1 /*
2 * Copyright (c) 2000, 2011, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "c1/c1_Compilation.hpp"
27 #include "c1/c1_LIRAssembler.hpp"
28 #include "c1/c1_MacroAssembler.hpp"
29 #include "c1/c1_Runtime1.hpp"
30 #include "c1/c1_ValueStack.hpp"
31 #include "ci/ciArrayKlass.hpp"
32 #include "ci/ciInstance.hpp"
33 #include "gc_interface/collectedHeap.hpp"
34 #include "memory/barrierSet.hpp"
35 #include "memory/cardTableModRefBS.hpp"
36 #include "nativeInst_sparc.hpp"
37 #include "oops/objArrayKlass.hpp"
38 #include "runtime/sharedRuntime.hpp"
39
40 #define __ _masm->
41
42
43 //------------------------------------------------------------
44
45
46 bool LIR_Assembler::is_small_constant(LIR_Opr opr) {
47 if (opr->is_constant()) {
48 LIR_Const* constant = opr->as_constant_ptr();
49 switch (constant->type()) {
50 case T_INT: {
51 jint value = constant->as_jint();
52 return Assembler::is_simm13(value);
53 }
54
55 default:
56 return false;
57 }
58 }
59 return false;
60 }
61
62
63 bool LIR_Assembler::is_single_instruction(LIR_Op* op) {
64 switch (op->code()) {
65 case lir_null_check:
66 return true;
67
68
69 case lir_add:
70 case lir_ushr:
71 case lir_shr:
72 case lir_shl:
73 // integer shifts and adds are always one instruction
74 return op->result_opr()->is_single_cpu();
75
76
77 case lir_move: {
78 LIR_Op1* op1 = op->as_Op1();
79 LIR_Opr src = op1->in_opr();
80 LIR_Opr dst = op1->result_opr();
81
82 if (src == dst) {
83 NEEDS_CLEANUP;
84 // this works around a problem where moves with the same src and dst
85 // end up in the delay slot and then the assembler swallows the mov
86 // since it has no effect and then it complains because the delay slot
87 // is empty. returning false stops the optimizer from putting this in
88 // the delay slot
89 return false;
90 }
91
92 // don't put moves involving oops into the delay slot since the VerifyOops code
93 // will make it much larger than a single instruction.
94 if (VerifyOops) {
95 return false;
96 }
97
98 if (src->is_double_cpu() || dst->is_double_cpu() || op1->patch_code() != lir_patch_none ||
99 ((src->is_double_fpu() || dst->is_double_fpu()) && op1->move_kind() != lir_move_normal)) {
100 return false;
101 }
102
103 if (UseCompressedOops) {
104 if (dst->is_address() && !dst->is_stack() && (dst->type() == T_OBJECT || dst->type() == T_ARRAY)) return false;
105 if (src->is_address() && !src->is_stack() && (src->type() == T_OBJECT || src->type() == T_ARRAY)) return false;
106 }
107
108 if (dst->is_register()) {
109 if (src->is_address() && Assembler::is_simm13(src->as_address_ptr()->disp())) {
110 return !PatchALot;
111 } else if (src->is_single_stack()) {
112 return true;
113 }
114 }
115
116 if (src->is_register()) {
117 if (dst->is_address() && Assembler::is_simm13(dst->as_address_ptr()->disp())) {
118 return !PatchALot;
119 } else if (dst->is_single_stack()) {
120 return true;
121 }
122 }
123
124 if (dst->is_register() &&
125 ((src->is_register() && src->is_single_word() && src->is_same_type(dst)) ||
126 (src->is_constant() && LIR_Assembler::is_small_constant(op->as_Op1()->in_opr())))) {
127 return true;
128 }
129
130 return false;
131 }
132
133 default:
134 return false;
135 }
136 ShouldNotReachHere();
137 }
138
139
140 LIR_Opr LIR_Assembler::receiverOpr() {
141 return FrameMap::O0_oop_opr;
142 }
143
144
145 LIR_Opr LIR_Assembler::incomingReceiverOpr() {
146 return FrameMap::I0_oop_opr;
147 }
148
149
150 LIR_Opr LIR_Assembler::osrBufferPointer() {
151 return FrameMap::I0_opr;
152 }
153
154
155 int LIR_Assembler::initial_frame_size_in_bytes() {
156 return in_bytes(frame_map()->framesize_in_bytes());
157 }
158
159
160 // inline cache check: the inline cached class is in G5_inline_cache_reg(G5);
161 // we fetch the class of the receiver (O0) and compare it with the cached class.
162 // If they do not match we jump to slow case.
163 int LIR_Assembler::check_icache() {
164 int offset = __ offset();
165 __ inline_cache_check(O0, G5_inline_cache_reg);
166 return offset;
167 }
168
169
170 void LIR_Assembler::osr_entry() {
171 // On-stack-replacement entry sequence (interpreter frame layout described in interpreter_sparc.cpp):
172 //
173 // 1. Create a new compiled activation.
174 // 2. Initialize local variables in the compiled activation. The expression stack must be empty
175 // at the osr_bci; it is not initialized.
176 // 3. Jump to the continuation address in compiled code to resume execution.
177
178 // OSR entry point
179 offsets()->set_value(CodeOffsets::OSR_Entry, code_offset());
180 BlockBegin* osr_entry = compilation()->hir()->osr_entry();
181 ValueStack* entry_state = osr_entry->end()->state();
182 int number_of_locks = entry_state->locks_size();
183
184 // Create a frame for the compiled activation.
185 __ build_frame(initial_frame_size_in_bytes());
186
187 // OSR buffer is
188 //
189 // locals[nlocals-1..0]
190 // monitors[number_of_locks-1..0]
191 //
192 // locals is a direct copy of the interpreter frame so in the osr buffer
193 // so first slot in the local array is the last local from the interpreter
194 // and last slot is local[0] (receiver) from the interpreter
195 //
196 // Similarly with locks. The first lock slot in the osr buffer is the nth lock
197 // from the interpreter frame, the nth lock slot in the osr buffer is 0th lock
198 // in the interpreter frame (the method lock if a sync method)
199
200 // Initialize monitors in the compiled activation.
201 // I0: pointer to osr buffer
202 //
203 // All other registers are dead at this point and the locals will be
204 // copied into place by code emitted in the IR.
205
206 Register OSR_buf = osrBufferPointer()->as_register();
207 { assert(frame::interpreter_frame_monitor_size() == BasicObjectLock::size(), "adjust code below");
208 int monitor_offset = BytesPerWord * method()->max_locals() +
209 (2 * BytesPerWord) * (number_of_locks - 1);
210 // SharedRuntime::OSR_migration_begin() packs BasicObjectLocks in
211 // the OSR buffer using 2 word entries: first the lock and then
212 // the oop.
213 for (int i = 0; i < number_of_locks; i++) {
214 int slot_offset = monitor_offset - ((i * 2) * BytesPerWord);
215 #ifdef ASSERT
216 // verify the interpreter's monitor has a non-null object
217 {
218 Label L;
219 __ ld_ptr(OSR_buf, slot_offset + 1*BytesPerWord, O7);
220 __ cmp(G0, O7);
221 __ br(Assembler::notEqual, false, Assembler::pt, L);
222 __ delayed()->nop();
223 __ stop("locked object is NULL");
224 __ bind(L);
225 }
226 #endif // ASSERT
227 // Copy the lock field into the compiled activation.
228 __ ld_ptr(OSR_buf, slot_offset + 0, O7);
229 __ st_ptr(O7, frame_map()->address_for_monitor_lock(i));
230 __ ld_ptr(OSR_buf, slot_offset + 1*BytesPerWord, O7);
231 __ st_ptr(O7, frame_map()->address_for_monitor_object(i));
232 }
233 }
234 }
235
236
237 // Optimized Library calls
238 // This is the fast version of java.lang.String.compare; it has not
239 // OSR-entry and therefore, we generate a slow version for OSR's
240 void LIR_Assembler::emit_string_compare(LIR_Opr left, LIR_Opr right, LIR_Opr dst, CodeEmitInfo* info) {
241 Register str0 = left->as_register();
242 Register str1 = right->as_register();
243
244 Label Ldone;
245
246 Register result = dst->as_register();
247 {
248 // Get a pointer to the first character of string0 in tmp0 and get string0.count in str0
249 // Get a pointer to the first character of string1 in tmp1 and get string1.count in str1
250 // Also, get string0.count-string1.count in o7 and get the condition code set
251 // Note: some instructions have been hoisted for better instruction scheduling
252
253 Register tmp0 = L0;
254 Register tmp1 = L1;
255 Register tmp2 = L2;
256
257 int value_offset = java_lang_String:: value_offset_in_bytes(); // char array
258 int offset_offset = java_lang_String::offset_offset_in_bytes(); // first character position
259 int count_offset = java_lang_String:: count_offset_in_bytes();
260
261 __ load_heap_oop(str0, value_offset, tmp0);
262 __ ld(str0, offset_offset, tmp2);
263 __ add(tmp0, arrayOopDesc::base_offset_in_bytes(T_CHAR), tmp0);
264 __ ld(str0, count_offset, str0);
265 __ sll(tmp2, exact_log2(sizeof(jchar)), tmp2);
266
267 // str1 may be null
268 add_debug_info_for_null_check_here(info);
269
270 __ load_heap_oop(str1, value_offset, tmp1);
271 __ add(tmp0, tmp2, tmp0);
272
273 __ ld(str1, offset_offset, tmp2);
274 __ add(tmp1, arrayOopDesc::base_offset_in_bytes(T_CHAR), tmp1);
275 __ ld(str1, count_offset, str1);
276 __ sll(tmp2, exact_log2(sizeof(jchar)), tmp2);
277 __ subcc(str0, str1, O7);
278 __ add(tmp1, tmp2, tmp1);
279 }
280
281 {
282 // Compute the minimum of the string lengths, scale it and store it in limit
283 Register count0 = I0;
284 Register count1 = I1;
285 Register limit = L3;
286
287 Label Lskip;
288 __ sll(count0, exact_log2(sizeof(jchar)), limit); // string0 is shorter
289 __ br(Assembler::greater, true, Assembler::pt, Lskip);
290 __ delayed()->sll(count1, exact_log2(sizeof(jchar)), limit); // string1 is shorter
291 __ bind(Lskip);
292
293 // If either string is empty (or both of them) the result is the difference in lengths
294 __ cmp(limit, 0);
295 __ br(Assembler::equal, true, Assembler::pn, Ldone);
296 __ delayed()->mov(O7, result); // result is difference in lengths
297 }
298
299 {
300 // Neither string is empty
301 Label Lloop;
302
303 Register base0 = L0;
304 Register base1 = L1;
305 Register chr0 = I0;
306 Register chr1 = I1;
307 Register limit = L3;
308
309 // Shift base0 and base1 to the end of the arrays, negate limit
310 __ add(base0, limit, base0);
311 __ add(base1, limit, base1);
312 __ neg(limit); // limit = -min{string0.count, strin1.count}
313
314 __ lduh(base0, limit, chr0);
315 __ bind(Lloop);
316 __ lduh(base1, limit, chr1);
317 __ subcc(chr0, chr1, chr0);
318 __ br(Assembler::notZero, false, Assembler::pn, Ldone);
319 assert(chr0 == result, "result must be pre-placed");
320 __ delayed()->inccc(limit, sizeof(jchar));
321 __ br(Assembler::notZero, true, Assembler::pt, Lloop);
322 __ delayed()->lduh(base0, limit, chr0);
323 }
324
325 // If strings are equal up to min length, return the length difference.
326 __ mov(O7, result);
327
328 // Otherwise, return the difference between the first mismatched chars.
329 __ bind(Ldone);
330 }
331
332
333 // --------------------------------------------------------------------------------------------
334
335 void LIR_Assembler::monitorexit(LIR_Opr obj_opr, LIR_Opr lock_opr, Register hdr, int monitor_no) {
336 if (!GenerateSynchronizationCode) return;
337
338 Register obj_reg = obj_opr->as_register();
339 Register lock_reg = lock_opr->as_register();
340
341 Address mon_addr = frame_map()->address_for_monitor_lock(monitor_no);
342 Register reg = mon_addr.base();
343 int offset = mon_addr.disp();
344 // compute pointer to BasicLock
345 if (mon_addr.is_simm13()) {
346 __ add(reg, offset, lock_reg);
347 }
348 else {
349 __ set(offset, lock_reg);
350 __ add(reg, lock_reg, lock_reg);
351 }
352 // unlock object
353 MonitorAccessStub* slow_case = new MonitorExitStub(lock_opr, UseFastLocking, monitor_no);
354 // _slow_case_stubs->append(slow_case);
355 // temporary fix: must be created after exceptionhandler, therefore as call stub
356 _slow_case_stubs->append(slow_case);
357 if (UseFastLocking) {
358 // try inlined fast unlocking first, revert to slow locking if it fails
359 // note: lock_reg points to the displaced header since the displaced header offset is 0!
360 assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
361 __ unlock_object(hdr, obj_reg, lock_reg, *slow_case->entry());
362 } else {
363 // always do slow unlocking
364 // note: the slow unlocking code could be inlined here, however if we use
365 // slow unlocking, speed doesn't matter anyway and this solution is
366 // simpler and requires less duplicated code - additionally, the
367 // slow unlocking code is the same in either case which simplifies
368 // debugging
369 __ br(Assembler::always, false, Assembler::pt, *slow_case->entry());
370 __ delayed()->nop();
371 }
372 // done
373 __ bind(*slow_case->continuation());
374 }
375
376
377 int LIR_Assembler::emit_exception_handler() {
378 // if the last instruction is a call (typically to do a throw which
379 // is coming at the end after block reordering) the return address
380 // must still point into the code area in order to avoid assertion
381 // failures when searching for the corresponding bci => add a nop
382 // (was bug 5/14/1999 - gri)
383 __ nop();
384
385 // generate code for exception handler
386 ciMethod* method = compilation()->method();
387
388 address handler_base = __ start_a_stub(exception_handler_size);
389
390 if (handler_base == NULL) {
391 // not enough space left for the handler
392 bailout("exception handler overflow");
393 return -1;
394 }
395
396 int offset = code_offset();
397
398 __ call(Runtime1::entry_for(Runtime1::handle_exception_from_callee_id), relocInfo::runtime_call_type);
399 __ delayed()->nop();
400 __ should_not_reach_here();
401 assert(code_offset() - offset <= exception_handler_size, "overflow");
402 __ end_a_stub();
403
404 return offset;
405 }
406
407
408 // Emit the code to remove the frame from the stack in the exception
409 // unwind path.
410 int LIR_Assembler::emit_unwind_handler() {
411 #ifndef PRODUCT
412 if (CommentedAssembly) {
413 _masm->block_comment("Unwind handler");
414 }
415 #endif
416
417 int offset = code_offset();
418
419 // Fetch the exception from TLS and clear out exception related thread state
420 __ ld_ptr(G2_thread, in_bytes(JavaThread::exception_oop_offset()), O0);
421 __ st_ptr(G0, G2_thread, in_bytes(JavaThread::exception_oop_offset()));
422 __ st_ptr(G0, G2_thread, in_bytes(JavaThread::exception_pc_offset()));
423
424 __ bind(_unwind_handler_entry);
425 __ verify_not_null_oop(O0);
426 if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
427 __ mov(O0, I0); // Preserve the exception
428 }
429
430 // Preform needed unlocking
431 MonitorExitStub* stub = NULL;
432 if (method()->is_synchronized()) {
433 monitor_address(0, FrameMap::I1_opr);
434 stub = new MonitorExitStub(FrameMap::I1_opr, true, 0);
435 __ unlock_object(I3, I2, I1, *stub->entry());
436 __ bind(*stub->continuation());
437 }
438
439 if (compilation()->env()->dtrace_method_probes()) {
440 __ mov(G2_thread, O0);
441 jobject2reg(method()->constant_encoding(), O1);
442 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), relocInfo::runtime_call_type);
443 __ delayed()->nop();
444 }
445
446 if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
447 __ mov(I0, O0); // Restore the exception
448 }
449
450 // dispatch to the unwind logic
451 __ call(Runtime1::entry_for(Runtime1::unwind_exception_id), relocInfo::runtime_call_type);
452 __ delayed()->nop();
453
454 // Emit the slow path assembly
455 if (stub != NULL) {
456 stub->emit_code(this);
457 }
458
459 return offset;
460 }
461
462
463 int LIR_Assembler::emit_deopt_handler() {
464 // if the last instruction is a call (typically to do a throw which
465 // is coming at the end after block reordering) the return address
466 // must still point into the code area in order to avoid assertion
467 // failures when searching for the corresponding bci => add a nop
468 // (was bug 5/14/1999 - gri)
469 __ nop();
470
471 // generate code for deopt handler
472 ciMethod* method = compilation()->method();
473 address handler_base = __ start_a_stub(deopt_handler_size);
474 if (handler_base == NULL) {
475 // not enough space left for the handler
476 bailout("deopt handler overflow");
477 return -1;
478 }
479
480 int offset = code_offset();
481 AddressLiteral deopt_blob(SharedRuntime::deopt_blob()->unpack());
482 __ JUMP(deopt_blob, G3_scratch, 0); // sethi;jmp
483 __ delayed()->nop();
484 assert(code_offset() - offset <= deopt_handler_size, "overflow");
485 debug_only(__ stop("should have gone to the caller");)
486 __ end_a_stub();
487
488 return offset;
489 }
490
491
492 void LIR_Assembler::jobject2reg(jobject o, Register reg) {
493 if (o == NULL) {
494 __ set(NULL_WORD, reg);
495 } else {
496 int oop_index = __ oop_recorder()->find_index(o);
497 RelocationHolder rspec = oop_Relocation::spec(oop_index);
498 __ set(NULL_WORD, reg, rspec); // Will be set when the nmethod is created
499 }
500 }
501
502
503 void LIR_Assembler::jobject2reg_with_patching(Register reg, CodeEmitInfo *info) {
504 // Allocate a new index in oop table to hold the oop once it's been patched
505 int oop_index = __ oop_recorder()->allocate_index((jobject)NULL);
506 PatchingStub* patch = new PatchingStub(_masm, PatchingStub::load_klass_id, oop_index);
507
508 AddressLiteral addrlit(NULL, oop_Relocation::spec(oop_index));
509 assert(addrlit.rspec().type() == relocInfo::oop_type, "must be an oop reloc");
510 // It may not seem necessary to use a sethi/add pair to load a NULL into dest, but the
511 // NULL will be dynamically patched later and the patched value may be large. We must
512 // therefore generate the sethi/add as a placeholders
513 __ patchable_set(addrlit, reg);
514
515 patching_epilog(patch, lir_patch_normal, reg, info);
516 }
517
518
519 void LIR_Assembler::emit_op3(LIR_Op3* op) {
520 Register Rdividend = op->in_opr1()->as_register();
521 Register Rdivisor = noreg;
522 Register Rscratch = op->in_opr3()->as_register();
523 Register Rresult = op->result_opr()->as_register();
524 int divisor = -1;
525
526 if (op->in_opr2()->is_register()) {
527 Rdivisor = op->in_opr2()->as_register();
528 } else {
529 divisor = op->in_opr2()->as_constant_ptr()->as_jint();
530 assert(Assembler::is_simm13(divisor), "can only handle simm13");
531 }
532
533 assert(Rdividend != Rscratch, "");
534 assert(Rdivisor != Rscratch, "");
535 assert(op->code() == lir_idiv || op->code() == lir_irem, "Must be irem or idiv");
536
537 if (Rdivisor == noreg && is_power_of_2(divisor)) {
538 // convert division by a power of two into some shifts and logical operations
539 if (op->code() == lir_idiv) {
540 if (divisor == 2) {
541 __ srl(Rdividend, 31, Rscratch);
542 } else {
543 __ sra(Rdividend, 31, Rscratch);
544 __ and3(Rscratch, divisor - 1, Rscratch);
545 }
546 __ add(Rdividend, Rscratch, Rscratch);
547 __ sra(Rscratch, log2_intptr(divisor), Rresult);
548 return;
549 } else {
550 if (divisor == 2) {
551 __ srl(Rdividend, 31, Rscratch);
552 } else {
553 __ sra(Rdividend, 31, Rscratch);
554 __ and3(Rscratch, divisor - 1,Rscratch);
555 }
556 __ add(Rdividend, Rscratch, Rscratch);
557 __ andn(Rscratch, divisor - 1,Rscratch);
558 __ sub(Rdividend, Rscratch, Rresult);
559 return;
560 }
561 }
562
563 __ sra(Rdividend, 31, Rscratch);
564 __ wry(Rscratch);
565 if (!VM_Version::v9_instructions_work()) {
566 // v9 doesn't require these nops
567 __ nop();
568 __ nop();
569 __ nop();
570 __ nop();
571 }
572
573 add_debug_info_for_div0_here(op->info());
574
575 if (Rdivisor != noreg) {
576 __ sdivcc(Rdividend, Rdivisor, (op->code() == lir_idiv ? Rresult : Rscratch));
577 } else {
578 assert(Assembler::is_simm13(divisor), "can only handle simm13");
579 __ sdivcc(Rdividend, divisor, (op->code() == lir_idiv ? Rresult : Rscratch));
580 }
581
582 Label skip;
583 __ br(Assembler::overflowSet, true, Assembler::pn, skip);
584 __ delayed()->Assembler::sethi(0x80000000, (op->code() == lir_idiv ? Rresult : Rscratch));
585 __ bind(skip);
586
587 if (op->code() == lir_irem) {
588 if (Rdivisor != noreg) {
589 __ smul(Rscratch, Rdivisor, Rscratch);
590 } else {
591 __ smul(Rscratch, divisor, Rscratch);
592 }
593 __ sub(Rdividend, Rscratch, Rresult);
594 }
595 }
596
597
598 void LIR_Assembler::emit_opBranch(LIR_OpBranch* op) {
599 #ifdef ASSERT
600 assert(op->block() == NULL || op->block()->label() == op->label(), "wrong label");
601 if (op->block() != NULL) _branch_target_blocks.append(op->block());
602 if (op->ublock() != NULL) _branch_target_blocks.append(op->ublock());
603 #endif
604 assert(op->info() == NULL, "shouldn't have CodeEmitInfo");
605
606 if (op->cond() == lir_cond_always) {
607 __ br(Assembler::always, false, Assembler::pt, *(op->label()));
608 } else if (op->code() == lir_cond_float_branch) {
609 assert(op->ublock() != NULL, "must have unordered successor");
610 bool is_unordered = (op->ublock() == op->block());
611 Assembler::Condition acond;
612 switch (op->cond()) {
613 case lir_cond_equal: acond = Assembler::f_equal; break;
614 case lir_cond_notEqual: acond = Assembler::f_notEqual; break;
615 case lir_cond_less: acond = (is_unordered ? Assembler::f_unorderedOrLess : Assembler::f_less); break;
616 case lir_cond_greater: acond = (is_unordered ? Assembler::f_unorderedOrGreater : Assembler::f_greater); break;
617 case lir_cond_lessEqual: acond = (is_unordered ? Assembler::f_unorderedOrLessOrEqual : Assembler::f_lessOrEqual); break;
618 case lir_cond_greaterEqual: acond = (is_unordered ? Assembler::f_unorderedOrGreaterOrEqual: Assembler::f_greaterOrEqual); break;
619 default : ShouldNotReachHere();
620 };
621
622 if (!VM_Version::v9_instructions_work()) {
623 __ nop();
624 }
625 __ fb( acond, false, Assembler::pn, *(op->label()));
626 } else {
627 assert (op->code() == lir_branch, "just checking");
628
629 Assembler::Condition acond;
630 switch (op->cond()) {
631 case lir_cond_equal: acond = Assembler::equal; break;
632 case lir_cond_notEqual: acond = Assembler::notEqual; break;
633 case lir_cond_less: acond = Assembler::less; break;
634 case lir_cond_lessEqual: acond = Assembler::lessEqual; break;
635 case lir_cond_greaterEqual: acond = Assembler::greaterEqual; break;
636 case lir_cond_greater: acond = Assembler::greater; break;
637 case lir_cond_aboveEqual: acond = Assembler::greaterEqualUnsigned; break;
638 case lir_cond_belowEqual: acond = Assembler::lessEqualUnsigned; break;
639 default: ShouldNotReachHere();
640 };
641
642 // sparc has different condition codes for testing 32-bit
643 // vs. 64-bit values. We could always test xcc is we could
644 // guarantee that 32-bit loads always sign extended but that isn't
645 // true and since sign extension isn't free, it would impose a
646 // slight cost.
647 #ifdef _LP64
648 if (op->type() == T_INT) {
649 __ br(acond, false, Assembler::pn, *(op->label()));
650 } else
651 #endif
652 __ brx(acond, false, Assembler::pn, *(op->label()));
653 }
654 // The peephole pass fills the delay slot
655 }
656
657
658 void LIR_Assembler::emit_opConvert(LIR_OpConvert* op) {
659 Bytecodes::Code code = op->bytecode();
660 LIR_Opr dst = op->result_opr();
661
662 switch(code) {
663 case Bytecodes::_i2l: {
664 Register rlo = dst->as_register_lo();
665 Register rhi = dst->as_register_hi();
666 Register rval = op->in_opr()->as_register();
667 #ifdef _LP64
668 __ sra(rval, 0, rlo);
669 #else
670 __ mov(rval, rlo);
671 __ sra(rval, BitsPerInt-1, rhi);
672 #endif
673 break;
674 }
675 case Bytecodes::_i2d:
676 case Bytecodes::_i2f: {
677 bool is_double = (code == Bytecodes::_i2d);
678 FloatRegister rdst = is_double ? dst->as_double_reg() : dst->as_float_reg();
679 FloatRegisterImpl::Width w = is_double ? FloatRegisterImpl::D : FloatRegisterImpl::S;
680 FloatRegister rsrc = op->in_opr()->as_float_reg();
681 if (rsrc != rdst) {
682 __ fmov(FloatRegisterImpl::S, rsrc, rdst);
683 }
684 __ fitof(w, rdst, rdst);
685 break;
686 }
687 case Bytecodes::_f2i:{
688 FloatRegister rsrc = op->in_opr()->as_float_reg();
689 Address addr = frame_map()->address_for_slot(dst->single_stack_ix());
690 Label L;
691 // result must be 0 if value is NaN; test by comparing value to itself
692 __ fcmp(FloatRegisterImpl::S, Assembler::fcc0, rsrc, rsrc);
693 if (!VM_Version::v9_instructions_work()) {
694 __ nop();
695 }
696 __ fb(Assembler::f_unordered, true, Assembler::pn, L);
697 __ delayed()->st(G0, addr); // annuled if contents of rsrc is not NaN
698 __ ftoi(FloatRegisterImpl::S, rsrc, rsrc);
699 // move integer result from float register to int register
700 __ stf(FloatRegisterImpl::S, rsrc, addr.base(), addr.disp());
701 __ bind (L);
702 break;
703 }
704 case Bytecodes::_l2i: {
705 Register rlo = op->in_opr()->as_register_lo();
706 Register rhi = op->in_opr()->as_register_hi();
707 Register rdst = dst->as_register();
708 #ifdef _LP64
709 __ sra(rlo, 0, rdst);
710 #else
711 __ mov(rlo, rdst);
712 #endif
713 break;
714 }
715 case Bytecodes::_d2f:
716 case Bytecodes::_f2d: {
717 bool is_double = (code == Bytecodes::_f2d);
718 assert((!is_double && dst->is_single_fpu()) || (is_double && dst->is_double_fpu()), "check");
719 LIR_Opr val = op->in_opr();
720 FloatRegister rval = (code == Bytecodes::_d2f) ? val->as_double_reg() : val->as_float_reg();
721 FloatRegister rdst = is_double ? dst->as_double_reg() : dst->as_float_reg();
722 FloatRegisterImpl::Width vw = is_double ? FloatRegisterImpl::S : FloatRegisterImpl::D;
723 FloatRegisterImpl::Width dw = is_double ? FloatRegisterImpl::D : FloatRegisterImpl::S;
724 __ ftof(vw, dw, rval, rdst);
725 break;
726 }
727 case Bytecodes::_i2s:
728 case Bytecodes::_i2b: {
729 Register rval = op->in_opr()->as_register();
730 Register rdst = dst->as_register();
731 int shift = (code == Bytecodes::_i2b) ? (BitsPerInt - T_BYTE_aelem_bytes * BitsPerByte) : (BitsPerInt - BitsPerShort);
732 __ sll (rval, shift, rdst);
733 __ sra (rdst, shift, rdst);
734 break;
735 }
736 case Bytecodes::_i2c: {
737 Register rval = op->in_opr()->as_register();
738 Register rdst = dst->as_register();
739 int shift = BitsPerInt - T_CHAR_aelem_bytes * BitsPerByte;
740 __ sll (rval, shift, rdst);
741 __ srl (rdst, shift, rdst);
742 break;
743 }
744
745 default: ShouldNotReachHere();
746 }
747 }
748
749
750 void LIR_Assembler::align_call(LIR_Code) {
751 // do nothing since all instructions are word aligned on sparc
752 }
753
754
755 void LIR_Assembler::call(LIR_OpJavaCall* op, relocInfo::relocType rtype) {
756 __ call(op->addr(), rtype);
757 // The peephole pass fills the delay slot, add_call_info is done in
758 // LIR_Assembler::emit_delay.
759 }
760
761
762 void LIR_Assembler::ic_call(LIR_OpJavaCall* op) {
763 RelocationHolder rspec = virtual_call_Relocation::spec(pc());
764 __ set_oop((jobject)Universe::non_oop_word(), G5_inline_cache_reg);
765 __ relocate(rspec);
766 __ call(op->addr(), relocInfo::none);
767 // The peephole pass fills the delay slot, add_call_info is done in
768 // LIR_Assembler::emit_delay.
769 }
770
771
772 void LIR_Assembler::vtable_call(LIR_OpJavaCall* op) {
773 add_debug_info_for_null_check_here(op->info());
774 __ load_klass(O0, G3_scratch);
775 if (__ is_simm13(op->vtable_offset())) {
776 __ ld_ptr(G3_scratch, op->vtable_offset(), G5_method);
777 } else {
778 // This will generate 2 instructions
779 __ set(op->vtable_offset(), G5_method);
780 // ld_ptr, set_hi, set
781 __ ld_ptr(G3_scratch, G5_method, G5_method);
782 }
783 __ ld_ptr(G5_method, methodOopDesc::from_compiled_offset(), G3_scratch);
784 __ callr(G3_scratch, G0);
785 // the peephole pass fills the delay slot
786 }
787
788 int LIR_Assembler::store(LIR_Opr from_reg, Register base, int offset, BasicType type, bool wide, bool unaligned) {
789 int store_offset;
790 if (!Assembler::is_simm13(offset + (type == T_LONG) ? wordSize : 0)) {
791 assert(!unaligned, "can't handle this");
792 // for offsets larger than a simm13 we setup the offset in O7
793 __ set(offset, O7);
794 store_offset = store(from_reg, base, O7, type, wide);
795 } else {
796 if (type == T_ARRAY || type == T_OBJECT) {
797 __ verify_oop(from_reg->as_register());
798 }
799 store_offset = code_offset();
800 switch (type) {
801 case T_BOOLEAN: // fall through
802 case T_BYTE : __ stb(from_reg->as_register(), base, offset); break;
803 case T_CHAR : __ sth(from_reg->as_register(), base, offset); break;
804 case T_SHORT : __ sth(from_reg->as_register(), base, offset); break;
805 case T_INT : __ stw(from_reg->as_register(), base, offset); break;
806 case T_LONG :
807 #ifdef _LP64
808 if (unaligned || PatchALot) {
809 __ srax(from_reg->as_register_lo(), 32, O7);
810 __ stw(from_reg->as_register_lo(), base, offset + lo_word_offset_in_bytes);
811 __ stw(O7, base, offset + hi_word_offset_in_bytes);
812 } else {
813 __ stx(from_reg->as_register_lo(), base, offset);
814 }
815 #else
816 assert(Assembler::is_simm13(offset + 4), "must be");
817 __ stw(from_reg->as_register_lo(), base, offset + lo_word_offset_in_bytes);
818 __ stw(from_reg->as_register_hi(), base, offset + hi_word_offset_in_bytes);
819 #endif
820 break;
821 case T_ADDRESS:
822 __ st_ptr(from_reg->as_register(), base, offset);
823 break;
824 case T_ARRAY : // fall through
825 case T_OBJECT:
826 {
827 if (UseCompressedOops && !wide) {
828 __ encode_heap_oop(from_reg->as_register(), G3_scratch);
829 store_offset = code_offset();
830 __ stw(G3_scratch, base, offset);
831 } else {
832 __ st_ptr(from_reg->as_register(), base, offset);
833 }
834 break;
835 }
836
837 case T_FLOAT : __ stf(FloatRegisterImpl::S, from_reg->as_float_reg(), base, offset); break;
838 case T_DOUBLE:
839 {
840 FloatRegister reg = from_reg->as_double_reg();
841 // split unaligned stores
842 if (unaligned || PatchALot) {
843 assert(Assembler::is_simm13(offset + 4), "must be");
844 __ stf(FloatRegisterImpl::S, reg->successor(), base, offset + 4);
845 __ stf(FloatRegisterImpl::S, reg, base, offset);
846 } else {
847 __ stf(FloatRegisterImpl::D, reg, base, offset);
848 }
849 break;
850 }
851 default : ShouldNotReachHere();
852 }
853 }
854 return store_offset;
855 }
856
857
858 int LIR_Assembler::store(LIR_Opr from_reg, Register base, Register disp, BasicType type, bool wide) {
859 if (type == T_ARRAY || type == T_OBJECT) {
860 __ verify_oop(from_reg->as_register());
861 }
862 int store_offset = code_offset();
863 switch (type) {
864 case T_BOOLEAN: // fall through
865 case T_BYTE : __ stb(from_reg->as_register(), base, disp); break;
866 case T_CHAR : __ sth(from_reg->as_register(), base, disp); break;
867 case T_SHORT : __ sth(from_reg->as_register(), base, disp); break;
868 case T_INT : __ stw(from_reg->as_register(), base, disp); break;
869 case T_LONG :
870 #ifdef _LP64
871 __ stx(from_reg->as_register_lo(), base, disp);
872 #else
873 assert(from_reg->as_register_hi()->successor() == from_reg->as_register_lo(), "must match");
874 __ std(from_reg->as_register_hi(), base, disp);
875 #endif
876 break;
877 case T_ADDRESS:
878 __ st_ptr(from_reg->as_register(), base, disp);
879 break;
880 case T_ARRAY : // fall through
881 case T_OBJECT:
882 {
883 if (UseCompressedOops && !wide) {
884 __ encode_heap_oop(from_reg->as_register(), G3_scratch);
885 store_offset = code_offset();
886 __ stw(G3_scratch, base, disp);
887 } else {
888 __ st_ptr(from_reg->as_register(), base, disp);
889 }
890 break;
891 }
892 case T_FLOAT : __ stf(FloatRegisterImpl::S, from_reg->as_float_reg(), base, disp); break;
893 case T_DOUBLE: __ stf(FloatRegisterImpl::D, from_reg->as_double_reg(), base, disp); break;
894 default : ShouldNotReachHere();
895 }
896 return store_offset;
897 }
898
899
900 int LIR_Assembler::load(Register base, int offset, LIR_Opr to_reg, BasicType type, bool wide, bool unaligned) {
901 int load_offset;
902 if (!Assembler::is_simm13(offset + (type == T_LONG) ? wordSize : 0)) {
903 assert(base != O7, "destroying register");
904 assert(!unaligned, "can't handle this");
905 // for offsets larger than a simm13 we setup the offset in O7
906 __ set(offset, O7);
907 load_offset = load(base, O7, to_reg, type, wide);
908 } else {
909 load_offset = code_offset();
910 switch(type) {
911 case T_BOOLEAN: // fall through
912 case T_BYTE : __ ldsb(base, offset, to_reg->as_register()); break;
913 case T_CHAR : __ lduh(base, offset, to_reg->as_register()); break;
914 case T_SHORT : __ ldsh(base, offset, to_reg->as_register()); break;
915 case T_INT : __ ld(base, offset, to_reg->as_register()); break;
916 case T_LONG :
917 if (!unaligned) {
918 #ifdef _LP64
919 __ ldx(base, offset, to_reg->as_register_lo());
920 #else
921 assert(to_reg->as_register_hi()->successor() == to_reg->as_register_lo(),
922 "must be sequential");
923 __ ldd(base, offset, to_reg->as_register_hi());
924 #endif
925 } else {
926 #ifdef _LP64
927 assert(base != to_reg->as_register_lo(), "can't handle this");
928 assert(O7 != to_reg->as_register_lo(), "can't handle this");
929 __ ld(base, offset + hi_word_offset_in_bytes, to_reg->as_register_lo());
930 __ lduw(base, offset + lo_word_offset_in_bytes, O7); // in case O7 is base or offset, use it last
931 __ sllx(to_reg->as_register_lo(), 32, to_reg->as_register_lo());
932 __ or3(to_reg->as_register_lo(), O7, to_reg->as_register_lo());
933 #else
934 if (base == to_reg->as_register_lo()) {
935 __ ld(base, offset + hi_word_offset_in_bytes, to_reg->as_register_hi());
936 __ ld(base, offset + lo_word_offset_in_bytes, to_reg->as_register_lo());
937 } else {
938 __ ld(base, offset + lo_word_offset_in_bytes, to_reg->as_register_lo());
939 __ ld(base, offset + hi_word_offset_in_bytes, to_reg->as_register_hi());
940 }
941 #endif
942 }
943 break;
944 case T_ADDRESS: __ ld_ptr(base, offset, to_reg->as_register()); break;
945 case T_ARRAY : // fall through
946 case T_OBJECT:
947 {
948 if (UseCompressedOops && !wide) {
949 __ lduw(base, offset, to_reg->as_register());
950 __ decode_heap_oop(to_reg->as_register());
951 } else {
952 __ ld_ptr(base, offset, to_reg->as_register());
953 }
954 break;
955 }
956 case T_FLOAT: __ ldf(FloatRegisterImpl::S, base, offset, to_reg->as_float_reg()); break;
957 case T_DOUBLE:
958 {
959 FloatRegister reg = to_reg->as_double_reg();
960 // split unaligned loads
961 if (unaligned || PatchALot) {
962 __ ldf(FloatRegisterImpl::S, base, offset + 4, reg->successor());
963 __ ldf(FloatRegisterImpl::S, base, offset, reg);
964 } else {
965 __ ldf(FloatRegisterImpl::D, base, offset, to_reg->as_double_reg());
966 }
967 break;
968 }
969 default : ShouldNotReachHere();
970 }
971 if (type == T_ARRAY || type == T_OBJECT) {
972 __ verify_oop(to_reg->as_register());
973 }
974 }
975 return load_offset;
976 }
977
978
979 int LIR_Assembler::load(Register base, Register disp, LIR_Opr to_reg, BasicType type, bool wide) {
980 int load_offset = code_offset();
981 switch(type) {
982 case T_BOOLEAN: // fall through
983 case T_BYTE : __ ldsb(base, disp, to_reg->as_register()); break;
984 case T_CHAR : __ lduh(base, disp, to_reg->as_register()); break;
985 case T_SHORT : __ ldsh(base, disp, to_reg->as_register()); break;
986 case T_INT : __ ld(base, disp, to_reg->as_register()); break;
987 case T_ADDRESS: __ ld_ptr(base, disp, to_reg->as_register()); break;
988 case T_ARRAY : // fall through
989 case T_OBJECT:
990 {
991 if (UseCompressedOops && !wide) {
992 __ lduw(base, disp, to_reg->as_register());
993 __ decode_heap_oop(to_reg->as_register());
994 } else {
995 __ ld_ptr(base, disp, to_reg->as_register());
996 }
997 break;
998 }
999 case T_FLOAT: __ ldf(FloatRegisterImpl::S, base, disp, to_reg->as_float_reg()); break;
1000 case T_DOUBLE: __ ldf(FloatRegisterImpl::D, base, disp, to_reg->as_double_reg()); break;
1001 case T_LONG :
1002 #ifdef _LP64
1003 __ ldx(base, disp, to_reg->as_register_lo());
1004 #else
1005 assert(to_reg->as_register_hi()->successor() == to_reg->as_register_lo(),
1006 "must be sequential");
1007 __ ldd(base, disp, to_reg->as_register_hi());
1008 #endif
1009 break;
1010 default : ShouldNotReachHere();
1011 }
1012 if (type == T_ARRAY || type == T_OBJECT) {
1013 __ verify_oop(to_reg->as_register());
1014 }
1015 return load_offset;
1016 }
1017
1018 void LIR_Assembler::const2stack(LIR_Opr src, LIR_Opr dest) {
1019 LIR_Const* c = src->as_constant_ptr();
1020 switch (c->type()) {
1021 case T_INT:
1022 case T_FLOAT: {
1023 Register src_reg = O7;
1024 int value = c->as_jint_bits();
1025 if (value == 0) {
1026 src_reg = G0;
1027 } else {
1028 __ set(value, O7);
1029 }
1030 Address addr = frame_map()->address_for_slot(dest->single_stack_ix());
1031 __ stw(src_reg, addr.base(), addr.disp());
1032 break;
1033 }
1034 case T_ADDRESS: {
1035 Register src_reg = O7;
1036 int value = c->as_jint_bits();
1037 if (value == 0) {
1038 src_reg = G0;
1039 } else {
1040 __ set(value, O7);
1041 }
1042 Address addr = frame_map()->address_for_slot(dest->single_stack_ix());
1043 __ st_ptr(src_reg, addr.base(), addr.disp());
1044 break;
1045 }
1046 case T_OBJECT: {
1047 Register src_reg = O7;
1048 jobject2reg(c->as_jobject(), src_reg);
1049 Address addr = frame_map()->address_for_slot(dest->single_stack_ix());
1050 __ st_ptr(src_reg, addr.base(), addr.disp());
1051 break;
1052 }
1053 case T_LONG:
1054 case T_DOUBLE: {
1055 Address addr = frame_map()->address_for_double_slot(dest->double_stack_ix());
1056
1057 Register tmp = O7;
1058 int value_lo = c->as_jint_lo_bits();
1059 if (value_lo == 0) {
1060 tmp = G0;
1061 } else {
1062 __ set(value_lo, O7);
1063 }
1064 __ stw(tmp, addr.base(), addr.disp() + lo_word_offset_in_bytes);
1065 int value_hi = c->as_jint_hi_bits();
1066 if (value_hi == 0) {
1067 tmp = G0;
1068 } else {
1069 __ set(value_hi, O7);
1070 }
1071 __ stw(tmp, addr.base(), addr.disp() + hi_word_offset_in_bytes);
1072 break;
1073 }
1074 default:
1075 Unimplemented();
1076 }
1077 }
1078
1079
1080 void LIR_Assembler::const2mem(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info, bool wide) {
1081 LIR_Const* c = src->as_constant_ptr();
1082 LIR_Address* addr = dest->as_address_ptr();
1083 Register base = addr->base()->as_pointer_register();
1084 int offset = -1;
1085
1086 switch (c->type()) {
1087 case T_INT:
1088 case T_FLOAT:
1089 case T_ADDRESS: {
1090 LIR_Opr tmp = FrameMap::O7_opr;
1091 int value = c->as_jint_bits();
1092 if (value == 0) {
1093 tmp = FrameMap::G0_opr;
1094 } else if (Assembler::is_simm13(value)) {
1095 __ set(value, O7);
1096 }
1097 if (addr->index()->is_valid()) {
1098 assert(addr->disp() == 0, "must be zero");
1099 offset = store(tmp, base, addr->index()->as_pointer_register(), type, wide);
1100 } else {
1101 assert(Assembler::is_simm13(addr->disp()), "can't handle larger addresses");
1102 offset = store(tmp, base, addr->disp(), type, wide, false);
1103 }
1104 break;
1105 }
1106 case T_LONG:
1107 case T_DOUBLE: {
1108 assert(!addr->index()->is_valid(), "can't handle reg reg address here");
1109 assert(Assembler::is_simm13(addr->disp()) &&
1110 Assembler::is_simm13(addr->disp() + 4), "can't handle larger addresses");
1111
1112 LIR_Opr tmp = FrameMap::O7_opr;
1113 int value_lo = c->as_jint_lo_bits();
1114 if (value_lo == 0) {
1115 tmp = FrameMap::G0_opr;
1116 } else {
1117 __ set(value_lo, O7);
1118 }
1119 offset = store(tmp, base, addr->disp() + lo_word_offset_in_bytes, T_INT, wide, false);
1120 int value_hi = c->as_jint_hi_bits();
1121 if (value_hi == 0) {
1122 tmp = FrameMap::G0_opr;
1123 } else {
1124 __ set(value_hi, O7);
1125 }
1126 offset = store(tmp, base, addr->disp() + hi_word_offset_in_bytes, T_INT, wide, false);
1127 break;
1128 }
1129 case T_OBJECT: {
1130 jobject obj = c->as_jobject();
1131 LIR_Opr tmp;
1132 if (obj == NULL) {
1133 tmp = FrameMap::G0_opr;
1134 } else {
1135 tmp = FrameMap::O7_opr;
1136 jobject2reg(c->as_jobject(), O7);
1137 }
1138 // handle either reg+reg or reg+disp address
1139 if (addr->index()->is_valid()) {
1140 assert(addr->disp() == 0, "must be zero");
1141 offset = store(tmp, base, addr->index()->as_pointer_register(), type, wide);
1142 } else {
1143 assert(Assembler::is_simm13(addr->disp()), "can't handle larger addresses");
1144 offset = store(tmp, base, addr->disp(), type, wide, false);
1145 }
1146
1147 break;
1148 }
1149 default:
1150 Unimplemented();
1151 }
1152 if (info != NULL) {
1153 assert(offset != -1, "offset should've been set");
1154 add_debug_info_for_null_check(offset, info);
1155 }
1156 }
1157
1158
1159 void LIR_Assembler::const2reg(LIR_Opr src, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
1160 LIR_Const* c = src->as_constant_ptr();
1161 LIR_Opr to_reg = dest;
1162
1163 switch (c->type()) {
1164 case T_INT:
1165 case T_ADDRESS:
1166 {
1167 jint con = c->as_jint();
1168 if (to_reg->is_single_cpu()) {
1169 assert(patch_code == lir_patch_none, "no patching handled here");
1170 __ set(con, to_reg->as_register());
1171 } else {
1172 ShouldNotReachHere();
1173 assert(to_reg->is_single_fpu(), "wrong register kind");
1174
1175 __ set(con, O7);
1176 Address temp_slot(SP, (frame::register_save_words * wordSize) + STACK_BIAS);
1177 __ st(O7, temp_slot);
1178 __ ldf(FloatRegisterImpl::S, temp_slot, to_reg->as_float_reg());
1179 }
1180 }
1181 break;
1182
1183 case T_LONG:
1184 {
1185 jlong con = c->as_jlong();
1186
1187 if (to_reg->is_double_cpu()) {
1188 #ifdef _LP64
1189 __ set(con, to_reg->as_register_lo());
1190 #else
1191 __ set(low(con), to_reg->as_register_lo());
1192 __ set(high(con), to_reg->as_register_hi());
1193 #endif
1194 #ifdef _LP64
1195 } else if (to_reg->is_single_cpu()) {
1196 __ set(con, to_reg->as_register());
1197 #endif
1198 } else {
1199 ShouldNotReachHere();
1200 assert(to_reg->is_double_fpu(), "wrong register kind");
1201 Address temp_slot_lo(SP, ((frame::register_save_words ) * wordSize) + STACK_BIAS);
1202 Address temp_slot_hi(SP, ((frame::register_save_words) * wordSize) + (longSize/2) + STACK_BIAS);
1203 __ set(low(con), O7);
1204 __ st(O7, temp_slot_lo);
1205 __ set(high(con), O7);
1206 __ st(O7, temp_slot_hi);
1207 __ ldf(FloatRegisterImpl::D, temp_slot_lo, to_reg->as_double_reg());
1208 }
1209 }
1210 break;
1211
1212 case T_OBJECT:
1213 {
1214 if (patch_code == lir_patch_none) {
1215 jobject2reg(c->as_jobject(), to_reg->as_register());
1216 } else {
1217 jobject2reg_with_patching(to_reg->as_register(), info);
1218 }
1219 }
1220 break;
1221
1222 case T_FLOAT:
1223 {
1224 address const_addr = __ float_constant(c->as_jfloat());
1225 if (const_addr == NULL) {
1226 bailout("const section overflow");
1227 break;
1228 }
1229 RelocationHolder rspec = internal_word_Relocation::spec(const_addr);
1230 AddressLiteral const_addrlit(const_addr, rspec);
1231 if (to_reg->is_single_fpu()) {
1232 __ patchable_sethi(const_addrlit, O7);
1233 __ relocate(rspec);
1234 __ ldf(FloatRegisterImpl::S, O7, const_addrlit.low10(), to_reg->as_float_reg());
1235
1236 } else {
1237 assert(to_reg->is_single_cpu(), "Must be a cpu register.");
1238
1239 __ set(const_addrlit, O7);
1240 __ ld(O7, 0, to_reg->as_register());
1241 }
1242 }
1243 break;
1244
1245 case T_DOUBLE:
1246 {
1247 address const_addr = __ double_constant(c->as_jdouble());
1248 if (const_addr == NULL) {
1249 bailout("const section overflow");
1250 break;
1251 }
1252 RelocationHolder rspec = internal_word_Relocation::spec(const_addr);
1253
1254 if (to_reg->is_double_fpu()) {
1255 AddressLiteral const_addrlit(const_addr, rspec);
1256 __ patchable_sethi(const_addrlit, O7);
1257 __ relocate(rspec);
1258 __ ldf (FloatRegisterImpl::D, O7, const_addrlit.low10(), to_reg->as_double_reg());
1259 } else {
1260 assert(to_reg->is_double_cpu(), "Must be a long register.");
1261 #ifdef _LP64
1262 __ set(jlong_cast(c->as_jdouble()), to_reg->as_register_lo());
1263 #else
1264 __ set(low(jlong_cast(c->as_jdouble())), to_reg->as_register_lo());
1265 __ set(high(jlong_cast(c->as_jdouble())), to_reg->as_register_hi());
1266 #endif
1267 }
1268
1269 }
1270 break;
1271
1272 default:
1273 ShouldNotReachHere();
1274 }
1275 }
1276
1277 Address LIR_Assembler::as_Address(LIR_Address* addr) {
1278 Register reg = addr->base()->as_register();
1279 return Address(reg, addr->disp());
1280 }
1281
1282
1283 void LIR_Assembler::stack2stack(LIR_Opr src, LIR_Opr dest, BasicType type) {
1284 switch (type) {
1285 case T_INT:
1286 case T_FLOAT: {
1287 Register tmp = O7;
1288 Address from = frame_map()->address_for_slot(src->single_stack_ix());
1289 Address to = frame_map()->address_for_slot(dest->single_stack_ix());
1290 __ lduw(from.base(), from.disp(), tmp);
1291 __ stw(tmp, to.base(), to.disp());
1292 break;
1293 }
1294 case T_OBJECT: {
1295 Register tmp = O7;
1296 Address from = frame_map()->address_for_slot(src->single_stack_ix());
1297 Address to = frame_map()->address_for_slot(dest->single_stack_ix());
1298 __ ld_ptr(from.base(), from.disp(), tmp);
1299 __ st_ptr(tmp, to.base(), to.disp());
1300 break;
1301 }
1302 case T_LONG:
1303 case T_DOUBLE: {
1304 Register tmp = O7;
1305 Address from = frame_map()->address_for_double_slot(src->double_stack_ix());
1306 Address to = frame_map()->address_for_double_slot(dest->double_stack_ix());
1307 __ lduw(from.base(), from.disp(), tmp);
1308 __ stw(tmp, to.base(), to.disp());
1309 __ lduw(from.base(), from.disp() + 4, tmp);
1310 __ stw(tmp, to.base(), to.disp() + 4);
1311 break;
1312 }
1313
1314 default:
1315 ShouldNotReachHere();
1316 }
1317 }
1318
1319
1320 Address LIR_Assembler::as_Address_hi(LIR_Address* addr) {
1321 Address base = as_Address(addr);
1322 return Address(base.base(), base.disp() + hi_word_offset_in_bytes);
1323 }
1324
1325
1326 Address LIR_Assembler::as_Address_lo(LIR_Address* addr) {
1327 Address base = as_Address(addr);
1328 return Address(base.base(), base.disp() + lo_word_offset_in_bytes);
1329 }
1330
1331
1332 void LIR_Assembler::mem2reg(LIR_Opr src_opr, LIR_Opr dest, BasicType type,
1333 LIR_PatchCode patch_code, CodeEmitInfo* info, bool wide, bool unaligned) {
1334
1335 LIR_Address* addr = src_opr->as_address_ptr();
1336 LIR_Opr to_reg = dest;
1337
1338 Register src = addr->base()->as_pointer_register();
1339 Register disp_reg = noreg;
1340 int disp_value = addr->disp();
1341 bool needs_patching = (patch_code != lir_patch_none);
1342
1343 if (addr->base()->type() == T_OBJECT) {
1344 __ verify_oop(src);
1345 }
1346
1347 PatchingStub* patch = NULL;
1348 if (needs_patching) {
1349 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1350 assert(!to_reg->is_double_cpu() ||
1351 patch_code == lir_patch_none ||
1352 patch_code == lir_patch_normal, "patching doesn't match register");
1353 }
1354
1355 if (addr->index()->is_illegal()) {
1356 if (!Assembler::is_simm13(disp_value) && (!unaligned || Assembler::is_simm13(disp_value + 4))) {
1357 if (needs_patching) {
1358 __ patchable_set(0, O7);
1359 } else {
1360 __ set(disp_value, O7);
1361 }
1362 disp_reg = O7;
1363 }
1364 } else if (unaligned || PatchALot) {
1365 __ add(src, addr->index()->as_register(), O7);
1366 src = O7;
1367 } else {
1368 disp_reg = addr->index()->as_pointer_register();
1369 assert(disp_value == 0, "can't handle 3 operand addresses");
1370 }
1371
1372 // remember the offset of the load. The patching_epilog must be done
1373 // before the call to add_debug_info, otherwise the PcDescs don't get
1374 // entered in increasing order.
1375 int offset = code_offset();
1376
1377 assert(disp_reg != noreg || Assembler::is_simm13(disp_value), "should have set this up");
1378 if (disp_reg == noreg) {
1379 offset = load(src, disp_value, to_reg, type, wide, unaligned);
1380 } else {
1381 assert(!unaligned, "can't handle this");
1382 offset = load(src, disp_reg, to_reg, type, wide);
1383 }
1384
1385 if (patch != NULL) {
1386 patching_epilog(patch, patch_code, src, info);
1387 }
1388 if (info != NULL) add_debug_info_for_null_check(offset, info);
1389 }
1390
1391
1392 void LIR_Assembler::prefetchr(LIR_Opr src) {
1393 LIR_Address* addr = src->as_address_ptr();
1394 Address from_addr = as_Address(addr);
1395
1396 if (VM_Version::has_v9()) {
1397 __ prefetch(from_addr, Assembler::severalReads);
1398 }
1399 }
1400
1401
1402 void LIR_Assembler::prefetchw(LIR_Opr src) {
1403 LIR_Address* addr = src->as_address_ptr();
1404 Address from_addr = as_Address(addr);
1405
1406 if (VM_Version::has_v9()) {
1407 __ prefetch(from_addr, Assembler::severalWritesAndPossiblyReads);
1408 }
1409 }
1410
1411
1412 void LIR_Assembler::stack2reg(LIR_Opr src, LIR_Opr dest, BasicType type) {
1413 Address addr;
1414 if (src->is_single_word()) {
1415 addr = frame_map()->address_for_slot(src->single_stack_ix());
1416 } else if (src->is_double_word()) {
1417 addr = frame_map()->address_for_double_slot(src->double_stack_ix());
1418 }
1419
1420 bool unaligned = (addr.disp() - STACK_BIAS) % 8 != 0;
1421 load(addr.base(), addr.disp(), dest, dest->type(), true /*wide*/, unaligned);
1422 }
1423
1424
1425 void LIR_Assembler::reg2stack(LIR_Opr from_reg, LIR_Opr dest, BasicType type, bool pop_fpu_stack) {
1426 Address addr;
1427 if (dest->is_single_word()) {
1428 addr = frame_map()->address_for_slot(dest->single_stack_ix());
1429 } else if (dest->is_double_word()) {
1430 addr = frame_map()->address_for_slot(dest->double_stack_ix());
1431 }
1432 bool unaligned = (addr.disp() - STACK_BIAS) % 8 != 0;
1433 store(from_reg, addr.base(), addr.disp(), from_reg->type(), true /*wide*/, unaligned);
1434 }
1435
1436
1437 void LIR_Assembler::reg2reg(LIR_Opr from_reg, LIR_Opr to_reg) {
1438 if (from_reg->is_float_kind() && to_reg->is_float_kind()) {
1439 if (from_reg->is_double_fpu()) {
1440 // double to double moves
1441 assert(to_reg->is_double_fpu(), "should match");
1442 __ fmov(FloatRegisterImpl::D, from_reg->as_double_reg(), to_reg->as_double_reg());
1443 } else {
1444 // float to float moves
1445 assert(to_reg->is_single_fpu(), "should match");
1446 __ fmov(FloatRegisterImpl::S, from_reg->as_float_reg(), to_reg->as_float_reg());
1447 }
1448 } else if (!from_reg->is_float_kind() && !to_reg->is_float_kind()) {
1449 if (from_reg->is_double_cpu()) {
1450 #ifdef _LP64
1451 __ mov(from_reg->as_pointer_register(), to_reg->as_pointer_register());
1452 #else
1453 assert(to_reg->is_double_cpu() &&
1454 from_reg->as_register_hi() != to_reg->as_register_lo() &&
1455 from_reg->as_register_lo() != to_reg->as_register_hi(),
1456 "should both be long and not overlap");
1457 // long to long moves
1458 __ mov(from_reg->as_register_hi(), to_reg->as_register_hi());
1459 __ mov(from_reg->as_register_lo(), to_reg->as_register_lo());
1460 #endif
1461 #ifdef _LP64
1462 } else if (to_reg->is_double_cpu()) {
1463 // int to int moves
1464 __ mov(from_reg->as_register(), to_reg->as_register_lo());
1465 #endif
1466 } else {
1467 // int to int moves
1468 __ mov(from_reg->as_register(), to_reg->as_register());
1469 }
1470 } else {
1471 ShouldNotReachHere();
1472 }
1473 if (to_reg->type() == T_OBJECT || to_reg->type() == T_ARRAY) {
1474 __ verify_oop(to_reg->as_register());
1475 }
1476 }
1477
1478
1479 void LIR_Assembler::reg2mem(LIR_Opr from_reg, LIR_Opr dest, BasicType type,
1480 LIR_PatchCode patch_code, CodeEmitInfo* info, bool pop_fpu_stack,
1481 bool wide, bool unaligned) {
1482 LIR_Address* addr = dest->as_address_ptr();
1483
1484 Register src = addr->base()->as_pointer_register();
1485 Register disp_reg = noreg;
1486 int disp_value = addr->disp();
1487 bool needs_patching = (patch_code != lir_patch_none);
1488
1489 if (addr->base()->is_oop_register()) {
1490 __ verify_oop(src);
1491 }
1492
1493 PatchingStub* patch = NULL;
1494 if (needs_patching) {
1495 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1496 assert(!from_reg->is_double_cpu() ||
1497 patch_code == lir_patch_none ||
1498 patch_code == lir_patch_normal, "patching doesn't match register");
1499 }
1500
1501 if (addr->index()->is_illegal()) {
1502 if (!Assembler::is_simm13(disp_value) && (!unaligned || Assembler::is_simm13(disp_value + 4))) {
1503 if (needs_patching) {
1504 __ patchable_set(0, O7);
1505 } else {
1506 __ set(disp_value, O7);
1507 }
1508 disp_reg = O7;
1509 }
1510 } else if (unaligned || PatchALot) {
1511 __ add(src, addr->index()->as_register(), O7);
1512 src = O7;
1513 } else {
1514 disp_reg = addr->index()->as_pointer_register();
1515 assert(disp_value == 0, "can't handle 3 operand addresses");
1516 }
1517
1518 // remember the offset of the store. The patching_epilog must be done
1519 // before the call to add_debug_info_for_null_check, otherwise the PcDescs don't get
1520 // entered in increasing order.
1521 int offset;
1522
1523 assert(disp_reg != noreg || Assembler::is_simm13(disp_value), "should have set this up");
1524 if (disp_reg == noreg) {
1525 offset = store(from_reg, src, disp_value, type, wide, unaligned);
1526 } else {
1527 assert(!unaligned, "can't handle this");
1528 offset = store(from_reg, src, disp_reg, type, wide);
1529 }
1530
1531 if (patch != NULL) {
1532 patching_epilog(patch, patch_code, src, info);
1533 }
1534
1535 if (info != NULL) add_debug_info_for_null_check(offset, info);
1536 }
1537
1538
1539 void LIR_Assembler::return_op(LIR_Opr result) {
1540 // the poll may need a register so just pick one that isn't the return register
1541 #if defined(TIERED) && !defined(_LP64)
1542 if (result->type_field() == LIR_OprDesc::long_type) {
1543 // Must move the result to G1
1544 // Must leave proper result in O0,O1 and G1 (TIERED only)
1545 __ sllx(I0, 32, G1); // Shift bits into high G1
1546 __ srl (I1, 0, I1); // Zero extend O1 (harmless?)
1547 __ or3 (I1, G1, G1); // OR 64 bits into G1
1548 #ifdef ASSERT
1549 // mangle it so any problems will show up
1550 __ set(0xdeadbeef, I0);
1551 __ set(0xdeadbeef, I1);
1552 #endif
1553 }
1554 #endif // TIERED
1555 __ set((intptr_t)os::get_polling_page(), L0);
1556 __ relocate(relocInfo::poll_return_type);
1557 __ ld_ptr(L0, 0, G0);
1558 __ ret();
1559 __ delayed()->restore();
1560 }
1561
1562
1563 int LIR_Assembler::safepoint_poll(LIR_Opr tmp, CodeEmitInfo* info) {
1564 __ set((intptr_t)os::get_polling_page(), tmp->as_register());
1565 if (info != NULL) {
1566 add_debug_info_for_branch(info);
1567 } else {
1568 __ relocate(relocInfo::poll_type);
1569 }
1570
1571 int offset = __ offset();
1572 __ ld_ptr(tmp->as_register(), 0, G0);
1573
1574 return offset;
1575 }
1576
1577
1578 void LIR_Assembler::emit_static_call_stub() {
1579 address call_pc = __ pc();
1580 address stub = __ start_a_stub(call_stub_size);
1581 if (stub == NULL) {
1582 bailout("static call stub overflow");
1583 return;
1584 }
1585
1586 int start = __ offset();
1587 __ relocate(static_stub_Relocation::spec(call_pc));
1588
1589 __ set_oop(NULL, G5);
1590 // must be set to -1 at code generation time
1591 AddressLiteral addrlit(-1);
1592 __ jump_to(addrlit, G3);
1593 __ delayed()->nop();
1594
1595 assert(__ offset() - start <= call_stub_size, "stub too big");
1596 __ end_a_stub();
1597 }
1598
1599
1600 void LIR_Assembler::comp_op(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Op2* op) {
1601 if (opr1->is_single_fpu()) {
1602 __ fcmp(FloatRegisterImpl::S, Assembler::fcc0, opr1->as_float_reg(), opr2->as_float_reg());
1603 } else if (opr1->is_double_fpu()) {
1604 __ fcmp(FloatRegisterImpl::D, Assembler::fcc0, opr1->as_double_reg(), opr2->as_double_reg());
1605 } else if (opr1->is_single_cpu()) {
1606 if (opr2->is_constant()) {
1607 switch (opr2->as_constant_ptr()->type()) {
1608 case T_INT:
1609 { jint con = opr2->as_constant_ptr()->as_jint();
1610 if (Assembler::is_simm13(con)) {
1611 __ cmp(opr1->as_register(), con);
1612 } else {
1613 __ set(con, O7);
1614 __ cmp(opr1->as_register(), O7);
1615 }
1616 }
1617 break;
1618
1619 case T_OBJECT:
1620 // there are only equal/notequal comparisions on objects
1621 { jobject con = opr2->as_constant_ptr()->as_jobject();
1622 if (con == NULL) {
1623 __ cmp(opr1->as_register(), 0);
1624 } else {
1625 jobject2reg(con, O7);
1626 __ cmp(opr1->as_register(), O7);
1627 }
1628 }
1629 break;
1630
1631 default:
1632 ShouldNotReachHere();
1633 break;
1634 }
1635 } else {
1636 if (opr2->is_address()) {
1637 LIR_Address * addr = opr2->as_address_ptr();
1638 BasicType type = addr->type();
1639 if ( type == T_OBJECT ) __ ld_ptr(as_Address(addr), O7);
1640 else __ ld(as_Address(addr), O7);
1641 __ cmp(opr1->as_register(), O7);
1642 } else {
1643 __ cmp(opr1->as_register(), opr2->as_register());
1644 }
1645 }
1646 } else if (opr1->is_double_cpu()) {
1647 Register xlo = opr1->as_register_lo();
1648 Register xhi = opr1->as_register_hi();
1649 if (opr2->is_constant() && opr2->as_jlong() == 0) {
1650 assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "only handles these cases");
1651 #ifdef _LP64
1652 __ orcc(xhi, G0, G0);
1653 #else
1654 __ orcc(xhi, xlo, G0);
1655 #endif
1656 } else if (opr2->is_register()) {
1657 Register ylo = opr2->as_register_lo();
1658 Register yhi = opr2->as_register_hi();
1659 #ifdef _LP64
1660 __ cmp(xlo, ylo);
1661 #else
1662 __ subcc(xlo, ylo, xlo);
1663 __ subccc(xhi, yhi, xhi);
1664 if (condition == lir_cond_equal || condition == lir_cond_notEqual) {
1665 __ orcc(xhi, xlo, G0);
1666 }
1667 #endif
1668 } else {
1669 ShouldNotReachHere();
1670 }
1671 } else if (opr1->is_address()) {
1672 LIR_Address * addr = opr1->as_address_ptr();
1673 BasicType type = addr->type();
1674 assert (opr2->is_constant(), "Checking");
1675 if ( type == T_OBJECT ) __ ld_ptr(as_Address(addr), O7);
1676 else __ ld(as_Address(addr), O7);
1677 __ cmp(O7, opr2->as_constant_ptr()->as_jint());
1678 } else {
1679 ShouldNotReachHere();
1680 }
1681 }
1682
1683
1684 void LIR_Assembler::comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst, LIR_Op2* op){
1685 if (code == lir_cmp_fd2i || code == lir_ucmp_fd2i) {
1686 bool is_unordered_less = (code == lir_ucmp_fd2i);
1687 if (left->is_single_fpu()) {
1688 __ float_cmp(true, is_unordered_less ? -1 : 1, left->as_float_reg(), right->as_float_reg(), dst->as_register());
1689 } else if (left->is_double_fpu()) {
1690 __ float_cmp(false, is_unordered_less ? -1 : 1, left->as_double_reg(), right->as_double_reg(), dst->as_register());
1691 } else {
1692 ShouldNotReachHere();
1693 }
1694 } else if (code == lir_cmp_l2i) {
1695 #ifdef _LP64
1696 __ lcmp(left->as_register_lo(), right->as_register_lo(), dst->as_register());
1697 #else
1698 __ lcmp(left->as_register_hi(), left->as_register_lo(),
1699 right->as_register_hi(), right->as_register_lo(),
1700 dst->as_register());
1701 #endif
1702 } else {
1703 ShouldNotReachHere();
1704 }
1705 }
1706
1707
1708 void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type) {
1709 Assembler::Condition acond;
1710 switch (condition) {
1711 case lir_cond_equal: acond = Assembler::equal; break;
1712 case lir_cond_notEqual: acond = Assembler::notEqual; break;
1713 case lir_cond_less: acond = Assembler::less; break;
1714 case lir_cond_lessEqual: acond = Assembler::lessEqual; break;
1715 case lir_cond_greaterEqual: acond = Assembler::greaterEqual; break;
1716 case lir_cond_greater: acond = Assembler::greater; break;
1717 case lir_cond_aboveEqual: acond = Assembler::greaterEqualUnsigned; break;
1718 case lir_cond_belowEqual: acond = Assembler::lessEqualUnsigned; break;
1719 default: ShouldNotReachHere();
1720 };
1721
1722 if (opr1->is_constant() && opr1->type() == T_INT) {
1723 Register dest = result->as_register();
1724 // load up first part of constant before branch
1725 // and do the rest in the delay slot.
1726 if (!Assembler::is_simm13(opr1->as_jint())) {
1727 __ sethi(opr1->as_jint(), dest);
1728 }
1729 } else if (opr1->is_constant()) {
1730 const2reg(opr1, result, lir_patch_none, NULL);
1731 } else if (opr1->is_register()) {
1732 reg2reg(opr1, result);
1733 } else if (opr1->is_stack()) {
1734 stack2reg(opr1, result, result->type());
1735 } else {
1736 ShouldNotReachHere();
1737 }
1738 Label skip;
1739 #ifdef _LP64
1740 if (type == T_INT) {
1741 __ br(acond, false, Assembler::pt, skip);
1742 } else
1743 #endif
1744 __ brx(acond, false, Assembler::pt, skip); // checks icc on 32bit and xcc on 64bit
1745 if (opr1->is_constant() && opr1->type() == T_INT) {
1746 Register dest = result->as_register();
1747 if (Assembler::is_simm13(opr1->as_jint())) {
1748 __ delayed()->or3(G0, opr1->as_jint(), dest);
1749 } else {
1750 // the sethi has been done above, so just put in the low 10 bits
1751 __ delayed()->or3(dest, opr1->as_jint() & 0x3ff, dest);
1752 }
1753 } else {
1754 // can't do anything useful in the delay slot
1755 __ delayed()->nop();
1756 }
1757 if (opr2->is_constant()) {
1758 const2reg(opr2, result, lir_patch_none, NULL);
1759 } else if (opr2->is_register()) {
1760 reg2reg(opr2, result);
1761 } else if (opr2->is_stack()) {
1762 stack2reg(opr2, result, result->type());
1763 } else {
1764 ShouldNotReachHere();
1765 }
1766 __ bind(skip);
1767 }
1768
1769
1770 void LIR_Assembler::arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest, CodeEmitInfo* info, bool pop_fpu_stack) {
1771 assert(info == NULL, "unused on this code path");
1772 assert(left->is_register(), "wrong items state");
1773 assert(dest->is_register(), "wrong items state");
1774
1775 if (right->is_register()) {
1776 if (dest->is_float_kind()) {
1777
1778 FloatRegister lreg, rreg, res;
1779 FloatRegisterImpl::Width w;
1780 if (right->is_single_fpu()) {
1781 w = FloatRegisterImpl::S;
1782 lreg = left->as_float_reg();
1783 rreg = right->as_float_reg();
1784 res = dest->as_float_reg();
1785 } else {
1786 w = FloatRegisterImpl::D;
1787 lreg = left->as_double_reg();
1788 rreg = right->as_double_reg();
1789 res = dest->as_double_reg();
1790 }
1791
1792 switch (code) {
1793 case lir_add: __ fadd(w, lreg, rreg, res); break;
1794 case lir_sub: __ fsub(w, lreg, rreg, res); break;
1795 case lir_mul: // fall through
1796 case lir_mul_strictfp: __ fmul(w, lreg, rreg, res); break;
1797 case lir_div: // fall through
1798 case lir_div_strictfp: __ fdiv(w, lreg, rreg, res); break;
1799 default: ShouldNotReachHere();
1800 }
1801
1802 } else if (dest->is_double_cpu()) {
1803 #ifdef _LP64
1804 Register dst_lo = dest->as_register_lo();
1805 Register op1_lo = left->as_pointer_register();
1806 Register op2_lo = right->as_pointer_register();
1807
1808 switch (code) {
1809 case lir_add:
1810 __ add(op1_lo, op2_lo, dst_lo);
1811 break;
1812
1813 case lir_sub:
1814 __ sub(op1_lo, op2_lo, dst_lo);
1815 break;
1816
1817 default: ShouldNotReachHere();
1818 }
1819 #else
1820 Register op1_lo = left->as_register_lo();
1821 Register op1_hi = left->as_register_hi();
1822 Register op2_lo = right->as_register_lo();
1823 Register op2_hi = right->as_register_hi();
1824 Register dst_lo = dest->as_register_lo();
1825 Register dst_hi = dest->as_register_hi();
1826
1827 switch (code) {
1828 case lir_add:
1829 __ addcc(op1_lo, op2_lo, dst_lo);
1830 __ addc (op1_hi, op2_hi, dst_hi);
1831 break;
1832
1833 case lir_sub:
1834 __ subcc(op1_lo, op2_lo, dst_lo);
1835 __ subc (op1_hi, op2_hi, dst_hi);
1836 break;
1837
1838 default: ShouldNotReachHere();
1839 }
1840 #endif
1841 } else {
1842 assert (right->is_single_cpu(), "Just Checking");
1843
1844 Register lreg = left->as_register();
1845 Register res = dest->as_register();
1846 Register rreg = right->as_register();
1847 switch (code) {
1848 case lir_add: __ add (lreg, rreg, res); break;
1849 case lir_sub: __ sub (lreg, rreg, res); break;
1850 case lir_mul: __ mult (lreg, rreg, res); break;
1851 default: ShouldNotReachHere();
1852 }
1853 }
1854 } else {
1855 assert (right->is_constant(), "must be constant");
1856
1857 if (dest->is_single_cpu()) {
1858 Register lreg = left->as_register();
1859 Register res = dest->as_register();
1860 int simm13 = right->as_constant_ptr()->as_jint();
1861
1862 switch (code) {
1863 case lir_add: __ add (lreg, simm13, res); break;
1864 case lir_sub: __ sub (lreg, simm13, res); break;
1865 case lir_mul: __ mult (lreg, simm13, res); break;
1866 default: ShouldNotReachHere();
1867 }
1868 } else {
1869 Register lreg = left->as_pointer_register();
1870 Register res = dest->as_register_lo();
1871 long con = right->as_constant_ptr()->as_jlong();
1872 assert(Assembler::is_simm13(con), "must be simm13");
1873
1874 switch (code) {
1875 case lir_add: __ add (lreg, (int)con, res); break;
1876 case lir_sub: __ sub (lreg, (int)con, res); break;
1877 case lir_mul: __ mult (lreg, (int)con, res); break;
1878 default: ShouldNotReachHere();
1879 }
1880 }
1881 }
1882 }
1883
1884
1885 void LIR_Assembler::fpop() {
1886 // do nothing
1887 }
1888
1889
1890 void LIR_Assembler::intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr thread, LIR_Opr dest, LIR_Op* op) {
1891 switch (code) {
1892 case lir_sin:
1893 case lir_tan:
1894 case lir_cos: {
1895 assert(thread->is_valid(), "preserve the thread object for performance reasons");
1896 assert(dest->as_double_reg() == F0, "the result will be in f0/f1");
1897 break;
1898 }
1899 case lir_sqrt: {
1900 assert(!thread->is_valid(), "there is no need for a thread_reg for dsqrt");
1901 FloatRegister src_reg = value->as_double_reg();
1902 FloatRegister dst_reg = dest->as_double_reg();
1903 __ fsqrt(FloatRegisterImpl::D, src_reg, dst_reg);
1904 break;
1905 }
1906 case lir_abs: {
1907 assert(!thread->is_valid(), "there is no need for a thread_reg for fabs");
1908 FloatRegister src_reg = value->as_double_reg();
1909 FloatRegister dst_reg = dest->as_double_reg();
1910 __ fabs(FloatRegisterImpl::D, src_reg, dst_reg);
1911 break;
1912 }
1913 default: {
1914 ShouldNotReachHere();
1915 break;
1916 }
1917 }
1918 }
1919
1920
1921 void LIR_Assembler::logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest) {
1922 if (right->is_constant()) {
1923 if (dest->is_single_cpu()) {
1924 int simm13 = right->as_constant_ptr()->as_jint();
1925 switch (code) {
1926 case lir_logic_and: __ and3 (left->as_register(), simm13, dest->as_register()); break;
1927 case lir_logic_or: __ or3 (left->as_register(), simm13, dest->as_register()); break;
1928 case lir_logic_xor: __ xor3 (left->as_register(), simm13, dest->as_register()); break;
1929 default: ShouldNotReachHere();
1930 }
1931 } else {
1932 long c = right->as_constant_ptr()->as_jlong();
1933 assert(c == (int)c && Assembler::is_simm13(c), "out of range");
1934 int simm13 = (int)c;
1935 switch (code) {
1936 case lir_logic_and:
1937 #ifndef _LP64
1938 __ and3 (left->as_register_hi(), 0, dest->as_register_hi());
1939 #endif
1940 __ and3 (left->as_register_lo(), simm13, dest->as_register_lo());
1941 break;
1942
1943 case lir_logic_or:
1944 #ifndef _LP64
1945 __ or3 (left->as_register_hi(), 0, dest->as_register_hi());
1946 #endif
1947 __ or3 (left->as_register_lo(), simm13, dest->as_register_lo());
1948 break;
1949
1950 case lir_logic_xor:
1951 #ifndef _LP64
1952 __ xor3 (left->as_register_hi(), 0, dest->as_register_hi());
1953 #endif
1954 __ xor3 (left->as_register_lo(), simm13, dest->as_register_lo());
1955 break;
1956
1957 default: ShouldNotReachHere();
1958 }
1959 }
1960 } else {
1961 assert(right->is_register(), "right should be in register");
1962
1963 if (dest->is_single_cpu()) {
1964 switch (code) {
1965 case lir_logic_and: __ and3 (left->as_register(), right->as_register(), dest->as_register()); break;
1966 case lir_logic_or: __ or3 (left->as_register(), right->as_register(), dest->as_register()); break;
1967 case lir_logic_xor: __ xor3 (left->as_register(), right->as_register(), dest->as_register()); break;
1968 default: ShouldNotReachHere();
1969 }
1970 } else {
1971 #ifdef _LP64
1972 Register l = (left->is_single_cpu() && left->is_oop_register()) ? left->as_register() :
1973 left->as_register_lo();
1974 Register r = (right->is_single_cpu() && right->is_oop_register()) ? right->as_register() :
1975 right->as_register_lo();
1976
1977 switch (code) {
1978 case lir_logic_and: __ and3 (l, r, dest->as_register_lo()); break;
1979 case lir_logic_or: __ or3 (l, r, dest->as_register_lo()); break;
1980 case lir_logic_xor: __ xor3 (l, r, dest->as_register_lo()); break;
1981 default: ShouldNotReachHere();
1982 }
1983 #else
1984 switch (code) {
1985 case lir_logic_and:
1986 __ and3 (left->as_register_hi(), right->as_register_hi(), dest->as_register_hi());
1987 __ and3 (left->as_register_lo(), right->as_register_lo(), dest->as_register_lo());
1988 break;
1989
1990 case lir_logic_or:
1991 __ or3 (left->as_register_hi(), right->as_register_hi(), dest->as_register_hi());
1992 __ or3 (left->as_register_lo(), right->as_register_lo(), dest->as_register_lo());
1993 break;
1994
1995 case lir_logic_xor:
1996 __ xor3 (left->as_register_hi(), right->as_register_hi(), dest->as_register_hi());
1997 __ xor3 (left->as_register_lo(), right->as_register_lo(), dest->as_register_lo());
1998 break;
1999
2000 default: ShouldNotReachHere();
2001 }
2002 #endif
2003 }
2004 }
2005 }
2006
2007
2008 int LIR_Assembler::shift_amount(BasicType t) {
2009 int elem_size = type2aelembytes(t);
2010 switch (elem_size) {
2011 case 1 : return 0;
2012 case 2 : return 1;
2013 case 4 : return 2;
2014 case 8 : return 3;
2015 }
2016 ShouldNotReachHere();
2017 return -1;
2018 }
2019
2020
2021 void LIR_Assembler::throw_op(LIR_Opr exceptionPC, LIR_Opr exceptionOop, CodeEmitInfo* info) {
2022 assert(exceptionOop->as_register() == Oexception, "should match");
2023 assert(exceptionPC->as_register() == Oissuing_pc, "should match");
2024
2025 info->add_register_oop(exceptionOop);
2026
2027 // reuse the debug info from the safepoint poll for the throw op itself
2028 address pc_for_athrow = __ pc();
2029 int pc_for_athrow_offset = __ offset();
2030 RelocationHolder rspec = internal_word_Relocation::spec(pc_for_athrow);
2031 __ set(pc_for_athrow, Oissuing_pc, rspec);
2032 add_call_info(pc_for_athrow_offset, info); // for exception handler
2033
2034 __ call(Runtime1::entry_for(Runtime1::handle_exception_id), relocInfo::runtime_call_type);
2035 __ delayed()->nop();
2036 }
2037
2038
2039 void LIR_Assembler::unwind_op(LIR_Opr exceptionOop) {
2040 assert(exceptionOop->as_register() == Oexception, "should match");
2041
2042 __ br(Assembler::always, false, Assembler::pt, _unwind_handler_entry);
2043 __ delayed()->nop();
2044 }
2045
2046
2047 void LIR_Assembler::emit_arraycopy(LIR_OpArrayCopy* op) {
2048 Register src = op->src()->as_register();
2049 Register dst = op->dst()->as_register();
2050 Register src_pos = op->src_pos()->as_register();
2051 Register dst_pos = op->dst_pos()->as_register();
2052 Register length = op->length()->as_register();
2053 Register tmp = op->tmp()->as_register();
2054 Register tmp2 = O7;
2055
2056 int flags = op->flags();
2057 ciArrayKlass* default_type = op->expected_type();
2058 BasicType basic_type = default_type != NULL ? default_type->element_type()->basic_type() : T_ILLEGAL;
2059 if (basic_type == T_ARRAY) basic_type = T_OBJECT;
2060
2061 #ifdef _LP64
2062 // higher 32bits must be null
2063 __ sra(dst_pos, 0, dst_pos);
2064 __ sra(src_pos, 0, src_pos);
2065 __ sra(length, 0, length);
2066 #endif
2067
2068 // set up the arraycopy stub information
2069 ArrayCopyStub* stub = op->stub();
2070
2071 // always do stub if no type information is available. it's ok if
2072 // the known type isn't loaded since the code sanity checks
2073 // in debug mode and the type isn't required when we know the exact type
2074 // also check that the type is an array type.
2075 if (op->expected_type() == NULL) {
2076 __ mov(src, O0);
2077 __ mov(src_pos, O1);
2078 __ mov(dst, O2);
2079 __ mov(dst_pos, O3);
2080 __ mov(length, O4);
2081 address copyfunc_addr = StubRoutines::generic_arraycopy();
2082
2083 if (copyfunc_addr == NULL) { // Use C version if stub was not generated
2084 __ call_VM_leaf(tmp, CAST_FROM_FN_PTR(address, Runtime1::arraycopy));
2085 } else {
2086 #ifndef PRODUCT
2087 if (PrintC1Statistics) {
2088 address counter = (address)&Runtime1::_generic_arraycopystub_cnt;
2089 __ inc_counter(counter, G1, G3);
2090 }
2091 #endif
2092 __ call_VM_leaf(tmp, copyfunc_addr);
2093 }
2094
2095 if (copyfunc_addr != NULL) {
2096 __ xor3(O0, -1, tmp);
2097 __ sub(length, tmp, length);
2098 __ add(src_pos, tmp, src_pos);
2099 __ br_zero(Assembler::less, false, Assembler::pn, O0, *stub->entry());
2100 __ delayed()->add(dst_pos, tmp, dst_pos);
2101 } else {
2102 __ br_zero(Assembler::less, false, Assembler::pn, O0, *stub->entry());
2103 __ delayed()->nop();
2104 }
2105 __ bind(*stub->continuation());
2106 return;
2107 }
2108
2109 assert(default_type != NULL && default_type->is_array_klass(), "must be true at this point");
2110
2111 // make sure src and dst are non-null and load array length
2112 if (flags & LIR_OpArrayCopy::src_null_check) {
2113 __ tst(src);
2114 __ brx(Assembler::equal, false, Assembler::pn, *stub->entry());
2115 __ delayed()->nop();
2116 }
2117
2118 if (flags & LIR_OpArrayCopy::dst_null_check) {
2119 __ tst(dst);
2120 __ brx(Assembler::equal, false, Assembler::pn, *stub->entry());
2121 __ delayed()->nop();
2122 }
2123
2124 if (flags & LIR_OpArrayCopy::src_pos_positive_check) {
2125 // test src_pos register
2126 __ tst(src_pos);
2127 __ br(Assembler::less, false, Assembler::pn, *stub->entry());
2128 __ delayed()->nop();
2129 }
2130
2131 if (flags & LIR_OpArrayCopy::dst_pos_positive_check) {
2132 // test dst_pos register
2133 __ tst(dst_pos);
2134 __ br(Assembler::less, false, Assembler::pn, *stub->entry());
2135 __ delayed()->nop();
2136 }
2137
2138 if (flags & LIR_OpArrayCopy::length_positive_check) {
2139 // make sure length isn't negative
2140 __ tst(length);
2141 __ br(Assembler::less, false, Assembler::pn, *stub->entry());
2142 __ delayed()->nop();
2143 }
2144
2145 if (flags & LIR_OpArrayCopy::src_range_check) {
2146 __ ld(src, arrayOopDesc::length_offset_in_bytes(), tmp2);
2147 __ add(length, src_pos, tmp);
2148 __ cmp(tmp2, tmp);
2149 __ br(Assembler::carrySet, false, Assembler::pn, *stub->entry());
2150 __ delayed()->nop();
2151 }
2152
2153 if (flags & LIR_OpArrayCopy::dst_range_check) {
2154 __ ld(dst, arrayOopDesc::length_offset_in_bytes(), tmp2);
2155 __ add(length, dst_pos, tmp);
2156 __ cmp(tmp2, tmp);
2157 __ br(Assembler::carrySet, false, Assembler::pn, *stub->entry());
2158 __ delayed()->nop();
2159 }
2160
2161 int shift = shift_amount(basic_type);
2162
2163 if (flags & LIR_OpArrayCopy::type_check) {
2164 // We don't know the array types are compatible
2165 if (basic_type != T_OBJECT) {
2166 // Simple test for basic type arrays
2167 if (UseCompressedOops) {
2168 // We don't need decode because we just need to compare
2169 __ lduw(src, oopDesc::klass_offset_in_bytes(), tmp);
2170 __ lduw(dst, oopDesc::klass_offset_in_bytes(), tmp2);
2171 __ cmp(tmp, tmp2);
2172 __ br(Assembler::notEqual, false, Assembler::pt, *stub->entry());
2173 } else {
2174 __ ld_ptr(src, oopDesc::klass_offset_in_bytes(), tmp);
2175 __ ld_ptr(dst, oopDesc::klass_offset_in_bytes(), tmp2);
2176 __ cmp(tmp, tmp2);
2177 __ brx(Assembler::notEqual, false, Assembler::pt, *stub->entry());
2178 }
2179 __ delayed()->nop();
2180 } else {
2181 // For object arrays, if src is a sub class of dst then we can
2182 // safely do the copy.
2183 address copyfunc_addr = StubRoutines::checkcast_arraycopy();
2184
2185 Label cont, slow;
2186 assert_different_registers(tmp, tmp2, G3, G1);
2187
2188 __ load_klass(src, G3);
2189 __ load_klass(dst, G1);
2190
2191 __ check_klass_subtype_fast_path(G3, G1, tmp, tmp2, &cont, copyfunc_addr == NULL ? stub->entry() : &slow, NULL);
2192
2193 __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type);
2194 __ delayed()->nop();
2195
2196 __ cmp(G3, 0);
2197 if (copyfunc_addr != NULL) { // use stub if available
2198 // src is not a sub class of dst so we have to do a
2199 // per-element check.
2200 __ br(Assembler::notEqual, false, Assembler::pt, cont);
2201 __ delayed()->nop();
2202
2203 __ bind(slow);
2204
2205 int mask = LIR_OpArrayCopy::src_objarray|LIR_OpArrayCopy::dst_objarray;
2206 if ((flags & mask) != mask) {
2207 // Check that at least both of them object arrays.
2208 assert(flags & mask, "one of the two should be known to be an object array");
2209
2210 if (!(flags & LIR_OpArrayCopy::src_objarray)) {
2211 __ load_klass(src, tmp);
2212 } else if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
2213 __ load_klass(dst, tmp);
2214 }
2215 int lh_offset = klassOopDesc::header_size() * HeapWordSize +
2216 Klass::layout_helper_offset_in_bytes();
2217
2218 __ lduw(tmp, lh_offset, tmp2);
2219
2220 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2221 __ set(objArray_lh, tmp);
2222 __ cmp(tmp, tmp2);
2223 __ br(Assembler::notEqual, false, Assembler::pt, *stub->entry());
2224 __ delayed()->nop();
2225 }
2226
2227 Register src_ptr = O0;
2228 Register dst_ptr = O1;
2229 Register len = O2;
2230 Register chk_off = O3;
2231 Register super_k = O4;
2232
2233 __ add(src, arrayOopDesc::base_offset_in_bytes(basic_type), src_ptr);
2234 if (shift == 0) {
2235 __ add(src_ptr, src_pos, src_ptr);
2236 } else {
2237 __ sll(src_pos, shift, tmp);
2238 __ add(src_ptr, tmp, src_ptr);
2239 }
2240
2241 __ add(dst, arrayOopDesc::base_offset_in_bytes(basic_type), dst_ptr);
2242 if (shift == 0) {
2243 __ add(dst_ptr, dst_pos, dst_ptr);
2244 } else {
2245 __ sll(dst_pos, shift, tmp);
2246 __ add(dst_ptr, tmp, dst_ptr);
2247 }
2248 __ mov(length, len);
2249 __ load_klass(dst, tmp);
2250
2251 int ek_offset = (klassOopDesc::header_size() * HeapWordSize +
2252 objArrayKlass::element_klass_offset_in_bytes());
2253 __ ld_ptr(tmp, ek_offset, super_k);
2254
2255 int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
2256 Klass::super_check_offset_offset_in_bytes());
2257 __ lduw(super_k, sco_offset, chk_off);
2258
2259 __ call_VM_leaf(tmp, copyfunc_addr);
2260
2261 #ifndef PRODUCT
2262 if (PrintC1Statistics) {
2263 Label failed;
2264 __ br_notnull(O0, false, Assembler::pn, failed);
2265 __ delayed()->nop();
2266 __ inc_counter((address)&Runtime1::_arraycopy_checkcast_cnt, G1, G3);
2267 __ bind(failed);
2268 }
2269 #endif
2270
2271 __ br_null(O0, false, Assembler::pt, *stub->continuation());
2272 __ delayed()->xor3(O0, -1, tmp);
2273
2274 #ifndef PRODUCT
2275 if (PrintC1Statistics) {
2276 __ inc_counter((address)&Runtime1::_arraycopy_checkcast_attempt_cnt, G1, G3);
2277 }
2278 #endif
2279
2280 __ sub(length, tmp, length);
2281 __ add(src_pos, tmp, src_pos);
2282 __ br(Assembler::always, false, Assembler::pt, *stub->entry());
2283 __ delayed()->add(dst_pos, tmp, dst_pos);
2284
2285 __ bind(cont);
2286 } else {
2287 __ br(Assembler::equal, false, Assembler::pn, *stub->entry());
2288 __ delayed()->nop();
2289 __ bind(cont);
2290 }
2291 }
2292 }
2293
2294 #ifdef ASSERT
2295 if (basic_type != T_OBJECT || !(flags & LIR_OpArrayCopy::type_check)) {
2296 // Sanity check the known type with the incoming class. For the
2297 // primitive case the types must match exactly with src.klass and
2298 // dst.klass each exactly matching the default type. For the
2299 // object array case, if no type check is needed then either the
2300 // dst type is exactly the expected type and the src type is a
2301 // subtype which we can't check or src is the same array as dst
2302 // but not necessarily exactly of type default_type.
2303 Label known_ok, halt;
2304 jobject2reg(op->expected_type()->constant_encoding(), tmp);
2305 if (UseCompressedOops) {
2306 // tmp holds the default type. It currently comes uncompressed after the
2307 // load of a constant, so encode it.
2308 __ encode_heap_oop(tmp);
2309 // load the raw value of the dst klass, since we will be comparing
2310 // uncompressed values directly.
2311 __ lduw(dst, oopDesc::klass_offset_in_bytes(), tmp2);
2312 if (basic_type != T_OBJECT) {
2313 __ cmp(tmp, tmp2);
2314 __ br(Assembler::notEqual, false, Assembler::pn, halt);
2315 // load the raw value of the src klass.
2316 __ delayed()->lduw(src, oopDesc::klass_offset_in_bytes(), tmp2);
2317 __ cmp(tmp, tmp2);
2318 __ br(Assembler::equal, false, Assembler::pn, known_ok);
2319 __ delayed()->nop();
2320 } else {
2321 __ cmp(tmp, tmp2);
2322 __ br(Assembler::equal, false, Assembler::pn, known_ok);
2323 __ delayed()->cmp(src, dst);
2324 __ brx(Assembler::equal, false, Assembler::pn, known_ok);
2325 __ delayed()->nop();
2326 }
2327 } else {
2328 __ ld_ptr(dst, oopDesc::klass_offset_in_bytes(), tmp2);
2329 if (basic_type != T_OBJECT) {
2330 __ cmp(tmp, tmp2);
2331 __ brx(Assembler::notEqual, false, Assembler::pn, halt);
2332 __ delayed()->ld_ptr(src, oopDesc::klass_offset_in_bytes(), tmp2);
2333 __ cmp(tmp, tmp2);
2334 __ brx(Assembler::equal, false, Assembler::pn, known_ok);
2335 __ delayed()->nop();
2336 } else {
2337 __ cmp(tmp, tmp2);
2338 __ brx(Assembler::equal, false, Assembler::pn, known_ok);
2339 __ delayed()->cmp(src, dst);
2340 __ brx(Assembler::equal, false, Assembler::pn, known_ok);
2341 __ delayed()->nop();
2342 }
2343 }
2344 __ bind(halt);
2345 __ stop("incorrect type information in arraycopy");
2346 __ bind(known_ok);
2347 }
2348 #endif
2349
2350 #ifndef PRODUCT
2351 if (PrintC1Statistics) {
2352 address counter = Runtime1::arraycopy_count_address(basic_type);
2353 __ inc_counter(counter, G1, G3);
2354 }
2355 #endif
2356
2357 Register src_ptr = O0;
2358 Register dst_ptr = O1;
2359 Register len = O2;
2360
2361 __ add(src, arrayOopDesc::base_offset_in_bytes(basic_type), src_ptr);
2362 if (shift == 0) {
2363 __ add(src_ptr, src_pos, src_ptr);
2364 } else {
2365 __ sll(src_pos, shift, tmp);
2366 __ add(src_ptr, tmp, src_ptr);
2367 }
2368
2369 __ add(dst, arrayOopDesc::base_offset_in_bytes(basic_type), dst_ptr);
2370 if (shift == 0) {
2371 __ add(dst_ptr, dst_pos, dst_ptr);
2372 } else {
2373 __ sll(dst_pos, shift, tmp);
2374 __ add(dst_ptr, tmp, dst_ptr);
2375 }
2376
2377 bool disjoint = (flags & LIR_OpArrayCopy::overlapping) == 0;
2378 bool aligned = (flags & LIR_OpArrayCopy::unaligned) == 0;
2379 const char *name;
2380 address entry = StubRoutines::select_arraycopy_function(basic_type, aligned, disjoint, name, false);
2381
2382 // arraycopy stubs takes a length in number of elements, so don't scale it.
2383 __ mov(length, len);
2384 __ call_VM_leaf(tmp, entry);
2385
2386 __ bind(*stub->continuation());
2387 }
2388
2389
2390 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
2391 if (dest->is_single_cpu()) {
2392 #ifdef _LP64
2393 if (left->type() == T_OBJECT) {
2394 switch (code) {
2395 case lir_shl: __ sllx (left->as_register(), count->as_register(), dest->as_register()); break;
2396 case lir_shr: __ srax (left->as_register(), count->as_register(), dest->as_register()); break;
2397 case lir_ushr: __ srl (left->as_register(), count->as_register(), dest->as_register()); break;
2398 default: ShouldNotReachHere();
2399 }
2400 } else
2401 #endif
2402 switch (code) {
2403 case lir_shl: __ sll (left->as_register(), count->as_register(), dest->as_register()); break;
2404 case lir_shr: __ sra (left->as_register(), count->as_register(), dest->as_register()); break;
2405 case lir_ushr: __ srl (left->as_register(), count->as_register(), dest->as_register()); break;
2406 default: ShouldNotReachHere();
2407 }
2408 } else {
2409 #ifdef _LP64
2410 switch (code) {
2411 case lir_shl: __ sllx (left->as_register_lo(), count->as_register(), dest->as_register_lo()); break;
2412 case lir_shr: __ srax (left->as_register_lo(), count->as_register(), dest->as_register_lo()); break;
2413 case lir_ushr: __ srlx (left->as_register_lo(), count->as_register(), dest->as_register_lo()); break;
2414 default: ShouldNotReachHere();
2415 }
2416 #else
2417 switch (code) {
2418 case lir_shl: __ lshl (left->as_register_hi(), left->as_register_lo(), count->as_register(), dest->as_register_hi(), dest->as_register_lo(), G3_scratch); break;
2419 case lir_shr: __ lshr (left->as_register_hi(), left->as_register_lo(), count->as_register(), dest->as_register_hi(), dest->as_register_lo(), G3_scratch); break;
2420 case lir_ushr: __ lushr (left->as_register_hi(), left->as_register_lo(), count->as_register(), dest->as_register_hi(), dest->as_register_lo(), G3_scratch); break;
2421 default: ShouldNotReachHere();
2422 }
2423 #endif
2424 }
2425 }
2426
2427
2428 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, jint count, LIR_Opr dest) {
2429 #ifdef _LP64
2430 if (left->type() == T_OBJECT) {
2431 count = count & 63; // shouldn't shift by more than sizeof(intptr_t)
2432 Register l = left->as_register();
2433 Register d = dest->as_register_lo();
2434 switch (code) {
2435 case lir_shl: __ sllx (l, count, d); break;
2436 case lir_shr: __ srax (l, count, d); break;
2437 case lir_ushr: __ srlx (l, count, d); break;
2438 default: ShouldNotReachHere();
2439 }
2440 return;
2441 }
2442 #endif
2443
2444 if (dest->is_single_cpu()) {
2445 count = count & 0x1F; // Java spec
2446 switch (code) {
2447 case lir_shl: __ sll (left->as_register(), count, dest->as_register()); break;
2448 case lir_shr: __ sra (left->as_register(), count, dest->as_register()); break;
2449 case lir_ushr: __ srl (left->as_register(), count, dest->as_register()); break;
2450 default: ShouldNotReachHere();
2451 }
2452 } else if (dest->is_double_cpu()) {
2453 count = count & 63; // Java spec
2454 switch (code) {
2455 case lir_shl: __ sllx (left->as_pointer_register(), count, dest->as_pointer_register()); break;
2456 case lir_shr: __ srax (left->as_pointer_register(), count, dest->as_pointer_register()); break;
2457 case lir_ushr: __ srlx (left->as_pointer_register(), count, dest->as_pointer_register()); break;
2458 default: ShouldNotReachHere();
2459 }
2460 } else {
2461 ShouldNotReachHere();
2462 }
2463 }
2464
2465
2466 void LIR_Assembler::emit_alloc_obj(LIR_OpAllocObj* op) {
2467 assert(op->tmp1()->as_register() == G1 &&
2468 op->tmp2()->as_register() == G3 &&
2469 op->tmp3()->as_register() == G4 &&
2470 op->obj()->as_register() == O0 &&
2471 op->klass()->as_register() == G5, "must be");
2472 if (op->init_check()) {
2473 __ ld(op->klass()->as_register(),
2474 instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc),
2475 op->tmp1()->as_register());
2476 add_debug_info_for_null_check_here(op->stub()->info());
2477 __ cmp(op->tmp1()->as_register(), instanceKlass::fully_initialized);
2478 __ br(Assembler::notEqual, false, Assembler::pn, *op->stub()->entry());
2479 __ delayed()->nop();
2480 }
2481 __ allocate_object(op->obj()->as_register(),
2482 op->tmp1()->as_register(),
2483 op->tmp2()->as_register(),
2484 op->tmp3()->as_register(),
2485 op->header_size(),
2486 op->object_size(),
2487 op->klass()->as_register(),
2488 *op->stub()->entry());
2489 __ bind(*op->stub()->continuation());
2490 __ verify_oop(op->obj()->as_register());
2491 }
2492
2493
2494 void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
2495 assert(op->tmp1()->as_register() == G1 &&
2496 op->tmp2()->as_register() == G3 &&
2497 op->tmp3()->as_register() == G4 &&
2498 op->tmp4()->as_register() == O1 &&
2499 op->klass()->as_register() == G5, "must be");
2500
2501 LP64_ONLY( __ signx(op->len()->as_register()); )
2502 if (UseSlowPath ||
2503 (!UseFastNewObjectArray && (op->type() == T_OBJECT || op->type() == T_ARRAY)) ||
2504 (!UseFastNewTypeArray && (op->type() != T_OBJECT && op->type() != T_ARRAY))) {
2505 __ br(Assembler::always, false, Assembler::pt, *op->stub()->entry());
2506 __ delayed()->nop();
2507 } else {
2508 __ allocate_array(op->obj()->as_register(),
2509 op->len()->as_register(),
2510 op->tmp1()->as_register(),
2511 op->tmp2()->as_register(),
2512 op->tmp3()->as_register(),
2513 arrayOopDesc::header_size(op->type()),
2514 type2aelembytes(op->type()),
2515 op->klass()->as_register(),
2516 *op->stub()->entry());
2517 }
2518 __ bind(*op->stub()->continuation());
2519 }
2520
2521
2522 void LIR_Assembler::type_profile_helper(Register mdo, int mdo_offset_bias,
2523 ciMethodData *md, ciProfileData *data,
2524 Register recv, Register tmp1, Label* update_done) {
2525 uint i;
2526 for (i = 0; i < VirtualCallData::row_limit(); i++) {
2527 Label next_test;
2528 // See if the receiver is receiver[n].
2529 Address receiver_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)) -
2530 mdo_offset_bias);
2531 __ ld_ptr(receiver_addr, tmp1);
2532 __ verify_oop(tmp1);
2533 __ cmp(recv, tmp1);
2534 __ brx(Assembler::notEqual, false, Assembler::pt, next_test);
2535 __ delayed()->nop();
2536 Address data_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)) -
2537 mdo_offset_bias);
2538 __ ld_ptr(data_addr, tmp1);
2539 __ add(tmp1, DataLayout::counter_increment, tmp1);
2540 __ st_ptr(tmp1, data_addr);
2541 __ ba(false, *update_done);
2542 __ delayed()->nop();
2543 __ bind(next_test);
2544 }
2545
2546 // Didn't find receiver; find next empty slot and fill it in
2547 for (i = 0; i < VirtualCallData::row_limit(); i++) {
2548 Label next_test;
2549 Address recv_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)) -
2550 mdo_offset_bias);
2551 __ ld_ptr(recv_addr, tmp1);
2552 __ br_notnull(tmp1, false, Assembler::pt, next_test);
2553 __ delayed()->nop();
2554 __ st_ptr(recv, recv_addr);
2555 __ set(DataLayout::counter_increment, tmp1);
2556 __ st_ptr(tmp1, mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)) -
2557 mdo_offset_bias);
2558 __ ba(false, *update_done);
2559 __ delayed()->nop();
2560 __ bind(next_test);
2561 }
2562 }
2563
2564
2565 void LIR_Assembler::setup_md_access(ciMethod* method, int bci,
2566 ciMethodData*& md, ciProfileData*& data, int& mdo_offset_bias) {
2567 md = method->method_data_or_null();
2568 assert(md != NULL, "Sanity");
2569 data = md->bci_to_data(bci);
2570 assert(data != NULL, "need data for checkcast");
2571 assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
2572 if (!Assembler::is_simm13(md->byte_offset_of_slot(data, DataLayout::header_offset()) + data->size_in_bytes())) {
2573 // The offset is large so bias the mdo by the base of the slot so
2574 // that the ld can use simm13s to reference the slots of the data
2575 mdo_offset_bias = md->byte_offset_of_slot(data, DataLayout::header_offset());
2576 }
2577 }
2578
2579 void LIR_Assembler::emit_typecheck_helper(LIR_OpTypeCheck *op, Label* success, Label* failure, Label* obj_is_null) {
2580 // we always need a stub for the failure case.
2581 CodeStub* stub = op->stub();
2582 Register obj = op->object()->as_register();
2583 Register k_RInfo = op->tmp1()->as_register();
2584 Register klass_RInfo = op->tmp2()->as_register();
2585 Register dst = op->result_opr()->as_register();
2586 Register Rtmp1 = op->tmp3()->as_register();
2587 ciKlass* k = op->klass();
2588
2589
2590 if (obj == k_RInfo) {
2591 k_RInfo = klass_RInfo;
2592 klass_RInfo = obj;
2593 }
2594
2595 ciMethodData* md;
2596 ciProfileData* data;
2597 int mdo_offset_bias = 0;
2598 if (op->should_profile()) {
2599 ciMethod* method = op->profiled_method();
2600 assert(method != NULL, "Should have method");
2601 setup_md_access(method, op->profiled_bci(), md, data, mdo_offset_bias);
2602
2603 Label not_null;
2604 __ br_notnull(obj, false, Assembler::pn, not_null);
2605 __ delayed()->nop();
2606 Register mdo = k_RInfo;
2607 Register data_val = Rtmp1;
2608 jobject2reg(md->constant_encoding(), mdo);
2609 if (mdo_offset_bias > 0) {
2610 __ set(mdo_offset_bias, data_val);
2611 __ add(mdo, data_val, mdo);
2612 }
2613 Address flags_addr(mdo, md->byte_offset_of_slot(data, DataLayout::flags_offset()) - mdo_offset_bias);
2614 __ ldub(flags_addr, data_val);
2615 __ or3(data_val, BitData::null_seen_byte_constant(), data_val);
2616 __ stb(data_val, flags_addr);
2617 __ ba(false, *obj_is_null);
2618 __ delayed()->nop();
2619 __ bind(not_null);
2620 } else {
2621 __ br_null(obj, false, Assembler::pn, *obj_is_null);
2622 __ delayed()->nop();
2623 }
2624
2625 Label profile_cast_failure, profile_cast_success;
2626 Label *failure_target = op->should_profile() ? &profile_cast_failure : failure;
2627 Label *success_target = op->should_profile() ? &profile_cast_success : success;
2628
2629 // patching may screw with our temporaries on sparc,
2630 // so let's do it before loading the class
2631 if (k->is_loaded()) {
2632 jobject2reg(k->constant_encoding(), k_RInfo);
2633 } else {
2634 jobject2reg_with_patching(k_RInfo, op->info_for_patch());
2635 }
2636 assert(obj != k_RInfo, "must be different");
2637
2638 // get object class
2639 // not a safepoint as obj null check happens earlier
2640 __ load_klass(obj, klass_RInfo);
2641 if (op->fast_check()) {
2642 assert_different_registers(klass_RInfo, k_RInfo);
2643 __ cmp(k_RInfo, klass_RInfo);
2644 __ brx(Assembler::notEqual, false, Assembler::pt, *failure_target);
2645 __ delayed()->nop();
2646 } else {
2647 bool need_slow_path = true;
2648 if (k->is_loaded()) {
2649 if (k->super_check_offset() != sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes())
2650 need_slow_path = false;
2651 // perform the fast part of the checking logic
2652 __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, noreg,
2653 (need_slow_path ? success_target : NULL),
2654 failure_target, NULL,
2655 RegisterOrConstant(k->super_check_offset()));
2656 } else {
2657 // perform the fast part of the checking logic
2658 __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, O7, success_target,
2659 failure_target, NULL);
2660 }
2661 if (need_slow_path) {
2662 // call out-of-line instance of __ check_klass_subtype_slow_path(...):
2663 assert(klass_RInfo == G3 && k_RInfo == G1, "incorrect call setup");
2664 __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type);
2665 __ delayed()->nop();
2666 __ cmp(G3, 0);
2667 __ br(Assembler::equal, false, Assembler::pn, *failure_target);
2668 __ delayed()->nop();
2669 // Fall through to success case
2670 }
2671 }
2672
2673 if (op->should_profile()) {
2674 Register mdo = klass_RInfo, recv = k_RInfo, tmp1 = Rtmp1;
2675 assert_different_registers(obj, mdo, recv, tmp1);
2676 __ bind(profile_cast_success);
2677 jobject2reg(md->constant_encoding(), mdo);
2678 if (mdo_offset_bias > 0) {
2679 __ set(mdo_offset_bias, tmp1);
2680 __ add(mdo, tmp1, mdo);
2681 }
2682 __ load_klass(obj, recv);
2683 type_profile_helper(mdo, mdo_offset_bias, md, data, recv, tmp1, success);
2684 // Jump over the failure case
2685 __ ba(false, *success);
2686 __ delayed()->nop();
2687 // Cast failure case
2688 __ bind(profile_cast_failure);
2689 jobject2reg(md->constant_encoding(), mdo);
2690 if (mdo_offset_bias > 0) {
2691 __ set(mdo_offset_bias, tmp1);
2692 __ add(mdo, tmp1, mdo);
2693 }
2694 Address data_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()) - mdo_offset_bias);
2695 __ ld_ptr(data_addr, tmp1);
2696 __ sub(tmp1, DataLayout::counter_increment, tmp1);
2697 __ st_ptr(tmp1, data_addr);
2698 __ ba(false, *failure);
2699 __ delayed()->nop();
2700 }
2701 __ ba(false, *success);
2702 __ delayed()->nop();
2703 }
2704
2705 void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
2706 LIR_Code code = op->code();
2707 if (code == lir_store_check) {
2708 Register value = op->object()->as_register();
2709 Register array = op->array()->as_register();
2710 Register k_RInfo = op->tmp1()->as_register();
2711 Register klass_RInfo = op->tmp2()->as_register();
2712 Register Rtmp1 = op->tmp3()->as_register();
2713
2714 __ verify_oop(value);
2715 CodeStub* stub = op->stub();
2716 // check if it needs to be profiled
2717 ciMethodData* md;
2718 ciProfileData* data;
2719 int mdo_offset_bias = 0;
2720 if (op->should_profile()) {
2721 ciMethod* method = op->profiled_method();
2722 assert(method != NULL, "Should have method");
2723 setup_md_access(method, op->profiled_bci(), md, data, mdo_offset_bias);
2724 }
2725 Label profile_cast_success, profile_cast_failure, done;
2726 Label *success_target = op->should_profile() ? &profile_cast_success : &done;
2727 Label *failure_target = op->should_profile() ? &profile_cast_failure : stub->entry();
2728
2729 if (op->should_profile()) {
2730 Label not_null;
2731 __ br_notnull(value, false, Assembler::pn, not_null);
2732 __ delayed()->nop();
2733 Register mdo = k_RInfo;
2734 Register data_val = Rtmp1;
2735 jobject2reg(md->constant_encoding(), mdo);
2736 if (mdo_offset_bias > 0) {
2737 __ set(mdo_offset_bias, data_val);
2738 __ add(mdo, data_val, mdo);
2739 }
2740 Address flags_addr(mdo, md->byte_offset_of_slot(data, DataLayout::flags_offset()) - mdo_offset_bias);
2741 __ ldub(flags_addr, data_val);
2742 __ or3(data_val, BitData::null_seen_byte_constant(), data_val);
2743 __ stb(data_val, flags_addr);
2744 __ ba(false, done);
2745 __ delayed()->nop();
2746 __ bind(not_null);
2747 } else {
2748 __ br_null(value, false, Assembler::pn, done);
2749 __ delayed()->nop();
2750 }
2751 add_debug_info_for_null_check_here(op->info_for_exception());
2752 __ load_klass(array, k_RInfo);
2753 __ load_klass(value, klass_RInfo);
2754
2755 // get instance klass
2756 __ ld_ptr(Address(k_RInfo, objArrayKlass::element_klass_offset_in_bytes() + sizeof(oopDesc)), k_RInfo);
2757 // perform the fast part of the checking logic
2758 __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, O7, success_target, failure_target, NULL);
2759
2760 // call out-of-line instance of __ check_klass_subtype_slow_path(...):
2761 assert(klass_RInfo == G3 && k_RInfo == G1, "incorrect call setup");
2762 __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type);
2763 __ delayed()->nop();
2764 __ cmp(G3, 0);
2765 __ br(Assembler::equal, false, Assembler::pn, *failure_target);
2766 __ delayed()->nop();
2767 // fall through to the success case
2768
2769 if (op->should_profile()) {
2770 Register mdo = klass_RInfo, recv = k_RInfo, tmp1 = Rtmp1;
2771 assert_different_registers(value, mdo, recv, tmp1);
2772 __ bind(profile_cast_success);
2773 jobject2reg(md->constant_encoding(), mdo);
2774 if (mdo_offset_bias > 0) {
2775 __ set(mdo_offset_bias, tmp1);
2776 __ add(mdo, tmp1, mdo);
2777 }
2778 __ load_klass(value, recv);
2779 type_profile_helper(mdo, mdo_offset_bias, md, data, recv, tmp1, &done);
2780 __ ba(false, done);
2781 __ delayed()->nop();
2782 // Cast failure case
2783 __ bind(profile_cast_failure);
2784 jobject2reg(md->constant_encoding(), mdo);
2785 if (mdo_offset_bias > 0) {
2786 __ set(mdo_offset_bias, tmp1);
2787 __ add(mdo, tmp1, mdo);
2788 }
2789 Address data_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()) - mdo_offset_bias);
2790 __ ld_ptr(data_addr, tmp1);
2791 __ sub(tmp1, DataLayout::counter_increment, tmp1);
2792 __ st_ptr(tmp1, data_addr);
2793 __ ba(false, *stub->entry());
2794 __ delayed()->nop();
2795 }
2796 __ bind(done);
2797 } else if (code == lir_checkcast) {
2798 Register obj = op->object()->as_register();
2799 Register dst = op->result_opr()->as_register();
2800 Label success;
2801 emit_typecheck_helper(op, &success, op->stub()->entry(), &success);
2802 __ bind(success);
2803 __ mov(obj, dst);
2804 } else if (code == lir_instanceof) {
2805 Register obj = op->object()->as_register();
2806 Register dst = op->result_opr()->as_register();
2807 Label success, failure, done;
2808 emit_typecheck_helper(op, &success, &failure, &failure);
2809 __ bind(failure);
2810 __ set(0, dst);
2811 __ ba(false, done);
2812 __ delayed()->nop();
2813 __ bind(success);
2814 __ set(1, dst);
2815 __ bind(done);
2816 } else {
2817 ShouldNotReachHere();
2818 }
2819
2820 }
2821
2822
2823 void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
2824 if (op->code() == lir_cas_long) {
2825 assert(VM_Version::supports_cx8(), "wrong machine");
2826 Register addr = op->addr()->as_pointer_register();
2827 Register cmp_value_lo = op->cmp_value()->as_register_lo();
2828 Register cmp_value_hi = op->cmp_value()->as_register_hi();
2829 Register new_value_lo = op->new_value()->as_register_lo();
2830 Register new_value_hi = op->new_value()->as_register_hi();
2831 Register t1 = op->tmp1()->as_register();
2832 Register t2 = op->tmp2()->as_register();
2833 #ifdef _LP64
2834 __ mov(cmp_value_lo, t1);
2835 __ mov(new_value_lo, t2);
2836 // perform the compare and swap operation
2837 __ casx(addr, t1, t2);
2838 // generate condition code - if the swap succeeded, t2 ("new value" reg) was
2839 // overwritten with the original value in "addr" and will be equal to t1.
2840 __ cmp(t1, t2);
2841 #else
2842 // move high and low halves of long values into single registers
2843 __ sllx(cmp_value_hi, 32, t1); // shift high half into temp reg
2844 __ srl(cmp_value_lo, 0, cmp_value_lo); // clear upper 32 bits of low half
2845 __ or3(t1, cmp_value_lo, t1); // t1 holds 64-bit compare value
2846 __ sllx(new_value_hi, 32, t2);
2847 __ srl(new_value_lo, 0, new_value_lo);
2848 __ or3(t2, new_value_lo, t2); // t2 holds 64-bit value to swap
2849 // perform the compare and swap operation
2850 __ casx(addr, t1, t2);
2851 // generate condition code - if the swap succeeded, t2 ("new value" reg) was
2852 // overwritten with the original value in "addr" and will be equal to t1.
2853 // Produce icc flag for 32bit.
2854 __ sub(t1, t2, t2);
2855 __ srlx(t2, 32, t1);
2856 __ orcc(t2, t1, G0);
2857 #endif
2858 } else if (op->code() == lir_cas_int || op->code() == lir_cas_obj) {
2859 Register addr = op->addr()->as_pointer_register();
2860 Register cmp_value = op->cmp_value()->as_register();
2861 Register new_value = op->new_value()->as_register();
2862 Register t1 = op->tmp1()->as_register();
2863 Register t2 = op->tmp2()->as_register();
2864 __ mov(cmp_value, t1);
2865 __ mov(new_value, t2);
2866 if (op->code() == lir_cas_obj) {
2867 if (UseCompressedOops) {
2868 __ encode_heap_oop(t1);
2869 __ encode_heap_oop(t2);
2870 __ cas(addr, t1, t2);
2871 } else {
2872 __ cas_ptr(addr, t1, t2);
2873 }
2874 } else {
2875 __ cas(addr, t1, t2);
2876 }
2877 __ cmp(t1, t2);
2878 } else {
2879 Unimplemented();
2880 }
2881 }
2882
2883 void LIR_Assembler::set_24bit_FPU() {
2884 Unimplemented();
2885 }
2886
2887
2888 void LIR_Assembler::reset_FPU() {
2889 Unimplemented();
2890 }
2891
2892
2893 void LIR_Assembler::breakpoint() {
2894 __ breakpoint_trap();
2895 }
2896
2897
2898 void LIR_Assembler::push(LIR_Opr opr) {
2899 Unimplemented();
2900 }
2901
2902
2903 void LIR_Assembler::pop(LIR_Opr opr) {
2904 Unimplemented();
2905 }
2906
2907
2908 void LIR_Assembler::monitor_address(int monitor_no, LIR_Opr dst_opr) {
2909 Address mon_addr = frame_map()->address_for_monitor_lock(monitor_no);
2910 Register dst = dst_opr->as_register();
2911 Register reg = mon_addr.base();
2912 int offset = mon_addr.disp();
2913 // compute pointer to BasicLock
2914 if (mon_addr.is_simm13()) {
2915 __ add(reg, offset, dst);
2916 } else {
2917 __ set(offset, dst);
2918 __ add(dst, reg, dst);
2919 }
2920 }
2921
2922
2923 void LIR_Assembler::emit_lock(LIR_OpLock* op) {
2924 Register obj = op->obj_opr()->as_register();
2925 Register hdr = op->hdr_opr()->as_register();
2926 Register lock = op->lock_opr()->as_register();
2927
2928 // obj may not be an oop
2929 if (op->code() == lir_lock) {
2930 MonitorEnterStub* stub = (MonitorEnterStub*)op->stub();
2931 if (UseFastLocking) {
2932 assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
2933 // add debug info for NullPointerException only if one is possible
2934 if (op->info() != NULL) {
2935 add_debug_info_for_null_check_here(op->info());
2936 }
2937 __ lock_object(hdr, obj, lock, op->scratch_opr()->as_register(), *op->stub()->entry());
2938 } else {
2939 // always do slow locking
2940 // note: the slow locking code could be inlined here, however if we use
2941 // slow locking, speed doesn't matter anyway and this solution is
2942 // simpler and requires less duplicated code - additionally, the
2943 // slow locking code is the same in either case which simplifies
2944 // debugging
2945 __ br(Assembler::always, false, Assembler::pt, *op->stub()->entry());
2946 __ delayed()->nop();
2947 }
2948 } else {
2949 assert (op->code() == lir_unlock, "Invalid code, expected lir_unlock");
2950 if (UseFastLocking) {
2951 assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
2952 __ unlock_object(hdr, obj, lock, *op->stub()->entry());
2953 } else {
2954 // always do slow unlocking
2955 // note: the slow unlocking code could be inlined here, however if we use
2956 // slow unlocking, speed doesn't matter anyway and this solution is
2957 // simpler and requires less duplicated code - additionally, the
2958 // slow unlocking code is the same in either case which simplifies
2959 // debugging
2960 __ br(Assembler::always, false, Assembler::pt, *op->stub()->entry());
2961 __ delayed()->nop();
2962 }
2963 }
2964 __ bind(*op->stub()->continuation());
2965 }
2966
2967
2968 void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
2969 ciMethod* method = op->profiled_method();
2970 int bci = op->profiled_bci();
2971
2972 // Update counter for all call types
2973 ciMethodData* md = method->method_data_or_null();
2974 assert(md != NULL, "Sanity");
2975 ciProfileData* data = md->bci_to_data(bci);
2976 assert(data->is_CounterData(), "need CounterData for calls");
2977 assert(op->mdo()->is_single_cpu(), "mdo must be allocated");
2978 Register mdo = op->mdo()->as_register();
2979 #ifdef _LP64
2980 assert(op->tmp1()->is_double_cpu(), "tmp1 must be allocated");
2981 Register tmp1 = op->tmp1()->as_register_lo();
2982 #else
2983 assert(op->tmp1()->is_single_cpu(), "tmp1 must be allocated");
2984 Register tmp1 = op->tmp1()->as_register();
2985 #endif
2986 jobject2reg(md->constant_encoding(), mdo);
2987 int mdo_offset_bias = 0;
2988 if (!Assembler::is_simm13(md->byte_offset_of_slot(data, CounterData::count_offset()) +
2989 data->size_in_bytes())) {
2990 // The offset is large so bias the mdo by the base of the slot so
2991 // that the ld can use simm13s to reference the slots of the data
2992 mdo_offset_bias = md->byte_offset_of_slot(data, CounterData::count_offset());
2993 __ set(mdo_offset_bias, O7);
2994 __ add(mdo, O7, mdo);
2995 }
2996
2997 Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()) - mdo_offset_bias);
2998 Bytecodes::Code bc = method->java_code_at_bci(bci);
2999 // Perform additional virtual call profiling for invokevirtual and
3000 // invokeinterface bytecodes
3001 if ((bc == Bytecodes::_invokevirtual || bc == Bytecodes::_invokeinterface) &&
3002 C1ProfileVirtualCalls) {
3003 assert(op->recv()->is_single_cpu(), "recv must be allocated");
3004 Register recv = op->recv()->as_register();
3005 assert_different_registers(mdo, tmp1, recv);
3006 assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
3007 ciKlass* known_klass = op->known_holder();
3008 if (C1OptimizeVirtualCallProfiling && known_klass != NULL) {
3009 // We know the type that will be seen at this call site; we can
3010 // statically update the methodDataOop rather than needing to do
3011 // dynamic tests on the receiver type
3012
3013 // NOTE: we should probably put a lock around this search to
3014 // avoid collisions by concurrent compilations
3015 ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
3016 uint i;
3017 for (i = 0; i < VirtualCallData::row_limit(); i++) {
3018 ciKlass* receiver = vc_data->receiver(i);
3019 if (known_klass->equals(receiver)) {
3020 Address data_addr(mdo, md->byte_offset_of_slot(data,
3021 VirtualCallData::receiver_count_offset(i)) -
3022 mdo_offset_bias);
3023 __ ld_ptr(data_addr, tmp1);
3024 __ add(tmp1, DataLayout::counter_increment, tmp1);
3025 __ st_ptr(tmp1, data_addr);
3026 return;
3027 }
3028 }
3029
3030 // Receiver type not found in profile data; select an empty slot
3031
3032 // Note that this is less efficient than it should be because it
3033 // always does a write to the receiver part of the
3034 // VirtualCallData rather than just the first time
3035 for (i = 0; i < VirtualCallData::row_limit(); i++) {
3036 ciKlass* receiver = vc_data->receiver(i);
3037 if (receiver == NULL) {
3038 Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)) -
3039 mdo_offset_bias);
3040 jobject2reg(known_klass->constant_encoding(), tmp1);
3041 __ st_ptr(tmp1, recv_addr);
3042 Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)) -
3043 mdo_offset_bias);
3044 __ ld_ptr(data_addr, tmp1);
3045 __ add(tmp1, DataLayout::counter_increment, tmp1);
3046 __ st_ptr(tmp1, data_addr);
3047 return;
3048 }
3049 }
3050 } else {
3051 __ load_klass(recv, recv);
3052 Label update_done;
3053 type_profile_helper(mdo, mdo_offset_bias, md, data, recv, tmp1, &update_done);
3054 // Receiver did not match any saved receiver and there is no empty row for it.
3055 // Increment total counter to indicate polymorphic case.
3056 __ ld_ptr(counter_addr, tmp1);
3057 __ add(tmp1, DataLayout::counter_increment, tmp1);
3058 __ st_ptr(tmp1, counter_addr);
3059
3060 __ bind(update_done);
3061 }
3062 } else {
3063 // Static call
3064 __ ld_ptr(counter_addr, tmp1);
3065 __ add(tmp1, DataLayout::counter_increment, tmp1);
3066 __ st_ptr(tmp1, counter_addr);
3067 }
3068 }
3069
3070 void LIR_Assembler::align_backward_branch_target() {
3071 __ align(OptoLoopAlignment);
3072 }
3073
3074
3075 void LIR_Assembler::emit_delay(LIR_OpDelay* op) {
3076 // make sure we are expecting a delay
3077 // this has the side effect of clearing the delay state
3078 // so we can use _masm instead of _masm->delayed() to do the
3079 // code generation.
3080 __ delayed();
3081
3082 // make sure we only emit one instruction
3083 int offset = code_offset();
3084 op->delay_op()->emit_code(this);
3085 #ifdef ASSERT
3086 if (code_offset() - offset != NativeInstruction::nop_instruction_size) {
3087 op->delay_op()->print();
3088 }
3089 assert(code_offset() - offset == NativeInstruction::nop_instruction_size,
3090 "only one instruction can go in a delay slot");
3091 #endif
3092
3093 // we may also be emitting the call info for the instruction
3094 // which we are the delay slot of.
3095 CodeEmitInfo* call_info = op->call_info();
3096 if (call_info) {
3097 add_call_info(code_offset(), call_info);
3098 }
3099
3100 if (VerifyStackAtCalls) {
3101 _masm->sub(FP, SP, O7);
3102 _masm->cmp(O7, initial_frame_size_in_bytes());
3103 _masm->trap(Assembler::notEqual, Assembler::ptr_cc, G0, ST_RESERVED_FOR_USER_0+2 );
3104 }
3105 }
3106
3107
3108 void LIR_Assembler::negate(LIR_Opr left, LIR_Opr dest) {
3109 assert(left->is_register(), "can only handle registers");
3110
3111 if (left->is_single_cpu()) {
3112 __ neg(left->as_register(), dest->as_register());
3113 } else if (left->is_single_fpu()) {
3114 __ fneg(FloatRegisterImpl::S, left->as_float_reg(), dest->as_float_reg());
3115 } else if (left->is_double_fpu()) {
3116 __ fneg(FloatRegisterImpl::D, left->as_double_reg(), dest->as_double_reg());
3117 } else {
3118 assert (left->is_double_cpu(), "Must be a long");
3119 Register Rlow = left->as_register_lo();
3120 Register Rhi = left->as_register_hi();
3121 #ifdef _LP64
3122 __ sub(G0, Rlow, dest->as_register_lo());
3123 #else
3124 __ subcc(G0, Rlow, dest->as_register_lo());
3125 __ subc (G0, Rhi, dest->as_register_hi());
3126 #endif
3127 }
3128 }
3129
3130
3131 void LIR_Assembler::fxch(int i) {
3132 Unimplemented();
3133 }
3134
3135 void LIR_Assembler::fld(int i) {
3136 Unimplemented();
3137 }
3138
3139 void LIR_Assembler::ffree(int i) {
3140 Unimplemented();
3141 }
3142
3143 void LIR_Assembler::rt_call(LIR_Opr result, address dest,
3144 const LIR_OprList* args, LIR_Opr tmp, CodeEmitInfo* info) {
3145
3146 // if tmp is invalid, then the function being called doesn't destroy the thread
3147 if (tmp->is_valid()) {
3148 __ save_thread(tmp->as_register());
3149 }
3150 __ call(dest, relocInfo::runtime_call_type);
3151 __ delayed()->nop();
3152 if (info != NULL) {
3153 add_call_info_here(info);
3154 }
3155 if (tmp->is_valid()) {
3156 __ restore_thread(tmp->as_register());
3157 }
3158
3159 #ifdef ASSERT
3160 __ verify_thread();
3161 #endif // ASSERT
3162 }
3163
3164
3165 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {
3166 #ifdef _LP64
3167 ShouldNotReachHere();
3168 #endif
3169
3170 NEEDS_CLEANUP;
3171 if (type == T_LONG) {
3172 LIR_Address* mem_addr = dest->is_address() ? dest->as_address_ptr() : src->as_address_ptr();
3173
3174 // (extended to allow indexed as well as constant displaced for JSR-166)
3175 Register idx = noreg; // contains either constant offset or index
3176
3177 int disp = mem_addr->disp();
3178 if (mem_addr->index() == LIR_OprFact::illegalOpr) {
3179 if (!Assembler::is_simm13(disp)) {
3180 idx = O7;
3181 __ set(disp, idx);
3182 }
3183 } else {
3184 assert(disp == 0, "not both indexed and disp");
3185 idx = mem_addr->index()->as_register();
3186 }
3187
3188 int null_check_offset = -1;
3189
3190 Register base = mem_addr->base()->as_register();
3191 if (src->is_register() && dest->is_address()) {
3192 // G4 is high half, G5 is low half
3193 if (VM_Version::v9_instructions_work()) {
3194 // clear the top bits of G5, and scale up G4
3195 __ srl (src->as_register_lo(), 0, G5);
3196 __ sllx(src->as_register_hi(), 32, G4);
3197 // combine the two halves into the 64 bits of G4
3198 __ or3(G4, G5, G4);
3199 null_check_offset = __ offset();
3200 if (idx == noreg) {
3201 __ stx(G4, base, disp);
3202 } else {
3203 __ stx(G4, base, idx);
3204 }
3205 } else {
3206 __ mov (src->as_register_hi(), G4);
3207 __ mov (src->as_register_lo(), G5);
3208 null_check_offset = __ offset();
3209 if (idx == noreg) {
3210 __ std(G4, base, disp);
3211 } else {
3212 __ std(G4, base, idx);
3213 }
3214 }
3215 } else if (src->is_address() && dest->is_register()) {
3216 null_check_offset = __ offset();
3217 if (VM_Version::v9_instructions_work()) {
3218 if (idx == noreg) {
3219 __ ldx(base, disp, G5);
3220 } else {
3221 __ ldx(base, idx, G5);
3222 }
3223 __ srax(G5, 32, dest->as_register_hi()); // fetch the high half into hi
3224 __ mov (G5, dest->as_register_lo()); // copy low half into lo
3225 } else {
3226 if (idx == noreg) {
3227 __ ldd(base, disp, G4);
3228 } else {
3229 __ ldd(base, idx, G4);
3230 }
3231 // G4 is high half, G5 is low half
3232 __ mov (G4, dest->as_register_hi());
3233 __ mov (G5, dest->as_register_lo());
3234 }
3235 } else {
3236 Unimplemented();
3237 }
3238 if (info != NULL) {
3239 add_debug_info_for_null_check(null_check_offset, info);
3240 }
3241
3242 } else {
3243 // use normal move for all other volatiles since they don't need
3244 // special handling to remain atomic.
3245 move_op(src, dest, type, lir_patch_none, info, false, false, false);
3246 }
3247 }
3248
3249 void LIR_Assembler::membar() {
3250 // only StoreLoad membars are ever explicitly needed on sparcs in TSO mode
3251 __ membar( Assembler::Membar_mask_bits(Assembler::StoreLoad) );
3252 }
3253
3254 void LIR_Assembler::membar_acquire() {
3255 // no-op on TSO
3256 }
3257
3258 void LIR_Assembler::membar_release() {
3259 // no-op on TSO
3260 }
3261
3262 // Pack two sequential registers containing 32 bit values
3263 // into a single 64 bit register.
3264 // src and src->successor() are packed into dst
3265 // src and dst may be the same register.
3266 // Note: src is destroyed
3267 void LIR_Assembler::pack64(LIR_Opr src, LIR_Opr dst) {
3268 Register rs = src->as_register();
3269 Register rd = dst->as_register_lo();
3270 __ sllx(rs, 32, rs);
3271 __ srl(rs->successor(), 0, rs->successor());
3272 __ or3(rs, rs->successor(), rd);
3273 }
3274
3275 // Unpack a 64 bit value in a register into
3276 // two sequential registers.
3277 // src is unpacked into dst and dst->successor()
3278 void LIR_Assembler::unpack64(LIR_Opr src, LIR_Opr dst) {
3279 Register rs = src->as_register_lo();
3280 Register rd = dst->as_register_hi();
3281 assert_different_registers(rs, rd, rd->successor());
3282 __ srlx(rs, 32, rd);
3283 __ srl (rs, 0, rd->successor());
3284 }
3285
3286
3287 void LIR_Assembler::leal(LIR_Opr addr_opr, LIR_Opr dest) {
3288 LIR_Address* addr = addr_opr->as_address_ptr();
3289 assert(addr->index()->is_illegal() && addr->scale() == LIR_Address::times_1 && Assembler::is_simm13(addr->disp()), "can't handle complex addresses yet");
3290
3291 __ add(addr->base()->as_pointer_register(), addr->disp(), dest->as_pointer_register());
3292 }
3293
3294
3295 void LIR_Assembler::get_thread(LIR_Opr result_reg) {
3296 assert(result_reg->is_register(), "check");
3297 __ mov(G2_thread, result_reg->as_register());
3298 }
3299
3300
3301 void LIR_Assembler::peephole(LIR_List* lir) {
3302 LIR_OpList* inst = lir->instructions_list();
3303 for (int i = 0; i < inst->length(); i++) {
3304 LIR_Op* op = inst->at(i);
3305 switch (op->code()) {
3306 case lir_cond_float_branch:
3307 case lir_branch: {
3308 LIR_OpBranch* branch = op->as_OpBranch();
3309 assert(branch->info() == NULL, "shouldn't be state on branches anymore");
3310 LIR_Op* delay_op = NULL;
3311 // we'd like to be able to pull following instructions into
3312 // this slot but we don't know enough to do it safely yet so
3313 // only optimize block to block control flow.
3314 if (LIRFillDelaySlots && branch->block()) {
3315 LIR_Op* prev = inst->at(i - 1);
3316 if (prev && LIR_Assembler::is_single_instruction(prev) && prev->info() == NULL) {
3317 // swap previous instruction into delay slot
3318 inst->at_put(i - 1, op);
3319 inst->at_put(i, new LIR_OpDelay(prev, op->info()));
3320 #ifndef PRODUCT
3321 if (LIRTracePeephole) {
3322 tty->print_cr("delayed");
3323 inst->at(i - 1)->print();
3324 inst->at(i)->print();
3325 tty->cr();
3326 }
3327 #endif
3328 continue;
3329 }
3330 }
3331
3332 if (!delay_op) {
3333 delay_op = new LIR_OpDelay(new LIR_Op0(lir_nop), NULL);
3334 }
3335 inst->insert_before(i + 1, delay_op);
3336 break;
3337 }
3338 case lir_static_call:
3339 case lir_virtual_call:
3340 case lir_icvirtual_call:
3341 case lir_optvirtual_call:
3342 case lir_dynamic_call: {
3343 LIR_Op* prev = inst->at(i - 1);
3344 if (LIRFillDelaySlots && prev && prev->code() == lir_move && prev->info() == NULL &&
3345 (op->code() != lir_virtual_call ||
3346 !prev->result_opr()->is_single_cpu() ||
3347 prev->result_opr()->as_register() != O0) &&
3348 LIR_Assembler::is_single_instruction(prev)) {
3349 // Only moves without info can be put into the delay slot.
3350 // Also don't allow the setup of the receiver in the delay
3351 // slot for vtable calls.
3352 inst->at_put(i - 1, op);
3353 inst->at_put(i, new LIR_OpDelay(prev, op->info()));
3354 #ifndef PRODUCT
3355 if (LIRTracePeephole) {
3356 tty->print_cr("delayed");
3357 inst->at(i - 1)->print();
3358 inst->at(i)->print();
3359 tty->cr();
3360 }
3361 #endif
3362 } else {
3363 LIR_Op* delay_op = new LIR_OpDelay(new LIR_Op0(lir_nop), op->as_OpJavaCall()->info());
3364 inst->insert_before(i + 1, delay_op);
3365 i++;
3366 }
3367
3368 #if defined(TIERED) && !defined(_LP64)
3369 // fixup the return value from G1 to O0/O1 for long returns.
3370 // It's done here instead of in LIRGenerator because there's
3371 // such a mismatch between the single reg and double reg
3372 // calling convention.
3373 LIR_OpJavaCall* callop = op->as_OpJavaCall();
3374 if (callop->result_opr() == FrameMap::out_long_opr) {
3375 LIR_OpJavaCall* call;
3376 LIR_OprList* arguments = new LIR_OprList(callop->arguments()->length());
3377 for (int a = 0; a < arguments->length(); a++) {
3378 arguments[a] = callop->arguments()[a];
3379 }
3380 if (op->code() == lir_virtual_call) {
3381 call = new LIR_OpJavaCall(op->code(), callop->method(), callop->receiver(), FrameMap::g1_long_single_opr,
3382 callop->vtable_offset(), arguments, callop->info());
3383 } else {
3384 call = new LIR_OpJavaCall(op->code(), callop->method(), callop->receiver(), FrameMap::g1_long_single_opr,
3385 callop->addr(), arguments, callop->info());
3386 }
3387 inst->at_put(i - 1, call);
3388 inst->insert_before(i + 1, new LIR_Op1(lir_unpack64, FrameMap::g1_long_single_opr, callop->result_opr(),
3389 T_LONG, lir_patch_none, NULL));
3390 }
3391 #endif
3392 break;
3393 }
3394 }
3395 }
3396 }
3397
3398
3399
3400
3401 #undef __