Thu, 25 Aug 2016 22:31:58 +0800
#4428: Use gsdmult to optimize the lmul template of the interpreter.
I have run the test program on 3A2000 computer
Effects:
[loongson@localhost project]$ ./LmulTest.sh
before
time:3225ms
after
time:3115ms
The test java program:
public class LmulTest{
public static void main(String args[]){
int count = 10000000;
long startTime = System.currentTimeMillis();
//long startTime = System.nanoTime();
for(int i = 0; i < count; i++){
long a, b, c;
a = 12345678;
a++;
b = 87654321;
b++;
c = a * b;
//System.out.println(c);
}
long endTime = System.currentTimeMillis();
//long endTime = System.nanoTime();
System.out.println("time:" + (endTime - startTime) + "ms");
}
}
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
52 // we use t8 as the local variables pointer register, by yjl 6/27/2005
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); }
74 //FIXME , can not use dadd and dsll
75 /*
76 static inline Address iaddress(Register r) {
77 return Address(r14, r, Address::times_8, Interpreter::value_offset_in_bytes());
78 }
80 static inline Address laddress(Register r) {
81 return Address(r14, r, Address::times_8, Interpreter::local_offset_in_bytes(1));
82 }
84 static inline Address faddress(Register r) {
85 return iaddress(r);
86 }
88 static inline Address daddress(Register r) {
89 return laddress(r);
90 }
92 static inline Address aaddress(Register r) {
93 return iaddress(r);
94 }
95 */
97 static inline Address at_sp() { return Address(SP, 0); }
98 static inline Address at_sp_p1() { return Address(SP, 1 * wordSize); }
99 static inline Address at_sp_p2() { return Address(SP, 2 * wordSize); }
101 // At top of Java expression stack which may be different than esp(). It
102 // isn't for category 1 objects.
103 static inline Address at_tos () {
104 Address tos = Address(SP, Interpreter::expr_offset_in_bytes(0));
105 return tos;
106 }
108 static inline Address at_tos_p1() {
109 return Address(SP, Interpreter::expr_offset_in_bytes(1));
110 }
112 static inline Address at_tos_p2() {
113 return Address(SP, Interpreter::expr_offset_in_bytes(2));
114 }
116 static inline Address at_tos_p3() {
117 return Address(SP, Interpreter::expr_offset_in_bytes(3));
118 }
120 // we use S0 as bcp, be sure you have bcp in S0 before you call any of the Template generator
121 Address TemplateTable::at_bcp(int offset) {
122 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
123 return Address(BCP, offset);
124 }
126 #define callee_saved_register(R) assert((R>=S0 && R<=S7), "should use callee saved registers!")
128 // bytecode folding
129 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
130 Register tmp_reg,
131 bool load_bc_into_bc_reg,/*=true*/
132 int byte_no) {
133 if (!RewriteBytecodes) {
134 return;
135 }
137 Label L_patch_done;
138 switch (bc) {
139 case Bytecodes::_fast_aputfield:
140 case Bytecodes::_fast_bputfield:
141 case Bytecodes::_fast_cputfield:
142 case Bytecodes::_fast_dputfield:
143 case Bytecodes::_fast_fputfield:
144 case Bytecodes::_fast_iputfield:
145 case Bytecodes::_fast_lputfield:
146 case Bytecodes::_fast_sputfield:
147 {
148 // We skip bytecode quickening for putfield instructions when the put_code written to the constant pool cache
149 // is zero. This is required so that every execution of this instruction calls out to
150 // InterpreterRuntime::resolve_get_put to do additional, required work.
151 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
152 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
153 __ get_cache_and_index_and_bytecode_at_bcp(tmp_reg, bc_reg, tmp_reg, byte_no, 1);
154 __ daddi(bc_reg, R0, bc);
155 __ beq(tmp_reg, R0, L_patch_done);
156 __ delayed()->nop();
157 }
158 break;
159 default:
160 assert(byte_no == -1, "sanity");
161 // the pair bytecodes have already done the load.
162 if (load_bc_into_bc_reg) {
163 __ move(bc_reg, bc);
164 }
166 }
167 if (JvmtiExport::can_post_breakpoint()) {
168 Label L_fast_patch;
169 // if a breakpoint is present we can't rewrite the stream directly
170 __ lbu(tmp_reg, at_bcp(0));
171 __ move(AT, Bytecodes::_breakpoint);
172 __ bne(tmp_reg, AT, L_fast_patch);
173 __ delayed()->nop();
175 __ get_method(tmp_reg);
176 // Let breakpoint table handling rewrite to quicker bytecode
177 __ call_VM(NOREG, CAST_FROM_FN_PTR(address,
178 InterpreterRuntime::set_original_bytecode_at), tmp_reg, BCP, bc_reg);
180 __ b(L_patch_done);
181 __ delayed()->nop();
182 __ bind(L_fast_patch);
183 }
185 #ifdef ASSERT
186 Label L_okay;
187 __ lbu(tmp_reg, at_bcp(0));
188 __ move(AT, (int)Bytecodes::java_code(bc));
189 __ beq(tmp_reg, AT, L_okay);
190 __ delayed()->nop();
191 __ beq(tmp_reg, bc_reg, L_patch_done);
192 __ delayed()->nop();
193 __ stop("patching the wrong bytecode");
194 __ bind(L_okay);
195 #endif
197 // patch bytecode
198 __ sb(bc_reg, at_bcp(0));
199 __ bind(L_patch_done);
200 }
203 // Individual instructions
205 void TemplateTable::nop() {
206 transition(vtos, vtos);
207 // nothing to do
208 }
210 void TemplateTable::shouldnotreachhere() {
211 transition(vtos, vtos);
212 __ stop("shouldnotreachhere bytecode");
213 }
215 void TemplateTable::aconst_null() {
216 transition(vtos, atos);
217 __ move(FSR, R0);
218 }
220 void TemplateTable::iconst(int value) {
221 transition(vtos, itos);
222 if (value == 0) {
223 __ move(FSR, R0);
224 } else {
225 __ move(FSR, value);
226 }
227 }
229 void TemplateTable::lconst(int value) {
230 transition(vtos, ltos);
231 if (value == 0) {
232 __ move(FSR, R0);
233 } else {
234 __ move(FSR, value);
235 }
236 assert(value >= 0, "check this code");
237 //__ move(SSR, R0);
238 }
240 void TemplateTable::fconst(int value) {
241 static float _f1 = 1.0, _f2 = 2.0;
242 transition(vtos, ftos);
243 float* p;
244 switch( value ) {
245 default: ShouldNotReachHere();
246 case 0: __ dmtc1(R0, FSF); return;
247 case 1: p = &_f1; break;
248 case 2: p = &_f2; break;
249 }
250 __ li(AT, (address)p);
251 __ lwc1(FSF, AT, 0);
252 }
254 void TemplateTable::dconst(int value) {
255 static double _d1 = 1.0;
256 transition(vtos, dtos);
257 double* p;
258 switch( value ) {
259 default: ShouldNotReachHere();
260 case 0: __ dmtc1(R0, FSF); return;
261 case 1: p = &_d1; break;
262 }
263 __ li(AT, (address)p);
264 __ ldc1(FSF, AT, 0);
265 }
267 void TemplateTable::bipush() {
268 transition(vtos, itos);
269 __ lb(FSR, at_bcp(1));
270 }
272 void TemplateTable::sipush() {
273 transition(vtos, itos);
274 __ get_2_byte_integer_at_bcp(FSR, AT, 1);
275 __ hswap(FSR);
276 }
278 // T1 : tags
279 // T2 : index
280 // T3 : cpool
281 // T8 : tag
282 void TemplateTable::ldc(bool wide) {
283 transition(vtos, vtos);
284 Label call_ldc, notFloat, notClass, Done;
285 // get index in cpool
286 if (wide) {
287 __ get_2_byte_integer_at_bcp(T2, AT, 1);
288 __ huswap(T2);
289 } else {
290 __ lbu(T2, at_bcp(1));
291 }
293 __ get_cpool_and_tags(T3, T1);
295 const int base_offset = ConstantPool::header_size() * wordSize;
296 const int tags_offset = Array<u1>::base_offset_in_bytes();
298 // get type
299 __ dadd(AT, T1, T2);
300 __ lb(T1, AT, tags_offset);
301 //now T1 is the tag
303 // unresolved string - get the resolved string
304 /*__ daddiu(AT, T1, - JVM_CONSTANT_UnresolvedString);
305 __ beq(AT, R0, call_ldc);
306 __ delayed()->nop();*/
308 // unresolved class - get the resolved class
309 __ daddiu(AT, T1, - JVM_CONSTANT_UnresolvedClass);
310 __ beq(AT, R0, call_ldc);
311 __ delayed()->nop();
313 // unresolved class in error (resolution failed) - call into runtime
314 // so that the same error from first resolution attempt is thrown.
315 __ daddiu(AT, T1, -JVM_CONSTANT_UnresolvedClassInError);
316 __ beq(AT, R0, call_ldc);
317 __ delayed()->nop();
319 // resolved class - need to call vm to get java mirror of the class
320 __ daddiu(AT, T1, - JVM_CONSTANT_Class);
321 __ bne(AT, R0, notClass);
322 __ delayed()->dsll(T2, T2, Address::times_8);
324 __ bind(call_ldc);
326 __ move(A1, wide);
327 call_VM(FSR, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), A1);
328 // __ sw(FSR, SP, - 1 * wordSize);
329 __ push(atos);
330 __ b(Done);
331 // __ delayed()->daddi(SP, SP, - 1 * wordSize);
332 __ delayed()->nop();
333 __ bind(notClass);
335 __ daddiu(AT, T1, -JVM_CONSTANT_Float);
336 __ bne(AT, R0, notFloat);
337 __ delayed()->nop();
338 // ftos
339 __ dadd(AT, T3, T2);
340 __ lwc1(FSF, AT, base_offset);
341 __ push_f();
342 __ b(Done);
343 __ delayed()->nop();
345 __ bind(notFloat);
346 #ifdef ASSERT
347 {
348 Label L;
349 __ daddiu(AT, T1, -JVM_CONSTANT_Integer);
350 __ beq(AT, R0, L);
351 __ delayed()->nop();
352 __ stop("unexpected tag type in ldc");
353 __ bind(L);
354 }
355 #endif
356 // atos and itos
357 __ dadd(T0, T3, T2);
358 __ lw(FSR, T0, base_offset);
359 __ push(itos);
360 __ b(Done);
361 __ delayed()->nop();
364 if (VerifyOops) {
365 __ verify_oop(FSR);
366 }
368 __ bind(Done);
369 }
371 // Fast path for caching oop constants.
372 void TemplateTable::fast_aldc(bool wide) {
373 transition(vtos, atos);
375 Register result = FSR;
376 Register tmp = SSR;
377 int index_size = wide ? sizeof(u2) : sizeof(u1);
379 Label resolved;
380 // We are resolved if the resolved reference cache entry contains a
381 // non-null object (String, MethodType, etc.)
382 assert_different_registers(result, tmp);
383 __ get_cache_index_at_bcp(tmp, 1, index_size);
384 __ load_resolved_reference_at_index(result, tmp);
385 __ bne(result, R0, resolved);
386 __ delayed()->nop();
388 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
389 // first time invocation - must resolve first
390 int i = (int)bytecode();
391 __ move(tmp, i);
392 __ call_VM(result, entry, tmp);
394 __ bind(resolved);
396 if (VerifyOops) {
397 __ verify_oop(result);
398 }
399 }
402 // used register: T2, T3, T1
403 // T2 : index
404 // T3 : cpool
405 // T1 : tag
406 void TemplateTable::ldc2_w() {
407 transition(vtos, vtos);
408 Label Long, Done;
410 // get index in cpool
411 __ get_2_byte_integer_at_bcp(T2, AT, 1);
412 __ huswap(T2);
414 __ get_cpool_and_tags(T3, T1);
416 const int base_offset = ConstantPool::header_size() * wordSize;
417 const int tags_offset = Array<u1>::base_offset_in_bytes();
419 // get type in T1
420 __ dadd(AT, T1, T2);
421 __ lb(T1, AT, tags_offset);
423 __ daddiu(AT, T1, - JVM_CONSTANT_Double);
424 __ bne(AT, R0, Long);
425 __ delayed()->dsll(T2, T2, Address::times_8);
426 // dtos
427 __ daddu(AT, T3, T2);
428 __ ldc1(FSF, AT, base_offset + 0 * wordSize);
429 __ sdc1(FSF, SP, - 2 * wordSize);
430 __ b(Done);
431 __ delayed()->daddi(SP, SP, - 2 * wordSize);
433 // ltos
434 __ bind(Long);
435 __ dadd(AT, T3, T2);
436 __ ld(FSR, AT, base_offset + 0 * wordSize);
437 __ push(ltos);
439 __ bind(Done);
440 }
442 // we compute the actual local variable address here
443 // the x86 dont do so for it has scaled index memory access model, we dont have, so do here
444 void TemplateTable::locals_index(Register reg, int offset) {
445 __ lbu(reg, at_bcp(offset));
446 __ dsll(reg, reg, Address::times_8);
447 __ dsub(reg, LVP, reg);
448 }
450 // this method will do bytecode folding of the two form:
451 // iload iload iload caload
452 // used register : T2, T3
453 // T2 : bytecode
454 // T3 : folded code
455 void TemplateTable::iload() {
456 transition(vtos, itos);
457 if (RewriteFrequentPairs) {
458 Label rewrite, done;
459 // get the next bytecode in T2
460 __ lbu(T2, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
461 // if _iload, wait to rewrite to iload2. We only want to rewrite the
462 // last two iloads in a pair. Comparing against fast_iload means that
463 // the next bytecode is neither an iload or a caload, and therefore
464 // an iload pair.
465 __ move(AT, Bytecodes::_iload);
466 __ beq(AT, T2, done);
467 __ delayed()->nop();
469 __ move(T3, Bytecodes::_fast_iload2);
470 __ move(AT, Bytecodes::_fast_iload);
471 __ beq(AT, T2, rewrite);
472 __ delayed()->nop();
474 // if _caload, rewrite to fast_icaload
475 __ move(T3, Bytecodes::_fast_icaload);
476 __ move(AT, Bytecodes::_caload);
477 __ beq(AT, T2, rewrite);
478 __ delayed()->nop();
480 // rewrite so iload doesn't check again.
481 __ move(T3, Bytecodes::_fast_iload);
483 // rewrite
484 // T3 : fast bytecode
485 __ bind(rewrite);
486 patch_bytecode(Bytecodes::_iload, T3, T2, false);
487 __ bind(done);
488 }
490 // Get the local value into tos
491 locals_index(T2);
492 __ lw(FSR, T2, 0);
493 }
495 // used register T2
496 // T2 : index
497 void TemplateTable::fast_iload2() {
498 transition(vtos, itos);
499 locals_index(T2);
500 __ lw(FSR, T2, 0);
501 __ push(itos);
502 locals_index(T2, 3);
503 __ lw(FSR, T2, 0);
504 }
506 // used register T2
507 // T2 : index
508 void TemplateTable::fast_iload() {
509 transition(vtos, itos);
510 locals_index(T2);
511 __ lw(FSR, T2, 0);
512 }
514 // used register T2
515 // T2 : index
516 void TemplateTable::lload() {
518 transition(vtos, ltos);
519 locals_index(T2);
520 __ ld(FSR, T2, -wordSize);
521 __ ld(SSR, T2, 0);
522 }
524 // used register T2
525 // T2 : index
526 void TemplateTable::fload() {
527 transition(vtos, ftos);
528 locals_index(T2);
529 //FIXME, aoqi. How should the high 32bits be when store a single float into a 64bits register.
530 //__ mtc1(R0, FSF);
531 __ lwc1(FSF, T2, 0);
532 }
534 // used register T2
535 // T2 : index
536 void TemplateTable::dload() {
538 transition(vtos, dtos);
539 locals_index(T2);
540 /* if (TaggedStackInterpreter) {
541 // Get double out of locals array, onto temp stack and load with
542 // float instruction into ST0
543 __ dsll(AT,T2,Interpreter::stackElementScale());
544 __ dadd(AT, LVP, AT);
545 __ ldc1(FSF, AT, Interpreter::local_offset_in_bytes(1));
546 } else {*/
547 __ ldc1(FSF, T2, -wordSize);
548 __ ldc1(SSF, T2, 0);
549 // }
550 }
552 // used register T2
553 // T2 : index
554 void TemplateTable::aload()
555 {
556 transition(vtos, atos);
557 locals_index(T2);
558 __ ld(FSR, T2, 0);
559 }
561 void TemplateTable::locals_index_wide(Register reg) {
562 __ get_2_byte_integer_at_bcp(reg, AT, 2);
563 __ huswap(reg);
564 __ dsll(reg, reg, Address::times_8);
565 __ dsub(reg, LVP, reg);
566 }
568 // used register T2
569 // T2 : index
570 void TemplateTable::wide_iload() {
571 transition(vtos, itos);
572 locals_index_wide(T2);
573 __ ld(FSR, T2, 0);
574 }
576 // used register T2
577 // T2 : index
578 void TemplateTable::wide_lload() {
579 transition(vtos, ltos);
580 locals_index_wide(T2);
581 __ ld(FSR, T2, -4);
582 }
584 // used register T2
585 // T2 : index
586 void TemplateTable::wide_fload() {
587 transition(vtos, ftos);
588 locals_index_wide(T2);
589 __ lwc1(FSF, T2, 0);
590 }
592 // used register T2
593 // T2 : index
594 void TemplateTable::wide_dload() {
595 transition(vtos, dtos);
596 locals_index_wide(T2);
597 /* if (TaggedStackInterpreter) {
598 // Get double out of locals array, onto temp stack and load with
599 // float instruction into ST0
600 // __ movl(eax, laddress(ebx));
601 // __ movl(edx, haddress(ebx));
602 __ dsll(AT,T2,Interpreter::stackElementScale());
603 __ dadd(AT, LVP, AT);
604 __ ldc1(FSF, AT, Interpreter::local_offset_in_bytes(1));
606 // __ pushl(edx); // push hi first
607 // __ pushl(eax);
608 // __ fld_d(Address(esp));
609 // __ addl(esp, 2*wordSize);
610 } else {*/
611 __ ldc1(FSF, T2, -4);
612 //}
613 }
615 // used register T2
616 // T2 : index
617 void TemplateTable::wide_aload() {
618 transition(vtos, atos);
619 locals_index_wide(T2);
620 __ ld(FSR, T2, 0);
621 }
623 // we use A2 as the regiser for index, BE CAREFUL!
624 // we dont use our tge 29 now, for later optimization
625 void TemplateTable::index_check(Register array, Register index) {
626 // Pop ptr into array
627 __ pop_ptr(array);
628 index_check_without_pop(array, index);
629 }
631 void TemplateTable::index_check_without_pop(Register array, Register index) {
632 // destroys ebx
633 // check array
634 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
636 // check index
637 Label ok;
638 __ lw(AT, array, arrayOopDesc::length_offset_in_bytes());
639 #ifndef OPT_RANGECHECK
640 __ sltu(AT, index, AT);
641 __ bne(AT, R0, ok);
642 __ delayed()->nop();
644 //throw_ArrayIndexOutOfBoundsException assume abberrant index in A2
645 if (A2 != index) __ move(A2, index);
646 __ jmp(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
647 __ delayed()->nop();
648 __ bind(ok);
649 #else
650 __ lw(AT, array, arrayOopDesc::length_offset_in_bytes());
651 __ move(A2, index);
652 __ tgeu(A2, AT, 29);
653 #endif
654 }
656 void TemplateTable::iaload() {
657 transition(itos, itos);
658 // __ pop(SSR);
659 index_check(SSR, FSR);
660 __ dsll(FSR, FSR, 2);
661 __ dadd(FSR, SSR, FSR);
662 //FSR: index
663 __ lw(FSR, FSR, arrayOopDesc::base_offset_in_bytes(T_INT));
664 }
667 void TemplateTable::laload() {
668 transition(itos, ltos);
669 // __ pop(SSR);
670 index_check(SSR, FSR);
671 __ dsll(AT, FSR, Address::times_8);
672 __ dadd(AT, SSR, AT);
673 __ ld(FSR, AT, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize);
674 }
676 void TemplateTable::faload() {
677 transition(itos, ftos);
678 // __ pop(SSR);
679 index_check(SSR, FSR);
680 __ shl(FSR, 2);
681 __ dadd(FSR, SSR, FSR);
682 __ lwc1(FSF, FSR, arrayOopDesc::base_offset_in_bytes(T_FLOAT));
683 }
685 void TemplateTable::daload() {
686 transition(itos, dtos);
687 //__ pop(SSR);
688 index_check(SSR, FSR);
689 __ dsll(AT, FSR, 3);
690 __ dadd(AT, SSR, AT);
691 __ ldc1(FSF, AT, arrayOopDesc::base_offset_in_bytes(T_DOUBLE) + 0 * wordSize);
692 }
694 void TemplateTable::aaload() {
695 transition(itos, atos);
696 //__ pop(SSR);
697 index_check(SSR, FSR);
698 __ dsll(FSR, FSR, UseCompressedOops ? Address::times_4 : Address::times_8);
699 __ dadd(FSR, SSR, FSR);
700 //add for compressedoops
701 __ load_heap_oop(FSR, Address(FSR, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
702 }
704 void TemplateTable::baload() {
705 transition(itos, itos);
706 //__ pop(SSR);
707 index_check(SSR, FSR);
708 __ dadd(FSR, SSR, FSR);
709 __ lb(FSR, FSR, arrayOopDesc::base_offset_in_bytes(T_BYTE));
710 }
712 void TemplateTable::caload() {
713 transition(itos, itos);
714 // __ pop(SSR);
715 index_check(SSR, FSR);
716 __ dsll(FSR, FSR, Address::times_2);
717 __ dadd(FSR, SSR, FSR);
718 __ lhu(FSR, FSR, arrayOopDesc::base_offset_in_bytes(T_CHAR));
719 }
721 // iload followed by caload frequent pair
722 // used register : T2
723 // T2 : index
724 void TemplateTable::fast_icaload() {
725 transition(vtos, itos);
726 // load index out of locals
727 locals_index(T2);
728 __ lw(FSR, T2, 0);
729 // __ pop(SSR);
730 index_check(SSR, FSR);
731 __ dsll(FSR, FSR, 1);
732 __ dadd(FSR, SSR, FSR);
733 __ lhu(FSR, FSR, arrayOopDesc::base_offset_in_bytes(T_CHAR));
734 }
736 void TemplateTable::saload() {
737 transition(itos, itos);
738 // __ pop(SSR);
739 index_check(SSR, FSR);
740 __ dsll(FSR, FSR, Address::times_2);
741 __ dadd(FSR, SSR, FSR);
742 __ lh(FSR, FSR, arrayOopDesc::base_offset_in_bytes(T_SHORT));
743 }
745 void TemplateTable::iload(int n) {
746 transition(vtos, itos);
747 __ lw(FSR, iaddress(n));
748 }
750 void TemplateTable::lload(int n) {
751 transition(vtos, ltos);
752 __ ld(FSR, laddress(n));
753 }
755 void TemplateTable::fload(int n) {
756 transition(vtos, ftos);
757 //__ mtc1(R0, FSF);
758 __ lwc1(FSF, faddress(n));
759 }
760 //FIXME here
761 void TemplateTable::dload(int n) {
762 transition(vtos, dtos);
763 __ ldc1(FSF, laddress(n));
764 }
766 void TemplateTable::aload(int n) {
767 transition(vtos, atos);
768 __ ld(FSR, aaddress(n));
769 }
771 // used register : T2, T3
772 // T2 : bytecode
773 // T3 : folded code
774 void TemplateTable::aload_0() {
775 transition(vtos, atos);
776 // According to bytecode histograms, the pairs:
777 //
778 // _aload_0, _fast_igetfield
779 // _aload_0, _fast_agetfield
780 // _aload_0, _fast_fgetfield
781 //
782 // occur frequently. If RewriteFrequentPairs is set, the (slow) _aload_0
783 // bytecode checks if the next bytecode is either _fast_igetfield,
784 // _fast_agetfield or _fast_fgetfield and then rewrites the
785 // current bytecode into a pair bytecode; otherwise it rewrites the current
786 // bytecode into _fast_aload_0 that doesn't do the pair check anymore.
787 //
788 // Note: If the next bytecode is _getfield, the rewrite must be delayed,
789 // otherwise we may miss an opportunity for a pair.
790 //
791 // Also rewrite frequent pairs
792 // aload_0, aload_1
793 // aload_0, iload_1
794 // These bytecodes with a small amount of code are most profitable to rewrite
795 if (RewriteFrequentPairs) {
796 Label rewrite, done;
797 // get the next bytecode in T2
798 __ lbu(T2, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
800 // do actual aload_0
801 aload(0);
803 // if _getfield then wait with rewrite
804 __ move(AT, Bytecodes::_getfield);
805 __ beq(AT, T2, done);
806 __ delayed()->nop();
808 // if _igetfield then reqrite to _fast_iaccess_0
809 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) ==
810 Bytecodes::_aload_0, "fix bytecode definition");
811 __ move(T3, Bytecodes::_fast_iaccess_0);
812 __ move(AT, Bytecodes::_fast_igetfield);
813 __ beq(AT, T2, rewrite);
814 __ delayed()->nop();
816 // if _agetfield then reqrite to _fast_aaccess_0
817 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) ==
818 Bytecodes::_aload_0, "fix bytecode definition");
819 __ move(T3, Bytecodes::_fast_aaccess_0);
820 __ move(AT, Bytecodes::_fast_agetfield);
821 __ beq(AT, T2, rewrite);
822 __ delayed()->nop();
824 // if _fgetfield then reqrite to _fast_faccess_0
825 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) ==
826 Bytecodes::_aload_0, "fix bytecode definition");
827 __ move(T3, Bytecodes::_fast_faccess_0);
828 __ move(AT, Bytecodes::_fast_fgetfield);
829 __ beq(AT, T2, rewrite);
830 __ delayed()->nop();
832 // else rewrite to _fast_aload0
833 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) ==
834 Bytecodes::_aload_0, "fix bytecode definition");
835 __ move(T3, Bytecodes::_fast_aload_0);
837 // rewrite
838 __ bind(rewrite);
839 patch_bytecode(Bytecodes::_aload_0, T3, T2, false);
841 __ bind(done);
842 } else {
843 aload(0);
844 }
845 }
847 void TemplateTable::istore() {
848 transition(itos, vtos);
849 locals_index(T2);
850 __ sw(FSR, T2, 0);
851 }
853 void TemplateTable::lstore() {
854 transition(ltos, vtos);
855 locals_index(T2);
856 __ sd(FSR, T2, -wordSize);
857 }
859 void TemplateTable::fstore() {
860 transition(ftos, vtos);
861 locals_index(T2);
862 __ swc1(FSF, T2, 0);
863 }
865 void TemplateTable::dstore() {
866 transition(dtos, vtos);
867 locals_index(T2);
868 __ sdc1(FSF, T2, -wordSize);
869 }
871 void TemplateTable::astore() {
872 transition(vtos, vtos);
873 // __ pop(FSR);
874 __ pop_ptr(FSR);
875 locals_index(T2);
876 __ sd(FSR, T2, 0);
877 }
879 void TemplateTable::wide_istore() {
880 transition(vtos, vtos);
881 // __ pop(FSR);
882 __ pop_i(FSR);
883 locals_index_wide(T2);
884 __ sd(FSR, T2, 0);
885 }
887 void TemplateTable::wide_lstore() {
888 transition(vtos, vtos);
889 //__ pop2(FSR, SSR);
890 //__ pop_l(FSR, SSR);
891 __ pop_l(FSR); //aoqi:FIXME Is this right?
892 locals_index_wide(T2);
893 __ sd(FSR, T2, -4);
894 }
896 void TemplateTable::wide_fstore() {
897 wide_istore();
898 }
900 void TemplateTable::wide_dstore() {
901 wide_lstore();
902 }
904 void TemplateTable::wide_astore() {
905 transition(vtos, vtos);
906 __ pop_ptr(FSR);
907 locals_index_wide(T2);
908 __ sd(FSR, T2, 0);
909 }
911 // used register : T2
912 void TemplateTable::iastore() {
913 transition(itos, vtos);
914 __ pop_i(SSR);
915 index_check(T2, SSR); // prefer index in ebx
916 __ dsll(SSR, SSR, Address::times_4);
917 __ dadd(T2, T2, SSR);
918 __ sw(FSR, T2, arrayOopDesc::base_offset_in_bytes(T_INT));
919 }
923 // used register T2, T3
924 void TemplateTable::lastore() {
925 transition(ltos, vtos);
926 __ pop_i (T2);
927 index_check(T3, T2);
928 __ dsll(T2, T2, Address::times_8);
929 __ dadd(T3, T3, T2);
930 __ sd(FSR, T3, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize);
931 }
933 // used register T2
934 void TemplateTable::fastore() {
935 transition(ftos, vtos);
936 __ pop_i(SSR);
937 index_check(T2, SSR);
938 __ dsll(SSR, SSR, Address::times_4);
939 __ dadd(T2, T2, SSR);
940 __ swc1(FSF, T2, arrayOopDesc::base_offset_in_bytes(T_FLOAT));
941 }
943 // used register T2, T3
944 void TemplateTable::dastore() {
945 transition(dtos, vtos);
946 __ pop_i (T2);
947 index_check(T3, T2);
948 __ dsll(T2, T2, Address::times_8);
949 __ daddu(T3, T3, T2);
950 __ sdc1(FSF, T3, arrayOopDesc::base_offset_in_bytes(T_DOUBLE) + 0 * wordSize);
952 }
954 // used register : T2, T3, T8
955 // T2 : array
956 // T3 : subklass
957 // T8 : supklass
958 void TemplateTable::aastore() {
959 Label is_null, ok_is_subtype, done;
960 transition(vtos, vtos);
961 // stack: ..., array, index, value
962 __ ld(FSR, at_tos()); // Value
963 __ lw(SSR, at_tos_p1()); // Index
964 __ ld(T2, at_tos_p2()); // Array
966 // index_check(T2, SSR);
967 index_check_without_pop(T2, SSR);
968 // do array store check - check for NULL value first
969 __ beq(FSR, R0, is_null);
970 __ delayed()->nop();
972 // Move subklass into T3
973 //__ ld(T3, Address(FSR, oopDesc::klass_offset_in_bytes()));
974 //add for compressedoops
975 __ load_klass(T3, FSR);
976 // Move superklass into T8
977 //__ ld(T8, Address(T2, oopDesc::klass_offset_in_bytes()));
978 //add for compressedoops
979 __ load_klass(T8, T2);
980 __ ld(T8, Address(T8, ObjArrayKlass::element_klass_offset()));
981 // Compress array+index*4+12 into a single register. T2
982 __ dsll(AT, SSR, UseCompressedOops? Address::times_4 : Address::times_8);
983 __ dadd(T2, T2, AT);
984 __ daddi(T2, T2, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
986 // Generate subtype check.
987 // Superklass in T8. Subklass in T3.
988 __ gen_subtype_check(T8, T3, ok_is_subtype); // <-- Jin
989 // Come here on failure
990 // object is at FSR
991 __ jmp(Interpreter::_throw_ArrayStoreException_entry); // <-- Jin
992 __ delayed()->nop();
993 // Come here on success
994 __ bind(ok_is_subtype);
995 //replace with do_oop_store->store_heap_oop
996 //__ sd(FSR, T2, 0);
997 __ store_heap_oop(Address(T2, 0), FSR); // <-- Jin
998 __ sync();
999 __ store_check(T2);
1000 __ b(done);
1001 __ delayed()->nop();
1003 // Have a NULL in FSR, EDX=T2, SSR=index. Store NULL at ary[idx]
1004 __ bind(is_null);
1005 __ profile_null_seen(T9);
1006 __ dsll(AT, SSR, UseCompressedOops? Address::times_4 : Address::times_8);
1007 __ dadd(T2, T2, AT);
1008 //__ sd(FSR, T2, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1009 __ store_heap_oop(Address(T2, arrayOopDesc::base_offset_in_bytes(T_OBJECT)), FSR); /* FSR is null here */
1010 __ sync();
1012 __ bind(done);
1013 __ daddi(SP, SP, 3 * Interpreter::stackElementSize);
1014 }
1016 void TemplateTable::bastore() {
1017 transition(itos, vtos);
1018 __ pop_i (SSR);
1019 index_check(T2, SSR);
1020 __ dadd(SSR, T2, SSR);
1021 __ sb(FSR, SSR, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1022 }
1024 void TemplateTable::castore() {
1025 transition(itos, vtos);
1026 __ pop_i(SSR);
1027 index_check(T2, SSR);
1028 __ dsll(SSR, SSR, Address::times_2);
1029 __ dadd(SSR, T2, SSR);
1030 __ sh(FSR, SSR, arrayOopDesc::base_offset_in_bytes(T_CHAR));
1031 }
1033 void TemplateTable::sastore() {
1034 castore();
1035 }
1037 void TemplateTable::istore(int n) {
1038 transition(itos, vtos);
1039 __ sw(FSR, iaddress(n));
1040 }
1042 void TemplateTable::lstore(int n) {
1043 transition(ltos, vtos);
1044 __ sd(FSR, laddress(n));
1045 }
1047 void TemplateTable::fstore(int n) {
1048 transition(ftos, vtos);
1049 __ swc1(FSF, faddress(n));
1050 }
1052 void TemplateTable::dstore(int n) {
1053 transition(dtos, vtos);
1054 __ sdc1(FSF, laddress(n));
1055 }
1057 void TemplateTable::astore(int n) {
1058 transition(vtos, vtos);
1059 __ pop_ptr(FSR);
1060 __ sd(FSR, aaddress(n));
1061 }
1063 void TemplateTable::pop() {
1064 transition(vtos, vtos);
1065 __ daddi(SP, SP, Interpreter::stackElementSize);
1066 }
1068 void TemplateTable::pop2() {
1069 transition(vtos, vtos);
1070 __ daddi(SP, SP, 2 * Interpreter::stackElementSize);
1071 }
1073 void TemplateTable::dup() {
1074 transition(vtos, vtos);
1075 // stack: ..., a
1076 __ load_ptr(0, FSR);
1077 __ push_ptr(FSR);
1078 // stack: ..., a, a
1079 }
1081 // blows FSR
1082 void TemplateTable::dup_x1() {
1083 transition(vtos, vtos);
1084 // stack: ..., a, b
1085 __ load_ptr(0, FSR); // load b
1086 __ load_ptr(1, A5); // load a
1087 __ store_ptr(1, FSR); // store b
1088 __ store_ptr(0, A5); // store a
1089 __ push_ptr(FSR); // push b
1090 // stack: ..., b, a, b
1091 }
1093 // blows FSR
1094 void TemplateTable::dup_x2() {
1095 transition(vtos, vtos);
1096 // stack: ..., a, b, c
1097 __ load_ptr(0, FSR); // load c
1098 __ load_ptr(2, A5); // load a
1099 __ store_ptr(2, FSR); // store c in a
1100 __ push_ptr(FSR); // push c
1101 // stack: ..., c, b, c, c
1102 __ load_ptr(2, FSR); // load b
1103 __ store_ptr(2, A5); // store a in b
1104 // stack: ..., c, a, c, c
1105 __ store_ptr(1, FSR); // store b in c
1106 // stack: ..., c, a, b, c
1107 }
1109 // blows FSR
1110 void TemplateTable::dup2() {
1111 transition(vtos, vtos);
1112 // stack: ..., a, b
1113 __ load_ptr(1, FSR); // load a
1114 __ push_ptr(FSR); // push a
1115 __ load_ptr(1, FSR); // load b
1116 __ push_ptr(FSR); // push b
1117 // stack: ..., a, b, a, b
1118 }
1120 // blows FSR
1121 void TemplateTable::dup2_x1() {
1122 transition(vtos, vtos);
1123 // stack: ..., a, b, c
1124 __ load_ptr(0, T2); // load c
1125 __ load_ptr(1, FSR); // load b
1126 __ push_ptr(FSR); // push b
1127 __ push_ptr(T2); // push c
1128 // stack: ..., a, b, c, b, c
1129 __ store_ptr(3, T2); // store c in b
1130 // stack: ..., a, c, c, b, c
1131 __ load_ptr(4, T2); // load a
1132 __ store_ptr(2, T2); // store a in 2nd c
1133 // stack: ..., a, c, a, b, c
1134 __ store_ptr(4, FSR); // store b in a
1135 // stack: ..., b, c, a, b, c
1137 // stack: ..., b, c, a, b, c
1138 }
1140 // blows FSR, SSR
1141 void TemplateTable::dup2_x2() {
1142 transition(vtos, vtos);
1143 // stack: ..., a, b, c, d
1144 // stack: ..., a, b, c, d
1145 __ load_ptr(0, T2); // load d
1146 __ load_ptr(1, FSR); // load c
1147 __ push_ptr(FSR); // push c
1148 __ push_ptr(T2); // push d
1149 // stack: ..., a, b, c, d, c, d
1150 __ load_ptr(4, FSR); // load b
1151 __ store_ptr(2, FSR); // store b in d
1152 __ store_ptr(4, T2); // store d in b
1153 // stack: ..., a, d, c, b, c, d
1154 __ load_ptr(5, T2); // load a
1155 __ load_ptr(3, FSR); // load c
1156 __ store_ptr(3, T2); // store a in c
1157 __ store_ptr(5, FSR); // store c in a
1158 // stack: ..., c, d, a, b, c, d
1160 // stack: ..., c, d, a, b, c, d
1161 }
1163 // blows FSR
1164 void TemplateTable::swap() {
1165 transition(vtos, vtos);
1166 // stack: ..., a, b
1168 __ load_ptr(1, A5); // load a
1169 __ load_ptr(0, FSR); // load b
1170 __ store_ptr(0, A5); // store a in b
1171 __ store_ptr(1, FSR); // store b in a
1173 // stack: ..., b, a
1174 }
1176 void TemplateTable::iop2(Operation op) {
1177 transition(itos, itos);
1178 switch (op) {
1179 case add :
1180 __ pop_i(SSR);
1181 __ addu32(FSR, SSR, FSR);
1182 break;
1183 case sub :
1184 __ pop_i(SSR);
1185 __ subu32(FSR, SSR, FSR);
1186 break;
1187 case mul :
1188 __ lw(SSR, SP, 0);
1189 __ daddi(SP, SP, wordSize);
1190 __ mul(FSR, SSR, FSR);
1191 break;
1192 case _and :
1193 __ pop_i(SSR);
1194 __ andr(FSR, SSR, FSR);
1195 break;
1196 case _or :
1197 __ pop_i(SSR);
1198 __ orr(FSR, SSR, FSR);
1199 break;
1200 case _xor :
1201 __ pop_i(SSR);
1202 __ xorr(FSR, SSR, FSR);
1203 break;
1204 case shl :
1205 __ pop_i(SSR);
1206 __ sllv(FSR, SSR, FSR);
1207 break; // implicit masking of lower 5 bits by Intel shift instr. mips also
1208 case shr :
1209 __ pop_i(SSR);
1210 __ srav(FSR, SSR, FSR);
1211 break; // implicit masking of lower 5 bits by Intel shift instr. mips also
1212 case ushr :
1213 __ pop_i(SSR);
1214 __ srlv(FSR, SSR, FSR);
1215 break; // implicit masking of lower 5 bits by Intel shift instr. mips also
1216 default : ShouldNotReachHere();
1217 }
1218 }
1220 // the result stored in FSR, SSR,
1221 // used registers : T2, T3
1222 //FIXME, aoqi
1223 void TemplateTable::lop2(Operation op) {
1224 transition(ltos, ltos);
1225 //__ pop2(T2, T3);
1226 __ pop_l(T2, T3);
1227 #ifdef ASSERT
1228 {
1229 Label L;
1230 __ beq(T3, R0, L);
1231 __ delayed()->nop();
1232 // FIXME: stack verification required
1233 // __ stop("lop2, wrong stack"); // <--- Fu 20130930
1234 __ bind(L);
1235 }
1236 #endif
1237 switch (op) {
1238 case add :
1239 __ daddu(FSR, T2, FSR);
1240 //__ sltu(AT, FSR, T2);
1241 //__ daddu(SSR, T3, SSR);
1242 //__ daddu(SSR, SSR, AT);
1243 break;
1244 case sub :
1245 __ dsubu(FSR, T2, FSR);
1246 //__ sltu(AT, T2, FSR);
1247 //__ dsubu(SSR, T3, SSR);
1248 //__ dsubu(SSR, SSR, AT);
1249 break;
1250 case _and:
1251 __ andr(FSR, T2, FSR);
1252 //__ andr(SSR, T3, SSR);
1253 break;
1254 case _or :
1255 __ orr(FSR, T2, FSR);
1256 //__ orr(SSR, T3, SSR);
1257 break;
1258 case _xor:
1259 __ xorr(FSR, T2, FSR);
1260 //__ xorr(SSR, T3, SSR);
1261 break;
1262 default : ShouldNotReachHere();
1263 }
1264 }
1266 // java require this bytecode could handle 0x80000000/-1, dont cause a overflow exception,
1267 // the result is 0x80000000
1268 // the godson2 cpu do the same, so we need not handle this specially like x86
1269 void TemplateTable::idiv() {
1270 transition(itos, itos);
1271 Label not_zero;
1273 __ bne(FSR, R0, not_zero);
1274 __ delayed()->nop();
1275 __ jmp(Interpreter::_throw_ArithmeticException_entry);
1276 __ delayed()->nop();
1277 __ bind(not_zero);
1279 __ pop_i(SSR);
1280 if (UseLoongsonISA) {
1281 __ gsdiv(FSR, SSR, FSR);
1282 } else {
1283 __ div(SSR, FSR);
1284 __ mflo(FSR);
1285 }
1286 }
1288 void TemplateTable::irem() {
1289 transition(itos, itos);
1290 Label not_zero;
1291 //__ pop(SSR);
1292 __ pop_i(SSR);
1293 __ div(SSR, FSR);
1295 __ bne(FSR, R0, not_zero);
1296 __ delayed()->nop();
1297 //__ brk(7);
1298 __ jmp(Interpreter::_throw_ArithmeticException_entry);
1299 __ delayed()->nop();
1301 __ bind(not_zero);
1302 __ mfhi(FSR);
1303 }
1305 void TemplateTable::lmul() {
1306 transition(ltos, ltos);
1307 __ pop_l(T2);
1308 if(UseLoongsonISA){
1309 __ gsdmult(FSR, T2, FSR);
1310 } else {
1311 __ dmult(T2, FSR);
1312 __ mflo(FSR);
1313 }
1314 }
1316 // NOTE: i DONT use the Interpreter::_throw_ArithmeticException_entry
1317 void TemplateTable::ldiv() {
1318 transition(ltos, ltos);
1319 Label normal;
1321 __ bne(FSR, R0, normal);
1322 __ delayed()->nop();
1324 //__ brk(7); //generate FPE
1325 __ jmp(Interpreter::_throw_ArithmeticException_entry);
1326 __ delayed()->nop();
1328 __ bind(normal);
1329 __ pop_l(A2, A3);
1330 if (UseLoongsonISA) {
1331 __ gsddiv(FSR, A2, FSR);
1332 } else {
1333 __ ddiv(A2, FSR);
1334 __ mflo(FSR);
1335 }
1336 }
1338 // NOTE: i DONT use the Interpreter::_throw_ArithmeticException_entry
1339 void TemplateTable::lrem() {
1340 transition(ltos, ltos);
1341 Label normal;
1343 __ bne(FSR, R0, normal);
1344 __ delayed()->nop();
1346 __ jmp(Interpreter::_throw_ArithmeticException_entry);
1347 __ delayed()->nop();
1349 __ bind(normal);
1350 __ pop_l (A2, A3);
1352 if(UseLoongsonISA){
1353 __ gsdmod(FSR, A2, FSR);
1354 } else {
1355 __ ddiv(A2, FSR);
1356 __ mfhi(FSR);
1357 }
1358 }
1360 // result in FSR
1361 // used registers : T0
1362 void TemplateTable::lshl() {
1363 transition(itos, ltos);
1364 __ pop_l(T0, T1);
1365 #ifdef ASSERT
1366 {
1367 Label L;
1368 __ beq(T1, R0, L);
1369 __ delayed()->nop();
1370 //__ stop("lshl, wrong stack"); // <-- Fu 20130930
1371 __ bind(L);
1372 }
1373 #endif
1374 __ andi(FSR, FSR, 0x3f); // the bit to be shifted
1375 __ dsllv(FSR, T0, FSR);
1376 }
1378 // used registers : T0
1379 void TemplateTable::lshr() {
1380 transition(itos, ltos);
1381 __ pop_l(T0, T1);
1382 #ifdef ASSERT
1383 {
1384 Label L;
1385 __ beq(T1, R0, L);
1386 __ delayed()->nop();
1387 __ stop("lshr, wrong stack");
1388 __ bind(L);
1389 }
1390 #endif
1391 __ andi(FSR, FSR, 0x3f); // the bit to be shifted
1392 __ dsrav(FSR, T0, FSR);
1393 }
1395 // used registers : T0
1396 void TemplateTable::lushr() {
1397 transition(itos, ltos);
1398 __ pop_l(T0, T1);
1399 #ifdef ASSERT
1400 {
1401 Label L;
1402 __ beq(T1, R0, L);
1403 __ delayed()->nop();
1404 __ stop("lushr, wrong stack");
1405 __ bind(L);
1406 }
1407 #endif
1408 __ andi(FSR, FSR, 0x3f); // the bit to be shifted
1409 __ dsrlv(FSR, T0, FSR);
1410 }
1412 // result in FSF
1413 void TemplateTable::fop2(Operation op) {
1414 transition(ftos, ftos);
1415 __ pop_ftos_to_esp(); // pop ftos into esp
1416 switch (op) {
1417 case add:
1418 __ lwc1(FTF, at_sp());
1419 __ add_s(FSF, FTF, FSF);
1420 break;
1421 case sub:
1422 __ lwc1(FTF, at_sp());
1423 __ sub_s(FSF, FTF, FSF);
1424 break;
1425 case mul:
1426 __ lwc1(FTF, at_sp());
1427 __ mul_s(FSF, FTF, FSF);
1428 break;
1429 case div:
1430 __ lwc1(FTF, at_sp());
1431 __ div_s(FSF, FTF, FSF);
1432 break;
1433 case rem:
1434 __ mfc1(FSR, FSF);
1435 __ mtc1(FSR, F12);
1436 __ lwc1(FTF, at_sp());
1437 __ rem_s(FSF, FTF, F12, FSF);
1438 break;
1439 default : ShouldNotReachHere();
1440 }
1442 __ daddi(SP, SP, 1 * wordSize);
1443 }
1445 // result in SSF||FSF
1446 // i dont handle the strict flags
1447 void TemplateTable::dop2(Operation op) {
1448 transition(dtos, dtos);
1449 __ pop_dtos_to_esp(); // pop dtos into esp
1450 switch (op) {
1451 case add:
1452 __ ldc1(FTF, at_sp());
1453 __ add_d(FSF, FTF, FSF);
1454 break;
1455 case sub:
1456 __ ldc1(FTF, at_sp());
1457 __ sub_d(FSF, FTF, FSF);
1458 break;
1459 case mul:
1460 __ ldc1(FTF, at_sp());
1461 __ mul_d(FSF, FTF, FSF);
1462 break;
1463 case div:
1464 __ ldc1(FTF, at_sp());
1465 __ div_d(FSF, FTF, FSF);
1466 break;
1467 case rem:
1468 __ dmfc1(FSR, FSF);
1469 __ dmtc1(FSR, F12);
1470 __ ldc1(FTF, at_sp());
1471 __ rem_d(FSF, FTF, F12, FSF);
1472 break;
1473 default : ShouldNotReachHere();
1474 }
1476 __ daddi(SP, SP, 2 * wordSize);
1477 }
1479 void TemplateTable::ineg() {
1480 transition(itos, itos);
1481 __ neg(FSR);
1482 }
1484 void TemplateTable::lneg() {
1485 transition(ltos, ltos);
1486 __ dsubu(FSR, R0, FSR);
1487 }
1488 /*
1489 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1490 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1491 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1492 // of 128-bits operands for SSE instructions.
1493 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1494 // Store the value to a 128-bits operand.
1495 operand[0] = lo;
1496 operand[1] = hi;
1497 return operand;
1498 }
1500 // Buffer for 128-bits masks used by SSE instructions.
1501 static jlong float_signflip_pool[2*2];
1502 static jlong double_signflip_pool[2*2];
1503 */
1504 void TemplateTable::fneg() {
1505 transition(ftos, ftos);
1506 __ neg_s(FSF, FSF);
1507 }
1509 void TemplateTable::dneg() {
1510 transition(dtos, dtos);
1511 __ neg_d(FSF, FSF);
1512 }
1514 // used registers : T2
1515 void TemplateTable::iinc() {
1516 transition(vtos, vtos);
1517 locals_index(T2);
1518 __ lw(FSR, T2, 0);
1519 __ lb(AT, at_bcp(2)); // get constant
1520 __ daddu(FSR, FSR, AT);
1521 __ sw(FSR, T2, 0);
1522 }
1524 // used register : T2
1525 void TemplateTable::wide_iinc() {
1526 transition(vtos, vtos);
1527 locals_index_wide(T2);
1528 __ get_2_byte_integer_at_bcp(FSR, AT, 4);
1529 __ hswap(FSR);
1530 __ lw(AT, T2, 0);
1531 __ daddu(FSR, AT, FSR);
1532 __ sw(FSR, T2, 0);
1533 }
1535 void TemplateTable::convert() {
1536 // Checking
1537 #ifdef ASSERT
1538 { TosState tos_in = ilgl;
1539 TosState tos_out = ilgl;
1540 switch (bytecode()) {
1541 case Bytecodes::_i2l: // fall through
1542 case Bytecodes::_i2f: // fall through
1543 case Bytecodes::_i2d: // fall through
1544 case Bytecodes::_i2b: // fall through
1545 case Bytecodes::_i2c: // fall through
1546 case Bytecodes::_i2s: tos_in = itos; break;
1547 case Bytecodes::_l2i: // fall through
1548 case Bytecodes::_l2f: // fall through
1549 case Bytecodes::_l2d: tos_in = ltos; break;
1550 case Bytecodes::_f2i: // fall through
1551 case Bytecodes::_f2l: // fall through
1552 case Bytecodes::_f2d: tos_in = ftos; break;
1553 case Bytecodes::_d2i: // fall through
1554 case Bytecodes::_d2l: // fall through
1555 case Bytecodes::_d2f: tos_in = dtos; break;
1556 default : ShouldNotReachHere();
1557 }
1558 switch (bytecode()) {
1559 case Bytecodes::_l2i: // fall through
1560 case Bytecodes::_f2i: // fall through
1561 case Bytecodes::_d2i: // fall through
1562 case Bytecodes::_i2b: // fall through
1563 case Bytecodes::_i2c: // fall through
1564 case Bytecodes::_i2s: tos_out = itos; break;
1565 case Bytecodes::_i2l: // fall through
1566 case Bytecodes::_f2l: // fall through
1567 case Bytecodes::_d2l: tos_out = ltos; break;
1568 case Bytecodes::_i2f: // fall through
1569 case Bytecodes::_l2f: // fall through
1570 case Bytecodes::_d2f: tos_out = ftos; break;
1571 case Bytecodes::_i2d: // fall through
1572 case Bytecodes::_l2d: // fall through
1573 case Bytecodes::_f2d: tos_out = dtos; break;
1574 default : ShouldNotReachHere();
1575 }
1576 transition(tos_in, tos_out);
1577 }
1578 #endif // ASSERT
1580 // Conversion
1581 // (Note: use pushl(ecx)/popl(ecx) for 1/2-word stack-ptr manipulation)
1582 switch (bytecode()) {
1583 case Bytecodes::_i2l:
1584 //__ extend_sign(SSR, FSR);
1585 __ sll(FSR, FSR, 0);
1586 break;
1587 case Bytecodes::_i2f:
1588 __ mtc1(FSR, FSF);
1589 __ cvt_s_w(FSF, FSF);
1590 break;
1591 case Bytecodes::_i2d:
1592 __ mtc1(FSR, FSF);
1593 __ cvt_d_w(FSF, FSF);
1594 break;
1595 case Bytecodes::_i2b:
1596 __ dsll32(FSR, FSR, 24);
1597 __ dsra32(FSR, FSR, 24);
1598 break;
1599 case Bytecodes::_i2c:
1600 __ andi(FSR, FSR, 0xFFFF); // truncate upper 56 bits
1601 break;
1602 case Bytecodes::_i2s:
1603 __ dsll32(FSR, FSR, 16);
1604 __ dsra32(FSR, FSR, 16);
1605 break;
1606 case Bytecodes::_l2i:
1607 __ dsll32(FSR, FSR, 0);
1608 __ dsra32(FSR, FSR, 0);
1609 break;
1610 case Bytecodes::_l2f:
1611 __ dmtc1(FSR, FSF);
1612 //__ mtc1(SSR, SSF);
1613 __ cvt_s_l(FSF, FSF);
1614 break;
1615 case Bytecodes::_l2d:
1616 __ dmtc1(FSR, FSF);
1617 //__ mtc1(SSR, SSF);
1618 __ cvt_d_l(FSF, FSF);
1619 break;
1620 case Bytecodes::_f2i:
1621 {
1622 Label L;
1623 /*
1624 __ c_un_s(FSF, FSF); //NaN?
1625 __ bc1t(L);
1626 __ delayed(); __ move(FSR, R0);
1627 */
1628 __ trunc_w_s(F12, FSF);
1629 __ cfc1(AT, 31);
1630 __ li(T0, 0x10000);
1631 __ andr(AT, AT, T0);
1632 __ beq(AT, R0, L);
1633 __ delayed()->mfc1(FSR, F12);
1635 __ mov_s(F12, FSF);
1636 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1637 __ bind(L);
1638 }
1639 break;
1640 case Bytecodes::_f2l:
1641 {
1642 Label L;
1643 /*
1644 __ move(SSR, R0);
1645 __ c_un_s(FSF, FSF); //NaN?
1646 __ bc1t(L);
1647 __ delayed();
1648 __ move(FSR, R0);
1649 */
1650 __ trunc_l_s(F12, FSF);
1651 __ cfc1(AT, 31);
1652 __ li(T0, 0x10000);
1653 __ andr(AT, AT, T0);
1654 __ beq(AT, R0, L);
1655 __ delayed()->dmfc1(FSR, F12);
1657 __ mov_s(F12, FSF);
1658 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1659 __ bind(L);
1660 }
1661 break;
1662 case Bytecodes::_f2d:
1663 __ cvt_d_s(FSF, FSF);
1664 break;
1665 case Bytecodes::_d2i:
1666 {
1667 Label L;
1668 /*
1669 __ c_un_d(FSF, FSF); //NaN?
1670 __ bc1t(L);
1671 __ delayed(); __ move(FSR, R0);
1672 */
1673 __ trunc_w_d(F12, FSF);
1674 __ cfc1(AT, 31);
1675 __ li(T0, 0x10000);
1676 __ andr(AT, AT, T0);
1677 __ beq(AT, R0, L);
1678 __ delayed()->mfc1(FSR, F12);
1680 __ mov_d(F12, FSF);
1681 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1682 __ bind(L);
1683 }
1684 break;
1685 case Bytecodes::_d2l:
1686 {
1687 Label L;
1688 /*
1689 __ move(SSR, R0);
1690 __ c_un_d(FSF, FSF); //NaN?
1691 __ bc1t(L);
1692 __ delayed(); __ move(FSR, R0);
1693 */
1694 __ trunc_l_d(F12, FSF);
1695 __ cfc1(AT, 31);
1696 __ li(T0, 0x10000);
1697 __ andr(AT, AT, T0);
1698 __ beq(AT, R0, L);
1699 __ delayed()->dmfc1(FSR, F12);
1701 __ mov_d(F12, FSF);
1702 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1703 __ bind(L);
1704 }
1705 break;
1706 case Bytecodes::_d2f:
1707 __ cvt_s_d(FSF, FSF);
1708 break;
1709 default :
1710 ShouldNotReachHere();
1711 }
1712 }
1714 void TemplateTable::lcmp() {
1715 transition(ltos, itos);
1717 Label low, high, done;
1718 __ pop(T0);
1719 __ pop(R0);
1720 __ slt(AT, T0, FSR);
1721 __ bne(AT, R0, low);
1722 __ delayed()->nop();
1724 __ bne(T0, FSR, high);
1725 __ delayed()->nop();
1727 __ li(FSR, (long)0);
1728 __ b(done);
1729 __ delayed()->nop();
1731 __ bind(low);
1732 __ li(FSR, (long)-1);
1733 __ b(done);
1734 __ delayed()->nop();
1736 __ bind(high);
1737 __ li(FSR, (long)1);
1738 __ b(done);
1739 __ delayed()->nop();
1741 __ bind(done);
1742 }
1744 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1745 Label less, done;
1747 __ move(FSR, R0);
1749 if (is_float) {
1750 __ pop_ftos_to_esp();
1751 __ lwc1(FTF, at_sp());
1752 __ c_eq_s(FTF, FSF);
1753 __ bc1t(done);
1754 __ delayed()->daddi(SP, SP, 1 * wordSize);
1756 if (unordered_result<0)
1757 __ c_ult_s(FTF, FSF);
1758 else
1759 __ c_olt_s(FTF, FSF);
1760 } else {
1761 __ pop_dtos_to_esp();
1762 __ ldc1(FTF, at_sp());
1763 __ c_eq_d(FTF, FSF);
1764 __ bc1t(done);
1765 __ delayed()->daddi(SP, SP, 2 * wordSize);
1767 if (unordered_result<0)
1768 __ c_ult_d(FTF, FSF);
1769 else
1770 __ c_olt_d(FTF, FSF);
1771 }
1772 __ bc1t(less);
1773 __ delayed()->nop();
1774 __ move(FSR, 1);
1775 __ b(done);
1776 __ delayed()->nop();
1777 __ bind(less);
1778 __ move(FSR, -1);
1779 __ bind(done);
1780 }
1783 // used registers : T3, A7, Rnext
1784 // FSR : return bci, this is defined by the vm specification
1785 // T2 : MDO taken count
1786 // T3 : method
1787 // A7 : offset
1788 // Rnext : next bytecode, this is required by dispatch_base
1789 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1790 __ get_method(T3);
1791 __ profile_taken_branch(A7, T2); // only C2 meaningful
1793 #ifndef CORE
1794 const ByteSize be_offset = MethodCounters::backedge_counter_offset()
1795 + InvocationCounter::counter_offset();
1796 const ByteSize inv_offset = MethodCounters::invocation_counter_offset()
1797 + InvocationCounter::counter_offset();
1798 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
1799 #endif // CORE
1801 // Load up T4 with the branch displacement
1802 if (!is_wide) {
1803 __ get_2_byte_integer_at_bcp(A7, AT, 1);
1804 __ hswap(A7);
1805 } else {
1806 __ get_4_byte_integer_at_bcp(A7, AT, 1);
1807 __ swap(A7);
1808 }
1810 // Handle all the JSR stuff here, then exit.
1811 // It's much shorter and cleaner than intermingling with the
1812 // non-JSR normal-branch stuff occuring below.
1813 if (is_jsr) {
1814 // Pre-load the next target bytecode into Rnext
1815 __ dadd(AT, BCP, A7);
1816 __ lbu(Rnext, AT, 0);
1818 // compute return address as bci in FSR
1819 __ daddi(FSR, BCP, (is_wide?5:3) - in_bytes(ConstMethod::codes_offset()));
1820 __ ld(AT, T3, in_bytes(Method::const_offset()));
1821 __ dsub(FSR, FSR, AT);
1822 // Adjust the bcp in BCP by the displacement in A7
1823 __ dadd(BCP, BCP, A7);
1824 // jsr returns atos that is not an oop
1825 // __ dispatch_only_noverify(atos);
1826 // Push return address
1827 __ push_i(FSR);
1828 // jsr returns vtos
1829 __ dispatch_only_noverify(vtos);
1831 return;
1832 }
1834 // Normal (non-jsr) branch handling
1836 // Adjust the bcp in S0 by the displacement in T4
1837 __ dadd(BCP, BCP, A7);
1839 #ifdef CORE
1840 // Pre-load the next target bytecode into EBX
1841 __ lbu(Rnext, BCP, 0);
1842 // continue with the bytecode @ target
1843 __ dispatch_only(vtos);
1844 #else
1845 assert(UseLoopCounter || !UseOnStackReplacement, "on-stack-replacement requires loop counters");
1846 Label backedge_counter_overflow;
1847 Label profile_method;
1848 Label dispatch;
1849 if (UseLoopCounter) {
1850 // increment backedge counter for backward branches
1851 // eax: MDO
1852 // ebx: MDO bumped taken-count
1853 // T3: method
1854 // T4: target offset
1855 // BCP: target bcp
1856 // LVP: locals pointer
1857 __ bgtz(A7, dispatch); // check if forward or backward branch
1858 __ delayed()->nop();
1860 // check if MethodCounters exists
1861 Label has_counters;
1862 __ ld(AT, T3, in_bytes(Method::method_counters_offset())); // use AT as MDO, TEMP
1863 __ bne(AT, R0, has_counters);
1864 __ nop();
1865 //__ push(T3);
1866 //__ push(A7);
1867 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
1868 T3);
1869 //__ pop(A7);
1870 //__ pop(T3);
1871 __ ld(AT, T3, in_bytes(Method::method_counters_offset())); // use AT as MDO, TEMP
1872 __ beq(AT, R0, dispatch);
1873 __ nop();
1874 __ bind(has_counters);
1876 // increment back edge counter
1877 __ ld(T1, T3, in_bytes(Method::method_counters_offset()));
1878 __ lw(T0, T1, in_bytes(be_offset));
1879 __ increment(T0, InvocationCounter::count_increment);
1880 __ sw(T0, T1, in_bytes(be_offset));
1882 // load invocation counter
1883 __ lw(T1, T1, in_bytes(inv_offset));
1884 // buffer bit added, mask no needed
1885 // by yjl 10/24/2005
1886 //__ move(AT, InvocationCounter::count_mask_value);
1887 //__ andr(T1, T1, AT);
1889 // dadd backedge counter & invocation counter
1890 __ dadd(T1, T1, T0);
1892 if (ProfileInterpreter) {
1893 // Test to see if we should create a method data oop
1894 //__ lui(AT, Assembler::split_high(int(&InvocationCounter::InterpreterProfileLimit)));
1895 //__ lw(AT, AT, Assembler::split_low(int(&InvocationCounter::InterpreterProfileLimit)));
1896 // T1 : backedge counter & invocation counter
1897 __ li(AT, (long)&InvocationCounter::InterpreterProfileLimit);
1898 __ lw(AT, AT, 0);
1899 __ slt(AT, T1, AT);
1900 __ bne(AT, R0, dispatch);
1901 __ delayed()->nop();
1903 // if no method data exists, go to profile method
1904 __ test_method_data_pointer(T1, profile_method);
1906 if (UseOnStackReplacement) {
1907 // check for overflow against ebx which is the MDO taken count
1908 //__ lui(AT, Assembler::split_high(int(&InvocationCounter::InterpreterBackwardBranchLimit)));
1909 //__ lw(AT, AT, Assembler::split_low(int(&InvocationCounter::InterpreterBackwardBranchLimit)));
1910 __ li(AT, (long)&InvocationCounter::InterpreterBackwardBranchLimit);
1911 __ lw(AT, AT, 0);
1912 // the value Rnext Is get from the beginning profile_taken_branch
1913 __ slt(AT, T2, AT);
1914 __ bne(AT, R0, dispatch);
1915 __ delayed()->nop();
1917 // When ProfileInterpreter is on, the backedge_count comes
1918 // from the methodDataOop, which value does not get reset on
1919 // the call to frequency_counter_overflow().
1920 // To avoid excessive calls to the overflow routine while
1921 // the method is being compiled, dadd a second test to make
1922 // sure the overflow function is called only once every
1923 // overflow_frequency.
1924 const int overflow_frequency = 1024;
1925 __ andi(AT, T2, overflow_frequency-1);
1926 __ beq(AT, R0, backedge_counter_overflow);
1927 __ delayed()->nop();
1928 }
1929 } else {
1930 if (UseOnStackReplacement) {
1931 // check for overflow against eax, which is the sum of the counters
1932 //__ lui(AT, Assembler::split_high(int(&InvocationCounter::InterpreterBackwardBranchLimit)));
1933 //__ lw(AT, AT, Assembler::split_low(int(&InvocationCounter::InterpreterBackwardBranchLimit)));
1934 __ li(AT, (long)&InvocationCounter::InterpreterBackwardBranchLimit);
1935 __ lw(AT, AT, 0);
1936 __ slt(AT, T1, AT);
1937 __ beq(AT, R0, backedge_counter_overflow);
1938 __ delayed()->nop();
1939 }
1940 }
1941 __ bind(dispatch);
1942 }
1944 // Pre-load the next target bytecode into Rnext
1945 __ lbu(Rnext, BCP, 0);
1947 // continue with the bytecode @ target
1948 // FSR: return bci for jsr's, unused otherwise
1949 // Rnext: target bytecode
1950 // BCP: target bcp
1951 __ dispatch_only(vtos);
1953 if (UseLoopCounter) {
1954 if (ProfileInterpreter) {
1955 // Out-of-line code to allocate method data oop.
1956 __ bind(profile_method);
1957 __ call_VM(NOREG, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1958 __ lbu(Rnext, BCP, 0);
1960 __ set_method_data_pointer_for_bcp();
1961 /*
1962 __ ld(T3, FP, method_offset);
1963 __ lw(T3, T3, in_bytes(Method::method_data_offset()));
1964 __ sw(T3, FP, frame::interpreter_frame_mdx_offset * wordSize);
1965 __ test_method_data_pointer(T3, dispatch);
1966 // offset non-null mdp by MDO::data_offset() + IR::profile_method()
1967 __ daddi(T3, T3, in_bytes(MethodData::data_offset()));
1968 __ dadd(T3, T3, T1);
1969 __ sw(T3, FP, frame::interpreter_frame_mdx_offset * wordSize);
1970 */
1971 __ b(dispatch);
1972 __ delayed()->nop();
1973 }
1975 if (UseOnStackReplacement) {
1976 // invocation counter overflow
1977 __ bind(backedge_counter_overflow);
1978 __ sub(A7, BCP, A7); // branch bcp
1979 call_VM(NOREG, CAST_FROM_FN_PTR(address,
1980 InterpreterRuntime::frequency_counter_overflow), A7);
1981 __ lbu(Rnext, BCP, 0);
1983 // V0: osr nmethod (osr ok) or NULL (osr not possible)
1984 // V1: osr adapter frame return address
1985 // Rnext: target bytecode
1986 // LVP: locals pointer
1987 // BCP: bcp
1988 __ beq(V0, R0, dispatch);
1989 __ delayed()->nop();
1990 // nmethod may have been invalidated (VM may block upon call_VM return)
1991 __ lw(T3, V0, nmethod::entry_bci_offset());
1992 __ move(AT, InvalidOSREntryBci);
1993 __ beq(AT, T3, dispatch);
1994 __ delayed()->nop();
1995 // We need to prepare to execute the OSR method. First we must
1996 // migrate the locals and monitors off of the stack.
1997 //eax V0: osr nmethod (osr ok) or NULL (osr not possible)
1998 //ebx V1: osr adapter frame return address
1999 //edx Rnext: target bytecode
2000 //edi LVP: locals pointer
2001 //esi BCP: bcp
2002 __ move(BCP, V0);
2003 // const Register thread = ecx;
2004 const Register thread = TREG;
2005 #ifndef OPT_THREAD
2006 __ get_thread(thread);
2007 #endif
2008 call_VM(noreg, CAST_FROM_FN_PTR(address,
2009 SharedRuntime::OSR_migration_begin));
2010 // eax is OSR buffer, move it to expected parameter location
2011 //refer to osrBufferPointer in c1_LIRAssembler_mips.cpp
2012 __ move(T0, V0);
2014 // pop the interpreter frame
2015 // __ movl(edx, Address(ebp, frame::interpreter_frame_sender_sp_offset
2016 // * wordSize)); // get sender sp
2017 __ ld(A7, Address(FP,
2018 frame::interpreter_frame_sender_sp_offset * wordSize));
2019 //FIXME, shall we keep the return address on the stack?
2020 __ leave(); // remove frame anchor
2021 // __ popl(edi); // get return address
2022 //__ daddi(SP, SP, wordSize); // get return address
2023 // __ pop(LVP);
2024 __ move(LVP, RA);
2025 // __ movl(esp, edx); // set sp to sender sp
2026 __ move(SP, A7);
2028 Label skip;
2029 Label chkint;
2031 // The interpreter frame we have removed may be returning to
2032 // either the callstub or the interpreter. Since we will
2033 // now be returning from a compiled (OSR) nmethod we must
2034 // adjust the return to the return were it can handler compiled
2035 // results and clean the fpu stack. This is very similar to
2036 // what a i2c adapter must do.
2038 // Are we returning to the call stub?
2039 #if 0
2040 // __ cmpl(edi, (int)StubRoutines::_call_stub_return_address);
2041 __ daddi(AT, LVP, -(int)StubRoutines::_call_stub_return_address);
2042 // __ jcc(Assembler::notEqual, chkint);
2043 __ bne(AT, R0, chkint);
2044 __ delayed()->nop();
2045 // yes adjust to the specialized call stub return.
2046 // assert(StubRoutines::i486::get_call_stub_compiled_return() != NULL,
2047 // "must be set");
2048 assert(StubRoutines::gs2::get_call_stub_compiled_return() != NULL,
2049 "must be set");
2050 // __ movl(edi, (intptr_t) StubRoutines::i486::get_call_stub_compiled_return());
2051 __ move(LVP, (intptr_t) StubRoutines::gs2::get_call_stub_compiled_return());
2052 // __ jmp(skip);
2053 __ b(skip);
2054 __ delayed()->nop();
2055 __ bind(chkint);
2057 // Are we returning to the interpreter? Look for sentinel
2059 //__ cmpl(Address(edi, -8), Interpreter::return_sentinel);
2060 __ lw(AT, LVP , -8);
2061 __ daddi(AT, AT, -Interpreter::return_sentinel);
2062 //__ jcc(Assembler::notEqual, skip);
2063 __ bne(AT, R0, skip);
2064 __ delayed()->nop();
2065 // Adjust to compiled return back to interpreter
2067 // __ movl(edi, Address(edi, -4));
2068 __ lw(LVP, LVP, -4);
2070 __ bind(skip);
2071 #endif
2072 // Align stack pointer for compiled code (note that caller is
2073 // responsible for undoing this fixup by remembering the old SP
2074 // in an ebp-relative location)
2075 // __ andl(esp, -(StackAlignmentInBytes));
2076 __ move(AT, -(StackAlignmentInBytes));
2077 __ andr(SP , SP , AT);
2078 // push the (possibly adjusted) return address
2079 // __ pushl(edi);
2080 //__ push(LVP);
2081 // __ move(RA, LVP);
2082 // and begin the OSR nmethod
2083 // __ jmp(Address(esi, nmethod::osr_entry_point_offset()));
2084 //refer to osr_entry in c1_LIRAssembler_mips.cpp
2085 __ ld(AT, BCP, nmethod::osr_entry_point_offset());
2086 __ jr(AT);
2087 __ delayed()->nop();
2088 }
2089 }
2090 #endif // not CORE
2091 }
2093 void TemplateTable::if_0cmp(Condition cc) {
2094 transition(itos, vtos);
2095 // assume branch is more often taken than not (loops use backward branches)
2096 Label not_taken;
2097 switch(cc) {
2098 case not_equal:
2099 __ beq(FSR, R0, not_taken);
2100 break;
2101 case equal:
2102 __ bne(FSR, R0, not_taken);
2103 break;
2104 case less:
2105 __ bgez(FSR, not_taken);
2106 break;
2107 case less_equal:
2108 __ bgtz(FSR, not_taken);
2109 break;
2110 case greater:
2111 __ blez(FSR, not_taken);
2112 break;
2113 case greater_equal:
2114 __ bltz(FSR, not_taken);
2115 break;
2116 }
2117 __ delayed()->nop();
2119 branch(false, false);
2121 __ bind(not_taken);
2122 __ profile_not_taken_branch(FSR);
2123 }
2126 void TemplateTable::if_icmp(Condition cc) {
2127 transition(itos, vtos);
2128 // assume branch is more often taken than not (loops use backward branches)
2129 Label not_taken;
2131 __ pop_i(SSR);
2132 switch(cc) {
2133 case not_equal:
2134 __ beq(SSR, FSR, not_taken);
2135 break;
2136 case equal:
2137 __ bne(SSR, FSR, not_taken);
2138 break;
2139 case less:
2140 __ slt(AT, SSR, FSR);
2141 __ beq(AT, R0, not_taken);
2142 break;
2143 case less_equal:
2144 __ slt(AT, FSR, SSR);
2145 __ bne(AT, R0, not_taken);
2146 break;
2147 case greater:
2148 __ slt(AT, FSR, SSR);
2149 __ beq(AT, R0, not_taken);
2150 break;
2151 case greater_equal:
2152 __ slt(AT, SSR, FSR);
2153 __ bne(AT, R0, not_taken);
2154 break;
2155 }
2156 __ delayed()->nop();
2158 branch(false, false);
2160 __ bind(not_taken);
2161 __ profile_not_taken_branch(FSR);
2162 }
2165 void TemplateTable::if_nullcmp(Condition cc) {
2166 transition(atos, vtos);
2167 // assume branch is more often taken than not (loops use backward branches)
2168 Label not_taken;
2169 switch(cc) {
2170 case not_equal:
2171 __ beq(FSR, R0, not_taken);
2172 break;
2173 case equal:
2174 __ bne(FSR, R0, not_taken);
2175 break;
2176 default:
2177 ShouldNotReachHere();
2178 }
2179 __ delayed()->nop();
2181 branch(false, false);
2183 __ bind(not_taken);
2184 __ profile_not_taken_branch(FSR);
2185 }
2188 void TemplateTable::if_acmp(Condition cc) {
2189 transition(atos, vtos);
2190 // assume branch is more often taken than not (loops use backward branches)
2191 Label not_taken;
2192 // __ lw(SSR, SP, 0);
2193 __ pop_ptr(SSR);
2194 switch(cc) {
2195 case not_equal:
2196 __ beq(SSR, FSR, not_taken);
2197 break;
2198 case equal:
2199 __ bne(SSR, FSR, not_taken);
2200 break;
2201 default:
2202 ShouldNotReachHere();
2203 }
2204 // __ delayed()->daddi(SP, SP, 4);
2205 __ delayed()->nop();
2207 branch(false, false);
2209 __ bind(not_taken);
2210 __ profile_not_taken_branch(FSR);
2211 }
2213 // used registers : T1, T2, T3
2214 // T1 : method
2215 // T2 : returb bci
2216 void TemplateTable::ret() {
2217 transition(vtos, vtos);
2219 locals_index(T2);
2220 __ ld(T2, T2, 0);
2221 __ profile_ret(T2, T3);
2223 __ get_method(T1);
2224 __ ld(BCP, T1, in_bytes(Method::const_offset()));
2225 __ dadd(BCP, BCP, T2);
2226 __ daddi(BCP, BCP, in_bytes(ConstMethod::codes_offset()));
2228 __ dispatch_next(vtos);
2229 }
2231 // used registers : T1, T2, T3
2232 // T1 : method
2233 // T2 : returb bci
2234 void TemplateTable::wide_ret() {
2235 transition(vtos, vtos);
2237 locals_index_wide(T2);
2238 __ ld(T2, T2, 0); // get return bci, compute return bcp
2239 __ profile_ret(T2, T3);
2241 __ get_method(T1);
2242 __ ld(BCP, T1, in_bytes(Method::const_offset()));
2243 __ dadd(BCP, BCP, T2);
2244 __ daddi(BCP, BCP, in_bytes(ConstMethod::codes_offset()));
2246 __ dispatch_next(vtos);
2247 }
2249 // used register T2, T3, A7, Rnext
2250 // T2 : bytecode pointer
2251 // T3 : low
2252 // A7 : high
2253 // Rnext : dest bytecode, required by dispatch_base
2254 void TemplateTable::tableswitch() {
2255 Label default_case, continue_execution;
2256 transition(itos, vtos);
2258 // align BCP
2259 __ daddi(T2, BCP, BytesPerInt);
2260 __ li(AT, -BytesPerInt);
2261 __ andr(T2, T2, AT);
2263 // load lo & hi
2264 __ lw(T3, T2, 1 * BytesPerInt);
2265 __ swap(T3);
2266 __ lw(A7, T2, 2 * BytesPerInt);
2267 __ swap(A7);
2269 // check against lo & hi
2270 __ slt(AT, FSR, T3);
2271 __ bne(AT, R0, default_case);
2272 __ delayed()->nop();
2274 __ slt(AT, A7, FSR);
2275 __ bne(AT, R0, default_case);
2276 __ delayed()->nop();
2278 // lookup dispatch offset, in A7 big endian
2279 __ dsub(FSR, FSR, T3);
2280 __ dsll(AT, FSR, Address::times_4);
2281 __ dadd(AT, T2, AT);
2282 __ lw(A7, AT, 3 * BytesPerInt);
2283 __ profile_switch_case(FSR, T9, T3);
2285 __ bind(continue_execution);
2286 __ swap(A7);
2287 __ dadd(BCP, BCP, A7);
2288 __ lbu(Rnext, BCP, 0);
2289 __ dispatch_only(vtos);
2291 // handle default
2292 __ bind(default_case);
2293 __ profile_switch_default(FSR);
2294 __ lw(A7, T2, 0);
2295 __ b(continue_execution);
2296 __ delayed()->nop();
2297 }
2299 void TemplateTable::lookupswitch() {
2300 transition(itos, itos);
2301 __ stop("lookupswitch bytecode should have been rewritten");
2302 }
2304 // used registers : T2, T3, A7, Rnext
2305 // T2 : bytecode pointer
2306 // T3 : pair index
2307 // A7 : offset
2308 // Rnext : dest bytecode
2309 // the data after the opcode is the same as lookupswitch
2310 // see Rewriter::rewrite_method for more information
2311 void TemplateTable::fast_linearswitch() {
2312 transition(itos, vtos);
2313 Label loop_entry, loop, found, continue_execution;
2315 // swap eax so we can avoid swapping the table entries
2316 __ swap(FSR);
2318 // align BCP
2319 __ daddi(T2, BCP, BytesPerInt);
2320 __ li(AT, -BytesPerInt);
2321 __ andr(T2, T2, AT);
2323 // set counter
2324 __ lw(T3, T2, BytesPerInt);
2325 __ swap(T3);
2326 __ b(loop_entry);
2327 __ delayed()->nop();
2329 // table search
2330 __ bind(loop);
2331 // get the entry value
2332 __ dsll(AT, T3, Address::times_8);
2333 __ dadd(AT, T2, AT);
2334 __ lw(AT, AT, 2 * BytesPerInt);
2336 // found?
2337 __ beq(FSR, AT, found);
2338 __ delayed()->nop();
2340 __ bind(loop_entry);
2341 __ bgtz(T3, loop);
2342 __ delayed()->daddiu(T3, T3, -1);
2344 // default case
2345 __ profile_switch_default(FSR);
2346 __ lw(A7, T2, 0);
2347 __ b(continue_execution);
2348 __ delayed()->nop();
2350 // entry found -> get offset
2351 __ bind(found);
2352 __ dsll(AT, T3, Address::times_8);
2353 __ dadd(AT, T2, AT);
2354 __ lw(A7, AT, 3 * BytesPerInt);
2355 __ profile_switch_case(T3, FSR, T2);
2357 // continue execution
2358 __ bind(continue_execution);
2359 __ swap(A7);
2360 __ dadd(BCP, BCP, A7);
2361 __ lbu(Rnext, BCP, 0);
2362 __ dispatch_only(vtos);
2363 }
2365 // used registers : T0, T1, T2, T3, A7, Rnext
2366 // T2 : pairs address(array)
2367 // Rnext : dest bytecode
2368 // the data after the opcode is the same as lookupswitch
2369 // see Rewriter::rewrite_method for more information
2370 void TemplateTable::fast_binaryswitch() {
2371 transition(itos, vtos);
2372 // Implementation using the following core algorithm:
2373 //
2374 // int binary_search(int key, LookupswitchPair* array, int n) {
2375 // // Binary search according to "Methodik des Programmierens" by
2376 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2377 // int i = 0;
2378 // int j = n;
2379 // while (i+1 < j) {
2380 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2381 // // with Q: for all i: 0 <= i < n: key < a[i]
2382 // // where a stands for the array and assuming that the (inexisting)
2383 // // element a[n] is infinitely big.
2384 // int h = (i + j) >> 1;
2385 // // i < h < j
2386 // if (key < array[h].fast_match()) {
2387 // j = h;
2388 // } else {
2389 // i = h;
2390 // }
2391 // }
2392 // // R: a[i] <= key < a[i+1] or Q
2393 // // (i.e., if key is within array, i is the correct index)
2394 // return i;
2395 // }
2397 // register allocation
2398 const Register array = T2;
2399 const Register i = T3, j = A7;
2400 const Register h = T1;
2401 const Register temp = T0;
2402 const Register key = FSR;
2404 // setup array
2405 __ daddi(array, BCP, 3*BytesPerInt);
2406 __ li(AT, -BytesPerInt);
2407 __ andr(array, array, AT);
2409 // initialize i & j
2410 __ move(i, R0);
2411 __ lw(j, array, - 1 * BytesPerInt);
2412 // Convert j into native byteordering
2413 __ swap(j);
2415 // and start
2416 Label entry;
2417 __ b(entry);
2418 __ delayed()->nop();
2420 // binary search loop
2421 {
2422 Label loop;
2423 __ bind(loop);
2424 // int h = (i + j) >> 1;
2425 __ dadd(h, i, j);
2426 __ dsrl(h, h, 1);
2427 // if (key < array[h].fast_match()) {
2428 // j = h;
2429 // } else {
2430 // i = h;
2431 // }
2432 // Convert array[h].match to native byte-ordering before compare
2433 __ dsll(AT, h, Address::times_8);
2434 __ dadd(AT, array, AT);
2435 __ lw(temp, AT, 0 * BytesPerInt);
2436 __ swap(temp);
2438 {
2439 Label set_i, end_of_if;
2440 __ slt(AT, key, temp);
2441 __ beq(AT, R0, set_i);
2442 __ delayed()->nop();
2444 __ b(end_of_if);
2445 __ delayed(); __ move(j, h);
2447 __ bind(set_i);
2448 __ move(i, h);
2450 __ bind(end_of_if);
2451 }
2452 // while (i+1 < j)
2453 __ bind(entry);
2454 __ daddi(h, i, 1);
2455 __ slt(AT, h, j);
2456 __ bne(AT, R0, loop);
2457 __ delayed()->nop();
2458 }
2460 // end of binary search, result index is i (must check again!)
2461 Label default_case;
2462 // Convert array[i].match to native byte-ordering before compare
2463 __ dsll(AT, i, Address::times_8);
2464 __ dadd(AT, array, AT);
2465 __ lw(temp, AT, 0 * BytesPerInt);
2466 __ swap(temp);
2467 __ bne(key, temp, default_case);
2468 __ delayed()->nop();
2470 // entry found -> j = offset
2471 __ dsll(AT, i, Address::times_8);
2472 __ dadd(AT, array, AT);
2473 __ lw(j, AT, 1 * BytesPerInt);
2474 __ profile_switch_case(i, key, array);
2475 __ swap(j);
2477 __ dadd(BCP, BCP, j);
2478 __ lbu(Rnext, BCP, 0);
2479 __ dispatch_only(vtos);
2481 // default case -> j = default offset
2482 __ bind(default_case);
2483 __ profile_switch_default(i);
2484 __ lw(j, array, - 2 * BytesPerInt);
2485 __ swap(j);
2486 __ dadd(BCP, BCP, j);
2487 __ lbu(Rnext, BCP, 0);
2488 __ dispatch_only(vtos);
2489 }
2491 void TemplateTable::_return(TosState state) {
2492 transition(state, state);
2493 assert(_desc->calls_vm(), "inconsistent calls_vm information"); // call in remove_activation
2494 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2495 assert(state == vtos, "only valid state");
2496 __ ld(T1, aaddress(0));
2497 //__ ld(LVP, T1, oopDesc::klass_offset_in_bytes());
2498 __ load_klass(LVP, T1);
2499 __ lw(LVP, LVP, in_bytes(Klass::access_flags_offset()));
2500 __ move(AT, JVM_ACC_HAS_FINALIZER);
2501 __ andr(AT, AT, LVP);//by_css
2502 Label skip_register_finalizer;
2503 __ beq(AT, R0, skip_register_finalizer);
2504 __ delayed()->nop();
2505 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2506 InterpreterRuntime::register_finalizer), T1);
2507 __ bind(skip_register_finalizer);
2508 }
2509 __ remove_activation(state, T9);
2510 __ sync();
2512 __ jr(T9);
2513 __ delayed()->nop();
2514 }
2516 // ----------------------------------------------------------------------------
2517 // Volatile variables demand their effects be made known to all CPU's
2518 // in order. Store buffers on most chips allow reads & writes to
2519 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2520 // without some kind of memory barrier (i.e., it's not sufficient that
2521 // the interpreter does not reorder volatile references, the hardware
2522 // also must not reorder them).
2523 //
2524 // According to the new Java Memory Model (JMM):
2525 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2526 // writes act as aquire & release, so:
2527 // (2) A read cannot let unrelated NON-volatile memory refs that
2528 // happen after the read float up to before the read. It's OK for
2529 // non-volatile memory refs that happen before the volatile read to
2530 // float down below it.
2531 // (3) Similar a volatile write cannot let unrelated NON-volatile
2532 // memory refs that happen BEFORE the write float down to after the
2533 // write. It's OK for non-volatile memory refs that happen after the
2534 // volatile write to float up before it.
2535 //
2536 // We only put in barriers around volatile refs (they are expensive),
2537 // not _between_ memory refs (that would require us to track the
2538 // flavor of the previous memory refs). Requirements (2) and (3)
2539 // require some barriers before volatile stores and after volatile
2540 // loads. These nearly cover requirement (1) but miss the
2541 // volatile-store-volatile-load case. This final case is placed after
2542 // volatile-stores although it could just as well go before
2543 // volatile-loads.
2544 //void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits
2545 // order_constraint) {
2546 void TemplateTable::volatile_barrier( ) {
2547 // Helper function to insert a is-volatile test and memory barrier
2548 //if (os::is_MP()) { // Not needed on single CPU
2549 // __ membar(order_constraint);
2550 //}
2551 if( !os::is_MP() ) return; // Not needed on single CPU
2552 __ sync();
2553 }
2555 // we dont shift left 2 bits in get_cache_and_index_at_bcp
2556 // for we always need shift the index we use it. the ConstantPoolCacheEntry
2557 // is 16-byte long, index is the index in
2558 // ConstantPoolCache, so cache + base_offset() + index * 16 is
2559 // the corresponding ConstantPoolCacheEntry
2560 // used registers : T2
2561 // NOTE : the returned index need also shift left 4 to get the address!
2562 void TemplateTable::resolve_cache_and_index(int byte_no,
2563 Register Rcache,
2564 Register index,
2565 size_t index_size) {
2566 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2567 const Register temp = A1;
2568 assert_different_registers(Rcache, index);
2569 const int shift_count = (1 + byte_no)*BitsPerByte;
2570 Label resolved;
2571 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2572 // is resolved?
2573 int i = (int)bytecode();
2574 __ addi(temp, temp, -i);
2575 __ beq(temp, R0, resolved);
2576 __ delayed()->nop();
2577 // resolve first time through
2578 address entry;
2579 switch (bytecode()) {
2580 case Bytecodes::_getstatic : // fall through
2581 case Bytecodes::_putstatic : // fall through
2582 case Bytecodes::_getfield : // fall through
2583 case Bytecodes::_putfield :
2584 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put);
2585 break;
2586 case Bytecodes::_invokevirtual : // fall through
2587 case Bytecodes::_invokespecial : // fall through
2588 case Bytecodes::_invokestatic : // fall through
2589 case Bytecodes::_invokeinterface:
2590 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke);
2591 break;
2592 case Bytecodes::_invokehandle:
2593 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokehandle);
2594 break;
2595 case Bytecodes::_invokedynamic:
2596 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic);
2597 break;
2598 default :
2599 fatal(err_msg("unexpected bytecode: %s", Bytecodes::name(bytecode())));
2600 }
2602 __ move(temp, i);
2603 __ call_VM(NOREG, entry, temp);
2605 // Update registers with resolved info
2606 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2607 __ bind(resolved);
2608 }
2610 // The Rcache and index registers must be set before call
2611 void TemplateTable::load_field_cp_cache_entry(Register obj,
2612 Register cache,
2613 Register index,
2614 Register off,
2615 Register flags,
2616 bool is_static = false) {
2617 assert_different_registers(cache, index, flags, off);
2618 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2619 // Field offset
2620 __ dsll(AT, index, Address::times_ptr);
2621 __ dadd(AT, cache, AT);
2622 __ ld(off, AT, in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset()));
2623 // Flags
2624 __ ld(flags, AT, in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset()));
2626 // klass overwrite register
2627 if (is_static) {
2628 __ ld(obj, AT, in_bytes(cp_base_offset + ConstantPoolCacheEntry::f1_offset()));
2629 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2630 __ ld(obj, Address(obj, mirror_offset));
2632 __ verify_oop(obj);
2633 }
2634 }
2636 // get the method, itable_index and flags of the current invoke
2637 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2638 Register method,
2639 Register itable_index,
2640 Register flags,
2641 bool is_invokevirtual,
2642 bool is_invokevfinal, /*unused*/
2643 bool is_invokedynamic) {
2644 // setup registers
2645 const Register cache = T3;
2646 const Register index = T1;
2647 assert_different_registers(method, flags);
2648 assert_different_registers(method, cache, index);
2649 assert_different_registers(itable_index, flags);
2650 assert_different_registers(itable_index, cache, index);
2651 assert(is_invokevirtual == (byte_no == f2_byte), "is invokevirtual flag redundant");
2652 // determine constant pool cache field offsets
2653 const int method_offset = in_bytes(
2654 ConstantPoolCache::base_offset() +
2655 ((byte_no == f2_byte)
2656 ? ConstantPoolCacheEntry::f2_offset()
2657 : ConstantPoolCacheEntry::f1_offset()
2658 )
2659 );
2660 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2661 ConstantPoolCacheEntry::flags_offset());
2662 // access constant pool cache fields
2663 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2664 ConstantPoolCacheEntry::f2_offset());
2665 size_t index_size = (is_invokedynamic ? sizeof(u4): sizeof(u2));
2666 resolve_cache_and_index(byte_no, cache, index, index_size);
2668 //assert(wordSize == 8, "adjust code below");
2669 // note we shift 4 not 2, for we get is the true inde
2670 // of ConstantPoolCacheEntry, not the shifted 2-bit index as x86 version
2671 __ dsll(AT, index, Address::times_ptr);
2672 __ dadd(AT, cache, AT);
2673 __ ld(method, AT, method_offset);
2676 if (itable_index != NOREG) {
2677 __ ld(itable_index, AT, index_offset);
2678 }
2679 __ ld(flags, AT, flags_offset);
2680 }
2683 // The registers cache and index expected to be set before call.
2684 // Correct values of the cache and index registers are preserved.
2685 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2686 bool is_static, bool has_tos) {
2687 // do the JVMTI work here to avoid disturbing the register state below
2688 // We use c_rarg registers here because we want to use the register used in
2689 // the call to the VM
2690 if (JvmtiExport::can_post_field_access()) {
2691 // Check to see if a field access watch has been set before we take
2692 // the time to call into the VM.
2693 Label L1;
2694 assert_different_registers(cache, index, FSR);
2695 __ li(AT, (intptr_t)JvmtiExport::get_field_access_count_addr());
2696 __ lw(FSR, AT, 0);
2697 __ beq(FSR, R0, L1);
2698 __ delayed()->nop();
2700 // We rely on the bytecode being resolved and the cpCache entry filled in.
2701 // cache entry pointer
2702 //__ get_cache_and_index_at_bcp(c_rarg2, c_rarg3, 1);
2703 __ daddi(cache, cache, in_bytes(ConstantPoolCache::base_offset()));
2704 __ shl(index, 4);
2705 __ dadd(cache, cache, index);
2706 if (is_static) {
2707 __ move(FSR, R0);
2708 } else {
2709 __ lw(FSR, SP, 0);
2710 __ verify_oop(FSR);
2711 }
2712 // FSR: object pointer or NULL
2713 // cache: cache entry pointer
2714 __ call_VM(NOREG, CAST_FROM_FN_PTR(address,
2715 InterpreterRuntime::post_field_access), FSR, cache);
2716 __ get_cache_and_index_at_bcp(cache, index, 1);
2717 __ bind(L1);
2718 }
2719 }
2721 void TemplateTable::pop_and_check_object(Register r) {
2722 __ pop_ptr(r);
2723 __ null_check(r); // for field access must check obj.
2724 __ verify_oop(r);
2725 }
2727 // used registers : T1, T2, T3, T1
2728 // T1 : flags
2729 // T2 : off
2730 // T3 : obj
2731 // T1 : field address
2732 // The flags 31, 30, 29, 28 together build a 4 bit number 0 to 8 with the
2733 // following mapping to the TosState states:
2734 // btos: 0
2735 // ctos: 1
2736 // stos: 2
2737 // itos: 3
2738 // ltos: 4
2739 // ftos: 5
2740 // dtos: 6
2741 // atos: 7
2742 // vtos: 8
2743 // see ConstantPoolCacheEntry::set_field for more info
2744 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2745 transition(vtos, vtos);
2747 const Register cache = T3;
2748 const Register index = T0;
2750 const Register obj = T3;
2751 const Register off = T2;
2752 const Register flags = T1;
2753 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2754 //jvmti_post_field_access(cache, index, is_static, false);
2756 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2758 if (!is_static) pop_and_check_object(obj);
2759 __ dadd(index, obj, off);
2762 Label Done, notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2764 assert(btos == 0, "change code, btos != 0");
2765 __ dsrl(flags, flags, ConstantPoolCacheEntry::tos_state_shift);
2766 __ andi(flags, flags, 0xf);
2767 __ bne(flags, R0, notByte);
2768 __ delayed()->nop();
2770 // btos
2771 __ sync();
2772 __ lb(FSR, index, 0);
2773 __ sd(FSR, SP, - wordSize);
2775 // Rewrite bytecode to be faster
2776 if (!is_static) {
2777 patch_bytecode(Bytecodes::_fast_bgetfield, T3, T2);
2778 }
2779 __ b(Done);
2780 __ delayed()->daddi(SP, SP, - wordSize);
2782 __ bind(notByte);
2783 __ move(AT, itos);
2784 __ bne(flags, AT, notInt);
2785 __ delayed()->nop();
2787 // itos
2788 __ sync();
2789 __ lw(FSR, index, 0);
2790 __ sd(FSR, SP, - wordSize);
2792 // Rewrite bytecode to be faster
2793 if (!is_static) {
2794 // patch_bytecode(Bytecodes::_fast_igetfield, T3, T2);
2795 patch_bytecode(Bytecodes::_fast_igetfield, T3, T2);
2796 }
2797 __ b(Done);
2798 __ delayed()->daddi(SP, SP, - wordSize);
2800 __ bind(notInt);
2801 __ move(AT, atos);
2802 __ bne(flags, AT, notObj);
2803 __ delayed()->nop();
2805 // atos
2806 //add for compressedoops
2807 __ sync();
2808 __ load_heap_oop(FSR, Address(index, 0));
2809 __ sd(FSR, SP, - wordSize);
2811 if (!is_static) {
2812 //patch_bytecode(Bytecodes::_fast_agetfield, T3, T2);
2813 patch_bytecode(Bytecodes::_fast_agetfield, T3, T2);
2814 }
2815 __ b(Done);
2816 __ delayed()->daddi(SP, SP, - wordSize);
2818 __ bind(notObj);
2819 __ move(AT, ctos);
2820 __ bne(flags, AT, notChar);
2821 __ delayed()->nop();
2823 // ctos
2824 __ sync();
2825 __ lhu(FSR, index, 0);
2826 __ sd(FSR, SP, - wordSize);
2828 if (!is_static) {
2829 patch_bytecode(Bytecodes::_fast_cgetfield, T3, T2);
2830 }
2831 __ b(Done);
2832 __ delayed()->daddi(SP, SP, - wordSize);
2834 __ bind(notChar);
2835 __ move(AT, stos);
2836 __ bne(flags, AT, notShort);
2837 __ delayed()->nop();
2839 // stos
2840 __ sync();
2841 __ lh(FSR, index, 0);
2842 __ sd(FSR, SP, - wordSize);
2844 if (!is_static) {
2845 // patch_bytecode(Bytecodes::_fast_sgetfield, T3, T2);
2846 patch_bytecode(Bytecodes::_fast_sgetfield, T3, T2);
2847 }
2848 __ b(Done);
2849 __ delayed()->daddi(SP, SP, - wordSize);
2851 __ bind(notShort);
2852 __ move(AT, ltos);
2853 __ bne(flags, AT, notLong);
2854 __ delayed()->nop();
2856 // FIXME : the load/store should be atomic, we have no simple method to do this in mips32
2857 // ltos
2858 __ sync();
2859 __ ld(FSR, index, 0 * wordSize);
2860 __ sd(FSR, SP, -2 * wordSize);
2861 __ sd(R0, SP, -1 * wordSize);
2863 // Don't rewrite to _fast_lgetfield for potential volatile case.
2864 __ b(Done);
2865 __ delayed()->daddi(SP, SP, - 2 * wordSize);
2867 __ bind(notLong);
2868 __ move(AT, ftos);
2869 __ bne(flags, AT, notFloat);
2870 __ delayed()->nop();
2872 // ftos
2873 __ sync();
2874 __ lwc1(FSF, index, 0);
2875 __ sdc1(FSF, SP, - wordSize);
2877 if (!is_static) {
2878 patch_bytecode(Bytecodes::_fast_fgetfield, T3, T2);
2879 }
2880 __ b(Done);
2881 __ delayed()->daddi(SP, SP, - wordSize);
2883 __ bind(notFloat);
2884 __ move(AT, dtos);
2885 __ bne(flags, AT, notDouble);
2886 __ delayed()->nop();
2888 // dtos
2889 __ sync();
2890 __ ldc1(FSF, index, 0 * wordSize);
2891 __ sdc1(FSF, SP, - 2 * wordSize);
2892 __ sd(R0, SP, - 1 * wordSize);
2894 if (!is_static) {
2895 patch_bytecode(Bytecodes::_fast_dgetfield, T3, T2);
2896 }
2897 __ b(Done);
2898 __ delayed()->daddi(SP, SP, - 2 * wordSize);
2900 __ bind(notDouble);
2902 __ stop("Bad state");
2904 __ bind(Done);
2905 }
2907 void TemplateTable::getfield(int byte_no) {
2908 getfield_or_static(byte_no, false);
2909 }
2911 void TemplateTable::getstatic(int byte_no) {
2912 getfield_or_static(byte_no, true);
2913 }
2914 /*
2915 // used registers : T1, T2, T3, T1
2916 // T1 : cache & cp entry
2917 // T2 : obj
2918 // T3 : flags & value pointer
2919 // T1 : index
2920 // see ConstantPoolCacheEntry::set_field for more info
2921 void TemplateTable::jvmti_post_field_mod(int byte_no, bool is_static) {
2922 */
2924 // The registers cache and index expected to be set before call.
2925 // The function may destroy various registers, just not the cache and index registers.
2926 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2927 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2929 if (JvmtiExport::can_post_field_modification()) {
2930 // Check to see if a field modification watch has been set before we take
2931 // the time to call into the VM.
2932 Label L1;
2933 assert_different_registers(cache, index, AT);
2935 //__ lui(AT, Assembler::split_high((int)JvmtiExport::get_field_modification_count_addr()));
2936 //__ lw(FSR, AT, Assembler::split_low((int)JvmtiExport::get_field_modification_count_addr()));
2937 __ li(AT, JvmtiExport::get_field_modification_count_addr());
2938 __ lw(FSR, AT, 0);
2939 __ beq(FSR, R0, L1);
2940 __ delayed()->nop();
2942 /* // We rely on the bytecode being resolved and the cpCache entry filled in.
2943 resolve_cache_and_index(byte_no, T1, T1);
2944 */
2945 // The cache and index registers have been already set.
2946 // This allows to eliminate this call but the cache and index
2947 // registers have to be correspondingly used after this line.
2948 // __ get_cache_and_index_at_bcp(eax, edx, 1);
2949 __ get_cache_and_index_at_bcp(T1, T9, 1);
2951 if (is_static) {
2952 __ move(T2, R0);
2953 } else {
2954 // Life is harder. The stack holds the value on top,
2955 // followed by the object.
2956 // We don't know the size of the value, though;
2957 // it could be one or two words
2958 // depending on its type. As a result, we must find
2959 // the type to determine where the object is.
2960 Label two_word, valsize_known;
2961 __ dsll(AT, T1, 4);
2962 __ dadd(AT, T1, AT);
2963 __ lw(T3, AT, in_bytes(cp_base_offset
2964 + ConstantPoolCacheEntry::flags_offset()));
2965 __ move(T2, SP);
2966 __ shr(T3, ConstantPoolCacheEntry::tos_state_shift);
2968 // Make sure we don't need to mask ecx for tos_state_shift
2969 // after the above shift
2970 ConstantPoolCacheEntry::verify_tos_state_shift();
2971 __ move(AT, ltos);
2972 __ beq(T3, AT, two_word);
2973 __ delayed()->nop();
2974 __ move(AT, dtos);
2975 __ beq(T3, AT, two_word);
2976 __ delayed()->nop();
2977 __ b(valsize_known);
2978 //__ delayed()->daddi(T2, T2, wordSize*1);
2979 __ delayed()->daddi(T2, T2,Interpreter::expr_offset_in_bytes(1) );
2981 __ bind(two_word);
2982 // __ daddi(T2, T2, wordSize*2);
2983 __ daddi(T2, T2,Interpreter::expr_offset_in_bytes(2));
2985 __ bind(valsize_known);
2986 // setup object pointer
2987 __ lw(T2, T2, 0*wordSize);
2988 }
2989 // cache entry pointer
2990 __ daddi(T1, T1, in_bytes(cp_base_offset));
2991 __ shl(T1, 4);
2992 __ daddu(T1, T1, T1);
2993 // object (tos)
2994 __ move(T3, SP);
2995 // T2: object pointer set up above (NULL if static)
2996 // T1: cache entry pointer
2997 // T3: jvalue object on the stack
2998 __ call_VM(NOREG, CAST_FROM_FN_PTR(address,
2999 InterpreterRuntime::post_field_modification), T2, T1, T3);
3000 __ get_cache_and_index_at_bcp(cache, index, 1);
3001 __ bind(L1);
3002 }
3003 }
3005 // used registers : T0, T1, T2, T3, T8
3006 // T1 : flags
3007 // T2 : off
3008 // T3 : obj
3009 // T8 : volatile bit
3010 // see ConstantPoolCacheEntry::set_field for more info
3011 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
3012 transition(vtos, vtos);
3014 const Register cache = T3;
3015 const Register index = T0;
3016 const Register obj = T3;
3017 const Register off = T2;
3018 const Register flags = T1;
3019 const Register bc = T3;
3021 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
3022 //TODO: LEE
3023 //jvmti_post_field_mod(cache, index, is_static);
3024 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
3025 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
3026 // volatile_barrier( );
3028 Label notVolatile, Done;
3029 __ move(AT, 1<<ConstantPoolCacheEntry::is_volatile_shift);
3030 __ andr(T8, flags, AT);
3032 Label notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
3034 assert(btos == 0, "change code, btos != 0");
3035 // btos
3036 __ dsrl(flags, flags, ConstantPoolCacheEntry::tos_state_shift);
3037 __ andi(flags, flags, ConstantPoolCacheEntry::tos_state_mask);
3038 __ bne(flags, R0, notByte);
3039 __ delayed()->nop();
3041 __ pop(btos);
3042 if (!is_static) {
3043 pop_and_check_object(obj);
3044 }
3045 __ dadd(AT, obj, off);
3046 __ sb(FSR, AT, 0);
3048 if (!is_static) {
3049 patch_bytecode(Bytecodes::_fast_bputfield, bc, off, true, byte_no);
3050 }
3051 __ b(Done);
3052 __ delayed()->nop();
3054 __ bind(notByte);
3055 // itos
3056 __ move(AT, itos);
3057 __ bne(flags, AT, notInt);
3058 __ delayed()->nop();
3060 __ pop(itos);
3061 if (!is_static) {
3062 pop_and_check_object(obj);
3063 }
3064 __ dadd(AT, obj, off);
3065 __ sw(FSR, AT, 0);
3067 if (!is_static) {
3068 patch_bytecode(Bytecodes::_fast_iputfield, bc, off, true, byte_no);
3069 }
3070 __ b(Done);
3071 __ delayed()->nop();
3072 __ bind(notInt);
3073 // atos
3074 __ move(AT, atos);
3075 __ bne(flags, AT, notObj);
3076 __ delayed()->nop();
3078 __ pop(atos);
3079 if (!is_static) {
3080 pop_and_check_object(obj);
3081 }
3083 __ dadd(AT, obj, off);
3084 //__ sd(FSR, AT, 0);
3085 __ store_heap_oop(Address(AT, 0), FSR);
3086 __ sync();
3087 __ store_check(obj);
3089 if (!is_static) {
3090 patch_bytecode(Bytecodes::_fast_aputfield, bc, off, true, byte_no);
3091 }
3092 __ b(Done);
3093 __ delayed()->nop();
3094 __ bind(notObj);
3095 // ctos
3096 __ move(AT, ctos);
3097 __ bne(flags, AT, notChar);
3098 __ delayed()->nop();
3100 __ pop(ctos);
3101 if (!is_static) {
3102 pop_and_check_object(obj);
3103 }
3104 __ dadd(AT, obj, off);
3105 __ sh(FSR, AT, 0);
3106 if (!is_static) {
3107 patch_bytecode(Bytecodes::_fast_cputfield, bc, off, true, byte_no);
3108 }
3109 __ b(Done);
3110 __ delayed()->nop();
3111 __ bind(notChar);
3112 // stos
3113 __ move(AT, stos);
3114 __ bne(flags, AT, notShort);
3115 __ delayed()->nop();
3117 __ pop(stos);
3118 if (!is_static) {
3119 pop_and_check_object(obj);
3120 }
3121 __ dadd(AT, obj, off);
3122 __ sh(FSR, AT, 0);
3123 if (!is_static) {
3124 patch_bytecode(Bytecodes::_fast_sputfield, bc, off, true, byte_no);
3125 }
3126 __ b(Done);
3127 __ delayed()->nop();
3128 __ bind(notShort);
3129 // ltos
3130 __ move(AT, ltos);
3131 __ bne(flags, AT, notLong);
3132 __ delayed()->nop();
3134 // FIXME: there is no simple method to load/store 64-bit data in a atomic operation
3135 // we just ignore the volatile flag.
3136 //Label notVolatileLong;
3137 //__ beq(T1, R0, notVolatileLong);
3138 //__ delayed()->nop();
3140 //addent = 2 * wordSize;
3141 // no need
3142 //__ lw(FSR, SP, 0);
3143 //__ lw(SSR, SP, 1 * wordSize);
3144 //if (!is_static) {
3145 // __ lw(T3, SP, addent);
3146 // addent += 1 * wordSize;
3147 // __ verify_oop(T3);
3148 //}
3150 //__ daddu(AT, T3, T2);
3152 // Replace with real volatile test
3153 // NOTE : we assume that sdc1&ldc1 operate in 32-bit, this is true for Godson2 even in 64-bit kernel
3154 // last modified by yjl 7/12/2005
3155 //__ ldc1(FSF, SP, 0);
3156 //__ sdc1(FSF, AT, 0);
3157 //volatile_barrier();
3159 // Don't rewrite volatile version
3160 //__ b(notVolatile);
3161 //__ delayed()->addiu(SP, SP, addent);
3163 //__ bind(notVolatileLong);
3165 //__ pop(ltos); // overwrites edx
3166 // __ lw(FSR, SP, 0 * wordSize);
3167 // __ lw(SSR, SP, 1 * wordSize);
3168 // __ daddi(SP, SP, 2*wordSize);
3169 __ pop(ltos);
3170 if (!is_static) {
3171 pop_and_check_object(obj);
3172 }
3173 __ dadd(AT, obj, off);
3174 __ sd(FSR, AT, 0);
3175 if (!is_static) {
3176 patch_bytecode(Bytecodes::_fast_lputfield, bc, off, true, byte_no);
3177 }
3178 __ b(notVolatile);
3179 __ delayed()->nop();
3181 __ bind(notLong);
3182 // ftos
3183 __ move(AT, ftos);
3184 __ bne(flags, AT, notFloat);
3185 __ delayed()->nop();
3187 __ pop(ftos);
3188 if (!is_static) {
3189 pop_and_check_object(obj);
3190 }
3191 __ dadd(AT, obj, off);
3192 __ swc1(FSF, AT, 0);
3193 if (!is_static) {
3194 patch_bytecode(Bytecodes::_fast_fputfield, bc, off, true, byte_no);
3195 }
3196 __ b(Done);
3197 __ delayed()->nop();
3198 __ bind(notFloat);
3199 // dtos
3200 __ move(AT, dtos);
3201 __ bne(flags, AT, notDouble);
3202 __ delayed()->nop();
3204 __ pop(dtos);
3205 if (!is_static) {
3206 pop_and_check_object(obj);
3207 }
3208 __ dadd(AT, obj, off);
3209 __ sdc1(FSF, AT, 0);
3210 if (!is_static) {
3211 patch_bytecode(Bytecodes::_fast_dputfield, bc, off, true, byte_no);
3212 }
3213 __ b(Done);
3214 __ delayed()->nop();
3215 __ bind(notDouble);
3217 __ stop("Bad state");
3219 __ bind(Done);
3221 // Check for volatile store
3222 __ beq(T8, R0, notVolatile);
3223 __ delayed()->nop();
3224 volatile_barrier( );
3225 __ bind(notVolatile);
3226 }
3228 void TemplateTable::putfield(int byte_no) {
3229 putfield_or_static(byte_no, false);
3230 }
3232 void TemplateTable::putstatic(int byte_no) {
3233 putfield_or_static(byte_no, true);
3234 }
3236 // used registers : T1, T2, T3
3237 // T1 : cp_entry
3238 // T2 : obj
3239 // T3 : value pointer
3240 void TemplateTable::jvmti_post_fast_field_mod() {
3241 if (JvmtiExport::can_post_field_modification()) {
3242 // Check to see if a field modification watch has been set before we take
3243 // the time to call into the VM.
3244 Label L2;
3245 //__ lui(AT, Assembler::split_high((intptr_t)JvmtiExport::get_field_modification_count_addr()));
3246 //__ lw(T3, AT, Assembler::split_low((intptr_t)JvmtiExport::get_field_modification_count_addr()));
3247 __ li(AT, JvmtiExport::get_field_modification_count_addr());
3248 __ lw(T3, AT, 0);
3249 __ beq(T3, R0, L2);
3250 __ delayed()->nop();
3251 //__ pop(T2);
3252 __ pop_ptr(T2);
3253 //__ lw(T2, SP, 0);
3254 __ verify_oop(T2);
3255 __ push_ptr(T2);
3256 __ li(AT, -sizeof(jvalue));
3257 __ daddu(SP, SP, AT);
3258 __ move(T3, SP);
3259 //__ push(T2);
3260 //__ move(T2, R0);
3262 switch (bytecode()) { // load values into the jvalue object
3263 case Bytecodes::_fast_bputfield:
3264 __ sb(FSR, SP, 0);
3265 break;
3266 case Bytecodes::_fast_sputfield:
3267 __ sh(FSR, SP, 0);
3268 break;
3269 case Bytecodes::_fast_cputfield:
3270 __ sh(FSR, SP, 0);
3271 break;
3272 case Bytecodes::_fast_iputfield:
3273 __ sw(FSR, SP, 0);
3274 break;
3275 case Bytecodes::_fast_lputfield:
3276 __ sd(FSR, SP, 0);
3277 break;
3278 case Bytecodes::_fast_fputfield:
3279 __ swc1(FSF, SP, 0);
3280 break;
3281 case Bytecodes::_fast_dputfield:
3282 __ sdc1(FSF, SP, 0);
3283 break;
3284 case Bytecodes::_fast_aputfield:
3285 __ sd(FSR, SP, 0);
3286 break;
3287 default: ShouldNotReachHere();
3288 }
3290 //__ pop(T2); // restore copy of object pointer
3292 // Save eax and sometimes edx because call_VM() will clobber them,
3293 // then use them for JVM/DI purposes
3294 __ push(FSR);
3295 if (bytecode() == Bytecodes::_fast_lputfield) __ push(SSR);
3296 // access constant pool cache entry
3297 __ get_cache_entry_pointer_at_bcp(T1, T2, 1);
3298 // no need, verified ahead
3299 __ verify_oop(T2);
3301 // ebx: object pointer copied above
3302 // eax: cache entry pointer
3303 // ecx: jvalue object on the stack
3304 __ call_VM(NOREG, CAST_FROM_FN_PTR(address,
3305 InterpreterRuntime::post_field_modification), T2, T1, T3);
3306 if (bytecode() == Bytecodes::_fast_lputfield) __ pop(SSR); // restore high value
3307 //__ pop(FSR); // restore lower value
3308 //__ daddi(SP, SP, sizeof(jvalue)); // release jvalue object space
3309 __ lw(FSR, SP, 0);
3310 __ daddiu(SP, SP, sizeof(jvalue) + 1 * wordSize);
3311 __ bind(L2);
3312 }
3313 }
3315 // used registers : T2, T3, T1
3316 // T2 : index & off & field address
3317 // T3 : cache & obj
3318 // T1 : flags
3319 void TemplateTable::fast_storefield(TosState state) {
3320 transition(state, vtos);
3322 ByteSize base = ConstantPoolCache::base_offset();
3324 jvmti_post_fast_field_mod();
3326 // access constant pool cache
3327 __ get_cache_and_index_at_bcp(T3, T2, 1);
3329 // test for volatile with edx but edx is tos register for lputfield.
3330 __ dsll(AT, T2, Address::times_8);
3331 __ dadd(AT, T3, AT);
3332 __ ld(T1, AT, in_bytes(base + ConstantPoolCacheEntry::flags_offset()));
3334 // replace index with field offset from cache entry
3335 __ ld(T2, AT, in_bytes(base + ConstantPoolCacheEntry::f2_offset()));
3337 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
3338 // volatile_barrier( );
3340 Label notVolatile, Done;
3341 // Check for volatile store
3342 __ move(AT, 1<<ConstantPoolCacheEntry::is_volatile_shift);
3343 __ andr(AT, T1, AT);
3344 __ beq(AT, R0, notVolatile);
3345 __ delayed()->nop();
3348 // Get object from stack
3349 // NOTE : the value in FSR/FSF now
3350 // __ pop(T3);
3351 // __ verify_oop(T3);
3352 pop_and_check_object(T3);
3353 // field addresses
3354 __ dadd(T2, T3, T2);
3356 // access field
3357 switch (bytecode()) {
3358 case Bytecodes::_fast_bputfield:
3359 __ sb(FSR, T2, 0);
3360 break;
3361 case Bytecodes::_fast_sputfield: // fall through
3362 case Bytecodes::_fast_cputfield:
3363 __ sh(FSR, T2, 0);
3364 break;
3365 case Bytecodes::_fast_iputfield:
3366 __ sw(FSR, T2, 0);
3367 break;
3368 case Bytecodes::_fast_lputfield:
3369 __ sd(FSR, T2, 0 * wordSize);
3370 break;
3371 case Bytecodes::_fast_fputfield:
3372 __ swc1(FSF, T2, 0);
3373 break;
3374 case Bytecodes::_fast_dputfield:
3375 __ sdc1(FSF, T2, 0 * wordSize);
3376 break;
3377 case Bytecodes::_fast_aputfield:
3378 __ store_heap_oop(Address(T2, 0), FSR);
3379 __ sync();
3380 __ store_check(T3);
3381 break;
3382 default:
3383 ShouldNotReachHere();
3384 }
3386 Label done;
3387 volatile_barrier( );
3388 __ b(done);
3389 __ delayed()->nop();
3391 // Same code as above, but don't need edx to test for volatile.
3392 __ bind(notVolatile);
3394 // Get object from stack
3395 // __ pop(T3);
3396 // __ verify_oop(T3);
3397 pop_and_check_object(T3);
3398 //get the field address
3399 __ dadd(T2, T3, T2);
3401 // access field
3402 switch (bytecode()) {
3403 case Bytecodes::_fast_bputfield:
3404 __ sb(FSR, T2, 0);
3405 break;
3406 case Bytecodes::_fast_sputfield: // fall through
3407 case Bytecodes::_fast_cputfield:
3408 __ sh(FSR, T2, 0);
3409 break;
3410 case Bytecodes::_fast_iputfield:
3411 __ sw(FSR, T2, 0);
3412 break;
3413 case Bytecodes::_fast_lputfield:
3414 __ sd(FSR, T2, 0 * wordSize);
3415 break;
3416 case Bytecodes::_fast_fputfield:
3417 __ swc1(FSF, T2, 0);
3418 break;
3419 case Bytecodes::_fast_dputfield:
3420 __ sdc1(FSF, T2, 0 * wordSize);
3421 break;
3422 case Bytecodes::_fast_aputfield:
3423 //add for compressedoops
3424 __ store_heap_oop(Address(T2, 0), FSR);
3425 __ sync();
3426 __ store_check(T3);
3427 break;
3428 default:
3429 ShouldNotReachHere();
3430 }
3431 __ bind(done);
3432 }
3434 // used registers : T2, T3, T1
3435 // T3 : cp_entry & cache
3436 // T2 : index & offset
3437 void TemplateTable::fast_accessfield(TosState state) {
3438 transition(atos, state);
3440 // do the JVMTI work here to avoid disturbing the register state below
3441 if (JvmtiExport::can_post_field_access()) {
3442 // Check to see if a field access watch has been set before we take
3443 // the time to call into the VM.
3444 Label L1;
3445 __ li(AT, (intptr_t)JvmtiExport::get_field_access_count_addr());
3446 __ lw(T3, AT, 0);
3447 __ beq(T3, R0, L1);
3448 __ delayed()->nop();
3449 // access constant pool cache entry
3450 __ get_cache_entry_pointer_at_bcp(T3, T1, 1);
3451 __ move(TSR, FSR);
3452 __ verify_oop(FSR);
3453 // FSR: object pointer copied above
3454 // T3: cache entry pointer
3455 __ call_VM(NOREG, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
3456 FSR, T3);
3457 __ move(FSR, TSR);
3458 __ bind(L1);
3459 }
3461 // access constant pool cache
3462 __ get_cache_and_index_at_bcp(T3, T2, 1);
3463 // replace index with field offset from cache entry
3464 __ dsll(AT, T2, Address::times_8);
3465 //__ dsll(AT, T2, 4);
3466 __ dadd(AT, T3, AT);
3467 __ ld(T2, AT, in_bytes(ConstantPoolCache::base_offset()
3468 + ConstantPoolCacheEntry::f2_offset()));
3470 // eax: object
3471 __ verify_oop(FSR);
3472 // __ null_check(FSR, 0);
3473 __ null_check(FSR);
3474 // field addresses
3475 __ dadd(FSR, FSR, T2);
3477 // access field
3478 switch (bytecode()) {
3479 case Bytecodes::_fast_bgetfield:
3480 __ lb(FSR, FSR, 0);
3481 break;
3482 case Bytecodes::_fast_sgetfield:
3483 __ lh(FSR, FSR, 0);
3484 break;
3485 case Bytecodes::_fast_cgetfield:
3486 __ lhu(FSR, FSR, 0);
3487 break;
3488 case Bytecodes::_fast_igetfield:
3489 __ lw(FSR, FSR, 0);
3490 break;
3491 case Bytecodes::_fast_lgetfield:
3492 __ stop("should not be rewritten");
3493 break;
3494 case Bytecodes::_fast_fgetfield:
3495 __ lwc1(FSF, FSR, 0);
3496 break;
3497 case Bytecodes::_fast_dgetfield:
3498 __ ldc1(FSF, FSR, 0);
3499 break;
3500 case Bytecodes::_fast_agetfield:
3501 //add for compressedoops
3502 __ load_heap_oop(FSR, Address(FSR, 0));
3503 __ verify_oop(FSR);
3504 break;
3505 default:
3506 ShouldNotReachHere();
3507 }
3509 // Doug Lea believes this is not needed with current Sparcs(TSO) and Intel(PSO)
3510 // volatile_barrier( );
3511 }
3513 // generator for _fast_iaccess_0, _fast_aaccess_0, _fast_faccess_0
3514 // used registers : T1, T2, T3, T1
3515 // T1 : obj & field address
3516 // T2 : off
3517 // T3 : cache
3518 // T1 : index
3519 void TemplateTable::fast_xaccess(TosState state) {
3520 transition(vtos, state);
3521 // get receiver
3522 __ ld(T1, aaddress(0));
3523 // access constant pool cache
3524 __ get_cache_and_index_at_bcp(T3, T2, 2);
3525 __ dsll(AT, T2, Address::times_8);
3526 __ dadd(AT, T3, AT);
3527 __ ld(T2, AT, in_bytes(ConstantPoolCache::base_offset()
3528 + ConstantPoolCacheEntry::f2_offset()));
3530 // make sure exception is reported in correct bcp range (getfield is next instruction)
3531 __ daddi(BCP, BCP, 1);
3532 // __ null_check(T1, 0);
3533 __ null_check(T1);
3534 __ dadd(T1, T1, T2);
3536 if (state == itos) {
3537 __ lw(FSR, T1, 0);
3538 } else if (state == atos) {
3539 //__ ld(FSR, T1, 0);
3540 __ load_heap_oop(FSR, Address(T1, 0));
3541 __ verify_oop(FSR);
3542 } else if (state == ftos) {
3543 __ lwc1(FSF, T1, 0);
3544 } else {
3545 ShouldNotReachHere();
3546 }
3547 __ daddi(BCP, BCP, -1);
3548 }
3550 //---------------------------------------------------
3551 //-------------------------------------------------
3552 // Calls
3554 void TemplateTable::count_calls(Register method, Register temp) {
3555 // implemented elsewhere
3556 ShouldNotReachHere();
3557 }
3559 // method, index, recv, flags: T1, T2, T3, T1
3560 // byte_no = 2 for _invokevirtual, 1 else
3561 // T0 : return address
3562 // get the method & index of the invoke, and push the return address of
3563 // the invoke(first word in the frame)
3564 // this address is where the return code jmp to.
3565 // NOTE : this method will set T3&T1 as recv&flags
3566 void TemplateTable::prepare_invoke(int byte_no,
3567 Register method, //linked method (or i-klass)
3568 Register index, //itable index, MethodType ,etc.
3569 Register recv, // if caller wants to see it
3570 Register flags // if caller wants to test it
3571 ) {
3572 // determine flags
3573 const Bytecodes::Code code = bytecode();
3574 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3575 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3576 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3577 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3578 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3579 const bool load_receiver = (recv != noreg);
3580 const bool save_flags = (flags != noreg);
3581 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic),"");
3582 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3583 assert(flags == noreg || flags == T1, "error flags reg.");
3584 assert(recv == noreg || recv == T3, "error recv reg.");
3585 // setup registers & access constant pool cache
3586 if(recv == noreg) recv = T3;
3587 if(flags == noreg) flags = T1;
3589 assert_different_registers(method, index, recv, flags);
3591 // save 'interpreter return address'
3592 __ save_bcp();
3594 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3595 if (is_invokedynamic || is_invokehandle) {
3596 Label L_no_push;
3597 __ move(AT, (1 << ConstantPoolCacheEntry::has_appendix_shift));
3598 __ andr(AT, AT, flags);
3599 __ beq(AT, R0, L_no_push);
3600 __ delayed()->nop();
3601 // Push the appendix as a trailing parameter.
3602 // This must be done before we get the receiver,
3603 // since the parameter_size includes it.
3604 Register tmp = SSR;
3605 __ push(tmp);
3606 __ move(tmp, index);
3607 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3608 __ load_resolved_reference_at_index(index, tmp);
3609 __ pop(tmp);
3610 __ push(index); // push appendix (MethodType, CallSite, etc.)
3611 __ bind(L_no_push);
3613 }
3615 // load receiver if needed (after appendix is pushed so parameter size is correct)
3616 // Note: no return address pushed yet
3617 if (load_receiver) {
3618 __ move(AT, ConstantPoolCacheEntry::parameter_size_mask);
3619 __ andr(recv, flags, AT);
3620 // 2014/07/31 Fu: Since we won't push RA on stack, no_return_pc_pushed_yet should be 0.
3621 const int no_return_pc_pushed_yet = 0; // argument slot correction before we push return address
3622 const int receiver_is_at_end = -1; // back off one slot to get receiver
3623 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3625 __ ld(recv, recv_addr);
3626 __ verify_oop(recv);
3627 }
3628 if(save_flags) {
3629 //__ movl(r13, flags);
3630 __ move(BCP, flags);
3631 }
3632 // compute return type
3633 __ dsrl(flags, flags, ConstantPoolCacheEntry::tos_state_shift);
3634 __ andi(flags, flags, 0xf);
3636 // Make sure we don't need to mask flags for tos_state_shift after the above shift
3637 ConstantPoolCacheEntry::verify_tos_state_shift();
3638 // load return address
3639 {
3640 const address table = (address) Interpreter::invoke_return_entry_table_for(code);
3641 __ li(AT, (long)table);
3642 __ dsll(flags, flags, LogBytesPerWord);
3643 __ dadd(AT, AT, flags);
3644 __ ld(RA, AT, 0);
3645 }
3647 if (save_flags) {
3648 __ move(flags, BCP);
3649 __ restore_bcp();
3650 }
3651 }
3653 // used registers : T0, T3, T1, T2
3654 // T3 : recv, this two register using convention is by prepare_invoke
3655 // T1 : flags, klass
3656 // Rmethod : method, index must be Rmethod
3657 void TemplateTable::invokevirtual_helper(Register index, Register recv,
3658 Register flags) {
3660 assert_different_registers(index, recv, flags, T2);
3662 // Test for an invoke of a final method
3663 Label notFinal;
3664 __ move(AT, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3665 __ andr(AT, flags, AT);
3666 __ beq(AT, R0, notFinal);
3667 __ delayed()->nop();
3669 Register method = index; // method must be Rmethod
3670 assert(method == Rmethod, "methodOop must be Rmethod for interpreter calling convention");
3672 // do the call - the index is actually the method to call
3673 // the index is indeed methodOop, for this is vfinal,
3674 // see ConstantPoolCacheEntry::set_method for more info
3676 __ verify_oop(method);
3678 // It's final, need a null check here!
3679 __ null_check(recv);
3681 // profile this call
3682 __ profile_final_call(T2);
3684 // 2014/11/24 Fu
3685 // T2: tmp, used for mdp
3686 // method: callee
3687 // T9: tmp
3688 // is_virtual: true
3689 __ profile_arguments_type(T2, method, T9, true);
3691 // __ move(T0, recv);
3692 __ jump_from_interpreted(method, T2);
3694 __ bind(notFinal);
3696 // get receiver klass
3697 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3698 // Keep recv in ecx for callee expects it there
3699 __ load_klass(T2, recv);
3700 __ verify_oop(T2);
3701 // profile this call
3702 __ profile_virtual_call(T2, T0, T1);
3704 // get target methodOop & entry point
3705 const int base = InstanceKlass::vtable_start_offset() * wordSize;
3706 assert(vtableEntry::size() * wordSize == 8, "adjust the scaling in the code below");
3707 __ dsll(AT, index, Address::times_8);
3708 __ dadd(AT, T2, AT);
3709 //this is a ualign read
3710 __ ld(method, AT, base + vtableEntry::method_offset_in_bytes());
3711 __ jump_from_interpreted(method, T2);
3713 }
3715 void TemplateTable::invokevirtual(int byte_no) {
3716 transition(vtos, vtos);
3717 assert(byte_no == f2_byte, "use this argument");
3718 prepare_invoke(byte_no, Rmethod, NOREG, T3, T1);
3719 // now recv & flags in T3, T1
3720 invokevirtual_helper(Rmethod, T3, T1);
3721 }
3723 // T9 : entry
3724 // Rmethod : method
3725 void TemplateTable::invokespecial(int byte_no) {
3726 transition(vtos, vtos);
3727 assert(byte_no == f1_byte, "use this argument");
3728 prepare_invoke(byte_no, Rmethod, NOREG, T3);
3729 // now recv & flags in T3, T1
3730 __ verify_oop(T3);
3731 __ null_check(T3);
3732 __ profile_call(T9);
3734 // 2014/11/24 Fu
3735 // T8: tmp, used for mdp
3736 // Rmethod: callee
3737 // T9: tmp
3738 // is_virtual: false
3739 __ profile_arguments_type(T8, Rmethod, T9, false);
3741 __ jump_from_interpreted(Rmethod, T9);
3742 __ move(T0, T3);//aoqi ?
3743 }
3745 void TemplateTable::invokestatic(int byte_no) {
3746 transition(vtos, vtos);
3747 assert(byte_no == f1_byte, "use this argument");
3748 prepare_invoke(byte_no, Rmethod, NOREG);
3749 __ verify_oop(Rmethod);
3751 __ profile_call(T9);
3753 // 2014/11/24 Fu
3754 // T8: tmp, used for mdp
3755 // Rmethod: callee
3756 // T9: tmp
3757 // is_virtual: false
3758 __ profile_arguments_type(T8, Rmethod, T9, false);
3760 __ jump_from_interpreted(Rmethod, T9);
3761 }
3763 // i have no idea what to do here, now. for future change. FIXME.
3764 void TemplateTable::fast_invokevfinal(int byte_no) {
3765 transition(vtos, vtos);
3766 assert(byte_no == f2_byte, "use this argument");
3767 __ stop("fast_invokevfinal not used on x86");
3768 }
3770 // used registers : T0, T1, T2, T3, T1, A7
3771 // T0 : itable, vtable, entry
3772 // T1 : interface
3773 // T3 : receiver
3774 // T1 : flags, klass
3775 // Rmethod : index, method, this is required by interpreter_entry
3776 void TemplateTable::invokeinterface(int byte_no) {
3777 transition(vtos, vtos);
3778 //this method will use T1-T4 and T0
3779 assert(byte_no == f1_byte, "use this argument");
3780 prepare_invoke(byte_no, T2, Rmethod, T3, T1);
3781 // T2: Interface
3782 // Rmethod: index
3783 // T3: receiver
3784 // T1: flags
3785 Label notMethod;
3786 __ move(AT, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3787 __ andr(AT, T1, AT);
3788 __ beq(AT, R0, notMethod);
3789 __ delayed()->nop();
3791 // Special case of invokeinterface called for virtual method of
3792 // java.lang.Object. See cpCacheOop.cpp for details.
3793 // This code isn't produced by javac, but could be produced by
3794 // another compliant java compiler.
3795 invokevirtual_helper(Rmethod, T3, T1);
3797 __ bind(notMethod);
3798 // Get receiver klass into T1 - also a null check
3799 //__ ld(T1, T3, oopDesc::klass_offset_in_bytes());
3800 //add for compressedoops
3801 //__ restore_locals();
3802 //__ null_check(T3, oopDesc::klass_offset_in_bytes());
3803 __ load_klass(T1, T3);
3804 __ verify_oop(T1);
3806 // profile this call
3807 __ profile_virtual_call(T1, T0, FSR);
3809 // Compute start of first itableOffsetEntry (which is at the end of the vtable)
3810 // TODO: x86 add a new method lookup_interface_method // LEE
3811 const int base = InstanceKlass::vtable_start_offset() * wordSize;
3812 assert(vtableEntry::size() * wordSize == 8, "adjust the scaling in the code below");
3813 __ lw(AT, T1, InstanceKlass::vtable_length_offset() * wordSize);
3814 __ dsll(AT, AT, Address::times_8);
3815 __ dadd(T0, T1, AT);
3816 __ daddi(T0, T0, base);
3817 if (HeapWordsPerLong > 1) {
3818 // Round up to align_object_offset boundary
3819 __ round_to(T0, BytesPerLong);
3820 }
3821 // now T0 is the begin of the itable
3823 Label entry, search, interface_ok;
3825 ///__ jmp(entry);
3826 __ b(entry);
3827 __ delayed()->nop();
3829 __ bind(search);
3830 __ increment(T0, itableOffsetEntry::size() * wordSize);
3832 __ bind(entry);
3834 // Check that the entry is non-null. A null entry means that the receiver
3835 // class doesn't implement the interface, and wasn't the same as the
3836 // receiver class checked when the interface was resolved.
3837 __ ld(AT, T0, itableOffsetEntry::interface_offset_in_bytes());
3838 __ bne(AT, R0, interface_ok);
3839 __ delayed()->nop();
3840 // throw exception
3841 // the call_VM checks for exception, so we should never return here.
3843 //__ pop();//FIXME here,
3844 // pop return address (pushed by prepare_invoke).
3845 // no need now, we just save the value in RA now
3847 __ call_VM(NOREG, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeError));
3848 __ should_not_reach_here();
3850 __ bind(interface_ok);
3851 //NOTICE here, no pop as x86 do
3852 //__ lw(AT, T0, itableOffsetEntry::interface_offset_in_bytes());
3853 __ bne(AT, T2, search);
3854 __ delayed()->nop();
3856 // now we get vtable of the interface
3857 __ ld(T0, T0, itableOffsetEntry::offset_offset_in_bytes());
3858 __ daddu(T0, T1, T0);
3859 assert(itableMethodEntry::size() * wordSize == 8, "adjust the scaling in the code below");
3860 __ dsll(AT, Rmethod, Address::times_8);
3861 __ daddu(AT, T0, AT);
3862 // now we get the method
3863 __ ld(Rmethod, AT, 0);
3864 // Rnext: methodOop to call
3865 // T3: receiver
3866 // Check for abstract method error
3867 // Note: This should be done more efficiently via a throw_abstract_method_error
3868 // interpreter entry point and a conditional jump to it in case of a null
3869 // method.
3870 {
3871 Label L;
3872 ///__ testl(ebx, ebx);
3873 ///__ jcc(Assembler::notZero, L);
3874 __ bne(Rmethod, R0, L);
3875 __ delayed()->nop();
3877 // throw exception
3878 // note: must restore interpreter registers to canonical
3879 // state for exception handling to work correctly!
3880 ///__ popl(ebx); // pop return address (pushed by prepare_invoke)
3881 //__ restore_bcp(); // esi must be correct for exception handler
3882 //(was destroyed)
3883 //__ restore_locals(); // make sure locals pointer
3884 //is correct as well (was destroyed)
3885 ///__ call_VM(noreg, CAST_FROM_FN_PTR(address,
3886 //InterpreterRuntime::throw_AbstractMethodError));
3887 __ call_VM(NOREG, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3888 // the call_VM checks for exception, so we should never return here.
3889 __ should_not_reach_here();
3890 __ bind(L);
3891 }
3893 // 2014/11/24 Fu
3894 // T8: tmp, used for mdp
3895 // Rmethod: callee
3896 // T9: tmp
3897 // is_virtual: true
3898 __ profile_arguments_type(T8, Rmethod, T9, true);
3900 __ jump_from_interpreted(Rmethod, T9);
3901 }
3903 void TemplateTable::invokehandle(int byte_no) {
3904 transition(vtos, vtos);
3905 assert(byte_no == f1_byte, "use this argument");
3906 const Register T2_method = Rmethod;
3907 const Register FSR_mtype = FSR;
3908 const Register T3_recv = T3;
3910 if (!EnableInvokeDynamic) {
3911 // rewriter does not generate this bytecode
3912 __ should_not_reach_here();
3913 return;
3914 }
3916 prepare_invoke(byte_no, T2_method, FSR_mtype, T3_recv);
3917 //??__ verify_method_ptr(T2_method);
3918 __ verify_oop(T3_recv);
3919 __ null_check(T3_recv);
3921 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3922 // rbx: MH.invokeExact_MT method (from f2)
3924 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke
3926 // FIXME: profile the LambdaForm also
3927 __ profile_final_call(T9);
3929 // 2014/11/24 Fu
3930 // T8: tmp, used for mdp
3931 // T2_method: callee
3932 // T9: tmp
3933 // is_virtual: true
3934 __ profile_arguments_type(T8, T2_method, T9, true);
3936 __ jump_from_interpreted(T2_method, T9);
3937 }
3939 void TemplateTable::invokedynamic(int byte_no) {
3940 transition(vtos, vtos);
3941 assert(byte_no == f1_byte, "use this argument");
3943 if (!EnableInvokeDynamic) {
3944 // We should not encounter this bytecode if !EnableInvokeDynamic.
3945 // The verifier will stop it. However, if we get past the verifier,
3946 // this will stop the thread in a reasonable way, without crashing the JVM.
3947 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3948 InterpreterRuntime::throw_IncompatibleClassChangeError));
3949 // the call_VM checks for exception, so we should never return here.
3950 __ should_not_reach_here();
3951 return;
3952 }
3954 //const Register Rmethod = T2;
3955 const Register T2_callsite = T2;
3957 prepare_invoke(byte_no, Rmethod, T2_callsite);
3959 // rax: CallSite object (from cpool->resolved_references[f1])
3960 // rbx: MH.linkToCallSite method (from f2)
3962 // Note: rax_callsite is already pushed by prepare_invoke
3963 // %%% should make a type profile for any invokedynamic that takes a ref argument
3964 // profile this call
3965 __ profile_call(T9);
3967 // 2014/11/24 Fu
3968 // T8: tmp, used for mdp
3969 // Rmethod: callee
3970 // T9: tmp
3971 // is_virtual: false
3972 __ profile_arguments_type(T8, Rmethod, T9, false);
3974 __ verify_oop(T2_callsite);
3976 __ jump_from_interpreted(Rmethod, T9);
3977 }
3979 //----------------------------------------------------------------------------------------------------
3980 // Allocation
3981 // T1 : tags & buffer end & thread
3982 // T2 : object end
3983 // T3 : klass
3984 // T1 : object size
3985 // A1 : cpool
3986 // A2 : cp index
3987 // return object in FSR
3988 void TemplateTable::_new() {
3989 transition(vtos, atos);
3990 __ get_2_byte_integer_at_bcp(A2, AT, 1);
3991 __ huswap(A2);
3993 Label slow_case;
3994 Label done;
3995 Label initialize_header;
3996 Label initialize_object; // including clearing the fields
3997 Label allocate_shared;
3999 // get InstanceKlass in T3
4000 __ get_cpool_and_tags(A1, T1);
4001 __ dsll(AT, A2, Address::times_8);
4002 __ dadd(AT, A1, AT);
4003 __ ld(T3, AT, sizeof(ConstantPool));
4005 // make sure the class we're about to instantiate has been resolved.
4006 // Note: slow_case does a pop of stack, which is why we loaded class/pushed above
4007 const int tags_offset = Array<u1>::base_offset_in_bytes();
4008 __ dadd(T1, T1, A2);
4009 __ lb(AT, T1, tags_offset);
4010 //__ addiu(AT, AT, - (int)JVM_CONSTANT_UnresolvedClass);
4011 __ daddiu(AT, AT, - (int)JVM_CONSTANT_Class);
4012 //__ beq(AT, R0, slow_case);
4013 __ bne(AT, R0, slow_case);
4014 __ delayed()->nop();
4016 /*make sure klass is initialized & doesn't have finalizer*/
4018 // make sure klass is fully initialized
4019 __ lhu(T1, T3, in_bytes(InstanceKlass::init_state_offset()));
4020 __ daddiu(AT, T1, - (int)InstanceKlass::fully_initialized);
4021 __ bne(AT, R0, slow_case);
4022 __ delayed()->nop();
4024 // has_finalizer
4025 //__ lw(T1, T3, Klass::access_flags_offset() + sizeof(oopDesc));
4026 //__ move(AT, JVM_ACC_CAN_BE_FASTPATH_ALLOCATED);
4027 //__ andr(AT, T1, AT);
4028 __ lw(T1, T3, in_bytes(Klass::layout_helper_offset()) );
4029 __ andi(AT, T1, Klass::_lh_instance_slow_path_bit);
4030 __ bne(AT, R0, slow_case);
4031 __ delayed()->nop();
4033 // get instance_size in InstanceKlass (already aligned) in T0,
4034 // be sure to preserve this value
4035 //__ lw(T0, T3, Klass::size_helper_offset_in_bytes() + sizeof(oopDesc));
4036 //Klass::_size_helper is renamed Klass::_layout_helper. aoqi
4037 __ lw(T0, T3, in_bytes(Klass::layout_helper_offset()) );
4039 //
4040 // Allocate the instance
4041 // 1) Try to allocate in the TLAB
4042 // 2) if fail and the object is large allocate in the shared Eden
4043 // 3) if the above fails (or is not applicable), go to a slow case
4044 // (creates a new TLAB, etc.)
4046 const bool allow_shared_alloc =
4047 Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
4049 if (UseTLAB) {
4050 #ifndef OPT_THREAD
4051 const Register thread = T8;
4052 __ get_thread(thread);
4053 #else
4054 const Register thread = TREG;
4055 #endif
4056 // get tlab_top
4057 __ ld(FSR, thread, in_bytes(JavaThread::tlab_top_offset()));
4058 __ dadd(T2, FSR, T0);
4059 // get tlab_end
4060 __ ld(AT, thread, in_bytes(JavaThread::tlab_end_offset()));
4061 __ slt(AT, AT, T2);
4062 // __ bne(AT, R0, allocate_shared);
4063 __ bne(AT, R0, allow_shared_alloc ? allocate_shared : slow_case);
4064 __ delayed()->nop();
4065 __ sd(T2, thread, in_bytes(JavaThread::tlab_top_offset()));
4067 if (ZeroTLAB) {
4068 // the fields have been already cleared
4069 __ b_far(initialize_header);
4070 } else {
4071 // initialize both the header and fields
4072 __ b_far(initialize_object);
4073 }
4074 __ delayed()->nop();
4075 /*
4077 if (CMSIncrementalMode) {
4078 // No allocation in shared eden.
4079 ///__ jmp(slow_case);
4080 __ b(slow_case);
4081 __ delayed()->nop();
4082 }
4083 */
4084 }
4086 // Allocation in the shared Eden , if allowed
4087 // T0 : instance size in words
4088 if(allow_shared_alloc){
4089 __ bind(allocate_shared);
4090 Label retry;
4091 //Address heap_top(T1, (int)Universe::heap()->top_addr());
4092 Address heap_top(T1);
4093 //__ lui(T1, Assembler::split_high((int)Universe::heap()->top_addr()));
4094 __ li(T1, (long)Universe::heap()->top_addr());
4096 __ ld(FSR, heap_top);
4097 __ bind(retry);
4098 __ dadd(T2, FSR, T0);
4099 //__ lui(AT, Assembler::split_high((int)Universe::heap()->end_addr()));
4100 //__ lw(AT, AT, Assembler::split_low((int)Universe::heap()->end_addr()));
4101 __ li(AT, (long)Universe::heap()->end_addr());
4102 __ ld(AT, AT, 0);
4103 __ slt(AT, AT, T2);
4104 __ bne(AT, R0, slow_case);
4105 __ delayed()->nop();
4107 // Compare FSR with the top addr, and if still equal, store the new
4108 // top addr in ebx at the address of the top addr pointer. Sets ZF if was
4109 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
4110 //
4111 // FSR: object begin
4112 // T2: object end
4113 // T0: instance size in words
4115 // if someone beat us on the allocation, try again, otherwise continue
4116 //__ lui(T1, Assembler::split_high((int)Universe::heap()->top_addr()));
4117 __ cmpxchg(T2, heap_top, FSR);
4118 __ beq(AT, R0, retry);
4119 __ delayed()->nop();
4120 }
4122 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
4123 // The object is initialized before the header. If the object size is
4124 // zero, go directly to the header initialization.
4125 __ bind(initialize_object);
4126 __ li(AT, - sizeof(oopDesc));
4127 __ daddu(T0, T0, AT);
4128 __ beq_far(T0, R0, initialize_header);
4129 __ delayed()->nop();
4132 // T0 must have been multiple of 2
4133 #ifdef ASSERT
4134 // make sure T0 was multiple of 2
4135 Label L;
4136 __ andi(AT, T0, 1);
4137 __ beq(AT, R0, L);
4138 __ delayed()->nop();
4139 __ stop("object size is not multiple of 2 - adjust this code");
4140 __ bind(L);
4141 // edx must be > 0, no extra check needed here
4142 #endif
4144 // initialize remaining object fields: T0 is a multiple of 2
4145 {
4146 Label loop;
4147 __ dadd(T1, FSR, T0);
4148 __ daddi(T1, T1, -oopSize);
4150 __ bind(loop);
4151 __ sd(R0, T1, sizeof(oopDesc) + 0 * oopSize);
4152 // __ sd(R0, T1, sizeof(oopDesc) + 1 * oopSize);
4153 __ bne(T1, FSR, loop); //dont clear header
4154 __ delayed()->daddi(T1, T1, -oopSize);
4155 // actually sizeof(oopDesc)==8, so we can move
4156 // __ addiu(AT, AT, -8) to delay slot, and compare FSR with T1
4157 }
4158 //klass in T3,
4159 // initialize object header only.
4160 __ bind(initialize_header);
4161 if (UseBiasedLocking) {
4162 // __ popl(ecx); // get saved klass back in the register.
4163 // __ movl(ebx, Address(ecx, Klass::prototype_header_offset_in_bytes()
4164 // + klassOopDesc::klass_part_offset_in_bytes()));
4165 __ ld(AT, T3, in_bytes(Klass::prototype_header_offset()));
4166 // __ movl(Address(eax, oopDesc::mark_offset_in_bytes ()), ebx);
4167 __ sd(AT, FSR, oopDesc::mark_offset_in_bytes ());
4168 } else {
4169 __ li(AT, (long)markOopDesc::prototype());
4170 __ sd(AT, FSR, oopDesc::mark_offset_in_bytes());
4171 }
4173 //__ sd(T3, FSR, oopDesc::klass_offset_in_bytes());
4174 __ store_klass_gap(FSR, R0);
4175 __ store_klass(FSR, T3);
4177 {
4178 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
4179 // Trigger dtrace event for fastpath
4180 __ push(atos);
4181 __ call_VM_leaf(
4182 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), FSR);
4183 __ pop(atos);
4184 }
4185 __ b(done);
4186 __ delayed()->nop();
4187 }
4188 // slow case
4189 __ bind(slow_case);
4190 call_VM(FSR, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), A1, A2);
4192 // continue
4193 __ bind(done);
4194 __ sync();
4195 }
4197 void TemplateTable::newarray() {
4198 transition(itos, atos);
4199 __ lbu(A1, at_bcp(1));
4200 //type, count
4201 call_VM(FSR, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), A1, FSR);
4202 __ sync();
4203 }
4205 void TemplateTable::anewarray() {
4206 transition(itos, atos);
4207 __ get_2_byte_integer_at_bcp(A2, AT, 1);
4208 __ huswap(A2);
4209 __ get_constant_pool(A1);
4210 // cp, index, count
4211 call_VM(FSR, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), A1, A2, FSR);
4212 __ sync();
4213 }
4215 void TemplateTable::arraylength() {
4216 transition(atos, itos);
4217 __ null_check(FSR, arrayOopDesc::length_offset_in_bytes());
4218 __ lw(FSR, FSR, arrayOopDesc::length_offset_in_bytes());
4219 }
4221 // i use T2 as ebx, T3 as ecx, T1 as edx
4222 // when invoke gen_subtype_check, super in T3, sub in T2, object in FSR(it's always)
4223 // T2 : sub klass
4224 // T3 : cpool
4225 // T3 : super klass
4226 void TemplateTable::checkcast() {
4227 transition(atos, atos);
4228 Label done, is_null, ok_is_subtype, quicked, resolved;
4229 __ beq(FSR, R0, is_null);
4230 __ delayed()->nop();
4232 // Get cpool & tags index
4233 __ get_cpool_and_tags(T3, T1);
4234 __ get_2_byte_integer_at_bcp(T2, AT, 1);
4235 __ huswap(T2);
4237 // See if bytecode has already been quicked
4238 __ dadd(AT, T1, T2);
4239 __ lb(AT, AT, Array<u1>::base_offset_in_bytes());
4240 __ daddiu(AT, AT, - (int)JVM_CONSTANT_Class);
4241 __ beq(AT, R0, quicked);
4242 __ delayed()->nop();
4244 /* 2012/6/2 Jin: In InterpreterRuntime::quicken_io_cc, lots of new classes may be loaded.
4245 * Then, GC will move the object in V0 to another places in heap.
4246 * Therefore, We should never save such an object in register.
4247 * Instead, we should save it in the stack. It can be modified automatically by the GC thread.
4248 * After GC, the object address in FSR is changed to a new place.
4249 */
4250 __ push(atos);
4251 const Register thread = TREG;
4252 #ifndef OPT_THREAD
4253 __ get_thread(thread);
4254 #endif
4255 call_VM(NOREG, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4256 __ get_vm_result_2(T3, thread);
4257 __ pop_ptr(FSR);
4258 __ b(resolved);
4259 __ delayed()->nop();
4261 // klass already in cp, get superklass in T3
4262 __ bind(quicked);
4263 __ dsll(AT, T2, Address::times_8);
4264 __ dadd(AT, T3, AT);
4265 __ ld(T3, AT, sizeof(ConstantPool));
4267 __ bind(resolved);
4269 // get subklass in T2
4270 //__ ld(T2, FSR, oopDesc::klass_offset_in_bytes());
4271 //add for compressedoops
4272 __ load_klass(T2, FSR);
4273 // Superklass in T3. Subklass in T2.
4274 __ gen_subtype_check(T3, T2, ok_is_subtype);
4276 // Come here on failure
4277 // object is at FSR
4278 __ jmp(Interpreter::_throw_ClassCastException_entry);
4279 __ delayed()->nop();
4281 // Come here on success
4282 __ bind(ok_is_subtype);
4284 // Collect counts on whether this check-cast sees NULLs a lot or not.
4285 if (ProfileInterpreter) {
4286 __ b(done);
4287 __ delayed()->nop();
4288 __ bind(is_null);
4289 __ profile_null_seen(T3);
4290 } else {
4291 __ bind(is_null);
4292 }
4293 __ bind(done);
4294 }
4296 // i use T3 as cpool, T1 as tags, T2 as index
4297 // object always in FSR, superklass in T3, subklass in T2
4298 void TemplateTable::instanceof() {
4299 transition(atos, itos);
4300 Label done, is_null, ok_is_subtype, quicked, resolved;
4302 __ beq(FSR, R0, is_null);
4303 __ delayed()->nop();
4305 // Get cpool & tags index
4306 __ get_cpool_and_tags(T3, T1);
4307 // get index
4308 __ get_2_byte_integer_at_bcp(T2, AT, 1);
4309 __ hswap(T2);
4311 // See if bytecode has already been quicked
4312 // quicked
4313 __ daddu(AT, T1, T2);
4314 __ lb(AT, AT, Array<u1>::base_offset_in_bytes());
4315 __ daddiu(AT, AT, - (int)JVM_CONSTANT_Class);
4316 __ beq(AT, R0, quicked);
4317 __ delayed()->nop();
4319 // get superklass in T3
4320 //__ move(TSR, FSR);
4321 // sometimes S2 may be changed during the call,
4322 // be careful if u use TSR as a saving place
4323 //__ push(FSR);
4324 __ push(atos);
4325 const Register thread = TREG;
4326 #ifndef OPT_THREAD
4327 __ get_thread(thread);
4328 #endif
4329 call_VM(NOREG, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4330 __ get_vm_result_2(T3, thread);
4331 //__ lw(FSR, SP, 0);
4332 __ pop_ptr(FSR);
4333 __ b(resolved);
4334 __ delayed()->nop();
4335 //__ move(FSR, TSR);
4337 // get superklass in T3, subklass in T2
4338 __ bind(quicked);
4339 __ dsll(AT, T2, Address::times_8);
4340 __ daddu(AT, T3, AT);
4341 __ ld(T3, AT, sizeof(ConstantPool));
4343 __ bind(resolved);
4344 // get subklass in T2
4345 //__ ld(T2, FSR, oopDesc::klass_offset_in_bytes());
4346 //add for compressedoops
4347 __ load_klass(T2, FSR);
4349 // Superklass in T3. Subklass in T2.
4350 __ gen_subtype_check(T3, T2, ok_is_subtype);
4351 // Come here on failure
4352 __ b(done);
4353 __ delayed(); __ move(FSR, R0);
4355 // Come here on success
4356 __ bind(ok_is_subtype);
4357 __ move(FSR, 1);
4359 // Collect counts on whether this test sees NULLs a lot or not.
4360 if (ProfileInterpreter) {
4361 __ beq(R0, R0, done);
4362 __ nop();
4363 __ bind(is_null);
4364 __ profile_null_seen(T3);
4365 } else {
4366 __ bind(is_null); // same as 'done'
4367 }
4368 __ bind(done);
4369 // FSR = 0: obj == NULL or obj is not an instanceof the specified klass
4370 // FSR = 1: obj != NULL and obj is an instanceof the specified klass
4371 }
4373 //--------------------------------------------------------
4374 //--------------------------------------------
4375 // Breakpoints
4376 void TemplateTable::_breakpoint() {
4378 // Note: We get here even if we are single stepping..
4379 // jbug inists on setting breakpoints at every bytecode
4380 // even if we are in single step mode.
4382 transition(vtos, vtos);
4384 // get the unpatched byte code
4385 ///__ get_method(ecx);
4386 ///__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at)
4387 //, ecx, esi);
4388 ///__ movl(ebx, eax);
4389 __ get_method(A1);
4390 __ call_VM(NOREG, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at),
4391 A1, BCP);
4392 __ move(Rnext, V0); // Jin: Rnext will be used in dispatch_only_normal
4394 // post the breakpoint event
4395 ///__ get_method(ecx);
4396 ///__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), ecx, esi);
4397 __ get_method(A1);
4398 __ call_VM(NOREG, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), A1, BCP);
4400 // complete the execution of original bytecode
4401 __ dispatch_only_normal(vtos);
4402 }
4404 //----------------------------------------------------------------------------------------------------
4405 // Exceptions
4407 void TemplateTable::athrow() {
4408 transition(atos, vtos);
4409 __ null_check(FSR);
4410 __ jmp(Interpreter::throw_exception_entry());
4411 __ delayed()->nop();
4412 }
4414 //----------------------------------------------------------------------------------------------------
4415 // Synchronization
4416 //
4417 // Note: monitorenter & exit are symmetric routines; which is reflected
4418 // in the assembly code structure as well
4419 //
4420 // Stack layout:
4421 //
4422 // [expressions ] <--- SP = expression stack top
4423 // ..
4424 // [expressions ]
4425 // [monitor entry] <--- monitor block top = expression stack bot
4426 // ..
4427 // [monitor entry]
4428 // [frame data ] <--- monitor block bot
4429 // ...
4430 // [return addr ] <--- FP
4432 // we use T2 as monitor entry pointer, T3 as monitor top pointer, c_rarg0 as free slot pointer
4433 // object always in FSR
4434 void TemplateTable::monitorenter() {
4435 transition(atos, vtos);
4436 // check for NULL object
4437 __ null_check(FSR);
4439 const Address monitor_block_top(FP, frame::interpreter_frame_monitor_block_top_offset
4440 * wordSize);
4441 const int entry_size = (frame::interpreter_frame_monitor_size()* wordSize);
4442 Label allocated;
4444 // initialize entry pointer
4445 __ move(c_rarg0, R0);
4447 // find a free slot in the monitor block (result in edx)
4448 {
4449 Label entry, loop, exit, next;
4450 __ ld(T2, monitor_block_top);
4451 __ b(entry);
4452 __ delayed()->daddi(T3, FP, frame::interpreter_frame_initial_sp_offset * wordSize);
4454 // free slot?
4455 __ bind(loop);
4456 __ ld(AT, T2, BasicObjectLock::obj_offset_in_bytes());
4457 __ bne(AT, R0, next);
4458 __ delayed()->nop();
4459 __ move(c_rarg0, T2);
4461 __ bind(next);
4462 __ beq(FSR, AT, exit);
4463 __ delayed()->nop();
4464 __ daddi(T2, T2, entry_size);
4466 __ bind(entry);
4467 __ bne(T3, T2, loop);
4468 __ delayed()->nop();
4469 __ bind(exit);
4470 }
4472 __ bne(c_rarg0, R0, allocated);
4473 __ delayed()->nop();
4475 // allocate one if there's no free slot
4476 {
4477 Label entry, loop;
4478 // 1. compute new pointers // SP: old expression stack top
4479 __ ld(c_rarg0, monitor_block_top);
4480 __ daddi(SP, SP, - entry_size);
4481 __ daddi(c_rarg0, c_rarg0, - entry_size);
4482 __ sd(c_rarg0, monitor_block_top);
4483 __ b(entry);
4484 __ delayed(); __ move(T3, SP);
4486 // 2. move expression stack contents
4487 __ bind(loop);
4488 __ ld(AT, T3, entry_size);
4489 __ sd(AT, T3, 0);
4490 __ daddi(T3, T3, wordSize);
4491 __ bind(entry);
4492 __ bne(T3, c_rarg0, loop);
4493 __ delayed()->nop();
4494 }
4496 __ bind(allocated);
4497 // Increment bcp to point to the next bytecode,
4498 // so exception handling for async. exceptions work correctly.
4499 // The object has already been poped from the stack, so the
4500 // expression stack looks correct.
4501 __ daddi(BCP, BCP, 1);
4502 __ sd(FSR, c_rarg0, BasicObjectLock::obj_offset_in_bytes());
4503 __ lock_object(c_rarg0);
4504 // check to make sure this monitor doesn't cause stack overflow after locking
4505 __ save_bcp(); // in case of exception
4506 __ generate_stack_overflow_check(0);
4507 // The bcp has already been incremented. Just need to dispatch to next instruction.
4509 __ dispatch_next(vtos);
4510 }
4512 // T2 : top
4513 // c_rarg0 : entry
4514 void TemplateTable::monitorexit() {
4515 transition(atos, vtos);
4517 __ null_check(FSR);
4519 const int entry_size =(frame::interpreter_frame_monitor_size()* wordSize);
4520 Label found;
4522 // find matching slot
4523 {
4524 Label entry, loop;
4525 __ ld(c_rarg0, FP, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4526 __ b(entry);
4527 __ delayed()->daddiu(T2, FP, frame::interpreter_frame_initial_sp_offset * wordSize);
4529 __ bind(loop);
4530 __ ld(AT, c_rarg0, BasicObjectLock::obj_offset_in_bytes());
4531 __ beq(FSR, AT, found);
4532 __ delayed()->nop();
4533 __ daddiu(c_rarg0, c_rarg0, entry_size);
4534 __ bind(entry);
4535 __ bne(T2, c_rarg0, loop);
4536 __ delayed()->nop();
4537 }
4539 // error handling. Unlocking was not block-structured
4540 Label end;
4541 __ call_VM(NOREG, CAST_FROM_FN_PTR(address,
4542 InterpreterRuntime::throw_illegal_monitor_state_exception));
4543 __ should_not_reach_here();
4545 // call run-time routine
4546 // c_rarg0: points to monitor entry
4547 __ bind(found);
4548 __ move(TSR, FSR);
4549 __ unlock_object(c_rarg0);
4550 __ move(FSR, TSR);
4551 __ bind(end);
4552 }
4554 //--------------------------------------------------------------------------------------------------// Wide instructions
4556 void TemplateTable::wide() {
4557 transition(vtos, vtos);
4558 // Note: the esi increment step is part of the individual wide bytecode implementations
4559 __ lbu(Rnext, at_bcp(1));
4560 __ dsll(T9, Rnext, Address::times_8);
4561 __ li(AT, (long)Interpreter::_wentry_point);
4562 __ dadd(AT, T9, AT);
4563 __ ld(T9, AT, 0);
4564 __ jr(T9);
4565 __ delayed()->nop();
4566 }
4568 //--------------------------------------------------------------------------------------------------// Multi arrays
4570 void TemplateTable::multianewarray() {
4571 transition(vtos, atos);
4572 // last dim is on top of stack; we want address of first one:
4573 // first_addr = last_addr + (ndims - 1) * wordSize
4574 __ lbu(A1, at_bcp(3)); // dimension
4575 __ daddi(A1, A1, -1);
4576 __ dsll(A1, A1, Address::times_8);
4577 __ dadd(A1, SP, A1); // now A1 pointer to the count array on the stack
4578 call_VM(FSR, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), A1);
4579 __ lbu(AT, at_bcp(3));
4580 __ dsll(AT, AT, Address::times_8);
4581 __ dadd(SP, SP, AT);
4582 __ sync();
4583 }
4585 #endif // !CC_INTERP