1 /*
2 * Copyright (c) 1997, 2024, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
4 * Copyright (c) 2020, 2023, Huawei Technologies Co., Ltd. All rights reserved.
5 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 *
7 * This code is free software; you can redistribute it and/or modify it
8 * under the terms of the GNU General Public License version 2 only, as
9 * published by the Free Software Foundation.
10 *
11 * This code is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 * version 2 for more details (a copy is included in the LICENSE file that
15 * accompanied this code).
16 *
17 * You should have received a copy of the GNU General Public License version
18 * 2 along with this work; if not, write to the Free Software Foundation,
19 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
20 *
21 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
22 * or visit www.oracle.com if you need additional information or have any
23 * questions.
24 *
25 */
26
27 #include "precompiled.hpp"
28 #include "asm/assembler.hpp"
29 #include "asm/assembler.inline.hpp"
30 #include "code/compiledIC.hpp"
31 #include "compiler/disassembler.hpp"
32 #include "gc/shared/barrierSet.hpp"
33 #include "gc/shared/barrierSetAssembler.hpp"
34 #include "gc/shared/cardTable.hpp"
35 #include "gc/shared/cardTableBarrierSet.hpp"
36 #include "gc/shared/collectedHeap.hpp"
37 #include "interpreter/bytecodeHistogram.hpp"
38 #include "interpreter/interpreter.hpp"
39 #include "memory/resourceArea.hpp"
40 #include "memory/universe.hpp"
41 #include "oops/accessDecorators.hpp"
42 #include "oops/compressedKlass.inline.hpp"
43 #include "oops/compressedOops.inline.hpp"
44 #include "oops/klass.inline.hpp"
45 #include "oops/oop.hpp"
46 #include "runtime/interfaceSupport.inline.hpp"
47 #include "runtime/javaThread.hpp"
48 #include "runtime/jniHandles.inline.hpp"
49 #include "runtime/sharedRuntime.hpp"
50 #include "runtime/stubRoutines.hpp"
51 #include "utilities/globalDefinitions.hpp"
52 #include "utilities/powerOfTwo.hpp"
53 #ifdef COMPILER2
54 #include "opto/compile.hpp"
55 #include "opto/node.hpp"
56 #include "opto/output.hpp"
57 #endif
58
59 #ifdef PRODUCT
60 #define BLOCK_COMMENT(str) /* nothing */
61 #else
62 #define BLOCK_COMMENT(str) block_comment(str)
63 #endif
64 #define STOP(str) stop(str);
65 #define BIND(label) bind(label); __ BLOCK_COMMENT(#label ":")
66
67
68
69 Register MacroAssembler::extract_rs1(address instr) {
70 assert_cond(instr != nullptr);
71 return as_Register(Assembler::extract(Assembler::ld_instr(instr), 19, 15));
72 }
73
74 Register MacroAssembler::extract_rs2(address instr) {
75 assert_cond(instr != nullptr);
76 return as_Register(Assembler::extract(Assembler::ld_instr(instr), 24, 20));
77 }
78
79 Register MacroAssembler::extract_rd(address instr) {
80 assert_cond(instr != nullptr);
81 return as_Register(Assembler::extract(Assembler::ld_instr(instr), 11, 7));
82 }
83
84 uint32_t MacroAssembler::extract_opcode(address instr) {
85 assert_cond(instr != nullptr);
86 return Assembler::extract(Assembler::ld_instr(instr), 6, 0);
87 }
88
89 uint32_t MacroAssembler::extract_funct3(address instr) {
90 assert_cond(instr != nullptr);
91 return Assembler::extract(Assembler::ld_instr(instr), 14, 12);
92 }
93
94 bool MacroAssembler::is_pc_relative_at(address instr) {
95 // auipc + jalr
96 // auipc + addi
97 // auipc + load
98 // auipc + fload_load
99 return (is_auipc_at(instr)) &&
100 (is_addi_at(instr + instruction_size) ||
101 is_jalr_at(instr + instruction_size) ||
102 is_load_at(instr + instruction_size) ||
103 is_float_load_at(instr + instruction_size)) &&
104 check_pc_relative_data_dependency(instr);
105 }
106
107 // ie:ld(Rd, Label)
108 bool MacroAssembler::is_load_pc_relative_at(address instr) {
109 return is_auipc_at(instr) && // auipc
110 is_ld_at(instr + instruction_size) && // ld
111 check_load_pc_relative_data_dependency(instr);
112 }
113
114 bool MacroAssembler::is_movptr1_at(address instr) {
115 return is_lui_at(instr) && // Lui
116 is_addi_at(instr + instruction_size) && // Addi
117 is_slli_shift_at(instr + instruction_size * 2, 11) && // Slli Rd, Rs, 11
118 is_addi_at(instr + instruction_size * 3) && // Addi
119 is_slli_shift_at(instr + instruction_size * 4, 6) && // Slli Rd, Rs, 6
120 (is_addi_at(instr + instruction_size * 5) ||
121 is_jalr_at(instr + instruction_size * 5) ||
122 is_load_at(instr + instruction_size * 5)) && // Addi/Jalr/Load
123 check_movptr1_data_dependency(instr);
124 }
125
126 bool MacroAssembler::is_movptr2_at(address instr) {
127 return is_lui_at(instr) && // lui
128 is_lui_at(instr + instruction_size) && // lui
129 is_slli_shift_at(instr + instruction_size * 2, 18) && // slli Rd, Rs, 18
130 is_add_at(instr + instruction_size * 3) &&
131 (is_addi_at(instr + instruction_size * 4) ||
132 is_jalr_at(instr + instruction_size * 4) ||
133 is_load_at(instr + instruction_size * 4)) && // Addi/Jalr/Load
134 check_movptr2_data_dependency(instr);
135 }
136
137 bool MacroAssembler::is_li16u_at(address instr) {
138 return is_lui_at(instr) && // lui
139 is_srli_at(instr + instruction_size) && // srli
140 check_li16u_data_dependency(instr);
141 }
142
143 bool MacroAssembler::is_li32_at(address instr) {
144 return is_lui_at(instr) && // lui
145 is_addiw_at(instr + instruction_size) && // addiw
146 check_li32_data_dependency(instr);
147 }
148
149 bool MacroAssembler::is_li64_at(address instr) {
150 return is_lui_at(instr) && // lui
151 is_addi_at(instr + instruction_size) && // addi
152 is_slli_shift_at(instr + instruction_size * 2, 12) && // Slli Rd, Rs, 12
153 is_addi_at(instr + instruction_size * 3) && // addi
154 is_slli_shift_at(instr + instruction_size * 4, 12) && // Slli Rd, Rs, 12
155 is_addi_at(instr + instruction_size * 5) && // addi
156 is_slli_shift_at(instr + instruction_size * 6, 8) && // Slli Rd, Rs, 8
157 is_addi_at(instr + instruction_size * 7) && // addi
158 check_li64_data_dependency(instr);
159 }
160
161 bool MacroAssembler::is_lwu_to_zr(address instr) {
162 assert_cond(instr != nullptr);
163 return (extract_opcode(instr) == 0b0000011 &&
164 extract_funct3(instr) == 0b110 &&
165 extract_rd(instr) == zr); // zr
166 }
167
168 uint32_t MacroAssembler::get_membar_kind(address addr) {
169 assert_cond(addr != nullptr);
170 assert(is_membar(addr), "no membar found");
171
172 uint32_t insn = Bytes::get_native_u4(addr);
173
174 uint32_t predecessor = Assembler::extract(insn, 27, 24);
175 uint32_t successor = Assembler::extract(insn, 23, 20);
176
177 return MacroAssembler::pred_succ_to_membar_mask(predecessor, successor);
178 }
179
180 void MacroAssembler::set_membar_kind(address addr, uint32_t order_kind) {
181 assert_cond(addr != nullptr);
182 assert(is_membar(addr), "no membar found");
183
184 uint32_t predecessor = 0;
185 uint32_t successor = 0;
186
187 MacroAssembler::membar_mask_to_pred_succ(order_kind, predecessor, successor);
188
189 uint32_t insn = Bytes::get_native_u4(addr);
190 address pInsn = (address) &insn;
191 Assembler::patch(pInsn, 27, 24, predecessor);
192 Assembler::patch(pInsn, 23, 20, successor);
193
194 address membar = addr;
195 Assembler::sd_instr(membar, insn);
196 }
197
198
199 static void pass_arg0(MacroAssembler* masm, Register arg) {
200 if (c_rarg0 != arg) {
201 masm->mv(c_rarg0, arg);
202 }
203 }
204
205 static void pass_arg1(MacroAssembler* masm, Register arg) {
206 if (c_rarg1 != arg) {
207 masm->mv(c_rarg1, arg);
208 }
209 }
210
211 static void pass_arg2(MacroAssembler* masm, Register arg) {
212 if (c_rarg2 != arg) {
213 masm->mv(c_rarg2, arg);
214 }
215 }
216
217 static void pass_arg3(MacroAssembler* masm, Register arg) {
218 if (c_rarg3 != arg) {
219 masm->mv(c_rarg3, arg);
220 }
221 }
222
223 void MacroAssembler::push_cont_fastpath(Register java_thread) {
224 if (!Continuations::enabled()) return;
225 Label done;
226 ld(t0, Address(java_thread, JavaThread::cont_fastpath_offset()));
227 bleu(sp, t0, done);
228 sd(sp, Address(java_thread, JavaThread::cont_fastpath_offset()));
229 bind(done);
230 }
231
232 void MacroAssembler::pop_cont_fastpath(Register java_thread) {
233 if (!Continuations::enabled()) return;
234 Label done;
235 ld(t0, Address(java_thread, JavaThread::cont_fastpath_offset()));
236 bltu(sp, t0, done);
237 sd(zr, Address(java_thread, JavaThread::cont_fastpath_offset()));
238 bind(done);
239 }
240
241 int MacroAssembler::align(int modulus, int extra_offset) {
242 CompressibleRegion cr(this);
243 intptr_t before = offset();
244 while ((offset() + extra_offset) % modulus != 0) { nop(); }
245 return (int)(offset() - before);
246 }
247
248 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
249 call_VM_base(oop_result, noreg, noreg, entry_point, number_of_arguments, check_exceptions);
250 }
251
252 // Implementation of call_VM versions
253
254 void MacroAssembler::call_VM(Register oop_result,
255 address entry_point,
256 bool check_exceptions) {
257 call_VM_helper(oop_result, entry_point, 0, check_exceptions);
258 }
259
260 void MacroAssembler::call_VM(Register oop_result,
261 address entry_point,
262 Register arg_1,
263 bool check_exceptions) {
264 pass_arg1(this, arg_1);
265 call_VM_helper(oop_result, entry_point, 1, check_exceptions);
266 }
267
268 void MacroAssembler::call_VM(Register oop_result,
269 address entry_point,
270 Register arg_1,
271 Register arg_2,
272 bool check_exceptions) {
273 assert_different_registers(arg_1, c_rarg2);
274 pass_arg2(this, arg_2);
275 pass_arg1(this, arg_1);
276 call_VM_helper(oop_result, entry_point, 2, check_exceptions);
277 }
278
279 void MacroAssembler::call_VM(Register oop_result,
280 address entry_point,
281 Register arg_1,
282 Register arg_2,
283 Register arg_3,
284 bool check_exceptions) {
285 assert_different_registers(arg_1, c_rarg2, c_rarg3);
286 assert_different_registers(arg_2, c_rarg3);
287 pass_arg3(this, arg_3);
288
289 pass_arg2(this, arg_2);
290
291 pass_arg1(this, arg_1);
292 call_VM_helper(oop_result, entry_point, 3, check_exceptions);
293 }
294
295 void MacroAssembler::call_VM(Register oop_result,
296 Register last_java_sp,
297 address entry_point,
298 int number_of_arguments,
299 bool check_exceptions) {
300 call_VM_base(oop_result, xthread, last_java_sp, entry_point, number_of_arguments, check_exceptions);
301 }
302
303 void MacroAssembler::call_VM(Register oop_result,
304 Register last_java_sp,
305 address entry_point,
306 Register arg_1,
307 bool check_exceptions) {
308 pass_arg1(this, arg_1);
309 call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
310 }
311
312 void MacroAssembler::call_VM(Register oop_result,
313 Register last_java_sp,
314 address entry_point,
315 Register arg_1,
316 Register arg_2,
317 bool check_exceptions) {
318
319 assert_different_registers(arg_1, c_rarg2);
320 pass_arg2(this, arg_2);
321 pass_arg1(this, arg_1);
322 call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
323 }
324
325 void MacroAssembler::call_VM(Register oop_result,
326 Register last_java_sp,
327 address entry_point,
328 Register arg_1,
329 Register arg_2,
330 Register arg_3,
331 bool check_exceptions) {
332 assert_different_registers(arg_1, c_rarg2, c_rarg3);
333 assert_different_registers(arg_2, c_rarg3);
334 pass_arg3(this, arg_3);
335 pass_arg2(this, arg_2);
336 pass_arg1(this, arg_1);
337 call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
338 }
339
340 void MacroAssembler::post_call_nop() {
341 if (!Continuations::enabled()) {
342 return;
343 }
344 relocate(post_call_nop_Relocation::spec(), [&] {
345 InlineSkippedInstructionsCounter skipCounter(this);
346 nop();
347 li32(zr, 0);
348 });
349 }
350
351 // these are no-ops overridden by InterpreterMacroAssembler
352 void MacroAssembler::check_and_handle_earlyret(Register java_thread) {}
353 void MacroAssembler::check_and_handle_popframe(Register java_thread) {}
354
355 // Calls to C land
356 //
357 // When entering C land, the fp, & esp of the last Java frame have to be recorded
358 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
359 // has to be reset to 0. This is required to allow proper stack traversal.
360 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
361 Register last_java_fp,
362 Register last_java_pc) {
363
364 if (last_java_pc->is_valid()) {
365 sd(last_java_pc, Address(xthread,
366 JavaThread::frame_anchor_offset() +
367 JavaFrameAnchor::last_Java_pc_offset()));
368 }
369
370 // determine last_java_sp register
371 if (!last_java_sp->is_valid()) {
372 last_java_sp = esp;
373 }
374
375 sd(last_java_sp, Address(xthread, JavaThread::last_Java_sp_offset()));
376
377 // last_java_fp is optional
378 if (last_java_fp->is_valid()) {
379 sd(last_java_fp, Address(xthread, JavaThread::last_Java_fp_offset()));
380 }
381 }
382
383 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
384 Register last_java_fp,
385 address last_java_pc,
386 Register tmp) {
387 assert(last_java_pc != nullptr, "must provide a valid PC");
388
389 la(tmp, last_java_pc);
390 sd(tmp, Address(xthread, JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset()));
391
392 set_last_Java_frame(last_java_sp, last_java_fp, noreg);
393 }
394
395 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
396 Register last_java_fp,
397 Label &L,
398 Register tmp) {
399 if (L.is_bound()) {
400 set_last_Java_frame(last_java_sp, last_java_fp, target(L), tmp);
401 } else {
402 L.add_patch_at(code(), locator());
403 IncompressibleRegion ir(this); // the label address will be patched back.
404 set_last_Java_frame(last_java_sp, last_java_fp, pc() /* Patched later */, tmp);
405 }
406 }
407
408 void MacroAssembler::reset_last_Java_frame(bool clear_fp) {
409 // we must set sp to zero to clear frame
410 sd(zr, Address(xthread, JavaThread::last_Java_sp_offset()));
411
412 // must clear fp, so that compiled frames are not confused; it is
413 // possible that we need it only for debugging
414 if (clear_fp) {
415 sd(zr, Address(xthread, JavaThread::last_Java_fp_offset()));
416 }
417
418 // Always clear the pc because it could have been set by make_walkable()
419 sd(zr, Address(xthread, JavaThread::last_Java_pc_offset()));
420 }
421
422 void MacroAssembler::call_VM_base(Register oop_result,
423 Register java_thread,
424 Register last_java_sp,
425 address entry_point,
426 int number_of_arguments,
427 bool check_exceptions) {
428 // determine java_thread register
429 if (!java_thread->is_valid()) {
430 java_thread = xthread;
431 }
432 // determine last_java_sp register
433 if (!last_java_sp->is_valid()) {
434 last_java_sp = esp;
435 }
436
437 // debugging support
438 assert(number_of_arguments >= 0 , "cannot have negative number of arguments");
439 assert(java_thread == xthread, "unexpected register");
440
441 assert(java_thread != oop_result , "cannot use the same register for java_thread & oop_result");
442 assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
443
444 // push java thread (becomes first argument of C function)
445 mv(c_rarg0, java_thread);
446
447 // set last Java frame before call
448 assert(last_java_sp != fp, "can't use fp");
449
450 Label l;
451 set_last_Java_frame(last_java_sp, fp, l, t0);
452
453 // do the call, remove parameters
454 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments, &l);
455
456 // reset last Java frame
457 // Only interpreter should have to clear fp
458 reset_last_Java_frame(true);
459
460 // C++ interp handles this in the interpreter
461 check_and_handle_popframe(java_thread);
462 check_and_handle_earlyret(java_thread);
463
464 if (check_exceptions) {
465 // check for pending exceptions (java_thread is set upon return)
466 ld(t0, Address(java_thread, in_bytes(Thread::pending_exception_offset())));
467 Label ok;
468 beqz(t0, ok);
469 RuntimeAddress target(StubRoutines::forward_exception_entry());
470 relocate(target.rspec(), [&] {
471 int32_t offset;
472 la(t0, target.target(), offset);
473 jr(t0, offset);
474 });
475 bind(ok);
476 }
477
478 // get oop result if there is one and reset the value in the thread
479 if (oop_result->is_valid()) {
480 get_vm_result(oop_result, java_thread);
481 }
482 }
483
484 void MacroAssembler::get_vm_result(Register oop_result, Register java_thread) {
485 ld(oop_result, Address(java_thread, JavaThread::vm_result_offset()));
486 sd(zr, Address(java_thread, JavaThread::vm_result_offset()));
487 verify_oop_msg(oop_result, "broken oop in call_VM_base");
488 }
489
490 void MacroAssembler::get_vm_result_2(Register metadata_result, Register java_thread) {
491 ld(metadata_result, Address(java_thread, JavaThread::vm_result_2_offset()));
492 sd(zr, Address(java_thread, JavaThread::vm_result_2_offset()));
493 }
494
495 void MacroAssembler::clinit_barrier(Register klass, Register tmp, Label* L_fast_path, Label* L_slow_path) {
496 assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
497 assert_different_registers(klass, xthread, tmp);
498
499 Label L_fallthrough, L_tmp;
500 if (L_fast_path == nullptr) {
501 L_fast_path = &L_fallthrough;
502 } else if (L_slow_path == nullptr) {
503 L_slow_path = &L_fallthrough;
504 }
505
506 // Fast path check: class is fully initialized
507 lbu(tmp, Address(klass, InstanceKlass::init_state_offset()));
508 sub(tmp, tmp, InstanceKlass::fully_initialized);
509 beqz(tmp, *L_fast_path);
510
511 // Fast path check: current thread is initializer thread
512 ld(tmp, Address(klass, InstanceKlass::init_thread_offset()));
513
514 if (L_slow_path == &L_fallthrough) {
515 beq(xthread, tmp, *L_fast_path);
516 bind(*L_slow_path);
517 } else if (L_fast_path == &L_fallthrough) {
518 bne(xthread, tmp, *L_slow_path);
519 bind(*L_fast_path);
520 } else {
521 Unimplemented();
522 }
523 }
524
525 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
526 if (!VerifyOops) { return; }
527
528 // Pass register number to verify_oop_subroutine
529 const char* b = nullptr;
530 {
531 ResourceMark rm;
532 stringStream ss;
533 ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
534 b = code_string(ss.as_string());
535 }
536 BLOCK_COMMENT("verify_oop {");
537
538 push_reg(RegSet::of(ra, t0, t1, c_rarg0), sp);
539
540 mv(c_rarg0, reg); // c_rarg0 : x10
541 {
542 // The length of the instruction sequence emitted should not depend
543 // on the address of the char buffer so that the size of mach nodes for
544 // scratch emit and normal emit matches.
545 IncompressibleRegion ir(this); // Fixed length
546 movptr(t0, (address) b);
547 }
548
549 // call indirectly to solve generation ordering problem
550 ExternalAddress target(StubRoutines::verify_oop_subroutine_entry_address());
551 relocate(target.rspec(), [&] {
552 int32_t offset;
553 la(t1, target.target(), offset);
554 ld(t1, Address(t1, offset));
555 });
556 jalr(t1);
557
558 pop_reg(RegSet::of(ra, t0, t1, c_rarg0), sp);
559
560 BLOCK_COMMENT("} verify_oop");
561 }
562
563 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
564 if (!VerifyOops) {
565 return;
566 }
567
568 const char* b = nullptr;
569 {
570 ResourceMark rm;
571 stringStream ss;
572 ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
573 b = code_string(ss.as_string());
574 }
575 BLOCK_COMMENT("verify_oop_addr {");
576
577 push_reg(RegSet::of(ra, t0, t1, c_rarg0), sp);
578
579 if (addr.uses(sp)) {
580 la(x10, addr);
581 ld(x10, Address(x10, 4 * wordSize));
582 } else {
583 ld(x10, addr);
584 }
585
586 {
587 // The length of the instruction sequence emitted should not depend
588 // on the address of the char buffer so that the size of mach nodes for
589 // scratch emit and normal emit matches.
590 IncompressibleRegion ir(this); // Fixed length
591 movptr(t0, (address) b);
592 }
593
594 // call indirectly to solve generation ordering problem
595 ExternalAddress target(StubRoutines::verify_oop_subroutine_entry_address());
596 relocate(target.rspec(), [&] {
597 int32_t offset;
598 la(t1, target.target(), offset);
599 ld(t1, Address(t1, offset));
600 });
601 jalr(t1);
602
603 pop_reg(RegSet::of(ra, t0, t1, c_rarg0), sp);
604
605 BLOCK_COMMENT("} verify_oop_addr");
606 }
607
608 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
609 int extra_slot_offset) {
610 // cf. TemplateTable::prepare_invoke(), if (load_receiver).
611 int stackElementSize = Interpreter::stackElementSize;
612 int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
613 #ifdef ASSERT
614 int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
615 assert(offset1 - offset == stackElementSize, "correct arithmetic");
616 #endif
617 if (arg_slot.is_constant()) {
618 return Address(esp, arg_slot.as_constant() * stackElementSize + offset);
619 } else {
620 assert_different_registers(t0, arg_slot.as_register());
621 shadd(t0, arg_slot.as_register(), esp, t0, exact_log2(stackElementSize));
622 return Address(t0, offset);
623 }
624 }
625
626 #ifndef PRODUCT
627 extern "C" void findpc(intptr_t x);
628 #endif
629
630 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[])
631 {
632 // In order to get locks to work, we need to fake a in_VM state
633 if (ShowMessageBoxOnError) {
634 JavaThread* thread = JavaThread::current();
635 JavaThreadState saved_state = thread->thread_state();
636 thread->set_thread_state(_thread_in_vm);
637 #ifndef PRODUCT
638 if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
639 ttyLocker ttyl;
640 BytecodeCounter::print();
641 }
642 #endif
643 if (os::message_box(msg, "Execution stopped, print registers?")) {
644 ttyLocker ttyl;
645 tty->print_cr(" pc = 0x%016lx", pc);
646 #ifndef PRODUCT
647 tty->cr();
648 findpc(pc);
649 tty->cr();
650 #endif
651 tty->print_cr(" x0 = 0x%016lx", regs[0]);
652 tty->print_cr(" x1 = 0x%016lx", regs[1]);
653 tty->print_cr(" x2 = 0x%016lx", regs[2]);
654 tty->print_cr(" x3 = 0x%016lx", regs[3]);
655 tty->print_cr(" x4 = 0x%016lx", regs[4]);
656 tty->print_cr(" x5 = 0x%016lx", regs[5]);
657 tty->print_cr(" x6 = 0x%016lx", regs[6]);
658 tty->print_cr(" x7 = 0x%016lx", regs[7]);
659 tty->print_cr(" x8 = 0x%016lx", regs[8]);
660 tty->print_cr(" x9 = 0x%016lx", regs[9]);
661 tty->print_cr("x10 = 0x%016lx", regs[10]);
662 tty->print_cr("x11 = 0x%016lx", regs[11]);
663 tty->print_cr("x12 = 0x%016lx", regs[12]);
664 tty->print_cr("x13 = 0x%016lx", regs[13]);
665 tty->print_cr("x14 = 0x%016lx", regs[14]);
666 tty->print_cr("x15 = 0x%016lx", regs[15]);
667 tty->print_cr("x16 = 0x%016lx", regs[16]);
668 tty->print_cr("x17 = 0x%016lx", regs[17]);
669 tty->print_cr("x18 = 0x%016lx", regs[18]);
670 tty->print_cr("x19 = 0x%016lx", regs[19]);
671 tty->print_cr("x20 = 0x%016lx", regs[20]);
672 tty->print_cr("x21 = 0x%016lx", regs[21]);
673 tty->print_cr("x22 = 0x%016lx", regs[22]);
674 tty->print_cr("x23 = 0x%016lx", regs[23]);
675 tty->print_cr("x24 = 0x%016lx", regs[24]);
676 tty->print_cr("x25 = 0x%016lx", regs[25]);
677 tty->print_cr("x26 = 0x%016lx", regs[26]);
678 tty->print_cr("x27 = 0x%016lx", regs[27]);
679 tty->print_cr("x28 = 0x%016lx", regs[28]);
680 tty->print_cr("x30 = 0x%016lx", regs[30]);
681 tty->print_cr("x31 = 0x%016lx", regs[31]);
682 BREAKPOINT;
683 }
684 }
685 fatal("DEBUG MESSAGE: %s", msg);
686 }
687
688 void MacroAssembler::resolve_jobject(Register value, Register tmp1, Register tmp2) {
689 assert_different_registers(value, tmp1, tmp2);
690 Label done, tagged, weak_tagged;
691
692 beqz(value, done); // Use null as-is.
693 // Test for tag.
694 andi(tmp1, value, JNIHandles::tag_mask);
695 bnez(tmp1, tagged);
696
697 // Resolve local handle
698 access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp1, tmp2);
699 verify_oop(value);
700 j(done);
701
702 bind(tagged);
703 // Test for jweak tag.
704 STATIC_ASSERT(JNIHandles::TypeTag::weak_global == 0b1);
705 test_bit(tmp1, value, exact_log2(JNIHandles::TypeTag::weak_global));
706 bnez(tmp1, weak_tagged);
707
708 // Resolve global handle
709 access_load_at(T_OBJECT, IN_NATIVE, value,
710 Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
711 verify_oop(value);
712 j(done);
713
714 bind(weak_tagged);
715 // Resolve jweak.
716 access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF, value,
717 Address(value, -JNIHandles::TypeTag::weak_global), tmp1, tmp2);
718 verify_oop(value);
719
720 bind(done);
721 }
722
723 void MacroAssembler::resolve_global_jobject(Register value, Register tmp1, Register tmp2) {
724 assert_different_registers(value, tmp1, tmp2);
725 Label done;
726
727 beqz(value, done); // Use null as-is.
728
729 #ifdef ASSERT
730 {
731 STATIC_ASSERT(JNIHandles::TypeTag::global == 0b10);
732 Label valid_global_tag;
733 test_bit(tmp1, value, exact_log2(JNIHandles::TypeTag::global)); // Test for global tag.
734 bnez(tmp1, valid_global_tag);
735 stop("non global jobject using resolve_global_jobject");
736 bind(valid_global_tag);
737 }
738 #endif
739
740 // Resolve global handle
741 access_load_at(T_OBJECT, IN_NATIVE, value,
742 Address(value, -JNIHandles::TypeTag::global), tmp1, tmp2);
743 verify_oop(value);
744
745 bind(done);
746 }
747
748 void MacroAssembler::stop(const char* msg) {
749 BLOCK_COMMENT(msg);
750 illegal_instruction(Assembler::csr::time);
751 emit_int64((uintptr_t)msg);
752 }
753
754 void MacroAssembler::unimplemented(const char* what) {
755 const char* buf = nullptr;
756 {
757 ResourceMark rm;
758 stringStream ss;
759 ss.print("unimplemented: %s", what);
760 buf = code_string(ss.as_string());
761 }
762 stop(buf);
763 }
764
765 void MacroAssembler::emit_static_call_stub() {
766 IncompressibleRegion ir(this); // Fixed length: see CompiledDirectCall::to_interp_stub_size().
767 // CompiledDirectCall::set_to_interpreted knows the
768 // exact layout of this stub.
769
770 mov_metadata(xmethod, (Metadata*)nullptr);
771
772 // Jump to the entry point of the c2i stub.
773 int32_t offset = 0;
774 movptr(t0, 0, offset, t1); // lui + lui + slli + add
775 jr(t0, offset);
776 }
777
778 void MacroAssembler::call_VM_leaf_base(address entry_point,
779 int number_of_arguments,
780 Label *retaddr) {
781 int32_t offset = 0;
782 push_reg(RegSet::of(t0, xmethod), sp); // push << t0 & xmethod >> to sp
783 mv(t0, entry_point, offset);
784 jalr(t0, offset);
785 if (retaddr != nullptr) {
786 bind(*retaddr);
787 }
788 pop_reg(RegSet::of(t0, xmethod), sp); // pop << t0 & xmethod >> from sp
789 }
790
791 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
792 call_VM_leaf_base(entry_point, number_of_arguments);
793 }
794
795 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
796 pass_arg0(this, arg_0);
797 call_VM_leaf_base(entry_point, 1);
798 }
799
800 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
801 assert_different_registers(arg_1, c_rarg0);
802 pass_arg0(this, arg_0);
803 pass_arg1(this, arg_1);
804 call_VM_leaf_base(entry_point, 2);
805 }
806
807 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0,
808 Register arg_1, Register arg_2) {
809 assert_different_registers(arg_1, c_rarg0);
810 assert_different_registers(arg_2, c_rarg0, c_rarg1);
811 pass_arg0(this, arg_0);
812 pass_arg1(this, arg_1);
813 pass_arg2(this, arg_2);
814 call_VM_leaf_base(entry_point, 3);
815 }
816
817 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
818 pass_arg0(this, arg_0);
819 MacroAssembler::call_VM_leaf_base(entry_point, 1);
820 }
821
822 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
823
824 assert_different_registers(arg_0, c_rarg1);
825 pass_arg1(this, arg_1);
826 pass_arg0(this, arg_0);
827 MacroAssembler::call_VM_leaf_base(entry_point, 2);
828 }
829
830 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
831 assert_different_registers(arg_0, c_rarg1, c_rarg2);
832 assert_different_registers(arg_1, c_rarg2);
833 pass_arg2(this, arg_2);
834 pass_arg1(this, arg_1);
835 pass_arg0(this, arg_0);
836 MacroAssembler::call_VM_leaf_base(entry_point, 3);
837 }
838
839 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
840 assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
841 assert_different_registers(arg_1, c_rarg2, c_rarg3);
842 assert_different_registers(arg_2, c_rarg3);
843
844 pass_arg3(this, arg_3);
845 pass_arg2(this, arg_2);
846 pass_arg1(this, arg_1);
847 pass_arg0(this, arg_0);
848 MacroAssembler::call_VM_leaf_base(entry_point, 4);
849 }
850
851 void MacroAssembler::la(Register Rd, const address addr) {
852 int32_t offset;
853 la(Rd, addr, offset);
854 addi(Rd, Rd, offset);
855 }
856
857 void MacroAssembler::la(Register Rd, const address addr, int32_t &offset) {
858 if (is_32bit_offset_from_codecache((int64_t)addr)) {
859 int64_t distance = addr - pc();
860 assert(is_valid_32bit_offset(distance), "Must be");
861 auipc(Rd, (int32_t)distance + 0x800);
862 offset = ((int32_t)distance << 20) >> 20;
863 } else {
864 assert(!CodeCache::contains(addr), "Must be");
865 movptr(Rd, addr, offset);
866 }
867 }
868
869 void MacroAssembler::la(Register Rd, const Address &adr) {
870 switch (adr.getMode()) {
871 case Address::literal: {
872 relocInfo::relocType rtype = adr.rspec().reloc()->type();
873 if (rtype == relocInfo::none) {
874 mv(Rd, (intptr_t)(adr.target()));
875 } else {
876 relocate(adr.rspec(), [&] {
877 movptr(Rd, adr.target());
878 });
879 }
880 break;
881 }
882 case Address::base_plus_offset: {
883 Address new_adr = legitimize_address(Rd, adr);
884 if (!(new_adr.base() == Rd && new_adr.offset() == 0)) {
885 addi(Rd, new_adr.base(), new_adr.offset());
886 }
887 break;
888 }
889 default:
890 ShouldNotReachHere();
891 }
892 }
893
894 void MacroAssembler::la(Register Rd, Label &label) {
895 IncompressibleRegion ir(this); // the label address may be patched back.
896 wrap_label(Rd, label, &MacroAssembler::la);
897 }
898
899 void MacroAssembler::li16u(Register Rd, uint16_t imm) {
900 lui(Rd, (uint32_t)imm << 12);
901 srli(Rd, Rd, 12);
902 }
903
904 void MacroAssembler::li32(Register Rd, int32_t imm) {
905 // int32_t is in range 0x8000 0000 ~ 0x7fff ffff, and imm[31] is the sign bit
906 int64_t upper = imm, lower = imm;
907 lower = (imm << 20) >> 20;
908 upper -= lower;
909 upper = (int32_t)upper;
910 // lui Rd, imm[31:12] + imm[11]
911 lui(Rd, upper);
912 // use addiw to distinguish li32 to li64
913 addiw(Rd, Rd, lower);
914 }
915
916 void MacroAssembler::li64(Register Rd, int64_t imm) {
917 // Load upper 32 bits. upper = imm[63:32], but if imm[31] == 1 or
918 // (imm[31:20] == 0x7ff && imm[19] == 1), upper = imm[63:32] + 1.
919 int64_t lower = imm & 0xffffffff;
920 lower -= ((lower << 44) >> 44);
921 int64_t tmp_imm = ((uint64_t)(imm & 0xffffffff00000000)) + (uint64_t)lower;
922 int32_t upper = (tmp_imm - (int32_t)lower) >> 32;
923
924 // Load upper 32 bits
925 int64_t up = upper, lo = upper;
926 lo = (lo << 52) >> 52;
927 up -= lo;
928 up = (int32_t)up;
929 lui(Rd, up);
930 addi(Rd, Rd, lo);
931
932 // Load the rest 32 bits.
933 slli(Rd, Rd, 12);
934 addi(Rd, Rd, (int32_t)lower >> 20);
935 slli(Rd, Rd, 12);
936 lower = ((int32_t)imm << 12) >> 20;
937 addi(Rd, Rd, lower);
938 slli(Rd, Rd, 8);
939 lower = imm & 0xff;
940 addi(Rd, Rd, lower);
941 }
942
943 void MacroAssembler::li(Register Rd, int64_t imm) {
944 // int64_t is in range 0x8000 0000 0000 0000 ~ 0x7fff ffff ffff ffff
945 // li -> c.li
946 if (do_compress() && (is_simm6(imm) && Rd != x0)) {
947 c_li(Rd, imm);
948 return;
949 }
950
951 int shift = 12;
952 int64_t upper = imm, lower = imm;
953 // Split imm to a lower 12-bit sign-extended part and the remainder,
954 // because addi will sign-extend the lower imm.
955 lower = ((int32_t)imm << 20) >> 20;
956 upper -= lower;
957
958 // Test whether imm is a 32-bit integer.
959 if (!(((imm) & ~(int64_t)0x7fffffff) == 0 ||
960 (((imm) & ~(int64_t)0x7fffffff) == ~(int64_t)0x7fffffff))) {
961 while (((upper >> shift) & 1) == 0) { shift++; }
962 upper >>= shift;
963 li(Rd, upper);
964 slli(Rd, Rd, shift);
965 if (lower != 0) {
966 addi(Rd, Rd, lower);
967 }
968 } else {
969 // 32-bit integer
970 Register hi_Rd = zr;
971 if (upper != 0) {
972 lui(Rd, (int32_t)upper);
973 hi_Rd = Rd;
974 }
975 if (lower != 0 || hi_Rd == zr) {
976 addiw(Rd, hi_Rd, lower);
977 }
978 }
979 }
980
981 void MacroAssembler::jump_link(const address dest, Register temp) {
982 assert_cond(dest != nullptr);
983 int64_t distance = dest - pc();
984 if (is_simm21(distance) && ((distance % 2) == 0)) {
985 Assembler::jal(x1, distance);
986 } else {
987 assert(temp != noreg && temp != x0, "expecting a register");
988 int32_t offset = 0;
989 la(temp, dest, offset);
990 jalr(temp, offset);
991 }
992 }
993
994 void MacroAssembler::jump_link(const Address &adr, Register temp) {
995 switch (adr.getMode()) {
996 case Address::literal: {
997 relocate(adr.rspec(), [&] {
998 jump_link(adr.target(), temp);
999 });
1000 break;
1001 }
1002 case Address::base_plus_offset: {
1003 int32_t offset = ((int32_t)adr.offset() << 20) >> 20;
1004 la(temp, Address(adr.base(), adr.offset() - offset));
1005 jalr(temp, offset);
1006 break;
1007 }
1008 default:
1009 ShouldNotReachHere();
1010 }
1011 }
1012
1013 void MacroAssembler::j(const address dest, Register temp) {
1014 assert(CodeCache::contains(dest), "Must be");
1015 assert_cond(dest != nullptr);
1016 int64_t distance = dest - pc();
1017
1018 // We can't patch C, i.e. if Label wasn't bound we need to patch this jump.
1019 IncompressibleRegion ir(this);
1020 if (is_simm21(distance) && ((distance % 2) == 0)) {
1021 Assembler::jal(x0, distance);
1022 } else {
1023 assert(temp != noreg && temp != x0, "expecting a register");
1024 int32_t offset = 0;
1025 la(temp, dest, offset);
1026 jr(temp, offset);
1027 }
1028 }
1029
1030 void MacroAssembler::j(const Address &adr, Register temp) {
1031 switch (adr.getMode()) {
1032 case Address::literal: {
1033 relocate(adr.rspec(), [&] {
1034 j(adr.target(), temp);
1035 });
1036 break;
1037 }
1038 case Address::base_plus_offset: {
1039 int32_t offset = ((int32_t)adr.offset() << 20) >> 20;
1040 la(temp, Address(adr.base(), adr.offset() - offset));
1041 jr(temp, offset);
1042 break;
1043 }
1044 default:
1045 ShouldNotReachHere();
1046 }
1047 }
1048
1049 void MacroAssembler::j(Label &lab, Register temp) {
1050 assert_different_registers(x0, temp);
1051 if (lab.is_bound()) {
1052 MacroAssembler::j(target(lab), temp);
1053 } else {
1054 lab.add_patch_at(code(), locator());
1055 MacroAssembler::j(pc(), temp);
1056 }
1057 }
1058
1059 void MacroAssembler::jr(Register Rd, int32_t offset) {
1060 assert(Rd != noreg, "expecting a register");
1061 Assembler::jalr(x0, Rd, offset);
1062 }
1063
1064 void MacroAssembler::call(const address dest, Register temp) {
1065 assert_cond(dest != nullptr);
1066 assert(temp != noreg, "expecting a register");
1067 int32_t offset = 0;
1068 la(temp, dest, offset);
1069 jalr(temp, offset);
1070 }
1071
1072 void MacroAssembler::jalr(Register Rs, int32_t offset) {
1073 assert(Rs != noreg, "expecting a register");
1074 Assembler::jalr(x1, Rs, offset);
1075 }
1076
1077 void MacroAssembler::rt_call(address dest, Register tmp) {
1078 CodeBlob *cb = CodeCache::find_blob(dest);
1079 RuntimeAddress target(dest);
1080 if (cb) {
1081 far_call(target, tmp);
1082 } else {
1083 relocate(target.rspec(), [&] {
1084 int32_t offset;
1085 la(tmp, target.target(), offset);
1086 jalr(tmp, offset);
1087 });
1088 }
1089 }
1090
1091 void MacroAssembler::wrap_label(Register Rt, Label &L, jal_jalr_insn insn) {
1092 if (L.is_bound()) {
1093 (this->*insn)(Rt, target(L));
1094 } else {
1095 L.add_patch_at(code(), locator());
1096 (this->*insn)(Rt, pc());
1097 }
1098 }
1099
1100 void MacroAssembler::wrap_label(Register r1, Register r2, Label &L,
1101 compare_and_branch_insn insn,
1102 compare_and_branch_label_insn neg_insn, bool is_far) {
1103 if (is_far) {
1104 Label done;
1105 (this->*neg_insn)(r1, r2, done, /* is_far */ false);
1106 j(L);
1107 bind(done);
1108 } else {
1109 if (L.is_bound()) {
1110 (this->*insn)(r1, r2, target(L));
1111 } else {
1112 L.add_patch_at(code(), locator());
1113 (this->*insn)(r1, r2, pc());
1114 }
1115 }
1116 }
1117
1118 #define INSN(NAME, NEG_INSN) \
1119 void MacroAssembler::NAME(Register Rs1, Register Rs2, Label &L, bool is_far) { \
1120 wrap_label(Rs1, Rs2, L, &MacroAssembler::NAME, &MacroAssembler::NEG_INSN, is_far); \
1121 }
1122
1123 INSN(beq, bne);
1124 INSN(bne, beq);
1125 INSN(blt, bge);
1126 INSN(bge, blt);
1127 INSN(bltu, bgeu);
1128 INSN(bgeu, bltu);
1129
1130 #undef INSN
1131
1132 #define INSN(NAME) \
1133 void MacroAssembler::NAME##z(Register Rs, const address dest) { \
1134 NAME(Rs, zr, dest); \
1135 } \
1136 void MacroAssembler::NAME##z(Register Rs, Label &l, bool is_far) { \
1137 NAME(Rs, zr, l, is_far); \
1138 } \
1139
1140 INSN(beq);
1141 INSN(bne);
1142 INSN(blt);
1143 INSN(ble);
1144 INSN(bge);
1145 INSN(bgt);
1146
1147 #undef INSN
1148
1149 #define INSN(NAME, NEG_INSN) \
1150 void MacroAssembler::NAME(Register Rs, Register Rt, const address dest) { \
1151 NEG_INSN(Rt, Rs, dest); \
1152 } \
1153 void MacroAssembler::NAME(Register Rs, Register Rt, Label &l, bool is_far) { \
1154 NEG_INSN(Rt, Rs, l, is_far); \
1155 }
1156
1157 INSN(bgt, blt);
1158 INSN(ble, bge);
1159 INSN(bgtu, bltu);
1160 INSN(bleu, bgeu);
1161
1162 #undef INSN
1163
1164 // Float compare branch instructions
1165
1166 #define INSN(NAME, FLOATCMP, BRANCH) \
1167 void MacroAssembler::float_##NAME(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far, bool is_unordered) { \
1168 FLOATCMP##_s(t0, Rs1, Rs2); \
1169 BRANCH(t0, l, is_far); \
1170 } \
1171 void MacroAssembler::double_##NAME(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far, bool is_unordered) { \
1172 FLOATCMP##_d(t0, Rs1, Rs2); \
1173 BRANCH(t0, l, is_far); \
1174 }
1175
1176 INSN(beq, feq, bnez);
1177 INSN(bne, feq, beqz);
1178
1179 #undef INSN
1180
1181
1182 #define INSN(NAME, FLOATCMP1, FLOATCMP2) \
1183 void MacroAssembler::float_##NAME(FloatRegister Rs1, FloatRegister Rs2, Label &l, \
1184 bool is_far, bool is_unordered) { \
1185 if (is_unordered) { \
1186 /* jump if either source is NaN or condition is expected */ \
1187 FLOATCMP2##_s(t0, Rs2, Rs1); \
1188 beqz(t0, l, is_far); \
1189 } else { \
1190 /* jump if no NaN in source and condition is expected */ \
1191 FLOATCMP1##_s(t0, Rs1, Rs2); \
1192 bnez(t0, l, is_far); \
1193 } \
1194 } \
1195 void MacroAssembler::double_##NAME(FloatRegister Rs1, FloatRegister Rs2, Label &l, \
1196 bool is_far, bool is_unordered) { \
1197 if (is_unordered) { \
1198 /* jump if either source is NaN or condition is expected */ \
1199 FLOATCMP2##_d(t0, Rs2, Rs1); \
1200 beqz(t0, l, is_far); \
1201 } else { \
1202 /* jump if no NaN in source and condition is expected */ \
1203 FLOATCMP1##_d(t0, Rs1, Rs2); \
1204 bnez(t0, l, is_far); \
1205 } \
1206 }
1207
1208 INSN(ble, fle, flt);
1209 INSN(blt, flt, fle);
1210
1211 #undef INSN
1212
1213 #define INSN(NAME, CMP) \
1214 void MacroAssembler::float_##NAME(FloatRegister Rs1, FloatRegister Rs2, Label &l, \
1215 bool is_far, bool is_unordered) { \
1216 float_##CMP(Rs2, Rs1, l, is_far, is_unordered); \
1217 } \
1218 void MacroAssembler::double_##NAME(FloatRegister Rs1, FloatRegister Rs2, Label &l, \
1219 bool is_far, bool is_unordered) { \
1220 double_##CMP(Rs2, Rs1, l, is_far, is_unordered); \
1221 }
1222
1223 INSN(bgt, blt);
1224 INSN(bge, ble);
1225
1226 #undef INSN
1227
1228
1229 #define INSN(NAME, CSR) \
1230 void MacroAssembler::NAME(Register Rd) { \
1231 csrr(Rd, CSR); \
1232 }
1233
1234 INSN(rdinstret, CSR_INSTRET);
1235 INSN(rdcycle, CSR_CYCLE);
1236 INSN(rdtime, CSR_TIME);
1237 INSN(frcsr, CSR_FCSR);
1238 INSN(frrm, CSR_FRM);
1239 INSN(frflags, CSR_FFLAGS);
1240
1241 #undef INSN
1242
1243 void MacroAssembler::csrr(Register Rd, unsigned csr) {
1244 csrrs(Rd, csr, x0);
1245 }
1246
1247 #define INSN(NAME, OPFUN) \
1248 void MacroAssembler::NAME(unsigned csr, Register Rs) { \
1249 OPFUN(x0, csr, Rs); \
1250 }
1251
1252 INSN(csrw, csrrw);
1253 INSN(csrs, csrrs);
1254 INSN(csrc, csrrc);
1255
1256 #undef INSN
1257
1258 #define INSN(NAME, OPFUN) \
1259 void MacroAssembler::NAME(unsigned csr, unsigned imm) { \
1260 OPFUN(x0, csr, imm); \
1261 }
1262
1263 INSN(csrwi, csrrwi);
1264 INSN(csrsi, csrrsi);
1265 INSN(csrci, csrrci);
1266
1267 #undef INSN
1268
1269 #define INSN(NAME, CSR) \
1270 void MacroAssembler::NAME(Register Rd, Register Rs) { \
1271 csrrw(Rd, CSR, Rs); \
1272 }
1273
1274 INSN(fscsr, CSR_FCSR);
1275 INSN(fsrm, CSR_FRM);
1276 INSN(fsflags, CSR_FFLAGS);
1277
1278 #undef INSN
1279
1280 #define INSN(NAME) \
1281 void MacroAssembler::NAME(Register Rs) { \
1282 NAME(x0, Rs); \
1283 }
1284
1285 INSN(fscsr);
1286 INSN(fsrm);
1287 INSN(fsflags);
1288
1289 #undef INSN
1290
1291 void MacroAssembler::fsrmi(Register Rd, unsigned imm) {
1292 guarantee(imm < 5, "Rounding Mode is invalid in Rounding Mode register");
1293 csrrwi(Rd, CSR_FRM, imm);
1294 }
1295
1296 void MacroAssembler::fsflagsi(Register Rd, unsigned imm) {
1297 csrrwi(Rd, CSR_FFLAGS, imm);
1298 }
1299
1300 #define INSN(NAME) \
1301 void MacroAssembler::NAME(unsigned imm) { \
1302 NAME(x0, imm); \
1303 }
1304
1305 INSN(fsrmi);
1306 INSN(fsflagsi);
1307
1308 #undef INSN
1309
1310 void MacroAssembler::restore_cpu_control_state_after_jni(Register tmp) {
1311 if (RestoreMXCSROnJNICalls) {
1312 Label skip_fsrmi;
1313 frrm(tmp);
1314 // Set FRM to the state we need. We do want Round to Nearest.
1315 // We don't want non-IEEE rounding modes.
1316 guarantee(RoundingMode::rne == 0, "must be");
1317 beqz(tmp, skip_fsrmi); // Only reset FRM if it's wrong
1318 fsrmi(RoundingMode::rne);
1319 bind(skip_fsrmi);
1320 }
1321 }
1322
1323 void MacroAssembler::push_reg(Register Rs)
1324 {
1325 addi(esp, esp, 0 - wordSize);
1326 sd(Rs, Address(esp, 0));
1327 }
1328
1329 void MacroAssembler::pop_reg(Register Rd)
1330 {
1331 ld(Rd, Address(esp, 0));
1332 addi(esp, esp, wordSize);
1333 }
1334
1335 int MacroAssembler::bitset_to_regs(unsigned int bitset, unsigned char* regs) {
1336 int count = 0;
1337 // Scan bitset to accumulate register pairs
1338 for (int reg = 31; reg >= 0; reg--) {
1339 if ((1U << 31) & bitset) {
1340 regs[count++] = reg;
1341 }
1342 bitset <<= 1;
1343 }
1344 return count;
1345 }
1346
1347 // Push integer registers in the bitset supplied. Don't push sp.
1348 // Return the number of words pushed
1349 int MacroAssembler::push_reg(unsigned int bitset, Register stack) {
1350 DEBUG_ONLY(int words_pushed = 0;)
1351 unsigned char regs[32];
1352 int count = bitset_to_regs(bitset, regs);
1353 // reserve one slot to align for odd count
1354 int offset = is_even(count) ? 0 : wordSize;
1355
1356 if (count) {
1357 addi(stack, stack, -count * wordSize - offset);
1358 }
1359 for (int i = count - 1; i >= 0; i--) {
1360 sd(as_Register(regs[i]), Address(stack, (count - 1 - i) * wordSize + offset));
1361 DEBUG_ONLY(words_pushed++;)
1362 }
1363
1364 assert(words_pushed == count, "oops, pushed != count");
1365
1366 return count;
1367 }
1368
1369 int MacroAssembler::pop_reg(unsigned int bitset, Register stack) {
1370 DEBUG_ONLY(int words_popped = 0;)
1371 unsigned char regs[32];
1372 int count = bitset_to_regs(bitset, regs);
1373 // reserve one slot to align for odd count
1374 int offset = is_even(count) ? 0 : wordSize;
1375
1376 for (int i = count - 1; i >= 0; i--) {
1377 ld(as_Register(regs[i]), Address(stack, (count - 1 - i) * wordSize + offset));
1378 DEBUG_ONLY(words_popped++;)
1379 }
1380
1381 if (count) {
1382 addi(stack, stack, count * wordSize + offset);
1383 }
1384 assert(words_popped == count, "oops, popped != count");
1385
1386 return count;
1387 }
1388
1389 // Push floating-point registers in the bitset supplied.
1390 // Return the number of words pushed
1391 int MacroAssembler::push_fp(unsigned int bitset, Register stack) {
1392 DEBUG_ONLY(int words_pushed = 0;)
1393 unsigned char regs[32];
1394 int count = bitset_to_regs(bitset, regs);
1395 int push_slots = count + (count & 1);
1396
1397 if (count) {
1398 addi(stack, stack, -push_slots * wordSize);
1399 }
1400
1401 for (int i = count - 1; i >= 0; i--) {
1402 fsd(as_FloatRegister(regs[i]), Address(stack, (push_slots - 1 - i) * wordSize));
1403 DEBUG_ONLY(words_pushed++;)
1404 }
1405
1406 assert(words_pushed == count, "oops, pushed(%d) != count(%d)", words_pushed, count);
1407
1408 return count;
1409 }
1410
1411 int MacroAssembler::pop_fp(unsigned int bitset, Register stack) {
1412 DEBUG_ONLY(int words_popped = 0;)
1413 unsigned char regs[32];
1414 int count = bitset_to_regs(bitset, regs);
1415 int pop_slots = count + (count & 1);
1416
1417 for (int i = count - 1; i >= 0; i--) {
1418 fld(as_FloatRegister(regs[i]), Address(stack, (pop_slots - 1 - i) * wordSize));
1419 DEBUG_ONLY(words_popped++;)
1420 }
1421
1422 if (count) {
1423 addi(stack, stack, pop_slots * wordSize);
1424 }
1425
1426 assert(words_popped == count, "oops, popped(%d) != count(%d)", words_popped, count);
1427
1428 return count;
1429 }
1430
1431 #ifdef COMPILER2
1432 // Push vector registers in the bitset supplied.
1433 // Return the number of words pushed
1434 int MacroAssembler::push_v(unsigned int bitset, Register stack) {
1435 int vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
1436
1437 // Scan bitset to accumulate register pairs
1438 unsigned char regs[32];
1439 int count = bitset_to_regs(bitset, regs);
1440
1441 for (int i = 0; i < count; i++) {
1442 sub(stack, stack, vector_size_in_bytes);
1443 vs1r_v(as_VectorRegister(regs[i]), stack);
1444 }
1445
1446 return count * vector_size_in_bytes / wordSize;
1447 }
1448
1449 int MacroAssembler::pop_v(unsigned int bitset, Register stack) {
1450 int vector_size_in_bytes = Matcher::scalable_vector_reg_size(T_BYTE);
1451
1452 // Scan bitset to accumulate register pairs
1453 unsigned char regs[32];
1454 int count = bitset_to_regs(bitset, regs);
1455
1456 for (int i = count - 1; i >= 0; i--) {
1457 vl1r_v(as_VectorRegister(regs[i]), stack);
1458 add(stack, stack, vector_size_in_bytes);
1459 }
1460
1461 return count * vector_size_in_bytes / wordSize;
1462 }
1463 #endif // COMPILER2
1464
1465 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude) {
1466 // Push integer registers x7, x10-x17, x28-x31.
1467 push_reg(RegSet::of(x7) + RegSet::range(x10, x17) + RegSet::range(x28, x31) - exclude, sp);
1468
1469 // Push float registers f0-f7, f10-f17, f28-f31.
1470 addi(sp, sp, - wordSize * 20);
1471 int offset = 0;
1472 for (int i = 0; i < 32; i++) {
1473 if (i <= f7->encoding() || i >= f28->encoding() || (i >= f10->encoding() && i <= f17->encoding())) {
1474 fsd(as_FloatRegister(i), Address(sp, wordSize * (offset++)));
1475 }
1476 }
1477 }
1478
1479 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude) {
1480 int offset = 0;
1481 for (int i = 0; i < 32; i++) {
1482 if (i <= f7->encoding() || i >= f28->encoding() || (i >= f10->encoding() && i <= f17->encoding())) {
1483 fld(as_FloatRegister(i), Address(sp, wordSize * (offset++)));
1484 }
1485 }
1486 addi(sp, sp, wordSize * 20);
1487
1488 pop_reg(RegSet::of(x7) + RegSet::range(x10, x17) + RegSet::range(x28, x31) - exclude, sp);
1489 }
1490
1491 void MacroAssembler::push_CPU_state(bool save_vectors, int vector_size_in_bytes) {
1492 // integer registers, except zr(x0) & ra(x1) & sp(x2) & gp(x3) & tp(x4)
1493 push_reg(RegSet::range(x5, x31), sp);
1494
1495 // float registers
1496 addi(sp, sp, - 32 * wordSize);
1497 for (int i = 0; i < 32; i++) {
1498 fsd(as_FloatRegister(i), Address(sp, i * wordSize));
1499 }
1500
1501 // vector registers
1502 if (save_vectors) {
1503 sub(sp, sp, vector_size_in_bytes * VectorRegister::number_of_registers);
1504 vsetvli(t0, x0, Assembler::e64, Assembler::m8);
1505 for (int i = 0; i < VectorRegister::number_of_registers; i += 8) {
1506 add(t0, sp, vector_size_in_bytes * i);
1507 vse64_v(as_VectorRegister(i), t0);
1508 }
1509 }
1510 }
1511
1512 void MacroAssembler::pop_CPU_state(bool restore_vectors, int vector_size_in_bytes) {
1513 // vector registers
1514 if (restore_vectors) {
1515 vsetvli(t0, x0, Assembler::e64, Assembler::m8);
1516 for (int i = 0; i < VectorRegister::number_of_registers; i += 8) {
1517 vle64_v(as_VectorRegister(i), sp);
1518 add(sp, sp, vector_size_in_bytes * 8);
1519 }
1520 }
1521
1522 // float registers
1523 for (int i = 0; i < 32; i++) {
1524 fld(as_FloatRegister(i), Address(sp, i * wordSize));
1525 }
1526 addi(sp, sp, 32 * wordSize);
1527
1528 // integer registers, except zr(x0) & ra(x1) & sp(x2) & gp(x3) & tp(x4)
1529 pop_reg(RegSet::range(x5, x31), sp);
1530 }
1531
1532 static int patch_offset_in_jal(address branch, int64_t offset) {
1533 assert(Assembler::is_simm21(offset) && ((offset % 2) == 0),
1534 "offset is too large to be patched in one jal instruction!\n");
1535 Assembler::patch(branch, 31, 31, (offset >> 20) & 0x1); // offset[20] ==> branch[31]
1536 Assembler::patch(branch, 30, 21, (offset >> 1) & 0x3ff); // offset[10:1] ==> branch[30:21]
1537 Assembler::patch(branch, 20, 20, (offset >> 11) & 0x1); // offset[11] ==> branch[20]
1538 Assembler::patch(branch, 19, 12, (offset >> 12) & 0xff); // offset[19:12] ==> branch[19:12]
1539 return MacroAssembler::instruction_size; // only one instruction
1540 }
1541
1542 static int patch_offset_in_conditional_branch(address branch, int64_t offset) {
1543 assert(Assembler::is_simm13(offset) && ((offset % 2) == 0),
1544 "offset is too large to be patched in one beq/bge/bgeu/blt/bltu/bne instruction!\n");
1545 Assembler::patch(branch, 31, 31, (offset >> 12) & 0x1); // offset[12] ==> branch[31]
1546 Assembler::patch(branch, 30, 25, (offset >> 5) & 0x3f); // offset[10:5] ==> branch[30:25]
1547 Assembler::patch(branch, 7, 7, (offset >> 11) & 0x1); // offset[11] ==> branch[7]
1548 Assembler::patch(branch, 11, 8, (offset >> 1) & 0xf); // offset[4:1] ==> branch[11:8]
1549 return MacroAssembler::instruction_size; // only one instruction
1550 }
1551
1552 static int patch_offset_in_pc_relative(address branch, int64_t offset) {
1553 const int PC_RELATIVE_INSTRUCTION_NUM = 2; // auipc, addi/jalr/load
1554 Assembler::patch(branch, 31, 12, ((offset + 0x800) >> 12) & 0xfffff); // Auipc. offset[31:12] ==> branch[31:12]
1555 Assembler::patch(branch + 4, 31, 20, offset & 0xfff); // Addi/Jalr/Load. offset[11:0] ==> branch[31:20]
1556 return PC_RELATIVE_INSTRUCTION_NUM * MacroAssembler::instruction_size;
1557 }
1558
1559 static int patch_addr_in_movptr1(address branch, address target) {
1560 int32_t lower = ((intptr_t)target << 35) >> 35;
1561 int64_t upper = ((intptr_t)target - lower) >> 29;
1562 Assembler::patch(branch + 0, 31, 12, upper & 0xfffff); // Lui. target[48:29] + target[28] ==> branch[31:12]
1563 Assembler::patch(branch + 4, 31, 20, (lower >> 17) & 0xfff); // Addi. target[28:17] ==> branch[31:20]
1564 Assembler::patch(branch + 12, 31, 20, (lower >> 6) & 0x7ff); // Addi. target[16: 6] ==> branch[31:20]
1565 Assembler::patch(branch + 20, 31, 20, lower & 0x3f); // Addi/Jalr/Load. target[ 5: 0] ==> branch[31:20]
1566 return MacroAssembler::movptr1_instruction_size;
1567 }
1568
1569 static int patch_addr_in_movptr2(address instruction_address, address target) {
1570 uintptr_t addr = (uintptr_t)target;
1571
1572 assert(addr < (1ull << 48), "48-bit overflow in address constant");
1573 unsigned int upper18 = (addr >> 30ull);
1574 int lower30 = (addr & 0x3fffffffu);
1575 int low12 = (lower30 << 20) >> 20;
1576 int mid18 = ((lower30 - low12) >> 12);
1577
1578 Assembler::patch(instruction_address + (MacroAssembler::instruction_size * 0), 31, 12, (upper18 & 0xfffff)); // Lui
1579 Assembler::patch(instruction_address + (MacroAssembler::instruction_size * 1), 31, 12, (mid18 & 0xfffff)); // Lui
1580 // Slli
1581 // Add
1582 Assembler::patch(instruction_address + (MacroAssembler::instruction_size * 4), 31, 20, low12 & 0xfff); // Addi/Jalr/Load
1583
1584 assert(MacroAssembler::target_addr_for_insn(instruction_address) == target, "Must be");
1585
1586 return MacroAssembler::movptr2_instruction_size;
1587 }
1588
1589 static int patch_imm_in_li64(address branch, address target) {
1590 const int LI64_INSTRUCTIONS_NUM = 8; // lui + addi + slli + addi + slli + addi + slli + addi
1591 int64_t lower = (intptr_t)target & 0xffffffff;
1592 lower = lower - ((lower << 44) >> 44);
1593 int64_t tmp_imm = ((uint64_t)((intptr_t)target & 0xffffffff00000000)) + (uint64_t)lower;
1594 int32_t upper = (tmp_imm - (int32_t)lower) >> 32;
1595 int64_t tmp_upper = upper, tmp_lower = upper;
1596 tmp_lower = (tmp_lower << 52) >> 52;
1597 tmp_upper -= tmp_lower;
1598 tmp_upper >>= 12;
1599 // Load upper 32 bits. Upper = target[63:32], but if target[31] = 1 or (target[31:20] == 0x7ff && target[19] == 1),
1600 // upper = target[63:32] + 1.
1601 Assembler::patch(branch + 0, 31, 12, tmp_upper & 0xfffff); // Lui.
1602 Assembler::patch(branch + 4, 31, 20, tmp_lower & 0xfff); // Addi.
1603 // Load the rest 32 bits.
1604 Assembler::patch(branch + 12, 31, 20, ((int32_t)lower >> 20) & 0xfff); // Addi.
1605 Assembler::patch(branch + 20, 31, 20, (((intptr_t)target << 44) >> 52) & 0xfff); // Addi.
1606 Assembler::patch(branch + 28, 31, 20, (intptr_t)target & 0xff); // Addi.
1607 return LI64_INSTRUCTIONS_NUM * MacroAssembler::instruction_size;
1608 }
1609
1610 static int patch_imm_in_li16u(address branch, uint16_t target) {
1611 Assembler::patch(branch, 31, 12, target); // patch lui only
1612 return MacroAssembler::instruction_size;
1613 }
1614
1615 int MacroAssembler::patch_imm_in_li32(address branch, int32_t target) {
1616 const int LI32_INSTRUCTIONS_NUM = 2; // lui + addiw
1617 int64_t upper = (intptr_t)target;
1618 int32_t lower = (((int32_t)target) << 20) >> 20;
1619 upper -= lower;
1620 upper = (int32_t)upper;
1621 Assembler::patch(branch + 0, 31, 12, (upper >> 12) & 0xfffff); // Lui.
1622 Assembler::patch(branch + 4, 31, 20, lower & 0xfff); // Addiw.
1623 return LI32_INSTRUCTIONS_NUM * MacroAssembler::instruction_size;
1624 }
1625
1626 static long get_offset_of_jal(address insn_addr) {
1627 assert_cond(insn_addr != nullptr);
1628 long offset = 0;
1629 unsigned insn = Assembler::ld_instr(insn_addr);
1630 long val = (long)Assembler::sextract(insn, 31, 12);
1631 offset |= ((val >> 19) & 0x1) << 20;
1632 offset |= (val & 0xff) << 12;
1633 offset |= ((val >> 8) & 0x1) << 11;
1634 offset |= ((val >> 9) & 0x3ff) << 1;
1635 offset = (offset << 43) >> 43;
1636 return offset;
1637 }
1638
1639 static long get_offset_of_conditional_branch(address insn_addr) {
1640 long offset = 0;
1641 assert_cond(insn_addr != nullptr);
1642 unsigned insn = Assembler::ld_instr(insn_addr);
1643 offset = (long)Assembler::sextract(insn, 31, 31);
1644 offset = (offset << 12) | (((long)(Assembler::sextract(insn, 7, 7) & 0x1)) << 11);
1645 offset = offset | (((long)(Assembler::sextract(insn, 30, 25) & 0x3f)) << 5);
1646 offset = offset | (((long)(Assembler::sextract(insn, 11, 8) & 0xf)) << 1);
1647 offset = (offset << 41) >> 41;
1648 return offset;
1649 }
1650
1651 static long get_offset_of_pc_relative(address insn_addr) {
1652 long offset = 0;
1653 assert_cond(insn_addr != nullptr);
1654 offset = ((long)(Assembler::sextract(Assembler::ld_instr(insn_addr), 31, 12))) << 12; // Auipc.
1655 offset += ((long)Assembler::sextract(Assembler::ld_instr(insn_addr + 4), 31, 20)); // Addi/Jalr/Load.
1656 offset = (offset << 32) >> 32;
1657 return offset;
1658 }
1659
1660 static address get_target_of_movptr1(address insn_addr) {
1661 assert_cond(insn_addr != nullptr);
1662 intptr_t target_address = (((int64_t)Assembler::sextract(Assembler::ld_instr(insn_addr), 31, 12)) & 0xfffff) << 29; // Lui.
1663 target_address += ((int64_t)Assembler::sextract(Assembler::ld_instr(insn_addr + 4), 31, 20)) << 17; // Addi.
1664 target_address += ((int64_t)Assembler::sextract(Assembler::ld_instr(insn_addr + 12), 31, 20)) << 6; // Addi.
1665 target_address += ((int64_t)Assembler::sextract(Assembler::ld_instr(insn_addr + 20), 31, 20)); // Addi/Jalr/Load.
1666 return (address) target_address;
1667 }
1668
1669 static address get_target_of_movptr2(address insn_addr) {
1670 assert_cond(insn_addr != nullptr);
1671 int32_t upper18 = ((Assembler::sextract(Assembler::ld_instr(insn_addr + MacroAssembler::instruction_size * 0), 31, 12)) & 0xfffff); // Lui
1672 int32_t mid18 = ((Assembler::sextract(Assembler::ld_instr(insn_addr + MacroAssembler::instruction_size * 1), 31, 12)) & 0xfffff); // Lui
1673 // 2 // Slli
1674 // 3 // Add
1675 int32_t low12 = ((Assembler::sextract(Assembler::ld_instr(insn_addr + MacroAssembler::instruction_size * 4), 31, 20))); // Addi/Jalr/Load.
1676 address ret = (address)(((intptr_t)upper18<<30ll) + ((intptr_t)mid18<<12ll) + low12);
1677 return ret;
1678 }
1679
1680 static address get_target_of_li64(address insn_addr) {
1681 assert_cond(insn_addr != nullptr);
1682 intptr_t target_address = (((int64_t)Assembler::sextract(Assembler::ld_instr(insn_addr), 31, 12)) & 0xfffff) << 44; // Lui.
1683 target_address += ((int64_t)Assembler::sextract(Assembler::ld_instr(insn_addr + 4), 31, 20)) << 32; // Addi.
1684 target_address += ((int64_t)Assembler::sextract(Assembler::ld_instr(insn_addr + 12), 31, 20)) << 20; // Addi.
1685 target_address += ((int64_t)Assembler::sextract(Assembler::ld_instr(insn_addr + 20), 31, 20)) << 8; // Addi.
1686 target_address += ((int64_t)Assembler::sextract(Assembler::ld_instr(insn_addr + 28), 31, 20)); // Addi.
1687 return (address)target_address;
1688 }
1689
1690 address MacroAssembler::get_target_of_li32(address insn_addr) {
1691 assert_cond(insn_addr != nullptr);
1692 intptr_t target_address = (((int64_t)Assembler::sextract(Assembler::ld_instr(insn_addr), 31, 12)) & 0xfffff) << 12; // Lui.
1693 target_address += ((int64_t)Assembler::sextract(Assembler::ld_instr(insn_addr + 4), 31, 20)); // Addiw.
1694 return (address)target_address;
1695 }
1696
1697 // Patch any kind of instruction; there may be several instructions.
1698 // Return the total length (in bytes) of the instructions.
1699 int MacroAssembler::pd_patch_instruction_size(address instruction_address, address target) {
1700 assert_cond(instruction_address != nullptr);
1701 int64_t offset = target - instruction_address;
1702 if (MacroAssembler::is_jal_at(instruction_address)) { // jal
1703 return patch_offset_in_jal(instruction_address, offset);
1704 } else if (MacroAssembler::is_branch_at(instruction_address)) { // beq/bge/bgeu/blt/bltu/bne
1705 return patch_offset_in_conditional_branch(instruction_address, offset);
1706 } else if (MacroAssembler::is_pc_relative_at(instruction_address)) { // auipc, addi/jalr/load
1707 return patch_offset_in_pc_relative(instruction_address, offset);
1708 } else if (MacroAssembler::is_movptr1_at(instruction_address)) { // movptr1
1709 return patch_addr_in_movptr1(instruction_address, target);
1710 } else if (MacroAssembler::is_movptr2_at(instruction_address)) { // movptr2
1711 return patch_addr_in_movptr2(instruction_address, target);
1712 } else if (MacroAssembler::is_li64_at(instruction_address)) { // li64
1713 return patch_imm_in_li64(instruction_address, target);
1714 } else if (MacroAssembler::is_li32_at(instruction_address)) { // li32
1715 int64_t imm = (intptr_t)target;
1716 return patch_imm_in_li32(instruction_address, (int32_t)imm);
1717 } else if (MacroAssembler::is_li16u_at(instruction_address)) {
1718 int64_t imm = (intptr_t)target;
1719 return patch_imm_in_li16u(instruction_address, (uint16_t)imm);
1720 } else {
1721 #ifdef ASSERT
1722 tty->print_cr("pd_patch_instruction_size: instruction 0x%x at " INTPTR_FORMAT " could not be patched!\n",
1723 Assembler::ld_instr(instruction_address), p2i(instruction_address));
1724 Disassembler::decode(instruction_address - 16, instruction_address + 16);
1725 #endif
1726 ShouldNotReachHere();
1727 return -1;
1728 }
1729 }
1730
1731 address MacroAssembler::target_addr_for_insn(address insn_addr) {
1732 long offset = 0;
1733 assert_cond(insn_addr != nullptr);
1734 if (MacroAssembler::is_jal_at(insn_addr)) { // jal
1735 offset = get_offset_of_jal(insn_addr);
1736 } else if (MacroAssembler::is_branch_at(insn_addr)) { // beq/bge/bgeu/blt/bltu/bne
1737 offset = get_offset_of_conditional_branch(insn_addr);
1738 } else if (MacroAssembler::is_pc_relative_at(insn_addr)) { // auipc, addi/jalr/load
1739 offset = get_offset_of_pc_relative(insn_addr);
1740 } else if (MacroAssembler::is_movptr1_at(insn_addr)) { // movptr1
1741 return get_target_of_movptr1(insn_addr);
1742 } else if (MacroAssembler::is_movptr2_at(insn_addr)) { // movptr2
1743 return get_target_of_movptr2(insn_addr);
1744 } else if (MacroAssembler::is_li64_at(insn_addr)) { // li64
1745 return get_target_of_li64(insn_addr);
1746 } else if (MacroAssembler::is_li32_at(insn_addr)) { // li32
1747 return get_target_of_li32(insn_addr);
1748 } else {
1749 ShouldNotReachHere();
1750 }
1751 return address(((uintptr_t)insn_addr + offset));
1752 }
1753
1754 int MacroAssembler::patch_oop(address insn_addr, address o) {
1755 // OOPs are either narrow (32 bits) or wide (48 bits). We encode
1756 // narrow OOPs by setting the upper 16 bits in the first
1757 // instruction.
1758 if (MacroAssembler::is_li32_at(insn_addr)) {
1759 // Move narrow OOP
1760 uint32_t n = CompressedOops::narrow_oop_value(cast_to_oop(o));
1761 return patch_imm_in_li32(insn_addr, (int32_t)n);
1762 } else if (MacroAssembler::is_movptr1_at(insn_addr)) {
1763 // Move wide OOP
1764 return patch_addr_in_movptr1(insn_addr, o);
1765 } else if (MacroAssembler::is_movptr2_at(insn_addr)) {
1766 // Move wide OOP
1767 return patch_addr_in_movptr2(insn_addr, o);
1768 }
1769 ShouldNotReachHere();
1770 return -1;
1771 }
1772
1773 void MacroAssembler::reinit_heapbase() {
1774 if (UseCompressedOops) {
1775 if (Universe::is_fully_initialized()) {
1776 mv(xheapbase, CompressedOops::ptrs_base());
1777 } else {
1778 ExternalAddress target(CompressedOops::ptrs_base_addr());
1779 relocate(target.rspec(), [&] {
1780 int32_t offset;
1781 la(xheapbase, target.target(), offset);
1782 ld(xheapbase, Address(xheapbase, offset));
1783 });
1784 }
1785 }
1786 }
1787
1788 void MacroAssembler::movptr(Register Rd, address addr, Register temp) {
1789 int offset = 0;
1790 movptr(Rd, addr, offset, temp);
1791 addi(Rd, Rd, offset);
1792 }
1793
1794 void MacroAssembler::movptr(Register Rd, address addr, int32_t &offset, Register temp) {
1795 uint64_t uimm64 = (uint64_t)addr;
1796 #ifndef PRODUCT
1797 {
1798 char buffer[64];
1799 snprintf(buffer, sizeof(buffer), "0x%" PRIx64, uimm64);
1800 block_comment(buffer);
1801 }
1802 #endif
1803 assert(uimm64 < (1ull << 48), "48-bit overflow in address constant");
1804
1805 if (temp == noreg) {
1806 movptr1(Rd, uimm64, offset);
1807 } else {
1808 movptr2(Rd, uimm64, offset, temp);
1809 }
1810 }
1811
1812 void MacroAssembler::movptr1(Register Rd, uint64_t imm64, int32_t &offset) {
1813 // Load upper 31 bits
1814 //
1815 // In case of 11th bit of `lower` is 0, it's straightforward to understand.
1816 // In case of 11th bit of `lower` is 1, it's a bit tricky, to help understand,
1817 // imagine divide both `upper` and `lower` into 2 parts respectively, i.e.
1818 // [upper_20, upper_12], [lower_20, lower_12], they are the same just before
1819 // `lower = (lower << 52) >> 52;`.
1820 // After `upper -= lower;`,
1821 // upper_20' = upper_20 - (-1) == upper_20 + 1
1822 // upper_12 = 0x000
1823 // After `lui(Rd, upper);`, `Rd` = upper_20' << 12
1824 // Also divide `Rd` into 2 parts [Rd_20, Rd_12],
1825 // Rd_20 == upper_20'
1826 // Rd_12 == 0x000
1827 // After `addi(Rd, Rd, lower);`,
1828 // Rd_20 = upper_20' + (-1) == upper_20 + 1 - 1 = upper_20
1829 // Rd_12 = lower_12
1830 // So, finally Rd == [upper_20, lower_12]
1831 int64_t imm = imm64 >> 17;
1832 int64_t upper = imm, lower = imm;
1833 lower = (lower << 52) >> 52;
1834 upper -= lower;
1835 upper = (int32_t)upper;
1836 lui(Rd, upper);
1837 addi(Rd, Rd, lower);
1838
1839 // Load the rest 17 bits.
1840 slli(Rd, Rd, 11);
1841 addi(Rd, Rd, (imm64 >> 6) & 0x7ff);
1842 slli(Rd, Rd, 6);
1843
1844 // This offset will be used by following jalr/ld.
1845 offset = imm64 & 0x3f;
1846 }
1847
1848 void MacroAssembler::movptr2(Register Rd, uint64_t addr, int32_t &offset, Register tmp) {
1849 assert_different_registers(Rd, tmp, noreg);
1850
1851 // addr: [upper18, lower30[mid18, lower12]]
1852
1853 int64_t upper18 = addr >> 18;
1854 lui(tmp, upper18);
1855
1856 int64_t lower30 = addr & 0x3fffffff;
1857 int64_t mid18 = lower30, lower12 = lower30;
1858 lower12 = (lower12 << 52) >> 52;
1859 // For this tricky part (`mid18 -= lower12;` + `offset = lower12;`),
1860 // please refer to movptr1 above.
1861 mid18 -= (int32_t)lower12;
1862 lui(Rd, mid18);
1863
1864 slli(tmp, tmp, 18);
1865 add(Rd, Rd, tmp);
1866
1867 offset = lower12;
1868 }
1869
1870 void MacroAssembler::add(Register Rd, Register Rn, int64_t increment, Register temp) {
1871 if (is_simm12(increment)) {
1872 addi(Rd, Rn, increment);
1873 } else {
1874 assert_different_registers(Rn, temp);
1875 li(temp, increment);
1876 add(Rd, Rn, temp);
1877 }
1878 }
1879
1880 void MacroAssembler::addw(Register Rd, Register Rn, int32_t increment, Register temp) {
1881 if (is_simm12(increment)) {
1882 addiw(Rd, Rn, increment);
1883 } else {
1884 assert_different_registers(Rn, temp);
1885 li(temp, increment);
1886 addw(Rd, Rn, temp);
1887 }
1888 }
1889
1890 void MacroAssembler::sub(Register Rd, Register Rn, int64_t decrement, Register temp) {
1891 if (is_simm12(-decrement)) {
1892 addi(Rd, Rn, -decrement);
1893 } else {
1894 assert_different_registers(Rn, temp);
1895 li(temp, decrement);
1896 sub(Rd, Rn, temp);
1897 }
1898 }
1899
1900 void MacroAssembler::subw(Register Rd, Register Rn, int32_t decrement, Register temp) {
1901 if (is_simm12(-decrement)) {
1902 addiw(Rd, Rn, -decrement);
1903 } else {
1904 assert_different_registers(Rn, temp);
1905 li(temp, decrement);
1906 subw(Rd, Rn, temp);
1907 }
1908 }
1909
1910 void MacroAssembler::andrw(Register Rd, Register Rs1, Register Rs2) {
1911 andr(Rd, Rs1, Rs2);
1912 sign_extend(Rd, Rd, 32);
1913 }
1914
1915 void MacroAssembler::orrw(Register Rd, Register Rs1, Register Rs2) {
1916 orr(Rd, Rs1, Rs2);
1917 sign_extend(Rd, Rd, 32);
1918 }
1919
1920 void MacroAssembler::xorrw(Register Rd, Register Rs1, Register Rs2) {
1921 xorr(Rd, Rs1, Rs2);
1922 sign_extend(Rd, Rd, 32);
1923 }
1924
1925 // Rd = Rs1 & (~Rd2)
1926 void MacroAssembler::andn(Register Rd, Register Rs1, Register Rs2) {
1927 if (UseZbb) {
1928 Assembler::andn(Rd, Rs1, Rs2);
1929 return;
1930 }
1931
1932 notr(Rd, Rs2);
1933 andr(Rd, Rs1, Rd);
1934 }
1935
1936 // Rd = Rs1 | (~Rd2)
1937 void MacroAssembler::orn(Register Rd, Register Rs1, Register Rs2) {
1938 if (UseZbb) {
1939 Assembler::orn(Rd, Rs1, Rs2);
1940 return;
1941 }
1942
1943 notr(Rd, Rs2);
1944 orr(Rd, Rs1, Rd);
1945 }
1946
1947 // Note: load_unsigned_short used to be called load_unsigned_word.
1948 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
1949 int off = offset();
1950 lhu(dst, src);
1951 return off;
1952 }
1953
1954 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
1955 int off = offset();
1956 lbu(dst, src);
1957 return off;
1958 }
1959
1960 int MacroAssembler::load_signed_short(Register dst, Address src) {
1961 int off = offset();
1962 lh(dst, src);
1963 return off;
1964 }
1965
1966 int MacroAssembler::load_signed_byte(Register dst, Address src) {
1967 int off = offset();
1968 lb(dst, src);
1969 return off;
1970 }
1971
1972 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed) {
1973 switch (size_in_bytes) {
1974 case 8: ld(dst, src); break;
1975 case 4: is_signed ? lw(dst, src) : lwu(dst, src); break;
1976 case 2: is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
1977 case 1: is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
1978 default: ShouldNotReachHere();
1979 }
1980 }
1981
1982 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes) {
1983 switch (size_in_bytes) {
1984 case 8: sd(src, dst); break;
1985 case 4: sw(src, dst); break;
1986 case 2: sh(src, dst); break;
1987 case 1: sb(src, dst); break;
1988 default: ShouldNotReachHere();
1989 }
1990 }
1991
1992 // granularity is 1 OR 2 bytes per load. dst and src.base() allowed to be the same register
1993 void MacroAssembler::load_short_misaligned(Register dst, Address src, Register tmp, bool is_signed, int granularity) {
1994 if (granularity != 1 && granularity != 2) {
1995 ShouldNotReachHere();
1996 }
1997 if (AvoidUnalignedAccesses && (granularity != 2)) {
1998 assert_different_registers(dst, tmp);
1999 assert_different_registers(tmp, src.base());
2000 is_signed ? lb(tmp, Address(src.base(), src.offset() + 1)) : lbu(tmp, Address(src.base(), src.offset() + 1));
2001 slli(tmp, tmp, 8);
2002 lbu(dst, src);
2003 add(dst, dst, tmp);
2004 } else {
2005 is_signed ? lh(dst, src) : lhu(dst, src);
2006 }
2007 }
2008
2009 // granularity is 1, 2 OR 4 bytes per load, if granularity 2 or 4 then dst and src.base() allowed to be the same register
2010 void MacroAssembler::load_int_misaligned(Register dst, Address src, Register tmp, bool is_signed, int granularity) {
2011 if (AvoidUnalignedAccesses && (granularity != 4)) {
2012 switch(granularity) {
2013 case 1:
2014 assert_different_registers(dst, tmp, src.base());
2015 lbu(dst, src);
2016 lbu(tmp, Address(src.base(), src.offset() + 1));
2017 slli(tmp, tmp, 8);
2018 add(dst, dst, tmp);
2019 lbu(tmp, Address(src.base(), src.offset() + 2));
2020 slli(tmp, tmp, 16);
2021 add(dst, dst, tmp);
2022 is_signed ? lb(tmp, Address(src.base(), src.offset() + 3)) : lbu(tmp, Address(src.base(), src.offset() + 3));
2023 slli(tmp, tmp, 24);
2024 add(dst, dst, tmp);
2025 break;
2026 case 2:
2027 assert_different_registers(dst, tmp);
2028 assert_different_registers(tmp, src.base());
2029 is_signed ? lh(tmp, Address(src.base(), src.offset() + 2)) : lhu(tmp, Address(src.base(), src.offset() + 2));
2030 slli(tmp, tmp, 16);
2031 lhu(dst, src);
2032 add(dst, dst, tmp);
2033 break;
2034 default:
2035 ShouldNotReachHere();
2036 }
2037 } else {
2038 is_signed ? lw(dst, src) : lwu(dst, src);
2039 }
2040 }
2041
2042 // granularity is 1, 2, 4 or 8 bytes per load, if granularity 4 or 8 then dst and src.base() allowed to be same register
2043 void MacroAssembler::load_long_misaligned(Register dst, Address src, Register tmp, int granularity) {
2044 if (AvoidUnalignedAccesses && (granularity != 8)) {
2045 switch(granularity){
2046 case 1:
2047 assert_different_registers(dst, tmp, src.base());
2048 lbu(dst, src);
2049 lbu(tmp, Address(src.base(), src.offset() + 1));
2050 slli(tmp, tmp, 8);
2051 add(dst, dst, tmp);
2052 lbu(tmp, Address(src.base(), src.offset() + 2));
2053 slli(tmp, tmp, 16);
2054 add(dst, dst, tmp);
2055 lbu(tmp, Address(src.base(), src.offset() + 3));
2056 slli(tmp, tmp, 24);
2057 add(dst, dst, tmp);
2058 lbu(tmp, Address(src.base(), src.offset() + 4));
2059 slli(tmp, tmp, 32);
2060 add(dst, dst, tmp);
2061 lbu(tmp, Address(src.base(), src.offset() + 5));
2062 slli(tmp, tmp, 40);
2063 add(dst, dst, tmp);
2064 lbu(tmp, Address(src.base(), src.offset() + 6));
2065 slli(tmp, tmp, 48);
2066 add(dst, dst, tmp);
2067 lbu(tmp, Address(src.base(), src.offset() + 7));
2068 slli(tmp, tmp, 56);
2069 add(dst, dst, tmp);
2070 break;
2071 case 2:
2072 assert_different_registers(dst, tmp, src.base());
2073 lhu(dst, src);
2074 lhu(tmp, Address(src.base(), src.offset() + 2));
2075 slli(tmp, tmp, 16);
2076 add(dst, dst, tmp);
2077 lhu(tmp, Address(src.base(), src.offset() + 4));
2078 slli(tmp, tmp, 32);
2079 add(dst, dst, tmp);
2080 lhu(tmp, Address(src.base(), src.offset() + 6));
2081 slli(tmp, tmp, 48);
2082 add(dst, dst, tmp);
2083 break;
2084 case 4:
2085 assert_different_registers(dst, tmp);
2086 assert_different_registers(tmp, src.base());
2087 lwu(tmp, Address(src.base(), src.offset() + 4));
2088 slli(tmp, tmp, 32);
2089 lwu(dst, src);
2090 add(dst, dst, tmp);
2091 break;
2092 default:
2093 ShouldNotReachHere();
2094 }
2095 } else {
2096 ld(dst, src);
2097 }
2098 }
2099
2100
2101 // reverse bytes in halfword in lower 16 bits and sign-extend
2102 // Rd[15:0] = Rs[7:0] Rs[15:8] (sign-extend to 64 bits)
2103 void MacroAssembler::revb_h_h(Register Rd, Register Rs, Register tmp) {
2104 if (UseZbb) {
2105 rev8(Rd, Rs);
2106 srai(Rd, Rd, 48);
2107 return;
2108 }
2109 assert_different_registers(Rs, tmp);
2110 assert_different_registers(Rd, tmp);
2111 srli(tmp, Rs, 8);
2112 andi(tmp, tmp, 0xFF);
2113 slli(Rd, Rs, 56);
2114 srai(Rd, Rd, 48); // sign-extend
2115 orr(Rd, Rd, tmp);
2116 }
2117
2118 // reverse bytes in lower word and sign-extend
2119 // Rd[31:0] = Rs[7:0] Rs[15:8] Rs[23:16] Rs[31:24] (sign-extend to 64 bits)
2120 void MacroAssembler::revb_w_w(Register Rd, Register Rs, Register tmp1, Register tmp2) {
2121 if (UseZbb) {
2122 rev8(Rd, Rs);
2123 srai(Rd, Rd, 32);
2124 return;
2125 }
2126 assert_different_registers(Rs, tmp1, tmp2);
2127 assert_different_registers(Rd, tmp1, tmp2);
2128 revb_h_w_u(Rd, Rs, tmp1, tmp2);
2129 slli(tmp2, Rd, 48);
2130 srai(tmp2, tmp2, 32); // sign-extend
2131 srli(Rd, Rd, 16);
2132 orr(Rd, Rd, tmp2);
2133 }
2134
2135 // reverse bytes in halfword in lower 16 bits and zero-extend
2136 // Rd[15:0] = Rs[7:0] Rs[15:8] (zero-extend to 64 bits)
2137 void MacroAssembler::revb_h_h_u(Register Rd, Register Rs, Register tmp) {
2138 if (UseZbb) {
2139 rev8(Rd, Rs);
2140 srli(Rd, Rd, 48);
2141 return;
2142 }
2143 assert_different_registers(Rs, tmp);
2144 assert_different_registers(Rd, tmp);
2145 srli(tmp, Rs, 8);
2146 andi(tmp, tmp, 0xFF);
2147 andi(Rd, Rs, 0xFF);
2148 slli(Rd, Rd, 8);
2149 orr(Rd, Rd, tmp);
2150 }
2151
2152 // reverse bytes in halfwords in lower 32 bits and zero-extend
2153 // Rd[31:0] = Rs[23:16] Rs[31:24] Rs[7:0] Rs[15:8] (zero-extend to 64 bits)
2154 void MacroAssembler::revb_h_w_u(Register Rd, Register Rs, Register tmp1, Register tmp2) {
2155 if (UseZbb) {
2156 rev8(Rd, Rs);
2157 rori(Rd, Rd, 32);
2158 roriw(Rd, Rd, 16);
2159 zero_extend(Rd, Rd, 32);
2160 return;
2161 }
2162 assert_different_registers(Rs, tmp1, tmp2);
2163 assert_different_registers(Rd, tmp1, tmp2);
2164 srli(tmp2, Rs, 16);
2165 revb_h_h_u(tmp2, tmp2, tmp1);
2166 revb_h_h_u(Rd, Rs, tmp1);
2167 slli(tmp2, tmp2, 16);
2168 orr(Rd, Rd, tmp2);
2169 }
2170
2171 // This method is only used for revb_h
2172 // Rd = Rs[47:0] Rs[55:48] Rs[63:56]
2173 void MacroAssembler::revb_h_helper(Register Rd, Register Rs, Register tmp1, Register tmp2) {
2174 assert_different_registers(Rs, tmp1, tmp2);
2175 assert_different_registers(Rd, tmp1);
2176 srli(tmp1, Rs, 48);
2177 andi(tmp2, tmp1, 0xFF);
2178 slli(tmp2, tmp2, 8);
2179 srli(tmp1, tmp1, 8);
2180 orr(tmp1, tmp1, tmp2);
2181 slli(Rd, Rs, 16);
2182 orr(Rd, Rd, tmp1);
2183 }
2184
2185 // reverse bytes in each halfword
2186 // Rd[63:0] = Rs[55:48] Rs[63:56] Rs[39:32] Rs[47:40] Rs[23:16] Rs[31:24] Rs[7:0] Rs[15:8]
2187 void MacroAssembler::revb_h(Register Rd, Register Rs, Register tmp1, Register tmp2) {
2188 if (UseZbb) {
2189 assert_different_registers(Rs, tmp1);
2190 assert_different_registers(Rd, tmp1);
2191 rev8(Rd, Rs);
2192 zero_extend(tmp1, Rd, 32);
2193 roriw(tmp1, tmp1, 16);
2194 slli(tmp1, tmp1, 32);
2195 srli(Rd, Rd, 32);
2196 roriw(Rd, Rd, 16);
2197 zero_extend(Rd, Rd, 32);
2198 orr(Rd, Rd, tmp1);
2199 return;
2200 }
2201 assert_different_registers(Rs, tmp1, tmp2);
2202 assert_different_registers(Rd, tmp1, tmp2);
2203 revb_h_helper(Rd, Rs, tmp1, tmp2);
2204 for (int i = 0; i < 3; ++i) {
2205 revb_h_helper(Rd, Rd, tmp1, tmp2);
2206 }
2207 }
2208
2209 // reverse bytes in each word
2210 // Rd[63:0] = Rs[39:32] Rs[47:40] Rs[55:48] Rs[63:56] Rs[7:0] Rs[15:8] Rs[23:16] Rs[31:24]
2211 void MacroAssembler::revb_w(Register Rd, Register Rs, Register tmp1, Register tmp2) {
2212 if (UseZbb) {
2213 rev8(Rd, Rs);
2214 rori(Rd, Rd, 32);
2215 return;
2216 }
2217 assert_different_registers(Rs, tmp1, tmp2);
2218 assert_different_registers(Rd, tmp1, tmp2);
2219 revb(Rd, Rs, tmp1, tmp2);
2220 ror_imm(Rd, Rd, 32);
2221 }
2222
2223 // reverse bytes in doubleword
2224 // Rd[63:0] = Rs[7:0] Rs[15:8] Rs[23:16] Rs[31:24] Rs[39:32] Rs[47,40] Rs[55,48] Rs[63:56]
2225 void MacroAssembler::revb(Register Rd, Register Rs, Register tmp1, Register tmp2) {
2226 if (UseZbb) {
2227 rev8(Rd, Rs);
2228 return;
2229 }
2230 assert_different_registers(Rs, tmp1, tmp2);
2231 assert_different_registers(Rd, tmp1, tmp2);
2232 andi(tmp1, Rs, 0xFF);
2233 slli(tmp1, tmp1, 8);
2234 for (int step = 8; step < 56; step += 8) {
2235 srli(tmp2, Rs, step);
2236 andi(tmp2, tmp2, 0xFF);
2237 orr(tmp1, tmp1, tmp2);
2238 slli(tmp1, tmp1, 8);
2239 }
2240 srli(Rd, Rs, 56);
2241 andi(Rd, Rd, 0xFF);
2242 orr(Rd, tmp1, Rd);
2243 }
2244
2245 // rotate right with shift bits
2246 void MacroAssembler::ror_imm(Register dst, Register src, uint32_t shift, Register tmp)
2247 {
2248 if (UseZbb) {
2249 rori(dst, src, shift);
2250 return;
2251 }
2252
2253 assert_different_registers(dst, tmp);
2254 assert_different_registers(src, tmp);
2255 assert(shift < 64, "shift amount must be < 64");
2256 slli(tmp, src, 64 - shift);
2257 srli(dst, src, shift);
2258 orr(dst, dst, tmp);
2259 }
2260
2261 // rotate left with shift bits, 32-bit version
2262 void MacroAssembler::rolw_imm(Register dst, Register src, uint32_t shift, Register tmp) {
2263 if (UseZbb) {
2264 // no roliw available
2265 roriw(dst, src, 32 - shift);
2266 return;
2267 }
2268
2269 assert_different_registers(dst, tmp);
2270 assert_different_registers(src, tmp);
2271 assert(shift < 32, "shift amount must be < 32");
2272 srliw(tmp, src, 32 - shift);
2273 slliw(dst, src, shift);
2274 orr(dst, dst, tmp);
2275 }
2276
2277 void MacroAssembler::andi(Register Rd, Register Rn, int64_t imm, Register tmp) {
2278 if (is_simm12(imm)) {
2279 and_imm12(Rd, Rn, imm);
2280 } else {
2281 assert_different_registers(Rn, tmp);
2282 mv(tmp, imm);
2283 andr(Rd, Rn, tmp);
2284 }
2285 }
2286
2287 void MacroAssembler::orptr(Address adr, RegisterOrConstant src, Register tmp1, Register tmp2) {
2288 ld(tmp1, adr);
2289 if (src.is_register()) {
2290 orr(tmp1, tmp1, src.as_register());
2291 } else {
2292 if (is_simm12(src.as_constant())) {
2293 ori(tmp1, tmp1, src.as_constant());
2294 } else {
2295 assert_different_registers(tmp1, tmp2);
2296 mv(tmp2, src.as_constant());
2297 orr(tmp1, tmp1, tmp2);
2298 }
2299 }
2300 sd(tmp1, adr);
2301 }
2302
2303 void MacroAssembler::cmp_klass(Register oop, Register trial_klass, Register tmp1, Register tmp2, Label &L) {
2304 assert_different_registers(oop, trial_klass, tmp1, tmp2);
2305 if (UseCompressedClassPointers) {
2306 lwu(tmp1, Address(oop, oopDesc::klass_offset_in_bytes()));
2307 if (CompressedKlassPointers::base() == nullptr) {
2308 slli(tmp1, tmp1, CompressedKlassPointers::shift());
2309 beq(trial_klass, tmp1, L);
2310 return;
2311 }
2312 decode_klass_not_null(tmp1, tmp2);
2313 } else {
2314 ld(tmp1, Address(oop, oopDesc::klass_offset_in_bytes()));
2315 }
2316 beq(trial_klass, tmp1, L);
2317 }
2318
2319 // Move an oop into a register.
2320 void MacroAssembler::movoop(Register dst, jobject obj) {
2321 int oop_index;
2322 if (obj == nullptr) {
2323 oop_index = oop_recorder()->allocate_oop_index(obj);
2324 } else {
2325 #ifdef ASSERT
2326 {
2327 ThreadInVMfromUnknown tiv;
2328 assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
2329 }
2330 #endif
2331 oop_index = oop_recorder()->find_index(obj);
2332 }
2333 RelocationHolder rspec = oop_Relocation::spec(oop_index);
2334
2335 if (BarrierSet::barrier_set()->barrier_set_assembler()->supports_instruction_patching()) {
2336 la(dst, Address((address)obj, rspec));
2337 } else {
2338 address dummy = address(uintptr_t(pc()) & -wordSize); // A nearby aligned address
2339 ld_constant(dst, Address(dummy, rspec));
2340 }
2341 }
2342
2343 // Move a metadata address into a register.
2344 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
2345 assert((uintptr_t)obj < (1ull << 48), "48-bit overflow in metadata");
2346 int oop_index;
2347 if (obj == nullptr) {
2348 oop_index = oop_recorder()->allocate_metadata_index(obj);
2349 } else {
2350 oop_index = oop_recorder()->find_index(obj);
2351 }
2352 RelocationHolder rspec = metadata_Relocation::spec(oop_index);
2353 la(dst, Address((address)obj, rspec));
2354 }
2355
2356 // Writes to stack successive pages until offset reached to check for
2357 // stack overflow + shadow pages. This clobbers tmp.
2358 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
2359 assert_different_registers(tmp, size, t0);
2360 // Bang stack for total size given plus shadow page size.
2361 // Bang one page at a time because large size can bang beyond yellow and
2362 // red zones.
2363 mv(t0, (int)os::vm_page_size());
2364 Label loop;
2365 bind(loop);
2366 sub(tmp, sp, t0);
2367 subw(size, size, t0);
2368 sd(size, Address(tmp));
2369 bgtz(size, loop);
2370
2371 // Bang down shadow pages too.
2372 // At this point, (tmp-0) is the last address touched, so don't
2373 // touch it again. (It was touched as (tmp-pagesize) but then tmp
2374 // was post-decremented.) Skip this address by starting at i=1, and
2375 // touch a few more pages below. N.B. It is important to touch all
2376 // the way down to and including i=StackShadowPages.
2377 for (int i = 0; i < (int)(StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()) - 1; i++) {
2378 // this could be any sized move but this is can be a debugging crumb
2379 // so the bigger the better.
2380 sub(tmp, tmp, (int)os::vm_page_size());
2381 sd(size, Address(tmp, 0));
2382 }
2383 }
2384
2385 SkipIfEqual::SkipIfEqual(MacroAssembler* masm, const bool* flag_addr, bool value) {
2386 int32_t offset = 0;
2387 _masm = masm;
2388 ExternalAddress target((address)flag_addr);
2389 _masm->relocate(target.rspec(), [&] {
2390 int32_t offset;
2391 _masm->la(t0, target.target(), offset);
2392 _masm->lbu(t0, Address(t0, offset));
2393 });
2394 if (value) {
2395 _masm->bnez(t0, _label);
2396 } else {
2397 _masm->beqz(t0, _label);
2398 }
2399 }
2400
2401 SkipIfEqual::~SkipIfEqual() {
2402 _masm->bind(_label);
2403 _masm = nullptr;
2404 }
2405
2406 void MacroAssembler::load_mirror(Register dst, Register method, Register tmp1, Register tmp2) {
2407 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2408 ld(dst, Address(xmethod, Method::const_offset()));
2409 ld(dst, Address(dst, ConstMethod::constants_offset()));
2410 ld(dst, Address(dst, ConstantPool::pool_holder_offset()));
2411 ld(dst, Address(dst, mirror_offset));
2412 resolve_oop_handle(dst, tmp1, tmp2);
2413 }
2414
2415 void MacroAssembler::resolve_oop_handle(Register result, Register tmp1, Register tmp2) {
2416 // OopHandle::resolve is an indirection.
2417 assert_different_registers(result, tmp1, tmp2);
2418 access_load_at(T_OBJECT, IN_NATIVE, result, Address(result, 0), tmp1, tmp2);
2419 }
2420
2421 // ((WeakHandle)result).resolve()
2422 void MacroAssembler::resolve_weak_handle(Register result, Register tmp1, Register tmp2) {
2423 assert_different_registers(result, tmp1, tmp2);
2424 Label resolved;
2425
2426 // A null weak handle resolves to null.
2427 beqz(result, resolved);
2428
2429 // Only 64 bit platforms support GCs that require a tmp register
2430 // Only IN_HEAP loads require a thread_tmp register
2431 // WeakHandle::resolve is an indirection like jweak.
2432 access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
2433 result, Address(result), tmp1, tmp2);
2434 bind(resolved);
2435 }
2436
2437 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators,
2438 Register dst, Address src,
2439 Register tmp1, Register tmp2) {
2440 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2441 decorators = AccessInternal::decorator_fixup(decorators, type);
2442 bool as_raw = (decorators & AS_RAW) != 0;
2443 if (as_raw) {
2444 bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, tmp2);
2445 } else {
2446 bs->load_at(this, decorators, type, dst, src, tmp1, tmp2);
2447 }
2448 }
2449
2450 void MacroAssembler::null_check(Register reg, int offset) {
2451 if (needs_explicit_null_check(offset)) {
2452 // provoke OS null exception if reg is null by
2453 // accessing M[reg] w/o changing any registers
2454 // NOTE: this is plenty to provoke a segv
2455 ld(zr, Address(reg, 0));
2456 } else {
2457 // nothing to do, (later) access of M[reg + offset]
2458 // will provoke OS null exception if reg is null
2459 }
2460 }
2461
2462 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
2463 Address dst, Register val,
2464 Register tmp1, Register tmp2, Register tmp3) {
2465 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2466 decorators = AccessInternal::decorator_fixup(decorators, type);
2467 bool as_raw = (decorators & AS_RAW) != 0;
2468 if (as_raw) {
2469 bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
2470 } else {
2471 bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
2472 }
2473 }
2474
2475 // Algorithm must match CompressedOops::encode.
2476 void MacroAssembler::encode_heap_oop(Register d, Register s) {
2477 verify_oop_msg(s, "broken oop in encode_heap_oop");
2478 if (CompressedOops::base() == nullptr) {
2479 if (CompressedOops::shift() != 0) {
2480 assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
2481 srli(d, s, LogMinObjAlignmentInBytes);
2482 } else {
2483 mv(d, s);
2484 }
2485 } else {
2486 Label notNull;
2487 sub(d, s, xheapbase);
2488 bgez(d, notNull);
2489 mv(d, zr);
2490 bind(notNull);
2491 if (CompressedOops::shift() != 0) {
2492 assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
2493 srli(d, d, CompressedOops::shift());
2494 }
2495 }
2496 }
2497
2498 void MacroAssembler::load_klass(Register dst, Register src, Register tmp) {
2499 assert_different_registers(dst, tmp);
2500 assert_different_registers(src, tmp);
2501 if (UseCompressedClassPointers) {
2502 lwu(dst, Address(src, oopDesc::klass_offset_in_bytes()));
2503 decode_klass_not_null(dst, tmp);
2504 } else {
2505 ld(dst, Address(src, oopDesc::klass_offset_in_bytes()));
2506 }
2507 }
2508
2509 void MacroAssembler::store_klass(Register dst, Register src, Register tmp) {
2510 // FIXME: Should this be a store release? concurrent gcs assumes
2511 // klass length is valid if klass field is not null.
2512 if (UseCompressedClassPointers) {
2513 encode_klass_not_null(src, tmp);
2514 sw(src, Address(dst, oopDesc::klass_offset_in_bytes()));
2515 } else {
2516 sd(src, Address(dst, oopDesc::klass_offset_in_bytes()));
2517 }
2518 }
2519
2520 void MacroAssembler::store_klass_gap(Register dst, Register src) {
2521 if (UseCompressedClassPointers) {
2522 // Store to klass gap in destination
2523 sw(src, Address(dst, oopDesc::klass_gap_offset_in_bytes()));
2524 }
2525 }
2526
2527 void MacroAssembler::decode_klass_not_null(Register r, Register tmp) {
2528 assert_different_registers(r, tmp);
2529 decode_klass_not_null(r, r, tmp);
2530 }
2531
2532 void MacroAssembler::decode_klass_not_null(Register dst, Register src, Register tmp) {
2533 assert(UseCompressedClassPointers, "should only be used for compressed headers");
2534
2535 if (CompressedKlassPointers::base() == nullptr) {
2536 if (CompressedKlassPointers::shift() != 0) {
2537 assert(LogKlassAlignmentInBytes == CompressedKlassPointers::shift(), "decode alg wrong");
2538 slli(dst, src, LogKlassAlignmentInBytes);
2539 } else {
2540 mv(dst, src);
2541 }
2542 return;
2543 }
2544
2545 Register xbase = dst;
2546 if (dst == src) {
2547 xbase = tmp;
2548 }
2549
2550 assert_different_registers(src, xbase);
2551 mv(xbase, (uintptr_t)CompressedKlassPointers::base());
2552
2553 if (CompressedKlassPointers::shift() != 0) {
2554 assert(LogKlassAlignmentInBytes == CompressedKlassPointers::shift(), "decode alg wrong");
2555 assert_different_registers(t0, xbase);
2556 shadd(dst, src, xbase, t0, LogKlassAlignmentInBytes);
2557 } else {
2558 add(dst, xbase, src);
2559 }
2560 }
2561
2562 void MacroAssembler::encode_klass_not_null(Register r, Register tmp) {
2563 assert_different_registers(r, tmp);
2564 encode_klass_not_null(r, r, tmp);
2565 }
2566
2567 void MacroAssembler::encode_klass_not_null(Register dst, Register src, Register tmp) {
2568 assert(UseCompressedClassPointers, "should only be used for compressed headers");
2569
2570 if (CompressedKlassPointers::base() == nullptr) {
2571 if (CompressedKlassPointers::shift() != 0) {
2572 assert(LogKlassAlignmentInBytes == CompressedKlassPointers::shift(), "decode alg wrong");
2573 srli(dst, src, LogKlassAlignmentInBytes);
2574 } else {
2575 mv(dst, src);
2576 }
2577 return;
2578 }
2579
2580 if (((uint64_t)CompressedKlassPointers::base() & 0xffffffff) == 0 &&
2581 CompressedKlassPointers::shift() == 0) {
2582 zero_extend(dst, src, 32);
2583 return;
2584 }
2585
2586 Register xbase = dst;
2587 if (dst == src) {
2588 xbase = tmp;
2589 }
2590
2591 assert_different_registers(src, xbase);
2592 mv(xbase, (uintptr_t)CompressedKlassPointers::base());
2593 sub(dst, src, xbase);
2594 if (CompressedKlassPointers::shift() != 0) {
2595 assert(LogKlassAlignmentInBytes == CompressedKlassPointers::shift(), "decode alg wrong");
2596 srli(dst, dst, LogKlassAlignmentInBytes);
2597 }
2598 }
2599
2600 void MacroAssembler::decode_heap_oop_not_null(Register r) {
2601 decode_heap_oop_not_null(r, r);
2602 }
2603
2604 void MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
2605 assert(UseCompressedOops, "should only be used for compressed headers");
2606 assert(Universe::heap() != nullptr, "java heap should be initialized");
2607 // Cannot assert, unverified entry point counts instructions (see .ad file)
2608 // vtableStubs also counts instructions in pd_code_size_limit.
2609 // Also do not verify_oop as this is called by verify_oop.
2610 if (CompressedOops::shift() != 0) {
2611 assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
2612 slli(dst, src, LogMinObjAlignmentInBytes);
2613 if (CompressedOops::base() != nullptr) {
2614 add(dst, xheapbase, dst);
2615 }
2616 } else {
2617 assert(CompressedOops::base() == nullptr, "sanity");
2618 mv(dst, src);
2619 }
2620 }
2621
2622 void MacroAssembler::decode_heap_oop(Register d, Register s) {
2623 if (CompressedOops::base() == nullptr) {
2624 if (CompressedOops::shift() != 0 || d != s) {
2625 slli(d, s, CompressedOops::shift());
2626 }
2627 } else {
2628 Label done;
2629 mv(d, s);
2630 beqz(s, done);
2631 shadd(d, s, xheapbase, d, LogMinObjAlignmentInBytes);
2632 bind(done);
2633 }
2634 verify_oop_msg(d, "broken oop in decode_heap_oop");
2635 }
2636
2637 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
2638 Register tmp2, Register tmp3, DecoratorSet decorators) {
2639 access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
2640 }
2641
2642 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
2643 Register tmp2, DecoratorSet decorators) {
2644 access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, tmp2);
2645 }
2646
2647 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
2648 Register tmp2, DecoratorSet decorators) {
2649 access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL, dst, src, tmp1, tmp2);
2650 }
2651
2652 // Used for storing nulls.
2653 void MacroAssembler::store_heap_oop_null(Address dst) {
2654 access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
2655 }
2656
2657 int MacroAssembler::corrected_idivl(Register result, Register rs1, Register rs2,
2658 bool want_remainder, bool is_signed)
2659 {
2660 // Full implementation of Java idiv and irem. The function
2661 // returns the (pc) offset of the div instruction - may be needed
2662 // for implicit exceptions.
2663 //
2664 // input : rs1: dividend
2665 // rs2: divisor
2666 //
2667 // result: either
2668 // quotient (= rs1 idiv rs2)
2669 // remainder (= rs1 irem rs2)
2670
2671
2672 int idivl_offset = offset();
2673 if (!want_remainder) {
2674 if (is_signed) {
2675 divw(result, rs1, rs2);
2676 } else {
2677 divuw(result, rs1, rs2);
2678 }
2679 } else {
2680 // result = rs1 % rs2;
2681 if (is_signed) {
2682 remw(result, rs1, rs2);
2683 } else {
2684 remuw(result, rs1, rs2);
2685 }
2686 }
2687 return idivl_offset;
2688 }
2689
2690 int MacroAssembler::corrected_idivq(Register result, Register rs1, Register rs2,
2691 bool want_remainder, bool is_signed)
2692 {
2693 // Full implementation of Java ldiv and lrem. The function
2694 // returns the (pc) offset of the div instruction - may be needed
2695 // for implicit exceptions.
2696 //
2697 // input : rs1: dividend
2698 // rs2: divisor
2699 //
2700 // result: either
2701 // quotient (= rs1 idiv rs2)
2702 // remainder (= rs1 irem rs2)
2703
2704 int idivq_offset = offset();
2705 if (!want_remainder) {
2706 if (is_signed) {
2707 div(result, rs1, rs2);
2708 } else {
2709 divu(result, rs1, rs2);
2710 }
2711 } else {
2712 // result = rs1 % rs2;
2713 if (is_signed) {
2714 rem(result, rs1, rs2);
2715 } else {
2716 remu(result, rs1, rs2);
2717 }
2718 }
2719 return idivq_offset;
2720 }
2721
2722 // Look up the method for a megamorpic invkkeinterface call.
2723 // The target method is determined by <intf_klass, itable_index>.
2724 // The receiver klass is in recv_klass.
2725 // On success, the result will be in method_result, and execution falls through.
2726 // On failure, execution transfers to the given label.
2727 void MacroAssembler::lookup_interface_method(Register recv_klass,
2728 Register intf_klass,
2729 RegisterOrConstant itable_index,
2730 Register method_result,
2731 Register scan_tmp,
2732 Label& L_no_such_interface,
2733 bool return_method) {
2734 assert_different_registers(recv_klass, intf_klass, scan_tmp);
2735 assert_different_registers(method_result, intf_klass, scan_tmp);
2736 assert(recv_klass != method_result || !return_method,
2737 "recv_klass can be destroyed when mehtid isn't needed");
2738 assert(itable_index.is_constant() || itable_index.as_register() == method_result,
2739 "caller must be same register for non-constant itable index as for method");
2740
2741 // Compute start of first itableOffsetEntry (which is at the end of the vtable).
2742 int vtable_base = in_bytes(Klass::vtable_start_offset());
2743 int itentry_off = in_bytes(itableMethodEntry::method_offset());
2744 int scan_step = itableOffsetEntry::size() * wordSize;
2745 int vte_size = vtableEntry::size_in_bytes();
2746 assert(vte_size == wordSize, "else adjust times_vte_scale");
2747
2748 lwu(scan_tmp, Address(recv_klass, Klass::vtable_length_offset()));
2749
2750 // Could store the aligned, prescaled offset in the klass.
2751 shadd(scan_tmp, scan_tmp, recv_klass, scan_tmp, 3);
2752 add(scan_tmp, scan_tmp, vtable_base);
2753
2754 if (return_method) {
2755 // Adjust recv_klass by scaled itable_index, so we can free itable_index.
2756 assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
2757 if (itable_index.is_register()) {
2758 slli(t0, itable_index.as_register(), 3);
2759 } else {
2760 mv(t0, itable_index.as_constant() << 3);
2761 }
2762 add(recv_klass, recv_klass, t0);
2763 if (itentry_off) {
2764 add(recv_klass, recv_klass, itentry_off);
2765 }
2766 }
2767
2768 Label search, found_method;
2769
2770 ld(method_result, Address(scan_tmp, itableOffsetEntry::interface_offset()));
2771 beq(intf_klass, method_result, found_method);
2772 bind(search);
2773 // Check that the previous entry is non-null. A null entry means that
2774 // the receiver class doesn't implement the interface, and wasn't the
2775 // same as when the caller was compiled.
2776 beqz(method_result, L_no_such_interface, /* is_far */ true);
2777 addi(scan_tmp, scan_tmp, scan_step);
2778 ld(method_result, Address(scan_tmp, itableOffsetEntry::interface_offset()));
2779 bne(intf_klass, method_result, search);
2780
2781 bind(found_method);
2782
2783 // Got a hit.
2784 if (return_method) {
2785 lwu(scan_tmp, Address(scan_tmp, itableOffsetEntry::offset_offset()));
2786 add(method_result, recv_klass, scan_tmp);
2787 ld(method_result, Address(method_result));
2788 }
2789 }
2790
2791 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
2792 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
2793 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
2794 // The target method is determined by <holder_klass, itable_index>.
2795 // The receiver klass is in recv_klass.
2796 // On success, the result will be in method_result, and execution falls through.
2797 // On failure, execution transfers to the given label.
2798 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
2799 Register holder_klass,
2800 Register resolved_klass,
2801 Register method_result,
2802 Register temp_itbl_klass,
2803 Register scan_temp,
2804 int itable_index,
2805 Label& L_no_such_interface) {
2806 // 'method_result' is only used as output register at the very end of this method.
2807 // Until then we can reuse it as 'holder_offset'.
2808 Register holder_offset = method_result;
2809 assert_different_registers(resolved_klass, recv_klass, holder_klass, temp_itbl_klass, scan_temp, holder_offset);
2810
2811 int vtable_start_offset_bytes = in_bytes(Klass::vtable_start_offset());
2812 int scan_step = itableOffsetEntry::size() * wordSize;
2813 int ioffset_bytes = in_bytes(itableOffsetEntry::interface_offset());
2814 int ooffset_bytes = in_bytes(itableOffsetEntry::offset_offset());
2815 int itmentry_off_bytes = in_bytes(itableMethodEntry::method_offset());
2816 const int vte_scale = exact_log2(vtableEntry::size_in_bytes());
2817
2818 Label L_loop_search_resolved_entry, L_resolved_found, L_holder_found;
2819
2820 lwu(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
2821 add(recv_klass, recv_klass, vtable_start_offset_bytes + ioffset_bytes);
2822 // itableOffsetEntry[] itable = recv_klass + Klass::vtable_start_offset()
2823 // + sizeof(vtableEntry) * (recv_klass->_vtable_len);
2824 // scan_temp = &(itable[0]._interface)
2825 // temp_itbl_klass = itable[0]._interface;
2826 shadd(scan_temp, scan_temp, recv_klass, scan_temp, vte_scale);
2827 ld(temp_itbl_klass, Address(scan_temp));
2828 mv(holder_offset, zr);
2829
2830 // Initial checks:
2831 // - if (holder_klass != resolved_klass), go to "scan for resolved"
2832 // - if (itable[0] == holder_klass), shortcut to "holder found"
2833 // - if (itable[0] == 0), no such interface
2834 bne(resolved_klass, holder_klass, L_loop_search_resolved_entry);
2835 beq(holder_klass, temp_itbl_klass, L_holder_found);
2836 beqz(temp_itbl_klass, L_no_such_interface);
2837
2838 // Loop: Look for holder_klass record in itable
2839 // do {
2840 // temp_itbl_klass = *(scan_temp += scan_step);
2841 // if (temp_itbl_klass == holder_klass) {
2842 // goto L_holder_found; // Found!
2843 // }
2844 // } while (temp_itbl_klass != 0);
2845 // goto L_no_such_interface // Not found.
2846 Label L_search_holder;
2847 bind(L_search_holder);
2848 add(scan_temp, scan_temp, scan_step);
2849 ld(temp_itbl_klass, Address(scan_temp));
2850 beq(holder_klass, temp_itbl_klass, L_holder_found);
2851 bnez(temp_itbl_klass, L_search_holder);
2852
2853 j(L_no_such_interface);
2854
2855 // Loop: Look for resolved_class record in itable
2856 // while (true) {
2857 // temp_itbl_klass = *(scan_temp += scan_step);
2858 // if (temp_itbl_klass == 0) {
2859 // goto L_no_such_interface;
2860 // }
2861 // if (temp_itbl_klass == resolved_klass) {
2862 // goto L_resolved_found; // Found!
2863 // }
2864 // if (temp_itbl_klass == holder_klass) {
2865 // holder_offset = scan_temp;
2866 // }
2867 // }
2868 //
2869 Label L_loop_search_resolved;
2870 bind(L_loop_search_resolved);
2871 add(scan_temp, scan_temp, scan_step);
2872 ld(temp_itbl_klass, Address(scan_temp));
2873 bind(L_loop_search_resolved_entry);
2874 beqz(temp_itbl_klass, L_no_such_interface);
2875 beq(resolved_klass, temp_itbl_klass, L_resolved_found);
2876 bne(holder_klass, temp_itbl_klass, L_loop_search_resolved);
2877 mv(holder_offset, scan_temp);
2878 j(L_loop_search_resolved);
2879
2880 // See if we already have a holder klass. If not, go and scan for it.
2881 bind(L_resolved_found);
2882 beqz(holder_offset, L_search_holder);
2883 mv(scan_temp, holder_offset);
2884
2885 // Finally, scan_temp contains holder_klass vtable offset
2886 bind(L_holder_found);
2887 lwu(method_result, Address(scan_temp, ooffset_bytes - ioffset_bytes));
2888 add(recv_klass, recv_klass, itable_index * wordSize + itmentry_off_bytes
2889 - vtable_start_offset_bytes - ioffset_bytes); // substract offsets to restore the original value of recv_klass
2890 add(method_result, recv_klass, method_result);
2891 ld(method_result, Address(method_result));
2892 }
2893
2894 // virtual method calling
2895 void MacroAssembler::lookup_virtual_method(Register recv_klass,
2896 RegisterOrConstant vtable_index,
2897 Register method_result) {
2898 const ByteSize base = Klass::vtable_start_offset();
2899 assert(vtableEntry::size() * wordSize == 8,
2900 "adjust the scaling in the code below");
2901 int vtable_offset_in_bytes = in_bytes(base + vtableEntry::method_offset());
2902
2903 if (vtable_index.is_register()) {
2904 shadd(method_result, vtable_index.as_register(), recv_klass, method_result, LogBytesPerWord);
2905 ld(method_result, Address(method_result, vtable_offset_in_bytes));
2906 } else {
2907 vtable_offset_in_bytes += vtable_index.as_constant() * wordSize;
2908 ld(method_result, form_address(method_result, recv_klass, vtable_offset_in_bytes));
2909 }
2910 }
2911
2912 void MacroAssembler::membar(uint32_t order_constraint) {
2913 address prev = pc() - MacroAssembler::instruction_size;
2914 address last = code()->last_insn();
2915
2916 if (last != nullptr && is_membar(last) && prev == last) {
2917 // We are merging two memory barrier instructions. On RISCV we
2918 // can do this simply by ORing them together.
2919 set_membar_kind(prev, get_membar_kind(prev) | order_constraint);
2920 BLOCK_COMMENT("merged membar");
2921 } else {
2922 code()->set_last_insn(pc());
2923
2924 uint32_t predecessor = 0;
2925 uint32_t successor = 0;
2926
2927 membar_mask_to_pred_succ(order_constraint, predecessor, successor);
2928 fence(predecessor, successor);
2929 }
2930 }
2931
2932 // Form an address from base + offset in Rd. Rd my or may not
2933 // actually be used: you must use the Address that is returned. It
2934 // is up to you to ensure that the shift provided matches the size
2935 // of your data.
2936 Address MacroAssembler::form_address(Register Rd, Register base, int64_t byte_offset) {
2937 if (is_simm12(byte_offset)) { // 12: imm in range 2^12
2938 return Address(base, byte_offset);
2939 }
2940
2941 assert_different_registers(Rd, base, noreg);
2942
2943 // Do it the hard way
2944 mv(Rd, byte_offset);
2945 add(Rd, base, Rd);
2946 return Address(Rd);
2947 }
2948
2949 void MacroAssembler::check_klass_subtype(Register sub_klass,
2950 Register super_klass,
2951 Register tmp_reg,
2952 Label& L_success) {
2953 Label L_failure;
2954 check_klass_subtype_fast_path(sub_klass, super_klass, tmp_reg, &L_success, &L_failure, nullptr);
2955 check_klass_subtype_slow_path(sub_klass, super_klass, tmp_reg, noreg, &L_success, nullptr);
2956 bind(L_failure);
2957 }
2958
2959 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool acquire, bool in_nmethod) {
2960 ld(t0, Address(xthread, JavaThread::polling_word_offset()));
2961 if (acquire) {
2962 membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
2963 }
2964 if (at_return) {
2965 bgtu(in_nmethod ? sp : fp, t0, slow_path, /* is_far */ true);
2966 } else {
2967 test_bit(t0, t0, exact_log2(SafepointMechanism::poll_bit()));
2968 bnez(t0, slow_path, true /* is_far */);
2969 }
2970 }
2971
2972 void MacroAssembler::cmpxchgptr(Register oldv, Register newv, Register addr, Register tmp,
2973 Label &succeed, Label *fail) {
2974 assert_different_registers(addr, tmp, t0);
2975 assert_different_registers(newv, tmp, t0);
2976 assert_different_registers(oldv, tmp, t0);
2977
2978 // oldv holds comparison value
2979 // newv holds value to write in exchange
2980 // addr identifies memory word to compare against/update
2981 if (UseZacas) {
2982 mv(tmp, oldv);
2983 atomic_cas(tmp, newv, addr, Assembler::int64, Assembler::aq, Assembler::rl);
2984 beq(tmp, oldv, succeed);
2985 } else {
2986 Label retry_load, nope;
2987 bind(retry_load);
2988 // Load reserved from the memory location
2989 load_reserved(tmp, addr, int64, Assembler::aqrl);
2990 // Fail and exit if it is not what we expect
2991 bne(tmp, oldv, nope);
2992 // If the store conditional succeeds, tmp will be zero
2993 store_conditional(tmp, newv, addr, int64, Assembler::rl);
2994 beqz(tmp, succeed);
2995 // Retry only when the store conditional failed
2996 j(retry_load);
2997
2998 bind(nope);
2999 }
3000
3001 // neither amocas nor lr/sc have an implied barrier in the failing case
3002 membar(AnyAny);
3003
3004 mv(oldv, tmp);
3005 if (fail != nullptr) {
3006 j(*fail);
3007 }
3008 }
3009
3010 void MacroAssembler::cmpxchg_obj_header(Register oldv, Register newv, Register obj, Register tmp,
3011 Label &succeed, Label *fail) {
3012 assert(oopDesc::mark_offset_in_bytes() == 0, "assumption");
3013 cmpxchgptr(oldv, newv, obj, tmp, succeed, fail);
3014 }
3015
3016 void MacroAssembler::load_reserved(Register dst,
3017 Register addr,
3018 enum operand_size size,
3019 Assembler::Aqrl acquire) {
3020 switch (size) {
3021 case int64:
3022 lr_d(dst, addr, acquire);
3023 break;
3024 case int32:
3025 lr_w(dst, addr, acquire);
3026 break;
3027 case uint32:
3028 lr_w(dst, addr, acquire);
3029 zero_extend(dst, dst, 32);
3030 break;
3031 default:
3032 ShouldNotReachHere();
3033 }
3034 }
3035
3036 void MacroAssembler::store_conditional(Register dst,
3037 Register new_val,
3038 Register addr,
3039 enum operand_size size,
3040 Assembler::Aqrl release) {
3041 switch (size) {
3042 case int64:
3043 sc_d(dst, new_val, addr, release);
3044 break;
3045 case int32:
3046 case uint32:
3047 sc_w(dst, new_val, addr, release);
3048 break;
3049 default:
3050 ShouldNotReachHere();
3051 }
3052 }
3053
3054
3055 void MacroAssembler::cmpxchg_narrow_value_helper(Register addr, Register expected,
3056 Register new_val,
3057 enum operand_size size,
3058 Register tmp1, Register tmp2, Register tmp3) {
3059 assert(size == int8 || size == int16, "unsupported operand size");
3060
3061 Register aligned_addr = t1, shift = tmp1, mask = tmp2, not_mask = tmp3;
3062
3063 andi(shift, addr, 3);
3064 slli(shift, shift, 3);
3065
3066 andi(aligned_addr, addr, ~3);
3067
3068 if (size == int8) {
3069 mv(mask, 0xff);
3070 } else {
3071 // size == int16 case
3072 mv(mask, -1);
3073 zero_extend(mask, mask, 16);
3074 }
3075 sll(mask, mask, shift);
3076
3077 notr(not_mask, mask);
3078
3079 sll(expected, expected, shift);
3080 andr(expected, expected, mask);
3081
3082 sll(new_val, new_val, shift);
3083 andr(new_val, new_val, mask);
3084 }
3085
3086 // cmpxchg_narrow_value will kill t0, t1, expected, new_val and tmps.
3087 // It's designed to implement compare and swap byte/boolean/char/short by lr.w/sc.w or amocas.w,
3088 // which are forced to work with 4-byte aligned address.
3089 void MacroAssembler::cmpxchg_narrow_value(Register addr, Register expected,
3090 Register new_val,
3091 enum operand_size size,
3092 Assembler::Aqrl acquire, Assembler::Aqrl release,
3093 Register result, bool result_as_bool,
3094 Register tmp1, Register tmp2, Register tmp3) {
3095 Register aligned_addr = t1, shift = tmp1, mask = tmp2, not_mask = tmp3, old = result, tmp = t0;
3096 assert_different_registers(addr, old, mask, not_mask, new_val, expected, shift, tmp);
3097 cmpxchg_narrow_value_helper(addr, expected, new_val, size, tmp1, tmp2, tmp3);
3098
3099 Label retry, fail, done;
3100
3101 bind(retry);
3102
3103 if (UseZacas) {
3104 lw(old, aligned_addr);
3105
3106 // if old & mask != expected
3107 andr(tmp, old, mask);
3108 bne(tmp, expected, fail);
3109
3110 andr(tmp, old, not_mask);
3111 orr(tmp, tmp, new_val);
3112
3113 atomic_cas(old, tmp, aligned_addr, operand_size::int32, acquire, release);
3114 bne(tmp, old, retry);
3115 } else {
3116 lr_w(old, aligned_addr, acquire);
3117 andr(tmp, old, mask);
3118 bne(tmp, expected, fail);
3119
3120 andr(tmp, old, not_mask);
3121 orr(tmp, tmp, new_val);
3122 sc_w(tmp, tmp, aligned_addr, release);
3123 bnez(tmp, retry);
3124 }
3125
3126 if (result_as_bool) {
3127 mv(result, 1);
3128 j(done);
3129
3130 bind(fail);
3131 mv(result, zr);
3132
3133 bind(done);
3134 } else {
3135 andr(tmp, old, mask);
3136
3137 bind(fail);
3138 srl(result, tmp, shift);
3139
3140 if (size == int8) {
3141 sign_extend(result, result, 8);
3142 } else {
3143 // size == int16 case
3144 sign_extend(result, result, 16);
3145 }
3146 }
3147 }
3148
3149 // weak_cmpxchg_narrow_value is a weak version of cmpxchg_narrow_value, to implement
3150 // the weak CAS stuff. The major difference is that it just failed when store conditional
3151 // failed.
3152 void MacroAssembler::weak_cmpxchg_narrow_value(Register addr, Register expected,
3153 Register new_val,
3154 enum operand_size size,
3155 Assembler::Aqrl acquire, Assembler::Aqrl release,
3156 Register result,
3157 Register tmp1, Register tmp2, Register tmp3) {
3158 Register aligned_addr = t1, shift = tmp1, mask = tmp2, not_mask = tmp3, old = result, tmp = t0;
3159 assert_different_registers(addr, old, mask, not_mask, new_val, expected, shift, tmp);
3160 cmpxchg_narrow_value_helper(addr, expected, new_val, size, tmp1, tmp2, tmp3);
3161
3162 Label fail, done;
3163
3164 if (UseZacas) {
3165 lw(old, aligned_addr);
3166
3167 // if old & mask != expected
3168 andr(tmp, old, mask);
3169 bne(tmp, expected, fail);
3170
3171 andr(tmp, old, not_mask);
3172 orr(tmp, tmp, new_val);
3173
3174 atomic_cas(tmp, new_val, addr, operand_size::int32, acquire, release);
3175 bne(tmp, old, fail);
3176 } else {
3177 lr_w(old, aligned_addr, acquire);
3178 andr(tmp, old, mask);
3179 bne(tmp, expected, fail);
3180
3181 andr(tmp, old, not_mask);
3182 orr(tmp, tmp, new_val);
3183 sc_w(tmp, tmp, aligned_addr, release);
3184 bnez(tmp, fail);
3185 }
3186
3187 // Success
3188 mv(result, 1);
3189 j(done);
3190
3191 // Fail
3192 bind(fail);
3193 mv(result, zr);
3194
3195 bind(done);
3196 }
3197
3198 void MacroAssembler::cmpxchg(Register addr, Register expected,
3199 Register new_val,
3200 enum operand_size size,
3201 Assembler::Aqrl acquire, Assembler::Aqrl release,
3202 Register result, bool result_as_bool) {
3203 assert(size != int8 && size != int16, "unsupported operand size");
3204 assert_different_registers(addr, t0);
3205 assert_different_registers(expected, t0);
3206 assert_different_registers(new_val, t0);
3207
3208 if (UseZacas) {
3209 if (result_as_bool) {
3210 mv(t0, expected);
3211 atomic_cas(t0, new_val, addr, size, acquire, release);
3212 xorr(t0, t0, expected);
3213 seqz(result, t0);
3214 } else {
3215 mv(result, expected);
3216 atomic_cas(result, new_val, addr, size, acquire, release);
3217 }
3218 return;
3219 }
3220
3221 Label retry_load, done, ne_done;
3222 bind(retry_load);
3223 load_reserved(t0, addr, size, acquire);
3224 bne(t0, expected, ne_done);
3225 store_conditional(t0, new_val, addr, size, release);
3226 bnez(t0, retry_load);
3227
3228 // equal, succeed
3229 if (result_as_bool) {
3230 mv(result, 1);
3231 } else {
3232 mv(result, expected);
3233 }
3234 j(done);
3235
3236 // not equal, failed
3237 bind(ne_done);
3238 if (result_as_bool) {
3239 mv(result, zr);
3240 } else {
3241 mv(result, t0);
3242 }
3243
3244 bind(done);
3245 }
3246
3247 void MacroAssembler::cmpxchg_weak(Register addr, Register expected,
3248 Register new_val,
3249 enum operand_size size,
3250 Assembler::Aqrl acquire, Assembler::Aqrl release,
3251 Register result) {
3252 if (UseZacas) {
3253 cmpxchg(addr, expected, new_val, size, acquire, release, result, true);
3254 return;
3255 }
3256
3257 assert_different_registers(addr, t0);
3258 assert_different_registers(expected, t0);
3259 assert_different_registers(new_val, t0);
3260
3261 Label fail, done;
3262 load_reserved(t0, addr, size, acquire);
3263 bne(t0, expected, fail);
3264 store_conditional(t0, new_val, addr, size, release);
3265 bnez(t0, fail);
3266
3267 // Success
3268 mv(result, 1);
3269 j(done);
3270
3271 // Fail
3272 bind(fail);
3273 mv(result, zr);
3274
3275 bind(done);
3276 }
3277
3278 #define ATOMIC_OP(NAME, AOP, ACQUIRE, RELEASE) \
3279 void MacroAssembler::atomic_##NAME(Register prev, RegisterOrConstant incr, Register addr) { \
3280 prev = prev->is_valid() ? prev : zr; \
3281 if (incr.is_register()) { \
3282 AOP(prev, addr, incr.as_register(), (Assembler::Aqrl)(ACQUIRE | RELEASE)); \
3283 } else { \
3284 mv(t0, incr.as_constant()); \
3285 AOP(prev, addr, t0, (Assembler::Aqrl)(ACQUIRE | RELEASE)); \
3286 } \
3287 return; \
3288 }
3289
3290 ATOMIC_OP(add, amoadd_d, Assembler::relaxed, Assembler::relaxed)
3291 ATOMIC_OP(addw, amoadd_w, Assembler::relaxed, Assembler::relaxed)
3292 ATOMIC_OP(addal, amoadd_d, Assembler::aq, Assembler::rl)
3293 ATOMIC_OP(addalw, amoadd_w, Assembler::aq, Assembler::rl)
3294
3295 #undef ATOMIC_OP
3296
3297 #define ATOMIC_XCHG(OP, AOP, ACQUIRE, RELEASE) \
3298 void MacroAssembler::atomic_##OP(Register prev, Register newv, Register addr) { \
3299 prev = prev->is_valid() ? prev : zr; \
3300 AOP(prev, addr, newv, (Assembler::Aqrl)(ACQUIRE | RELEASE)); \
3301 return; \
3302 }
3303
3304 ATOMIC_XCHG(xchg, amoswap_d, Assembler::relaxed, Assembler::relaxed)
3305 ATOMIC_XCHG(xchgw, amoswap_w, Assembler::relaxed, Assembler::relaxed)
3306 ATOMIC_XCHG(xchgal, amoswap_d, Assembler::aq, Assembler::rl)
3307 ATOMIC_XCHG(xchgalw, amoswap_w, Assembler::aq, Assembler::rl)
3308
3309 #undef ATOMIC_XCHG
3310
3311 #define ATOMIC_XCHGU(OP1, OP2) \
3312 void MacroAssembler::atomic_##OP1(Register prev, Register newv, Register addr) { \
3313 atomic_##OP2(prev, newv, addr); \
3314 zero_extend(prev, prev, 32); \
3315 return; \
3316 }
3317
3318 ATOMIC_XCHGU(xchgwu, xchgw)
3319 ATOMIC_XCHGU(xchgalwu, xchgalw)
3320
3321 #undef ATOMIC_XCHGU
3322
3323 #define ATOMIC_CAS(OP, AOP, ACQUIRE, RELEASE) \
3324 void MacroAssembler::atomic_##OP(Register prev, Register newv, Register addr) { \
3325 assert(UseZacas, "invariant"); \
3326 prev = prev->is_valid() ? prev : zr; \
3327 AOP(prev, addr, newv, (Assembler::Aqrl)(ACQUIRE | RELEASE)); \
3328 return; \
3329 }
3330
3331 ATOMIC_CAS(cas, amocas_d, Assembler::relaxed, Assembler::relaxed)
3332 ATOMIC_CAS(casw, amocas_w, Assembler::relaxed, Assembler::relaxed)
3333 ATOMIC_CAS(casl, amocas_d, Assembler::relaxed, Assembler::rl)
3334 ATOMIC_CAS(caslw, amocas_w, Assembler::relaxed, Assembler::rl)
3335 ATOMIC_CAS(casal, amocas_d, Assembler::aq, Assembler::rl)
3336 ATOMIC_CAS(casalw, amocas_w, Assembler::aq, Assembler::rl)
3337
3338 #undef ATOMIC_CAS
3339
3340 #define ATOMIC_CASU(OP1, OP2) \
3341 void MacroAssembler::atomic_##OP1(Register prev, Register newv, Register addr) { \
3342 atomic_##OP2(prev, newv, addr); \
3343 zero_extend(prev, prev, 32); \
3344 return; \
3345 }
3346
3347 ATOMIC_CASU(caswu, casw)
3348 ATOMIC_CASU(caslwu, caslw)
3349 ATOMIC_CASU(casalwu, casalw)
3350
3351 #undef ATOMIC_CASU
3352
3353 void MacroAssembler::atomic_cas(
3354 Register prev, Register newv, Register addr, enum operand_size size, Assembler::Aqrl acquire, Assembler::Aqrl release) {
3355 switch (size) {
3356 case int64:
3357 switch ((Assembler::Aqrl)(acquire | release)) {
3358 case Assembler::relaxed:
3359 atomic_cas(prev, newv, addr);
3360 break;
3361 case Assembler::rl:
3362 atomic_casl(prev, newv, addr);
3363 break;
3364 case Assembler::aqrl:
3365 atomic_casal(prev, newv, addr);
3366 break;
3367 default:
3368 ShouldNotReachHere();
3369 }
3370 break;
3371 case int32:
3372 switch ((Assembler::Aqrl)(acquire | release)) {
3373 case Assembler::relaxed:
3374 atomic_casw(prev, newv, addr);
3375 break;
3376 case Assembler::rl:
3377 atomic_caslw(prev, newv, addr);
3378 break;
3379 case Assembler::aqrl:
3380 atomic_casalw(prev, newv, addr);
3381 break;
3382 default:
3383 ShouldNotReachHere();
3384 }
3385 break;
3386 case uint32:
3387 switch ((Assembler::Aqrl)(acquire | release)) {
3388 case Assembler::relaxed:
3389 atomic_caswu(prev, newv, addr);
3390 break;
3391 case Assembler::rl:
3392 atomic_caslwu(prev, newv, addr);
3393 break;
3394 case Assembler::aqrl:
3395 atomic_casalwu(prev, newv, addr);
3396 break;
3397 default:
3398 ShouldNotReachHere();
3399 }
3400 break;
3401 default:
3402 ShouldNotReachHere();
3403 }
3404 }
3405
3406 void MacroAssembler::far_jump(const Address &entry, Register tmp) {
3407 assert(CodeCache::find_blob(entry.target()) != nullptr,
3408 "destination of far call not found in code cache");
3409 assert(entry.rspec().type() == relocInfo::external_word_type
3410 || entry.rspec().type() == relocInfo::runtime_call_type
3411 || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
3412 // Fixed length: see MacroAssembler::far_branch_size()
3413 relocate(entry.rspec(), [&] {
3414 int32_t offset;
3415 la(tmp, entry.target(), offset);
3416 jr(tmp, offset);
3417 });
3418 }
3419
3420 void MacroAssembler::far_call(const Address &entry, Register tmp) {
3421 assert(CodeCache::find_blob(entry.target()) != nullptr,
3422 "destination of far call not found in code cache");
3423 assert(entry.rspec().type() == relocInfo::external_word_type
3424 || entry.rspec().type() == relocInfo::runtime_call_type
3425 || entry.rspec().type() == relocInfo::none, "wrong entry relocInfo type");
3426 // Fixed length: see MacroAssembler::far_branch_size()
3427 // We can use auipc + jalr here because we know that the total size of
3428 // the code cache cannot exceed 2Gb.
3429 relocate(entry.rspec(), [&] {
3430 assert(is_valid_32bit_offset(entry.target() - pc()), "Far call using wrong instructions.");
3431 call(entry.target(), tmp);
3432 });
3433 }
3434
3435 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
3436 Register super_klass,
3437 Register tmp_reg,
3438 Label* L_success,
3439 Label* L_failure,
3440 Label* L_slow_path,
3441 Register super_check_offset) {
3442 assert_different_registers(sub_klass, super_klass, tmp_reg);
3443 bool must_load_sco = (super_check_offset == noreg);
3444 if (must_load_sco) {
3445 assert(tmp_reg != noreg, "supply either a temp or a register offset");
3446 } else {
3447 assert_different_registers(sub_klass, super_klass, super_check_offset);
3448 }
3449
3450 Label L_fallthrough;
3451 int label_nulls = 0;
3452 if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
3453 if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
3454 if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
3455 assert(label_nulls <= 1, "at most one null in batch");
3456
3457 int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
3458 int sco_offset = in_bytes(Klass::super_check_offset_offset());
3459 Address super_check_offset_addr(super_klass, sco_offset);
3460
3461 // Hacked jmp, which may only be used just before L_fallthrough.
3462 #define final_jmp(label) \
3463 if (&(label) == &L_fallthrough) { /*do nothing*/ } \
3464 else j(label) /*omit semi*/
3465
3466 // If the pointers are equal, we are done (e.g., String[] elements).
3467 // This self-check enables sharing of secondary supertype arrays among
3468 // non-primary types such as array-of-interface. Otherwise, each such
3469 // type would need its own customized SSA.
3470 // We move this check to the front of the fast path because many
3471 // type checks are in fact trivially successful in this manner,
3472 // so we get a nicely predicted branch right at the start of the check.
3473 beq(sub_klass, super_klass, *L_success);
3474
3475 // Check the supertype display:
3476 if (must_load_sco) {
3477 lwu(tmp_reg, super_check_offset_addr);
3478 super_check_offset = tmp_reg;
3479 }
3480 add(t0, sub_klass, super_check_offset);
3481 Address super_check_addr(t0);
3482 ld(t0, super_check_addr); // load displayed supertype
3483
3484 // This check has worked decisively for primary supers.
3485 // Secondary supers are sought in the super_cache ('super_cache_addr').
3486 // (Secondary supers are interfaces and very deeply nested subtypes.)
3487 // This works in the same check above because of a tricky aliasing
3488 // between the super_Cache and the primary super display elements.
3489 // (The 'super_check_addr' can address either, as the case requires.)
3490 // Note that the cache is updated below if it does not help us find
3491 // what we need immediately.
3492 // So if it was a primary super, we can just fail immediately.
3493 // Otherwise, it's the slow path for us (no success at this point).
3494
3495 beq(super_klass, t0, *L_success);
3496 mv(t1, sc_offset);
3497 if (L_failure == &L_fallthrough) {
3498 beq(super_check_offset, t1, *L_slow_path);
3499 } else {
3500 bne(super_check_offset, t1, *L_failure, /* is_far */ true);
3501 final_jmp(*L_slow_path);
3502 }
3503
3504 bind(L_fallthrough);
3505
3506 #undef final_jmp
3507 }
3508
3509 // Scans count pointer sized words at [addr] for occurrence of value,
3510 // generic
3511 void MacroAssembler::repne_scan(Register addr, Register value, Register count,
3512 Register tmp) {
3513 Label Lloop, Lexit;
3514 beqz(count, Lexit);
3515 bind(Lloop);
3516 ld(tmp, addr);
3517 beq(value, tmp, Lexit);
3518 add(addr, addr, wordSize);
3519 sub(count, count, 1);
3520 bnez(count, Lloop);
3521 bind(Lexit);
3522 }
3523
3524 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
3525 Register super_klass,
3526 Register tmp1_reg,
3527 Register tmp2_reg,
3528 Label* L_success,
3529 Label* L_failure) {
3530 assert_different_registers(sub_klass, super_klass, tmp1_reg);
3531 if (tmp2_reg != noreg) {
3532 assert_different_registers(sub_klass, super_klass, tmp1_reg, tmp2_reg, t0);
3533 }
3534 #define IS_A_TEMP(reg) ((reg) == tmp1_reg || (reg) == tmp2_reg)
3535
3536 Label L_fallthrough;
3537 int label_nulls = 0;
3538 if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
3539 if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
3540
3541 assert(label_nulls <= 1, "at most one null in the batch");
3542
3543 // A couple of useful fields in sub_klass:
3544 int ss_offset = in_bytes(Klass::secondary_supers_offset());
3545 int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
3546 Address secondary_supers_addr(sub_klass, ss_offset);
3547 Address super_cache_addr( sub_klass, sc_offset);
3548
3549 BLOCK_COMMENT("check_klass_subtype_slow_path");
3550
3551 // Do a linear scan of the secondary super-klass chain.
3552 // This code is rarely used, so simplicity is a virtue here.
3553 // The repne_scan instruction uses fixed registers, which we must spill.
3554 // Don't worry too much about pre-existing connections with the input regs.
3555
3556 assert(sub_klass != x10, "killed reg"); // killed by mv(x10, super)
3557 assert(sub_klass != x12, "killed reg"); // killed by la(x12, &pst_counter)
3558
3559 RegSet pushed_registers;
3560 if (!IS_A_TEMP(x12)) {
3561 pushed_registers += x12;
3562 }
3563 if (!IS_A_TEMP(x15)) {
3564 pushed_registers += x15;
3565 }
3566
3567 if (super_klass != x10) {
3568 if (!IS_A_TEMP(x10)) {
3569 pushed_registers += x10;
3570 }
3571 }
3572
3573 push_reg(pushed_registers, sp);
3574
3575 // Get super_klass value into x10 (even if it was in x15 or x12)
3576 mv(x10, super_klass);
3577
3578 #ifndef PRODUCT
3579 incrementw(ExternalAddress((address)&SharedRuntime::_partial_subtype_ctr));
3580 #endif // PRODUCT
3581
3582 // We will consult the secondary-super array.
3583 ld(x15, secondary_supers_addr);
3584 // Load the array length.
3585 lwu(x12, Address(x15, Array<Klass*>::length_offset_in_bytes()));
3586 // Skip to start of data.
3587 add(x15, x15, Array<Klass*>::base_offset_in_bytes());
3588
3589 // Set t0 to an obvious invalid value, falling through by default
3590 mv(t0, -1);
3591 // Scan X12 words at [X15] for an occurrence of X10.
3592 repne_scan(x15, x10, x12, t0);
3593
3594 // pop will restore x10, so we should use a temp register to keep its value
3595 mv(t1, x10);
3596
3597 // Unspill the temp registers:
3598 pop_reg(pushed_registers, sp);
3599
3600 bne(t1, t0, *L_failure);
3601
3602 // Success. Cache the super we found an proceed in triumph.
3603 sd(super_klass, super_cache_addr);
3604
3605 if (L_success != &L_fallthrough) {
3606 j(*L_success);
3607 }
3608
3609 #undef IS_A_TEMP
3610
3611 bind(L_fallthrough);
3612 }
3613
3614 // Defines obj, preserves var_size_in_bytes, okay for tmp2 == var_size_in_bytes.
3615 void MacroAssembler::tlab_allocate(Register obj,
3616 Register var_size_in_bytes,
3617 int con_size_in_bytes,
3618 Register tmp1,
3619 Register tmp2,
3620 Label& slow_case,
3621 bool is_far) {
3622 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
3623 bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, tmp1, tmp2, slow_case, is_far);
3624 }
3625
3626 // get_thread() can be called anywhere inside generated code so we
3627 // need to save whatever non-callee save context might get clobbered
3628 // by the call to Thread::current() or, indeed, the call setup code.
3629 void MacroAssembler::get_thread(Register thread) {
3630 // save all call-clobbered regs except thread
3631 RegSet saved_regs = RegSet::range(x5, x7) + RegSet::range(x10, x17) +
3632 RegSet::range(x28, x31) + ra - thread;
3633 push_reg(saved_regs, sp);
3634
3635 mv(ra, CAST_FROM_FN_PTR(address, Thread::current));
3636 jalr(ra);
3637 if (thread != c_rarg0) {
3638 mv(thread, c_rarg0);
3639 }
3640
3641 // restore pushed registers
3642 pop_reg(saved_regs, sp);
3643 }
3644
3645 void MacroAssembler::load_byte_map_base(Register reg) {
3646 CardTable::CardValue* byte_map_base =
3647 ((CardTableBarrierSet*)(BarrierSet::barrier_set()))->card_table()->byte_map_base();
3648 mv(reg, (uint64_t)byte_map_base);
3649 }
3650
3651 void MacroAssembler::build_frame(int framesize) {
3652 assert(framesize >= 2, "framesize must include space for FP/RA");
3653 assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
3654 sub(sp, sp, framesize);
3655 sd(fp, Address(sp, framesize - 2 * wordSize));
3656 sd(ra, Address(sp, framesize - wordSize));
3657 if (PreserveFramePointer) { add(fp, sp, framesize); }
3658 }
3659
3660 void MacroAssembler::remove_frame(int framesize) {
3661 assert(framesize >= 2, "framesize must include space for FP/RA");
3662 assert(framesize % (2*wordSize) == 0, "must preserve 2*wordSize alignment");
3663 ld(fp, Address(sp, framesize - 2 * wordSize));
3664 ld(ra, Address(sp, framesize - wordSize));
3665 add(sp, sp, framesize);
3666 }
3667
3668 void MacroAssembler::reserved_stack_check() {
3669 // testing if reserved zone needs to be enabled
3670 Label no_reserved_zone_enabling;
3671
3672 ld(t0, Address(xthread, JavaThread::reserved_stack_activation_offset()));
3673 bltu(sp, t0, no_reserved_zone_enabling);
3674
3675 enter(); // RA and FP are live.
3676 mv(c_rarg0, xthread);
3677 rt_call(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone));
3678 leave();
3679
3680 // We have already removed our own frame.
3681 // throw_delayed_StackOverflowError will think that it's been
3682 // called by our caller.
3683 RuntimeAddress target(StubRoutines::throw_delayed_StackOverflowError_entry());
3684 relocate(target.rspec(), [&] {
3685 int32_t offset;
3686 movptr(t0, target.target(), offset);
3687 jr(t0, offset);
3688 });
3689 should_not_reach_here();
3690
3691 bind(no_reserved_zone_enabling);
3692 }
3693
3694 // Move the address of the polling page into dest.
3695 void MacroAssembler::get_polling_page(Register dest, relocInfo::relocType rtype) {
3696 ld(dest, Address(xthread, JavaThread::polling_page_offset()));
3697 }
3698
3699 // Read the polling page. The address of the polling page must
3700 // already be in r.
3701 void MacroAssembler::read_polling_page(Register r, int32_t offset, relocInfo::relocType rtype) {
3702 relocate(rtype, [&] {
3703 lwu(zr, Address(r, offset));
3704 });
3705 }
3706
3707 void MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
3708 #ifdef ASSERT
3709 {
3710 ThreadInVMfromUnknown tiv;
3711 assert (UseCompressedOops, "should only be used for compressed oops");
3712 assert (Universe::heap() != nullptr, "java heap should be initialized");
3713 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
3714 assert(Universe::heap()->is_in(JNIHandles::resolve(obj)), "should be real oop");
3715 }
3716 #endif
3717 int oop_index = oop_recorder()->find_index(obj);
3718 relocate(oop_Relocation::spec(oop_index), [&] {
3719 li32(dst, 0xDEADBEEF);
3720 });
3721 zero_extend(dst, dst, 32);
3722 }
3723
3724 void MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
3725 assert (UseCompressedClassPointers, "should only be used for compressed headers");
3726 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
3727 int index = oop_recorder()->find_index(k);
3728 assert(!Universe::heap()->is_in(k), "should not be an oop");
3729
3730 narrowKlass nk = CompressedKlassPointers::encode(k);
3731 relocate(metadata_Relocation::spec(index), [&] {
3732 li32(dst, nk);
3733 });
3734 zero_extend(dst, dst, 32);
3735 }
3736
3737 // Maybe emit a call via a trampoline. If the code cache is small
3738 // trampolines won't be emitted.
3739 address MacroAssembler::trampoline_call(Address entry) {
3740 assert(entry.rspec().type() == relocInfo::runtime_call_type ||
3741 entry.rspec().type() == relocInfo::opt_virtual_call_type ||
3742 entry.rspec().type() == relocInfo::static_call_type ||
3743 entry.rspec().type() == relocInfo::virtual_call_type, "wrong reloc type");
3744
3745 address target = entry.target();
3746
3747 // We need a trampoline if branches are far.
3748 if (!in_scratch_emit_size()) {
3749 if (entry.rspec().type() == relocInfo::runtime_call_type) {
3750 assert(CodeBuffer::supports_shared_stubs(), "must support shared stubs");
3751 code()->share_trampoline_for(entry.target(), offset());
3752 } else {
3753 address stub = emit_trampoline_stub(offset(), target);
3754 if (stub == nullptr) {
3755 postcond(pc() == badAddress);
3756 return nullptr; // CodeCache is full
3757 }
3758 }
3759 }
3760 target = pc();
3761
3762 address call_pc = pc();
3763 #ifdef ASSERT
3764 if (entry.rspec().type() != relocInfo::runtime_call_type) {
3765 assert_alignment(call_pc);
3766 }
3767 #endif
3768 relocate(entry.rspec(), [&] {
3769 jump_link(target, t0);
3770 });
3771
3772 postcond(pc() != badAddress);
3773 return call_pc;
3774 }
3775
3776 address MacroAssembler::ic_call(address entry, jint method_index) {
3777 RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
3778 IncompressibleRegion ir(this); // relocations
3779 movptr(t1, (address)Universe::non_oop_word(), t0);
3780 assert_cond(entry != nullptr);
3781 return trampoline_call(Address(entry, rh));
3782 }
3783
3784 int MacroAssembler::ic_check_size() {
3785 // No compressed
3786 return (MacroAssembler::instruction_size * (2 /* 2 loads */ + 1 /* branch */)) +
3787 far_branch_size();
3788 }
3789
3790 int MacroAssembler::ic_check(int end_alignment) {
3791 IncompressibleRegion ir(this);
3792 Register receiver = j_rarg0;
3793 Register data = t1;
3794
3795 Register tmp1 = t0; // t0 always scratch
3796 // t2 is saved on call, thus should have been saved before this check.
3797 // Hence we can clobber it.
3798 Register tmp2 = t2;
3799
3800 // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
3801 // before the inline cache check, so we don't have to execute any nop instructions when dispatching
3802 // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
3803 // before the inline cache check here, and not after
3804 align(end_alignment, ic_check_size());
3805 int uep_offset = offset();
3806
3807 if (UseCompressedClassPointers) {
3808 lwu(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
3809 lwu(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
3810 } else {
3811 ld(tmp1, Address(receiver, oopDesc::klass_offset_in_bytes()));
3812 ld(tmp2, Address(data, CompiledICData::speculated_klass_offset()));
3813 }
3814
3815 Label ic_hit;
3816 beq(tmp1, tmp2, ic_hit);
3817 // Note, far_jump is not fixed size.
3818 // Is this ever generates a movptr alignment/size will be off.
3819 far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
3820 bind(ic_hit);
3821
3822 assert((offset() % end_alignment) == 0, "Misaligned verified entry point.");
3823 return uep_offset;
3824 }
3825
3826 // Emit a trampoline stub for a call to a target which is too far away.
3827 //
3828 // code sequences:
3829 //
3830 // call-site:
3831 // branch-and-link to <destination> or <trampoline stub>
3832 //
3833 // Related trampoline stub for this call site in the stub section:
3834 // load the call target from the constant pool
3835 // branch (RA still points to the call site above)
3836
3837 address MacroAssembler::emit_trampoline_stub(int insts_call_instruction_offset,
3838 address dest) {
3839 // Max stub size: alignment nop, TrampolineStub.
3840 address stub = start_a_stub(max_trampoline_stub_size());
3841 if (stub == nullptr) {
3842 return nullptr; // CodeBuffer::expand failed
3843 }
3844
3845 // We are always 4-byte aligned here.
3846 assert_alignment(pc());
3847
3848 // Create a trampoline stub relocation which relates this trampoline stub
3849 // with the call instruction at insts_call_instruction_offset in the
3850 // instructions code-section.
3851
3852 // Make sure the address of destination 8-byte aligned after 3 instructions.
3853 align(wordSize, MacroAssembler::trampoline_stub_data_offset);
3854
3855 RelocationHolder rh = trampoline_stub_Relocation::spec(code()->insts()->start() +
3856 insts_call_instruction_offset);
3857 const int stub_start_offset = offset();
3858 relocate(rh, [&] {
3859 // Now, create the trampoline stub's code:
3860 // - load the call
3861 // - call
3862 Label target;
3863 ld(t0, target); // auipc + ld
3864 jr(t0); // jalr
3865 bind(target);
3866 assert(offset() - stub_start_offset == MacroAssembler::trampoline_stub_data_offset,
3867 "should be");
3868 assert(offset() % wordSize == 0, "bad alignment");
3869 emit_int64((int64_t)dest);
3870 });
3871
3872 const address stub_start_addr = addr_at(stub_start_offset);
3873
3874 assert(MacroAssembler::is_trampoline_stub_at(stub_start_addr), "doesn't look like a trampoline");
3875
3876 end_a_stub();
3877 return stub_start_addr;
3878 }
3879
3880 int MacroAssembler::max_trampoline_stub_size() {
3881 // Max stub size: alignment nop, TrampolineStub.
3882 return MacroAssembler::instruction_size + MacroAssembler::trampoline_stub_instruction_size;
3883 }
3884
3885 int MacroAssembler::static_call_stub_size() {
3886 // (lui, addi, slli, addi, slli, addi) + (lui + lui + slli + add) + jalr
3887 return 11 * MacroAssembler::instruction_size;
3888 }
3889
3890 Address MacroAssembler::add_memory_helper(const Address dst, Register tmp) {
3891 switch (dst.getMode()) {
3892 case Address::base_plus_offset:
3893 // This is the expected mode, although we allow all the other
3894 // forms below.
3895 return form_address(tmp, dst.base(), dst.offset());
3896 default:
3897 la(tmp, dst);
3898 return Address(tmp);
3899 }
3900 }
3901
3902 void MacroAssembler::increment(const Address dst, int64_t value, Register tmp1, Register tmp2) {
3903 assert(((dst.getMode() == Address::base_plus_offset &&
3904 is_simm12(dst.offset())) || is_simm12(value)),
3905 "invalid value and address mode combination");
3906 Address adr = add_memory_helper(dst, tmp2);
3907 assert(!adr.uses(tmp1), "invalid dst for address increment");
3908 ld(tmp1, adr);
3909 add(tmp1, tmp1, value, tmp2);
3910 sd(tmp1, adr);
3911 }
3912
3913 void MacroAssembler::incrementw(const Address dst, int32_t value, Register tmp1, Register tmp2) {
3914 assert(((dst.getMode() == Address::base_plus_offset &&
3915 is_simm12(dst.offset())) || is_simm12(value)),
3916 "invalid value and address mode combination");
3917 Address adr = add_memory_helper(dst, tmp2);
3918 assert(!adr.uses(tmp1), "invalid dst for address increment");
3919 lwu(tmp1, adr);
3920 addw(tmp1, tmp1, value, tmp2);
3921 sw(tmp1, adr);
3922 }
3923
3924 void MacroAssembler::decrement(const Address dst, int64_t value, Register tmp1, Register tmp2) {
3925 assert(((dst.getMode() == Address::base_plus_offset &&
3926 is_simm12(dst.offset())) || is_simm12(value)),
3927 "invalid value and address mode combination");
3928 Address adr = add_memory_helper(dst, tmp2);
3929 assert(!adr.uses(tmp1), "invalid dst for address decrement");
3930 ld(tmp1, adr);
3931 sub(tmp1, tmp1, value, tmp2);
3932 sd(tmp1, adr);
3933 }
3934
3935 void MacroAssembler::decrementw(const Address dst, int32_t value, Register tmp1, Register tmp2) {
3936 assert(((dst.getMode() == Address::base_plus_offset &&
3937 is_simm12(dst.offset())) || is_simm12(value)),
3938 "invalid value and address mode combination");
3939 Address adr = add_memory_helper(dst, tmp2);
3940 assert(!adr.uses(tmp1), "invalid dst for address decrement");
3941 lwu(tmp1, adr);
3942 subw(tmp1, tmp1, value, tmp2);
3943 sw(tmp1, adr);
3944 }
3945
3946 void MacroAssembler::cmpptr(Register src1, Address src2, Label& equal) {
3947 assert_different_registers(src1, t0);
3948 relocate(src2.rspec(), [&] {
3949 int32_t offset;
3950 la(t0, src2.target(), offset);
3951 ld(t0, Address(t0, offset));
3952 });
3953 beq(src1, t0, equal);
3954 }
3955
3956 void MacroAssembler::load_method_holder_cld(Register result, Register method) {
3957 load_method_holder(result, method);
3958 ld(result, Address(result, InstanceKlass::class_loader_data_offset()));
3959 }
3960
3961 void MacroAssembler::load_method_holder(Register holder, Register method) {
3962 ld(holder, Address(method, Method::const_offset())); // ConstMethod*
3963 ld(holder, Address(holder, ConstMethod::constants_offset())); // ConstantPool*
3964 ld(holder, Address(holder, ConstantPool::pool_holder_offset())); // InstanceKlass*
3965 }
3966
3967 // string indexof
3968 // compute index by trailing zeros
3969 void MacroAssembler::compute_index(Register haystack, Register trailing_zeros,
3970 Register match_mask, Register result,
3971 Register ch2, Register tmp,
3972 bool haystack_isL) {
3973 int haystack_chr_shift = haystack_isL ? 0 : 1;
3974 srl(match_mask, match_mask, trailing_zeros);
3975 srli(match_mask, match_mask, 1);
3976 srli(tmp, trailing_zeros, LogBitsPerByte);
3977 if (!haystack_isL) andi(tmp, tmp, 0xE);
3978 add(haystack, haystack, tmp);
3979 ld(ch2, Address(haystack));
3980 if (!haystack_isL) srli(tmp, tmp, haystack_chr_shift);
3981 add(result, result, tmp);
3982 }
3983
3984 // string indexof
3985 // Find pattern element in src, compute match mask,
3986 // only the first occurrence of 0x80/0x8000 at low bits is the valid match index
3987 // match mask patterns and corresponding indices would be like:
3988 // - 0x8080808080808080 (Latin1)
3989 // - 7 6 5 4 3 2 1 0 (match index)
3990 // - 0x8000800080008000 (UTF16)
3991 // - 3 2 1 0 (match index)
3992 void MacroAssembler::compute_match_mask(Register src, Register pattern, Register match_mask,
3993 Register mask1, Register mask2) {
3994 xorr(src, pattern, src);
3995 sub(match_mask, src, mask1);
3996 orr(src, src, mask2);
3997 notr(src, src);
3998 andr(match_mask, match_mask, src);
3999 }
4000
4001 #ifdef COMPILER2
4002 // Code for BigInteger::mulAdd intrinsic
4003 // out = x10
4004 // in = x11
4005 // offset = x12 (already out.length-offset)
4006 // len = x13
4007 // k = x14
4008 // tmp = x28
4009 //
4010 // pseudo code from java implementation:
4011 // long kLong = k & LONG_MASK;
4012 // carry = 0;
4013 // offset = out.length-offset - 1;
4014 // for (int j = len - 1; j >= 0; j--) {
4015 // product = (in[j] & LONG_MASK) * kLong + (out[offset] & LONG_MASK) + carry;
4016 // out[offset--] = (int)product;
4017 // carry = product >>> 32;
4018 // }
4019 // return (int)carry;
4020 void MacroAssembler::mul_add(Register out, Register in, Register offset,
4021 Register len, Register k, Register tmp) {
4022 Label L_tail_loop, L_unroll, L_end;
4023 mv(tmp, out);
4024 mv(out, zr);
4025 blez(len, L_end);
4026 zero_extend(k, k, 32);
4027 slliw(t0, offset, LogBytesPerInt);
4028 add(offset, tmp, t0);
4029 slliw(t0, len, LogBytesPerInt);
4030 add(in, in, t0);
4031
4032 const int unroll = 8;
4033 mv(tmp, unroll);
4034 blt(len, tmp, L_tail_loop);
4035 bind(L_unroll);
4036 for (int i = 0; i < unroll; i++) {
4037 sub(in, in, BytesPerInt);
4038 lwu(t0, Address(in, 0));
4039 mul(t1, t0, k);
4040 add(t0, t1, out);
4041 sub(offset, offset, BytesPerInt);
4042 lwu(t1, Address(offset, 0));
4043 add(t0, t0, t1);
4044 sw(t0, Address(offset, 0));
4045 srli(out, t0, 32);
4046 }
4047 subw(len, len, tmp);
4048 bge(len, tmp, L_unroll);
4049
4050 bind(L_tail_loop);
4051 blez(len, L_end);
4052 sub(in, in, BytesPerInt);
4053 lwu(t0, Address(in, 0));
4054 mul(t1, t0, k);
4055 add(t0, t1, out);
4056 sub(offset, offset, BytesPerInt);
4057 lwu(t1, Address(offset, 0));
4058 add(t0, t0, t1);
4059 sw(t0, Address(offset, 0));
4060 srli(out, t0, 32);
4061 subw(len, len, 1);
4062 j(L_tail_loop);
4063
4064 bind(L_end);
4065 }
4066
4067 // Multiply and multiply-accumulate unsigned 64-bit registers.
4068 void MacroAssembler::wide_mul(Register prod_lo, Register prod_hi, Register n, Register m) {
4069 assert_different_registers(prod_lo, prod_hi);
4070
4071 mul(prod_lo, n, m);
4072 mulhu(prod_hi, n, m);
4073 }
4074
4075 void MacroAssembler::wide_madd(Register sum_lo, Register sum_hi, Register n,
4076 Register m, Register tmp1, Register tmp2) {
4077 assert_different_registers(sum_lo, sum_hi);
4078 assert_different_registers(sum_hi, tmp2);
4079
4080 wide_mul(tmp1, tmp2, n, m);
4081 cad(sum_lo, sum_lo, tmp1, tmp1); // Add tmp1 to sum_lo with carry output to tmp1
4082 adc(sum_hi, sum_hi, tmp2, tmp1); // Add tmp2 with carry to sum_hi
4083 }
4084
4085 // add two unsigned input and output carry
4086 void MacroAssembler::cad(Register dst, Register src1, Register src2, Register carry)
4087 {
4088 assert_different_registers(dst, carry);
4089 assert_different_registers(dst, src2);
4090 add(dst, src1, src2);
4091 sltu(carry, dst, src2);
4092 }
4093
4094 // add two input with carry
4095 void MacroAssembler::adc(Register dst, Register src1, Register src2, Register carry) {
4096 assert_different_registers(dst, carry);
4097 add(dst, src1, src2);
4098 add(dst, dst, carry);
4099 }
4100
4101 // add two unsigned input with carry and output carry
4102 void MacroAssembler::cadc(Register dst, Register src1, Register src2, Register carry) {
4103 assert_different_registers(dst, src2);
4104 adc(dst, src1, src2, carry);
4105 sltu(carry, dst, src2);
4106 }
4107
4108 void MacroAssembler::add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
4109 Register src1, Register src2, Register carry) {
4110 cad(dest_lo, dest_lo, src1, carry);
4111 add(dest_hi, dest_hi, carry);
4112 cad(dest_lo, dest_lo, src2, carry);
4113 add(final_dest_hi, dest_hi, carry);
4114 }
4115
4116 /**
4117 * Multiply 32 bit by 32 bit first loop.
4118 */
4119 void MacroAssembler::multiply_32_x_32_loop(Register x, Register xstart, Register x_xstart,
4120 Register y, Register y_idx, Register z,
4121 Register carry, Register product,
4122 Register idx, Register kdx) {
4123 // jlong carry, x[], y[], z[];
4124 // for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx--, kdx--) {
4125 // long product = y[idx] * x[xstart] + carry;
4126 // z[kdx] = (int)product;
4127 // carry = product >>> 32;
4128 // }
4129 // z[xstart] = (int)carry;
4130
4131 Label L_first_loop, L_first_loop_exit;
4132 blez(idx, L_first_loop_exit);
4133
4134 shadd(t0, xstart, x, t0, LogBytesPerInt);
4135 lwu(x_xstart, Address(t0, 0));
4136
4137 bind(L_first_loop);
4138 subw(idx, idx, 1);
4139 shadd(t0, idx, y, t0, LogBytesPerInt);
4140 lwu(y_idx, Address(t0, 0));
4141 mul(product, x_xstart, y_idx);
4142 add(product, product, carry);
4143 srli(carry, product, 32);
4144 subw(kdx, kdx, 1);
4145 shadd(t0, kdx, z, t0, LogBytesPerInt);
4146 sw(product, Address(t0, 0));
4147 bgtz(idx, L_first_loop);
4148
4149 bind(L_first_loop_exit);
4150 }
4151
4152 /**
4153 * Multiply 64 bit by 64 bit first loop.
4154 */
4155 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
4156 Register y, Register y_idx, Register z,
4157 Register carry, Register product,
4158 Register idx, Register kdx) {
4159 //
4160 // jlong carry, x[], y[], z[];
4161 // for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx--, kdx--) {
4162 // huge_128 product = y[idx] * x[xstart] + carry;
4163 // z[kdx] = (jlong)product;
4164 // carry = (jlong)(product >>> 64);
4165 // }
4166 // z[xstart] = carry;
4167 //
4168
4169 Label L_first_loop, L_first_loop_exit;
4170 Label L_one_x, L_one_y, L_multiply;
4171
4172 subw(xstart, xstart, 1);
4173 bltz(xstart, L_one_x);
4174
4175 shadd(t0, xstart, x, t0, LogBytesPerInt);
4176 ld(x_xstart, Address(t0, 0));
4177 ror_imm(x_xstart, x_xstart, 32); // convert big-endian to little-endian
4178
4179 bind(L_first_loop);
4180 subw(idx, idx, 1);
4181 bltz(idx, L_first_loop_exit);
4182 subw(idx, idx, 1);
4183 bltz(idx, L_one_y);
4184
4185 shadd(t0, idx, y, t0, LogBytesPerInt);
4186 ld(y_idx, Address(t0, 0));
4187 ror_imm(y_idx, y_idx, 32); // convert big-endian to little-endian
4188 bind(L_multiply);
4189
4190 mulhu(t0, x_xstart, y_idx);
4191 mul(product, x_xstart, y_idx);
4192 cad(product, product, carry, t1);
4193 adc(carry, t0, zr, t1);
4194
4195 subw(kdx, kdx, 2);
4196 ror_imm(product, product, 32); // back to big-endian
4197 shadd(t0, kdx, z, t0, LogBytesPerInt);
4198 sd(product, Address(t0, 0));
4199
4200 j(L_first_loop);
4201
4202 bind(L_one_y);
4203 lwu(y_idx, Address(y, 0));
4204 j(L_multiply);
4205
4206 bind(L_one_x);
4207 lwu(x_xstart, Address(x, 0));
4208 j(L_first_loop);
4209
4210 bind(L_first_loop_exit);
4211 }
4212
4213 /**
4214 * Multiply 128 bit by 128 bit. Unrolled inner loop.
4215 *
4216 */
4217 void MacroAssembler::multiply_128_x_128_loop(Register y, Register z,
4218 Register carry, Register carry2,
4219 Register idx, Register jdx,
4220 Register yz_idx1, Register yz_idx2,
4221 Register tmp, Register tmp3, Register tmp4,
4222 Register tmp6, Register product_hi) {
4223 // jlong carry, x[], y[], z[];
4224 // int kdx = xstart+1;
4225 // for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
4226 // huge_128 tmp3 = (y[idx+1] * product_hi) + z[kdx+idx+1] + carry;
4227 // jlong carry2 = (jlong)(tmp3 >>> 64);
4228 // huge_128 tmp4 = (y[idx] * product_hi) + z[kdx+idx] + carry2;
4229 // carry = (jlong)(tmp4 >>> 64);
4230 // z[kdx+idx+1] = (jlong)tmp3;
4231 // z[kdx+idx] = (jlong)tmp4;
4232 // }
4233 // idx += 2;
4234 // if (idx > 0) {
4235 // yz_idx1 = (y[idx] * product_hi) + z[kdx+idx] + carry;
4236 // z[kdx+idx] = (jlong)yz_idx1;
4237 // carry = (jlong)(yz_idx1 >>> 64);
4238 // }
4239 //
4240
4241 Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
4242
4243 srliw(jdx, idx, 2);
4244
4245 bind(L_third_loop);
4246
4247 subw(jdx, jdx, 1);
4248 bltz(jdx, L_third_loop_exit);
4249 subw(idx, idx, 4);
4250
4251 shadd(t0, idx, y, t0, LogBytesPerInt);
4252 ld(yz_idx2, Address(t0, 0));
4253 ld(yz_idx1, Address(t0, wordSize));
4254
4255 shadd(tmp6, idx, z, t0, LogBytesPerInt);
4256
4257 ror_imm(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
4258 ror_imm(yz_idx2, yz_idx2, 32);
4259
4260 ld(t1, Address(tmp6, 0));
4261 ld(t0, Address(tmp6, wordSize));
4262
4263 mul(tmp3, product_hi, yz_idx1); // yz_idx1 * product_hi -> tmp4:tmp3
4264 mulhu(tmp4, product_hi, yz_idx1);
4265
4266 ror_imm(t0, t0, 32, tmp); // convert big-endian to little-endian
4267 ror_imm(t1, t1, 32, tmp);
4268
4269 mul(tmp, product_hi, yz_idx2); // yz_idx2 * product_hi -> carry2:tmp
4270 mulhu(carry2, product_hi, yz_idx2);
4271
4272 cad(tmp3, tmp3, carry, carry);
4273 adc(tmp4, tmp4, zr, carry);
4274 cad(tmp3, tmp3, t0, t0);
4275 cadc(tmp4, tmp4, tmp, t0);
4276 adc(carry, carry2, zr, t0);
4277 cad(tmp4, tmp4, t1, carry2);
4278 adc(carry, carry, zr, carry2);
4279
4280 ror_imm(tmp3, tmp3, 32); // convert little-endian to big-endian
4281 ror_imm(tmp4, tmp4, 32);
4282 sd(tmp4, Address(tmp6, 0));
4283 sd(tmp3, Address(tmp6, wordSize));
4284
4285 j(L_third_loop);
4286
4287 bind(L_third_loop_exit);
4288
4289 andi(idx, idx, 0x3);
4290 beqz(idx, L_post_third_loop_done);
4291
4292 Label L_check_1;
4293 subw(idx, idx, 2);
4294 bltz(idx, L_check_1);
4295
4296 shadd(t0, idx, y, t0, LogBytesPerInt);
4297 ld(yz_idx1, Address(t0, 0));
4298 ror_imm(yz_idx1, yz_idx1, 32);
4299
4300 mul(tmp3, product_hi, yz_idx1); // yz_idx1 * product_hi -> tmp4:tmp3
4301 mulhu(tmp4, product_hi, yz_idx1);
4302
4303 shadd(t0, idx, z, t0, LogBytesPerInt);
4304 ld(yz_idx2, Address(t0, 0));
4305 ror_imm(yz_idx2, yz_idx2, 32, tmp);
4306
4307 add2_with_carry(carry, tmp4, tmp3, carry, yz_idx2, tmp);
4308
4309 ror_imm(tmp3, tmp3, 32, tmp);
4310 sd(tmp3, Address(t0, 0));
4311
4312 bind(L_check_1);
4313
4314 andi(idx, idx, 0x1);
4315 subw(idx, idx, 1);
4316 bltz(idx, L_post_third_loop_done);
4317 shadd(t0, idx, y, t0, LogBytesPerInt);
4318 lwu(tmp4, Address(t0, 0));
4319 mul(tmp3, tmp4, product_hi); // tmp4 * product_hi -> carry2:tmp3
4320 mulhu(carry2, tmp4, product_hi);
4321
4322 shadd(t0, idx, z, t0, LogBytesPerInt);
4323 lwu(tmp4, Address(t0, 0));
4324
4325 add2_with_carry(carry2, carry2, tmp3, tmp4, carry, t0);
4326
4327 shadd(t0, idx, z, t0, LogBytesPerInt);
4328 sw(tmp3, Address(t0, 0));
4329
4330 slli(t0, carry2, 32);
4331 srli(carry, tmp3, 32);
4332 orr(carry, carry, t0);
4333
4334 bind(L_post_third_loop_done);
4335 }
4336
4337 /**
4338 * Code for BigInteger::multiplyToLen() intrinsic.
4339 *
4340 * x10: x
4341 * x11: xlen
4342 * x12: y
4343 * x13: ylen
4344 * x14: z
4345 * x15: tmp0
4346 * x16: tmp1
4347 * x17: tmp2
4348 * x7: tmp3
4349 * x28: tmp4
4350 * x29: tmp5
4351 * x30: tmp6
4352 * x31: tmp7
4353 */
4354 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen,
4355 Register z, Register tmp0,
4356 Register tmp1, Register tmp2, Register tmp3, Register tmp4,
4357 Register tmp5, Register tmp6, Register product_hi) {
4358 assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6);
4359
4360 const Register idx = tmp1;
4361 const Register kdx = tmp2;
4362 const Register xstart = tmp3;
4363
4364 const Register y_idx = tmp4;
4365 const Register carry = tmp5;
4366 const Register product = xlen;
4367 const Register x_xstart = tmp0;
4368
4369 mv(idx, ylen); // idx = ylen;
4370 addw(kdx, xlen, ylen); // kdx = xlen+ylen;
4371 mv(carry, zr); // carry = 0;
4372
4373 Label L_multiply_64_x_64_loop, L_done;
4374
4375 subw(xstart, xlen, 1);
4376 bltz(xstart, L_done);
4377
4378 const Register jdx = tmp1;
4379
4380 if (AvoidUnalignedAccesses) {
4381 // Check if x and y are both 8-byte aligned.
4382 orr(t0, xlen, ylen);
4383 test_bit(t0, t0, 0);
4384 beqz(t0, L_multiply_64_x_64_loop);
4385
4386 multiply_32_x_32_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
4387 shadd(t0, xstart, z, t0, LogBytesPerInt);
4388 sw(carry, Address(t0, 0));
4389
4390 Label L_second_loop_unaligned;
4391 bind(L_second_loop_unaligned);
4392 mv(carry, zr);
4393 mv(jdx, ylen);
4394 subw(xstart, xstart, 1);
4395 bltz(xstart, L_done);
4396 sub(sp, sp, 2 * wordSize);
4397 sd(z, Address(sp, 0));
4398 sd(zr, Address(sp, wordSize));
4399 shadd(t0, xstart, z, t0, LogBytesPerInt);
4400 addi(z, t0, 4);
4401 shadd(t0, xstart, x, t0, LogBytesPerInt);
4402 lwu(product, Address(t0, 0));
4403 Label L_third_loop, L_third_loop_exit;
4404
4405 blez(jdx, L_third_loop_exit);
4406
4407 bind(L_third_loop);
4408 subw(jdx, jdx, 1);
4409 shadd(t0, jdx, y, t0, LogBytesPerInt);
4410 lwu(t0, Address(t0, 0));
4411 mul(t1, t0, product);
4412 add(t0, t1, carry);
4413 shadd(tmp6, jdx, z, t1, LogBytesPerInt);
4414 lwu(t1, Address(tmp6, 0));
4415 add(t0, t0, t1);
4416 sw(t0, Address(tmp6, 0));
4417 srli(carry, t0, 32);
4418 bgtz(jdx, L_third_loop);
4419
4420 bind(L_third_loop_exit);
4421 ld(z, Address(sp, 0));
4422 addi(sp, sp, 2 * wordSize);
4423 shadd(t0, xstart, z, t0, LogBytesPerInt);
4424 sw(carry, Address(t0, 0));
4425
4426 j(L_second_loop_unaligned);
4427 }
4428
4429 bind(L_multiply_64_x_64_loop);
4430 multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
4431
4432 Label L_second_loop_aligned;
4433 beqz(kdx, L_second_loop_aligned);
4434
4435 Label L_carry;
4436 subw(kdx, kdx, 1);
4437 beqz(kdx, L_carry);
4438
4439 shadd(t0, kdx, z, t0, LogBytesPerInt);
4440 sw(carry, Address(t0, 0));
4441 srli(carry, carry, 32);
4442 subw(kdx, kdx, 1);
4443
4444 bind(L_carry);
4445 shadd(t0, kdx, z, t0, LogBytesPerInt);
4446 sw(carry, Address(t0, 0));
4447
4448 // Second and third (nested) loops.
4449 //
4450 // for (int i = xstart-1; i >= 0; i--) { // Second loop
4451 // carry = 0;
4452 // for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
4453 // long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
4454 // (z[k] & LONG_MASK) + carry;
4455 // z[k] = (int)product;
4456 // carry = product >>> 32;
4457 // }
4458 // z[i] = (int)carry;
4459 // }
4460 //
4461 // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = product_hi
4462
4463 bind(L_second_loop_aligned);
4464 mv(carry, zr); // carry = 0;
4465 mv(jdx, ylen); // j = ystart+1
4466
4467 subw(xstart, xstart, 1); // i = xstart-1;
4468 bltz(xstart, L_done);
4469
4470 sub(sp, sp, 4 * wordSize);
4471 sd(z, Address(sp, 0));
4472
4473 Label L_last_x;
4474 shadd(t0, xstart, z, t0, LogBytesPerInt);
4475 addi(z, t0, 4);
4476 subw(xstart, xstart, 1); // i = xstart-1;
4477 bltz(xstart, L_last_x);
4478
4479 shadd(t0, xstart, x, t0, LogBytesPerInt);
4480 ld(product_hi, Address(t0, 0));
4481 ror_imm(product_hi, product_hi, 32); // convert big-endian to little-endian
4482
4483 Label L_third_loop_prologue;
4484 bind(L_third_loop_prologue);
4485
4486 sd(ylen, Address(sp, wordSize));
4487 sd(x, Address(sp, 2 * wordSize));
4488 sd(xstart, Address(sp, 3 * wordSize));
4489 multiply_128_x_128_loop(y, z, carry, x, jdx, ylen, product,
4490 tmp2, x_xstart, tmp3, tmp4, tmp6, product_hi);
4491 ld(z, Address(sp, 0));
4492 ld(ylen, Address(sp, wordSize));
4493 ld(x, Address(sp, 2 * wordSize));
4494 ld(xlen, Address(sp, 3 * wordSize)); // copy old xstart -> xlen
4495 addi(sp, sp, 4 * wordSize);
4496
4497 addiw(tmp3, xlen, 1);
4498 shadd(t0, tmp3, z, t0, LogBytesPerInt);
4499 sw(carry, Address(t0, 0));
4500
4501 subw(tmp3, tmp3, 1);
4502 bltz(tmp3, L_done);
4503
4504 srli(carry, carry, 32);
4505 shadd(t0, tmp3, z, t0, LogBytesPerInt);
4506 sw(carry, Address(t0, 0));
4507 j(L_second_loop_aligned);
4508
4509 // Next infrequent code is moved outside loops.
4510 bind(L_last_x);
4511 lwu(product_hi, Address(x, 0));
4512 j(L_third_loop_prologue);
4513
4514 bind(L_done);
4515 }
4516 #endif
4517
4518 // Count bits of trailing zero chars from lsb to msb until first non-zero element.
4519 // For LL case, one byte for one element, so shift 8 bits once, and for other case,
4520 // shift 16 bits once.
4521 void MacroAssembler::ctzc_bit(Register Rd, Register Rs, bool isLL, Register tmp1, Register tmp2) {
4522 if (UseZbb) {
4523 assert_different_registers(Rd, Rs, tmp1);
4524 int step = isLL ? 8 : 16;
4525 ctz(Rd, Rs);
4526 andi(tmp1, Rd, step - 1);
4527 sub(Rd, Rd, tmp1);
4528 return;
4529 }
4530
4531 assert_different_registers(Rd, Rs, tmp1, tmp2);
4532 Label Loop;
4533 int step = isLL ? 8 : 16;
4534 mv(Rd, -step);
4535 mv(tmp2, Rs);
4536
4537 bind(Loop);
4538 addi(Rd, Rd, step);
4539 andi(tmp1, tmp2, ((1 << step) - 1));
4540 srli(tmp2, tmp2, step);
4541 beqz(tmp1, Loop);
4542 }
4543
4544 // This instruction reads adjacent 4 bytes from the lower half of source register,
4545 // inflate into a register, for example:
4546 // Rs: A7A6A5A4A3A2A1A0
4547 // Rd: 00A300A200A100A0
4548 void MacroAssembler::inflate_lo32(Register Rd, Register Rs, Register tmp1, Register tmp2) {
4549 assert_different_registers(Rd, Rs, tmp1, tmp2);
4550
4551 mv(tmp1, 0xFF000000); // first byte mask at lower word
4552 andr(Rd, Rs, tmp1);
4553 for (int i = 0; i < 2; i++) {
4554 slli(Rd, Rd, wordSize);
4555 srli(tmp1, tmp1, wordSize);
4556 andr(tmp2, Rs, tmp1);
4557 orr(Rd, Rd, tmp2);
4558 }
4559 slli(Rd, Rd, wordSize);
4560 andi(tmp2, Rs, 0xFF); // last byte mask at lower word
4561 orr(Rd, Rd, tmp2);
4562 }
4563
4564 // This instruction reads adjacent 4 bytes from the upper half of source register,
4565 // inflate into a register, for example:
4566 // Rs: A7A6A5A4A3A2A1A0
4567 // Rd: 00A700A600A500A4
4568 void MacroAssembler::inflate_hi32(Register Rd, Register Rs, Register tmp1, Register tmp2) {
4569 assert_different_registers(Rd, Rs, tmp1, tmp2);
4570 srli(Rs, Rs, 32); // only upper 32 bits are needed
4571 inflate_lo32(Rd, Rs, tmp1, tmp2);
4572 }
4573
4574 // The size of the blocks erased by the zero_blocks stub. We must
4575 // handle anything smaller than this ourselves in zero_words().
4576 const int MacroAssembler::zero_words_block_size = 8;
4577
4578 // zero_words() is used by C2 ClearArray patterns. It is as small as
4579 // possible, handling small word counts locally and delegating
4580 // anything larger to the zero_blocks stub. It is expanded many times
4581 // in compiled code, so it is important to keep it short.
4582
4583 // ptr: Address of a buffer to be zeroed.
4584 // cnt: Count in HeapWords.
4585 //
4586 // ptr, cnt, and t0 are clobbered.
4587 address MacroAssembler::zero_words(Register ptr, Register cnt) {
4588 assert(is_power_of_2(zero_words_block_size), "adjust this");
4589 assert(ptr == x28 && cnt == x29, "mismatch in register usage");
4590 assert_different_registers(cnt, t0);
4591
4592 BLOCK_COMMENT("zero_words {");
4593
4594 mv(t0, zero_words_block_size);
4595 Label around, done, done16;
4596 bltu(cnt, t0, around);
4597 {
4598 RuntimeAddress zero_blocks(StubRoutines::riscv::zero_blocks());
4599 assert(zero_blocks.target() != nullptr, "zero_blocks stub has not been generated");
4600 if (StubRoutines::riscv::complete()) {
4601 address tpc = trampoline_call(zero_blocks);
4602 if (tpc == nullptr) {
4603 DEBUG_ONLY(reset_labels(around));
4604 postcond(pc() == badAddress);
4605 return nullptr;
4606 }
4607 } else {
4608 jump_link(zero_blocks, t0);
4609 }
4610 }
4611 bind(around);
4612 for (int i = zero_words_block_size >> 1; i > 1; i >>= 1) {
4613 Label l;
4614 test_bit(t0, cnt, exact_log2(i));
4615 beqz(t0, l);
4616 for (int j = 0; j < i; j++) {
4617 sd(zr, Address(ptr, j * wordSize));
4618 }
4619 addi(ptr, ptr, i * wordSize);
4620 bind(l);
4621 }
4622 {
4623 Label l;
4624 test_bit(t0, cnt, 0);
4625 beqz(t0, l);
4626 sd(zr, Address(ptr, 0));
4627 bind(l);
4628 }
4629
4630 BLOCK_COMMENT("} zero_words");
4631 postcond(pc() != badAddress);
4632 return pc();
4633 }
4634
4635 #define SmallArraySize (18 * BytesPerLong)
4636
4637 // base: Address of a buffer to be zeroed, 8 bytes aligned.
4638 // cnt: Immediate count in HeapWords.
4639 void MacroAssembler::zero_words(Register base, uint64_t cnt) {
4640 assert_different_registers(base, t0, t1);
4641
4642 BLOCK_COMMENT("zero_words {");
4643
4644 if (cnt <= SmallArraySize / BytesPerLong) {
4645 for (int i = 0; i < (int)cnt; i++) {
4646 sd(zr, Address(base, i * wordSize));
4647 }
4648 } else {
4649 const int unroll = 8; // Number of sd(zr, adr), instructions we'll unroll
4650 int remainder = cnt % unroll;
4651 for (int i = 0; i < remainder; i++) {
4652 sd(zr, Address(base, i * wordSize));
4653 }
4654
4655 Label loop;
4656 Register cnt_reg = t0;
4657 Register loop_base = t1;
4658 cnt = cnt - remainder;
4659 mv(cnt_reg, cnt);
4660 add(loop_base, base, remainder * wordSize);
4661 bind(loop);
4662 sub(cnt_reg, cnt_reg, unroll);
4663 for (int i = 0; i < unroll; i++) {
4664 sd(zr, Address(loop_base, i * wordSize));
4665 }
4666 add(loop_base, loop_base, unroll * wordSize);
4667 bnez(cnt_reg, loop);
4668 }
4669
4670 BLOCK_COMMENT("} zero_words");
4671 }
4672
4673 // base: Address of a buffer to be filled, 8 bytes aligned.
4674 // cnt: Count in 8-byte unit.
4675 // value: Value to be filled with.
4676 // base will point to the end of the buffer after filling.
4677 void MacroAssembler::fill_words(Register base, Register cnt, Register value) {
4678 // Algorithm:
4679 //
4680 // t0 = cnt & 7
4681 // cnt -= t0
4682 // p += t0
4683 // switch (t0):
4684 // switch start:
4685 // do while cnt
4686 // cnt -= 8
4687 // p[-8] = value
4688 // case 7:
4689 // p[-7] = value
4690 // case 6:
4691 // p[-6] = value
4692 // // ...
4693 // case 1:
4694 // p[-1] = value
4695 // case 0:
4696 // p += 8
4697 // do-while end
4698 // switch end
4699
4700 assert_different_registers(base, cnt, value, t0, t1);
4701
4702 Label fini, skip, entry, loop;
4703 const int unroll = 8; // Number of sd instructions we'll unroll
4704
4705 beqz(cnt, fini);
4706
4707 andi(t0, cnt, unroll - 1);
4708 sub(cnt, cnt, t0);
4709 // align 8, so first sd n % 8 = mod, next loop sd 8 * n.
4710 shadd(base, t0, base, t1, 3);
4711 la(t1, entry);
4712 slli(t0, t0, 2); // sd_inst_nums * 4; t0 is cnt % 8, so t1 = t1 - sd_inst_nums * 4, 4 is sizeof(inst)
4713 sub(t1, t1, t0);
4714 jr(t1);
4715
4716 bind(loop);
4717 add(base, base, unroll * 8);
4718 for (int i = -unroll; i < 0; i++) {
4719 sd(value, Address(base, i * 8));
4720 }
4721 bind(entry);
4722 sub(cnt, cnt, unroll);
4723 bgez(cnt, loop);
4724
4725 bind(fini);
4726 }
4727
4728 // Zero blocks of memory by using CBO.ZERO.
4729 //
4730 // Aligns the base address first sufficiently for CBO.ZERO, then uses
4731 // CBO.ZERO repeatedly for every full block. cnt is the size to be
4732 // zeroed in HeapWords. Returns the count of words left to be zeroed
4733 // in cnt.
4734 //
4735 // NOTE: This is intended to be used in the zero_blocks() stub. If
4736 // you want to use it elsewhere, note that cnt must be >= CacheLineSize.
4737 void MacroAssembler::zero_dcache_blocks(Register base, Register cnt, Register tmp1, Register tmp2) {
4738 Label initial_table_end, loop;
4739
4740 // Align base with cache line size.
4741 neg(tmp1, base);
4742 andi(tmp1, tmp1, CacheLineSize - 1);
4743
4744 // tmp1: the number of bytes to be filled to align the base with cache line size.
4745 add(base, base, tmp1);
4746 srai(tmp2, tmp1, 3);
4747 sub(cnt, cnt, tmp2);
4748 srli(tmp2, tmp1, 1);
4749 la(tmp1, initial_table_end);
4750 sub(tmp2, tmp1, tmp2);
4751 jr(tmp2);
4752 for (int i = -CacheLineSize + wordSize; i < 0; i += wordSize) {
4753 sd(zr, Address(base, i));
4754 }
4755 bind(initial_table_end);
4756
4757 mv(tmp1, CacheLineSize / wordSize);
4758 bind(loop);
4759 cbo_zero(base);
4760 sub(cnt, cnt, tmp1);
4761 add(base, base, CacheLineSize);
4762 bge(cnt, tmp1, loop);
4763 }
4764
4765 // java.lang.Math.round(float a)
4766 // Returns the closest int to the argument, with ties rounding to positive infinity.
4767 void MacroAssembler::java_round_float(Register dst, FloatRegister src, FloatRegister ftmp) {
4768 // this instructions calling sequence provides performance improvement on all tested devices;
4769 // don't change it without re-verification
4770 Label done;
4771 mv(t0, jint_cast(0.5f));
4772 fmv_w_x(ftmp, t0);
4773
4774 // dst = 0 if NaN
4775 feq_s(t0, src, src); // replacing fclass with feq as performance optimization
4776 mv(dst, zr);
4777 beqz(t0, done);
4778
4779 // dst = (src + 0.5f) rounded down towards negative infinity
4780 // Adding 0.5f to some floats exceeds the precision limits for a float and rounding takes place.
4781 // RDN is required for fadd_s, RNE gives incorrect results:
4782 // --------------------------------------------------------------------
4783 // fadd.s rne (src + 0.5f): src = 8388609.000000 ftmp = 8388610.000000
4784 // fcvt.w.s rdn: ftmp = 8388610.000000 dst = 8388610
4785 // --------------------------------------------------------------------
4786 // fadd.s rdn (src + 0.5f): src = 8388609.000000 ftmp = 8388609.000000
4787 // fcvt.w.s rdn: ftmp = 8388609.000000 dst = 8388609
4788 // --------------------------------------------------------------------
4789 fadd_s(ftmp, src, ftmp, RoundingMode::rdn);
4790 fcvt_w_s(dst, ftmp, RoundingMode::rdn);
4791
4792 bind(done);
4793 }
4794
4795 // java.lang.Math.round(double a)
4796 // Returns the closest long to the argument, with ties rounding to positive infinity.
4797 void MacroAssembler::java_round_double(Register dst, FloatRegister src, FloatRegister ftmp) {
4798 // this instructions calling sequence provides performance improvement on all tested devices;
4799 // don't change it without re-verification
4800 Label done;
4801 mv(t0, julong_cast(0.5));
4802 fmv_d_x(ftmp, t0);
4803
4804 // dst = 0 if NaN
4805 feq_d(t0, src, src); // replacing fclass with feq as performance optimization
4806 mv(dst, zr);
4807 beqz(t0, done);
4808
4809 // dst = (src + 0.5) rounded down towards negative infinity
4810 fadd_d(ftmp, src, ftmp, RoundingMode::rdn); // RDN is required here otherwise some inputs produce incorrect results
4811 fcvt_l_d(dst, ftmp, RoundingMode::rdn);
4812
4813 bind(done);
4814 }
4815
4816 #define FCVT_SAFE(FLOATCVT, FLOATSIG) \
4817 void MacroAssembler::FLOATCVT##_safe(Register dst, FloatRegister src, Register tmp) { \
4818 Label done; \
4819 assert_different_registers(dst, tmp); \
4820 fclass_##FLOATSIG(tmp, src); \
4821 mv(dst, zr); \
4822 /* check if src is NaN */ \
4823 andi(tmp, tmp, fclass_mask::nan); \
4824 bnez(tmp, done); \
4825 FLOATCVT(dst, src); \
4826 bind(done); \
4827 }
4828
4829 FCVT_SAFE(fcvt_w_s, s);
4830 FCVT_SAFE(fcvt_l_s, s);
4831 FCVT_SAFE(fcvt_w_d, d);
4832 FCVT_SAFE(fcvt_l_d, d);
4833
4834 #undef FCVT_SAFE
4835
4836 #define FCMP(FLOATTYPE, FLOATSIG) \
4837 void MacroAssembler::FLOATTYPE##_compare(Register result, FloatRegister Rs1, \
4838 FloatRegister Rs2, int unordered_result) { \
4839 Label Ldone; \
4840 if (unordered_result < 0) { \
4841 /* we want -1 for unordered or less than, 0 for equal and 1 for greater than. */ \
4842 /* installs 1 if gt else 0 */ \
4843 flt_##FLOATSIG(result, Rs2, Rs1); \
4844 /* Rs1 > Rs2, install 1 */ \
4845 bgtz(result, Ldone); \
4846 feq_##FLOATSIG(result, Rs1, Rs2); \
4847 addi(result, result, -1); \
4848 /* Rs1 = Rs2, install 0 */ \
4849 /* NaN or Rs1 < Rs2, install -1 */ \
4850 bind(Ldone); \
4851 } else { \
4852 /* we want -1 for less than, 0 for equal and 1 for unordered or greater than. */ \
4853 /* installs 1 if gt or unordered else 0 */ \
4854 flt_##FLOATSIG(result, Rs1, Rs2); \
4855 /* Rs1 < Rs2, install -1 */ \
4856 bgtz(result, Ldone); \
4857 feq_##FLOATSIG(result, Rs1, Rs2); \
4858 addi(result, result, -1); \
4859 /* Rs1 = Rs2, install 0 */ \
4860 /* NaN or Rs1 > Rs2, install 1 */ \
4861 bind(Ldone); \
4862 neg(result, result); \
4863 } \
4864 }
4865
4866 FCMP(float, s);
4867 FCMP(double, d);
4868
4869 #undef FCMP
4870
4871 // Zero words; len is in bytes
4872 // Destroys all registers except addr
4873 // len must be a nonzero multiple of wordSize
4874 void MacroAssembler::zero_memory(Register addr, Register len, Register tmp) {
4875 assert_different_registers(addr, len, tmp, t0, t1);
4876
4877 #ifdef ASSERT
4878 {
4879 Label L;
4880 andi(t0, len, BytesPerWord - 1);
4881 beqz(t0, L);
4882 stop("len is not a multiple of BytesPerWord");
4883 bind(L);
4884 }
4885 #endif // ASSERT
4886
4887 #ifndef PRODUCT
4888 block_comment("zero memory");
4889 #endif // PRODUCT
4890
4891 Label loop;
4892 Label entry;
4893
4894 // Algorithm:
4895 //
4896 // t0 = cnt & 7
4897 // cnt -= t0
4898 // p += t0
4899 // switch (t0) {
4900 // do {
4901 // cnt -= 8
4902 // p[-8] = 0
4903 // case 7:
4904 // p[-7] = 0
4905 // case 6:
4906 // p[-6] = 0
4907 // ...
4908 // case 1:
4909 // p[-1] = 0
4910 // case 0:
4911 // p += 8
4912 // } while (cnt)
4913 // }
4914
4915 const int unroll = 8; // Number of sd(zr) instructions we'll unroll
4916
4917 srli(len, len, LogBytesPerWord);
4918 andi(t0, len, unroll - 1); // t0 = cnt % unroll
4919 sub(len, len, t0); // cnt -= unroll
4920 // tmp always points to the end of the region we're about to zero
4921 shadd(tmp, t0, addr, t1, LogBytesPerWord);
4922 la(t1, entry);
4923 slli(t0, t0, 2);
4924 sub(t1, t1, t0);
4925 jr(t1);
4926 bind(loop);
4927 sub(len, len, unroll);
4928 for (int i = -unroll; i < 0; i++) {
4929 sd(zr, Address(tmp, i * wordSize));
4930 }
4931 bind(entry);
4932 add(tmp, tmp, unroll * wordSize);
4933 bnez(len, loop);
4934 }
4935
4936 // shift left by shamt and add
4937 // Rd = (Rs1 << shamt) + Rs2
4938 void MacroAssembler::shadd(Register Rd, Register Rs1, Register Rs2, Register tmp, int shamt) {
4939 if (UseZba) {
4940 if (shamt == 1) {
4941 sh1add(Rd, Rs1, Rs2);
4942 return;
4943 } else if (shamt == 2) {
4944 sh2add(Rd, Rs1, Rs2);
4945 return;
4946 } else if (shamt == 3) {
4947 sh3add(Rd, Rs1, Rs2);
4948 return;
4949 }
4950 }
4951
4952 if (shamt != 0) {
4953 assert_different_registers(Rs2, tmp);
4954 slli(tmp, Rs1, shamt);
4955 add(Rd, Rs2, tmp);
4956 } else {
4957 add(Rd, Rs1, Rs2);
4958 }
4959 }
4960
4961 void MacroAssembler::zero_extend(Register dst, Register src, int bits) {
4962 switch (bits) {
4963 case 32:
4964 if (UseZba) {
4965 zext_w(dst, src);
4966 return;
4967 }
4968 break;
4969 case 16:
4970 if (UseZbb) {
4971 zext_h(dst, src);
4972 return;
4973 }
4974 break;
4975 case 8:
4976 if (UseZbb) {
4977 zext_b(dst, src);
4978 return;
4979 }
4980 break;
4981 default:
4982 break;
4983 }
4984 slli(dst, src, XLEN - bits);
4985 srli(dst, dst, XLEN - bits);
4986 }
4987
4988 void MacroAssembler::sign_extend(Register dst, Register src, int bits) {
4989 switch (bits) {
4990 case 32:
4991 sext_w(dst, src);
4992 return;
4993 case 16:
4994 if (UseZbb) {
4995 sext_h(dst, src);
4996 return;
4997 }
4998 break;
4999 case 8:
5000 if (UseZbb) {
5001 sext_b(dst, src);
5002 return;
5003 }
5004 break;
5005 default:
5006 break;
5007 }
5008 slli(dst, src, XLEN - bits);
5009 srai(dst, dst, XLEN - bits);
5010 }
5011
5012 void MacroAssembler::cmp_x2i(Register dst, Register src1, Register src2,
5013 Register tmp, bool is_signed) {
5014 if (src1 == src2) {
5015 mv(dst, zr);
5016 return;
5017 }
5018 Label done;
5019 Register left = src1;
5020 Register right = src2;
5021 if (dst == src1) {
5022 assert_different_registers(dst, src2, tmp);
5023 mv(tmp, src1);
5024 left = tmp;
5025 } else if (dst == src2) {
5026 assert_different_registers(dst, src1, tmp);
5027 mv(tmp, src2);
5028 right = tmp;
5029 }
5030
5031 // installs 1 if gt else 0
5032 if (is_signed) {
5033 slt(dst, right, left);
5034 } else {
5035 sltu(dst, right, left);
5036 }
5037 bnez(dst, done);
5038 if (is_signed) {
5039 slt(dst, left, right);
5040 } else {
5041 sltu(dst, left, right);
5042 }
5043 // dst = -1 if lt; else if eq , dst = 0
5044 neg(dst, dst);
5045 bind(done);
5046 }
5047
5048 void MacroAssembler::cmp_l2i(Register dst, Register src1, Register src2, Register tmp)
5049 {
5050 cmp_x2i(dst, src1, src2, tmp);
5051 }
5052
5053 void MacroAssembler::cmp_ul2i(Register dst, Register src1, Register src2, Register tmp) {
5054 cmp_x2i(dst, src1, src2, tmp, false);
5055 }
5056
5057 void MacroAssembler::cmp_uw2i(Register dst, Register src1, Register src2, Register tmp) {
5058 cmp_x2i(dst, src1, src2, tmp, false);
5059 }
5060
5061 // The java_calling_convention describes stack locations as ideal slots on
5062 // a frame with no abi restrictions. Since we must observe abi restrictions
5063 // (like the placement of the register window) the slots must be biased by
5064 // the following value.
5065 static int reg2offset_in(VMReg r) {
5066 // Account for saved fp and ra
5067 // This should really be in_preserve_stack_slots
5068 return r->reg2stack() * VMRegImpl::stack_slot_size;
5069 }
5070
5071 static int reg2offset_out(VMReg r) {
5072 return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
5073 }
5074
5075 // On 64 bit we will store integer like items to the stack as
5076 // 64 bits items (riscv64 abi) even though java would only store
5077 // 32bits for a parameter. On 32bit it will simply be 32 bits
5078 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
5079 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp) {
5080 if (src.first()->is_stack()) {
5081 if (dst.first()->is_stack()) {
5082 // stack to stack
5083 ld(tmp, Address(fp, reg2offset_in(src.first())));
5084 sd(tmp, Address(sp, reg2offset_out(dst.first())));
5085 } else {
5086 // stack to reg
5087 lw(dst.first()->as_Register(), Address(fp, reg2offset_in(src.first())));
5088 }
5089 } else if (dst.first()->is_stack()) {
5090 // reg to stack
5091 sd(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
5092 } else {
5093 if (dst.first() != src.first()) {
5094 sign_extend(dst.first()->as_Register(), src.first()->as_Register(), 32);
5095 }
5096 }
5097 }
5098
5099 // An oop arg. Must pass a handle not the oop itself
5100 void MacroAssembler::object_move(OopMap* map,
5101 int oop_handle_offset,
5102 int framesize_in_slots,
5103 VMRegPair src,
5104 VMRegPair dst,
5105 bool is_receiver,
5106 int* receiver_offset) {
5107 assert_cond(map != nullptr && receiver_offset != nullptr);
5108
5109 // must pass a handle. First figure out the location we use as a handle
5110 Register rHandle = dst.first()->is_stack() ? t1 : dst.first()->as_Register();
5111
5112 // See if oop is null if it is we need no handle
5113
5114 if (src.first()->is_stack()) {
5115 // Oop is already on the stack as an argument
5116 int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
5117 map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
5118 if (is_receiver) {
5119 *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
5120 }
5121
5122 ld(t0, Address(fp, reg2offset_in(src.first())));
5123 la(rHandle, Address(fp, reg2offset_in(src.first())));
5124 // conditionally move a null
5125 Label notZero1;
5126 bnez(t0, notZero1);
5127 mv(rHandle, zr);
5128 bind(notZero1);
5129 } else {
5130
5131 // Oop is in a register we must store it to the space we reserve
5132 // on the stack for oop_handles and pass a handle if oop is non-null
5133
5134 const Register rOop = src.first()->as_Register();
5135 int oop_slot = -1;
5136 if (rOop == j_rarg0) {
5137 oop_slot = 0;
5138 } else if (rOop == j_rarg1) {
5139 oop_slot = 1;
5140 } else if (rOop == j_rarg2) {
5141 oop_slot = 2;
5142 } else if (rOop == j_rarg3) {
5143 oop_slot = 3;
5144 } else if (rOop == j_rarg4) {
5145 oop_slot = 4;
5146 } else if (rOop == j_rarg5) {
5147 oop_slot = 5;
5148 } else if (rOop == j_rarg6) {
5149 oop_slot = 6;
5150 } else {
5151 assert(rOop == j_rarg7, "wrong register");
5152 oop_slot = 7;
5153 }
5154
5155 oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
5156 int offset = oop_slot * VMRegImpl::stack_slot_size;
5157
5158 map->set_oop(VMRegImpl::stack2reg(oop_slot));
5159 // Store oop in handle area, may be null
5160 sd(rOop, Address(sp, offset));
5161 if (is_receiver) {
5162 *receiver_offset = offset;
5163 }
5164
5165 //rOop maybe the same as rHandle
5166 if (rOop == rHandle) {
5167 Label isZero;
5168 beqz(rOop, isZero);
5169 la(rHandle, Address(sp, offset));
5170 bind(isZero);
5171 } else {
5172 Label notZero2;
5173 la(rHandle, Address(sp, offset));
5174 bnez(rOop, notZero2);
5175 mv(rHandle, zr);
5176 bind(notZero2);
5177 }
5178 }
5179
5180 // If arg is on the stack then place it otherwise it is already in correct reg.
5181 if (dst.first()->is_stack()) {
5182 sd(rHandle, Address(sp, reg2offset_out(dst.first())));
5183 }
5184 }
5185
5186 // A float arg may have to do float reg int reg conversion
5187 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp) {
5188 assert((src.first()->is_stack() && dst.first()->is_stack()) ||
5189 (src.first()->is_reg() && dst.first()->is_reg()) ||
5190 (src.first()->is_stack() && dst.first()->is_reg()), "Unexpected error");
5191 if (src.first()->is_stack()) {
5192 if (dst.first()->is_stack()) {
5193 lwu(tmp, Address(fp, reg2offset_in(src.first())));
5194 sw(tmp, Address(sp, reg2offset_out(dst.first())));
5195 } else if (dst.first()->is_Register()) {
5196 lwu(dst.first()->as_Register(), Address(fp, reg2offset_in(src.first())));
5197 } else {
5198 ShouldNotReachHere();
5199 }
5200 } else if (src.first() != dst.first()) {
5201 if (src.is_single_phys_reg() && dst.is_single_phys_reg()) {
5202 fmv_s(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
5203 } else {
5204 ShouldNotReachHere();
5205 }
5206 }
5207 }
5208
5209 // A long move
5210 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp) {
5211 if (src.first()->is_stack()) {
5212 if (dst.first()->is_stack()) {
5213 // stack to stack
5214 ld(tmp, Address(fp, reg2offset_in(src.first())));
5215 sd(tmp, Address(sp, reg2offset_out(dst.first())));
5216 } else {
5217 // stack to reg
5218 ld(dst.first()->as_Register(), Address(fp, reg2offset_in(src.first())));
5219 }
5220 } else if (dst.first()->is_stack()) {
5221 // reg to stack
5222 sd(src.first()->as_Register(), Address(sp, reg2offset_out(dst.first())));
5223 } else {
5224 if (dst.first() != src.first()) {
5225 mv(dst.first()->as_Register(), src.first()->as_Register());
5226 }
5227 }
5228 }
5229
5230 // A double move
5231 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp) {
5232 assert((src.first()->is_stack() && dst.first()->is_stack()) ||
5233 (src.first()->is_reg() && dst.first()->is_reg()) ||
5234 (src.first()->is_stack() && dst.first()->is_reg()), "Unexpected error");
5235 if (src.first()->is_stack()) {
5236 if (dst.first()->is_stack()) {
5237 ld(tmp, Address(fp, reg2offset_in(src.first())));
5238 sd(tmp, Address(sp, reg2offset_out(dst.first())));
5239 } else if (dst.first()-> is_Register()) {
5240 ld(dst.first()->as_Register(), Address(fp, reg2offset_in(src.first())));
5241 } else {
5242 ShouldNotReachHere();
5243 }
5244 } else if (src.first() != dst.first()) {
5245 if (src.is_single_phys_reg() && dst.is_single_phys_reg()) {
5246 fmv_d(dst.first()->as_FloatRegister(), src.first()->as_FloatRegister());
5247 } else {
5248 ShouldNotReachHere();
5249 }
5250 }
5251 }
5252
5253 void MacroAssembler::test_bit(Register Rd, Register Rs, uint32_t bit_pos) {
5254 assert(bit_pos < 64, "invalid bit range");
5255 if (UseZbs) {
5256 bexti(Rd, Rs, bit_pos);
5257 return;
5258 }
5259 int64_t imm = (int64_t)(1UL << bit_pos);
5260 if (is_simm12(imm)) {
5261 and_imm12(Rd, Rs, imm);
5262 } else {
5263 srli(Rd, Rs, bit_pos);
5264 and_imm12(Rd, Rd, 1);
5265 }
5266 }
5267
5268 // Implements lightweight-locking.
5269 //
5270 // - obj: the object to be locked
5271 // - tmp1, tmp2, tmp3: temporary registers, will be destroyed
5272 // - slow: branched to if locking fails
5273 void MacroAssembler::lightweight_lock(Register obj, Register tmp1, Register tmp2, Register tmp3, Label& slow) {
5274 assert(LockingMode == LM_LIGHTWEIGHT, "only used with new lightweight locking");
5275 assert_different_registers(obj, tmp1, tmp2, tmp3, t0);
5276
5277 Label push;
5278 const Register top = tmp1;
5279 const Register mark = tmp2;
5280 const Register t = tmp3;
5281
5282 // Preload the markWord. It is important that this is the first
5283 // instruction emitted as it is part of C1's null check semantics.
5284 ld(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
5285
5286 // Check if the lock-stack is full.
5287 lwu(top, Address(xthread, JavaThread::lock_stack_top_offset()));
5288 mv(t, (unsigned)LockStack::end_offset());
5289 bge(top, t, slow, /* is_far */ true);
5290
5291 // Check for recursion.
5292 add(t, xthread, top);
5293 ld(t, Address(t, -oopSize));
5294 beq(obj, t, push);
5295
5296 // Check header for monitor (0b10).
5297 test_bit(t, mark, exact_log2(markWord::monitor_value));
5298 bnez(t, slow, /* is_far */ true);
5299
5300 // Try to lock. Transition lock-bits 0b01 => 0b00
5301 assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid a la");
5302 ori(mark, mark, markWord::unlocked_value);
5303 xori(t, mark, markWord::unlocked_value);
5304 cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::int64,
5305 /*acquire*/ Assembler::aq, /*release*/ Assembler::relaxed, /*result*/ t);
5306 bne(mark, t, slow, /* is_far */ true);
5307
5308 bind(push);
5309 // After successful lock, push object on lock-stack.
5310 add(t, xthread, top);
5311 sd(obj, Address(t));
5312 addw(top, top, oopSize);
5313 sw(top, Address(xthread, JavaThread::lock_stack_top_offset()));
5314 }
5315
5316 // Implements ligthweight-unlocking.
5317 //
5318 // - obj: the object to be unlocked
5319 // - tmp1, tmp2, tmp3: temporary registers
5320 // - slow: branched to if unlocking fails
5321 void MacroAssembler::lightweight_unlock(Register obj, Register tmp1, Register tmp2, Register tmp3, Label& slow) {
5322 assert(LockingMode == LM_LIGHTWEIGHT, "only used with new lightweight locking");
5323 assert_different_registers(obj, tmp1, tmp2, tmp3, t0);
5324
5325 #ifdef ASSERT
5326 {
5327 // Check for lock-stack underflow.
5328 Label stack_ok;
5329 lwu(tmp1, Address(xthread, JavaThread::lock_stack_top_offset()));
5330 mv(tmp2, (unsigned)LockStack::start_offset());
5331 bge(tmp1, tmp2, stack_ok);
5332 STOP("Lock-stack underflow");
5333 bind(stack_ok);
5334 }
5335 #endif
5336
5337 Label unlocked, push_and_slow;
5338 const Register top = tmp1;
5339 const Register mark = tmp2;
5340 const Register t = tmp3;
5341
5342 // Check if obj is top of lock-stack.
5343 lwu(top, Address(xthread, JavaThread::lock_stack_top_offset()));
5344 subw(top, top, oopSize);
5345 add(t, xthread, top);
5346 ld(t, Address(t));
5347 bne(obj, t, slow, /* is_far */ true);
5348
5349 // Pop lock-stack.
5350 DEBUG_ONLY(add(t, xthread, top);)
5351 DEBUG_ONLY(sd(zr, Address(t));)
5352 sw(top, Address(xthread, JavaThread::lock_stack_top_offset()));
5353
5354 // Check if recursive.
5355 add(t, xthread, top);
5356 ld(t, Address(t, -oopSize));
5357 beq(obj, t, unlocked);
5358
5359 // Not recursive. Check header for monitor (0b10).
5360 ld(mark, Address(obj, oopDesc::mark_offset_in_bytes()));
5361 test_bit(t, mark, exact_log2(markWord::monitor_value));
5362 bnez(t, push_and_slow);
5363
5364 #ifdef ASSERT
5365 // Check header not unlocked (0b01).
5366 Label not_unlocked;
5367 test_bit(t, mark, exact_log2(markWord::unlocked_value));
5368 beqz(t, not_unlocked);
5369 stop("lightweight_unlock already unlocked");
5370 bind(not_unlocked);
5371 #endif
5372
5373 // Try to unlock. Transition lock bits 0b00 => 0b01
5374 assert(oopDesc::mark_offset_in_bytes() == 0, "required to avoid lea");
5375 ori(t, mark, markWord::unlocked_value);
5376 cmpxchg(/*addr*/ obj, /*expected*/ mark, /*new*/ t, Assembler::int64,
5377 /*acquire*/ Assembler::relaxed, /*release*/ Assembler::rl, /*result*/ t);
5378 beq(mark, t, unlocked);
5379
5380 bind(push_and_slow);
5381 // Restore lock-stack and handle the unlock in runtime.
5382 DEBUG_ONLY(add(t, xthread, top);)
5383 DEBUG_ONLY(sd(obj, Address(t));)
5384 addw(top, top, oopSize);
5385 sw(top, Address(xthread, JavaThread::lock_stack_top_offset()));
5386 j(slow);
5387
5388 bind(unlocked);
5389 }