1 /*
2 * Copyright (c) 2000, 2025, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2012, 2025 SAP SE. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "asm/macroAssembler.inline.hpp"
27 #include "c1/c1_Compilation.hpp"
28 #include "c1/c1_LIRAssembler.hpp"
29 #include "c1/c1_MacroAssembler.hpp"
30 #include "c1/c1_Runtime1.hpp"
31 #include "c1/c1_ValueStack.hpp"
32 #include "ci/ciArrayKlass.hpp"
33 #include "ci/ciInstance.hpp"
34 #include "gc/shared/collectedHeap.hpp"
35 #include "memory/universe.hpp"
36 #include "nativeInst_ppc.hpp"
37 #include "oops/compressedOops.hpp"
38 #include "oops/objArrayKlass.hpp"
39 #include "runtime/frame.inline.hpp"
40 #include "runtime/os.inline.hpp"
41 #include "runtime/safepointMechanism.inline.hpp"
42 #include "runtime/sharedRuntime.hpp"
43 #include "runtime/stubRoutines.hpp"
44 #include "runtime/vm_version.hpp"
45 #include "utilities/macros.hpp"
46 #include "utilities/powerOfTwo.hpp"
47
48 #define __ _masm->
49
50
51 const ConditionRegister LIR_Assembler::BOOL_RESULT = CR5;
52
53
54 bool LIR_Assembler::is_small_constant(LIR_Opr opr) {
55 Unimplemented(); return false; // Currently not used on this platform.
56 }
57
58
59 LIR_Opr LIR_Assembler::receiverOpr() {
60 return FrameMap::R3_oop_opr;
61 }
62
63
64 LIR_Opr LIR_Assembler::osrBufferPointer() {
65 return FrameMap::R3_opr;
66 }
67
68
69 // This specifies the stack pointer decrement needed to build the frame.
70 int LIR_Assembler::initial_frame_size_in_bytes() const {
71 return in_bytes(frame_map()->framesize_in_bytes());
72 }
73
74
75 // Inline cache check: the inline cached class is in inline_cache_reg;
76 // we fetch the class of the receiver and compare it with the cached class.
77 // If they do not match we jump to slow case.
78 int LIR_Assembler::check_icache() {
79 return __ ic_check(CodeEntryAlignment);
80 }
81
82 void LIR_Assembler::clinit_barrier(ciMethod* method) {
83 assert(!method->holder()->is_not_initialized(), "initialization should have been started");
84
85 Label L_skip_barrier;
86 Register klass = R20;
87
88 metadata2reg(method->holder()->constant_encoding(), klass);
89 __ clinit_barrier(klass, R16_thread, &L_skip_barrier /*L_fast_path*/);
90
91 __ load_const_optimized(klass, SharedRuntime::get_handle_wrong_method_stub(), R0);
92 __ mtctr(klass);
93 __ bctr();
94
95 __ bind(L_skip_barrier);
96 }
97
98 void LIR_Assembler::osr_entry() {
99 // On-stack-replacement entry sequence:
100 //
101 // 1. Create a new compiled activation.
102 // 2. Initialize local variables in the compiled activation. The expression
103 // stack must be empty at the osr_bci; it is not initialized.
104 // 3. Jump to the continuation address in compiled code to resume execution.
105
106 // OSR entry point
107 offsets()->set_value(CodeOffsets::OSR_Entry, code_offset());
108 BlockBegin* osr_entry = compilation()->hir()->osr_entry();
109 ValueStack* entry_state = osr_entry->end()->state();
110 int number_of_locks = entry_state->locks_size();
111
112 // Create a frame for the compiled activation.
113 __ build_frame(initial_frame_size_in_bytes(), bang_size_in_bytes());
114
115 // OSR buffer is
116 //
117 // locals[nlocals-1..0]
118 // monitors[number_of_locks-1..0]
119 //
120 // Locals is a direct copy of the interpreter frame so in the osr buffer
121 // the first slot in the local array is the last local from the interpreter
122 // and the last slot is local[0] (receiver) from the interpreter.
123 //
124 // Similarly with locks. The first lock slot in the osr buffer is the nth lock
125 // from the interpreter frame, the nth lock slot in the osr buffer is 0th lock
126 // in the interpreter frame (the method lock if a sync method).
127
128 // Initialize monitors in the compiled activation.
129 // R3: pointer to osr buffer
130 //
131 // All other registers are dead at this point and the locals will be
132 // copied into place by code emitted in the IR.
133
134 Register OSR_buf = osrBufferPointer()->as_register();
135 {
136 assert(frame::interpreter_frame_monitor_size() == BasicObjectLock::size(), "adjust code below");
137
138 const int locals_space = BytesPerWord * method()->max_locals();
139 int monitor_offset = locals_space + (2 * BytesPerWord) * (number_of_locks - 1);
140 bool use_OSR_bias = false;
141
142 if (!Assembler::is_simm16(monitor_offset + BytesPerWord) && number_of_locks > 0) {
143 // Offsets too large for ld instructions. Use bias.
144 __ add_const_optimized(OSR_buf, OSR_buf, locals_space);
145 monitor_offset -= locals_space;
146 use_OSR_bias = true;
147 }
148
149 // SharedRuntime::OSR_migration_begin() packs BasicObjectLocks in
150 // the OSR buffer using 2 word entries: first the lock and then
151 // the oop.
152 for (int i = 0; i < number_of_locks; i++) {
153 int slot_offset = monitor_offset - ((i * 2) * BytesPerWord);
154 #ifdef ASSERT
155 // Verify the interpreter's monitor has a non-null object.
156 {
157 Label L;
158 __ ld(R0, slot_offset + 1*BytesPerWord, OSR_buf);
159 __ cmpdi(CR0, R0, 0);
160 __ bne(CR0, L);
161 __ stop("locked object is null");
162 __ bind(L);
163 }
164 #endif // ASSERT
165 // Copy the lock field into the compiled activation.
166 Address ml = frame_map()->address_for_monitor_lock(i),
167 mo = frame_map()->address_for_monitor_object(i);
168 assert(ml.index() == noreg && mo.index() == noreg, "sanity");
169 __ ld(R0, slot_offset + 0, OSR_buf);
170 __ std(R0, ml);
171 __ ld(R0, slot_offset + 1*BytesPerWord, OSR_buf);
172 __ std(R0, mo);
173 }
174
175 if (use_OSR_bias) {
176 // Restore.
177 __ sub_const_optimized(OSR_buf, OSR_buf, locals_space);
178 }
179 }
180 }
181
182
183 int LIR_Assembler::emit_exception_handler() {
184 // Generate code for the exception handler.
185 address handler_base = __ start_a_stub(exception_handler_size());
186
187 if (handler_base == nullptr) {
188 // Not enough space left for the handler.
189 bailout("exception handler overflow");
190 return -1;
191 }
192
193 int offset = code_offset();
194 address entry_point = CAST_FROM_FN_PTR(address, Runtime1::entry_for(StubId::c1_handle_exception_from_callee_id));
195 //__ load_const_optimized(R0, entry_point);
196 __ add_const_optimized(R0, R29_TOC, MacroAssembler::offset_to_global_toc(entry_point));
197 __ mtctr(R0);
198 __ bctr();
199
200 guarantee(code_offset() - offset <= exception_handler_size(), "overflow");
201 __ end_a_stub();
202
203 return offset;
204 }
205
206
207 // Emit the code to remove the frame from the stack in the exception
208 // unwind path.
209 int LIR_Assembler::emit_unwind_handler() {
210 _masm->block_comment("Unwind handler");
211
212 int offset = code_offset();
213 bool preserve_exception = method()->is_synchronized() || compilation()->env()->dtrace_method_probes();
214 const Register Rexception = R3 /*LIRGenerator::exceptionOopOpr()*/, Rexception_save = R31;
215
216 // Fetch the exception from TLS and clear out exception related thread state.
217 __ ld(Rexception, in_bytes(JavaThread::exception_oop_offset()), R16_thread);
218 __ li(R0, 0);
219 __ std(R0, in_bytes(JavaThread::exception_oop_offset()), R16_thread);
220 __ std(R0, in_bytes(JavaThread::exception_pc_offset()), R16_thread);
221
222 __ bind(_unwind_handler_entry);
223 __ verify_not_null_oop(Rexception);
224 if (preserve_exception) { __ mr(Rexception_save, Rexception); }
225
226 // Perform needed unlocking
227 MonitorExitStub* stub = nullptr;
228 if (method()->is_synchronized()) {
229 monitor_address(0, FrameMap::R4_opr);
230 stub = new MonitorExitStub(FrameMap::R4_opr, 0);
231 __ unlock_object(R5, R6, R4, *stub->entry());
232 __ bind(*stub->continuation());
233 }
234
235 if (compilation()->env()->dtrace_method_probes()) {
236 Unimplemented();
237 }
238
239 // Dispatch to the unwind logic.
240 address unwind_stub = Runtime1::entry_for(StubId::c1_unwind_exception_id);
241 //__ load_const_optimized(R0, unwind_stub);
242 __ add_const_optimized(R0, R29_TOC, MacroAssembler::offset_to_global_toc(unwind_stub));
243 if (preserve_exception) { __ mr(Rexception, Rexception_save); }
244 __ mtctr(R0);
245 __ bctr();
246
247 // Emit the slow path assembly.
248 if (stub != nullptr) {
249 stub->emit_code(this);
250 }
251
252 return offset;
253 }
254
255
256 int LIR_Assembler::emit_deopt_handler() {
257 // Generate code for deopt handler.
258 address handler_base = __ start_a_stub(deopt_handler_size());
259
260 if (handler_base == nullptr) {
261 // Not enough space left for the handler.
262 bailout("deopt handler overflow");
263 return -1;
264 }
265
266 int offset = code_offset();
267 Label start;
268
269 __ bind(start);
270 __ bl64_patchable(SharedRuntime::deopt_blob()->unpack(), relocInfo::runtime_call_type);
271 int entry_offset = __ offset();
272 __ b(start);
273
274 guarantee(code_offset() - offset <= deopt_handler_size(), "overflow");
275 __ end_a_stub();
276
277 return entry_offset;
278 }
279
280
281 void LIR_Assembler::jobject2reg(jobject o, Register reg) {
282 if (o == nullptr) {
283 __ li(reg, 0);
284 } else {
285 AddressLiteral addrlit = __ constant_oop_address(o);
286 __ load_const(reg, addrlit, (reg != R0) ? R0 : noreg);
287 }
288 }
289
290
291 void LIR_Assembler::jobject2reg_with_patching(Register reg, CodeEmitInfo *info) {
292 // Allocate a new index in table to hold the object once it's been patched.
293 int oop_index = __ oop_recorder()->allocate_oop_index(nullptr);
294 PatchingStub* patch = new PatchingStub(_masm, patching_id(info), oop_index);
295
296 AddressLiteral addrlit((address)nullptr, oop_Relocation::spec(oop_index));
297 __ load_const(reg, addrlit, R0);
298
299 patching_epilog(patch, lir_patch_normal, reg, info);
300 }
301
302
303 void LIR_Assembler::metadata2reg(Metadata* o, Register reg) {
304 AddressLiteral md = __ constant_metadata_address(o); // Notify OOP recorder (don't need the relocation)
305 __ load_const_optimized(reg, md.value(), (reg != R0) ? R0 : noreg);
306 }
307
308
309 void LIR_Assembler::klass2reg_with_patching(Register reg, CodeEmitInfo *info) {
310 // Allocate a new index in table to hold the klass once it's been patched.
311 int index = __ oop_recorder()->allocate_metadata_index(nullptr);
312 PatchingStub* patch = new PatchingStub(_masm, PatchingStub::load_klass_id, index);
313
314 AddressLiteral addrlit((address)nullptr, metadata_Relocation::spec(index));
315 assert(addrlit.rspec().type() == relocInfo::metadata_type, "must be an metadata reloc");
316 __ load_const(reg, addrlit, R0);
317
318 patching_epilog(patch, lir_patch_normal, reg, info);
319 }
320
321
322 void LIR_Assembler::arithmetic_idiv(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr temp, LIR_Opr result, CodeEmitInfo* info) {
323 const bool is_int = result->is_single_cpu();
324 Register Rdividend = is_int ? left->as_register() : left->as_register_lo();
325 Register Rdivisor = noreg;
326 Register Rscratch = temp->as_register();
327 Register Rresult = is_int ? result->as_register() : result->as_register_lo();
328 long divisor = -1;
329
330 if (right->is_register()) {
331 Rdivisor = is_int ? right->as_register() : right->as_register_lo();
332 } else {
333 divisor = is_int ? right->as_constant_ptr()->as_jint()
334 : right->as_constant_ptr()->as_jlong();
335 }
336
337 assert(Rdividend != Rscratch, "");
338 assert(Rdivisor != Rscratch, "");
339 assert(code == lir_idiv || code == lir_irem, "Must be irem or idiv");
340
341 if (Rdivisor == noreg) {
342 if (divisor == 1) { // stupid, but can happen
343 if (code == lir_idiv) {
344 __ mr_if_needed(Rresult, Rdividend);
345 } else {
346 __ li(Rresult, 0);
347 }
348
349 } else if (is_power_of_2(divisor)) {
350 // Convert division by a power of two into some shifts and logical operations.
351 int log2 = log2i_exact(divisor);
352
353 // Round towards 0.
354 if (divisor == 2) {
355 if (is_int) {
356 __ srwi(Rscratch, Rdividend, 31);
357 } else {
358 __ srdi(Rscratch, Rdividend, 63);
359 }
360 } else {
361 if (is_int) {
362 __ srawi(Rscratch, Rdividend, 31);
363 } else {
364 __ sradi(Rscratch, Rdividend, 63);
365 }
366 __ clrldi(Rscratch, Rscratch, 64-log2);
367 }
368 __ add(Rscratch, Rdividend, Rscratch);
369
370 if (code == lir_idiv) {
371 if (is_int) {
372 __ srawi(Rresult, Rscratch, log2);
373 } else {
374 __ sradi(Rresult, Rscratch, log2);
375 }
376 } else { // lir_irem
377 __ clrrdi(Rscratch, Rscratch, log2);
378 __ sub(Rresult, Rdividend, Rscratch);
379 }
380
381 } else if (divisor == -1) {
382 if (code == lir_idiv) {
383 __ neg(Rresult, Rdividend);
384 } else {
385 __ li(Rresult, 0);
386 }
387
388 } else {
389 __ load_const_optimized(Rscratch, divisor);
390 if (code == lir_idiv) {
391 if (is_int) {
392 __ divw(Rresult, Rdividend, Rscratch); // Can't divide minint/-1.
393 } else {
394 __ divd(Rresult, Rdividend, Rscratch); // Can't divide minint/-1.
395 }
396 } else {
397 assert(Rscratch != R0, "need both");
398 if (is_int) {
399 __ divw(R0, Rdividend, Rscratch); // Can't divide minint/-1.
400 __ mullw(Rscratch, R0, Rscratch);
401 } else {
402 __ divd(R0, Rdividend, Rscratch); // Can't divide minint/-1.
403 __ mulld(Rscratch, R0, Rscratch);
404 }
405 __ sub(Rresult, Rdividend, Rscratch);
406 }
407
408 }
409 return;
410 }
411
412 Label regular, done;
413 if (is_int) {
414 __ cmpwi(CR0, Rdivisor, -1);
415 } else {
416 __ cmpdi(CR0, Rdivisor, -1);
417 }
418 __ bne(CR0, regular);
419 if (code == lir_idiv) {
420 __ neg(Rresult, Rdividend);
421 __ b(done);
422 __ bind(regular);
423 if (is_int) {
424 __ divw(Rresult, Rdividend, Rdivisor); // Can't divide minint/-1.
425 } else {
426 __ divd(Rresult, Rdividend, Rdivisor); // Can't divide minint/-1.
427 }
428 } else { // lir_irem
429 __ li(Rresult, 0);
430 __ b(done);
431 __ bind(regular);
432 if (is_int) {
433 __ divw(Rscratch, Rdividend, Rdivisor); // Can't divide minint/-1.
434 __ mullw(Rscratch, Rscratch, Rdivisor);
435 } else {
436 __ divd(Rscratch, Rdividend, Rdivisor); // Can't divide minint/-1.
437 __ mulld(Rscratch, Rscratch, Rdivisor);
438 }
439 __ sub(Rresult, Rdividend, Rscratch);
440 }
441 __ bind(done);
442 }
443
444
445 void LIR_Assembler::emit_op3(LIR_Op3* op) {
446 switch (op->code()) {
447 case lir_idiv:
448 case lir_irem:
449 arithmetic_idiv(op->code(), op->in_opr1(), op->in_opr2(), op->in_opr3(),
450 op->result_opr(), op->info());
451 break;
452 case lir_fmad:
453 __ fmadd(op->result_opr()->as_double_reg(), op->in_opr1()->as_double_reg(),
454 op->in_opr2()->as_double_reg(), op->in_opr3()->as_double_reg());
455 break;
456 case lir_fmaf:
457 __ fmadds(op->result_opr()->as_float_reg(), op->in_opr1()->as_float_reg(),
458 op->in_opr2()->as_float_reg(), op->in_opr3()->as_float_reg());
459 break;
460 default: ShouldNotReachHere(); break;
461 }
462 }
463
464
465 void LIR_Assembler::emit_opBranch(LIR_OpBranch* op) {
466 #ifdef ASSERT
467 assert(op->block() == nullptr || op->block()->label() == op->label(), "wrong label");
468 if (op->block() != nullptr) _branch_target_blocks.append(op->block());
469 if (op->ublock() != nullptr) _branch_target_blocks.append(op->ublock());
470 assert(op->info() == nullptr, "shouldn't have CodeEmitInfo");
471 #endif
472
473 Label *L = op->label();
474 if (op->cond() == lir_cond_always) {
475 __ b(*L);
476 } else {
477 Label done;
478 bool is_unordered = false;
479 if (op->code() == lir_cond_float_branch) {
480 assert(op->ublock() != nullptr, "must have unordered successor");
481 is_unordered = true;
482 } else {
483 assert(op->code() == lir_branch, "just checking");
484 }
485
486 bool positive = false;
487 Assembler::Condition cond = Assembler::equal;
488 switch (op->cond()) {
489 case lir_cond_equal: positive = true ; cond = Assembler::equal ; is_unordered = false; break;
490 case lir_cond_notEqual: positive = false; cond = Assembler::equal ; is_unordered = false; break;
491 case lir_cond_less: positive = true ; cond = Assembler::less ; break;
492 case lir_cond_belowEqual: assert(op->code() != lir_cond_float_branch, ""); // fallthru
493 case lir_cond_lessEqual: positive = false; cond = Assembler::greater; break;
494 case lir_cond_greater: positive = true ; cond = Assembler::greater; break;
495 case lir_cond_aboveEqual: assert(op->code() != lir_cond_float_branch, ""); // fallthru
496 case lir_cond_greaterEqual: positive = false; cond = Assembler::less ; break;
497 default: ShouldNotReachHere();
498 }
499 int bo = positive ? Assembler::bcondCRbiIs1 : Assembler::bcondCRbiIs0;
500 int bi = Assembler::bi0(BOOL_RESULT, cond);
501 if (is_unordered) {
502 if (positive) {
503 if (op->ublock() == op->block()) {
504 __ bc_far_optimized(Assembler::bcondCRbiIs1, __ bi0(BOOL_RESULT, Assembler::summary_overflow), *L);
505 }
506 } else {
507 if (op->ublock() != op->block()) { __ bso(BOOL_RESULT, done); }
508 }
509 }
510 __ bc_far_optimized(bo, bi, *L);
511 __ bind(done);
512 }
513 }
514
515
516 void LIR_Assembler::emit_opConvert(LIR_OpConvert* op) {
517 Bytecodes::Code code = op->bytecode();
518 LIR_Opr src = op->in_opr(),
519 dst = op->result_opr();
520
521 switch(code) {
522 case Bytecodes::_i2l: {
523 __ extsw(dst->as_register_lo(), src->as_register());
524 break;
525 }
526 case Bytecodes::_l2i: {
527 __ mr_if_needed(dst->as_register(), src->as_register_lo()); // high bits are garbage
528 break;
529 }
530 case Bytecodes::_i2b: {
531 __ extsb(dst->as_register(), src->as_register());
532 break;
533 }
534 case Bytecodes::_i2c: {
535 __ clrldi(dst->as_register(), src->as_register(), 64-16);
536 break;
537 }
538 case Bytecodes::_i2s: {
539 __ extsh(dst->as_register(), src->as_register());
540 break;
541 }
542 case Bytecodes::_i2d:{
543 FloatRegister rdst = dst->as_double_reg();
544 // move src to dst register
545 __ mtfprwa(rdst, src->as_register());
546 __ fcfid(rdst, rdst);
547 break;
548 }
549 case Bytecodes::_l2d: {
550 FloatRegister rdst = dst->as_double_reg();
551 // move src to dst register
552 __ mtfprd(rdst, src->as_register_lo());
553 __ fcfid(rdst, rdst);
554 break;
555 }
556 case Bytecodes::_i2f:{
557 FloatRegister rdst = dst->as_float_reg();
558 // move src to dst register
559 __ mtfprwa(rdst, src->as_register());
560 __ fcfids(rdst, rdst);
561 break;
562 }
563 case Bytecodes::_l2f: {
564 FloatRegister rdst = dst->as_float_reg();
565 // move src to dst register
566 __ mtfprd(rdst, src->as_register_lo());
567 __ fcfids(rdst, rdst);
568 break;
569 }
570 case Bytecodes::_f2d: {
571 __ fmr_if_needed(dst->as_double_reg(), src->as_float_reg());
572 break;
573 }
574 case Bytecodes::_d2f: {
575 __ frsp(dst->as_float_reg(), src->as_double_reg());
576 break;
577 }
578 case Bytecodes::_d2i:
579 case Bytecodes::_f2i: {
580 FloatRegister rsrc = (code == Bytecodes::_d2i) ? src->as_double_reg() : src->as_float_reg();
581 Label L;
582 // Result must be 0 if value is NaN; test by comparing value to itself.
583 __ fcmpu(CR0, rsrc, rsrc);
584 __ li(dst->as_register(), 0);
585 __ bso(CR0, L);
586 __ fctiwz(rsrc, rsrc); // USE_KILL
587 __ mffprd(dst->as_register(), rsrc);
588 __ bind(L);
589 break;
590 }
591 case Bytecodes::_d2l:
592 case Bytecodes::_f2l: {
593 FloatRegister rsrc = (code == Bytecodes::_d2l) ? src->as_double_reg() : src->as_float_reg();
594 Label L;
595 // Result must be 0 if value is NaN; test by comparing value to itself.
596 __ fcmpu(CR0, rsrc, rsrc);
597 __ li(dst->as_register_lo(), 0);
598 __ bso(CR0, L);
599 __ fctidz(rsrc, rsrc); // USE_KILL
600 __ mffprd(dst->as_register_lo(), rsrc);
601 __ bind(L);
602 break;
603 }
604
605 default: ShouldNotReachHere();
606 }
607 }
608
609
610 void LIR_Assembler::align_call(LIR_Code) {
611 // do nothing since all instructions are word aligned on ppc
612 }
613
614
615 bool LIR_Assembler::emit_trampoline_stub_for_call(address target, Register Rtoc) {
616 int start_offset = __ offset();
617 // Put the entry point as a constant into the constant pool.
618 const address entry_point_toc_addr = __ address_constant(target, RelocationHolder::none);
619 if (entry_point_toc_addr == nullptr) {
620 bailout("const section overflow");
621 return false;
622 }
623 const int entry_point_toc_offset = __ offset_to_method_toc(entry_point_toc_addr);
624
625 // Emit the trampoline stub which will be related to the branch-and-link below.
626 address stub = __ emit_trampoline_stub(entry_point_toc_offset, start_offset, Rtoc);
627 if (!stub) {
628 bailout("no space for trampoline stub");
629 return false;
630 }
631 return true;
632 }
633
634
635 void LIR_Assembler::call(LIR_OpJavaCall* op, relocInfo::relocType rtype) {
636 assert(rtype==relocInfo::opt_virtual_call_type || rtype==relocInfo::static_call_type, "unexpected rtype");
637
638 bool success = emit_trampoline_stub_for_call(op->addr());
639 if (!success) { return; }
640
641 __ relocate(rtype);
642 // Note: At this point we do not have the address of the trampoline
643 // stub, and the entry point might be too far away for bl, so __ pc()
644 // serves as dummy and the bl will be patched later.
645 __ code()->set_insts_mark();
646 __ bl(__ pc());
647 add_call_info(code_offset(), op->info());
648 __ post_call_nop();
649 }
650
651
652 void LIR_Assembler::ic_call(LIR_OpJavaCall* op) {
653 __ calculate_address_from_global_toc(R2_TOC, __ method_toc());
654
655 // Virtual call relocation will point to ic load.
656 address virtual_call_meta_addr = __ pc();
657 // Load a clear inline cache.
658 AddressLiteral empty_ic((address) Universe::non_oop_word());
659 bool success = __ load_const_from_method_toc(R19_inline_cache_reg, empty_ic, R2_TOC);
660 if (!success) {
661 bailout("const section overflow");
662 return;
663 }
664 // Call to fixup routine. Fixup routine uses ScopeDesc info
665 // to determine who we intended to call.
666 __ relocate(virtual_call_Relocation::spec(virtual_call_meta_addr));
667
668 success = emit_trampoline_stub_for_call(op->addr(), R2_TOC);
669 if (!success) { return; }
670
671 // Note: At this point we do not have the address of the trampoline
672 // stub, and the entry point might be too far away for bl, so __ pc()
673 // serves as dummy and the bl will be patched later.
674 __ bl(__ pc());
675 add_call_info(code_offset(), op->info());
676 __ post_call_nop();
677 }
678
679 void LIR_Assembler::explicit_null_check(Register addr, CodeEmitInfo* info) {
680 ImplicitNullCheckStub* stub = new ImplicitNullCheckStub(code_offset(), info);
681 __ null_check(addr, stub->entry());
682 append_code_stub(stub);
683 }
684
685
686 // Attention: caller must encode oop if needed
687 int LIR_Assembler::store(LIR_Opr from_reg, Register base, int offset, BasicType type, bool wide) {
688 int store_offset;
689 if (!Assembler::is_simm16(offset)) {
690 // For offsets larger than a simm16 we setup the offset.
691 assert(wide && !from_reg->is_same_register(FrameMap::R0_opr), "large offset only supported in special case");
692 __ load_const_optimized(R0, offset);
693 store_offset = store(from_reg, base, R0, type, wide);
694 } else {
695 store_offset = code_offset();
696 switch (type) {
697 case T_BOOLEAN: // fall through
698 case T_BYTE : __ stb(from_reg->as_register(), offset, base); break;
699 case T_CHAR :
700 case T_SHORT : __ sth(from_reg->as_register(), offset, base); break;
701 case T_INT : __ stw(from_reg->as_register(), offset, base); break;
702 case T_LONG : __ std(from_reg->as_register_lo(), offset, base); break;
703 case T_ADDRESS:
704 case T_METADATA: __ std(from_reg->as_register(), offset, base); break;
705 case T_ARRAY : // fall through
706 case T_OBJECT:
707 {
708 if (UseCompressedOops && !wide) {
709 // Encoding done in caller
710 __ stw(from_reg->as_register(), offset, base);
711 __ verify_coop(from_reg->as_register(), FILE_AND_LINE);
712 } else {
713 __ std(from_reg->as_register(), offset, base);
714 if (VerifyOops) {
715 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
716 bs->check_oop(_masm, from_reg->as_register(), FILE_AND_LINE); // kills R0
717 }
718 }
719 break;
720 }
721 case T_FLOAT : __ stfs(from_reg->as_float_reg(), offset, base); break;
722 case T_DOUBLE: __ stfd(from_reg->as_double_reg(), offset, base); break;
723 default : ShouldNotReachHere();
724 }
725 }
726 return store_offset;
727 }
728
729
730 // Attention: caller must encode oop if needed
731 int LIR_Assembler::store(LIR_Opr from_reg, Register base, Register disp, BasicType type, bool wide) {
732 int store_offset = code_offset();
733 switch (type) {
734 case T_BOOLEAN: // fall through
735 case T_BYTE : __ stbx(from_reg->as_register(), base, disp); break;
736 case T_CHAR :
737 case T_SHORT : __ sthx(from_reg->as_register(), base, disp); break;
738 case T_INT : __ stwx(from_reg->as_register(), base, disp); break;
739 case T_LONG :
740 #ifdef _LP64
741 __ stdx(from_reg->as_register_lo(), base, disp);
742 #else
743 Unimplemented();
744 #endif
745 break;
746 case T_ADDRESS:
747 __ stdx(from_reg->as_register(), base, disp);
748 break;
749 case T_ARRAY : // fall through
750 case T_OBJECT:
751 {
752 if (UseCompressedOops && !wide) {
753 // Encoding done in caller.
754 __ stwx(from_reg->as_register(), base, disp);
755 __ verify_coop(from_reg->as_register(), FILE_AND_LINE); // kills R0
756 } else {
757 __ stdx(from_reg->as_register(), base, disp);
758 if (VerifyOops) {
759 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
760 bs->check_oop(_masm, from_reg->as_register(), FILE_AND_LINE); // kills R0
761 }
762 }
763 break;
764 }
765 case T_FLOAT : __ stfsx(from_reg->as_float_reg(), base, disp); break;
766 case T_DOUBLE: __ stfdx(from_reg->as_double_reg(), base, disp); break;
767 default : ShouldNotReachHere();
768 }
769 return store_offset;
770 }
771
772
773 int LIR_Assembler::load(Register base, int offset, LIR_Opr to_reg, BasicType type, bool wide) {
774 int load_offset;
775 if (!Assembler::is_simm16(offset)) {
776 // For offsets larger than a simm16 we setup the offset.
777 __ load_const_optimized(R0, offset);
778 load_offset = load(base, R0, to_reg, type, wide);
779 } else {
780 load_offset = code_offset();
781 switch(type) {
782 case T_BOOLEAN: // fall through
783 case T_BYTE : __ lbz(to_reg->as_register(), offset, base);
784 __ extsb(to_reg->as_register(), to_reg->as_register()); break;
785 case T_CHAR : __ lhz(to_reg->as_register(), offset, base); break;
786 case T_SHORT : __ lha(to_reg->as_register(), offset, base); break;
787 case T_INT : __ lwa(to_reg->as_register(), offset, base); break;
788 case T_LONG : __ ld(to_reg->as_register_lo(), offset, base); break;
789 case T_METADATA: __ ld(to_reg->as_register(), offset, base); break;
790 case T_ADDRESS:
791 __ ld(to_reg->as_register(), offset, base);
792 break;
793 case T_ARRAY : // fall through
794 case T_OBJECT:
795 {
796 if (UseCompressedOops && !wide) {
797 __ lwz(to_reg->as_register(), offset, base);
798 __ decode_heap_oop(to_reg->as_register());
799 } else {
800 __ ld(to_reg->as_register(), offset, base);
801 }
802 break;
803 }
804 case T_FLOAT: __ lfs(to_reg->as_float_reg(), offset, base); break;
805 case T_DOUBLE: __ lfd(to_reg->as_double_reg(), offset, base); break;
806 default : ShouldNotReachHere();
807 }
808 }
809 return load_offset;
810 }
811
812
813 int LIR_Assembler::load(Register base, Register disp, LIR_Opr to_reg, BasicType type, bool wide) {
814 int load_offset = code_offset();
815 switch(type) {
816 case T_BOOLEAN: // fall through
817 case T_BYTE : __ lbzx(to_reg->as_register(), base, disp);
818 __ extsb(to_reg->as_register(), to_reg->as_register()); break;
819 case T_CHAR : __ lhzx(to_reg->as_register(), base, disp); break;
820 case T_SHORT : __ lhax(to_reg->as_register(), base, disp); break;
821 case T_INT : __ lwax(to_reg->as_register(), base, disp); break;
822 case T_ADDRESS: __ ldx(to_reg->as_register(), base, disp); break;
823 case T_ARRAY : // fall through
824 case T_OBJECT:
825 {
826 if (UseCompressedOops && !wide) {
827 __ lwzx(to_reg->as_register(), base, disp);
828 __ decode_heap_oop(to_reg->as_register());
829 } else {
830 __ ldx(to_reg->as_register(), base, disp);
831 }
832 break;
833 }
834 case T_FLOAT: __ lfsx(to_reg->as_float_reg() , base, disp); break;
835 case T_DOUBLE: __ lfdx(to_reg->as_double_reg(), base, disp); break;
836 case T_LONG :
837 #ifdef _LP64
838 __ ldx(to_reg->as_register_lo(), base, disp);
839 #else
840 Unimplemented();
841 #endif
842 break;
843 default : ShouldNotReachHere();
844 }
845 return load_offset;
846 }
847
848
849 void LIR_Assembler::const2stack(LIR_Opr src, LIR_Opr dest) {
850 LIR_Const* c = src->as_constant_ptr();
851 Register src_reg = R0;
852 switch (c->type()) {
853 case T_INT:
854 case T_FLOAT: {
855 int value = c->as_jint_bits();
856 __ load_const_optimized(src_reg, value);
857 Address addr = frame_map()->address_for_slot(dest->single_stack_ix());
858 __ stw(src_reg, addr);
859 break;
860 }
861 case T_ADDRESS: {
862 int value = c->as_jint_bits();
863 __ load_const_optimized(src_reg, value);
864 Address addr = frame_map()->address_for_slot(dest->single_stack_ix());
865 __ std(src_reg, addr);
866 break;
867 }
868 case T_OBJECT: {
869 jobject2reg(c->as_jobject(), src_reg);
870 Address addr = frame_map()->address_for_slot(dest->single_stack_ix());
871 __ std(src_reg, addr);
872 break;
873 }
874 case T_LONG:
875 case T_DOUBLE: {
876 int value = c->as_jlong_bits();
877 __ load_const_optimized(src_reg, value);
878 Address addr = frame_map()->address_for_double_slot(dest->double_stack_ix());
879 __ std(src_reg, addr);
880 break;
881 }
882 default:
883 Unimplemented();
884 }
885 }
886
887
888 void LIR_Assembler::const2mem(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info, bool wide) {
889 LIR_Const* c = src->as_constant_ptr();
890 LIR_Address* addr = dest->as_address_ptr();
891 Register base = addr->base()->as_pointer_register();
892 LIR_Opr tmp = LIR_OprFact::illegalOpr;
893 int offset = -1;
894 // Null check for large offsets in LIRGenerator::do_StoreField.
895 bool needs_explicit_null_check = !ImplicitNullChecks;
896
897 if (info != nullptr && needs_explicit_null_check) {
898 explicit_null_check(base, info);
899 }
900
901 switch (c->type()) {
902 case T_FLOAT: type = T_INT;
903 case T_INT:
904 case T_ADDRESS: {
905 tmp = FrameMap::R0_opr;
906 __ load_const_optimized(tmp->as_register(), c->as_jint_bits());
907 break;
908 }
909 case T_DOUBLE: type = T_LONG;
910 case T_LONG: {
911 tmp = FrameMap::R0_long_opr;
912 __ load_const_optimized(tmp->as_register_lo(), c->as_jlong_bits());
913 break;
914 }
915 case T_OBJECT: {
916 tmp = FrameMap::R0_opr;
917 if (UseCompressedOops && !wide && c->as_jobject() != nullptr) {
918 AddressLiteral oop_addr = __ constant_oop_address(c->as_jobject());
919 // Don't care about sign extend (will use stw).
920 __ lis(R0, 0); // Will get patched.
921 __ relocate(oop_addr.rspec(), /*compressed format*/ 1);
922 __ ori(R0, R0, 0); // Will get patched.
923 } else {
924 jobject2reg(c->as_jobject(), R0);
925 }
926 break;
927 }
928 default:
929 Unimplemented();
930 }
931
932 // Handle either reg+reg or reg+disp address.
933 if (addr->index()->is_valid()) {
934 assert(addr->disp() == 0, "must be zero");
935 offset = store(tmp, base, addr->index()->as_pointer_register(), type, wide);
936 } else {
937 assert(Assembler::is_simm16(addr->disp()), "can't handle larger addresses");
938 offset = store(tmp, base, addr->disp(), type, wide);
939 }
940
941 if (info != nullptr) {
942 assert(offset != -1, "offset should've been set");
943 if (!needs_explicit_null_check) {
944 add_debug_info_for_null_check(offset, info);
945 }
946 }
947 }
948
949
950 void LIR_Assembler::const2reg(LIR_Opr src, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
951 LIR_Const* c = src->as_constant_ptr();
952 LIR_Opr to_reg = dest;
953
954 switch (c->type()) {
955 case T_INT: {
956 assert(patch_code == lir_patch_none, "no patching handled here");
957 __ load_const_optimized(dest->as_register(), c->as_jint(), R0);
958 break;
959 }
960 case T_ADDRESS: {
961 assert(patch_code == lir_patch_none, "no patching handled here");
962 __ load_const_optimized(dest->as_register(), c->as_jint(), R0); // Yes, as_jint ...
963 break;
964 }
965 case T_LONG: {
966 assert(patch_code == lir_patch_none, "no patching handled here");
967 __ load_const_optimized(dest->as_register_lo(), c->as_jlong(), R0);
968 break;
969 }
970
971 case T_OBJECT: {
972 if (patch_code == lir_patch_none) {
973 jobject2reg(c->as_jobject(), to_reg->as_register());
974 } else {
975 jobject2reg_with_patching(to_reg->as_register(), info);
976 }
977 break;
978 }
979
980 case T_METADATA:
981 {
982 if (patch_code == lir_patch_none) {
983 metadata2reg(c->as_metadata(), to_reg->as_register());
984 } else {
985 klass2reg_with_patching(to_reg->as_register(), info);
986 }
987 }
988 break;
989
990 case T_FLOAT:
991 {
992 if (to_reg->is_single_fpu()) {
993 address const_addr = __ float_constant(c->as_jfloat());
994 if (const_addr == nullptr) {
995 bailout("const section overflow");
996 break;
997 }
998 RelocationHolder rspec = internal_word_Relocation::spec(const_addr);
999 __ relocate(rspec);
1000 __ load_const(R0, const_addr);
1001 __ lfsx(to_reg->as_float_reg(), R0);
1002 } else {
1003 assert(to_reg->is_single_cpu(), "Must be a cpu register.");
1004 __ load_const_optimized(to_reg->as_register(), jint_cast(c->as_jfloat()), R0);
1005 }
1006 }
1007 break;
1008
1009 case T_DOUBLE:
1010 {
1011 if (to_reg->is_double_fpu()) {
1012 address const_addr = __ double_constant(c->as_jdouble());
1013 if (const_addr == nullptr) {
1014 bailout("const section overflow");
1015 break;
1016 }
1017 RelocationHolder rspec = internal_word_Relocation::spec(const_addr);
1018 __ relocate(rspec);
1019 __ load_const(R0, const_addr);
1020 __ lfdx(to_reg->as_double_reg(), R0);
1021 } else {
1022 assert(to_reg->is_double_cpu(), "Must be a long register.");
1023 __ load_const_optimized(to_reg->as_register_lo(), jlong_cast(c->as_jdouble()), R0);
1024 }
1025 }
1026 break;
1027
1028 default:
1029 ShouldNotReachHere();
1030 }
1031 }
1032
1033
1034 Address LIR_Assembler::as_Address(LIR_Address* addr) {
1035 Unimplemented(); return Address();
1036 }
1037
1038
1039 inline RegisterOrConstant index_or_disp(LIR_Address* addr) {
1040 if (addr->index()->is_illegal()) {
1041 return (RegisterOrConstant)(addr->disp());
1042 } else {
1043 return (RegisterOrConstant)(addr->index()->as_pointer_register());
1044 }
1045 }
1046
1047
1048 void LIR_Assembler::stack2stack(LIR_Opr src, LIR_Opr dest, BasicType type) {
1049 const Register tmp = R0;
1050 switch (type) {
1051 case T_INT:
1052 case T_FLOAT: {
1053 Address from = frame_map()->address_for_slot(src->single_stack_ix());
1054 Address to = frame_map()->address_for_slot(dest->single_stack_ix());
1055 __ lwz(tmp, from);
1056 __ stw(tmp, to);
1057 break;
1058 }
1059 case T_ADDRESS:
1060 case T_OBJECT: {
1061 Address from = frame_map()->address_for_slot(src->single_stack_ix());
1062 Address to = frame_map()->address_for_slot(dest->single_stack_ix());
1063 __ ld(tmp, from);
1064 __ std(tmp, to);
1065 break;
1066 }
1067 case T_LONG:
1068 case T_DOUBLE: {
1069 Address from = frame_map()->address_for_double_slot(src->double_stack_ix());
1070 Address to = frame_map()->address_for_double_slot(dest->double_stack_ix());
1071 __ ld(tmp, from);
1072 __ std(tmp, to);
1073 break;
1074 }
1075
1076 default:
1077 ShouldNotReachHere();
1078 }
1079 }
1080
1081
1082 Address LIR_Assembler::as_Address_hi(LIR_Address* addr) {
1083 Unimplemented(); return Address();
1084 }
1085
1086
1087 Address LIR_Assembler::as_Address_lo(LIR_Address* addr) {
1088 Unimplemented(); return Address();
1089 }
1090
1091
1092 void LIR_Assembler::mem2reg(LIR_Opr src_opr, LIR_Opr dest, BasicType type,
1093 LIR_PatchCode patch_code, CodeEmitInfo* info, bool wide) {
1094
1095 assert(type != T_METADATA, "load of metadata ptr not supported");
1096 LIR_Address* addr = src_opr->as_address_ptr();
1097 LIR_Opr to_reg = dest;
1098
1099 Register src = addr->base()->as_pointer_register();
1100 Register disp_reg = noreg;
1101 int disp_value = addr->disp();
1102 bool needs_patching = (patch_code != lir_patch_none);
1103 // null check for large offsets in LIRGenerator::do_LoadField
1104 bool needs_explicit_null_check = !os::zero_page_read_protected() || !ImplicitNullChecks;
1105
1106 if (info != nullptr && needs_explicit_null_check) {
1107 explicit_null_check(src, info);
1108 }
1109
1110 if (addr->base()->type() == T_OBJECT) {
1111 __ verify_oop(src, FILE_AND_LINE);
1112 }
1113
1114 PatchingStub* patch = nullptr;
1115 if (needs_patching) {
1116 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1117 assert(!to_reg->is_double_cpu() ||
1118 patch_code == lir_patch_none ||
1119 patch_code == lir_patch_normal, "patching doesn't match register");
1120 }
1121
1122 if (addr->index()->is_illegal()) {
1123 if (!Assembler::is_simm16(disp_value)) {
1124 if (needs_patching) {
1125 __ load_const32(R0, 0); // patchable int
1126 } else {
1127 __ load_const_optimized(R0, disp_value);
1128 }
1129 disp_reg = R0;
1130 }
1131 } else {
1132 disp_reg = addr->index()->as_pointer_register();
1133 assert(disp_value == 0, "can't handle 3 operand addresses");
1134 }
1135
1136 // Remember the offset of the load. The patching_epilog must be done
1137 // before the call to add_debug_info, otherwise the PcDescs don't get
1138 // entered in increasing order.
1139 int offset;
1140
1141 if (disp_reg == noreg) {
1142 assert(Assembler::is_simm16(disp_value), "should have set this up");
1143 offset = load(src, disp_value, to_reg, type, wide);
1144 } else {
1145 offset = load(src, disp_reg, to_reg, type, wide);
1146 }
1147
1148 if (patch != nullptr) {
1149 patching_epilog(patch, patch_code, src, info);
1150 }
1151 if (info != nullptr && !needs_explicit_null_check) {
1152 add_debug_info_for_null_check(offset, info);
1153 }
1154 }
1155
1156
1157 void LIR_Assembler::stack2reg(LIR_Opr src, LIR_Opr dest, BasicType type) {
1158 Address addr;
1159 if (src->is_single_word()) {
1160 addr = frame_map()->address_for_slot(src->single_stack_ix());
1161 } else if (src->is_double_word()) {
1162 addr = frame_map()->address_for_double_slot(src->double_stack_ix());
1163 }
1164
1165 load(addr.base(), addr.disp(), dest, dest->type(), true /*wide*/);
1166 }
1167
1168
1169 void LIR_Assembler::reg2stack(LIR_Opr from_reg, LIR_Opr dest, BasicType type) {
1170 Address addr;
1171 if (dest->is_single_word()) {
1172 addr = frame_map()->address_for_slot(dest->single_stack_ix());
1173 } else if (dest->is_double_word()) {
1174 addr = frame_map()->address_for_slot(dest->double_stack_ix());
1175 }
1176
1177 store(from_reg, addr.base(), addr.disp(), from_reg->type(), true /*wide*/);
1178 }
1179
1180
1181 void LIR_Assembler::reg2reg(LIR_Opr from_reg, LIR_Opr to_reg) {
1182 if (from_reg->is_float_kind() && to_reg->is_float_kind()) {
1183 if (from_reg->is_double_fpu()) {
1184 // double to double moves
1185 assert(to_reg->is_double_fpu(), "should match");
1186 __ fmr_if_needed(to_reg->as_double_reg(), from_reg->as_double_reg());
1187 } else {
1188 // float to float moves
1189 assert(to_reg->is_single_fpu(), "should match");
1190 __ fmr_if_needed(to_reg->as_float_reg(), from_reg->as_float_reg());
1191 }
1192 } else if (!from_reg->is_float_kind() && !to_reg->is_float_kind()) {
1193 if (from_reg->is_double_cpu()) {
1194 __ mr_if_needed(to_reg->as_pointer_register(), from_reg->as_pointer_register());
1195 } else if (to_reg->is_double_cpu()) {
1196 // int to int moves
1197 __ mr_if_needed(to_reg->as_register_lo(), from_reg->as_register());
1198 } else {
1199 // int to int moves
1200 __ mr_if_needed(to_reg->as_register(), from_reg->as_register());
1201 }
1202 } else {
1203 ShouldNotReachHere();
1204 }
1205 if (is_reference_type(to_reg->type())) {
1206 __ verify_oop(to_reg->as_register(), FILE_AND_LINE);
1207 }
1208 }
1209
1210
1211 void LIR_Assembler::reg2mem(LIR_Opr from_reg, LIR_Opr dest, BasicType type,
1212 LIR_PatchCode patch_code, CodeEmitInfo* info,
1213 bool wide) {
1214 assert(type != T_METADATA, "store of metadata ptr not supported");
1215 LIR_Address* addr = dest->as_address_ptr();
1216
1217 Register src = addr->base()->as_pointer_register();
1218 Register disp_reg = noreg;
1219 int disp_value = addr->disp();
1220 bool needs_patching = (patch_code != lir_patch_none);
1221 bool compress_oop = (is_reference_type(type)) && UseCompressedOops && !wide &&
1222 CompressedOops::mode() != CompressedOops::UnscaledNarrowOop;
1223 bool load_disp = addr->index()->is_illegal() && !Assembler::is_simm16(disp_value);
1224 bool use_R29 = compress_oop && load_disp; // Avoid register conflict, also do null check before killing R29.
1225 // Null check for large offsets in LIRGenerator::do_StoreField.
1226 bool needs_explicit_null_check = !ImplicitNullChecks || use_R29;
1227
1228 if (info != nullptr && needs_explicit_null_check) {
1229 explicit_null_check(src, info);
1230 }
1231
1232 if (addr->base()->is_oop_register()) {
1233 __ verify_oop(src, FILE_AND_LINE);
1234 }
1235
1236 PatchingStub* patch = nullptr;
1237 if (needs_patching) {
1238 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1239 assert(!from_reg->is_double_cpu() ||
1240 patch_code == lir_patch_none ||
1241 patch_code == lir_patch_normal, "patching doesn't match register");
1242 }
1243
1244 if (addr->index()->is_illegal()) {
1245 if (load_disp) {
1246 disp_reg = use_R29 ? R29_TOC : R0;
1247 if (needs_patching) {
1248 __ load_const32(disp_reg, 0); // patchable int
1249 } else {
1250 __ load_const_optimized(disp_reg, disp_value);
1251 }
1252 }
1253 } else {
1254 disp_reg = addr->index()->as_pointer_register();
1255 assert(disp_value == 0, "can't handle 3 operand addresses");
1256 }
1257
1258 // remember the offset of the store. The patching_epilog must be done
1259 // before the call to add_debug_info_for_null_check, otherwise the PcDescs don't get
1260 // entered in increasing order.
1261 int offset;
1262
1263 if (compress_oop) {
1264 Register co = __ encode_heap_oop(R0, from_reg->as_register());
1265 from_reg = FrameMap::as_opr(co);
1266 }
1267
1268 if (disp_reg == noreg) {
1269 assert(Assembler::is_simm16(disp_value), "should have set this up");
1270 offset = store(from_reg, src, disp_value, type, wide);
1271 } else {
1272 offset = store(from_reg, src, disp_reg, type, wide);
1273 }
1274
1275 if (use_R29) {
1276 __ load_const_optimized(R29_TOC, MacroAssembler::global_toc(), R0); // reinit
1277 }
1278
1279 if (patch != nullptr) {
1280 patching_epilog(patch, patch_code, src, info);
1281 }
1282
1283 if (info != nullptr && !needs_explicit_null_check) {
1284 add_debug_info_for_null_check(offset, info);
1285 }
1286 }
1287
1288
1289 void LIR_Assembler::return_op(LIR_Opr result, C1SafepointPollStub* code_stub) {
1290 const Register return_pc = R31; // Must survive C-call to enable_stack_reserved_zone().
1291 const Register temp = R12;
1292
1293 // Pop the stack before the safepoint code.
1294 int frame_size = initial_frame_size_in_bytes();
1295 if (Assembler::is_simm(frame_size, 16)) {
1296 __ addi(R1_SP, R1_SP, frame_size);
1297 } else {
1298 __ pop_frame();
1299 }
1300
1301 // Restore return pc relative to callers' sp.
1302 __ ld(return_pc, _abi0(lr), R1_SP);
1303 // Move return pc to LR.
1304 __ mtlr(return_pc);
1305
1306 if (StackReservedPages > 0 && compilation()->has_reserved_stack_access()) {
1307 __ reserved_stack_check(return_pc);
1308 }
1309
1310 // We need to mark the code position where the load from the safepoint
1311 // polling page was emitted as relocInfo::poll_return_type here.
1312 if (!UseSIGTRAP) {
1313 code_stub->set_safepoint_offset(__ offset());
1314 __ relocate(relocInfo::poll_return_type);
1315 }
1316 __ safepoint_poll(*code_stub->entry(), temp, true /* at_return */, true /* in_nmethod */);
1317
1318 // Return.
1319 __ blr();
1320 }
1321
1322
1323 int LIR_Assembler::safepoint_poll(LIR_Opr tmp, CodeEmitInfo* info) {
1324 const Register poll_addr = tmp->as_register();
1325 __ ld(poll_addr, in_bytes(JavaThread::polling_page_offset()), R16_thread);
1326 if (info != nullptr) {
1327 add_debug_info_for_branch(info);
1328 }
1329 int offset = __ offset();
1330 __ relocate(relocInfo::poll_type);
1331 __ load_from_polling_page(poll_addr);
1332
1333 return offset;
1334 }
1335
1336
1337 void LIR_Assembler::emit_static_call_stub() {
1338 address call_pc = __ pc();
1339 address stub = __ start_a_stub(static_call_stub_size());
1340 if (stub == nullptr) {
1341 bailout("static call stub overflow");
1342 return;
1343 }
1344
1345 // For java_to_interp stubs we use R11_scratch1 as scratch register
1346 // and in call trampoline stubs we use R12_scratch2. This way we
1347 // can distinguish them (see is_NativeCallTrampolineStub_at()).
1348 const Register reg_scratch = R11_scratch1;
1349
1350 // Create a static stub relocation which relates this stub
1351 // with the call instruction at insts_call_instruction_offset in the
1352 // instructions code-section.
1353 int start = __ offset();
1354 __ relocate(static_stub_Relocation::spec(call_pc));
1355
1356 // Now, create the stub's code:
1357 // - load the TOC
1358 // - load the inline cache oop from the constant pool
1359 // - load the call target from the constant pool
1360 // - call
1361 __ calculate_address_from_global_toc(reg_scratch, __ method_toc());
1362 AddressLiteral ic = __ allocate_metadata_address((Metadata *)nullptr);
1363 bool success = __ load_const_from_method_toc(R19_inline_cache_reg, ic, reg_scratch, /*fixed_size*/ true);
1364
1365 if (ReoptimizeCallSequences) {
1366 __ b64_patchable((address)-1, relocInfo::none);
1367 } else {
1368 AddressLiteral a((address)-1);
1369 success = success && __ load_const_from_method_toc(reg_scratch, a, reg_scratch, /*fixed_size*/ true);
1370 __ mtctr(reg_scratch);
1371 __ bctr();
1372 }
1373 if (!success) {
1374 bailout("const section overflow");
1375 return;
1376 }
1377
1378 assert(__ offset() - start <= static_call_stub_size(), "stub too big");
1379 __ end_a_stub();
1380 }
1381
1382
1383 void LIR_Assembler::comp_op(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Op2* op) {
1384 bool unsigned_comp = (condition == lir_cond_belowEqual || condition == lir_cond_aboveEqual);
1385 if (opr1->is_single_fpu()) {
1386 __ fcmpu(BOOL_RESULT, opr1->as_float_reg(), opr2->as_float_reg());
1387 } else if (opr1->is_double_fpu()) {
1388 __ fcmpu(BOOL_RESULT, opr1->as_double_reg(), opr2->as_double_reg());
1389 } else if (opr1->is_single_cpu()) {
1390 if (opr2->is_constant()) {
1391 switch (opr2->as_constant_ptr()->type()) {
1392 case T_INT:
1393 {
1394 jint con = opr2->as_constant_ptr()->as_jint();
1395 if (unsigned_comp) {
1396 if (Assembler::is_uimm(con, 16)) {
1397 __ cmplwi(BOOL_RESULT, opr1->as_register(), con);
1398 } else {
1399 __ load_const_optimized(R0, con);
1400 __ cmplw(BOOL_RESULT, opr1->as_register(), R0);
1401 }
1402 } else {
1403 if (Assembler::is_simm(con, 16)) {
1404 __ cmpwi(BOOL_RESULT, opr1->as_register(), con);
1405 } else {
1406 __ load_const_optimized(R0, con);
1407 __ cmpw(BOOL_RESULT, opr1->as_register(), R0);
1408 }
1409 }
1410 }
1411 break;
1412
1413 case T_OBJECT:
1414 // There are only equal/notequal comparisons on objects.
1415 {
1416 assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "oops");
1417 jobject con = opr2->as_constant_ptr()->as_jobject();
1418 if (con == nullptr) {
1419 __ cmpdi(BOOL_RESULT, opr1->as_register(), 0);
1420 } else {
1421 jobject2reg(con, R0);
1422 __ cmpd(BOOL_RESULT, opr1->as_register(), R0);
1423 }
1424 }
1425 break;
1426
1427 case T_METADATA:
1428 // We only need, for now, comparison with null for metadata.
1429 {
1430 assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "oops");
1431 Metadata* p = opr2->as_constant_ptr()->as_metadata();
1432 if (p == nullptr) {
1433 __ cmpdi(BOOL_RESULT, opr1->as_register(), 0);
1434 } else {
1435 ShouldNotReachHere();
1436 }
1437 }
1438 break;
1439
1440 default:
1441 ShouldNotReachHere();
1442 break;
1443 }
1444 } else {
1445 assert(opr1->type() != T_ADDRESS && opr2->type() != T_ADDRESS, "currently unsupported");
1446 if (is_reference_type(opr1->type())) {
1447 // There are only equal/notequal comparisons on objects.
1448 assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "oops");
1449 __ cmpd(BOOL_RESULT, opr1->as_register(), opr2->as_register());
1450 } else {
1451 if (unsigned_comp) {
1452 __ cmplw(BOOL_RESULT, opr1->as_register(), opr2->as_register());
1453 } else {
1454 __ cmpw(BOOL_RESULT, opr1->as_register(), opr2->as_register());
1455 }
1456 }
1457 }
1458 } else if (opr1->is_double_cpu()) {
1459 if (opr2->is_constant()) {
1460 jlong con = opr2->as_constant_ptr()->as_jlong();
1461 if (unsigned_comp) {
1462 if (Assembler::is_uimm(con, 16)) {
1463 __ cmpldi(BOOL_RESULT, opr1->as_register_lo(), con);
1464 } else {
1465 __ load_const_optimized(R0, con);
1466 __ cmpld(BOOL_RESULT, opr1->as_register_lo(), R0);
1467 }
1468 } else {
1469 if (Assembler::is_simm(con, 16)) {
1470 __ cmpdi(BOOL_RESULT, opr1->as_register_lo(), con);
1471 } else {
1472 __ load_const_optimized(R0, con);
1473 __ cmpd(BOOL_RESULT, opr1->as_register_lo(), R0);
1474 }
1475 }
1476 } else if (opr2->is_register()) {
1477 if (unsigned_comp) {
1478 __ cmpld(BOOL_RESULT, opr1->as_register_lo(), opr2->as_register_lo());
1479 } else {
1480 __ cmpd(BOOL_RESULT, opr1->as_register_lo(), opr2->as_register_lo());
1481 }
1482 } else {
1483 ShouldNotReachHere();
1484 }
1485 } else {
1486 ShouldNotReachHere();
1487 }
1488 }
1489
1490
1491 void LIR_Assembler::comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst, LIR_Op2* op){
1492 const Register Rdst = dst->as_register();
1493 if (code == lir_cmp_fd2i || code == lir_ucmp_fd2i) {
1494 bool is_unordered_less = (code == lir_ucmp_fd2i);
1495 if (left->is_single_fpu()) {
1496 __ fcmpu(CR0, left->as_float_reg(), right->as_float_reg());
1497 } else if (left->is_double_fpu()) {
1498 __ fcmpu(CR0, left->as_double_reg(), right->as_double_reg());
1499 } else {
1500 ShouldNotReachHere();
1501 }
1502 __ set_cmpu3(Rdst, is_unordered_less); // is_unordered_less ? -1 : 1
1503 } else if (code == lir_cmp_l2i) {
1504 __ cmpd(CR0, left->as_register_lo(), right->as_register_lo());
1505 __ set_cmp3(Rdst); // set result as follows: <: -1, =: 0, >: 1
1506 } else {
1507 ShouldNotReachHere();
1508 }
1509 }
1510
1511
1512 inline void load_to_reg(LIR_Assembler *lasm, LIR_Opr src, LIR_Opr dst) {
1513 if (src->is_constant()) {
1514 lasm->const2reg(src, dst, lir_patch_none, nullptr);
1515 } else if (src->is_register()) {
1516 lasm->reg2reg(src, dst);
1517 } else if (src->is_stack()) {
1518 lasm->stack2reg(src, dst, dst->type());
1519 } else {
1520 ShouldNotReachHere();
1521 }
1522 }
1523
1524 void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type,
1525 LIR_Opr cmp_opr1, LIR_Opr cmp_opr2) {
1526 assert(cmp_opr1 == LIR_OprFact::illegalOpr && cmp_opr2 == LIR_OprFact::illegalOpr, "unnecessary cmp oprs on ppc");
1527
1528 if (opr1->is_equal(opr2) || opr1->is_same_register(opr2)) {
1529 load_to_reg(this, opr1, result); // Condition doesn't matter.
1530 return;
1531 }
1532
1533 bool positive = false;
1534 Assembler::Condition cond = Assembler::equal;
1535 switch (condition) {
1536 case lir_cond_equal: positive = true ; cond = Assembler::equal ; break;
1537 case lir_cond_notEqual: positive = false; cond = Assembler::equal ; break;
1538 case lir_cond_less: positive = true ; cond = Assembler::less ; break;
1539 case lir_cond_belowEqual:
1540 case lir_cond_lessEqual: positive = false; cond = Assembler::greater; break;
1541 case lir_cond_greater: positive = true ; cond = Assembler::greater; break;
1542 case lir_cond_aboveEqual:
1543 case lir_cond_greaterEqual: positive = false; cond = Assembler::less ; break;
1544 default: ShouldNotReachHere();
1545 }
1546
1547 if (result->is_cpu_register()) {
1548 bool o1_is_reg = opr1->is_cpu_register(), o2_is_reg = opr2->is_cpu_register();
1549 const Register result_reg = result->is_single_cpu() ? result->as_register() : result->as_register_lo();
1550
1551 // We can use result_reg to load one operand if not already in register.
1552 Register first = o1_is_reg ? (opr1->is_single_cpu() ? opr1->as_register() : opr1->as_register_lo()) : result_reg,
1553 second = o2_is_reg ? (opr2->is_single_cpu() ? opr2->as_register() : opr2->as_register_lo()) : result_reg;
1554
1555 if (first != second) {
1556 if (!o1_is_reg) {
1557 load_to_reg(this, opr1, result);
1558 }
1559
1560 if (!o2_is_reg) {
1561 load_to_reg(this, opr2, result);
1562 }
1563
1564 __ isel(result_reg, BOOL_RESULT, cond, !positive, first, second);
1565 return;
1566 }
1567 } // isel
1568
1569 load_to_reg(this, opr1, result);
1570
1571 Label skip;
1572 int bo = positive ? Assembler::bcondCRbiIs1 : Assembler::bcondCRbiIs0;
1573 int bi = Assembler::bi0(BOOL_RESULT, cond);
1574 __ bc(bo, bi, skip);
1575
1576 load_to_reg(this, opr2, result);
1577 __ bind(skip);
1578 }
1579
1580
1581 void LIR_Assembler::arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest,
1582 CodeEmitInfo* info) {
1583 assert(info == nullptr, "unused on this code path");
1584 assert(left->is_register(), "wrong items state");
1585 assert(dest->is_register(), "wrong items state");
1586
1587 if (right->is_register()) {
1588 if (dest->is_float_kind()) {
1589
1590 FloatRegister lreg, rreg, res;
1591 if (right->is_single_fpu()) {
1592 lreg = left->as_float_reg();
1593 rreg = right->as_float_reg();
1594 res = dest->as_float_reg();
1595 switch (code) {
1596 case lir_add: __ fadds(res, lreg, rreg); break;
1597 case lir_sub: __ fsubs(res, lreg, rreg); break;
1598 case lir_mul: __ fmuls(res, lreg, rreg); break;
1599 case lir_div: __ fdivs(res, lreg, rreg); break;
1600 default: ShouldNotReachHere();
1601 }
1602 } else {
1603 lreg = left->as_double_reg();
1604 rreg = right->as_double_reg();
1605 res = dest->as_double_reg();
1606 switch (code) {
1607 case lir_add: __ fadd(res, lreg, rreg); break;
1608 case lir_sub: __ fsub(res, lreg, rreg); break;
1609 case lir_mul: __ fmul(res, lreg, rreg); break;
1610 case lir_div: __ fdiv(res, lreg, rreg); break;
1611 default: ShouldNotReachHere();
1612 }
1613 }
1614
1615 } else if (dest->is_double_cpu()) {
1616
1617 Register dst_lo = dest->as_register_lo();
1618 Register op1_lo = left->as_pointer_register();
1619 Register op2_lo = right->as_pointer_register();
1620
1621 switch (code) {
1622 case lir_add: __ add(dst_lo, op1_lo, op2_lo); break;
1623 case lir_sub: __ sub(dst_lo, op1_lo, op2_lo); break;
1624 case lir_mul: __ mulld(dst_lo, op1_lo, op2_lo); break;
1625 default: ShouldNotReachHere();
1626 }
1627 } else {
1628 assert (right->is_single_cpu(), "Just Checking");
1629
1630 Register lreg = left->as_register();
1631 Register res = dest->as_register();
1632 Register rreg = right->as_register();
1633 switch (code) {
1634 case lir_add: __ add (res, lreg, rreg); break;
1635 case lir_sub: __ sub (res, lreg, rreg); break;
1636 case lir_mul: __ mullw(res, lreg, rreg); break;
1637 default: ShouldNotReachHere();
1638 }
1639 }
1640 } else {
1641 assert (right->is_constant(), "must be constant");
1642
1643 if (dest->is_single_cpu()) {
1644 Register lreg = left->as_register();
1645 Register res = dest->as_register();
1646 int simm16 = right->as_constant_ptr()->as_jint();
1647
1648 switch (code) {
1649 case lir_sub: assert(Assembler::is_simm16(-simm16), "cannot encode"); // see do_ArithmeticOp_Int
1650 simm16 = -simm16;
1651 case lir_add: if (res == lreg && simm16 == 0) break;
1652 __ addi(res, lreg, simm16); break;
1653 case lir_mul: if (res == lreg && simm16 == 1) break;
1654 __ mulli(res, lreg, simm16); break;
1655 default: ShouldNotReachHere();
1656 }
1657 } else {
1658 Register lreg = left->as_pointer_register();
1659 Register res = dest->as_register_lo();
1660 long con = right->as_constant_ptr()->as_jlong();
1661 assert(Assembler::is_simm16(con), "must be simm16");
1662
1663 switch (code) {
1664 case lir_sub: assert(Assembler::is_simm16(-con), "cannot encode"); // see do_ArithmeticOp_Long
1665 con = -con;
1666 case lir_add: if (res == lreg && con == 0) break;
1667 __ addi(res, lreg, (int)con); break;
1668 case lir_mul: if (res == lreg && con == 1) break;
1669 __ mulli(res, lreg, (int)con); break;
1670 default: ShouldNotReachHere();
1671 }
1672 }
1673 }
1674 }
1675
1676
1677 void LIR_Assembler::intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr tmp, LIR_Opr dest, LIR_Op* op) {
1678 switch (code) {
1679 case lir_sqrt: {
1680 __ fsqrt(dest->as_double_reg(), value->as_double_reg());
1681 break;
1682 }
1683 case lir_abs: {
1684 __ fabs(dest->as_double_reg(), value->as_double_reg());
1685 break;
1686 }
1687 case lir_f2hf: {
1688 __ f2hf(dest.as_register(), value.as_float_reg(), tmp.as_float_reg());
1689 break;
1690 }
1691 case lir_hf2f: {
1692 __ hf2f(dest->as_float_reg(), value.as_register());
1693 break;
1694 }
1695 default: {
1696 ShouldNotReachHere();
1697 break;
1698 }
1699 }
1700 }
1701
1702
1703 void LIR_Assembler::logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest) {
1704 if (right->is_constant()) { // see do_LogicOp
1705 long uimm;
1706 Register d, l;
1707 if (dest->is_single_cpu()) {
1708 uimm = right->as_constant_ptr()->as_jint();
1709 d = dest->as_register();
1710 l = left->as_register();
1711 } else {
1712 uimm = right->as_constant_ptr()->as_jlong();
1713 d = dest->as_register_lo();
1714 l = left->as_register_lo();
1715 }
1716 long uimms = (unsigned long)uimm >> 16,
1717 uimmss = (unsigned long)uimm >> 32;
1718
1719 switch (code) {
1720 case lir_logic_and:
1721 if (uimmss != 0 || (uimms != 0 && (uimm & 0xFFFF) != 0) || is_power_of_2(uimm)) {
1722 __ andi(d, l, uimm); // special cases
1723 } else if (uimms != 0) { __ andis_(d, l, uimms); }
1724 else { __ andi_(d, l, uimm); }
1725 break;
1726
1727 case lir_logic_or:
1728 if (uimms != 0) { assert((uimm & 0xFFFF) == 0, "sanity"); __ oris(d, l, uimms); }
1729 else { __ ori(d, l, uimm); }
1730 break;
1731
1732 case lir_logic_xor:
1733 if (uimm == -1) { __ nand(d, l, l); } // special case
1734 else if (uimms != 0) { assert((uimm & 0xFFFF) == 0, "sanity"); __ xoris(d, l, uimms); }
1735 else { __ xori(d, l, uimm); }
1736 break;
1737
1738 default: ShouldNotReachHere();
1739 }
1740 } else {
1741 assert(right->is_register(), "right should be in register");
1742
1743 if (dest->is_single_cpu()) {
1744 switch (code) {
1745 case lir_logic_and: __ andr(dest->as_register(), left->as_register(), right->as_register()); break;
1746 case lir_logic_or: __ orr (dest->as_register(), left->as_register(), right->as_register()); break;
1747 case lir_logic_xor: __ xorr(dest->as_register(), left->as_register(), right->as_register()); break;
1748 default: ShouldNotReachHere();
1749 }
1750 } else {
1751 Register l = (left->is_single_cpu() && left->is_oop_register()) ? left->as_register() :
1752 left->as_register_lo();
1753 Register r = (right->is_single_cpu() && right->is_oop_register()) ? right->as_register() :
1754 right->as_register_lo();
1755
1756 switch (code) {
1757 case lir_logic_and: __ andr(dest->as_register_lo(), l, r); break;
1758 case lir_logic_or: __ orr (dest->as_register_lo(), l, r); break;
1759 case lir_logic_xor: __ xorr(dest->as_register_lo(), l, r); break;
1760 default: ShouldNotReachHere();
1761 }
1762 }
1763 }
1764 }
1765
1766
1767 int LIR_Assembler::shift_amount(BasicType t) {
1768 int elem_size = type2aelembytes(t);
1769 switch (elem_size) {
1770 case 1 : return 0;
1771 case 2 : return 1;
1772 case 4 : return 2;
1773 case 8 : return 3;
1774 }
1775 ShouldNotReachHere();
1776 return -1;
1777 }
1778
1779
1780 void LIR_Assembler::throw_op(LIR_Opr exceptionPC, LIR_Opr exceptionOop, CodeEmitInfo* info) {
1781 info->add_register_oop(exceptionOop);
1782
1783 // Reuse the debug info from the safepoint poll for the throw op itself.
1784 address pc_for_athrow = __ pc();
1785 int pc_for_athrow_offset = __ offset();
1786 //RelocationHolder rspec = internal_word_Relocation::spec(pc_for_athrow);
1787 //__ relocate(rspec);
1788 //__ load_const(exceptionPC->as_register(), pc_for_athrow, R0);
1789 __ calculate_address_from_global_toc(exceptionPC->as_register(), pc_for_athrow, true, true, /*add_relocation*/ true);
1790 add_call_info(pc_for_athrow_offset, info); // for exception handler
1791
1792 address stub = Runtime1::entry_for(compilation()->has_fpu_code() ? StubId::c1_handle_exception_id
1793 : StubId::c1_handle_exception_nofpu_id);
1794 //__ load_const_optimized(R0, stub);
1795 __ add_const_optimized(R0, R29_TOC, MacroAssembler::offset_to_global_toc(stub));
1796 __ mtctr(R0);
1797 __ bctr();
1798 }
1799
1800
1801 void LIR_Assembler::unwind_op(LIR_Opr exceptionOop) {
1802 // Note: Not used with EnableDebuggingOnDemand.
1803 assert(exceptionOop->as_register() == R3, "should match");
1804 __ b(_unwind_handler_entry);
1805 }
1806
1807
1808 void LIR_Assembler::emit_arraycopy(LIR_OpArrayCopy* op) {
1809 Register src = op->src()->as_register();
1810 Register dst = op->dst()->as_register();
1811 Register src_pos = op->src_pos()->as_register();
1812 Register dst_pos = op->dst_pos()->as_register();
1813 Register length = op->length()->as_register();
1814 Register tmp = op->tmp()->as_register();
1815 Register tmp2 = R0;
1816
1817 int flags = op->flags();
1818 ciArrayKlass* default_type = op->expected_type();
1819 BasicType basic_type = (default_type != nullptr) ? default_type->element_type()->basic_type() : T_ILLEGAL;
1820 if (basic_type == T_ARRAY) basic_type = T_OBJECT;
1821
1822 // Set up the arraycopy stub information.
1823 ArrayCopyStub* stub = op->stub();
1824
1825 // Always do stub if no type information is available. It's ok if
1826 // the known type isn't loaded since the code sanity checks
1827 // in debug mode and the type isn't required when we know the exact type
1828 // also check that the type is an array type.
1829 if (default_type == nullptr) {
1830 assert(src->is_nonvolatile() && src_pos->is_nonvolatile() && dst->is_nonvolatile() && dst_pos->is_nonvolatile() &&
1831 length->is_nonvolatile(), "must preserve");
1832 address copyfunc_addr = StubRoutines::generic_arraycopy();
1833 assert(copyfunc_addr != nullptr, "generic arraycopy stub required");
1834
1835 // 3 parms are int. Convert to long.
1836 __ mr(R3_ARG1, src);
1837 __ extsw(R4_ARG2, src_pos);
1838 __ mr(R5_ARG3, dst);
1839 __ extsw(R6_ARG4, dst_pos);
1840 __ extsw(R7_ARG5, length);
1841
1842 #ifndef PRODUCT
1843 if (PrintC1Statistics) {
1844 address counter = (address)&Runtime1::_generic_arraycopystub_cnt;
1845 int simm16_offs = __ load_const_optimized(tmp, counter, tmp2, true);
1846 __ lwz(R11_scratch1, simm16_offs, tmp);
1847 __ addi(R11_scratch1, R11_scratch1, 1);
1848 __ stw(R11_scratch1, simm16_offs, tmp);
1849 }
1850 #endif
1851 __ call_c(copyfunc_addr, relocInfo::runtime_call_type);
1852
1853 __ nand(tmp, R3_RET, R3_RET);
1854 __ subf(length, tmp, length);
1855 __ add(src_pos, tmp, src_pos);
1856 __ add(dst_pos, tmp, dst_pos);
1857
1858 __ cmpwi(CR0, R3_RET, 0);
1859 __ bc_far_optimized(Assembler::bcondCRbiIs1, __ bi0(CR0, Assembler::less), *stub->entry());
1860 __ bind(*stub->continuation());
1861 return;
1862 }
1863
1864 assert(default_type != nullptr && default_type->is_array_klass() && default_type->is_loaded(), "must be true at this point");
1865 Label cont, slow, copyfunc;
1866
1867 bool simple_check_flag_set = flags & (LIR_OpArrayCopy::src_null_check |
1868 LIR_OpArrayCopy::dst_null_check |
1869 LIR_OpArrayCopy::src_pos_positive_check |
1870 LIR_OpArrayCopy::dst_pos_positive_check |
1871 LIR_OpArrayCopy::length_positive_check);
1872
1873 // Use only one conditional branch for simple checks.
1874 if (simple_check_flag_set) {
1875 ConditionRegister combined_check = CR1, tmp_check = CR1;
1876
1877 // Make sure src and dst are non-null.
1878 if (flags & LIR_OpArrayCopy::src_null_check) {
1879 __ cmpdi(combined_check, src, 0);
1880 tmp_check = CR0;
1881 }
1882
1883 if (flags & LIR_OpArrayCopy::dst_null_check) {
1884 __ cmpdi(tmp_check, dst, 0);
1885 if (tmp_check != combined_check) {
1886 __ cror(combined_check, Assembler::equal, tmp_check, Assembler::equal);
1887 }
1888 tmp_check = CR0;
1889 }
1890
1891 // Clear combined_check.eq if not already used.
1892 if (tmp_check == combined_check) {
1893 __ crandc(combined_check, Assembler::equal, combined_check, Assembler::equal);
1894 tmp_check = CR0;
1895 }
1896
1897 if (flags & LIR_OpArrayCopy::src_pos_positive_check) {
1898 // Test src_pos register.
1899 __ cmpwi(tmp_check, src_pos, 0);
1900 __ cror(combined_check, Assembler::equal, tmp_check, Assembler::less);
1901 }
1902
1903 if (flags & LIR_OpArrayCopy::dst_pos_positive_check) {
1904 // Test dst_pos register.
1905 __ cmpwi(tmp_check, dst_pos, 0);
1906 __ cror(combined_check, Assembler::equal, tmp_check, Assembler::less);
1907 }
1908
1909 if (flags & LIR_OpArrayCopy::length_positive_check) {
1910 // Make sure length isn't negative.
1911 __ cmpwi(tmp_check, length, 0);
1912 __ cror(combined_check, Assembler::equal, tmp_check, Assembler::less);
1913 }
1914
1915 __ beq(combined_check, slow);
1916 }
1917
1918 // If the compiler was not able to prove that exact type of the source or the destination
1919 // of the arraycopy is an array type, check at runtime if the source or the destination is
1920 // an instance type.
1921 if (flags & LIR_OpArrayCopy::type_check) {
1922 if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
1923 __ load_klass(tmp, dst);
1924 __ lwz(tmp2, in_bytes(Klass::layout_helper_offset()), tmp);
1925 __ cmpwi(CR0, tmp2, Klass::_lh_neutral_value);
1926 __ bge(CR0, slow);
1927 }
1928
1929 if (!(flags & LIR_OpArrayCopy::src_objarray)) {
1930 __ load_klass(tmp, src);
1931 __ lwz(tmp2, in_bytes(Klass::layout_helper_offset()), tmp);
1932 __ cmpwi(CR0, tmp2, Klass::_lh_neutral_value);
1933 __ bge(CR0, slow);
1934 }
1935 }
1936
1937 // Higher 32bits must be null.
1938 __ extsw(length, length);
1939
1940 __ extsw(src_pos, src_pos);
1941 if (flags & LIR_OpArrayCopy::src_range_check) {
1942 __ lwz(tmp2, arrayOopDesc::length_offset_in_bytes(), src);
1943 __ add(tmp, length, src_pos);
1944 __ cmpld(CR0, tmp2, tmp);
1945 __ ble(CR0, slow);
1946 }
1947
1948 __ extsw(dst_pos, dst_pos);
1949 if (flags & LIR_OpArrayCopy::dst_range_check) {
1950 __ lwz(tmp2, arrayOopDesc::length_offset_in_bytes(), dst);
1951 __ add(tmp, length, dst_pos);
1952 __ cmpld(CR0, tmp2, tmp);
1953 __ ble(CR0, slow);
1954 }
1955
1956 int shift = shift_amount(basic_type);
1957
1958 if (!(flags & LIR_OpArrayCopy::type_check)) {
1959 if (stub != nullptr) {
1960 __ b(cont);
1961 __ bind(slow);
1962 __ b(*stub->entry());
1963 }
1964 } else {
1965 // We don't know the array types are compatible.
1966 if (basic_type != T_OBJECT) {
1967 // Simple test for basic type arrays.
1968 __ cmp_klasses_from_objects(CR0, src, dst, tmp, tmp2);
1969 __ beq(CR0, cont);
1970 } else {
1971 // For object arrays, if src is a sub class of dst then we can
1972 // safely do the copy.
1973 address copyfunc_addr = StubRoutines::checkcast_arraycopy();
1974
1975 const Register sub_klass = R5, super_klass = R4; // like CheckCast/InstanceOf
1976 assert_different_registers(tmp, tmp2, sub_klass, super_klass);
1977
1978 __ load_klass(sub_klass, src);
1979 __ load_klass(super_klass, dst);
1980
1981 __ check_klass_subtype_fast_path(sub_klass, super_klass, tmp, tmp2,
1982 &cont, copyfunc_addr != nullptr ? ©func : &slow, nullptr);
1983
1984 address slow_stc = Runtime1::entry_for(StubId::c1_slow_subtype_check_id);
1985 //__ load_const_optimized(tmp, slow_stc, tmp2);
1986 __ calculate_address_from_global_toc(tmp, slow_stc, true, true, false);
1987 __ mtctr(tmp);
1988 __ bctrl(); // sets CR0
1989 __ beq(CR0, cont);
1990
1991 if (copyfunc_addr != nullptr) { // Use stub if available.
1992 __ bind(copyfunc);
1993 // Src is not a sub class of dst so we have to do a
1994 // per-element check.
1995 int mask = LIR_OpArrayCopy::src_objarray|LIR_OpArrayCopy::dst_objarray;
1996 if ((flags & mask) != mask) {
1997 assert(flags & mask, "one of the two should be known to be an object array");
1998
1999 if (!(flags & LIR_OpArrayCopy::src_objarray)) {
2000 __ load_klass(tmp, src);
2001 } else if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
2002 __ load_klass(tmp, dst);
2003 }
2004
2005 __ lwz(tmp2, in_bytes(Klass::layout_helper_offset()), tmp);
2006
2007 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2008 __ load_const_optimized(tmp, objArray_lh);
2009 __ cmpw(CR0, tmp, tmp2);
2010 __ bne(CR0, slow);
2011 }
2012
2013 Register src_ptr = R3_ARG1;
2014 Register dst_ptr = R4_ARG2;
2015 Register len = R5_ARG3;
2016 Register chk_off = R6_ARG4;
2017 Register super_k = R7_ARG5;
2018
2019 __ addi(src_ptr, src, arrayOopDesc::base_offset_in_bytes(basic_type));
2020 __ addi(dst_ptr, dst, arrayOopDesc::base_offset_in_bytes(basic_type));
2021 if (shift == 0) {
2022 __ add(src_ptr, src_pos, src_ptr);
2023 __ add(dst_ptr, dst_pos, dst_ptr);
2024 } else {
2025 __ sldi(tmp, src_pos, shift);
2026 __ sldi(tmp2, dst_pos, shift);
2027 __ add(src_ptr, tmp, src_ptr);
2028 __ add(dst_ptr, tmp2, dst_ptr);
2029 }
2030
2031 __ load_klass(tmp, dst);
2032 __ mr(len, length);
2033
2034 int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
2035 __ ld(super_k, ek_offset, tmp);
2036
2037 int sco_offset = in_bytes(Klass::super_check_offset_offset());
2038 __ lwz(chk_off, sco_offset, super_k);
2039
2040 __ call_c(copyfunc_addr, relocInfo::runtime_call_type);
2041
2042 #ifndef PRODUCT
2043 if (PrintC1Statistics) {
2044 Label failed;
2045 __ cmpwi(CR0, R3_RET, 0);
2046 __ bne(CR0, failed);
2047 address counter = (address)&Runtime1::_arraycopy_checkcast_cnt;
2048 int simm16_offs = __ load_const_optimized(tmp, counter, tmp2, true);
2049 __ lwz(R11_scratch1, simm16_offs, tmp);
2050 __ addi(R11_scratch1, R11_scratch1, 1);
2051 __ stw(R11_scratch1, simm16_offs, tmp);
2052 __ bind(failed);
2053 }
2054 #endif
2055
2056 __ nand(tmp, R3_RET, R3_RET);
2057 __ cmpwi(CR0, R3_RET, 0);
2058 __ beq(CR0, *stub->continuation());
2059
2060 #ifndef PRODUCT
2061 if (PrintC1Statistics) {
2062 address counter = (address)&Runtime1::_arraycopy_checkcast_attempt_cnt;
2063 int simm16_offs = __ load_const_optimized(tmp, counter, tmp2, true);
2064 __ lwz(R11_scratch1, simm16_offs, tmp);
2065 __ addi(R11_scratch1, R11_scratch1, 1);
2066 __ stw(R11_scratch1, simm16_offs, tmp);
2067 }
2068 #endif
2069
2070 __ subf(length, tmp, length);
2071 __ add(src_pos, tmp, src_pos);
2072 __ add(dst_pos, tmp, dst_pos);
2073 }
2074 }
2075 __ bind(slow);
2076 __ b(*stub->entry());
2077 }
2078 __ bind(cont);
2079
2080 #ifdef ASSERT
2081 if (basic_type != T_OBJECT || !(flags & LIR_OpArrayCopy::type_check)) {
2082 // Sanity check the known type with the incoming class. For the
2083 // primitive case the types must match exactly with src.klass and
2084 // dst.klass each exactly matching the default type. For the
2085 // object array case, if no type check is needed then either the
2086 // dst type is exactly the expected type and the src type is a
2087 // subtype which we can't check or src is the same array as dst
2088 // but not necessarily exactly of type default_type.
2089 Label known_ok, halt;
2090 metadata2reg(default_type->constant_encoding(), tmp);
2091 __ cmp_klass(CR0, dst, tmp, R11_scratch1, R12_scratch2);
2092 if (basic_type != T_OBJECT) {
2093 __ bne(CR0, halt);
2094 __ cmp_klass(CR0, src, tmp, R11_scratch1, R12_scratch2);
2095 __ beq(CR0, known_ok);
2096 } else {
2097 __ beq(CR0, known_ok);
2098 __ cmpw(CR0, src, dst);
2099 __ beq(CR0, known_ok);
2100 }
2101 __ bind(halt);
2102 __ stop("incorrect type information in arraycopy");
2103 __ bind(known_ok);
2104 }
2105 #endif
2106
2107 #ifndef PRODUCT
2108 if (PrintC1Statistics) {
2109 address counter = Runtime1::arraycopy_count_address(basic_type);
2110 int simm16_offs = __ load_const_optimized(tmp, counter, tmp2, true);
2111 __ lwz(R11_scratch1, simm16_offs, tmp);
2112 __ addi(R11_scratch1, R11_scratch1, 1);
2113 __ stw(R11_scratch1, simm16_offs, tmp);
2114 }
2115 #endif
2116
2117 Register src_ptr = R3_ARG1;
2118 Register dst_ptr = R4_ARG2;
2119 Register len = R5_ARG3;
2120
2121 __ addi(src_ptr, src, arrayOopDesc::base_offset_in_bytes(basic_type));
2122 __ addi(dst_ptr, dst, arrayOopDesc::base_offset_in_bytes(basic_type));
2123 if (shift == 0) {
2124 __ add(src_ptr, src_pos, src_ptr);
2125 __ add(dst_ptr, dst_pos, dst_ptr);
2126 } else {
2127 __ sldi(tmp, src_pos, shift);
2128 __ sldi(tmp2, dst_pos, shift);
2129 __ add(src_ptr, tmp, src_ptr);
2130 __ add(dst_ptr, tmp2, dst_ptr);
2131 }
2132
2133 bool disjoint = (flags & LIR_OpArrayCopy::overlapping) == 0;
2134 bool aligned = (flags & LIR_OpArrayCopy::unaligned) == 0;
2135 const char *name;
2136 address entry = StubRoutines::select_arraycopy_function(basic_type, aligned, disjoint, name, false);
2137
2138 // Arraycopy stubs takes a length in number of elements, so don't scale it.
2139 __ mr(len, length);
2140 __ call_c(entry, relocInfo::runtime_call_type);
2141
2142 if (stub != nullptr) {
2143 __ bind(*stub->continuation());
2144 }
2145 }
2146
2147
2148 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
2149 if (dest->is_single_cpu()) {
2150 __ rldicl(tmp->as_register(), count->as_register(), 0, 64-5);
2151 #ifdef _LP64
2152 if (left->type() == T_OBJECT) {
2153 switch (code) {
2154 case lir_shl: __ sld(dest->as_register(), left->as_register(), tmp->as_register()); break;
2155 case lir_shr: __ srad(dest->as_register(), left->as_register(), tmp->as_register()); break;
2156 case lir_ushr: __ srd(dest->as_register(), left->as_register(), tmp->as_register()); break;
2157 default: ShouldNotReachHere();
2158 }
2159 } else
2160 #endif
2161 switch (code) {
2162 case lir_shl: __ slw(dest->as_register(), left->as_register(), tmp->as_register()); break;
2163 case lir_shr: __ sraw(dest->as_register(), left->as_register(), tmp->as_register()); break;
2164 case lir_ushr: __ srw(dest->as_register(), left->as_register(), tmp->as_register()); break;
2165 default: ShouldNotReachHere();
2166 }
2167 } else {
2168 __ rldicl(tmp->as_register(), count->as_register(), 0, 64-6);
2169 switch (code) {
2170 case lir_shl: __ sld(dest->as_register_lo(), left->as_register_lo(), tmp->as_register()); break;
2171 case lir_shr: __ srad(dest->as_register_lo(), left->as_register_lo(), tmp->as_register()); break;
2172 case lir_ushr: __ srd(dest->as_register_lo(), left->as_register_lo(), tmp->as_register()); break;
2173 default: ShouldNotReachHere();
2174 }
2175 }
2176 }
2177
2178
2179 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, jint count, LIR_Opr dest) {
2180 #ifdef _LP64
2181 if (left->type() == T_OBJECT) {
2182 count = count & 63; // Shouldn't shift by more than sizeof(intptr_t).
2183 if (count == 0) { __ mr_if_needed(dest->as_register_lo(), left->as_register()); }
2184 else {
2185 switch (code) {
2186 case lir_shl: __ sldi(dest->as_register_lo(), left->as_register(), count); break;
2187 case lir_shr: __ sradi(dest->as_register_lo(), left->as_register(), count); break;
2188 case lir_ushr: __ srdi(dest->as_register_lo(), left->as_register(), count); break;
2189 default: ShouldNotReachHere();
2190 }
2191 }
2192 return;
2193 }
2194 #endif
2195
2196 if (dest->is_single_cpu()) {
2197 count = count & 0x1F; // Java spec
2198 if (count == 0) { __ mr_if_needed(dest->as_register(), left->as_register()); }
2199 else {
2200 switch (code) {
2201 case lir_shl: __ slwi(dest->as_register(), left->as_register(), count); break;
2202 case lir_shr: __ srawi(dest->as_register(), left->as_register(), count); break;
2203 case lir_ushr: __ srwi(dest->as_register(), left->as_register(), count); break;
2204 default: ShouldNotReachHere();
2205 }
2206 }
2207 } else if (dest->is_double_cpu()) {
2208 count = count & 63; // Java spec
2209 if (count == 0) { __ mr_if_needed(dest->as_pointer_register(), left->as_pointer_register()); }
2210 else {
2211 switch (code) {
2212 case lir_shl: __ sldi(dest->as_pointer_register(), left->as_pointer_register(), count); break;
2213 case lir_shr: __ sradi(dest->as_pointer_register(), left->as_pointer_register(), count); break;
2214 case lir_ushr: __ srdi(dest->as_pointer_register(), left->as_pointer_register(), count); break;
2215 default: ShouldNotReachHere();
2216 }
2217 }
2218 } else {
2219 ShouldNotReachHere();
2220 }
2221 }
2222
2223
2224 void LIR_Assembler::emit_alloc_obj(LIR_OpAllocObj* op) {
2225 if (op->init_check()) {
2226 if (!os::zero_page_read_protected() || !ImplicitNullChecks) {
2227 explicit_null_check(op->klass()->as_register(), op->stub()->info());
2228 } else {
2229 add_debug_info_for_null_check_here(op->stub()->info());
2230 }
2231 __ lbz(op->tmp1()->as_register(),
2232 in_bytes(InstanceKlass::init_state_offset()), op->klass()->as_register());
2233 // acquire barrier included in membar_storestore() which follows the allocation immediately.
2234 __ cmpwi(CR0, op->tmp1()->as_register(), InstanceKlass::fully_initialized);
2235 __ bc_far_optimized(Assembler::bcondCRbiIs0, __ bi0(CR0, Assembler::equal), *op->stub()->entry());
2236 }
2237 __ allocate_object(op->obj()->as_register(),
2238 op->tmp1()->as_register(),
2239 op->tmp2()->as_register(),
2240 op->tmp3()->as_register(),
2241 op->header_size(),
2242 op->object_size(),
2243 op->klass()->as_register(),
2244 *op->stub()->entry());
2245
2246 __ bind(*op->stub()->continuation());
2247 __ verify_oop(op->obj()->as_register(), FILE_AND_LINE);
2248 }
2249
2250
2251 void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
2252 LP64_ONLY( __ extsw(op->len()->as_register(), op->len()->as_register()); )
2253 if (UseSlowPath ||
2254 (!UseFastNewObjectArray && (is_reference_type(op->type()))) ||
2255 (!UseFastNewTypeArray && (!is_reference_type(op->type())))) {
2256 __ b(*op->stub()->entry());
2257 } else {
2258 __ allocate_array(op->obj()->as_register(),
2259 op->len()->as_register(),
2260 op->tmp1()->as_register(),
2261 op->tmp2()->as_register(),
2262 op->tmp3()->as_register(),
2263 arrayOopDesc::base_offset_in_bytes(op->type()),
2264 type2aelembytes(op->type()),
2265 op->klass()->as_register(),
2266 *op->stub()->entry(),
2267 op->zero_array());
2268 }
2269 __ bind(*op->stub()->continuation());
2270 }
2271
2272
2273 void LIR_Assembler::type_profile_helper(Register mdo, int mdo_offset_bias,
2274 ciMethodData *md, ciProfileData *data,
2275 Register recv, Register tmp1, Label* update_done) {
2276 uint i;
2277 for (i = 0; i < VirtualCallData::row_limit(); i++) {
2278 Label next_test;
2279 // See if the receiver is receiver[n].
2280 __ ld(tmp1, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)) - mdo_offset_bias, mdo);
2281 __ verify_klass_ptr(tmp1);
2282 __ cmpd(CR0, recv, tmp1);
2283 __ bne(CR0, next_test);
2284
2285 __ ld(tmp1, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)) - mdo_offset_bias, mdo);
2286 __ addi(tmp1, tmp1, DataLayout::counter_increment);
2287 __ std(tmp1, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)) - mdo_offset_bias, mdo);
2288 __ b(*update_done);
2289
2290 __ bind(next_test);
2291 }
2292
2293 // Didn't find receiver; find next empty slot and fill it in.
2294 for (i = 0; i < VirtualCallData::row_limit(); i++) {
2295 Label next_test;
2296 __ ld(tmp1, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)) - mdo_offset_bias, mdo);
2297 __ cmpdi(CR0, tmp1, 0);
2298 __ bne(CR0, next_test);
2299 __ li(tmp1, DataLayout::counter_increment);
2300 __ std(recv, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)) - mdo_offset_bias, mdo);
2301 __ std(tmp1, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)) - mdo_offset_bias, mdo);
2302 __ b(*update_done);
2303
2304 __ bind(next_test);
2305 }
2306 }
2307
2308
2309 void LIR_Assembler::setup_md_access(ciMethod* method, int bci,
2310 ciMethodData*& md, ciProfileData*& data, int& mdo_offset_bias) {
2311 md = method->method_data_or_null();
2312 assert(md != nullptr, "Sanity");
2313 data = md->bci_to_data(bci);
2314 assert(data != nullptr, "need data for checkcast");
2315 assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
2316 if (!Assembler::is_simm16(md->byte_offset_of_slot(data, DataLayout::header_offset()) + data->size_in_bytes())) {
2317 // The offset is large so bias the mdo by the base of the slot so
2318 // that the ld can use simm16s to reference the slots of the data.
2319 mdo_offset_bias = md->byte_offset_of_slot(data, DataLayout::header_offset());
2320 }
2321 }
2322
2323
2324 void LIR_Assembler::emit_typecheck_helper(LIR_OpTypeCheck *op, Label* success, Label* failure, Label* obj_is_null) {
2325 const Register obj = op->object()->as_register(); // Needs to live in this register at safepoint (patching stub).
2326 Register k_RInfo = op->tmp1()->as_register();
2327 Register klass_RInfo = op->tmp2()->as_register();
2328 Register Rtmp1 = op->tmp3()->as_register();
2329 Register dst = op->result_opr()->as_register();
2330 ciKlass* k = op->klass();
2331 bool should_profile = op->should_profile();
2332 // Attention: do_temp(opTypeCheck->_object) is not used, i.e. obj may be same as one of the temps.
2333 bool reg_conflict = false;
2334 if (obj == k_RInfo) {
2335 k_RInfo = dst;
2336 reg_conflict = true;
2337 } else if (obj == klass_RInfo) {
2338 klass_RInfo = dst;
2339 reg_conflict = true;
2340 } else if (obj == Rtmp1) {
2341 Rtmp1 = dst;
2342 reg_conflict = true;
2343 }
2344 assert_different_registers(obj, k_RInfo, klass_RInfo, Rtmp1);
2345
2346 ciMethodData* md = nullptr;
2347 ciProfileData* data = nullptr;
2348 int mdo_offset_bias = 0;
2349 if (should_profile) {
2350 ciMethod* method = op->profiled_method();
2351 assert(method != nullptr, "Should have method");
2352 setup_md_access(method, op->profiled_bci(), md, data, mdo_offset_bias);
2353
2354 Register mdo = k_RInfo;
2355 Register data_val = Rtmp1;
2356 Label not_null;
2357 metadata2reg(md->constant_encoding(), mdo);
2358 __ add_const_optimized(mdo, mdo, mdo_offset_bias, R0);
2359 __ cmpdi(CR0, obj, 0);
2360 __ bne(CR0, not_null);
2361 __ lbz(data_val, md->byte_offset_of_slot(data, DataLayout::flags_offset()) - mdo_offset_bias, mdo);
2362 __ ori(data_val, data_val, BitData::null_seen_byte_constant());
2363 __ stb(data_val, md->byte_offset_of_slot(data, DataLayout::flags_offset()) - mdo_offset_bias, mdo);
2364 __ b(*obj_is_null);
2365 __ bind(not_null);
2366
2367 Label update_done;
2368 Register recv = klass_RInfo;
2369 __ load_klass(recv, obj);
2370 type_profile_helper(mdo, mdo_offset_bias, md, data, recv, Rtmp1, &update_done);
2371 const int slot_offset = md->byte_offset_of_slot(data, CounterData::count_offset()) - mdo_offset_bias;
2372 __ ld(Rtmp1, slot_offset, mdo);
2373 __ addi(Rtmp1, Rtmp1, DataLayout::counter_increment);
2374 __ std(Rtmp1, slot_offset, mdo);
2375 __ bind(update_done);
2376 } else {
2377 __ cmpdi(CR0, obj, 0);
2378 __ beq(CR0, *obj_is_null);
2379 }
2380
2381 // get object class
2382 __ load_klass(klass_RInfo, obj);
2383
2384 if (k->is_loaded()) {
2385 metadata2reg(k->constant_encoding(), k_RInfo);
2386 } else {
2387 klass2reg_with_patching(k_RInfo, op->info_for_patch());
2388 }
2389
2390 if (op->fast_check()) {
2391 assert_different_registers(klass_RInfo, k_RInfo);
2392 __ cmpd(CR0, k_RInfo, klass_RInfo);
2393 __ beq(CR0, *success);
2394 // Fall through to failure case.
2395 } else {
2396 bool need_slow_path = true;
2397 if (k->is_loaded()) {
2398 if ((int) k->super_check_offset() != in_bytes(Klass::secondary_super_cache_offset())) {
2399 need_slow_path = false;
2400 }
2401 // Perform the fast part of the checking logic.
2402 __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, R0, (need_slow_path ? success : nullptr),
2403 failure, nullptr, RegisterOrConstant(k->super_check_offset()));
2404 } else {
2405 // Perform the fast part of the checking logic.
2406 __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, R0, success, failure);
2407 }
2408 if (!need_slow_path) {
2409 __ b(*success);
2410 } else {
2411 // Call out-of-line instance of __ check_klass_subtype_slow_path(...):
2412 address entry = Runtime1::entry_for(StubId::c1_slow_subtype_check_id);
2413 // Stub needs fixed registers (tmp1-3).
2414 Register original_k_RInfo = op->tmp1()->as_register();
2415 Register original_klass_RInfo = op->tmp2()->as_register();
2416 Register original_Rtmp1 = op->tmp3()->as_register();
2417 bool keep_obj_alive = reg_conflict && (op->code() == lir_checkcast);
2418 if (keep_obj_alive && (obj != original_Rtmp1)) { __ mr(R0, obj); }
2419 __ mr_if_needed(original_k_RInfo, k_RInfo);
2420 __ mr_if_needed(original_klass_RInfo, klass_RInfo);
2421 if (keep_obj_alive) { __ mr(dst, (obj == original_Rtmp1) ? obj : R0); }
2422 //__ load_const_optimized(original_Rtmp1, entry, R0);
2423 __ calculate_address_from_global_toc(original_Rtmp1, entry, true, true, false);
2424 __ mtctr(original_Rtmp1);
2425 __ bctrl(); // sets CR0
2426 if (keep_obj_alive) { __ mr(obj, dst); }
2427 __ beq(CR0, *success);
2428 // Fall through to failure case.
2429 }
2430 }
2431
2432 __ bind(*failure);
2433 }
2434
2435
2436 void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
2437 LIR_Code code = op->code();
2438 if (code == lir_store_check) {
2439 Register value = op->object()->as_register();
2440 Register array = op->array()->as_register();
2441 Register k_RInfo = op->tmp1()->as_register();
2442 Register klass_RInfo = op->tmp2()->as_register();
2443 Register Rtmp1 = op->tmp3()->as_register();
2444 bool should_profile = op->should_profile();
2445
2446 __ verify_oop(value, FILE_AND_LINE);
2447 CodeStub* stub = op->stub();
2448 // Check if it needs to be profiled.
2449 ciMethodData* md = nullptr;
2450 ciProfileData* data = nullptr;
2451 int mdo_offset_bias = 0;
2452 if (should_profile) {
2453 ciMethod* method = op->profiled_method();
2454 assert(method != nullptr, "Should have method");
2455 setup_md_access(method, op->profiled_bci(), md, data, mdo_offset_bias);
2456 }
2457
2458 Label done;
2459
2460 if (should_profile) {
2461 Label not_null;
2462 Register mdo = k_RInfo;
2463 Register data_val = Rtmp1;
2464 metadata2reg(md->constant_encoding(), mdo);
2465 __ add_const_optimized(mdo, mdo, mdo_offset_bias, R0);
2466 __ cmpdi(CR0, value, 0);
2467 __ bne(CR0, not_null);
2468 __ lbz(data_val, md->byte_offset_of_slot(data, DataLayout::flags_offset()) - mdo_offset_bias, mdo);
2469 __ ori(data_val, data_val, BitData::null_seen_byte_constant());
2470 __ stb(data_val, md->byte_offset_of_slot(data, DataLayout::flags_offset()) - mdo_offset_bias, mdo);
2471 __ b(done);
2472 __ bind(not_null);
2473
2474 Label update_done;
2475 Register recv = klass_RInfo;
2476 __ load_klass(recv, value);
2477 type_profile_helper(mdo, mdo_offset_bias, md, data, recv, Rtmp1, &update_done);
2478 const int slot_offset = md->byte_offset_of_slot(data, CounterData::count_offset()) - mdo_offset_bias;
2479 __ ld(Rtmp1, slot_offset, mdo);
2480 __ addi(Rtmp1, Rtmp1, DataLayout::counter_increment);
2481 __ std(Rtmp1, slot_offset, mdo);
2482 __ bind(update_done);
2483 } else {
2484 __ cmpdi(CR0, value, 0);
2485 __ beq(CR0, done);
2486 }
2487 if (!os::zero_page_read_protected() || !ImplicitNullChecks) {
2488 explicit_null_check(array, op->info_for_exception());
2489 } else {
2490 add_debug_info_for_null_check_here(op->info_for_exception());
2491 }
2492 __ load_klass(k_RInfo, array);
2493 __ load_klass(klass_RInfo, value);
2494
2495 Label failure;
2496
2497 // Get instance klass.
2498 __ ld(k_RInfo, in_bytes(ObjArrayKlass::element_klass_offset()), k_RInfo);
2499 // Perform the fast part of the checking logic.
2500 __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, R0, &done, &failure, nullptr);
2501
2502 // Call out-of-line instance of __ check_klass_subtype_slow_path(...):
2503 const address slow_path = Runtime1::entry_for(StubId::c1_slow_subtype_check_id);
2504 //__ load_const_optimized(R0, slow_path);
2505 __ add_const_optimized(R0, R29_TOC, MacroAssembler::offset_to_global_toc(slow_path));
2506 __ mtctr(R0);
2507 __ bctrl(); // sets CR0
2508 __ beq(CR0, done);
2509
2510 __ bind(failure);
2511 __ b(*stub->entry());
2512 __ align(32, 12);
2513 __ bind(done);
2514
2515 } else if (code == lir_checkcast) {
2516 Label success, failure;
2517 emit_typecheck_helper(op, &success, /*fallthru*/&failure, &success);
2518 __ b(*op->stub()->entry());
2519 __ align(32, 12);
2520 __ bind(success);
2521 __ mr_if_needed(op->result_opr()->as_register(), op->object()->as_register());
2522 } else if (code == lir_instanceof) {
2523 Register dst = op->result_opr()->as_register();
2524 Label success, failure, done;
2525 emit_typecheck_helper(op, &success, /*fallthru*/&failure, &failure);
2526 __ li(dst, 0);
2527 __ b(done);
2528 __ align(32, 12);
2529 __ bind(success);
2530 __ li(dst, 1);
2531 __ bind(done);
2532 } else {
2533 ShouldNotReachHere();
2534 }
2535 }
2536
2537
2538 void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
2539 Register addr = op->addr()->as_pointer_register();
2540 Register cmp_value = noreg, new_value = noreg;
2541 bool is_64bit = false;
2542
2543 if (op->code() == lir_cas_long) {
2544 cmp_value = op->cmp_value()->as_register_lo();
2545 new_value = op->new_value()->as_register_lo();
2546 is_64bit = true;
2547 } else if (op->code() == lir_cas_int || op->code() == lir_cas_obj) {
2548 cmp_value = op->cmp_value()->as_register();
2549 new_value = op->new_value()->as_register();
2550 if (op->code() == lir_cas_obj) {
2551 if (UseCompressedOops) {
2552 Register t1 = op->tmp1()->as_register();
2553 Register t2 = op->tmp2()->as_register();
2554 cmp_value = __ encode_heap_oop(t1, cmp_value);
2555 new_value = __ encode_heap_oop(t2, new_value);
2556 } else {
2557 is_64bit = true;
2558 }
2559 }
2560 } else {
2561 Unimplemented();
2562 }
2563
2564 // There might be a volatile load before this Unsafe CAS.
2565 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2566 __ sync();
2567 } else {
2568 __ lwsync();
2569 }
2570
2571 if (is_64bit) {
2572 __ cmpxchgd(BOOL_RESULT, /*current_value=*/R0, cmp_value, new_value, addr,
2573 MacroAssembler::MemBarNone,
2574 MacroAssembler::cmpxchgx_hint_atomic_update(),
2575 noreg, nullptr, /*check without ldarx first*/true);
2576 } else {
2577 __ cmpxchgw(BOOL_RESULT, /*current_value=*/R0, cmp_value, new_value, addr,
2578 MacroAssembler::MemBarNone,
2579 MacroAssembler::cmpxchgx_hint_atomic_update(),
2580 noreg, nullptr, /*check without ldarx first*/true);
2581 }
2582
2583 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2584 __ isync();
2585 } else {
2586 __ sync();
2587 }
2588 }
2589
2590 void LIR_Assembler::breakpoint() {
2591 __ illtrap();
2592 }
2593
2594
2595 void LIR_Assembler::push(LIR_Opr opr) {
2596 Unimplemented();
2597 }
2598
2599 void LIR_Assembler::pop(LIR_Opr opr) {
2600 Unimplemented();
2601 }
2602
2603
2604 void LIR_Assembler::monitor_address(int monitor_no, LIR_Opr dst_opr) {
2605 Address mon_addr = frame_map()->address_for_monitor_lock(monitor_no);
2606 Register dst = dst_opr->as_register();
2607 Register reg = mon_addr.base();
2608 int offset = mon_addr.disp();
2609 // Compute pointer to BasicLock.
2610 __ add_const_optimized(dst, reg, offset);
2611 }
2612
2613
2614 void LIR_Assembler::emit_lock(LIR_OpLock* op) {
2615 Register obj = op->obj_opr()->as_register();
2616 Register hdr = op->hdr_opr()->as_register();
2617 Register lock = op->lock_opr()->as_register();
2618
2619 // Obj may not be an oop.
2620 if (op->code() == lir_lock) {
2621 MonitorEnterStub* stub = (MonitorEnterStub*)op->stub();
2622 // Add debug info for NullPointerException only if one is possible.
2623 if (op->info() != nullptr) {
2624 if (!os::zero_page_read_protected() || !ImplicitNullChecks) {
2625 explicit_null_check(obj, op->info());
2626 } else {
2627 add_debug_info_for_null_check_here(op->info());
2628 }
2629 }
2630 __ lock_object(hdr, obj, lock, op->scratch_opr()->as_register(), *op->stub()->entry());
2631 } else {
2632 assert (op->code() == lir_unlock, "Invalid code, expected lir_unlock");
2633 __ unlock_object(hdr, obj, lock, *op->stub()->entry());
2634 }
2635 __ bind(*op->stub()->continuation());
2636 }
2637
2638 void LIR_Assembler::emit_load_klass(LIR_OpLoadKlass* op) {
2639 Register obj = op->obj()->as_pointer_register();
2640 Register result = op->result_opr()->as_pointer_register();
2641
2642 CodeEmitInfo* info = op->info();
2643 if (info != nullptr) {
2644 if (!os::zero_page_read_protected() || !ImplicitNullChecks) {
2645 explicit_null_check(obj, info);
2646 } else {
2647 add_debug_info_for_null_check_here(info);
2648 }
2649 }
2650
2651 __ load_klass(result, obj);
2652 }
2653
2654 void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
2655 ciMethod* method = op->profiled_method();
2656 int bci = op->profiled_bci();
2657 ciMethod* callee = op->profiled_callee();
2658
2659 // Update counter for all call types.
2660 ciMethodData* md = method->method_data_or_null();
2661 assert(md != nullptr, "Sanity");
2662 ciProfileData* data = md->bci_to_data(bci);
2663 assert(data != nullptr && data->is_CounterData(), "need CounterData for calls");
2664 assert(op->mdo()->is_single_cpu(), "mdo must be allocated");
2665 Register mdo = op->mdo()->as_register();
2666 #ifdef _LP64
2667 assert(op->tmp1()->is_double_cpu(), "tmp1 must be allocated");
2668 Register tmp1 = op->tmp1()->as_register_lo();
2669 #else
2670 assert(op->tmp1()->is_single_cpu(), "tmp1 must be allocated");
2671 Register tmp1 = op->tmp1()->as_register();
2672 #endif
2673 metadata2reg(md->constant_encoding(), mdo);
2674 int mdo_offset_bias = 0;
2675 if (!Assembler::is_simm16(md->byte_offset_of_slot(data, CounterData::count_offset()) +
2676 data->size_in_bytes())) {
2677 // The offset is large so bias the mdo by the base of the slot so
2678 // that the ld can use simm16s to reference the slots of the data.
2679 mdo_offset_bias = md->byte_offset_of_slot(data, CounterData::count_offset());
2680 __ add_const_optimized(mdo, mdo, mdo_offset_bias, R0);
2681 }
2682
2683 // Perform additional virtual call profiling for invokevirtual and
2684 // invokeinterface bytecodes
2685 if (op->should_profile_receiver_type()) {
2686 assert(op->recv()->is_single_cpu(), "recv must be allocated");
2687 Register recv = op->recv()->as_register();
2688 assert_different_registers(mdo, tmp1, recv);
2689 assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
2690 ciKlass* known_klass = op->known_holder();
2691 if (C1OptimizeVirtualCallProfiling && known_klass != nullptr) {
2692 // We know the type that will be seen at this call site; we can
2693 // statically update the MethodData* rather than needing to do
2694 // dynamic tests on the receiver type.
2695
2696 // NOTE: we should probably put a lock around this search to
2697 // avoid collisions by concurrent compilations.
2698 ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
2699 uint i;
2700 for (i = 0; i < VirtualCallData::row_limit(); i++) {
2701 ciKlass* receiver = vc_data->receiver(i);
2702 if (known_klass->equals(receiver)) {
2703 __ ld(tmp1, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)) - mdo_offset_bias, mdo);
2704 __ addi(tmp1, tmp1, DataLayout::counter_increment);
2705 __ std(tmp1, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)) - mdo_offset_bias, mdo);
2706 return;
2707 }
2708 }
2709
2710 // Receiver type not found in profile data; select an empty slot.
2711
2712 // Note that this is less efficient than it should be because it
2713 // always does a write to the receiver part of the
2714 // VirtualCallData rather than just the first time.
2715 for (i = 0; i < VirtualCallData::row_limit(); i++) {
2716 ciKlass* receiver = vc_data->receiver(i);
2717 if (receiver == nullptr) {
2718 metadata2reg(known_klass->constant_encoding(), tmp1);
2719 __ std(tmp1, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)) - mdo_offset_bias, mdo);
2720
2721 __ ld(tmp1, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)) - mdo_offset_bias, mdo);
2722 __ addi(tmp1, tmp1, DataLayout::counter_increment);
2723 __ std(tmp1, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)) - mdo_offset_bias, mdo);
2724 return;
2725 }
2726 }
2727 } else {
2728 __ load_klass(recv, recv);
2729 Label update_done;
2730 type_profile_helper(mdo, mdo_offset_bias, md, data, recv, tmp1, &update_done);
2731 // Receiver did not match any saved receiver and there is no empty row for it.
2732 // Increment total counter to indicate polymorphic case.
2733 __ ld(tmp1, md->byte_offset_of_slot(data, CounterData::count_offset()) - mdo_offset_bias, mdo);
2734 __ addi(tmp1, tmp1, DataLayout::counter_increment);
2735 __ std(tmp1, md->byte_offset_of_slot(data, CounterData::count_offset()) - mdo_offset_bias, mdo);
2736
2737 __ bind(update_done);
2738 }
2739 } else {
2740 // Static call
2741 __ ld(tmp1, md->byte_offset_of_slot(data, CounterData::count_offset()) - mdo_offset_bias, mdo);
2742 __ addi(tmp1, tmp1, DataLayout::counter_increment);
2743 __ std(tmp1, md->byte_offset_of_slot(data, CounterData::count_offset()) - mdo_offset_bias, mdo);
2744 }
2745 }
2746
2747
2748 void LIR_Assembler::align_backward_branch_target() {
2749 __ align(32, 12); // Insert up to 3 nops to align with 32 byte boundary.
2750 }
2751
2752
2753 void LIR_Assembler::negate(LIR_Opr left, LIR_Opr dest, LIR_Opr tmp) {
2754 // tmp must be unused
2755 assert(tmp->is_illegal(), "wasting a register if tmp is allocated");
2756 assert(left->is_register(), "can only handle registers");
2757
2758 if (left->is_single_cpu()) {
2759 __ neg(dest->as_register(), left->as_register());
2760 } else if (left->is_single_fpu()) {
2761 __ fneg(dest->as_float_reg(), left->as_float_reg());
2762 } else if (left->is_double_fpu()) {
2763 __ fneg(dest->as_double_reg(), left->as_double_reg());
2764 } else {
2765 assert (left->is_double_cpu(), "Must be a long");
2766 __ neg(dest->as_register_lo(), left->as_register_lo());
2767 }
2768 }
2769
2770
2771 void LIR_Assembler::rt_call(LIR_Opr result, address dest,
2772 const LIR_OprList* args, LIR_Opr tmp, CodeEmitInfo* info) {
2773 // Stubs: Called via rt_call, but dest is a stub address (no FunctionDescriptor).
2774 if (dest == Runtime1::entry_for(StubId::c1_register_finalizer_id) ||
2775 dest == Runtime1::entry_for(StubId::c1_new_multi_array_id ) ||
2776 dest == Runtime1::entry_for(StubId::c1_is_instance_of_id )) {
2777 assert(CodeCache::contains(dest), "simplified call is only for special C1 stubs");
2778 //__ load_const_optimized(R0, dest);
2779 __ add_const_optimized(R0, R29_TOC, MacroAssembler::offset_to_global_toc(dest));
2780 __ mtctr(R0);
2781 __ bctrl();
2782 if (info != nullptr) {
2783 add_call_info_here(info);
2784 __ post_call_nop();
2785 }
2786 return;
2787 }
2788
2789 __ call_c(dest, relocInfo::runtime_call_type);
2790 assert(__ last_calls_return_pc() == __ pc(), "pcn not at return pc");
2791 if (info != nullptr) {
2792 add_call_info_here(info);
2793 __ post_call_nop();
2794 }
2795 }
2796
2797
2798 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {
2799 ShouldNotReachHere(); // Not needed on _LP64.
2800 }
2801
2802 void LIR_Assembler::membar() {
2803 __ fence();
2804 }
2805
2806 void LIR_Assembler::membar_acquire() {
2807 __ acquire();
2808 }
2809
2810 void LIR_Assembler::membar_release() {
2811 __ release();
2812 }
2813
2814 void LIR_Assembler::membar_loadload() {
2815 __ membar(Assembler::LoadLoad);
2816 }
2817
2818 void LIR_Assembler::membar_storestore() {
2819 __ membar(Assembler::StoreStore);
2820 }
2821
2822 void LIR_Assembler::membar_loadstore() {
2823 __ membar(Assembler::LoadStore);
2824 }
2825
2826 void LIR_Assembler::membar_storeload() {
2827 __ membar(Assembler::StoreLoad);
2828 }
2829
2830 void LIR_Assembler::on_spin_wait() {
2831 Unimplemented();
2832 }
2833
2834 void LIR_Assembler::leal(LIR_Opr addr_opr, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
2835 LIR_Address* addr = addr_opr->as_address_ptr();
2836 assert(addr->scale() == LIR_Address::times_1, "no scaling on this platform");
2837
2838 if (addr->index()->is_illegal()) {
2839 if (patch_code != lir_patch_none) {
2840 PatchingStub* patch = new PatchingStub(_masm, PatchingStub::access_field_id);
2841 __ load_const32(R0, 0); // patchable int
2842 __ add(dest->as_pointer_register(), addr->base()->as_pointer_register(), R0);
2843 patching_epilog(patch, patch_code, addr->base()->as_register(), info);
2844 } else {
2845 __ add_const_optimized(dest->as_pointer_register(), addr->base()->as_pointer_register(), addr->disp());
2846 }
2847 } else {
2848 assert(patch_code == lir_patch_none, "Patch code not supported");
2849 assert(addr->disp() == 0, "can't have both: index and disp");
2850 __ add(dest->as_pointer_register(), addr->index()->as_pointer_register(), addr->base()->as_pointer_register());
2851 }
2852 }
2853
2854
2855 void LIR_Assembler::get_thread(LIR_Opr result_reg) {
2856 ShouldNotReachHere();
2857 }
2858
2859
2860 #ifdef ASSERT
2861 // Emit run-time assertion.
2862 void LIR_Assembler::emit_assert(LIR_OpAssert* op) {
2863 Unimplemented();
2864 }
2865 #endif
2866
2867
2868 void LIR_Assembler::peephole(LIR_List* lir) {
2869 // Optimize instruction pairs before emitting.
2870 LIR_OpList* inst = lir->instructions_list();
2871 for (int i = 1; i < inst->length(); i++) {
2872 LIR_Op* op = inst->at(i);
2873
2874 // 2 register-register-moves
2875 if (op->code() == lir_move) {
2876 LIR_Opr in2 = ((LIR_Op1*)op)->in_opr(),
2877 res2 = ((LIR_Op1*)op)->result_opr();
2878 if (in2->is_register() && res2->is_register()) {
2879 LIR_Op* prev = inst->at(i - 1);
2880 if (prev && prev->code() == lir_move) {
2881 LIR_Opr in1 = ((LIR_Op1*)prev)->in_opr(),
2882 res1 = ((LIR_Op1*)prev)->result_opr();
2883 if (in1->is_same_register(res2) && in2->is_same_register(res1)) {
2884 inst->remove_at(i);
2885 }
2886 }
2887 }
2888 }
2889
2890 }
2891 return;
2892 }
2893
2894
2895 void LIR_Assembler::atomic_op(LIR_Code code, LIR_Opr src, LIR_Opr data, LIR_Opr dest, LIR_Opr tmp) {
2896 const LIR_Address *addr = src->as_address_ptr();
2897 assert(addr->disp() == 0 && addr->index()->is_illegal(), "use leal!");
2898 const Register Rptr = addr->base()->as_pointer_register(),
2899 Rtmp = tmp->as_register();
2900 Register Robj = noreg;
2901 if (data->is_oop()) {
2902 if (UseCompressedOops) {
2903 Robj = __ encode_heap_oop(Rtmp, data->as_register());
2904 } else {
2905 Robj = data->as_register();
2906 if (Robj == dest->as_register()) { // May happen with ZGC.
2907 __ mr(Rtmp, Robj);
2908 Robj = Rtmp;
2909 }
2910 }
2911 }
2912
2913 // There might be a volatile load before this Unsafe OP.
2914 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2915 __ sync();
2916 } else {
2917 __ lwsync();
2918 }
2919
2920 Label Lretry;
2921 __ bind(Lretry);
2922
2923 if (data->type() == T_INT) {
2924 const Register Rold = dest->as_register(),
2925 Rsrc = data->as_register();
2926 assert_different_registers(Rptr, Rtmp, Rold, Rsrc);
2927 __ lwarx(Rold, Rptr, MacroAssembler::cmpxchgx_hint_atomic_update());
2928 if (code == lir_xadd) {
2929 __ add(Rtmp, Rsrc, Rold);
2930 __ stwcx_(Rtmp, Rptr);
2931 } else {
2932 __ stwcx_(Rsrc, Rptr);
2933 }
2934 } else if (data->is_oop()) {
2935 assert(code == lir_xchg, "xadd for oops");
2936 const Register Rold = dest->as_register();
2937 assert_different_registers(Rptr, Rold, Robj);
2938 if (UseCompressedOops) {
2939 __ lwarx(Rold, Rptr, MacroAssembler::cmpxchgx_hint_atomic_update());
2940 __ stwcx_(Robj, Rptr);
2941 } else {
2942 __ ldarx(Rold, Rptr, MacroAssembler::cmpxchgx_hint_atomic_update());
2943 __ stdcx_(Robj, Rptr);
2944 }
2945 } else if (data->type() == T_LONG) {
2946 const Register Rold = dest->as_register_lo(),
2947 Rsrc = data->as_register_lo();
2948 assert_different_registers(Rptr, Rtmp, Rold, Rsrc);
2949 __ ldarx(Rold, Rptr, MacroAssembler::cmpxchgx_hint_atomic_update());
2950 if (code == lir_xadd) {
2951 __ add(Rtmp, Rsrc, Rold);
2952 __ stdcx_(Rtmp, Rptr);
2953 } else {
2954 __ stdcx_(Rsrc, Rptr);
2955 }
2956 } else {
2957 ShouldNotReachHere();
2958 }
2959
2960 if (UseStaticBranchPredictionInCompareAndSwapPPC64) {
2961 __ bne_predict_not_taken(CR0, Lretry);
2962 } else {
2963 __ bne( CR0, Lretry);
2964 }
2965
2966 if (UseCompressedOops && data->is_oop()) {
2967 __ decode_heap_oop(dest->as_register());
2968 }
2969
2970 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2971 __ isync();
2972 } else {
2973 __ sync();
2974 }
2975 }
2976
2977
2978 void LIR_Assembler::emit_profile_type(LIR_OpProfileType* op) {
2979 Register obj = op->obj()->as_register();
2980 Register tmp = op->tmp()->as_pointer_register();
2981 LIR_Address* mdo_addr = op->mdp()->as_address_ptr();
2982 ciKlass* exact_klass = op->exact_klass();
2983 intptr_t current_klass = op->current_klass();
2984 bool not_null = op->not_null();
2985 bool no_conflict = op->no_conflict();
2986
2987 Label Lupdate, Ldo_update, Ldone;
2988
2989 bool do_null = !not_null;
2990 bool exact_klass_set = exact_klass != nullptr && ciTypeEntries::valid_ciklass(current_klass) == exact_klass;
2991 bool do_update = !TypeEntries::is_type_unknown(current_klass) && !exact_klass_set;
2992
2993 assert(do_null || do_update, "why are we here?");
2994 assert(!TypeEntries::was_null_seen(current_klass) || do_update, "why are we here?");
2995
2996 __ verify_oop(obj, FILE_AND_LINE);
2997
2998 if (do_null) {
2999 if (!TypeEntries::was_null_seen(current_klass)) {
3000 __ cmpdi(CR0, obj, 0);
3001 __ bne(CR0, Lupdate);
3002 __ ld(R0, index_or_disp(mdo_addr), mdo_addr->base()->as_pointer_register());
3003 __ ori(R0, R0, TypeEntries::null_seen);
3004 if (do_update) {
3005 __ b(Ldo_update);
3006 } else {
3007 __ std(R0, index_or_disp(mdo_addr), mdo_addr->base()->as_pointer_register());
3008 }
3009 } else {
3010 if (do_update) {
3011 __ cmpdi(CR0, obj, 0);
3012 __ beq(CR0, Ldone);
3013 }
3014 }
3015 #ifdef ASSERT
3016 } else {
3017 __ cmpdi(CR0, obj, 0);
3018 __ bne(CR0, Lupdate);
3019 __ stop("unexpected null obj");
3020 #endif
3021 }
3022
3023 __ bind(Lupdate);
3024 if (do_update) {
3025 Label Lnext;
3026 const Register klass = R29_TOC; // kill and reload
3027 bool klass_reg_used = false;
3028 #ifdef ASSERT
3029 if (exact_klass != nullptr) {
3030 Label ok;
3031 klass_reg_used = true;
3032 __ load_klass(klass, obj);
3033 metadata2reg(exact_klass->constant_encoding(), R0);
3034 __ cmpd(CR0, klass, R0);
3035 __ beq(CR0, ok);
3036 __ stop("exact klass and actual klass differ");
3037 __ bind(ok);
3038 }
3039 #endif
3040
3041 if (!no_conflict) {
3042 if (exact_klass == nullptr || TypeEntries::is_type_none(current_klass)) {
3043 klass_reg_used = true;
3044 if (exact_klass != nullptr) {
3045 __ ld(tmp, index_or_disp(mdo_addr), mdo_addr->base()->as_pointer_register());
3046 metadata2reg(exact_klass->constant_encoding(), klass);
3047 } else {
3048 __ load_klass(klass, obj);
3049 __ ld(tmp, index_or_disp(mdo_addr), mdo_addr->base()->as_pointer_register()); // may kill obj
3050 }
3051
3052 // Like InterpreterMacroAssembler::profile_obj_type
3053 __ clrrdi(R0, tmp, exact_log2(-TypeEntries::type_klass_mask));
3054 // Basically same as andi(R0, tmp, TypeEntries::type_klass_mask);
3055 __ cmpd(CR1, R0, klass);
3056 // Klass seen before, nothing to do (regardless of unknown bit).
3057 //beq(CR1, do_nothing);
3058
3059 __ andi_(R0, tmp, TypeEntries::type_unknown);
3060 // Already unknown. Nothing to do anymore.
3061 //bne(CR0, do_nothing);
3062 __ crorc(CR0, Assembler::equal, CR1, Assembler::equal); // cr0 eq = cr1 eq or cr0 ne
3063 __ beq(CR0, Lnext);
3064
3065 if (TypeEntries::is_type_none(current_klass)) {
3066 __ clrrdi_(R0, tmp, exact_log2(-TypeEntries::type_mask));
3067 __ orr(R0, klass, tmp); // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
3068 __ beq(CR0, Ldo_update); // First time here. Set profile type.
3069 }
3070
3071 } else {
3072 assert(ciTypeEntries::valid_ciklass(current_klass) != nullptr &&
3073 ciTypeEntries::valid_ciklass(current_klass) != exact_klass, "conflict only");
3074
3075 __ ld(tmp, index_or_disp(mdo_addr), mdo_addr->base()->as_pointer_register());
3076 __ andi_(R0, tmp, TypeEntries::type_unknown);
3077 // Already unknown. Nothing to do anymore.
3078 __ bne(CR0, Lnext);
3079 }
3080
3081 // Different than before. Cannot keep accurate profile.
3082 __ ori(R0, tmp, TypeEntries::type_unknown);
3083 } else {
3084 // There's a single possible klass at this profile point
3085 assert(exact_klass != nullptr, "should be");
3086 __ ld(tmp, index_or_disp(mdo_addr), mdo_addr->base()->as_pointer_register());
3087
3088 if (TypeEntries::is_type_none(current_klass)) {
3089 klass_reg_used = true;
3090 metadata2reg(exact_klass->constant_encoding(), klass);
3091
3092 __ clrrdi(R0, tmp, exact_log2(-TypeEntries::type_klass_mask));
3093 // Basically same as andi(R0, tmp, TypeEntries::type_klass_mask);
3094 __ cmpd(CR1, R0, klass);
3095 // Klass seen before, nothing to do (regardless of unknown bit).
3096 __ beq(CR1, Lnext);
3097 #ifdef ASSERT
3098 {
3099 Label ok;
3100 __ clrrdi_(R0, tmp, exact_log2(-TypeEntries::type_mask));
3101 __ beq(CR0, ok); // First time here.
3102
3103 __ stop("unexpected profiling mismatch");
3104 __ bind(ok);
3105 }
3106 #endif
3107 // First time here. Set profile type.
3108 __ orr(R0, klass, tmp); // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
3109 } else {
3110 assert(ciTypeEntries::valid_ciklass(current_klass) != nullptr &&
3111 ciTypeEntries::valid_ciklass(current_klass) != exact_klass, "inconsistent");
3112
3113 // Already unknown. Nothing to do anymore.
3114 __ andi_(R0, tmp, TypeEntries::type_unknown);
3115 __ bne(CR0, Lnext);
3116
3117 // Different than before. Cannot keep accurate profile.
3118 __ ori(R0, tmp, TypeEntries::type_unknown);
3119 }
3120 }
3121
3122 __ bind(Ldo_update);
3123 __ std(R0, index_or_disp(mdo_addr), mdo_addr->base()->as_pointer_register());
3124
3125 __ bind(Lnext);
3126 if (klass_reg_used) { __ load_const_optimized(R29_TOC, MacroAssembler::global_toc(), R0); } // reinit
3127 }
3128 __ bind(Ldone);
3129 }
3130
3131
3132 void LIR_Assembler::emit_updatecrc32(LIR_OpUpdateCRC32* op) {
3133 assert(op->crc()->is_single_cpu(), "crc must be register");
3134 assert(op->val()->is_single_cpu(), "byte value must be register");
3135 assert(op->result_opr()->is_single_cpu(), "result must be register");
3136 Register crc = op->crc()->as_register();
3137 Register val = op->val()->as_register();
3138 Register res = op->result_opr()->as_register();
3139
3140 assert_different_registers(val, crc, res);
3141
3142 __ load_const_optimized(res, StubRoutines::crc_table_addr(), R0);
3143 __ kernel_crc32_singleByteReg(crc, val, res, true);
3144 __ mr(res, crc);
3145 }
3146
3147 #undef __