1 /*
2 * Copyright (c) 1997, 2023, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/assembler.hpp"
27 #include "asm/assembler.inline.hpp"
28 #include "compiler/compiler_globals.hpp"
29 #include "compiler/disassembler.hpp"
30 #include "crc32c.h"
31 #include "gc/shared/barrierSet.hpp"
32 #include "gc/shared/barrierSetAssembler.hpp"
33 #include "gc/shared/collectedHeap.inline.hpp"
34 #include "gc/shared/tlab_globals.hpp"
35 #include "interpreter/bytecodeHistogram.hpp"
36 #include "interpreter/interpreter.hpp"
37 #include "jvm.h"
38 #include "memory/resourceArea.hpp"
39 #include "memory/universe.hpp"
40 #include "oops/accessDecorators.hpp"
41 #include "oops/compressedOops.inline.hpp"
42 #include "oops/klass.inline.hpp"
43 #include "prims/methodHandles.hpp"
44 #include "runtime/continuation.hpp"
45 #include "runtime/interfaceSupport.inline.hpp"
46 #include "runtime/javaThread.hpp"
47 #include "runtime/jniHandles.hpp"
48 #include "runtime/objectMonitor.hpp"
49 #include "runtime/os.hpp"
50 #include "runtime/safepoint.hpp"
51 #include "runtime/safepointMechanism.hpp"
52 #include "runtime/sharedRuntime.hpp"
53 #include "runtime/stubRoutines.hpp"
54 #include "utilities/macros.hpp"
55
56 #ifdef PRODUCT
57 #define BLOCK_COMMENT(str) /* nothing */
58 #define STOP(error) stop(error)
59 #else
60 #define BLOCK_COMMENT(str) block_comment(str)
61 #define STOP(error) block_comment(error); stop(error)
62 #endif
63
64 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
65
66 #ifdef ASSERT
67 bool AbstractAssembler::pd_check_instruction_mark() { return true; }
68 #endif
69
70 static const Assembler::Condition reverse[] = {
71 Assembler::noOverflow /* overflow = 0x0 */ ,
72 Assembler::overflow /* noOverflow = 0x1 */ ,
73 Assembler::aboveEqual /* carrySet = 0x2, below = 0x2 */ ,
74 Assembler::below /* aboveEqual = 0x3, carryClear = 0x3 */ ,
75 Assembler::notZero /* zero = 0x4, equal = 0x4 */ ,
76 Assembler::zero /* notZero = 0x5, notEqual = 0x5 */ ,
77 Assembler::above /* belowEqual = 0x6 */ ,
78 Assembler::belowEqual /* above = 0x7 */ ,
79 Assembler::positive /* negative = 0x8 */ ,
80 Assembler::negative /* positive = 0x9 */ ,
81 Assembler::noParity /* parity = 0xa */ ,
82 Assembler::parity /* noParity = 0xb */ ,
83 Assembler::greaterEqual /* less = 0xc */ ,
84 Assembler::less /* greaterEqual = 0xd */ ,
85 Assembler::greater /* lessEqual = 0xe */ ,
86 Assembler::lessEqual /* greater = 0xf, */
87
88 };
89
90
91 // Implementation of MacroAssembler
92
93 // First all the versions that have distinct versions depending on 32/64 bit
94 // Unless the difference is trivial (1 line or so).
95
96 #ifndef _LP64
97
98 // 32bit versions
99
100 Address MacroAssembler::as_Address(AddressLiteral adr) {
101 return Address(adr.target(), adr.rspec());
102 }
103
104 Address MacroAssembler::as_Address(ArrayAddress adr, Register rscratch) {
105 assert(rscratch == noreg, "");
106 return Address::make_array(adr);
107 }
108
109 void MacroAssembler::call_VM_leaf_base(address entry_point,
110 int number_of_arguments) {
111 call(RuntimeAddress(entry_point));
112 increment(rsp, number_of_arguments * wordSize);
113 }
114
115 void MacroAssembler::cmpklass(Address src1, Metadata* obj) {
116 cmp_literal32(src1, (int32_t)obj, metadata_Relocation::spec_for_immediate());
117 }
118
119
120 void MacroAssembler::cmpklass(Register src1, Metadata* obj) {
121 cmp_literal32(src1, (int32_t)obj, metadata_Relocation::spec_for_immediate());
122 }
123
124 void MacroAssembler::cmpoop(Address src1, jobject obj) {
125 cmp_literal32(src1, (int32_t)obj, oop_Relocation::spec_for_immediate());
126 }
127
128 void MacroAssembler::cmpoop(Register src1, jobject obj, Register rscratch) {
129 assert(rscratch == noreg, "redundant");
130 cmp_literal32(src1, (int32_t)obj, oop_Relocation::spec_for_immediate());
131 }
132
133 void MacroAssembler::extend_sign(Register hi, Register lo) {
134 // According to Intel Doc. AP-526, "Integer Divide", p.18.
135 if (VM_Version::is_P6() && hi == rdx && lo == rax) {
136 cdql();
137 } else {
138 movl(hi, lo);
139 sarl(hi, 31);
140 }
141 }
142
143 void MacroAssembler::jC2(Register tmp, Label& L) {
144 // set parity bit if FPU flag C2 is set (via rax)
145 save_rax(tmp);
146 fwait(); fnstsw_ax();
147 sahf();
148 restore_rax(tmp);
149 // branch
150 jcc(Assembler::parity, L);
151 }
152
153 void MacroAssembler::jnC2(Register tmp, Label& L) {
154 // set parity bit if FPU flag C2 is set (via rax)
155 save_rax(tmp);
156 fwait(); fnstsw_ax();
157 sahf();
158 restore_rax(tmp);
159 // branch
160 jcc(Assembler::noParity, L);
161 }
162
163 // 32bit can do a case table jump in one instruction but we no longer allow the base
164 // to be installed in the Address class
165 void MacroAssembler::jump(ArrayAddress entry, Register rscratch) {
166 assert(rscratch == noreg, "not needed");
167 jmp(as_Address(entry, noreg));
168 }
169
170 // Note: y_lo will be destroyed
171 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) {
172 // Long compare for Java (semantics as described in JVM spec.)
173 Label high, low, done;
174
175 cmpl(x_hi, y_hi);
176 jcc(Assembler::less, low);
177 jcc(Assembler::greater, high);
178 // x_hi is the return register
179 xorl(x_hi, x_hi);
180 cmpl(x_lo, y_lo);
181 jcc(Assembler::below, low);
182 jcc(Assembler::equal, done);
183
184 bind(high);
185 xorl(x_hi, x_hi);
186 increment(x_hi);
187 jmp(done);
188
189 bind(low);
190 xorl(x_hi, x_hi);
191 decrementl(x_hi);
192
193 bind(done);
194 }
195
196 void MacroAssembler::lea(Register dst, AddressLiteral src) {
197 mov_literal32(dst, (int32_t)src.target(), src.rspec());
198 }
199
200 void MacroAssembler::lea(Address dst, AddressLiteral adr, Register rscratch) {
201 assert(rscratch == noreg, "not needed");
202
203 // leal(dst, as_Address(adr));
204 // see note in movl as to why we must use a move
205 mov_literal32(dst, (int32_t)adr.target(), adr.rspec());
206 }
207
208 void MacroAssembler::leave() {
209 mov(rsp, rbp);
210 pop(rbp);
211 }
212
213 void MacroAssembler::lmul(int x_rsp_offset, int y_rsp_offset) {
214 // Multiplication of two Java long values stored on the stack
215 // as illustrated below. Result is in rdx:rax.
216 //
217 // rsp ---> [ ?? ] \ \
218 // .... | y_rsp_offset |
219 // [ y_lo ] / (in bytes) | x_rsp_offset
220 // [ y_hi ] | (in bytes)
221 // .... |
222 // [ x_lo ] /
223 // [ x_hi ]
224 // ....
225 //
226 // Basic idea: lo(result) = lo(x_lo * y_lo)
227 // hi(result) = hi(x_lo * y_lo) + lo(x_hi * y_lo) + lo(x_lo * y_hi)
228 Address x_hi(rsp, x_rsp_offset + wordSize); Address x_lo(rsp, x_rsp_offset);
229 Address y_hi(rsp, y_rsp_offset + wordSize); Address y_lo(rsp, y_rsp_offset);
230 Label quick;
231 // load x_hi, y_hi and check if quick
232 // multiplication is possible
233 movl(rbx, x_hi);
234 movl(rcx, y_hi);
235 movl(rax, rbx);
236 orl(rbx, rcx); // rbx, = 0 <=> x_hi = 0 and y_hi = 0
237 jcc(Assembler::zero, quick); // if rbx, = 0 do quick multiply
238 // do full multiplication
239 // 1st step
240 mull(y_lo); // x_hi * y_lo
241 movl(rbx, rax); // save lo(x_hi * y_lo) in rbx,
242 // 2nd step
243 movl(rax, x_lo);
244 mull(rcx); // x_lo * y_hi
245 addl(rbx, rax); // add lo(x_lo * y_hi) to rbx,
246 // 3rd step
247 bind(quick); // note: rbx, = 0 if quick multiply!
248 movl(rax, x_lo);
249 mull(y_lo); // x_lo * y_lo
250 addl(rdx, rbx); // correct hi(x_lo * y_lo)
251 }
252
253 void MacroAssembler::lneg(Register hi, Register lo) {
254 negl(lo);
255 adcl(hi, 0);
256 negl(hi);
257 }
258
259 void MacroAssembler::lshl(Register hi, Register lo) {
260 // Java shift left long support (semantics as described in JVM spec., p.305)
261 // (basic idea for shift counts s >= n: x << s == (x << n) << (s - n))
262 // shift value is in rcx !
263 assert(hi != rcx, "must not use rcx");
264 assert(lo != rcx, "must not use rcx");
265 const Register s = rcx; // shift count
266 const int n = BitsPerWord;
267 Label L;
268 andl(s, 0x3f); // s := s & 0x3f (s < 0x40)
269 cmpl(s, n); // if (s < n)
270 jcc(Assembler::less, L); // else (s >= n)
271 movl(hi, lo); // x := x << n
272 xorl(lo, lo);
273 // Note: subl(s, n) is not needed since the Intel shift instructions work rcx mod n!
274 bind(L); // s (mod n) < n
275 shldl(hi, lo); // x := x << s
276 shll(lo);
277 }
278
279
280 void MacroAssembler::lshr(Register hi, Register lo, bool sign_extension) {
281 // Java shift right long support (semantics as described in JVM spec., p.306 & p.310)
282 // (basic idea for shift counts s >= n: x >> s == (x >> n) >> (s - n))
283 assert(hi != rcx, "must not use rcx");
284 assert(lo != rcx, "must not use rcx");
285 const Register s = rcx; // shift count
286 const int n = BitsPerWord;
287 Label L;
288 andl(s, 0x3f); // s := s & 0x3f (s < 0x40)
289 cmpl(s, n); // if (s < n)
290 jcc(Assembler::less, L); // else (s >= n)
291 movl(lo, hi); // x := x >> n
292 if (sign_extension) sarl(hi, 31);
293 else xorl(hi, hi);
294 // Note: subl(s, n) is not needed since the Intel shift instructions work rcx mod n!
295 bind(L); // s (mod n) < n
296 shrdl(lo, hi); // x := x >> s
297 if (sign_extension) sarl(hi);
298 else shrl(hi);
299 }
300
301 void MacroAssembler::movoop(Register dst, jobject obj) {
302 mov_literal32(dst, (int32_t)obj, oop_Relocation::spec_for_immediate());
303 }
304
305 void MacroAssembler::movoop(Address dst, jobject obj, Register rscratch) {
306 assert(rscratch == noreg, "redundant");
307 mov_literal32(dst, (int32_t)obj, oop_Relocation::spec_for_immediate());
308 }
309
310 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
311 mov_literal32(dst, (int32_t)obj, metadata_Relocation::spec_for_immediate());
312 }
313
314 void MacroAssembler::mov_metadata(Address dst, Metadata* obj, Register rscratch) {
315 assert(rscratch == noreg, "redundant");
316 mov_literal32(dst, (int32_t)obj, metadata_Relocation::spec_for_immediate());
317 }
318
319 void MacroAssembler::movptr(Register dst, AddressLiteral src) {
320 if (src.is_lval()) {
321 mov_literal32(dst, (intptr_t)src.target(), src.rspec());
322 } else {
323 movl(dst, as_Address(src));
324 }
325 }
326
327 void MacroAssembler::movptr(ArrayAddress dst, Register src, Register rscratch) {
328 assert(rscratch == noreg, "redundant");
329 movl(as_Address(dst, noreg), src);
330 }
331
332 void MacroAssembler::movptr(Register dst, ArrayAddress src) {
333 movl(dst, as_Address(src, noreg));
334 }
335
336 void MacroAssembler::movptr(Address dst, intptr_t src, Register rscratch) {
337 assert(rscratch == noreg, "redundant");
338 movl(dst, src);
339 }
340
341 void MacroAssembler::pushoop(jobject obj, Register rscratch) {
342 assert(rscratch == noreg, "redundant");
343 push_literal32((int32_t)obj, oop_Relocation::spec_for_immediate());
344 }
345
346 void MacroAssembler::pushklass(Metadata* obj, Register rscratch) {
347 assert(rscratch == noreg, "redundant");
348 push_literal32((int32_t)obj, metadata_Relocation::spec_for_immediate());
349 }
350
351 void MacroAssembler::pushptr(AddressLiteral src, Register rscratch) {
352 assert(rscratch == noreg, "redundant");
353 if (src.is_lval()) {
354 push_literal32((int32_t)src.target(), src.rspec());
355 } else {
356 pushl(as_Address(src));
357 }
358 }
359
360 static void pass_arg0(MacroAssembler* masm, Register arg) {
361 masm->push(arg);
362 }
363
364 static void pass_arg1(MacroAssembler* masm, Register arg) {
365 masm->push(arg);
366 }
367
368 static void pass_arg2(MacroAssembler* masm, Register arg) {
369 masm->push(arg);
370 }
371
372 static void pass_arg3(MacroAssembler* masm, Register arg) {
373 masm->push(arg);
374 }
375
376 #ifndef PRODUCT
377 extern "C" void findpc(intptr_t x);
378 #endif
379
380 void MacroAssembler::debug32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip, char* msg) {
381 // In order to get locks to work, we need to fake a in_VM state
382 JavaThread* thread = JavaThread::current();
383 JavaThreadState saved_state = thread->thread_state();
384 thread->set_thread_state(_thread_in_vm);
385 if (ShowMessageBoxOnError) {
386 JavaThread* thread = JavaThread::current();
387 JavaThreadState saved_state = thread->thread_state();
388 thread->set_thread_state(_thread_in_vm);
389 if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
390 ttyLocker ttyl;
391 BytecodeCounter::print();
392 }
393 // To see where a verify_oop failed, get $ebx+40/X for this frame.
394 // This is the value of eip which points to where verify_oop will return.
395 if (os::message_box(msg, "Execution stopped, print registers?")) {
396 print_state32(rdi, rsi, rbp, rsp, rbx, rdx, rcx, rax, eip);
397 BREAKPOINT;
398 }
399 }
400 fatal("DEBUG MESSAGE: %s", msg);
401 }
402
403 void MacroAssembler::print_state32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip) {
404 ttyLocker ttyl;
405 DebuggingContext debugging{};
406 tty->print_cr("eip = 0x%08x", eip);
407 #ifndef PRODUCT
408 if ((WizardMode || Verbose) && PrintMiscellaneous) {
409 tty->cr();
410 findpc(eip);
411 tty->cr();
412 }
413 #endif
414 #define PRINT_REG(rax) \
415 { tty->print("%s = ", #rax); os::print_location(tty, rax); }
416 PRINT_REG(rax);
417 PRINT_REG(rbx);
418 PRINT_REG(rcx);
419 PRINT_REG(rdx);
420 PRINT_REG(rdi);
421 PRINT_REG(rsi);
422 PRINT_REG(rbp);
423 PRINT_REG(rsp);
424 #undef PRINT_REG
425 // Print some words near top of staack.
426 int* dump_sp = (int*) rsp;
427 for (int col1 = 0; col1 < 8; col1++) {
428 tty->print("(rsp+0x%03x) 0x%08x: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
429 os::print_location(tty, *dump_sp++);
430 }
431 for (int row = 0; row < 16; row++) {
432 tty->print("(rsp+0x%03x) 0x%08x: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
433 for (int col = 0; col < 8; col++) {
434 tty->print(" 0x%08x", *dump_sp++);
435 }
436 tty->cr();
437 }
438 // Print some instructions around pc:
439 Disassembler::decode((address)eip-64, (address)eip);
440 tty->print_cr("--------");
441 Disassembler::decode((address)eip, (address)eip+32);
442 }
443
444 void MacroAssembler::stop(const char* msg) {
445 // push address of message
446 ExternalAddress message((address)msg);
447 pushptr(message.addr(), noreg);
448 { Label L; call(L, relocInfo::none); bind(L); } // push eip
449 pusha(); // push registers
450 call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32)));
451 hlt();
452 }
453
454 void MacroAssembler::warn(const char* msg) {
455 push_CPU_state();
456
457 // push address of message
458 ExternalAddress message((address)msg);
459 pushptr(message.addr(), noreg);
460
461 call(RuntimeAddress(CAST_FROM_FN_PTR(address, warning)));
462 addl(rsp, wordSize); // discard argument
463 pop_CPU_state();
464 }
465
466 void MacroAssembler::print_state() {
467 { Label L; call(L, relocInfo::none); bind(L); } // push eip
468 pusha(); // push registers
469
470 push_CPU_state();
471 call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::print_state32)));
472 pop_CPU_state();
473
474 popa();
475 addl(rsp, wordSize);
476 }
477
478 #else // _LP64
479
480 // 64 bit versions
481
482 Address MacroAssembler::as_Address(AddressLiteral adr) {
483 // amd64 always does this as a pc-rel
484 // we can be absolute or disp based on the instruction type
485 // jmp/call are displacements others are absolute
486 assert(!adr.is_lval(), "must be rval");
487 assert(reachable(adr), "must be");
488 return Address(checked_cast<int32_t>(adr.target() - pc()), adr.target(), adr.reloc());
489
490 }
491
492 Address MacroAssembler::as_Address(ArrayAddress adr, Register rscratch) {
493 AddressLiteral base = adr.base();
494 lea(rscratch, base);
495 Address index = adr.index();
496 assert(index._disp == 0, "must not have disp"); // maybe it can?
497 Address array(rscratch, index._index, index._scale, index._disp);
498 return array;
499 }
500
501 void MacroAssembler::call_VM_leaf_base(address entry_point, int num_args) {
502 Label L, E;
503
504 #ifdef _WIN64
505 // Windows always allocates space for it's register args
506 assert(num_args <= 4, "only register arguments supported");
507 subq(rsp, frame::arg_reg_save_area_bytes);
508 #endif
509
510 // Align stack if necessary
511 testl(rsp, 15);
512 jcc(Assembler::zero, L);
513
514 subq(rsp, 8);
515 call(RuntimeAddress(entry_point));
516 addq(rsp, 8);
517 jmp(E);
518
519 bind(L);
520 call(RuntimeAddress(entry_point));
521
522 bind(E);
523
524 #ifdef _WIN64
525 // restore stack pointer
526 addq(rsp, frame::arg_reg_save_area_bytes);
527 #endif
528
529 }
530
531 void MacroAssembler::cmp64(Register src1, AddressLiteral src2, Register rscratch) {
532 assert(!src2.is_lval(), "should use cmpptr");
533 assert(rscratch != noreg || always_reachable(src2), "missing");
534
535 if (reachable(src2)) {
536 cmpq(src1, as_Address(src2));
537 } else {
538 lea(rscratch, src2);
539 Assembler::cmpq(src1, Address(rscratch, 0));
540 }
541 }
542
543 int MacroAssembler::corrected_idivq(Register reg) {
544 // Full implementation of Java ldiv and lrem; checks for special
545 // case as described in JVM spec., p.243 & p.271. The function
546 // returns the (pc) offset of the idivl instruction - may be needed
547 // for implicit exceptions.
548 //
549 // normal case special case
550 //
551 // input : rax: dividend min_long
552 // reg: divisor (may not be eax/edx) -1
553 //
554 // output: rax: quotient (= rax idiv reg) min_long
555 // rdx: remainder (= rax irem reg) 0
556 assert(reg != rax && reg != rdx, "reg cannot be rax or rdx register");
557 static const int64_t min_long = 0x8000000000000000;
558 Label normal_case, special_case;
559
560 // check for special case
561 cmp64(rax, ExternalAddress((address) &min_long), rdx /*rscratch*/);
562 jcc(Assembler::notEqual, normal_case);
563 xorl(rdx, rdx); // prepare rdx for possible special case (where
564 // remainder = 0)
565 cmpq(reg, -1);
566 jcc(Assembler::equal, special_case);
567
568 // handle normal case
569 bind(normal_case);
570 cdqq();
571 int idivq_offset = offset();
572 idivq(reg);
573
574 // normal and special case exit
575 bind(special_case);
576
577 return idivq_offset;
578 }
579
580 void MacroAssembler::decrementq(Register reg, int value) {
581 if (value == min_jint) { subq(reg, value); return; }
582 if (value < 0) { incrementq(reg, -value); return; }
583 if (value == 0) { ; return; }
584 if (value == 1 && UseIncDec) { decq(reg) ; return; }
585 /* else */ { subq(reg, value) ; return; }
586 }
587
588 void MacroAssembler::decrementq(Address dst, int value) {
589 if (value == min_jint) { subq(dst, value); return; }
590 if (value < 0) { incrementq(dst, -value); return; }
591 if (value == 0) { ; return; }
592 if (value == 1 && UseIncDec) { decq(dst) ; return; }
593 /* else */ { subq(dst, value) ; return; }
594 }
595
596 void MacroAssembler::incrementq(AddressLiteral dst, Register rscratch) {
597 assert(rscratch != noreg || always_reachable(dst), "missing");
598
599 if (reachable(dst)) {
600 incrementq(as_Address(dst));
601 } else {
602 lea(rscratch, dst);
603 incrementq(Address(rscratch, 0));
604 }
605 }
606
607 void MacroAssembler::incrementq(Register reg, int value) {
608 if (value == min_jint) { addq(reg, value); return; }
609 if (value < 0) { decrementq(reg, -value); return; }
610 if (value == 0) { ; return; }
611 if (value == 1 && UseIncDec) { incq(reg) ; return; }
612 /* else */ { addq(reg, value) ; return; }
613 }
614
615 void MacroAssembler::incrementq(Address dst, int value) {
616 if (value == min_jint) { addq(dst, value); return; }
617 if (value < 0) { decrementq(dst, -value); return; }
618 if (value == 0) { ; return; }
619 if (value == 1 && UseIncDec) { incq(dst) ; return; }
620 /* else */ { addq(dst, value) ; return; }
621 }
622
623 // 32bit can do a case table jump in one instruction but we no longer allow the base
624 // to be installed in the Address class
625 void MacroAssembler::jump(ArrayAddress entry, Register rscratch) {
626 lea(rscratch, entry.base());
627 Address dispatch = entry.index();
628 assert(dispatch._base == noreg, "must be");
629 dispatch._base = rscratch;
630 jmp(dispatch);
631 }
632
633 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) {
634 ShouldNotReachHere(); // 64bit doesn't use two regs
635 cmpq(x_lo, y_lo);
636 }
637
638 void MacroAssembler::lea(Register dst, AddressLiteral src) {
639 mov_literal64(dst, (intptr_t)src.target(), src.rspec());
640 }
641
642 void MacroAssembler::lea(Address dst, AddressLiteral adr, Register rscratch) {
643 lea(rscratch, adr);
644 movptr(dst, rscratch);
645 }
646
647 void MacroAssembler::leave() {
648 // %%% is this really better? Why not on 32bit too?
649 emit_int8((unsigned char)0xC9); // LEAVE
650 }
651
652 void MacroAssembler::lneg(Register hi, Register lo) {
653 ShouldNotReachHere(); // 64bit doesn't use two regs
654 negq(lo);
655 }
656
657 void MacroAssembler::movoop(Register dst, jobject obj) {
658 mov_literal64(dst, (intptr_t)obj, oop_Relocation::spec_for_immediate());
659 }
660
661 void MacroAssembler::movoop(Address dst, jobject obj, Register rscratch) {
662 mov_literal64(rscratch, (intptr_t)obj, oop_Relocation::spec_for_immediate());
663 movq(dst, rscratch);
664 }
665
666 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
667 mov_literal64(dst, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
668 }
669
670 void MacroAssembler::mov_metadata(Address dst, Metadata* obj, Register rscratch) {
671 mov_literal64(rscratch, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
672 movq(dst, rscratch);
673 }
674
675 void MacroAssembler::movptr(Register dst, AddressLiteral src) {
676 if (src.is_lval()) {
677 mov_literal64(dst, (intptr_t)src.target(), src.rspec());
678 } else {
679 if (reachable(src)) {
680 movq(dst, as_Address(src));
681 } else {
682 lea(dst, src);
683 movq(dst, Address(dst, 0));
684 }
685 }
686 }
687
688 void MacroAssembler::movptr(ArrayAddress dst, Register src, Register rscratch) {
689 movq(as_Address(dst, rscratch), src);
690 }
691
692 void MacroAssembler::movptr(Register dst, ArrayAddress src) {
693 movq(dst, as_Address(src, dst /*rscratch*/));
694 }
695
696 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
697 void MacroAssembler::movptr(Address dst, intptr_t src, Register rscratch) {
698 if (is_simm32(src)) {
699 movptr(dst, checked_cast<int32_t>(src));
700 } else {
701 mov64(rscratch, src);
702 movq(dst, rscratch);
703 }
704 }
705
706 void MacroAssembler::pushoop(jobject obj, Register rscratch) {
707 movoop(rscratch, obj);
708 push(rscratch);
709 }
710
711 void MacroAssembler::pushklass(Metadata* obj, Register rscratch) {
712 mov_metadata(rscratch, obj);
713 push(rscratch);
714 }
715
716 void MacroAssembler::pushptr(AddressLiteral src, Register rscratch) {
717 lea(rscratch, src);
718 if (src.is_lval()) {
719 push(rscratch);
720 } else {
721 pushq(Address(rscratch, 0));
722 }
723 }
724
725 void MacroAssembler::reset_last_Java_frame(bool clear_fp) {
726 reset_last_Java_frame(r15_thread, clear_fp);
727 }
728
729 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
730 Register last_java_fp,
731 address last_java_pc,
732 Register rscratch) {
733 set_last_Java_frame(r15_thread, last_java_sp, last_java_fp, last_java_pc, rscratch);
734 }
735
736 static void pass_arg0(MacroAssembler* masm, Register arg) {
737 if (c_rarg0 != arg ) {
738 masm->mov(c_rarg0, arg);
739 }
740 }
741
742 static void pass_arg1(MacroAssembler* masm, Register arg) {
743 if (c_rarg1 != arg ) {
744 masm->mov(c_rarg1, arg);
745 }
746 }
747
748 static void pass_arg2(MacroAssembler* masm, Register arg) {
749 if (c_rarg2 != arg ) {
750 masm->mov(c_rarg2, arg);
751 }
752 }
753
754 static void pass_arg3(MacroAssembler* masm, Register arg) {
755 if (c_rarg3 != arg ) {
756 masm->mov(c_rarg3, arg);
757 }
758 }
759
760 void MacroAssembler::stop(const char* msg) {
761 if (ShowMessageBoxOnError) {
762 address rip = pc();
763 pusha(); // get regs on stack
764 lea(c_rarg1, InternalAddress(rip));
765 movq(c_rarg2, rsp); // pass pointer to regs array
766 }
767 lea(c_rarg0, ExternalAddress((address) msg));
768 andq(rsp, -16); // align stack as required by ABI
769 call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
770 hlt();
771 }
772
773 void MacroAssembler::warn(const char* msg) {
774 push(rbp);
775 movq(rbp, rsp);
776 andq(rsp, -16); // align stack as required by push_CPU_state and call
777 push_CPU_state(); // keeps alignment at 16 bytes
778
779 lea(c_rarg0, ExternalAddress((address) msg));
780 call(RuntimeAddress(CAST_FROM_FN_PTR(address, warning)));
781
782 pop_CPU_state();
783 mov(rsp, rbp);
784 pop(rbp);
785 }
786
787 void MacroAssembler::print_state() {
788 address rip = pc();
789 pusha(); // get regs on stack
790 push(rbp);
791 movq(rbp, rsp);
792 andq(rsp, -16); // align stack as required by push_CPU_state and call
793 push_CPU_state(); // keeps alignment at 16 bytes
794
795 lea(c_rarg0, InternalAddress(rip));
796 lea(c_rarg1, Address(rbp, wordSize)); // pass pointer to regs array
797 call_VM_leaf(CAST_FROM_FN_PTR(address, MacroAssembler::print_state64), c_rarg0, c_rarg1);
798
799 pop_CPU_state();
800 mov(rsp, rbp);
801 pop(rbp);
802 popa();
803 }
804
805 #ifndef PRODUCT
806 extern "C" void findpc(intptr_t x);
807 #endif
808
809 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[]) {
810 // In order to get locks to work, we need to fake a in_VM state
811 if (ShowMessageBoxOnError) {
812 JavaThread* thread = JavaThread::current();
813 JavaThreadState saved_state = thread->thread_state();
814 thread->set_thread_state(_thread_in_vm);
815 #ifndef PRODUCT
816 if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
817 ttyLocker ttyl;
818 BytecodeCounter::print();
819 }
820 #endif
821 // To see where a verify_oop failed, get $ebx+40/X for this frame.
822 // XXX correct this offset for amd64
823 // This is the value of eip which points to where verify_oop will return.
824 if (os::message_box(msg, "Execution stopped, print registers?")) {
825 print_state64(pc, regs);
826 BREAKPOINT;
827 }
828 }
829 fatal("DEBUG MESSAGE: %s", msg);
830 }
831
832 void MacroAssembler::print_state64(int64_t pc, int64_t regs[]) {
833 ttyLocker ttyl;
834 DebuggingContext debugging{};
835 tty->print_cr("rip = 0x%016lx", (intptr_t)pc);
836 #ifndef PRODUCT
837 tty->cr();
838 findpc(pc);
839 tty->cr();
840 #endif
841 #define PRINT_REG(rax, value) \
842 { tty->print("%s = ", #rax); os::print_location(tty, value); }
843 PRINT_REG(rax, regs[15]);
844 PRINT_REG(rbx, regs[12]);
845 PRINT_REG(rcx, regs[14]);
846 PRINT_REG(rdx, regs[13]);
847 PRINT_REG(rdi, regs[8]);
848 PRINT_REG(rsi, regs[9]);
849 PRINT_REG(rbp, regs[10]);
850 // rsp is actually not stored by pusha(), compute the old rsp from regs (rsp after pusha): regs + 16 = old rsp
851 PRINT_REG(rsp, (intptr_t)(®s[16]));
852 PRINT_REG(r8 , regs[7]);
853 PRINT_REG(r9 , regs[6]);
854 PRINT_REG(r10, regs[5]);
855 PRINT_REG(r11, regs[4]);
856 PRINT_REG(r12, regs[3]);
857 PRINT_REG(r13, regs[2]);
858 PRINT_REG(r14, regs[1]);
859 PRINT_REG(r15, regs[0]);
860 #undef PRINT_REG
861 // Print some words near the top of the stack.
862 int64_t* rsp = ®s[16];
863 int64_t* dump_sp = rsp;
864 for (int col1 = 0; col1 < 8; col1++) {
865 tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
866 os::print_location(tty, *dump_sp++);
867 }
868 for (int row = 0; row < 25; row++) {
869 tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
870 for (int col = 0; col < 4; col++) {
871 tty->print(" 0x%016lx", (intptr_t)*dump_sp++);
872 }
873 tty->cr();
874 }
875 // Print some instructions around pc:
876 Disassembler::decode((address)pc-64, (address)pc);
877 tty->print_cr("--------");
878 Disassembler::decode((address)pc, (address)pc+32);
879 }
880
881 // The java_calling_convention describes stack locations as ideal slots on
882 // a frame with no abi restrictions. Since we must observe abi restrictions
883 // (like the placement of the register window) the slots must be biased by
884 // the following value.
885 static int reg2offset_in(VMReg r) {
886 // Account for saved rbp and return address
887 // This should really be in_preserve_stack_slots
888 return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
889 }
890
891 static int reg2offset_out(VMReg r) {
892 return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
893 }
894
895 // A long move
896 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
897
898 // The calling conventions assures us that each VMregpair is either
899 // all really one physical register or adjacent stack slots.
900
901 if (src.is_single_phys_reg() ) {
902 if (dst.is_single_phys_reg()) {
903 if (dst.first() != src.first()) {
904 mov(dst.first()->as_Register(), src.first()->as_Register());
905 }
906 } else {
907 assert(dst.is_single_reg(), "not a stack pair: (%s, %s), (%s, %s)",
908 src.first()->name(), src.second()->name(), dst.first()->name(), dst.second()->name());
909 movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
910 }
911 } else if (dst.is_single_phys_reg()) {
912 assert(src.is_single_reg(), "not a stack pair");
913 movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
914 } else {
915 assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
916 movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
917 movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
918 }
919 }
920
921 // A double move
922 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
923
924 // The calling conventions assures us that each VMregpair is either
925 // all really one physical register or adjacent stack slots.
926
927 if (src.is_single_phys_reg() ) {
928 if (dst.is_single_phys_reg()) {
929 // In theory these overlap but the ordering is such that this is likely a nop
930 if ( src.first() != dst.first()) {
931 movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister());
932 }
933 } else {
934 assert(dst.is_single_reg(), "not a stack pair");
935 movdbl(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
936 }
937 } else if (dst.is_single_phys_reg()) {
938 assert(src.is_single_reg(), "not a stack pair");
939 movdbl(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
940 } else {
941 assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
942 movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
943 movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
944 }
945 }
946
947
948 // A float arg may have to do float reg int reg conversion
949 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
950 assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move");
951
952 // The calling conventions assures us that each VMregpair is either
953 // all really one physical register or adjacent stack slots.
954
955 if (src.first()->is_stack()) {
956 if (dst.first()->is_stack()) {
957 movl(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
958 movptr(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
959 } else {
960 // stack to reg
961 assert(dst.first()->is_XMMRegister(), "only expect xmm registers as parameters");
962 movflt(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
963 }
964 } else if (dst.first()->is_stack()) {
965 // reg to stack
966 assert(src.first()->is_XMMRegister(), "only expect xmm registers as parameters");
967 movflt(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
968 } else {
969 // reg to reg
970 // In theory these overlap but the ordering is such that this is likely a nop
971 if ( src.first() != dst.first()) {
972 movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister());
973 }
974 }
975 }
976
977 // On 64 bit we will store integer like items to the stack as
978 // 64 bits items (x86_32/64 abi) even though java would only store
979 // 32bits for a parameter. On 32bit it will simply be 32 bits
980 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
981 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
982 if (src.first()->is_stack()) {
983 if (dst.first()->is_stack()) {
984 // stack to stack
985 movslq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
986 movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
987 } else {
988 // stack to reg
989 movslq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
990 }
991 } else if (dst.first()->is_stack()) {
992 // reg to stack
993 // Do we really have to sign extend???
994 // __ movslq(src.first()->as_Register(), src.first()->as_Register());
995 movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
996 } else {
997 // Do we really have to sign extend???
998 // __ movslq(dst.first()->as_Register(), src.first()->as_Register());
999 if (dst.first() != src.first()) {
1000 movq(dst.first()->as_Register(), src.first()->as_Register());
1001 }
1002 }
1003 }
1004
1005 void MacroAssembler::move_ptr(VMRegPair src, VMRegPair dst) {
1006 if (src.first()->is_stack()) {
1007 if (dst.first()->is_stack()) {
1008 // stack to stack
1009 movq(rax, Address(rbp, reg2offset_in(src.first())));
1010 movq(Address(rsp, reg2offset_out(dst.first())), rax);
1011 } else {
1012 // stack to reg
1013 movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first())));
1014 }
1015 } else if (dst.first()->is_stack()) {
1016 // reg to stack
1017 movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register());
1018 } else {
1019 if (dst.first() != src.first()) {
1020 movq(dst.first()->as_Register(), src.first()->as_Register());
1021 }
1022 }
1023 }
1024
1025 // An oop arg. Must pass a handle not the oop itself
1026 void MacroAssembler::object_move(OopMap* map,
1027 int oop_handle_offset,
1028 int framesize_in_slots,
1029 VMRegPair src,
1030 VMRegPair dst,
1031 bool is_receiver,
1032 int* receiver_offset) {
1033
1034 // must pass a handle. First figure out the location we use as a handle
1035
1036 Register rHandle = dst.first()->is_stack() ? rax : dst.first()->as_Register();
1037
1038 // See if oop is null if it is we need no handle
1039
1040 if (src.first()->is_stack()) {
1041
1042 // Oop is already on the stack as an argument
1043 int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
1044 map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
1045 if (is_receiver) {
1046 *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
1047 }
1048
1049 cmpptr(Address(rbp, reg2offset_in(src.first())), NULL_WORD);
1050 lea(rHandle, Address(rbp, reg2offset_in(src.first())));
1051 // conditionally move a null
1052 cmovptr(Assembler::equal, rHandle, Address(rbp, reg2offset_in(src.first())));
1053 } else {
1054
1055 // Oop is in a register we must store it to the space we reserve
1056 // on the stack for oop_handles and pass a handle if oop is non-null
1057
1058 const Register rOop = src.first()->as_Register();
1059 int oop_slot;
1060 if (rOop == j_rarg0)
1061 oop_slot = 0;
1062 else if (rOop == j_rarg1)
1063 oop_slot = 1;
1064 else if (rOop == j_rarg2)
1065 oop_slot = 2;
1066 else if (rOop == j_rarg3)
1067 oop_slot = 3;
1068 else if (rOop == j_rarg4)
1069 oop_slot = 4;
1070 else {
1071 assert(rOop == j_rarg5, "wrong register");
1072 oop_slot = 5;
1073 }
1074
1075 oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
1076 int offset = oop_slot*VMRegImpl::stack_slot_size;
1077
1078 map->set_oop(VMRegImpl::stack2reg(oop_slot));
1079 // Store oop in handle area, may be null
1080 movptr(Address(rsp, offset), rOop);
1081 if (is_receiver) {
1082 *receiver_offset = offset;
1083 }
1084
1085 cmpptr(rOop, NULL_WORD);
1086 lea(rHandle, Address(rsp, offset));
1087 // conditionally move a null from the handle area where it was just stored
1088 cmovptr(Assembler::equal, rHandle, Address(rsp, offset));
1089 }
1090
1091 // If arg is on the stack then place it otherwise it is already in correct reg.
1092 if (dst.first()->is_stack()) {
1093 movptr(Address(rsp, reg2offset_out(dst.first())), rHandle);
1094 }
1095 }
1096
1097 #endif // _LP64
1098
1099 // Now versions that are common to 32/64 bit
1100
1101 void MacroAssembler::addptr(Register dst, int32_t imm32) {
1102 LP64_ONLY(addq(dst, imm32)) NOT_LP64(addl(dst, imm32));
1103 }
1104
1105 void MacroAssembler::addptr(Register dst, Register src) {
1106 LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src));
1107 }
1108
1109 void MacroAssembler::addptr(Address dst, Register src) {
1110 LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src));
1111 }
1112
1113 void MacroAssembler::addsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
1114 assert(rscratch != noreg || always_reachable(src), "missing");
1115
1116 if (reachable(src)) {
1117 Assembler::addsd(dst, as_Address(src));
1118 } else {
1119 lea(rscratch, src);
1120 Assembler::addsd(dst, Address(rscratch, 0));
1121 }
1122 }
1123
1124 void MacroAssembler::addss(XMMRegister dst, AddressLiteral src, Register rscratch) {
1125 assert(rscratch != noreg || always_reachable(src), "missing");
1126
1127 if (reachable(src)) {
1128 addss(dst, as_Address(src));
1129 } else {
1130 lea(rscratch, src);
1131 addss(dst, Address(rscratch, 0));
1132 }
1133 }
1134
1135 void MacroAssembler::addpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
1136 assert(rscratch != noreg || always_reachable(src), "missing");
1137
1138 if (reachable(src)) {
1139 Assembler::addpd(dst, as_Address(src));
1140 } else {
1141 lea(rscratch, src);
1142 Assembler::addpd(dst, Address(rscratch, 0));
1143 }
1144 }
1145
1146 // See 8273459. Function for ensuring 64-byte alignment, intended for stubs only.
1147 // Stub code is generated once and never copied.
1148 // NMethods can't use this because they get copied and we can't force alignment > 32 bytes.
1149 void MacroAssembler::align64() {
1150 align(64, (unsigned long long) pc());
1151 }
1152
1153 void MacroAssembler::align32() {
1154 align(32, (unsigned long long) pc());
1155 }
1156
1157 void MacroAssembler::align(int modulus) {
1158 // 8273459: Ensure alignment is possible with current segment alignment
1159 assert(modulus <= CodeEntryAlignment, "Alignment must be <= CodeEntryAlignment");
1160 align(modulus, offset());
1161 }
1162
1163 void MacroAssembler::align(int modulus, int target) {
1164 if (target % modulus != 0) {
1165 nop(modulus - (target % modulus));
1166 }
1167 }
1168
1169 void MacroAssembler::push_f(XMMRegister r) {
1170 subptr(rsp, wordSize);
1171 movflt(Address(rsp, 0), r);
1172 }
1173
1174 void MacroAssembler::pop_f(XMMRegister r) {
1175 movflt(r, Address(rsp, 0));
1176 addptr(rsp, wordSize);
1177 }
1178
1179 void MacroAssembler::push_d(XMMRegister r) {
1180 subptr(rsp, 2 * wordSize);
1181 movdbl(Address(rsp, 0), r);
1182 }
1183
1184 void MacroAssembler::pop_d(XMMRegister r) {
1185 movdbl(r, Address(rsp, 0));
1186 addptr(rsp, 2 * Interpreter::stackElementSize);
1187 }
1188
1189 void MacroAssembler::andpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
1190 // Used in sign-masking with aligned address.
1191 assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
1192 assert(rscratch != noreg || always_reachable(src), "missing");
1193
1194 if (reachable(src)) {
1195 Assembler::andpd(dst, as_Address(src));
1196 } else {
1197 lea(rscratch, src);
1198 Assembler::andpd(dst, Address(rscratch, 0));
1199 }
1200 }
1201
1202 void MacroAssembler::andps(XMMRegister dst, AddressLiteral src, Register rscratch) {
1203 // Used in sign-masking with aligned address.
1204 assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
1205 assert(rscratch != noreg || always_reachable(src), "missing");
1206
1207 if (reachable(src)) {
1208 Assembler::andps(dst, as_Address(src));
1209 } else {
1210 lea(rscratch, src);
1211 Assembler::andps(dst, Address(rscratch, 0));
1212 }
1213 }
1214
1215 void MacroAssembler::andptr(Register dst, int32_t imm32) {
1216 LP64_ONLY(andq(dst, imm32)) NOT_LP64(andl(dst, imm32));
1217 }
1218
1219 #ifdef _LP64
1220 void MacroAssembler::andq(Register dst, AddressLiteral src, Register rscratch) {
1221 assert(rscratch != noreg || always_reachable(src), "missing");
1222
1223 if (reachable(src)) {
1224 andq(dst, as_Address(src));
1225 } else {
1226 lea(rscratch, src);
1227 andq(dst, Address(rscratch, 0));
1228 }
1229 }
1230 #endif
1231
1232 void MacroAssembler::atomic_incl(Address counter_addr) {
1233 lock();
1234 incrementl(counter_addr);
1235 }
1236
1237 void MacroAssembler::atomic_incl(AddressLiteral counter_addr, Register rscratch) {
1238 assert(rscratch != noreg || always_reachable(counter_addr), "missing");
1239
1240 if (reachable(counter_addr)) {
1241 atomic_incl(as_Address(counter_addr));
1242 } else {
1243 lea(rscratch, counter_addr);
1244 atomic_incl(Address(rscratch, 0));
1245 }
1246 }
1247
1248 #ifdef _LP64
1249 void MacroAssembler::atomic_incq(Address counter_addr) {
1250 lock();
1251 incrementq(counter_addr);
1252 }
1253
1254 void MacroAssembler::atomic_incq(AddressLiteral counter_addr, Register rscratch) {
1255 assert(rscratch != noreg || always_reachable(counter_addr), "missing");
1256
1257 if (reachable(counter_addr)) {
1258 atomic_incq(as_Address(counter_addr));
1259 } else {
1260 lea(rscratch, counter_addr);
1261 atomic_incq(Address(rscratch, 0));
1262 }
1263 }
1264 #endif
1265
1266 // Writes to stack successive pages until offset reached to check for
1267 // stack overflow + shadow pages. This clobbers tmp.
1268 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
1269 movptr(tmp, rsp);
1270 // Bang stack for total size given plus shadow page size.
1271 // Bang one page at a time because large size can bang beyond yellow and
1272 // red zones.
1273 Label loop;
1274 bind(loop);
1275 movl(Address(tmp, (-(int)os::vm_page_size())), size );
1276 subptr(tmp, (int)os::vm_page_size());
1277 subl(size, (int)os::vm_page_size());
1278 jcc(Assembler::greater, loop);
1279
1280 // Bang down shadow pages too.
1281 // At this point, (tmp-0) is the last address touched, so don't
1282 // touch it again. (It was touched as (tmp-pagesize) but then tmp
1283 // was post-decremented.) Skip this address by starting at i=1, and
1284 // touch a few more pages below. N.B. It is important to touch all
1285 // the way down including all pages in the shadow zone.
1286 for (int i = 1; i < ((int)StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()); i++) {
1287 // this could be any sized move but this is can be a debugging crumb
1288 // so the bigger the better.
1289 movptr(Address(tmp, (-i*(int)os::vm_page_size())), size );
1290 }
1291 }
1292
1293 void MacroAssembler::reserved_stack_check() {
1294 // testing if reserved zone needs to be enabled
1295 Label no_reserved_zone_enabling;
1296 Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
1297 NOT_LP64(get_thread(rsi);)
1298
1299 cmpptr(rsp, Address(thread, JavaThread::reserved_stack_activation_offset()));
1300 jcc(Assembler::below, no_reserved_zone_enabling);
1301
1302 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), thread);
1303 jump(RuntimeAddress(StubRoutines::throw_delayed_StackOverflowError_entry()));
1304 should_not_reach_here();
1305
1306 bind(no_reserved_zone_enabling);
1307 }
1308
1309 void MacroAssembler::c2bool(Register x) {
1310 // implements x == 0 ? 0 : 1
1311 // note: must only look at least-significant byte of x
1312 // since C-style booleans are stored in one byte
1313 // only! (was bug)
1314 andl(x, 0xFF);
1315 setb(Assembler::notZero, x);
1316 }
1317
1318 // Wouldn't need if AddressLiteral version had new name
1319 void MacroAssembler::call(Label& L, relocInfo::relocType rtype) {
1320 Assembler::call(L, rtype);
1321 }
1322
1323 void MacroAssembler::call(Register entry) {
1324 Assembler::call(entry);
1325 }
1326
1327 void MacroAssembler::call(AddressLiteral entry, Register rscratch) {
1328 assert(rscratch != noreg || always_reachable(entry), "missing");
1329
1330 if (reachable(entry)) {
1331 Assembler::call_literal(entry.target(), entry.rspec());
1332 } else {
1333 lea(rscratch, entry);
1334 Assembler::call(rscratch);
1335 }
1336 }
1337
1338 void MacroAssembler::ic_call(address entry, jint method_index) {
1339 RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
1340 movptr(rax, (intptr_t)Universe::non_oop_word());
1341 call(AddressLiteral(entry, rh));
1342 }
1343
1344 void MacroAssembler::emit_static_call_stub() {
1345 // Static stub relocation also tags the Method* in the code-stream.
1346 mov_metadata(rbx, (Metadata*) nullptr); // Method is zapped till fixup time.
1347 // This is recognized as unresolved by relocs/nativeinst/ic code.
1348 jump(RuntimeAddress(pc()));
1349 }
1350
1351 // Implementation of call_VM versions
1352
1353 void MacroAssembler::call_VM(Register oop_result,
1354 address entry_point,
1355 bool check_exceptions) {
1356 Label C, E;
1357 call(C, relocInfo::none);
1358 jmp(E);
1359
1360 bind(C);
1361 call_VM_helper(oop_result, entry_point, 0, check_exceptions);
1362 ret(0);
1363
1364 bind(E);
1365 }
1366
1367 void MacroAssembler::call_VM(Register oop_result,
1368 address entry_point,
1369 Register arg_1,
1370 bool check_exceptions) {
1371 Label C, E;
1372 call(C, relocInfo::none);
1373 jmp(E);
1374
1375 bind(C);
1376 pass_arg1(this, arg_1);
1377 call_VM_helper(oop_result, entry_point, 1, check_exceptions);
1378 ret(0);
1379
1380 bind(E);
1381 }
1382
1383 void MacroAssembler::call_VM(Register oop_result,
1384 address entry_point,
1385 Register arg_1,
1386 Register arg_2,
1387 bool check_exceptions) {
1388 Label C, E;
1389 call(C, relocInfo::none);
1390 jmp(E);
1391
1392 bind(C);
1393
1394 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
1395
1396 pass_arg2(this, arg_2);
1397 pass_arg1(this, arg_1);
1398 call_VM_helper(oop_result, entry_point, 2, check_exceptions);
1399 ret(0);
1400
1401 bind(E);
1402 }
1403
1404 void MacroAssembler::call_VM(Register oop_result,
1405 address entry_point,
1406 Register arg_1,
1407 Register arg_2,
1408 Register arg_3,
1409 bool check_exceptions) {
1410 Label C, E;
1411 call(C, relocInfo::none);
1412 jmp(E);
1413
1414 bind(C);
1415
1416 LP64_ONLY(assert(arg_1 != c_rarg3, "smashed arg"));
1417 LP64_ONLY(assert(arg_2 != c_rarg3, "smashed arg"));
1418 pass_arg3(this, arg_3);
1419
1420 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
1421 pass_arg2(this, arg_2);
1422
1423 pass_arg1(this, arg_1);
1424 call_VM_helper(oop_result, entry_point, 3, check_exceptions);
1425 ret(0);
1426
1427 bind(E);
1428 }
1429
1430 void MacroAssembler::call_VM(Register oop_result,
1431 Register last_java_sp,
1432 address entry_point,
1433 int number_of_arguments,
1434 bool check_exceptions) {
1435 Register thread = LP64_ONLY(r15_thread) NOT_LP64(noreg);
1436 call_VM_base(oop_result, thread, last_java_sp, entry_point, number_of_arguments, check_exceptions);
1437 }
1438
1439 void MacroAssembler::call_VM(Register oop_result,
1440 Register last_java_sp,
1441 address entry_point,
1442 Register arg_1,
1443 bool check_exceptions) {
1444 pass_arg1(this, arg_1);
1445 call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
1446 }
1447
1448 void MacroAssembler::call_VM(Register oop_result,
1449 Register last_java_sp,
1450 address entry_point,
1451 Register arg_1,
1452 Register arg_2,
1453 bool check_exceptions) {
1454
1455 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
1456 pass_arg2(this, arg_2);
1457 pass_arg1(this, arg_1);
1458 call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
1459 }
1460
1461 void MacroAssembler::call_VM(Register oop_result,
1462 Register last_java_sp,
1463 address entry_point,
1464 Register arg_1,
1465 Register arg_2,
1466 Register arg_3,
1467 bool check_exceptions) {
1468 LP64_ONLY(assert(arg_1 != c_rarg3, "smashed arg"));
1469 LP64_ONLY(assert(arg_2 != c_rarg3, "smashed arg"));
1470 pass_arg3(this, arg_3);
1471 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
1472 pass_arg2(this, arg_2);
1473 pass_arg1(this, arg_1);
1474 call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
1475 }
1476
1477 void MacroAssembler::super_call_VM(Register oop_result,
1478 Register last_java_sp,
1479 address entry_point,
1480 int number_of_arguments,
1481 bool check_exceptions) {
1482 Register thread = LP64_ONLY(r15_thread) NOT_LP64(noreg);
1483 MacroAssembler::call_VM_base(oop_result, thread, last_java_sp, entry_point, number_of_arguments, check_exceptions);
1484 }
1485
1486 void MacroAssembler::super_call_VM(Register oop_result,
1487 Register last_java_sp,
1488 address entry_point,
1489 Register arg_1,
1490 bool check_exceptions) {
1491 pass_arg1(this, arg_1);
1492 super_call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
1493 }
1494
1495 void MacroAssembler::super_call_VM(Register oop_result,
1496 Register last_java_sp,
1497 address entry_point,
1498 Register arg_1,
1499 Register arg_2,
1500 bool check_exceptions) {
1501
1502 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
1503 pass_arg2(this, arg_2);
1504 pass_arg1(this, arg_1);
1505 super_call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
1506 }
1507
1508 void MacroAssembler::super_call_VM(Register oop_result,
1509 Register last_java_sp,
1510 address entry_point,
1511 Register arg_1,
1512 Register arg_2,
1513 Register arg_3,
1514 bool check_exceptions) {
1515 LP64_ONLY(assert(arg_1 != c_rarg3, "smashed arg"));
1516 LP64_ONLY(assert(arg_2 != c_rarg3, "smashed arg"));
1517 pass_arg3(this, arg_3);
1518 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
1519 pass_arg2(this, arg_2);
1520 pass_arg1(this, arg_1);
1521 super_call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
1522 }
1523
1524 void MacroAssembler::call_VM_base(Register oop_result,
1525 Register java_thread,
1526 Register last_java_sp,
1527 address entry_point,
1528 int number_of_arguments,
1529 bool check_exceptions) {
1530 // determine java_thread register
1531 if (!java_thread->is_valid()) {
1532 #ifdef _LP64
1533 java_thread = r15_thread;
1534 #else
1535 java_thread = rdi;
1536 get_thread(java_thread);
1537 #endif // LP64
1538 }
1539 // determine last_java_sp register
1540 if (!last_java_sp->is_valid()) {
1541 last_java_sp = rsp;
1542 }
1543 // debugging support
1544 assert(number_of_arguments >= 0 , "cannot have negative number of arguments");
1545 LP64_ONLY(assert(java_thread == r15_thread, "unexpected register"));
1546 #ifdef ASSERT
1547 // TraceBytecodes does not use r12 but saves it over the call, so don't verify
1548 // r12 is the heapbase.
1549 LP64_ONLY(if (UseCompressedOops && !TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");)
1550 #endif // ASSERT
1551
1552 assert(java_thread != oop_result , "cannot use the same register for java_thread & oop_result");
1553 assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
1554
1555 // push java thread (becomes first argument of C function)
1556
1557 NOT_LP64(push(java_thread); number_of_arguments++);
1558 LP64_ONLY(mov(c_rarg0, r15_thread));
1559
1560 // set last Java frame before call
1561 assert(last_java_sp != rbp, "can't use ebp/rbp");
1562
1563 // Only interpreter should have to set fp
1564 set_last_Java_frame(java_thread, last_java_sp, rbp, nullptr, rscratch1);
1565
1566 // do the call, remove parameters
1567 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1568
1569 // restore the thread (cannot use the pushed argument since arguments
1570 // may be overwritten by C code generated by an optimizing compiler);
1571 // however can use the register value directly if it is callee saved.
1572 if (LP64_ONLY(true ||) java_thread == rdi || java_thread == rsi) {
1573 // rdi & rsi (also r15) are callee saved -> nothing to do
1574 #ifdef ASSERT
1575 guarantee(java_thread != rax, "change this code");
1576 push(rax);
1577 { Label L;
1578 get_thread(rax);
1579 cmpptr(java_thread, rax);
1580 jcc(Assembler::equal, L);
1581 STOP("MacroAssembler::call_VM_base: rdi not callee saved?");
1582 bind(L);
1583 }
1584 pop(rax);
1585 #endif
1586 } else {
1587 get_thread(java_thread);
1588 }
1589 // reset last Java frame
1590 // Only interpreter should have to clear fp
1591 reset_last_Java_frame(java_thread, true);
1592
1593 // C++ interp handles this in the interpreter
1594 check_and_handle_popframe(java_thread);
1595 check_and_handle_earlyret(java_thread);
1596
1597 if (check_exceptions) {
1598 // check for pending exceptions (java_thread is set upon return)
1599 cmpptr(Address(java_thread, Thread::pending_exception_offset()), NULL_WORD);
1600 #ifndef _LP64
1601 jump_cc(Assembler::notEqual,
1602 RuntimeAddress(StubRoutines::forward_exception_entry()));
1603 #else
1604 // This used to conditionally jump to forward_exception however it is
1605 // possible if we relocate that the branch will not reach. So we must jump
1606 // around so we can always reach
1607
1608 Label ok;
1609 jcc(Assembler::equal, ok);
1610 jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
1611 bind(ok);
1612 #endif // LP64
1613 }
1614
1615 // get oop result if there is one and reset the value in the thread
1616 if (oop_result->is_valid()) {
1617 get_vm_result(oop_result, java_thread);
1618 }
1619 }
1620
1621 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
1622
1623 // Calculate the value for last_Java_sp
1624 // somewhat subtle. call_VM does an intermediate call
1625 // which places a return address on the stack just under the
1626 // stack pointer as the user finished with it. This allows
1627 // use to retrieve last_Java_pc from last_Java_sp[-1].
1628 // On 32bit we then have to push additional args on the stack to accomplish
1629 // the actual requested call. On 64bit call_VM only can use register args
1630 // so the only extra space is the return address that call_VM created.
1631 // This hopefully explains the calculations here.
1632
1633 #ifdef _LP64
1634 // We've pushed one address, correct last_Java_sp
1635 lea(rax, Address(rsp, wordSize));
1636 #else
1637 lea(rax, Address(rsp, (1 + number_of_arguments) * wordSize));
1638 #endif // LP64
1639
1640 call_VM_base(oop_result, noreg, rax, entry_point, number_of_arguments, check_exceptions);
1641
1642 }
1643
1644 // Use this method when MacroAssembler version of call_VM_leaf_base() should be called from Interpreter.
1645 void MacroAssembler::call_VM_leaf0(address entry_point) {
1646 MacroAssembler::call_VM_leaf_base(entry_point, 0);
1647 }
1648
1649 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
1650 call_VM_leaf_base(entry_point, number_of_arguments);
1651 }
1652
1653 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
1654 pass_arg0(this, arg_0);
1655 call_VM_leaf(entry_point, 1);
1656 }
1657
1658 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
1659
1660 LP64_ONLY(assert(arg_0 != c_rarg1, "smashed arg"));
1661 pass_arg1(this, arg_1);
1662 pass_arg0(this, arg_0);
1663 call_VM_leaf(entry_point, 2);
1664 }
1665
1666 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
1667 LP64_ONLY(assert(arg_0 != c_rarg2, "smashed arg"));
1668 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
1669 pass_arg2(this, arg_2);
1670 LP64_ONLY(assert(arg_0 != c_rarg1, "smashed arg"));
1671 pass_arg1(this, arg_1);
1672 pass_arg0(this, arg_0);
1673 call_VM_leaf(entry_point, 3);
1674 }
1675
1676 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
1677 LP64_ONLY(assert(arg_0 != c_rarg3, "smashed arg"));
1678 LP64_ONLY(assert(arg_1 != c_rarg3, "smashed arg"));
1679 LP64_ONLY(assert(arg_2 != c_rarg3, "smashed arg"));
1680 pass_arg3(this, arg_3);
1681 LP64_ONLY(assert(arg_0 != c_rarg2, "smashed arg"));
1682 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
1683 pass_arg2(this, arg_2);
1684 LP64_ONLY(assert(arg_0 != c_rarg1, "smashed arg"));
1685 pass_arg1(this, arg_1);
1686 pass_arg0(this, arg_0);
1687 call_VM_leaf(entry_point, 3);
1688 }
1689
1690 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
1691 pass_arg0(this, arg_0);
1692 MacroAssembler::call_VM_leaf_base(entry_point, 1);
1693 }
1694
1695 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
1696
1697 LP64_ONLY(assert(arg_0 != c_rarg1, "smashed arg"));
1698 pass_arg1(this, arg_1);
1699 pass_arg0(this, arg_0);
1700 MacroAssembler::call_VM_leaf_base(entry_point, 2);
1701 }
1702
1703 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
1704 LP64_ONLY(assert(arg_0 != c_rarg2, "smashed arg"));
1705 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
1706 pass_arg2(this, arg_2);
1707 LP64_ONLY(assert(arg_0 != c_rarg1, "smashed arg"));
1708 pass_arg1(this, arg_1);
1709 pass_arg0(this, arg_0);
1710 MacroAssembler::call_VM_leaf_base(entry_point, 3);
1711 }
1712
1713 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
1714 LP64_ONLY(assert(arg_0 != c_rarg3, "smashed arg"));
1715 LP64_ONLY(assert(arg_1 != c_rarg3, "smashed arg"));
1716 LP64_ONLY(assert(arg_2 != c_rarg3, "smashed arg"));
1717 pass_arg3(this, arg_3);
1718 LP64_ONLY(assert(arg_0 != c_rarg2, "smashed arg"));
1719 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg"));
1720 pass_arg2(this, arg_2);
1721 LP64_ONLY(assert(arg_0 != c_rarg1, "smashed arg"));
1722 pass_arg1(this, arg_1);
1723 pass_arg0(this, arg_0);
1724 MacroAssembler::call_VM_leaf_base(entry_point, 4);
1725 }
1726
1727 void MacroAssembler::get_vm_result(Register oop_result, Register java_thread) {
1728 movptr(oop_result, Address(java_thread, JavaThread::vm_result_offset()));
1729 movptr(Address(java_thread, JavaThread::vm_result_offset()), NULL_WORD);
1730 verify_oop_msg(oop_result, "broken oop in call_VM_base");
1731 }
1732
1733 void MacroAssembler::get_vm_result_2(Register metadata_result, Register java_thread) {
1734 movptr(metadata_result, Address(java_thread, JavaThread::vm_result_2_offset()));
1735 movptr(Address(java_thread, JavaThread::vm_result_2_offset()), NULL_WORD);
1736 }
1737
1738 void MacroAssembler::check_and_handle_earlyret(Register java_thread) {
1739 }
1740
1741 void MacroAssembler::check_and_handle_popframe(Register java_thread) {
1742 }
1743
1744 void MacroAssembler::cmp32(AddressLiteral src1, int32_t imm, Register rscratch) {
1745 assert(rscratch != noreg || always_reachable(src1), "missing");
1746
1747 if (reachable(src1)) {
1748 cmpl(as_Address(src1), imm);
1749 } else {
1750 lea(rscratch, src1);
1751 cmpl(Address(rscratch, 0), imm);
1752 }
1753 }
1754
1755 void MacroAssembler::cmp32(Register src1, AddressLiteral src2, Register rscratch) {
1756 assert(!src2.is_lval(), "use cmpptr");
1757 assert(rscratch != noreg || always_reachable(src2), "missing");
1758
1759 if (reachable(src2)) {
1760 cmpl(src1, as_Address(src2));
1761 } else {
1762 lea(rscratch, src2);
1763 cmpl(src1, Address(rscratch, 0));
1764 }
1765 }
1766
1767 void MacroAssembler::cmp32(Register src1, int32_t imm) {
1768 Assembler::cmpl(src1, imm);
1769 }
1770
1771 void MacroAssembler::cmp32(Register src1, Address src2) {
1772 Assembler::cmpl(src1, src2);
1773 }
1774
1775 void MacroAssembler::cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
1776 ucomisd(opr1, opr2);
1777
1778 Label L;
1779 if (unordered_is_less) {
1780 movl(dst, -1);
1781 jcc(Assembler::parity, L);
1782 jcc(Assembler::below , L);
1783 movl(dst, 0);
1784 jcc(Assembler::equal , L);
1785 increment(dst);
1786 } else { // unordered is greater
1787 movl(dst, 1);
1788 jcc(Assembler::parity, L);
1789 jcc(Assembler::above , L);
1790 movl(dst, 0);
1791 jcc(Assembler::equal , L);
1792 decrementl(dst);
1793 }
1794 bind(L);
1795 }
1796
1797 void MacroAssembler::cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
1798 ucomiss(opr1, opr2);
1799
1800 Label L;
1801 if (unordered_is_less) {
1802 movl(dst, -1);
1803 jcc(Assembler::parity, L);
1804 jcc(Assembler::below , L);
1805 movl(dst, 0);
1806 jcc(Assembler::equal , L);
1807 increment(dst);
1808 } else { // unordered is greater
1809 movl(dst, 1);
1810 jcc(Assembler::parity, L);
1811 jcc(Assembler::above , L);
1812 movl(dst, 0);
1813 jcc(Assembler::equal , L);
1814 decrementl(dst);
1815 }
1816 bind(L);
1817 }
1818
1819
1820 void MacroAssembler::cmp8(AddressLiteral src1, int imm, Register rscratch) {
1821 assert(rscratch != noreg || always_reachable(src1), "missing");
1822
1823 if (reachable(src1)) {
1824 cmpb(as_Address(src1), imm);
1825 } else {
1826 lea(rscratch, src1);
1827 cmpb(Address(rscratch, 0), imm);
1828 }
1829 }
1830
1831 void MacroAssembler::cmpptr(Register src1, AddressLiteral src2, Register rscratch) {
1832 #ifdef _LP64
1833 assert(rscratch != noreg || always_reachable(src2), "missing");
1834
1835 if (src2.is_lval()) {
1836 movptr(rscratch, src2);
1837 Assembler::cmpq(src1, rscratch);
1838 } else if (reachable(src2)) {
1839 cmpq(src1, as_Address(src2));
1840 } else {
1841 lea(rscratch, src2);
1842 Assembler::cmpq(src1, Address(rscratch, 0));
1843 }
1844 #else
1845 assert(rscratch == noreg, "not needed");
1846 if (src2.is_lval()) {
1847 cmp_literal32(src1, (int32_t)src2.target(), src2.rspec());
1848 } else {
1849 cmpl(src1, as_Address(src2));
1850 }
1851 #endif // _LP64
1852 }
1853
1854 void MacroAssembler::cmpptr(Address src1, AddressLiteral src2, Register rscratch) {
1855 assert(src2.is_lval(), "not a mem-mem compare");
1856 #ifdef _LP64
1857 // moves src2's literal address
1858 movptr(rscratch, src2);
1859 Assembler::cmpq(src1, rscratch);
1860 #else
1861 assert(rscratch == noreg, "not needed");
1862 cmp_literal32(src1, (int32_t)src2.target(), src2.rspec());
1863 #endif // _LP64
1864 }
1865
1866 void MacroAssembler::cmpoop(Register src1, Register src2) {
1867 cmpptr(src1, src2);
1868 }
1869
1870 void MacroAssembler::cmpoop(Register src1, Address src2) {
1871 cmpptr(src1, src2);
1872 }
1873
1874 #ifdef _LP64
1875 void MacroAssembler::cmpoop(Register src1, jobject src2, Register rscratch) {
1876 movoop(rscratch, src2);
1877 cmpptr(src1, rscratch);
1878 }
1879 #endif
1880
1881 void MacroAssembler::locked_cmpxchgptr(Register reg, AddressLiteral adr, Register rscratch) {
1882 assert(rscratch != noreg || always_reachable(adr), "missing");
1883
1884 if (reachable(adr)) {
1885 lock();
1886 cmpxchgptr(reg, as_Address(adr));
1887 } else {
1888 lea(rscratch, adr);
1889 lock();
1890 cmpxchgptr(reg, Address(rscratch, 0));
1891 }
1892 }
1893
1894 void MacroAssembler::cmpxchgptr(Register reg, Address adr) {
1895 LP64_ONLY(cmpxchgq(reg, adr)) NOT_LP64(cmpxchgl(reg, adr));
1896 }
1897
1898 void MacroAssembler::comisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
1899 assert(rscratch != noreg || always_reachable(src), "missing");
1900
1901 if (reachable(src)) {
1902 Assembler::comisd(dst, as_Address(src));
1903 } else {
1904 lea(rscratch, src);
1905 Assembler::comisd(dst, Address(rscratch, 0));
1906 }
1907 }
1908
1909 void MacroAssembler::comiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
1910 assert(rscratch != noreg || always_reachable(src), "missing");
1911
1912 if (reachable(src)) {
1913 Assembler::comiss(dst, as_Address(src));
1914 } else {
1915 lea(rscratch, src);
1916 Assembler::comiss(dst, Address(rscratch, 0));
1917 }
1918 }
1919
1920
1921 void MacroAssembler::cond_inc32(Condition cond, AddressLiteral counter_addr, Register rscratch) {
1922 assert(rscratch != noreg || always_reachable(counter_addr), "missing");
1923
1924 Condition negated_cond = negate_condition(cond);
1925 Label L;
1926 jcc(negated_cond, L);
1927 pushf(); // Preserve flags
1928 atomic_incl(counter_addr, rscratch);
1929 popf();
1930 bind(L);
1931 }
1932
1933 int MacroAssembler::corrected_idivl(Register reg) {
1934 // Full implementation of Java idiv and irem; checks for
1935 // special case as described in JVM spec., p.243 & p.271.
1936 // The function returns the (pc) offset of the idivl
1937 // instruction - may be needed for implicit exceptions.
1938 //
1939 // normal case special case
1940 //
1941 // input : rax,: dividend min_int
1942 // reg: divisor (may not be rax,/rdx) -1
1943 //
1944 // output: rax,: quotient (= rax, idiv reg) min_int
1945 // rdx: remainder (= rax, irem reg) 0
1946 assert(reg != rax && reg != rdx, "reg cannot be rax, or rdx register");
1947 const int min_int = 0x80000000;
1948 Label normal_case, special_case;
1949
1950 // check for special case
1951 cmpl(rax, min_int);
1952 jcc(Assembler::notEqual, normal_case);
1953 xorl(rdx, rdx); // prepare rdx for possible special case (where remainder = 0)
1954 cmpl(reg, -1);
1955 jcc(Assembler::equal, special_case);
1956
1957 // handle normal case
1958 bind(normal_case);
1959 cdql();
1960 int idivl_offset = offset();
1961 idivl(reg);
1962
1963 // normal and special case exit
1964 bind(special_case);
1965
1966 return idivl_offset;
1967 }
1968
1969
1970
1971 void MacroAssembler::decrementl(Register reg, int value) {
1972 if (value == min_jint) {subl(reg, value) ; return; }
1973 if (value < 0) { incrementl(reg, -value); return; }
1974 if (value == 0) { ; return; }
1975 if (value == 1 && UseIncDec) { decl(reg) ; return; }
1976 /* else */ { subl(reg, value) ; return; }
1977 }
1978
1979 void MacroAssembler::decrementl(Address dst, int value) {
1980 if (value == min_jint) {subl(dst, value) ; return; }
1981 if (value < 0) { incrementl(dst, -value); return; }
1982 if (value == 0) { ; return; }
1983 if (value == 1 && UseIncDec) { decl(dst) ; return; }
1984 /* else */ { subl(dst, value) ; return; }
1985 }
1986
1987 void MacroAssembler::division_with_shift (Register reg, int shift_value) {
1988 assert(shift_value > 0, "illegal shift value");
1989 Label _is_positive;
1990 testl (reg, reg);
1991 jcc (Assembler::positive, _is_positive);
1992 int offset = (1 << shift_value) - 1 ;
1993
1994 if (offset == 1) {
1995 incrementl(reg);
1996 } else {
1997 addl(reg, offset);
1998 }
1999
2000 bind (_is_positive);
2001 sarl(reg, shift_value);
2002 }
2003
2004 void MacroAssembler::divsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
2005 assert(rscratch != noreg || always_reachable(src), "missing");
2006
2007 if (reachable(src)) {
2008 Assembler::divsd(dst, as_Address(src));
2009 } else {
2010 lea(rscratch, src);
2011 Assembler::divsd(dst, Address(rscratch, 0));
2012 }
2013 }
2014
2015 void MacroAssembler::divss(XMMRegister dst, AddressLiteral src, Register rscratch) {
2016 assert(rscratch != noreg || always_reachable(src), "missing");
2017
2018 if (reachable(src)) {
2019 Assembler::divss(dst, as_Address(src));
2020 } else {
2021 lea(rscratch, src);
2022 Assembler::divss(dst, Address(rscratch, 0));
2023 }
2024 }
2025
2026 void MacroAssembler::enter() {
2027 push(rbp);
2028 mov(rbp, rsp);
2029 }
2030
2031 void MacroAssembler::post_call_nop() {
2032 if (!Continuations::enabled()) {
2033 return;
2034 }
2035 InstructionMark im(this);
2036 relocate(post_call_nop_Relocation::spec());
2037 InlineSkippedInstructionsCounter skipCounter(this);
2038 emit_int8((int8_t)0x0f);
2039 emit_int8((int8_t)0x1f);
2040 emit_int8((int8_t)0x84);
2041 emit_int8((int8_t)0x00);
2042 emit_int32(0x00);
2043 }
2044
2045 // A 5 byte nop that is safe for patching (see patch_verified_entry)
2046 void MacroAssembler::fat_nop() {
2047 if (UseAddressNop) {
2048 addr_nop_5();
2049 } else {
2050 emit_int8((int8_t)0x26); // es:
2051 emit_int8((int8_t)0x2e); // cs:
2052 emit_int8((int8_t)0x64); // fs:
2053 emit_int8((int8_t)0x65); // gs:
2054 emit_int8((int8_t)0x90);
2055 }
2056 }
2057
2058 #ifndef _LP64
2059 void MacroAssembler::fcmp(Register tmp) {
2060 fcmp(tmp, 1, true, true);
2061 }
2062
2063 void MacroAssembler::fcmp(Register tmp, int index, bool pop_left, bool pop_right) {
2064 assert(!pop_right || pop_left, "usage error");
2065 if (VM_Version::supports_cmov()) {
2066 assert(tmp == noreg, "unneeded temp");
2067 if (pop_left) {
2068 fucomip(index);
2069 } else {
2070 fucomi(index);
2071 }
2072 if (pop_right) {
2073 fpop();
2074 }
2075 } else {
2076 assert(tmp != noreg, "need temp");
2077 if (pop_left) {
2078 if (pop_right) {
2079 fcompp();
2080 } else {
2081 fcomp(index);
2082 }
2083 } else {
2084 fcom(index);
2085 }
2086 // convert FPU condition into eflags condition via rax,
2087 save_rax(tmp);
2088 fwait(); fnstsw_ax();
2089 sahf();
2090 restore_rax(tmp);
2091 }
2092 // condition codes set as follows:
2093 //
2094 // CF (corresponds to C0) if x < y
2095 // PF (corresponds to C2) if unordered
2096 // ZF (corresponds to C3) if x = y
2097 }
2098
2099 void MacroAssembler::fcmp2int(Register dst, bool unordered_is_less) {
2100 fcmp2int(dst, unordered_is_less, 1, true, true);
2101 }
2102
2103 void MacroAssembler::fcmp2int(Register dst, bool unordered_is_less, int index, bool pop_left, bool pop_right) {
2104 fcmp(VM_Version::supports_cmov() ? noreg : dst, index, pop_left, pop_right);
2105 Label L;
2106 if (unordered_is_less) {
2107 movl(dst, -1);
2108 jcc(Assembler::parity, L);
2109 jcc(Assembler::below , L);
2110 movl(dst, 0);
2111 jcc(Assembler::equal , L);
2112 increment(dst);
2113 } else { // unordered is greater
2114 movl(dst, 1);
2115 jcc(Assembler::parity, L);
2116 jcc(Assembler::above , L);
2117 movl(dst, 0);
2118 jcc(Assembler::equal , L);
2119 decrementl(dst);
2120 }
2121 bind(L);
2122 }
2123
2124 void MacroAssembler::fld_d(AddressLiteral src) {
2125 fld_d(as_Address(src));
2126 }
2127
2128 void MacroAssembler::fld_s(AddressLiteral src) {
2129 fld_s(as_Address(src));
2130 }
2131
2132 void MacroAssembler::fldcw(AddressLiteral src) {
2133 fldcw(as_Address(src));
2134 }
2135
2136 void MacroAssembler::fpop() {
2137 ffree();
2138 fincstp();
2139 }
2140
2141 void MacroAssembler::fremr(Register tmp) {
2142 save_rax(tmp);
2143 { Label L;
2144 bind(L);
2145 fprem();
2146 fwait(); fnstsw_ax();
2147 sahf();
2148 jcc(Assembler::parity, L);
2149 }
2150 restore_rax(tmp);
2151 // Result is in ST0.
2152 // Note: fxch & fpop to get rid of ST1
2153 // (otherwise FPU stack could overflow eventually)
2154 fxch(1);
2155 fpop();
2156 }
2157
2158 void MacroAssembler::empty_FPU_stack() {
2159 if (VM_Version::supports_mmx()) {
2160 emms();
2161 } else {
2162 for (int i = 8; i-- > 0; ) ffree(i);
2163 }
2164 }
2165 #endif // !LP64
2166
2167 void MacroAssembler::mulpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
2168 assert(rscratch != noreg || always_reachable(src), "missing");
2169 if (reachable(src)) {
2170 Assembler::mulpd(dst, as_Address(src));
2171 } else {
2172 lea(rscratch, src);
2173 Assembler::mulpd(dst, Address(rscratch, 0));
2174 }
2175 }
2176
2177 void MacroAssembler::load_float(Address src) {
2178 #ifdef _LP64
2179 movflt(xmm0, src);
2180 #else
2181 if (UseSSE >= 1) {
2182 movflt(xmm0, src);
2183 } else {
2184 fld_s(src);
2185 }
2186 #endif // LP64
2187 }
2188
2189 void MacroAssembler::store_float(Address dst) {
2190 #ifdef _LP64
2191 movflt(dst, xmm0);
2192 #else
2193 if (UseSSE >= 1) {
2194 movflt(dst, xmm0);
2195 } else {
2196 fstp_s(dst);
2197 }
2198 #endif // LP64
2199 }
2200
2201 void MacroAssembler::load_double(Address src) {
2202 #ifdef _LP64
2203 movdbl(xmm0, src);
2204 #else
2205 if (UseSSE >= 2) {
2206 movdbl(xmm0, src);
2207 } else {
2208 fld_d(src);
2209 }
2210 #endif // LP64
2211 }
2212
2213 void MacroAssembler::store_double(Address dst) {
2214 #ifdef _LP64
2215 movdbl(dst, xmm0);
2216 #else
2217 if (UseSSE >= 2) {
2218 movdbl(dst, xmm0);
2219 } else {
2220 fstp_d(dst);
2221 }
2222 #endif // LP64
2223 }
2224
2225 // dst = c = a * b + c
2226 void MacroAssembler::fmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
2227 Assembler::vfmadd231sd(c, a, b);
2228 if (dst != c) {
2229 movdbl(dst, c);
2230 }
2231 }
2232
2233 // dst = c = a * b + c
2234 void MacroAssembler::fmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
2235 Assembler::vfmadd231ss(c, a, b);
2236 if (dst != c) {
2237 movflt(dst, c);
2238 }
2239 }
2240
2241 // dst = c = a * b + c
2242 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
2243 Assembler::vfmadd231pd(c, a, b, vector_len);
2244 if (dst != c) {
2245 vmovdqu(dst, c);
2246 }
2247 }
2248
2249 // dst = c = a * b + c
2250 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
2251 Assembler::vfmadd231ps(c, a, b, vector_len);
2252 if (dst != c) {
2253 vmovdqu(dst, c);
2254 }
2255 }
2256
2257 // dst = c = a * b + c
2258 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
2259 Assembler::vfmadd231pd(c, a, b, vector_len);
2260 if (dst != c) {
2261 vmovdqu(dst, c);
2262 }
2263 }
2264
2265 // dst = c = a * b + c
2266 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
2267 Assembler::vfmadd231ps(c, a, b, vector_len);
2268 if (dst != c) {
2269 vmovdqu(dst, c);
2270 }
2271 }
2272
2273 void MacroAssembler::incrementl(AddressLiteral dst, Register rscratch) {
2274 assert(rscratch != noreg || always_reachable(dst), "missing");
2275
2276 if (reachable(dst)) {
2277 incrementl(as_Address(dst));
2278 } else {
2279 lea(rscratch, dst);
2280 incrementl(Address(rscratch, 0));
2281 }
2282 }
2283
2284 void MacroAssembler::incrementl(ArrayAddress dst, Register rscratch) {
2285 incrementl(as_Address(dst, rscratch));
2286 }
2287
2288 void MacroAssembler::incrementl(Register reg, int value) {
2289 if (value == min_jint) {addl(reg, value) ; return; }
2290 if (value < 0) { decrementl(reg, -value); return; }
2291 if (value == 0) { ; return; }
2292 if (value == 1 && UseIncDec) { incl(reg) ; return; }
2293 /* else */ { addl(reg, value) ; return; }
2294 }
2295
2296 void MacroAssembler::incrementl(Address dst, int value) {
2297 if (value == min_jint) {addl(dst, value) ; return; }
2298 if (value < 0) { decrementl(dst, -value); return; }
2299 if (value == 0) { ; return; }
2300 if (value == 1 && UseIncDec) { incl(dst) ; return; }
2301 /* else */ { addl(dst, value) ; return; }
2302 }
2303
2304 void MacroAssembler::jump(AddressLiteral dst, Register rscratch) {
2305 assert(rscratch != noreg || always_reachable(dst), "missing");
2306
2307 if (reachable(dst)) {
2308 jmp_literal(dst.target(), dst.rspec());
2309 } else {
2310 lea(rscratch, dst);
2311 jmp(rscratch);
2312 }
2313 }
2314
2315 void MacroAssembler::jump_cc(Condition cc, AddressLiteral dst, Register rscratch) {
2316 assert(rscratch != noreg || always_reachable(dst), "missing");
2317
2318 if (reachable(dst)) {
2319 InstructionMark im(this);
2320 relocate(dst.reloc());
2321 const int short_size = 2;
2322 const int long_size = 6;
2323 int offs = (intptr_t)dst.target() - ((intptr_t)pc());
2324 if (dst.reloc() == relocInfo::none && is8bit(offs - short_size)) {
2325 // 0111 tttn #8-bit disp
2326 emit_int8(0x70 | cc);
2327 emit_int8((offs - short_size) & 0xFF);
2328 } else {
2329 // 0000 1111 1000 tttn #32-bit disp
2330 emit_int8(0x0F);
2331 emit_int8((unsigned char)(0x80 | cc));
2332 emit_int32(offs - long_size);
2333 }
2334 } else {
2335 #ifdef ASSERT
2336 warning("reversing conditional branch");
2337 #endif /* ASSERT */
2338 Label skip;
2339 jccb(reverse[cc], skip);
2340 lea(rscratch, dst);
2341 Assembler::jmp(rscratch);
2342 bind(skip);
2343 }
2344 }
2345
2346 void MacroAssembler::ldmxcsr(AddressLiteral src, Register rscratch) {
2347 assert(rscratch != noreg || always_reachable(src), "missing");
2348
2349 if (reachable(src)) {
2350 Assembler::ldmxcsr(as_Address(src));
2351 } else {
2352 lea(rscratch, src);
2353 Assembler::ldmxcsr(Address(rscratch, 0));
2354 }
2355 }
2356
2357 int MacroAssembler::load_signed_byte(Register dst, Address src) {
2358 int off;
2359 if (LP64_ONLY(true ||) VM_Version::is_P6()) {
2360 off = offset();
2361 movsbl(dst, src); // movsxb
2362 } else {
2363 off = load_unsigned_byte(dst, src);
2364 shll(dst, 24);
2365 sarl(dst, 24);
2366 }
2367 return off;
2368 }
2369
2370 // Note: load_signed_short used to be called load_signed_word.
2371 // Although the 'w' in x86 opcodes refers to the term "word" in the assembler
2372 // manual, which means 16 bits, that usage is found nowhere in HotSpot code.
2373 // The term "word" in HotSpot means a 32- or 64-bit machine word.
2374 int MacroAssembler::load_signed_short(Register dst, Address src) {
2375 int off;
2376 if (LP64_ONLY(true ||) VM_Version::is_P6()) {
2377 // This is dubious to me since it seems safe to do a signed 16 => 64 bit
2378 // version but this is what 64bit has always done. This seems to imply
2379 // that users are only using 32bits worth.
2380 off = offset();
2381 movswl(dst, src); // movsxw
2382 } else {
2383 off = load_unsigned_short(dst, src);
2384 shll(dst, 16);
2385 sarl(dst, 16);
2386 }
2387 return off;
2388 }
2389
2390 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
2391 // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
2392 // and "3.9 Partial Register Penalties", p. 22).
2393 int off;
2394 if (LP64_ONLY(true || ) VM_Version::is_P6() || src.uses(dst)) {
2395 off = offset();
2396 movzbl(dst, src); // movzxb
2397 } else {
2398 xorl(dst, dst);
2399 off = offset();
2400 movb(dst, src);
2401 }
2402 return off;
2403 }
2404
2405 // Note: load_unsigned_short used to be called load_unsigned_word.
2406 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
2407 // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
2408 // and "3.9 Partial Register Penalties", p. 22).
2409 int off;
2410 if (LP64_ONLY(true ||) VM_Version::is_P6() || src.uses(dst)) {
2411 off = offset();
2412 movzwl(dst, src); // movzxw
2413 } else {
2414 xorl(dst, dst);
2415 off = offset();
2416 movw(dst, src);
2417 }
2418 return off;
2419 }
2420
2421 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2) {
2422 switch (size_in_bytes) {
2423 #ifndef _LP64
2424 case 8:
2425 assert(dst2 != noreg, "second dest register required");
2426 movl(dst, src);
2427 movl(dst2, src.plus_disp(BytesPerInt));
2428 break;
2429 #else
2430 case 8: movq(dst, src); break;
2431 #endif
2432 case 4: movl(dst, src); break;
2433 case 2: is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
2434 case 1: is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
2435 default: ShouldNotReachHere();
2436 }
2437 }
2438
2439 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2) {
2440 switch (size_in_bytes) {
2441 #ifndef _LP64
2442 case 8:
2443 assert(src2 != noreg, "second source register required");
2444 movl(dst, src);
2445 movl(dst.plus_disp(BytesPerInt), src2);
2446 break;
2447 #else
2448 case 8: movq(dst, src); break;
2449 #endif
2450 case 4: movl(dst, src); break;
2451 case 2: movw(dst, src); break;
2452 case 1: movb(dst, src); break;
2453 default: ShouldNotReachHere();
2454 }
2455 }
2456
2457 void MacroAssembler::mov32(AddressLiteral dst, Register src, Register rscratch) {
2458 assert(rscratch != noreg || always_reachable(dst), "missing");
2459
2460 if (reachable(dst)) {
2461 movl(as_Address(dst), src);
2462 } else {
2463 lea(rscratch, dst);
2464 movl(Address(rscratch, 0), src);
2465 }
2466 }
2467
2468 void MacroAssembler::mov32(Register dst, AddressLiteral src) {
2469 if (reachable(src)) {
2470 movl(dst, as_Address(src));
2471 } else {
2472 lea(dst, src);
2473 movl(dst, Address(dst, 0));
2474 }
2475 }
2476
2477 // C++ bool manipulation
2478
2479 void MacroAssembler::movbool(Register dst, Address src) {
2480 if(sizeof(bool) == 1)
2481 movb(dst, src);
2482 else if(sizeof(bool) == 2)
2483 movw(dst, src);
2484 else if(sizeof(bool) == 4)
2485 movl(dst, src);
2486 else
2487 // unsupported
2488 ShouldNotReachHere();
2489 }
2490
2491 void MacroAssembler::movbool(Address dst, bool boolconst) {
2492 if(sizeof(bool) == 1)
2493 movb(dst, (int) boolconst);
2494 else if(sizeof(bool) == 2)
2495 movw(dst, (int) boolconst);
2496 else if(sizeof(bool) == 4)
2497 movl(dst, (int) boolconst);
2498 else
2499 // unsupported
2500 ShouldNotReachHere();
2501 }
2502
2503 void MacroAssembler::movbool(Address dst, Register src) {
2504 if(sizeof(bool) == 1)
2505 movb(dst, src);
2506 else if(sizeof(bool) == 2)
2507 movw(dst, src);
2508 else if(sizeof(bool) == 4)
2509 movl(dst, src);
2510 else
2511 // unsupported
2512 ShouldNotReachHere();
2513 }
2514
2515 void MacroAssembler::movdl(XMMRegister dst, AddressLiteral src, Register rscratch) {
2516 assert(rscratch != noreg || always_reachable(src), "missing");
2517
2518 if (reachable(src)) {
2519 movdl(dst, as_Address(src));
2520 } else {
2521 lea(rscratch, src);
2522 movdl(dst, Address(rscratch, 0));
2523 }
2524 }
2525
2526 void MacroAssembler::movq(XMMRegister dst, AddressLiteral src, Register rscratch) {
2527 assert(rscratch != noreg || always_reachable(src), "missing");
2528
2529 if (reachable(src)) {
2530 movq(dst, as_Address(src));
2531 } else {
2532 lea(rscratch, src);
2533 movq(dst, Address(rscratch, 0));
2534 }
2535 }
2536
2537 void MacroAssembler::movdbl(XMMRegister dst, AddressLiteral src, Register rscratch) {
2538 assert(rscratch != noreg || always_reachable(src), "missing");
2539
2540 if (reachable(src)) {
2541 if (UseXmmLoadAndClearUpper) {
2542 movsd (dst, as_Address(src));
2543 } else {
2544 movlpd(dst, as_Address(src));
2545 }
2546 } else {
2547 lea(rscratch, src);
2548 if (UseXmmLoadAndClearUpper) {
2549 movsd (dst, Address(rscratch, 0));
2550 } else {
2551 movlpd(dst, Address(rscratch, 0));
2552 }
2553 }
2554 }
2555
2556 void MacroAssembler::movflt(XMMRegister dst, AddressLiteral src, Register rscratch) {
2557 assert(rscratch != noreg || always_reachable(src), "missing");
2558
2559 if (reachable(src)) {
2560 movss(dst, as_Address(src));
2561 } else {
2562 lea(rscratch, src);
2563 movss(dst, Address(rscratch, 0));
2564 }
2565 }
2566
2567 void MacroAssembler::movptr(Register dst, Register src) {
2568 LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src));
2569 }
2570
2571 void MacroAssembler::movptr(Register dst, Address src) {
2572 LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src));
2573 }
2574
2575 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
2576 void MacroAssembler::movptr(Register dst, intptr_t src) {
2577 LP64_ONLY(mov64(dst, src)) NOT_LP64(movl(dst, src));
2578 }
2579
2580 void MacroAssembler::movptr(Address dst, Register src) {
2581 LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src));
2582 }
2583
2584 void MacroAssembler::movptr(Address dst, int32_t src) {
2585 LP64_ONLY(movslq(dst, src)) NOT_LP64(movl(dst, src));
2586 }
2587
2588 void MacroAssembler::movdqu(Address dst, XMMRegister src) {
2589 assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
2590 Assembler::movdqu(dst, src);
2591 }
2592
2593 void MacroAssembler::movdqu(XMMRegister dst, Address src) {
2594 assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
2595 Assembler::movdqu(dst, src);
2596 }
2597
2598 void MacroAssembler::movdqu(XMMRegister dst, XMMRegister src) {
2599 assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
2600 Assembler::movdqu(dst, src);
2601 }
2602
2603 void MacroAssembler::movdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
2604 assert(rscratch != noreg || always_reachable(src), "missing");
2605
2606 if (reachable(src)) {
2607 movdqu(dst, as_Address(src));
2608 } else {
2609 lea(rscratch, src);
2610 movdqu(dst, Address(rscratch, 0));
2611 }
2612 }
2613
2614 void MacroAssembler::vmovdqu(Address dst, XMMRegister src) {
2615 assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
2616 Assembler::vmovdqu(dst, src);
2617 }
2618
2619 void MacroAssembler::vmovdqu(XMMRegister dst, Address src) {
2620 assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
2621 Assembler::vmovdqu(dst, src);
2622 }
2623
2624 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src) {
2625 assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
2626 Assembler::vmovdqu(dst, src);
2627 }
2628
2629 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
2630 assert(rscratch != noreg || always_reachable(src), "missing");
2631
2632 if (reachable(src)) {
2633 vmovdqu(dst, as_Address(src));
2634 }
2635 else {
2636 lea(rscratch, src);
2637 vmovdqu(dst, Address(rscratch, 0));
2638 }
2639 }
2640
2641 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
2642 assert(rscratch != noreg || always_reachable(src), "missing");
2643
2644 if (vector_len == AVX_512bit) {
2645 evmovdquq(dst, src, AVX_512bit, rscratch);
2646 } else if (vector_len == AVX_256bit) {
2647 vmovdqu(dst, src, rscratch);
2648 } else {
2649 movdqu(dst, src, rscratch);
2650 }
2651 }
2652
2653 void MacroAssembler::kmov(KRegister dst, Address src) {
2654 if (VM_Version::supports_avx512bw()) {
2655 kmovql(dst, src);
2656 } else {
2657 assert(VM_Version::supports_evex(), "");
2658 kmovwl(dst, src);
2659 }
2660 }
2661
2662 void MacroAssembler::kmov(Address dst, KRegister src) {
2663 if (VM_Version::supports_avx512bw()) {
2664 kmovql(dst, src);
2665 } else {
2666 assert(VM_Version::supports_evex(), "");
2667 kmovwl(dst, src);
2668 }
2669 }
2670
2671 void MacroAssembler::kmov(KRegister dst, KRegister src) {
2672 if (VM_Version::supports_avx512bw()) {
2673 kmovql(dst, src);
2674 } else {
2675 assert(VM_Version::supports_evex(), "");
2676 kmovwl(dst, src);
2677 }
2678 }
2679
2680 void MacroAssembler::kmov(Register dst, KRegister src) {
2681 if (VM_Version::supports_avx512bw()) {
2682 kmovql(dst, src);
2683 } else {
2684 assert(VM_Version::supports_evex(), "");
2685 kmovwl(dst, src);
2686 }
2687 }
2688
2689 void MacroAssembler::kmov(KRegister dst, Register src) {
2690 if (VM_Version::supports_avx512bw()) {
2691 kmovql(dst, src);
2692 } else {
2693 assert(VM_Version::supports_evex(), "");
2694 kmovwl(dst, src);
2695 }
2696 }
2697
2698 void MacroAssembler::kmovql(KRegister dst, AddressLiteral src, Register rscratch) {
2699 assert(rscratch != noreg || always_reachable(src), "missing");
2700
2701 if (reachable(src)) {
2702 kmovql(dst, as_Address(src));
2703 } else {
2704 lea(rscratch, src);
2705 kmovql(dst, Address(rscratch, 0));
2706 }
2707 }
2708
2709 void MacroAssembler::kmovwl(KRegister dst, AddressLiteral src, Register rscratch) {
2710 assert(rscratch != noreg || always_reachable(src), "missing");
2711
2712 if (reachable(src)) {
2713 kmovwl(dst, as_Address(src));
2714 } else {
2715 lea(rscratch, src);
2716 kmovwl(dst, Address(rscratch, 0));
2717 }
2718 }
2719
2720 void MacroAssembler::evmovdqub(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
2721 int vector_len, Register rscratch) {
2722 assert(rscratch != noreg || always_reachable(src), "missing");
2723
2724 if (reachable(src)) {
2725 Assembler::evmovdqub(dst, mask, as_Address(src), merge, vector_len);
2726 } else {
2727 lea(rscratch, src);
2728 Assembler::evmovdqub(dst, mask, Address(rscratch, 0), merge, vector_len);
2729 }
2730 }
2731
2732 void MacroAssembler::evmovdquw(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
2733 int vector_len, Register rscratch) {
2734 assert(rscratch != noreg || always_reachable(src), "missing");
2735
2736 if (reachable(src)) {
2737 Assembler::evmovdquw(dst, mask, as_Address(src), merge, vector_len);
2738 } else {
2739 lea(rscratch, src);
2740 Assembler::evmovdquw(dst, mask, Address(rscratch, 0), merge, vector_len);
2741 }
2742 }
2743
2744 void MacroAssembler::evmovdqul(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
2745 assert(rscratch != noreg || always_reachable(src), "missing");
2746
2747 if (reachable(src)) {
2748 Assembler::evmovdqul(dst, mask, as_Address(src), merge, vector_len);
2749 } else {
2750 lea(rscratch, src);
2751 Assembler::evmovdqul(dst, mask, Address(rscratch, 0), merge, vector_len);
2752 }
2753 }
2754
2755 void MacroAssembler::evmovdquq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
2756 assert(rscratch != noreg || always_reachable(src), "missing");
2757
2758 if (reachable(src)) {
2759 Assembler::evmovdquq(dst, mask, as_Address(src), merge, vector_len);
2760 } else {
2761 lea(rscratch, src);
2762 Assembler::evmovdquq(dst, mask, Address(rscratch, 0), merge, vector_len);
2763 }
2764 }
2765
2766 void MacroAssembler::evmovdquq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
2767 assert(rscratch != noreg || always_reachable(src), "missing");
2768
2769 if (reachable(src)) {
2770 Assembler::evmovdquq(dst, as_Address(src), vector_len);
2771 } else {
2772 lea(rscratch, src);
2773 Assembler::evmovdquq(dst, Address(rscratch, 0), vector_len);
2774 }
2775 }
2776
2777 void MacroAssembler::movdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
2778 assert(rscratch != noreg || always_reachable(src), "missing");
2779
2780 if (reachable(src)) {
2781 Assembler::movdqa(dst, as_Address(src));
2782 } else {
2783 lea(rscratch, src);
2784 Assembler::movdqa(dst, Address(rscratch, 0));
2785 }
2786 }
2787
2788 void MacroAssembler::movsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
2789 assert(rscratch != noreg || always_reachable(src), "missing");
2790
2791 if (reachable(src)) {
2792 Assembler::movsd(dst, as_Address(src));
2793 } else {
2794 lea(rscratch, src);
2795 Assembler::movsd(dst, Address(rscratch, 0));
2796 }
2797 }
2798
2799 void MacroAssembler::movss(XMMRegister dst, AddressLiteral src, Register rscratch) {
2800 assert(rscratch != noreg || always_reachable(src), "missing");
2801
2802 if (reachable(src)) {
2803 Assembler::movss(dst, as_Address(src));
2804 } else {
2805 lea(rscratch, src);
2806 Assembler::movss(dst, Address(rscratch, 0));
2807 }
2808 }
2809
2810 void MacroAssembler::movddup(XMMRegister dst, AddressLiteral src, Register rscratch) {
2811 assert(rscratch != noreg || always_reachable(src), "missing");
2812
2813 if (reachable(src)) {
2814 Assembler::movddup(dst, as_Address(src));
2815 } else {
2816 lea(rscratch, src);
2817 Assembler::movddup(dst, Address(rscratch, 0));
2818 }
2819 }
2820
2821 void MacroAssembler::vmovddup(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
2822 assert(rscratch != noreg || always_reachable(src), "missing");
2823
2824 if (reachable(src)) {
2825 Assembler::vmovddup(dst, as_Address(src), vector_len);
2826 } else {
2827 lea(rscratch, src);
2828 Assembler::vmovddup(dst, Address(rscratch, 0), vector_len);
2829 }
2830 }
2831
2832 void MacroAssembler::mulsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
2833 assert(rscratch != noreg || always_reachable(src), "missing");
2834
2835 if (reachable(src)) {
2836 Assembler::mulsd(dst, as_Address(src));
2837 } else {
2838 lea(rscratch, src);
2839 Assembler::mulsd(dst, Address(rscratch, 0));
2840 }
2841 }
2842
2843 void MacroAssembler::mulss(XMMRegister dst, AddressLiteral src, Register rscratch) {
2844 assert(rscratch != noreg || always_reachable(src), "missing");
2845
2846 if (reachable(src)) {
2847 Assembler::mulss(dst, as_Address(src));
2848 } else {
2849 lea(rscratch, src);
2850 Assembler::mulss(dst, Address(rscratch, 0));
2851 }
2852 }
2853
2854 void MacroAssembler::null_check(Register reg, int offset) {
2855 if (needs_explicit_null_check(offset)) {
2856 // provoke OS null exception if reg is null by
2857 // accessing M[reg] w/o changing any (non-CC) registers
2858 // NOTE: cmpl is plenty here to provoke a segv
2859 cmpptr(rax, Address(reg, 0));
2860 // Note: should probably use testl(rax, Address(reg, 0));
2861 // may be shorter code (however, this version of
2862 // testl needs to be implemented first)
2863 } else {
2864 // nothing to do, (later) access of M[reg + offset]
2865 // will provoke OS null exception if reg is null
2866 }
2867 }
2868
2869 void MacroAssembler::os_breakpoint() {
2870 // instead of directly emitting a breakpoint, call os:breakpoint for better debugability
2871 // (e.g., MSVC can't call ps() otherwise)
2872 call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint)));
2873 }
2874
2875 void MacroAssembler::unimplemented(const char* what) {
2876 const char* buf = nullptr;
2877 {
2878 ResourceMark rm;
2879 stringStream ss;
2880 ss.print("unimplemented: %s", what);
2881 buf = code_string(ss.as_string());
2882 }
2883 stop(buf);
2884 }
2885
2886 #ifdef _LP64
2887 #define XSTATE_BV 0x200
2888 #endif
2889
2890 void MacroAssembler::pop_CPU_state() {
2891 pop_FPU_state();
2892 pop_IU_state();
2893 }
2894
2895 void MacroAssembler::pop_FPU_state() {
2896 #ifndef _LP64
2897 frstor(Address(rsp, 0));
2898 #else
2899 fxrstor(Address(rsp, 0));
2900 #endif
2901 addptr(rsp, FPUStateSizeInWords * wordSize);
2902 }
2903
2904 void MacroAssembler::pop_IU_state() {
2905 popa();
2906 LP64_ONLY(addq(rsp, 8));
2907 popf();
2908 }
2909
2910 // Save Integer and Float state
2911 // Warning: Stack must be 16 byte aligned (64bit)
2912 void MacroAssembler::push_CPU_state() {
2913 push_IU_state();
2914 push_FPU_state();
2915 }
2916
2917 void MacroAssembler::push_FPU_state() {
2918 subptr(rsp, FPUStateSizeInWords * wordSize);
2919 #ifndef _LP64
2920 fnsave(Address(rsp, 0));
2921 fwait();
2922 #else
2923 fxsave(Address(rsp, 0));
2924 #endif // LP64
2925 }
2926
2927 void MacroAssembler::push_IU_state() {
2928 // Push flags first because pusha kills them
2929 pushf();
2930 // Make sure rsp stays 16-byte aligned
2931 LP64_ONLY(subq(rsp, 8));
2932 pusha();
2933 }
2934
2935 void MacroAssembler::push_cont_fastpath() {
2936 if (!Continuations::enabled()) return;
2937
2938 #ifndef _LP64
2939 Register rthread = rax;
2940 Register rrealsp = rbx;
2941 push(rthread);
2942 push(rrealsp);
2943
2944 get_thread(rthread);
2945
2946 // The code below wants the original RSP.
2947 // Move it back after the pushes above.
2948 movptr(rrealsp, rsp);
2949 addptr(rrealsp, 2*wordSize);
2950 #else
2951 Register rthread = r15_thread;
2952 Register rrealsp = rsp;
2953 #endif
2954
2955 Label done;
2956 cmpptr(rrealsp, Address(rthread, JavaThread::cont_fastpath_offset()));
2957 jccb(Assembler::belowEqual, done);
2958 movptr(Address(rthread, JavaThread::cont_fastpath_offset()), rrealsp);
2959 bind(done);
2960
2961 #ifndef _LP64
2962 pop(rrealsp);
2963 pop(rthread);
2964 #endif
2965 }
2966
2967 void MacroAssembler::pop_cont_fastpath() {
2968 if (!Continuations::enabled()) return;
2969
2970 #ifndef _LP64
2971 Register rthread = rax;
2972 Register rrealsp = rbx;
2973 push(rthread);
2974 push(rrealsp);
2975
2976 get_thread(rthread);
2977
2978 // The code below wants the original RSP.
2979 // Move it back after the pushes above.
2980 movptr(rrealsp, rsp);
2981 addptr(rrealsp, 2*wordSize);
2982 #else
2983 Register rthread = r15_thread;
2984 Register rrealsp = rsp;
2985 #endif
2986
2987 Label done;
2988 cmpptr(rrealsp, Address(rthread, JavaThread::cont_fastpath_offset()));
2989 jccb(Assembler::below, done);
2990 movptr(Address(rthread, JavaThread::cont_fastpath_offset()), 0);
2991 bind(done);
2992
2993 #ifndef _LP64
2994 pop(rrealsp);
2995 pop(rthread);
2996 #endif
2997 }
2998
2999 void MacroAssembler::inc_held_monitor_count() {
3000 #ifndef _LP64
3001 Register thread = rax;
3002 push(thread);
3003 get_thread(thread);
3004 incrementl(Address(thread, JavaThread::held_monitor_count_offset()));
3005 pop(thread);
3006 #else // LP64
3007 incrementq(Address(r15_thread, JavaThread::held_monitor_count_offset()));
3008 #endif
3009 }
3010
3011 void MacroAssembler::dec_held_monitor_count() {
3012 #ifndef _LP64
3013 Register thread = rax;
3014 push(thread);
3015 get_thread(thread);
3016 decrementl(Address(thread, JavaThread::held_monitor_count_offset()));
3017 pop(thread);
3018 #else // LP64
3019 decrementq(Address(r15_thread, JavaThread::held_monitor_count_offset()));
3020 #endif
3021 }
3022
3023 #ifdef ASSERT
3024 void MacroAssembler::stop_if_in_cont(Register cont, const char* name) {
3025 #ifdef _LP64
3026 Label no_cont;
3027 movptr(cont, Address(r15_thread, JavaThread::cont_entry_offset()));
3028 testl(cont, cont);
3029 jcc(Assembler::zero, no_cont);
3030 stop(name);
3031 bind(no_cont);
3032 #else
3033 Unimplemented();
3034 #endif
3035 }
3036 #endif
3037
3038 void MacroAssembler::reset_last_Java_frame(Register java_thread, bool clear_fp) { // determine java_thread register
3039 if (!java_thread->is_valid()) {
3040 java_thread = rdi;
3041 get_thread(java_thread);
3042 }
3043 // we must set sp to zero to clear frame
3044 movptr(Address(java_thread, JavaThread::last_Java_sp_offset()), NULL_WORD);
3045 // must clear fp, so that compiled frames are not confused; it is
3046 // possible that we need it only for debugging
3047 if (clear_fp) {
3048 movptr(Address(java_thread, JavaThread::last_Java_fp_offset()), NULL_WORD);
3049 }
3050 // Always clear the pc because it could have been set by make_walkable()
3051 movptr(Address(java_thread, JavaThread::last_Java_pc_offset()), NULL_WORD);
3052 vzeroupper();
3053 }
3054
3055 void MacroAssembler::restore_rax(Register tmp) {
3056 if (tmp == noreg) pop(rax);
3057 else if (tmp != rax) mov(rax, tmp);
3058 }
3059
3060 void MacroAssembler::round_to(Register reg, int modulus) {
3061 addptr(reg, modulus - 1);
3062 andptr(reg, -modulus);
3063 }
3064
3065 void MacroAssembler::save_rax(Register tmp) {
3066 if (tmp == noreg) push(rax);
3067 else if (tmp != rax) mov(tmp, rax);
3068 }
3069
3070 void MacroAssembler::safepoint_poll(Label& slow_path, Register thread_reg, bool at_return, bool in_nmethod) {
3071 if (at_return) {
3072 // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
3073 // we may safely use rsp instead to perform the stack watermark check.
3074 cmpptr(in_nmethod ? rsp : rbp, Address(thread_reg, JavaThread::polling_word_offset()));
3075 jcc(Assembler::above, slow_path);
3076 return;
3077 }
3078 testb(Address(thread_reg, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
3079 jcc(Assembler::notZero, slow_path); // handshake bit set implies poll
3080 }
3081
3082 // Calls to C land
3083 //
3084 // When entering C land, the rbp, & rsp of the last Java frame have to be recorded
3085 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
3086 // has to be reset to 0. This is required to allow proper stack traversal.
3087 void MacroAssembler::set_last_Java_frame(Register java_thread,
3088 Register last_java_sp,
3089 Register last_java_fp,
3090 address last_java_pc,
3091 Register rscratch) {
3092 vzeroupper();
3093 // determine java_thread register
3094 if (!java_thread->is_valid()) {
3095 java_thread = rdi;
3096 get_thread(java_thread);
3097 }
3098 // determine last_java_sp register
3099 if (!last_java_sp->is_valid()) {
3100 last_java_sp = rsp;
3101 }
3102 // last_java_fp is optional
3103 if (last_java_fp->is_valid()) {
3104 movptr(Address(java_thread, JavaThread::last_Java_fp_offset()), last_java_fp);
3105 }
3106 // last_java_pc is optional
3107 if (last_java_pc != nullptr) {
3108 Address java_pc(java_thread,
3109 JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset());
3110 lea(java_pc, InternalAddress(last_java_pc), rscratch);
3111 }
3112 movptr(Address(java_thread, JavaThread::last_Java_sp_offset()), last_java_sp);
3113 }
3114
3115 void MacroAssembler::shlptr(Register dst, int imm8) {
3116 LP64_ONLY(shlq(dst, imm8)) NOT_LP64(shll(dst, imm8));
3117 }
3118
3119 void MacroAssembler::shrptr(Register dst, int imm8) {
3120 LP64_ONLY(shrq(dst, imm8)) NOT_LP64(shrl(dst, imm8));
3121 }
3122
3123 void MacroAssembler::sign_extend_byte(Register reg) {
3124 if (LP64_ONLY(true ||) (VM_Version::is_P6() && reg->has_byte_register())) {
3125 movsbl(reg, reg); // movsxb
3126 } else {
3127 shll(reg, 24);
3128 sarl(reg, 24);
3129 }
3130 }
3131
3132 void MacroAssembler::sign_extend_short(Register reg) {
3133 if (LP64_ONLY(true ||) VM_Version::is_P6()) {
3134 movswl(reg, reg); // movsxw
3135 } else {
3136 shll(reg, 16);
3137 sarl(reg, 16);
3138 }
3139 }
3140
3141 void MacroAssembler::testl(Address dst, int32_t imm32) {
3142 if (imm32 >= 0 && is8bit(imm32)) {
3143 testb(dst, imm32);
3144 } else {
3145 Assembler::testl(dst, imm32);
3146 }
3147 }
3148
3149 void MacroAssembler::testl(Register dst, int32_t imm32) {
3150 if (imm32 >= 0 && is8bit(imm32) && dst->has_byte_register()) {
3151 testb(dst, imm32);
3152 } else {
3153 Assembler::testl(dst, imm32);
3154 }
3155 }
3156
3157 void MacroAssembler::testl(Register dst, AddressLiteral src) {
3158 assert(always_reachable(src), "Address should be reachable");
3159 testl(dst, as_Address(src));
3160 }
3161
3162 #ifdef _LP64
3163
3164 void MacroAssembler::testq(Address dst, int32_t imm32) {
3165 if (imm32 >= 0) {
3166 testl(dst, imm32);
3167 } else {
3168 Assembler::testq(dst, imm32);
3169 }
3170 }
3171
3172 void MacroAssembler::testq(Register dst, int32_t imm32) {
3173 if (imm32 >= 0) {
3174 testl(dst, imm32);
3175 } else {
3176 Assembler::testq(dst, imm32);
3177 }
3178 }
3179
3180 #endif
3181
3182 void MacroAssembler::pcmpeqb(XMMRegister dst, XMMRegister src) {
3183 assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3184 Assembler::pcmpeqb(dst, src);
3185 }
3186
3187 void MacroAssembler::pcmpeqw(XMMRegister dst, XMMRegister src) {
3188 assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3189 Assembler::pcmpeqw(dst, src);
3190 }
3191
3192 void MacroAssembler::pcmpestri(XMMRegister dst, Address src, int imm8) {
3193 assert((dst->encoding() < 16),"XMM register should be 0-15");
3194 Assembler::pcmpestri(dst, src, imm8);
3195 }
3196
3197 void MacroAssembler::pcmpestri(XMMRegister dst, XMMRegister src, int imm8) {
3198 assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
3199 Assembler::pcmpestri(dst, src, imm8);
3200 }
3201
3202 void MacroAssembler::pmovzxbw(XMMRegister dst, XMMRegister src) {
3203 assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3204 Assembler::pmovzxbw(dst, src);
3205 }
3206
3207 void MacroAssembler::pmovzxbw(XMMRegister dst, Address src) {
3208 assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3209 Assembler::pmovzxbw(dst, src);
3210 }
3211
3212 void MacroAssembler::pmovmskb(Register dst, XMMRegister src) {
3213 assert((src->encoding() < 16),"XMM register should be 0-15");
3214 Assembler::pmovmskb(dst, src);
3215 }
3216
3217 void MacroAssembler::ptest(XMMRegister dst, XMMRegister src) {
3218 assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
3219 Assembler::ptest(dst, src);
3220 }
3221
3222 void MacroAssembler::sqrtss(XMMRegister dst, AddressLiteral src, Register rscratch) {
3223 assert(rscratch != noreg || always_reachable(src), "missing");
3224
3225 if (reachable(src)) {
3226 Assembler::sqrtss(dst, as_Address(src));
3227 } else {
3228 lea(rscratch, src);
3229 Assembler::sqrtss(dst, Address(rscratch, 0));
3230 }
3231 }
3232
3233 void MacroAssembler::subsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
3234 assert(rscratch != noreg || always_reachable(src), "missing");
3235
3236 if (reachable(src)) {
3237 Assembler::subsd(dst, as_Address(src));
3238 } else {
3239 lea(rscratch, src);
3240 Assembler::subsd(dst, Address(rscratch, 0));
3241 }
3242 }
3243
3244 void MacroAssembler::roundsd(XMMRegister dst, AddressLiteral src, int32_t rmode, Register rscratch) {
3245 assert(rscratch != noreg || always_reachable(src), "missing");
3246
3247 if (reachable(src)) {
3248 Assembler::roundsd(dst, as_Address(src), rmode);
3249 } else {
3250 lea(rscratch, src);
3251 Assembler::roundsd(dst, Address(rscratch, 0), rmode);
3252 }
3253 }
3254
3255 void MacroAssembler::subss(XMMRegister dst, AddressLiteral src, Register rscratch) {
3256 assert(rscratch != noreg || always_reachable(src), "missing");
3257
3258 if (reachable(src)) {
3259 Assembler::subss(dst, as_Address(src));
3260 } else {
3261 lea(rscratch, src);
3262 Assembler::subss(dst, Address(rscratch, 0));
3263 }
3264 }
3265
3266 void MacroAssembler::ucomisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
3267 assert(rscratch != noreg || always_reachable(src), "missing");
3268
3269 if (reachable(src)) {
3270 Assembler::ucomisd(dst, as_Address(src));
3271 } else {
3272 lea(rscratch, src);
3273 Assembler::ucomisd(dst, Address(rscratch, 0));
3274 }
3275 }
3276
3277 void MacroAssembler::ucomiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
3278 assert(rscratch != noreg || always_reachable(src), "missing");
3279
3280 if (reachable(src)) {
3281 Assembler::ucomiss(dst, as_Address(src));
3282 } else {
3283 lea(rscratch, src);
3284 Assembler::ucomiss(dst, Address(rscratch, 0));
3285 }
3286 }
3287
3288 void MacroAssembler::xorpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
3289 assert(rscratch != noreg || always_reachable(src), "missing");
3290
3291 // Used in sign-bit flipping with aligned address.
3292 assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
3293 if (reachable(src)) {
3294 Assembler::xorpd(dst, as_Address(src));
3295 } else {
3296 lea(rscratch, src);
3297 Assembler::xorpd(dst, Address(rscratch, 0));
3298 }
3299 }
3300
3301 void MacroAssembler::xorpd(XMMRegister dst, XMMRegister src) {
3302 if (UseAVX > 2 && !VM_Version::supports_avx512dq() && (dst->encoding() == src->encoding())) {
3303 Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
3304 }
3305 else {
3306 Assembler::xorpd(dst, src);
3307 }
3308 }
3309
3310 void MacroAssembler::xorps(XMMRegister dst, XMMRegister src) {
3311 if (UseAVX > 2 && !VM_Version::supports_avx512dq() && (dst->encoding() == src->encoding())) {
3312 Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
3313 } else {
3314 Assembler::xorps(dst, src);
3315 }
3316 }
3317
3318 void MacroAssembler::xorps(XMMRegister dst, AddressLiteral src, Register rscratch) {
3319 assert(rscratch != noreg || always_reachable(src), "missing");
3320
3321 // Used in sign-bit flipping with aligned address.
3322 assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
3323 if (reachable(src)) {
3324 Assembler::xorps(dst, as_Address(src));
3325 } else {
3326 lea(rscratch, src);
3327 Assembler::xorps(dst, Address(rscratch, 0));
3328 }
3329 }
3330
3331 void MacroAssembler::pshufb(XMMRegister dst, AddressLiteral src, Register rscratch) {
3332 assert(rscratch != noreg || always_reachable(src), "missing");
3333
3334 // Used in sign-bit flipping with aligned address.
3335 bool aligned_adr = (((intptr_t)src.target() & 15) == 0);
3336 assert((UseAVX > 0) || aligned_adr, "SSE mode requires address alignment 16 bytes");
3337 if (reachable(src)) {
3338 Assembler::pshufb(dst, as_Address(src));
3339 } else {
3340 lea(rscratch, src);
3341 Assembler::pshufb(dst, Address(rscratch, 0));
3342 }
3343 }
3344
3345 // AVX 3-operands instructions
3346
3347 void MacroAssembler::vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3348 assert(rscratch != noreg || always_reachable(src), "missing");
3349
3350 if (reachable(src)) {
3351 vaddsd(dst, nds, as_Address(src));
3352 } else {
3353 lea(rscratch, src);
3354 vaddsd(dst, nds, Address(rscratch, 0));
3355 }
3356 }
3357
3358 void MacroAssembler::vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3359 assert(rscratch != noreg || always_reachable(src), "missing");
3360
3361 if (reachable(src)) {
3362 vaddss(dst, nds, as_Address(src));
3363 } else {
3364 lea(rscratch, src);
3365 vaddss(dst, nds, Address(rscratch, 0));
3366 }
3367 }
3368
3369 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3370 assert(UseAVX > 0, "requires some form of AVX");
3371 assert(rscratch != noreg || always_reachable(src), "missing");
3372
3373 if (reachable(src)) {
3374 Assembler::vpaddb(dst, nds, as_Address(src), vector_len);
3375 } else {
3376 lea(rscratch, src);
3377 Assembler::vpaddb(dst, nds, Address(rscratch, 0), vector_len);
3378 }
3379 }
3380
3381 void MacroAssembler::vpaddd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3382 assert(UseAVX > 0, "requires some form of AVX");
3383 assert(rscratch != noreg || always_reachable(src), "missing");
3384
3385 if (reachable(src)) {
3386 Assembler::vpaddd(dst, nds, as_Address(src), vector_len);
3387 } else {
3388 lea(rscratch, src);
3389 Assembler::vpaddd(dst, nds, Address(rscratch, 0), vector_len);
3390 }
3391 }
3392
3393 void MacroAssembler::vabsss(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
3394 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
3395 assert(rscratch != noreg || always_reachable(negate_field), "missing");
3396
3397 vandps(dst, nds, negate_field, vector_len, rscratch);
3398 }
3399
3400 void MacroAssembler::vabssd(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
3401 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
3402 assert(rscratch != noreg || always_reachable(negate_field), "missing");
3403
3404 vandpd(dst, nds, negate_field, vector_len, rscratch);
3405 }
3406
3407 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
3408 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3409 Assembler::vpaddb(dst, nds, src, vector_len);
3410 }
3411
3412 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
3413 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3414 Assembler::vpaddb(dst, nds, src, vector_len);
3415 }
3416
3417 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
3418 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3419 Assembler::vpaddw(dst, nds, src, vector_len);
3420 }
3421
3422 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
3423 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3424 Assembler::vpaddw(dst, nds, src, vector_len);
3425 }
3426
3427 void MacroAssembler::vpand(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3428 assert(rscratch != noreg || always_reachable(src), "missing");
3429
3430 if (reachable(src)) {
3431 Assembler::vpand(dst, nds, as_Address(src), vector_len);
3432 } else {
3433 lea(rscratch, src);
3434 Assembler::vpand(dst, nds, Address(rscratch, 0), vector_len);
3435 }
3436 }
3437
3438 void MacroAssembler::vpbroadcastd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
3439 assert(rscratch != noreg || always_reachable(src), "missing");
3440
3441 if (reachable(src)) {
3442 Assembler::vpbroadcastd(dst, as_Address(src), vector_len);
3443 } else {
3444 lea(rscratch, src);
3445 Assembler::vpbroadcastd(dst, Address(rscratch, 0), vector_len);
3446 }
3447 }
3448
3449 void MacroAssembler::vpbroadcastq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
3450 assert(rscratch != noreg || always_reachable(src), "missing");
3451
3452 if (reachable(src)) {
3453 Assembler::vpbroadcastq(dst, as_Address(src), vector_len);
3454 } else {
3455 lea(rscratch, src);
3456 Assembler::vpbroadcastq(dst, Address(rscratch, 0), vector_len);
3457 }
3458 }
3459
3460 void MacroAssembler::vbroadcastsd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
3461 assert(rscratch != noreg || always_reachable(src), "missing");
3462
3463 if (reachable(src)) {
3464 Assembler::vbroadcastsd(dst, as_Address(src), vector_len);
3465 } else {
3466 lea(rscratch, src);
3467 Assembler::vbroadcastsd(dst, Address(rscratch, 0), vector_len);
3468 }
3469 }
3470
3471 void MacroAssembler::vbroadcastss(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
3472 assert(rscratch != noreg || always_reachable(src), "missing");
3473
3474 if (reachable(src)) {
3475 Assembler::vbroadcastss(dst, as_Address(src), vector_len);
3476 } else {
3477 lea(rscratch, src);
3478 Assembler::vbroadcastss(dst, Address(rscratch, 0), vector_len);
3479 }
3480 }
3481
3482 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
3483 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3484 Assembler::vpcmpeqb(dst, nds, src, vector_len);
3485 }
3486
3487 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
3488 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3489 Assembler::vpcmpeqw(dst, nds, src, vector_len);
3490 }
3491
3492 void MacroAssembler::evpcmpeqd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3493 assert(rscratch != noreg || always_reachable(src), "missing");
3494
3495 if (reachable(src)) {
3496 Assembler::evpcmpeqd(kdst, mask, nds, as_Address(src), vector_len);
3497 } else {
3498 lea(rscratch, src);
3499 Assembler::evpcmpeqd(kdst, mask, nds, Address(rscratch, 0), vector_len);
3500 }
3501 }
3502
3503 void MacroAssembler::evpcmpd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
3504 int comparison, bool is_signed, int vector_len, Register rscratch) {
3505 assert(rscratch != noreg || always_reachable(src), "missing");
3506
3507 if (reachable(src)) {
3508 Assembler::evpcmpd(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
3509 } else {
3510 lea(rscratch, src);
3511 Assembler::evpcmpd(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
3512 }
3513 }
3514
3515 void MacroAssembler::evpcmpq(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
3516 int comparison, bool is_signed, int vector_len, Register rscratch) {
3517 assert(rscratch != noreg || always_reachable(src), "missing");
3518
3519 if (reachable(src)) {
3520 Assembler::evpcmpq(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
3521 } else {
3522 lea(rscratch, src);
3523 Assembler::evpcmpq(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
3524 }
3525 }
3526
3527 void MacroAssembler::evpcmpb(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
3528 int comparison, bool is_signed, int vector_len, Register rscratch) {
3529 assert(rscratch != noreg || always_reachable(src), "missing");
3530
3531 if (reachable(src)) {
3532 Assembler::evpcmpb(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
3533 } else {
3534 lea(rscratch, src);
3535 Assembler::evpcmpb(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
3536 }
3537 }
3538
3539 void MacroAssembler::evpcmpw(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
3540 int comparison, bool is_signed, int vector_len, Register rscratch) {
3541 assert(rscratch != noreg || always_reachable(src), "missing");
3542
3543 if (reachable(src)) {
3544 Assembler::evpcmpw(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
3545 } else {
3546 lea(rscratch, src);
3547 Assembler::evpcmpw(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
3548 }
3549 }
3550
3551 void MacroAssembler::vpcmpCC(XMMRegister dst, XMMRegister nds, XMMRegister src, int cond_encoding, Width width, int vector_len) {
3552 if (width == Assembler::Q) {
3553 Assembler::vpcmpCCq(dst, nds, src, cond_encoding, vector_len);
3554 } else {
3555 Assembler::vpcmpCCbwd(dst, nds, src, cond_encoding, vector_len);
3556 }
3557 }
3558
3559 void MacroAssembler::vpcmpCCW(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister xtmp, ComparisonPredicate cond, Width width, int vector_len) {
3560 int eq_cond_enc = 0x29;
3561 int gt_cond_enc = 0x37;
3562 if (width != Assembler::Q) {
3563 eq_cond_enc = 0x74 + width;
3564 gt_cond_enc = 0x64 + width;
3565 }
3566 switch (cond) {
3567 case eq:
3568 vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
3569 break;
3570 case neq:
3571 vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
3572 vallones(xtmp, vector_len);
3573 vpxor(dst, xtmp, dst, vector_len);
3574 break;
3575 case le:
3576 vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
3577 vallones(xtmp, vector_len);
3578 vpxor(dst, xtmp, dst, vector_len);
3579 break;
3580 case nlt:
3581 vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
3582 vallones(xtmp, vector_len);
3583 vpxor(dst, xtmp, dst, vector_len);
3584 break;
3585 case lt:
3586 vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
3587 break;
3588 case nle:
3589 vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
3590 break;
3591 default:
3592 assert(false, "Should not reach here");
3593 }
3594 }
3595
3596 void MacroAssembler::vpmovzxbw(XMMRegister dst, Address src, int vector_len) {
3597 assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3598 Assembler::vpmovzxbw(dst, src, vector_len);
3599 }
3600
3601 void MacroAssembler::vpmovmskb(Register dst, XMMRegister src, int vector_len) {
3602 assert((src->encoding() < 16),"XMM register should be 0-15");
3603 Assembler::vpmovmskb(dst, src, vector_len);
3604 }
3605
3606 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
3607 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3608 Assembler::vpmullw(dst, nds, src, vector_len);
3609 }
3610
3611 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
3612 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3613 Assembler::vpmullw(dst, nds, src, vector_len);
3614 }
3615
3616 void MacroAssembler::vpmulld(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3617 assert((UseAVX > 0), "AVX support is needed");
3618 assert(rscratch != noreg || always_reachable(src), "missing");
3619
3620 if (reachable(src)) {
3621 Assembler::vpmulld(dst, nds, as_Address(src), vector_len);
3622 } else {
3623 lea(rscratch, src);
3624 Assembler::vpmulld(dst, nds, Address(rscratch, 0), vector_len);
3625 }
3626 }
3627
3628 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
3629 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3630 Assembler::vpsubb(dst, nds, src, vector_len);
3631 }
3632
3633 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
3634 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3635 Assembler::vpsubb(dst, nds, src, vector_len);
3636 }
3637
3638 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
3639 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3640 Assembler::vpsubw(dst, nds, src, vector_len);
3641 }
3642
3643 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
3644 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3645 Assembler::vpsubw(dst, nds, src, vector_len);
3646 }
3647
3648 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
3649 assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3650 Assembler::vpsraw(dst, nds, shift, vector_len);
3651 }
3652
3653 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
3654 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3655 Assembler::vpsraw(dst, nds, shift, vector_len);
3656 }
3657
3658 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
3659 assert(UseAVX > 2,"");
3660 if (!VM_Version::supports_avx512vl() && vector_len < 2) {
3661 vector_len = 2;
3662 }
3663 Assembler::evpsraq(dst, nds, shift, vector_len);
3664 }
3665
3666 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
3667 assert(UseAVX > 2,"");
3668 if (!VM_Version::supports_avx512vl() && vector_len < 2) {
3669 vector_len = 2;
3670 }
3671 Assembler::evpsraq(dst, nds, shift, vector_len);
3672 }
3673
3674 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
3675 assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3676 Assembler::vpsrlw(dst, nds, shift, vector_len);
3677 }
3678
3679 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
3680 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3681 Assembler::vpsrlw(dst, nds, shift, vector_len);
3682 }
3683
3684 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
3685 assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3686 Assembler::vpsllw(dst, nds, shift, vector_len);
3687 }
3688
3689 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
3690 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3691 Assembler::vpsllw(dst, nds, shift, vector_len);
3692 }
3693
3694 void MacroAssembler::vptest(XMMRegister dst, XMMRegister src) {
3695 assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
3696 Assembler::vptest(dst, src);
3697 }
3698
3699 void MacroAssembler::punpcklbw(XMMRegister dst, XMMRegister src) {
3700 assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3701 Assembler::punpcklbw(dst, src);
3702 }
3703
3704 void MacroAssembler::pshufd(XMMRegister dst, Address src, int mode) {
3705 assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
3706 Assembler::pshufd(dst, src, mode);
3707 }
3708
3709 void MacroAssembler::pshuflw(XMMRegister dst, XMMRegister src, int mode) {
3710 assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3711 Assembler::pshuflw(dst, src, mode);
3712 }
3713
3714 void MacroAssembler::vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3715 assert(rscratch != noreg || always_reachable(src), "missing");
3716
3717 if (reachable(src)) {
3718 vandpd(dst, nds, as_Address(src), vector_len);
3719 } else {
3720 lea(rscratch, src);
3721 vandpd(dst, nds, Address(rscratch, 0), vector_len);
3722 }
3723 }
3724
3725 void MacroAssembler::vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3726 assert(rscratch != noreg || always_reachable(src), "missing");
3727
3728 if (reachable(src)) {
3729 vandps(dst, nds, as_Address(src), vector_len);
3730 } else {
3731 lea(rscratch, src);
3732 vandps(dst, nds, Address(rscratch, 0), vector_len);
3733 }
3734 }
3735
3736 void MacroAssembler::evpord(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src,
3737 bool merge, int vector_len, Register rscratch) {
3738 assert(rscratch != noreg || always_reachable(src), "missing");
3739
3740 if (reachable(src)) {
3741 Assembler::evpord(dst, mask, nds, as_Address(src), merge, vector_len);
3742 } else {
3743 lea(rscratch, src);
3744 Assembler::evpord(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
3745 }
3746 }
3747
3748 void MacroAssembler::vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3749 assert(rscratch != noreg || always_reachable(src), "missing");
3750
3751 if (reachable(src)) {
3752 vdivsd(dst, nds, as_Address(src));
3753 } else {
3754 lea(rscratch, src);
3755 vdivsd(dst, nds, Address(rscratch, 0));
3756 }
3757 }
3758
3759 void MacroAssembler::vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3760 assert(rscratch != noreg || always_reachable(src), "missing");
3761
3762 if (reachable(src)) {
3763 vdivss(dst, nds, as_Address(src));
3764 } else {
3765 lea(rscratch, src);
3766 vdivss(dst, nds, Address(rscratch, 0));
3767 }
3768 }
3769
3770 void MacroAssembler::vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3771 assert(rscratch != noreg || always_reachable(src), "missing");
3772
3773 if (reachable(src)) {
3774 vmulsd(dst, nds, as_Address(src));
3775 } else {
3776 lea(rscratch, src);
3777 vmulsd(dst, nds, Address(rscratch, 0));
3778 }
3779 }
3780
3781 void MacroAssembler::vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3782 assert(rscratch != noreg || always_reachable(src), "missing");
3783
3784 if (reachable(src)) {
3785 vmulss(dst, nds, as_Address(src));
3786 } else {
3787 lea(rscratch, src);
3788 vmulss(dst, nds, Address(rscratch, 0));
3789 }
3790 }
3791
3792 void MacroAssembler::vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3793 assert(rscratch != noreg || always_reachable(src), "missing");
3794
3795 if (reachable(src)) {
3796 vsubsd(dst, nds, as_Address(src));
3797 } else {
3798 lea(rscratch, src);
3799 vsubsd(dst, nds, Address(rscratch, 0));
3800 }
3801 }
3802
3803 void MacroAssembler::vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3804 assert(rscratch != noreg || always_reachable(src), "missing");
3805
3806 if (reachable(src)) {
3807 vsubss(dst, nds, as_Address(src));
3808 } else {
3809 lea(rscratch, src);
3810 vsubss(dst, nds, Address(rscratch, 0));
3811 }
3812 }
3813
3814 void MacroAssembler::vnegatess(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3815 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
3816 assert(rscratch != noreg || always_reachable(src), "missing");
3817
3818 vxorps(dst, nds, src, Assembler::AVX_128bit, rscratch);
3819 }
3820
3821 void MacroAssembler::vnegatesd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3822 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
3823 assert(rscratch != noreg || always_reachable(src), "missing");
3824
3825 vxorpd(dst, nds, src, Assembler::AVX_128bit, rscratch);
3826 }
3827
3828 void MacroAssembler::vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3829 assert(rscratch != noreg || always_reachable(src), "missing");
3830
3831 if (reachable(src)) {
3832 vxorpd(dst, nds, as_Address(src), vector_len);
3833 } else {
3834 lea(rscratch, src);
3835 vxorpd(dst, nds, Address(rscratch, 0), vector_len);
3836 }
3837 }
3838
3839 void MacroAssembler::vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3840 assert(rscratch != noreg || always_reachable(src), "missing");
3841
3842 if (reachable(src)) {
3843 vxorps(dst, nds, as_Address(src), vector_len);
3844 } else {
3845 lea(rscratch, src);
3846 vxorps(dst, nds, Address(rscratch, 0), vector_len);
3847 }
3848 }
3849
3850 void MacroAssembler::vpxor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3851 assert(rscratch != noreg || always_reachable(src), "missing");
3852
3853 if (UseAVX > 1 || (vector_len < 1)) {
3854 if (reachable(src)) {
3855 Assembler::vpxor(dst, nds, as_Address(src), vector_len);
3856 } else {
3857 lea(rscratch, src);
3858 Assembler::vpxor(dst, nds, Address(rscratch, 0), vector_len);
3859 }
3860 } else {
3861 MacroAssembler::vxorpd(dst, nds, src, vector_len, rscratch);
3862 }
3863 }
3864
3865 void MacroAssembler::vpermd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3866 assert(rscratch != noreg || always_reachable(src), "missing");
3867
3868 if (reachable(src)) {
3869 Assembler::vpermd(dst, nds, as_Address(src), vector_len);
3870 } else {
3871 lea(rscratch, src);
3872 Assembler::vpermd(dst, nds, Address(rscratch, 0), vector_len);
3873 }
3874 }
3875
3876 void MacroAssembler::clear_jobject_tag(Register possibly_non_local) {
3877 const int32_t inverted_mask = ~static_cast<int32_t>(JNIHandles::tag_mask);
3878 STATIC_ASSERT(inverted_mask == -4); // otherwise check this code
3879 // The inverted mask is sign-extended
3880 andptr(possibly_non_local, inverted_mask);
3881 }
3882
3883 void MacroAssembler::resolve_jobject(Register value,
3884 Register thread,
3885 Register tmp) {
3886 assert_different_registers(value, thread, tmp);
3887 Label done, tagged, weak_tagged;
3888 testptr(value, value);
3889 jcc(Assembler::zero, done); // Use null as-is.
3890 testptr(value, JNIHandles::tag_mask); // Test for tag.
3891 jcc(Assembler::notZero, tagged);
3892
3893 // Resolve local handle
3894 access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp, thread);
3895 verify_oop(value);
3896 jmp(done);
3897
3898 bind(tagged);
3899 testptr(value, JNIHandles::TypeTag::weak_global); // Test for weak tag.
3900 jcc(Assembler::notZero, weak_tagged);
3901
3902 // Resolve global handle
3903 access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp, thread);
3904 verify_oop(value);
3905 jmp(done);
3906
3907 bind(weak_tagged);
3908 // Resolve jweak.
3909 access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
3910 value, Address(value, -JNIHandles::TypeTag::weak_global), tmp, thread);
3911 verify_oop(value);
3912
3913 bind(done);
3914 }
3915
3916 void MacroAssembler::resolve_global_jobject(Register value,
3917 Register thread,
3918 Register tmp) {
3919 assert_different_registers(value, thread, tmp);
3920 Label done;
3921
3922 testptr(value, value);
3923 jcc(Assembler::zero, done); // Use null as-is.
3924
3925 #ifdef ASSERT
3926 {
3927 Label valid_global_tag;
3928 testptr(value, JNIHandles::TypeTag::global); // Test for global tag.
3929 jcc(Assembler::notZero, valid_global_tag);
3930 stop("non global jobject using resolve_global_jobject");
3931 bind(valid_global_tag);
3932 }
3933 #endif
3934
3935 // Resolve global handle
3936 access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp, thread);
3937 verify_oop(value);
3938
3939 bind(done);
3940 }
3941
3942 void MacroAssembler::subptr(Register dst, int32_t imm32) {
3943 LP64_ONLY(subq(dst, imm32)) NOT_LP64(subl(dst, imm32));
3944 }
3945
3946 // Force generation of a 4 byte immediate value even if it fits into 8bit
3947 void MacroAssembler::subptr_imm32(Register dst, int32_t imm32) {
3948 LP64_ONLY(subq_imm32(dst, imm32)) NOT_LP64(subl_imm32(dst, imm32));
3949 }
3950
3951 void MacroAssembler::subptr(Register dst, Register src) {
3952 LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src));
3953 }
3954
3955 // C++ bool manipulation
3956 void MacroAssembler::testbool(Register dst) {
3957 if(sizeof(bool) == 1)
3958 testb(dst, 0xff);
3959 else if(sizeof(bool) == 2) {
3960 // testw implementation needed for two byte bools
3961 ShouldNotReachHere();
3962 } else if(sizeof(bool) == 4)
3963 testl(dst, dst);
3964 else
3965 // unsupported
3966 ShouldNotReachHere();
3967 }
3968
3969 void MacroAssembler::testptr(Register dst, Register src) {
3970 LP64_ONLY(testq(dst, src)) NOT_LP64(testl(dst, src));
3971 }
3972
3973 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
3974 void MacroAssembler::tlab_allocate(Register thread, Register obj,
3975 Register var_size_in_bytes,
3976 int con_size_in_bytes,
3977 Register t1,
3978 Register t2,
3979 Label& slow_case) {
3980 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
3981 bs->tlab_allocate(this, thread, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
3982 }
3983
3984 RegSet MacroAssembler::call_clobbered_gp_registers() {
3985 RegSet regs;
3986 #ifdef _LP64
3987 regs += RegSet::of(rax, rcx, rdx);
3988 #ifndef WINDOWS
3989 regs += RegSet::of(rsi, rdi);
3990 #endif
3991 regs += RegSet::range(r8, r11);
3992 #else
3993 regs += RegSet::of(rax, rcx, rdx);
3994 #endif
3995 return regs;
3996 }
3997
3998 XMMRegSet MacroAssembler::call_clobbered_xmm_registers() {
3999 int num_xmm_registers = XMMRegister::available_xmm_registers();
4000 #if defined(WINDOWS) && defined(_LP64)
4001 XMMRegSet result = XMMRegSet::range(xmm0, xmm5);
4002 if (num_xmm_registers > 16) {
4003 result += XMMRegSet::range(xmm16, as_XMMRegister(num_xmm_registers - 1));
4004 }
4005 return result;
4006 #else
4007 return XMMRegSet::range(xmm0, as_XMMRegister(num_xmm_registers - 1));
4008 #endif
4009 }
4010
4011 static int FPUSaveAreaSize = align_up(108, StackAlignmentInBytes); // 108 bytes needed for FPU state by fsave/frstor
4012
4013 #ifndef _LP64
4014 static bool use_x87_registers() { return UseSSE < 2; }
4015 #endif
4016 static bool use_xmm_registers() { return UseSSE >= 1; }
4017
4018 // C1 only ever uses the first double/float of the XMM register.
4019 static int xmm_save_size() { return UseSSE >= 2 ? sizeof(double) : sizeof(float); }
4020
4021 static void save_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
4022 if (UseSSE == 1) {
4023 masm->movflt(Address(rsp, offset), reg);
4024 } else {
4025 masm->movdbl(Address(rsp, offset), reg);
4026 }
4027 }
4028
4029 static void restore_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
4030 if (UseSSE == 1) {
4031 masm->movflt(reg, Address(rsp, offset));
4032 } else {
4033 masm->movdbl(reg, Address(rsp, offset));
4034 }
4035 }
4036
4037 int register_section_sizes(RegSet gp_registers, XMMRegSet xmm_registers, bool save_fpu,
4038 int& gp_area_size, int& fp_area_size, int& xmm_area_size) {
4039
4040 gp_area_size = align_up(gp_registers.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size,
4041 StackAlignmentInBytes);
4042 #ifdef _LP64
4043 fp_area_size = 0;
4044 #else
4045 fp_area_size = (save_fpu && use_x87_registers()) ? FPUSaveAreaSize : 0;
4046 #endif
4047 xmm_area_size = (save_fpu && use_xmm_registers()) ? xmm_registers.size() * xmm_save_size() : 0;
4048
4049 return gp_area_size + fp_area_size + xmm_area_size;
4050 }
4051
4052 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude, bool save_fpu) {
4053 block_comment("push_call_clobbered_registers start");
4054 // Regular registers
4055 RegSet gp_registers_to_push = call_clobbered_gp_registers() - exclude;
4056
4057 int gp_area_size;
4058 int fp_area_size;
4059 int xmm_area_size;
4060 int total_save_size = register_section_sizes(gp_registers_to_push, call_clobbered_xmm_registers(), save_fpu,
4061 gp_area_size, fp_area_size, xmm_area_size);
4062 subptr(rsp, total_save_size);
4063
4064 push_set(gp_registers_to_push, 0);
4065
4066 #ifndef _LP64
4067 if (save_fpu && use_x87_registers()) {
4068 fnsave(Address(rsp, gp_area_size));
4069 fwait();
4070 }
4071 #endif
4072 if (save_fpu && use_xmm_registers()) {
4073 push_set(call_clobbered_xmm_registers(), gp_area_size + fp_area_size);
4074 }
4075
4076 block_comment("push_call_clobbered_registers end");
4077 }
4078
4079 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude, bool restore_fpu) {
4080 block_comment("pop_call_clobbered_registers start");
4081
4082 RegSet gp_registers_to_pop = call_clobbered_gp_registers() - exclude;
4083
4084 int gp_area_size;
4085 int fp_area_size;
4086 int xmm_area_size;
4087 int total_save_size = register_section_sizes(gp_registers_to_pop, call_clobbered_xmm_registers(), restore_fpu,
4088 gp_area_size, fp_area_size, xmm_area_size);
4089
4090 if (restore_fpu && use_xmm_registers()) {
4091 pop_set(call_clobbered_xmm_registers(), gp_area_size + fp_area_size);
4092 }
4093 #ifndef _LP64
4094 if (restore_fpu && use_x87_registers()) {
4095 frstor(Address(rsp, gp_area_size));
4096 }
4097 #endif
4098
4099 pop_set(gp_registers_to_pop, 0);
4100
4101 addptr(rsp, total_save_size);
4102
4103 vzeroupper();
4104
4105 block_comment("pop_call_clobbered_registers end");
4106 }
4107
4108 void MacroAssembler::push_set(XMMRegSet set, int offset) {
4109 assert(is_aligned(set.size() * xmm_save_size(), StackAlignmentInBytes), "must be");
4110 int spill_offset = offset;
4111
4112 for (RegSetIterator<XMMRegister> it = set.begin(); *it != xnoreg; ++it) {
4113 save_xmm_register(this, spill_offset, *it);
4114 spill_offset += xmm_save_size();
4115 }
4116 }
4117
4118 void MacroAssembler::pop_set(XMMRegSet set, int offset) {
4119 int restore_size = set.size() * xmm_save_size();
4120 assert(is_aligned(restore_size, StackAlignmentInBytes), "must be");
4121
4122 int restore_offset = offset + restore_size - xmm_save_size();
4123
4124 for (ReverseRegSetIterator<XMMRegister> it = set.rbegin(); *it != xnoreg; ++it) {
4125 restore_xmm_register(this, restore_offset, *it);
4126 restore_offset -= xmm_save_size();
4127 }
4128 }
4129
4130 void MacroAssembler::push_set(RegSet set, int offset) {
4131 int spill_offset;
4132 if (offset == -1) {
4133 int register_push_size = set.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
4134 int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
4135 subptr(rsp, aligned_size);
4136 spill_offset = 0;
4137 } else {
4138 spill_offset = offset;
4139 }
4140
4141 for (RegSetIterator<Register> it = set.begin(); *it != noreg; ++it) {
4142 movptr(Address(rsp, spill_offset), *it);
4143 spill_offset += Register::max_slots_per_register * VMRegImpl::stack_slot_size;
4144 }
4145 }
4146
4147 void MacroAssembler::pop_set(RegSet set, int offset) {
4148
4149 int gp_reg_size = Register::max_slots_per_register * VMRegImpl::stack_slot_size;
4150 int restore_size = set.size() * gp_reg_size;
4151 int aligned_size = align_up(restore_size, StackAlignmentInBytes);
4152
4153 int restore_offset;
4154 if (offset == -1) {
4155 restore_offset = restore_size - gp_reg_size;
4156 } else {
4157 restore_offset = offset + restore_size - gp_reg_size;
4158 }
4159 for (ReverseRegSetIterator<Register> it = set.rbegin(); *it != noreg; ++it) {
4160 movptr(*it, Address(rsp, restore_offset));
4161 restore_offset -= gp_reg_size;
4162 }
4163
4164 if (offset == -1) {
4165 addptr(rsp, aligned_size);
4166 }
4167 }
4168
4169 // Preserves the contents of address, destroys the contents length_in_bytes and temp.
4170 void MacroAssembler::zero_memory(Register address, Register length_in_bytes, int offset_in_bytes, Register temp) {
4171 assert(address != length_in_bytes && address != temp && temp != length_in_bytes, "registers must be different");
4172 assert((offset_in_bytes & (BytesPerWord - 1)) == 0, "offset must be a multiple of BytesPerWord");
4173 Label done;
4174
4175 testptr(length_in_bytes, length_in_bytes);
4176 jcc(Assembler::zero, done);
4177
4178 // initialize topmost word, divide index by 2, check if odd and test if zero
4179 // note: for the remaining code to work, index must be a multiple of BytesPerWord
4180 #ifdef ASSERT
4181 {
4182 Label L;
4183 testptr(length_in_bytes, BytesPerWord - 1);
4184 jcc(Assembler::zero, L);
4185 stop("length must be a multiple of BytesPerWord");
4186 bind(L);
4187 }
4188 #endif
4189 Register index = length_in_bytes;
4190 xorptr(temp, temp); // use _zero reg to clear memory (shorter code)
4191 if (UseIncDec) {
4192 shrptr(index, 3); // divide by 8/16 and set carry flag if bit 2 was set
4193 } else {
4194 shrptr(index, 2); // use 2 instructions to avoid partial flag stall
4195 shrptr(index, 1);
4196 }
4197 #ifndef _LP64
4198 // index could have not been a multiple of 8 (i.e., bit 2 was set)
4199 {
4200 Label even;
4201 // note: if index was a multiple of 8, then it cannot
4202 // be 0 now otherwise it must have been 0 before
4203 // => if it is even, we don't need to check for 0 again
4204 jcc(Assembler::carryClear, even);
4205 // clear topmost word (no jump would be needed if conditional assignment worked here)
4206 movptr(Address(address, index, Address::times_8, offset_in_bytes - 0*BytesPerWord), temp);
4207 // index could be 0 now, must check again
4208 jcc(Assembler::zero, done);
4209 bind(even);
4210 }
4211 #endif // !_LP64
4212 // initialize remaining object fields: index is a multiple of 2 now
4213 {
4214 Label loop;
4215 bind(loop);
4216 movptr(Address(address, index, Address::times_8, offset_in_bytes - 1*BytesPerWord), temp);
4217 NOT_LP64(movptr(Address(address, index, Address::times_8, offset_in_bytes - 2*BytesPerWord), temp);)
4218 decrement(index);
4219 jcc(Assembler::notZero, loop);
4220 }
4221
4222 bind(done);
4223 }
4224
4225 // Look up the method for a megamorphic invokeinterface call.
4226 // The target method is determined by <intf_klass, itable_index>.
4227 // The receiver klass is in recv_klass.
4228 // On success, the result will be in method_result, and execution falls through.
4229 // On failure, execution transfers to the given label.
4230 void MacroAssembler::lookup_interface_method(Register recv_klass,
4231 Register intf_klass,
4232 RegisterOrConstant itable_index,
4233 Register method_result,
4234 Register scan_temp,
4235 Label& L_no_such_interface,
4236 bool return_method) {
4237 assert_different_registers(recv_klass, intf_klass, scan_temp);
4238 assert_different_registers(method_result, intf_klass, scan_temp);
4239 assert(recv_klass != method_result || !return_method,
4240 "recv_klass can be destroyed when method isn't needed");
4241
4242 assert(itable_index.is_constant() || itable_index.as_register() == method_result,
4243 "caller must use same register for non-constant itable index as for method");
4244
4245 // Compute start of first itableOffsetEntry (which is at the end of the vtable)
4246 int vtable_base = in_bytes(Klass::vtable_start_offset());
4247 int itentry_off = in_bytes(itableMethodEntry::method_offset());
4248 int scan_step = itableOffsetEntry::size() * wordSize;
4249 int vte_size = vtableEntry::size_in_bytes();
4250 Address::ScaleFactor times_vte_scale = Address::times_ptr;
4251 assert(vte_size == wordSize, "else adjust times_vte_scale");
4252
4253 movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
4254
4255 // %%% Could store the aligned, prescaled offset in the klassoop.
4256 lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
4257
4258 if (return_method) {
4259 // Adjust recv_klass by scaled itable_index, so we can free itable_index.
4260 assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
4261 lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
4262 }
4263
4264 // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
4265 // if (scan->interface() == intf) {
4266 // result = (klass + scan->offset() + itable_index);
4267 // }
4268 // }
4269 Label search, found_method;
4270
4271 for (int peel = 1; peel >= 0; peel--) {
4272 movptr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
4273 cmpptr(intf_klass, method_result);
4274
4275 if (peel) {
4276 jccb(Assembler::equal, found_method);
4277 } else {
4278 jccb(Assembler::notEqual, search);
4279 // (invert the test to fall through to found_method...)
4280 }
4281
4282 if (!peel) break;
4283
4284 bind(search);
4285
4286 // Check that the previous entry is non-null. A null entry means that
4287 // the receiver class doesn't implement the interface, and wasn't the
4288 // same as when the caller was compiled.
4289 testptr(method_result, method_result);
4290 jcc(Assembler::zero, L_no_such_interface);
4291 addptr(scan_temp, scan_step);
4292 }
4293
4294 bind(found_method);
4295
4296 if (return_method) {
4297 // Got a hit.
4298 movl(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
4299 movptr(method_result, Address(recv_klass, scan_temp, Address::times_1));
4300 }
4301 }
4302
4303 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
4304 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICHolder
4305 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
4306 // The target method is determined by <holder_klass, itable_index>.
4307 // The receiver klass is in recv_klass.
4308 // On success, the result will be in method_result, and execution falls through.
4309 // On failure, execution transfers to the given label.
4310 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
4311 Register holder_klass,
4312 Register resolved_klass,
4313 Register method_result,
4314 Register scan_temp,
4315 Register temp_reg2,
4316 Register receiver,
4317 int itable_index,
4318 Label& L_no_such_interface) {
4319 assert_different_registers(recv_klass, method_result, holder_klass, resolved_klass, scan_temp, temp_reg2, receiver);
4320 Register temp_itbl_klass = method_result;
4321 Register temp_reg = (temp_reg2 == noreg ? recv_klass : temp_reg2); // reuse recv_klass register on 32-bit x86 impl
4322
4323 int vtable_base = in_bytes(Klass::vtable_start_offset());
4324 int itentry_off = in_bytes(itableMethodEntry::method_offset());
4325 int scan_step = itableOffsetEntry::size() * wordSize;
4326 int vte_size = vtableEntry::size_in_bytes();
4327 int ioffset = in_bytes(itableOffsetEntry::interface_offset());
4328 int ooffset = in_bytes(itableOffsetEntry::offset_offset());
4329 Address::ScaleFactor times_vte_scale = Address::times_ptr;
4330 assert(vte_size == wordSize, "adjust times_vte_scale");
4331
4332 Label L_loop_scan_resolved_entry, L_resolved_found, L_holder_found;
4333
4334 // temp_itbl_klass = recv_klass.itable[0]
4335 // scan_temp = &recv_klass.itable[0] + step
4336 movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
4337 movptr(temp_itbl_klass, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset));
4338 lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset + scan_step));
4339 xorptr(temp_reg, temp_reg);
4340
4341 // Initial checks:
4342 // - if (holder_klass != resolved_klass), go to "scan for resolved"
4343 // - if (itable[0] == 0), no such interface
4344 // - if (itable[0] == holder_klass), shortcut to "holder found"
4345 cmpptr(holder_klass, resolved_klass);
4346 jccb(Assembler::notEqual, L_loop_scan_resolved_entry);
4347 testptr(temp_itbl_klass, temp_itbl_klass);
4348 jccb(Assembler::zero, L_no_such_interface);
4349 cmpptr(holder_klass, temp_itbl_klass);
4350 jccb(Assembler::equal, L_holder_found);
4351
4352 // Loop: Look for holder_klass record in itable
4353 // do {
4354 // tmp = itable[index];
4355 // index += step;
4356 // if (tmp == holder_klass) {
4357 // goto L_holder_found; // Found!
4358 // }
4359 // } while (tmp != 0);
4360 // goto L_no_such_interface // Not found.
4361 Label L_scan_holder;
4362 bind(L_scan_holder);
4363 movptr(temp_itbl_klass, Address(scan_temp, 0));
4364 addptr(scan_temp, scan_step);
4365 cmpptr(holder_klass, temp_itbl_klass);
4366 jccb(Assembler::equal, L_holder_found);
4367 testptr(temp_itbl_klass, temp_itbl_klass);
4368 jccb(Assembler::notZero, L_scan_holder);
4369
4370 jmpb(L_no_such_interface);
4371
4372 // Loop: Look for resolved_class record in itable
4373 // do {
4374 // tmp = itable[index];
4375 // index += step;
4376 // if (tmp == holder_klass) {
4377 // // Also check if we have met a holder klass
4378 // holder_tmp = itable[index-step-ioffset];
4379 // }
4380 // if (tmp == resolved_klass) {
4381 // goto L_resolved_found; // Found!
4382 // }
4383 // } while (tmp != 0);
4384 // goto L_no_such_interface // Not found.
4385 //
4386 Label L_loop_scan_resolved;
4387 bind(L_loop_scan_resolved);
4388 movptr(temp_itbl_klass, Address(scan_temp, 0));
4389 addptr(scan_temp, scan_step);
4390 bind(L_loop_scan_resolved_entry);
4391 cmpptr(holder_klass, temp_itbl_klass);
4392 cmovl(Assembler::equal, temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
4393 cmpptr(resolved_klass, temp_itbl_klass);
4394 jccb(Assembler::equal, L_resolved_found);
4395 testptr(temp_itbl_klass, temp_itbl_klass);
4396 jccb(Assembler::notZero, L_loop_scan_resolved);
4397
4398 jmpb(L_no_such_interface);
4399
4400 Label L_ready;
4401
4402 // See if we already have a holder klass. If not, go and scan for it.
4403 bind(L_resolved_found);
4404 testptr(temp_reg, temp_reg);
4405 jccb(Assembler::zero, L_scan_holder);
4406 jmpb(L_ready);
4407
4408 bind(L_holder_found);
4409 movl(temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
4410
4411 // Finally, temp_reg contains holder_klass vtable offset
4412 bind(L_ready);
4413 assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
4414 if (temp_reg2 == noreg) { // recv_klass register is clobbered for 32-bit x86 impl
4415 load_klass(scan_temp, receiver, noreg);
4416 movptr(method_result, Address(scan_temp, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
4417 } else {
4418 movptr(method_result, Address(recv_klass, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
4419 }
4420 }
4421
4422
4423 // virtual method calling
4424 void MacroAssembler::lookup_virtual_method(Register recv_klass,
4425 RegisterOrConstant vtable_index,
4426 Register method_result) {
4427 const ByteSize base = Klass::vtable_start_offset();
4428 assert(vtableEntry::size() * wordSize == wordSize, "else adjust the scaling in the code below");
4429 Address vtable_entry_addr(recv_klass,
4430 vtable_index, Address::times_ptr,
4431 base + vtableEntry::method_offset());
4432 movptr(method_result, vtable_entry_addr);
4433 }
4434
4435
4436 void MacroAssembler::check_klass_subtype(Register sub_klass,
4437 Register super_klass,
4438 Register temp_reg,
4439 Label& L_success) {
4440 Label L_failure;
4441 check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg, &L_success, &L_failure, nullptr);
4442 check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
4443 bind(L_failure);
4444 }
4445
4446
4447 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
4448 Register super_klass,
4449 Register temp_reg,
4450 Label* L_success,
4451 Label* L_failure,
4452 Label* L_slow_path,
4453 RegisterOrConstant super_check_offset) {
4454 assert_different_registers(sub_klass, super_klass, temp_reg);
4455 bool must_load_sco = (super_check_offset.constant_or_zero() == -1);
4456 if (super_check_offset.is_register()) {
4457 assert_different_registers(sub_klass, super_klass,
4458 super_check_offset.as_register());
4459 } else if (must_load_sco) {
4460 assert(temp_reg != noreg, "supply either a temp or a register offset");
4461 }
4462
4463 Label L_fallthrough;
4464 int label_nulls = 0;
4465 if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
4466 if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
4467 if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
4468 assert(label_nulls <= 1, "at most one null in the batch");
4469
4470 int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
4471 int sco_offset = in_bytes(Klass::super_check_offset_offset());
4472 Address super_check_offset_addr(super_klass, sco_offset);
4473
4474 // Hacked jcc, which "knows" that L_fallthrough, at least, is in
4475 // range of a jccb. If this routine grows larger, reconsider at
4476 // least some of these.
4477 #define local_jcc(assembler_cond, label) \
4478 if (&(label) == &L_fallthrough) jccb(assembler_cond, label); \
4479 else jcc( assembler_cond, label) /*omit semi*/
4480
4481 // Hacked jmp, which may only be used just before L_fallthrough.
4482 #define final_jmp(label) \
4483 if (&(label) == &L_fallthrough) { /*do nothing*/ } \
4484 else jmp(label) /*omit semi*/
4485
4486 // If the pointers are equal, we are done (e.g., String[] elements).
4487 // This self-check enables sharing of secondary supertype arrays among
4488 // non-primary types such as array-of-interface. Otherwise, each such
4489 // type would need its own customized SSA.
4490 // We move this check to the front of the fast path because many
4491 // type checks are in fact trivially successful in this manner,
4492 // so we get a nicely predicted branch right at the start of the check.
4493 cmpptr(sub_klass, super_klass);
4494 local_jcc(Assembler::equal, *L_success);
4495
4496 // Check the supertype display:
4497 if (must_load_sco) {
4498 // Positive movl does right thing on LP64.
4499 movl(temp_reg, super_check_offset_addr);
4500 super_check_offset = RegisterOrConstant(temp_reg);
4501 }
4502 Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0);
4503 cmpptr(super_klass, super_check_addr); // load displayed supertype
4504
4505 // This check has worked decisively for primary supers.
4506 // Secondary supers are sought in the super_cache ('super_cache_addr').
4507 // (Secondary supers are interfaces and very deeply nested subtypes.)
4508 // This works in the same check above because of a tricky aliasing
4509 // between the super_cache and the primary super display elements.
4510 // (The 'super_check_addr' can address either, as the case requires.)
4511 // Note that the cache is updated below if it does not help us find
4512 // what we need immediately.
4513 // So if it was a primary super, we can just fail immediately.
4514 // Otherwise, it's the slow path for us (no success at this point).
4515
4516 if (super_check_offset.is_register()) {
4517 local_jcc(Assembler::equal, *L_success);
4518 cmpl(super_check_offset.as_register(), sc_offset);
4519 if (L_failure == &L_fallthrough) {
4520 local_jcc(Assembler::equal, *L_slow_path);
4521 } else {
4522 local_jcc(Assembler::notEqual, *L_failure);
4523 final_jmp(*L_slow_path);
4524 }
4525 } else if (super_check_offset.as_constant() == sc_offset) {
4526 // Need a slow path; fast failure is impossible.
4527 if (L_slow_path == &L_fallthrough) {
4528 local_jcc(Assembler::equal, *L_success);
4529 } else {
4530 local_jcc(Assembler::notEqual, *L_slow_path);
4531 final_jmp(*L_success);
4532 }
4533 } else {
4534 // No slow path; it's a fast decision.
4535 if (L_failure == &L_fallthrough) {
4536 local_jcc(Assembler::equal, *L_success);
4537 } else {
4538 local_jcc(Assembler::notEqual, *L_failure);
4539 final_jmp(*L_success);
4540 }
4541 }
4542
4543 bind(L_fallthrough);
4544
4545 #undef local_jcc
4546 #undef final_jmp
4547 }
4548
4549
4550 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
4551 Register super_klass,
4552 Register temp_reg,
4553 Register temp2_reg,
4554 Label* L_success,
4555 Label* L_failure,
4556 bool set_cond_codes) {
4557 assert_different_registers(sub_klass, super_klass, temp_reg);
4558 if (temp2_reg != noreg)
4559 assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg);
4560 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
4561
4562 Label L_fallthrough;
4563 int label_nulls = 0;
4564 if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
4565 if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
4566 assert(label_nulls <= 1, "at most one null in the batch");
4567
4568 // a couple of useful fields in sub_klass:
4569 int ss_offset = in_bytes(Klass::secondary_supers_offset());
4570 int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
4571 Address secondary_supers_addr(sub_klass, ss_offset);
4572 Address super_cache_addr( sub_klass, sc_offset);
4573
4574 // Do a linear scan of the secondary super-klass chain.
4575 // This code is rarely used, so simplicity is a virtue here.
4576 // The repne_scan instruction uses fixed registers, which we must spill.
4577 // Don't worry too much about pre-existing connections with the input regs.
4578
4579 assert(sub_klass != rax, "killed reg"); // killed by mov(rax, super)
4580 assert(sub_klass != rcx, "killed reg"); // killed by lea(rcx, &pst_counter)
4581
4582 // Get super_klass value into rax (even if it was in rdi or rcx).
4583 bool pushed_rax = false, pushed_rcx = false, pushed_rdi = false;
4584 if (super_klass != rax) {
4585 if (!IS_A_TEMP(rax)) { push(rax); pushed_rax = true; }
4586 mov(rax, super_klass);
4587 }
4588 if (!IS_A_TEMP(rcx)) { push(rcx); pushed_rcx = true; }
4589 if (!IS_A_TEMP(rdi)) { push(rdi); pushed_rdi = true; }
4590
4591 #ifndef PRODUCT
4592 int* pst_counter = &SharedRuntime::_partial_subtype_ctr;
4593 ExternalAddress pst_counter_addr((address) pst_counter);
4594 NOT_LP64( incrementl(pst_counter_addr) );
4595 LP64_ONLY( lea(rcx, pst_counter_addr) );
4596 LP64_ONLY( incrementl(Address(rcx, 0)) );
4597 #endif //PRODUCT
4598
4599 // We will consult the secondary-super array.
4600 movptr(rdi, secondary_supers_addr);
4601 // Load the array length. (Positive movl does right thing on LP64.)
4602 movl(rcx, Address(rdi, Array<Klass*>::length_offset_in_bytes()));
4603 // Skip to start of data.
4604 addptr(rdi, Array<Klass*>::base_offset_in_bytes());
4605
4606 // Scan RCX words at [RDI] for an occurrence of RAX.
4607 // Set NZ/Z based on last compare.
4608 // Z flag value will not be set by 'repne' if RCX == 0 since 'repne' does
4609 // not change flags (only scas instruction which is repeated sets flags).
4610 // Set Z = 0 (not equal) before 'repne' to indicate that class was not found.
4611
4612 testptr(rax,rax); // Set Z = 0
4613 repne_scan();
4614
4615 // Unspill the temp. registers:
4616 if (pushed_rdi) pop(rdi);
4617 if (pushed_rcx) pop(rcx);
4618 if (pushed_rax) pop(rax);
4619
4620 if (set_cond_codes) {
4621 // Special hack for the AD files: rdi is guaranteed non-zero.
4622 assert(!pushed_rdi, "rdi must be left non-null");
4623 // Also, the condition codes are properly set Z/NZ on succeed/failure.
4624 }
4625
4626 if (L_failure == &L_fallthrough)
4627 jccb(Assembler::notEqual, *L_failure);
4628 else jcc(Assembler::notEqual, *L_failure);
4629
4630 // Success. Cache the super we found and proceed in triumph.
4631 movptr(super_cache_addr, super_klass);
4632
4633 if (L_success != &L_fallthrough) {
4634 jmp(*L_success);
4635 }
4636
4637 #undef IS_A_TEMP
4638
4639 bind(L_fallthrough);
4640 }
4641
4642 void MacroAssembler::clinit_barrier(Register klass, Register thread, Label* L_fast_path, Label* L_slow_path) {
4643 assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
4644
4645 Label L_fallthrough;
4646 if (L_fast_path == nullptr) {
4647 L_fast_path = &L_fallthrough;
4648 } else if (L_slow_path == nullptr) {
4649 L_slow_path = &L_fallthrough;
4650 }
4651
4652 // Fast path check: class is fully initialized
4653 cmpb(Address(klass, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
4654 jcc(Assembler::equal, *L_fast_path);
4655
4656 // Fast path check: current thread is initializer thread
4657 cmpptr(thread, Address(klass, InstanceKlass::init_thread_offset()));
4658 if (L_slow_path == &L_fallthrough) {
4659 jcc(Assembler::equal, *L_fast_path);
4660 bind(*L_slow_path);
4661 } else if (L_fast_path == &L_fallthrough) {
4662 jcc(Assembler::notEqual, *L_slow_path);
4663 bind(*L_fast_path);
4664 } else {
4665 Unimplemented();
4666 }
4667 }
4668
4669 void MacroAssembler::cmov32(Condition cc, Register dst, Address src) {
4670 if (VM_Version::supports_cmov()) {
4671 cmovl(cc, dst, src);
4672 } else {
4673 Label L;
4674 jccb(negate_condition(cc), L);
4675 movl(dst, src);
4676 bind(L);
4677 }
4678 }
4679
4680 void MacroAssembler::cmov32(Condition cc, Register dst, Register src) {
4681 if (VM_Version::supports_cmov()) {
4682 cmovl(cc, dst, src);
4683 } else {
4684 Label L;
4685 jccb(negate_condition(cc), L);
4686 movl(dst, src);
4687 bind(L);
4688 }
4689 }
4690
4691 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
4692 if (!VerifyOops) return;
4693
4694 BLOCK_COMMENT("verify_oop {");
4695 #ifdef _LP64
4696 push(rscratch1);
4697 #endif
4698 push(rax); // save rax
4699 push(reg); // pass register argument
4700
4701 // Pass register number to verify_oop_subroutine
4702 const char* b = nullptr;
4703 {
4704 ResourceMark rm;
4705 stringStream ss;
4706 ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
4707 b = code_string(ss.as_string());
4708 }
4709 ExternalAddress buffer((address) b);
4710 pushptr(buffer.addr(), rscratch1);
4711
4712 // call indirectly to solve generation ordering problem
4713 movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
4714 call(rax);
4715 // Caller pops the arguments (oop, message) and restores rax, r10
4716 BLOCK_COMMENT("} verify_oop");
4717 }
4718
4719 void MacroAssembler::vallones(XMMRegister dst, int vector_len) {
4720 if (UseAVX > 2 && (vector_len == Assembler::AVX_512bit || VM_Version::supports_avx512vl())) {
4721 // Only pcmpeq has dependency breaking treatment (i.e the execution can begin without
4722 // waiting for the previous result on dst), not vpcmpeqd, so just use vpternlog
4723 vpternlogd(dst, 0xFF, dst, dst, vector_len);
4724 } else if (VM_Version::supports_avx()) {
4725 vpcmpeqd(dst, dst, dst, vector_len);
4726 } else {
4727 assert(VM_Version::supports_sse2(), "");
4728 pcmpeqd(dst, dst);
4729 }
4730 }
4731
4732 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
4733 int extra_slot_offset) {
4734 // cf. TemplateTable::prepare_invoke(), if (load_receiver).
4735 int stackElementSize = Interpreter::stackElementSize;
4736 int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
4737 #ifdef ASSERT
4738 int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
4739 assert(offset1 - offset == stackElementSize, "correct arithmetic");
4740 #endif
4741 Register scale_reg = noreg;
4742 Address::ScaleFactor scale_factor = Address::no_scale;
4743 if (arg_slot.is_constant()) {
4744 offset += arg_slot.as_constant() * stackElementSize;
4745 } else {
4746 scale_reg = arg_slot.as_register();
4747 scale_factor = Address::times(stackElementSize);
4748 }
4749 offset += wordSize; // return PC is on stack
4750 return Address(rsp, scale_reg, scale_factor, offset);
4751 }
4752
4753 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
4754 if (!VerifyOops) return;
4755
4756 #ifdef _LP64
4757 push(rscratch1);
4758 #endif
4759 push(rax); // save rax,
4760 // addr may contain rsp so we will have to adjust it based on the push
4761 // we just did (and on 64 bit we do two pushes)
4762 // NOTE: 64bit seemed to have had a bug in that it did movq(addr, rax); which
4763 // stores rax into addr which is backwards of what was intended.
4764 if (addr.uses(rsp)) {
4765 lea(rax, addr);
4766 pushptr(Address(rax, LP64_ONLY(2 *) BytesPerWord));
4767 } else {
4768 pushptr(addr);
4769 }
4770
4771 // Pass register number to verify_oop_subroutine
4772 const char* b = nullptr;
4773 {
4774 ResourceMark rm;
4775 stringStream ss;
4776 ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
4777 b = code_string(ss.as_string());
4778 }
4779 ExternalAddress buffer((address) b);
4780 pushptr(buffer.addr(), rscratch1);
4781
4782 // call indirectly to solve generation ordering problem
4783 movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
4784 call(rax);
4785 // Caller pops the arguments (addr, message) and restores rax, r10.
4786 }
4787
4788 void MacroAssembler::verify_tlab() {
4789 #ifdef ASSERT
4790 if (UseTLAB && VerifyOops) {
4791 Label next, ok;
4792 Register t1 = rsi;
4793 Register thread_reg = NOT_LP64(rbx) LP64_ONLY(r15_thread);
4794
4795 push(t1);
4796 NOT_LP64(push(thread_reg));
4797 NOT_LP64(get_thread(thread_reg));
4798
4799 movptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset())));
4800 cmpptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_start_offset())));
4801 jcc(Assembler::aboveEqual, next);
4802 STOP("assert(top >= start)");
4803 should_not_reach_here();
4804
4805 bind(next);
4806 movptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_end_offset())));
4807 cmpptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset())));
4808 jcc(Assembler::aboveEqual, ok);
4809 STOP("assert(top <= end)");
4810 should_not_reach_here();
4811
4812 bind(ok);
4813 NOT_LP64(pop(thread_reg));
4814 pop(t1);
4815 }
4816 #endif
4817 }
4818
4819 class ControlWord {
4820 public:
4821 int32_t _value;
4822
4823 int rounding_control() const { return (_value >> 10) & 3 ; }
4824 int precision_control() const { return (_value >> 8) & 3 ; }
4825 bool precision() const { return ((_value >> 5) & 1) != 0; }
4826 bool underflow() const { return ((_value >> 4) & 1) != 0; }
4827 bool overflow() const { return ((_value >> 3) & 1) != 0; }
4828 bool zero_divide() const { return ((_value >> 2) & 1) != 0; }
4829 bool denormalized() const { return ((_value >> 1) & 1) != 0; }
4830 bool invalid() const { return ((_value >> 0) & 1) != 0; }
4831
4832 void print() const {
4833 // rounding control
4834 const char* rc;
4835 switch (rounding_control()) {
4836 case 0: rc = "round near"; break;
4837 case 1: rc = "round down"; break;
4838 case 2: rc = "round up "; break;
4839 case 3: rc = "chop "; break;
4840 default:
4841 rc = nullptr; // silence compiler warnings
4842 fatal("Unknown rounding control: %d", rounding_control());
4843 };
4844 // precision control
4845 const char* pc;
4846 switch (precision_control()) {
4847 case 0: pc = "24 bits "; break;
4848 case 1: pc = "reserved"; break;
4849 case 2: pc = "53 bits "; break;
4850 case 3: pc = "64 bits "; break;
4851 default:
4852 pc = nullptr; // silence compiler warnings
4853 fatal("Unknown precision control: %d", precision_control());
4854 };
4855 // flags
4856 char f[9];
4857 f[0] = ' ';
4858 f[1] = ' ';
4859 f[2] = (precision ()) ? 'P' : 'p';
4860 f[3] = (underflow ()) ? 'U' : 'u';
4861 f[4] = (overflow ()) ? 'O' : 'o';
4862 f[5] = (zero_divide ()) ? 'Z' : 'z';
4863 f[6] = (denormalized()) ? 'D' : 'd';
4864 f[7] = (invalid ()) ? 'I' : 'i';
4865 f[8] = '\x0';
4866 // output
4867 printf("%04x masks = %s, %s, %s", _value & 0xFFFF, f, rc, pc);
4868 }
4869
4870 };
4871
4872 class StatusWord {
4873 public:
4874 int32_t _value;
4875
4876 bool busy() const { return ((_value >> 15) & 1) != 0; }
4877 bool C3() const { return ((_value >> 14) & 1) != 0; }
4878 bool C2() const { return ((_value >> 10) & 1) != 0; }
4879 bool C1() const { return ((_value >> 9) & 1) != 0; }
4880 bool C0() const { return ((_value >> 8) & 1) != 0; }
4881 int top() const { return (_value >> 11) & 7 ; }
4882 bool error_status() const { return ((_value >> 7) & 1) != 0; }
4883 bool stack_fault() const { return ((_value >> 6) & 1) != 0; }
4884 bool precision() const { return ((_value >> 5) & 1) != 0; }
4885 bool underflow() const { return ((_value >> 4) & 1) != 0; }
4886 bool overflow() const { return ((_value >> 3) & 1) != 0; }
4887 bool zero_divide() const { return ((_value >> 2) & 1) != 0; }
4888 bool denormalized() const { return ((_value >> 1) & 1) != 0; }
4889 bool invalid() const { return ((_value >> 0) & 1) != 0; }
4890
4891 void print() const {
4892 // condition codes
4893 char c[5];
4894 c[0] = (C3()) ? '3' : '-';
4895 c[1] = (C2()) ? '2' : '-';
4896 c[2] = (C1()) ? '1' : '-';
4897 c[3] = (C0()) ? '0' : '-';
4898 c[4] = '\x0';
4899 // flags
4900 char f[9];
4901 f[0] = (error_status()) ? 'E' : '-';
4902 f[1] = (stack_fault ()) ? 'S' : '-';
4903 f[2] = (precision ()) ? 'P' : '-';
4904 f[3] = (underflow ()) ? 'U' : '-';
4905 f[4] = (overflow ()) ? 'O' : '-';
4906 f[5] = (zero_divide ()) ? 'Z' : '-';
4907 f[6] = (denormalized()) ? 'D' : '-';
4908 f[7] = (invalid ()) ? 'I' : '-';
4909 f[8] = '\x0';
4910 // output
4911 printf("%04x flags = %s, cc = %s, top = %d", _value & 0xFFFF, f, c, top());
4912 }
4913
4914 };
4915
4916 class TagWord {
4917 public:
4918 int32_t _value;
4919
4920 int tag_at(int i) const { return (_value >> (i*2)) & 3; }
4921
4922 void print() const {
4923 printf("%04x", _value & 0xFFFF);
4924 }
4925
4926 };
4927
4928 class FPU_Register {
4929 public:
4930 int32_t _m0;
4931 int32_t _m1;
4932 int16_t _ex;
4933
4934 bool is_indefinite() const {
4935 return _ex == -1 && _m1 == (int32_t)0xC0000000 && _m0 == 0;
4936 }
4937
4938 void print() const {
4939 char sign = (_ex < 0) ? '-' : '+';
4940 const char* kind = (_ex == 0x7FFF || _ex == (int16_t)-1) ? "NaN" : " ";
4941 printf("%c%04hx.%08x%08x %s", sign, _ex, _m1, _m0, kind);
4942 };
4943
4944 };
4945
4946 class FPU_State {
4947 public:
4948 enum {
4949 register_size = 10,
4950 number_of_registers = 8,
4951 register_mask = 7
4952 };
4953
4954 ControlWord _control_word;
4955 StatusWord _status_word;
4956 TagWord _tag_word;
4957 int32_t _error_offset;
4958 int32_t _error_selector;
4959 int32_t _data_offset;
4960 int32_t _data_selector;
4961 int8_t _register[register_size * number_of_registers];
4962
4963 int tag_for_st(int i) const { return _tag_word.tag_at((_status_word.top() + i) & register_mask); }
4964 FPU_Register* st(int i) const { return (FPU_Register*)&_register[register_size * i]; }
4965
4966 const char* tag_as_string(int tag) const {
4967 switch (tag) {
4968 case 0: return "valid";
4969 case 1: return "zero";
4970 case 2: return "special";
4971 case 3: return "empty";
4972 }
4973 ShouldNotReachHere();
4974 return nullptr;
4975 }
4976
4977 void print() const {
4978 // print computation registers
4979 { int t = _status_word.top();
4980 for (int i = 0; i < number_of_registers; i++) {
4981 int j = (i - t) & register_mask;
4982 printf("%c r%d = ST%d = ", (j == 0 ? '*' : ' '), i, j);
4983 st(j)->print();
4984 printf(" %s\n", tag_as_string(_tag_word.tag_at(i)));
4985 }
4986 }
4987 printf("\n");
4988 // print control registers
4989 printf("ctrl = "); _control_word.print(); printf("\n");
4990 printf("stat = "); _status_word .print(); printf("\n");
4991 printf("tags = "); _tag_word .print(); printf("\n");
4992 }
4993
4994 };
4995
4996 class Flag_Register {
4997 public:
4998 int32_t _value;
4999
5000 bool overflow() const { return ((_value >> 11) & 1) != 0; }
5001 bool direction() const { return ((_value >> 10) & 1) != 0; }
5002 bool sign() const { return ((_value >> 7) & 1) != 0; }
5003 bool zero() const { return ((_value >> 6) & 1) != 0; }
5004 bool auxiliary_carry() const { return ((_value >> 4) & 1) != 0; }
5005 bool parity() const { return ((_value >> 2) & 1) != 0; }
5006 bool carry() const { return ((_value >> 0) & 1) != 0; }
5007
5008 void print() const {
5009 // flags
5010 char f[8];
5011 f[0] = (overflow ()) ? 'O' : '-';
5012 f[1] = (direction ()) ? 'D' : '-';
5013 f[2] = (sign ()) ? 'S' : '-';
5014 f[3] = (zero ()) ? 'Z' : '-';
5015 f[4] = (auxiliary_carry()) ? 'A' : '-';
5016 f[5] = (parity ()) ? 'P' : '-';
5017 f[6] = (carry ()) ? 'C' : '-';
5018 f[7] = '\x0';
5019 // output
5020 printf("%08x flags = %s", _value, f);
5021 }
5022
5023 };
5024
5025 class IU_Register {
5026 public:
5027 int32_t _value;
5028
5029 void print() const {
5030 printf("%08x %11d", _value, _value);
5031 }
5032
5033 };
5034
5035 class IU_State {
5036 public:
5037 Flag_Register _eflags;
5038 IU_Register _rdi;
5039 IU_Register _rsi;
5040 IU_Register _rbp;
5041 IU_Register _rsp;
5042 IU_Register _rbx;
5043 IU_Register _rdx;
5044 IU_Register _rcx;
5045 IU_Register _rax;
5046
5047 void print() const {
5048 // computation registers
5049 printf("rax, = "); _rax.print(); printf("\n");
5050 printf("rbx, = "); _rbx.print(); printf("\n");
5051 printf("rcx = "); _rcx.print(); printf("\n");
5052 printf("rdx = "); _rdx.print(); printf("\n");
5053 printf("rdi = "); _rdi.print(); printf("\n");
5054 printf("rsi = "); _rsi.print(); printf("\n");
5055 printf("rbp, = "); _rbp.print(); printf("\n");
5056 printf("rsp = "); _rsp.print(); printf("\n");
5057 printf("\n");
5058 // control registers
5059 printf("flgs = "); _eflags.print(); printf("\n");
5060 }
5061 };
5062
5063
5064 class CPU_State {
5065 public:
5066 FPU_State _fpu_state;
5067 IU_State _iu_state;
5068
5069 void print() const {
5070 printf("--------------------------------------------------\n");
5071 _iu_state .print();
5072 printf("\n");
5073 _fpu_state.print();
5074 printf("--------------------------------------------------\n");
5075 }
5076
5077 };
5078
5079
5080 static void _print_CPU_state(CPU_State* state) {
5081 state->print();
5082 };
5083
5084
5085 void MacroAssembler::print_CPU_state() {
5086 push_CPU_state();
5087 push(rsp); // pass CPU state
5088 call(RuntimeAddress(CAST_FROM_FN_PTR(address, _print_CPU_state)));
5089 addptr(rsp, wordSize); // discard argument
5090 pop_CPU_state();
5091 }
5092
5093
5094 #ifndef _LP64
5095 static bool _verify_FPU(int stack_depth, char* s, CPU_State* state) {
5096 static int counter = 0;
5097 FPU_State* fs = &state->_fpu_state;
5098 counter++;
5099 // For leaf calls, only verify that the top few elements remain empty.
5100 // We only need 1 empty at the top for C2 code.
5101 if( stack_depth < 0 ) {
5102 if( fs->tag_for_st(7) != 3 ) {
5103 printf("FPR7 not empty\n");
5104 state->print();
5105 assert(false, "error");
5106 return false;
5107 }
5108 return true; // All other stack states do not matter
5109 }
5110
5111 assert((fs->_control_word._value & 0xffff) == StubRoutines::x86::fpu_cntrl_wrd_std(),
5112 "bad FPU control word");
5113
5114 // compute stack depth
5115 int i = 0;
5116 while (i < FPU_State::number_of_registers && fs->tag_for_st(i) < 3) i++;
5117 int d = i;
5118 while (i < FPU_State::number_of_registers && fs->tag_for_st(i) == 3) i++;
5119 // verify findings
5120 if (i != FPU_State::number_of_registers) {
5121 // stack not contiguous
5122 printf("%s: stack not contiguous at ST%d\n", s, i);
5123 state->print();
5124 assert(false, "error");
5125 return false;
5126 }
5127 // check if computed stack depth corresponds to expected stack depth
5128 if (stack_depth < 0) {
5129 // expected stack depth is -stack_depth or less
5130 if (d > -stack_depth) {
5131 // too many elements on the stack
5132 printf("%s: <= %d stack elements expected but found %d\n", s, -stack_depth, d);
5133 state->print();
5134 assert(false, "error");
5135 return false;
5136 }
5137 } else {
5138 // expected stack depth is stack_depth
5139 if (d != stack_depth) {
5140 // wrong stack depth
5141 printf("%s: %d stack elements expected but found %d\n", s, stack_depth, d);
5142 state->print();
5143 assert(false, "error");
5144 return false;
5145 }
5146 }
5147 // everything is cool
5148 return true;
5149 }
5150
5151 void MacroAssembler::verify_FPU(int stack_depth, const char* s) {
5152 if (!VerifyFPU) return;
5153 push_CPU_state();
5154 push(rsp); // pass CPU state
5155 ExternalAddress msg((address) s);
5156 // pass message string s
5157 pushptr(msg.addr(), noreg);
5158 push(stack_depth); // pass stack depth
5159 call(RuntimeAddress(CAST_FROM_FN_PTR(address, _verify_FPU)));
5160 addptr(rsp, 3 * wordSize); // discard arguments
5161 // check for error
5162 { Label L;
5163 testl(rax, rax);
5164 jcc(Assembler::notZero, L);
5165 int3(); // break if error condition
5166 bind(L);
5167 }
5168 pop_CPU_state();
5169 }
5170 #endif // _LP64
5171
5172 void MacroAssembler::restore_cpu_control_state_after_jni(Register rscratch) {
5173 // Either restore the MXCSR register after returning from the JNI Call
5174 // or verify that it wasn't changed (with -Xcheck:jni flag).
5175 if (VM_Version::supports_sse()) {
5176 if (RestoreMXCSROnJNICalls) {
5177 ldmxcsr(ExternalAddress(StubRoutines::x86::addr_mxcsr_std()), rscratch);
5178 } else if (CheckJNICalls) {
5179 call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
5180 }
5181 }
5182 // Clear upper bits of YMM registers to avoid SSE <-> AVX transition penalty.
5183 vzeroupper();
5184
5185 #ifndef _LP64
5186 // Either restore the x87 floating pointer control word after returning
5187 // from the JNI call or verify that it wasn't changed.
5188 if (CheckJNICalls) {
5189 call(RuntimeAddress(StubRoutines::x86::verify_fpu_cntrl_wrd_entry()));
5190 }
5191 #endif // _LP64
5192 }
5193
5194 // ((OopHandle)result).resolve();
5195 void MacroAssembler::resolve_oop_handle(Register result, Register tmp) {
5196 assert_different_registers(result, tmp);
5197
5198 // Only 64 bit platforms support GCs that require a tmp register
5199 // Only IN_HEAP loads require a thread_tmp register
5200 // OopHandle::resolve is an indirection like jobject.
5201 access_load_at(T_OBJECT, IN_NATIVE,
5202 result, Address(result, 0), tmp, /*tmp_thread*/noreg);
5203 }
5204
5205 // ((WeakHandle)result).resolve();
5206 void MacroAssembler::resolve_weak_handle(Register rresult, Register rtmp) {
5207 assert_different_registers(rresult, rtmp);
5208 Label resolved;
5209
5210 // A null weak handle resolves to null.
5211 cmpptr(rresult, 0);
5212 jcc(Assembler::equal, resolved);
5213
5214 // Only 64 bit platforms support GCs that require a tmp register
5215 // Only IN_HEAP loads require a thread_tmp register
5216 // WeakHandle::resolve is an indirection like jweak.
5217 access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
5218 rresult, Address(rresult, 0), rtmp, /*tmp_thread*/noreg);
5219 bind(resolved);
5220 }
5221
5222 void MacroAssembler::load_mirror(Register mirror, Register method, Register tmp) {
5223 // get mirror
5224 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
5225 load_method_holder(mirror, method);
5226 movptr(mirror, Address(mirror, mirror_offset));
5227 resolve_oop_handle(mirror, tmp);
5228 }
5229
5230 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5231 load_method_holder(rresult, rmethod);
5232 movptr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5233 }
5234
5235 void MacroAssembler::load_method_holder(Register holder, Register method) {
5236 movptr(holder, Address(method, Method::const_offset())); // ConstMethod*
5237 movptr(holder, Address(holder, ConstMethod::constants_offset())); // ConstantPool*
5238 movptr(holder, Address(holder, ConstantPool::pool_holder_offset())); // InstanceKlass*
5239 }
5240
5241 void MacroAssembler::load_klass(Register dst, Register src, Register tmp) {
5242 assert_different_registers(src, tmp);
5243 assert_different_registers(dst, tmp);
5244 #ifdef _LP64
5245 if (UseCompressedClassPointers) {
5246 movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5247 decode_klass_not_null(dst, tmp);
5248 } else
5249 #endif
5250 movptr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5251 }
5252
5253 void MacroAssembler::store_klass(Register dst, Register src, Register tmp) {
5254 assert_different_registers(src, tmp);
5255 assert_different_registers(dst, tmp);
5256 #ifdef _LP64
5257 if (UseCompressedClassPointers) {
5258 encode_klass_not_null(src, tmp);
5259 movl(Address(dst, oopDesc::klass_offset_in_bytes()), src);
5260 } else
5261 #endif
5262 movptr(Address(dst, oopDesc::klass_offset_in_bytes()), src);
5263 }
5264
5265 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
5266 Register tmp1, Register thread_tmp) {
5267 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
5268 decorators = AccessInternal::decorator_fixup(decorators, type);
5269 bool as_raw = (decorators & AS_RAW) != 0;
5270 if (as_raw) {
5271 bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, thread_tmp);
5272 } else {
5273 bs->load_at(this, decorators, type, dst, src, tmp1, thread_tmp);
5274 }
5275 }
5276
5277 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
5278 Register tmp1, Register tmp2, Register tmp3) {
5279 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
5280 decorators = AccessInternal::decorator_fixup(decorators, type);
5281 bool as_raw = (decorators & AS_RAW) != 0;
5282 if (as_raw) {
5283 bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5284 } else {
5285 bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5286 }
5287 }
5288
5289 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
5290 Register thread_tmp, DecoratorSet decorators) {
5291 access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, thread_tmp);
5292 }
5293
5294 // Doesn't do verification, generates fixed size code
5295 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
5296 Register thread_tmp, DecoratorSet decorators) {
5297 access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1, thread_tmp);
5298 }
5299
5300 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5301 Register tmp2, Register tmp3, DecoratorSet decorators) {
5302 access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5303 }
5304
5305 // Used for storing nulls.
5306 void MacroAssembler::store_heap_oop_null(Address dst) {
5307 access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5308 }
5309
5310 #ifdef _LP64
5311 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5312 if (UseCompressedClassPointers) {
5313 // Store to klass gap in destination
5314 movl(Address(dst, oopDesc::klass_gap_offset_in_bytes()), src);
5315 }
5316 }
5317
5318 #ifdef ASSERT
5319 void MacroAssembler::verify_heapbase(const char* msg) {
5320 assert (UseCompressedOops, "should be compressed");
5321 assert (Universe::heap() != nullptr, "java heap should be initialized");
5322 if (CheckCompressedOops) {
5323 Label ok;
5324 ExternalAddress src2(CompressedOops::ptrs_base_addr());
5325 const bool is_src2_reachable = reachable(src2);
5326 if (!is_src2_reachable) {
5327 push(rscratch1); // cmpptr trashes rscratch1
5328 }
5329 cmpptr(r12_heapbase, src2, rscratch1);
5330 jcc(Assembler::equal, ok);
5331 STOP(msg);
5332 bind(ok);
5333 if (!is_src2_reachable) {
5334 pop(rscratch1);
5335 }
5336 }
5337 }
5338 #endif
5339
5340 // Algorithm must match oop.inline.hpp encode_heap_oop.
5341 void MacroAssembler::encode_heap_oop(Register r) {
5342 #ifdef ASSERT
5343 verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5344 #endif
5345 verify_oop_msg(r, "broken oop in encode_heap_oop");
5346 if (CompressedOops::base() == nullptr) {
5347 if (CompressedOops::shift() != 0) {
5348 assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5349 shrq(r, LogMinObjAlignmentInBytes);
5350 }
5351 return;
5352 }
5353 testq(r, r);
5354 cmovq(Assembler::equal, r, r12_heapbase);
5355 subq(r, r12_heapbase);
5356 shrq(r, LogMinObjAlignmentInBytes);
5357 }
5358
5359 void MacroAssembler::encode_heap_oop_not_null(Register r) {
5360 #ifdef ASSERT
5361 verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
5362 if (CheckCompressedOops) {
5363 Label ok;
5364 testq(r, r);
5365 jcc(Assembler::notEqual, ok);
5366 STOP("null oop passed to encode_heap_oop_not_null");
5367 bind(ok);
5368 }
5369 #endif
5370 verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
5371 if (CompressedOops::base() != nullptr) {
5372 subq(r, r12_heapbase);
5373 }
5374 if (CompressedOops::shift() != 0) {
5375 assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5376 shrq(r, LogMinObjAlignmentInBytes);
5377 }
5378 }
5379
5380 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
5381 #ifdef ASSERT
5382 verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
5383 if (CheckCompressedOops) {
5384 Label ok;
5385 testq(src, src);
5386 jcc(Assembler::notEqual, ok);
5387 STOP("null oop passed to encode_heap_oop_not_null2");
5388 bind(ok);
5389 }
5390 #endif
5391 verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
5392 if (dst != src) {
5393 movq(dst, src);
5394 }
5395 if (CompressedOops::base() != nullptr) {
5396 subq(dst, r12_heapbase);
5397 }
5398 if (CompressedOops::shift() != 0) {
5399 assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5400 shrq(dst, LogMinObjAlignmentInBytes);
5401 }
5402 }
5403
5404 void MacroAssembler::decode_heap_oop(Register r) {
5405 #ifdef ASSERT
5406 verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
5407 #endif
5408 if (CompressedOops::base() == nullptr) {
5409 if (CompressedOops::shift() != 0) {
5410 assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5411 shlq(r, LogMinObjAlignmentInBytes);
5412 }
5413 } else {
5414 Label done;
5415 shlq(r, LogMinObjAlignmentInBytes);
5416 jccb(Assembler::equal, done);
5417 addq(r, r12_heapbase);
5418 bind(done);
5419 }
5420 verify_oop_msg(r, "broken oop in decode_heap_oop");
5421 }
5422
5423 void MacroAssembler::decode_heap_oop_not_null(Register r) {
5424 // Note: it will change flags
5425 assert (UseCompressedOops, "should only be used for compressed headers");
5426 assert (Universe::heap() != nullptr, "java heap should be initialized");
5427 // Cannot assert, unverified entry point counts instructions (see .ad file)
5428 // vtableStubs also counts instructions in pd_code_size_limit.
5429 // Also do not verify_oop as this is called by verify_oop.
5430 if (CompressedOops::shift() != 0) {
5431 assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5432 shlq(r, LogMinObjAlignmentInBytes);
5433 if (CompressedOops::base() != nullptr) {
5434 addq(r, r12_heapbase);
5435 }
5436 } else {
5437 assert (CompressedOops::base() == nullptr, "sanity");
5438 }
5439 }
5440
5441 void MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
5442 // Note: it will change flags
5443 assert (UseCompressedOops, "should only be used for compressed headers");
5444 assert (Universe::heap() != nullptr, "java heap should be initialized");
5445 // Cannot assert, unverified entry point counts instructions (see .ad file)
5446 // vtableStubs also counts instructions in pd_code_size_limit.
5447 // Also do not verify_oop as this is called by verify_oop.
5448 if (CompressedOops::shift() != 0) {
5449 assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5450 if (LogMinObjAlignmentInBytes == Address::times_8) {
5451 leaq(dst, Address(r12_heapbase, src, Address::times_8, 0));
5452 } else {
5453 if (dst != src) {
5454 movq(dst, src);
5455 }
5456 shlq(dst, LogMinObjAlignmentInBytes);
5457 if (CompressedOops::base() != nullptr) {
5458 addq(dst, r12_heapbase);
5459 }
5460 }
5461 } else {
5462 assert (CompressedOops::base() == nullptr, "sanity");
5463 if (dst != src) {
5464 movq(dst, src);
5465 }
5466 }
5467 }
5468
5469 void MacroAssembler::encode_klass_not_null(Register r, Register tmp) {
5470 assert_different_registers(r, tmp);
5471 if (CompressedKlassPointers::base() != nullptr) {
5472 mov64(tmp, (int64_t)CompressedKlassPointers::base());
5473 subq(r, tmp);
5474 }
5475 if (CompressedKlassPointers::shift() != 0) {
5476 assert (LogKlassAlignmentInBytes == CompressedKlassPointers::shift(), "decode alg wrong");
5477 shrq(r, LogKlassAlignmentInBytes);
5478 }
5479 }
5480
5481 void MacroAssembler::encode_and_move_klass_not_null(Register dst, Register src) {
5482 assert_different_registers(src, dst);
5483 if (CompressedKlassPointers::base() != nullptr) {
5484 mov64(dst, -(int64_t)CompressedKlassPointers::base());
5485 addq(dst, src);
5486 } else {
5487 movptr(dst, src);
5488 }
5489 if (CompressedKlassPointers::shift() != 0) {
5490 assert (LogKlassAlignmentInBytes == CompressedKlassPointers::shift(), "decode alg wrong");
5491 shrq(dst, LogKlassAlignmentInBytes);
5492 }
5493 }
5494
5495 void MacroAssembler::decode_klass_not_null(Register r, Register tmp) {
5496 assert_different_registers(r, tmp);
5497 // Note: it will change flags
5498 assert(UseCompressedClassPointers, "should only be used for compressed headers");
5499 // Cannot assert, unverified entry point counts instructions (see .ad file)
5500 // vtableStubs also counts instructions in pd_code_size_limit.
5501 // Also do not verify_oop as this is called by verify_oop.
5502 if (CompressedKlassPointers::shift() != 0) {
5503 assert(LogKlassAlignmentInBytes == CompressedKlassPointers::shift(), "decode alg wrong");
5504 shlq(r, LogKlassAlignmentInBytes);
5505 }
5506 if (CompressedKlassPointers::base() != nullptr) {
5507 mov64(tmp, (int64_t)CompressedKlassPointers::base());
5508 addq(r, tmp);
5509 }
5510 }
5511
5512 void MacroAssembler::decode_and_move_klass_not_null(Register dst, Register src) {
5513 assert_different_registers(src, dst);
5514 // Note: it will change flags
5515 assert (UseCompressedClassPointers, "should only be used for compressed headers");
5516 // Cannot assert, unverified entry point counts instructions (see .ad file)
5517 // vtableStubs also counts instructions in pd_code_size_limit.
5518 // Also do not verify_oop as this is called by verify_oop.
5519
5520 if (CompressedKlassPointers::base() == nullptr &&
5521 CompressedKlassPointers::shift() == 0) {
5522 // The best case scenario is that there is no base or shift. Then it is already
5523 // a pointer that needs nothing but a register rename.
5524 movl(dst, src);
5525 } else {
5526 if (CompressedKlassPointers::base() != nullptr) {
5527 mov64(dst, (int64_t)CompressedKlassPointers::base());
5528 } else {
5529 xorq(dst, dst);
5530 }
5531 if (CompressedKlassPointers::shift() != 0) {
5532 assert(LogKlassAlignmentInBytes == CompressedKlassPointers::shift(), "decode alg wrong");
5533 assert(LogKlassAlignmentInBytes == Address::times_8, "klass not aligned on 64bits?");
5534 leaq(dst, Address(dst, src, Address::times_8, 0));
5535 } else {
5536 addq(dst, src);
5537 }
5538 }
5539 }
5540
5541 void MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
5542 assert (UseCompressedOops, "should only be used for compressed headers");
5543 assert (Universe::heap() != nullptr, "java heap should be initialized");
5544 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5545 int oop_index = oop_recorder()->find_index(obj);
5546 RelocationHolder rspec = oop_Relocation::spec(oop_index);
5547 mov_narrow_oop(dst, oop_index, rspec);
5548 }
5549
5550 void MacroAssembler::set_narrow_oop(Address dst, jobject obj) {
5551 assert (UseCompressedOops, "should only be used for compressed headers");
5552 assert (Universe::heap() != nullptr, "java heap should be initialized");
5553 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5554 int oop_index = oop_recorder()->find_index(obj);
5555 RelocationHolder rspec = oop_Relocation::spec(oop_index);
5556 mov_narrow_oop(dst, oop_index, rspec);
5557 }
5558
5559 void MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
5560 assert (UseCompressedClassPointers, "should only be used for compressed headers");
5561 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5562 int klass_index = oop_recorder()->find_index(k);
5563 RelocationHolder rspec = metadata_Relocation::spec(klass_index);
5564 mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
5565 }
5566
5567 void MacroAssembler::set_narrow_klass(Address dst, Klass* k) {
5568 assert (UseCompressedClassPointers, "should only be used for compressed headers");
5569 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5570 int klass_index = oop_recorder()->find_index(k);
5571 RelocationHolder rspec = metadata_Relocation::spec(klass_index);
5572 mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
5573 }
5574
5575 void MacroAssembler::cmp_narrow_oop(Register dst, jobject obj) {
5576 assert (UseCompressedOops, "should only be used for compressed headers");
5577 assert (Universe::heap() != nullptr, "java heap should be initialized");
5578 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5579 int oop_index = oop_recorder()->find_index(obj);
5580 RelocationHolder rspec = oop_Relocation::spec(oop_index);
5581 Assembler::cmp_narrow_oop(dst, oop_index, rspec);
5582 }
5583
5584 void MacroAssembler::cmp_narrow_oop(Address dst, jobject obj) {
5585 assert (UseCompressedOops, "should only be used for compressed headers");
5586 assert (Universe::heap() != nullptr, "java heap should be initialized");
5587 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5588 int oop_index = oop_recorder()->find_index(obj);
5589 RelocationHolder rspec = oop_Relocation::spec(oop_index);
5590 Assembler::cmp_narrow_oop(dst, oop_index, rspec);
5591 }
5592
5593 void MacroAssembler::cmp_narrow_klass(Register dst, Klass* k) {
5594 assert (UseCompressedClassPointers, "should only be used for compressed headers");
5595 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5596 int klass_index = oop_recorder()->find_index(k);
5597 RelocationHolder rspec = metadata_Relocation::spec(klass_index);
5598 Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
5599 }
5600
5601 void MacroAssembler::cmp_narrow_klass(Address dst, Klass* k) {
5602 assert (UseCompressedClassPointers, "should only be used for compressed headers");
5603 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5604 int klass_index = oop_recorder()->find_index(k);
5605 RelocationHolder rspec = metadata_Relocation::spec(klass_index);
5606 Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
5607 }
5608
5609 void MacroAssembler::reinit_heapbase() {
5610 if (UseCompressedOops) {
5611 if (Universe::heap() != nullptr) {
5612 if (CompressedOops::base() == nullptr) {
5613 MacroAssembler::xorptr(r12_heapbase, r12_heapbase);
5614 } else {
5615 mov64(r12_heapbase, (int64_t)CompressedOops::ptrs_base());
5616 }
5617 } else {
5618 movptr(r12_heapbase, ExternalAddress(CompressedOops::ptrs_base_addr()));
5619 }
5620 }
5621 }
5622
5623 #endif // _LP64
5624
5625 #if COMPILER2_OR_JVMCI
5626
5627 // clear memory of size 'cnt' qwords, starting at 'base' using XMM/YMM/ZMM registers
5628 void MacroAssembler::xmm_clear_mem(Register base, Register cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
5629 // cnt - number of qwords (8-byte words).
5630 // base - start address, qword aligned.
5631 Label L_zero_64_bytes, L_loop, L_sloop, L_tail, L_end;
5632 bool use64byteVector = (MaxVectorSize == 64) && (VM_Version::avx3_threshold() == 0);
5633 if (use64byteVector) {
5634 vpxor(xtmp, xtmp, xtmp, AVX_512bit);
5635 } else if (MaxVectorSize >= 32) {
5636 vpxor(xtmp, xtmp, xtmp, AVX_256bit);
5637 } else {
5638 pxor(xtmp, xtmp);
5639 }
5640 jmp(L_zero_64_bytes);
5641
5642 BIND(L_loop);
5643 if (MaxVectorSize >= 32) {
5644 fill64(base, 0, xtmp, use64byteVector);
5645 } else {
5646 movdqu(Address(base, 0), xtmp);
5647 movdqu(Address(base, 16), xtmp);
5648 movdqu(Address(base, 32), xtmp);
5649 movdqu(Address(base, 48), xtmp);
5650 }
5651 addptr(base, 64);
5652
5653 BIND(L_zero_64_bytes);
5654 subptr(cnt, 8);
5655 jccb(Assembler::greaterEqual, L_loop);
5656
5657 // Copy trailing 64 bytes
5658 if (use64byteVector) {
5659 addptr(cnt, 8);
5660 jccb(Assembler::equal, L_end);
5661 fill64_masked(3, base, 0, xtmp, mask, cnt, rtmp, true);
5662 jmp(L_end);
5663 } else {
5664 addptr(cnt, 4);
5665 jccb(Assembler::less, L_tail);
5666 if (MaxVectorSize >= 32) {
5667 vmovdqu(Address(base, 0), xtmp);
5668 } else {
5669 movdqu(Address(base, 0), xtmp);
5670 movdqu(Address(base, 16), xtmp);
5671 }
5672 }
5673 addptr(base, 32);
5674 subptr(cnt, 4);
5675
5676 BIND(L_tail);
5677 addptr(cnt, 4);
5678 jccb(Assembler::lessEqual, L_end);
5679 if (UseAVX > 2 && MaxVectorSize >= 32 && VM_Version::supports_avx512vl()) {
5680 fill32_masked(3, base, 0, xtmp, mask, cnt, rtmp);
5681 } else {
5682 decrement(cnt);
5683
5684 BIND(L_sloop);
5685 movq(Address(base, 0), xtmp);
5686 addptr(base, 8);
5687 decrement(cnt);
5688 jccb(Assembler::greaterEqual, L_sloop);
5689 }
5690 BIND(L_end);
5691 }
5692
5693 // Clearing constant sized memory using YMM/ZMM registers.
5694 void MacroAssembler::clear_mem(Register base, int cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
5695 assert(UseAVX > 2 && VM_Version::supports_avx512vlbw(), "");
5696 bool use64byteVector = (MaxVectorSize > 32) && (VM_Version::avx3_threshold() == 0);
5697
5698 int vector64_count = (cnt & (~0x7)) >> 3;
5699 cnt = cnt & 0x7;
5700 const int fill64_per_loop = 4;
5701 const int max_unrolled_fill64 = 8;
5702
5703 // 64 byte initialization loop.
5704 vpxor(xtmp, xtmp, xtmp, use64byteVector ? AVX_512bit : AVX_256bit);
5705 int start64 = 0;
5706 if (vector64_count > max_unrolled_fill64) {
5707 Label LOOP;
5708 Register index = rtmp;
5709
5710 start64 = vector64_count - (vector64_count % fill64_per_loop);
5711
5712 movl(index, 0);
5713 BIND(LOOP);
5714 for (int i = 0; i < fill64_per_loop; i++) {
5715 fill64(Address(base, index, Address::times_1, i * 64), xtmp, use64byteVector);
5716 }
5717 addl(index, fill64_per_loop * 64);
5718 cmpl(index, start64 * 64);
5719 jccb(Assembler::less, LOOP);
5720 }
5721 for (int i = start64; i < vector64_count; i++) {
5722 fill64(base, i * 64, xtmp, use64byteVector);
5723 }
5724
5725 // Clear remaining 64 byte tail.
5726 int disp = vector64_count * 64;
5727 if (cnt) {
5728 switch (cnt) {
5729 case 1:
5730 movq(Address(base, disp), xtmp);
5731 break;
5732 case 2:
5733 evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_128bit);
5734 break;
5735 case 3:
5736 movl(rtmp, 0x7);
5737 kmovwl(mask, rtmp);
5738 evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_256bit);
5739 break;
5740 case 4:
5741 evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
5742 break;
5743 case 5:
5744 if (use64byteVector) {
5745 movl(rtmp, 0x1F);
5746 kmovwl(mask, rtmp);
5747 evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
5748 } else {
5749 evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
5750 movq(Address(base, disp + 32), xtmp);
5751 }
5752 break;
5753 case 6:
5754 if (use64byteVector) {
5755 movl(rtmp, 0x3F);
5756 kmovwl(mask, rtmp);
5757 evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
5758 } else {
5759 evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
5760 evmovdqu(T_LONG, k0, Address(base, disp + 32), xtmp, false, Assembler::AVX_128bit);
5761 }
5762 break;
5763 case 7:
5764 if (use64byteVector) {
5765 movl(rtmp, 0x7F);
5766 kmovwl(mask, rtmp);
5767 evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
5768 } else {
5769 evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
5770 movl(rtmp, 0x7);
5771 kmovwl(mask, rtmp);
5772 evmovdqu(T_LONG, mask, Address(base, disp + 32), xtmp, true, Assembler::AVX_256bit);
5773 }
5774 break;
5775 default:
5776 fatal("Unexpected length : %d\n",cnt);
5777 break;
5778 }
5779 }
5780 }
5781
5782 void MacroAssembler::clear_mem(Register base, Register cnt, Register tmp, XMMRegister xtmp,
5783 bool is_large, KRegister mask) {
5784 // cnt - number of qwords (8-byte words).
5785 // base - start address, qword aligned.
5786 // is_large - if optimizers know cnt is larger than InitArrayShortSize
5787 assert(base==rdi, "base register must be edi for rep stos");
5788 assert(tmp==rax, "tmp register must be eax for rep stos");
5789 assert(cnt==rcx, "cnt register must be ecx for rep stos");
5790 assert(InitArrayShortSize % BytesPerLong == 0,
5791 "InitArrayShortSize should be the multiple of BytesPerLong");
5792
5793 Label DONE;
5794 if (!is_large || !UseXMMForObjInit) {
5795 xorptr(tmp, tmp);
5796 }
5797
5798 if (!is_large) {
5799 Label LOOP, LONG;
5800 cmpptr(cnt, InitArrayShortSize/BytesPerLong);
5801 jccb(Assembler::greater, LONG);
5802
5803 NOT_LP64(shlptr(cnt, 1);) // convert to number of 32-bit words for 32-bit VM
5804
5805 decrement(cnt);
5806 jccb(Assembler::negative, DONE); // Zero length
5807
5808 // Use individual pointer-sized stores for small counts:
5809 BIND(LOOP);
5810 movptr(Address(base, cnt, Address::times_ptr), tmp);
5811 decrement(cnt);
5812 jccb(Assembler::greaterEqual, LOOP);
5813 jmpb(DONE);
5814
5815 BIND(LONG);
5816 }
5817
5818 // Use longer rep-prefixed ops for non-small counts:
5819 if (UseFastStosb) {
5820 shlptr(cnt, 3); // convert to number of bytes
5821 rep_stosb();
5822 } else if (UseXMMForObjInit) {
5823 xmm_clear_mem(base, cnt, tmp, xtmp, mask);
5824 } else {
5825 NOT_LP64(shlptr(cnt, 1);) // convert to number of 32-bit words for 32-bit VM
5826 rep_stos();
5827 }
5828
5829 BIND(DONE);
5830 }
5831
5832 #endif //COMPILER2_OR_JVMCI
5833
5834
5835 void MacroAssembler::generate_fill(BasicType t, bool aligned,
5836 Register to, Register value, Register count,
5837 Register rtmp, XMMRegister xtmp) {
5838 ShortBranchVerifier sbv(this);
5839 assert_different_registers(to, value, count, rtmp);
5840 Label L_exit;
5841 Label L_fill_2_bytes, L_fill_4_bytes;
5842
5843 #if defined(COMPILER2) && defined(_LP64)
5844 if(MaxVectorSize >=32 &&
5845 VM_Version::supports_avx512vlbw() &&
5846 VM_Version::supports_bmi2()) {
5847 generate_fill_avx3(t, to, value, count, rtmp, xtmp);
5848 return;
5849 }
5850 #endif
5851
5852 int shift = -1;
5853 switch (t) {
5854 case T_BYTE:
5855 shift = 2;
5856 break;
5857 case T_SHORT:
5858 shift = 1;
5859 break;
5860 case T_INT:
5861 shift = 0;
5862 break;
5863 default: ShouldNotReachHere();
5864 }
5865
5866 if (t == T_BYTE) {
5867 andl(value, 0xff);
5868 movl(rtmp, value);
5869 shll(rtmp, 8);
5870 orl(value, rtmp);
5871 }
5872 if (t == T_SHORT) {
5873 andl(value, 0xffff);
5874 }
5875 if (t == T_BYTE || t == T_SHORT) {
5876 movl(rtmp, value);
5877 shll(rtmp, 16);
5878 orl(value, rtmp);
5879 }
5880
5881 cmpl(count, 2<<shift); // Short arrays (< 8 bytes) fill by element
5882 jcc(Assembler::below, L_fill_4_bytes); // use unsigned cmp
5883 if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
5884 Label L_skip_align2;
5885 // align source address at 4 bytes address boundary
5886 if (t == T_BYTE) {
5887 Label L_skip_align1;
5888 // One byte misalignment happens only for byte arrays
5889 testptr(to, 1);
5890 jccb(Assembler::zero, L_skip_align1);
5891 movb(Address(to, 0), value);
5892 increment(to);
5893 decrement(count);
5894 BIND(L_skip_align1);
5895 }
5896 // Two bytes misalignment happens only for byte and short (char) arrays
5897 testptr(to, 2);
5898 jccb(Assembler::zero, L_skip_align2);
5899 movw(Address(to, 0), value);
5900 addptr(to, 2);
5901 subl(count, 1<<(shift-1));
5902 BIND(L_skip_align2);
5903 }
5904 if (UseSSE < 2) {
5905 Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
5906 // Fill 32-byte chunks
5907 subl(count, 8 << shift);
5908 jcc(Assembler::less, L_check_fill_8_bytes);
5909 align(16);
5910
5911 BIND(L_fill_32_bytes_loop);
5912
5913 for (int i = 0; i < 32; i += 4) {
5914 movl(Address(to, i), value);
5915 }
5916
5917 addptr(to, 32);
5918 subl(count, 8 << shift);
5919 jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
5920 BIND(L_check_fill_8_bytes);
5921 addl(count, 8 << shift);
5922 jccb(Assembler::zero, L_exit);
5923 jmpb(L_fill_8_bytes);
5924
5925 //
5926 // length is too short, just fill qwords
5927 //
5928 BIND(L_fill_8_bytes_loop);
5929 movl(Address(to, 0), value);
5930 movl(Address(to, 4), value);
5931 addptr(to, 8);
5932 BIND(L_fill_8_bytes);
5933 subl(count, 1 << (shift + 1));
5934 jcc(Assembler::greaterEqual, L_fill_8_bytes_loop);
5935 // fall through to fill 4 bytes
5936 } else {
5937 Label L_fill_32_bytes;
5938 if (!UseUnalignedLoadStores) {
5939 // align to 8 bytes, we know we are 4 byte aligned to start
5940 testptr(to, 4);
5941 jccb(Assembler::zero, L_fill_32_bytes);
5942 movl(Address(to, 0), value);
5943 addptr(to, 4);
5944 subl(count, 1<<shift);
5945 }
5946 BIND(L_fill_32_bytes);
5947 {
5948 assert( UseSSE >= 2, "supported cpu only" );
5949 Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
5950 movdl(xtmp, value);
5951 if (UseAVX >= 2 && UseUnalignedLoadStores) {
5952 Label L_check_fill_32_bytes;
5953 if (UseAVX > 2) {
5954 // Fill 64-byte chunks
5955 Label L_fill_64_bytes_loop_avx3, L_check_fill_64_bytes_avx2;
5956
5957 // If number of bytes to fill < VM_Version::avx3_threshold(), perform fill using AVX2
5958 cmpl(count, VM_Version::avx3_threshold());
5959 jccb(Assembler::below, L_check_fill_64_bytes_avx2);
5960
5961 vpbroadcastd(xtmp, xtmp, Assembler::AVX_512bit);
5962
5963 subl(count, 16 << shift);
5964 jccb(Assembler::less, L_check_fill_32_bytes);
5965 align(16);
5966
5967 BIND(L_fill_64_bytes_loop_avx3);
5968 evmovdqul(Address(to, 0), xtmp, Assembler::AVX_512bit);
5969 addptr(to, 64);
5970 subl(count, 16 << shift);
5971 jcc(Assembler::greaterEqual, L_fill_64_bytes_loop_avx3);
5972 jmpb(L_check_fill_32_bytes);
5973
5974 BIND(L_check_fill_64_bytes_avx2);
5975 }
5976 // Fill 64-byte chunks
5977 Label L_fill_64_bytes_loop;
5978 vpbroadcastd(xtmp, xtmp, Assembler::AVX_256bit);
5979
5980 subl(count, 16 << shift);
5981 jcc(Assembler::less, L_check_fill_32_bytes);
5982 align(16);
5983
5984 BIND(L_fill_64_bytes_loop);
5985 vmovdqu(Address(to, 0), xtmp);
5986 vmovdqu(Address(to, 32), xtmp);
5987 addptr(to, 64);
5988 subl(count, 16 << shift);
5989 jcc(Assembler::greaterEqual, L_fill_64_bytes_loop);
5990
5991 BIND(L_check_fill_32_bytes);
5992 addl(count, 8 << shift);
5993 jccb(Assembler::less, L_check_fill_8_bytes);
5994 vmovdqu(Address(to, 0), xtmp);
5995 addptr(to, 32);
5996 subl(count, 8 << shift);
5997
5998 BIND(L_check_fill_8_bytes);
5999 // clean upper bits of YMM registers
6000 movdl(xtmp, value);
6001 pshufd(xtmp, xtmp, 0);
6002 } else {
6003 // Fill 32-byte chunks
6004 pshufd(xtmp, xtmp, 0);
6005
6006 subl(count, 8 << shift);
6007 jcc(Assembler::less, L_check_fill_8_bytes);
6008 align(16);
6009
6010 BIND(L_fill_32_bytes_loop);
6011
6012 if (UseUnalignedLoadStores) {
6013 movdqu(Address(to, 0), xtmp);
6014 movdqu(Address(to, 16), xtmp);
6015 } else {
6016 movq(Address(to, 0), xtmp);
6017 movq(Address(to, 8), xtmp);
6018 movq(Address(to, 16), xtmp);
6019 movq(Address(to, 24), xtmp);
6020 }
6021
6022 addptr(to, 32);
6023 subl(count, 8 << shift);
6024 jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
6025
6026 BIND(L_check_fill_8_bytes);
6027 }
6028 addl(count, 8 << shift);
6029 jccb(Assembler::zero, L_exit);
6030 jmpb(L_fill_8_bytes);
6031
6032 //
6033 // length is too short, just fill qwords
6034 //
6035 BIND(L_fill_8_bytes_loop);
6036 movq(Address(to, 0), xtmp);
6037 addptr(to, 8);
6038 BIND(L_fill_8_bytes);
6039 subl(count, 1 << (shift + 1));
6040 jcc(Assembler::greaterEqual, L_fill_8_bytes_loop);
6041 }
6042 }
6043 // fill trailing 4 bytes
6044 BIND(L_fill_4_bytes);
6045 testl(count, 1<<shift);
6046 jccb(Assembler::zero, L_fill_2_bytes);
6047 movl(Address(to, 0), value);
6048 if (t == T_BYTE || t == T_SHORT) {
6049 Label L_fill_byte;
6050 addptr(to, 4);
6051 BIND(L_fill_2_bytes);
6052 // fill trailing 2 bytes
6053 testl(count, 1<<(shift-1));
6054 jccb(Assembler::zero, L_fill_byte);
6055 movw(Address(to, 0), value);
6056 if (t == T_BYTE) {
6057 addptr(to, 2);
6058 BIND(L_fill_byte);
6059 // fill trailing byte
6060 testl(count, 1);
6061 jccb(Assembler::zero, L_exit);
6062 movb(Address(to, 0), value);
6063 } else {
6064 BIND(L_fill_byte);
6065 }
6066 } else {
6067 BIND(L_fill_2_bytes);
6068 }
6069 BIND(L_exit);
6070 }
6071
6072 void MacroAssembler::evpbroadcast(BasicType type, XMMRegister dst, Register src, int vector_len) {
6073 switch(type) {
6074 case T_BYTE:
6075 case T_BOOLEAN:
6076 evpbroadcastb(dst, src, vector_len);
6077 break;
6078 case T_SHORT:
6079 case T_CHAR:
6080 evpbroadcastw(dst, src, vector_len);
6081 break;
6082 case T_INT:
6083 case T_FLOAT:
6084 evpbroadcastd(dst, src, vector_len);
6085 break;
6086 case T_LONG:
6087 case T_DOUBLE:
6088 evpbroadcastq(dst, src, vector_len);
6089 break;
6090 default:
6091 fatal("Unhandled type : %s", type2name(type));
6092 break;
6093 }
6094 }
6095
6096 // encode char[] to byte[] in ISO_8859_1 or ASCII
6097 //@IntrinsicCandidate
6098 //private static int implEncodeISOArray(byte[] sa, int sp,
6099 //byte[] da, int dp, int len) {
6100 // int i = 0;
6101 // for (; i < len; i++) {
6102 // char c = StringUTF16.getChar(sa, sp++);
6103 // if (c > '\u00FF')
6104 // break;
6105 // da[dp++] = (byte)c;
6106 // }
6107 // return i;
6108 //}
6109 //
6110 //@IntrinsicCandidate
6111 //private static int implEncodeAsciiArray(char[] sa, int sp,
6112 // byte[] da, int dp, int len) {
6113 // int i = 0;
6114 // for (; i < len; i++) {
6115 // char c = sa[sp++];
6116 // if (c >= '\u0080')
6117 // break;
6118 // da[dp++] = (byte)c;
6119 // }
6120 // return i;
6121 //}
6122 void MacroAssembler::encode_iso_array(Register src, Register dst, Register len,
6123 XMMRegister tmp1Reg, XMMRegister tmp2Reg,
6124 XMMRegister tmp3Reg, XMMRegister tmp4Reg,
6125 Register tmp5, Register result, bool ascii) {
6126
6127 // rsi: src
6128 // rdi: dst
6129 // rdx: len
6130 // rcx: tmp5
6131 // rax: result
6132 ShortBranchVerifier sbv(this);
6133 assert_different_registers(src, dst, len, tmp5, result);
6134 Label L_done, L_copy_1_char, L_copy_1_char_exit;
6135
6136 int mask = ascii ? 0xff80ff80 : 0xff00ff00;
6137 int short_mask = ascii ? 0xff80 : 0xff00;
6138
6139 // set result
6140 xorl(result, result);
6141 // check for zero length
6142 testl(len, len);
6143 jcc(Assembler::zero, L_done);
6144
6145 movl(result, len);
6146
6147 // Setup pointers
6148 lea(src, Address(src, len, Address::times_2)); // char[]
6149 lea(dst, Address(dst, len, Address::times_1)); // byte[]
6150 negptr(len);
6151
6152 if (UseSSE42Intrinsics || UseAVX >= 2) {
6153 Label L_copy_8_chars, L_copy_8_chars_exit;
6154 Label L_chars_16_check, L_copy_16_chars, L_copy_16_chars_exit;
6155
6156 if (UseAVX >= 2) {
6157 Label L_chars_32_check, L_copy_32_chars, L_copy_32_chars_exit;
6158 movl(tmp5, mask); // create mask to test for Unicode or non-ASCII chars in vector
6159 movdl(tmp1Reg, tmp5);
6160 vpbroadcastd(tmp1Reg, tmp1Reg, Assembler::AVX_256bit);
6161 jmp(L_chars_32_check);
6162
6163 bind(L_copy_32_chars);
6164 vmovdqu(tmp3Reg, Address(src, len, Address::times_2, -64));
6165 vmovdqu(tmp4Reg, Address(src, len, Address::times_2, -32));
6166 vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
6167 vptest(tmp2Reg, tmp1Reg); // check for Unicode or non-ASCII chars in vector
6168 jccb(Assembler::notZero, L_copy_32_chars_exit);
6169 vpackuswb(tmp3Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
6170 vpermq(tmp4Reg, tmp3Reg, 0xD8, /* vector_len */ 1);
6171 vmovdqu(Address(dst, len, Address::times_1, -32), tmp4Reg);
6172
6173 bind(L_chars_32_check);
6174 addptr(len, 32);
6175 jcc(Assembler::lessEqual, L_copy_32_chars);
6176
6177 bind(L_copy_32_chars_exit);
6178 subptr(len, 16);
6179 jccb(Assembler::greater, L_copy_16_chars_exit);
6180
6181 } else if (UseSSE42Intrinsics) {
6182 movl(tmp5, mask); // create mask to test for Unicode or non-ASCII chars in vector
6183 movdl(tmp1Reg, tmp5);
6184 pshufd(tmp1Reg, tmp1Reg, 0);
6185 jmpb(L_chars_16_check);
6186 }
6187
6188 bind(L_copy_16_chars);
6189 if (UseAVX >= 2) {
6190 vmovdqu(tmp2Reg, Address(src, len, Address::times_2, -32));
6191 vptest(tmp2Reg, tmp1Reg);
6192 jcc(Assembler::notZero, L_copy_16_chars_exit);
6193 vpackuswb(tmp2Reg, tmp2Reg, tmp1Reg, /* vector_len */ 1);
6194 vpermq(tmp3Reg, tmp2Reg, 0xD8, /* vector_len */ 1);
6195 } else {
6196 if (UseAVX > 0) {
6197 movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
6198 movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
6199 vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 0);
6200 } else {
6201 movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
6202 por(tmp2Reg, tmp3Reg);
6203 movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
6204 por(tmp2Reg, tmp4Reg);
6205 }
6206 ptest(tmp2Reg, tmp1Reg); // check for Unicode or non-ASCII chars in vector
6207 jccb(Assembler::notZero, L_copy_16_chars_exit);
6208 packuswb(tmp3Reg, tmp4Reg);
6209 }
6210 movdqu(Address(dst, len, Address::times_1, -16), tmp3Reg);
6211
6212 bind(L_chars_16_check);
6213 addptr(len, 16);
6214 jcc(Assembler::lessEqual, L_copy_16_chars);
6215
6216 bind(L_copy_16_chars_exit);
6217 if (UseAVX >= 2) {
6218 // clean upper bits of YMM registers
6219 vpxor(tmp2Reg, tmp2Reg);
6220 vpxor(tmp3Reg, tmp3Reg);
6221 vpxor(tmp4Reg, tmp4Reg);
6222 movdl(tmp1Reg, tmp5);
6223 pshufd(tmp1Reg, tmp1Reg, 0);
6224 }
6225 subptr(len, 8);
6226 jccb(Assembler::greater, L_copy_8_chars_exit);
6227
6228 bind(L_copy_8_chars);
6229 movdqu(tmp3Reg, Address(src, len, Address::times_2, -16));
6230 ptest(tmp3Reg, tmp1Reg);
6231 jccb(Assembler::notZero, L_copy_8_chars_exit);
6232 packuswb(tmp3Reg, tmp1Reg);
6233 movq(Address(dst, len, Address::times_1, -8), tmp3Reg);
6234 addptr(len, 8);
6235 jccb(Assembler::lessEqual, L_copy_8_chars);
6236
6237 bind(L_copy_8_chars_exit);
6238 subptr(len, 8);
6239 jccb(Assembler::zero, L_done);
6240 }
6241
6242 bind(L_copy_1_char);
6243 load_unsigned_short(tmp5, Address(src, len, Address::times_2, 0));
6244 testl(tmp5, short_mask); // check if Unicode or non-ASCII char
6245 jccb(Assembler::notZero, L_copy_1_char_exit);
6246 movb(Address(dst, len, Address::times_1, 0), tmp5);
6247 addptr(len, 1);
6248 jccb(Assembler::less, L_copy_1_char);
6249
6250 bind(L_copy_1_char_exit);
6251 addptr(result, len); // len is negative count of not processed elements
6252
6253 bind(L_done);
6254 }
6255
6256 #ifdef _LP64
6257 /**
6258 * Helper for multiply_to_len().
6259 */
6260 void MacroAssembler::add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
6261 addq(dest_lo, src1);
6262 adcq(dest_hi, 0);
6263 addq(dest_lo, src2);
6264 adcq(dest_hi, 0);
6265 }
6266
6267 /**
6268 * Multiply 64 bit by 64 bit first loop.
6269 */
6270 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
6271 Register y, Register y_idx, Register z,
6272 Register carry, Register product,
6273 Register idx, Register kdx) {
6274 //
6275 // jlong carry, x[], y[], z[];
6276 // for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
6277 // huge_128 product = y[idx] * x[xstart] + carry;
6278 // z[kdx] = (jlong)product;
6279 // carry = (jlong)(product >>> 64);
6280 // }
6281 // z[xstart] = carry;
6282 //
6283
6284 Label L_first_loop, L_first_loop_exit;
6285 Label L_one_x, L_one_y, L_multiply;
6286
6287 decrementl(xstart);
6288 jcc(Assembler::negative, L_one_x);
6289
6290 movq(x_xstart, Address(x, xstart, Address::times_4, 0));
6291 rorq(x_xstart, 32); // convert big-endian to little-endian
6292
6293 bind(L_first_loop);
6294 decrementl(idx);
6295 jcc(Assembler::negative, L_first_loop_exit);
6296 decrementl(idx);
6297 jcc(Assembler::negative, L_one_y);
6298 movq(y_idx, Address(y, idx, Address::times_4, 0));
6299 rorq(y_idx, 32); // convert big-endian to little-endian
6300 bind(L_multiply);
6301 movq(product, x_xstart);
6302 mulq(y_idx); // product(rax) * y_idx -> rdx:rax
6303 addq(product, carry);
6304 adcq(rdx, 0);
6305 subl(kdx, 2);
6306 movl(Address(z, kdx, Address::times_4, 4), product);
6307 shrq(product, 32);
6308 movl(Address(z, kdx, Address::times_4, 0), product);
6309 movq(carry, rdx);
6310 jmp(L_first_loop);
6311
6312 bind(L_one_y);
6313 movl(y_idx, Address(y, 0));
6314 jmp(L_multiply);
6315
6316 bind(L_one_x);
6317 movl(x_xstart, Address(x, 0));
6318 jmp(L_first_loop);
6319
6320 bind(L_first_loop_exit);
6321 }
6322
6323 /**
6324 * Multiply 64 bit by 64 bit and add 128 bit.
6325 */
6326 void MacroAssembler::multiply_add_128_x_128(Register x_xstart, Register y, Register z,
6327 Register yz_idx, Register idx,
6328 Register carry, Register product, int offset) {
6329 // huge_128 product = (y[idx] * x_xstart) + z[kdx] + carry;
6330 // z[kdx] = (jlong)product;
6331
6332 movq(yz_idx, Address(y, idx, Address::times_4, offset));
6333 rorq(yz_idx, 32); // convert big-endian to little-endian
6334 movq(product, x_xstart);
6335 mulq(yz_idx); // product(rax) * yz_idx -> rdx:product(rax)
6336 movq(yz_idx, Address(z, idx, Address::times_4, offset));
6337 rorq(yz_idx, 32); // convert big-endian to little-endian
6338
6339 add2_with_carry(rdx, product, carry, yz_idx);
6340
6341 movl(Address(z, idx, Address::times_4, offset+4), product);
6342 shrq(product, 32);
6343 movl(Address(z, idx, Address::times_4, offset), product);
6344
6345 }
6346
6347 /**
6348 * Multiply 128 bit by 128 bit. Unrolled inner loop.
6349 */
6350 void MacroAssembler::multiply_128_x_128_loop(Register x_xstart, Register y, Register z,
6351 Register yz_idx, Register idx, Register jdx,
6352 Register carry, Register product,
6353 Register carry2) {
6354 // jlong carry, x[], y[], z[];
6355 // int kdx = ystart+1;
6356 // for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
6357 // huge_128 product = (y[idx+1] * x_xstart) + z[kdx+idx+1] + carry;
6358 // z[kdx+idx+1] = (jlong)product;
6359 // jlong carry2 = (jlong)(product >>> 64);
6360 // product = (y[idx] * x_xstart) + z[kdx+idx] + carry2;
6361 // z[kdx+idx] = (jlong)product;
6362 // carry = (jlong)(product >>> 64);
6363 // }
6364 // idx += 2;
6365 // if (idx > 0) {
6366 // product = (y[idx] * x_xstart) + z[kdx+idx] + carry;
6367 // z[kdx+idx] = (jlong)product;
6368 // carry = (jlong)(product >>> 64);
6369 // }
6370 //
6371
6372 Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
6373
6374 movl(jdx, idx);
6375 andl(jdx, 0xFFFFFFFC);
6376 shrl(jdx, 2);
6377
6378 bind(L_third_loop);
6379 subl(jdx, 1);
6380 jcc(Assembler::negative, L_third_loop_exit);
6381 subl(idx, 4);
6382
6383 multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 8);
6384 movq(carry2, rdx);
6385
6386 multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry2, product, 0);
6387 movq(carry, rdx);
6388 jmp(L_third_loop);
6389
6390 bind (L_third_loop_exit);
6391
6392 andl (idx, 0x3);
6393 jcc(Assembler::zero, L_post_third_loop_done);
6394
6395 Label L_check_1;
6396 subl(idx, 2);
6397 jcc(Assembler::negative, L_check_1);
6398
6399 multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 0);
6400 movq(carry, rdx);
6401
6402 bind (L_check_1);
6403 addl (idx, 0x2);
6404 andl (idx, 0x1);
6405 subl(idx, 1);
6406 jcc(Assembler::negative, L_post_third_loop_done);
6407
6408 movl(yz_idx, Address(y, idx, Address::times_4, 0));
6409 movq(product, x_xstart);
6410 mulq(yz_idx); // product(rax) * yz_idx -> rdx:product(rax)
6411 movl(yz_idx, Address(z, idx, Address::times_4, 0));
6412
6413 add2_with_carry(rdx, product, yz_idx, carry);
6414
6415 movl(Address(z, idx, Address::times_4, 0), product);
6416 shrq(product, 32);
6417
6418 shlq(rdx, 32);
6419 orq(product, rdx);
6420 movq(carry, product);
6421
6422 bind(L_post_third_loop_done);
6423 }
6424
6425 /**
6426 * Multiply 128 bit by 128 bit using BMI2. Unrolled inner loop.
6427 *
6428 */
6429 void MacroAssembler::multiply_128_x_128_bmi2_loop(Register y, Register z,
6430 Register carry, Register carry2,
6431 Register idx, Register jdx,
6432 Register yz_idx1, Register yz_idx2,
6433 Register tmp, Register tmp3, Register tmp4) {
6434 assert(UseBMI2Instructions, "should be used only when BMI2 is available");
6435
6436 // jlong carry, x[], y[], z[];
6437 // int kdx = ystart+1;
6438 // for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
6439 // huge_128 tmp3 = (y[idx+1] * rdx) + z[kdx+idx+1] + carry;
6440 // jlong carry2 = (jlong)(tmp3 >>> 64);
6441 // huge_128 tmp4 = (y[idx] * rdx) + z[kdx+idx] + carry2;
6442 // carry = (jlong)(tmp4 >>> 64);
6443 // z[kdx+idx+1] = (jlong)tmp3;
6444 // z[kdx+idx] = (jlong)tmp4;
6445 // }
6446 // idx += 2;
6447 // if (idx > 0) {
6448 // yz_idx1 = (y[idx] * rdx) + z[kdx+idx] + carry;
6449 // z[kdx+idx] = (jlong)yz_idx1;
6450 // carry = (jlong)(yz_idx1 >>> 64);
6451 // }
6452 //
6453
6454 Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
6455
6456 movl(jdx, idx);
6457 andl(jdx, 0xFFFFFFFC);
6458 shrl(jdx, 2);
6459
6460 bind(L_third_loop);
6461 subl(jdx, 1);
6462 jcc(Assembler::negative, L_third_loop_exit);
6463 subl(idx, 4);
6464
6465 movq(yz_idx1, Address(y, idx, Address::times_4, 8));
6466 rorxq(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
6467 movq(yz_idx2, Address(y, idx, Address::times_4, 0));
6468 rorxq(yz_idx2, yz_idx2, 32);
6469
6470 mulxq(tmp4, tmp3, yz_idx1); // yz_idx1 * rdx -> tmp4:tmp3
6471 mulxq(carry2, tmp, yz_idx2); // yz_idx2 * rdx -> carry2:tmp
6472
6473 movq(yz_idx1, Address(z, idx, Address::times_4, 8));
6474 rorxq(yz_idx1, yz_idx1, 32);
6475 movq(yz_idx2, Address(z, idx, Address::times_4, 0));
6476 rorxq(yz_idx2, yz_idx2, 32);
6477
6478 if (VM_Version::supports_adx()) {
6479 adcxq(tmp3, carry);
6480 adoxq(tmp3, yz_idx1);
6481
6482 adcxq(tmp4, tmp);
6483 adoxq(tmp4, yz_idx2);
6484
6485 movl(carry, 0); // does not affect flags
6486 adcxq(carry2, carry);
6487 adoxq(carry2, carry);
6488 } else {
6489 add2_with_carry(tmp4, tmp3, carry, yz_idx1);
6490 add2_with_carry(carry2, tmp4, tmp, yz_idx2);
6491 }
6492 movq(carry, carry2);
6493
6494 movl(Address(z, idx, Address::times_4, 12), tmp3);
6495 shrq(tmp3, 32);
6496 movl(Address(z, idx, Address::times_4, 8), tmp3);
6497
6498 movl(Address(z, idx, Address::times_4, 4), tmp4);
6499 shrq(tmp4, 32);
6500 movl(Address(z, idx, Address::times_4, 0), tmp4);
6501
6502 jmp(L_third_loop);
6503
6504 bind (L_third_loop_exit);
6505
6506 andl (idx, 0x3);
6507 jcc(Assembler::zero, L_post_third_loop_done);
6508
6509 Label L_check_1;
6510 subl(idx, 2);
6511 jcc(Assembler::negative, L_check_1);
6512
6513 movq(yz_idx1, Address(y, idx, Address::times_4, 0));
6514 rorxq(yz_idx1, yz_idx1, 32);
6515 mulxq(tmp4, tmp3, yz_idx1); // yz_idx1 * rdx -> tmp4:tmp3
6516 movq(yz_idx2, Address(z, idx, Address::times_4, 0));
6517 rorxq(yz_idx2, yz_idx2, 32);
6518
6519 add2_with_carry(tmp4, tmp3, carry, yz_idx2);
6520
6521 movl(Address(z, idx, Address::times_4, 4), tmp3);
6522 shrq(tmp3, 32);
6523 movl(Address(z, idx, Address::times_4, 0), tmp3);
6524 movq(carry, tmp4);
6525
6526 bind (L_check_1);
6527 addl (idx, 0x2);
6528 andl (idx, 0x1);
6529 subl(idx, 1);
6530 jcc(Assembler::negative, L_post_third_loop_done);
6531 movl(tmp4, Address(y, idx, Address::times_4, 0));
6532 mulxq(carry2, tmp3, tmp4); // tmp4 * rdx -> carry2:tmp3
6533 movl(tmp4, Address(z, idx, Address::times_4, 0));
6534
6535 add2_with_carry(carry2, tmp3, tmp4, carry);
6536
6537 movl(Address(z, idx, Address::times_4, 0), tmp3);
6538 shrq(tmp3, 32);
6539
6540 shlq(carry2, 32);
6541 orq(tmp3, carry2);
6542 movq(carry, tmp3);
6543
6544 bind(L_post_third_loop_done);
6545 }
6546
6547 /**
6548 * Code for BigInteger::multiplyToLen() intrinsic.
6549 *
6550 * rdi: x
6551 * rax: xlen
6552 * rsi: y
6553 * rcx: ylen
6554 * r8: z
6555 * r11: zlen
6556 * r12: tmp1
6557 * r13: tmp2
6558 * r14: tmp3
6559 * r15: tmp4
6560 * rbx: tmp5
6561 *
6562 */
6563 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register zlen,
6564 Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5) {
6565 ShortBranchVerifier sbv(this);
6566 assert_different_registers(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, rdx);
6567
6568 push(tmp1);
6569 push(tmp2);
6570 push(tmp3);
6571 push(tmp4);
6572 push(tmp5);
6573
6574 push(xlen);
6575 push(zlen);
6576
6577 const Register idx = tmp1;
6578 const Register kdx = tmp2;
6579 const Register xstart = tmp3;
6580
6581 const Register y_idx = tmp4;
6582 const Register carry = tmp5;
6583 const Register product = xlen;
6584 const Register x_xstart = zlen; // reuse register
6585
6586 // First Loop.
6587 //
6588 // final static long LONG_MASK = 0xffffffffL;
6589 // int xstart = xlen - 1;
6590 // int ystart = ylen - 1;
6591 // long carry = 0;
6592 // for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
6593 // long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
6594 // z[kdx] = (int)product;
6595 // carry = product >>> 32;
6596 // }
6597 // z[xstart] = (int)carry;
6598 //
6599
6600 movl(idx, ylen); // idx = ylen;
6601 movl(kdx, zlen); // kdx = xlen+ylen;
6602 xorq(carry, carry); // carry = 0;
6603
6604 Label L_done;
6605
6606 movl(xstart, xlen);
6607 decrementl(xstart);
6608 jcc(Assembler::negative, L_done);
6609
6610 multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
6611
6612 Label L_second_loop;
6613 testl(kdx, kdx);
6614 jcc(Assembler::zero, L_second_loop);
6615
6616 Label L_carry;
6617 subl(kdx, 1);
6618 jcc(Assembler::zero, L_carry);
6619
6620 movl(Address(z, kdx, Address::times_4, 0), carry);
6621 shrq(carry, 32);
6622 subl(kdx, 1);
6623
6624 bind(L_carry);
6625 movl(Address(z, kdx, Address::times_4, 0), carry);
6626
6627 // Second and third (nested) loops.
6628 //
6629 // for (int i = xstart-1; i >= 0; i--) { // Second loop
6630 // carry = 0;
6631 // for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
6632 // long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
6633 // (z[k] & LONG_MASK) + carry;
6634 // z[k] = (int)product;
6635 // carry = product >>> 32;
6636 // }
6637 // z[i] = (int)carry;
6638 // }
6639 //
6640 // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = rdx
6641
6642 const Register jdx = tmp1;
6643
6644 bind(L_second_loop);
6645 xorl(carry, carry); // carry = 0;
6646 movl(jdx, ylen); // j = ystart+1
6647
6648 subl(xstart, 1); // i = xstart-1;
6649 jcc(Assembler::negative, L_done);
6650
6651 push (z);
6652
6653 Label L_last_x;
6654 lea(z, Address(z, xstart, Address::times_4, 4)); // z = z + k - j
6655 subl(xstart, 1); // i = xstart-1;
6656 jcc(Assembler::negative, L_last_x);
6657
6658 if (UseBMI2Instructions) {
6659 movq(rdx, Address(x, xstart, Address::times_4, 0));
6660 rorxq(rdx, rdx, 32); // convert big-endian to little-endian
6661 } else {
6662 movq(x_xstart, Address(x, xstart, Address::times_4, 0));
6663 rorq(x_xstart, 32); // convert big-endian to little-endian
6664 }
6665
6666 Label L_third_loop_prologue;
6667 bind(L_third_loop_prologue);
6668
6669 push (x);
6670 push (xstart);
6671 push (ylen);
6672
6673
6674 if (UseBMI2Instructions) {
6675 multiply_128_x_128_bmi2_loop(y, z, carry, x, jdx, ylen, product, tmp2, x_xstart, tmp3, tmp4);
6676 } else { // !UseBMI2Instructions
6677 multiply_128_x_128_loop(x_xstart, y, z, y_idx, jdx, ylen, carry, product, x);
6678 }
6679
6680 pop(ylen);
6681 pop(xlen);
6682 pop(x);
6683 pop(z);
6684
6685 movl(tmp3, xlen);
6686 addl(tmp3, 1);
6687 movl(Address(z, tmp3, Address::times_4, 0), carry);
6688 subl(tmp3, 1);
6689 jccb(Assembler::negative, L_done);
6690
6691 shrq(carry, 32);
6692 movl(Address(z, tmp3, Address::times_4, 0), carry);
6693 jmp(L_second_loop);
6694
6695 // Next infrequent code is moved outside loops.
6696 bind(L_last_x);
6697 if (UseBMI2Instructions) {
6698 movl(rdx, Address(x, 0));
6699 } else {
6700 movl(x_xstart, Address(x, 0));
6701 }
6702 jmp(L_third_loop_prologue);
6703
6704 bind(L_done);
6705
6706 pop(zlen);
6707 pop(xlen);
6708
6709 pop(tmp5);
6710 pop(tmp4);
6711 pop(tmp3);
6712 pop(tmp2);
6713 pop(tmp1);
6714 }
6715
6716 void MacroAssembler::vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale,
6717 Register result, Register tmp1, Register tmp2, XMMRegister rymm0, XMMRegister rymm1, XMMRegister rymm2){
6718 assert(UseSSE42Intrinsics, "SSE4.2 must be enabled.");
6719 Label VECTOR16_LOOP, VECTOR8_LOOP, VECTOR4_LOOP;
6720 Label VECTOR8_TAIL, VECTOR4_TAIL;
6721 Label VECTOR32_NOT_EQUAL, VECTOR16_NOT_EQUAL, VECTOR8_NOT_EQUAL, VECTOR4_NOT_EQUAL;
6722 Label SAME_TILL_END, DONE;
6723 Label BYTES_LOOP, BYTES_TAIL, BYTES_NOT_EQUAL;
6724
6725 //scale is in rcx in both Win64 and Unix
6726 ShortBranchVerifier sbv(this);
6727
6728 shlq(length);
6729 xorq(result, result);
6730
6731 if ((AVX3Threshold == 0) && (UseAVX > 2) &&
6732 VM_Version::supports_avx512vlbw()) {
6733 Label VECTOR64_LOOP, VECTOR64_NOT_EQUAL, VECTOR32_TAIL;
6734
6735 cmpq(length, 64);
6736 jcc(Assembler::less, VECTOR32_TAIL);
6737
6738 movq(tmp1, length);
6739 andq(tmp1, 0x3F); // tail count
6740 andq(length, ~(0x3F)); //vector count
6741
6742 bind(VECTOR64_LOOP);
6743 // AVX512 code to compare 64 byte vectors.
6744 evmovdqub(rymm0, Address(obja, result), Assembler::AVX_512bit);
6745 evpcmpeqb(k7, rymm0, Address(objb, result), Assembler::AVX_512bit);
6746 kortestql(k7, k7);
6747 jcc(Assembler::aboveEqual, VECTOR64_NOT_EQUAL); // mismatch
6748 addq(result, 64);
6749 subq(length, 64);
6750 jccb(Assembler::notZero, VECTOR64_LOOP);
6751
6752 //bind(VECTOR64_TAIL);
6753 testq(tmp1, tmp1);
6754 jcc(Assembler::zero, SAME_TILL_END);
6755
6756 //bind(VECTOR64_TAIL);
6757 // AVX512 code to compare up to 63 byte vectors.
6758 mov64(tmp2, 0xFFFFFFFFFFFFFFFF);
6759 shlxq(tmp2, tmp2, tmp1);
6760 notq(tmp2);
6761 kmovql(k3, tmp2);
6762
6763 evmovdqub(rymm0, k3, Address(obja, result), false, Assembler::AVX_512bit);
6764 evpcmpeqb(k7, k3, rymm0, Address(objb, result), Assembler::AVX_512bit);
6765
6766 ktestql(k7, k3);
6767 jcc(Assembler::below, SAME_TILL_END); // not mismatch
6768
6769 bind(VECTOR64_NOT_EQUAL);
6770 kmovql(tmp1, k7);
6771 notq(tmp1);
6772 tzcntq(tmp1, tmp1);
6773 addq(result, tmp1);
6774 shrq(result);
6775 jmp(DONE);
6776 bind(VECTOR32_TAIL);
6777 }
6778
6779 cmpq(length, 8);
6780 jcc(Assembler::equal, VECTOR8_LOOP);
6781 jcc(Assembler::less, VECTOR4_TAIL);
6782
6783 if (UseAVX >= 2) {
6784 Label VECTOR16_TAIL, VECTOR32_LOOP;
6785
6786 cmpq(length, 16);
6787 jcc(Assembler::equal, VECTOR16_LOOP);
6788 jcc(Assembler::less, VECTOR8_LOOP);
6789
6790 cmpq(length, 32);
6791 jccb(Assembler::less, VECTOR16_TAIL);
6792
6793 subq(length, 32);
6794 bind(VECTOR32_LOOP);
6795 vmovdqu(rymm0, Address(obja, result));
6796 vmovdqu(rymm1, Address(objb, result));
6797 vpxor(rymm2, rymm0, rymm1, Assembler::AVX_256bit);
6798 vptest(rymm2, rymm2);
6799 jcc(Assembler::notZero, VECTOR32_NOT_EQUAL);//mismatch found
6800 addq(result, 32);
6801 subq(length, 32);
6802 jcc(Assembler::greaterEqual, VECTOR32_LOOP);
6803 addq(length, 32);
6804 jcc(Assembler::equal, SAME_TILL_END);
6805 //falling through if less than 32 bytes left //close the branch here.
6806
6807 bind(VECTOR16_TAIL);
6808 cmpq(length, 16);
6809 jccb(Assembler::less, VECTOR8_TAIL);
6810 bind(VECTOR16_LOOP);
6811 movdqu(rymm0, Address(obja, result));
6812 movdqu(rymm1, Address(objb, result));
6813 vpxor(rymm2, rymm0, rymm1, Assembler::AVX_128bit);
6814 ptest(rymm2, rymm2);
6815 jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
6816 addq(result, 16);
6817 subq(length, 16);
6818 jcc(Assembler::equal, SAME_TILL_END);
6819 //falling through if less than 16 bytes left
6820 } else {//regular intrinsics
6821
6822 cmpq(length, 16);
6823 jccb(Assembler::less, VECTOR8_TAIL);
6824
6825 subq(length, 16);
6826 bind(VECTOR16_LOOP);
6827 movdqu(rymm0, Address(obja, result));
6828 movdqu(rymm1, Address(objb, result));
6829 pxor(rymm0, rymm1);
6830 ptest(rymm0, rymm0);
6831 jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
6832 addq(result, 16);
6833 subq(length, 16);
6834 jccb(Assembler::greaterEqual, VECTOR16_LOOP);
6835 addq(length, 16);
6836 jcc(Assembler::equal, SAME_TILL_END);
6837 //falling through if less than 16 bytes left
6838 }
6839
6840 bind(VECTOR8_TAIL);
6841 cmpq(length, 8);
6842 jccb(Assembler::less, VECTOR4_TAIL);
6843 bind(VECTOR8_LOOP);
6844 movq(tmp1, Address(obja, result));
6845 movq(tmp2, Address(objb, result));
6846 xorq(tmp1, tmp2);
6847 testq(tmp1, tmp1);
6848 jcc(Assembler::notZero, VECTOR8_NOT_EQUAL);//mismatch found
6849 addq(result, 8);
6850 subq(length, 8);
6851 jcc(Assembler::equal, SAME_TILL_END);
6852 //falling through if less than 8 bytes left
6853
6854 bind(VECTOR4_TAIL);
6855 cmpq(length, 4);
6856 jccb(Assembler::less, BYTES_TAIL);
6857 bind(VECTOR4_LOOP);
6858 movl(tmp1, Address(obja, result));
6859 xorl(tmp1, Address(objb, result));
6860 testl(tmp1, tmp1);
6861 jcc(Assembler::notZero, VECTOR4_NOT_EQUAL);//mismatch found
6862 addq(result, 4);
6863 subq(length, 4);
6864 jcc(Assembler::equal, SAME_TILL_END);
6865 //falling through if less than 4 bytes left
6866
6867 bind(BYTES_TAIL);
6868 bind(BYTES_LOOP);
6869 load_unsigned_byte(tmp1, Address(obja, result));
6870 load_unsigned_byte(tmp2, Address(objb, result));
6871 xorl(tmp1, tmp2);
6872 testl(tmp1, tmp1);
6873 jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
6874 decq(length);
6875 jcc(Assembler::zero, SAME_TILL_END);
6876 incq(result);
6877 load_unsigned_byte(tmp1, Address(obja, result));
6878 load_unsigned_byte(tmp2, Address(objb, result));
6879 xorl(tmp1, tmp2);
6880 testl(tmp1, tmp1);
6881 jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
6882 decq(length);
6883 jcc(Assembler::zero, SAME_TILL_END);
6884 incq(result);
6885 load_unsigned_byte(tmp1, Address(obja, result));
6886 load_unsigned_byte(tmp2, Address(objb, result));
6887 xorl(tmp1, tmp2);
6888 testl(tmp1, tmp1);
6889 jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
6890 jmp(SAME_TILL_END);
6891
6892 if (UseAVX >= 2) {
6893 bind(VECTOR32_NOT_EQUAL);
6894 vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_256bit);
6895 vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_256bit);
6896 vpxor(rymm0, rymm0, rymm2, Assembler::AVX_256bit);
6897 vpmovmskb(tmp1, rymm0);
6898 bsfq(tmp1, tmp1);
6899 addq(result, tmp1);
6900 shrq(result);
6901 jmp(DONE);
6902 }
6903
6904 bind(VECTOR16_NOT_EQUAL);
6905 if (UseAVX >= 2) {
6906 vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_128bit);
6907 vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_128bit);
6908 pxor(rymm0, rymm2);
6909 } else {
6910 pcmpeqb(rymm2, rymm2);
6911 pxor(rymm0, rymm1);
6912 pcmpeqb(rymm0, rymm1);
6913 pxor(rymm0, rymm2);
6914 }
6915 pmovmskb(tmp1, rymm0);
6916 bsfq(tmp1, tmp1);
6917 addq(result, tmp1);
6918 shrq(result);
6919 jmpb(DONE);
6920
6921 bind(VECTOR8_NOT_EQUAL);
6922 bind(VECTOR4_NOT_EQUAL);
6923 bsfq(tmp1, tmp1);
6924 shrq(tmp1, 3);
6925 addq(result, tmp1);
6926 bind(BYTES_NOT_EQUAL);
6927 shrq(result);
6928 jmpb(DONE);
6929
6930 bind(SAME_TILL_END);
6931 mov64(result, -1);
6932
6933 bind(DONE);
6934 }
6935
6936 //Helper functions for square_to_len()
6937
6938 /**
6939 * Store the squares of x[], right shifted one bit (divided by 2) into z[]
6940 * Preserves x and z and modifies rest of the registers.
6941 */
6942 void MacroAssembler::square_rshift(Register x, Register xlen, Register z, Register tmp1, Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
6943 // Perform square and right shift by 1
6944 // Handle odd xlen case first, then for even xlen do the following
6945 // jlong carry = 0;
6946 // for (int j=0, i=0; j < xlen; j+=2, i+=4) {
6947 // huge_128 product = x[j:j+1] * x[j:j+1];
6948 // z[i:i+1] = (carry << 63) | (jlong)(product >>> 65);
6949 // z[i+2:i+3] = (jlong)(product >>> 1);
6950 // carry = (jlong)product;
6951 // }
6952
6953 xorq(tmp5, tmp5); // carry
6954 xorq(rdxReg, rdxReg);
6955 xorl(tmp1, tmp1); // index for x
6956 xorl(tmp4, tmp4); // index for z
6957
6958 Label L_first_loop, L_first_loop_exit;
6959
6960 testl(xlen, 1);
6961 jccb(Assembler::zero, L_first_loop); //jump if xlen is even
6962
6963 // Square and right shift by 1 the odd element using 32 bit multiply
6964 movl(raxReg, Address(x, tmp1, Address::times_4, 0));
6965 imulq(raxReg, raxReg);
6966 shrq(raxReg, 1);
6967 adcq(tmp5, 0);
6968 movq(Address(z, tmp4, Address::times_4, 0), raxReg);
6969 incrementl(tmp1);
6970 addl(tmp4, 2);
6971
6972 // Square and right shift by 1 the rest using 64 bit multiply
6973 bind(L_first_loop);
6974 cmpptr(tmp1, xlen);
6975 jccb(Assembler::equal, L_first_loop_exit);
6976
6977 // Square
6978 movq(raxReg, Address(x, tmp1, Address::times_4, 0));
6979 rorq(raxReg, 32); // convert big-endian to little-endian
6980 mulq(raxReg); // 64-bit multiply rax * rax -> rdx:rax
6981
6982 // Right shift by 1 and save carry
6983 shrq(tmp5, 1); // rdx:rax:tmp5 = (tmp5:rdx:rax) >>> 1
6984 rcrq(rdxReg, 1);
6985 rcrq(raxReg, 1);
6986 adcq(tmp5, 0);
6987
6988 // Store result in z
6989 movq(Address(z, tmp4, Address::times_4, 0), rdxReg);
6990 movq(Address(z, tmp4, Address::times_4, 8), raxReg);
6991
6992 // Update indices for x and z
6993 addl(tmp1, 2);
6994 addl(tmp4, 4);
6995 jmp(L_first_loop);
6996
6997 bind(L_first_loop_exit);
6998 }
6999
7000
7001 /**
7002 * Perform the following multiply add operation using BMI2 instructions
7003 * carry:sum = sum + op1*op2 + carry
7004 * op2 should be in rdx
7005 * op2 is preserved, all other registers are modified
7006 */
7007 void MacroAssembler::multiply_add_64_bmi2(Register sum, Register op1, Register op2, Register carry, Register tmp2) {
7008 // assert op2 is rdx
7009 mulxq(tmp2, op1, op1); // op1 * op2 -> tmp2:op1
7010 addq(sum, carry);
7011 adcq(tmp2, 0);
7012 addq(sum, op1);
7013 adcq(tmp2, 0);
7014 movq(carry, tmp2);
7015 }
7016
7017 /**
7018 * Perform the following multiply add operation:
7019 * carry:sum = sum + op1*op2 + carry
7020 * Preserves op1, op2 and modifies rest of registers
7021 */
7022 void MacroAssembler::multiply_add_64(Register sum, Register op1, Register op2, Register carry, Register rdxReg, Register raxReg) {
7023 // rdx:rax = op1 * op2
7024 movq(raxReg, op2);
7025 mulq(op1);
7026
7027 // rdx:rax = sum + carry + rdx:rax
7028 addq(sum, carry);
7029 adcq(rdxReg, 0);
7030 addq(sum, raxReg);
7031 adcq(rdxReg, 0);
7032
7033 // carry:sum = rdx:sum
7034 movq(carry, rdxReg);
7035 }
7036
7037 /**
7038 * Add 64 bit long carry into z[] with carry propagation.
7039 * Preserves z and carry register values and modifies rest of registers.
7040 *
7041 */
7042 void MacroAssembler::add_one_64(Register z, Register zlen, Register carry, Register tmp1) {
7043 Label L_fourth_loop, L_fourth_loop_exit;
7044
7045 movl(tmp1, 1);
7046 subl(zlen, 2);
7047 addq(Address(z, zlen, Address::times_4, 0), carry);
7048
7049 bind(L_fourth_loop);
7050 jccb(Assembler::carryClear, L_fourth_loop_exit);
7051 subl(zlen, 2);
7052 jccb(Assembler::negative, L_fourth_loop_exit);
7053 addq(Address(z, zlen, Address::times_4, 0), tmp1);
7054 jmp(L_fourth_loop);
7055 bind(L_fourth_loop_exit);
7056 }
7057
7058 /**
7059 * Shift z[] left by 1 bit.
7060 * Preserves x, len, z and zlen registers and modifies rest of the registers.
7061 *
7062 */
7063 void MacroAssembler::lshift_by_1(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
7064
7065 Label L_fifth_loop, L_fifth_loop_exit;
7066
7067 // Fifth loop
7068 // Perform primitiveLeftShift(z, zlen, 1)
7069
7070 const Register prev_carry = tmp1;
7071 const Register new_carry = tmp4;
7072 const Register value = tmp2;
7073 const Register zidx = tmp3;
7074
7075 // int zidx, carry;
7076 // long value;
7077 // carry = 0;
7078 // for (zidx = zlen-2; zidx >=0; zidx -= 2) {
7079 // (carry:value) = (z[i] << 1) | carry ;
7080 // z[i] = value;
7081 // }
7082
7083 movl(zidx, zlen);
7084 xorl(prev_carry, prev_carry); // clear carry flag and prev_carry register
7085
7086 bind(L_fifth_loop);
7087 decl(zidx); // Use decl to preserve carry flag
7088 decl(zidx);
7089 jccb(Assembler::negative, L_fifth_loop_exit);
7090
7091 if (UseBMI2Instructions) {
7092 movq(value, Address(z, zidx, Address::times_4, 0));
7093 rclq(value, 1);
7094 rorxq(value, value, 32);
7095 movq(Address(z, zidx, Address::times_4, 0), value); // Store back in big endian form
7096 }
7097 else {
7098 // clear new_carry
7099 xorl(new_carry, new_carry);
7100
7101 // Shift z[i] by 1, or in previous carry and save new carry
7102 movq(value, Address(z, zidx, Address::times_4, 0));
7103 shlq(value, 1);
7104 adcl(new_carry, 0);
7105
7106 orq(value, prev_carry);
7107 rorq(value, 0x20);
7108 movq(Address(z, zidx, Address::times_4, 0), value); // Store back in big endian form
7109
7110 // Set previous carry = new carry
7111 movl(prev_carry, new_carry);
7112 }
7113 jmp(L_fifth_loop);
7114
7115 bind(L_fifth_loop_exit);
7116 }
7117
7118
7119 /**
7120 * Code for BigInteger::squareToLen() intrinsic
7121 *
7122 * rdi: x
7123 * rsi: len
7124 * r8: z
7125 * rcx: zlen
7126 * r12: tmp1
7127 * r13: tmp2
7128 * r14: tmp3
7129 * r15: tmp4
7130 * rbx: tmp5
7131 *
7132 */
7133 void MacroAssembler::square_to_len(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
7134
7135 Label L_second_loop, L_second_loop_exit, L_third_loop, L_third_loop_exit, L_last_x, L_multiply;
7136 push(tmp1);
7137 push(tmp2);
7138 push(tmp3);
7139 push(tmp4);
7140 push(tmp5);
7141
7142 // First loop
7143 // Store the squares, right shifted one bit (i.e., divided by 2).
7144 square_rshift(x, len, z, tmp1, tmp3, tmp4, tmp5, rdxReg, raxReg);
7145
7146 // Add in off-diagonal sums.
7147 //
7148 // Second, third (nested) and fourth loops.
7149 // zlen +=2;
7150 // for (int xidx=len-2,zidx=zlen-4; xidx > 0; xidx-=2,zidx-=4) {
7151 // carry = 0;
7152 // long op2 = x[xidx:xidx+1];
7153 // for (int j=xidx-2,k=zidx; j >= 0; j-=2) {
7154 // k -= 2;
7155 // long op1 = x[j:j+1];
7156 // long sum = z[k:k+1];
7157 // carry:sum = multiply_add_64(sum, op1, op2, carry, tmp_regs);
7158 // z[k:k+1] = sum;
7159 // }
7160 // add_one_64(z, k, carry, tmp_regs);
7161 // }
7162
7163 const Register carry = tmp5;
7164 const Register sum = tmp3;
7165 const Register op1 = tmp4;
7166 Register op2 = tmp2;
7167
7168 push(zlen);
7169 push(len);
7170 addl(zlen,2);
7171 bind(L_second_loop);
7172 xorq(carry, carry);
7173 subl(zlen, 4);
7174 subl(len, 2);
7175 push(zlen);
7176 push(len);
7177 cmpl(len, 0);
7178 jccb(Assembler::lessEqual, L_second_loop_exit);
7179
7180 // Multiply an array by one 64 bit long.
7181 if (UseBMI2Instructions) {
7182 op2 = rdxReg;
7183 movq(op2, Address(x, len, Address::times_4, 0));
7184 rorxq(op2, op2, 32);
7185 }
7186 else {
7187 movq(op2, Address(x, len, Address::times_4, 0));
7188 rorq(op2, 32);
7189 }
7190
7191 bind(L_third_loop);
7192 decrementl(len);
7193 jccb(Assembler::negative, L_third_loop_exit);
7194 decrementl(len);
7195 jccb(Assembler::negative, L_last_x);
7196
7197 movq(op1, Address(x, len, Address::times_4, 0));
7198 rorq(op1, 32);
7199
7200 bind(L_multiply);
7201 subl(zlen, 2);
7202 movq(sum, Address(z, zlen, Address::times_4, 0));
7203
7204 // Multiply 64 bit by 64 bit and add 64 bits lower half and upper 64 bits as carry.
7205 if (UseBMI2Instructions) {
7206 multiply_add_64_bmi2(sum, op1, op2, carry, tmp2);
7207 }
7208 else {
7209 multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
7210 }
7211
7212 movq(Address(z, zlen, Address::times_4, 0), sum);
7213
7214 jmp(L_third_loop);
7215 bind(L_third_loop_exit);
7216
7217 // Fourth loop
7218 // Add 64 bit long carry into z with carry propagation.
7219 // Uses offsetted zlen.
7220 add_one_64(z, zlen, carry, tmp1);
7221
7222 pop(len);
7223 pop(zlen);
7224 jmp(L_second_loop);
7225
7226 // Next infrequent code is moved outside loops.
7227 bind(L_last_x);
7228 movl(op1, Address(x, 0));
7229 jmp(L_multiply);
7230
7231 bind(L_second_loop_exit);
7232 pop(len);
7233 pop(zlen);
7234 pop(len);
7235 pop(zlen);
7236
7237 // Fifth loop
7238 // Shift z left 1 bit.
7239 lshift_by_1(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4);
7240
7241 // z[zlen-1] |= x[len-1] & 1;
7242 movl(tmp3, Address(x, len, Address::times_4, -4));
7243 andl(tmp3, 1);
7244 orl(Address(z, zlen, Address::times_4, -4), tmp3);
7245
7246 pop(tmp5);
7247 pop(tmp4);
7248 pop(tmp3);
7249 pop(tmp2);
7250 pop(tmp1);
7251 }
7252
7253 /**
7254 * Helper function for mul_add()
7255 * Multiply the in[] by int k and add to out[] starting at offset offs using
7256 * 128 bit by 32 bit multiply and return the carry in tmp5.
7257 * Only quad int aligned length of in[] is operated on in this function.
7258 * k is in rdxReg for BMI2Instructions, for others it is in tmp2.
7259 * This function preserves out, in and k registers.
7260 * len and offset point to the appropriate index in "in" & "out" correspondingly
7261 * tmp5 has the carry.
7262 * other registers are temporary and are modified.
7263 *
7264 */
7265 void MacroAssembler::mul_add_128_x_32_loop(Register out, Register in,
7266 Register offset, Register len, Register tmp1, Register tmp2, Register tmp3,
7267 Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
7268
7269 Label L_first_loop, L_first_loop_exit;
7270
7271 movl(tmp1, len);
7272 shrl(tmp1, 2);
7273
7274 bind(L_first_loop);
7275 subl(tmp1, 1);
7276 jccb(Assembler::negative, L_first_loop_exit);
7277
7278 subl(len, 4);
7279 subl(offset, 4);
7280
7281 Register op2 = tmp2;
7282 const Register sum = tmp3;
7283 const Register op1 = tmp4;
7284 const Register carry = tmp5;
7285
7286 if (UseBMI2Instructions) {
7287 op2 = rdxReg;
7288 }
7289
7290 movq(op1, Address(in, len, Address::times_4, 8));
7291 rorq(op1, 32);
7292 movq(sum, Address(out, offset, Address::times_4, 8));
7293 rorq(sum, 32);
7294 if (UseBMI2Instructions) {
7295 multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
7296 }
7297 else {
7298 multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
7299 }
7300 // Store back in big endian from little endian
7301 rorq(sum, 0x20);
7302 movq(Address(out, offset, Address::times_4, 8), sum);
7303
7304 movq(op1, Address(in, len, Address::times_4, 0));
7305 rorq(op1, 32);
7306 movq(sum, Address(out, offset, Address::times_4, 0));
7307 rorq(sum, 32);
7308 if (UseBMI2Instructions) {
7309 multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
7310 }
7311 else {
7312 multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
7313 }
7314 // Store back in big endian from little endian
7315 rorq(sum, 0x20);
7316 movq(Address(out, offset, Address::times_4, 0), sum);
7317
7318 jmp(L_first_loop);
7319 bind(L_first_loop_exit);
7320 }
7321
7322 /**
7323 * Code for BigInteger::mulAdd() intrinsic
7324 *
7325 * rdi: out
7326 * rsi: in
7327 * r11: offs (out.length - offset)
7328 * rcx: len
7329 * r8: k
7330 * r12: tmp1
7331 * r13: tmp2
7332 * r14: tmp3
7333 * r15: tmp4
7334 * rbx: tmp5
7335 * Multiply the in[] by word k and add to out[], return the carry in rax
7336 */
7337 void MacroAssembler::mul_add(Register out, Register in, Register offs,
7338 Register len, Register k, Register tmp1, Register tmp2, Register tmp3,
7339 Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
7340
7341 Label L_carry, L_last_in, L_done;
7342
7343 // carry = 0;
7344 // for (int j=len-1; j >= 0; j--) {
7345 // long product = (in[j] & LONG_MASK) * kLong +
7346 // (out[offs] & LONG_MASK) + carry;
7347 // out[offs--] = (int)product;
7348 // carry = product >>> 32;
7349 // }
7350 //
7351 push(tmp1);
7352 push(tmp2);
7353 push(tmp3);
7354 push(tmp4);
7355 push(tmp5);
7356
7357 Register op2 = tmp2;
7358 const Register sum = tmp3;
7359 const Register op1 = tmp4;
7360 const Register carry = tmp5;
7361
7362 if (UseBMI2Instructions) {
7363 op2 = rdxReg;
7364 movl(op2, k);
7365 }
7366 else {
7367 movl(op2, k);
7368 }
7369
7370 xorq(carry, carry);
7371
7372 //First loop
7373
7374 //Multiply in[] by k in a 4 way unrolled loop using 128 bit by 32 bit multiply
7375 //The carry is in tmp5
7376 mul_add_128_x_32_loop(out, in, offs, len, tmp1, tmp2, tmp3, tmp4, tmp5, rdxReg, raxReg);
7377
7378 //Multiply the trailing in[] entry using 64 bit by 32 bit, if any
7379 decrementl(len);
7380 jccb(Assembler::negative, L_carry);
7381 decrementl(len);
7382 jccb(Assembler::negative, L_last_in);
7383
7384 movq(op1, Address(in, len, Address::times_4, 0));
7385 rorq(op1, 32);
7386
7387 subl(offs, 2);
7388 movq(sum, Address(out, offs, Address::times_4, 0));
7389 rorq(sum, 32);
7390
7391 if (UseBMI2Instructions) {
7392 multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
7393 }
7394 else {
7395 multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
7396 }
7397
7398 // Store back in big endian from little endian
7399 rorq(sum, 0x20);
7400 movq(Address(out, offs, Address::times_4, 0), sum);
7401
7402 testl(len, len);
7403 jccb(Assembler::zero, L_carry);
7404
7405 //Multiply the last in[] entry, if any
7406 bind(L_last_in);
7407 movl(op1, Address(in, 0));
7408 movl(sum, Address(out, offs, Address::times_4, -4));
7409
7410 movl(raxReg, k);
7411 mull(op1); //tmp4 * eax -> edx:eax
7412 addl(sum, carry);
7413 adcl(rdxReg, 0);
7414 addl(sum, raxReg);
7415 adcl(rdxReg, 0);
7416 movl(carry, rdxReg);
7417
7418 movl(Address(out, offs, Address::times_4, -4), sum);
7419
7420 bind(L_carry);
7421 //return tmp5/carry as carry in rax
7422 movl(rax, carry);
7423
7424 bind(L_done);
7425 pop(tmp5);
7426 pop(tmp4);
7427 pop(tmp3);
7428 pop(tmp2);
7429 pop(tmp1);
7430 }
7431 #endif
7432
7433 /**
7434 * Emits code to update CRC-32 with a byte value according to constants in table
7435 *
7436 * @param [in,out]crc Register containing the crc.
7437 * @param [in]val Register containing the byte to fold into the CRC.
7438 * @param [in]table Register containing the table of crc constants.
7439 *
7440 * uint32_t crc;
7441 * val = crc_table[(val ^ crc) & 0xFF];
7442 * crc = val ^ (crc >> 8);
7443 *
7444 */
7445 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
7446 xorl(val, crc);
7447 andl(val, 0xFF);
7448 shrl(crc, 8); // unsigned shift
7449 xorl(crc, Address(table, val, Address::times_4, 0));
7450 }
7451
7452 /**
7453 * Fold 128-bit data chunk
7454 */
7455 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset) {
7456 if (UseAVX > 0) {
7457 vpclmulhdq(xtmp, xK, xcrc); // [123:64]
7458 vpclmulldq(xcrc, xK, xcrc); // [63:0]
7459 vpxor(xcrc, xcrc, Address(buf, offset), 0 /* vector_len */);
7460 pxor(xcrc, xtmp);
7461 } else {
7462 movdqa(xtmp, xcrc);
7463 pclmulhdq(xtmp, xK); // [123:64]
7464 pclmulldq(xcrc, xK); // [63:0]
7465 pxor(xcrc, xtmp);
7466 movdqu(xtmp, Address(buf, offset));
7467 pxor(xcrc, xtmp);
7468 }
7469 }
7470
7471 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, XMMRegister xbuf) {
7472 if (UseAVX > 0) {
7473 vpclmulhdq(xtmp, xK, xcrc);
7474 vpclmulldq(xcrc, xK, xcrc);
7475 pxor(xcrc, xbuf);
7476 pxor(xcrc, xtmp);
7477 } else {
7478 movdqa(xtmp, xcrc);
7479 pclmulhdq(xtmp, xK);
7480 pclmulldq(xcrc, xK);
7481 pxor(xcrc, xbuf);
7482 pxor(xcrc, xtmp);
7483 }
7484 }
7485
7486 /**
7487 * 8-bit folds to compute 32-bit CRC
7488 *
7489 * uint64_t xcrc;
7490 * timesXtoThe32[xcrc & 0xFF] ^ (xcrc >> 8);
7491 */
7492 void MacroAssembler::fold_8bit_crc32(XMMRegister xcrc, Register table, XMMRegister xtmp, Register tmp) {
7493 movdl(tmp, xcrc);
7494 andl(tmp, 0xFF);
7495 movdl(xtmp, Address(table, tmp, Address::times_4, 0));
7496 psrldq(xcrc, 1); // unsigned shift one byte
7497 pxor(xcrc, xtmp);
7498 }
7499
7500 /**
7501 * uint32_t crc;
7502 * timesXtoThe32[crc & 0xFF] ^ (crc >> 8);
7503 */
7504 void MacroAssembler::fold_8bit_crc32(Register crc, Register table, Register tmp) {
7505 movl(tmp, crc);
7506 andl(tmp, 0xFF);
7507 shrl(crc, 8);
7508 xorl(crc, Address(table, tmp, Address::times_4, 0));
7509 }
7510
7511 /**
7512 * @param crc register containing existing CRC (32-bit)
7513 * @param buf register pointing to input byte buffer (byte*)
7514 * @param len register containing number of bytes
7515 * @param table register that will contain address of CRC table
7516 * @param tmp scratch register
7517 */
7518 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len, Register table, Register tmp) {
7519 assert_different_registers(crc, buf, len, table, tmp, rax);
7520
7521 Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
7522 Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
7523
7524 // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
7525 // context for the registers used, where all instructions below are using 128-bit mode
7526 // On EVEX without VL and BW, these instructions will all be AVX.
7527 lea(table, ExternalAddress(StubRoutines::crc_table_addr()));
7528 notl(crc); // ~crc
7529 cmpl(len, 16);
7530 jcc(Assembler::less, L_tail);
7531
7532 // Align buffer to 16 bytes
7533 movl(tmp, buf);
7534 andl(tmp, 0xF);
7535 jccb(Assembler::zero, L_aligned);
7536 subl(tmp, 16);
7537 addl(len, tmp);
7538
7539 align(4);
7540 BIND(L_align_loop);
7541 movsbl(rax, Address(buf, 0)); // load byte with sign extension
7542 update_byte_crc32(crc, rax, table);
7543 increment(buf);
7544 incrementl(tmp);
7545 jccb(Assembler::less, L_align_loop);
7546
7547 BIND(L_aligned);
7548 movl(tmp, len); // save
7549 shrl(len, 4);
7550 jcc(Assembler::zero, L_tail_restore);
7551
7552 // Fold crc into first bytes of vector
7553 movdqa(xmm1, Address(buf, 0));
7554 movdl(rax, xmm1);
7555 xorl(crc, rax);
7556 if (VM_Version::supports_sse4_1()) {
7557 pinsrd(xmm1, crc, 0);
7558 } else {
7559 pinsrw(xmm1, crc, 0);
7560 shrl(crc, 16);
7561 pinsrw(xmm1, crc, 1);
7562 }
7563 addptr(buf, 16);
7564 subl(len, 4); // len > 0
7565 jcc(Assembler::less, L_fold_tail);
7566
7567 movdqa(xmm2, Address(buf, 0));
7568 movdqa(xmm3, Address(buf, 16));
7569 movdqa(xmm4, Address(buf, 32));
7570 addptr(buf, 48);
7571 subl(len, 3);
7572 jcc(Assembler::lessEqual, L_fold_512b);
7573
7574 // Fold total 512 bits of polynomial on each iteration,
7575 // 128 bits per each of 4 parallel streams.
7576 movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 32), rscratch1);
7577
7578 align32();
7579 BIND(L_fold_512b_loop);
7580 fold_128bit_crc32(xmm1, xmm0, xmm5, buf, 0);
7581 fold_128bit_crc32(xmm2, xmm0, xmm5, buf, 16);
7582 fold_128bit_crc32(xmm3, xmm0, xmm5, buf, 32);
7583 fold_128bit_crc32(xmm4, xmm0, xmm5, buf, 48);
7584 addptr(buf, 64);
7585 subl(len, 4);
7586 jcc(Assembler::greater, L_fold_512b_loop);
7587
7588 // Fold 512 bits to 128 bits.
7589 BIND(L_fold_512b);
7590 movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
7591 fold_128bit_crc32(xmm1, xmm0, xmm5, xmm2);
7592 fold_128bit_crc32(xmm1, xmm0, xmm5, xmm3);
7593 fold_128bit_crc32(xmm1, xmm0, xmm5, xmm4);
7594
7595 // Fold the rest of 128 bits data chunks
7596 BIND(L_fold_tail);
7597 addl(len, 3);
7598 jccb(Assembler::lessEqual, L_fold_128b);
7599 movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
7600
7601 BIND(L_fold_tail_loop);
7602 fold_128bit_crc32(xmm1, xmm0, xmm5, buf, 0);
7603 addptr(buf, 16);
7604 decrementl(len);
7605 jccb(Assembler::greater, L_fold_tail_loop);
7606
7607 // Fold 128 bits in xmm1 down into 32 bits in crc register.
7608 BIND(L_fold_128b);
7609 movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr()), rscratch1);
7610 if (UseAVX > 0) {
7611 vpclmulqdq(xmm2, xmm0, xmm1, 0x1);
7612 vpand(xmm3, xmm0, xmm2, 0 /* vector_len */);
7613 vpclmulqdq(xmm0, xmm0, xmm3, 0x1);
7614 } else {
7615 movdqa(xmm2, xmm0);
7616 pclmulqdq(xmm2, xmm1, 0x1);
7617 movdqa(xmm3, xmm0);
7618 pand(xmm3, xmm2);
7619 pclmulqdq(xmm0, xmm3, 0x1);
7620 }
7621 psrldq(xmm1, 8);
7622 psrldq(xmm2, 4);
7623 pxor(xmm0, xmm1);
7624 pxor(xmm0, xmm2);
7625
7626 // 8 8-bit folds to compute 32-bit CRC.
7627 for (int j = 0; j < 4; j++) {
7628 fold_8bit_crc32(xmm0, table, xmm1, rax);
7629 }
7630 movdl(crc, xmm0); // mov 32 bits to general register
7631 for (int j = 0; j < 4; j++) {
7632 fold_8bit_crc32(crc, table, rax);
7633 }
7634
7635 BIND(L_tail_restore);
7636 movl(len, tmp); // restore
7637 BIND(L_tail);
7638 andl(len, 0xf);
7639 jccb(Assembler::zero, L_exit);
7640
7641 // Fold the rest of bytes
7642 align(4);
7643 BIND(L_tail_loop);
7644 movsbl(rax, Address(buf, 0)); // load byte with sign extension
7645 update_byte_crc32(crc, rax, table);
7646 increment(buf);
7647 decrementl(len);
7648 jccb(Assembler::greater, L_tail_loop);
7649
7650 BIND(L_exit);
7651 notl(crc); // ~c
7652 }
7653
7654 #ifdef _LP64
7655 // Helper function for AVX 512 CRC32
7656 // Fold 512-bit data chunks
7657 void MacroAssembler::fold512bit_crc32_avx512(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf,
7658 Register pos, int offset) {
7659 evmovdquq(xmm3, Address(buf, pos, Address::times_1, offset), Assembler::AVX_512bit);
7660 evpclmulqdq(xtmp, xcrc, xK, 0x10, Assembler::AVX_512bit); // [123:64]
7661 evpclmulqdq(xmm2, xcrc, xK, 0x01, Assembler::AVX_512bit); // [63:0]
7662 evpxorq(xcrc, xtmp, xmm2, Assembler::AVX_512bit /* vector_len */);
7663 evpxorq(xcrc, xcrc, xmm3, Assembler::AVX_512bit /* vector_len */);
7664 }
7665
7666 // Helper function for AVX 512 CRC32
7667 // Compute CRC32 for < 256B buffers
7668 void MacroAssembler::kernel_crc32_avx512_256B(Register crc, Register buf, Register len, Register table, Register pos,
7669 Register tmp1, Register tmp2, Label& L_barrett, Label& L_16B_reduction_loop,
7670 Label& L_get_last_two_xmms, Label& L_128_done, Label& L_cleanup) {
7671
7672 Label L_less_than_32, L_exact_16_left, L_less_than_16_left;
7673 Label L_less_than_8_left, L_less_than_4_left, L_less_than_2_left, L_zero_left;
7674 Label L_only_less_than_4, L_only_less_than_3, L_only_less_than_2;
7675
7676 // check if there is enough buffer to be able to fold 16B at a time
7677 cmpl(len, 32);
7678 jcc(Assembler::less, L_less_than_32);
7679
7680 // if there is, load the constants
7681 movdqu(xmm10, Address(table, 1 * 16)); //rk1 and rk2 in xmm10
7682 movdl(xmm0, crc); // get the initial crc value
7683 movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
7684 pxor(xmm7, xmm0);
7685
7686 // update the buffer pointer
7687 addl(pos, 16);
7688 //update the counter.subtract 32 instead of 16 to save one instruction from the loop
7689 subl(len, 32);
7690 jmp(L_16B_reduction_loop);
7691
7692 bind(L_less_than_32);
7693 //mov initial crc to the return value. this is necessary for zero - length buffers.
7694 movl(rax, crc);
7695 testl(len, len);
7696 jcc(Assembler::equal, L_cleanup);
7697
7698 movdl(xmm0, crc); //get the initial crc value
7699
7700 cmpl(len, 16);
7701 jcc(Assembler::equal, L_exact_16_left);
7702 jcc(Assembler::less, L_less_than_16_left);
7703
7704 movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
7705 pxor(xmm7, xmm0); //xor the initial crc value
7706 addl(pos, 16);
7707 subl(len, 16);
7708 movdqu(xmm10, Address(table, 1 * 16)); // rk1 and rk2 in xmm10
7709 jmp(L_get_last_two_xmms);
7710
7711 bind(L_less_than_16_left);
7712 //use stack space to load data less than 16 bytes, zero - out the 16B in memory first.
7713 pxor(xmm1, xmm1);
7714 movptr(tmp1, rsp);
7715 movdqu(Address(tmp1, 0 * 16), xmm1);
7716
7717 cmpl(len, 4);
7718 jcc(Assembler::less, L_only_less_than_4);
7719
7720 //backup the counter value
7721 movl(tmp2, len);
7722 cmpl(len, 8);
7723 jcc(Assembler::less, L_less_than_8_left);
7724
7725 //load 8 Bytes
7726 movq(rax, Address(buf, pos, Address::times_1, 0 * 16));
7727 movq(Address(tmp1, 0 * 16), rax);
7728 addptr(tmp1, 8);
7729 subl(len, 8);
7730 addl(pos, 8);
7731
7732 bind(L_less_than_8_left);
7733 cmpl(len, 4);
7734 jcc(Assembler::less, L_less_than_4_left);
7735
7736 //load 4 Bytes
7737 movl(rax, Address(buf, pos, Address::times_1, 0));
7738 movl(Address(tmp1, 0 * 16), rax);
7739 addptr(tmp1, 4);
7740 subl(len, 4);
7741 addl(pos, 4);
7742
7743 bind(L_less_than_4_left);
7744 cmpl(len, 2);
7745 jcc(Assembler::less, L_less_than_2_left);
7746
7747 // load 2 Bytes
7748 movw(rax, Address(buf, pos, Address::times_1, 0));
7749 movl(Address(tmp1, 0 * 16), rax);
7750 addptr(tmp1, 2);
7751 subl(len, 2);
7752 addl(pos, 2);
7753
7754 bind(L_less_than_2_left);
7755 cmpl(len, 1);
7756 jcc(Assembler::less, L_zero_left);
7757
7758 // load 1 Byte
7759 movb(rax, Address(buf, pos, Address::times_1, 0));
7760 movb(Address(tmp1, 0 * 16), rax);
7761
7762 bind(L_zero_left);
7763 movdqu(xmm7, Address(rsp, 0));
7764 pxor(xmm7, xmm0); //xor the initial crc value
7765
7766 lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
7767 movdqu(xmm0, Address(rax, tmp2));
7768 pshufb(xmm7, xmm0);
7769 jmp(L_128_done);
7770
7771 bind(L_exact_16_left);
7772 movdqu(xmm7, Address(buf, pos, Address::times_1, 0));
7773 pxor(xmm7, xmm0); //xor the initial crc value
7774 jmp(L_128_done);
7775
7776 bind(L_only_less_than_4);
7777 cmpl(len, 3);
7778 jcc(Assembler::less, L_only_less_than_3);
7779
7780 // load 3 Bytes
7781 movb(rax, Address(buf, pos, Address::times_1, 0));
7782 movb(Address(tmp1, 0), rax);
7783
7784 movb(rax, Address(buf, pos, Address::times_1, 1));
7785 movb(Address(tmp1, 1), rax);
7786
7787 movb(rax, Address(buf, pos, Address::times_1, 2));
7788 movb(Address(tmp1, 2), rax);
7789
7790 movdqu(xmm7, Address(rsp, 0));
7791 pxor(xmm7, xmm0); //xor the initial crc value
7792
7793 pslldq(xmm7, 0x5);
7794 jmp(L_barrett);
7795 bind(L_only_less_than_3);
7796 cmpl(len, 2);
7797 jcc(Assembler::less, L_only_less_than_2);
7798
7799 // load 2 Bytes
7800 movb(rax, Address(buf, pos, Address::times_1, 0));
7801 movb(Address(tmp1, 0), rax);
7802
7803 movb(rax, Address(buf, pos, Address::times_1, 1));
7804 movb(Address(tmp1, 1), rax);
7805
7806 movdqu(xmm7, Address(rsp, 0));
7807 pxor(xmm7, xmm0); //xor the initial crc value
7808
7809 pslldq(xmm7, 0x6);
7810 jmp(L_barrett);
7811
7812 bind(L_only_less_than_2);
7813 //load 1 Byte
7814 movb(rax, Address(buf, pos, Address::times_1, 0));
7815 movb(Address(tmp1, 0), rax);
7816
7817 movdqu(xmm7, Address(rsp, 0));
7818 pxor(xmm7, xmm0); //xor the initial crc value
7819
7820 pslldq(xmm7, 0x7);
7821 }
7822
7823 /**
7824 * Compute CRC32 using AVX512 instructions
7825 * param crc register containing existing CRC (32-bit)
7826 * param buf register pointing to input byte buffer (byte*)
7827 * param len register containing number of bytes
7828 * param table address of crc or crc32c table
7829 * param tmp1 scratch register
7830 * param tmp2 scratch register
7831 * return rax result register
7832 *
7833 * This routine is identical for crc32c with the exception of the precomputed constant
7834 * table which will be passed as the table argument. The calculation steps are
7835 * the same for both variants.
7836 */
7837 void MacroAssembler::kernel_crc32_avx512(Register crc, Register buf, Register len, Register table, Register tmp1, Register tmp2) {
7838 assert_different_registers(crc, buf, len, table, tmp1, tmp2, rax, r12);
7839
7840 Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
7841 Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
7842 Label L_less_than_256, L_fold_128_B_loop, L_fold_256_B_loop;
7843 Label L_fold_128_B_register, L_final_reduction_for_128, L_16B_reduction_loop;
7844 Label L_128_done, L_get_last_two_xmms, L_barrett, L_cleanup;
7845
7846 const Register pos = r12;
7847 push(r12);
7848 subptr(rsp, 16 * 2 + 8);
7849
7850 // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
7851 // context for the registers used, where all instructions below are using 128-bit mode
7852 // On EVEX without VL and BW, these instructions will all be AVX.
7853 movl(pos, 0);
7854
7855 // check if smaller than 256B
7856 cmpl(len, 256);
7857 jcc(Assembler::less, L_less_than_256);
7858
7859 // load the initial crc value
7860 movdl(xmm10, crc);
7861
7862 // receive the initial 64B data, xor the initial crc value
7863 evmovdquq(xmm0, Address(buf, pos, Address::times_1, 0 * 64), Assembler::AVX_512bit);
7864 evmovdquq(xmm4, Address(buf, pos, Address::times_1, 1 * 64), Assembler::AVX_512bit);
7865 evpxorq(xmm0, xmm0, xmm10, Assembler::AVX_512bit);
7866 evbroadcasti32x4(xmm10, Address(table, 2 * 16), Assembler::AVX_512bit); //zmm10 has rk3 and rk4
7867
7868 subl(len, 256);
7869 cmpl(len, 256);
7870 jcc(Assembler::less, L_fold_128_B_loop);
7871
7872 evmovdquq(xmm7, Address(buf, pos, Address::times_1, 2 * 64), Assembler::AVX_512bit);
7873 evmovdquq(xmm8, Address(buf, pos, Address::times_1, 3 * 64), Assembler::AVX_512bit);
7874 evbroadcasti32x4(xmm16, Address(table, 0 * 16), Assembler::AVX_512bit); //zmm16 has rk-1 and rk-2
7875 subl(len, 256);
7876
7877 bind(L_fold_256_B_loop);
7878 addl(pos, 256);
7879 fold512bit_crc32_avx512(xmm0, xmm16, xmm1, buf, pos, 0 * 64);
7880 fold512bit_crc32_avx512(xmm4, xmm16, xmm1, buf, pos, 1 * 64);
7881 fold512bit_crc32_avx512(xmm7, xmm16, xmm1, buf, pos, 2 * 64);
7882 fold512bit_crc32_avx512(xmm8, xmm16, xmm1, buf, pos, 3 * 64);
7883
7884 subl(len, 256);
7885 jcc(Assembler::greaterEqual, L_fold_256_B_loop);
7886
7887 // Fold 256 into 128
7888 addl(pos, 256);
7889 evpclmulqdq(xmm1, xmm0, xmm10, 0x01, Assembler::AVX_512bit);
7890 evpclmulqdq(xmm2, xmm0, xmm10, 0x10, Assembler::AVX_512bit);
7891 vpternlogq(xmm7, 0x96, xmm1, xmm2, Assembler::AVX_512bit); // xor ABC
7892
7893 evpclmulqdq(xmm5, xmm4, xmm10, 0x01, Assembler::AVX_512bit);
7894 evpclmulqdq(xmm6, xmm4, xmm10, 0x10, Assembler::AVX_512bit);
7895 vpternlogq(xmm8, 0x96, xmm5, xmm6, Assembler::AVX_512bit); // xor ABC
7896
7897 evmovdquq(xmm0, xmm7, Assembler::AVX_512bit);
7898 evmovdquq(xmm4, xmm8, Assembler::AVX_512bit);
7899
7900 addl(len, 128);
7901 jmp(L_fold_128_B_register);
7902
7903 // at this section of the code, there is 128 * x + y(0 <= y<128) bytes of buffer.The fold_128_B_loop
7904 // loop will fold 128B at a time until we have 128 + y Bytes of buffer
7905
7906 // fold 128B at a time.This section of the code folds 8 xmm registers in parallel
7907 bind(L_fold_128_B_loop);
7908 addl(pos, 128);
7909 fold512bit_crc32_avx512(xmm0, xmm10, xmm1, buf, pos, 0 * 64);
7910 fold512bit_crc32_avx512(xmm4, xmm10, xmm1, buf, pos, 1 * 64);
7911
7912 subl(len, 128);
7913 jcc(Assembler::greaterEqual, L_fold_128_B_loop);
7914
7915 addl(pos, 128);
7916
7917 // at this point, the buffer pointer is pointing at the last y Bytes of the buffer, where 0 <= y < 128
7918 // the 128B of folded data is in 8 of the xmm registers : xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7
7919 bind(L_fold_128_B_register);
7920 evmovdquq(xmm16, Address(table, 5 * 16), Assembler::AVX_512bit); // multiply by rk9-rk16
7921 evmovdquq(xmm11, Address(table, 9 * 16), Assembler::AVX_512bit); // multiply by rk17-rk20, rk1,rk2, 0,0
7922 evpclmulqdq(xmm1, xmm0, xmm16, 0x01, Assembler::AVX_512bit);
7923 evpclmulqdq(xmm2, xmm0, xmm16, 0x10, Assembler::AVX_512bit);
7924 // save last that has no multiplicand
7925 vextracti64x2(xmm7, xmm4, 3);
7926
7927 evpclmulqdq(xmm5, xmm4, xmm11, 0x01, Assembler::AVX_512bit);
7928 evpclmulqdq(xmm6, xmm4, xmm11, 0x10, Assembler::AVX_512bit);
7929 // Needed later in reduction loop
7930 movdqu(xmm10, Address(table, 1 * 16));
7931 vpternlogq(xmm1, 0x96, xmm2, xmm5, Assembler::AVX_512bit); // xor ABC
7932 vpternlogq(xmm1, 0x96, xmm6, xmm7, Assembler::AVX_512bit); // xor ABC
7933
7934 // Swap 1,0,3,2 - 01 00 11 10
7935 evshufi64x2(xmm8, xmm1, xmm1, 0x4e, Assembler::AVX_512bit);
7936 evpxorq(xmm8, xmm8, xmm1, Assembler::AVX_256bit);
7937 vextracti128(xmm5, xmm8, 1);
7938 evpxorq(xmm7, xmm5, xmm8, Assembler::AVX_128bit);
7939
7940 // instead of 128, we add 128 - 16 to the loop counter to save 1 instruction from the loop
7941 // instead of a cmp instruction, we use the negative flag with the jl instruction
7942 addl(len, 128 - 16);
7943 jcc(Assembler::less, L_final_reduction_for_128);
7944
7945 bind(L_16B_reduction_loop);
7946 vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
7947 vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
7948 vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
7949 movdqu(xmm0, Address(buf, pos, Address::times_1, 0 * 16));
7950 vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
7951 addl(pos, 16);
7952 subl(len, 16);
7953 jcc(Assembler::greaterEqual, L_16B_reduction_loop);
7954
7955 bind(L_final_reduction_for_128);
7956 addl(len, 16);
7957 jcc(Assembler::equal, L_128_done);
7958
7959 bind(L_get_last_two_xmms);
7960 movdqu(xmm2, xmm7);
7961 addl(pos, len);
7962 movdqu(xmm1, Address(buf, pos, Address::times_1, -16));
7963 subl(pos, len);
7964
7965 // get rid of the extra data that was loaded before
7966 // load the shift constant
7967 lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
7968 movdqu(xmm0, Address(rax, len));
7969 addl(rax, len);
7970
7971 vpshufb(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
7972 //Change mask to 512
7973 vpxor(xmm0, xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 2 * 16), Assembler::AVX_128bit, tmp2);
7974 vpshufb(xmm2, xmm2, xmm0, Assembler::AVX_128bit);
7975
7976 blendvpb(xmm2, xmm2, xmm1, xmm0, Assembler::AVX_128bit);
7977 vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
7978 vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
7979 vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
7980 vpxor(xmm7, xmm7, xmm2, Assembler::AVX_128bit);
7981
7982 bind(L_128_done);
7983 // compute crc of a 128-bit value
7984 movdqu(xmm10, Address(table, 3 * 16));
7985 movdqu(xmm0, xmm7);
7986
7987 // 64b fold
7988 vpclmulqdq(xmm7, xmm7, xmm10, 0x0);
7989 vpsrldq(xmm0, xmm0, 0x8, Assembler::AVX_128bit);
7990 vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
7991
7992 // 32b fold
7993 movdqu(xmm0, xmm7);
7994 vpslldq(xmm7, xmm7, 0x4, Assembler::AVX_128bit);
7995 vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
7996 vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
7997 jmp(L_barrett);
7998
7999 bind(L_less_than_256);
8000 kernel_crc32_avx512_256B(crc, buf, len, table, pos, tmp1, tmp2, L_barrett, L_16B_reduction_loop, L_get_last_two_xmms, L_128_done, L_cleanup);
8001
8002 //barrett reduction
8003 bind(L_barrett);
8004 vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 1 * 16), Assembler::AVX_128bit, tmp2);
8005 movdqu(xmm1, xmm7);
8006 movdqu(xmm2, xmm7);
8007 movdqu(xmm10, Address(table, 4 * 16));
8008
8009 pclmulqdq(xmm7, xmm10, 0x0);
8010 pxor(xmm7, xmm2);
8011 vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr()), Assembler::AVX_128bit, tmp2);
8012 movdqu(xmm2, xmm7);
8013 pclmulqdq(xmm7, xmm10, 0x10);
8014 pxor(xmm7, xmm2);
8015 pxor(xmm7, xmm1);
8016 pextrd(crc, xmm7, 2);
8017
8018 bind(L_cleanup);
8019 addptr(rsp, 16 * 2 + 8);
8020 pop(r12);
8021 }
8022
8023 // S. Gueron / Information Processing Letters 112 (2012) 184
8024 // Algorithm 4: Computing carry-less multiplication using a precomputed lookup table.
8025 // Input: A 32 bit value B = [byte3, byte2, byte1, byte0].
8026 // Output: the 64-bit carry-less product of B * CONST
8027 void MacroAssembler::crc32c_ipl_alg4(Register in, uint32_t n,
8028 Register tmp1, Register tmp2, Register tmp3) {
8029 lea(tmp3, ExternalAddress(StubRoutines::crc32c_table_addr()));
8030 if (n > 0) {
8031 addq(tmp3, n * 256 * 8);
8032 }
8033 // Q1 = TABLEExt[n][B & 0xFF];
8034 movl(tmp1, in);
8035 andl(tmp1, 0x000000FF);
8036 shll(tmp1, 3);
8037 addq(tmp1, tmp3);
8038 movq(tmp1, Address(tmp1, 0));
8039
8040 // Q2 = TABLEExt[n][B >> 8 & 0xFF];
8041 movl(tmp2, in);
8042 shrl(tmp2, 8);
8043 andl(tmp2, 0x000000FF);
8044 shll(tmp2, 3);
8045 addq(tmp2, tmp3);
8046 movq(tmp2, Address(tmp2, 0));
8047
8048 shlq(tmp2, 8);
8049 xorq(tmp1, tmp2);
8050
8051 // Q3 = TABLEExt[n][B >> 16 & 0xFF];
8052 movl(tmp2, in);
8053 shrl(tmp2, 16);
8054 andl(tmp2, 0x000000FF);
8055 shll(tmp2, 3);
8056 addq(tmp2, tmp3);
8057 movq(tmp2, Address(tmp2, 0));
8058
8059 shlq(tmp2, 16);
8060 xorq(tmp1, tmp2);
8061
8062 // Q4 = TABLEExt[n][B >> 24 & 0xFF];
8063 shrl(in, 24);
8064 andl(in, 0x000000FF);
8065 shll(in, 3);
8066 addq(in, tmp3);
8067 movq(in, Address(in, 0));
8068
8069 shlq(in, 24);
8070 xorq(in, tmp1);
8071 // return Q1 ^ Q2 << 8 ^ Q3 << 16 ^ Q4 << 24;
8072 }
8073
8074 void MacroAssembler::crc32c_pclmulqdq(XMMRegister w_xtmp1,
8075 Register in_out,
8076 uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported,
8077 XMMRegister w_xtmp2,
8078 Register tmp1,
8079 Register n_tmp2, Register n_tmp3) {
8080 if (is_pclmulqdq_supported) {
8081 movdl(w_xtmp1, in_out); // modified blindly
8082
8083 movl(tmp1, const_or_pre_comp_const_index);
8084 movdl(w_xtmp2, tmp1);
8085 pclmulqdq(w_xtmp1, w_xtmp2, 0);
8086
8087 movdq(in_out, w_xtmp1);
8088 } else {
8089 crc32c_ipl_alg4(in_out, const_or_pre_comp_const_index, tmp1, n_tmp2, n_tmp3);
8090 }
8091 }
8092
8093 // Recombination Alternative 2: No bit-reflections
8094 // T1 = (CRC_A * U1) << 1
8095 // T2 = (CRC_B * U2) << 1
8096 // C1 = T1 >> 32
8097 // C2 = T2 >> 32
8098 // T1 = T1 & 0xFFFFFFFF
8099 // T2 = T2 & 0xFFFFFFFF
8100 // T1 = CRC32(0, T1)
8101 // T2 = CRC32(0, T2)
8102 // C1 = C1 ^ T1
8103 // C2 = C2 ^ T2
8104 // CRC = C1 ^ C2 ^ CRC_C
8105 void MacroAssembler::crc32c_rec_alt2(uint32_t const_or_pre_comp_const_index_u1, uint32_t const_or_pre_comp_const_index_u2, bool is_pclmulqdq_supported, Register in_out, Register in1, Register in2,
8106 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
8107 Register tmp1, Register tmp2,
8108 Register n_tmp3) {
8109 crc32c_pclmulqdq(w_xtmp1, in_out, const_or_pre_comp_const_index_u1, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
8110 crc32c_pclmulqdq(w_xtmp2, in1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
8111 shlq(in_out, 1);
8112 movl(tmp1, in_out);
8113 shrq(in_out, 32);
8114 xorl(tmp2, tmp2);
8115 crc32(tmp2, tmp1, 4);
8116 xorl(in_out, tmp2); // we don't care about upper 32 bit contents here
8117 shlq(in1, 1);
8118 movl(tmp1, in1);
8119 shrq(in1, 32);
8120 xorl(tmp2, tmp2);
8121 crc32(tmp2, tmp1, 4);
8122 xorl(in1, tmp2);
8123 xorl(in_out, in1);
8124 xorl(in_out, in2);
8125 }
8126
8127 // Set N to predefined value
8128 // Subtract from a length of a buffer
8129 // execute in a loop:
8130 // CRC_A = 0xFFFFFFFF, CRC_B = 0, CRC_C = 0
8131 // for i = 1 to N do
8132 // CRC_A = CRC32(CRC_A, A[i])
8133 // CRC_B = CRC32(CRC_B, B[i])
8134 // CRC_C = CRC32(CRC_C, C[i])
8135 // end for
8136 // Recombine
8137 void MacroAssembler::crc32c_proc_chunk(uint32_t size, uint32_t const_or_pre_comp_const_index_u1, uint32_t const_or_pre_comp_const_index_u2, bool is_pclmulqdq_supported,
8138 Register in_out1, Register in_out2, Register in_out3,
8139 Register tmp1, Register tmp2, Register tmp3,
8140 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
8141 Register tmp4, Register tmp5,
8142 Register n_tmp6) {
8143 Label L_processPartitions;
8144 Label L_processPartition;
8145 Label L_exit;
8146
8147 bind(L_processPartitions);
8148 cmpl(in_out1, 3 * size);
8149 jcc(Assembler::less, L_exit);
8150 xorl(tmp1, tmp1);
8151 xorl(tmp2, tmp2);
8152 movq(tmp3, in_out2);
8153 addq(tmp3, size);
8154
8155 bind(L_processPartition);
8156 crc32(in_out3, Address(in_out2, 0), 8);
8157 crc32(tmp1, Address(in_out2, size), 8);
8158 crc32(tmp2, Address(in_out2, size * 2), 8);
8159 addq(in_out2, 8);
8160 cmpq(in_out2, tmp3);
8161 jcc(Assembler::less, L_processPartition);
8162 crc32c_rec_alt2(const_or_pre_comp_const_index_u1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, in_out3, tmp1, tmp2,
8163 w_xtmp1, w_xtmp2, w_xtmp3,
8164 tmp4, tmp5,
8165 n_tmp6);
8166 addq(in_out2, 2 * size);
8167 subl(in_out1, 3 * size);
8168 jmp(L_processPartitions);
8169
8170 bind(L_exit);
8171 }
8172 #else
8173 void MacroAssembler::crc32c_ipl_alg4(Register in_out, uint32_t n,
8174 Register tmp1, Register tmp2, Register tmp3,
8175 XMMRegister xtmp1, XMMRegister xtmp2) {
8176 lea(tmp3, ExternalAddress(StubRoutines::crc32c_table_addr()));
8177 if (n > 0) {
8178 addl(tmp3, n * 256 * 8);
8179 }
8180 // Q1 = TABLEExt[n][B & 0xFF];
8181 movl(tmp1, in_out);
8182 andl(tmp1, 0x000000FF);
8183 shll(tmp1, 3);
8184 addl(tmp1, tmp3);
8185 movq(xtmp1, Address(tmp1, 0));
8186
8187 // Q2 = TABLEExt[n][B >> 8 & 0xFF];
8188 movl(tmp2, in_out);
8189 shrl(tmp2, 8);
8190 andl(tmp2, 0x000000FF);
8191 shll(tmp2, 3);
8192 addl(tmp2, tmp3);
8193 movq(xtmp2, Address(tmp2, 0));
8194
8195 psllq(xtmp2, 8);
8196 pxor(xtmp1, xtmp2);
8197
8198 // Q3 = TABLEExt[n][B >> 16 & 0xFF];
8199 movl(tmp2, in_out);
8200 shrl(tmp2, 16);
8201 andl(tmp2, 0x000000FF);
8202 shll(tmp2, 3);
8203 addl(tmp2, tmp3);
8204 movq(xtmp2, Address(tmp2, 0));
8205
8206 psllq(xtmp2, 16);
8207 pxor(xtmp1, xtmp2);
8208
8209 // Q4 = TABLEExt[n][B >> 24 & 0xFF];
8210 shrl(in_out, 24);
8211 andl(in_out, 0x000000FF);
8212 shll(in_out, 3);
8213 addl(in_out, tmp3);
8214 movq(xtmp2, Address(in_out, 0));
8215
8216 psllq(xtmp2, 24);
8217 pxor(xtmp1, xtmp2); // Result in CXMM
8218 // return Q1 ^ Q2 << 8 ^ Q3 << 16 ^ Q4 << 24;
8219 }
8220
8221 void MacroAssembler::crc32c_pclmulqdq(XMMRegister w_xtmp1,
8222 Register in_out,
8223 uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported,
8224 XMMRegister w_xtmp2,
8225 Register tmp1,
8226 Register n_tmp2, Register n_tmp3) {
8227 if (is_pclmulqdq_supported) {
8228 movdl(w_xtmp1, in_out);
8229
8230 movl(tmp1, const_or_pre_comp_const_index);
8231 movdl(w_xtmp2, tmp1);
8232 pclmulqdq(w_xtmp1, w_xtmp2, 0);
8233 // Keep result in XMM since GPR is 32 bit in length
8234 } else {
8235 crc32c_ipl_alg4(in_out, const_or_pre_comp_const_index, tmp1, n_tmp2, n_tmp3, w_xtmp1, w_xtmp2);
8236 }
8237 }
8238
8239 void MacroAssembler::crc32c_rec_alt2(uint32_t const_or_pre_comp_const_index_u1, uint32_t const_or_pre_comp_const_index_u2, bool is_pclmulqdq_supported, Register in_out, Register in1, Register in2,
8240 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
8241 Register tmp1, Register tmp2,
8242 Register n_tmp3) {
8243 crc32c_pclmulqdq(w_xtmp1, in_out, const_or_pre_comp_const_index_u1, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
8244 crc32c_pclmulqdq(w_xtmp2, in1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
8245
8246 psllq(w_xtmp1, 1);
8247 movdl(tmp1, w_xtmp1);
8248 psrlq(w_xtmp1, 32);
8249 movdl(in_out, w_xtmp1);
8250
8251 xorl(tmp2, tmp2);
8252 crc32(tmp2, tmp1, 4);
8253 xorl(in_out, tmp2);
8254
8255 psllq(w_xtmp2, 1);
8256 movdl(tmp1, w_xtmp2);
8257 psrlq(w_xtmp2, 32);
8258 movdl(in1, w_xtmp2);
8259
8260 xorl(tmp2, tmp2);
8261 crc32(tmp2, tmp1, 4);
8262 xorl(in1, tmp2);
8263 xorl(in_out, in1);
8264 xorl(in_out, in2);
8265 }
8266
8267 void MacroAssembler::crc32c_proc_chunk(uint32_t size, uint32_t const_or_pre_comp_const_index_u1, uint32_t const_or_pre_comp_const_index_u2, bool is_pclmulqdq_supported,
8268 Register in_out1, Register in_out2, Register in_out3,
8269 Register tmp1, Register tmp2, Register tmp3,
8270 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
8271 Register tmp4, Register tmp5,
8272 Register n_tmp6) {
8273 Label L_processPartitions;
8274 Label L_processPartition;
8275 Label L_exit;
8276
8277 bind(L_processPartitions);
8278 cmpl(in_out1, 3 * size);
8279 jcc(Assembler::less, L_exit);
8280 xorl(tmp1, tmp1);
8281 xorl(tmp2, tmp2);
8282 movl(tmp3, in_out2);
8283 addl(tmp3, size);
8284
8285 bind(L_processPartition);
8286 crc32(in_out3, Address(in_out2, 0), 4);
8287 crc32(tmp1, Address(in_out2, size), 4);
8288 crc32(tmp2, Address(in_out2, size*2), 4);
8289 crc32(in_out3, Address(in_out2, 0+4), 4);
8290 crc32(tmp1, Address(in_out2, size+4), 4);
8291 crc32(tmp2, Address(in_out2, size*2+4), 4);
8292 addl(in_out2, 8);
8293 cmpl(in_out2, tmp3);
8294 jcc(Assembler::less, L_processPartition);
8295
8296 push(tmp3);
8297 push(in_out1);
8298 push(in_out2);
8299 tmp4 = tmp3;
8300 tmp5 = in_out1;
8301 n_tmp6 = in_out2;
8302
8303 crc32c_rec_alt2(const_or_pre_comp_const_index_u1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, in_out3, tmp1, tmp2,
8304 w_xtmp1, w_xtmp2, w_xtmp3,
8305 tmp4, tmp5,
8306 n_tmp6);
8307
8308 pop(in_out2);
8309 pop(in_out1);
8310 pop(tmp3);
8311
8312 addl(in_out2, 2 * size);
8313 subl(in_out1, 3 * size);
8314 jmp(L_processPartitions);
8315
8316 bind(L_exit);
8317 }
8318 #endif //LP64
8319
8320 #ifdef _LP64
8321 // Algorithm 2: Pipelined usage of the CRC32 instruction.
8322 // Input: A buffer I of L bytes.
8323 // Output: the CRC32C value of the buffer.
8324 // Notations:
8325 // Write L = 24N + r, with N = floor (L/24).
8326 // r = L mod 24 (0 <= r < 24).
8327 // Consider I as the concatenation of A|B|C|R, where A, B, C, each,
8328 // N quadwords, and R consists of r bytes.
8329 // A[j] = I [8j+7:8j], j= 0, 1, ..., N-1
8330 // B[j] = I [N + 8j+7:N + 8j], j= 0, 1, ..., N-1
8331 // C[j] = I [2N + 8j+7:2N + 8j], j= 0, 1, ..., N-1
8332 // if r > 0 R[j] = I [3N +j], j= 0, 1, ...,r-1
8333 void MacroAssembler::crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2,
8334 Register tmp1, Register tmp2, Register tmp3,
8335 Register tmp4, Register tmp5, Register tmp6,
8336 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
8337 bool is_pclmulqdq_supported) {
8338 uint32_t const_or_pre_comp_const_index[CRC32C_NUM_PRECOMPUTED_CONSTANTS];
8339 Label L_wordByWord;
8340 Label L_byteByByteProlog;
8341 Label L_byteByByte;
8342 Label L_exit;
8343
8344 if (is_pclmulqdq_supported ) {
8345 const_or_pre_comp_const_index[1] = *(uint32_t *)StubRoutines::_crc32c_table_addr;
8346 const_or_pre_comp_const_index[0] = *((uint32_t *)StubRoutines::_crc32c_table_addr+1);
8347
8348 const_or_pre_comp_const_index[3] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 2);
8349 const_or_pre_comp_const_index[2] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 3);
8350
8351 const_or_pre_comp_const_index[5] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 4);
8352 const_or_pre_comp_const_index[4] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 5);
8353 assert((CRC32C_NUM_PRECOMPUTED_CONSTANTS - 1 ) == 5, "Checking whether you declared all of the constants based on the number of \"chunks\"");
8354 } else {
8355 const_or_pre_comp_const_index[0] = 1;
8356 const_or_pre_comp_const_index[1] = 0;
8357
8358 const_or_pre_comp_const_index[2] = 3;
8359 const_or_pre_comp_const_index[3] = 2;
8360
8361 const_or_pre_comp_const_index[4] = 5;
8362 const_or_pre_comp_const_index[5] = 4;
8363 }
8364 crc32c_proc_chunk(CRC32C_HIGH, const_or_pre_comp_const_index[0], const_or_pre_comp_const_index[1], is_pclmulqdq_supported,
8365 in2, in1, in_out,
8366 tmp1, tmp2, tmp3,
8367 w_xtmp1, w_xtmp2, w_xtmp3,
8368 tmp4, tmp5,
8369 tmp6);
8370 crc32c_proc_chunk(CRC32C_MIDDLE, const_or_pre_comp_const_index[2], const_or_pre_comp_const_index[3], is_pclmulqdq_supported,
8371 in2, in1, in_out,
8372 tmp1, tmp2, tmp3,
8373 w_xtmp1, w_xtmp2, w_xtmp3,
8374 tmp4, tmp5,
8375 tmp6);
8376 crc32c_proc_chunk(CRC32C_LOW, const_or_pre_comp_const_index[4], const_or_pre_comp_const_index[5], is_pclmulqdq_supported,
8377 in2, in1, in_out,
8378 tmp1, tmp2, tmp3,
8379 w_xtmp1, w_xtmp2, w_xtmp3,
8380 tmp4, tmp5,
8381 tmp6);
8382 movl(tmp1, in2);
8383 andl(tmp1, 0x00000007);
8384 negl(tmp1);
8385 addl(tmp1, in2);
8386 addq(tmp1, in1);
8387
8388 cmpq(in1, tmp1);
8389 jccb(Assembler::greaterEqual, L_byteByByteProlog);
8390 align(16);
8391 BIND(L_wordByWord);
8392 crc32(in_out, Address(in1, 0), 8);
8393 addq(in1, 8);
8394 cmpq(in1, tmp1);
8395 jcc(Assembler::less, L_wordByWord);
8396
8397 BIND(L_byteByByteProlog);
8398 andl(in2, 0x00000007);
8399 movl(tmp2, 1);
8400
8401 cmpl(tmp2, in2);
8402 jccb(Assembler::greater, L_exit);
8403 BIND(L_byteByByte);
8404 crc32(in_out, Address(in1, 0), 1);
8405 incq(in1);
8406 incl(tmp2);
8407 cmpl(tmp2, in2);
8408 jcc(Assembler::lessEqual, L_byteByByte);
8409
8410 BIND(L_exit);
8411 }
8412 #else
8413 void MacroAssembler::crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2,
8414 Register tmp1, Register tmp2, Register tmp3,
8415 Register tmp4, Register tmp5, Register tmp6,
8416 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
8417 bool is_pclmulqdq_supported) {
8418 uint32_t const_or_pre_comp_const_index[CRC32C_NUM_PRECOMPUTED_CONSTANTS];
8419 Label L_wordByWord;
8420 Label L_byteByByteProlog;
8421 Label L_byteByByte;
8422 Label L_exit;
8423
8424 if (is_pclmulqdq_supported) {
8425 const_or_pre_comp_const_index[1] = *(uint32_t *)StubRoutines::_crc32c_table_addr;
8426 const_or_pre_comp_const_index[0] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 1);
8427
8428 const_or_pre_comp_const_index[3] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 2);
8429 const_or_pre_comp_const_index[2] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 3);
8430
8431 const_or_pre_comp_const_index[5] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 4);
8432 const_or_pre_comp_const_index[4] = *((uint32_t *)StubRoutines::_crc32c_table_addr + 5);
8433 } else {
8434 const_or_pre_comp_const_index[0] = 1;
8435 const_or_pre_comp_const_index[1] = 0;
8436
8437 const_or_pre_comp_const_index[2] = 3;
8438 const_or_pre_comp_const_index[3] = 2;
8439
8440 const_or_pre_comp_const_index[4] = 5;
8441 const_or_pre_comp_const_index[5] = 4;
8442 }
8443 crc32c_proc_chunk(CRC32C_HIGH, const_or_pre_comp_const_index[0], const_or_pre_comp_const_index[1], is_pclmulqdq_supported,
8444 in2, in1, in_out,
8445 tmp1, tmp2, tmp3,
8446 w_xtmp1, w_xtmp2, w_xtmp3,
8447 tmp4, tmp5,
8448 tmp6);
8449 crc32c_proc_chunk(CRC32C_MIDDLE, const_or_pre_comp_const_index[2], const_or_pre_comp_const_index[3], is_pclmulqdq_supported,
8450 in2, in1, in_out,
8451 tmp1, tmp2, tmp3,
8452 w_xtmp1, w_xtmp2, w_xtmp3,
8453 tmp4, tmp5,
8454 tmp6);
8455 crc32c_proc_chunk(CRC32C_LOW, const_or_pre_comp_const_index[4], const_or_pre_comp_const_index[5], is_pclmulqdq_supported,
8456 in2, in1, in_out,
8457 tmp1, tmp2, tmp3,
8458 w_xtmp1, w_xtmp2, w_xtmp3,
8459 tmp4, tmp5,
8460 tmp6);
8461 movl(tmp1, in2);
8462 andl(tmp1, 0x00000007);
8463 negl(tmp1);
8464 addl(tmp1, in2);
8465 addl(tmp1, in1);
8466
8467 BIND(L_wordByWord);
8468 cmpl(in1, tmp1);
8469 jcc(Assembler::greaterEqual, L_byteByByteProlog);
8470 crc32(in_out, Address(in1,0), 4);
8471 addl(in1, 4);
8472 jmp(L_wordByWord);
8473
8474 BIND(L_byteByByteProlog);
8475 andl(in2, 0x00000007);
8476 movl(tmp2, 1);
8477
8478 BIND(L_byteByByte);
8479 cmpl(tmp2, in2);
8480 jccb(Assembler::greater, L_exit);
8481 movb(tmp1, Address(in1, 0));
8482 crc32(in_out, tmp1, 1);
8483 incl(in1);
8484 incl(tmp2);
8485 jmp(L_byteByByte);
8486
8487 BIND(L_exit);
8488 }
8489 #endif // LP64
8490 #undef BIND
8491 #undef BLOCK_COMMENT
8492
8493 // Compress char[] array to byte[].
8494 // ..\jdk\src\java.base\share\classes\java\lang\StringUTF16.java
8495 // @IntrinsicCandidate
8496 // private static int compress(char[] src, int srcOff, byte[] dst, int dstOff, int len) {
8497 // for (int i = 0; i < len; i++) {
8498 // int c = src[srcOff++];
8499 // if (c >>> 8 != 0) {
8500 // return 0;
8501 // }
8502 // dst[dstOff++] = (byte)c;
8503 // }
8504 // return len;
8505 // }
8506 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
8507 XMMRegister tmp1Reg, XMMRegister tmp2Reg,
8508 XMMRegister tmp3Reg, XMMRegister tmp4Reg,
8509 Register tmp5, Register result, KRegister mask1, KRegister mask2) {
8510 Label copy_chars_loop, return_length, return_zero, done;
8511
8512 // rsi: src
8513 // rdi: dst
8514 // rdx: len
8515 // rcx: tmp5
8516 // rax: result
8517
8518 // rsi holds start addr of source char[] to be compressed
8519 // rdi holds start addr of destination byte[]
8520 // rdx holds length
8521
8522 assert(len != result, "");
8523
8524 // save length for return
8525 push(len);
8526
8527 if ((AVX3Threshold == 0) && (UseAVX > 2) && // AVX512
8528 VM_Version::supports_avx512vlbw() &&
8529 VM_Version::supports_bmi2()) {
8530
8531 Label copy_32_loop, copy_loop_tail, below_threshold;
8532
8533 // alignment
8534 Label post_alignment;
8535
8536 // if length of the string is less than 16, handle it in an old fashioned way
8537 testl(len, -32);
8538 jcc(Assembler::zero, below_threshold);
8539
8540 // First check whether a character is compressible ( <= 0xFF).
8541 // Create mask to test for Unicode chars inside zmm vector
8542 movl(result, 0x00FF);
8543 evpbroadcastw(tmp2Reg, result, Assembler::AVX_512bit);
8544
8545 testl(len, -64);
8546 jcc(Assembler::zero, post_alignment);
8547
8548 movl(tmp5, dst);
8549 andl(tmp5, (32 - 1));
8550 negl(tmp5);
8551 andl(tmp5, (32 - 1));
8552
8553 // bail out when there is nothing to be done
8554 testl(tmp5, 0xFFFFFFFF);
8555 jcc(Assembler::zero, post_alignment);
8556
8557 // ~(~0 << len), where len is the # of remaining elements to process
8558 movl(result, 0xFFFFFFFF);
8559 shlxl(result, result, tmp5);
8560 notl(result);
8561 kmovdl(mask2, result);
8562
8563 evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
8564 evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
8565 ktestd(mask1, mask2);
8566 jcc(Assembler::carryClear, return_zero);
8567
8568 evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
8569
8570 addptr(src, tmp5);
8571 addptr(src, tmp5);
8572 addptr(dst, tmp5);
8573 subl(len, tmp5);
8574
8575 bind(post_alignment);
8576 // end of alignment
8577
8578 movl(tmp5, len);
8579 andl(tmp5, (32 - 1)); // tail count (in chars)
8580 andl(len, ~(32 - 1)); // vector count (in chars)
8581 jcc(Assembler::zero, copy_loop_tail);
8582
8583 lea(src, Address(src, len, Address::times_2));
8584 lea(dst, Address(dst, len, Address::times_1));
8585 negptr(len);
8586
8587 bind(copy_32_loop);
8588 evmovdquw(tmp1Reg, Address(src, len, Address::times_2), Assembler::AVX_512bit);
8589 evpcmpuw(mask1, tmp1Reg, tmp2Reg, Assembler::le, Assembler::AVX_512bit);
8590 kortestdl(mask1, mask1);
8591 jcc(Assembler::carryClear, return_zero);
8592
8593 // All elements in current processed chunk are valid candidates for
8594 // compression. Write a truncated byte elements to the memory.
8595 evpmovwb(Address(dst, len, Address::times_1), tmp1Reg, Assembler::AVX_512bit);
8596 addptr(len, 32);
8597 jcc(Assembler::notZero, copy_32_loop);
8598
8599 bind(copy_loop_tail);
8600 // bail out when there is nothing to be done
8601 testl(tmp5, 0xFFFFFFFF);
8602 jcc(Assembler::zero, return_length);
8603
8604 movl(len, tmp5);
8605
8606 // ~(~0 << len), where len is the # of remaining elements to process
8607 movl(result, 0xFFFFFFFF);
8608 shlxl(result, result, len);
8609 notl(result);
8610
8611 kmovdl(mask2, result);
8612
8613 evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
8614 evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
8615 ktestd(mask1, mask2);
8616 jcc(Assembler::carryClear, return_zero);
8617
8618 evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
8619 jmp(return_length);
8620
8621 bind(below_threshold);
8622 }
8623
8624 if (UseSSE42Intrinsics) {
8625 Label copy_32_loop, copy_16, copy_tail;
8626
8627 movl(result, len);
8628
8629 movl(tmp5, 0xff00ff00); // create mask to test for Unicode chars in vectors
8630
8631 // vectored compression
8632 andl(len, 0xfffffff0); // vector count (in chars)
8633 andl(result, 0x0000000f); // tail count (in chars)
8634 testl(len, len);
8635 jcc(Assembler::zero, copy_16);
8636
8637 // compress 16 chars per iter
8638 movdl(tmp1Reg, tmp5);
8639 pshufd(tmp1Reg, tmp1Reg, 0); // store Unicode mask in tmp1Reg
8640 pxor(tmp4Reg, tmp4Reg);
8641
8642 lea(src, Address(src, len, Address::times_2));
8643 lea(dst, Address(dst, len, Address::times_1));
8644 negptr(len);
8645
8646 bind(copy_32_loop);
8647 movdqu(tmp2Reg, Address(src, len, Address::times_2)); // load 1st 8 characters
8648 por(tmp4Reg, tmp2Reg);
8649 movdqu(tmp3Reg, Address(src, len, Address::times_2, 16)); // load next 8 characters
8650 por(tmp4Reg, tmp3Reg);
8651 ptest(tmp4Reg, tmp1Reg); // check for Unicode chars in next vector
8652 jcc(Assembler::notZero, return_zero);
8653 packuswb(tmp2Reg, tmp3Reg); // only ASCII chars; compress each to 1 byte
8654 movdqu(Address(dst, len, Address::times_1), tmp2Reg);
8655 addptr(len, 16);
8656 jcc(Assembler::notZero, copy_32_loop);
8657
8658 // compress next vector of 8 chars (if any)
8659 bind(copy_16);
8660 movl(len, result);
8661 andl(len, 0xfffffff8); // vector count (in chars)
8662 andl(result, 0x00000007); // tail count (in chars)
8663 testl(len, len);
8664 jccb(Assembler::zero, copy_tail);
8665
8666 movdl(tmp1Reg, tmp5);
8667 pshufd(tmp1Reg, tmp1Reg, 0); // store Unicode mask in tmp1Reg
8668 pxor(tmp3Reg, tmp3Reg);
8669
8670 movdqu(tmp2Reg, Address(src, 0));
8671 ptest(tmp2Reg, tmp1Reg); // check for Unicode chars in vector
8672 jccb(Assembler::notZero, return_zero);
8673 packuswb(tmp2Reg, tmp3Reg); // only LATIN1 chars; compress each to 1 byte
8674 movq(Address(dst, 0), tmp2Reg);
8675 addptr(src, 16);
8676 addptr(dst, 8);
8677
8678 bind(copy_tail);
8679 movl(len, result);
8680 }
8681 // compress 1 char per iter
8682 testl(len, len);
8683 jccb(Assembler::zero, return_length);
8684 lea(src, Address(src, len, Address::times_2));
8685 lea(dst, Address(dst, len, Address::times_1));
8686 negptr(len);
8687
8688 bind(copy_chars_loop);
8689 load_unsigned_short(result, Address(src, len, Address::times_2));
8690 testl(result, 0xff00); // check if Unicode char
8691 jccb(Assembler::notZero, return_zero);
8692 movb(Address(dst, len, Address::times_1), result); // ASCII char; compress to 1 byte
8693 increment(len);
8694 jcc(Assembler::notZero, copy_chars_loop);
8695
8696 // if compression succeeded, return length
8697 bind(return_length);
8698 pop(result);
8699 jmpb(done);
8700
8701 // if compression failed, return 0
8702 bind(return_zero);
8703 xorl(result, result);
8704 addptr(rsp, wordSize);
8705
8706 bind(done);
8707 }
8708
8709 // Inflate byte[] array to char[].
8710 // ..\jdk\src\java.base\share\classes\java\lang\StringLatin1.java
8711 // @IntrinsicCandidate
8712 // private static void inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len) {
8713 // for (int i = 0; i < len; i++) {
8714 // dst[dstOff++] = (char)(src[srcOff++] & 0xff);
8715 // }
8716 // }
8717 void MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
8718 XMMRegister tmp1, Register tmp2, KRegister mask) {
8719 Label copy_chars_loop, done, below_threshold, avx3_threshold;
8720 // rsi: src
8721 // rdi: dst
8722 // rdx: len
8723 // rcx: tmp2
8724
8725 // rsi holds start addr of source byte[] to be inflated
8726 // rdi holds start addr of destination char[]
8727 // rdx holds length
8728 assert_different_registers(src, dst, len, tmp2);
8729 movl(tmp2, len);
8730 if ((UseAVX > 2) && // AVX512
8731 VM_Version::supports_avx512vlbw() &&
8732 VM_Version::supports_bmi2()) {
8733
8734 Label copy_32_loop, copy_tail;
8735 Register tmp3_aliased = len;
8736
8737 // if length of the string is less than 16, handle it in an old fashioned way
8738 testl(len, -16);
8739 jcc(Assembler::zero, below_threshold);
8740
8741 testl(len, -1 * AVX3Threshold);
8742 jcc(Assembler::zero, avx3_threshold);
8743
8744 // In order to use only one arithmetic operation for the main loop we use
8745 // this pre-calculation
8746 andl(tmp2, (32 - 1)); // tail count (in chars), 32 element wide loop
8747 andl(len, -32); // vector count
8748 jccb(Assembler::zero, copy_tail);
8749
8750 lea(src, Address(src, len, Address::times_1));
8751 lea(dst, Address(dst, len, Address::times_2));
8752 negptr(len);
8753
8754
8755 // inflate 32 chars per iter
8756 bind(copy_32_loop);
8757 vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_512bit);
8758 evmovdquw(Address(dst, len, Address::times_2), tmp1, Assembler::AVX_512bit);
8759 addptr(len, 32);
8760 jcc(Assembler::notZero, copy_32_loop);
8761
8762 bind(copy_tail);
8763 // bail out when there is nothing to be done
8764 testl(tmp2, -1); // we don't destroy the contents of tmp2 here
8765 jcc(Assembler::zero, done);
8766
8767 // ~(~0 << length), where length is the # of remaining elements to process
8768 movl(tmp3_aliased, -1);
8769 shlxl(tmp3_aliased, tmp3_aliased, tmp2);
8770 notl(tmp3_aliased);
8771 kmovdl(mask, tmp3_aliased);
8772 evpmovzxbw(tmp1, mask, Address(src, 0), Assembler::AVX_512bit);
8773 evmovdquw(Address(dst, 0), mask, tmp1, /*merge*/ true, Assembler::AVX_512bit);
8774
8775 jmp(done);
8776 bind(avx3_threshold);
8777 }
8778 if (UseSSE42Intrinsics) {
8779 Label copy_16_loop, copy_8_loop, copy_bytes, copy_new_tail, copy_tail;
8780
8781 if (UseAVX > 1) {
8782 andl(tmp2, (16 - 1));
8783 andl(len, -16);
8784 jccb(Assembler::zero, copy_new_tail);
8785 } else {
8786 andl(tmp2, 0x00000007); // tail count (in chars)
8787 andl(len, 0xfffffff8); // vector count (in chars)
8788 jccb(Assembler::zero, copy_tail);
8789 }
8790
8791 // vectored inflation
8792 lea(src, Address(src, len, Address::times_1));
8793 lea(dst, Address(dst, len, Address::times_2));
8794 negptr(len);
8795
8796 if (UseAVX > 1) {
8797 bind(copy_16_loop);
8798 vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_256bit);
8799 vmovdqu(Address(dst, len, Address::times_2), tmp1);
8800 addptr(len, 16);
8801 jcc(Assembler::notZero, copy_16_loop);
8802
8803 bind(below_threshold);
8804 bind(copy_new_tail);
8805 movl(len, tmp2);
8806 andl(tmp2, 0x00000007);
8807 andl(len, 0xFFFFFFF8);
8808 jccb(Assembler::zero, copy_tail);
8809
8810 pmovzxbw(tmp1, Address(src, 0));
8811 movdqu(Address(dst, 0), tmp1);
8812 addptr(src, 8);
8813 addptr(dst, 2 * 8);
8814
8815 jmp(copy_tail, true);
8816 }
8817
8818 // inflate 8 chars per iter
8819 bind(copy_8_loop);
8820 pmovzxbw(tmp1, Address(src, len, Address::times_1)); // unpack to 8 words
8821 movdqu(Address(dst, len, Address::times_2), tmp1);
8822 addptr(len, 8);
8823 jcc(Assembler::notZero, copy_8_loop);
8824
8825 bind(copy_tail);
8826 movl(len, tmp2);
8827
8828 cmpl(len, 4);
8829 jccb(Assembler::less, copy_bytes);
8830
8831 movdl(tmp1, Address(src, 0)); // load 4 byte chars
8832 pmovzxbw(tmp1, tmp1);
8833 movq(Address(dst, 0), tmp1);
8834 subptr(len, 4);
8835 addptr(src, 4);
8836 addptr(dst, 8);
8837
8838 bind(copy_bytes);
8839 } else {
8840 bind(below_threshold);
8841 }
8842
8843 testl(len, len);
8844 jccb(Assembler::zero, done);
8845 lea(src, Address(src, len, Address::times_1));
8846 lea(dst, Address(dst, len, Address::times_2));
8847 negptr(len);
8848
8849 // inflate 1 char per iter
8850 bind(copy_chars_loop);
8851 load_unsigned_byte(tmp2, Address(src, len, Address::times_1)); // load byte char
8852 movw(Address(dst, len, Address::times_2), tmp2); // inflate byte char to word
8853 increment(len);
8854 jcc(Assembler::notZero, copy_chars_loop);
8855
8856 bind(done);
8857 }
8858
8859
8860 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address src, bool merge, int vector_len) {
8861 switch(type) {
8862 case T_BYTE:
8863 case T_BOOLEAN:
8864 evmovdqub(dst, kmask, src, merge, vector_len);
8865 break;
8866 case T_CHAR:
8867 case T_SHORT:
8868 evmovdquw(dst, kmask, src, merge, vector_len);
8869 break;
8870 case T_INT:
8871 case T_FLOAT:
8872 evmovdqul(dst, kmask, src, merge, vector_len);
8873 break;
8874 case T_LONG:
8875 case T_DOUBLE:
8876 evmovdquq(dst, kmask, src, merge, vector_len);
8877 break;
8878 default:
8879 fatal("Unexpected type argument %s", type2name(type));
8880 break;
8881 }
8882 }
8883
8884 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister src, bool merge, int vector_len) {
8885 switch(type) {
8886 case T_BYTE:
8887 case T_BOOLEAN:
8888 evmovdqub(dst, kmask, src, merge, vector_len);
8889 break;
8890 case T_CHAR:
8891 case T_SHORT:
8892 evmovdquw(dst, kmask, src, merge, vector_len);
8893 break;
8894 case T_INT:
8895 case T_FLOAT:
8896 evmovdqul(dst, kmask, src, merge, vector_len);
8897 break;
8898 case T_LONG:
8899 case T_DOUBLE:
8900 evmovdquq(dst, kmask, src, merge, vector_len);
8901 break;
8902 default:
8903 fatal("Unexpected type argument %s", type2name(type));
8904 break;
8905 }
8906 }
8907
8908 void MacroAssembler::knot(uint masklen, KRegister dst, KRegister src, KRegister ktmp, Register rtmp) {
8909 switch(masklen) {
8910 case 2:
8911 knotbl(dst, src);
8912 movl(rtmp, 3);
8913 kmovbl(ktmp, rtmp);
8914 kandbl(dst, ktmp, dst);
8915 break;
8916 case 4:
8917 knotbl(dst, src);
8918 movl(rtmp, 15);
8919 kmovbl(ktmp, rtmp);
8920 kandbl(dst, ktmp, dst);
8921 break;
8922 case 8:
8923 knotbl(dst, src);
8924 break;
8925 case 16:
8926 knotwl(dst, src);
8927 break;
8928 case 32:
8929 knotdl(dst, src);
8930 break;
8931 case 64:
8932 knotql(dst, src);
8933 break;
8934 default:
8935 fatal("Unexpected vector length %d", masklen);
8936 break;
8937 }
8938 }
8939
8940 void MacroAssembler::kand(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
8941 switch(type) {
8942 case T_BOOLEAN:
8943 case T_BYTE:
8944 kandbl(dst, src1, src2);
8945 break;
8946 case T_CHAR:
8947 case T_SHORT:
8948 kandwl(dst, src1, src2);
8949 break;
8950 case T_INT:
8951 case T_FLOAT:
8952 kanddl(dst, src1, src2);
8953 break;
8954 case T_LONG:
8955 case T_DOUBLE:
8956 kandql(dst, src1, src2);
8957 break;
8958 default:
8959 fatal("Unexpected type argument %s", type2name(type));
8960 break;
8961 }
8962 }
8963
8964 void MacroAssembler::kor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
8965 switch(type) {
8966 case T_BOOLEAN:
8967 case T_BYTE:
8968 korbl(dst, src1, src2);
8969 break;
8970 case T_CHAR:
8971 case T_SHORT:
8972 korwl(dst, src1, src2);
8973 break;
8974 case T_INT:
8975 case T_FLOAT:
8976 kordl(dst, src1, src2);
8977 break;
8978 case T_LONG:
8979 case T_DOUBLE:
8980 korql(dst, src1, src2);
8981 break;
8982 default:
8983 fatal("Unexpected type argument %s", type2name(type));
8984 break;
8985 }
8986 }
8987
8988 void MacroAssembler::kxor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
8989 switch(type) {
8990 case T_BOOLEAN:
8991 case T_BYTE:
8992 kxorbl(dst, src1, src2);
8993 break;
8994 case T_CHAR:
8995 case T_SHORT:
8996 kxorwl(dst, src1, src2);
8997 break;
8998 case T_INT:
8999 case T_FLOAT:
9000 kxordl(dst, src1, src2);
9001 break;
9002 case T_LONG:
9003 case T_DOUBLE:
9004 kxorql(dst, src1, src2);
9005 break;
9006 default:
9007 fatal("Unexpected type argument %s", type2name(type));
9008 break;
9009 }
9010 }
9011
9012 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
9013 switch(type) {
9014 case T_BOOLEAN:
9015 case T_BYTE:
9016 evpermb(dst, mask, nds, src, merge, vector_len); break;
9017 case T_CHAR:
9018 case T_SHORT:
9019 evpermw(dst, mask, nds, src, merge, vector_len); break;
9020 case T_INT:
9021 case T_FLOAT:
9022 evpermd(dst, mask, nds, src, merge, vector_len); break;
9023 case T_LONG:
9024 case T_DOUBLE:
9025 evpermq(dst, mask, nds, src, merge, vector_len); break;
9026 default:
9027 fatal("Unexpected type argument %s", type2name(type)); break;
9028 }
9029 }
9030
9031 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
9032 switch(type) {
9033 case T_BOOLEAN:
9034 case T_BYTE:
9035 evpermb(dst, mask, nds, src, merge, vector_len); break;
9036 case T_CHAR:
9037 case T_SHORT:
9038 evpermw(dst, mask, nds, src, merge, vector_len); break;
9039 case T_INT:
9040 case T_FLOAT:
9041 evpermd(dst, mask, nds, src, merge, vector_len); break;
9042 case T_LONG:
9043 case T_DOUBLE:
9044 evpermq(dst, mask, nds, src, merge, vector_len); break;
9045 default:
9046 fatal("Unexpected type argument %s", type2name(type)); break;
9047 }
9048 }
9049
9050 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
9051 switch(type) {
9052 case T_BYTE:
9053 evpminsb(dst, mask, nds, src, merge, vector_len); break;
9054 case T_SHORT:
9055 evpminsw(dst, mask, nds, src, merge, vector_len); break;
9056 case T_INT:
9057 evpminsd(dst, mask, nds, src, merge, vector_len); break;
9058 case T_LONG:
9059 evpminsq(dst, mask, nds, src, merge, vector_len); break;
9060 default:
9061 fatal("Unexpected type argument %s", type2name(type)); break;
9062 }
9063 }
9064
9065 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
9066 switch(type) {
9067 case T_BYTE:
9068 evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
9069 case T_SHORT:
9070 evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
9071 case T_INT:
9072 evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
9073 case T_LONG:
9074 evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
9075 default:
9076 fatal("Unexpected type argument %s", type2name(type)); break;
9077 }
9078 }
9079
9080 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
9081 switch(type) {
9082 case T_BYTE:
9083 evpminsb(dst, mask, nds, src, merge, vector_len); break;
9084 case T_SHORT:
9085 evpminsw(dst, mask, nds, src, merge, vector_len); break;
9086 case T_INT:
9087 evpminsd(dst, mask, nds, src, merge, vector_len); break;
9088 case T_LONG:
9089 evpminsq(dst, mask, nds, src, merge, vector_len); break;
9090 default:
9091 fatal("Unexpected type argument %s", type2name(type)); break;
9092 }
9093 }
9094
9095 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
9096 switch(type) {
9097 case T_BYTE:
9098 evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
9099 case T_SHORT:
9100 evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
9101 case T_INT:
9102 evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
9103 case T_LONG:
9104 evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
9105 default:
9106 fatal("Unexpected type argument %s", type2name(type)); break;
9107 }
9108 }
9109
9110 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
9111 switch(type) {
9112 case T_INT:
9113 evpxord(dst, mask, nds, src, merge, vector_len); break;
9114 case T_LONG:
9115 evpxorq(dst, mask, nds, src, merge, vector_len); break;
9116 default:
9117 fatal("Unexpected type argument %s", type2name(type)); break;
9118 }
9119 }
9120
9121 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
9122 switch(type) {
9123 case T_INT:
9124 evpxord(dst, mask, nds, src, merge, vector_len); break;
9125 case T_LONG:
9126 evpxorq(dst, mask, nds, src, merge, vector_len); break;
9127 default:
9128 fatal("Unexpected type argument %s", type2name(type)); break;
9129 }
9130 }
9131
9132 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
9133 switch(type) {
9134 case T_INT:
9135 Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
9136 case T_LONG:
9137 evporq(dst, mask, nds, src, merge, vector_len); break;
9138 default:
9139 fatal("Unexpected type argument %s", type2name(type)); break;
9140 }
9141 }
9142
9143 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
9144 switch(type) {
9145 case T_INT:
9146 Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
9147 case T_LONG:
9148 evporq(dst, mask, nds, src, merge, vector_len); break;
9149 default:
9150 fatal("Unexpected type argument %s", type2name(type)); break;
9151 }
9152 }
9153
9154 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
9155 switch(type) {
9156 case T_INT:
9157 evpandd(dst, mask, nds, src, merge, vector_len); break;
9158 case T_LONG:
9159 evpandq(dst, mask, nds, src, merge, vector_len); break;
9160 default:
9161 fatal("Unexpected type argument %s", type2name(type)); break;
9162 }
9163 }
9164
9165 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
9166 switch(type) {
9167 case T_INT:
9168 evpandd(dst, mask, nds, src, merge, vector_len); break;
9169 case T_LONG:
9170 evpandq(dst, mask, nds, src, merge, vector_len); break;
9171 default:
9172 fatal("Unexpected type argument %s", type2name(type)); break;
9173 }
9174 }
9175
9176 void MacroAssembler::kortest(uint masklen, KRegister src1, KRegister src2) {
9177 switch(masklen) {
9178 case 8:
9179 kortestbl(src1, src2);
9180 break;
9181 case 16:
9182 kortestwl(src1, src2);
9183 break;
9184 case 32:
9185 kortestdl(src1, src2);
9186 break;
9187 case 64:
9188 kortestql(src1, src2);
9189 break;
9190 default:
9191 fatal("Unexpected mask length %d", masklen);
9192 break;
9193 }
9194 }
9195
9196
9197 void MacroAssembler::ktest(uint masklen, KRegister src1, KRegister src2) {
9198 switch(masklen) {
9199 case 8:
9200 ktestbl(src1, src2);
9201 break;
9202 case 16:
9203 ktestwl(src1, src2);
9204 break;
9205 case 32:
9206 ktestdl(src1, src2);
9207 break;
9208 case 64:
9209 ktestql(src1, src2);
9210 break;
9211 default:
9212 fatal("Unexpected mask length %d", masklen);
9213 break;
9214 }
9215 }
9216
9217 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
9218 switch(type) {
9219 case T_INT:
9220 evprold(dst, mask, src, shift, merge, vlen_enc); break;
9221 case T_LONG:
9222 evprolq(dst, mask, src, shift, merge, vlen_enc); break;
9223 default:
9224 fatal("Unexpected type argument %s", type2name(type)); break;
9225 break;
9226 }
9227 }
9228
9229 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
9230 switch(type) {
9231 case T_INT:
9232 evprord(dst, mask, src, shift, merge, vlen_enc); break;
9233 case T_LONG:
9234 evprorq(dst, mask, src, shift, merge, vlen_enc); break;
9235 default:
9236 fatal("Unexpected type argument %s", type2name(type)); break;
9237 }
9238 }
9239
9240 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
9241 switch(type) {
9242 case T_INT:
9243 evprolvd(dst, mask, src1, src2, merge, vlen_enc); break;
9244 case T_LONG:
9245 evprolvq(dst, mask, src1, src2, merge, vlen_enc); break;
9246 default:
9247 fatal("Unexpected type argument %s", type2name(type)); break;
9248 }
9249 }
9250
9251 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
9252 switch(type) {
9253 case T_INT:
9254 evprorvd(dst, mask, src1, src2, merge, vlen_enc); break;
9255 case T_LONG:
9256 evprorvq(dst, mask, src1, src2, merge, vlen_enc); break;
9257 default:
9258 fatal("Unexpected type argument %s", type2name(type)); break;
9259 }
9260 }
9261
9262 void MacroAssembler::evpandq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
9263 assert(rscratch != noreg || always_reachable(src), "missing");
9264
9265 if (reachable(src)) {
9266 evpandq(dst, nds, as_Address(src), vector_len);
9267 } else {
9268 lea(rscratch, src);
9269 evpandq(dst, nds, Address(rscratch, 0), vector_len);
9270 }
9271 }
9272
9273 void MacroAssembler::evpaddq(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
9274 assert(rscratch != noreg || always_reachable(src), "missing");
9275
9276 if (reachable(src)) {
9277 Assembler::evpaddq(dst, mask, nds, as_Address(src), merge, vector_len);
9278 } else {
9279 lea(rscratch, src);
9280 Assembler::evpaddq(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
9281 }
9282 }
9283
9284 void MacroAssembler::evporq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
9285 assert(rscratch != noreg || always_reachable(src), "missing");
9286
9287 if (reachable(src)) {
9288 evporq(dst, nds, as_Address(src), vector_len);
9289 } else {
9290 lea(rscratch, src);
9291 evporq(dst, nds, Address(rscratch, 0), vector_len);
9292 }
9293 }
9294
9295 void MacroAssembler::vpternlogq(XMMRegister dst, int imm8, XMMRegister src2, AddressLiteral src3, int vector_len, Register rscratch) {
9296 assert(rscratch != noreg || always_reachable(src3), "missing");
9297
9298 if (reachable(src3)) {
9299 vpternlogq(dst, imm8, src2, as_Address(src3), vector_len);
9300 } else {
9301 lea(rscratch, src3);
9302 vpternlogq(dst, imm8, src2, Address(rscratch, 0), vector_len);
9303 }
9304 }
9305
9306 #if COMPILER2_OR_JVMCI
9307
9308 void MacroAssembler::fill_masked(BasicType bt, Address dst, XMMRegister xmm, KRegister mask,
9309 Register length, Register temp, int vec_enc) {
9310 // Computing mask for predicated vector store.
9311 movptr(temp, -1);
9312 bzhiq(temp, temp, length);
9313 kmov(mask, temp);
9314 evmovdqu(bt, mask, dst, xmm, true, vec_enc);
9315 }
9316
9317 // Set memory operation for length "less than" 64 bytes.
9318 void MacroAssembler::fill64_masked(uint shift, Register dst, int disp,
9319 XMMRegister xmm, KRegister mask, Register length,
9320 Register temp, bool use64byteVector) {
9321 assert(MaxVectorSize >= 32, "vector length should be >= 32");
9322 const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
9323 if (!use64byteVector) {
9324 fill32(dst, disp, xmm);
9325 subptr(length, 32 >> shift);
9326 fill32_masked(shift, dst, disp + 32, xmm, mask, length, temp);
9327 } else {
9328 assert(MaxVectorSize == 64, "vector length != 64");
9329 fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_512bit);
9330 }
9331 }
9332
9333
9334 void MacroAssembler::fill32_masked(uint shift, Register dst, int disp,
9335 XMMRegister xmm, KRegister mask, Register length,
9336 Register temp) {
9337 assert(MaxVectorSize >= 32, "vector length should be >= 32");
9338 const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
9339 fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_256bit);
9340 }
9341
9342
9343 void MacroAssembler::fill32(Address dst, XMMRegister xmm) {
9344 assert(MaxVectorSize >= 32, "vector length should be >= 32");
9345 vmovdqu(dst, xmm);
9346 }
9347
9348 void MacroAssembler::fill32(Register dst, int disp, XMMRegister xmm) {
9349 fill32(Address(dst, disp), xmm);
9350 }
9351
9352 void MacroAssembler::fill64(Address dst, XMMRegister xmm, bool use64byteVector) {
9353 assert(MaxVectorSize >= 32, "vector length should be >= 32");
9354 if (!use64byteVector) {
9355 fill32(dst, xmm);
9356 fill32(dst.plus_disp(32), xmm);
9357 } else {
9358 evmovdquq(dst, xmm, Assembler::AVX_512bit);
9359 }
9360 }
9361
9362 void MacroAssembler::fill64(Register dst, int disp, XMMRegister xmm, bool use64byteVector) {
9363 fill64(Address(dst, disp), xmm, use64byteVector);
9364 }
9365
9366 #ifdef _LP64
9367 void MacroAssembler::generate_fill_avx3(BasicType type, Register to, Register value,
9368 Register count, Register rtmp, XMMRegister xtmp) {
9369 Label L_exit;
9370 Label L_fill_start;
9371 Label L_fill_64_bytes;
9372 Label L_fill_96_bytes;
9373 Label L_fill_128_bytes;
9374 Label L_fill_128_bytes_loop;
9375 Label L_fill_128_loop_header;
9376 Label L_fill_128_bytes_loop_header;
9377 Label L_fill_128_bytes_loop_pre_header;
9378 Label L_fill_zmm_sequence;
9379
9380 int shift = -1;
9381 int avx3threshold = VM_Version::avx3_threshold();
9382 switch(type) {
9383 case T_BYTE: shift = 0;
9384 break;
9385 case T_SHORT: shift = 1;
9386 break;
9387 case T_INT: shift = 2;
9388 break;
9389 /* Uncomment when LONG fill stubs are supported.
9390 case T_LONG: shift = 3;
9391 break;
9392 */
9393 default:
9394 fatal("Unhandled type: %s\n", type2name(type));
9395 }
9396
9397 if ((avx3threshold != 0) || (MaxVectorSize == 32)) {
9398
9399 if (MaxVectorSize == 64) {
9400 cmpq(count, avx3threshold >> shift);
9401 jcc(Assembler::greater, L_fill_zmm_sequence);
9402 }
9403
9404 evpbroadcast(type, xtmp, value, Assembler::AVX_256bit);
9405
9406 bind(L_fill_start);
9407
9408 cmpq(count, 32 >> shift);
9409 jccb(Assembler::greater, L_fill_64_bytes);
9410 fill32_masked(shift, to, 0, xtmp, k2, count, rtmp);
9411 jmp(L_exit);
9412
9413 bind(L_fill_64_bytes);
9414 cmpq(count, 64 >> shift);
9415 jccb(Assembler::greater, L_fill_96_bytes);
9416 fill64_masked(shift, to, 0, xtmp, k2, count, rtmp);
9417 jmp(L_exit);
9418
9419 bind(L_fill_96_bytes);
9420 cmpq(count, 96 >> shift);
9421 jccb(Assembler::greater, L_fill_128_bytes);
9422 fill64(to, 0, xtmp);
9423 subq(count, 64 >> shift);
9424 fill32_masked(shift, to, 64, xtmp, k2, count, rtmp);
9425 jmp(L_exit);
9426
9427 bind(L_fill_128_bytes);
9428 cmpq(count, 128 >> shift);
9429 jccb(Assembler::greater, L_fill_128_bytes_loop_pre_header);
9430 fill64(to, 0, xtmp);
9431 fill32(to, 64, xtmp);
9432 subq(count, 96 >> shift);
9433 fill32_masked(shift, to, 96, xtmp, k2, count, rtmp);
9434 jmp(L_exit);
9435
9436 bind(L_fill_128_bytes_loop_pre_header);
9437 {
9438 mov(rtmp, to);
9439 andq(rtmp, 31);
9440 jccb(Assembler::zero, L_fill_128_bytes_loop_header);
9441 negq(rtmp);
9442 addq(rtmp, 32);
9443 mov64(r8, -1L);
9444 bzhiq(r8, r8, rtmp);
9445 kmovql(k2, r8);
9446 evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_256bit);
9447 addq(to, rtmp);
9448 shrq(rtmp, shift);
9449 subq(count, rtmp);
9450 }
9451
9452 cmpq(count, 128 >> shift);
9453 jcc(Assembler::less, L_fill_start);
9454
9455 bind(L_fill_128_bytes_loop_header);
9456 subq(count, 128 >> shift);
9457
9458 align32();
9459 bind(L_fill_128_bytes_loop);
9460 fill64(to, 0, xtmp);
9461 fill64(to, 64, xtmp);
9462 addq(to, 128);
9463 subq(count, 128 >> shift);
9464 jccb(Assembler::greaterEqual, L_fill_128_bytes_loop);
9465
9466 addq(count, 128 >> shift);
9467 jcc(Assembler::zero, L_exit);
9468 jmp(L_fill_start);
9469 }
9470
9471 if (MaxVectorSize == 64) {
9472 // Sequence using 64 byte ZMM register.
9473 Label L_fill_128_bytes_zmm;
9474 Label L_fill_192_bytes_zmm;
9475 Label L_fill_192_bytes_loop_zmm;
9476 Label L_fill_192_bytes_loop_header_zmm;
9477 Label L_fill_192_bytes_loop_pre_header_zmm;
9478 Label L_fill_start_zmm_sequence;
9479
9480 bind(L_fill_zmm_sequence);
9481 evpbroadcast(type, xtmp, value, Assembler::AVX_512bit);
9482
9483 bind(L_fill_start_zmm_sequence);
9484 cmpq(count, 64 >> shift);
9485 jccb(Assembler::greater, L_fill_128_bytes_zmm);
9486 fill64_masked(shift, to, 0, xtmp, k2, count, rtmp, true);
9487 jmp(L_exit);
9488
9489 bind(L_fill_128_bytes_zmm);
9490 cmpq(count, 128 >> shift);
9491 jccb(Assembler::greater, L_fill_192_bytes_zmm);
9492 fill64(to, 0, xtmp, true);
9493 subq(count, 64 >> shift);
9494 fill64_masked(shift, to, 64, xtmp, k2, count, rtmp, true);
9495 jmp(L_exit);
9496
9497 bind(L_fill_192_bytes_zmm);
9498 cmpq(count, 192 >> shift);
9499 jccb(Assembler::greater, L_fill_192_bytes_loop_pre_header_zmm);
9500 fill64(to, 0, xtmp, true);
9501 fill64(to, 64, xtmp, true);
9502 subq(count, 128 >> shift);
9503 fill64_masked(shift, to, 128, xtmp, k2, count, rtmp, true);
9504 jmp(L_exit);
9505
9506 bind(L_fill_192_bytes_loop_pre_header_zmm);
9507 {
9508 movq(rtmp, to);
9509 andq(rtmp, 63);
9510 jccb(Assembler::zero, L_fill_192_bytes_loop_header_zmm);
9511 negq(rtmp);
9512 addq(rtmp, 64);
9513 mov64(r8, -1L);
9514 bzhiq(r8, r8, rtmp);
9515 kmovql(k2, r8);
9516 evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_512bit);
9517 addq(to, rtmp);
9518 shrq(rtmp, shift);
9519 subq(count, rtmp);
9520 }
9521
9522 cmpq(count, 192 >> shift);
9523 jcc(Assembler::less, L_fill_start_zmm_sequence);
9524
9525 bind(L_fill_192_bytes_loop_header_zmm);
9526 subq(count, 192 >> shift);
9527
9528 align32();
9529 bind(L_fill_192_bytes_loop_zmm);
9530 fill64(to, 0, xtmp, true);
9531 fill64(to, 64, xtmp, true);
9532 fill64(to, 128, xtmp, true);
9533 addq(to, 192);
9534 subq(count, 192 >> shift);
9535 jccb(Assembler::greaterEqual, L_fill_192_bytes_loop_zmm);
9536
9537 addq(count, 192 >> shift);
9538 jcc(Assembler::zero, L_exit);
9539 jmp(L_fill_start_zmm_sequence);
9540 }
9541 bind(L_exit);
9542 }
9543 #endif
9544 #endif //COMPILER2_OR_JVMCI
9545
9546
9547 #ifdef _LP64
9548 void MacroAssembler::convert_f2i(Register dst, XMMRegister src) {
9549 Label done;
9550 cvttss2sil(dst, src);
9551 // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
9552 cmpl(dst, 0x80000000); // float_sign_flip
9553 jccb(Assembler::notEqual, done);
9554 subptr(rsp, 8);
9555 movflt(Address(rsp, 0), src);
9556 call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2i_fixup())));
9557 pop(dst);
9558 bind(done);
9559 }
9560
9561 void MacroAssembler::convert_d2i(Register dst, XMMRegister src) {
9562 Label done;
9563 cvttsd2sil(dst, src);
9564 // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
9565 cmpl(dst, 0x80000000); // float_sign_flip
9566 jccb(Assembler::notEqual, done);
9567 subptr(rsp, 8);
9568 movdbl(Address(rsp, 0), src);
9569 call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2i_fixup())));
9570 pop(dst);
9571 bind(done);
9572 }
9573
9574 void MacroAssembler::convert_f2l(Register dst, XMMRegister src) {
9575 Label done;
9576 cvttss2siq(dst, src);
9577 cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
9578 jccb(Assembler::notEqual, done);
9579 subptr(rsp, 8);
9580 movflt(Address(rsp, 0), src);
9581 call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2l_fixup())));
9582 pop(dst);
9583 bind(done);
9584 }
9585
9586 void MacroAssembler::round_float(Register dst, XMMRegister src, Register rtmp, Register rcx) {
9587 // Following code is line by line assembly translation rounding algorithm.
9588 // Please refer to java.lang.Math.round(float) algorithm for details.
9589 const int32_t FloatConsts_EXP_BIT_MASK = 0x7F800000;
9590 const int32_t FloatConsts_SIGNIFICAND_WIDTH = 24;
9591 const int32_t FloatConsts_EXP_BIAS = 127;
9592 const int32_t FloatConsts_SIGNIF_BIT_MASK = 0x007FFFFF;
9593 const int32_t MINUS_32 = 0xFFFFFFE0;
9594 Label L_special_case, L_block1, L_exit;
9595 movl(rtmp, FloatConsts_EXP_BIT_MASK);
9596 movdl(dst, src);
9597 andl(dst, rtmp);
9598 sarl(dst, FloatConsts_SIGNIFICAND_WIDTH - 1);
9599 movl(rtmp, FloatConsts_SIGNIFICAND_WIDTH - 2 + FloatConsts_EXP_BIAS);
9600 subl(rtmp, dst);
9601 movl(rcx, rtmp);
9602 movl(dst, MINUS_32);
9603 testl(rtmp, dst);
9604 jccb(Assembler::notEqual, L_special_case);
9605 movdl(dst, src);
9606 andl(dst, FloatConsts_SIGNIF_BIT_MASK);
9607 orl(dst, FloatConsts_SIGNIF_BIT_MASK + 1);
9608 movdl(rtmp, src);
9609 testl(rtmp, rtmp);
9610 jccb(Assembler::greaterEqual, L_block1);
9611 negl(dst);
9612 bind(L_block1);
9613 sarl(dst);
9614 addl(dst, 0x1);
9615 sarl(dst, 0x1);
9616 jmp(L_exit);
9617 bind(L_special_case);
9618 convert_f2i(dst, src);
9619 bind(L_exit);
9620 }
9621
9622 void MacroAssembler::round_double(Register dst, XMMRegister src, Register rtmp, Register rcx) {
9623 // Following code is line by line assembly translation rounding algorithm.
9624 // Please refer to java.lang.Math.round(double) algorithm for details.
9625 const int64_t DoubleConsts_EXP_BIT_MASK = 0x7FF0000000000000L;
9626 const int64_t DoubleConsts_SIGNIFICAND_WIDTH = 53;
9627 const int64_t DoubleConsts_EXP_BIAS = 1023;
9628 const int64_t DoubleConsts_SIGNIF_BIT_MASK = 0x000FFFFFFFFFFFFFL;
9629 const int64_t MINUS_64 = 0xFFFFFFFFFFFFFFC0L;
9630 Label L_special_case, L_block1, L_exit;
9631 mov64(rtmp, DoubleConsts_EXP_BIT_MASK);
9632 movq(dst, src);
9633 andq(dst, rtmp);
9634 sarq(dst, DoubleConsts_SIGNIFICAND_WIDTH - 1);
9635 mov64(rtmp, DoubleConsts_SIGNIFICAND_WIDTH - 2 + DoubleConsts_EXP_BIAS);
9636 subq(rtmp, dst);
9637 movq(rcx, rtmp);
9638 mov64(dst, MINUS_64);
9639 testq(rtmp, dst);
9640 jccb(Assembler::notEqual, L_special_case);
9641 movq(dst, src);
9642 mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK);
9643 andq(dst, rtmp);
9644 mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK + 1);
9645 orq(dst, rtmp);
9646 movq(rtmp, src);
9647 testq(rtmp, rtmp);
9648 jccb(Assembler::greaterEqual, L_block1);
9649 negq(dst);
9650 bind(L_block1);
9651 sarq(dst);
9652 addq(dst, 0x1);
9653 sarq(dst, 0x1);
9654 jmp(L_exit);
9655 bind(L_special_case);
9656 convert_d2l(dst, src);
9657 bind(L_exit);
9658 }
9659
9660 void MacroAssembler::convert_d2l(Register dst, XMMRegister src) {
9661 Label done;
9662 cvttsd2siq(dst, src);
9663 cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
9664 jccb(Assembler::notEqual, done);
9665 subptr(rsp, 8);
9666 movdbl(Address(rsp, 0), src);
9667 call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2l_fixup())));
9668 pop(dst);
9669 bind(done);
9670 }
9671
9672 void MacroAssembler::cache_wb(Address line)
9673 {
9674 // 64 bit cpus always support clflush
9675 assert(VM_Version::supports_clflush(), "clflush should be available");
9676 bool optimized = VM_Version::supports_clflushopt();
9677 bool no_evict = VM_Version::supports_clwb();
9678
9679 // prefer clwb (writeback without evict) otherwise
9680 // prefer clflushopt (potentially parallel writeback with evict)
9681 // otherwise fallback on clflush (serial writeback with evict)
9682
9683 if (optimized) {
9684 if (no_evict) {
9685 clwb(line);
9686 } else {
9687 clflushopt(line);
9688 }
9689 } else {
9690 // no need for fence when using CLFLUSH
9691 clflush(line);
9692 }
9693 }
9694
9695 void MacroAssembler::cache_wbsync(bool is_pre)
9696 {
9697 assert(VM_Version::supports_clflush(), "clflush should be available");
9698 bool optimized = VM_Version::supports_clflushopt();
9699 bool no_evict = VM_Version::supports_clwb();
9700
9701 // pick the correct implementation
9702
9703 if (!is_pre && (optimized || no_evict)) {
9704 // need an sfence for post flush when using clflushopt or clwb
9705 // otherwise no no need for any synchroniaztion
9706
9707 sfence();
9708 }
9709 }
9710
9711 #endif // _LP64
9712
9713 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) {
9714 switch (cond) {
9715 // Note some conditions are synonyms for others
9716 case Assembler::zero: return Assembler::notZero;
9717 case Assembler::notZero: return Assembler::zero;
9718 case Assembler::less: return Assembler::greaterEqual;
9719 case Assembler::lessEqual: return Assembler::greater;
9720 case Assembler::greater: return Assembler::lessEqual;
9721 case Assembler::greaterEqual: return Assembler::less;
9722 case Assembler::below: return Assembler::aboveEqual;
9723 case Assembler::belowEqual: return Assembler::above;
9724 case Assembler::above: return Assembler::belowEqual;
9725 case Assembler::aboveEqual: return Assembler::below;
9726 case Assembler::overflow: return Assembler::noOverflow;
9727 case Assembler::noOverflow: return Assembler::overflow;
9728 case Assembler::negative: return Assembler::positive;
9729 case Assembler::positive: return Assembler::negative;
9730 case Assembler::parity: return Assembler::noParity;
9731 case Assembler::noParity: return Assembler::parity;
9732 }
9733 ShouldNotReachHere(); return Assembler::overflow;
9734 }
9735
9736 SkipIfEqual::SkipIfEqual(
9737 MacroAssembler* masm, const bool* flag_addr, bool value, Register rscratch) {
9738 _masm = masm;
9739 _masm->cmp8(ExternalAddress((address)flag_addr), value, rscratch);
9740 _masm->jcc(Assembler::equal, _label);
9741 }
9742
9743 SkipIfEqual::~SkipIfEqual() {
9744 _masm->bind(_label);
9745 }
9746
9747 // 32-bit Windows has its own fast-path implementation
9748 // of get_thread
9749 #if !defined(WIN32) || defined(_LP64)
9750
9751 // This is simply a call to Thread::current()
9752 void MacroAssembler::get_thread(Register thread) {
9753 if (thread != rax) {
9754 push(rax);
9755 }
9756 LP64_ONLY(push(rdi);)
9757 LP64_ONLY(push(rsi);)
9758 push(rdx);
9759 push(rcx);
9760 #ifdef _LP64
9761 push(r8);
9762 push(r9);
9763 push(r10);
9764 push(r11);
9765 #endif
9766
9767 MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, Thread::current), 0);
9768
9769 #ifdef _LP64
9770 pop(r11);
9771 pop(r10);
9772 pop(r9);
9773 pop(r8);
9774 #endif
9775 pop(rcx);
9776 pop(rdx);
9777 LP64_ONLY(pop(rsi);)
9778 LP64_ONLY(pop(rdi);)
9779 if (thread != rax) {
9780 mov(thread, rax);
9781 pop(rax);
9782 }
9783 }
9784
9785
9786 #endif // !WIN32 || _LP64
9787
9788 void MacroAssembler::check_stack_alignment(Register sp, const char* msg, unsigned bias, Register tmp) {
9789 Label L_stack_ok;
9790 if (bias == 0) {
9791 testptr(sp, 2 * wordSize - 1);
9792 } else {
9793 // lea(tmp, Address(rsp, bias);
9794 mov(tmp, sp);
9795 addptr(tmp, bias);
9796 testptr(tmp, 2 * wordSize - 1);
9797 }
9798 jcc(Assembler::equal, L_stack_ok);
9799 block_comment(msg);
9800 stop(msg);
9801 bind(L_stack_ok);
9802 }
9803
9804 // Implements fast-locking.
9805 // Branches to slow upon failure to lock the object, with ZF cleared.
9806 // Falls through upon success with unspecified ZF.
9807 //
9808 // obj: the object to be locked
9809 // hdr: the (pre-loaded) header of the object, must be rax
9810 // thread: the thread which attempts to lock obj
9811 // tmp: a temporary register
9812 void MacroAssembler::fast_lock_impl(Register obj, Register hdr, Register thread, Register tmp, Label& slow) {
9813 assert(hdr == rax, "header must be in rax for cmpxchg");
9814 assert_different_registers(obj, hdr, thread, tmp);
9815
9816 // First we need to check if the lock-stack has room for pushing the object reference.
9817 // Note: we subtract 1 from the end-offset so that we can do a 'greater' comparison, instead
9818 // of 'greaterEqual' below, which readily clears the ZF. This makes C2 code a little simpler and
9819 // avoids one branch.
9820 cmpl(Address(thread, JavaThread::lock_stack_top_offset()), LockStack::end_offset() - 1);
9821 jcc(Assembler::greater, slow);
9822
9823 // Now we attempt to take the fast-lock.
9824 // Clear lock_mask bits (locked state).
9825 andptr(hdr, ~(int32_t)markWord::lock_mask_in_place);
9826 movptr(tmp, hdr);
9827 // Set unlocked_value bit.
9828 orptr(hdr, markWord::unlocked_value);
9829 lock();
9830 cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
9831 jcc(Assembler::notEqual, slow);
9832
9833 // If successful, push object to lock-stack.
9834 movl(tmp, Address(thread, JavaThread::lock_stack_top_offset()));
9835 movptr(Address(thread, tmp), obj);
9836 incrementl(tmp, oopSize);
9837 movl(Address(thread, JavaThread::lock_stack_top_offset()), tmp);
9838 }
9839
9840 // Implements fast-unlocking.
9841 // Branches to slow upon failure, with ZF cleared.
9842 // Falls through upon success, with unspecified ZF.
9843 //
9844 // obj: the object to be unlocked
9845 // hdr: the (pre-loaded) header of the object, must be rax
9846 // tmp: a temporary register
9847 void MacroAssembler::fast_unlock_impl(Register obj, Register hdr, Register tmp, Label& slow) {
9848 assert(hdr == rax, "header must be in rax for cmpxchg");
9849 assert_different_registers(obj, hdr, tmp);
9850
9851 // Mark-word must be lock_mask now, try to swing it back to unlocked_value.
9852 movptr(tmp, hdr); // The expected old value
9853 orptr(tmp, markWord::unlocked_value);
9854 lock();
9855 cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
9856 jcc(Assembler::notEqual, slow);
9857 // Pop the lock object from the lock-stack.
9858 #ifdef _LP64
9859 const Register thread = r15_thread;
9860 #else
9861 const Register thread = rax;
9862 get_thread(thread);
9863 #endif
9864 subl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
9865 #ifdef ASSERT
9866 movl(tmp, Address(thread, JavaThread::lock_stack_top_offset()));
9867 movptr(Address(thread, tmp), 0);
9868 #endif
9869 }