Fri, 22 Sep 2017 14:09:57 +0800
[Interpreter] Optimize TemplateTable::xaload.
1 /*
2 * Copyright (c) 2003, 2013, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2015, 2016, Loongson Technology. 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.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "interpreter/interpreterRuntime.hpp"
30 #include "interpreter/templateTable.hpp"
31 #include "memory/universe.inline.hpp"
32 #include "oops/methodData.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"
41 #ifndef CC_INTERP
43 #define __ _masm->
45 // Platform-dependent initialization
47 void TemplateTable::pd_initialize() {
48 // No mips specific initialization
49 }
51 // Address computation: local variables
53 static inline Address iaddress(int n) {
54 return Address(LVP, Interpreter::local_offset_in_bytes(n));
55 }
57 static inline Address laddress(int n) {
58 return iaddress(n + 1);
59 }
61 static inline Address faddress(int n) {
62 return iaddress(n);
63 }
65 static inline Address daddress(int n) {
66 return laddress(n);
67 }
69 static inline Address aaddress(int n) {
70 return iaddress(n);
71 }
72 static inline Address haddress(int n) { return iaddress(n + 0); }
75 static inline Address at_sp() { return Address(SP, 0); }
76 static inline Address at_sp_p1() { return Address(SP, 1 * wordSize); }
77 static inline Address at_sp_p2() { return Address(SP, 2 * wordSize); }
79 // At top of Java expression stack which may be different than esp(). It
80 // isn't for category 1 objects.
81 static inline Address at_tos () {
82 Address tos = Address(SP, Interpreter::expr_offset_in_bytes(0));
83 return tos;
84 }
86 static inline Address at_tos_p1() {
87 return Address(SP, Interpreter::expr_offset_in_bytes(1));
88 }
90 static inline Address at_tos_p2() {
91 return Address(SP, Interpreter::expr_offset_in_bytes(2));
92 }
94 static inline Address at_tos_p3() {
95 return Address(SP, Interpreter::expr_offset_in_bytes(3));
96 }
98 // we use S0 as bcp, be sure you have bcp in S0 before you call any of the Template generator
99 Address TemplateTable::at_bcp(int offset) {
100 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
101 return Address(BCP, offset);
102 }
104 // bytecode folding
105 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
106 Register tmp_reg, bool load_bc_into_bc_reg/*=true*/,
107 int byte_no) {
108 if (!RewriteBytecodes) return;
109 Label L_patch_done;
111 switch (bc) {
112 case Bytecodes::_fast_aputfield:
113 case Bytecodes::_fast_bputfield:
114 case Bytecodes::_fast_cputfield:
115 case Bytecodes::_fast_dputfield:
116 case Bytecodes::_fast_fputfield:
117 case Bytecodes::_fast_iputfield:
118 case Bytecodes::_fast_lputfield:
119 case Bytecodes::_fast_sputfield:
120 {
121 // We skip bytecode quickening for putfield instructions when
122 // the put_code written to the constant pool cache is zero.
123 // This is required so that every execution of this instruction
124 // calls out to InterpreterRuntime::resolve_get_put to do
125 // additional, required work.
126 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
127 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
128 __ get_cache_and_index_and_bytecode_at_bcp(tmp_reg, bc_reg, tmp_reg, byte_no, 1);
129 __ daddi(bc_reg, R0, bc);
130 __ beq(tmp_reg, R0, L_patch_done);
131 __ delayed()->nop();
132 }
133 break;
134 default:
135 assert(byte_no == -1, "sanity");
136 // the pair bytecodes have already done the load.
137 if (load_bc_into_bc_reg) {
138 __ move(bc_reg, bc);
139 }
140 }
142 if (JvmtiExport::can_post_breakpoint()) {
143 Label L_fast_patch;
144 // if a breakpoint is present we can't rewrite the stream directly
145 __ lbu(tmp_reg, at_bcp(0));
146 __ move(AT, Bytecodes::_breakpoint);
147 __ bne(tmp_reg, AT, L_fast_patch);
148 __ delayed()->nop();
150 __ get_method(tmp_reg);
151 // Let breakpoint table handling rewrite to quicker bytecode
152 __ call_VM(NOREG, CAST_FROM_FN_PTR(address,
153 InterpreterRuntime::set_original_bytecode_at), tmp_reg, BCP, bc_reg);
155 __ b(L_patch_done);
156 __ delayed()->nop();
157 __ bind(L_fast_patch);
158 }
160 #ifdef ASSERT
161 Label L_okay;
162 __ lbu(tmp_reg, at_bcp(0));
163 __ move(AT, (int)Bytecodes::java_code(bc));
164 __ beq(tmp_reg, AT, L_okay);
165 __ delayed()->nop();
166 __ beq(tmp_reg, bc_reg, L_patch_done);
167 __ delayed()->nop();
168 __ stop("patching the wrong bytecode");
169 __ bind(L_okay);
170 #endif
172 // patch bytecode
173 __ sb(bc_reg, at_bcp(0));
174 __ bind(L_patch_done);
175 }
178 // Individual instructions
180 void TemplateTable::nop() {
181 transition(vtos, vtos);
182 // nothing to do
183 }
185 void TemplateTable::shouldnotreachhere() {
186 transition(vtos, vtos);
187 __ stop("shouldnotreachhere bytecode");
188 }
190 void TemplateTable::aconst_null() {
191 transition(vtos, atos);
192 __ move(FSR, R0);
193 }
195 void TemplateTable::iconst(int value) {
196 transition(vtos, itos);
197 if (value == 0) {
198 __ move(FSR, R0);
199 } else {
200 __ move(FSR, value);
201 }
202 }
204 void TemplateTable::lconst(int value) {
205 transition(vtos, ltos);
206 if (value == 0) {
207 __ move(FSR, R0);
208 } else {
209 __ move(FSR, value);
210 }
211 }
213 void TemplateTable::fconst(int value) {
214 transition(vtos, ftos);
215 switch( value ) {
216 case 0: __ mtc1(R0, FSF); return;
217 case 1: __ addiu(AT, R0, 1); break;
218 case 2: __ addiu(AT, R0, 2); break;
219 default: ShouldNotReachHere();
220 }
221 __ mtc1(AT, FSF);
222 __ cvt_s_w(FSF, FSF);
223 }
225 void TemplateTable::dconst(int value) {
226 transition(vtos, dtos);
227 switch( value ) {
228 case 0: __ dmtc1(R0, FSF);
229 return;
230 case 1: __ daddiu(AT, R0, 1);
231 __ dmtc1(AT, FSF);
232 __ cvt_d_w(FSF, FSF);
233 break;
234 default: ShouldNotReachHere();
235 }
236 }
238 void TemplateTable::bipush() {
239 transition(vtos, itos);
240 __ lb(FSR, at_bcp(1));
241 }
243 void TemplateTable::sipush() {
244 transition(vtos, itos);
245 __ lb(FSR, BCP, 1);
246 __ lbu(AT, BCP, 2);
247 __ dsll(FSR, FSR, 8);
248 __ orr(FSR, FSR, AT);
249 }
251 // T1 : tags
252 // T2 : index
253 // T3 : cpool
254 // T8 : tag
255 void TemplateTable::ldc(bool wide) {
256 transition(vtos, vtos);
257 Label call_ldc, notFloat, notClass, Done;
258 // get index in cpool
259 if (wide) {
260 __ get_unsigned_2_byte_index_at_bcp(T2, 1);
261 } else {
262 __ lbu(T2, at_bcp(1));
263 }
265 __ get_cpool_and_tags(T3, T1);
267 const int base_offset = ConstantPool::header_size() * wordSize;
268 const int tags_offset = Array<u1>::base_offset_in_bytes();
270 // get type
271 if (UseLoongsonISA && Assembler::is_simm(sizeof(tags_offset), 8)) {
272 __ gslbx(T1, T1, T2, tags_offset);
273 } else {
274 __ dadd(AT, T1, T2);
275 __ lb(T1, AT, tags_offset);
276 }
277 //now T1 is the tag
279 // unresolved class - get the resolved class
280 __ daddiu(AT, T1, - JVM_CONSTANT_UnresolvedClass);
281 __ beq(AT, R0, call_ldc);
282 __ delayed()->nop();
284 // unresolved class in error (resolution failed) - call into runtime
285 // so that the same error from first resolution attempt is thrown.
286 __ daddiu(AT, T1, -JVM_CONSTANT_UnresolvedClassInError);
287 __ beq(AT, R0, call_ldc);
288 __ delayed()->nop();
290 // resolved class - need to call vm to get java mirror of the class
291 __ daddiu(AT, T1, - JVM_CONSTANT_Class);
292 __ bne(AT, R0, notClass);
293 __ delayed()->dsll(T2, T2, Address::times_8);
295 __ bind(call_ldc);
296 __ move(A1, wide);
297 call_VM(FSR, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), A1);
298 //__ push(atos);
299 __ sd(FSR, SP, - Interpreter::stackElementSize);
300 __ b(Done);
301 __ delayed()->daddiu(SP, SP, - Interpreter::stackElementSize);
302 __ nop(); // added for performance issue
304 __ bind(notClass);
305 __ daddiu(AT, T1, -JVM_CONSTANT_Float);
306 __ bne(AT, R0, notFloat);
307 __ delayed()->nop();
308 // ftos
309 if (UseLoongsonISA && Assembler::is_simm(sizeof(base_offset), 8)) {
310 __ gslwxc1(FSF, T3, T2, base_offset);
311 } else {
312 __ dadd(AT, T3, T2);
313 __ lwc1(FSF, AT, base_offset);
314 }
315 //__ push_f();
316 __ swc1(FSF, SP, - Interpreter::stackElementSize);
317 __ b(Done);
318 __ delayed()->daddiu(SP, SP, - Interpreter::stackElementSize);
320 __ bind(notFloat);
321 #ifdef ASSERT
322 {
323 Label L;
324 __ daddiu(AT, T1, -JVM_CONSTANT_Integer);
325 __ beq(AT, R0, L);
326 __ delayed()->nop();
327 __ stop("unexpected tag type in ldc");
328 __ bind(L);
329 }
330 #endif
331 // itos JVM_CONSTANT_Integer only
332 if (UseLoongsonISA && Assembler::is_simm(sizeof(base_offset), 8)) {
333 __ gslwx(FSR, T3, T2, base_offset);
334 } else {
335 __ dadd(T0, T3, T2);
336 __ lw(FSR, T0, base_offset);
337 }
338 __ push(itos);
339 __ bind(Done);
340 }
342 // Fast path for caching oop constants.
343 void TemplateTable::fast_aldc(bool wide) {
344 transition(vtos, atos);
346 Register result = FSR;
347 Register tmp = SSR;
348 int index_size = wide ? sizeof(u2) : sizeof(u1);
350 Label resolved;
352 // We are resolved if the resolved reference cache entry contains a
353 // non-null object (String, MethodType, etc.)
354 assert_different_registers(result, tmp);
355 __ get_cache_index_at_bcp(tmp, 1, index_size);
356 __ load_resolved_reference_at_index(result, tmp);
357 __ bne(result, R0, resolved);
358 __ delayed()->nop();
360 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
361 // first time invocation - must resolve first
362 int i = (int)bytecode();
363 __ move(tmp, i);
364 __ call_VM(result, entry, tmp);
366 __ bind(resolved);
368 if (VerifyOops) {
369 __ verify_oop(result);
370 }
371 }
374 // used register: T2, T3, T1
375 // T2 : index
376 // T3 : cpool
377 // T1 : tag
378 void TemplateTable::ldc2_w() {
379 transition(vtos, vtos);
380 Label Long, Done;
382 // get index in cpool
383 __ get_unsigned_2_byte_index_at_bcp(T2, 1);
385 __ get_cpool_and_tags(T3, T1);
387 const int base_offset = ConstantPool::header_size() * wordSize;
388 const int tags_offset = Array<u1>::base_offset_in_bytes();
390 // get type in T1
391 if (UseLoongsonISA && Assembler::is_simm(tags_offset, 8)) {
392 __ gslbx(T1, T1, T2, tags_offset);
393 } else {
394 __ dadd(AT, T1, T2);
395 __ lb(T1, AT, tags_offset);
396 }
398 __ daddiu(AT, T1, - JVM_CONSTANT_Double);
399 __ bne(AT, R0, Long);
400 __ delayed()->dsll(T2, T2, Address::times_8);
402 // dtos
403 if (UseLoongsonISA && Assembler::is_simm(base_offset, 8)) {
404 __ gsldxc1(FSF, T3, T2, base_offset);
405 } else {
406 __ daddu(AT, T3, T2);
407 __ ldc1(FSF, AT, base_offset);
408 }
409 __ sdc1(FSF, SP, - 2 * wordSize);
410 __ b(Done);
411 __ delayed()->daddi(SP, SP, - 2 * wordSize);
413 // ltos
414 __ bind(Long);
415 if (UseLoongsonISA && Assembler::is_simm(base_offset, 8)) {
416 __ gsldx(FSR, T3, T2, base_offset);
417 } else {
418 __ dadd(AT, T3, T2);
419 __ ld(FSR, AT, base_offset);
420 }
421 __ push(ltos);
423 __ bind(Done);
424 }
426 // we compute the actual local variable address here
427 // the x86 dont do so for it has scaled index memory access model, we dont have, so do here
428 void TemplateTable::locals_index(Register reg, int offset) {
429 __ lbu(reg, at_bcp(offset));
430 __ dsll(reg, reg, Address::times_8);
431 __ dsub(reg, LVP, reg);
432 }
434 // this method will do bytecode folding of the two form:
435 // iload iload iload caload
436 // used register : T2, T3
437 // T2 : bytecode
438 // T3 : folded code
439 void TemplateTable::iload() {
440 transition(vtos, itos);
441 if (RewriteFrequentPairs) {
442 Label rewrite, done;
443 // get the next bytecode in T2
444 __ lbu(T2, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
445 // if _iload, wait to rewrite to iload2. We only want to rewrite the
446 // last two iloads in a pair. Comparing against fast_iload means that
447 // the next bytecode is neither an iload or a caload, and therefore
448 // an iload pair.
449 __ move(AT, Bytecodes::_iload);
450 __ beq(AT, T2, done);
451 __ delayed()->nop();
453 __ move(T3, Bytecodes::_fast_iload2);
454 __ move(AT, Bytecodes::_fast_iload);
455 __ beq(AT, T2, rewrite);
456 __ delayed()->nop();
458 // if _caload, rewrite to fast_icaload
459 __ move(T3, Bytecodes::_fast_icaload);
460 __ move(AT, Bytecodes::_caload);
461 __ beq(AT, T2, rewrite);
462 __ delayed()->nop();
464 // rewrite so iload doesn't check again.
465 __ move(T3, Bytecodes::_fast_iload);
467 // rewrite
468 // T3 : fast bytecode
469 __ bind(rewrite);
470 patch_bytecode(Bytecodes::_iload, T3, T2, false);
471 __ bind(done);
472 }
474 // Get the local value into tos
475 locals_index(T2);
476 __ lw(FSR, T2, 0);
477 }
479 // used register T2
480 // T2 : index
481 void TemplateTable::fast_iload2() {
482 transition(vtos, itos);
483 locals_index(T2);
484 __ lw(FSR, T2, 0);
485 __ push(itos);
486 locals_index(T2, 3);
487 __ lw(FSR, T2, 0);
488 }
490 // used register T2
491 // T2 : index
492 void TemplateTable::fast_iload() {
493 transition(vtos, itos);
494 locals_index(T2);
495 __ lw(FSR, T2, 0);
496 }
498 // used register T2
499 // T2 : index
500 void TemplateTable::lload() {
501 transition(vtos, ltos);
502 locals_index(T2);
503 __ ld(FSR, T2, -wordSize);
504 }
506 // used register T2
507 // T2 : index
508 void TemplateTable::fload() {
509 transition(vtos, ftos);
510 locals_index(T2);
511 __ lwc1(FSF, T2, 0);
512 }
514 // used register T2
515 // T2 : index
516 void TemplateTable::dload() {
517 transition(vtos, dtos);
518 locals_index(T2);
519 __ ldc1(FSF, T2, -wordSize);
520 }
522 // used register T2
523 // T2 : index
524 void TemplateTable::aload() {
525 transition(vtos, atos);
526 locals_index(T2);
527 __ ld(FSR, T2, 0);
528 }
530 void TemplateTable::locals_index_wide(Register reg) {
531 __ get_unsigned_2_byte_index_at_bcp(reg, 2);
532 __ dsll(reg, reg, Address::times_8);
533 __ dsub(reg, LVP, reg);
534 }
536 // used register T2
537 // T2 : index
538 void TemplateTable::wide_iload() {
539 transition(vtos, itos);
540 locals_index_wide(T2);
541 __ ld(FSR, T2, 0);
542 }
544 // used register T2
545 // T2 : index
546 void TemplateTable::wide_lload() {
547 transition(vtos, ltos);
548 locals_index_wide(T2);
549 __ ld(FSR, T2, -wordSize);
550 }
552 // used register T2
553 // T2 : index
554 void TemplateTable::wide_fload() {
555 transition(vtos, ftos);
556 locals_index_wide(T2);
557 __ lwc1(FSF, T2, 0);
558 }
560 // used register T2
561 // T2 : index
562 void TemplateTable::wide_dload() {
563 transition(vtos, dtos);
564 locals_index_wide(T2);
565 __ ldc1(FSF, T2, -wordSize);
566 }
568 // used register T2
569 // T2 : index
570 void TemplateTable::wide_aload() {
571 transition(vtos, atos);
572 locals_index_wide(T2);
573 __ ld(FSR, T2, 0);
574 }
576 // we use A2 as the regiser for index, BE CAREFUL!
577 // we dont use our tge 29 now, for later optimization
578 void TemplateTable::index_check(Register array, Register index) {
579 // Pop ptr into array
580 __ pop_ptr(array);
581 index_check_without_pop(array, index);
582 }
584 void TemplateTable::index_check_without_pop(Register array, Register index) {
585 // destroys ebx
586 // check array
587 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
589 #ifdef _LP64
590 // sign extend since tos (index) might contain garbage in upper bits
591 __ sll(index, index, 0);
592 #endif // _LP64
594 // check index
595 Label ok;
596 __ lw(AT, array, arrayOopDesc::length_offset_in_bytes());
597 #ifndef OPT_RANGECHECK
598 __ sltu(AT, index, AT);
599 __ bne(AT, R0, ok);
600 __ delayed()->nop();
602 //throw_ArrayIndexOutOfBoundsException assume abberrant index in A2
603 if (A2 != index) __ move(A2, index);
604 __ jmp(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
605 __ delayed()->nop();
606 __ bind(ok);
607 #else
608 __ lw(AT, array, arrayOopDesc::length_offset_in_bytes());
609 __ move(A2, index);
610 __ tgeu(A2, AT, 29);
611 #endif
612 }
614 void TemplateTable::iaload() {
615 transition(itos, itos);
616 if(UseBoundCheckInstruction) {
617 __ pop(SSR); //SSR:array FSR: index
618 __ dsll(FSR, FSR, 2);
619 __ dadd(FSR, SSR, FSR);
620 __ addi(FSR, FSR, arrayOopDesc::base_offset_in_bytes(T_INT));
622 __ lw(AT, SSR, arrayOopDesc::length_offset_in_bytes()); //bound
623 __ dsll(AT, AT, 2);
624 __ dadd(AT, SSR, AT);
625 __ addi(AT, AT, arrayOopDesc::base_offset_in_bytes(T_INT));
627 __ gslwle(FSR, FSR, AT);
628 } else {
629 index_check(SSR, FSR);
630 __ dsll(FSR, FSR, 2);
631 if (UseLoongsonISA && Assembler::is_simm(arrayOopDesc::base_offset_in_bytes(T_INT), 8)) {
632 __ gslwx(FSR, FSR, SSR, arrayOopDesc::base_offset_in_bytes(T_INT));
633 } else {
634 __ dadd(FSR, SSR, FSR);
635 __ lw(FSR, FSR, arrayOopDesc::base_offset_in_bytes(T_INT));
636 }
637 }
638 }
640 void TemplateTable::laload() {
641 transition(itos, ltos);
642 if(UseBoundCheckInstruction) {
643 __ pop(SSR); //SSR:array FSR: index
644 __ dsll(FSR, FSR, Address::times_8);
645 __ dadd(FSR, SSR, FSR);
646 __ addi(FSR, FSR, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize);
648 __ lw(AT, SSR, arrayOopDesc::length_offset_in_bytes()); //bound
649 __ dsll(AT, AT, Address::times_8);
650 __ dadd(AT, SSR, AT);
651 __ addi(AT, AT, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize);
653 __ gsldle(FSR, FSR, AT);
654 } else {
655 index_check(SSR, FSR);
656 __ dsll(AT, FSR, Address::times_8);
657 if (UseLoongsonISA && Assembler::is_simm(arrayOopDesc::base_offset_in_bytes(T_LONG), 8)) {
658 __ gsldx(FSR, SSR, AT, arrayOopDesc::base_offset_in_bytes(T_LONG));
659 } else {
660 __ dadd(AT, SSR, AT);
661 __ ld(FSR, AT, arrayOopDesc::base_offset_in_bytes(T_LONG));
662 }
663 }
664 }
666 void TemplateTable::faload() {
667 transition(itos, ftos);
668 if(UseBoundCheckInstruction) {
669 __ pop(SSR); //SSR:array FSR: index
670 __ shl(FSR, 2);
671 __ dadd(FSR, SSR, FSR);
672 __ addi(FSR, FSR, arrayOopDesc::base_offset_in_bytes(T_FLOAT));
674 __ lw(AT, SSR, arrayOopDesc::length_offset_in_bytes()); //bound
675 __ shl(AT, 2);
676 __ dadd(AT, SSR, AT);
677 __ addi(AT, AT, arrayOopDesc::base_offset_in_bytes(T_FLOAT));
679 __ gslwlec1(FSF, FSR, AT);
680 } else {
681 index_check(SSR, FSR);
682 __ shl(FSR, 2);
683 if (UseLoongsonISA && Assembler::is_simm(arrayOopDesc::base_offset_in_bytes(T_FLOAT), 8)) {
684 __ gslwxc1(FSF, SSR, FSR, arrayOopDesc::base_offset_in_bytes(T_FLOAT));
685 } else {
686 __ dadd(FSR, SSR, FSR);
687 __ lwc1(FSF, FSR, arrayOopDesc::base_offset_in_bytes(T_FLOAT));
688 }
689 }
690 }
692 void TemplateTable::daload() {
693 transition(itos, dtos);
694 if(UseBoundCheckInstruction) {
695 __ pop(SSR); //SSR:array FSR: index
696 __ dsll(FSR, FSR, 3);
697 __ dadd(FSR, SSR, FSR);
698 __ addi(FSR, FSR, arrayOopDesc::base_offset_in_bytes(T_DOUBLE) + 0 * wordSize);
700 __ lw(AT, SSR, arrayOopDesc::length_offset_in_bytes()); //bound
701 __ dsll(AT, AT, 3);
702 __ dadd(AT, SSR, AT);
703 __ addi(AT, AT, arrayOopDesc::base_offset_in_bytes(T_DOUBLE) + 0 * wordSize);
705 __ gsldlec1(FSF, FSR, AT);
706 } else {
707 index_check(SSR, FSR);
708 __ dsll(AT, FSR, 3);
709 if (UseLoongsonISA && Assembler::is_simm(arrayOopDesc::base_offset_in_bytes(T_DOUBLE), 8)) {
710 __ gsldxc1(FSF, SSR, AT, arrayOopDesc::base_offset_in_bytes(T_DOUBLE));
711 } else {
712 __ dadd(AT, SSR, AT);
713 __ ldc1(FSF, AT, arrayOopDesc::base_offset_in_bytes(T_DOUBLE));
714 }
715 }
716 }
718 void TemplateTable::aaload() {
719 transition(itos, atos);
720 index_check(SSR, FSR);
721 __ dsll(FSR, FSR, UseCompressedOops ? Address::times_4 : Address::times_8);
722 __ dadd(FSR, SSR, FSR);
723 //add for compressedoops
724 __ load_heap_oop(FSR, Address(FSR, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
725 }
727 void TemplateTable::baload() {
728 transition(itos, itos);
729 if(UseBoundCheckInstruction) {
730 __ pop(SSR); //SSR:array FSR:index
731 __ dadd(FSR, SSR, FSR);
732 __ addi(FSR, FSR, arrayOopDesc::base_offset_in_bytes(T_BYTE)); //base
734 __ lw(AT, SSR, arrayOopDesc::length_offset_in_bytes());
735 __ dadd(AT, SSR, AT);
736 __ addi(AT, AT, arrayOopDesc::base_offset_in_bytes(T_BYTE)); //bound
738 __ gslble(FSR, FSR, AT);
739 } else {
740 index_check(SSR, FSR);
741 if (UseLoongsonISA && Assembler::is_simm(arrayOopDesc::base_offset_in_bytes(T_BYTE), 8)) {
742 __ gslbx(FSR, SSR, FSR, arrayOopDesc::base_offset_in_bytes(T_BYTE));
743 } else {
744 __ dadd(FSR, SSR, FSR);
745 __ lb(FSR, FSR, arrayOopDesc::base_offset_in_bytes(T_BYTE));
746 }
747 }
748 }
750 void TemplateTable::caload() {
751 transition(itos, itos);
752 index_check(SSR, FSR);
753 __ dsll(FSR, FSR, Address::times_2);
754 __ dadd(FSR, SSR, FSR);
755 __ lhu(FSR, FSR, arrayOopDesc::base_offset_in_bytes(T_CHAR));
756 }
758 // iload followed by caload frequent pair
759 // used register : T2
760 // T2 : index
761 void TemplateTable::fast_icaload() {
762 transition(vtos, itos);
763 // load index out of locals
764 locals_index(T2);
765 __ lw(FSR, T2, 0);
766 index_check(SSR, FSR);
767 __ dsll(FSR, FSR, 1);
768 __ dadd(FSR, SSR, FSR);
769 __ lhu(FSR, FSR, arrayOopDesc::base_offset_in_bytes(T_CHAR));
770 }
772 void TemplateTable::saload() {
773 transition(itos, itos);
774 if(UseBoundCheckInstruction) {
775 __ pop(SSR); //SSR:array FSR: index
776 __ dsll(FSR, FSR, Address::times_2);
777 __ dadd(FSR, SSR, FSR);
778 __ addi(FSR, FSR, arrayOopDesc::base_offset_in_bytes(T_SHORT));
780 __ lw(AT, SSR, arrayOopDesc::length_offset_in_bytes()); //bound
781 __ dsll(AT, AT, Address::times_2);
782 __ dadd(AT, SSR, AT);
783 __ addi(AT, AT, arrayOopDesc::base_offset_in_bytes(T_SHORT));
785 __ gslhle(FSR, FSR, AT);
786 } else {
787 index_check(SSR, FSR);
788 __ dsll(FSR, FSR, Address::times_2);
789 if (UseLoongsonISA && Assembler::is_simm(arrayOopDesc::base_offset_in_bytes(T_SHORT), 8)) {
790 __ gslhx(FSR, SSR, FSR, arrayOopDesc::base_offset_in_bytes(T_SHORT));
791 } else {
792 __ dadd(FSR, SSR, FSR);
793 __ lh(FSR, FSR, arrayOopDesc::base_offset_in_bytes(T_SHORT));
794 }
795 }
796 }
798 void TemplateTable::iload(int n) {
799 transition(vtos, itos);
800 __ lw(FSR, iaddress(n));
801 }
803 void TemplateTable::lload(int n) {
804 transition(vtos, ltos);
805 __ ld(FSR, laddress(n));
806 }
808 void TemplateTable::fload(int n) {
809 transition(vtos, ftos);
810 __ lwc1(FSF, faddress(n));
811 }
813 void TemplateTable::dload(int n) {
814 transition(vtos, dtos);
815 __ ldc1(FSF, laddress(n));
816 }
818 void TemplateTable::aload(int n) {
819 transition(vtos, atos);
820 __ ld(FSR, aaddress(n));
821 }
823 // used register : T2, T3
824 // T2 : bytecode
825 // T3 : folded code
826 void TemplateTable::aload_0() {
827 transition(vtos, atos);
828 // According to bytecode histograms, the pairs:
829 //
830 // _aload_0, _fast_igetfield
831 // _aload_0, _fast_agetfield
832 // _aload_0, _fast_fgetfield
833 //
834 // occur frequently. If RewriteFrequentPairs is set, the (slow)
835 // _aload_0 bytecode checks if the next bytecode is either
836 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
837 // rewrites the current bytecode into a pair bytecode; otherwise it
838 // rewrites the current bytecode into _fast_aload_0 that doesn't do
839 // the pair check anymore.
840 //
841 // Note: If the next bytecode is _getfield, the rewrite must be
842 // delayed, otherwise we may miss an opportunity for a pair.
843 //
844 // Also rewrite frequent pairs
845 // aload_0, aload_1
846 // aload_0, iload_1
847 // These bytecodes with a small amount of code are most profitable
848 // to rewrite
849 if (RewriteFrequentPairs) {
850 Label rewrite, done;
851 // get the next bytecode in T2
852 __ lbu(T2, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
854 // do actual aload_0
855 aload(0);
857 // if _getfield then wait with rewrite
858 __ move(AT, Bytecodes::_getfield);
859 __ beq(AT, T2, done);
860 __ delayed()->nop();
862 // if _igetfield then reqrite to _fast_iaccess_0
863 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) ==
864 Bytecodes::_aload_0,
865 "fix bytecode definition");
866 __ move(T3, Bytecodes::_fast_iaccess_0);
867 __ move(AT, Bytecodes::_fast_igetfield);
868 __ beq(AT, T2, rewrite);
869 __ delayed()->nop();
871 // if _agetfield then reqrite to _fast_aaccess_0
872 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) ==
873 Bytecodes::_aload_0,
874 "fix bytecode definition");
875 __ move(T3, Bytecodes::_fast_aaccess_0);
876 __ move(AT, Bytecodes::_fast_agetfield);
877 __ beq(AT, T2, rewrite);
878 __ delayed()->nop();
880 // if _fgetfield then reqrite to _fast_faccess_0
881 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) ==
882 Bytecodes::_aload_0,
883 "fix bytecode definition");
884 __ move(T3, Bytecodes::_fast_faccess_0);
885 __ move(AT, Bytecodes::_fast_fgetfield);
886 __ beq(AT, T2, rewrite);
887 __ delayed()->nop();
889 // else rewrite to _fast_aload0
890 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) ==
891 Bytecodes::_aload_0,
892 "fix bytecode definition");
893 __ move(T3, Bytecodes::_fast_aload_0);
895 // rewrite
896 __ bind(rewrite);
897 patch_bytecode(Bytecodes::_aload_0, T3, T2, false);
899 __ bind(done);
900 } else {
901 aload(0);
902 }
903 }
905 void TemplateTable::istore() {
906 transition(itos, vtos);
907 locals_index(T2);
908 __ sw(FSR, T2, 0);
909 }
911 void TemplateTable::lstore() {
912 transition(ltos, vtos);
913 locals_index(T2);
914 __ sd(FSR, T2, -wordSize);
915 }
917 void TemplateTable::fstore() {
918 transition(ftos, vtos);
919 locals_index(T2);
920 __ swc1(FSF, T2, 0);
921 }
923 void TemplateTable::dstore() {
924 transition(dtos, vtos);
925 locals_index(T2);
926 __ sdc1(FSF, T2, -wordSize);
927 }
929 void TemplateTable::astore() {
930 transition(vtos, vtos);
931 __ pop_ptr(FSR);
932 locals_index(T2);
933 __ sd(FSR, T2, 0);
934 }
936 void TemplateTable::wide_istore() {
937 transition(vtos, vtos);
938 __ pop_i(FSR);
939 locals_index_wide(T2);
940 __ sd(FSR, T2, 0);
941 }
943 void TemplateTable::wide_lstore() {
944 transition(vtos, vtos);
945 __ pop_l(FSR);
946 locals_index_wide(T2);
947 __ sd(FSR, T2, -wordSize);
948 }
950 void TemplateTable::wide_fstore() {
951 wide_istore();
952 }
954 void TemplateTable::wide_dstore() {
955 wide_lstore();
956 }
958 void TemplateTable::wide_astore() {
959 transition(vtos, vtos);
960 __ pop_ptr(FSR);
961 locals_index_wide(T2);
962 __ sd(FSR, T2, 0);
963 }
965 // used register : T2
966 void TemplateTable::iastore() {
967 transition(itos, vtos);
968 __ pop_i(SSR); // T2: array SSR: index
969 if(UseBoundCheckInstruction) {
970 __ pop_ptr(T2);
971 __ dsll(SSR, SSR, Address::times_4);
972 __ dadd(SSR, T2, SSR);
973 __ addi(SSR, SSR, arrayOopDesc::base_offset_in_bytes(T_INT)); // base
975 __ lw(AT, T2, arrayOopDesc::length_offset_in_bytes());
976 __ dsll(AT, AT, Address::times_4);
977 __ dadd(AT, T2, AT);
978 __ addi(AT, AT, arrayOopDesc::base_offset_in_bytes(T_INT)); //bound
980 __ gsswle(FSR, SSR, AT);
981 } else {
982 index_check(T2, SSR); // prefer index in ebx
983 __ dsll(SSR, SSR, Address::times_4);
984 __ dadd(T2, T2, SSR);
985 __ sw(FSR, T2, arrayOopDesc::base_offset_in_bytes(T_INT));
986 }
987 }
991 // used register T2, T3
992 void TemplateTable::lastore() {
993 transition(ltos, vtos);
994 __ pop_i (T2);
995 if(UseBoundCheckInstruction) {
996 __ pop_ptr(T3);
997 __ dsll(T2, T2, Address::times_8);
998 __ dadd(T2, T3, T2);
999 __ addi(T2, T2, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize); // base
1001 __ lw(AT, T3, arrayOopDesc::length_offset_in_bytes());
1002 __ dsll(AT, AT, Address::times_8);
1003 __ dadd(AT, T3, AT);
1004 __ addi(AT, AT, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize); //bound
1006 __ gssdle(FSR, T2, AT);
1007 } else {
1008 index_check(T3, T2);
1009 __ dsll(T2, T2, Address::times_8);
1010 __ dadd(T3, T3, T2);
1011 __ sd(FSR, T3, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize);
1012 }
1013 }
1015 // used register T2
1016 void TemplateTable::fastore() {
1017 transition(ftos, vtos);
1018 __ pop_i(SSR);
1019 if(UseBoundCheckInstruction) {
1020 __ pop_ptr(T2);
1021 __ dsll(SSR, SSR, Address::times_4);
1022 __ dadd(SSR, T2, SSR);
1023 __ addi(SSR, SSR, arrayOopDesc::base_offset_in_bytes(T_FLOAT)); // base
1025 __ lw(AT, T2, arrayOopDesc::length_offset_in_bytes());
1026 __ dsll(AT, AT, Address::times_4);
1027 __ dadd(AT, T2, AT);
1028 __ addi(AT, AT, arrayOopDesc::base_offset_in_bytes(T_FLOAT)); //bound
1030 __ gsswlec1(FSF, SSR, AT);
1031 } else {
1032 index_check(T2, SSR);
1033 __ dsll(SSR, SSR, Address::times_4);
1034 __ dadd(T2, T2, SSR);
1035 __ swc1(FSF, T2, arrayOopDesc::base_offset_in_bytes(T_FLOAT));
1036 }
1037 }
1039 // used register T2, T3
1040 void TemplateTable::dastore() {
1041 transition(dtos, vtos);
1042 __ pop_i (T2);
1043 if(UseBoundCheckInstruction) {
1044 __ pop_ptr(T3);
1045 __ dsll(T2, T2, Address::times_8);
1046 __ dadd(T2, T3, T2);
1047 __ addi(T2, T2, arrayOopDesc::base_offset_in_bytes(T_DOUBLE) + 0 * wordSize); // base
1049 __ lw(AT, T3, arrayOopDesc::length_offset_in_bytes());
1050 __ dsll(AT, AT, Address::times_8);
1051 __ dadd(AT, T3, AT);
1052 __ addi(AT, AT, arrayOopDesc::base_offset_in_bytes(T_DOUBLE) + 0 * wordSize); //bound
1054 __ gssdlec1(FSF, T2, AT);
1055 } else {
1056 index_check(T3, T2);
1057 __ dsll(T2, T2, Address::times_8);
1058 __ daddu(T3, T3, T2);
1059 __ sdc1(FSF, T3, arrayOopDesc::base_offset_in_bytes(T_DOUBLE) + 0 * wordSize);
1060 }
1061 }
1063 // used register : T2, T3, T8
1064 // T2 : array
1065 // T3 : subklass
1066 // T8 : supklass
1067 void TemplateTable::aastore() {
1068 Label is_null, ok_is_subtype, done;
1069 transition(vtos, vtos);
1070 // stack: ..., array, index, value
1071 __ ld(FSR, at_tos()); // Value
1072 __ lw(SSR, at_tos_p1()); // Index
1073 __ ld(T2, at_tos_p2()); // Array
1075 // index_check(T2, SSR);
1076 index_check_without_pop(T2, SSR);
1077 // do array store check - check for NULL value first
1078 __ beq(FSR, R0, is_null);
1079 __ delayed()->nop();
1081 // Move subklass into T3
1082 //add for compressedoops
1083 __ load_klass(T3, FSR);
1084 // Move superklass into T8
1085 //add for compressedoops
1086 __ load_klass(T8, T2);
1087 __ ld(T8, Address(T8, ObjArrayKlass::element_klass_offset()));
1088 // Compress array+index*4+12 into a single register. T2
1089 __ dsll(AT, SSR, UseCompressedOops? Address::times_4 : Address::times_8);
1090 __ dadd(T2, T2, AT);
1091 __ daddi(T2, T2, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1093 // Generate subtype check.
1094 // Superklass in T8. Subklass in T3.
1095 __ gen_subtype_check(T8, T3, ok_is_subtype); // <-- Jin
1096 // Come here on failure
1097 // object is at FSR
1098 __ jmp(Interpreter::_throw_ArrayStoreException_entry); // <-- Jin
1099 __ delayed()->nop();
1100 // Come here on success
1101 __ bind(ok_is_subtype);
1102 //replace with do_oop_store->store_heap_oop
1103 __ store_heap_oop(Address(T2, 0), FSR); // <-- Jin
1104 __ store_check(T2);
1105 __ b(done);
1106 __ delayed()->nop();
1108 // Have a NULL in FSR, EDX=T2, SSR=index. Store NULL at ary[idx]
1109 __ bind(is_null);
1110 __ profile_null_seen(T9);
1111 __ dsll(AT, SSR, UseCompressedOops? Address::times_4 : Address::times_8);
1112 __ dadd(T2, T2, AT);
1113 __ store_heap_oop(Address(T2, arrayOopDesc::base_offset_in_bytes(T_OBJECT)), FSR); /* FSR is null here */
1115 __ bind(done);
1116 __ daddi(SP, SP, 3 * Interpreter::stackElementSize);
1117 }
1119 void TemplateTable::bastore() {
1120 transition(itos, vtos);
1121 __ pop_i(SSR);
1122 if(UseBoundCheckInstruction) {
1123 __ pop_ptr(T2);
1124 __ dadd(SSR, T2, SSR);
1125 __ addi(SSR, SSR, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // base
1127 __ lw(AT, T2, arrayOopDesc::length_offset_in_bytes());
1128 __ dadd(AT, T2, AT);
1129 __ addi(AT, AT, arrayOopDesc::base_offset_in_bytes(T_BYTE)); //bound
1131 __ gssble(FSR, SSR, AT);
1132 } else {
1133 index_check(T2, SSR);
1134 __ dadd(SSR, T2, SSR);
1135 __ sb(FSR, SSR, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1136 }
1137 }
1139 void TemplateTable::castore() {
1140 transition(itos, vtos);
1141 __ pop_i(SSR);
1142 if(UseBoundCheckInstruction) {
1143 __ pop_ptr(T2);
1144 __ dsll(SSR, SSR, Address::times_2);
1145 __ dadd(SSR, T2, SSR);
1146 __ addi(SSR, SSR, arrayOopDesc::base_offset_in_bytes(T_CHAR)); // base
1148 __ lw(AT, T2, arrayOopDesc::length_offset_in_bytes());
1149 __ dsll(AT, AT, Address::times_2);
1150 __ dadd(AT, T2, AT);
1151 __ addi(AT, AT, arrayOopDesc::base_offset_in_bytes(T_CHAR)); //bound
1153 __ gsshle(FSR, SSR, AT);
1154 } else {
1155 index_check(T2, SSR);
1156 __ dsll(SSR, SSR, Address::times_2);
1157 __ dadd(SSR, T2, SSR);
1158 __ sh(FSR, SSR, arrayOopDesc::base_offset_in_bytes(T_CHAR));
1159 }
1160 }
1162 void TemplateTable::sastore() {
1163 castore();
1164 }
1166 void TemplateTable::istore(int n) {
1167 transition(itos, vtos);
1168 __ sw(FSR, iaddress(n));
1169 }
1171 void TemplateTable::lstore(int n) {
1172 transition(ltos, vtos);
1173 __ sd(FSR, laddress(n));
1174 }
1176 void TemplateTable::fstore(int n) {
1177 transition(ftos, vtos);
1178 __ swc1(FSF, faddress(n));
1179 }
1181 void TemplateTable::dstore(int n) {
1182 transition(dtos, vtos);
1183 __ sdc1(FSF, laddress(n));
1184 }
1186 void TemplateTable::astore(int n) {
1187 transition(vtos, vtos);
1188 __ pop_ptr(FSR);
1189 __ sd(FSR, aaddress(n));
1190 }
1192 void TemplateTable::pop() {
1193 transition(vtos, vtos);
1194 __ daddi(SP, SP, Interpreter::stackElementSize);
1195 }
1197 void TemplateTable::pop2() {
1198 transition(vtos, vtos);
1199 __ daddi(SP, SP, 2 * Interpreter::stackElementSize);
1200 }
1202 void TemplateTable::dup() {
1203 transition(vtos, vtos);
1204 // stack: ..., a
1205 __ load_ptr(0, FSR);
1206 __ push_ptr(FSR);
1207 // stack: ..., a, a
1208 }
1210 // blows FSR
1211 void TemplateTable::dup_x1() {
1212 transition(vtos, vtos);
1213 // stack: ..., a, b
1214 __ load_ptr(0, FSR); // load b
1215 __ load_ptr(1, A5); // load a
1216 __ store_ptr(1, FSR); // store b
1217 __ store_ptr(0, A5); // store a
1218 __ push_ptr(FSR); // push b
1219 // stack: ..., b, a, b
1220 }
1222 // blows FSR
1223 void TemplateTable::dup_x2() {
1224 transition(vtos, vtos);
1225 // stack: ..., a, b, c
1226 __ load_ptr(0, FSR); // load c
1227 __ load_ptr(2, A5); // load a
1228 __ store_ptr(2, FSR); // store c in a
1229 __ push_ptr(FSR); // push c
1230 // stack: ..., c, b, c, c
1231 __ load_ptr(2, FSR); // load b
1232 __ store_ptr(2, A5); // store a in b
1233 // stack: ..., c, a, c, c
1234 __ store_ptr(1, FSR); // store b in c
1235 // stack: ..., c, a, b, c
1236 }
1238 // blows FSR
1239 void TemplateTable::dup2() {
1240 transition(vtos, vtos);
1241 // stack: ..., a, b
1242 __ load_ptr(1, FSR); // load a
1243 __ push_ptr(FSR); // push a
1244 __ load_ptr(1, FSR); // load b
1245 __ push_ptr(FSR); // push b
1246 // stack: ..., a, b, a, b
1247 }
1249 // blows FSR
1250 void TemplateTable::dup2_x1() {
1251 transition(vtos, vtos);
1252 // stack: ..., a, b, c
1253 __ load_ptr(0, T2); // load c
1254 __ load_ptr(1, FSR); // load b
1255 __ push_ptr(FSR); // push b
1256 __ push_ptr(T2); // push c
1257 // stack: ..., a, b, c, b, c
1258 __ store_ptr(3, T2); // store c in b
1259 // stack: ..., a, c, c, b, c
1260 __ load_ptr(4, T2); // load a
1261 __ store_ptr(2, T2); // store a in 2nd c
1262 // stack: ..., a, c, a, b, c
1263 __ store_ptr(4, FSR); // store b in a
1264 // stack: ..., b, c, a, b, c
1266 // stack: ..., b, c, a, b, c
1267 }
1269 // blows FSR, SSR
1270 void TemplateTable::dup2_x2() {
1271 transition(vtos, vtos);
1272 // stack: ..., a, b, c, d
1273 // stack: ..., a, b, c, d
1274 __ load_ptr(0, T2); // load d
1275 __ load_ptr(1, FSR); // load c
1276 __ push_ptr(FSR); // push c
1277 __ push_ptr(T2); // push d
1278 // stack: ..., a, b, c, d, c, d
1279 __ load_ptr(4, FSR); // load b
1280 __ store_ptr(2, FSR); // store b in d
1281 __ store_ptr(4, T2); // store d in b
1282 // stack: ..., a, d, c, b, c, d
1283 __ load_ptr(5, T2); // load a
1284 __ load_ptr(3, FSR); // load c
1285 __ store_ptr(3, T2); // store a in c
1286 __ store_ptr(5, FSR); // store c in a
1287 // stack: ..., c, d, a, b, c, d
1289 // stack: ..., c, d, a, b, c, d
1290 }
1292 // blows FSR
1293 void TemplateTable::swap() {
1294 transition(vtos, vtos);
1295 // stack: ..., a, b
1297 __ load_ptr(1, A5); // load a
1298 __ load_ptr(0, FSR); // load b
1299 __ store_ptr(0, A5); // store a in b
1300 __ store_ptr(1, FSR); // store b in a
1302 // stack: ..., b, a
1303 }
1305 void TemplateTable::iop2(Operation op) {
1306 transition(itos, itos);
1308 __ pop_i(SSR);
1309 switch (op) {
1310 case add : __ addu32(FSR, SSR, FSR); break;
1311 case sub : __ subu32(FSR, SSR, FSR); break;
1312 case mul : __ mul(FSR, SSR, FSR); break;
1313 case _and : __ andr(FSR, SSR, FSR); break;
1314 case _or : __ orr(FSR, SSR, FSR); break;
1315 case _xor : __ xorr(FSR, SSR, FSR); break;
1316 case shl : __ sllv(FSR, SSR, FSR); break; // implicit masking of lower 5 bits by Intel shift instr. mips also
1317 case shr : __ srav(FSR, SSR, FSR); break; // implicit masking of lower 5 bits by Intel shift instr. mips also
1318 case ushr : __ srlv(FSR, SSR, FSR); break; // implicit masking of lower 5 bits by Intel shift instr. mips also
1319 default : ShouldNotReachHere();
1320 }
1321 }
1323 // the result stored in FSR, SSR,
1324 // used registers : T2, T3
1325 void TemplateTable::lop2(Operation op) {
1326 transition(ltos, ltos);
1327 __ pop_l(T2, T3);
1328 #ifdef ASSERT
1329 {
1330 Label L;
1331 __ beq(T3, R0, L);
1332 __ delayed()->nop();
1333 __ bind(L);
1334 }
1335 #endif
1336 switch (op) {
1337 case add : __ daddu(FSR, T2, FSR); break;
1338 case sub : __ dsubu(FSR, T2, FSR); break;
1339 case _and: __ andr(FSR, T2, FSR); break;
1340 case _or : __ orr(FSR, T2, FSR); break;
1341 case _xor: __ xorr(FSR, T2, FSR); break;
1342 default : ShouldNotReachHere();
1343 }
1344 }
1346 // java require this bytecode could handle 0x80000000/-1, dont cause a overflow exception,
1347 // the result is 0x80000000
1348 // the godson2 cpu do the same, so we need not handle this specially like x86
1349 void TemplateTable::idiv() {
1350 transition(itos, itos);
1351 Label not_zero;
1353 __ bne(FSR, R0, not_zero);
1354 __ delayed()->nop();
1355 __ jmp(Interpreter::_throw_ArithmeticException_entry);
1356 __ delayed()->nop();
1357 __ bind(not_zero);
1359 __ pop_i(SSR);
1360 if (UseLoongsonISA) {
1361 __ gsdiv(FSR, SSR, FSR);
1362 } else {
1363 __ div(SSR, FSR);
1364 __ mflo(FSR);
1365 }
1366 }
1368 void TemplateTable::irem() {
1369 transition(itos, itos);
1370 Label not_zero;
1371 __ pop_i(SSR);
1372 __ div(SSR, FSR);
1374 __ bne(FSR, R0, not_zero);
1375 __ delayed()->nop();
1376 //__ brk(7);
1377 __ jmp(Interpreter::_throw_ArithmeticException_entry);
1378 __ delayed()->nop();
1380 __ bind(not_zero);
1381 __ mfhi(FSR);
1382 }
1384 void TemplateTable::lmul() {
1385 transition(ltos, ltos);
1386 __ pop_l(T2);
1387 if(UseLoongsonISA){
1388 __ gsdmult(FSR, T2, FSR);
1389 } else {
1390 __ dmult(T2, FSR);
1391 __ mflo(FSR);
1392 }
1393 }
1395 // NOTE: i DONT use the Interpreter::_throw_ArithmeticException_entry
1396 void TemplateTable::ldiv() {
1397 transition(ltos, ltos);
1398 Label normal;
1400 __ bne(FSR, R0, normal);
1401 __ delayed()->nop();
1403 //__ brk(7); //generate FPE
1404 __ jmp(Interpreter::_throw_ArithmeticException_entry);
1405 __ delayed()->nop();
1407 __ bind(normal);
1408 __ pop_l(A2, A3);
1409 if (UseLoongsonISA) {
1410 __ gsddiv(FSR, A2, FSR);
1411 } else {
1412 __ ddiv(A2, FSR);
1413 __ mflo(FSR);
1414 }
1415 }
1417 // NOTE: i DONT use the Interpreter::_throw_ArithmeticException_entry
1418 void TemplateTable::lrem() {
1419 transition(ltos, ltos);
1420 Label normal;
1422 __ bne(FSR, R0, normal);
1423 __ delayed()->nop();
1425 __ jmp(Interpreter::_throw_ArithmeticException_entry);
1426 __ delayed()->nop();
1428 __ bind(normal);
1429 __ pop_l (A2, A3);
1431 if(UseLoongsonISA){
1432 __ gsdmod(FSR, A2, FSR);
1433 } else {
1434 __ ddiv(A2, FSR);
1435 __ mfhi(FSR);
1436 }
1437 }
1439 // result in FSR
1440 // used registers : T0
1441 void TemplateTable::lshl() {
1442 transition(itos, ltos);
1443 __ pop_l(T0, T1);
1444 #ifdef ASSERT
1445 {
1446 Label L;
1447 __ beq(T1, R0, L);
1448 __ delayed()->nop();
1449 //__ stop("lshl, wrong stack"); // <-- Fu 20130930
1450 __ bind(L);
1451 }
1452 #endif
1453 __ dsllv(FSR, T0, FSR);
1454 }
1456 // used registers : T0
1457 void TemplateTable::lshr() {
1458 transition(itos, ltos);
1459 __ pop_l(T0, T1);
1460 #ifdef ASSERT
1461 {
1462 Label L;
1463 __ beq(T1, R0, L);
1464 __ delayed()->nop();
1465 __ stop("lshr, wrong stack");
1466 __ bind(L);
1467 }
1468 #endif
1469 __ dsrav(FSR, T0, FSR);
1470 }
1472 // used registers : T0
1473 void TemplateTable::lushr() {
1474 transition(itos, ltos);
1475 __ pop_l(T0, T1);
1476 #ifdef ASSERT
1477 {
1478 Label L;
1479 __ beq(T1, R0, L);
1480 __ delayed()->nop();
1481 __ stop("lushr, wrong stack");
1482 __ bind(L);
1483 }
1484 #endif
1485 __ dsrlv(FSR, T0, FSR);
1486 }
1488 // result in FSF
1489 void TemplateTable::fop2(Operation op) {
1490 transition(ftos, ftos);
1491 switch (op) {
1492 case add:
1493 __ lwc1(FTF, at_sp());
1494 __ add_s(FSF, FTF, FSF);
1495 break;
1496 case sub:
1497 __ lwc1(FTF, at_sp());
1498 __ sub_s(FSF, FTF, FSF);
1499 break;
1500 case mul:
1501 __ lwc1(FTF, at_sp());
1502 __ mul_s(FSF, FTF, FSF);
1503 break;
1504 case div:
1505 __ lwc1(FTF, at_sp());
1506 __ div_s(FSF, FTF, FSF);
1507 break;
1508 case rem:
1509 __ mov_s(F13, FSF);
1510 __ lwc1(F12, at_sp());
1511 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1512 break;
1513 default : ShouldNotReachHere();
1514 }
1516 __ daddi(SP, SP, 1 * wordSize);
1517 }
1519 // result in SSF||FSF
1520 // i dont handle the strict flags
1521 void TemplateTable::dop2(Operation op) {
1522 transition(dtos, dtos);
1523 switch (op) {
1524 case add:
1525 __ ldc1(FTF, at_sp());
1526 __ add_d(FSF, FTF, FSF);
1527 break;
1528 case sub:
1529 __ ldc1(FTF, at_sp());
1530 __ sub_d(FSF, FTF, FSF);
1531 break;
1532 case mul:
1533 __ ldc1(FTF, at_sp());
1534 __ mul_d(FSF, FTF, FSF);
1535 break;
1536 case div:
1537 __ ldc1(FTF, at_sp());
1538 __ div_d(FSF, FTF, FSF);
1539 break;
1540 case rem:
1541 __ mov_d(F13, FSF);
1542 __ ldc1(F12, at_sp());
1543 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1544 break;
1545 default : ShouldNotReachHere();
1546 }
1548 __ daddi(SP, SP, 2 * wordSize);
1549 }
1551 void TemplateTable::ineg() {
1552 transition(itos, itos);
1553 __ neg(FSR);
1554 }
1556 void TemplateTable::lneg() {
1557 transition(ltos, ltos);
1558 __ dsubu(FSR, R0, FSR);
1559 }
1561 void TemplateTable::fneg() {
1562 transition(ftos, ftos);
1563 __ neg_s(FSF, FSF);
1564 }
1566 void TemplateTable::dneg() {
1567 transition(dtos, dtos);
1568 __ neg_d(FSF, FSF);
1569 }
1571 // used registers : T2
1572 void TemplateTable::iinc() {
1573 transition(vtos, vtos);
1574 locals_index(T2);
1575 __ lw(FSR, T2, 0);
1576 __ lb(AT, at_bcp(2)); // get constant
1577 __ daddu(FSR, FSR, AT);
1578 __ sw(FSR, T2, 0);
1579 }
1581 // used register : T2
1582 void TemplateTable::wide_iinc() {
1583 transition(vtos, vtos);
1584 locals_index_wide(T2);
1585 __ get_2_byte_integer_at_bcp(FSR, AT, 4);
1586 __ hswap(FSR);
1587 __ lw(AT, T2, 0);
1588 __ daddu(FSR, AT, FSR);
1589 __ sw(FSR, T2, 0);
1590 }
1592 void TemplateTable::convert() {
1593 // Checking
1594 #ifdef ASSERT
1595 {
1596 TosState tos_in = ilgl;
1597 TosState tos_out = ilgl;
1598 switch (bytecode()) {
1599 case Bytecodes::_i2l: // fall through
1600 case Bytecodes::_i2f: // fall through
1601 case Bytecodes::_i2d: // fall through
1602 case Bytecodes::_i2b: // fall through
1603 case Bytecodes::_i2c: // fall through
1604 case Bytecodes::_i2s: tos_in = itos; break;
1605 case Bytecodes::_l2i: // fall through
1606 case Bytecodes::_l2f: // fall through
1607 case Bytecodes::_l2d: tos_in = ltos; break;
1608 case Bytecodes::_f2i: // fall through
1609 case Bytecodes::_f2l: // fall through
1610 case Bytecodes::_f2d: tos_in = ftos; break;
1611 case Bytecodes::_d2i: // fall through
1612 case Bytecodes::_d2l: // fall through
1613 case Bytecodes::_d2f: tos_in = dtos; break;
1614 default : ShouldNotReachHere();
1615 }
1616 switch (bytecode()) {
1617 case Bytecodes::_l2i: // fall through
1618 case Bytecodes::_f2i: // fall through
1619 case Bytecodes::_d2i: // fall through
1620 case Bytecodes::_i2b: // fall through
1621 case Bytecodes::_i2c: // fall through
1622 case Bytecodes::_i2s: tos_out = itos; break;
1623 case Bytecodes::_i2l: // fall through
1624 case Bytecodes::_f2l: // fall through
1625 case Bytecodes::_d2l: tos_out = ltos; break;
1626 case Bytecodes::_i2f: // fall through
1627 case Bytecodes::_l2f: // fall through
1628 case Bytecodes::_d2f: tos_out = ftos; break;
1629 case Bytecodes::_i2d: // fall through
1630 case Bytecodes::_l2d: // fall through
1631 case Bytecodes::_f2d: tos_out = dtos; break;
1632 default : ShouldNotReachHere();
1633 }
1634 transition(tos_in, tos_out);
1635 }
1636 #endif // ASSERT
1638 // Conversion
1639 // (Note: use pushl(ecx)/popl(ecx) for 1/2-word stack-ptr manipulation)
1640 switch (bytecode()) {
1641 case Bytecodes::_i2l:
1642 __ sll(FSR, FSR, 0);
1643 break;
1644 case Bytecodes::_i2f:
1645 __ mtc1(FSR, FSF);
1646 __ cvt_s_w(FSF, FSF);
1647 break;
1648 case Bytecodes::_i2d:
1649 __ mtc1(FSR, FSF);
1650 __ cvt_d_w(FSF, FSF);
1651 break;
1652 case Bytecodes::_i2b:
1653 __ seb(FSR, FSR);
1654 break;
1655 case Bytecodes::_i2c:
1656 __ andi(FSR, FSR, 0xFFFF); // truncate upper 56 bits
1657 break;
1658 case Bytecodes::_i2s:
1659 __ seh(FSR, FSR);
1660 break;
1661 case Bytecodes::_l2i:
1662 __ sll(FSR, FSR, 0);
1663 break;
1664 case Bytecodes::_l2f:
1665 __ dmtc1(FSR, FSF);
1666 __ cvt_s_l(FSF, FSF);
1667 break;
1668 case Bytecodes::_l2d:
1669 __ dmtc1(FSR, FSF);
1670 __ cvt_d_l(FSF, FSF);
1671 break;
1672 case Bytecodes::_f2i:
1673 {
1674 Label L;
1676 __ trunc_w_s(F12, FSF);
1677 __ move(AT, 0x7fffffff);
1678 __ mfc1(FSR, F12);
1679 __ c_un_s(FSF, FSF); //NaN?
1680 __ movt(FSR, R0);
1682 __ bne(AT, FSR, L);
1683 __ delayed()->lui(T9, 0x8000);
1685 __ mfc1(AT, FSF);
1686 __ andr(AT, AT, T9);
1688 __ movn(FSR, T9, AT);
1690 __ bind(L);
1691 }
1692 break;
1693 case Bytecodes::_f2l:
1694 {
1695 Label L;
1697 __ trunc_l_s(F12, FSF);
1698 __ daddiu(AT, R0, -1);
1699 __ dsrl(AT, AT, 1);
1700 __ dmfc1(FSR, F12);
1701 __ c_un_s(FSF, FSF); //NaN?
1702 __ movt(FSR, R0);
1704 __ bne(AT, FSR, L);
1705 __ delayed()->lui(T9, 0x8000);
1707 __ mfc1(AT, FSF);
1708 __ andr(AT, AT, T9);
1710 __ dsll32(T9, T9, 0);
1711 __ movn(FSR, T9, AT);
1713 __ bind(L);
1714 }
1715 break;
1716 case Bytecodes::_f2d:
1717 __ cvt_d_s(FSF, FSF);
1718 break;
1719 case Bytecodes::_d2i:
1720 {
1721 Label L;
1723 __ trunc_w_d(F12, FSF);
1724 __ move(AT, 0x7fffffff);
1725 __ mfc1(FSR, F12);
1727 __ bne(FSR, AT, L);
1728 __ delayed()->mtc1(R0, F12);
1730 __ cvt_d_w(F12, F12);
1731 __ c_ult_d(FSF, F12);
1732 __ bc1f(L);
1733 __ delayed()->addiu(T9, R0, -1);
1735 __ c_un_d(FSF, FSF); //NaN?
1736 __ subu32(FSR, T9, AT);
1737 __ movt(FSR, R0);
1739 __ bind(L);
1740 }
1741 break;
1742 case Bytecodes::_d2l:
1743 {
1744 Label L;
1746 __ trunc_l_d(F12, FSF);
1747 __ daddiu(AT, R0, -1);
1748 __ dsrl(AT, AT, 1);
1749 __ dmfc1(FSR, F12);
1751 __ bne(FSR, AT, L);
1752 __ delayed()->mtc1(R0, F12);
1754 __ cvt_d_w(F12, F12);
1755 __ c_ult_d(FSF, F12);
1756 __ bc1f(L);
1757 __ delayed()->daddiu(T9, R0, -1);
1759 __ c_un_d(FSF, FSF); //NaN?
1760 __ subu(FSR, T9, AT);
1761 __ movt(FSR, R0);
1763 __ bind(L);
1764 }
1765 break;
1766 case Bytecodes::_d2f:
1767 __ cvt_s_d(FSF, FSF);
1768 break;
1769 default :
1770 ShouldNotReachHere();
1771 }
1772 }
1774 void TemplateTable::lcmp() {
1775 transition(ltos, itos);
1777 Label low, high, done;
1778 __ pop(T0);
1779 __ pop(R0);
1780 __ slt(AT, T0, FSR);
1781 __ bne(AT, R0, low);
1782 __ delayed()->nop();
1784 __ bne(T0, FSR, high);
1785 __ delayed()->nop();
1787 __ li(FSR, (long)0);
1788 __ b(done);
1789 __ delayed()->nop();
1791 __ bind(low);
1792 __ li(FSR, (long)-1);
1793 __ b(done);
1794 __ delayed()->nop();
1796 __ bind(high);
1797 __ li(FSR, (long)1);
1798 __ b(done);
1799 __ delayed()->nop();
1801 __ bind(done);
1802 }
1804 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1805 Label less, done;
1807 __ move(FSR, R0);
1809 if (is_float) {
1810 __ lwc1(FTF, at_sp());
1811 __ c_eq_s(FTF, FSF);
1812 __ bc1t(done);
1813 __ delayed()->daddi(SP, SP, 1 * wordSize);
1815 if (unordered_result<0)
1816 __ c_ult_s(FTF, FSF);
1817 else
1818 __ c_olt_s(FTF, FSF);
1819 } else {
1820 __ ldc1(FTF, at_sp());
1821 __ c_eq_d(FTF, FSF);
1822 __ bc1t(done);
1823 __ delayed()->daddi(SP, SP, 2 * wordSize);
1825 if (unordered_result<0)
1826 __ c_ult_d(FTF, FSF);
1827 else
1828 __ c_olt_d(FTF, FSF);
1829 }
1830 __ bc1t(less);
1831 __ delayed()->nop();
1832 __ move(FSR, 1);
1833 __ b(done);
1834 __ delayed()->nop();
1835 __ bind(less);
1836 __ move(FSR, -1);
1837 __ bind(done);
1838 }
1841 // used registers : T3, A7, Rnext
1842 // FSR : return bci, this is defined by the vm specification
1843 // T2 : MDO taken count
1844 // T3 : method
1845 // A7 : offset
1846 // Rnext : next bytecode, this is required by dispatch_base
1847 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1848 __ get_method(T3);
1849 __ profile_taken_branch(A7, T2); // only C2 meaningful
1851 #ifndef CORE
1852 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
1853 InvocationCounter::counter_offset();
1854 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
1855 InvocationCounter::counter_offset();
1856 #endif // CORE
1858 // Load up T4 with the branch displacement
1859 if (!is_wide) {
1860 __ get_2_byte_integer_at_bcp(A7, AT, 1);
1861 __ hswap(A7);
1862 } else {
1863 __ get_4_byte_integer_at_bcp(A7, AT, 1);
1864 __ swap(A7);
1865 }
1867 // Handle all the JSR stuff here, then exit.
1868 // It's much shorter and cleaner than intermingling with the non-JSR
1869 // normal-branch stuff occuring below.
1870 if (is_jsr) {
1871 // Pre-load the next target bytecode into Rnext
1872 __ dadd(AT, BCP, A7);
1873 __ lbu(Rnext, AT, 0);
1875 // compute return address as bci in FSR
1876 __ daddi(FSR, BCP, (is_wide?5:3) - in_bytes(ConstMethod::codes_offset()));
1877 __ ld(AT, T3, in_bytes(Method::const_offset()));
1878 __ dsub(FSR, FSR, AT);
1879 // Adjust the bcp in BCP by the displacement in A7
1880 __ dadd(BCP, BCP, A7);
1881 // jsr returns atos that is not an oop
1882 // Push return address
1883 __ push_i(FSR);
1884 // jsr returns vtos
1885 __ dispatch_only_noverify(vtos);
1887 return;
1888 }
1890 // Normal (non-jsr) branch handling
1892 // Adjust the bcp in S0 by the displacement in T4
1893 __ dadd(BCP, BCP, A7);
1895 #ifdef CORE
1896 // Pre-load the next target bytecode into EBX
1897 __ lbu(Rnext, BCP, 0);
1898 // continue with the bytecode @ target
1899 __ dispatch_only(vtos);
1900 #else
1901 assert(UseLoopCounter || !UseOnStackReplacement, "on-stack-replacement requires loop counters");
1902 Label backedge_counter_overflow;
1903 Label profile_method;
1904 Label dispatch;
1905 if (UseLoopCounter) {
1906 // increment backedge counter for backward branches
1907 // eax: MDO
1908 // ebx: MDO bumped taken-count
1909 // T3: method
1910 // T4: target offset
1911 // BCP: target bcp
1912 // LVP: locals pointer
1913 __ bgtz(A7, dispatch); // check if forward or backward branch
1914 __ delayed()->nop();
1916 // check if MethodCounters exists
1917 Label has_counters;
1918 __ ld(AT, T3, in_bytes(Method::method_counters_offset())); // use AT as MDO, TEMP
1919 __ bne(AT, R0, has_counters);
1920 __ nop();
1921 __ push(T3);
1922 //__ push(A7);
1923 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
1924 T3);
1925 //__ pop(A7);
1926 __ pop(T3);
1927 __ ld(AT, T3, in_bytes(Method::method_counters_offset())); // use AT as MDO, TEMP
1928 __ beq(AT, R0, dispatch);
1929 __ nop();
1930 __ bind(has_counters);
1932 // increment back edge counter
1933 __ ld(T1, T3, in_bytes(Method::method_counters_offset()));
1934 __ lw(T0, T1, in_bytes(be_offset));
1935 __ increment(T0, InvocationCounter::count_increment);
1936 __ sw(T0, T1, in_bytes(be_offset));
1938 // load invocation counter
1939 __ lw(T1, T1, in_bytes(inv_offset));
1940 // buffer bit added, mask no needed
1942 // dadd backedge counter & invocation counter
1943 __ dadd(T1, T1, T0);
1945 if (ProfileInterpreter) {
1946 // Test to see if we should create a method data oop
1947 //__ lui(AT, Assembler::split_high(int(&InvocationCounter::InterpreterProfileLimit)));
1948 //__ lw(AT, AT, Assembler::split_low(int(&InvocationCounter::InterpreterProfileLimit)));
1949 // T1 : backedge counter & invocation counter
1950 __ li(AT, (long)&InvocationCounter::InterpreterProfileLimit);
1951 __ lw(AT, AT, 0);
1952 __ slt(AT, T1, AT);
1953 __ bne(AT, R0, dispatch);
1954 __ delayed()->nop();
1956 // if no method data exists, go to profile method
1957 __ test_method_data_pointer(T1, profile_method);
1959 if (UseOnStackReplacement) {
1960 // check for overflow against ebx which is the MDO taken count
1961 __ li(AT, (long)&InvocationCounter::InterpreterBackwardBranchLimit);
1962 __ lw(AT, AT, 0);
1963 // the value Rnext Is get from the beginning profile_taken_branch
1964 __ slt(AT, T2, AT);
1965 __ bne(AT, R0, dispatch);
1966 __ delayed()->nop();
1968 // When ProfileInterpreter is on, the backedge_count comes
1969 // from the methodDataOop, which value does not get reset on
1970 // the call to frequency_counter_overflow().
1971 // To avoid excessive calls to the overflow routine while
1972 // the method is being compiled, dadd a second test to make
1973 // sure the overflow function is called only once every
1974 // overflow_frequency.
1975 const int overflow_frequency = 1024;
1976 __ andi(AT, T2, overflow_frequency-1);
1977 __ beq(AT, R0, backedge_counter_overflow);
1978 __ delayed()->nop();
1979 }
1980 } else {
1981 if (UseOnStackReplacement) {
1982 // check for overflow against eax, which is the sum of the counters
1983 __ li(AT, (long)&InvocationCounter::InterpreterBackwardBranchLimit);
1984 __ lw(AT, AT, 0);
1985 __ slt(AT, T1, AT);
1986 __ beq(AT, R0, backedge_counter_overflow);
1987 __ delayed()->nop();
1988 }
1989 }
1990 __ bind(dispatch);
1991 }
1993 // Pre-load the next target bytecode into Rnext
1994 __ lbu(Rnext, BCP, 0);
1996 // continue with the bytecode @ target
1997 // FSR: return bci for jsr's, unused otherwise
1998 // Rnext: target bytecode
1999 // BCP: target bcp
2000 __ dispatch_only(vtos);
2002 if (UseLoopCounter) {
2003 if (ProfileInterpreter) {
2004 // Out-of-line code to allocate method data oop.
2005 __ bind(profile_method);
2006 __ call_VM(NOREG, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
2007 __ lbu(Rnext, BCP, 0);
2008 __ set_method_data_pointer_for_bcp();
2009 __ b(dispatch);
2010 __ delayed()->nop();
2011 }
2013 if (UseOnStackReplacement) {
2014 // invocation counter overflow
2015 __ bind(backedge_counter_overflow);
2016 __ sub(A7, BCP, A7); // branch bcp
2017 call_VM(NOREG, CAST_FROM_FN_PTR(address,
2018 InterpreterRuntime::frequency_counter_overflow), A7);
2019 __ lbu(Rnext, BCP, 0);
2021 // V0: osr nmethod (osr ok) or NULL (osr not possible)
2022 // V1: osr adapter frame return address
2023 // Rnext: target bytecode
2024 // LVP: locals pointer
2025 // BCP: bcp
2026 __ beq(V0, R0, dispatch);
2027 __ delayed()->nop();
2028 // nmethod may have been invalidated (VM may block upon call_VM return)
2029 __ lw(T3, V0, nmethod::entry_bci_offset());
2030 __ move(AT, InvalidOSREntryBci);
2031 __ beq(AT, T3, dispatch);
2032 __ delayed()->nop();
2033 // We need to prepare to execute the OSR method. First we must
2034 // migrate the locals and monitors off of the stack.
2035 //eax V0: osr nmethod (osr ok) or NULL (osr not possible)
2036 //ebx V1: osr adapter frame return address
2037 //edx Rnext: target bytecode
2038 //edi LVP: locals pointer
2039 //esi BCP: bcp
2040 __ move(BCP, V0);
2041 // const Register thread = ecx;
2042 const Register thread = TREG;
2043 #ifndef OPT_THREAD
2044 __ get_thread(thread);
2045 #endif
2046 call_VM(noreg, CAST_FROM_FN_PTR(address,
2047 SharedRuntime::OSR_migration_begin));
2048 // eax is OSR buffer, move it to expected parameter location
2049 //refer to osrBufferPointer in c1_LIRAssembler_mips.cpp
2050 __ move(T0, V0);
2052 // pop the interpreter frame
2053 __ ld(A7, Address(FP, frame::interpreter_frame_sender_sp_offset * wordSize));
2054 //FIXME, shall we keep the return address on the stack?
2055 __ leave(); // remove frame anchor
2056 __ move(LVP, RA);
2057 __ move(SP, A7);
2059 __ move(AT, -(StackAlignmentInBytes));
2060 __ andr(SP , SP , AT);
2062 // push the (possibly adjusted) return address
2063 //refer to osr_entry in c1_LIRAssembler_mips.cpp
2064 __ ld(AT, BCP, nmethod::osr_entry_point_offset());
2065 __ jr(AT);
2066 __ delayed()->nop();
2067 }
2068 }
2069 #endif // not CORE
2070 }
2073 void TemplateTable::if_0cmp(Condition cc) {
2074 transition(itos, vtos);
2075 // assume branch is more often taken than not (loops use backward branches)
2076 Label not_taken;
2077 switch(cc) {
2078 case not_equal:
2079 __ beq(FSR, R0, not_taken);
2080 break;
2081 case equal:
2082 __ bne(FSR, R0, not_taken);
2083 break;
2084 case less:
2085 __ bgez(FSR, not_taken);
2086 break;
2087 case less_equal:
2088 __ bgtz(FSR, not_taken);
2089 break;
2090 case greater:
2091 __ blez(FSR, not_taken);
2092 break;
2093 case greater_equal:
2094 __ bltz(FSR, not_taken);
2095 break;
2096 }
2097 __ delayed()->nop();
2099 branch(false, false);
2101 __ bind(not_taken);
2102 __ profile_not_taken_branch(FSR);
2103 }
2105 void TemplateTable::if_icmp(Condition cc) {
2106 transition(itos, vtos);
2107 // assume branch is more often taken than not (loops use backward branches)
2108 Label not_taken;
2110 __ pop_i(SSR);
2111 switch(cc) {
2112 case not_equal:
2113 __ beq(SSR, FSR, not_taken);
2114 break;
2115 case equal:
2116 __ bne(SSR, FSR, not_taken);
2117 break;
2118 case less:
2119 __ slt(AT, SSR, FSR);
2120 __ beq(AT, R0, not_taken);
2121 break;
2122 case less_equal:
2123 __ slt(AT, FSR, SSR);
2124 __ bne(AT, R0, not_taken);
2125 break;
2126 case greater:
2127 __ slt(AT, FSR, SSR);
2128 __ beq(AT, R0, not_taken);
2129 break;
2130 case greater_equal:
2131 __ slt(AT, SSR, FSR);
2132 __ bne(AT, R0, not_taken);
2133 break;
2134 }
2135 __ delayed()->nop();
2137 branch(false, false);
2138 __ bind(not_taken);
2139 __ profile_not_taken_branch(FSR);
2140 }
2142 void TemplateTable::if_nullcmp(Condition cc) {
2143 transition(atos, vtos);
2144 // assume branch is more often taken than not (loops use backward branches)
2145 Label not_taken;
2146 switch(cc) {
2147 case not_equal:
2148 __ beq(FSR, R0, not_taken);
2149 break;
2150 case equal:
2151 __ bne(FSR, R0, not_taken);
2152 break;
2153 default:
2154 ShouldNotReachHere();
2155 }
2156 __ delayed()->nop();
2158 branch(false, false);
2159 __ bind(not_taken);
2160 __ profile_not_taken_branch(FSR);
2161 }
2164 void TemplateTable::if_acmp(Condition cc) {
2165 transition(atos, vtos);
2166 // assume branch is more often taken than not (loops use backward branches)
2167 Label not_taken;
2168 // __ lw(SSR, SP, 0);
2169 __ pop_ptr(SSR);
2170 switch(cc) {
2171 case not_equal:
2172 __ beq(SSR, FSR, not_taken);
2173 break;
2174 case equal:
2175 __ bne(SSR, FSR, not_taken);
2176 break;
2177 default:
2178 ShouldNotReachHere();
2179 }
2180 __ delayed()->nop();
2182 branch(false, false);
2184 __ bind(not_taken);
2185 __ profile_not_taken_branch(FSR);
2186 }
2188 // used registers : T1, T2, T3
2189 // T1 : method
2190 // T2 : returb bci
2191 void TemplateTable::ret() {
2192 transition(vtos, vtos);
2194 locals_index(T2);
2195 __ ld(T2, T2, 0);
2196 __ profile_ret(T2, T3);
2198 __ get_method(T1);
2199 __ ld(BCP, T1, in_bytes(Method::const_offset()));
2200 __ dadd(BCP, BCP, T2);
2201 __ daddi(BCP, BCP, in_bytes(ConstMethod::codes_offset()));
2203 __ dispatch_next(vtos);
2204 }
2206 // used registers : T1, T2, T3
2207 // T1 : method
2208 // T2 : returb bci
2209 void TemplateTable::wide_ret() {
2210 transition(vtos, vtos);
2212 locals_index_wide(T2);
2213 __ ld(T2, T2, 0); // get return bci, compute return bcp
2214 __ profile_ret(T2, T3);
2216 __ get_method(T1);
2217 __ ld(BCP, T1, in_bytes(Method::const_offset()));
2218 __ dadd(BCP, BCP, T2);
2219 __ daddi(BCP, BCP, in_bytes(ConstMethod::codes_offset()));
2221 __ dispatch_next(vtos);
2222 }
2224 // used register T2, T3, A7, Rnext
2225 // T2 : bytecode pointer
2226 // T3 : low
2227 // A7 : high
2228 // Rnext : dest bytecode, required by dispatch_base
2229 void TemplateTable::tableswitch() {
2230 Label default_case, continue_execution;
2231 transition(itos, vtos);
2233 // align BCP
2234 __ daddi(T2, BCP, BytesPerInt);
2235 __ li(AT, -BytesPerInt);
2236 __ andr(T2, T2, AT);
2238 // load lo & hi
2239 __ lw(T3, T2, 1 * BytesPerInt);
2240 __ swap(T3);
2241 __ lw(A7, T2, 2 * BytesPerInt);
2242 __ swap(A7);
2244 // check against lo & hi
2245 __ slt(AT, FSR, T3);
2246 __ bne(AT, R0, default_case);
2247 __ delayed()->nop();
2249 __ slt(AT, A7, FSR);
2250 __ bne(AT, R0, default_case);
2251 __ delayed()->nop();
2253 // lookup dispatch offset, in A7 big endian
2254 __ dsub(FSR, FSR, T3);
2255 __ dsll(AT, FSR, Address::times_4);
2256 __ dadd(AT, T2, AT);
2257 __ lw(A7, AT, 3 * BytesPerInt);
2258 __ profile_switch_case(FSR, T9, T3);
2260 __ bind(continue_execution);
2261 __ swap(A7);
2262 __ dadd(BCP, BCP, A7);
2263 __ lbu(Rnext, BCP, 0);
2264 __ dispatch_only(vtos);
2266 // handle default
2267 __ bind(default_case);
2268 __ profile_switch_default(FSR);
2269 __ lw(A7, T2, 0);
2270 __ b(continue_execution);
2271 __ delayed()->nop();
2272 }
2274 void TemplateTable::lookupswitch() {
2275 transition(itos, itos);
2276 __ stop("lookupswitch bytecode should have been rewritten");
2277 }
2279 // used registers : T2, T3, A7, Rnext
2280 // T2 : bytecode pointer
2281 // T3 : pair index
2282 // A7 : offset
2283 // Rnext : dest bytecode
2284 // the data after the opcode is the same as lookupswitch
2285 // see Rewriter::rewrite_method for more information
2286 void TemplateTable::fast_linearswitch() {
2287 transition(itos, vtos);
2288 Label loop_entry, loop, found, continue_execution;
2290 // swap eax so we can avoid swapping the table entries
2291 __ swap(FSR);
2293 // align BCP
2294 __ daddi(T2, BCP, BytesPerInt);
2295 __ li(AT, -BytesPerInt);
2296 __ andr(T2, T2, AT);
2298 // set counter
2299 __ lw(T3, T2, BytesPerInt);
2300 __ swap(T3);
2301 __ b(loop_entry);
2302 __ delayed()->nop();
2304 // table search
2305 __ bind(loop);
2306 // get the entry value
2307 __ dsll(AT, T3, Address::times_8);
2308 __ dadd(AT, T2, AT);
2309 __ lw(AT, AT, 2 * BytesPerInt);
2311 // found?
2312 __ beq(FSR, AT, found);
2313 __ delayed()->nop();
2315 __ bind(loop_entry);
2316 __ bgtz(T3, loop);
2317 __ delayed()->daddiu(T3, T3, -1);
2319 // default case
2320 __ profile_switch_default(FSR);
2321 __ lw(A7, T2, 0);
2322 __ b(continue_execution);
2323 __ delayed()->nop();
2325 // entry found -> get offset
2326 __ bind(found);
2327 __ dsll(AT, T3, Address::times_8);
2328 __ dadd(AT, T2, AT);
2329 __ lw(A7, AT, 3 * BytesPerInt);
2330 __ profile_switch_case(T3, FSR, T2);
2332 // continue execution
2333 __ bind(continue_execution);
2334 __ swap(A7);
2335 __ dadd(BCP, BCP, A7);
2336 __ lbu(Rnext, BCP, 0);
2337 __ dispatch_only(vtos);
2338 }
2340 // used registers : T0, T1, T2, T3, A7, Rnext
2341 // T2 : pairs address(array)
2342 // Rnext : dest bytecode
2343 // the data after the opcode is the same as lookupswitch
2344 // see Rewriter::rewrite_method for more information
2345 void TemplateTable::fast_binaryswitch() {
2346 transition(itos, vtos);
2347 // Implementation using the following core algorithm:
2348 //
2349 // int binary_search(int key, LookupswitchPair* array, int n) {
2350 // // Binary search according to "Methodik des Programmierens" by
2351 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2352 // int i = 0;
2353 // int j = n;
2354 // while (i+1 < j) {
2355 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2356 // // with Q: for all i: 0 <= i < n: key < a[i]
2357 // // where a stands for the array and assuming that the (inexisting)
2358 // // element a[n] is infinitely big.
2359 // int h = (i + j) >> 1;
2360 // // i < h < j
2361 // if (key < array[h].fast_match()) {
2362 // j = h;
2363 // } else {
2364 // i = h;
2365 // }
2366 // }
2367 // // R: a[i] <= key < a[i+1] or Q
2368 // // (i.e., if key is within array, i is the correct index)
2369 // return i;
2370 // }
2372 // register allocation
2373 const Register array = T2;
2374 const Register i = T3, j = A7;
2375 const Register h = T1;
2376 const Register temp = T0;
2377 const Register key = FSR;
2379 // setup array
2380 __ daddi(array, BCP, 3*BytesPerInt);
2381 __ li(AT, -BytesPerInt);
2382 __ andr(array, array, AT);
2384 // initialize i & j
2385 __ move(i, R0);
2386 __ lw(j, array, - 1 * BytesPerInt);
2387 // Convert j into native byteordering
2388 __ swap(j);
2390 // and start
2391 Label entry;
2392 __ b(entry);
2393 __ delayed()->nop();
2395 // binary search loop
2396 {
2397 Label loop;
2398 __ bind(loop);
2399 // int h = (i + j) >> 1;
2400 __ dadd(h, i, j);
2401 __ dsrl(h, h, 1);
2402 // if (key < array[h].fast_match()) {
2403 // j = h;
2404 // } else {
2405 // i = h;
2406 // }
2407 // Convert array[h].match to native byte-ordering before compare
2408 __ dsll(AT, h, Address::times_8);
2409 __ dadd(AT, array, AT);
2410 __ lw(temp, AT, 0 * BytesPerInt);
2411 __ swap(temp);
2413 {
2414 Label set_i, end_of_if;
2415 __ slt(AT, key, temp);
2416 __ beq(AT, R0, set_i);
2417 __ delayed()->nop();
2419 __ b(end_of_if);
2420 __ delayed(); __ move(j, h);
2422 __ bind(set_i);
2423 __ move(i, h);
2425 __ bind(end_of_if);
2426 }
2427 // while (i+1 < j)
2428 __ bind(entry);
2429 __ daddi(h, i, 1);
2430 __ slt(AT, h, j);
2431 __ bne(AT, R0, loop);
2432 __ delayed()->nop();
2433 }
2435 // end of binary search, result index is i (must check again!)
2436 Label default_case;
2437 // Convert array[i].match to native byte-ordering before compare
2438 __ dsll(AT, i, Address::times_8);
2439 __ dadd(AT, array, AT);
2440 __ lw(temp, AT, 0 * BytesPerInt);
2441 __ swap(temp);
2442 __ bne(key, temp, default_case);
2443 __ delayed()->nop();
2445 // entry found -> j = offset
2446 __ dsll(AT, i, Address::times_8);
2447 __ dadd(AT, array, AT);
2448 __ lw(j, AT, 1 * BytesPerInt);
2449 __ profile_switch_case(i, key, array);
2450 __ swap(j);
2452 __ dadd(BCP, BCP, j);
2453 __ lbu(Rnext, BCP, 0);
2454 __ dispatch_only(vtos);
2456 // default case -> j = default offset
2457 __ bind(default_case);
2458 __ profile_switch_default(i);
2459 __ lw(j, array, - 2 * BytesPerInt);
2460 __ swap(j);
2461 __ dadd(BCP, BCP, j);
2462 __ lbu(Rnext, BCP, 0);
2463 __ dispatch_only(vtos);
2464 }
2466 void TemplateTable::_return(TosState state) {
2467 transition(state, state);
2468 assert(_desc->calls_vm(),
2469 "inconsistent calls_vm information"); // call in remove_activation
2471 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2472 assert(state == vtos, "only valid state");
2473 __ ld(T1, aaddress(0));
2474 __ load_klass(LVP, T1);
2475 __ lw(LVP, LVP, in_bytes(Klass::access_flags_offset()));
2476 __ move(AT, JVM_ACC_HAS_FINALIZER);
2477 __ andr(AT, AT, LVP);//by_css
2478 Label skip_register_finalizer;
2479 __ beq(AT, R0, skip_register_finalizer);
2480 __ delayed()->nop();
2481 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2482 InterpreterRuntime::register_finalizer), T1);
2483 __ bind(skip_register_finalizer);
2484 }
2485 __ remove_activation(state, T9);
2486 __ sync();
2488 __ jr(T9);
2489 __ delayed()->nop();
2490 }
2492 // ----------------------------------------------------------------------------
2493 // Volatile variables demand their effects be made known to all CPU's
2494 // in order. Store buffers on most chips allow reads & writes to
2495 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2496 // without some kind of memory barrier (i.e., it's not sufficient that
2497 // the interpreter does not reorder volatile references, the hardware
2498 // also must not reorder them).
2499 //
2500 // According to the new Java Memory Model (JMM):
2501 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2502 // writes act as aquire & release, so:
2503 // (2) A read cannot let unrelated NON-volatile memory refs that
2504 // happen after the read float up to before the read. It's OK for
2505 // non-volatile memory refs that happen before the volatile read to
2506 // float down below it.
2507 // (3) Similar a volatile write cannot let unrelated NON-volatile
2508 // memory refs that happen BEFORE the write float down to after the
2509 // write. It's OK for non-volatile memory refs that happen after the
2510 // volatile write to float up before it.
2511 //
2512 // We only put in barriers around volatile refs (they are expensive),
2513 // not _between_ memory refs (that would require us to track the
2514 // flavor of the previous memory refs). Requirements (2) and (3)
2515 // require some barriers before volatile stores and after volatile
2516 // loads. These nearly cover requirement (1) but miss the
2517 // volatile-store-volatile-load case. This final case is placed after
2518 // volatile-stores although it could just as well go before
2519 // volatile-loads.
2520 //void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits
2521 // order_constraint) {
2522 void TemplateTable::volatile_barrier( ) {
2523 // Helper function to insert a is-volatile test and memory barrier
2524 //if (os::is_MP()) { // Not needed on single CPU
2525 // __ membar(order_constraint);
2526 //}
2527 if( !os::is_MP() ) return; // Not needed on single CPU
2528 __ sync();
2529 }
2531 // we dont shift left 2 bits in get_cache_and_index_at_bcp
2532 // for we always need shift the index we use it. the ConstantPoolCacheEntry
2533 // is 16-byte long, index is the index in
2534 // ConstantPoolCache, so cache + base_offset() + index * 16 is
2535 // the corresponding ConstantPoolCacheEntry
2536 // used registers : T2
2537 // NOTE : the returned index need also shift left 4 to get the address!
2538 void TemplateTable::resolve_cache_and_index(int byte_no,
2539 Register Rcache,
2540 Register index,
2541 size_t index_size) {
2542 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2543 const Register temp = A1;
2544 assert_different_registers(Rcache, index);
2546 Label resolved;
2547 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2548 // is resolved?
2549 int i = (int)bytecode();
2550 __ addi(temp, temp, -i);
2551 __ beq(temp, R0, resolved);
2552 __ delayed()->nop();
2553 // resolve first time through
2554 address entry;
2555 switch (bytecode()) {
2556 case Bytecodes::_getstatic : // fall through
2557 case Bytecodes::_putstatic : // fall through
2558 case Bytecodes::_getfield : // fall through
2559 case Bytecodes::_putfield :
2560 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put);
2561 break;
2562 case Bytecodes::_invokevirtual : // fall through
2563 case Bytecodes::_invokespecial : // fall through
2564 case Bytecodes::_invokestatic : // fall through
2565 case Bytecodes::_invokeinterface:
2566 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke);
2567 break;
2568 case Bytecodes::_invokehandle:
2569 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokehandle);
2570 break;
2571 case Bytecodes::_invokedynamic:
2572 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic);
2573 break;
2574 default :
2575 fatal(err_msg("unexpected bytecode: %s", Bytecodes::name(bytecode())));
2576 break;
2577 }
2579 __ move(temp, i);
2580 __ call_VM(NOREG, entry, temp);
2582 // Update registers with resolved info
2583 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2584 __ bind(resolved);
2585 }
2587 // The Rcache and index registers must be set before call
2588 void TemplateTable::load_field_cp_cache_entry(Register obj,
2589 Register cache,
2590 Register index,
2591 Register off,
2592 Register flags,
2593 bool is_static = false) {
2594 assert_different_registers(cache, index, flags, off);
2596 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2597 // Field offset
2598 __ dsll(AT, index, Address::times_ptr);
2599 __ dadd(AT, cache, AT);
2600 __ ld(off, AT, in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset()));
2601 // Flags
2602 __ ld(flags, AT, in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset()));
2604 // klass overwrite register
2605 if (is_static) {
2606 __ ld(obj, AT, in_bytes(cp_base_offset + ConstantPoolCacheEntry::f1_offset()));
2607 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2608 __ ld(obj, Address(obj, mirror_offset));
2610 __ verify_oop(obj);
2611 }
2612 }
2614 // get the method, itable_index and flags of the current invoke
2615 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2616 Register method,
2617 Register itable_index,
2618 Register flags,
2619 bool is_invokevirtual,
2620 bool is_invokevfinal, /*unused*/
2621 bool is_invokedynamic) {
2622 // setup registers
2623 const Register cache = T3;
2624 const Register index = T1;
2625 assert_different_registers(method, flags);
2626 assert_different_registers(method, cache, index);
2627 assert_different_registers(itable_index, flags);
2628 assert_different_registers(itable_index, cache, index);
2629 assert(is_invokevirtual == (byte_no == f2_byte), "is invokevirtual flag redundant");
2630 // determine constant pool cache field offsets
2631 const int method_offset = in_bytes(
2632 ConstantPoolCache::base_offset() +
2633 ((byte_no == f2_byte)
2634 ? ConstantPoolCacheEntry::f2_offset()
2635 : ConstantPoolCacheEntry::f1_offset()));
2636 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2637 ConstantPoolCacheEntry::flags_offset());
2638 // access constant pool cache fields
2639 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2640 ConstantPoolCacheEntry::f2_offset());
2642 size_t index_size = (is_invokedynamic ? sizeof(u4): sizeof(u2));
2643 resolve_cache_and_index(byte_no, cache, index, index_size);
2645 //assert(wordSize == 8, "adjust code below");
2646 // note we shift 4 not 2, for we get is the true inde
2647 // of ConstantPoolCacheEntry, not the shifted 2-bit index as x86 version
2648 __ dsll(AT, index, Address::times_ptr);
2649 __ dadd(AT, cache, AT);
2650 __ ld(method, AT, method_offset);
2652 if (itable_index != NOREG) {
2653 __ ld(itable_index, AT, index_offset);
2654 }
2655 __ ld(flags, AT, flags_offset);
2656 }
2658 // The registers cache and index expected to be set before call.
2659 // Correct values of the cache and index registers are preserved.
2660 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2661 bool is_static, bool has_tos) {
2662 // do the JVMTI work here to avoid disturbing the register state below
2663 // We use c_rarg registers here because we want to use the register used in
2664 // the call to the VM
2665 if (JvmtiExport::can_post_field_access()) {
2666 // Check to see if a field access watch has been set before we
2667 // take the time to call into the VM.
2668 Label L1;
2669 // kill FSR
2670 Register tmp1 = T2;
2671 Register tmp2 = T1;
2672 Register tmp3 = T3;
2673 assert_different_registers(cache, index, AT);
2674 __ li(AT, (intptr_t)JvmtiExport::get_field_access_count_addr());
2675 __ lw(AT, AT, 0);
2676 __ beq(AT, R0, L1);
2677 __ delayed()->nop();
2679 __ get_cache_and_index_at_bcp(tmp2, tmp3, 1);
2681 // cache entry pointer
2682 __ daddi(tmp2, tmp2, in_bytes(ConstantPoolCache::base_offset()));
2683 __ shl(tmp3, LogBytesPerWord);
2684 __ dadd(tmp2, tmp2, tmp3);
2685 if (is_static) {
2686 __ move(tmp1, R0);
2687 } else {
2688 __ ld(tmp1, SP, 0);
2689 __ verify_oop(tmp1);
2690 }
2691 // tmp1: object pointer or NULL
2692 // tmp2: cache entry pointer
2693 // tmp3: jvalue object on the stack
2694 __ call_VM(NOREG, CAST_FROM_FN_PTR(address,
2695 InterpreterRuntime::post_field_access),
2696 tmp1, tmp2, tmp3);
2697 __ get_cache_and_index_at_bcp(cache, index, 1);
2698 __ bind(L1);
2699 }
2700 }
2702 void TemplateTable::pop_and_check_object(Register r) {
2703 __ pop_ptr(r);
2704 __ null_check(r); // for field access must check obj.
2705 __ verify_oop(r);
2706 }
2708 // used registers : T1, T2, T3, T1
2709 // T1 : flags
2710 // T2 : off
2711 // T3 : obj
2712 // T1 : field address
2713 // The flags 31, 30, 29, 28 together build a 4 bit number 0 to 8 with the
2714 // following mapping to the TosState states:
2715 // btos: 0
2716 // ctos: 1
2717 // stos: 2
2718 // itos: 3
2719 // ltos: 4
2720 // ftos: 5
2721 // dtos: 6
2722 // atos: 7
2723 // vtos: 8
2724 // see ConstantPoolCacheEntry::set_field for more info
2725 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2726 transition(vtos, vtos);
2728 const Register cache = T3;
2729 const Register index = T0;
2731 const Register obj = T3;
2732 const Register off = T2;
2733 const Register flags = T1;
2734 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2735 jvmti_post_field_access(cache, index, is_static, false);
2736 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2738 if (!is_static) pop_and_check_object(obj);
2739 __ dadd(index, obj, off);
2742 Label Done, notByte, notInt, notShort, notChar,
2743 notLong, notFloat, notObj, notDouble;
2745 assert(btos == 0, "change code, btos != 0");
2746 __ dsrl(flags, flags, ConstantPoolCacheEntry::tos_state_shift);
2747 __ andi(flags, flags, 0xf);
2748 __ bne(flags, R0, notByte);
2749 __ delayed()->nop();
2751 // btos
2752 __ lb(FSR, index, 0);
2753 __ sd(FSR, SP, - wordSize);
2755 // Rewrite bytecode to be faster
2756 if (!is_static) {
2757 patch_bytecode(Bytecodes::_fast_bgetfield, T3, T2);
2758 }
2759 __ b(Done);
2760 __ delayed()->daddi(SP, SP, - wordSize);
2762 __ bind(notByte);
2763 __ move(AT, itos);
2764 __ bne(flags, AT, notInt);
2765 __ delayed()->nop();
2767 // itos
2768 __ lw(FSR, index, 0);
2769 __ sd(FSR, SP, - wordSize);
2771 // Rewrite bytecode to be faster
2772 if (!is_static) {
2773 // patch_bytecode(Bytecodes::_fast_igetfield, T3, T2);
2774 patch_bytecode(Bytecodes::_fast_igetfield, T3, T2);
2775 }
2776 __ b(Done);
2777 __ delayed()->daddi(SP, SP, - wordSize);
2779 __ bind(notInt);
2780 __ move(AT, atos);
2781 __ bne(flags, AT, notObj);
2782 __ delayed()->nop();
2784 // atos
2785 //add for compressedoops
2786 __ load_heap_oop(FSR, Address(index, 0));
2787 __ sd(FSR, SP, - wordSize);
2789 if (!is_static) {
2790 //patch_bytecode(Bytecodes::_fast_agetfield, T3, T2);
2791 patch_bytecode(Bytecodes::_fast_agetfield, T3, T2);
2792 }
2793 __ b(Done);
2794 __ delayed()->daddi(SP, SP, - wordSize);
2796 __ bind(notObj);
2797 __ move(AT, ctos);
2798 __ bne(flags, AT, notChar);
2799 __ delayed()->nop();
2801 // ctos
2802 __ lhu(FSR, index, 0);
2803 __ sd(FSR, SP, - wordSize);
2805 if (!is_static) {
2806 patch_bytecode(Bytecodes::_fast_cgetfield, T3, T2);
2807 }
2808 __ b(Done);
2809 __ delayed()->daddi(SP, SP, - wordSize);
2811 __ bind(notChar);
2812 __ move(AT, stos);
2813 __ bne(flags, AT, notShort);
2814 __ delayed()->nop();
2816 // stos
2817 __ lh(FSR, index, 0);
2818 __ sd(FSR, SP, - wordSize);
2820 if (!is_static) {
2821 patch_bytecode(Bytecodes::_fast_sgetfield, T3, T2);
2822 }
2823 __ b(Done);
2824 __ delayed()->daddi(SP, SP, - wordSize);
2826 __ bind(notShort);
2827 __ move(AT, ltos);
2828 __ bne(flags, AT, notLong);
2829 __ delayed()->nop();
2831 // FIXME : the load/store should be atomic, we have no simple method to do this in mips32
2832 // ltos
2833 __ ld(FSR, index, 0 * wordSize);
2834 __ sd(FSR, SP, -2 * wordSize);
2835 __ sd(R0, SP, -1 * wordSize);
2837 // Don't rewrite to _fast_lgetfield for potential volatile case.
2838 __ b(Done);
2839 __ delayed()->daddi(SP, SP, - 2 * wordSize);
2841 __ bind(notLong);
2842 __ move(AT, ftos);
2843 __ bne(flags, AT, notFloat);
2844 __ delayed()->nop();
2846 // ftos
2847 __ lwc1(FSF, index, 0);
2848 __ sdc1(FSF, SP, - wordSize);
2850 if (!is_static) {
2851 patch_bytecode(Bytecodes::_fast_fgetfield, T3, T2);
2852 }
2853 __ b(Done);
2854 __ delayed()->daddi(SP, SP, - wordSize);
2856 __ bind(notFloat);
2857 __ move(AT, dtos);
2858 __ bne(flags, AT, notDouble);
2859 __ delayed()->nop();
2861 // dtos
2862 __ ldc1(FSF, index, 0 * wordSize);
2863 __ sdc1(FSF, SP, - 2 * wordSize);
2864 __ sd(R0, SP, - 1 * wordSize);
2866 if (!is_static) {
2867 patch_bytecode(Bytecodes::_fast_dgetfield, T3, T2);
2868 }
2869 __ b(Done);
2870 __ delayed()->daddi(SP, SP, - 2 * wordSize);
2872 __ bind(notDouble);
2874 __ stop("Bad state");
2876 __ bind(Done);
2877 }
2880 void TemplateTable::getfield(int byte_no) {
2881 getfield_or_static(byte_no, false);
2882 }
2884 void TemplateTable::getstatic(int byte_no) {
2885 getfield_or_static(byte_no, true);
2886 }
2888 // The registers cache and index expected to be set before call.
2889 // The function may destroy various registers, just not the cache and index registers.
2890 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2891 transition(vtos, vtos);
2893 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2895 if (JvmtiExport::can_post_field_modification()) {
2896 // Check to see if a field modification watch has been set before
2897 // we take the time to call into the VM.
2898 Label L1;
2899 //kill AT, T1, T2, T3, T9
2900 Register tmp1 = T2;
2901 Register tmp2 = T1;
2902 Register tmp3 = T3;
2903 Register tmp4 = T9;
2904 assert_different_registers(cache, index, tmp4);
2906 __ li(AT, JvmtiExport::get_field_modification_count_addr());
2907 __ lw(AT, AT, 0);
2908 __ beq(AT, R0, L1);
2909 __ delayed()->nop();
2911 __ get_cache_and_index_at_bcp(tmp2, tmp4, 1);
2913 if (is_static) {
2914 __ move(tmp1, R0);
2915 } else {
2916 // Life is harder. The stack holds the value on top, followed by
2917 // the object. We don't know the size of the value, though; it
2918 // could be one or two words depending on its type. As a result,
2919 // we must find the type to determine where the object is.
2920 Label two_word, valsize_known;
2921 __ dsll(AT, tmp4, Address::times_8);
2922 __ dadd(AT, tmp2, AT);
2923 __ ld(tmp3, AT, in_bytes(cp_base_offset +
2924 ConstantPoolCacheEntry::flags_offset()));
2925 __ shr(tmp3, ConstantPoolCacheEntry::tos_state_shift);
2927 // Make sure we don't need to mask ecx for tos_state_shift
2928 // after the above shift
2929 ConstantPoolCacheEntry::verify_tos_state_shift();
2930 __ move(tmp1, SP);
2931 __ move(AT, ltos);
2932 __ beq(tmp3, AT, two_word);
2933 __ delayed()->nop();
2934 __ move(AT, dtos);
2935 __ beq(tmp3, AT, two_word);
2936 __ delayed()->nop();
2937 __ b(valsize_known);
2938 __ delayed()->daddi(tmp1, tmp1, Interpreter::expr_offset_in_bytes(1) );
2940 __ bind(two_word);
2941 __ daddi(tmp1, tmp1, Interpreter::expr_offset_in_bytes(2));
2943 __ bind(valsize_known);
2944 // setup object pointer
2945 __ ld(tmp1, tmp1, 0*wordSize);
2946 }
2947 // cache entry pointer
2948 __ daddi(tmp2, tmp2, in_bytes(cp_base_offset));
2949 __ shl(tmp4, LogBytesPerWord);
2950 __ daddu(tmp2, tmp2, tmp4);
2951 // object (tos)
2952 __ move(tmp3, SP);
2953 // tmp1: object pointer set up above (NULL if static)
2954 // tmp2: cache entry pointer
2955 // tmp3: jvalue object on the stack
2956 __ call_VM(NOREG,
2957 CAST_FROM_FN_PTR(address,
2958 InterpreterRuntime::post_field_modification),
2959 tmp1, tmp2, tmp3);
2960 __ get_cache_and_index_at_bcp(cache, index, 1);
2961 __ bind(L1);
2962 }
2963 }
2965 // used registers : T0, T1, T2, T3, T8
2966 // T1 : flags
2967 // T2 : off
2968 // T3 : obj
2969 // T8 : volatile bit
2970 // see ConstantPoolCacheEntry::set_field for more info
2971 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2972 transition(vtos, vtos);
2974 const Register cache = T3;
2975 const Register index = T0;
2976 const Register obj = T3;
2977 const Register off = T2;
2978 const Register flags = T1;
2979 const Register bc = T3;
2981 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2982 jvmti_post_field_mod(cache, index, is_static);
2983 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2985 Label notVolatile, Done;
2986 __ move(AT, 1<<ConstantPoolCacheEntry::is_volatile_shift);
2987 __ andr(T8, flags, AT);
2989 Label notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2991 assert(btos == 0, "change code, btos != 0");
2992 // btos
2993 __ dsrl(flags, flags, ConstantPoolCacheEntry::tos_state_shift);
2994 __ andi(flags, flags, ConstantPoolCacheEntry::tos_state_mask);
2995 __ bne(flags, R0, notByte);
2996 __ delayed()->nop();
2998 __ pop(btos);
2999 if (!is_static) {
3000 pop_and_check_object(obj);
3001 }
3002 __ dadd(AT, obj, off);
3003 __ sb(FSR, AT, 0);
3005 if (!is_static) {
3006 patch_bytecode(Bytecodes::_fast_bputfield, bc, off, true, byte_no);
3007 }
3008 __ b(Done);
3009 __ delayed()->nop();
3011 __ bind(notByte);
3012 // itos
3013 __ move(AT, itos);
3014 __ bne(flags, AT, notInt);
3015 __ delayed()->nop();
3017 __ pop(itos);
3018 if (!is_static) {
3019 pop_and_check_object(obj);
3020 }
3021 __ dadd(AT, obj, off);
3022 __ sw(FSR, AT, 0);
3024 if (!is_static) {
3025 patch_bytecode(Bytecodes::_fast_iputfield, bc, off, true, byte_no);
3026 }
3027 __ b(Done);
3028 __ delayed()->nop();
3029 __ bind(notInt);
3030 // atos
3031 __ move(AT, atos);
3032 __ bne(flags, AT, notObj);
3033 __ delayed()->nop();
3035 __ pop(atos);
3036 if (!is_static) {
3037 pop_and_check_object(obj);
3038 }
3040 __ dadd(AT, obj, off);
3041 __ store_heap_oop(Address(AT, 0), FSR);
3042 __ store_check(obj);
3044 if (!is_static) {
3045 patch_bytecode(Bytecodes::_fast_aputfield, bc, off, true, byte_no);
3046 }
3047 __ b(Done);
3048 __ delayed()->nop();
3049 __ bind(notObj);
3050 // ctos
3051 __ move(AT, ctos);
3052 __ bne(flags, AT, notChar);
3053 __ delayed()->nop();
3055 __ pop(ctos);
3056 if (!is_static) {
3057 pop_and_check_object(obj);
3058 }
3059 __ dadd(AT, obj, off);
3060 __ sh(FSR, AT, 0);
3061 if (!is_static) {
3062 patch_bytecode(Bytecodes::_fast_cputfield, bc, off, true, byte_no);
3063 }
3064 __ b(Done);
3065 __ delayed()->nop();
3066 __ bind(notChar);
3067 // stos
3068 __ move(AT, stos);
3069 __ bne(flags, AT, notShort);
3070 __ delayed()->nop();
3072 __ pop(stos);
3073 if (!is_static) {
3074 pop_and_check_object(obj);
3075 }
3076 __ dadd(AT, obj, off);
3077 __ sh(FSR, AT, 0);
3078 if (!is_static) {
3079 patch_bytecode(Bytecodes::_fast_sputfield, bc, off, true, byte_no);
3080 }
3081 __ b(Done);
3082 __ delayed()->nop();
3083 __ bind(notShort);
3084 // ltos
3085 __ move(AT, ltos);
3086 __ bne(flags, AT, notLong);
3087 __ delayed()->nop();
3089 // FIXME: there is no simple method to load/store 64-bit data in a atomic operation
3090 // we just ignore the volatile flag.
3091 //Label notVolatileLong;
3092 //__ beq(T1, R0, notVolatileLong);
3093 //__ delayed()->nop();
3095 //addent = 2 * wordSize;
3096 // no need
3097 //__ lw(FSR, SP, 0);
3098 //__ lw(SSR, SP, 1 * wordSize);
3099 //if (!is_static) {
3100 // __ lw(T3, SP, addent);
3101 // addent += 1 * wordSize;
3102 // __ verify_oop(T3);
3103 //}
3105 //__ daddu(AT, T3, T2);
3107 // Replace with real volatile test
3108 // NOTE : we assume that sdc1&ldc1 operate in 32-bit, this is true for Godson2 even in 64-bit kernel
3109 // last modified by yjl 7/12/2005
3110 //__ ldc1(FSF, SP, 0);
3111 //__ sdc1(FSF, AT, 0);
3112 //volatile_barrier();
3114 // Don't rewrite volatile version
3115 //__ b(notVolatile);
3116 //__ delayed()->addiu(SP, SP, addent);
3118 //__ bind(notVolatileLong);
3120 //__ pop(ltos); // overwrites edx
3121 // __ lw(FSR, SP, 0 * wordSize);
3122 // __ lw(SSR, SP, 1 * wordSize);
3123 // __ daddi(SP, SP, 2*wordSize);
3124 __ pop(ltos);
3125 if (!is_static) {
3126 pop_and_check_object(obj);
3127 }
3128 __ dadd(AT, obj, off);
3129 __ sd(FSR, AT, 0);
3130 if (!is_static) {
3131 patch_bytecode(Bytecodes::_fast_lputfield, bc, off, true, byte_no);
3132 }
3133 __ b(notVolatile);
3134 __ delayed()->nop();
3136 __ bind(notLong);
3137 // ftos
3138 __ move(AT, ftos);
3139 __ bne(flags, AT, notFloat);
3140 __ delayed()->nop();
3142 __ pop(ftos);
3143 if (!is_static) {
3144 pop_and_check_object(obj);
3145 }
3146 __ dadd(AT, obj, off);
3147 __ swc1(FSF, AT, 0);
3148 if (!is_static) {
3149 patch_bytecode(Bytecodes::_fast_fputfield, bc, off, true, byte_no);
3150 }
3151 __ b(Done);
3152 __ delayed()->nop();
3153 __ bind(notFloat);
3154 // dtos
3155 __ move(AT, dtos);
3156 __ bne(flags, AT, notDouble);
3157 __ delayed()->nop();
3159 __ pop(dtos);
3160 if (!is_static) {
3161 pop_and_check_object(obj);
3162 }
3163 __ dadd(AT, obj, off);
3164 __ sdc1(FSF, AT, 0);
3165 if (!is_static) {
3166 patch_bytecode(Bytecodes::_fast_dputfield, bc, off, true, byte_no);
3167 }
3169 #ifdef ASSERT
3170 __ b(Done);
3171 __ delayed()->nop();
3173 __ bind(notDouble);
3174 __ stop("Bad state");
3175 #endif
3177 __ bind(Done);
3179 // Check for volatile store
3180 __ beq(T8, R0, notVolatile);
3181 __ delayed()->nop();
3182 volatile_barrier( );
3183 __ bind(notVolatile);
3184 }
3186 void TemplateTable::putfield(int byte_no) {
3187 putfield_or_static(byte_no, false);
3188 }
3190 void TemplateTable::putstatic(int byte_no) {
3191 putfield_or_static(byte_no, true);
3192 }
3194 // used registers : T1, T2, T3
3195 // T1 : cp_entry
3196 // T2 : obj
3197 // T3 : value pointer
3198 void TemplateTable::jvmti_post_fast_field_mod() {
3199 if (JvmtiExport::can_post_field_modification()) {
3200 // Check to see if a field modification watch has been set before
3201 // we take the time to call into the VM.
3202 Label L2;
3203 //kill AT, T1, T2, T3, T9
3204 Register tmp1 = T2;
3205 Register tmp2 = T1;
3206 Register tmp3 = T3;
3207 Register tmp4 = T9;
3208 __ li(AT, JvmtiExport::get_field_modification_count_addr());
3209 __ lw(tmp3, AT, 0);
3210 __ beq(tmp3, R0, L2);
3211 __ delayed()->nop();
3212 __ pop_ptr(tmp1);
3213 __ verify_oop(tmp1);
3214 __ push_ptr(tmp1);
3215 switch (bytecode()) { // load values into the jvalue object
3216 case Bytecodes::_fast_aputfield: __ push_ptr(FSR); break;
3217 case Bytecodes::_fast_bputfield: // fall through
3218 case Bytecodes::_fast_sputfield: // fall through
3219 case Bytecodes::_fast_cputfield: // fall through
3220 case Bytecodes::_fast_iputfield: __ push_i(FSR); break;
3221 case Bytecodes::_fast_dputfield: __ push_d(FSF); break;
3222 case Bytecodes::_fast_fputfield: __ push_f(); break;
3223 case Bytecodes::_fast_lputfield: __ push_l(FSR); break;
3224 default: ShouldNotReachHere();
3225 }
3226 __ move(tmp3, SP);
3227 // access constant pool cache entry
3228 __ get_cache_entry_pointer_at_bcp(tmp2, FSR, 1);
3229 __ verify_oop(tmp1);
3230 // tmp1: object pointer copied above
3231 // tmp2: cache entry pointer
3232 // tmp3: jvalue object on the stack
3233 __ call_VM(NOREG,
3234 CAST_FROM_FN_PTR(address,
3235 InterpreterRuntime::post_field_modification),
3236 tmp1, tmp2, tmp3);
3238 switch (bytecode()) { // restore tos values
3239 case Bytecodes::_fast_aputfield: __ pop_ptr(FSR); break;
3240 case Bytecodes::_fast_bputfield: // fall through
3241 case Bytecodes::_fast_sputfield: // fall through
3242 case Bytecodes::_fast_cputfield: // fall through
3243 case Bytecodes::_fast_iputfield: __ pop_i(FSR); break;
3244 case Bytecodes::_fast_dputfield: __ pop_d(); break;
3245 case Bytecodes::_fast_fputfield: __ pop_f(); break;
3246 case Bytecodes::_fast_lputfield: __ pop_l(FSR); break;
3247 }
3248 __ bind(L2);
3249 }
3250 }
3252 // used registers : T2, T3, T1
3253 // T2 : index & off & field address
3254 // T3 : cache & obj
3255 // T1 : flags
3256 void TemplateTable::fast_storefield(TosState state) {
3257 transition(state, vtos);
3259 ByteSize base = ConstantPoolCache::base_offset();
3261 jvmti_post_fast_field_mod();
3263 // access constant pool cache
3264 __ get_cache_and_index_at_bcp(T3, T2, 1);
3266 // test for volatile with edx but edx is tos register for lputfield.
3267 __ dsll(AT, T2, Address::times_8);
3268 __ dadd(AT, T3, AT);
3269 __ ld(T1, AT, in_bytes(base + ConstantPoolCacheEntry::flags_offset()));
3271 // replace index with field offset from cache entry
3272 __ ld(T2, AT, in_bytes(base + ConstantPoolCacheEntry::f2_offset()));
3274 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
3275 // volatile_barrier( );
3277 Label notVolatile, Done;
3278 // Check for volatile store
3279 __ move(AT, 1<<ConstantPoolCacheEntry::is_volatile_shift);
3280 __ andr(AT, T1, AT);
3281 __ beq(AT, R0, notVolatile);
3282 __ delayed()->nop();
3285 // Get object from stack
3286 pop_and_check_object(T3);
3288 // field address
3289 __ dadd(T2, T3, T2);
3291 // access field
3292 switch (bytecode()) {
3293 case Bytecodes::_fast_bputfield:
3294 __ sb(FSR, T2, 0);
3295 break;
3296 case Bytecodes::_fast_sputfield: // fall through
3297 case Bytecodes::_fast_cputfield:
3298 __ sh(FSR, T2, 0);
3299 break;
3300 case Bytecodes::_fast_iputfield:
3301 __ sw(FSR, T2, 0);
3302 break;
3303 case Bytecodes::_fast_lputfield:
3304 __ sd(FSR, T2, 0 * wordSize);
3305 break;
3306 case Bytecodes::_fast_fputfield:
3307 __ swc1(FSF, T2, 0);
3308 break;
3309 case Bytecodes::_fast_dputfield:
3310 __ sdc1(FSF, T2, 0 * wordSize);
3311 break;
3312 case Bytecodes::_fast_aputfield:
3313 __ store_heap_oop(Address(T2, 0), FSR);
3314 __ store_check(T3);
3315 break;
3316 default:
3317 ShouldNotReachHere();
3318 }
3320 Label done;
3321 volatile_barrier( );
3322 __ b(done);
3323 __ delayed()->nop();
3325 // Same code as above, but don't need edx to test for volatile.
3326 __ bind(notVolatile);
3327 pop_and_check_object(T3);
3328 //get the field address
3329 __ dadd(T2, T3, T2);
3331 // access field
3332 switch (bytecode()) {
3333 case Bytecodes::_fast_bputfield:
3334 __ sb(FSR, T2, 0);
3335 break;
3336 case Bytecodes::_fast_sputfield: // fall through
3337 case Bytecodes::_fast_cputfield:
3338 __ sh(FSR, T2, 0);
3339 break;
3340 case Bytecodes::_fast_iputfield:
3341 __ sw(FSR, T2, 0);
3342 break;
3343 case Bytecodes::_fast_lputfield:
3344 __ sd(FSR, T2, 0 * wordSize);
3345 break;
3346 case Bytecodes::_fast_fputfield:
3347 __ swc1(FSF, T2, 0);
3348 break;
3349 case Bytecodes::_fast_dputfield:
3350 __ sdc1(FSF, T2, 0 * wordSize);
3351 break;
3352 case Bytecodes::_fast_aputfield:
3353 //add for compressedoops
3354 __ store_heap_oop(Address(T2, 0), FSR);
3355 __ store_check(T3);
3356 break;
3357 default:
3358 ShouldNotReachHere();
3359 }
3360 __ bind(done);
3361 }
3363 // used registers : T2, T3, T1
3364 // T3 : cp_entry & cache
3365 // T2 : index & offset
3366 void TemplateTable::fast_accessfield(TosState state) {
3367 transition(atos, state);
3369 // do the JVMTI work here to avoid disturbing the register state below
3370 if (JvmtiExport::can_post_field_access()) {
3371 // Check to see if a field access watch has been set before we take
3372 // the time to call into the VM.
3373 Label L1;
3374 __ li(AT, (intptr_t)JvmtiExport::get_field_access_count_addr());
3375 __ lw(T3, AT, 0);
3376 __ beq(T3, R0, L1);
3377 __ delayed()->nop();
3378 // access constant pool cache entry
3379 __ get_cache_entry_pointer_at_bcp(T3, T1, 1);
3380 __ move(TSR, FSR);
3381 __ verify_oop(FSR);
3382 // FSR: object pointer copied above
3383 // T3: cache entry pointer
3384 __ call_VM(NOREG,
3385 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
3386 FSR, T3);
3387 __ move(FSR, TSR);
3388 __ bind(L1);
3389 }
3391 // access constant pool cache
3392 __ get_cache_and_index_at_bcp(T3, T2, 1);
3393 // replace index with field offset from cache entry
3394 __ dsll(AT, T2, Address::times_8);
3395 __ dadd(AT, T3, AT);
3396 __ ld(T2, AT, in_bytes(ConstantPoolCache::base_offset()
3397 + ConstantPoolCacheEntry::f2_offset()));
3399 // eax: object
3400 __ verify_oop(FSR);
3401 __ null_check(FSR);
3402 // field addresses
3403 __ dadd(FSR, FSR, T2);
3405 // access field
3406 switch (bytecode()) {
3407 case Bytecodes::_fast_bgetfield:
3408 __ lb(FSR, FSR, 0);
3409 break;
3410 case Bytecodes::_fast_sgetfield:
3411 __ lh(FSR, FSR, 0);
3412 break;
3413 case Bytecodes::_fast_cgetfield:
3414 __ lhu(FSR, FSR, 0);
3415 break;
3416 case Bytecodes::_fast_igetfield:
3417 __ lw(FSR, FSR, 0);
3418 break;
3419 case Bytecodes::_fast_lgetfield:
3420 __ stop("should not be rewritten");
3421 break;
3422 case Bytecodes::_fast_fgetfield:
3423 __ lwc1(FSF, FSR, 0);
3424 break;
3425 case Bytecodes::_fast_dgetfield:
3426 __ ldc1(FSF, FSR, 0);
3427 break;
3428 case Bytecodes::_fast_agetfield:
3429 //add for compressedoops
3430 __ load_heap_oop(FSR, Address(FSR, 0));
3431 __ verify_oop(FSR);
3432 break;
3433 default:
3434 ShouldNotReachHere();
3435 }
3437 // Doug Lea believes this is not needed with current Sparcs(TSO) and Intel(PSO)
3438 // volatile_barrier( );
3439 }
3441 // generator for _fast_iaccess_0, _fast_aaccess_0, _fast_faccess_0
3442 // used registers : T1, T2, T3, T1
3443 // T1 : obj & field address
3444 // T2 : off
3445 // T3 : cache
3446 // T1 : index
3447 void TemplateTable::fast_xaccess(TosState state) {
3448 transition(vtos, state);
3450 // get receiver
3451 __ ld(T1, aaddress(0));
3452 // access constant pool cache
3453 __ get_cache_and_index_at_bcp(T3, T2, 2);
3454 __ dsll(AT, T2, Address::times_8);
3455 __ dadd(AT, T3, AT);
3456 __ ld(T2, AT, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset()));
3458 // make sure exception is reported in correct bcp range (getfield is
3459 // next instruction)
3460 __ daddi(BCP, BCP, 1);
3461 __ null_check(T1);
3462 __ dadd(T1, T1, T2);
3464 if (state == itos) {
3465 __ lw(FSR, T1, 0);
3466 } else if (state == atos) {
3467 __ load_heap_oop(FSR, Address(T1, 0));
3468 __ verify_oop(FSR);
3469 } else if (state == ftos) {
3470 __ lwc1(FSF, T1, 0);
3471 } else {
3472 ShouldNotReachHere();
3473 }
3474 __ daddi(BCP, BCP, -1);
3475 }
3479 //-----------------------------------------------------------------------------
3480 // Calls
3482 void TemplateTable::count_calls(Register method, Register temp) {
3483 // implemented elsewhere
3484 ShouldNotReachHere();
3485 }
3487 // method, index, recv, flags: T1, T2, T3, T1
3488 // byte_no = 2 for _invokevirtual, 1 else
3489 // T0 : return address
3490 // get the method & index of the invoke, and push the return address of
3491 // the invoke(first word in the frame)
3492 // this address is where the return code jmp to.
3493 // NOTE : this method will set T3&T1 as recv&flags
3494 void TemplateTable::prepare_invoke(int byte_no,
3495 Register method, // linked method (or i-klass)
3496 Register index, // itable index, MethodType, etc.
3497 Register recv, // if caller wants to see it
3498 Register flags // if caller wants to test it
3499 ) {
3500 // determine flags
3501 const Bytecodes::Code code = bytecode();
3502 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3503 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3504 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3505 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3506 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3507 const bool load_receiver = (recv != noreg);
3508 const bool save_flags = (flags != noreg);
3509 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic),"");
3510 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3511 assert(flags == noreg || flags == T1, "error flags reg.");
3512 assert(recv == noreg || recv == T3, "error recv reg.");
3514 // setup registers & access constant pool cache
3515 if(recv == noreg) recv = T3;
3516 if(flags == noreg) flags = T1;
3517 assert_different_registers(method, index, recv, flags);
3519 // save 'interpreter return address'
3520 __ save_bcp();
3522 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3524 if (is_invokedynamic || is_invokehandle) {
3525 Label L_no_push;
3526 __ move(AT, (1 << ConstantPoolCacheEntry::has_appendix_shift));
3527 __ andr(AT, AT, flags);
3528 __ beq(AT, R0, L_no_push);
3529 __ delayed()->nop();
3530 // Push the appendix as a trailing parameter.
3531 // This must be done before we get the receiver,
3532 // since the parameter_size includes it.
3533 Register tmp = SSR;
3534 __ push(tmp);
3535 __ move(tmp, index);
3536 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3537 __ load_resolved_reference_at_index(index, tmp);
3538 __ pop(tmp);
3539 __ push(index); // push appendix (MethodType, CallSite, etc.)
3540 __ bind(L_no_push);
3541 }
3543 // load receiver if needed (after appendix is pushed so parameter size is correct)
3544 // Note: no return address pushed yet
3545 if (load_receiver) {
3546 __ move(AT, ConstantPoolCacheEntry::parameter_size_mask);
3547 __ andr(recv, flags, AT);
3548 // 2014/07/31 Fu: Since we won't push RA on stack, no_return_pc_pushed_yet should be 0.
3549 const int no_return_pc_pushed_yet = 0; // argument slot correction before we push return address
3550 const int receiver_is_at_end = -1; // back off one slot to get receiver
3551 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3552 __ ld(recv, recv_addr);
3553 __ verify_oop(recv);
3554 }
3555 if(save_flags) {
3556 __ move(BCP, flags);
3557 }
3559 // compute return type
3560 __ dsrl(flags, flags, ConstantPoolCacheEntry::tos_state_shift);
3561 __ andi(flags, flags, 0xf);
3563 // Make sure we don't need to mask flags for tos_state_shift after the above shift
3564 ConstantPoolCacheEntry::verify_tos_state_shift();
3565 // load return address
3566 {
3567 const address table = (address) Interpreter::invoke_return_entry_table_for(code);
3568 __ li(AT, (long)table);
3569 __ dsll(flags, flags, LogBytesPerWord);
3570 __ dadd(AT, AT, flags);
3571 __ ld(RA, AT, 0);
3572 }
3574 if (save_flags) {
3575 __ move(flags, BCP);
3576 __ restore_bcp();
3577 }
3578 }
3580 // used registers : T0, T3, T1, T2
3581 // T3 : recv, this two register using convention is by prepare_invoke
3582 // T1 : flags, klass
3583 // Rmethod : method, index must be Rmethod
3584 void TemplateTable::invokevirtual_helper(Register index,
3585 Register recv,
3586 Register flags) {
3588 assert_different_registers(index, recv, flags, T2);
3590 // Test for an invoke of a final method
3591 Label notFinal;
3592 __ move(AT, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3593 __ andr(AT, flags, AT);
3594 __ beq(AT, R0, notFinal);
3595 __ delayed()->nop();
3597 Register method = index; // method must be Rmethod
3598 assert(method == Rmethod, "methodOop must be Rmethod for interpreter calling convention");
3600 // do the call - the index is actually the method to call
3601 // the index is indeed methodOop, for this is vfinal,
3602 // see ConstantPoolCacheEntry::set_method for more info
3604 __ verify_oop(method);
3606 // It's final, need a null check here!
3607 __ null_check(recv);
3609 // profile this call
3610 __ profile_final_call(T2);
3612 // 2014/11/24 Fu
3613 // T2: tmp, used for mdp
3614 // method: callee
3615 // T9: tmp
3616 // is_virtual: true
3617 __ profile_arguments_type(T2, method, T9, true);
3619 __ jump_from_interpreted(method, T2);
3621 __ bind(notFinal);
3623 // get receiver klass
3624 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3625 __ load_klass(T2, recv);
3626 __ verify_oop(T2);
3628 // profile this call
3629 __ profile_virtual_call(T2, T0, T1);
3631 // get target methodOop & entry point
3632 const int base = InstanceKlass::vtable_start_offset() * wordSize;
3633 assert(vtableEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
3634 __ dsll(AT, index, Address::times_ptr);
3635 // T2: receiver
3636 __ dadd(AT, T2, AT);
3637 //this is a ualign read
3638 __ ld(method, AT, base + vtableEntry::method_offset_in_bytes());
3639 __ profile_arguments_type(T2, method, T9, true);
3640 __ jump_from_interpreted(method, T2);
3642 }
3644 void TemplateTable::invokevirtual(int byte_no) {
3645 transition(vtos, vtos);
3646 assert(byte_no == f2_byte, "use this argument");
3647 prepare_invoke(byte_no, Rmethod, NOREG, T3, T1);
3648 // now recv & flags in T3, T1
3649 invokevirtual_helper(Rmethod, T3, T1);
3650 }
3652 // T9 : entry
3653 // Rmethod : method
3654 void TemplateTable::invokespecial(int byte_no) {
3655 transition(vtos, vtos);
3656 assert(byte_no == f1_byte, "use this argument");
3657 prepare_invoke(byte_no, Rmethod, NOREG, T3);
3658 // now recv & flags in T3, T1
3659 __ verify_oop(T3);
3660 __ null_check(T3);
3661 __ profile_call(T9);
3663 // 2014/11/24 Fu
3664 // T8: tmp, used for mdp
3665 // Rmethod: callee
3666 // T9: tmp
3667 // is_virtual: false
3668 __ profile_arguments_type(T8, Rmethod, T9, false);
3670 __ jump_from_interpreted(Rmethod, T9);
3671 __ move(T0, T3);//aoqi ?
3672 }
3674 void TemplateTable::invokestatic(int byte_no) {
3675 transition(vtos, vtos);
3676 assert(byte_no == f1_byte, "use this argument");
3677 prepare_invoke(byte_no, Rmethod, NOREG);
3678 __ verify_oop(Rmethod);
3680 __ profile_call(T9);
3682 // 2014/11/24 Fu
3683 // T8: tmp, used for mdp
3684 // Rmethod: callee
3685 // T9: tmp
3686 // is_virtual: false
3687 __ profile_arguments_type(T8, Rmethod, T9, false);
3689 __ jump_from_interpreted(Rmethod, T9);
3690 }
3692 // i have no idea what to do here, now. for future change. FIXME.
3693 void TemplateTable::fast_invokevfinal(int byte_no) {
3694 transition(vtos, vtos);
3695 assert(byte_no == f2_byte, "use this argument");
3696 __ stop("fast_invokevfinal not used on mips64");
3697 }
3699 // used registers : T0, T1, T2, T3, T1, A7
3700 // T0 : itable, vtable, entry
3701 // T1 : interface
3702 // T3 : receiver
3703 // T1 : flags, klass
3704 // Rmethod : index, method, this is required by interpreter_entry
3705 void TemplateTable::invokeinterface(int byte_no) {
3706 transition(vtos, vtos);
3707 //this method will use T1-T4 and T0
3708 assert(byte_no == f1_byte, "use this argument");
3709 prepare_invoke(byte_no, T2, Rmethod, T3, T1);
3710 // T2: Interface
3711 // Rmethod: index
3712 // T3: receiver
3713 // T1: flags
3715 // Special case of invokeinterface called for virtual method of
3716 // java.lang.Object. See cpCacheOop.cpp for details.
3717 // This code isn't produced by javac, but could be produced by
3718 // another compliant java compiler.
3719 Label notMethod;
3720 __ move(AT, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3721 __ andr(AT, T1, AT);
3722 __ beq(AT, R0, notMethod);
3723 __ delayed()->nop();
3725 invokevirtual_helper(Rmethod, T3, T1);
3726 __ bind(notMethod);
3727 // Get receiver klass into T1 - also a null check
3728 //add for compressedoops
3729 __ load_klass(T1, T3);
3730 __ verify_oop(T1);
3732 // profile this call
3733 __ profile_virtual_call(T1, T0, FSR);
3735 // Compute start of first itableOffsetEntry (which is at the end of the vtable)
3736 // TODO: x86 add a new method lookup_interface_method // LEE
3737 const int base = InstanceKlass::vtable_start_offset() * wordSize;
3738 assert(vtableEntry::size() * wordSize == 8, "adjust the scaling in the code below");
3739 __ lw(AT, T1, InstanceKlass::vtable_length_offset() * wordSize);
3740 __ dsll(AT, AT, Address::times_8);
3741 __ dadd(T0, T1, AT);
3742 __ daddi(T0, T0, base);
3743 if (HeapWordsPerLong > 1) {
3744 // Round up to align_object_offset boundary
3745 __ round_to(T0, BytesPerLong);
3746 }
3747 // now T0 is the begin of the itable
3749 Label entry, search, interface_ok;
3751 ///__ jmp(entry);
3752 __ b(entry);
3753 __ delayed()->nop();
3755 __ bind(search);
3756 __ increment(T0, itableOffsetEntry::size() * wordSize);
3758 __ bind(entry);
3760 // Check that the entry is non-null. A null entry means that the receiver
3761 // class doesn't implement the interface, and wasn't the same as the
3762 // receiver class checked when the interface was resolved.
3763 __ ld(AT, T0, itableOffsetEntry::interface_offset_in_bytes());
3764 __ bne(AT, R0, interface_ok);
3765 __ delayed()->nop();
3766 // throw exception
3767 // the call_VM checks for exception, so we should never return here.
3769 //__ pop();//FIXME here,
3770 // pop return address (pushed by prepare_invoke).
3771 // no need now, we just save the value in RA now
3773 __ call_VM(NOREG, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeError));
3774 __ should_not_reach_here();
3776 __ bind(interface_ok);
3777 //NOTICE here, no pop as x86 do
3778 __ bne(AT, T2, search);
3779 __ delayed()->nop();
3781 // now we get vtable of the interface
3782 __ ld(T0, T0, itableOffsetEntry::offset_offset_in_bytes());
3783 __ daddu(T0, T1, T0);
3784 assert(itableMethodEntry::size() * wordSize == 8, "adjust the scaling in the code below");
3785 __ dsll(AT, Rmethod, Address::times_8);
3786 __ daddu(AT, T0, AT);
3787 // now we get the method
3788 __ ld(Rmethod, AT, 0);
3789 // Rnext: methodOop to call
3790 // T3: receiver
3791 // Check for abstract method error
3792 // Note: This should be done more efficiently via a throw_abstract_method_error
3793 // interpreter entry point and a conditional jump to it in case of a null
3794 // method.
3795 {
3796 Label L;
3797 __ bne(Rmethod, R0, L);
3798 __ delayed()->nop();
3800 // throw exception
3801 // note: must restore interpreter registers to canonical
3802 // state for exception handling to work correctly!
3803 ///__ popl(ebx); // pop return address (pushed by prepare_invoke)
3804 //__ restore_bcp(); // esi must be correct for exception handler
3805 //(was destroyed)
3806 //__ restore_locals(); // make sure locals pointer
3807 //is correct as well (was destroyed)
3808 ///__ call_VM(noreg, CAST_FROM_FN_PTR(address,
3809 //InterpreterRuntime::throw_AbstractMethodError));
3810 __ call_VM(NOREG, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3811 // the call_VM checks for exception, so we should never return here.
3812 __ should_not_reach_here();
3813 __ bind(L);
3814 }
3816 // 2014/11/24 Fu
3817 // T8: tmp, used for mdp
3818 // Rmethod: callee
3819 // T9: tmp
3820 // is_virtual: true
3821 __ profile_arguments_type(T8, Rmethod, T9, true);
3823 __ jump_from_interpreted(Rmethod, T9);
3824 }
3827 void TemplateTable::invokehandle(int byte_no) {
3828 transition(vtos, vtos);
3829 assert(byte_no == f1_byte, "use this argument");
3830 const Register T2_method = Rmethod;
3831 const Register FSR_mtype = FSR;
3832 const Register T3_recv = T3;
3834 if (!EnableInvokeDynamic) {
3835 // rewriter does not generate this bytecode
3836 __ should_not_reach_here();
3837 return;
3838 }
3840 prepare_invoke(byte_no, T2_method, FSR_mtype, T3_recv);
3841 //??__ verify_method_ptr(T2_method);
3842 __ verify_oop(T3_recv);
3843 __ null_check(T3_recv);
3845 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3846 // rbx: MH.invokeExact_MT method (from f2)
3848 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke
3850 // FIXME: profile the LambdaForm also
3851 __ profile_final_call(T9);
3853 // 2014/11/24 Fu
3854 // T8: tmp, used for mdp
3855 // T2_method: callee
3856 // T9: tmp
3857 // is_virtual: true
3858 __ profile_arguments_type(T8, T2_method, T9, true);
3860 __ jump_from_interpreted(T2_method, T9);
3861 }
3863 void TemplateTable::invokedynamic(int byte_no) {
3864 transition(vtos, vtos);
3865 assert(byte_no == f1_byte, "use this argument");
3867 if (!EnableInvokeDynamic) {
3868 // We should not encounter this bytecode if !EnableInvokeDynamic.
3869 // The verifier will stop it. However, if we get past the verifier,
3870 // this will stop the thread in a reasonable way, without crashing the JVM.
3871 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3872 InterpreterRuntime::throw_IncompatibleClassChangeError));
3873 // the call_VM checks for exception, so we should never return here.
3874 __ should_not_reach_here();
3875 return;
3876 }
3878 //const Register Rmethod = T2;
3879 const Register T2_callsite = T2;
3881 prepare_invoke(byte_no, Rmethod, T2_callsite);
3883 // rax: CallSite object (from cpool->resolved_references[f1])
3884 // rbx: MH.linkToCallSite method (from f2)
3886 // Note: rax_callsite is already pushed by prepare_invoke
3887 // %%% should make a type profile for any invokedynamic that takes a ref argument
3888 // profile this call
3889 __ profile_call(T9);
3891 // 2014/11/24 Fu
3892 // T8: tmp, used for mdp
3893 // Rmethod: callee
3894 // T9: tmp
3895 // is_virtual: false
3896 __ profile_arguments_type(T8, Rmethod, T9, false);
3898 __ verify_oop(T2_callsite);
3900 __ jump_from_interpreted(Rmethod, T9);
3901 }
3903 //-----------------------------------------------------------------------------
3904 // Allocation
3905 // T1 : tags & buffer end & thread
3906 // T2 : object end
3907 // T3 : klass
3908 // T1 : object size
3909 // A1 : cpool
3910 // A2 : cp index
3911 // return object in FSR
3912 void TemplateTable::_new() {
3913 transition(vtos, atos);
3914 __ get_unsigned_2_byte_index_at_bcp(A2, 1);
3916 Label slow_case;
3917 Label done;
3918 Label initialize_header;
3919 Label initialize_object; // including clearing the fields
3920 Label allocate_shared;
3922 // get InstanceKlass in T3
3923 __ get_cpool_and_tags(A1, T1);
3925 __ dsll(AT, A2, Address::times_8);
3926 if (UseLoongsonISA && Assembler::is_simm(sizeof(ConstantPool), 8)) {
3927 __ gsldx(T3, A1, AT, sizeof(ConstantPool));
3928 } else {
3929 __ dadd(AT, A1, AT);
3930 __ ld(T3, AT, sizeof(ConstantPool));
3931 }
3933 // make sure the class we're about to instantiate has been resolved.
3934 // Note: slow_case does a pop of stack, which is why we loaded class/pushed above
3935 const int tags_offset = Array<u1>::base_offset_in_bytes();
3936 if (UseLoongsonISA && Assembler::is_simm(tags_offset, 8)) {
3937 __ gslbx(AT, T1, A2, tags_offset);
3938 } else {
3939 __ dadd(T1, T1, A2);
3940 __ lb(AT, T1, tags_offset);
3941 }
3942 __ daddiu(AT, AT, - (int)JVM_CONSTANT_Class);
3943 __ bne(AT, R0, slow_case);
3944 //__ delayed()->nop();
3947 // make sure klass is initialized & doesn't have finalizer
3948 // make sure klass is fully initialized
3949 __ lhu(T1, T3, in_bytes(InstanceKlass::init_state_offset()));
3950 __ daddiu(AT, T1, - (int)InstanceKlass::fully_initialized);
3951 __ bne(AT, R0, slow_case);
3952 //__ delayed()->nop();
3954 // has_finalizer
3955 __ lw(T0, T3, in_bytes(Klass::layout_helper_offset()) );
3956 __ andi(AT, T0, Klass::_lh_instance_slow_path_bit);
3957 __ bne(AT, R0, slow_case);
3958 //__ delayed()->nop();
3960 // Allocate the instance
3961 // 1) Try to allocate in the TLAB
3962 // 2) if fail and the object is large allocate in the shared Eden
3963 // 3) if the above fails (or is not applicable), go to a slow case
3964 // (creates a new TLAB, etc.)
3966 const bool allow_shared_alloc =
3967 Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
3969 if (UseTLAB) {
3970 #ifndef OPT_THREAD
3971 const Register thread = T8;
3972 __ get_thread(thread);
3973 #else
3974 const Register thread = TREG;
3975 #endif
3976 // get tlab_top
3977 __ ld(FSR, thread, in_bytes(JavaThread::tlab_top_offset()));
3978 // get tlab_end
3979 __ ld(AT, thread, in_bytes(JavaThread::tlab_end_offset()));
3980 __ dadd(T2, FSR, T0);
3981 __ slt(AT, AT, T2);
3982 __ bne(AT, R0, allow_shared_alloc ? allocate_shared : slow_case);
3983 __ delayed()->nop();
3984 __ sd(T2, thread, in_bytes(JavaThread::tlab_top_offset()));
3986 if (ZeroTLAB) {
3987 // the fields have been already cleared
3988 __ beq(R0, R0, initialize_header);
3989 } else {
3990 // initialize both the header and fields
3991 __ beq(R0, R0, initialize_object);
3992 }
3993 __ delayed()->nop();
3994 }
3996 // Allocation in the shared Eden , if allowed
3997 // T0 : instance size in words
3998 if(allow_shared_alloc){
3999 __ bind(allocate_shared);
4001 Label retry;
4002 Address heap_top(T1);
4003 __ set64(T1, (long)Universe::heap()->top_addr());
4004 __ ld(FSR, heap_top);
4006 __ bind(retry);
4007 __ set64(AT, (long)Universe::heap()->end_addr());
4008 __ ld(AT, AT, 0);
4009 __ dadd(T2, FSR, T0);
4010 __ slt(AT, AT, T2);
4011 __ bne(AT, R0, slow_case);
4012 __ delayed()->nop();
4014 // Compare FSR with the top addr, and if still equal, store the new
4015 // top addr in ebx at the address of the top addr pointer. Sets ZF if was
4016 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
4017 //
4018 // FSR: object begin
4019 // T2: object end
4020 // T0: instance size in words
4022 // if someone beat us on the allocation, try again, otherwise continue
4023 __ cmpxchg(T2, heap_top, FSR);
4024 __ beq(AT, R0, retry);
4025 __ delayed()->nop();
4026 }
4028 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
4029 // The object is initialized before the header. If the object size is
4030 // zero, go directly to the header initialization.
4031 __ bind(initialize_object);
4032 __ set64(AT, - sizeof(oopDesc));
4033 __ daddu(T0, T0, AT);
4034 __ beq(T0, R0, initialize_header);
4035 __ delayed()->nop();
4037 // initialize remaining object fields: T0 is a multiple of 2
4038 {
4039 Label loop;
4040 __ dadd(T1, FSR, T0);
4041 __ daddi(T1, T1, -oopSize);
4043 __ bind(loop);
4044 __ sd(R0, T1, sizeof(oopDesc) + 0 * oopSize);
4045 __ bne(T1, FSR, loop); //dont clear header
4046 __ delayed()->daddi(T1, T1, -oopSize);
4047 }
4049 //klass in T3,
4050 // initialize object header only.
4051 __ bind(initialize_header);
4052 if (UseBiasedLocking) {
4053 __ ld(AT, T3, in_bytes(Klass::prototype_header_offset()));
4054 __ sd(AT, FSR, oopDesc::mark_offset_in_bytes ());
4055 } else {
4056 __ set64(AT, (long)markOopDesc::prototype());
4057 __ sd(AT, FSR, oopDesc::mark_offset_in_bytes());
4058 }
4060 __ store_klass_gap(FSR, R0);
4061 __ store_klass(FSR, T3);
4063 {
4064 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
4065 // Trigger dtrace event for fastpath
4066 __ push(atos);
4067 __ call_VM_leaf(
4068 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), FSR);
4069 __ pop(atos);
4071 }
4072 __ b(done);
4073 __ delayed()->nop();
4074 }
4076 // slow case
4077 __ bind(slow_case);
4078 call_VM(FSR, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), A1, A2);
4080 // continue
4081 __ bind(done);
4082 __ sync();
4083 }
4085 void TemplateTable::newarray() {
4086 transition(itos, atos);
4087 __ lbu(A1, at_bcp(1));
4088 //type, count
4089 call_VM(FSR, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), A1, FSR);
4090 __ sync();
4091 }
4093 void TemplateTable::anewarray() {
4094 transition(itos, atos);
4095 __ get_2_byte_integer_at_bcp(A2, AT, 1);
4096 __ huswap(A2);
4097 __ get_constant_pool(A1);
4098 // cp, index, count
4099 call_VM(FSR, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), A1, A2, FSR);
4100 __ sync();
4101 }
4103 void TemplateTable::arraylength() {
4104 transition(atos, itos);
4105 __ null_check(FSR, arrayOopDesc::length_offset_in_bytes());
4106 __ lw(FSR, FSR, arrayOopDesc::length_offset_in_bytes());
4107 }
4109 // i use T2 as ebx, T3 as ecx, T1 as edx
4110 // when invoke gen_subtype_check, super in T3, sub in T2, object in FSR(it's always)
4111 // T2 : sub klass
4112 // T3 : cpool
4113 // T3 : super klass
4114 void TemplateTable::checkcast() {
4115 transition(atos, atos);
4116 Label done, is_null, ok_is_subtype, quicked, resolved;
4117 __ beq(FSR, R0, is_null);
4118 __ delayed()->nop();
4120 // Get cpool & tags index
4121 __ get_cpool_and_tags(T3, T1);
4122 __ get_2_byte_integer_at_bcp(T2, AT, 1);
4123 __ huswap(T2);
4125 // See if bytecode has already been quicked
4126 __ dadd(AT, T1, T2);
4127 __ lb(AT, AT, Array<u1>::base_offset_in_bytes());
4128 __ daddiu(AT, AT, - (int)JVM_CONSTANT_Class);
4129 __ beq(AT, R0, quicked);
4130 __ delayed()->nop();
4132 /* 2012/6/2 Jin: In InterpreterRuntime::quicken_io_cc, lots of new classes may be loaded.
4133 * Then, GC will move the object in V0 to another places in heap.
4134 * Therefore, We should never save such an object in register.
4135 * Instead, we should save it in the stack. It can be modified automatically by the GC thread.
4136 * After GC, the object address in FSR is changed to a new place.
4137 */
4138 __ push(atos);
4139 const Register thread = TREG;
4140 #ifndef OPT_THREAD
4141 __ get_thread(thread);
4142 #endif
4143 call_VM(NOREG, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4144 __ get_vm_result_2(T3, thread);
4145 __ pop_ptr(FSR);
4146 __ b(resolved);
4147 __ delayed()->nop();
4149 // klass already in cp, get superklass in T3
4150 __ bind(quicked);
4151 __ dsll(AT, T2, Address::times_8);
4152 __ dadd(AT, T3, AT);
4153 __ ld(T3, AT, sizeof(ConstantPool));
4155 __ bind(resolved);
4157 // get subklass in T2
4158 //add for compressedoops
4159 __ load_klass(T2, FSR);
4160 // Superklass in T3. Subklass in T2.
4161 __ gen_subtype_check(T3, T2, ok_is_subtype);
4163 // Come here on failure
4164 // object is at FSR
4165 __ jmp(Interpreter::_throw_ClassCastException_entry);
4166 __ delayed()->nop();
4168 // Come here on success
4169 __ bind(ok_is_subtype);
4171 // Collect counts on whether this check-cast sees NULLs a lot or not.
4172 if (ProfileInterpreter) {
4173 __ b(done);
4174 __ delayed()->nop();
4175 __ bind(is_null);
4176 __ profile_null_seen(T3);
4177 } else {
4178 __ bind(is_null);
4179 }
4180 __ bind(done);
4181 }
4183 // i use T3 as cpool, T1 as tags, T2 as index
4184 // object always in FSR, superklass in T3, subklass in T2
4185 void TemplateTable::instanceof() {
4186 transition(atos, itos);
4187 Label done, is_null, ok_is_subtype, quicked, resolved;
4189 __ beq(FSR, R0, is_null);
4190 __ delayed()->nop();
4192 // Get cpool & tags index
4193 __ get_cpool_and_tags(T3, T1);
4194 // get index
4195 __ get_2_byte_integer_at_bcp(T2, AT, 1);
4196 __ hswap(T2);
4198 // See if bytecode has already been quicked
4199 // quicked
4200 __ daddu(AT, T1, T2);
4201 __ lb(AT, AT, Array<u1>::base_offset_in_bytes());
4202 __ daddiu(AT, AT, - (int)JVM_CONSTANT_Class);
4203 __ beq(AT, R0, quicked);
4204 __ delayed()->nop();
4206 __ push(atos);
4207 const Register thread = TREG;
4208 #ifndef OPT_THREAD
4209 __ get_thread(thread);
4210 #endif
4211 call_VM(NOREG, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4212 __ get_vm_result_2(T3, thread);
4213 __ pop_ptr(FSR);
4214 __ b(resolved);
4215 __ delayed()->nop();
4217 // get superklass in T3, subklass in T2
4218 __ bind(quicked);
4219 __ dsll(AT, T2, Address::times_8);
4220 __ daddu(AT, T3, AT);
4221 __ ld(T3, AT, sizeof(ConstantPool));
4223 __ bind(resolved);
4224 // get subklass in T2
4225 //add for compressedoops
4226 __ load_klass(T2, FSR);
4228 // Superklass in T3. Subklass in T2.
4229 __ gen_subtype_check(T3, T2, ok_is_subtype);
4230 // Come here on failure
4231 __ b(done);
4232 __ delayed(); __ move(FSR, R0);
4234 // Come here on success
4235 __ bind(ok_is_subtype);
4236 __ move(FSR, 1);
4238 // Collect counts on whether this test sees NULLs a lot or not.
4239 if (ProfileInterpreter) {
4240 __ beq(R0, R0, done);
4241 __ nop();
4242 __ bind(is_null);
4243 __ profile_null_seen(T3);
4244 } else {
4245 __ bind(is_null); // same as 'done'
4246 }
4247 __ bind(done);
4248 // FSR = 0: obj == NULL or obj is not an instanceof the specified klass
4249 // FSR = 1: obj != NULL and obj is an instanceof the specified klass
4250 }
4252 //--------------------------------------------------------
4253 //--------------------------------------------
4254 // Breakpoints
4255 void TemplateTable::_breakpoint() {
4256 // Note: We get here even if we are single stepping..
4257 // jbug inists on setting breakpoints at every bytecode
4258 // even if we are in single step mode.
4260 transition(vtos, vtos);
4262 // get the unpatched byte code
4263 __ get_method(A1);
4264 __ call_VM(NOREG,
4265 CAST_FROM_FN_PTR(address,
4266 InterpreterRuntime::get_original_bytecode_at),
4267 A1, BCP);
4268 __ move(Rnext, V0); // Jin: Rnext will be used in dispatch_only_normal
4270 // post the breakpoint event
4271 __ get_method(A1);
4272 __ call_VM(NOREG, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), A1, BCP);
4274 // complete the execution of original bytecode
4275 __ dispatch_only_normal(vtos);
4276 }
4278 //-----------------------------------------------------------------------------
4279 // Exceptions
4281 void TemplateTable::athrow() {
4282 transition(atos, vtos);
4283 __ null_check(FSR);
4284 __ jmp(Interpreter::throw_exception_entry());
4285 __ delayed()->nop();
4286 }
4288 //-----------------------------------------------------------------------------
4289 // Synchronization
4290 //
4291 // Note: monitorenter & exit are symmetric routines; which is reflected
4292 // in the assembly code structure as well
4293 //
4294 // Stack layout:
4295 //
4296 // [expressions ] <--- SP = expression stack top
4297 // ..
4298 // [expressions ]
4299 // [monitor entry] <--- monitor block top = expression stack bot
4300 // ..
4301 // [monitor entry]
4302 // [frame data ] <--- monitor block bot
4303 // ...
4304 // [return addr ] <--- FP
4306 // we use T2 as monitor entry pointer, T3 as monitor top pointer, c_rarg0 as free slot pointer
4307 // object always in FSR
4308 void TemplateTable::monitorenter() {
4309 transition(atos, vtos);
4311 // check for NULL object
4312 __ null_check(FSR);
4314 const Address monitor_block_top(FP, frame::interpreter_frame_monitor_block_top_offset
4315 * wordSize);
4316 const int entry_size = (frame::interpreter_frame_monitor_size()* wordSize);
4317 Label allocated;
4319 // initialize entry pointer
4320 __ move(c_rarg0, R0);
4322 // find a free slot in the monitor block (result in edx)
4323 {
4324 Label entry, loop, exit, next;
4325 __ ld(T2, monitor_block_top);
4326 __ b(entry);
4327 __ delayed()->daddi(T3, FP, frame::interpreter_frame_initial_sp_offset * wordSize);
4329 // free slot?
4330 __ bind(loop);
4331 __ ld(AT, T2, BasicObjectLock::obj_offset_in_bytes());
4332 __ bne(AT, R0, next);
4333 __ delayed()->nop();
4334 __ move(c_rarg0, T2);
4336 __ bind(next);
4337 __ beq(FSR, AT, exit);
4338 __ delayed()->nop();
4339 __ daddi(T2, T2, entry_size);
4341 __ bind(entry);
4342 __ bne(T3, T2, loop);
4343 __ delayed()->nop();
4344 __ bind(exit);
4345 }
4347 __ bne(c_rarg0, R0, allocated);
4348 __ delayed()->nop();
4350 // allocate one if there's no free slot
4351 {
4352 Label entry, loop;
4353 // 1. compute new pointers // SP: old expression stack top
4354 __ ld(c_rarg0, monitor_block_top);
4355 __ daddi(SP, SP, - entry_size);
4356 __ daddi(c_rarg0, c_rarg0, - entry_size);
4357 __ sd(c_rarg0, monitor_block_top);
4358 __ b(entry);
4359 __ delayed(); __ move(T3, SP);
4361 // 2. move expression stack contents
4362 __ bind(loop);
4363 __ ld(AT, T3, entry_size);
4364 __ sd(AT, T3, 0);
4365 __ daddi(T3, T3, wordSize);
4366 __ bind(entry);
4367 __ bne(T3, c_rarg0, loop);
4368 __ delayed()->nop();
4369 }
4371 __ bind(allocated);
4372 // Increment bcp to point to the next bytecode,
4373 // so exception handling for async. exceptions work correctly.
4374 // The object has already been poped from the stack, so the
4375 // expression stack looks correct.
4376 __ daddi(BCP, BCP, 1);
4377 __ sd(FSR, c_rarg0, BasicObjectLock::obj_offset_in_bytes());
4378 __ lock_object(c_rarg0);
4379 // check to make sure this monitor doesn't cause stack overflow after locking
4380 __ save_bcp(); // in case of exception
4381 __ generate_stack_overflow_check(0);
4382 // The bcp has already been incremented. Just need to dispatch to next instruction.
4384 __ dispatch_next(vtos);
4385 }
4387 // T2 : top
4388 // c_rarg0 : entry
4389 void TemplateTable::monitorexit() {
4390 transition(atos, vtos);
4392 __ null_check(FSR);
4394 const int entry_size =(frame::interpreter_frame_monitor_size()* wordSize);
4395 Label found;
4397 // find matching slot
4398 {
4399 Label entry, loop;
4400 __ ld(c_rarg0, FP, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4401 __ b(entry);
4402 __ delayed()->daddiu(T2, FP, frame::interpreter_frame_initial_sp_offset * wordSize);
4404 __ bind(loop);
4405 __ ld(AT, c_rarg0, BasicObjectLock::obj_offset_in_bytes());
4406 __ beq(FSR, AT, found);
4407 __ delayed()->nop();
4408 __ daddiu(c_rarg0, c_rarg0, entry_size);
4409 __ bind(entry);
4410 __ bne(T2, c_rarg0, loop);
4411 __ delayed()->nop();
4412 }
4414 // error handling. Unlocking was not block-structured
4415 Label end;
4416 __ call_VM(NOREG, CAST_FROM_FN_PTR(address,
4417 InterpreterRuntime::throw_illegal_monitor_state_exception));
4418 __ should_not_reach_here();
4420 // call run-time routine
4421 // c_rarg0: points to monitor entry
4422 __ bind(found);
4423 __ move(TSR, FSR);
4424 __ unlock_object(c_rarg0);
4425 __ move(FSR, TSR);
4426 __ bind(end);
4427 }
4430 // Wide instructions
4431 void TemplateTable::wide() {
4432 transition(vtos, vtos);
4433 // Note: the esi increment step is part of the individual wide bytecode implementations
4434 __ lbu(Rnext, at_bcp(1));
4435 __ dsll(T9, Rnext, Address::times_8);
4436 __ li(AT, (long)Interpreter::_wentry_point);
4437 __ dadd(AT, T9, AT);
4438 __ ld(T9, AT, 0);
4439 __ jr(T9);
4440 __ delayed()->nop();
4441 }
4444 void TemplateTable::multianewarray() {
4445 transition(vtos, atos);
4446 // last dim is on top of stack; we want address of first one:
4447 // first_addr = last_addr + (ndims - 1) * wordSize
4448 __ lbu(A1, at_bcp(3)); // dimension
4449 __ daddi(A1, A1, -1);
4450 __ dsll(A1, A1, Address::times_8);
4451 __ dadd(A1, SP, A1); // now A1 pointer to the count array on the stack
4452 call_VM(FSR, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), A1);
4453 __ lbu(AT, at_bcp(3));
4454 __ dsll(AT, AT, Address::times_8);
4455 __ dadd(SP, SP, AT);
4456 __ sync();
4457 }
4458 #endif // !CC_INTERP