_bytes(MethodCounters::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
1666 __ bind(Lno_mdo);
1667 __ lwz(Rscratch2, mo_bc_offs, R4_counters);
1668 __ load_const_optimized(Rscratch3, mask, R0);
1669 __ addi(Rscratch2, Rscratch2, increment);
1670 __ stw(Rscratch2, mo_bc_offs, R19_method);
1671 __ and_(Rscratch3, Rscratch2, Rscratch3);
1672 __ bne(CCR0, Lforward);
1674 __ bind(Loverflow);
1676 // Notify point for loop, pass branch bytecode.
1677 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R14_bcp, true);
1679 // Was an OSR adapter generated?
1680 // O0 = osr nmethod
1681 __ cmpdi(CCR0, R3_RET, 0);
1682 __ beq(CCR0, Lforward);
1684 // Has the nmethod been invalidated already?
1685 __ lwz(R0, nmethod::entry_bci_offset(), R3_RET);
1686 __ cmpwi(CCR0, R0, InvalidOSREntryBci);
1687 __ beq(CCR0, Lforward);
1689 // Migrate the interpreter frame off of the stack.
1690 // We can use all registers because we will not return to interpreter from this point.
1692 // Save nmethod.
1693 const Register osr_nmethod = R31;
1694 __ mr(osr_nmethod, R3_RET);
1695 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R11_scratch1);
1696 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), R16_thread);
1697 __ reset_last_Java_frame();
1698 // OSR buffer is in ARG1.
1700 // Remove the interpreter frame.
1701 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1703 // Jump to the osr code.
1704 __ ld(R11_scratch1, nmethod::osr_entry_point_offset(), osr_nmethod);
1705 __ mtlr(R0);
1706 __ mtctr(R11_scratch1);
1707 __ bctr();
1709 } else {
1711 const Register invoke_ctr = Rscratch1;
1712 // Update Backedge branch separately from invocations.
1713 __ increment_backedge_counter(R4_counters, invoke_ctr, Rscratch2, Rscratch3);
1715 if (ProfileInterpreter) {
1716 __ test_invocation_counter_for_mdp(invoke_ctr, Rscratch2, Lforward);
1717 if (UseOnStackReplacement) {
1718 __ test_backedge_count_for_osr(bumped_count, R14_bcp, Rscratch2);
1719 }
1720 } else {
1721 if (UseOnStackReplacement) {
1722 __ test_backedge_count_for_osr(invoke_ctr, R14_bcp, Rscratch2);
1723 }
1724 }
1725 }
1727 __ bind(Lforward);
1729 } else {
1730 // Bump bytecode pointer by displacement (take the branch).
1731 __ add(R14_bcp, Rdisp, R14_bcp); // Add to bc addr.
1732 }
1733 // Continue with bytecode @ target.
1734 // %%%%% Like Intel, could speed things up by moving bytecode fetch to code above,
1735 // %%%%% and changing dispatch_next to dispatch_only.
1736 __ dispatch_next(vtos);
1737 }
1739 // Helper function for if_cmp* methods below.
1740 // Factored out common compare and branch code.
1741 void TemplateTable::if_cmp_common(Register Rfirst, Register Rsecond, Register Rscratch1, Register Rscratch2, Condition cc, bool is_jint, bool cmp0) {
1742 Label Lnot_taken;
1743 // Note: The condition code we get is the condition under which we
1744 // *fall through*! So we have to inverse the CC here.
1746 if (is_jint) {
1747 if (cmp0) {
1748 __ cmpwi(CCR0, Rfirst, 0);
1749 } else {
1750 __ cmpw(CCR0, Rfirst, Rsecond);
1751 }
1752 } else {
1753 if (cmp0) {
1754 __ cmpdi(CCR0, Rfirst, 0);
1755 } else {
1756 __ cmpd(CCR0, Rfirst, Rsecond);
1757 }
1758 }
1759 branch_conditional(CCR0, cc, Lnot_taken, /*invert*/ true);
1761 // Conition is false => Jump!
1762 branch(false, false);
1764 // Condition is not true => Continue.
1765 __ align(32, 12);
1766 __ bind(Lnot_taken);
1767 __ profile_not_taken_branch(Rscratch1, Rscratch2);
1768 }
1770 // Compare integer values with zero and fall through if CC holds, branch away otherwise.
1771 void TemplateTable::if_0cmp(Condition cc) {
1772 transition(itos, vtos);
1774 if_cmp_common(R17_tos, noreg, R11_scratch1, R12_scratch2, cc, true, true);
1775 }
1777 // Compare integer values and fall through if CC holds, branch away otherwise.
1778 //
1779 // Interface:
1780 // - Rfirst: First operand (older stack value)
1781 // - tos: Second operand (younger stack value)
1782 void TemplateTable::if_icmp(Condition cc) {
1783 transition(itos, vtos);
1785 const Register Rfirst = R0,
1786 Rsecond = R17_tos;
1788 __ pop_i(Rfirst);
1789 if_cmp_common(Rfirst, Rsecond, R11_scratch1, R12_scratch2, cc, true, false);
1790 }
1792 void TemplateTable::if_nullcmp(Condition cc) {
1793 transition(atos, vtos);
1795 if_cmp_common(R17_tos, noreg, R11_scratch1, R12_scratch2, cc, false, true);
1796 }
1798 void TemplateTable::if_acmp(Condition cc) {
1799 transition(atos, vtos);
1801 const Register Rfirst = R0,
1802 Rsecond = R17_tos;
1804 __ pop_ptr(Rfirst);
1805 if_cmp_common(Rfirst, Rsecond, R11_scratch1, R12_scratch2, cc, false, false);
1806 }
1808 void TemplateTable::ret() {
1809 locals_index(R11_scratch1);
1810 __ load_local_ptr(R17_tos, R11_scratch1, R11_scratch1);
1812 __ profile_ret(vtos, R17_tos, R11_scratch1, R12_scratch2);
1814 __ ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method);
1815 __ add(R11_scratch1, R17_tos, R11_scratch1);
1816 __ addi(R14_bcp, R11_scratch1, in_bytes(ConstMethod::codes_offset()));
1817 __ dispatch_next(vtos);
1818 }
1820 void TemplateTable::wide_ret() {
1821 transition(vtos, vtos);
1823 const Register Rindex = R3_ARG1,
1824 Rscratch1 = R11_scratch1,
1825 Rscratch2 = R12_scratch2;
1827 locals_index_wide(Rindex);
1828 __ load_local_ptr(R17_tos, R17_tos, Rindex);
1829 __ profile_ret(vtos, R17_tos, Rscratch1, R12_scratch2);
1830 // Tos now contains the bci, compute the bcp from that.
1831 __ ld(Rscratch1, in_bytes(Method::const_offset()), R19_method);
1832 __ addi(Rscratch2, R17_tos, in_bytes(ConstMethod::codes_offset()));
1833 __ add(R14_bcp, Rscratch1, Rscratch2);
1834 __ dispatch_next(vtos);
1835 }
1837 void TemplateTable::tableswitch() {
1838 transition(itos, vtos);
1840 Label Ldispatch, Ldefault_case;
1841 Register Rlow_byte = R3_ARG1,
1842 Rindex = Rlow_byte,
1843 Rhigh_byte = R4_ARG2,
1844 Rdef_offset_addr = R5_ARG3, // is going to contain address of default offset
1845 Rscratch1 = R11_scratch1,
1846 Rscratch2 = R12_scratch2,
1847 Roffset = R6_ARG4;
1849 // Align bcp.
1850 __ addi(Rdef_offset_addr, R14_bcp, BytesPerInt);
1851 __ clrrdi(Rdef_offset_addr, Rdef_offset_addr, log2_long((jlong)BytesPerInt));
1853 // Load lo & hi.
1854 __ get_u4(Rlow_byte, Rdef_offset_addr, BytesPerInt, InterpreterMacroAssembler::Unsigned);
1855 __ get_u4(Rhigh_byte, Rdef_offset_addr, 2 *BytesPerInt, InterpreterMacroAssembler::Unsigned);
1857 // Check for default case (=index outside [low,high]).
1858 __ cmpw(CCR0, R17_tos, Rlow_byte);
1859 __ cmpw(CCR1, R17_tos, Rhigh_byte);
1860 __ blt(CCR0, Ldefault_case);
1861 __ bgt(CCR1, Ldefault_case);
1863 // Lookup dispatch offset.
1864 __ sub(Rindex, R17_tos, Rlow_byte);
1865 __ extsw(Rindex, Rindex);
1866 __ profile_switch_case(Rindex, Rhigh_byte /* scratch */, Rscratch1, Rscratch2);
1867 __ sldi(Rindex, Rindex, LogBytesPerInt);
1868 __ addi(Rindex, Rindex, 3 * BytesPerInt);
1869 #if defined(VM_LITTLE_ENDIAN)
1870 __ lwbrx(Roffset, Rdef_offset_addr, Rindex);
1871 __ extsw(Roffset, Roffset);
1872 #else
1873 __ lwax(Roffset, Rdef_offset_addr, Rindex);
1874 #endif
1875 __ b(Ldispatch);
1877 __ bind(Ldefault_case);
1878 __ profile_switch_default(Rhigh_byte, Rscratch1);
1879 __ get_u4(Roffset, Rdef_offset_addr, 0, InterpreterMacroAssembler::Signed);
1881 __ bind(Ldispatch);
1883 __ add(R14_bcp, Roffset, R14_bcp);
1884 __ dispatch_next(vtos);
1885 }
1887 void TemplateTable::lookupswitch() {
1888 transition(itos, itos);
1889 __ stop("lookupswitch bytecode should have been rewritten");
1890 }
1892 // Table switch using linear search through cases.
1893 // Bytecode stream format:
1894 // Bytecode (1) | 4-byte padding | default offset (4) | count (4) | value/offset pair1 (8) | value/offset pair2 (8) | ...
1895 // Note: Everything is big-endian format here.
1896 void TemplateTable::fast_linearswitch() {
1897 transition(itos, vtos);
1899 Label Lloop_entry, Lsearch_loop, Lcontinue_execution, Ldefault_case;
1900 Register Rcount = R3_ARG1,
1901 Rcurrent_pair = R4_ARG2,
1902 Rdef_offset_addr = R5_ARG3, // Is going to contain address of default offset.
1903 Roffset = R31, // Might need to survive C call.
1904 Rvalue = R12_scratch2,
1905 Rscratch = R11_scratch1,
1906 Rcmp_value = R17_tos;
1908 // Align bcp.
1909 __ addi(Rdef_offset_addr, R14_bcp, BytesPerInt);
1910 __ clrrdi(Rdef_offset_addr, Rdef_offset_addr, log2_long((jlong)BytesPerInt));
1912 // Setup loop counter and limit.
1913 __ get_u4(Rcount, Rdef_offset_addr, BytesPerInt, InterpreterMacroAssembler::Unsigned);
1914 __ addi(Rcurrent_pair, Rdef_offset_addr, 2 * BytesPerInt); // Rcurrent_pair now points to first pair.
1916 __ mtctr(Rcount);
1917 __ cmpwi(CCR0, Rcount, 0);
1918 __ bne(CCR0, Lloop_entry);
1920 // Default case
1921 __ bind(Ldefault_case);
1922 __ get_u4(Roffset, Rdef_offset_addr, 0, InterpreterMacroAssembler::Signed);
1923 if (ProfileInterpreter) {
1924 __ profile_switch_default(Rdef_offset_addr, Rcount/* scratch */);
1925 }
1926 __ b(Lcontinue_execution);
1928 // Next iteration
1929 __ bind(Lsearch_loop);
1930 __ bdz(Ldefault_case);
1931 __ addi(Rcurrent_pair, Rcurrent_pair, 2 * BytesPerInt);
1932 __ bind(Lloop_entry);
1933 __ get_u4(Rvalue, Rcurrent_pair, 0, InterpreterMacroAssembler::Unsigned);
1934 __ cmpw(CCR0, Rvalue, Rcmp_value);
1935 __ bne(CCR0, Lsearch_loop);
1937 // Found, load offset.
1938 __ get_u4(Roffset, Rcurrent_pair, BytesPerInt, InterpreterMacroAssembler::Signed);
1939 // Calculate case index and profile
1940 __ mfctr(Rcurrent_pair);
1941 if (ProfileInterpreter) {
1942 __ sub(Rcurrent_pair, Rcount, Rcurrent_pair);
1943 __ profile_switch_case(Rcurrent_pair, Rcount /*scratch*/, Rdef_offset_addr/*scratch*/, Rscratch);
1944 }
1946 __ bind(Lcontinue_execution);
1947 __ add(R14_bcp, Roffset, R14_bcp);
1948 __ dispatch_next(vtos);
1949 }
1951 // Table switch using binary search (value/offset pairs are ordered).
1952 // Bytecode stream format:
1953 // Bytecode (1) | 4-byte padding | default offset (4) | count (4) | value/offset pair1 (8) | value/offset pair2 (8) | ...
1954 // Note: Everything is big-endian format here. So on little endian machines, we have to revers offset and count and cmp value.
1955 void TemplateTable::fast_binaryswitch() {
1957 transition(itos, vtos);
1958 // Implementation using the following core algorithm: (copied from Intel)
1959 //
1960 // int binary_search(int key, LookupswitchPair* array, int n) {
1961 // // Binary search according to "Methodik des Programmierens" by
1962 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1963 // int i = 0;
1964 // int j = n;
1965 // while (i+1 < j) {
1966 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1967 // // with Q: for all i: 0 <= i < n: key < a[i]
1968 // // where a stands for the array and assuming that the (inexisting)
1969 // // element a[n] is infinitely big.
1970 // int h = (i + j) >> 1;
1971 // // i < h < j
1972 // if (key < array[h].fast_match()) {
1973 // j = h;
1974 // } else {
1975 // i = h;
1976 // }
1977 // }
1978 // // R: a[i] <= key < a[i+1] or Q
1979 // // (i.e., if key is within array, i is the correct index)
1980 // return i;
1981 // }
1983 // register allocation
1984 const Register Rkey = R17_tos; // already set (tosca)
1985 const Register Rarray = R3_ARG1;
1986 const Register Ri = R4_ARG2;
1987 const Register Rj = R5_ARG3;
1988 const Register Rh = R6_ARG4;
1989 const Register Rscratch = R11_scratch1;
1991 const int log_entry_size = 3;
1992 const int entry_size = 1 << log_entry_size;
1994 Label found;
1996 // Find Array start,
1997 __ addi(Rarray, R14_bcp, 3 * BytesPerInt);
1998 __ clrrdi(Rarray, Rarray, log2_long((jlong)BytesPerInt));
2000 // initialize i & j
2001 __ li(Ri,0);
2002 __ get_u4(Rj, Rarray, -BytesPerInt, InterpreterMacroAssembler::Unsigned);
2004 // and start.
2005 Label entry;
2006 __ b(entry);
2008 // binary search loop
2009 { Label loop;
2010 __ bind(loop);
2011 // int h = (i + j) >> 1;
2012 __ srdi(Rh, Rh, 1);
2013 // if (key < array[h].fast_match()) {
2014 // j = h;
2015 // } else {
2016 // i = h;
2017 // }
2018 __ sldi(Rscratch, Rh, log_entry_size);
2019 #if defined(VM_LITTLE_ENDIAN)
2020 __ lwbrx(Rscratch, Rscratch, Rarray);
2021 #else
2022 __ lwzx(Rscratch, Rscratch, Rarray);
2023 #endif
2025 // if (key < current value)
2026 // Rh = Rj
2027 // else
2028 // Rh = Ri
2029 Label Lgreater;
2030 __ cmpw(CCR0, Rkey, Rscratch);
2031 __ bge(CCR0, Lgreater);
2032 __ mr(Rj, Rh);
2033 __ b(entry);
2034 __ bind(Lgreater);
2035 __ mr(Ri, Rh);
2037 // while (i+1 < j)
2038 __ bind(entry);
2039 __ addi(Rscratch, Ri, 1);
2040 __ cmpw(CCR0, Rscratch, Rj);
2041 __ add(Rh, Ri, Rj); // start h = i + j >> 1;
2043 __ blt(CCR0, loop);
2044 }
2046 // End of binary search, result index is i (must check again!).
2047 Label default_case;
2048 Label continue_execution;
2049 if (ProfileInterpreter) {
2050 __ mr(Rh, Ri); // Save index in i for profiling.
2051 }
2052 // Ri = value offset
2053 __ sldi(Ri, Ri, log_entry_size);
2054 __ add(Ri, Ri, Rarray);
2055 __ get_u4(Rscratch, Ri, 0, InterpreterMacroAssembler::Unsigned);
2057 Label not_found;
2058 // Ri = offset offset
2059 __ cmpw(CCR0, Rkey, Rscratch);
2060 __ beq(CCR0, not_found);
2061 // entry not found -> j = default offset
2062 __ get_u4(Rj, Rarray, -2 * BytesPerInt, InterpreterMacroAssembler::Unsigned);
2063 __ b(default_case);
2065 __ bind(not_found);
2066 // entry found -> j = offset
2067 __ profile_switch_case(Rh, Rj, Rscratch, Rkey);
2068 __ get_u4(Rj, Ri, BytesPerInt, InterpreterMacroAssembler::Unsigned);
2070 if (ProfileInterpreter) {
2071 __ b(continue_execution);
2072 }
2074 __ bind(default_case); // fall through (if not profiling)
2075 __ profile_switch_default(Ri, Rscratch);
2077 __ bind(continue_execution);
2079 __ extsw(Rj, Rj);
2080 __ add(R14_bcp, Rj, R14_bcp);
2081 __ dispatch_next(vtos);
2082 }
2084 void TemplateTable::_return(TosState state) {
2085 transition(state, state);
2086 assert(_desc->calls_vm(),
2087 "inconsistent calls_vm information"); // call in remove_activation
2089 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2091 Register Rscratch = R11_scratch1,
2092 Rklass = R12_scratch2,
2093 Rklass_flags = Rklass;
2094 Label Lskip_register_finalizer;
2096 // Check if the method has the FINALIZER flag set and call into the VM to finalize in this case.
2097 assert(state == vtos, "only valid state");
2098 __ ld(R17_tos, 0, R18_locals);
2100 // Load klass of this obj.
2101 __ load_klass(Rklass, R17_tos);
2102 __ lwz(Rklass_flags, in_bytes(Klass::access_flags_offset()), Rklass);
2103 __ testbitdi(CCR0, R0, Rklass_flags, exact_log2(JVM_ACC_HAS_FINALIZER));
2104 __ bfalse(CCR0, Lskip_register_finalizer);
2106 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), R17_tos /* obj */);
2108 __ align(32, 12);
2109 __ bind(Lskip_register_finalizer);
2110 }
2112 // Move the result value into the correct register and remove memory stack frame.
2113 __ remove_activation(state, /* throw_monitor_exception */ true);
2114 // Restoration of lr done by remove_activation.
2115 switch (state) {
2116 // Narrow result if state is itos but result type is smaller.
2117 // Need to narrow in the return bytecode rather than in generate_return_entry
2118 // since compiled code callers expect the result to already be narrowed.
2119 case itos: __ narrow(R17_tos); /* fall through */
2120 case ltos:
2121 case btos:
2122 case ztos:
2123 case ctos:
2124 case stos:
2125 case atos: __ mr(R3_RET, R17_tos); break;
2126 case ftos:
2127 case dtos: __ fmr(F1_RET, F15_ftos); break;
2128 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
2129 // to get visible before the reference to the object gets stored anywhere.
2130 __ membar(Assembler::StoreStore); break;
2131 default : ShouldNotReachHere();
2132 }
2133 __ blr();
2134 }
2136 // ============================================================================
2137 // Constant pool cache access
2138 //
2139 // Memory ordering:
2140 //
2141 // Like done in C++ interpreter, we load the fields
2142 // - _indices
2143 // - _f12_oop
2144 // acquired, because these are asked if the cache is already resolved. We don't
2145 // want to float loads above this check.
2146 // See also comments in ConstantPoolCacheEntry::bytecode_1(),
2147 // ConstantPoolCacheEntry::bytecode_2() and ConstantPoolCacheEntry::f1();
2149 // Call into the VM if call site is not yet resolved
2150 //
2151 // Input regs:
2152 // - None, all passed regs are outputs.
2153 //
2154 // Returns:
2155 // - Rcache: The const pool cache entry that contains the resolved result.
2156 // - Rresult: Either noreg or output for f1/f2.
2157 //
2158 // Kills:
2159 // - Rscratch
2160 void TemplateTable::resolve_cache_and_index(int byte_no, Register Rcache, Register Rscratch, size_t index_size) {
2162 __ get_cache_and_index_at_bcp(Rcache, 1, index_size);
2163 Label Lresolved, Ldone;
2165 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2166 // We are resolved if the indices offset contains the current bytecode.
2167 #if defined(VM_LITTLE_ENDIAN)
2168 __ lbz(Rscratch, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + byte_no + 1, Rcache);
2169 #else
2170 __ lbz(Rscratch, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 7 - (byte_no + 1), Rcache);
2171 #endif
2172 // Acquire by cmp-br-isync (see below).
2173 __ cmpdi(CCR0, Rscratch, (int)bytecode());
2174 __ beq(CCR0, Lresolved);
2176 address entry = NULL;
2177 switch (bytecode()) {
2178 case Bytecodes::_getstatic : // fall through
2179 case Bytecodes::_putstatic : // fall through
2180 case Bytecodes::_getfield : // fall through
2181 case Bytecodes::_putfield : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break;
2182 case Bytecodes::_invokevirtual : // fall through
2183 case Bytecodes::_invokespecial : // fall through
2184 case Bytecodes::_invokestatic : // fall through
2185 case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke); break;
2186 case Bytecodes::_invokehandle : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokehandle); break;
2187 case Bytecodes::_invokedynamic : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic); break;
2188 default : ShouldNotReachHere(); break;
2189 }
2190 __ li(R4_ARG2, (int)bytecode());
2191 __ call_VM(noreg, entry, R4_ARG2, true);
2193 // Update registers with resolved info.
2194 __ get_cache_and_index_at_bcp(Rcache, 1, index_size);
2195 __ b(Ldone);
2197 __ bind(Lresolved);
2198 __ isync(); // Order load wrt. succeeding loads.
2199 __ bind(Ldone);
2200 }
2202 // Load the constant pool cache entry at field accesses into registers.
2203 // The Rcache and Rindex registers must be set before call.
2204 // Input:
2205 // - Rcache, Rindex
2206 // Output:
2207 // - Robj, Roffset, Rflags
2208 void TemplateTable::load_field_cp_cache_entry(Register Robj,
2209 Register Rcache,
2210 Register Rindex /* unused on PPC64 */,
2211 Register Roffset,
2212 Register Rflags,
2213 bool is_static = false) {
2214 assert_different_registers(Rcache, Rflags, Roffset);
2215 // assert(Rindex == noreg, "parameter not used on PPC64");
2217 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2218 __ ld(Rflags, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::flags_offset()), Rcache);
2219 __ ld(Roffset, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::f2_offset()), Rcache);
2220 if (is_static) {
2221 __ ld(Robj, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::f1_offset()), Rcache);
2222 __ ld(Robj, in_bytes(Klass::java_mirror_offset()), Robj);
2223 // Acquire not needed here. Following access has an address dependency on this value.
2224 }
2225 }
2227 // Load the constant pool cache entry at invokes into registers.
2228 // Resolve if necessary.
2230 // Input Registers:
2231 // - None, bcp is used, though
2232 //
2233 // Return registers:
2234 // - Rmethod (f1 field or f2 if invokevirtual)
2235 // - Ritable_index (f2 field)
2236 // - Rflags (flags field)
2237 //
2238 // Kills:
2239 // - R21
2240 //
2241 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2242 Register Rmethod,
2243 Register Ritable_index,
2244 Register Rflags,
2245 bool is_invokevirtual,
2246 bool is_invokevfinal,
2247 bool is_invokedynamic) {
2249 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2250 // Determine constant pool cache field offsets.
2251 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2252 const int method_offset = in_bytes(cp_base_offset + (is_invokevirtual ? ConstantPoolCacheEntry::f2_offset() : ConstantPoolCacheEntry::f1_offset()));
2253 const int flags_offset = in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset());
2254 // Access constant pool cache fields.
2255 const int index_offset = in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset());
2257 Register Rcache = R21_tmp1; // Note: same register as R21_sender_SP.
2259 if (is_invokevfinal) {
2260 assert(Ritable_index == noreg, "register not used");
2261 // Already resolved.
2262 __ get_cache_and_index_at_bcp(Rcache, 1);
2263 } else {
2264 resolve_cache_and_index(byte_no, Rcache, R0, is_invokedynamic ? sizeof(u4) : sizeof(u2));
2265 }
2267 __ ld(Rmethod, method_offset, Rcache);
2268 __ ld(Rflags, flags_offset, Rcache);
2270 if (Ritable_index != noreg) {
2271 __ ld(Ritable_index, index_offset, Rcache);
2272 }
2273 }
2275 // ============================================================================
2276 // Field access
2278 // Volatile variables demand their effects be made known to all CPU's
2279 // in order. Store buffers on most chips allow reads & writes to
2280 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2281 // without some kind of memory barrier (i.e., it's not sufficient that
2282 // the interpreter does not reorder volatile references, the hardware
2283 // also must not reorder them).
2284 //
2285 // According to the new Java Memory Model (JMM):
2286 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2287 // writes act as aquire & release, so:
2288 // (2) A read cannot let unrelated NON-volatile memory refs that
2289 // happen after the read float up to before the read. It's OK for
2290 // non-volatile memory refs that happen before the volatile read to
2291 // float down below it.
2292 // (3) Similar a volatile write cannot let unrelated NON-volatile
2293 // memory refs that happen BEFORE the write float down to after the
2294 // write. It's OK for non-volatile memory refs that happen after the
2295 // volatile write to float up before it.
2296 //
2297 // We only put in barriers around volatile refs (they are expensive),
2298 // not _between_ memory refs (that would require us to track the
2299 // flavor of the previous memory refs). Requirements (2) and (3)
2300 // require some barriers before volatile stores and after volatile
2301 // loads. These nearly cover requirement (1) but miss the
2302 // volatile-store-volatile-load case. This final case is placed after
2303 // volatile-stores although it could just as well go before
2304 // volatile-loads.
2306 // The registers cache and index expected to be set before call.
2307 // Correct values of the cache and index registers are preserved.
2308 // Kills:
2309 // Rcache (if has_tos)
2310 // Rscratch
2311 void TemplateTable::jvmti_post_field_access(Register Rcache, Register Rscratch, bool is_static, bool has_tos) {
2313 assert_different_registers(Rcache, Rscratch);
2315 if (JvmtiExport::can_post_field_access()) {
2316 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2317 Label Lno_field_access_post;
2319 // Check if post field access in enabled.
2320 int offs = __ load_const_optimized(Rscratch, JvmtiExport::get_field_access_count_addr(), R0, true);
2321 __ lwz(Rscratch, offs, Rscratch);
2323 __ cmpwi(CCR0, Rscratch, 0);
2324 __ beq(CCR0, Lno_field_access_post);
2326 // Post access enabled - do it!
2327 __ addi(Rcache, Rcache, in_bytes(cp_base_offset));
2328 if (is_static) {
2329 __ li(R17_tos, 0);
2330 } else {
2331 if (has_tos) {
2332 // The fast bytecode versions have obj ptr in register.
2333 // Thus, save object pointer before call_VM() clobbers it
2334 // put object on tos where GC wants it.
2335 __ push_ptr(R17_tos);
2336 } else {
2337 // Load top of stack (do not pop the value off the stack).
2338 __ ld(R17_tos, Interpreter::expr_offset_in_bytes(0), R15_esp);
2339 }
2340 __ verify_oop(R17_tos);
2341 }
2342 // tos: object pointer or NULL if static
2343 // cache: cache entry pointer
2344 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), R17_tos, Rcache);
2345 if (!is_static && has_tos) {
2346 // Restore object pointer.
2347 __ pop_ptr(R17_tos);
2348 __ verify_oop(R17_tos);
2349 } else {
2350 // Cache is still needed to get class or obj.
2351 __ get_cache_and_index_at_bcp(Rcache, 1);
2352 }
2354 __ align(32, 12);
2355 __ bind(Lno_field_access_post);
2356 }
2357 }
2359 // kills R11_scratch1
2360 void TemplateTable::pop_and_check_object(Register Roop) {
2361 Register Rtmp = R11_scratch1;
2363 assert_different_registers(Rtmp, Roop);
2364 __ pop_ptr(Roop);
2365 // For field access must check obj.
2366 __ null_check_throw(Roop, -1, Rtmp);
2367 __ verify_oop(Roop);
2368 }
2370 // PPC64: implement volatile loads as fence-store-acquire.
2371 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2372 transition(vtos, vtos);
2374 Label Lacquire, Lisync;
2376 const Register Rcache = R3_ARG1,
2377 Rclass_or_obj = R22_tmp2,
2378 Roffset = R23_tmp3,
2379 Rflags = R31,
2380 Rbtable = R5_ARG3,
2381 Rbc = R6_ARG4,
2382 Rscratch = R12_scratch2;
2384 static address field_branch_table[number_of_states],
2385 static_branch_table[number_of_states];
2387 address* branch_table = is_static ? static_branch_table : field_branch_table;
2389 // Get field offset.
2390 resolve_cache_and_index(byte_no, Rcache, Rscratch, sizeof(u2));
2392 // JVMTI support
2393 jvmti_post_field_access(Rcache, Rscratch, is_static, false);
2395 // Load after possible GC.
2396 load_field_cp_cache_entry(Rclass_or_obj, Rcache, noreg, Roffset, Rflags, is_static);
2398 // Load pointer to branch table.
2399 __ load_const_optimized(Rbtable, (address)branch_table, Rscratch);
2401 // Get volatile flag.
2402 __ rldicl(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2403 // Note: sync is needed before volatile load on PPC64.
2405 // Check field type.
2406 __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2408 #ifdef ASSERT
2409 Label LFlagInvalid;
2410 __ cmpldi(CCR0, Rflags, number_of_states);
2411 __ bge(CCR0, LFlagInvalid);
2412 #endif
2414 // Load from branch table and dispatch (volatile case: one instruction ahead).
2415 __ sldi(Rflags, Rflags, LogBytesPerWord);
2416 __ cmpwi(CCR6, Rscratch, 1); // Volatile?
2417 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2418 __ sldi(Rscratch, Rscratch, exact_log2(BytesPerInstWord)); // Volatile ? size of 1 instruction : 0.
2419 }
2420 __ ldx(Rbtable, Rbtable, Rflags);
2422 // Get the obj from stack.
2423 if (!is_static) {
2424 pop_and_check_object(Rclass_or_obj); // Kills R11_scratch1.
2425 } else {
2426 __ verify_oop(Rclass_or_obj);
2427 }
2429 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2430 __ subf(Rbtable, Rscratch, Rbtable); // Point to volatile/non-volatile entry point.
2431 }
2432 __ mtctr(Rbtable);
2433 __ bctr();
2435 #ifdef ASSERT
2436 __ bind(LFlagInvalid);
2437 __ stop("got invalid flag", 0x654);
2439 // __ bind(Lvtos);
2440 address pc_before_fence = __ pc();
2441 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2442 assert(__ pc() - pc_before_fence == (ptrdiff_t)BytesPerInstWord, "must be single instruction");
2443 assert(branch_table[vtos] == 0, "can't compute twice");
2444 branch_table[vtos] = __ pc(); // non-volatile_entry point
2445 __ stop("vtos unexpected", 0x655);
2446 #endif
2448 __ align(32, 28, 28); // Align load.
2449 // __ bind(Ldtos);
2450 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2451 assert(branch_table[dtos] == 0, "can't compute twice");
2452 branch_table[dtos] = __ pc(); // non-volatile_entry point
2453 __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
2454 __ push(dtos);
2455 if (!is_static) patch_bytecode(Bytecodes::_fast_dgetfield, Rbc, Rscratch);
2456 {
2457 Label acquire_double;
2458 __ beq(CCR6, acquire_double); // Volatile?
2459 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2461 __ bind(acquire_double);
2462 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
2463 __ beq_predict_taken(CCR0, Lisync);
2464 __ b(Lisync); // In case of NAN.
2465 }
2467 __ align(32, 28, 28); // Align load.
2468 // __ bind(Lftos);
2469 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2470 assert(branch_table[ftos] == 0, "can't compute twice");
2471 branch_table[ftos] = __ pc(); // non-volatile_entry point
2472 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
2473 __ push(ftos);
2474 if (!is_static) { patch_bytecode(Bytecodes::_fast_fgetfield, Rbc, Rscratch); }
2475 {
2476 Label acquire_float;
2477 __ beq(CCR6, acquire_float); // Volatile?
2478 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2480 __ bind(acquire_float);
2481 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
2482 __ beq_predict_taken(CCR0, Lisync);
2483 __ b(Lisync); // In case of NAN.
2484 }
2486 __ align(32, 28, 28); // Align load.
2487 // __ bind(Litos);
2488 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2489 assert(branch_table[itos] == 0, "can't compute twice");
2490 branch_table[itos] = __ pc(); // non-volatile_entry point
2491 __ lwax(R17_tos, Rclass_or_obj, Roffset);
2492 __ push(itos);
2493 if (!is_static) patch_bytecode(Bytecodes::_fast_igetfield, Rbc, Rscratch);
2494 __ beq(CCR6, Lacquire); // Volatile?
2495 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2497 __ align(32, 28, 28); // Align load.
2498 // __ bind(Lltos);
2499 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2500 assert(branch_table[ltos] == 0, "can't compute twice");
2501 branch_table[ltos] = __ pc(); // non-volatile_entry point
2502 __ ldx(R17_tos, Rclass_or_obj, Roffset);
2503 __ push(ltos);
2504 if (!is_static) patch_bytecode(Bytecodes::_fast_lgetfield, Rbc, Rscratch);
2505 __ beq(CCR6, Lacquire); // Volatile?
2506 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2508 __ align(32, 28, 28); // Align load.
2509 // __ bind(Lbtos);
2510 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2511 assert(branch_table[btos] == 0, "can't compute twice");
2512 branch_table[btos] = __ pc(); // non-volatile_entry point
2513 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
2514 __ extsb(R17_tos, R17_tos);
2515 __ push(btos);
2516 if (!is_static) patch_bytecode(Bytecodes::_fast_bgetfield, Rbc, Rscratch);
2517 __ beq(CCR6, Lacquire); // Volatile?
2518 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2520 __ align(32, 28, 28); // Align load.
2521 // __ bind(Lztos); (same code as btos)
2522 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2523 assert(branch_table[ztos] == 0, "can't compute twice");
2524 branch_table[ztos] = __ pc(); // non-volatile_entry point
2525 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
2526 __ extsb(R17_tos, R17_tos);
2527 __ push(ztos);
2528 if (!is_static) {
2529 // use btos rewriting, no truncating to t/f bit is needed for getfield.
2530 patch_bytecode(Bytecodes::_fast_bgetfield, Rbc, Rscratch);
2531 }
2532 __ beq(CCR6, Lacquire); // Volatile?
2533 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2535 __ align(32, 28, 28); // Align load.
2536 // __ bind(Lctos);
2537 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2538 assert(branch_table[ctos] == 0, "can't compute twice");
2539 branch_table[ctos] = __ pc(); // non-volatile_entry point
2540 __ lhzx(R17_tos, Rclass_or_obj, Roffset);
2541 __ push(ctos);
2542 if (!is_static) patch_bytecode(Bytecodes::_fast_cgetfield, Rbc, Rscratch);
2543 __ beq(CCR6, Lacquire); // Volatile?
2544 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2546 __ align(32, 28, 28); // Align load.
2547 // __ bind(Lstos);
2548 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2549 assert(branch_table[stos] == 0, "can't compute twice");
2550 branch_table[stos] = __ pc(); // non-volatile_entry point
2551 __ lhax(R17_tos, Rclass_or_obj, Roffset);
2552 __ push(stos);
2553 if (!is_static) patch_bytecode(Bytecodes::_fast_sgetfield, Rbc, Rscratch);
2554 __ beq(CCR6, Lacquire); // Volatile?
2555 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2557 __ align(32, 28, 28); // Align load.
2558 // __ bind(Latos);
2559 __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2560 assert(branch_table[atos] == 0, "can't compute twice");
2561 branch_table[atos] = __ pc(); // non-volatile_entry point
2562 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
2563 __ verify_oop(R17_tos);
2564 __ push(atos);
2565 //__ dcbt(R17_tos); // prefetch
2566 if (!is_static) patch_bytecode(Bytecodes::_fast_agetfield, Rbc, Rscratch);
2567 __ beq(CCR6, Lacquire); // Volatile?
2568 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2570 __ align(32, 12);
2571 __ bind(Lacquire);
2572 __ twi_0(R17_tos);
2573 __ bind(Lisync);
2574 __ isync(); // acquire
2576 #ifdef ASSERT
2577 for (int i = 0; i<number_of_states; ++i) {
2578 assert(branch_table[i], "get initialization");
2579 //tty->print_cr("get: %s_branch_table[%d] = 0x%llx (opcode 0x%llx)",
2580 // is_static ? "static" : "field", i, branch_table[i], *((unsigned int*)branch_table[i]));
2581 }
2582 #endif
2583 }
2585 void TemplateTable::getfield(int byte_no) {
2586 getfield_or_static(byte_no, false);
2587 }
2589 void TemplateTable::getstatic(int byte_no) {
2590 getfield_or_static(byte_no, true);
2591 }
2593 // The registers cache and index expected to be set before call.
2594 // The function may destroy various registers, just not the cache and index registers.
2595 void TemplateTable::jvmti_post_field_mod(Register Rcache, Register Rscratch, bool is_static) {
2597 assert_different_registers(Rcache, Rscratch, R6_ARG4);
2599 if (JvmtiExport::can_post_field_modification()) {
2600 Label Lno_field_mod_post;
2602 // Check if post field access in enabled.
2603 int offs = __ load_const_optimized(Rscratch, JvmtiExport::get_field_modification_count_addr(), R0, true);
2604 __ lwz(Rscratch, offs, Rscratch);
2606 __ cmpwi(CCR0, Rscratch, 0);
2607 __ beq(CCR0, Lno_field_mod_post);
2609 // Do the post
2610 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2611 const Register Robj = Rscratch;
2613 __ addi(Rcache, Rcache, in_bytes(cp_base_offset));
2614 if (is_static) {
2615 // Life is simple. Null out the object pointer.
2616 __ li(Robj, 0);
2617 } else {
2618 // In case of the fast versions, value lives in registers => put it back on tos.
2619 int offs = Interpreter::expr_offset_in_bytes(0);
2620 Register base = R15_esp;
2621 switch(bytecode()) {
2622 case Bytecodes::_fast_aputfield: __ push_ptr(); offs+= Interpreter::stackElementSize; break;
2623 case Bytecodes::_fast_iputfield: // Fall through
2624 case Bytecodes::_fast_bputfield: // Fall through
2625 case Bytecodes::_fast_zputfield: // Fall through
2626 case Bytecodes::_fast_cputfield: // Fall through
2627 case Bytecodes::_fast_sputfield: __ push_i(); offs+= Interpreter::stackElementSize; break;
2628 case Bytecodes::_fast_lputfield: __ push_l(); offs+=2*Interpreter::stackElementSize; break;
2629 case Bytecodes::_fast_fputfield: __ push_f(); offs+= Interpreter::stackElementSize; break;
2630 case Bytecodes::_fast_dputfield: __ push_d(); offs+=2*Interpreter::stackElementSize; break;
2631 default: {
2632 offs = 0;
2633 base = Robj;
2634 const Register Rflags = Robj;
2635 Label is_one_slot;
2636 // Life is harder. The stack holds the value on top, followed by the
2637 // object. We don't know the size of the value, though; it could be
2638 // one or two words depending on its type. As a result, we must find
2639 // the type to determine where the object is.
2640 __ ld(Rflags, in_bytes(ConstantPoolCacheEntry::flags_offset()), Rcache); // Big Endian
2641 __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2643 __ cmpwi(CCR0, Rflags, ltos);
2644 __ cmpwi(CCR1, Rflags, dtos);
2645 __ addi(base, R15_esp, Interpreter::expr_offset_in_bytes(1));
2646 __ crnor(/*CR0 eq*/2, /*CR1 eq*/4+2, /*CR0 eq*/2);
2647 __ beq(CCR0, is_one_slot);
2648 __ addi(base, R15_esp, Interpreter::expr_offset_in_bytes(2));
2649 __ bind(is_one_slot);
2650 break;
2651 }
2652 }
2653 __ ld(Robj, offs, base);
2654 __ verify_oop(Robj);
2655 }
2657 __ addi(R6_ARG4, R15_esp, Interpreter::expr_offset_in_bytes(0));
2658 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), Robj, Rcache, R6_ARG4);
2659 __ get_cache_and_index_at_bcp(Rcache, 1);
2661 // In case of the fast versions, value lives in registers => put it back on tos.
2662 switch(bytecode()) {
2663 case Bytecodes::_fast_aputfield: __ pop_ptr(); break;
2664 case Bytecodes::_fast_iputfield: // Fall through
2665 case Bytecodes::_fast_bputfield: // Fall through
2666 case Bytecodes::_fast_zputfield: // Fall through
2667 case Bytecodes::_fast_cputfield: // Fall through
2668 case Bytecodes::_fast_sputfield: __ pop_i(); break;
2669 case Bytecodes::_fast_lputfield: __ pop_l(); break;
2670 case Bytecodes::_fast_fputfield: __ pop_f(); break;
2671 case Bytecodes::_fast_dputfield: __ pop_d(); break;
2672 default: break; // Nothin' to do.
2673 }
2675 __ align(32, 12);
2676 __ bind(Lno_field_mod_post);
2677 }
2678 }
2680 // PPC64: implement volatile stores as release-store (return bytecode contains an additional release).
2681 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2682 Label Lvolatile;
2684 const Register Rcache = R5_ARG3, // Do not use ARG1/2 (causes trouble in jvmti_post_field_mod).
2685 Rclass_or_obj = R31, // Needs to survive C call.
2686 Roffset = R22_tmp2, // Needs to survive C call.
2687 Rflags = R3_ARG1,
2688 Rbtable = R4_ARG2,
2689 Rscratch = R11_scratch1,
2690 Rscratch2 = R12_scratch2,
2691 Rscratch3 = R6_ARG4,
2692 Rbc = Rscratch3;
2693 const ConditionRegister CR_is_vol = CCR2; // Non-volatile condition register (survives runtime call in do_oop_store).
2695 static address field_branch_table[number_of_states],
2696 static_branch_table[number_of_states];
2698 address* branch_table = is_static ? static_branch_table : field_branch_table;
2700 // Stack (grows up):
2701 // value
2702 // obj
2704 // Load the field offset.
2705 resolve_cache_and_index(byte_no, Rcache, Rscratch, sizeof(u2));
2706 jvmti_post_field_mod(Rcache, Rscratch, is_static);
2707 load_field_cp_cache_entry(Rclass_or_obj, Rcache, noreg, Roffset, Rflags, is_static);
2709 // Load pointer to branch table.
2710 __ load_const_optimized(Rbtable, (address)branch_table, Rscratch);
2712 // Get volatile flag.
2713 __ rldicl(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2715 // Check the field type.
2716 __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2718 #ifdef ASSERT
2719 Label LFlagInvalid;
2720 __ cmpldi(CCR0, Rflags, number_of_states);
2721 __ bge(CCR0, LFlagInvalid);
2722 #endif
2724 // Load from branch table and dispatch (volatile case: one instruction ahead).
2725 __ sldi(Rflags, Rflags, LogBytesPerWord);
2726 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) { __ cmpwi(CR_is_vol, Rscratch, 1); } // Volatile?
2727 __ sldi(Rscratch, Rscratch, exact_log2(BytesPerInstWord)); // Volatile? size of instruction 1 : 0.
2728 __ ldx(Rbtable, Rbtable, Rflags);
2730 __ subf(Rbtable, Rscratch, Rbtable); // Point to volatile/non-volatile entry point.
2731 __ mtctr(Rbtable);
2732 __ bctr();
2734 #ifdef ASSERT
2735 __ bind(LFlagInvalid);
2736 __ stop("got invalid flag", 0x656);
2738 // __ bind(Lvtos);
2739 address pc_before_release = __ pc();
2740 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2741 assert(__ pc() - pc_before_release == (ptrdiff_t)BytesPerInstWord, "must be single instruction");
2742 assert(branch_table[vtos] == 0, "can't compute twice");
2743 branch_table[vtos] = __ pc(); // non-volatile_entry point
2744 __ stop("vtos unexpected", 0x657);
2745 #endif
2747 __ align(32, 28, 28); // Align pop.
2748 // __ bind(Ldtos);
2749 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2750 assert(branch_table[dtos] == 0, "can't compute twice");
2751 branch_table[dtos] = __ pc(); // non-volatile_entry point
2752 __ pop(dtos);
2753 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2754 __ stfdx(F15_ftos, Rclass_or_obj, Roffset);
2755 if (!is_static) { patch_bytecode(Bytecodes::_fast_dputfield, Rbc, Rscratch, true, byte_no); }
2756 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2757 __ beq(CR_is_vol, Lvolatile); // Volatile?
2758 }
2759 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2761 __ align(32, 28, 28); // Align pop.
2762 // __ bind(Lftos);
2763 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2764 assert(branch_table[ftos] == 0, "can't compute twice");
2765 branch_table[ftos] = __ pc(); // non-volatile_entry point
2766 __ pop(ftos);
2767 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2768 __ stfsx(F15_ftos, Rclass_or_obj, Roffset);
2769 if (!is_static) { patch_bytecode(Bytecodes::_fast_fputfield, Rbc, Rscratch, true, byte_no); }
2770 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2771 __ beq(CR_is_vol, Lvolatile); // Volatile?
2772 }
2773 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2775 __ align(32, 28, 28); // Align pop.
2776 // __ bind(Litos);
2777 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2778 assert(branch_table[itos] == 0, "can't compute twice");
2779 branch_table[itos] = __ pc(); // non-volatile_entry point
2780 __ pop(itos);
2781 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2782 __ stwx(R17_tos, Rclass_or_obj, Roffset);
2783 if (!is_static) { patch_bytecode(Bytecodes::_fast_iputfield, Rbc, Rscratch, true, byte_no); }
2784 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2785 __ beq(CR_is_vol, Lvolatile); // Volatile?
2786 }
2787 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2789 __ align(32, 28, 28); // Align pop.
2790 // __ bind(Lltos);
2791 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2792 assert(branch_table[ltos] == 0, "can't compute twice");
2793 branch_table[ltos] = __ pc(); // non-volatile_entry point
2794 __ pop(ltos);
2795 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2796 __ stdx(R17_tos, Rclass_or_obj, Roffset);
2797 if (!is_static) { patch_bytecode(Bytecodes::_fast_lputfield, Rbc, Rscratch, true, byte_no); }
2798 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2799 __ beq(CR_is_vol, Lvolatile); // Volatile?
2800 }
2801 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2803 __ align(32, 28, 28); // Align pop.
2804 // __ bind(Lbtos);
2805 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2806 assert(branch_table[btos] == 0, "can't compute twice");
2807 branch_table[btos] = __ pc(); // non-volatile_entry point
2808 __ pop(btos);
2809 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2810 __ stbx(R17_tos, Rclass_or_obj, Roffset);
2811 if (!is_static) { patch_bytecode(Bytecodes::_fast_bputfield, Rbc, Rscratch, true, byte_no); }
2812 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2813 __ beq(CR_is_vol, Lvolatile); // Volatile?
2814 }
2815 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2817 __ align(32, 28, 28); // Align pop.
2818 // __ bind(Lztos);
2819 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2820 assert(branch_table[ztos] == 0, "can't compute twice");
2821 branch_table[ztos] = __ pc(); // non-volatile_entry point
2822 __ pop(ztos);
2823 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2824 __ andi(R17_tos, R17_tos, 0x1);
2825 __ stbx(R17_tos, Rclass_or_obj, Roffset);
2826 if (!is_static) { patch_bytecode(Bytecodes::_fast_zputfield, Rbc, Rscratch, true, byte_no); }
2827 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2828 __ beq(CR_is_vol, Lvolatile); // Volatile?
2829 }
2830 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2832 __ align(32, 28, 28); // Align pop.
2833 // __ bind(Lctos);
2834 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2835 assert(branch_table[ctos] == 0, "can't compute twice");
2836 branch_table[ctos] = __ pc(); // non-volatile_entry point
2837 __ pop(ctos);
2838 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1..
2839 __ sthx(R17_tos, Rclass_or_obj, Roffset);
2840 if (!is_static) { patch_bytecode(Bytecodes::_fast_cputfield, Rbc, Rscratch, true, byte_no); }
2841 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2842 __ beq(CR_is_vol, Lvolatile); // Volatile?
2843 }
2844 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2846 __ align(32, 28, 28); // Align pop.
2847 // __ bind(Lstos);
2848 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2849 assert(branch_table[stos] == 0, "can't compute twice");
2850 branch_table[stos] = __ pc(); // non-volatile_entry point
2851 __ pop(stos);
2852 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2853 __ sthx(R17_tos, Rclass_or_obj, Roffset);
2854 if (!is_static) { patch_bytecode(Bytecodes::_fast_sputfield, Rbc, Rscratch, true, byte_no); }
2855 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2856 __ beq(CR_is_vol, Lvolatile); // Volatile?
2857 }
2858 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2860 __ align(32, 28, 28); // Align pop.
2861 // __ bind(Latos);
2862 __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2863 assert(branch_table[atos] == 0, "can't compute twice");
2864 branch_table[atos] = __ pc(); // non-volatile_entry point
2865 __ pop(atos);
2866 if (!is_static) { pop_and_check_object(Rclass_or_obj); } // kills R11_scratch1
2867 do_oop_store(_masm, Rclass_or_obj, Roffset, R17_tos, Rscratch, Rscratch2, Rscratch3, _bs->kind(), false /* precise */, true /* check null */);
2868 if (!is_static) { patch_bytecode(Bytecodes::_fast_aputfield, Rbc, Rscratch, true, byte_no); }
2869 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2870 __ beq(CR_is_vol, Lvolatile); // Volatile?
2871 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2873 __ align(32, 12);
2874 __ bind(Lvolatile);
2875 __ fence();
2876 }
2877 // fallthru: __ b(Lexit);
2879 #ifdef ASSERT
2880 for (int i = 0; i<number_of_states; ++i) {
2881 assert(branch_table[i], "put initialization");
2882 //tty->print_cr("put: %s_branch_table[%d] = 0x%llx (opcode 0x%llx)",
2883 // is_static ? "static" : "field", i, branch_table[i], *((unsigned int*)branch_table[i]));
2884 }
2885 #endif
2886 }
2888 void TemplateTable::putfield(int byte_no) {
2889 putfield_or_static(byte_no, false);
2890 }
2892 void TemplateTable::putstatic(int byte_no) {
2893 putfield_or_static(byte_no, true);
2894 }
2896 // See SPARC. On PPC64, we have a different jvmti_post_field_mod which does the job.
2897 void TemplateTable::jvmti_post_fast_field_mod() {
2898 __ should_not_reach_here();
2899 }
2901 void TemplateTable::fast_storefield(TosState state) {
2902 transition(state, vtos);
2904 const Register Rcache = R5_ARG3, // Do not use ARG1/2 (causes trouble in jvmti_post_field_mod).
2905 Rclass_or_obj = R31, // Needs to survive C call.
2906 Roffset = R22_tmp2, // Needs to survive C call.
2907 Rflags = R3_ARG1,
2908 Rscratch = R11_scratch1,
2909 Rscratch2 = R12_scratch2,
2910 Rscratch3 = R4_ARG2;
2911 const ConditionRegister CR_is_vol = CCR2; // Non-volatile condition register (survives runtime call in do_oop_store).
2913 // Constant pool already resolved => Load flags and offset of field.
2914 __ get_cache_and_index_at_bcp(Rcache, 1);
2915 jvmti_post_field_mod(Rcache, Rscratch, false /* not static */);
2916 load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
2918 // Get the obj and the final store addr.
2919 pop_and_check_object(Rclass_or_obj); // Kills R11_scratch1.
2921 // Get volatile flag.
2922 __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2923 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) { __ cmpdi(CR_is_vol, Rscratch, 1); }
2924 {
2925 Label LnotVolatile;
2926 __ beq(CCR0, LnotVolatile);
2927 __ release();
2928 __ align(32, 12);
2929 __ bind(LnotVolatile);
2930 }
2932 // Do the store and fencing.
2933 switch(bytecode()) {
2934 case Bytecodes::_fast_aputfield:
2935 // Store into the field.
2936 do_oop_store(_masm, Rclass_or_obj, Roffset, R17_tos, Rscratch, Rscratch2, Rscratch3, _bs->kind(), false /* precise */, true /* check null */);
2937 break;
2939 case Bytecodes::_fast_iputfield:
2940 __ stwx(R17_tos, Rclass_or_obj, Roffset);
2941 break;
2943 case Bytecodes::_fast_lputfield:
2944 __ stdx(R17_tos, Rclass_or_obj, Roffset);
2945 break;
2947 case Bytecodes::_fast_zputfield:
2948 __ andi(R17_tos, R17_tos, 0x1); // boolean is true if LSB is 1
2949 // fall through to bputfield
2950 case Bytecodes::_fast_bputfield:
2951 __ stbx(R17_tos, Rclass_or_obj, Roffset);
2952 break;
2954 case Bytecodes::_fast_cputfield:
2955 case Bytecodes::_fast_sputfield:
2956 __ sthx(R17_tos, Rclass_or_obj, Roffset);
2957 break;
2959 case Bytecodes::_fast_fputfield:
2960 __ stfsx(F15_ftos, Rclass_or_obj, Roffset);
2961 break;
2963 case Bytecodes::_fast_dputfield:
2964 __ stfdx(F15_ftos, Rclass_or_obj, Roffset);
2965 break;
2967 default: ShouldNotReachHere();
2968 }
2970 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2971 Label LVolatile;
2972 __ beq(CR_is_vol, LVolatile);
2973 __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2975 __ align(32, 12);
2976 __ bind(LVolatile);
2977 __ fence();
2978 }
2979 }
2981 void TemplateTable::fast_accessfield(TosState state) {
2982 transition(atos, state);
2984 Label LisVolatile;
2985 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2987 const Register Rcache = R3_ARG1,
2988 Rclass_or_obj = R17_tos,
2989 Roffset = R22_tmp2,
2990 Rflags = R23_tmp3,
2991 Rscratch = R12_scratch2;
2993 // Constant pool already resolved. Get the field offset.
2994 __ get_cache_and_index_at_bcp(Rcache, 1);
2995 load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
2997 // JVMTI support
2998 jvmti_post_field_access(Rcache, Rscratch, false, true);
3000 // Get the load address.
3001 __ null_check_throw(Rclass_or_obj, -1, Rscratch);
3003 // Get volatile flag.
3004 __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
3005 __ bne(CCR0, LisVolatile);
3007 switch(bytecode()) {
3008 case Bytecodes::_fast_agetfield:
3009 {
3010 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3011 __ verify_oop(R17_tos);
3012 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3014 __ bind(LisVolatile);
3015 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3016 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3017 __ verify_oop(R17_tos);
3018 __ twi_0(R17_tos);
3019 __ isync();
3020 break;
3021 }
3022 case Bytecodes::_fast_igetfield:
3023 {
3024 __ lwax(R17_tos, Rclass_or_obj, Roffset);
3025 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3027 __ bind(LisVolatile);
3028 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3029 __ lwax(R17_tos, Rclass_or_obj, Roffset);
3030 __ twi_0(R17_tos);
3031 __ isync();
3032 break;
3033 }
3034 case Bytecodes::_fast_lgetfield:
3035 {
3036 __ ldx(R17_tos, Rclass_or_obj, Roffset);
3037 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3039 __ bind(LisVolatile);
3040 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3041 __ ldx(R17_tos, Rclass_or_obj, Roffset);
3042 __ twi_0(R17_tos);
3043 __ isync();
3044 break;
3045 }
3046 case Bytecodes::_fast_bgetfield:
3047 {
3048 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
3049 __ extsb(R17_tos, R17_tos);
3050 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3052 __ bind(LisVolatile);
3053 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3054 __ lbzx(R17_tos, Rclass_or_obj, Roffset);
3055 __ twi_0(R17_tos);
3056 __ extsb(R17_tos, R17_tos);
3057 __ isync();
3058 break;
3059 }
3060 case Bytecodes::_fast_cgetfield:
3061 {
3062 __ lhzx(R17_tos, Rclass_or_obj, Roffset);
3063 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3065 __ bind(LisVolatile);
3066 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3067 __ lhzx(R17_tos, Rclass_or_obj, Roffset);
3068 __ twi_0(R17_tos);
3069 __ isync();
3070 break;
3071 }
3072 case Bytecodes::_fast_sgetfield:
3073 {
3074 __ lhax(R17_tos, Rclass_or_obj, Roffset);
3075 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3077 __ bind(LisVolatile);
3078 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3079 __ lhax(R17_tos, Rclass_or_obj, Roffset);
3080 __ twi_0(R17_tos);
3081 __ isync();
3082 break;
3083 }
3084 case Bytecodes::_fast_fgetfield:
3085 {
3086 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3087 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3089 __ bind(LisVolatile);
3090 Label Ldummy;
3091 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3092 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3093 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3094 __ bne_predict_not_taken(CCR0, Ldummy);
3095 __ bind(Ldummy);
3096 __ isync();
3097 break;
3098 }
3099 case Bytecodes::_fast_dgetfield:
3100 {
3101 __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
3102 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3104 __ bind(LisVolatile);
3105 Label Ldummy;
3106 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3107 __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
3108 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3109 __ bne_predict_not_taken(CCR0, Ldummy);
3110 __ bind(Ldummy);
3111 __ isync();
3112 break;
3113 }
3114 default: ShouldNotReachHere();
3115 }
3116 }
3118 void TemplateTable::fast_xaccess(TosState state) {
3119 transition(vtos, state);
3121 Label LisVolatile;
3122 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
3123 const Register Rcache = R3_ARG1,
3124 Rclass_or_obj = R17_tos,
3125 Roffset = R22_tmp2,
3126 Rflags = R23_tmp3,
3127 Rscratch = R12_scratch2;
3129 __ ld(Rclass_or_obj, 0, R18_locals);
3131 // Constant pool already resolved. Get the field offset.
3132 __ get_cache_and_index_at_bcp(Rcache, 2);
3133 load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
3135 // JVMTI support not needed, since we switch back to single bytecode as soon as debugger attaches.
3137 // Needed to report exception at the correct bcp.
3138 __ addi(R14_bcp, R14_bcp, 1);
3140 // Get the load address.
3141 __ null_check_throw(Rclass_or_obj, -1, Rscratch);
3143 // Get volatile flag.
3144 __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
3145 __ bne(CCR0, LisVolatile);
3147 switch(state) {
3148 case atos:
3149 {
3150 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3151 __ verify_oop(R17_tos);
3152 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3154 __ bind(LisVolatile);
3155 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3156 __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3157 __ verify_oop(R17_tos);
3158 __ twi_0(R17_tos);
3159 __ isync();
3160 break;
3161 }
3162 case itos:
3163 {
3164 __ lwax(R17_tos, Rclass_or_obj, Roffset);
3165 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3167 __ bind(LisVolatile);
3168 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3169 __ lwax(R17_tos, Rclass_or_obj, Roffset);
3170 __ twi_0(R17_tos);
3171 __ isync();
3172 break;
3173 }
3174 case ftos:
3175 {
3176 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3177 __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3179 __ bind(LisVolatile);
3180 Label Ldummy;
3181 if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3182 __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3183 __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3184 __ bne_predict_not_taken(CCR0, Ldummy);
3185 __ bind(Ldummy);
3186 __ isync();
3187 break;
3188 }
3189 default: ShouldNotReachHere();
3190 }
3191 __ addi(R14_bcp, R14_bcp, -1);
3192 }
3194 // ============================================================================
3195 // Calls
3197 // Common code for invoke
3198 //
3199 // Input:
3200 // - byte_no
3201 //
3202 // Output:
3203 // - Rmethod: The method to invoke next.
3204 // - Rret_addr: The return address to return to.
3205 // - Rindex: MethodType (invokehandle) or CallSite obj (invokedynamic)
3206 // - Rrecv: Cache for "this" pointer, might be noreg if static call.
3207 // - Rflags: Method flags from const pool cache.
3208 //
3209 // Kills:
3210 // - Rscratch1
3211 //
3212 void TemplateTable::prepare_invoke(int byte_no,
3213 Register Rmethod, // linked method (or i-klass)
3214 Register Rret_addr,// return address
3215 Register Rindex, // itable index, MethodType, etc.
3216 Register Rrecv, // If caller wants to see it.
3217 Register Rflags, // If caller wants to test it.
3218 Register Rscratch
3219 ) {
3220 // Determine flags.
3221 const Bytecodes::Code code = bytecode();
3222 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3223 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3224 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3225 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3226 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3227 const bool load_receiver = (Rrecv != noreg);
3228 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3230 assert_different_registers(Rmethod, Rindex, Rflags, Rscratch);
3231 assert_different_registers(Rmethod, Rrecv, Rflags, Rscratch);
3232 assert_different_registers(Rret_addr, Rscratch);
3234 load_invoke_cp_cache_entry(byte_no, Rmethod, Rindex, Rflags, is_invokevirtual, false, is_invokedynamic);
3236 // Saving of SP done in call_from_interpreter.
3238 // Maybe push "appendix" to arguments.
3239 if (is_invokedynamic || is_invokehandle) {
3240 Label Ldone;
3241 __ rldicl_(R0, Rflags, 64-ConstantPoolCacheEntry::has_appendix_shift, 63);
3242 __ beq(CCR0, Ldone);
3243 // Push "appendix" (MethodType, CallSite, etc.).
3244 // This must be done before we get the receiver,
3245 // since the parameter_size includes it.
3246 __ load_resolved_reference_at_index(Rscratch, Rindex);
3247 __ verify_oop(Rscratch);
3248 __ push_ptr(Rscratch);
3249 __ bind(Ldone);
3250 }
3252 // Load receiver if needed (after appendix is pushed so parameter size is correct).
3253 if (load_receiver) {
3254 const Register Rparam_count = Rscratch;
3255 __ andi(Rparam_count, Rflags, ConstantPoolCacheEntry::parameter_size_mask);
3256 __ load_receiver(Rparam_count, Rrecv);
3257 __ verify_oop(Rrecv);
3258 }
3260 // Get return address.
3261 {
3262 Register Rtable_addr = Rscratch;
3263 Register Rret_type = Rret_addr;
3264 address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3266 // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3267 __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3268 __ load_dispatch_table(Rtable_addr, (address*)table_addr);
3269 __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3270 // Get return address.
3271 __ ldx(Rret_addr, Rtable_addr, Rret_type);
3272 }
3273 }
3275 // Helper for virtual calls. Load target out of vtable and jump off!
3276 // Kills all passed registers.
3277 void TemplateTable::generate_vtable_call(Register Rrecv_klass, Register Rindex, Register Rret, Register Rtemp) {
3279 assert_different_registers(Rrecv_klass, Rtemp, Rret);
3280 const Register Rtarget_method = Rindex;
3282 // Get target method & entry point.
3283 const int base = InstanceKlass::vtable_start_offset() * wordSize;
3284 // Calc vtable addr scale the vtable index by 8.
3285 __ sldi(Rindex, Rindex, exact_log2(vtableEntry::size() * wordSize));
3286 // Load target.
3287 __ addi(Rrecv_klass, Rrecv_klass, base + vtableEntry::method_offset_in_bytes());
3288 __ ldx(Rtarget_method, Rindex, Rrecv_klass);
3289 // Argument and return type profiling.
3290 __ profile_arguments_type(Rtarget_method, Rrecv_klass /* scratch1 */, Rtemp /* scratch2 */, true);
3291 __ call_from_interpreter(Rtarget_method, Rret, Rrecv_klass /* scratch1 */, Rtemp /* scratch2 */);
3292 }
3294 // Virtual or final call. Final calls are rewritten on the fly to run through "fast_finalcall" next time.
3295 void TemplateTable::invokevirtual(int byte_no) {
3296 transition(vtos, vtos);
3298 Register Rtable_addr = R11_scratch1,
3299 Rret_type = R12_scratch2,
3300 Rret_addr = R5_ARG3,
3301 Rflags = R22_tmp2, // Should survive C call.
3302 Rrecv = R3_ARG1,
3303 Rrecv_klass = Rrecv,
3304 Rvtableindex_or_method = R31, // Should survive C call.
3305 Rnum_params = R4_ARG2,
3306 Rnew_bc = R6_ARG4;
3308 Label LnotFinal;
3310 load_invoke_cp_cache_entry(byte_no, Rvtableindex_or_method, noreg, Rflags, /*virtual*/ true, false, false);
3312 __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_vfinal_shift);
3313 __ bfalse(CCR0, LnotFinal);
3315 patch_bytecode(Bytecodes::_fast_invokevfinal, Rnew_bc, R12_scratch2);
3316 invokevfinal_helper(Rvtableindex_or_method, Rflags, R11_scratch1, R12_scratch2);
3318 __ align(32, 12);
3319 __ bind(LnotFinal);
3320 // Load "this" pointer (receiver).
3321 __ rldicl(Rnum_params, Rflags, 64, 48);
3322 __ load_receiver(Rnum_params, Rrecv);
3323 __ verify_oop(Rrecv);
3325 // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3326 __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3327 __ load_dispatch_table(Rtable_addr, Interpreter::invoke_return_entry_table());
3328 __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3329 __ ldx(Rret_addr, Rret_type, Rtable_addr);
3330 __ null_check_throw(Rrecv, oopDesc::klass_offset_in_bytes(), R11_scratch1);
3331 __ load_klass(Rrecv_klass, Rrecv);
3332 __ verify_klass_ptr(Rrecv_klass);
3333 __ profile_virtual_call(Rrecv_klass, R11_scratch1, R12_scratch2, false);
3335 generate_vtable_call(Rrecv_klass, Rvtableindex_or_method, Rret_addr, R11_scratch1);
3336 }
3338 void TemplateTable::fast_invokevfinal(int byte_no) {
3339 transition(vtos, vtos);
3341 assert(byte_no == f2_byte, "use this argument");
3342 Register Rflags = R22_tmp2,
3343 Rmethod = R31;
3344 load_invoke_cp_cache_entry(byte_no, Rmethod, noreg, Rflags, /*virtual*/ true, /*is_invokevfinal*/ true, false);
3345 invokevfinal_helper(Rmethod, Rflags, R11_scratch1, R12_scratch2);
3346 }
3348 void TemplateTable::invokevfinal_helper(Register Rmethod, Register Rflags, Register Rscratch1, Register Rscratch2) {
3350 assert_different_registers(Rmethod, Rflags, Rscratch1, Rscratch2);
3352 // Load receiver from stack slot.
3353 Register Rrecv = Rscratch2;
3354 Register Rnum_params = Rrecv;
3356 __ ld(Rnum_params, in_bytes(Method::const_offset()), Rmethod);
3357 __ lhz(Rnum_params /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), Rnum_params);
3359 // Get return address.
3360 Register Rtable_addr = Rscratch1,
3361 Rret_addr = Rflags,
3362 Rret_type = Rret_addr;
3363 // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3364 __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3365 __ load_dispatch_table(Rtable_addr, Interpreter::invoke_return_entry_table());
3366 __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3367 __ ldx(Rret_addr, Rret_type, Rtable_addr);
3369 // Load receiver and receiver NULL check.
3370 __ load_receiver(Rnum_params, Rrecv);
3371 __ null_check_throw(Rrecv, -1, Rscratch1);
3373 __ profile_final_call(Rrecv, Rscratch1);
3374 // Argument and return type profiling.
3375 __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, true);
3377 // Do the call.
3378 __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1, Rscratch2);
3379 }
3381 void TemplateTable::invokespecial(int byte_no) {
3382 assert(byte_no == f1_byte, "use this argument");
3383 transition(vtos, vtos);
3385 Register Rtable_addr = R3_ARG1,
3386 Rret_addr = R4_ARG2,
3387 Rflags = R5_ARG3,
3388 Rreceiver = R6_ARG4,
3389 Rmethod = R31;
3391 prepare_invoke(byte_no, Rmethod, Rret_addr, noreg, Rreceiver, Rflags, R11_scratch1);
3393 // Receiver NULL check.
3394 __ null_check_throw(Rreceiver, -1, R11_scratch1);
3396 __ profile_call(R11_scratch1, R12_scratch2);
3397 // Argument and return type profiling.
3398 __ profile_arguments_type(Rmethod, R11_scratch1, R12_scratch2, false);
3399 __ call_from_interpreter(Rmethod, Rret_addr, R11_scratch1, R12_scratch2);
3400 }
3402 void TemplateTable::invokestatic(int byte_no) {
3403 assert(byte_no == f1_byte, "use this argument");
3404 transition(vtos, vtos);
3406 Register Rtable_addr = R3_ARG1,
3407 Rret_addr = R4_ARG2,
3408 Rflags = R5_ARG3;
3410 prepare_invoke(byte_no, R19_method, Rret_addr, noreg, noreg, Rflags, R11_scratch1);
3412 __ profile_call(R11_scratch1, R12_scratch2);
3413 // Argument and return type profiling.
3414 __ profile_arguments_type(R19_method, R11_scratch1, R12_scratch2, false);
3415 __ call_from_interpreter(R19_method, Rret_addr, R11_scratch1, R12_scratch2);
3416 }
3418 void TemplateTable::invokeinterface_object_method(Register Rrecv_klass,
3419 Register Rret,
3420 Register Rflags,
3421 Register Rmethod,
3422 Register Rtemp1,
3423 Register Rtemp2) {
3425 assert_different_registers(Rmethod, Rret, Rrecv_klass, Rflags, Rtemp1, Rtemp2);
3426 Label LnotFinal;
3428 // Check for vfinal.
3429 __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_vfinal_shift);
3430 __ bfalse(CCR0, LnotFinal);
3432 Register Rscratch = Rflags; // Rflags is dead now.
3434 // Final call case.
3435 __ profile_final_call(Rtemp1, Rscratch);
3436 // Argument and return type profiling.
3437 __ profile_arguments_type(Rmethod, Rscratch, Rrecv_klass /* scratch */, true);
3438 // Do the final call - the index (f2) contains the method.
3439 __ call_from_interpreter(Rmethod, Rret, Rscratch, Rrecv_klass /* scratch */);
3441 // Non-final callc case.
3442 __ bind(LnotFinal);
3443 __ profile_virtual_call(Rrecv_klass, Rtemp1, Rscratch, false);
3444 generate_vtable_call(Rrecv_klass, Rmethod, Rret, Rscratch);
3445 }
3447 void TemplateTable::invokeinterface(int byte_no) {
3448 assert(byte_no == f1_byte, "use this argument");
3449 transition(vtos, vtos);
3451 const Register Rscratch1 = R11_scratch1,
3452 Rscratch2 = R12_scratch2,
3453 Rmethod = R6_ARG4,
3454 Rmethod2 = R9_ARG7,
3455 Rinterface_klass = R5_ARG3,
3456 Rret_addr = R8_ARG6,
3457 Rindex = R10_ARG8,
3458 Rreceiver = R3_ARG1,
3459 Rrecv_klass = R4_ARG2,
3460 Rflags = R7_ARG5;
3462 prepare_invoke(byte_no, Rinterface_klass, Rret_addr, Rmethod, Rreceiver, Rflags, Rscratch1);
3464 // Get receiver klass.
3465 __ null_check_throw(Rreceiver, oopDesc::klass_offset_in_bytes(), Rscratch2);
3466 __ load_klass(Rrecv_klass, Rreceiver);
3468 // Check corner case object method.
3469 Label LobjectMethod, L_no_such_interface, Lthrow_ame;
3470 __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_forced_virtual_shift);
3471 __ btrue(CCR0, LobjectMethod);
3473 __ lookup_interface_method(Rrecv_klass, Rinterface_klass, noreg, noreg, Rscratch1, Rscratch2,
3474 L_no_such_interface, /*return_method=*/false);
3476 __ profile_virtual_call(Rrecv_klass, Rscratch1, Rscratch2, false);
3478 // Find entry point to call.
3480 // Get declaring interface class from method
3481 __ ld(Rinterface_klass, in_bytes(Method::const_offset()), Rmethod);
3482 __ ld(Rinterface_klass, in_bytes(ConstMethod::constants_offset()), Rinterface_klass);
3483 __ ld(Rinterface_klass, ConstantPool::pool_holder_offset_in_bytes(), Rinterface_klass);
3485 // Get itable index from method
3486 __ lwa(Rindex, in_bytes(Method::itable_index_offset()), Rmethod);
3487 __ subfic(Rindex, Rindex, Method::itable_index_max);
3489 __ lookup_interface_method(Rrecv_klass, Rinterface_klass, Rindex, Rmethod2, Rscratch1, Rscratch2,
3490 L_no_such_interface);
3492 __ cmpdi(CCR0, Rmethod2, 0);
3493 __ beq(CCR0, Lthrow_ame);
3494 // Found entry. Jump off!
3495 // Argument and return type profiling.
3496 __ profile_arguments_type(Rmethod2, Rscratch1, Rscratch2, true);
3497 //__ profile_called_method(Rindex, Rscratch1);
3498 __ call_from_interpreter(Rmethod2, Rret_addr, Rscratch1, Rscratch2);
3500 // Vtable entry was NULL => Throw abstract method error.
3501 __ bind(Lthrow_ame);
3502 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3504 // Interface was not found => Throw incompatible class change error.
3505 __ bind(L_no_such_interface);
3506 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeError));
3507 DEBUG_ONLY( __ should_not_reach_here(); )
3509 // Special case of invokeinterface called for virtual method of
3510 // java.lang.Object. See ConstantPoolCacheEntry::set_method() for details:
3511 // The invokeinterface was rewritten to a invokevirtual, hence we have
3512 // to handle this corner case. This code isn't produced by javac, but could
3513 // be produced by another compliant java compiler.
3514 __ bind(LobjectMethod);
3515 invokeinterface_object_method(Rrecv_klass, Rret_addr, Rflags, Rmethod, Rscratch1, Rscratch2);
3516 }
3518 void TemplateTable::invokedynamic(int byte_no) {
3519 transition(vtos, vtos);
3521 const Register Rret_addr = R3_ARG1,
3522 Rflags = R4_ARG2,
3523 Rmethod = R22_tmp2,
3524 Rscratch1 = R11_scratch1,
3525 Rscratch2 = R12_scratch2;
3527 if (!EnableInvokeDynamic) {
3528 // We should not encounter this bytecode if !EnableInvokeDynamic.
3529 // The verifier will stop it. However, if we get past the verifier,
3530 // this will stop the thread in a reasonable way, without crashing the JVM.
3531 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeError));
3532 // The call_VM checks for exception, so we should never return here.
3533 __ should_not_reach_here();
3534 return;
3535 }
3537 prepare_invoke(byte_no, Rmethod, Rret_addr, Rscratch1, noreg, Rflags, Rscratch2);
3539 // Profile this call.
3540 __ profile_call(Rscratch1, Rscratch2);
3542 // Off we go. With the new method handles, we don't jump to a method handle
3543 // entry any more. Instead, we pushed an "appendix" in prepare invoke, which happens
3544 // to be the callsite object the bootstrap method returned. This is passed to a
3545 // "link" method which does the dispatch (Most likely just grabs the MH stored
3546 // inside the callsite and does an invokehandle).
3547 // Argument and return type profiling.
3548 __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, false);
3549 __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1 /* scratch1 */, Rscratch2 /* scratch2 */);
3550 }
3552 void TemplateTable::invokehandle(int byte_no) {
3553 transition(vtos, vtos);
3555 const Register Rret_addr = R3_ARG1,
3556 Rflags = R4_ARG2,
3557 Rrecv = R5_ARG3,
3558 Rmethod = R22_tmp2,
3559 Rscratch1 = R11_scratch1,
3560 Rscratch2 = R12_scratch2;
3562 if (!EnableInvokeDynamic) {
3563 // Rewriter does not generate this bytecode.
3564 __ should_not_reach_here();
3565 return;
3566 }
3568 prepare_invoke(byte_no, Rmethod, Rret_addr, Rscratch1, Rrecv, Rflags, Rscratch2);
3569 __ verify_method_ptr(Rmethod);
3570 __ null_check_throw(Rrecv, -1, Rscratch2);
3572 __ profile_final_call(Rrecv, Rscratch1);
3574 // Still no call from handle => We call the method handle interpreter here.
3575 // Argument and return type profiling.
3576 __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, true);
3577 __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1 /* scratch1 */, Rscratch2 /* scratch2 */);
3578 }
3580 // =============================================================================
3581 // Allocation
3583 // Puts allocated obj ref onto the expression stack.
3584 void TemplateTable::_new() {
3585 transition(vtos, atos);
3587 Label Lslow_case,
3588 Ldone,
3589 Linitialize_header,
3590 Lallocate_shared,
3591 Linitialize_object; // Including clearing the fields.
3593 const Register RallocatedObject = R17_tos,
3594 RinstanceKlass = R9_ARG7,
3595 Rscratch = R11_scratch1,
3596 Roffset = R8_ARG6,
3597 Rinstance_size = Roffset,
3598 Rcpool = R4_ARG2,
3599 Rtags = R3_ARG1,
3600 Rindex = R5_ARG3;
3602 const bool allow_shared_alloc = Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
3604 // --------------------------------------------------------------------------
3605 // Check if fast case is possible.
3607 // Load pointers to const pool and const pool's tags array.
3608 __ get_cpool_and_tags(Rcpool, Rtags);
3609 // Load index of constant pool entry.
3610 __ get_2_byte_integer_at_bcp(1, Rindex, InterpreterMacroAssembler::Unsigned);
3612 if (UseTLAB) {
3613 // Make sure the class we're about to instantiate has been resolved
3614 // This is done before loading instanceKlass to be consistent with the order
3615 // how Constant Pool is updated (see ConstantPoolCache::klass_at_put).
3616 __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3617 __ lbzx(Rtags, Rindex, Rtags);
3619 __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3620 __ bne(CCR0, Lslow_case);
3622 // Get instanceKlass (load from Rcpool + sizeof(ConstantPool) + Rindex*BytesPerWord).
3623 __ sldi(Roffset, Rindex, LogBytesPerWord);
3624 __ addi(Rscratch, Rcpool, sizeof(ConstantPool));
3625 __ isync(); // Order load of instance Klass wrt. tags.
3626 __ ldx(RinstanceKlass, Roffset, Rscratch);
3628 // Make sure klass is fully initialized and get instance_size.
3629 __ lbz(Rscratch, in_bytes(InstanceKlass::init_state_offset()), RinstanceKlass);
3630 __ lwz(Rinstance_size, in_bytes(Klass::layout_helper_offset()), RinstanceKlass);
3632 __ cmpdi(CCR1, Rscratch, InstanceKlass::fully_initialized);
3633 // Make sure klass does not have has_finalizer, or is abstract, or interface or java/lang/Class.
3634 __ andi_(R0, Rinstance_size, Klass::_lh_instance_slow_path_bit); // slow path bit equals 0?
3636 __ crnand(/*CR0 eq*/2, /*CR1 eq*/4+2, /*CR0 eq*/2); // slow path bit set or not fully initialized?
3637 __ beq(CCR0, Lslow_case);
3639 // --------------------------------------------------------------------------
3640 // Fast case:
3641 // Allocate the instance.
3642 // 1) Try to allocate in the TLAB.
3643 // 2) If fail, and the TLAB is not full enough to discard, allocate in the shared Eden.
3644 // 3) If the above fails (or is not applicable), go to a slow case (creates a new TLAB, etc.).
3646 Register RoldTopValue = RallocatedObject; // Object will be allocated here if it fits.
3647 Register RnewTopValue = R6_ARG4;
3648 Register RendValue = R7_ARG5;
3650 // Check if we can allocate in the TLAB.
3651 __ ld(RoldTopValue, in_bytes(JavaThread::tlab_top_offset()), R16_thread);
3652 __ ld(RendValue, in_bytes(JavaThread::tlab_end_offset()), R16_thread);
3654 __ add(RnewTopValue, Rinstance_size, RoldTopValue);
3656 // If there is enough space, we do not CAS and do not clear.
3657 __ cmpld(CCR0, RnewTopValue, RendValue);
3658 __ bgt(CCR0, allow_shared_alloc ? Lallocate_shared : Lslow_case);
3660 __ std(RnewTopValue, in_bytes(JavaThread::tlab_top_offset()), R16_thread);
3662 if (ZeroTLAB) {
3663 // The fields have already been cleared.
3664 __ b(Linitialize_header);
3665 } else {
3666 // Initialize both the header and fields.
3667 __ b(Linitialize_object);
3668 }
3670 // Fall through: TLAB was too small.
3671 if (allow_shared_alloc) {
3672 Register RtlabWasteLimitValue = R10_ARG8;
3673 Register RfreeValue = RnewTopValue;
3675 __ bind(Lallocate_shared);
3676 // Check if tlab should be discarded (refill_waste_limit >= free).
3677 __ ld(RtlabWasteLimitValue, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), R16_thread);
3678 __ subf(RfreeValue, RoldTopValue, RendValue);
3679 __ srdi(RfreeValue, RfreeValue, LogHeapWordSize); // in dwords
3680 __ cmpld(CCR0, RtlabWasteLimitValue, RfreeValue);
3681 __ bge(CCR0, Lslow_case);
3683 // Increment waste limit to prevent getting stuck on this slow path.
3684 __ addi(RtlabWasteLimitValue, RtlabWasteLimitValue, (int)ThreadLocalAllocBuffer::refill_waste_limit_increment());
3685 __ std(RtlabWasteLimitValue, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), R16_thread);
3686 }
3687 // else: No allocation in the shared eden. // fallthru: __ b(Lslow_case);
3688 }
3689 // else: Always go the slow path.
3691 // --------------------------------------------------------------------------
3692 // slow case
3693 __ bind(Lslow_case);
3694 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), Rcpool, Rindex);
3696 if (UseTLAB) {
3697 __ b(Ldone);
3698 // --------------------------------------------------------------------------
3699 // Init1: Zero out newly allocated memory.
3701 if (!ZeroTLAB || allow_shared_alloc) {
3702 // Clear object fields.
3703 __ bind(Linitialize_object);
3705 // Initialize remaining object fields.
3706 Register Rbase = Rtags;
3707 __ addi(Rinstance_size, Rinstance_size, 7 - (int)sizeof(oopDesc));
3708 __ addi(Rbase, RallocatedObject, sizeof(oopDesc));
3709 __ srdi(Rinstance_size, Rinstance_size, 3);
3711 // Clear out object skipping header. Takes also care of the zero length case.
3712 __ clear_memory_doubleword(Rbase, Rinstance_size);
3713 // fallthru: __ b(Linitialize_header);
3714 }
3716 // --------------------------------------------------------------------------
3717 // Init2: Initialize the header: mark, klass
3718 __ bind(Linitialize_header);
3720 // Init mark.
3721 if (UseBiasedLocking) {
3722 __ ld(Rscratch, in_bytes(Klass::prototype_header_offset()), RinstanceKlass);
3723 } else {
3724 __ load_const_optimized(Rscratch, markOopDesc::prototype(), R0);
3725 }
3726 __ std(Rscratch, oopDesc::mark_offset_in_bytes(), RallocatedObject);
3728 // Init klass.
3729 __ store_klass_gap(RallocatedObject);
3730 __ store_klass(RallocatedObject, RinstanceKlass, Rscratch); // klass (last for cms)
3732 // Check and trigger dtrace event.
3733 {
3734 SkipIfEqualZero skip_if(_masm, Rscratch, &DTraceAllocProbes);
3735 __ push(atos);
3736 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc));
3737 __ pop(atos);
3738 }
3739 }
3741 // continue
3742 __ bind(Ldone);
3744 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3745 __ membar(Assembler::StoreStore);
3746 }
3748 void TemplateTable::newarray() {
3749 transition(itos, atos);
3751 __ lbz(R4, 1, R14_bcp);
3752 __ extsw(R5, R17_tos);
3753 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), R4, R5 /* size */);
3755 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3756 __ membar(Assembler::StoreStore);
3757 }
3759 void TemplateTable::anewarray() {
3760 transition(itos, atos);
3762 __ get_constant_pool(R4);
3763 __ get_2_byte_integer_at_bcp(1, R5, InterpreterMacroAssembler::Unsigned);
3764 __ extsw(R6, R17_tos); // size
3765 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), R4 /* pool */, R5 /* index */, R6 /* size */);
3767 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3768 __ membar(Assembler::StoreStore);
3769 }
3771 // Allocate a multi dimensional array
3772 void TemplateTable::multianewarray() {
3773 transition(vtos, atos);
3775 Register Rptr = R31; // Needs to survive C call.
3777 // Put ndims * wordSize into frame temp slot
3778 __ lbz(Rptr, 3, R14_bcp);
3779 __ sldi(Rptr, Rptr, Interpreter::logStackElementSize);
3780 // Esp points past last_dim, so set to R4 to first_dim address.
3781 __ add(R4, Rptr, R15_esp);
3782 call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), R4 /* first_size_address */);
3783 // Pop all dimensions off the stack.
3784 __ add(R15_esp, Rptr, R15_esp);
3786 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3787 __ membar(Assembler::StoreStore);
3788 }
3790 void TemplateTable::arraylength() {
3791 transition(atos, itos);
3793 Label LnoException;
3794 __ verify_oop(R17_tos);
3795 __ null_check_throw(R17_tos, arrayOopDesc::length_offset_in_bytes(), R11_scratch1);
3796 __ lwa(R17_tos, arrayOopDesc::length_offset_in_bytes(), R17_tos);
3797 }
3799 // ============================================================================
3800 // Typechecks
3802 void TemplateTable::checkcast() {
3803 transition(atos, atos);
3805 Label Ldone, Lis_null, Lquicked, Lresolved;
3806 Register Roffset = R6_ARG4,
3807 RobjKlass = R4_ARG2,
3808 RspecifiedKlass = R5_ARG3, // Generate_ClassCastException_verbose_handler will read value from this register.
3809 Rcpool = R11_scratch1,
3810 Rtags = R12_scratch2;
3812 // Null does not pass.
3813 __ cmpdi(CCR0, R17_tos, 0);
3814 __ beq(CCR0, Lis_null);
3816 // Get constant pool tag to find out if the bytecode has already been "quickened".
3817 __ get_cpool_and_tags(Rcpool, Rtags);
3819 __ get_2_byte_integer_at_bcp(1, Roffset, InterpreterMacroAssembler::Unsigned);
3821 __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3822 __ lbzx(Rtags, Rtags, Roffset);
3824 __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3825 __ beq(CCR0, Lquicked);
3827 // Call into the VM to "quicken" instanceof.
3828 __ push_ptr(); // for GC
3829 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3830 __ get_vm_result_2(RspecifiedKlass);
3831 __ pop_ptr(); // Restore receiver.
3832 __ b(Lresolved);
3834 // Extract target class from constant pool.
3835 __ bind(Lquicked);
3836 __ sldi(Roffset, Roffset, LogBytesPerWord);
3837 __ addi(Rcpool, Rcpool, sizeof(ConstantPool));
3838 __ isync(); // Order load of specified Klass wrt. tags.
3839 __ ldx(RspecifiedKlass, Rcpool, Roffset);
3841 // Do the checkcast.
3842 __ bind(Lresolved);
3843 // Get value klass in RobjKlass.
3844 __ load_klass(RobjKlass, R17_tos);
3845 // Generate a fast subtype check. Branch to cast_ok if no failure. Return 0 if failure.
3846 __ gen_subtype_check(RobjKlass, RspecifiedKlass, /*3 temp regs*/ Roffset, Rcpool, Rtags, /*target if subtype*/ Ldone);
3848 // Not a subtype; so must throw exception
3849 // Target class oop is in register R6_ARG4 == RspecifiedKlass by convention.
3850 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ClassCastException_entry);
3851 __ mtctr(R11_scratch1);
3852 __ bctr();
3854 // Profile the null case.
3855 __ align(32, 12);
3856 __ bind(Lis_null);
3857 __ profile_null_seen(R11_scratch1, Rtags); // Rtags used as scratch.
3859 __ align(32, 12);
3860 __ bind(Ldone);
3861 }
3863 // Output:
3864 // - tos == 0: Obj was null or not an instance of class.
3865 // - tos == 1: Obj was an instance of class.
3866 void TemplateTable::instanceof() {
3867 transition(atos, itos);
3869 Label Ldone, Lis_null, Lquicked, Lresolved;
3870 Register Roffset = R5_ARG3,
3871 RobjKlass = R4_ARG2,
3872 RspecifiedKlass = R6_ARG4, // Generate_ClassCastException_verbose_handler will expect the value in this register.
3873 Rcpool = R11_scratch1,
3874 Rtags = R12_scratch2;
3876 // Null does not pass.
3877 __ cmpdi(CCR0, R17_tos, 0);
3878 __ beq(CCR0, Lis_null);
3880 // Get constant pool tag to find out if the bytecode has already been "quickened".
3881 __ get_cpool_and_tags(Rcpool, Rtags);
3883 __ get_2_byte_integer_at_bcp(1, Roffset, InterpreterMacroAssembler::Unsigned);
3885 __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3886 __ lbzx(Rtags, Rtags, Roffset);
3888 __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3889 __ beq(CCR0, Lquicked);
3891 // Call into the VM to "quicken" instanceof.
3892 __ push_ptr(); // for GC
3893 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3894 __ get_vm_result_2(RspecifiedKlass);
3895 __ pop_ptr(); // Restore receiver.
3896 __ b(Lresolved);
3898 // Extract target class from constant pool.
3899 __ bind(Lquicked);
3900 __ sldi(Roffset, Roffset, LogBytesPerWord);
3901 __ addi(Rcpool, Rcpool, sizeof(ConstantPool));
3902 __ isync(); // Order load of specified Klass wrt. tags.
3903 __ ldx(RspecifiedKlass, Rcpool, Roffset);
3905 // Do the checkcast.
3906 __ bind(Lresolved);
3907 // Get value klass in RobjKlass.
3908 __ load_klass(RobjKlass, R17_tos);
3909 // Generate a fast subtype check. Branch to cast_ok if no failure. Return 0 if failure.
3910 __ li(R17_tos, 1);
3911 __ gen_subtype_check(RobjKlass, RspecifiedKlass, /*3 temp regs*/ Roffset, Rcpool, Rtags, /*target if subtype*/ Ldone);
3912 __ li(R17_tos, 0);
3914 if (ProfileInterpreter) {
3915 __ b(Ldone);
3916 }
3918 // Profile the null case.
3919 __ align(32, 12);
3920 __ bind(Lis_null);
3921 __ profile_null_seen(Rcpool, Rtags); // Rcpool and Rtags used as scratch.
3923 __ align(32, 12);
3924 __ bind(Ldone);
3925 }
3927 // =============================================================================
3928 // Breakpoints
3930 void TemplateTable::_breakpoint() {
3931 transition(vtos, vtos);
3933 // Get the unpatched byte code.
3934 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), R19_method, R14_bcp);
3935 __ mr(R31, R3_RET);
3937 // Post the breakpoint event.
3938 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), R19_method, R14_bcp);
3940 // Complete the execution of original bytecode.
3941 __ dispatch_Lbyte_code(vtos, R31, Interpreter::normal_table(vtos));
3942 }
3944 // =============================================================================
3945 // Exceptions
3947 void TemplateTable::athrow() {
3948 transition(atos, vtos);
3950 // Exception oop is in tos
3951 __ verify_oop(R17_tos);
3953 __ null_check_throw(R17_tos, -1, R11_scratch1);
3955 // Throw exception interpreter entry expects exception oop to be in R3.
3956 __ mr(R3_RET, R17_tos);
3957 __ load_dispatch_table(R11_scratch1, (address*)Interpreter::throw_exception_entry());
3958 __ mtctr(R11_scratch1);
3959 __ bctr();
3960 }
3962 // =============================================================================
3963 // Synchronization
3964 // Searches the basic object lock list on the stack for a free slot
3965 // and uses it to lock the obect in tos.
3966 //
3967 // Recursive locking is enabled by exiting the search if the same
3968 // object is already found in the list. Thus, a new basic lock obj lock
3969 // is allocated "higher up" in the stack and thus is found first
3970 // at next monitor exit.
3971 void TemplateTable::monitorenter() {
3972 transition(atos, vtos);
3974 __ verify_oop(R17_tos);
3976 Register Rcurrent_monitor = R11_scratch1,
3977 Rcurrent_obj = R12_scratch2,
3978 Robj_to_lock = R17_tos,
3979 Rscratch1 = R3_ARG1,
3980 Rscratch2 = R4_ARG2,
3981 Rscratch3 = R5_ARG3,
3982 Rcurrent_obj_addr = R6_ARG4;
3984 // ------------------------------------------------------------------------------
3985 // Null pointer exception.
3986 __ null_check_throw(Robj_to_lock, -1, R11_scratch1);
3988 // Try to acquire a lock on the object.
3989 // Repeat until succeeded (i.e., until monitorenter returns true).
3991 // ------------------------------------------------------------------------------
3992 // Find a free slot in the monitor block.
3993 Label Lfound, Lexit, Lallocate_new;
3994 ConditionRegister found_free_slot = CCR0,
3995 found_same_obj = CCR1,
3996 reached_limit = CCR6;
3997 {
3998 Label Lloop, Lentry;
3999 Register Rlimit = Rcurrent_monitor;
4001 // Set up search loop - start with topmost monitor.
4002 __ add(Rcurrent_obj_addr, BasicObjectLock::obj_offset_in_bytes(), R26_monitor);
4004 __ ld(Rlimit, 0, R1_SP);
4005 __ addi(Rlimit, Rlimit, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes() - BasicObjectLock::obj_offset_in_bytes())); // Monitor base
4007 // Check if any slot is present => short cut to allocation if not.
4008 __ cmpld(reached_limit, Rcurrent_obj_addr, Rlimit);
4009 __ bgt(reached_limit, Lallocate_new);
4011 // Pre-load topmost slot.
4012 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4013 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4014 // The search loop.
4015 __ bind(Lloop);
4016 // Found free slot?
4017 __ cmpdi(found_free_slot, Rcurrent_obj, 0);
4018 // Is this entry for same obj? If so, stop the search and take the found
4019 // free slot or allocate a new one to enable recursive locking.
4020 __ cmpd(found_same_obj, Rcurrent_obj, Robj_to_lock);
4021 __ cmpld(reached_limit, Rcurrent_obj_addr, Rlimit);
4022 __ beq(found_free_slot, Lexit);
4023 __ beq(found_same_obj, Lallocate_new);
4024 __ bgt(reached_limit, Lallocate_new);
4025 // Check if last allocated BasicLockObj reached.
4026 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4027 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4028 // Next iteration if unchecked BasicObjectLocks exist on the stack.
4029 __ b(Lloop);
4030 }
4032 // ------------------------------------------------------------------------------
4033 // Check if we found a free slot.
4034 __ bind(Lexit);
4036 __ addi(Rcurrent_monitor, Rcurrent_obj_addr, -(frame::interpreter_frame_monitor_size() * wordSize) - BasicObjectLock::obj_offset_in_bytes());
4037 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, - frame::interpreter_frame_monitor_size() * wordSize);
4038 __ b(Lfound);
4040 // We didn't find a free BasicObjLock => allocate one.
4041 __ align(32, 12);
4042 __ bind(Lallocate_new);
4043 __ add_monitor_to_stack(false, Rscratch1, Rscratch2);
4044 __ mr(Rcurrent_monitor, R26_monitor);
4045 __ addi(Rcurrent_obj_addr, R26_monitor, BasicObjectLock::obj_offset_in_bytes());
4047 // ------------------------------------------------------------------------------
4048 // We now have a slot to lock.
4049 __ bind(Lfound);
4051 // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
4052 // The object has already been poped from the stack, so the expression stack looks correct.
4053 __ addi(R14_bcp, R14_bcp, 1);
4055 __ std(Robj_to_lock, 0, Rcurrent_obj_addr);
4056 __ lock_object(Rcurrent_monitor, Robj_to_lock);
4058 // Check if there's enough space on the stack for the monitors after locking.
4059 Label Lskip_stack_check;
4060 // Optimization: If the monitors stack section is less then a std page size (4K) don't run
4061 // the stack check. There should be enough shadow pages to fit that in.
4062 __ ld(Rscratch3, 0, R1_SP);
4063 __ sub(Rscratch3, Rscratch3, R26_monitor);
4064 __ cmpdi(CCR0, Rscratch3, 4*K);
4065 __ blt(CCR0, Lskip_stack_check);
4067 DEBUG_ONLY(__ untested("stack overflow check during monitor enter");)
4068 __ li(Rscratch1, 0);
4069 __ generate_stack_overflow_check_with_compare_and_throw(Rscratch1, Rscratch2);
4071 __ align(32, 12);
4072 __ bind(Lskip_stack_check);
4074 // The bcp has already been incremented. Just need to dispatch to next instruction.
4075 __ dispatch_next(vtos);
4076 }
4078 void TemplateTable::monitorexit() {
4079 transition(atos, vtos);
4080 __ verify_oop(R17_tos);
4082 Register Rcurrent_monitor = R11_scratch1,
4083 Rcurrent_obj = R12_scratch2,
4084 Robj_to_lock = R17_tos,
4085 Rcurrent_obj_addr = R3_ARG1,
4086 Rlimit = R4_ARG2;
4087 Label Lfound, Lillegal_monitor_state;
4089 // Check corner case: unbalanced monitorEnter / Exit.
4090 __ ld(Rlimit, 0, R1_SP);
4091 __ addi(Rlimit, Rlimit, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base
4093 // Null pointer check.
4094 __ null_check_throw(Robj_to_lock, -1, R11_scratch1);
4096 __ cmpld(CCR0, R26_monitor, Rlimit);
4097 __ bgt(CCR0, Lillegal_monitor_state);
4099 // Find the corresponding slot in the monitors stack section.
4100 {
4101 Label Lloop;
4103 // Start with topmost monitor.
4104 __ addi(Rcurrent_obj_addr, R26_monitor, BasicObjectLock::obj_offset_in_bytes());
4105 __ addi(Rlimit, Rlimit, BasicObjectLock::obj_offset_in_bytes());
4106 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4107 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4109 __ bind(Lloop);
4110 // Is this entry for same obj?
4111 __ cmpd(CCR0, Rcurrent_obj, Robj_to_lock);
4112 __ beq(CCR0, Lfound);
4114 // Check if last allocated BasicLockObj reached.
4116 __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4117 __ cmpld(CCR0, Rcurrent_obj_addr, Rlimit);
4118 __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4120 // Next iteration if unchecked BasicObjectLocks exist on the stack.
4121 __ ble(CCR0, Lloop);
4122 }
4124 // Fell through without finding the basic obj lock => throw up!
4125 __ bind(Lillegal_monitor_state);
4126 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
4127 __ should_not_reach_here();
4129 __ align(32, 12);
4130 __ bind(Lfound);
4131 __ addi(Rcurrent_monitor, Rcurrent_obj_addr,
4132 -(frame::interpreter_frame_monitor_size() * wordSize) - BasicObjectLock::obj_offset_in_bytes());
4133 __ unlock_object(Rcurrent_monitor);
4134 }
4136 // ============================================================================
4137 // Wide bytecodes
4139 // Wide instructions. Simply redirects to the wide entry point for that instruction.
4140 void TemplateTable::wide() {
4141 transition(vtos, vtos);
4143 const Register Rtable = R11_scratch1,
4144 Rindex = R12_scratch2,
4145 Rtmp = R0;
4147 __ lbz(Rindex, 1, R14_bcp);
4149 __ load_dispatch_table(Rtable, Interpreter::_wentry_point);
4151 __ slwi(Rindex, Rindex, LogBytesPerWord);
4152 __ ldx(Rtmp, Rtable, Rindex);
4153 __ mtctr(Rtmp);
4154 __ bctr();
4155 // Note: the bcp increment step is part of the individual wide bytecode implementations.
4156 }
4157 #endif // !CC_INTERP