1 /*
2 * Copyright (c) 1997, 2025, 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 "asm/assembler.hpp"
26 #include "asm/assembler.inline.hpp"
27 #include "code/compiledIC.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 "interpreter/interpreterRuntime.hpp"
38 #include "jvm.h"
39 #include "memory/resourceArea.hpp"
40 #include "memory/universe.hpp"
41 #include "oops/accessDecorators.hpp"
42 #include "oops/compressedKlass.inline.hpp"
43 #include "oops/compressedOops.inline.hpp"
44 #include "oops/klass.inline.hpp"
45 #include "prims/methodHandles.hpp"
46 #include "runtime/continuation.hpp"
47 #include "runtime/interfaceSupport.inline.hpp"
48 #include "runtime/javaThread.hpp"
49 #include "runtime/jniHandles.hpp"
50 #include "runtime/objectMonitor.hpp"
51 #include "runtime/os.hpp"
52 #include "runtime/safepoint.hpp"
53 #include "runtime/safepointMechanism.hpp"
54 #include "runtime/sharedRuntime.hpp"
55 #include "runtime/stubRoutines.hpp"
56 #include "utilities/checkedCast.hpp"
57 #include "utilities/macros.hpp"
58
59 #ifdef PRODUCT
60 #define BLOCK_COMMENT(str) /* nothing */
61 #define STOP(error) stop(error)
62 #else
63 #define BLOCK_COMMENT(str) block_comment(str)
64 #define STOP(error) block_comment(error); stop(error)
65 #endif
66
67 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
68
69 #ifdef ASSERT
70 bool AbstractAssembler::pd_check_instruction_mark() { return true; }
71 #endif
72
73 static const Assembler::Condition reverse[] = {
74 Assembler::noOverflow /* overflow = 0x0 */ ,
75 Assembler::overflow /* noOverflow = 0x1 */ ,
76 Assembler::aboveEqual /* carrySet = 0x2, below = 0x2 */ ,
77 Assembler::below /* aboveEqual = 0x3, carryClear = 0x3 */ ,
78 Assembler::notZero /* zero = 0x4, equal = 0x4 */ ,
79 Assembler::zero /* notZero = 0x5, notEqual = 0x5 */ ,
80 Assembler::above /* belowEqual = 0x6 */ ,
81 Assembler::belowEqual /* above = 0x7 */ ,
82 Assembler::positive /* negative = 0x8 */ ,
83 Assembler::negative /* positive = 0x9 */ ,
84 Assembler::noParity /* parity = 0xa */ ,
85 Assembler::parity /* noParity = 0xb */ ,
86 Assembler::greaterEqual /* less = 0xc */ ,
87 Assembler::less /* greaterEqual = 0xd */ ,
88 Assembler::greater /* lessEqual = 0xe */ ,
89 Assembler::lessEqual /* greater = 0xf, */
90
91 };
92
93
94 // Implementation of MacroAssembler
95
96 Address MacroAssembler::as_Address(AddressLiteral adr) {
97 // amd64 always does this as a pc-rel
98 // we can be absolute or disp based on the instruction type
99 // jmp/call are displacements others are absolute
100 assert(!adr.is_lval(), "must be rval");
101 assert(reachable(adr), "must be");
102 return Address(checked_cast<int32_t>(adr.target() - pc()), adr.target(), adr.reloc());
103
104 }
105
106 Address MacroAssembler::as_Address(ArrayAddress adr, Register rscratch) {
107 AddressLiteral base = adr.base();
108 lea(rscratch, base);
109 Address index = adr.index();
110 assert(index._disp == 0, "must not have disp"); // maybe it can?
111 Address array(rscratch, index._index, index._scale, index._disp);
112 return array;
113 }
114
115 void MacroAssembler::call_VM_leaf_base(address entry_point, int num_args) {
116 Label L, E;
117
118 #ifdef _WIN64
119 // Windows always allocates space for it's register args
120 assert(num_args <= 4, "only register arguments supported");
121 subq(rsp, frame::arg_reg_save_area_bytes);
122 #endif
123
124 // Align stack if necessary
125 testl(rsp, 15);
126 jcc(Assembler::zero, L);
127
128 subq(rsp, 8);
129 call(RuntimeAddress(entry_point));
130 addq(rsp, 8);
131 jmp(E);
132
133 bind(L);
134 call(RuntimeAddress(entry_point));
135
136 bind(E);
137
138 #ifdef _WIN64
139 // restore stack pointer
140 addq(rsp, frame::arg_reg_save_area_bytes);
141 #endif
142 }
143
144 void MacroAssembler::cmp64(Register src1, AddressLiteral src2, Register rscratch) {
145 assert(!src2.is_lval(), "should use cmpptr");
146 assert(rscratch != noreg || always_reachable(src2), "missing");
147
148 if (reachable(src2)) {
149 cmpq(src1, as_Address(src2));
150 } else {
151 lea(rscratch, src2);
152 Assembler::cmpq(src1, Address(rscratch, 0));
153 }
154 }
155
156 int MacroAssembler::corrected_idivq(Register reg) {
157 // Full implementation of Java ldiv and lrem; checks for special
158 // case as described in JVM spec., p.243 & p.271. The function
159 // returns the (pc) offset of the idivl instruction - may be needed
160 // for implicit exceptions.
161 //
162 // normal case special case
163 //
164 // input : rax: dividend min_long
165 // reg: divisor (may not be eax/edx) -1
166 //
167 // output: rax: quotient (= rax idiv reg) min_long
168 // rdx: remainder (= rax irem reg) 0
169 assert(reg != rax && reg != rdx, "reg cannot be rax or rdx register");
170 static const int64_t min_long = 0x8000000000000000;
171 Label normal_case, special_case;
172
173 // check for special case
174 cmp64(rax, ExternalAddress((address) &min_long), rdx /*rscratch*/);
175 jcc(Assembler::notEqual, normal_case);
176 xorl(rdx, rdx); // prepare rdx for possible special case (where
177 // remainder = 0)
178 cmpq(reg, -1);
179 jcc(Assembler::equal, special_case);
180
181 // handle normal case
182 bind(normal_case);
183 cdqq();
184 int idivq_offset = offset();
185 idivq(reg);
186
187 // normal and special case exit
188 bind(special_case);
189
190 return idivq_offset;
191 }
192
193 void MacroAssembler::decrementq(Register reg, int value) {
194 if (value == min_jint) { subq(reg, value); return; }
195 if (value < 0) { incrementq(reg, -value); return; }
196 if (value == 0) { ; return; }
197 if (value == 1 && UseIncDec) { decq(reg) ; return; }
198 /* else */ { subq(reg, value) ; return; }
199 }
200
201 void MacroAssembler::decrementq(Address dst, int value) {
202 if (value == min_jint) { subq(dst, value); return; }
203 if (value < 0) { incrementq(dst, -value); return; }
204 if (value == 0) { ; return; }
205 if (value == 1 && UseIncDec) { decq(dst) ; return; }
206 /* else */ { subq(dst, value) ; return; }
207 }
208
209 void MacroAssembler::incrementq(AddressLiteral dst, Register rscratch) {
210 assert(rscratch != noreg || always_reachable(dst), "missing");
211
212 if (reachable(dst)) {
213 incrementq(as_Address(dst));
214 } else {
215 lea(rscratch, dst);
216 incrementq(Address(rscratch, 0));
217 }
218 }
219
220 void MacroAssembler::incrementq(Register reg, int value) {
221 if (value == min_jint) { addq(reg, value); return; }
222 if (value < 0) { decrementq(reg, -value); return; }
223 if (value == 0) { ; return; }
224 if (value == 1 && UseIncDec) { incq(reg) ; return; }
225 /* else */ { addq(reg, value) ; return; }
226 }
227
228 void MacroAssembler::incrementq(Address dst, int value) {
229 if (value == min_jint) { addq(dst, value); return; }
230 if (value < 0) { decrementq(dst, -value); return; }
231 if (value == 0) { ; return; }
232 if (value == 1 && UseIncDec) { incq(dst) ; return; }
233 /* else */ { addq(dst, value) ; return; }
234 }
235
236 // 32bit can do a case table jump in one instruction but we no longer allow the base
237 // to be installed in the Address class
238 void MacroAssembler::jump(ArrayAddress entry, Register rscratch) {
239 lea(rscratch, entry.base());
240 Address dispatch = entry.index();
241 assert(dispatch._base == noreg, "must be");
242 dispatch._base = rscratch;
243 jmp(dispatch);
244 }
245
246 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) {
247 ShouldNotReachHere(); // 64bit doesn't use two regs
248 cmpq(x_lo, y_lo);
249 }
250
251 void MacroAssembler::lea(Register dst, AddressLiteral src) {
252 mov_literal64(dst, (intptr_t)src.target(), src.rspec());
253 }
254
255 void MacroAssembler::lea(Address dst, AddressLiteral adr, Register rscratch) {
256 lea(rscratch, adr);
257 movptr(dst, rscratch);
258 }
259
260 void MacroAssembler::leave() {
261 // %%% is this really better? Why not on 32bit too?
262 emit_int8((unsigned char)0xC9); // LEAVE
263 }
264
265 void MacroAssembler::lneg(Register hi, Register lo) {
266 ShouldNotReachHere(); // 64bit doesn't use two regs
267 negq(lo);
268 }
269
270 void MacroAssembler::movoop(Register dst, jobject obj) {
271 mov_literal64(dst, (intptr_t)obj, oop_Relocation::spec_for_immediate());
272 }
273
274 void MacroAssembler::movoop(Address dst, jobject obj, Register rscratch) {
275 mov_literal64(rscratch, (intptr_t)obj, oop_Relocation::spec_for_immediate());
276 movq(dst, rscratch);
277 }
278
279 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
280 mov_literal64(dst, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
281 }
282
283 void MacroAssembler::mov_metadata(Address dst, Metadata* obj, Register rscratch) {
284 mov_literal64(rscratch, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
285 movq(dst, rscratch);
286 }
287
288 void MacroAssembler::movptr(Register dst, AddressLiteral src) {
289 if (src.is_lval()) {
290 mov_literal64(dst, (intptr_t)src.target(), src.rspec());
291 } else {
292 if (reachable(src)) {
293 movq(dst, as_Address(src));
294 } else {
295 lea(dst, src);
296 movq(dst, Address(dst, 0));
297 }
298 }
299 }
300
301 void MacroAssembler::movptr(ArrayAddress dst, Register src, Register rscratch) {
302 movq(as_Address(dst, rscratch), src);
303 }
304
305 void MacroAssembler::movptr(Register dst, ArrayAddress src) {
306 movq(dst, as_Address(src, dst /*rscratch*/));
307 }
308
309 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
310 void MacroAssembler::movptr(Address dst, intptr_t src, Register rscratch) {
311 if (is_simm32(src)) {
312 movptr(dst, checked_cast<int32_t>(src));
313 } else {
314 mov64(rscratch, src);
315 movq(dst, rscratch);
316 }
317 }
318
319 void MacroAssembler::pushoop(jobject obj, Register rscratch) {
320 movoop(rscratch, obj);
321 push(rscratch);
322 }
323
324 void MacroAssembler::pushklass(Metadata* obj, Register rscratch) {
325 mov_metadata(rscratch, obj);
326 push(rscratch);
327 }
328
329 void MacroAssembler::pushptr(AddressLiteral src, Register rscratch) {
330 lea(rscratch, src);
331 if (src.is_lval()) {
332 push(rscratch);
333 } else {
334 pushq(Address(rscratch, 0));
335 }
336 }
337
338 static void pass_arg0(MacroAssembler* masm, Register arg) {
339 if (c_rarg0 != arg ) {
340 masm->mov(c_rarg0, arg);
341 }
342 }
343
344 static void pass_arg1(MacroAssembler* masm, Register arg) {
345 if (c_rarg1 != arg ) {
346 masm->mov(c_rarg1, arg);
347 }
348 }
349
350 static void pass_arg2(MacroAssembler* masm, Register arg) {
351 if (c_rarg2 != arg ) {
352 masm->mov(c_rarg2, arg);
353 }
354 }
355
356 static void pass_arg3(MacroAssembler* masm, Register arg) {
357 if (c_rarg3 != arg ) {
358 masm->mov(c_rarg3, arg);
359 }
360 }
361
362 void MacroAssembler::stop(const char* msg) {
363 if (ShowMessageBoxOnError) {
364 address rip = pc();
365 pusha(); // get regs on stack
366 lea(c_rarg1, InternalAddress(rip));
367 movq(c_rarg2, rsp); // pass pointer to regs array
368 }
369 lea(c_rarg0, ExternalAddress((address) msg));
370 andq(rsp, -16); // align stack as required by ABI
371 call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
372 hlt();
373 }
374
375 void MacroAssembler::warn(const char* msg) {
376 push(rbp);
377 movq(rbp, rsp);
378 andq(rsp, -16); // align stack as required by push_CPU_state and call
379 push_CPU_state(); // keeps alignment at 16 bytes
380
381 #ifdef _WIN64
382 // Windows always allocates space for its register args
383 subq(rsp, frame::arg_reg_save_area_bytes);
384 #endif
385 lea(c_rarg0, ExternalAddress((address) msg));
386 call(RuntimeAddress(CAST_FROM_FN_PTR(address, warning)));
387
388 #ifdef _WIN64
389 // restore stack pointer
390 addq(rsp, frame::arg_reg_save_area_bytes);
391 #endif
392 pop_CPU_state();
393 mov(rsp, rbp);
394 pop(rbp);
395 }
396
397 void MacroAssembler::print_state() {
398 address rip = pc();
399 pusha(); // get regs on stack
400 push(rbp);
401 movq(rbp, rsp);
402 andq(rsp, -16); // align stack as required by push_CPU_state and call
403 push_CPU_state(); // keeps alignment at 16 bytes
404
405 lea(c_rarg0, InternalAddress(rip));
406 lea(c_rarg1, Address(rbp, wordSize)); // pass pointer to regs array
407 call_VM_leaf(CAST_FROM_FN_PTR(address, MacroAssembler::print_state64), c_rarg0, c_rarg1);
408
409 pop_CPU_state();
410 mov(rsp, rbp);
411 pop(rbp);
412 popa();
413 }
414
415 #ifndef PRODUCT
416 extern "C" void findpc(intptr_t x);
417 #endif
418
419 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[]) {
420 // In order to get locks to work, we need to fake a in_VM state
421 if (ShowMessageBoxOnError) {
422 JavaThread* thread = JavaThread::current();
423 JavaThreadState saved_state = thread->thread_state();
424 thread->set_thread_state(_thread_in_vm);
425 #ifndef PRODUCT
426 if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
427 ttyLocker ttyl;
428 BytecodeCounter::print();
429 }
430 #endif
431 // To see where a verify_oop failed, get $ebx+40/X for this frame.
432 // XXX correct this offset for amd64
433 // This is the value of eip which points to where verify_oop will return.
434 if (os::message_box(msg, "Execution stopped, print registers?")) {
435 print_state64(pc, regs);
436 BREAKPOINT;
437 }
438 }
439 fatal("DEBUG MESSAGE: %s", msg);
440 }
441
442 void MacroAssembler::print_state64(int64_t pc, int64_t regs[]) {
443 ttyLocker ttyl;
444 DebuggingContext debugging{};
445 tty->print_cr("rip = 0x%016lx", (intptr_t)pc);
446 #ifndef PRODUCT
447 tty->cr();
448 findpc(pc);
449 tty->cr();
450 #endif
451 #define PRINT_REG(rax, value) \
452 { tty->print("%s = ", #rax); os::print_location(tty, value); }
453 PRINT_REG(rax, regs[15]);
454 PRINT_REG(rbx, regs[12]);
455 PRINT_REG(rcx, regs[14]);
456 PRINT_REG(rdx, regs[13]);
457 PRINT_REG(rdi, regs[8]);
458 PRINT_REG(rsi, regs[9]);
459 PRINT_REG(rbp, regs[10]);
460 // rsp is actually not stored by pusha(), compute the old rsp from regs (rsp after pusha): regs + 16 = old rsp
461 PRINT_REG(rsp, (intptr_t)(®s[16]));
462 PRINT_REG(r8 , regs[7]);
463 PRINT_REG(r9 , regs[6]);
464 PRINT_REG(r10, regs[5]);
465 PRINT_REG(r11, regs[4]);
466 PRINT_REG(r12, regs[3]);
467 PRINT_REG(r13, regs[2]);
468 PRINT_REG(r14, regs[1]);
469 PRINT_REG(r15, regs[0]);
470 #undef PRINT_REG
471 // Print some words near the top of the stack.
472 int64_t* rsp = ®s[16];
473 int64_t* dump_sp = rsp;
474 for (int col1 = 0; col1 < 8; col1++) {
475 tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
476 os::print_location(tty, *dump_sp++);
477 }
478 for (int row = 0; row < 25; row++) {
479 tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
480 for (int col = 0; col < 4; col++) {
481 tty->print(" 0x%016lx", (intptr_t)*dump_sp++);
482 }
483 tty->cr();
484 }
485 // Print some instructions around pc:
486 Disassembler::decode((address)pc-64, (address)pc);
487 tty->print_cr("--------");
488 Disassembler::decode((address)pc, (address)pc+32);
489 }
490
491 // The java_calling_convention describes stack locations as ideal slots on
492 // a frame with no abi restrictions. Since we must observe abi restrictions
493 // (like the placement of the register window) the slots must be biased by
494 // the following value.
495 static int reg2offset_in(VMReg r) {
496 // Account for saved rbp and return address
497 // This should really be in_preserve_stack_slots
498 return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
499 }
500
501 static int reg2offset_out(VMReg r) {
502 return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
503 }
504
505 // A long move
506 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
507
508 // The calling conventions assures us that each VMregpair is either
509 // all really one physical register or adjacent stack slots.
510
511 if (src.is_single_phys_reg() ) {
512 if (dst.is_single_phys_reg()) {
513 if (dst.first() != src.first()) {
514 mov(dst.first()->as_Register(), src.first()->as_Register());
515 }
516 } else {
517 assert(dst.is_single_reg(), "not a stack pair: (%s, %s), (%s, %s)",
518 src.first()->name(), src.second()->name(), dst.first()->name(), dst.second()->name());
519 movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
520 }
521 } else if (dst.is_single_phys_reg()) {
522 assert(src.is_single_reg(), "not a stack pair");
523 movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
524 } else {
525 assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
526 movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
527 movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
528 }
529 }
530
531 // A double move
532 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
533
534 // The calling conventions assures us that each VMregpair is either
535 // all really one physical register or adjacent stack slots.
536
537 if (src.is_single_phys_reg() ) {
538 if (dst.is_single_phys_reg()) {
539 // In theory these overlap but the ordering is such that this is likely a nop
540 if ( src.first() != dst.first()) {
541 movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister());
542 }
543 } else {
544 assert(dst.is_single_reg(), "not a stack pair");
545 movdbl(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
546 }
547 } else if (dst.is_single_phys_reg()) {
548 assert(src.is_single_reg(), "not a stack pair");
549 movdbl(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
550 } else {
551 assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
552 movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
553 movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
554 }
555 }
556
557
558 // A float arg may have to do float reg int reg conversion
559 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
560 assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move");
561
562 // The calling conventions assures us that each VMregpair is either
563 // all really one physical register or adjacent stack slots.
564
565 if (src.first()->is_stack()) {
566 if (dst.first()->is_stack()) {
567 movl(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
568 movptr(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
569 } else {
570 // stack to reg
571 assert(dst.first()->is_XMMRegister(), "only expect xmm registers as parameters");
572 movflt(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
573 }
574 } else if (dst.first()->is_stack()) {
575 // reg to stack
576 assert(src.first()->is_XMMRegister(), "only expect xmm registers as parameters");
577 movflt(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
578 } else {
579 // reg to reg
580 // In theory these overlap but the ordering is such that this is likely a nop
581 if ( src.first() != dst.first()) {
582 movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister());
583 }
584 }
585 }
586
587 // On 64 bit we will store integer like items to the stack as
588 // 64 bits items (x86_32/64 abi) even though java would only store
589 // 32bits for a parameter. On 32bit it will simply be 32 bits
590 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
591 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
592 if (src.first()->is_stack()) {
593 if (dst.first()->is_stack()) {
594 // stack to stack
595 movslq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
596 movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
597 } else {
598 // stack to reg
599 movslq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
600 }
601 } else if (dst.first()->is_stack()) {
602 // reg to stack
603 // Do we really have to sign extend???
604 // __ movslq(src.first()->as_Register(), src.first()->as_Register());
605 movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
606 } else {
607 // Do we really have to sign extend???
608 // __ movslq(dst.first()->as_Register(), src.first()->as_Register());
609 if (dst.first() != src.first()) {
610 movq(dst.first()->as_Register(), src.first()->as_Register());
611 }
612 }
613 }
614
615 void MacroAssembler::move_ptr(VMRegPair src, VMRegPair dst) {
616 if (src.first()->is_stack()) {
617 if (dst.first()->is_stack()) {
618 // stack to stack
619 movq(rax, Address(rbp, reg2offset_in(src.first())));
620 movq(Address(rsp, reg2offset_out(dst.first())), rax);
621 } else {
622 // stack to reg
623 movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first())));
624 }
625 } else if (dst.first()->is_stack()) {
626 // reg to stack
627 movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register());
628 } else {
629 if (dst.first() != src.first()) {
630 movq(dst.first()->as_Register(), src.first()->as_Register());
631 }
632 }
633 }
634
635 // An oop arg. Must pass a handle not the oop itself
636 void MacroAssembler::object_move(OopMap* map,
637 int oop_handle_offset,
638 int framesize_in_slots,
639 VMRegPair src,
640 VMRegPair dst,
641 bool is_receiver,
642 int* receiver_offset) {
643
644 // must pass a handle. First figure out the location we use as a handle
645
646 Register rHandle = dst.first()->is_stack() ? rax : dst.first()->as_Register();
647
648 // See if oop is null if it is we need no handle
649
650 if (src.first()->is_stack()) {
651
652 // Oop is already on the stack as an argument
653 int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
654 map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
655 if (is_receiver) {
656 *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
657 }
658
659 cmpptr(Address(rbp, reg2offset_in(src.first())), NULL_WORD);
660 lea(rHandle, Address(rbp, reg2offset_in(src.first())));
661 // conditionally move a null
662 cmovptr(Assembler::equal, rHandle, Address(rbp, reg2offset_in(src.first())));
663 } else {
664
665 // Oop is in a register we must store it to the space we reserve
666 // on the stack for oop_handles and pass a handle if oop is non-null
667
668 const Register rOop = src.first()->as_Register();
669 int oop_slot;
670 if (rOop == j_rarg0)
671 oop_slot = 0;
672 else if (rOop == j_rarg1)
673 oop_slot = 1;
674 else if (rOop == j_rarg2)
675 oop_slot = 2;
676 else if (rOop == j_rarg3)
677 oop_slot = 3;
678 else if (rOop == j_rarg4)
679 oop_slot = 4;
680 else {
681 assert(rOop == j_rarg5, "wrong register");
682 oop_slot = 5;
683 }
684
685 oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
686 int offset = oop_slot*VMRegImpl::stack_slot_size;
687
688 map->set_oop(VMRegImpl::stack2reg(oop_slot));
689 // Store oop in handle area, may be null
690 movptr(Address(rsp, offset), rOop);
691 if (is_receiver) {
692 *receiver_offset = offset;
693 }
694
695 cmpptr(rOop, NULL_WORD);
696 lea(rHandle, Address(rsp, offset));
697 // conditionally move a null from the handle area where it was just stored
698 cmovptr(Assembler::equal, rHandle, Address(rsp, offset));
699 }
700
701 // If arg is on the stack then place it otherwise it is already in correct reg.
702 if (dst.first()->is_stack()) {
703 movptr(Address(rsp, reg2offset_out(dst.first())), rHandle);
704 }
705 }
706
707 void MacroAssembler::addptr(Register dst, int32_t imm32) {
708 addq(dst, imm32);
709 }
710
711 void MacroAssembler::addptr(Register dst, Register src) {
712 addq(dst, src);
713 }
714
715 void MacroAssembler::addptr(Address dst, Register src) {
716 addq(dst, src);
717 }
718
719 void MacroAssembler::addsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
720 assert(rscratch != noreg || always_reachable(src), "missing");
721
722 if (reachable(src)) {
723 Assembler::addsd(dst, as_Address(src));
724 } else {
725 lea(rscratch, src);
726 Assembler::addsd(dst, Address(rscratch, 0));
727 }
728 }
729
730 void MacroAssembler::addss(XMMRegister dst, AddressLiteral src, Register rscratch) {
731 assert(rscratch != noreg || always_reachable(src), "missing");
732
733 if (reachable(src)) {
734 addss(dst, as_Address(src));
735 } else {
736 lea(rscratch, src);
737 addss(dst, Address(rscratch, 0));
738 }
739 }
740
741 void MacroAssembler::addpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
742 assert(rscratch != noreg || always_reachable(src), "missing");
743
744 if (reachable(src)) {
745 Assembler::addpd(dst, as_Address(src));
746 } else {
747 lea(rscratch, src);
748 Assembler::addpd(dst, Address(rscratch, 0));
749 }
750 }
751
752 // See 8273459. Function for ensuring 64-byte alignment, intended for stubs only.
753 // Stub code is generated once and never copied.
754 // NMethods can't use this because they get copied and we can't force alignment > 32 bytes.
755 void MacroAssembler::align64() {
756 align(64, (uint)(uintptr_t)pc());
757 }
758
759 void MacroAssembler::align32() {
760 align(32, (uint)(uintptr_t)pc());
761 }
762
763 void MacroAssembler::align(uint modulus) {
764 // 8273459: Ensure alignment is possible with current segment alignment
765 assert(modulus <= (uintx)CodeEntryAlignment, "Alignment must be <= CodeEntryAlignment");
766 align(modulus, offset());
767 }
768
769 void MacroAssembler::align(uint modulus, uint target) {
770 if (target % modulus != 0) {
771 nop(modulus - (target % modulus));
772 }
773 }
774
775 void MacroAssembler::push_f(XMMRegister r) {
776 subptr(rsp, wordSize);
777 movflt(Address(rsp, 0), r);
778 }
779
780 void MacroAssembler::pop_f(XMMRegister r) {
781 movflt(r, Address(rsp, 0));
782 addptr(rsp, wordSize);
783 }
784
785 void MacroAssembler::push_d(XMMRegister r) {
786 subptr(rsp, 2 * wordSize);
787 movdbl(Address(rsp, 0), r);
788 }
789
790 void MacroAssembler::pop_d(XMMRegister r) {
791 movdbl(r, Address(rsp, 0));
792 addptr(rsp, 2 * Interpreter::stackElementSize);
793 }
794
795 void MacroAssembler::andpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
796 // Used in sign-masking with aligned address.
797 assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
798 assert(rscratch != noreg || always_reachable(src), "missing");
799
800 if (UseAVX > 2 &&
801 (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
802 (dst->encoding() >= 16)) {
803 vpand(dst, dst, src, AVX_512bit, rscratch);
804 } else if (reachable(src)) {
805 Assembler::andpd(dst, as_Address(src));
806 } else {
807 lea(rscratch, src);
808 Assembler::andpd(dst, Address(rscratch, 0));
809 }
810 }
811
812 void MacroAssembler::andps(XMMRegister dst, AddressLiteral src, Register rscratch) {
813 // Used in sign-masking with aligned address.
814 assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
815 assert(rscratch != noreg || always_reachable(src), "missing");
816
817 if (reachable(src)) {
818 Assembler::andps(dst, as_Address(src));
819 } else {
820 lea(rscratch, src);
821 Assembler::andps(dst, Address(rscratch, 0));
822 }
823 }
824
825 void MacroAssembler::andptr(Register dst, int32_t imm32) {
826 andq(dst, imm32);
827 }
828
829 void MacroAssembler::andq(Register dst, AddressLiteral src, Register rscratch) {
830 assert(rscratch != noreg || always_reachable(src), "missing");
831
832 if (reachable(src)) {
833 andq(dst, as_Address(src));
834 } else {
835 lea(rscratch, src);
836 andq(dst, Address(rscratch, 0));
837 }
838 }
839
840 void MacroAssembler::atomic_incl(Address counter_addr) {
841 lock();
842 incrementl(counter_addr);
843 }
844
845 void MacroAssembler::atomic_incl(AddressLiteral counter_addr, Register rscratch) {
846 assert(rscratch != noreg || always_reachable(counter_addr), "missing");
847
848 if (reachable(counter_addr)) {
849 atomic_incl(as_Address(counter_addr));
850 } else {
851 lea(rscratch, counter_addr);
852 atomic_incl(Address(rscratch, 0));
853 }
854 }
855
856 void MacroAssembler::atomic_incq(Address counter_addr) {
857 lock();
858 incrementq(counter_addr);
859 }
860
861 void MacroAssembler::atomic_incq(AddressLiteral counter_addr, Register rscratch) {
862 assert(rscratch != noreg || always_reachable(counter_addr), "missing");
863
864 if (reachable(counter_addr)) {
865 atomic_incq(as_Address(counter_addr));
866 } else {
867 lea(rscratch, counter_addr);
868 atomic_incq(Address(rscratch, 0));
869 }
870 }
871
872 // Writes to stack successive pages until offset reached to check for
873 // stack overflow + shadow pages. This clobbers tmp.
874 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
875 movptr(tmp, rsp);
876 // Bang stack for total size given plus shadow page size.
877 // Bang one page at a time because large size can bang beyond yellow and
878 // red zones.
879 Label loop;
880 bind(loop);
881 movl(Address(tmp, (-(int)os::vm_page_size())), size );
882 subptr(tmp, (int)os::vm_page_size());
883 subl(size, (int)os::vm_page_size());
884 jcc(Assembler::greater, loop);
885
886 // Bang down shadow pages too.
887 // At this point, (tmp-0) is the last address touched, so don't
888 // touch it again. (It was touched as (tmp-pagesize) but then tmp
889 // was post-decremented.) Skip this address by starting at i=1, and
890 // touch a few more pages below. N.B. It is important to touch all
891 // the way down including all pages in the shadow zone.
892 for (int i = 1; i < ((int)StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()); i++) {
893 // this could be any sized move but this is can be a debugging crumb
894 // so the bigger the better.
895 movptr(Address(tmp, (-i*(int)os::vm_page_size())), size );
896 }
897 }
898
899 void MacroAssembler::reserved_stack_check() {
900 // testing if reserved zone needs to be enabled
901 Label no_reserved_zone_enabling;
902
903 cmpptr(rsp, Address(r15_thread, JavaThread::reserved_stack_activation_offset()));
904 jcc(Assembler::below, no_reserved_zone_enabling);
905
906 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), r15_thread);
907 jump(RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
908 should_not_reach_here();
909
910 bind(no_reserved_zone_enabling);
911 }
912
913 void MacroAssembler::c2bool(Register x) {
914 // implements x == 0 ? 0 : 1
915 // note: must only look at least-significant byte of x
916 // since C-style booleans are stored in one byte
917 // only! (was bug)
918 andl(x, 0xFF);
919 setb(Assembler::notZero, x);
920 }
921
922 // Wouldn't need if AddressLiteral version had new name
923 void MacroAssembler::call(Label& L, relocInfo::relocType rtype) {
924 Assembler::call(L, rtype);
925 }
926
927 void MacroAssembler::call(Register entry) {
928 Assembler::call(entry);
929 }
930
931 void MacroAssembler::call(AddressLiteral entry, Register rscratch) {
932 assert(rscratch != noreg || always_reachable(entry), "missing");
933
934 if (reachable(entry)) {
935 Assembler::call_literal(entry.target(), entry.rspec());
936 } else {
937 lea(rscratch, entry);
938 Assembler::call(rscratch);
939 }
940 }
941
942 void MacroAssembler::ic_call(address entry, jint method_index) {
943 RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
944 // Needs full 64-bit immediate for later patching.
945 mov64(rax, (int64_t)Universe::non_oop_word());
946 call(AddressLiteral(entry, rh));
947 }
948
949 int MacroAssembler::ic_check_size() {
950 return UseCompactObjectHeaders ? 17 : 14;
951 }
952
953 int MacroAssembler::ic_check(int end_alignment) {
954 Register receiver = j_rarg0;
955 Register data = rax;
956 Register temp = rscratch1;
957
958 // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
959 // before the inline cache check, so we don't have to execute any nop instructions when dispatching
960 // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
961 // before the inline cache check here, and not after
962 align(end_alignment, offset() + ic_check_size());
963
964 int uep_offset = offset();
965
966 if (UseCompactObjectHeaders) {
967 load_narrow_klass_compact(temp, receiver);
968 cmpl(temp, Address(data, CompiledICData::speculated_klass_offset()));
969 } else if (UseCompressedClassPointers) {
970 movl(temp, Address(receiver, oopDesc::klass_offset_in_bytes()));
971 cmpl(temp, Address(data, CompiledICData::speculated_klass_offset()));
972 } else {
973 movptr(temp, Address(receiver, oopDesc::klass_offset_in_bytes()));
974 cmpptr(temp, Address(data, CompiledICData::speculated_klass_offset()));
975 }
976
977 // if inline cache check fails, then jump to runtime routine
978 jump_cc(Assembler::notEqual, RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
979 assert((offset() % end_alignment) == 0, "Misaligned verified entry point (%d, %d, %d)", uep_offset, offset(), end_alignment);
980
981 return uep_offset;
982 }
983
984 void MacroAssembler::emit_static_call_stub() {
985 // Static stub relocation also tags the Method* in the code-stream.
986 mov_metadata(rbx, (Metadata*) nullptr); // Method is zapped till fixup time.
987 // This is recognized as unresolved by relocs/nativeinst/ic code.
988 jump(RuntimeAddress(pc()));
989 }
990
991 // Implementation of call_VM versions
992
993 void MacroAssembler::call_VM(Register oop_result,
994 address entry_point,
995 bool check_exceptions) {
996 Label C, E;
997 call(C, relocInfo::none);
998 jmp(E);
999
1000 bind(C);
1001 call_VM_helper(oop_result, entry_point, 0, check_exceptions);
1002 ret(0);
1003
1004 bind(E);
1005 }
1006
1007 void MacroAssembler::call_VM(Register oop_result,
1008 address entry_point,
1009 Register arg_1,
1010 bool check_exceptions) {
1011 Label C, E;
1012 call(C, relocInfo::none);
1013 jmp(E);
1014
1015 bind(C);
1016 pass_arg1(this, arg_1);
1017 call_VM_helper(oop_result, entry_point, 1, check_exceptions);
1018 ret(0);
1019
1020 bind(E);
1021 }
1022
1023 void MacroAssembler::call_VM(Register oop_result,
1024 address entry_point,
1025 Register arg_1,
1026 Register arg_2,
1027 bool check_exceptions) {
1028 Label C, E;
1029 call(C, relocInfo::none);
1030 jmp(E);
1031
1032 bind(C);
1033
1034 assert_different_registers(arg_1, c_rarg2);
1035
1036 pass_arg2(this, arg_2);
1037 pass_arg1(this, arg_1);
1038 call_VM_helper(oop_result, entry_point, 2, check_exceptions);
1039 ret(0);
1040
1041 bind(E);
1042 }
1043
1044 void MacroAssembler::call_VM(Register oop_result,
1045 address entry_point,
1046 Register arg_1,
1047 Register arg_2,
1048 Register arg_3,
1049 bool check_exceptions) {
1050 Label C, E;
1051 call(C, relocInfo::none);
1052 jmp(E);
1053
1054 bind(C);
1055
1056 assert_different_registers(arg_1, c_rarg2, c_rarg3);
1057 assert_different_registers(arg_2, c_rarg3);
1058 pass_arg3(this, arg_3);
1059 pass_arg2(this, arg_2);
1060 pass_arg1(this, arg_1);
1061 call_VM_helper(oop_result, entry_point, 3, check_exceptions);
1062 ret(0);
1063
1064 bind(E);
1065 }
1066
1067 void MacroAssembler::call_VM(Register oop_result,
1068 Register last_java_sp,
1069 address entry_point,
1070 int number_of_arguments,
1071 bool check_exceptions) {
1072 call_VM_base(oop_result, last_java_sp, entry_point, number_of_arguments, check_exceptions);
1073 }
1074
1075 void MacroAssembler::call_VM(Register oop_result,
1076 Register last_java_sp,
1077 address entry_point,
1078 Register arg_1,
1079 bool check_exceptions) {
1080 pass_arg1(this, arg_1);
1081 call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
1082 }
1083
1084 void MacroAssembler::call_VM(Register oop_result,
1085 Register last_java_sp,
1086 address entry_point,
1087 Register arg_1,
1088 Register arg_2,
1089 bool check_exceptions) {
1090
1091 assert_different_registers(arg_1, c_rarg2);
1092 pass_arg2(this, arg_2);
1093 pass_arg1(this, arg_1);
1094 call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
1095 }
1096
1097 void MacroAssembler::call_VM(Register oop_result,
1098 Register last_java_sp,
1099 address entry_point,
1100 Register arg_1,
1101 Register arg_2,
1102 Register arg_3,
1103 bool check_exceptions) {
1104 assert_different_registers(arg_1, c_rarg2, c_rarg3);
1105 assert_different_registers(arg_2, c_rarg3);
1106 pass_arg3(this, arg_3);
1107 pass_arg2(this, arg_2);
1108 pass_arg1(this, arg_1);
1109 call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
1110 }
1111
1112 void MacroAssembler::super_call_VM(Register oop_result,
1113 Register last_java_sp,
1114 address entry_point,
1115 int number_of_arguments,
1116 bool check_exceptions) {
1117 MacroAssembler::call_VM_base(oop_result, last_java_sp, entry_point, number_of_arguments, check_exceptions);
1118 }
1119
1120 void MacroAssembler::super_call_VM(Register oop_result,
1121 Register last_java_sp,
1122 address entry_point,
1123 Register arg_1,
1124 bool check_exceptions) {
1125 pass_arg1(this, arg_1);
1126 super_call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
1127 }
1128
1129 void MacroAssembler::super_call_VM(Register oop_result,
1130 Register last_java_sp,
1131 address entry_point,
1132 Register arg_1,
1133 Register arg_2,
1134 bool check_exceptions) {
1135
1136 assert_different_registers(arg_1, c_rarg2);
1137 pass_arg2(this, arg_2);
1138 pass_arg1(this, arg_1);
1139 super_call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
1140 }
1141
1142 void MacroAssembler::super_call_VM(Register oop_result,
1143 Register last_java_sp,
1144 address entry_point,
1145 Register arg_1,
1146 Register arg_2,
1147 Register arg_3,
1148 bool check_exceptions) {
1149 assert_different_registers(arg_1, c_rarg2, c_rarg3);
1150 assert_different_registers(arg_2, c_rarg3);
1151 pass_arg3(this, arg_3);
1152 pass_arg2(this, arg_2);
1153 pass_arg1(this, arg_1);
1154 super_call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
1155 }
1156
1157 void MacroAssembler::call_VM_base(Register oop_result,
1158 Register last_java_sp,
1159 address entry_point,
1160 int number_of_arguments,
1161 bool check_exceptions) {
1162 Register java_thread = r15_thread;
1163
1164 // determine last_java_sp register
1165 if (!last_java_sp->is_valid()) {
1166 last_java_sp = rsp;
1167 }
1168 // debugging support
1169 assert(number_of_arguments >= 0 , "cannot have negative number of arguments");
1170 #ifdef ASSERT
1171 // TraceBytecodes does not use r12 but saves it over the call, so don't verify
1172 // r12 is the heapbase.
1173 if (UseCompressedOops && !TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
1174 #endif // ASSERT
1175
1176 assert(java_thread != oop_result , "cannot use the same register for java_thread & oop_result");
1177 assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
1178
1179 // push java thread (becomes first argument of C function)
1180
1181 mov(c_rarg0, r15_thread);
1182
1183 // set last Java frame before call
1184 assert(last_java_sp != rbp, "can't use ebp/rbp");
1185
1186 // Only interpreter should have to set fp
1187 set_last_Java_frame(last_java_sp, rbp, nullptr, rscratch1);
1188
1189 // do the call, remove parameters
1190 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1191
1192 #ifdef ASSERT
1193 // Check that thread register is not clobbered.
1194 guarantee(java_thread != rax, "change this code");
1195 push(rax);
1196 { Label L;
1197 get_thread_slow(rax);
1198 cmpptr(java_thread, rax);
1199 jcc(Assembler::equal, L);
1200 STOP("MacroAssembler::call_VM_base: java_thread not callee saved?");
1201 bind(L);
1202 }
1203 pop(rax);
1204 #endif
1205
1206 // reset last Java frame
1207 // Only interpreter should have to clear fp
1208 reset_last_Java_frame(true);
1209
1210 // C++ interp handles this in the interpreter
1211 check_and_handle_popframe();
1212 check_and_handle_earlyret();
1213
1214 if (check_exceptions) {
1215 // check for pending exceptions (java_thread is set upon return)
1216 cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
1217 // This used to conditionally jump to forward_exception however it is
1218 // possible if we relocate that the branch will not reach. So we must jump
1219 // around so we can always reach
1220
1221 Label ok;
1222 jcc(Assembler::equal, ok);
1223 jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
1224 bind(ok);
1225 }
1226
1227 // get oop result if there is one and reset the value in the thread
1228 if (oop_result->is_valid()) {
1229 get_vm_result_oop(oop_result);
1230 }
1231 }
1232
1233 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
1234 // Calculate the value for last_Java_sp somewhat subtle.
1235 // call_VM does an intermediate call which places a return address on
1236 // the stack just under the stack pointer as the user finished with it.
1237 // This allows use to retrieve last_Java_pc from last_Java_sp[-1].
1238
1239 // We've pushed one address, correct last_Java_sp
1240 lea(rax, Address(rsp, wordSize));
1241
1242 call_VM_base(oop_result, rax, entry_point, number_of_arguments, check_exceptions);
1243 }
1244
1245 // Use this method when MacroAssembler version of call_VM_leaf_base() should be called from Interpreter.
1246 void MacroAssembler::call_VM_leaf0(address entry_point) {
1247 MacroAssembler::call_VM_leaf_base(entry_point, 0);
1248 }
1249
1250 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
1251 call_VM_leaf_base(entry_point, number_of_arguments);
1252 }
1253
1254 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
1255 pass_arg0(this, arg_0);
1256 call_VM_leaf(entry_point, 1);
1257 }
1258
1259 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
1260
1261 assert_different_registers(arg_0, c_rarg1);
1262 pass_arg1(this, arg_1);
1263 pass_arg0(this, arg_0);
1264 call_VM_leaf(entry_point, 2);
1265 }
1266
1267 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
1268 assert_different_registers(arg_0, c_rarg1, c_rarg2);
1269 assert_different_registers(arg_1, c_rarg2);
1270 pass_arg2(this, arg_2);
1271 pass_arg1(this, arg_1);
1272 pass_arg0(this, arg_0);
1273 call_VM_leaf(entry_point, 3);
1274 }
1275
1276 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
1277 assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
1278 assert_different_registers(arg_1, c_rarg2, c_rarg3);
1279 assert_different_registers(arg_2, c_rarg3);
1280 pass_arg3(this, arg_3);
1281 pass_arg2(this, arg_2);
1282 pass_arg1(this, arg_1);
1283 pass_arg0(this, arg_0);
1284 call_VM_leaf(entry_point, 3);
1285 }
1286
1287 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
1288 pass_arg0(this, arg_0);
1289 MacroAssembler::call_VM_leaf_base(entry_point, 1);
1290 }
1291
1292 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
1293 assert_different_registers(arg_0, c_rarg1);
1294 pass_arg1(this, arg_1);
1295 pass_arg0(this, arg_0);
1296 MacroAssembler::call_VM_leaf_base(entry_point, 2);
1297 }
1298
1299 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
1300 assert_different_registers(arg_0, c_rarg1, c_rarg2);
1301 assert_different_registers(arg_1, c_rarg2);
1302 pass_arg2(this, arg_2);
1303 pass_arg1(this, arg_1);
1304 pass_arg0(this, arg_0);
1305 MacroAssembler::call_VM_leaf_base(entry_point, 3);
1306 }
1307
1308 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
1309 assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
1310 assert_different_registers(arg_1, c_rarg2, c_rarg3);
1311 assert_different_registers(arg_2, c_rarg3);
1312 pass_arg3(this, arg_3);
1313 pass_arg2(this, arg_2);
1314 pass_arg1(this, arg_1);
1315 pass_arg0(this, arg_0);
1316 MacroAssembler::call_VM_leaf_base(entry_point, 4);
1317 }
1318
1319 void MacroAssembler::get_vm_result_oop(Register oop_result) {
1320 movptr(oop_result, Address(r15_thread, JavaThread::vm_result_oop_offset()));
1321 movptr(Address(r15_thread, JavaThread::vm_result_oop_offset()), NULL_WORD);
1322 verify_oop_msg(oop_result, "broken oop in call_VM_base");
1323 }
1324
1325 void MacroAssembler::get_vm_result_metadata(Register metadata_result) {
1326 movptr(metadata_result, Address(r15_thread, JavaThread::vm_result_metadata_offset()));
1327 movptr(Address(r15_thread, JavaThread::vm_result_metadata_offset()), NULL_WORD);
1328 }
1329
1330 void MacroAssembler::check_and_handle_earlyret() {
1331 }
1332
1333 void MacroAssembler::check_and_handle_popframe() {
1334 }
1335
1336 void MacroAssembler::cmp32(AddressLiteral src1, int32_t imm, Register rscratch) {
1337 assert(rscratch != noreg || always_reachable(src1), "missing");
1338
1339 if (reachable(src1)) {
1340 cmpl(as_Address(src1), imm);
1341 } else {
1342 lea(rscratch, src1);
1343 cmpl(Address(rscratch, 0), imm);
1344 }
1345 }
1346
1347 void MacroAssembler::cmp32(Register src1, AddressLiteral src2, Register rscratch) {
1348 assert(!src2.is_lval(), "use cmpptr");
1349 assert(rscratch != noreg || always_reachable(src2), "missing");
1350
1351 if (reachable(src2)) {
1352 cmpl(src1, as_Address(src2));
1353 } else {
1354 lea(rscratch, src2);
1355 cmpl(src1, Address(rscratch, 0));
1356 }
1357 }
1358
1359 void MacroAssembler::cmp32(Register src1, int32_t imm) {
1360 Assembler::cmpl(src1, imm);
1361 }
1362
1363 void MacroAssembler::cmp32(Register src1, Address src2) {
1364 Assembler::cmpl(src1, src2);
1365 }
1366
1367 void MacroAssembler::cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
1368 ucomisd(opr1, opr2);
1369
1370 Label L;
1371 if (unordered_is_less) {
1372 movl(dst, -1);
1373 jcc(Assembler::parity, L);
1374 jcc(Assembler::below , L);
1375 movl(dst, 0);
1376 jcc(Assembler::equal , L);
1377 increment(dst);
1378 } else { // unordered is greater
1379 movl(dst, 1);
1380 jcc(Assembler::parity, L);
1381 jcc(Assembler::above , L);
1382 movl(dst, 0);
1383 jcc(Assembler::equal , L);
1384 decrementl(dst);
1385 }
1386 bind(L);
1387 }
1388
1389 void MacroAssembler::cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
1390 ucomiss(opr1, opr2);
1391
1392 Label L;
1393 if (unordered_is_less) {
1394 movl(dst, -1);
1395 jcc(Assembler::parity, L);
1396 jcc(Assembler::below , L);
1397 movl(dst, 0);
1398 jcc(Assembler::equal , L);
1399 increment(dst);
1400 } else { // unordered is greater
1401 movl(dst, 1);
1402 jcc(Assembler::parity, L);
1403 jcc(Assembler::above , L);
1404 movl(dst, 0);
1405 jcc(Assembler::equal , L);
1406 decrementl(dst);
1407 }
1408 bind(L);
1409 }
1410
1411
1412 void MacroAssembler::cmp8(AddressLiteral src1, int imm, Register rscratch) {
1413 assert(rscratch != noreg || always_reachable(src1), "missing");
1414
1415 if (reachable(src1)) {
1416 cmpb(as_Address(src1), imm);
1417 } else {
1418 lea(rscratch, src1);
1419 cmpb(Address(rscratch, 0), imm);
1420 }
1421 }
1422
1423 void MacroAssembler::cmpptr(Register src1, AddressLiteral src2, Register rscratch) {
1424 assert(rscratch != noreg || always_reachable(src2), "missing");
1425
1426 if (src2.is_lval()) {
1427 movptr(rscratch, src2);
1428 Assembler::cmpq(src1, rscratch);
1429 } else if (reachable(src2)) {
1430 cmpq(src1, as_Address(src2));
1431 } else {
1432 lea(rscratch, src2);
1433 Assembler::cmpq(src1, Address(rscratch, 0));
1434 }
1435 }
1436
1437 void MacroAssembler::cmpptr(Address src1, AddressLiteral src2, Register rscratch) {
1438 assert(src2.is_lval(), "not a mem-mem compare");
1439 // moves src2's literal address
1440 movptr(rscratch, src2);
1441 Assembler::cmpq(src1, rscratch);
1442 }
1443
1444 void MacroAssembler::cmpoop(Register src1, Register src2) {
1445 cmpptr(src1, src2);
1446 }
1447
1448 void MacroAssembler::cmpoop(Register src1, Address src2) {
1449 cmpptr(src1, src2);
1450 }
1451
1452 void MacroAssembler::cmpoop(Register src1, jobject src2, Register rscratch) {
1453 movoop(rscratch, src2);
1454 cmpptr(src1, rscratch);
1455 }
1456
1457 void MacroAssembler::locked_cmpxchgptr(Register reg, AddressLiteral adr, Register rscratch) {
1458 assert(rscratch != noreg || always_reachable(adr), "missing");
1459
1460 if (reachable(adr)) {
1461 lock();
1462 cmpxchgptr(reg, as_Address(adr));
1463 } else {
1464 lea(rscratch, adr);
1465 lock();
1466 cmpxchgptr(reg, Address(rscratch, 0));
1467 }
1468 }
1469
1470 void MacroAssembler::cmpxchgptr(Register reg, Address adr) {
1471 cmpxchgq(reg, adr);
1472 }
1473
1474 void MacroAssembler::comisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
1475 assert(rscratch != noreg || always_reachable(src), "missing");
1476
1477 if (reachable(src)) {
1478 Assembler::comisd(dst, as_Address(src));
1479 } else {
1480 lea(rscratch, src);
1481 Assembler::comisd(dst, Address(rscratch, 0));
1482 }
1483 }
1484
1485 void MacroAssembler::comiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
1486 assert(rscratch != noreg || always_reachable(src), "missing");
1487
1488 if (reachable(src)) {
1489 Assembler::comiss(dst, as_Address(src));
1490 } else {
1491 lea(rscratch, src);
1492 Assembler::comiss(dst, Address(rscratch, 0));
1493 }
1494 }
1495
1496
1497 void MacroAssembler::cond_inc32(Condition cond, AddressLiteral counter_addr, Register rscratch) {
1498 assert(rscratch != noreg || always_reachable(counter_addr), "missing");
1499
1500 Condition negated_cond = negate_condition(cond);
1501 Label L;
1502 jcc(negated_cond, L);
1503 pushf(); // Preserve flags
1504 atomic_incl(counter_addr, rscratch);
1505 popf();
1506 bind(L);
1507 }
1508
1509 int MacroAssembler::corrected_idivl(Register reg) {
1510 // Full implementation of Java idiv and irem; checks for
1511 // special case as described in JVM spec., p.243 & p.271.
1512 // The function returns the (pc) offset of the idivl
1513 // instruction - may be needed for implicit exceptions.
1514 //
1515 // normal case special case
1516 //
1517 // input : rax,: dividend min_int
1518 // reg: divisor (may not be rax,/rdx) -1
1519 //
1520 // output: rax,: quotient (= rax, idiv reg) min_int
1521 // rdx: remainder (= rax, irem reg) 0
1522 assert(reg != rax && reg != rdx, "reg cannot be rax, or rdx register");
1523 const int min_int = 0x80000000;
1524 Label normal_case, special_case;
1525
1526 // check for special case
1527 cmpl(rax, min_int);
1528 jcc(Assembler::notEqual, normal_case);
1529 xorl(rdx, rdx); // prepare rdx for possible special case (where remainder = 0)
1530 cmpl(reg, -1);
1531 jcc(Assembler::equal, special_case);
1532
1533 // handle normal case
1534 bind(normal_case);
1535 cdql();
1536 int idivl_offset = offset();
1537 idivl(reg);
1538
1539 // normal and special case exit
1540 bind(special_case);
1541
1542 return idivl_offset;
1543 }
1544
1545
1546
1547 void MacroAssembler::decrementl(Register reg, int value) {
1548 if (value == min_jint) {subl(reg, value) ; return; }
1549 if (value < 0) { incrementl(reg, -value); return; }
1550 if (value == 0) { ; return; }
1551 if (value == 1 && UseIncDec) { decl(reg) ; return; }
1552 /* else */ { subl(reg, value) ; return; }
1553 }
1554
1555 void MacroAssembler::decrementl(Address dst, int value) {
1556 if (value == min_jint) {subl(dst, value) ; return; }
1557 if (value < 0) { incrementl(dst, -value); return; }
1558 if (value == 0) { ; return; }
1559 if (value == 1 && UseIncDec) { decl(dst) ; return; }
1560 /* else */ { subl(dst, value) ; return; }
1561 }
1562
1563 void MacroAssembler::division_with_shift (Register reg, int shift_value) {
1564 assert(shift_value > 0, "illegal shift value");
1565 Label _is_positive;
1566 testl (reg, reg);
1567 jcc (Assembler::positive, _is_positive);
1568 int offset = (1 << shift_value) - 1 ;
1569
1570 if (offset == 1) {
1571 incrementl(reg);
1572 } else {
1573 addl(reg, offset);
1574 }
1575
1576 bind (_is_positive);
1577 sarl(reg, shift_value);
1578 }
1579
1580 void MacroAssembler::divsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
1581 assert(rscratch != noreg || always_reachable(src), "missing");
1582
1583 if (reachable(src)) {
1584 Assembler::divsd(dst, as_Address(src));
1585 } else {
1586 lea(rscratch, src);
1587 Assembler::divsd(dst, Address(rscratch, 0));
1588 }
1589 }
1590
1591 void MacroAssembler::divss(XMMRegister dst, AddressLiteral src, Register rscratch) {
1592 assert(rscratch != noreg || always_reachable(src), "missing");
1593
1594 if (reachable(src)) {
1595 Assembler::divss(dst, as_Address(src));
1596 } else {
1597 lea(rscratch, src);
1598 Assembler::divss(dst, Address(rscratch, 0));
1599 }
1600 }
1601
1602 void MacroAssembler::enter() {
1603 push(rbp);
1604 mov(rbp, rsp);
1605 }
1606
1607 void MacroAssembler::post_call_nop() {
1608 if (!Continuations::enabled()) {
1609 return;
1610 }
1611 InstructionMark im(this);
1612 relocate(post_call_nop_Relocation::spec());
1613 InlineSkippedInstructionsCounter skipCounter(this);
1614 emit_int8((uint8_t)0x0f);
1615 emit_int8((uint8_t)0x1f);
1616 emit_int8((uint8_t)0x84);
1617 emit_int8((uint8_t)0x00);
1618 emit_int32(0x00);
1619 }
1620
1621 // A 5 byte nop that is safe for patching (see patch_verified_entry)
1622 void MacroAssembler::fat_nop() {
1623 if (UseAddressNop) {
1624 addr_nop_5();
1625 } else {
1626 emit_int8((uint8_t)0x26); // es:
1627 emit_int8((uint8_t)0x2e); // cs:
1628 emit_int8((uint8_t)0x64); // fs:
1629 emit_int8((uint8_t)0x65); // gs:
1630 emit_int8((uint8_t)0x90);
1631 }
1632 }
1633
1634 void MacroAssembler::mulpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
1635 assert(rscratch != noreg || always_reachable(src), "missing");
1636 if (reachable(src)) {
1637 Assembler::mulpd(dst, as_Address(src));
1638 } else {
1639 lea(rscratch, src);
1640 Assembler::mulpd(dst, Address(rscratch, 0));
1641 }
1642 }
1643
1644 // dst = c = a * b + c
1645 void MacroAssembler::fmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
1646 Assembler::vfmadd231sd(c, a, b);
1647 if (dst != c) {
1648 movdbl(dst, c);
1649 }
1650 }
1651
1652 // dst = c = a * b + c
1653 void MacroAssembler::fmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
1654 Assembler::vfmadd231ss(c, a, b);
1655 if (dst != c) {
1656 movflt(dst, c);
1657 }
1658 }
1659
1660 // dst = c = a * b + c
1661 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
1662 Assembler::vfmadd231pd(c, a, b, vector_len);
1663 if (dst != c) {
1664 vmovdqu(dst, c);
1665 }
1666 }
1667
1668 // dst = c = a * b + c
1669 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
1670 Assembler::vfmadd231ps(c, a, b, vector_len);
1671 if (dst != c) {
1672 vmovdqu(dst, c);
1673 }
1674 }
1675
1676 // dst = c = a * b + c
1677 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
1678 Assembler::vfmadd231pd(c, a, b, vector_len);
1679 if (dst != c) {
1680 vmovdqu(dst, c);
1681 }
1682 }
1683
1684 // dst = c = a * b + c
1685 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
1686 Assembler::vfmadd231ps(c, a, b, vector_len);
1687 if (dst != c) {
1688 vmovdqu(dst, c);
1689 }
1690 }
1691
1692 void MacroAssembler::incrementl(AddressLiteral dst, Register rscratch) {
1693 assert(rscratch != noreg || always_reachable(dst), "missing");
1694
1695 if (reachable(dst)) {
1696 incrementl(as_Address(dst));
1697 } else {
1698 lea(rscratch, dst);
1699 incrementl(Address(rscratch, 0));
1700 }
1701 }
1702
1703 void MacroAssembler::incrementl(ArrayAddress dst, Register rscratch) {
1704 incrementl(as_Address(dst, rscratch));
1705 }
1706
1707 void MacroAssembler::incrementl(Register reg, int value) {
1708 if (value == min_jint) {addl(reg, value) ; return; }
1709 if (value < 0) { decrementl(reg, -value); return; }
1710 if (value == 0) { ; return; }
1711 if (value == 1 && UseIncDec) { incl(reg) ; return; }
1712 /* else */ { addl(reg, value) ; return; }
1713 }
1714
1715 void MacroAssembler::incrementl(Address dst, int value) {
1716 if (value == min_jint) {addl(dst, value) ; return; }
1717 if (value < 0) { decrementl(dst, -value); return; }
1718 if (value == 0) { ; return; }
1719 if (value == 1 && UseIncDec) { incl(dst) ; return; }
1720 /* else */ { addl(dst, value) ; return; }
1721 }
1722
1723 void MacroAssembler::jump(AddressLiteral dst, Register rscratch) {
1724 assert(rscratch != noreg || always_reachable(dst), "missing");
1725 assert(!dst.rspec().reloc()->is_data(), "should not use ExternalAddress for jump");
1726 if (reachable(dst)) {
1727 jmp_literal(dst.target(), dst.rspec());
1728 } else {
1729 lea(rscratch, dst);
1730 jmp(rscratch);
1731 }
1732 }
1733
1734 void MacroAssembler::jump_cc(Condition cc, AddressLiteral dst, Register rscratch) {
1735 assert(rscratch != noreg || always_reachable(dst), "missing");
1736 assert(!dst.rspec().reloc()->is_data(), "should not use ExternalAddress for jump_cc");
1737 if (reachable(dst)) {
1738 InstructionMark im(this);
1739 relocate(dst.reloc());
1740 const int short_size = 2;
1741 const int long_size = 6;
1742 int offs = (intptr_t)dst.target() - ((intptr_t)pc());
1743 if (dst.reloc() == relocInfo::none && is8bit(offs - short_size)) {
1744 // 0111 tttn #8-bit disp
1745 emit_int8(0x70 | cc);
1746 emit_int8((offs - short_size) & 0xFF);
1747 } else {
1748 // 0000 1111 1000 tttn #32-bit disp
1749 emit_int8(0x0F);
1750 emit_int8((unsigned char)(0x80 | cc));
1751 emit_int32(offs - long_size);
1752 }
1753 } else {
1754 #ifdef ASSERT
1755 warning("reversing conditional branch");
1756 #endif /* ASSERT */
1757 Label skip;
1758 jccb(reverse[cc], skip);
1759 lea(rscratch, dst);
1760 Assembler::jmp(rscratch);
1761 bind(skip);
1762 }
1763 }
1764
1765 void MacroAssembler::cmp32_mxcsr_std(Address mxcsr_save, Register tmp, Register rscratch) {
1766 ExternalAddress mxcsr_std(StubRoutines::x86::addr_mxcsr_std());
1767 assert(rscratch != noreg || always_reachable(mxcsr_std), "missing");
1768
1769 stmxcsr(mxcsr_save);
1770 movl(tmp, mxcsr_save);
1771 if (EnableX86ECoreOpts) {
1772 // The mxcsr_std has status bits set for performance on ECore
1773 orl(tmp, 0x003f);
1774 } else {
1775 // Mask out status bits (only check control and mask bits)
1776 andl(tmp, 0xFFC0);
1777 }
1778 cmp32(tmp, mxcsr_std, rscratch);
1779 }
1780
1781 void MacroAssembler::ldmxcsr(AddressLiteral src, Register rscratch) {
1782 assert(rscratch != noreg || always_reachable(src), "missing");
1783
1784 if (reachable(src)) {
1785 Assembler::ldmxcsr(as_Address(src));
1786 } else {
1787 lea(rscratch, src);
1788 Assembler::ldmxcsr(Address(rscratch, 0));
1789 }
1790 }
1791
1792 int MacroAssembler::load_signed_byte(Register dst, Address src) {
1793 int off = offset();
1794 movsbl(dst, src); // movsxb
1795 return off;
1796 }
1797
1798 // Note: load_signed_short used to be called load_signed_word.
1799 // Although the 'w' in x86 opcodes refers to the term "word" in the assembler
1800 // manual, which means 16 bits, that usage is found nowhere in HotSpot code.
1801 // The term "word" in HotSpot means a 32- or 64-bit machine word.
1802 int MacroAssembler::load_signed_short(Register dst, Address src) {
1803 // This is dubious to me since it seems safe to do a signed 16 => 64 bit
1804 // version but this is what 64bit has always done. This seems to imply
1805 // that users are only using 32bits worth.
1806 int off = offset();
1807 movswl(dst, src); // movsxw
1808 return off;
1809 }
1810
1811 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
1812 // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
1813 // and "3.9 Partial Register Penalties", p. 22).
1814 int off = offset();
1815 movzbl(dst, src); // movzxb
1816 return off;
1817 }
1818
1819 // Note: load_unsigned_short used to be called load_unsigned_word.
1820 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
1821 // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
1822 // and "3.9 Partial Register Penalties", p. 22).
1823 int off = offset();
1824 movzwl(dst, src); // movzxw
1825 return off;
1826 }
1827
1828 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2) {
1829 switch (size_in_bytes) {
1830 case 8: movq(dst, src); break;
1831 case 4: movl(dst, src); break;
1832 case 2: is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
1833 case 1: is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
1834 default: ShouldNotReachHere();
1835 }
1836 }
1837
1838 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2) {
1839 switch (size_in_bytes) {
1840 case 8: movq(dst, src); break;
1841 case 4: movl(dst, src); break;
1842 case 2: movw(dst, src); break;
1843 case 1: movb(dst, src); break;
1844 default: ShouldNotReachHere();
1845 }
1846 }
1847
1848 void MacroAssembler::mov32(AddressLiteral dst, Register src, Register rscratch) {
1849 assert(rscratch != noreg || always_reachable(dst), "missing");
1850
1851 if (reachable(dst)) {
1852 movl(as_Address(dst), src);
1853 } else {
1854 lea(rscratch, dst);
1855 movl(Address(rscratch, 0), src);
1856 }
1857 }
1858
1859 void MacroAssembler::mov32(Register dst, AddressLiteral src) {
1860 if (reachable(src)) {
1861 movl(dst, as_Address(src));
1862 } else {
1863 lea(dst, src);
1864 movl(dst, Address(dst, 0));
1865 }
1866 }
1867
1868 // C++ bool manipulation
1869
1870 void MacroAssembler::movbool(Register dst, Address src) {
1871 if(sizeof(bool) == 1)
1872 movb(dst, src);
1873 else if(sizeof(bool) == 2)
1874 movw(dst, src);
1875 else if(sizeof(bool) == 4)
1876 movl(dst, src);
1877 else
1878 // unsupported
1879 ShouldNotReachHere();
1880 }
1881
1882 void MacroAssembler::movbool(Address dst, bool boolconst) {
1883 if(sizeof(bool) == 1)
1884 movb(dst, (int) boolconst);
1885 else if(sizeof(bool) == 2)
1886 movw(dst, (int) boolconst);
1887 else if(sizeof(bool) == 4)
1888 movl(dst, (int) boolconst);
1889 else
1890 // unsupported
1891 ShouldNotReachHere();
1892 }
1893
1894 void MacroAssembler::movbool(Address dst, Register src) {
1895 if(sizeof(bool) == 1)
1896 movb(dst, src);
1897 else if(sizeof(bool) == 2)
1898 movw(dst, src);
1899 else if(sizeof(bool) == 4)
1900 movl(dst, src);
1901 else
1902 // unsupported
1903 ShouldNotReachHere();
1904 }
1905
1906 void MacroAssembler::movdl(XMMRegister dst, AddressLiteral src, Register rscratch) {
1907 assert(rscratch != noreg || always_reachable(src), "missing");
1908
1909 if (reachable(src)) {
1910 movdl(dst, as_Address(src));
1911 } else {
1912 lea(rscratch, src);
1913 movdl(dst, Address(rscratch, 0));
1914 }
1915 }
1916
1917 void MacroAssembler::movq(XMMRegister dst, AddressLiteral src, Register rscratch) {
1918 assert(rscratch != noreg || always_reachable(src), "missing");
1919
1920 if (reachable(src)) {
1921 movq(dst, as_Address(src));
1922 } else {
1923 lea(rscratch, src);
1924 movq(dst, Address(rscratch, 0));
1925 }
1926 }
1927
1928 void MacroAssembler::movdbl(XMMRegister dst, AddressLiteral src, Register rscratch) {
1929 assert(rscratch != noreg || always_reachable(src), "missing");
1930
1931 if (reachable(src)) {
1932 if (UseXmmLoadAndClearUpper) {
1933 movsd (dst, as_Address(src));
1934 } else {
1935 movlpd(dst, as_Address(src));
1936 }
1937 } else {
1938 lea(rscratch, src);
1939 if (UseXmmLoadAndClearUpper) {
1940 movsd (dst, Address(rscratch, 0));
1941 } else {
1942 movlpd(dst, Address(rscratch, 0));
1943 }
1944 }
1945 }
1946
1947 void MacroAssembler::movflt(XMMRegister dst, AddressLiteral src, Register rscratch) {
1948 assert(rscratch != noreg || always_reachable(src), "missing");
1949
1950 if (reachable(src)) {
1951 movss(dst, as_Address(src));
1952 } else {
1953 lea(rscratch, src);
1954 movss(dst, Address(rscratch, 0));
1955 }
1956 }
1957
1958 void MacroAssembler::movptr(Register dst, Register src) {
1959 movq(dst, src);
1960 }
1961
1962 void MacroAssembler::movptr(Register dst, Address src) {
1963 movq(dst, src);
1964 }
1965
1966 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
1967 void MacroAssembler::movptr(Register dst, intptr_t src) {
1968 if (is_uimm32(src)) {
1969 movl(dst, checked_cast<uint32_t>(src));
1970 } else if (is_simm32(src)) {
1971 movq(dst, checked_cast<int32_t>(src));
1972 } else {
1973 mov64(dst, src);
1974 }
1975 }
1976
1977 void MacroAssembler::movptr(Address dst, Register src) {
1978 movq(dst, src);
1979 }
1980
1981 void MacroAssembler::movptr(Address dst, int32_t src) {
1982 movslq(dst, src);
1983 }
1984
1985 void MacroAssembler::movdqu(Address dst, XMMRegister src) {
1986 assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
1987 Assembler::movdqu(dst, src);
1988 }
1989
1990 void MacroAssembler::movdqu(XMMRegister dst, Address src) {
1991 assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
1992 Assembler::movdqu(dst, src);
1993 }
1994
1995 void MacroAssembler::movdqu(XMMRegister dst, XMMRegister src) {
1996 assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
1997 Assembler::movdqu(dst, src);
1998 }
1999
2000 void MacroAssembler::movdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
2001 assert(rscratch != noreg || always_reachable(src), "missing");
2002
2003 if (reachable(src)) {
2004 movdqu(dst, as_Address(src));
2005 } else {
2006 lea(rscratch, src);
2007 movdqu(dst, Address(rscratch, 0));
2008 }
2009 }
2010
2011 void MacroAssembler::vmovdqu(Address dst, XMMRegister src) {
2012 assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
2013 Assembler::vmovdqu(dst, src);
2014 }
2015
2016 void MacroAssembler::vmovdqu(XMMRegister dst, Address src) {
2017 assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
2018 Assembler::vmovdqu(dst, src);
2019 }
2020
2021 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src) {
2022 assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
2023 Assembler::vmovdqu(dst, src);
2024 }
2025
2026 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
2027 assert(rscratch != noreg || always_reachable(src), "missing");
2028
2029 if (reachable(src)) {
2030 vmovdqu(dst, as_Address(src));
2031 }
2032 else {
2033 lea(rscratch, src);
2034 vmovdqu(dst, Address(rscratch, 0));
2035 }
2036 }
2037
2038 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
2039 assert(rscratch != noreg || always_reachable(src), "missing");
2040
2041 if (vector_len == AVX_512bit) {
2042 evmovdquq(dst, src, AVX_512bit, rscratch);
2043 } else if (vector_len == AVX_256bit) {
2044 vmovdqu(dst, src, rscratch);
2045 } else {
2046 movdqu(dst, src, rscratch);
2047 }
2048 }
2049
2050 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src, int vector_len) {
2051 if (vector_len == AVX_512bit) {
2052 evmovdquq(dst, src, AVX_512bit);
2053 } else if (vector_len == AVX_256bit) {
2054 vmovdqu(dst, src);
2055 } else {
2056 movdqu(dst, src);
2057 }
2058 }
2059
2060 void MacroAssembler::vmovdqu(Address dst, XMMRegister src, int vector_len) {
2061 if (vector_len == AVX_512bit) {
2062 evmovdquq(dst, src, AVX_512bit);
2063 } else if (vector_len == AVX_256bit) {
2064 vmovdqu(dst, src);
2065 } else {
2066 movdqu(dst, src);
2067 }
2068 }
2069
2070 void MacroAssembler::vmovdqu(XMMRegister dst, Address src, int vector_len) {
2071 if (vector_len == AVX_512bit) {
2072 evmovdquq(dst, src, AVX_512bit);
2073 } else if (vector_len == AVX_256bit) {
2074 vmovdqu(dst, src);
2075 } else {
2076 movdqu(dst, src);
2077 }
2078 }
2079
2080 void MacroAssembler::vmovdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
2081 assert(rscratch != noreg || always_reachable(src), "missing");
2082
2083 if (reachable(src)) {
2084 vmovdqa(dst, as_Address(src));
2085 }
2086 else {
2087 lea(rscratch, src);
2088 vmovdqa(dst, Address(rscratch, 0));
2089 }
2090 }
2091
2092 void MacroAssembler::vmovdqa(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
2093 assert(rscratch != noreg || always_reachable(src), "missing");
2094
2095 if (vector_len == AVX_512bit) {
2096 evmovdqaq(dst, src, AVX_512bit, rscratch);
2097 } else if (vector_len == AVX_256bit) {
2098 vmovdqa(dst, src, rscratch);
2099 } else {
2100 movdqa(dst, src, rscratch);
2101 }
2102 }
2103
2104 void MacroAssembler::kmov(KRegister dst, Address src) {
2105 if (VM_Version::supports_avx512bw()) {
2106 kmovql(dst, src);
2107 } else {
2108 assert(VM_Version::supports_evex(), "");
2109 kmovwl(dst, src);
2110 }
2111 }
2112
2113 void MacroAssembler::kmov(Address dst, KRegister src) {
2114 if (VM_Version::supports_avx512bw()) {
2115 kmovql(dst, src);
2116 } else {
2117 assert(VM_Version::supports_evex(), "");
2118 kmovwl(dst, src);
2119 }
2120 }
2121
2122 void MacroAssembler::kmov(KRegister dst, KRegister src) {
2123 if (VM_Version::supports_avx512bw()) {
2124 kmovql(dst, src);
2125 } else {
2126 assert(VM_Version::supports_evex(), "");
2127 kmovwl(dst, src);
2128 }
2129 }
2130
2131 void MacroAssembler::kmov(Register dst, KRegister src) {
2132 if (VM_Version::supports_avx512bw()) {
2133 kmovql(dst, src);
2134 } else {
2135 assert(VM_Version::supports_evex(), "");
2136 kmovwl(dst, src);
2137 }
2138 }
2139
2140 void MacroAssembler::kmov(KRegister dst, Register src) {
2141 if (VM_Version::supports_avx512bw()) {
2142 kmovql(dst, src);
2143 } else {
2144 assert(VM_Version::supports_evex(), "");
2145 kmovwl(dst, src);
2146 }
2147 }
2148
2149 void MacroAssembler::kmovql(KRegister dst, AddressLiteral src, Register rscratch) {
2150 assert(rscratch != noreg || always_reachable(src), "missing");
2151
2152 if (reachable(src)) {
2153 kmovql(dst, as_Address(src));
2154 } else {
2155 lea(rscratch, src);
2156 kmovql(dst, Address(rscratch, 0));
2157 }
2158 }
2159
2160 void MacroAssembler::kmovwl(KRegister dst, AddressLiteral src, Register rscratch) {
2161 assert(rscratch != noreg || always_reachable(src), "missing");
2162
2163 if (reachable(src)) {
2164 kmovwl(dst, as_Address(src));
2165 } else {
2166 lea(rscratch, src);
2167 kmovwl(dst, Address(rscratch, 0));
2168 }
2169 }
2170
2171 void MacroAssembler::evmovdqub(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
2172 int vector_len, Register rscratch) {
2173 assert(rscratch != noreg || always_reachable(src), "missing");
2174
2175 if (reachable(src)) {
2176 Assembler::evmovdqub(dst, mask, as_Address(src), merge, vector_len);
2177 } else {
2178 lea(rscratch, src);
2179 Assembler::evmovdqub(dst, mask, Address(rscratch, 0), merge, vector_len);
2180 }
2181 }
2182
2183 void MacroAssembler::evmovdquw(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
2184 int vector_len, Register rscratch) {
2185 assert(rscratch != noreg || always_reachable(src), "missing");
2186
2187 if (reachable(src)) {
2188 Assembler::evmovdquw(dst, mask, as_Address(src), merge, vector_len);
2189 } else {
2190 lea(rscratch, src);
2191 Assembler::evmovdquw(dst, mask, Address(rscratch, 0), merge, vector_len);
2192 }
2193 }
2194
2195 void MacroAssembler::evmovdqul(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
2196 assert(rscratch != noreg || always_reachable(src), "missing");
2197
2198 if (reachable(src)) {
2199 Assembler::evmovdqul(dst, mask, as_Address(src), merge, vector_len);
2200 } else {
2201 lea(rscratch, src);
2202 Assembler::evmovdqul(dst, mask, Address(rscratch, 0), merge, vector_len);
2203 }
2204 }
2205
2206 void MacroAssembler::evmovdquq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
2207 assert(rscratch != noreg || always_reachable(src), "missing");
2208
2209 if (reachable(src)) {
2210 Assembler::evmovdquq(dst, mask, as_Address(src), merge, vector_len);
2211 } else {
2212 lea(rscratch, src);
2213 Assembler::evmovdquq(dst, mask, Address(rscratch, 0), merge, vector_len);
2214 }
2215 }
2216
2217 void MacroAssembler::evmovdquq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
2218 assert(rscratch != noreg || always_reachable(src), "missing");
2219
2220 if (reachable(src)) {
2221 Assembler::evmovdquq(dst, as_Address(src), vector_len);
2222 } else {
2223 lea(rscratch, src);
2224 Assembler::evmovdquq(dst, Address(rscratch, 0), vector_len);
2225 }
2226 }
2227
2228 void MacroAssembler::evmovdqaq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
2229 assert(rscratch != noreg || always_reachable(src), "missing");
2230
2231 if (reachable(src)) {
2232 Assembler::evmovdqaq(dst, mask, as_Address(src), merge, vector_len);
2233 } else {
2234 lea(rscratch, src);
2235 Assembler::evmovdqaq(dst, mask, Address(rscratch, 0), merge, vector_len);
2236 }
2237 }
2238
2239 void MacroAssembler::evmovdqaq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
2240 assert(rscratch != noreg || always_reachable(src), "missing");
2241
2242 if (reachable(src)) {
2243 Assembler::evmovdqaq(dst, as_Address(src), vector_len);
2244 } else {
2245 lea(rscratch, src);
2246 Assembler::evmovdqaq(dst, Address(rscratch, 0), vector_len);
2247 }
2248 }
2249
2250
2251 void MacroAssembler::movdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
2252 assert(rscratch != noreg || always_reachable(src), "missing");
2253
2254 if (reachable(src)) {
2255 Assembler::movdqa(dst, as_Address(src));
2256 } else {
2257 lea(rscratch, src);
2258 Assembler::movdqa(dst, Address(rscratch, 0));
2259 }
2260 }
2261
2262 void MacroAssembler::movsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
2263 assert(rscratch != noreg || always_reachable(src), "missing");
2264
2265 if (reachable(src)) {
2266 Assembler::movsd(dst, as_Address(src));
2267 } else {
2268 lea(rscratch, src);
2269 Assembler::movsd(dst, Address(rscratch, 0));
2270 }
2271 }
2272
2273 void MacroAssembler::movss(XMMRegister dst, AddressLiteral src, Register rscratch) {
2274 assert(rscratch != noreg || always_reachable(src), "missing");
2275
2276 if (reachable(src)) {
2277 Assembler::movss(dst, as_Address(src));
2278 } else {
2279 lea(rscratch, src);
2280 Assembler::movss(dst, Address(rscratch, 0));
2281 }
2282 }
2283
2284 void MacroAssembler::movddup(XMMRegister dst, AddressLiteral src, Register rscratch) {
2285 assert(rscratch != noreg || always_reachable(src), "missing");
2286
2287 if (reachable(src)) {
2288 Assembler::movddup(dst, as_Address(src));
2289 } else {
2290 lea(rscratch, src);
2291 Assembler::movddup(dst, Address(rscratch, 0));
2292 }
2293 }
2294
2295 void MacroAssembler::vmovddup(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
2296 assert(rscratch != noreg || always_reachable(src), "missing");
2297
2298 if (reachable(src)) {
2299 Assembler::vmovddup(dst, as_Address(src), vector_len);
2300 } else {
2301 lea(rscratch, src);
2302 Assembler::vmovddup(dst, Address(rscratch, 0), vector_len);
2303 }
2304 }
2305
2306 void MacroAssembler::mulsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
2307 assert(rscratch != noreg || always_reachable(src), "missing");
2308
2309 if (reachable(src)) {
2310 Assembler::mulsd(dst, as_Address(src));
2311 } else {
2312 lea(rscratch, src);
2313 Assembler::mulsd(dst, Address(rscratch, 0));
2314 }
2315 }
2316
2317 void MacroAssembler::mulss(XMMRegister dst, AddressLiteral src, Register rscratch) {
2318 assert(rscratch != noreg || always_reachable(src), "missing");
2319
2320 if (reachable(src)) {
2321 Assembler::mulss(dst, as_Address(src));
2322 } else {
2323 lea(rscratch, src);
2324 Assembler::mulss(dst, Address(rscratch, 0));
2325 }
2326 }
2327
2328 void MacroAssembler::null_check(Register reg, int offset) {
2329 if (needs_explicit_null_check(offset)) {
2330 // provoke OS null exception if reg is null by
2331 // accessing M[reg] w/o changing any (non-CC) registers
2332 // NOTE: cmpl is plenty here to provoke a segv
2333 cmpptr(rax, Address(reg, 0));
2334 // Note: should probably use testl(rax, Address(reg, 0));
2335 // may be shorter code (however, this version of
2336 // testl needs to be implemented first)
2337 } else {
2338 // nothing to do, (later) access of M[reg + offset]
2339 // will provoke OS null exception if reg is null
2340 }
2341 }
2342
2343 void MacroAssembler::os_breakpoint() {
2344 // instead of directly emitting a breakpoint, call os:breakpoint for better debugability
2345 // (e.g., MSVC can't call ps() otherwise)
2346 call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint)));
2347 }
2348
2349 void MacroAssembler::unimplemented(const char* what) {
2350 const char* buf = nullptr;
2351 {
2352 ResourceMark rm;
2353 stringStream ss;
2354 ss.print("unimplemented: %s", what);
2355 buf = code_string(ss.as_string());
2356 }
2357 stop(buf);
2358 }
2359
2360 #define XSTATE_BV 0x200
2361
2362 void MacroAssembler::pop_CPU_state() {
2363 pop_FPU_state();
2364 pop_IU_state();
2365 }
2366
2367 void MacroAssembler::pop_FPU_state() {
2368 fxrstor(Address(rsp, 0));
2369 addptr(rsp, FPUStateSizeInWords * wordSize);
2370 }
2371
2372 void MacroAssembler::pop_IU_state() {
2373 popa();
2374 addq(rsp, 8);
2375 popf();
2376 }
2377
2378 // Save Integer and Float state
2379 // Warning: Stack must be 16 byte aligned (64bit)
2380 void MacroAssembler::push_CPU_state() {
2381 push_IU_state();
2382 push_FPU_state();
2383 }
2384
2385 void MacroAssembler::push_FPU_state() {
2386 subptr(rsp, FPUStateSizeInWords * wordSize);
2387 fxsave(Address(rsp, 0));
2388 }
2389
2390 void MacroAssembler::push_IU_state() {
2391 // Push flags first because pusha kills them
2392 pushf();
2393 // Make sure rsp stays 16-byte aligned
2394 subq(rsp, 8);
2395 pusha();
2396 }
2397
2398 void MacroAssembler::push_cont_fastpath() {
2399 if (!Continuations::enabled()) return;
2400
2401 Label L_done;
2402 cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
2403 jccb(Assembler::belowEqual, L_done);
2404 movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), rsp);
2405 bind(L_done);
2406 }
2407
2408 void MacroAssembler::pop_cont_fastpath() {
2409 if (!Continuations::enabled()) return;
2410
2411 Label L_done;
2412 cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
2413 jccb(Assembler::below, L_done);
2414 movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), 0);
2415 bind(L_done);
2416 }
2417
2418 void MacroAssembler::inc_held_monitor_count() {
2419 incrementq(Address(r15_thread, JavaThread::held_monitor_count_offset()));
2420 }
2421
2422 void MacroAssembler::dec_held_monitor_count() {
2423 decrementq(Address(r15_thread, JavaThread::held_monitor_count_offset()));
2424 }
2425
2426 #ifdef ASSERT
2427 void MacroAssembler::stop_if_in_cont(Register cont, const char* name) {
2428 Label no_cont;
2429 movptr(cont, Address(r15_thread, JavaThread::cont_entry_offset()));
2430 testl(cont, cont);
2431 jcc(Assembler::zero, no_cont);
2432 stop(name);
2433 bind(no_cont);
2434 }
2435 #endif
2436
2437 void MacroAssembler::reset_last_Java_frame(bool clear_fp) { // determine java_thread register
2438 // we must set sp to zero to clear frame
2439 movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), NULL_WORD);
2440 // must clear fp, so that compiled frames are not confused; it is
2441 // possible that we need it only for debugging
2442 if (clear_fp) {
2443 movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), NULL_WORD);
2444 }
2445 // Always clear the pc because it could have been set by make_walkable()
2446 movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), NULL_WORD);
2447 vzeroupper();
2448 }
2449
2450 void MacroAssembler::round_to(Register reg, int modulus) {
2451 addptr(reg, modulus - 1);
2452 andptr(reg, -modulus);
2453 }
2454
2455 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod) {
2456 if (at_return) {
2457 // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
2458 // we may safely use rsp instead to perform the stack watermark check.
2459 cmpptr(in_nmethod ? rsp : rbp, Address(r15_thread, JavaThread::polling_word_offset()));
2460 jcc(Assembler::above, slow_path);
2461 return;
2462 }
2463 testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
2464 jcc(Assembler::notZero, slow_path); // handshake bit set implies poll
2465 }
2466
2467 // Calls to C land
2468 //
2469 // When entering C land, the rbp, & rsp of the last Java frame have to be recorded
2470 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
2471 // has to be reset to 0. This is required to allow proper stack traversal.
2472 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
2473 Register last_java_fp,
2474 address last_java_pc,
2475 Register rscratch) {
2476 vzeroupper();
2477 // determine last_java_sp register
2478 if (!last_java_sp->is_valid()) {
2479 last_java_sp = rsp;
2480 }
2481 // last_java_fp is optional
2482 if (last_java_fp->is_valid()) {
2483 movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), last_java_fp);
2484 }
2485 // last_java_pc is optional
2486 if (last_java_pc != nullptr) {
2487 Address java_pc(r15_thread,
2488 JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset());
2489 lea(java_pc, InternalAddress(last_java_pc), rscratch);
2490 }
2491 movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), last_java_sp);
2492 }
2493
2494 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
2495 Register last_java_fp,
2496 Label &L,
2497 Register scratch) {
2498 lea(scratch, L);
2499 movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), scratch);
2500 set_last_Java_frame(last_java_sp, last_java_fp, nullptr, scratch);
2501 }
2502
2503 void MacroAssembler::shlptr(Register dst, int imm8) {
2504 shlq(dst, imm8);
2505 }
2506
2507 void MacroAssembler::shrptr(Register dst, int imm8) {
2508 shrq(dst, imm8);
2509 }
2510
2511 void MacroAssembler::sign_extend_byte(Register reg) {
2512 movsbl(reg, reg); // movsxb
2513 }
2514
2515 void MacroAssembler::sign_extend_short(Register reg) {
2516 movswl(reg, reg); // movsxw
2517 }
2518
2519 void MacroAssembler::testl(Address dst, int32_t imm32) {
2520 if (imm32 >= 0 && is8bit(imm32)) {
2521 testb(dst, imm32);
2522 } else {
2523 Assembler::testl(dst, imm32);
2524 }
2525 }
2526
2527 void MacroAssembler::testl(Register dst, int32_t imm32) {
2528 if (imm32 >= 0 && is8bit(imm32) && dst->has_byte_register()) {
2529 testb(dst, imm32);
2530 } else {
2531 Assembler::testl(dst, imm32);
2532 }
2533 }
2534
2535 void MacroAssembler::testl(Register dst, AddressLiteral src) {
2536 assert(always_reachable(src), "Address should be reachable");
2537 testl(dst, as_Address(src));
2538 }
2539
2540 void MacroAssembler::testq(Address dst, int32_t imm32) {
2541 if (imm32 >= 0) {
2542 testl(dst, imm32);
2543 } else {
2544 Assembler::testq(dst, imm32);
2545 }
2546 }
2547
2548 void MacroAssembler::testq(Register dst, int32_t imm32) {
2549 if (imm32 >= 0) {
2550 testl(dst, imm32);
2551 } else {
2552 Assembler::testq(dst, imm32);
2553 }
2554 }
2555
2556 void MacroAssembler::pcmpeqb(XMMRegister dst, XMMRegister src) {
2557 assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2558 Assembler::pcmpeqb(dst, src);
2559 }
2560
2561 void MacroAssembler::pcmpeqw(XMMRegister dst, XMMRegister src) {
2562 assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2563 Assembler::pcmpeqw(dst, src);
2564 }
2565
2566 void MacroAssembler::pcmpestri(XMMRegister dst, Address src, int imm8) {
2567 assert((dst->encoding() < 16),"XMM register should be 0-15");
2568 Assembler::pcmpestri(dst, src, imm8);
2569 }
2570
2571 void MacroAssembler::pcmpestri(XMMRegister dst, XMMRegister src, int imm8) {
2572 assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
2573 Assembler::pcmpestri(dst, src, imm8);
2574 }
2575
2576 void MacroAssembler::pmovzxbw(XMMRegister dst, XMMRegister src) {
2577 assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2578 Assembler::pmovzxbw(dst, src);
2579 }
2580
2581 void MacroAssembler::pmovzxbw(XMMRegister dst, Address src) {
2582 assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2583 Assembler::pmovzxbw(dst, src);
2584 }
2585
2586 void MacroAssembler::pmovmskb(Register dst, XMMRegister src) {
2587 assert((src->encoding() < 16),"XMM register should be 0-15");
2588 Assembler::pmovmskb(dst, src);
2589 }
2590
2591 void MacroAssembler::ptest(XMMRegister dst, XMMRegister src) {
2592 assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
2593 Assembler::ptest(dst, src);
2594 }
2595
2596 void MacroAssembler::sqrtss(XMMRegister dst, AddressLiteral src, Register rscratch) {
2597 assert(rscratch != noreg || always_reachable(src), "missing");
2598
2599 if (reachable(src)) {
2600 Assembler::sqrtss(dst, as_Address(src));
2601 } else {
2602 lea(rscratch, src);
2603 Assembler::sqrtss(dst, Address(rscratch, 0));
2604 }
2605 }
2606
2607 void MacroAssembler::subsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
2608 assert(rscratch != noreg || always_reachable(src), "missing");
2609
2610 if (reachable(src)) {
2611 Assembler::subsd(dst, as_Address(src));
2612 } else {
2613 lea(rscratch, src);
2614 Assembler::subsd(dst, Address(rscratch, 0));
2615 }
2616 }
2617
2618 void MacroAssembler::roundsd(XMMRegister dst, AddressLiteral src, int32_t rmode, Register rscratch) {
2619 assert(rscratch != noreg || always_reachable(src), "missing");
2620
2621 if (reachable(src)) {
2622 Assembler::roundsd(dst, as_Address(src), rmode);
2623 } else {
2624 lea(rscratch, src);
2625 Assembler::roundsd(dst, Address(rscratch, 0), rmode);
2626 }
2627 }
2628
2629 void MacroAssembler::subss(XMMRegister dst, AddressLiteral src, Register rscratch) {
2630 assert(rscratch != noreg || always_reachable(src), "missing");
2631
2632 if (reachable(src)) {
2633 Assembler::subss(dst, as_Address(src));
2634 } else {
2635 lea(rscratch, src);
2636 Assembler::subss(dst, Address(rscratch, 0));
2637 }
2638 }
2639
2640 void MacroAssembler::ucomisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
2641 assert(rscratch != noreg || always_reachable(src), "missing");
2642
2643 if (reachable(src)) {
2644 Assembler::ucomisd(dst, as_Address(src));
2645 } else {
2646 lea(rscratch, src);
2647 Assembler::ucomisd(dst, Address(rscratch, 0));
2648 }
2649 }
2650
2651 void MacroAssembler::ucomiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
2652 assert(rscratch != noreg || always_reachable(src), "missing");
2653
2654 if (reachable(src)) {
2655 Assembler::ucomiss(dst, as_Address(src));
2656 } else {
2657 lea(rscratch, src);
2658 Assembler::ucomiss(dst, Address(rscratch, 0));
2659 }
2660 }
2661
2662 void MacroAssembler::xorpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
2663 assert(rscratch != noreg || always_reachable(src), "missing");
2664
2665 // Used in sign-bit flipping with aligned address.
2666 assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
2667
2668 if (UseAVX > 2 &&
2669 (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
2670 (dst->encoding() >= 16)) {
2671 vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
2672 } else if (reachable(src)) {
2673 Assembler::xorpd(dst, as_Address(src));
2674 } else {
2675 lea(rscratch, src);
2676 Assembler::xorpd(dst, Address(rscratch, 0));
2677 }
2678 }
2679
2680 void MacroAssembler::xorpd(XMMRegister dst, XMMRegister src) {
2681 if (UseAVX > 2 &&
2682 (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
2683 ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
2684 Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
2685 } else {
2686 Assembler::xorpd(dst, src);
2687 }
2688 }
2689
2690 void MacroAssembler::xorps(XMMRegister dst, XMMRegister src) {
2691 if (UseAVX > 2 &&
2692 (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
2693 ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
2694 Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
2695 } else {
2696 Assembler::xorps(dst, src);
2697 }
2698 }
2699
2700 void MacroAssembler::xorps(XMMRegister dst, AddressLiteral src, Register rscratch) {
2701 assert(rscratch != noreg || always_reachable(src), "missing");
2702
2703 // Used in sign-bit flipping with aligned address.
2704 assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
2705
2706 if (UseAVX > 2 &&
2707 (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
2708 (dst->encoding() >= 16)) {
2709 vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
2710 } else if (reachable(src)) {
2711 Assembler::xorps(dst, as_Address(src));
2712 } else {
2713 lea(rscratch, src);
2714 Assembler::xorps(dst, Address(rscratch, 0));
2715 }
2716 }
2717
2718 void MacroAssembler::pshufb(XMMRegister dst, AddressLiteral src, Register rscratch) {
2719 assert(rscratch != noreg || always_reachable(src), "missing");
2720
2721 // Used in sign-bit flipping with aligned address.
2722 bool aligned_adr = (((intptr_t)src.target() & 15) == 0);
2723 assert((UseAVX > 0) || aligned_adr, "SSE mode requires address alignment 16 bytes");
2724 if (reachable(src)) {
2725 Assembler::pshufb(dst, as_Address(src));
2726 } else {
2727 lea(rscratch, src);
2728 Assembler::pshufb(dst, Address(rscratch, 0));
2729 }
2730 }
2731
2732 // AVX 3-operands instructions
2733
2734 void MacroAssembler::vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
2735 assert(rscratch != noreg || always_reachable(src), "missing");
2736
2737 if (reachable(src)) {
2738 vaddsd(dst, nds, as_Address(src));
2739 } else {
2740 lea(rscratch, src);
2741 vaddsd(dst, nds, Address(rscratch, 0));
2742 }
2743 }
2744
2745 void MacroAssembler::vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
2746 assert(rscratch != noreg || always_reachable(src), "missing");
2747
2748 if (reachable(src)) {
2749 vaddss(dst, nds, as_Address(src));
2750 } else {
2751 lea(rscratch, src);
2752 vaddss(dst, nds, Address(rscratch, 0));
2753 }
2754 }
2755
2756 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
2757 assert(UseAVX > 0, "requires some form of AVX");
2758 assert(rscratch != noreg || always_reachable(src), "missing");
2759
2760 if (reachable(src)) {
2761 Assembler::vpaddb(dst, nds, as_Address(src), vector_len);
2762 } else {
2763 lea(rscratch, src);
2764 Assembler::vpaddb(dst, nds, Address(rscratch, 0), vector_len);
2765 }
2766 }
2767
2768 void MacroAssembler::vpaddd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
2769 assert(UseAVX > 0, "requires some form of AVX");
2770 assert(rscratch != noreg || always_reachable(src), "missing");
2771
2772 if (reachable(src)) {
2773 Assembler::vpaddd(dst, nds, as_Address(src), vector_len);
2774 } else {
2775 lea(rscratch, src);
2776 Assembler::vpaddd(dst, nds, Address(rscratch, 0), vector_len);
2777 }
2778 }
2779
2780 void MacroAssembler::vabsss(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
2781 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
2782 assert(rscratch != noreg || always_reachable(negate_field), "missing");
2783
2784 vandps(dst, nds, negate_field, vector_len, rscratch);
2785 }
2786
2787 void MacroAssembler::vabssd(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
2788 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
2789 assert(rscratch != noreg || always_reachable(negate_field), "missing");
2790
2791 vandpd(dst, nds, negate_field, vector_len, rscratch);
2792 }
2793
2794 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
2795 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2796 Assembler::vpaddb(dst, nds, src, vector_len);
2797 }
2798
2799 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
2800 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2801 Assembler::vpaddb(dst, nds, src, vector_len);
2802 }
2803
2804 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
2805 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2806 Assembler::vpaddw(dst, nds, src, vector_len);
2807 }
2808
2809 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
2810 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2811 Assembler::vpaddw(dst, nds, src, vector_len);
2812 }
2813
2814 void MacroAssembler::vpand(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
2815 assert(rscratch != noreg || always_reachable(src), "missing");
2816
2817 if (reachable(src)) {
2818 Assembler::vpand(dst, nds, as_Address(src), vector_len);
2819 } else {
2820 lea(rscratch, src);
2821 Assembler::vpand(dst, nds, Address(rscratch, 0), vector_len);
2822 }
2823 }
2824
2825 void MacroAssembler::vpbroadcastd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
2826 assert(rscratch != noreg || always_reachable(src), "missing");
2827
2828 if (reachable(src)) {
2829 Assembler::vpbroadcastd(dst, as_Address(src), vector_len);
2830 } else {
2831 lea(rscratch, src);
2832 Assembler::vpbroadcastd(dst, Address(rscratch, 0), vector_len);
2833 }
2834 }
2835
2836 void MacroAssembler::vbroadcasti128(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
2837 assert(rscratch != noreg || always_reachable(src), "missing");
2838
2839 if (reachable(src)) {
2840 Assembler::vbroadcasti128(dst, as_Address(src), vector_len);
2841 } else {
2842 lea(rscratch, src);
2843 Assembler::vbroadcasti128(dst, Address(rscratch, 0), vector_len);
2844 }
2845 }
2846
2847 void MacroAssembler::vpbroadcastq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
2848 assert(rscratch != noreg || always_reachable(src), "missing");
2849
2850 if (reachable(src)) {
2851 Assembler::vpbroadcastq(dst, as_Address(src), vector_len);
2852 } else {
2853 lea(rscratch, src);
2854 Assembler::vpbroadcastq(dst, Address(rscratch, 0), vector_len);
2855 }
2856 }
2857
2858 void MacroAssembler::vbroadcastsd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
2859 assert(rscratch != noreg || always_reachable(src), "missing");
2860
2861 if (reachable(src)) {
2862 Assembler::vbroadcastsd(dst, as_Address(src), vector_len);
2863 } else {
2864 lea(rscratch, src);
2865 Assembler::vbroadcastsd(dst, Address(rscratch, 0), vector_len);
2866 }
2867 }
2868
2869 void MacroAssembler::vbroadcastss(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
2870 assert(rscratch != noreg || always_reachable(src), "missing");
2871
2872 if (reachable(src)) {
2873 Assembler::vbroadcastss(dst, as_Address(src), vector_len);
2874 } else {
2875 lea(rscratch, src);
2876 Assembler::vbroadcastss(dst, Address(rscratch, 0), vector_len);
2877 }
2878 }
2879
2880 // Vector float blend
2881 // vblendvps(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
2882 void MacroAssembler::vblendvps(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
2883 // WARN: Allow dst == (src1|src2), mask == scratch
2884 bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1;
2885 bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst;
2886 bool dst_available = dst != mask && (dst != src1 || dst != src2);
2887 if (blend_emulation && scratch_available && dst_available) {
2888 if (compute_mask) {
2889 vpsrad(scratch, mask, 32, vector_len);
2890 mask = scratch;
2891 }
2892 if (dst == src1) {
2893 vpandn(dst, mask, src1, vector_len); // if mask == 0, src1
2894 vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
2895 } else {
2896 vpand (dst, mask, src2, vector_len); // if mask == 1, src2
2897 vpandn(scratch, mask, src1, vector_len); // if mask == 0, src1
2898 }
2899 vpor(dst, dst, scratch, vector_len);
2900 } else {
2901 Assembler::vblendvps(dst, src1, src2, mask, vector_len);
2902 }
2903 }
2904
2905 // vblendvpd(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
2906 void MacroAssembler::vblendvpd(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
2907 // WARN: Allow dst == (src1|src2), mask == scratch
2908 bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1;
2909 bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst && (!compute_mask || scratch != mask);
2910 bool dst_available = dst != mask && (dst != src1 || dst != src2);
2911 if (blend_emulation && scratch_available && dst_available) {
2912 if (compute_mask) {
2913 vpxor(scratch, scratch, scratch, vector_len);
2914 vpcmpgtq(scratch, scratch, mask, vector_len);
2915 mask = scratch;
2916 }
2917 if (dst == src1) {
2918 vpandn(dst, mask, src1, vector_len); // if mask == 0, src
2919 vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
2920 } else {
2921 vpand (dst, mask, src2, vector_len); // if mask == 1, src2
2922 vpandn(scratch, mask, src1, vector_len); // if mask == 0, src
2923 }
2924 vpor(dst, dst, scratch, vector_len);
2925 } else {
2926 Assembler::vblendvpd(dst, src1, src2, mask, vector_len);
2927 }
2928 }
2929
2930 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
2931 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2932 Assembler::vpcmpeqb(dst, nds, src, vector_len);
2933 }
2934
2935 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister src1, Address src2, int vector_len) {
2936 assert(((dst->encoding() < 16 && src1->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2937 Assembler::vpcmpeqb(dst, src1, src2, vector_len);
2938 }
2939
2940 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
2941 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2942 Assembler::vpcmpeqw(dst, nds, src, vector_len);
2943 }
2944
2945 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
2946 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2947 Assembler::vpcmpeqw(dst, nds, src, vector_len);
2948 }
2949
2950 void MacroAssembler::evpcmpeqd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
2951 assert(rscratch != noreg || always_reachable(src), "missing");
2952
2953 if (reachable(src)) {
2954 Assembler::evpcmpeqd(kdst, mask, nds, as_Address(src), vector_len);
2955 } else {
2956 lea(rscratch, src);
2957 Assembler::evpcmpeqd(kdst, mask, nds, Address(rscratch, 0), vector_len);
2958 }
2959 }
2960
2961 void MacroAssembler::evpcmpd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
2962 int comparison, bool is_signed, int vector_len, Register rscratch) {
2963 assert(rscratch != noreg || always_reachable(src), "missing");
2964
2965 if (reachable(src)) {
2966 Assembler::evpcmpd(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
2967 } else {
2968 lea(rscratch, src);
2969 Assembler::evpcmpd(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
2970 }
2971 }
2972
2973 void MacroAssembler::evpcmpq(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
2974 int comparison, bool is_signed, int vector_len, Register rscratch) {
2975 assert(rscratch != noreg || always_reachable(src), "missing");
2976
2977 if (reachable(src)) {
2978 Assembler::evpcmpq(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
2979 } else {
2980 lea(rscratch, src);
2981 Assembler::evpcmpq(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
2982 }
2983 }
2984
2985 void MacroAssembler::evpcmpb(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
2986 int comparison, bool is_signed, int vector_len, Register rscratch) {
2987 assert(rscratch != noreg || always_reachable(src), "missing");
2988
2989 if (reachable(src)) {
2990 Assembler::evpcmpb(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
2991 } else {
2992 lea(rscratch, src);
2993 Assembler::evpcmpb(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
2994 }
2995 }
2996
2997 void MacroAssembler::evpcmpw(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
2998 int comparison, bool is_signed, int vector_len, Register rscratch) {
2999 assert(rscratch != noreg || always_reachable(src), "missing");
3000
3001 if (reachable(src)) {
3002 Assembler::evpcmpw(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
3003 } else {
3004 lea(rscratch, src);
3005 Assembler::evpcmpw(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
3006 }
3007 }
3008
3009 void MacroAssembler::vpcmpCC(XMMRegister dst, XMMRegister nds, XMMRegister src, int cond_encoding, Width width, int vector_len) {
3010 if (width == Assembler::Q) {
3011 Assembler::vpcmpCCq(dst, nds, src, cond_encoding, vector_len);
3012 } else {
3013 Assembler::vpcmpCCbwd(dst, nds, src, cond_encoding, vector_len);
3014 }
3015 }
3016
3017 void MacroAssembler::vpcmpCCW(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister xtmp, ComparisonPredicate cond, Width width, int vector_len) {
3018 int eq_cond_enc = 0x29;
3019 int gt_cond_enc = 0x37;
3020 if (width != Assembler::Q) {
3021 eq_cond_enc = 0x74 + width;
3022 gt_cond_enc = 0x64 + width;
3023 }
3024 switch (cond) {
3025 case eq:
3026 vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
3027 break;
3028 case neq:
3029 vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
3030 vallones(xtmp, vector_len);
3031 vpxor(dst, xtmp, dst, vector_len);
3032 break;
3033 case le:
3034 vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
3035 vallones(xtmp, vector_len);
3036 vpxor(dst, xtmp, dst, vector_len);
3037 break;
3038 case nlt:
3039 vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
3040 vallones(xtmp, vector_len);
3041 vpxor(dst, xtmp, dst, vector_len);
3042 break;
3043 case lt:
3044 vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
3045 break;
3046 case nle:
3047 vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
3048 break;
3049 default:
3050 assert(false, "Should not reach here");
3051 }
3052 }
3053
3054 void MacroAssembler::vpmovzxbw(XMMRegister dst, Address src, int vector_len) {
3055 assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3056 Assembler::vpmovzxbw(dst, src, vector_len);
3057 }
3058
3059 void MacroAssembler::vpmovmskb(Register dst, XMMRegister src, int vector_len) {
3060 assert((src->encoding() < 16),"XMM register should be 0-15");
3061 Assembler::vpmovmskb(dst, src, vector_len);
3062 }
3063
3064 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
3065 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3066 Assembler::vpmullw(dst, nds, src, vector_len);
3067 }
3068
3069 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
3070 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3071 Assembler::vpmullw(dst, nds, src, vector_len);
3072 }
3073
3074 void MacroAssembler::vpmulld(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3075 assert((UseAVX > 0), "AVX support is needed");
3076 assert(rscratch != noreg || always_reachable(src), "missing");
3077
3078 if (reachable(src)) {
3079 Assembler::vpmulld(dst, nds, as_Address(src), vector_len);
3080 } else {
3081 lea(rscratch, src);
3082 Assembler::vpmulld(dst, nds, Address(rscratch, 0), vector_len);
3083 }
3084 }
3085
3086 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
3087 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3088 Assembler::vpsubb(dst, nds, src, vector_len);
3089 }
3090
3091 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
3092 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3093 Assembler::vpsubb(dst, nds, src, vector_len);
3094 }
3095
3096 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
3097 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3098 Assembler::vpsubw(dst, nds, src, vector_len);
3099 }
3100
3101 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
3102 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3103 Assembler::vpsubw(dst, nds, src, vector_len);
3104 }
3105
3106 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
3107 assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3108 Assembler::vpsraw(dst, nds, shift, vector_len);
3109 }
3110
3111 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
3112 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3113 Assembler::vpsraw(dst, nds, shift, vector_len);
3114 }
3115
3116 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
3117 assert(UseAVX > 2,"");
3118 if (!VM_Version::supports_avx512vl() && vector_len < 2) {
3119 vector_len = 2;
3120 }
3121 Assembler::evpsraq(dst, nds, shift, vector_len);
3122 }
3123
3124 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
3125 assert(UseAVX > 2,"");
3126 if (!VM_Version::supports_avx512vl() && vector_len < 2) {
3127 vector_len = 2;
3128 }
3129 Assembler::evpsraq(dst, nds, shift, vector_len);
3130 }
3131
3132 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
3133 assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3134 Assembler::vpsrlw(dst, nds, shift, vector_len);
3135 }
3136
3137 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
3138 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3139 Assembler::vpsrlw(dst, nds, shift, vector_len);
3140 }
3141
3142 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
3143 assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3144 Assembler::vpsllw(dst, nds, shift, vector_len);
3145 }
3146
3147 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
3148 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3149 Assembler::vpsllw(dst, nds, shift, vector_len);
3150 }
3151
3152 void MacroAssembler::vptest(XMMRegister dst, XMMRegister src) {
3153 assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
3154 Assembler::vptest(dst, src);
3155 }
3156
3157 void MacroAssembler::punpcklbw(XMMRegister dst, XMMRegister src) {
3158 assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3159 Assembler::punpcklbw(dst, src);
3160 }
3161
3162 void MacroAssembler::pshufd(XMMRegister dst, Address src, int mode) {
3163 assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
3164 Assembler::pshufd(dst, src, mode);
3165 }
3166
3167 void MacroAssembler::pshuflw(XMMRegister dst, XMMRegister src, int mode) {
3168 assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3169 Assembler::pshuflw(dst, src, mode);
3170 }
3171
3172 void MacroAssembler::vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3173 assert(rscratch != noreg || always_reachable(src), "missing");
3174
3175 if (reachable(src)) {
3176 vandpd(dst, nds, as_Address(src), vector_len);
3177 } else {
3178 lea(rscratch, src);
3179 vandpd(dst, nds, Address(rscratch, 0), vector_len);
3180 }
3181 }
3182
3183 void MacroAssembler::vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3184 assert(rscratch != noreg || always_reachable(src), "missing");
3185
3186 if (reachable(src)) {
3187 vandps(dst, nds, as_Address(src), vector_len);
3188 } else {
3189 lea(rscratch, src);
3190 vandps(dst, nds, Address(rscratch, 0), vector_len);
3191 }
3192 }
3193
3194 void MacroAssembler::evpord(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src,
3195 bool merge, int vector_len, Register rscratch) {
3196 assert(rscratch != noreg || always_reachable(src), "missing");
3197
3198 if (reachable(src)) {
3199 Assembler::evpord(dst, mask, nds, as_Address(src), merge, vector_len);
3200 } else {
3201 lea(rscratch, src);
3202 Assembler::evpord(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
3203 }
3204 }
3205
3206 void MacroAssembler::vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3207 assert(rscratch != noreg || always_reachable(src), "missing");
3208
3209 if (reachable(src)) {
3210 vdivsd(dst, nds, as_Address(src));
3211 } else {
3212 lea(rscratch, src);
3213 vdivsd(dst, nds, Address(rscratch, 0));
3214 }
3215 }
3216
3217 void MacroAssembler::vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3218 assert(rscratch != noreg || always_reachable(src), "missing");
3219
3220 if (reachable(src)) {
3221 vdivss(dst, nds, as_Address(src));
3222 } else {
3223 lea(rscratch, src);
3224 vdivss(dst, nds, Address(rscratch, 0));
3225 }
3226 }
3227
3228 void MacroAssembler::vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3229 assert(rscratch != noreg || always_reachable(src), "missing");
3230
3231 if (reachable(src)) {
3232 vmulsd(dst, nds, as_Address(src));
3233 } else {
3234 lea(rscratch, src);
3235 vmulsd(dst, nds, Address(rscratch, 0));
3236 }
3237 }
3238
3239 void MacroAssembler::vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3240 assert(rscratch != noreg || always_reachable(src), "missing");
3241
3242 if (reachable(src)) {
3243 vmulss(dst, nds, as_Address(src));
3244 } else {
3245 lea(rscratch, src);
3246 vmulss(dst, nds, Address(rscratch, 0));
3247 }
3248 }
3249
3250 void MacroAssembler::vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3251 assert(rscratch != noreg || always_reachable(src), "missing");
3252
3253 if (reachable(src)) {
3254 vsubsd(dst, nds, as_Address(src));
3255 } else {
3256 lea(rscratch, src);
3257 vsubsd(dst, nds, Address(rscratch, 0));
3258 }
3259 }
3260
3261 void MacroAssembler::vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3262 assert(rscratch != noreg || always_reachable(src), "missing");
3263
3264 if (reachable(src)) {
3265 vsubss(dst, nds, as_Address(src));
3266 } else {
3267 lea(rscratch, src);
3268 vsubss(dst, nds, Address(rscratch, 0));
3269 }
3270 }
3271
3272 void MacroAssembler::vnegatess(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3273 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
3274 assert(rscratch != noreg || always_reachable(src), "missing");
3275
3276 vxorps(dst, nds, src, Assembler::AVX_128bit, rscratch);
3277 }
3278
3279 void MacroAssembler::vnegatesd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3280 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
3281 assert(rscratch != noreg || always_reachable(src), "missing");
3282
3283 vxorpd(dst, nds, src, Assembler::AVX_128bit, rscratch);
3284 }
3285
3286 void MacroAssembler::vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3287 assert(rscratch != noreg || always_reachable(src), "missing");
3288
3289 if (reachable(src)) {
3290 vxorpd(dst, nds, as_Address(src), vector_len);
3291 } else {
3292 lea(rscratch, src);
3293 vxorpd(dst, nds, Address(rscratch, 0), vector_len);
3294 }
3295 }
3296
3297 void MacroAssembler::vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3298 assert(rscratch != noreg || always_reachable(src), "missing");
3299
3300 if (reachable(src)) {
3301 vxorps(dst, nds, as_Address(src), vector_len);
3302 } else {
3303 lea(rscratch, src);
3304 vxorps(dst, nds, Address(rscratch, 0), vector_len);
3305 }
3306 }
3307
3308 void MacroAssembler::vpxor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3309 assert(rscratch != noreg || always_reachable(src), "missing");
3310
3311 if (UseAVX > 1 || (vector_len < 1)) {
3312 if (reachable(src)) {
3313 Assembler::vpxor(dst, nds, as_Address(src), vector_len);
3314 } else {
3315 lea(rscratch, src);
3316 Assembler::vpxor(dst, nds, Address(rscratch, 0), vector_len);
3317 }
3318 } else {
3319 MacroAssembler::vxorpd(dst, nds, src, vector_len, rscratch);
3320 }
3321 }
3322
3323 void MacroAssembler::vpermd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3324 assert(rscratch != noreg || always_reachable(src), "missing");
3325
3326 if (reachable(src)) {
3327 Assembler::vpermd(dst, nds, as_Address(src), vector_len);
3328 } else {
3329 lea(rscratch, src);
3330 Assembler::vpermd(dst, nds, Address(rscratch, 0), vector_len);
3331 }
3332 }
3333
3334 void MacroAssembler::clear_jobject_tag(Register possibly_non_local) {
3335 const int32_t inverted_mask = ~static_cast<int32_t>(JNIHandles::tag_mask);
3336 STATIC_ASSERT(inverted_mask == -4); // otherwise check this code
3337 // The inverted mask is sign-extended
3338 andptr(possibly_non_local, inverted_mask);
3339 }
3340
3341 void MacroAssembler::resolve_jobject(Register value,
3342 Register tmp) {
3343 Register thread = r15_thread;
3344 assert_different_registers(value, thread, tmp);
3345 Label done, tagged, weak_tagged;
3346 testptr(value, value);
3347 jcc(Assembler::zero, done); // Use null as-is.
3348 testptr(value, JNIHandles::tag_mask); // Test for tag.
3349 jcc(Assembler::notZero, tagged);
3350
3351 // Resolve local handle
3352 access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp);
3353 verify_oop(value);
3354 jmp(done);
3355
3356 bind(tagged);
3357 testptr(value, JNIHandles::TypeTag::weak_global); // Test for weak tag.
3358 jcc(Assembler::notZero, weak_tagged);
3359
3360 // Resolve global handle
3361 access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
3362 verify_oop(value);
3363 jmp(done);
3364
3365 bind(weak_tagged);
3366 // Resolve jweak.
3367 access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
3368 value, Address(value, -JNIHandles::TypeTag::weak_global), tmp);
3369 verify_oop(value);
3370
3371 bind(done);
3372 }
3373
3374 void MacroAssembler::resolve_global_jobject(Register value,
3375 Register tmp) {
3376 Register thread = r15_thread;
3377 assert_different_registers(value, thread, tmp);
3378 Label done;
3379
3380 testptr(value, value);
3381 jcc(Assembler::zero, done); // Use null as-is.
3382
3383 #ifdef ASSERT
3384 {
3385 Label valid_global_tag;
3386 testptr(value, JNIHandles::TypeTag::global); // Test for global tag.
3387 jcc(Assembler::notZero, valid_global_tag);
3388 stop("non global jobject using resolve_global_jobject");
3389 bind(valid_global_tag);
3390 }
3391 #endif
3392
3393 // Resolve global handle
3394 access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
3395 verify_oop(value);
3396
3397 bind(done);
3398 }
3399
3400 void MacroAssembler::subptr(Register dst, int32_t imm32) {
3401 subq(dst, imm32);
3402 }
3403
3404 // Force generation of a 4 byte immediate value even if it fits into 8bit
3405 void MacroAssembler::subptr_imm32(Register dst, int32_t imm32) {
3406 subq_imm32(dst, imm32);
3407 }
3408
3409 void MacroAssembler::subptr(Register dst, Register src) {
3410 subq(dst, src);
3411 }
3412
3413 // C++ bool manipulation
3414 void MacroAssembler::testbool(Register dst) {
3415 if(sizeof(bool) == 1)
3416 testb(dst, 0xff);
3417 else if(sizeof(bool) == 2) {
3418 // testw implementation needed for two byte bools
3419 ShouldNotReachHere();
3420 } else if(sizeof(bool) == 4)
3421 testl(dst, dst);
3422 else
3423 // unsupported
3424 ShouldNotReachHere();
3425 }
3426
3427 void MacroAssembler::testptr(Register dst, Register src) {
3428 testq(dst, src);
3429 }
3430
3431 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
3432 void MacroAssembler::tlab_allocate(Register obj,
3433 Register var_size_in_bytes,
3434 int con_size_in_bytes,
3435 Register t1,
3436 Register t2,
3437 Label& slow_case) {
3438 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
3439 bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
3440 }
3441
3442 RegSet MacroAssembler::call_clobbered_gp_registers() {
3443 RegSet regs;
3444 regs += RegSet::of(rax, rcx, rdx);
3445 #ifndef _WINDOWS
3446 regs += RegSet::of(rsi, rdi);
3447 #endif
3448 regs += RegSet::range(r8, r11);
3449 if (UseAPX) {
3450 regs += RegSet::range(r16, as_Register(Register::number_of_registers - 1));
3451 }
3452 return regs;
3453 }
3454
3455 XMMRegSet MacroAssembler::call_clobbered_xmm_registers() {
3456 int num_xmm_registers = XMMRegister::available_xmm_registers();
3457 #if defined(_WINDOWS)
3458 XMMRegSet result = XMMRegSet::range(xmm0, xmm5);
3459 if (num_xmm_registers > 16) {
3460 result += XMMRegSet::range(xmm16, as_XMMRegister(num_xmm_registers - 1));
3461 }
3462 return result;
3463 #else
3464 return XMMRegSet::range(xmm0, as_XMMRegister(num_xmm_registers - 1));
3465 #endif
3466 }
3467
3468 // C1 only ever uses the first double/float of the XMM register.
3469 static int xmm_save_size() { return sizeof(double); }
3470
3471 static void save_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
3472 masm->movdbl(Address(rsp, offset), reg);
3473 }
3474
3475 static void restore_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
3476 masm->movdbl(reg, Address(rsp, offset));
3477 }
3478
3479 static int register_section_sizes(RegSet gp_registers, XMMRegSet xmm_registers,
3480 bool save_fpu, int& gp_area_size, int& xmm_area_size) {
3481
3482 gp_area_size = align_up(gp_registers.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size,
3483 StackAlignmentInBytes);
3484 xmm_area_size = save_fpu ? xmm_registers.size() * xmm_save_size() : 0;
3485
3486 return gp_area_size + xmm_area_size;
3487 }
3488
3489 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude, bool save_fpu) {
3490 block_comment("push_call_clobbered_registers start");
3491 // Regular registers
3492 RegSet gp_registers_to_push = call_clobbered_gp_registers() - exclude;
3493
3494 int gp_area_size;
3495 int xmm_area_size;
3496 int total_save_size = register_section_sizes(gp_registers_to_push, call_clobbered_xmm_registers(), save_fpu,
3497 gp_area_size, xmm_area_size);
3498 subptr(rsp, total_save_size);
3499
3500 push_set(gp_registers_to_push, 0);
3501
3502 if (save_fpu) {
3503 push_set(call_clobbered_xmm_registers(), gp_area_size);
3504 }
3505
3506 block_comment("push_call_clobbered_registers end");
3507 }
3508
3509 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude, bool restore_fpu) {
3510 block_comment("pop_call_clobbered_registers start");
3511
3512 RegSet gp_registers_to_pop = call_clobbered_gp_registers() - exclude;
3513
3514 int gp_area_size;
3515 int xmm_area_size;
3516 int total_save_size = register_section_sizes(gp_registers_to_pop, call_clobbered_xmm_registers(), restore_fpu,
3517 gp_area_size, xmm_area_size);
3518
3519 if (restore_fpu) {
3520 pop_set(call_clobbered_xmm_registers(), gp_area_size);
3521 }
3522
3523 pop_set(gp_registers_to_pop, 0);
3524
3525 addptr(rsp, total_save_size);
3526
3527 vzeroupper();
3528
3529 block_comment("pop_call_clobbered_registers end");
3530 }
3531
3532 void MacroAssembler::push_set(XMMRegSet set, int offset) {
3533 assert(is_aligned(set.size() * xmm_save_size(), StackAlignmentInBytes), "must be");
3534 int spill_offset = offset;
3535
3536 for (RegSetIterator<XMMRegister> it = set.begin(); *it != xnoreg; ++it) {
3537 save_xmm_register(this, spill_offset, *it);
3538 spill_offset += xmm_save_size();
3539 }
3540 }
3541
3542 void MacroAssembler::pop_set(XMMRegSet set, int offset) {
3543 int restore_size = set.size() * xmm_save_size();
3544 assert(is_aligned(restore_size, StackAlignmentInBytes), "must be");
3545
3546 int restore_offset = offset + restore_size - xmm_save_size();
3547
3548 for (ReverseRegSetIterator<XMMRegister> it = set.rbegin(); *it != xnoreg; ++it) {
3549 restore_xmm_register(this, restore_offset, *it);
3550 restore_offset -= xmm_save_size();
3551 }
3552 }
3553
3554 void MacroAssembler::push_set(RegSet set, int offset) {
3555 int spill_offset;
3556 if (offset == -1) {
3557 int register_push_size = set.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
3558 int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
3559 subptr(rsp, aligned_size);
3560 spill_offset = 0;
3561 } else {
3562 spill_offset = offset;
3563 }
3564
3565 for (RegSetIterator<Register> it = set.begin(); *it != noreg; ++it) {
3566 movptr(Address(rsp, spill_offset), *it);
3567 spill_offset += Register::max_slots_per_register * VMRegImpl::stack_slot_size;
3568 }
3569 }
3570
3571 void MacroAssembler::pop_set(RegSet set, int offset) {
3572
3573 int gp_reg_size = Register::max_slots_per_register * VMRegImpl::stack_slot_size;
3574 int restore_size = set.size() * gp_reg_size;
3575 int aligned_size = align_up(restore_size, StackAlignmentInBytes);
3576
3577 int restore_offset;
3578 if (offset == -1) {
3579 restore_offset = restore_size - gp_reg_size;
3580 } else {
3581 restore_offset = offset + restore_size - gp_reg_size;
3582 }
3583 for (ReverseRegSetIterator<Register> it = set.rbegin(); *it != noreg; ++it) {
3584 movptr(*it, Address(rsp, restore_offset));
3585 restore_offset -= gp_reg_size;
3586 }
3587
3588 if (offset == -1) {
3589 addptr(rsp, aligned_size);
3590 }
3591 }
3592
3593 // Preserves the contents of address, destroys the contents length_in_bytes and temp.
3594 void MacroAssembler::zero_memory(Register address, Register length_in_bytes, int offset_in_bytes, Register temp) {
3595 assert(address != length_in_bytes && address != temp && temp != length_in_bytes, "registers must be different");
3596 assert((offset_in_bytes & (BytesPerWord - 1)) == 0, "offset must be a multiple of BytesPerWord");
3597 Label done;
3598
3599 testptr(length_in_bytes, length_in_bytes);
3600 jcc(Assembler::zero, done);
3601
3602 // initialize topmost word, divide index by 2, check if odd and test if zero
3603 // note: for the remaining code to work, index must be a multiple of BytesPerWord
3604 #ifdef ASSERT
3605 {
3606 Label L;
3607 testptr(length_in_bytes, BytesPerWord - 1);
3608 jcc(Assembler::zero, L);
3609 stop("length must be a multiple of BytesPerWord");
3610 bind(L);
3611 }
3612 #endif
3613 Register index = length_in_bytes;
3614 xorptr(temp, temp); // use _zero reg to clear memory (shorter code)
3615 if (UseIncDec) {
3616 shrptr(index, 3); // divide by 8/16 and set carry flag if bit 2 was set
3617 } else {
3618 shrptr(index, 2); // use 2 instructions to avoid partial flag stall
3619 shrptr(index, 1);
3620 }
3621
3622 // initialize remaining object fields: index is a multiple of 2 now
3623 {
3624 Label loop;
3625 bind(loop);
3626 movptr(Address(address, index, Address::times_8, offset_in_bytes - 1*BytesPerWord), temp);
3627 decrement(index);
3628 jcc(Assembler::notZero, loop);
3629 }
3630
3631 bind(done);
3632 }
3633
3634 // Look up the method for a megamorphic invokeinterface call.
3635 // The target method is determined by <intf_klass, itable_index>.
3636 // The receiver klass is in recv_klass.
3637 // On success, the result will be in method_result, and execution falls through.
3638 // On failure, execution transfers to the given label.
3639 void MacroAssembler::lookup_interface_method(Register recv_klass,
3640 Register intf_klass,
3641 RegisterOrConstant itable_index,
3642 Register method_result,
3643 Register scan_temp,
3644 Label& L_no_such_interface,
3645 bool return_method) {
3646 assert_different_registers(recv_klass, intf_klass, scan_temp);
3647 assert_different_registers(method_result, intf_klass, scan_temp);
3648 assert(recv_klass != method_result || !return_method,
3649 "recv_klass can be destroyed when method isn't needed");
3650
3651 assert(itable_index.is_constant() || itable_index.as_register() == method_result,
3652 "caller must use same register for non-constant itable index as for method");
3653
3654 // Compute start of first itableOffsetEntry (which is at the end of the vtable)
3655 int vtable_base = in_bytes(Klass::vtable_start_offset());
3656 int itentry_off = in_bytes(itableMethodEntry::method_offset());
3657 int scan_step = itableOffsetEntry::size() * wordSize;
3658 int vte_size = vtableEntry::size_in_bytes();
3659 Address::ScaleFactor times_vte_scale = Address::times_ptr;
3660 assert(vte_size == wordSize, "else adjust times_vte_scale");
3661
3662 movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
3663
3664 // Could store the aligned, prescaled offset in the klass.
3665 lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
3666
3667 if (return_method) {
3668 // Adjust recv_klass by scaled itable_index, so we can free itable_index.
3669 assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
3670 lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
3671 }
3672
3673 // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
3674 // if (scan->interface() == intf) {
3675 // result = (klass + scan->offset() + itable_index);
3676 // }
3677 // }
3678 Label search, found_method;
3679
3680 for (int peel = 1; peel >= 0; peel--) {
3681 movptr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
3682 cmpptr(intf_klass, method_result);
3683
3684 if (peel) {
3685 jccb(Assembler::equal, found_method);
3686 } else {
3687 jccb(Assembler::notEqual, search);
3688 // (invert the test to fall through to found_method...)
3689 }
3690
3691 if (!peel) break;
3692
3693 bind(search);
3694
3695 // Check that the previous entry is non-null. A null entry means that
3696 // the receiver class doesn't implement the interface, and wasn't the
3697 // same as when the caller was compiled.
3698 testptr(method_result, method_result);
3699 jcc(Assembler::zero, L_no_such_interface);
3700 addptr(scan_temp, scan_step);
3701 }
3702
3703 bind(found_method);
3704
3705 if (return_method) {
3706 // Got a hit.
3707 movl(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
3708 movptr(method_result, Address(recv_klass, scan_temp, Address::times_1));
3709 }
3710 }
3711
3712 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
3713 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
3714 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
3715 // The target method is determined by <holder_klass, itable_index>.
3716 // The receiver klass is in recv_klass.
3717 // On success, the result will be in method_result, and execution falls through.
3718 // On failure, execution transfers to the given label.
3719 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
3720 Register holder_klass,
3721 Register resolved_klass,
3722 Register method_result,
3723 Register scan_temp,
3724 Register temp_reg2,
3725 Register receiver,
3726 int itable_index,
3727 Label& L_no_such_interface) {
3728 assert_different_registers(recv_klass, method_result, holder_klass, resolved_klass, scan_temp, temp_reg2, receiver);
3729 Register temp_itbl_klass = method_result;
3730 Register temp_reg = (temp_reg2 == noreg ? recv_klass : temp_reg2); // reuse recv_klass register on 32-bit x86 impl
3731
3732 int vtable_base = in_bytes(Klass::vtable_start_offset());
3733 int itentry_off = in_bytes(itableMethodEntry::method_offset());
3734 int scan_step = itableOffsetEntry::size() * wordSize;
3735 int vte_size = vtableEntry::size_in_bytes();
3736 int ioffset = in_bytes(itableOffsetEntry::interface_offset());
3737 int ooffset = in_bytes(itableOffsetEntry::offset_offset());
3738 Address::ScaleFactor times_vte_scale = Address::times_ptr;
3739 assert(vte_size == wordSize, "adjust times_vte_scale");
3740
3741 Label L_loop_scan_resolved_entry, L_resolved_found, L_holder_found;
3742
3743 // temp_itbl_klass = recv_klass.itable[0]
3744 // scan_temp = &recv_klass.itable[0] + step
3745 movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
3746 movptr(temp_itbl_klass, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset));
3747 lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset + scan_step));
3748 xorptr(temp_reg, temp_reg);
3749
3750 // Initial checks:
3751 // - if (holder_klass != resolved_klass), go to "scan for resolved"
3752 // - if (itable[0] == 0), no such interface
3753 // - if (itable[0] == holder_klass), shortcut to "holder found"
3754 cmpptr(holder_klass, resolved_klass);
3755 jccb(Assembler::notEqual, L_loop_scan_resolved_entry);
3756 testptr(temp_itbl_klass, temp_itbl_klass);
3757 jccb(Assembler::zero, L_no_such_interface);
3758 cmpptr(holder_klass, temp_itbl_klass);
3759 jccb(Assembler::equal, L_holder_found);
3760
3761 // Loop: Look for holder_klass record in itable
3762 // do {
3763 // tmp = itable[index];
3764 // index += step;
3765 // if (tmp == holder_klass) {
3766 // goto L_holder_found; // Found!
3767 // }
3768 // } while (tmp != 0);
3769 // goto L_no_such_interface // Not found.
3770 Label L_scan_holder;
3771 bind(L_scan_holder);
3772 movptr(temp_itbl_klass, Address(scan_temp, 0));
3773 addptr(scan_temp, scan_step);
3774 cmpptr(holder_klass, temp_itbl_klass);
3775 jccb(Assembler::equal, L_holder_found);
3776 testptr(temp_itbl_klass, temp_itbl_klass);
3777 jccb(Assembler::notZero, L_scan_holder);
3778
3779 jmpb(L_no_such_interface);
3780
3781 // Loop: Look for resolved_class record in itable
3782 // do {
3783 // tmp = itable[index];
3784 // index += step;
3785 // if (tmp == holder_klass) {
3786 // // Also check if we have met a holder klass
3787 // holder_tmp = itable[index-step-ioffset];
3788 // }
3789 // if (tmp == resolved_klass) {
3790 // goto L_resolved_found; // Found!
3791 // }
3792 // } while (tmp != 0);
3793 // goto L_no_such_interface // Not found.
3794 //
3795 Label L_loop_scan_resolved;
3796 bind(L_loop_scan_resolved);
3797 movptr(temp_itbl_klass, Address(scan_temp, 0));
3798 addptr(scan_temp, scan_step);
3799 bind(L_loop_scan_resolved_entry);
3800 cmpptr(holder_klass, temp_itbl_klass);
3801 cmovl(Assembler::equal, temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
3802 cmpptr(resolved_klass, temp_itbl_klass);
3803 jccb(Assembler::equal, L_resolved_found);
3804 testptr(temp_itbl_klass, temp_itbl_klass);
3805 jccb(Assembler::notZero, L_loop_scan_resolved);
3806
3807 jmpb(L_no_such_interface);
3808
3809 Label L_ready;
3810
3811 // See if we already have a holder klass. If not, go and scan for it.
3812 bind(L_resolved_found);
3813 testptr(temp_reg, temp_reg);
3814 jccb(Assembler::zero, L_scan_holder);
3815 jmpb(L_ready);
3816
3817 bind(L_holder_found);
3818 movl(temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
3819
3820 // Finally, temp_reg contains holder_klass vtable offset
3821 bind(L_ready);
3822 assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
3823 if (temp_reg2 == noreg) { // recv_klass register is clobbered for 32-bit x86 impl
3824 load_klass(scan_temp, receiver, noreg);
3825 movptr(method_result, Address(scan_temp, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
3826 } else {
3827 movptr(method_result, Address(recv_klass, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
3828 }
3829 }
3830
3831
3832 // virtual method calling
3833 void MacroAssembler::lookup_virtual_method(Register recv_klass,
3834 RegisterOrConstant vtable_index,
3835 Register method_result) {
3836 const ByteSize base = Klass::vtable_start_offset();
3837 assert(vtableEntry::size() * wordSize == wordSize, "else adjust the scaling in the code below");
3838 Address vtable_entry_addr(recv_klass,
3839 vtable_index, Address::times_ptr,
3840 base + vtableEntry::method_offset());
3841 movptr(method_result, vtable_entry_addr);
3842 }
3843
3844
3845 void MacroAssembler::check_klass_subtype(Register sub_klass,
3846 Register super_klass,
3847 Register temp_reg,
3848 Label& L_success) {
3849 Label L_failure;
3850 check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg, &L_success, &L_failure, nullptr);
3851 check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
3852 bind(L_failure);
3853 }
3854
3855
3856 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
3857 Register super_klass,
3858 Register temp_reg,
3859 Label* L_success,
3860 Label* L_failure,
3861 Label* L_slow_path,
3862 RegisterOrConstant super_check_offset) {
3863 assert_different_registers(sub_klass, super_klass, temp_reg);
3864 bool must_load_sco = (super_check_offset.constant_or_zero() == -1);
3865 if (super_check_offset.is_register()) {
3866 assert_different_registers(sub_klass, super_klass,
3867 super_check_offset.as_register());
3868 } else if (must_load_sco) {
3869 assert(temp_reg != noreg, "supply either a temp or a register offset");
3870 }
3871
3872 Label L_fallthrough;
3873 int label_nulls = 0;
3874 if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
3875 if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
3876 if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
3877 assert(label_nulls <= 1, "at most one null in the batch");
3878
3879 int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
3880 int sco_offset = in_bytes(Klass::super_check_offset_offset());
3881 Address super_check_offset_addr(super_klass, sco_offset);
3882
3883 // Hacked jcc, which "knows" that L_fallthrough, at least, is in
3884 // range of a jccb. If this routine grows larger, reconsider at
3885 // least some of these.
3886 #define local_jcc(assembler_cond, label) \
3887 if (&(label) == &L_fallthrough) jccb(assembler_cond, label); \
3888 else jcc( assembler_cond, label) /*omit semi*/
3889
3890 // Hacked jmp, which may only be used just before L_fallthrough.
3891 #define final_jmp(label) \
3892 if (&(label) == &L_fallthrough) { /*do nothing*/ } \
3893 else jmp(label) /*omit semi*/
3894
3895 // If the pointers are equal, we are done (e.g., String[] elements).
3896 // This self-check enables sharing of secondary supertype arrays among
3897 // non-primary types such as array-of-interface. Otherwise, each such
3898 // type would need its own customized SSA.
3899 // We move this check to the front of the fast path because many
3900 // type checks are in fact trivially successful in this manner,
3901 // so we get a nicely predicted branch right at the start of the check.
3902 cmpptr(sub_klass, super_klass);
3903 local_jcc(Assembler::equal, *L_success);
3904
3905 // Check the supertype display:
3906 if (must_load_sco) {
3907 // Positive movl does right thing on LP64.
3908 movl(temp_reg, super_check_offset_addr);
3909 super_check_offset = RegisterOrConstant(temp_reg);
3910 }
3911 Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0);
3912 cmpptr(super_klass, super_check_addr); // load displayed supertype
3913
3914 // This check has worked decisively for primary supers.
3915 // Secondary supers are sought in the super_cache ('super_cache_addr').
3916 // (Secondary supers are interfaces and very deeply nested subtypes.)
3917 // This works in the same check above because of a tricky aliasing
3918 // between the super_cache and the primary super display elements.
3919 // (The 'super_check_addr' can address either, as the case requires.)
3920 // Note that the cache is updated below if it does not help us find
3921 // what we need immediately.
3922 // So if it was a primary super, we can just fail immediately.
3923 // Otherwise, it's the slow path for us (no success at this point).
3924
3925 if (super_check_offset.is_register()) {
3926 local_jcc(Assembler::equal, *L_success);
3927 cmpl(super_check_offset.as_register(), sc_offset);
3928 if (L_failure == &L_fallthrough) {
3929 local_jcc(Assembler::equal, *L_slow_path);
3930 } else {
3931 local_jcc(Assembler::notEqual, *L_failure);
3932 final_jmp(*L_slow_path);
3933 }
3934 } else if (super_check_offset.as_constant() == sc_offset) {
3935 // Need a slow path; fast failure is impossible.
3936 if (L_slow_path == &L_fallthrough) {
3937 local_jcc(Assembler::equal, *L_success);
3938 } else {
3939 local_jcc(Assembler::notEqual, *L_slow_path);
3940 final_jmp(*L_success);
3941 }
3942 } else {
3943 // No slow path; it's a fast decision.
3944 if (L_failure == &L_fallthrough) {
3945 local_jcc(Assembler::equal, *L_success);
3946 } else {
3947 local_jcc(Assembler::notEqual, *L_failure);
3948 final_jmp(*L_success);
3949 }
3950 }
3951
3952 bind(L_fallthrough);
3953
3954 #undef local_jcc
3955 #undef final_jmp
3956 }
3957
3958
3959 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
3960 Register super_klass,
3961 Register temp_reg,
3962 Register temp2_reg,
3963 Label* L_success,
3964 Label* L_failure,
3965 bool set_cond_codes) {
3966 assert_different_registers(sub_klass, super_klass, temp_reg);
3967 if (temp2_reg != noreg)
3968 assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg);
3969 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
3970
3971 Label L_fallthrough;
3972 int label_nulls = 0;
3973 if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
3974 if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
3975 assert(label_nulls <= 1, "at most one null in the batch");
3976
3977 // a couple of useful fields in sub_klass:
3978 int ss_offset = in_bytes(Klass::secondary_supers_offset());
3979 int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
3980 Address secondary_supers_addr(sub_klass, ss_offset);
3981 Address super_cache_addr( sub_klass, sc_offset);
3982
3983 // Do a linear scan of the secondary super-klass chain.
3984 // This code is rarely used, so simplicity is a virtue here.
3985 // The repne_scan instruction uses fixed registers, which we must spill.
3986 // Don't worry too much about pre-existing connections with the input regs.
3987
3988 assert(sub_klass != rax, "killed reg"); // killed by mov(rax, super)
3989 assert(sub_klass != rcx, "killed reg"); // killed by lea(rcx, &pst_counter)
3990
3991 // Get super_klass value into rax (even if it was in rdi or rcx).
3992 bool pushed_rax = false, pushed_rcx = false, pushed_rdi = false;
3993 if (super_klass != rax) {
3994 if (!IS_A_TEMP(rax)) { push(rax); pushed_rax = true; }
3995 mov(rax, super_klass);
3996 }
3997 if (!IS_A_TEMP(rcx)) { push(rcx); pushed_rcx = true; }
3998 if (!IS_A_TEMP(rdi)) { push(rdi); pushed_rdi = true; }
3999
4000 #ifndef PRODUCT
4001 uint* pst_counter = &SharedRuntime::_partial_subtype_ctr;
4002 ExternalAddress pst_counter_addr((address) pst_counter);
4003 lea(rcx, pst_counter_addr);
4004 incrementl(Address(rcx, 0));
4005 #endif //PRODUCT
4006
4007 // We will consult the secondary-super array.
4008 movptr(rdi, secondary_supers_addr);
4009 // Load the array length. (Positive movl does right thing on LP64.)
4010 movl(rcx, Address(rdi, Array<Klass*>::length_offset_in_bytes()));
4011 // Skip to start of data.
4012 addptr(rdi, Array<Klass*>::base_offset_in_bytes());
4013
4014 // Scan RCX words at [RDI] for an occurrence of RAX.
4015 // Set NZ/Z based on last compare.
4016 // Z flag value will not be set by 'repne' if RCX == 0 since 'repne' does
4017 // not change flags (only scas instruction which is repeated sets flags).
4018 // Set Z = 0 (not equal) before 'repne' to indicate that class was not found.
4019
4020 testptr(rax,rax); // Set Z = 0
4021 repne_scan();
4022
4023 // Unspill the temp. registers:
4024 if (pushed_rdi) pop(rdi);
4025 if (pushed_rcx) pop(rcx);
4026 if (pushed_rax) pop(rax);
4027
4028 if (set_cond_codes) {
4029 // Special hack for the AD files: rdi is guaranteed non-zero.
4030 assert(!pushed_rdi, "rdi must be left non-null");
4031 // Also, the condition codes are properly set Z/NZ on succeed/failure.
4032 }
4033
4034 if (L_failure == &L_fallthrough)
4035 jccb(Assembler::notEqual, *L_failure);
4036 else jcc(Assembler::notEqual, *L_failure);
4037
4038 // Success. Cache the super we found and proceed in triumph.
4039 movptr(super_cache_addr, super_klass);
4040
4041 if (L_success != &L_fallthrough) {
4042 jmp(*L_success);
4043 }
4044
4045 #undef IS_A_TEMP
4046
4047 bind(L_fallthrough);
4048 }
4049
4050 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
4051 Register super_klass,
4052 Register temp_reg,
4053 Register temp2_reg,
4054 Label* L_success,
4055 Label* L_failure,
4056 bool set_cond_codes) {
4057 assert(set_cond_codes == false, "must be false on 64-bit x86");
4058 check_klass_subtype_slow_path
4059 (sub_klass, super_klass, temp_reg, temp2_reg, noreg, noreg,
4060 L_success, L_failure);
4061 }
4062
4063 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
4064 Register super_klass,
4065 Register temp_reg,
4066 Register temp2_reg,
4067 Register temp3_reg,
4068 Register temp4_reg,
4069 Label* L_success,
4070 Label* L_failure) {
4071 if (UseSecondarySupersTable) {
4072 check_klass_subtype_slow_path_table
4073 (sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, temp4_reg,
4074 L_success, L_failure);
4075 } else {
4076 check_klass_subtype_slow_path_linear
4077 (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, /*set_cond_codes*/false);
4078 }
4079 }
4080
4081 Register MacroAssembler::allocate_if_noreg(Register r,
4082 RegSetIterator<Register> &available_regs,
4083 RegSet ®s_to_push) {
4084 if (!r->is_valid()) {
4085 r = *available_regs++;
4086 regs_to_push += r;
4087 }
4088 return r;
4089 }
4090
4091 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
4092 Register super_klass,
4093 Register temp_reg,
4094 Register temp2_reg,
4095 Register temp3_reg,
4096 Register result_reg,
4097 Label* L_success,
4098 Label* L_failure) {
4099 // NB! Callers may assume that, when temp2_reg is a valid register,
4100 // this code sets it to a nonzero value.
4101 bool temp2_reg_was_valid = temp2_reg->is_valid();
4102
4103 RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
4104
4105 Label L_fallthrough;
4106 int label_nulls = 0;
4107 if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
4108 if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
4109 assert(label_nulls <= 1, "at most one null in the batch");
4110
4111 BLOCK_COMMENT("check_klass_subtype_slow_path_table");
4112
4113 RegSetIterator<Register> available_regs
4114 = (RegSet::of(rax, rcx, rdx, r8) + r9 + r10 + r11 + r12 - temps - sub_klass - super_klass).begin();
4115
4116 RegSet pushed_regs;
4117
4118 temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
4119 temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
4120 temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
4121 result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
4122 Register temp4_reg = allocate_if_noreg(noreg, available_regs, pushed_regs);
4123
4124 assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, result_reg);
4125
4126 {
4127
4128 int register_push_size = pushed_regs.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
4129 int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
4130 subptr(rsp, aligned_size);
4131 push_set(pushed_regs, 0);
4132
4133 lookup_secondary_supers_table_var(sub_klass,
4134 super_klass,
4135 temp_reg, temp2_reg, temp3_reg, temp4_reg, result_reg);
4136 cmpq(result_reg, 0);
4137
4138 // Unspill the temp. registers:
4139 pop_set(pushed_regs, 0);
4140 // Increment SP but do not clobber flags.
4141 lea(rsp, Address(rsp, aligned_size));
4142 }
4143
4144 if (temp2_reg_was_valid) {
4145 movq(temp2_reg, 1);
4146 }
4147
4148 jcc(Assembler::notEqual, *L_failure);
4149
4150 if (L_success != &L_fallthrough) {
4151 jmp(*L_success);
4152 }
4153
4154 bind(L_fallthrough);
4155 }
4156
4157 // population_count variant for running without the POPCNT
4158 // instruction, which was introduced with SSE4.2 in 2008.
4159 void MacroAssembler::population_count(Register dst, Register src,
4160 Register scratch1, Register scratch2) {
4161 assert_different_registers(src, scratch1, scratch2);
4162 if (UsePopCountInstruction) {
4163 Assembler::popcntq(dst, src);
4164 } else {
4165 assert_different_registers(src, scratch1, scratch2);
4166 assert_different_registers(dst, scratch1, scratch2);
4167 Label loop, done;
4168
4169 mov(scratch1, src);
4170 // dst = 0;
4171 // while(scratch1 != 0) {
4172 // dst++;
4173 // scratch1 &= (scratch1 - 1);
4174 // }
4175 xorl(dst, dst);
4176 testq(scratch1, scratch1);
4177 jccb(Assembler::equal, done);
4178 {
4179 bind(loop);
4180 incq(dst);
4181 movq(scratch2, scratch1);
4182 decq(scratch2);
4183 andq(scratch1, scratch2);
4184 jccb(Assembler::notEqual, loop);
4185 }
4186 bind(done);
4187 }
4188 #ifdef ASSERT
4189 mov64(scratch1, 0xCafeBabeDeadBeef);
4190 movq(scratch2, scratch1);
4191 #endif
4192 }
4193
4194 // Ensure that the inline code and the stub are using the same registers.
4195 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS \
4196 do { \
4197 assert(r_super_klass == rax, "mismatch"); \
4198 assert(r_array_base == rbx, "mismatch"); \
4199 assert(r_array_length == rcx, "mismatch"); \
4200 assert(r_array_index == rdx, "mismatch"); \
4201 assert(r_sub_klass == rsi || r_sub_klass == noreg, "mismatch"); \
4202 assert(r_bitmap == r11 || r_bitmap == noreg, "mismatch"); \
4203 assert(result == rdi || result == noreg, "mismatch"); \
4204 } while(0)
4205
4206 // Versions of salq and rorq that don't need count to be in rcx
4207
4208 void MacroAssembler::salq(Register dest, Register count) {
4209 if (count == rcx) {
4210 Assembler::salq(dest);
4211 } else {
4212 assert_different_registers(rcx, dest);
4213 xchgq(rcx, count);
4214 Assembler::salq(dest);
4215 xchgq(rcx, count);
4216 }
4217 }
4218
4219 void MacroAssembler::rorq(Register dest, Register count) {
4220 if (count == rcx) {
4221 Assembler::rorq(dest);
4222 } else {
4223 assert_different_registers(rcx, dest);
4224 xchgq(rcx, count);
4225 Assembler::rorq(dest);
4226 xchgq(rcx, count);
4227 }
4228 }
4229
4230 // Return true: we succeeded in generating this code
4231 //
4232 // At runtime, return 0 in result if r_super_klass is a superclass of
4233 // r_sub_klass, otherwise return nonzero. Use this if you know the
4234 // super_klass_slot of the class you're looking for. This is always
4235 // the case for instanceof and checkcast.
4236 void MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
4237 Register r_super_klass,
4238 Register temp1,
4239 Register temp2,
4240 Register temp3,
4241 Register temp4,
4242 Register result,
4243 u1 super_klass_slot) {
4244 assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
4245
4246 Label L_fallthrough, L_success, L_failure;
4247
4248 BLOCK_COMMENT("lookup_secondary_supers_table {");
4249
4250 const Register
4251 r_array_index = temp1,
4252 r_array_length = temp2,
4253 r_array_base = temp3,
4254 r_bitmap = temp4;
4255
4256 LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
4257
4258 xorq(result, result); // = 0
4259
4260 movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
4261 movq(r_array_index, r_bitmap);
4262
4263 // First check the bitmap to see if super_klass might be present. If
4264 // the bit is zero, we are certain that super_klass is not one of
4265 // the secondary supers.
4266 u1 bit = super_klass_slot;
4267 {
4268 // NB: If the count in a x86 shift instruction is 0, the flags are
4269 // not affected, so we do a testq instead.
4270 int shift_count = Klass::SECONDARY_SUPERS_TABLE_MASK - bit;
4271 if (shift_count != 0) {
4272 salq(r_array_index, shift_count);
4273 } else {
4274 testq(r_array_index, r_array_index);
4275 }
4276 }
4277 // We test the MSB of r_array_index, i.e. its sign bit
4278 jcc(Assembler::positive, L_failure);
4279
4280 // Get the first array index that can contain super_klass into r_array_index.
4281 if (bit != 0) {
4282 population_count(r_array_index, r_array_index, temp2, temp3);
4283 } else {
4284 movl(r_array_index, 1);
4285 }
4286 // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
4287
4288 // We will consult the secondary-super array.
4289 movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
4290
4291 // We're asserting that the first word in an Array<Klass*> is the
4292 // length, and the second word is the first word of the data. If
4293 // that ever changes, r_array_base will have to be adjusted here.
4294 assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
4295 assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
4296
4297 cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
4298 jccb(Assembler::equal, L_success);
4299
4300 // Is there another entry to check? Consult the bitmap.
4301 btq(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK);
4302 jccb(Assembler::carryClear, L_failure);
4303
4304 // Linear probe. Rotate the bitmap so that the next bit to test is
4305 // in Bit 1.
4306 if (bit != 0) {
4307 rorq(r_bitmap, bit);
4308 }
4309
4310 // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
4311 // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
4312 // Kills: r_array_length.
4313 // Returns: result.
4314 call(RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub()));
4315 // Result (0/1) is in rdi
4316 jmpb(L_fallthrough);
4317
4318 bind(L_failure);
4319 incq(result); // 0 => 1
4320
4321 bind(L_success);
4322 // result = 0;
4323
4324 bind(L_fallthrough);
4325 BLOCK_COMMENT("} lookup_secondary_supers_table");
4326
4327 if (VerifySecondarySupers) {
4328 verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
4329 temp1, temp2, temp3);
4330 }
4331 }
4332
4333 // At runtime, return 0 in result if r_super_klass is a superclass of
4334 // r_sub_klass, otherwise return nonzero. Use this version of
4335 // lookup_secondary_supers_table() if you don't know ahead of time
4336 // which superclass will be searched for. Used by interpreter and
4337 // runtime stubs. It is larger and has somewhat greater latency than
4338 // the version above, which takes a constant super_klass_slot.
4339 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
4340 Register r_super_klass,
4341 Register temp1,
4342 Register temp2,
4343 Register temp3,
4344 Register temp4,
4345 Register result) {
4346 assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
4347 assert_different_registers(r_sub_klass, r_super_klass, rcx);
4348 RegSet temps = RegSet::of(temp1, temp2, temp3, temp4);
4349
4350 Label L_fallthrough, L_success, L_failure;
4351
4352 BLOCK_COMMENT("lookup_secondary_supers_table {");
4353
4354 RegSetIterator<Register> available_regs = (temps - rcx).begin();
4355
4356 // FIXME. Once we are sure that all paths reaching this point really
4357 // do pass rcx as one of our temps we can get rid of the following
4358 // workaround.
4359 assert(temps.contains(rcx), "fix this code");
4360
4361 // We prefer to have our shift count in rcx. If rcx is one of our
4362 // temps, use it for slot. If not, pick any of our temps.
4363 Register slot;
4364 if (!temps.contains(rcx)) {
4365 slot = *available_regs++;
4366 } else {
4367 slot = rcx;
4368 }
4369
4370 const Register r_array_index = *available_regs++;
4371 const Register r_bitmap = *available_regs++;
4372
4373 // The logic above guarantees this property, but we state it here.
4374 assert_different_registers(r_array_index, r_bitmap, rcx);
4375
4376 movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
4377 movq(r_array_index, r_bitmap);
4378
4379 // First check the bitmap to see if super_klass might be present. If
4380 // the bit is zero, we are certain that super_klass is not one of
4381 // the secondary supers.
4382 movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
4383 xorl(slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1)); // slot ^ 63 === 63 - slot (mod 64)
4384 salq(r_array_index, slot);
4385
4386 testq(r_array_index, r_array_index);
4387 // We test the MSB of r_array_index, i.e. its sign bit
4388 jcc(Assembler::positive, L_failure);
4389
4390 const Register r_array_base = *available_regs++;
4391
4392 // Get the first array index that can contain super_klass into r_array_index.
4393 // Note: Clobbers r_array_base and slot.
4394 population_count(r_array_index, r_array_index, /*temp2*/r_array_base, /*temp3*/slot);
4395
4396 // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
4397
4398 // We will consult the secondary-super array.
4399 movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
4400
4401 // We're asserting that the first word in an Array<Klass*> is the
4402 // length, and the second word is the first word of the data. If
4403 // that ever changes, r_array_base will have to be adjusted here.
4404 assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
4405 assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
4406
4407 cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
4408 jccb(Assembler::equal, L_success);
4409
4410 // Restore slot to its true value
4411 movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
4412
4413 // Linear probe. Rotate the bitmap so that the next bit to test is
4414 // in Bit 1.
4415 rorq(r_bitmap, slot);
4416
4417 // Is there another entry to check? Consult the bitmap.
4418 btq(r_bitmap, 1);
4419 jccb(Assembler::carryClear, L_failure);
4420
4421 // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
4422 // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
4423 // Kills: r_array_length.
4424 // Returns: result.
4425 lookup_secondary_supers_table_slow_path(r_super_klass,
4426 r_array_base,
4427 r_array_index,
4428 r_bitmap,
4429 /*temp1*/result,
4430 /*temp2*/slot,
4431 &L_success,
4432 nullptr);
4433
4434 bind(L_failure);
4435 movq(result, 1);
4436 jmpb(L_fallthrough);
4437
4438 bind(L_success);
4439 xorq(result, result); // = 0
4440
4441 bind(L_fallthrough);
4442 BLOCK_COMMENT("} lookup_secondary_supers_table");
4443
4444 if (VerifySecondarySupers) {
4445 verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
4446 temp1, temp2, temp3);
4447 }
4448 }
4449
4450 void MacroAssembler::repne_scanq(Register addr, Register value, Register count, Register limit,
4451 Label* L_success, Label* L_failure) {
4452 Label L_loop, L_fallthrough;
4453 {
4454 int label_nulls = 0;
4455 if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
4456 if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
4457 assert(label_nulls <= 1, "at most one null in the batch");
4458 }
4459 bind(L_loop);
4460 cmpq(value, Address(addr, count, Address::times_8));
4461 jcc(Assembler::equal, *L_success);
4462 addl(count, 1);
4463 cmpl(count, limit);
4464 jcc(Assembler::less, L_loop);
4465
4466 if (&L_fallthrough != L_failure) {
4467 jmp(*L_failure);
4468 }
4469 bind(L_fallthrough);
4470 }
4471
4472 // Called by code generated by check_klass_subtype_slow_path
4473 // above. This is called when there is a collision in the hashed
4474 // lookup in the secondary supers array.
4475 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
4476 Register r_array_base,
4477 Register r_array_index,
4478 Register r_bitmap,
4479 Register temp1,
4480 Register temp2,
4481 Label* L_success,
4482 Label* L_failure) {
4483 assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, temp2);
4484
4485 const Register
4486 r_array_length = temp1,
4487 r_sub_klass = noreg,
4488 result = noreg;
4489
4490 Label L_fallthrough;
4491 int label_nulls = 0;
4492 if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
4493 if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
4494 assert(label_nulls <= 1, "at most one null in the batch");
4495
4496 // Load the array length.
4497 movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
4498 // And adjust the array base to point to the data.
4499 // NB! Effectively increments current slot index by 1.
4500 assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
4501 addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
4502
4503 // Linear probe
4504 Label L_huge;
4505
4506 // The bitmap is full to bursting.
4507 // Implicit invariant: BITMAP_FULL implies (length > 0)
4508 cmpl(r_array_length, (int32_t)Klass::SECONDARY_SUPERS_TABLE_SIZE - 2);
4509 jcc(Assembler::greater, L_huge);
4510
4511 // NB! Our caller has checked bits 0 and 1 in the bitmap. The
4512 // current slot (at secondary_supers[r_array_index]) has not yet
4513 // been inspected, and r_array_index may be out of bounds if we
4514 // wrapped around the end of the array.
4515
4516 { // This is conventional linear probing, but instead of terminating
4517 // when a null entry is found in the table, we maintain a bitmap
4518 // in which a 0 indicates missing entries.
4519 // The check above guarantees there are 0s in the bitmap, so the loop
4520 // eventually terminates.
4521
4522 xorl(temp2, temp2); // = 0;
4523
4524 Label L_again;
4525 bind(L_again);
4526
4527 // Check for array wraparound.
4528 cmpl(r_array_index, r_array_length);
4529 cmovl(Assembler::greaterEqual, r_array_index, temp2);
4530
4531 cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
4532 jcc(Assembler::equal, *L_success);
4533
4534 // If the next bit in bitmap is zero, we're done.
4535 btq(r_bitmap, 2); // look-ahead check (Bit 2); Bits 0 and 1 are tested by now
4536 jcc(Assembler::carryClear, *L_failure);
4537
4538 rorq(r_bitmap, 1); // Bits 1/2 => 0/1
4539 addl(r_array_index, 1);
4540
4541 jmp(L_again);
4542 }
4543
4544 { // Degenerate case: more than 64 secondary supers.
4545 // FIXME: We could do something smarter here, maybe a vectorized
4546 // comparison or a binary search, but is that worth any added
4547 // complexity?
4548 bind(L_huge);
4549 xorl(r_array_index, r_array_index); // = 0
4550 repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length,
4551 L_success,
4552 (&L_fallthrough != L_failure ? L_failure : nullptr));
4553
4554 bind(L_fallthrough);
4555 }
4556 }
4557
4558 struct VerifyHelperArguments {
4559 Klass* _super;
4560 Klass* _sub;
4561 intptr_t _linear_result;
4562 intptr_t _table_result;
4563 };
4564
4565 static void verify_secondary_supers_table_helper(const char* msg, VerifyHelperArguments* args) {
4566 Klass::on_secondary_supers_verification_failure(args->_super,
4567 args->_sub,
4568 args->_linear_result,
4569 args->_table_result,
4570 msg);
4571 }
4572
4573 // Make sure that the hashed lookup and a linear scan agree.
4574 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
4575 Register r_super_klass,
4576 Register result,
4577 Register temp1,
4578 Register temp2,
4579 Register temp3) {
4580 const Register
4581 r_array_index = temp1,
4582 r_array_length = temp2,
4583 r_array_base = temp3,
4584 r_bitmap = noreg;
4585
4586 BLOCK_COMMENT("verify_secondary_supers_table {");
4587
4588 Label L_success, L_failure, L_check, L_done;
4589
4590 movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
4591 movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
4592 // And adjust the array base to point to the data.
4593 addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
4594
4595 testl(r_array_length, r_array_length); // array_length == 0?
4596 jcc(Assembler::zero, L_failure);
4597
4598 movl(r_array_index, 0);
4599 repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length, &L_success);
4600 // fall through to L_failure
4601
4602 const Register linear_result = r_array_index; // reuse temp1
4603
4604 bind(L_failure); // not present
4605 movl(linear_result, 1);
4606 jmp(L_check);
4607
4608 bind(L_success); // present
4609 movl(linear_result, 0);
4610
4611 bind(L_check);
4612 cmpl(linear_result, result);
4613 jcc(Assembler::equal, L_done);
4614
4615 { // To avoid calling convention issues, build a record on the stack
4616 // and pass the pointer to that instead.
4617 push(result);
4618 push(linear_result);
4619 push(r_sub_klass);
4620 push(r_super_klass);
4621 movptr(c_rarg1, rsp);
4622 movptr(c_rarg0, (uintptr_t) "mismatch");
4623 call(RuntimeAddress(CAST_FROM_FN_PTR(address, verify_secondary_supers_table_helper)));
4624 should_not_reach_here();
4625 }
4626 bind(L_done);
4627
4628 BLOCK_COMMENT("} verify_secondary_supers_table");
4629 }
4630
4631 #undef LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS
4632
4633 void MacroAssembler::clinit_barrier(Register klass, Label* L_fast_path, Label* L_slow_path) {
4634 assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
4635
4636 Label L_fallthrough;
4637 if (L_fast_path == nullptr) {
4638 L_fast_path = &L_fallthrough;
4639 } else if (L_slow_path == nullptr) {
4640 L_slow_path = &L_fallthrough;
4641 }
4642
4643 // Fast path check: class is fully initialized.
4644 // init_state needs acquire, but x86 is TSO, and so we are already good.
4645 cmpb(Address(klass, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
4646 jcc(Assembler::equal, *L_fast_path);
4647
4648 // Fast path check: current thread is initializer thread
4649 cmpptr(r15_thread, Address(klass, InstanceKlass::init_thread_offset()));
4650 if (L_slow_path == &L_fallthrough) {
4651 jcc(Assembler::equal, *L_fast_path);
4652 bind(*L_slow_path);
4653 } else if (L_fast_path == &L_fallthrough) {
4654 jcc(Assembler::notEqual, *L_slow_path);
4655 bind(*L_fast_path);
4656 } else {
4657 Unimplemented();
4658 }
4659 }
4660
4661 void MacroAssembler::cmov32(Condition cc, Register dst, Address src) {
4662 if (VM_Version::supports_cmov()) {
4663 cmovl(cc, dst, src);
4664 } else {
4665 Label L;
4666 jccb(negate_condition(cc), L);
4667 movl(dst, src);
4668 bind(L);
4669 }
4670 }
4671
4672 void MacroAssembler::cmov32(Condition cc, Register dst, Register src) {
4673 if (VM_Version::supports_cmov()) {
4674 cmovl(cc, dst, src);
4675 } else {
4676 Label L;
4677 jccb(negate_condition(cc), L);
4678 movl(dst, src);
4679 bind(L);
4680 }
4681 }
4682
4683 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
4684 if (!VerifyOops) return;
4685
4686 BLOCK_COMMENT("verify_oop {");
4687 push(rscratch1);
4688 push(rax); // save rax
4689 push(reg); // pass register argument
4690
4691 // Pass register number to verify_oop_subroutine
4692 const char* b = nullptr;
4693 {
4694 ResourceMark rm;
4695 stringStream ss;
4696 ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
4697 b = code_string(ss.as_string());
4698 }
4699 AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
4700 pushptr(buffer.addr(), rscratch1);
4701
4702 // call indirectly to solve generation ordering problem
4703 movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
4704 call(rax);
4705 // Caller pops the arguments (oop, message) and restores rax, r10
4706 BLOCK_COMMENT("} verify_oop");
4707 }
4708
4709 void MacroAssembler::vallones(XMMRegister dst, int vector_len) {
4710 if (UseAVX > 2 && (vector_len == Assembler::AVX_512bit || VM_Version::supports_avx512vl())) {
4711 // Only pcmpeq has dependency breaking treatment (i.e the execution can begin without
4712 // waiting for the previous result on dst), not vpcmpeqd, so just use vpternlog
4713 vpternlogd(dst, 0xFF, dst, dst, vector_len);
4714 } else if (VM_Version::supports_avx()) {
4715 vpcmpeqd(dst, dst, dst, vector_len);
4716 } else {
4717 pcmpeqd(dst, dst);
4718 }
4719 }
4720
4721 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
4722 int extra_slot_offset) {
4723 // cf. TemplateTable::prepare_invoke(), if (load_receiver).
4724 int stackElementSize = Interpreter::stackElementSize;
4725 int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
4726 #ifdef ASSERT
4727 int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
4728 assert(offset1 - offset == stackElementSize, "correct arithmetic");
4729 #endif
4730 Register scale_reg = noreg;
4731 Address::ScaleFactor scale_factor = Address::no_scale;
4732 if (arg_slot.is_constant()) {
4733 offset += arg_slot.as_constant() * stackElementSize;
4734 } else {
4735 scale_reg = arg_slot.as_register();
4736 scale_factor = Address::times(stackElementSize);
4737 }
4738 offset += wordSize; // return PC is on stack
4739 return Address(rsp, scale_reg, scale_factor, offset);
4740 }
4741
4742 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
4743 if (!VerifyOops) return;
4744
4745 push(rscratch1);
4746 push(rax); // save rax,
4747 // addr may contain rsp so we will have to adjust it based on the push
4748 // we just did (and on 64 bit we do two pushes)
4749 // NOTE: 64bit seemed to have had a bug in that it did movq(addr, rax); which
4750 // stores rax into addr which is backwards of what was intended.
4751 if (addr.uses(rsp)) {
4752 lea(rax, addr);
4753 pushptr(Address(rax, 2 * BytesPerWord));
4754 } else {
4755 pushptr(addr);
4756 }
4757
4758 // Pass register number to verify_oop_subroutine
4759 const char* b = nullptr;
4760 {
4761 ResourceMark rm;
4762 stringStream ss;
4763 ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
4764 b = code_string(ss.as_string());
4765 }
4766 AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
4767 pushptr(buffer.addr(), rscratch1);
4768
4769 // call indirectly to solve generation ordering problem
4770 movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
4771 call(rax);
4772 // Caller pops the arguments (addr, message) and restores rax, r10.
4773 }
4774
4775 void MacroAssembler::verify_tlab() {
4776 #ifdef ASSERT
4777 if (UseTLAB && VerifyOops) {
4778 Label next, ok;
4779 Register t1 = rsi;
4780
4781 push(t1);
4782
4783 movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
4784 cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_start_offset())));
4785 jcc(Assembler::aboveEqual, next);
4786 STOP("assert(top >= start)");
4787 should_not_reach_here();
4788
4789 bind(next);
4790 movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_end_offset())));
4791 cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
4792 jcc(Assembler::aboveEqual, ok);
4793 STOP("assert(top <= end)");
4794 should_not_reach_here();
4795
4796 bind(ok);
4797 pop(t1);
4798 }
4799 #endif
4800 }
4801
4802 class ControlWord {
4803 public:
4804 int32_t _value;
4805
4806 int rounding_control() const { return (_value >> 10) & 3 ; }
4807 int precision_control() const { return (_value >> 8) & 3 ; }
4808 bool precision() const { return ((_value >> 5) & 1) != 0; }
4809 bool underflow() const { return ((_value >> 4) & 1) != 0; }
4810 bool overflow() const { return ((_value >> 3) & 1) != 0; }
4811 bool zero_divide() const { return ((_value >> 2) & 1) != 0; }
4812 bool denormalized() const { return ((_value >> 1) & 1) != 0; }
4813 bool invalid() const { return ((_value >> 0) & 1) != 0; }
4814
4815 void print() const {
4816 // rounding control
4817 const char* rc;
4818 switch (rounding_control()) {
4819 case 0: rc = "round near"; break;
4820 case 1: rc = "round down"; break;
4821 case 2: rc = "round up "; break;
4822 case 3: rc = "chop "; break;
4823 default:
4824 rc = nullptr; // silence compiler warnings
4825 fatal("Unknown rounding control: %d", rounding_control());
4826 };
4827 // precision control
4828 const char* pc;
4829 switch (precision_control()) {
4830 case 0: pc = "24 bits "; break;
4831 case 1: pc = "reserved"; break;
4832 case 2: pc = "53 bits "; break;
4833 case 3: pc = "64 bits "; break;
4834 default:
4835 pc = nullptr; // silence compiler warnings
4836 fatal("Unknown precision control: %d", precision_control());
4837 };
4838 // flags
4839 char f[9];
4840 f[0] = ' ';
4841 f[1] = ' ';
4842 f[2] = (precision ()) ? 'P' : 'p';
4843 f[3] = (underflow ()) ? 'U' : 'u';
4844 f[4] = (overflow ()) ? 'O' : 'o';
4845 f[5] = (zero_divide ()) ? 'Z' : 'z';
4846 f[6] = (denormalized()) ? 'D' : 'd';
4847 f[7] = (invalid ()) ? 'I' : 'i';
4848 f[8] = '\x0';
4849 // output
4850 printf("%04x masks = %s, %s, %s", _value & 0xFFFF, f, rc, pc);
4851 }
4852
4853 };
4854
4855 class StatusWord {
4856 public:
4857 int32_t _value;
4858
4859 bool busy() const { return ((_value >> 15) & 1) != 0; }
4860 bool C3() const { return ((_value >> 14) & 1) != 0; }
4861 bool C2() const { return ((_value >> 10) & 1) != 0; }
4862 bool C1() const { return ((_value >> 9) & 1) != 0; }
4863 bool C0() const { return ((_value >> 8) & 1) != 0; }
4864 int top() const { return (_value >> 11) & 7 ; }
4865 bool error_status() const { return ((_value >> 7) & 1) != 0; }
4866 bool stack_fault() const { return ((_value >> 6) & 1) != 0; }
4867 bool precision() const { return ((_value >> 5) & 1) != 0; }
4868 bool underflow() const { return ((_value >> 4) & 1) != 0; }
4869 bool overflow() const { return ((_value >> 3) & 1) != 0; }
4870 bool zero_divide() const { return ((_value >> 2) & 1) != 0; }
4871 bool denormalized() const { return ((_value >> 1) & 1) != 0; }
4872 bool invalid() const { return ((_value >> 0) & 1) != 0; }
4873
4874 void print() const {
4875 // condition codes
4876 char c[5];
4877 c[0] = (C3()) ? '3' : '-';
4878 c[1] = (C2()) ? '2' : '-';
4879 c[2] = (C1()) ? '1' : '-';
4880 c[3] = (C0()) ? '0' : '-';
4881 c[4] = '\x0';
4882 // flags
4883 char f[9];
4884 f[0] = (error_status()) ? 'E' : '-';
4885 f[1] = (stack_fault ()) ? 'S' : '-';
4886 f[2] = (precision ()) ? 'P' : '-';
4887 f[3] = (underflow ()) ? 'U' : '-';
4888 f[4] = (overflow ()) ? 'O' : '-';
4889 f[5] = (zero_divide ()) ? 'Z' : '-';
4890 f[6] = (denormalized()) ? 'D' : '-';
4891 f[7] = (invalid ()) ? 'I' : '-';
4892 f[8] = '\x0';
4893 // output
4894 printf("%04x flags = %s, cc = %s, top = %d", _value & 0xFFFF, f, c, top());
4895 }
4896
4897 };
4898
4899 class TagWord {
4900 public:
4901 int32_t _value;
4902
4903 int tag_at(int i) const { return (_value >> (i*2)) & 3; }
4904
4905 void print() const {
4906 printf("%04x", _value & 0xFFFF);
4907 }
4908
4909 };
4910
4911 class FPU_Register {
4912 public:
4913 int32_t _m0;
4914 int32_t _m1;
4915 int16_t _ex;
4916
4917 bool is_indefinite() const {
4918 return _ex == -1 && _m1 == (int32_t)0xC0000000 && _m0 == 0;
4919 }
4920
4921 void print() const {
4922 char sign = (_ex < 0) ? '-' : '+';
4923 const char* kind = (_ex == 0x7FFF || _ex == (int16_t)-1) ? "NaN" : " ";
4924 printf("%c%04hx.%08x%08x %s", sign, _ex, _m1, _m0, kind);
4925 };
4926
4927 };
4928
4929 class FPU_State {
4930 public:
4931 enum {
4932 register_size = 10,
4933 number_of_registers = 8,
4934 register_mask = 7
4935 };
4936
4937 ControlWord _control_word;
4938 StatusWord _status_word;
4939 TagWord _tag_word;
4940 int32_t _error_offset;
4941 int32_t _error_selector;
4942 int32_t _data_offset;
4943 int32_t _data_selector;
4944 int8_t _register[register_size * number_of_registers];
4945
4946 int tag_for_st(int i) const { return _tag_word.tag_at((_status_word.top() + i) & register_mask); }
4947 FPU_Register* st(int i) const { return (FPU_Register*)&_register[register_size * i]; }
4948
4949 const char* tag_as_string(int tag) const {
4950 switch (tag) {
4951 case 0: return "valid";
4952 case 1: return "zero";
4953 case 2: return "special";
4954 case 3: return "empty";
4955 }
4956 ShouldNotReachHere();
4957 return nullptr;
4958 }
4959
4960 void print() const {
4961 // print computation registers
4962 { int t = _status_word.top();
4963 for (int i = 0; i < number_of_registers; i++) {
4964 int j = (i - t) & register_mask;
4965 printf("%c r%d = ST%d = ", (j == 0 ? '*' : ' '), i, j);
4966 st(j)->print();
4967 printf(" %s\n", tag_as_string(_tag_word.tag_at(i)));
4968 }
4969 }
4970 printf("\n");
4971 // print control registers
4972 printf("ctrl = "); _control_word.print(); printf("\n");
4973 printf("stat = "); _status_word .print(); printf("\n");
4974 printf("tags = "); _tag_word .print(); printf("\n");
4975 }
4976
4977 };
4978
4979 class Flag_Register {
4980 public:
4981 int32_t _value;
4982
4983 bool overflow() const { return ((_value >> 11) & 1) != 0; }
4984 bool direction() const { return ((_value >> 10) & 1) != 0; }
4985 bool sign() const { return ((_value >> 7) & 1) != 0; }
4986 bool zero() const { return ((_value >> 6) & 1) != 0; }
4987 bool auxiliary_carry() const { return ((_value >> 4) & 1) != 0; }
4988 bool parity() const { return ((_value >> 2) & 1) != 0; }
4989 bool carry() const { return ((_value >> 0) & 1) != 0; }
4990
4991 void print() const {
4992 // flags
4993 char f[8];
4994 f[0] = (overflow ()) ? 'O' : '-';
4995 f[1] = (direction ()) ? 'D' : '-';
4996 f[2] = (sign ()) ? 'S' : '-';
4997 f[3] = (zero ()) ? 'Z' : '-';
4998 f[4] = (auxiliary_carry()) ? 'A' : '-';
4999 f[5] = (parity ()) ? 'P' : '-';
5000 f[6] = (carry ()) ? 'C' : '-';
5001 f[7] = '\x0';
5002 // output
5003 printf("%08x flags = %s", _value, f);
5004 }
5005
5006 };
5007
5008 class IU_Register {
5009 public:
5010 int32_t _value;
5011
5012 void print() const {
5013 printf("%08x %11d", _value, _value);
5014 }
5015
5016 };
5017
5018 class IU_State {
5019 public:
5020 Flag_Register _eflags;
5021 IU_Register _rdi;
5022 IU_Register _rsi;
5023 IU_Register _rbp;
5024 IU_Register _rsp;
5025 IU_Register _rbx;
5026 IU_Register _rdx;
5027 IU_Register _rcx;
5028 IU_Register _rax;
5029
5030 void print() const {
5031 // computation registers
5032 printf("rax, = "); _rax.print(); printf("\n");
5033 printf("rbx, = "); _rbx.print(); printf("\n");
5034 printf("rcx = "); _rcx.print(); printf("\n");
5035 printf("rdx = "); _rdx.print(); printf("\n");
5036 printf("rdi = "); _rdi.print(); printf("\n");
5037 printf("rsi = "); _rsi.print(); printf("\n");
5038 printf("rbp, = "); _rbp.print(); printf("\n");
5039 printf("rsp = "); _rsp.print(); printf("\n");
5040 printf("\n");
5041 // control registers
5042 printf("flgs = "); _eflags.print(); printf("\n");
5043 }
5044 };
5045
5046
5047 class CPU_State {
5048 public:
5049 FPU_State _fpu_state;
5050 IU_State _iu_state;
5051
5052 void print() const {
5053 printf("--------------------------------------------------\n");
5054 _iu_state .print();
5055 printf("\n");
5056 _fpu_state.print();
5057 printf("--------------------------------------------------\n");
5058 }
5059
5060 };
5061
5062
5063 static void _print_CPU_state(CPU_State* state) {
5064 state->print();
5065 };
5066
5067
5068 void MacroAssembler::print_CPU_state() {
5069 push_CPU_state();
5070 push(rsp); // pass CPU state
5071 call(RuntimeAddress(CAST_FROM_FN_PTR(address, _print_CPU_state)));
5072 addptr(rsp, wordSize); // discard argument
5073 pop_CPU_state();
5074 }
5075
5076 void MacroAssembler::restore_cpu_control_state_after_jni(Register rscratch) {
5077 // Either restore the MXCSR register after returning from the JNI Call
5078 // or verify that it wasn't changed (with -Xcheck:jni flag).
5079 if (VM_Version::supports_sse()) {
5080 if (RestoreMXCSROnJNICalls) {
5081 ldmxcsr(ExternalAddress(StubRoutines::x86::addr_mxcsr_std()), rscratch);
5082 } else if (CheckJNICalls) {
5083 call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
5084 }
5085 }
5086 // Clear upper bits of YMM registers to avoid SSE <-> AVX transition penalty.
5087 vzeroupper();
5088 }
5089
5090 // ((OopHandle)result).resolve();
5091 void MacroAssembler::resolve_oop_handle(Register result, Register tmp) {
5092 assert_different_registers(result, tmp);
5093
5094 // Only 64 bit platforms support GCs that require a tmp register
5095 // Only IN_HEAP loads require a thread_tmp register
5096 // OopHandle::resolve is an indirection like jobject.
5097 access_load_at(T_OBJECT, IN_NATIVE,
5098 result, Address(result, 0), tmp);
5099 }
5100
5101 // ((WeakHandle)result).resolve();
5102 void MacroAssembler::resolve_weak_handle(Register rresult, Register rtmp) {
5103 assert_different_registers(rresult, rtmp);
5104 Label resolved;
5105
5106 // A null weak handle resolves to null.
5107 cmpptr(rresult, 0);
5108 jcc(Assembler::equal, resolved);
5109
5110 // Only 64 bit platforms support GCs that require a tmp register
5111 // Only IN_HEAP loads require a thread_tmp register
5112 // WeakHandle::resolve is an indirection like jweak.
5113 access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
5114 rresult, Address(rresult, 0), rtmp);
5115 bind(resolved);
5116 }
5117
5118 void MacroAssembler::load_mirror(Register mirror, Register method, Register tmp) {
5119 // get mirror
5120 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
5121 load_method_holder(mirror, method);
5122 movptr(mirror, Address(mirror, mirror_offset));
5123 resolve_oop_handle(mirror, tmp);
5124 }
5125
5126 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5127 load_method_holder(rresult, rmethod);
5128 movptr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5129 }
5130
5131 void MacroAssembler::load_method_holder(Register holder, Register method) {
5132 movptr(holder, Address(method, Method::const_offset())); // ConstMethod*
5133 movptr(holder, Address(holder, ConstMethod::constants_offset())); // ConstantPool*
5134 movptr(holder, Address(holder, ConstantPool::pool_holder_offset())); // InstanceKlass*
5135 }
5136
5137 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5138 assert(UseCompactObjectHeaders, "expect compact object headers");
5139 movq(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5140 shrq(dst, markWord::klass_shift);
5141 }
5142
5143 void MacroAssembler::load_klass(Register dst, Register src, Register tmp) {
5144 assert_different_registers(src, tmp);
5145 assert_different_registers(dst, tmp);
5146
5147 if (UseCompactObjectHeaders) {
5148 load_narrow_klass_compact(dst, src);
5149 decode_klass_not_null(dst, tmp);
5150 } else if (UseCompressedClassPointers) {
5151 movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5152 decode_klass_not_null(dst, tmp);
5153 } else {
5154 movptr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5155 }
5156 }
5157
5158 void MacroAssembler::store_klass(Register dst, Register src, Register tmp) {
5159 assert(!UseCompactObjectHeaders, "not with compact headers");
5160 assert_different_registers(src, tmp);
5161 assert_different_registers(dst, tmp);
5162 if (UseCompressedClassPointers) {
5163 encode_klass_not_null(src, tmp);
5164 movl(Address(dst, oopDesc::klass_offset_in_bytes()), src);
5165 } else {
5166 movptr(Address(dst, oopDesc::klass_offset_in_bytes()), src);
5167 }
5168 }
5169
5170 void MacroAssembler::cmp_klass(Register klass, Register obj, Register tmp) {
5171 if (UseCompactObjectHeaders) {
5172 assert(tmp != noreg, "need tmp");
5173 assert_different_registers(klass, obj, tmp);
5174 load_narrow_klass_compact(tmp, obj);
5175 cmpl(klass, tmp);
5176 } else if (UseCompressedClassPointers) {
5177 cmpl(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
5178 } else {
5179 cmpptr(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
5180 }
5181 }
5182
5183 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5184 if (UseCompactObjectHeaders) {
5185 assert(tmp2 != noreg, "need tmp2");
5186 assert_different_registers(obj1, obj2, tmp1, tmp2);
5187 load_narrow_klass_compact(tmp1, obj1);
5188 load_narrow_klass_compact(tmp2, obj2);
5189 cmpl(tmp1, tmp2);
5190 } else if (UseCompressedClassPointers) {
5191 movl(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5192 cmpl(tmp1, Address(obj2, oopDesc::klass_offset_in_bytes()));
5193 } else {
5194 movptr(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5195 cmpptr(tmp1, Address(obj2, oopDesc::klass_offset_in_bytes()));
5196 }
5197 }
5198
5199 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
5200 Register tmp1) {
5201 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
5202 decorators = AccessInternal::decorator_fixup(decorators, type);
5203 bool as_raw = (decorators & AS_RAW) != 0;
5204 if (as_raw) {
5205 bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1);
5206 } else {
5207 bs->load_at(this, decorators, type, dst, src, tmp1);
5208 }
5209 }
5210
5211 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
5212 Register tmp1, Register tmp2, Register tmp3) {
5213 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
5214 decorators = AccessInternal::decorator_fixup(decorators, type);
5215 bool as_raw = (decorators & AS_RAW) != 0;
5216 if (as_raw) {
5217 bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5218 } else {
5219 bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5220 }
5221 }
5222
5223 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
5224 access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1);
5225 }
5226
5227 // Doesn't do verification, generates fixed size code
5228 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
5229 access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1);
5230 }
5231
5232 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5233 Register tmp2, Register tmp3, DecoratorSet decorators) {
5234 access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5235 }
5236
5237 // Used for storing nulls.
5238 void MacroAssembler::store_heap_oop_null(Address dst) {
5239 access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5240 }
5241
5242 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5243 assert(!UseCompactObjectHeaders, "Don't use with compact headers");
5244 if (UseCompressedClassPointers) {
5245 // Store to klass gap in destination
5246 movl(Address(dst, oopDesc::klass_gap_offset_in_bytes()), src);
5247 }
5248 }
5249
5250 #ifdef ASSERT
5251 void MacroAssembler::verify_heapbase(const char* msg) {
5252 assert (UseCompressedOops, "should be compressed");
5253 assert (Universe::heap() != nullptr, "java heap should be initialized");
5254 if (CheckCompressedOops) {
5255 Label ok;
5256 ExternalAddress src2(CompressedOops::base_addr());
5257 const bool is_src2_reachable = reachable(src2);
5258 if (!is_src2_reachable) {
5259 push(rscratch1); // cmpptr trashes rscratch1
5260 }
5261 cmpptr(r12_heapbase, src2, rscratch1);
5262 jcc(Assembler::equal, ok);
5263 STOP(msg);
5264 bind(ok);
5265 if (!is_src2_reachable) {
5266 pop(rscratch1);
5267 }
5268 }
5269 }
5270 #endif
5271
5272 // Algorithm must match oop.inline.hpp encode_heap_oop.
5273 void MacroAssembler::encode_heap_oop(Register r) {
5274 #ifdef ASSERT
5275 verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5276 #endif
5277 verify_oop_msg(r, "broken oop in encode_heap_oop");
5278 if (CompressedOops::base() == nullptr) {
5279 if (CompressedOops::shift() != 0) {
5280 assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5281 shrq(r, LogMinObjAlignmentInBytes);
5282 }
5283 return;
5284 }
5285 testq(r, r);
5286 cmovq(Assembler::equal, r, r12_heapbase);
5287 subq(r, r12_heapbase);
5288 shrq(r, LogMinObjAlignmentInBytes);
5289 }
5290
5291 void MacroAssembler::encode_heap_oop_not_null(Register r) {
5292 #ifdef ASSERT
5293 verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
5294 if (CheckCompressedOops) {
5295 Label ok;
5296 testq(r, r);
5297 jcc(Assembler::notEqual, ok);
5298 STOP("null oop passed to encode_heap_oop_not_null");
5299 bind(ok);
5300 }
5301 #endif
5302 verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
5303 if (CompressedOops::base() != nullptr) {
5304 subq(r, r12_heapbase);
5305 }
5306 if (CompressedOops::shift() != 0) {
5307 assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5308 shrq(r, LogMinObjAlignmentInBytes);
5309 }
5310 }
5311
5312 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
5313 #ifdef ASSERT
5314 verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
5315 if (CheckCompressedOops) {
5316 Label ok;
5317 testq(src, src);
5318 jcc(Assembler::notEqual, ok);
5319 STOP("null oop passed to encode_heap_oop_not_null2");
5320 bind(ok);
5321 }
5322 #endif
5323 verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
5324 if (dst != src) {
5325 movq(dst, src);
5326 }
5327 if (CompressedOops::base() != nullptr) {
5328 subq(dst, r12_heapbase);
5329 }
5330 if (CompressedOops::shift() != 0) {
5331 assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5332 shrq(dst, LogMinObjAlignmentInBytes);
5333 }
5334 }
5335
5336 void MacroAssembler::decode_heap_oop(Register r) {
5337 #ifdef ASSERT
5338 verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
5339 #endif
5340 if (CompressedOops::base() == nullptr) {
5341 if (CompressedOops::shift() != 0) {
5342 assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5343 shlq(r, LogMinObjAlignmentInBytes);
5344 }
5345 } else {
5346 Label done;
5347 shlq(r, LogMinObjAlignmentInBytes);
5348 jccb(Assembler::equal, done);
5349 addq(r, r12_heapbase);
5350 bind(done);
5351 }
5352 verify_oop_msg(r, "broken oop in decode_heap_oop");
5353 }
5354
5355 void MacroAssembler::decode_heap_oop_not_null(Register r) {
5356 // Note: it will change flags
5357 assert (UseCompressedOops, "should only be used for compressed headers");
5358 assert (Universe::heap() != nullptr, "java heap should be initialized");
5359 // Cannot assert, unverified entry point counts instructions (see .ad file)
5360 // vtableStubs also counts instructions in pd_code_size_limit.
5361 // Also do not verify_oop as this is called by verify_oop.
5362 if (CompressedOops::shift() != 0) {
5363 assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5364 shlq(r, LogMinObjAlignmentInBytes);
5365 if (CompressedOops::base() != nullptr) {
5366 addq(r, r12_heapbase);
5367 }
5368 } else {
5369 assert (CompressedOops::base() == nullptr, "sanity");
5370 }
5371 }
5372
5373 void MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
5374 // Note: it will change flags
5375 assert (UseCompressedOops, "should only be used for compressed headers");
5376 assert (Universe::heap() != nullptr, "java heap should be initialized");
5377 // Cannot assert, unverified entry point counts instructions (see .ad file)
5378 // vtableStubs also counts instructions in pd_code_size_limit.
5379 // Also do not verify_oop as this is called by verify_oop.
5380 if (CompressedOops::shift() != 0) {
5381 assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5382 if (LogMinObjAlignmentInBytes == Address::times_8) {
5383 leaq(dst, Address(r12_heapbase, src, Address::times_8, 0));
5384 } else {
5385 if (dst != src) {
5386 movq(dst, src);
5387 }
5388 shlq(dst, LogMinObjAlignmentInBytes);
5389 if (CompressedOops::base() != nullptr) {
5390 addq(dst, r12_heapbase);
5391 }
5392 }
5393 } else {
5394 assert (CompressedOops::base() == nullptr, "sanity");
5395 if (dst != src) {
5396 movq(dst, src);
5397 }
5398 }
5399 }
5400
5401 void MacroAssembler::encode_klass_not_null(Register r, Register tmp) {
5402 assert_different_registers(r, tmp);
5403 if (CompressedKlassPointers::base() != nullptr) {
5404 mov64(tmp, (int64_t)CompressedKlassPointers::base());
5405 subq(r, tmp);
5406 }
5407 if (CompressedKlassPointers::shift() != 0) {
5408 shrq(r, CompressedKlassPointers::shift());
5409 }
5410 }
5411
5412 void MacroAssembler::encode_and_move_klass_not_null(Register dst, Register src) {
5413 assert_different_registers(src, dst);
5414 if (CompressedKlassPointers::base() != nullptr) {
5415 mov64(dst, -(int64_t)CompressedKlassPointers::base());
5416 addq(dst, src);
5417 } else {
5418 movptr(dst, src);
5419 }
5420 if (CompressedKlassPointers::shift() != 0) {
5421 shrq(dst, CompressedKlassPointers::shift());
5422 }
5423 }
5424
5425 void MacroAssembler::decode_klass_not_null(Register r, Register tmp) {
5426 assert_different_registers(r, tmp);
5427 // Note: it will change flags
5428 assert(UseCompressedClassPointers, "should only be used for compressed headers");
5429 // Cannot assert, unverified entry point counts instructions (see .ad file)
5430 // vtableStubs also counts instructions in pd_code_size_limit.
5431 // Also do not verify_oop as this is called by verify_oop.
5432 if (CompressedKlassPointers::shift() != 0) {
5433 shlq(r, CompressedKlassPointers::shift());
5434 }
5435 if (CompressedKlassPointers::base() != nullptr) {
5436 mov64(tmp, (int64_t)CompressedKlassPointers::base());
5437 addq(r, tmp);
5438 }
5439 }
5440
5441 void MacroAssembler::decode_and_move_klass_not_null(Register dst, Register src) {
5442 assert_different_registers(src, dst);
5443 // Note: it will change flags
5444 assert (UseCompressedClassPointers, "should only be used for compressed headers");
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
5449 if (CompressedKlassPointers::base() == nullptr &&
5450 CompressedKlassPointers::shift() == 0) {
5451 // The best case scenario is that there is no base or shift. Then it is already
5452 // a pointer that needs nothing but a register rename.
5453 movl(dst, src);
5454 } else {
5455 if (CompressedKlassPointers::shift() <= Address::times_8) {
5456 if (CompressedKlassPointers::base() != nullptr) {
5457 mov64(dst, (int64_t)CompressedKlassPointers::base());
5458 } else {
5459 xorq(dst, dst);
5460 }
5461 if (CompressedKlassPointers::shift() != 0) {
5462 assert(CompressedKlassPointers::shift() == Address::times_8, "klass not aligned on 64bits?");
5463 leaq(dst, Address(dst, src, Address::times_8, 0));
5464 } else {
5465 addq(dst, src);
5466 }
5467 } else {
5468 if (CompressedKlassPointers::base() != nullptr) {
5469 const uint64_t base_right_shifted =
5470 (uint64_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
5471 mov64(dst, base_right_shifted);
5472 } else {
5473 xorq(dst, dst);
5474 }
5475 addq(dst, src);
5476 shlq(dst, CompressedKlassPointers::shift());
5477 }
5478 }
5479 }
5480
5481 void MacroAssembler::set_narrow_oop(Register dst, jobject obj) {
5482 assert (UseCompressedOops, "should only be used for compressed headers");
5483 assert (Universe::heap() != nullptr, "java heap should be initialized");
5484 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5485 int oop_index = oop_recorder()->find_index(obj);
5486 RelocationHolder rspec = oop_Relocation::spec(oop_index);
5487 mov_narrow_oop(dst, oop_index, rspec);
5488 }
5489
5490 void MacroAssembler::set_narrow_oop(Address dst, jobject obj) {
5491 assert (UseCompressedOops, "should only be used for compressed headers");
5492 assert (Universe::heap() != nullptr, "java heap should be initialized");
5493 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5494 int oop_index = oop_recorder()->find_index(obj);
5495 RelocationHolder rspec = oop_Relocation::spec(oop_index);
5496 mov_narrow_oop(dst, oop_index, rspec);
5497 }
5498
5499 void MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
5500 assert (UseCompressedClassPointers, "should only be used for compressed headers");
5501 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5502 int klass_index = oop_recorder()->find_index(k);
5503 RelocationHolder rspec = metadata_Relocation::spec(klass_index);
5504 mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
5505 }
5506
5507 void MacroAssembler::set_narrow_klass(Address dst, Klass* k) {
5508 assert (UseCompressedClassPointers, "should only be used for compressed headers");
5509 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5510 int klass_index = oop_recorder()->find_index(k);
5511 RelocationHolder rspec = metadata_Relocation::spec(klass_index);
5512 mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
5513 }
5514
5515 void MacroAssembler::cmp_narrow_oop(Register dst, jobject obj) {
5516 assert (UseCompressedOops, "should only be used for compressed headers");
5517 assert (Universe::heap() != nullptr, "java heap should be initialized");
5518 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5519 int oop_index = oop_recorder()->find_index(obj);
5520 RelocationHolder rspec = oop_Relocation::spec(oop_index);
5521 Assembler::cmp_narrow_oop(dst, oop_index, rspec);
5522 }
5523
5524 void MacroAssembler::cmp_narrow_oop(Address dst, jobject obj) {
5525 assert (UseCompressedOops, "should only be used for compressed headers");
5526 assert (Universe::heap() != nullptr, "java heap should be initialized");
5527 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5528 int oop_index = oop_recorder()->find_index(obj);
5529 RelocationHolder rspec = oop_Relocation::spec(oop_index);
5530 Assembler::cmp_narrow_oop(dst, oop_index, rspec);
5531 }
5532
5533 void MacroAssembler::cmp_narrow_klass(Register dst, Klass* k) {
5534 assert (UseCompressedClassPointers, "should only be used for compressed headers");
5535 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5536 int klass_index = oop_recorder()->find_index(k);
5537 RelocationHolder rspec = metadata_Relocation::spec(klass_index);
5538 Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
5539 }
5540
5541 void MacroAssembler::cmp_narrow_klass(Address dst, Klass* k) {
5542 assert (UseCompressedClassPointers, "should only be used for compressed headers");
5543 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5544 int klass_index = oop_recorder()->find_index(k);
5545 RelocationHolder rspec = metadata_Relocation::spec(klass_index);
5546 Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
5547 }
5548
5549 void MacroAssembler::reinit_heapbase() {
5550 if (UseCompressedOops) {
5551 if (Universe::heap() != nullptr) {
5552 if (CompressedOops::base() == nullptr) {
5553 MacroAssembler::xorptr(r12_heapbase, r12_heapbase);
5554 } else {
5555 mov64(r12_heapbase, (int64_t)CompressedOops::base());
5556 }
5557 } else {
5558 movptr(r12_heapbase, ExternalAddress(CompressedOops::base_addr()));
5559 }
5560 }
5561 }
5562
5563 #if COMPILER2_OR_JVMCI
5564
5565 // clear memory of size 'cnt' qwords, starting at 'base' using XMM/YMM/ZMM registers
5566 void MacroAssembler::xmm_clear_mem(Register base, Register cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
5567 // cnt - number of qwords (8-byte words).
5568 // base - start address, qword aligned.
5569 Label L_zero_64_bytes, L_loop, L_sloop, L_tail, L_end;
5570 bool use64byteVector = (MaxVectorSize == 64) && (VM_Version::avx3_threshold() == 0);
5571 if (use64byteVector) {
5572 vpxor(xtmp, xtmp, xtmp, AVX_512bit);
5573 } else if (MaxVectorSize >= 32) {
5574 vpxor(xtmp, xtmp, xtmp, AVX_256bit);
5575 } else {
5576 pxor(xtmp, xtmp);
5577 }
5578 jmp(L_zero_64_bytes);
5579
5580 BIND(L_loop);
5581 if (MaxVectorSize >= 32) {
5582 fill64(base, 0, xtmp, use64byteVector);
5583 } else {
5584 movdqu(Address(base, 0), xtmp);
5585 movdqu(Address(base, 16), xtmp);
5586 movdqu(Address(base, 32), xtmp);
5587 movdqu(Address(base, 48), xtmp);
5588 }
5589 addptr(base, 64);
5590
5591 BIND(L_zero_64_bytes);
5592 subptr(cnt, 8);
5593 jccb(Assembler::greaterEqual, L_loop);
5594
5595 // Copy trailing 64 bytes
5596 if (use64byteVector) {
5597 addptr(cnt, 8);
5598 jccb(Assembler::equal, L_end);
5599 fill64_masked(3, base, 0, xtmp, mask, cnt, rtmp, true);
5600 jmp(L_end);
5601 } else {
5602 addptr(cnt, 4);
5603 jccb(Assembler::less, L_tail);
5604 if (MaxVectorSize >= 32) {
5605 vmovdqu(Address(base, 0), xtmp);
5606 } else {
5607 movdqu(Address(base, 0), xtmp);
5608 movdqu(Address(base, 16), xtmp);
5609 }
5610 }
5611 addptr(base, 32);
5612 subptr(cnt, 4);
5613
5614 BIND(L_tail);
5615 addptr(cnt, 4);
5616 jccb(Assembler::lessEqual, L_end);
5617 if (UseAVX > 2 && MaxVectorSize >= 32 && VM_Version::supports_avx512vl()) {
5618 fill32_masked(3, base, 0, xtmp, mask, cnt, rtmp);
5619 } else {
5620 decrement(cnt);
5621
5622 BIND(L_sloop);
5623 movq(Address(base, 0), xtmp);
5624 addptr(base, 8);
5625 decrement(cnt);
5626 jccb(Assembler::greaterEqual, L_sloop);
5627 }
5628 BIND(L_end);
5629 }
5630
5631 // Clearing constant sized memory using YMM/ZMM registers.
5632 void MacroAssembler::clear_mem(Register base, int cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
5633 assert(UseAVX > 2 && VM_Version::supports_avx512vl(), "");
5634 bool use64byteVector = (MaxVectorSize > 32) && (VM_Version::avx3_threshold() == 0);
5635
5636 int vector64_count = (cnt & (~0x7)) >> 3;
5637 cnt = cnt & 0x7;
5638 const int fill64_per_loop = 4;
5639 const int max_unrolled_fill64 = 8;
5640
5641 // 64 byte initialization loop.
5642 vpxor(xtmp, xtmp, xtmp, use64byteVector ? AVX_512bit : AVX_256bit);
5643 int start64 = 0;
5644 if (vector64_count > max_unrolled_fill64) {
5645 Label LOOP;
5646 Register index = rtmp;
5647
5648 start64 = vector64_count - (vector64_count % fill64_per_loop);
5649
5650 movl(index, 0);
5651 BIND(LOOP);
5652 for (int i = 0; i < fill64_per_loop; i++) {
5653 fill64(Address(base, index, Address::times_1, i * 64), xtmp, use64byteVector);
5654 }
5655 addl(index, fill64_per_loop * 64);
5656 cmpl(index, start64 * 64);
5657 jccb(Assembler::less, LOOP);
5658 }
5659 for (int i = start64; i < vector64_count; i++) {
5660 fill64(base, i * 64, xtmp, use64byteVector);
5661 }
5662
5663 // Clear remaining 64 byte tail.
5664 int disp = vector64_count * 64;
5665 if (cnt) {
5666 switch (cnt) {
5667 case 1:
5668 movq(Address(base, disp), xtmp);
5669 break;
5670 case 2:
5671 evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_128bit);
5672 break;
5673 case 3:
5674 movl(rtmp, 0x7);
5675 kmovwl(mask, rtmp);
5676 evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_256bit);
5677 break;
5678 case 4:
5679 evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
5680 break;
5681 case 5:
5682 if (use64byteVector) {
5683 movl(rtmp, 0x1F);
5684 kmovwl(mask, rtmp);
5685 evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
5686 } else {
5687 evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
5688 movq(Address(base, disp + 32), xtmp);
5689 }
5690 break;
5691 case 6:
5692 if (use64byteVector) {
5693 movl(rtmp, 0x3F);
5694 kmovwl(mask, rtmp);
5695 evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
5696 } else {
5697 evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
5698 evmovdqu(T_LONG, k0, Address(base, disp + 32), xtmp, false, Assembler::AVX_128bit);
5699 }
5700 break;
5701 case 7:
5702 if (use64byteVector) {
5703 movl(rtmp, 0x7F);
5704 kmovwl(mask, rtmp);
5705 evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
5706 } else {
5707 evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
5708 movl(rtmp, 0x7);
5709 kmovwl(mask, rtmp);
5710 evmovdqu(T_LONG, mask, Address(base, disp + 32), xtmp, true, Assembler::AVX_256bit);
5711 }
5712 break;
5713 default:
5714 fatal("Unexpected length : %d\n",cnt);
5715 break;
5716 }
5717 }
5718 }
5719
5720 void MacroAssembler::clear_mem(Register base, Register cnt, Register tmp, XMMRegister xtmp,
5721 bool is_large, KRegister mask) {
5722 // cnt - number of qwords (8-byte words).
5723 // base - start address, qword aligned.
5724 // is_large - if optimizers know cnt is larger than InitArrayShortSize
5725 assert(base==rdi, "base register must be edi for rep stos");
5726 assert(tmp==rax, "tmp register must be eax for rep stos");
5727 assert(cnt==rcx, "cnt register must be ecx for rep stos");
5728 assert(InitArrayShortSize % BytesPerLong == 0,
5729 "InitArrayShortSize should be the multiple of BytesPerLong");
5730
5731 Label DONE;
5732 if (!is_large || !UseXMMForObjInit) {
5733 xorptr(tmp, tmp);
5734 }
5735
5736 if (!is_large) {
5737 Label LOOP, LONG;
5738 cmpptr(cnt, InitArrayShortSize/BytesPerLong);
5739 jccb(Assembler::greater, LONG);
5740
5741 decrement(cnt);
5742 jccb(Assembler::negative, DONE); // Zero length
5743
5744 // Use individual pointer-sized stores for small counts:
5745 BIND(LOOP);
5746 movptr(Address(base, cnt, Address::times_ptr), tmp);
5747 decrement(cnt);
5748 jccb(Assembler::greaterEqual, LOOP);
5749 jmpb(DONE);
5750
5751 BIND(LONG);
5752 }
5753
5754 // Use longer rep-prefixed ops for non-small counts:
5755 if (UseFastStosb) {
5756 shlptr(cnt, 3); // convert to number of bytes
5757 rep_stosb();
5758 } else if (UseXMMForObjInit) {
5759 xmm_clear_mem(base, cnt, tmp, xtmp, mask);
5760 } else {
5761 rep_stos();
5762 }
5763
5764 BIND(DONE);
5765 }
5766
5767 #endif //COMPILER2_OR_JVMCI
5768
5769
5770 void MacroAssembler::generate_fill(BasicType t, bool aligned,
5771 Register to, Register value, Register count,
5772 Register rtmp, XMMRegister xtmp) {
5773 ShortBranchVerifier sbv(this);
5774 assert_different_registers(to, value, count, rtmp);
5775 Label L_exit;
5776 Label L_fill_2_bytes, L_fill_4_bytes;
5777
5778 #if defined(COMPILER2)
5779 if(MaxVectorSize >=32 &&
5780 VM_Version::supports_avx512vlbw() &&
5781 VM_Version::supports_bmi2()) {
5782 generate_fill_avx3(t, to, value, count, rtmp, xtmp);
5783 return;
5784 }
5785 #endif
5786
5787 int shift = -1;
5788 switch (t) {
5789 case T_BYTE:
5790 shift = 2;
5791 break;
5792 case T_SHORT:
5793 shift = 1;
5794 break;
5795 case T_INT:
5796 shift = 0;
5797 break;
5798 default: ShouldNotReachHere();
5799 }
5800
5801 if (t == T_BYTE) {
5802 andl(value, 0xff);
5803 movl(rtmp, value);
5804 shll(rtmp, 8);
5805 orl(value, rtmp);
5806 }
5807 if (t == T_SHORT) {
5808 andl(value, 0xffff);
5809 }
5810 if (t == T_BYTE || t == T_SHORT) {
5811 movl(rtmp, value);
5812 shll(rtmp, 16);
5813 orl(value, rtmp);
5814 }
5815
5816 cmpptr(count, 2<<shift); // Short arrays (< 8 bytes) fill by element
5817 jcc(Assembler::below, L_fill_4_bytes); // use unsigned cmp
5818 if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
5819 Label L_skip_align2;
5820 // align source address at 4 bytes address boundary
5821 if (t == T_BYTE) {
5822 Label L_skip_align1;
5823 // One byte misalignment happens only for byte arrays
5824 testptr(to, 1);
5825 jccb(Assembler::zero, L_skip_align1);
5826 movb(Address(to, 0), value);
5827 increment(to);
5828 decrement(count);
5829 BIND(L_skip_align1);
5830 }
5831 // Two bytes misalignment happens only for byte and short (char) arrays
5832 testptr(to, 2);
5833 jccb(Assembler::zero, L_skip_align2);
5834 movw(Address(to, 0), value);
5835 addptr(to, 2);
5836 subptr(count, 1<<(shift-1));
5837 BIND(L_skip_align2);
5838 }
5839 {
5840 Label L_fill_32_bytes;
5841 if (!UseUnalignedLoadStores) {
5842 // align to 8 bytes, we know we are 4 byte aligned to start
5843 testptr(to, 4);
5844 jccb(Assembler::zero, L_fill_32_bytes);
5845 movl(Address(to, 0), value);
5846 addptr(to, 4);
5847 subptr(count, 1<<shift);
5848 }
5849 BIND(L_fill_32_bytes);
5850 {
5851 Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
5852 movdl(xtmp, value);
5853 if (UseAVX >= 2 && UseUnalignedLoadStores) {
5854 Label L_check_fill_32_bytes;
5855 if (UseAVX > 2) {
5856 // Fill 64-byte chunks
5857 Label L_fill_64_bytes_loop_avx3, L_check_fill_64_bytes_avx2;
5858
5859 // If number of bytes to fill < VM_Version::avx3_threshold(), perform fill using AVX2
5860 cmpptr(count, VM_Version::avx3_threshold());
5861 jccb(Assembler::below, L_check_fill_64_bytes_avx2);
5862
5863 vpbroadcastd(xtmp, xtmp, Assembler::AVX_512bit);
5864
5865 subptr(count, 16 << shift);
5866 jccb(Assembler::less, L_check_fill_32_bytes);
5867 align(16);
5868
5869 BIND(L_fill_64_bytes_loop_avx3);
5870 evmovdqul(Address(to, 0), xtmp, Assembler::AVX_512bit);
5871 addptr(to, 64);
5872 subptr(count, 16 << shift);
5873 jcc(Assembler::greaterEqual, L_fill_64_bytes_loop_avx3);
5874 jmpb(L_check_fill_32_bytes);
5875
5876 BIND(L_check_fill_64_bytes_avx2);
5877 }
5878 // Fill 64-byte chunks
5879 Label L_fill_64_bytes_loop;
5880 vpbroadcastd(xtmp, xtmp, Assembler::AVX_256bit);
5881
5882 subptr(count, 16 << shift);
5883 jcc(Assembler::less, L_check_fill_32_bytes);
5884 align(16);
5885
5886 BIND(L_fill_64_bytes_loop);
5887 vmovdqu(Address(to, 0), xtmp);
5888 vmovdqu(Address(to, 32), xtmp);
5889 addptr(to, 64);
5890 subptr(count, 16 << shift);
5891 jcc(Assembler::greaterEqual, L_fill_64_bytes_loop);
5892
5893 BIND(L_check_fill_32_bytes);
5894 addptr(count, 8 << shift);
5895 jccb(Assembler::less, L_check_fill_8_bytes);
5896 vmovdqu(Address(to, 0), xtmp);
5897 addptr(to, 32);
5898 subptr(count, 8 << shift);
5899
5900 BIND(L_check_fill_8_bytes);
5901 // clean upper bits of YMM registers
5902 movdl(xtmp, value);
5903 pshufd(xtmp, xtmp, 0);
5904 } else {
5905 // Fill 32-byte chunks
5906 pshufd(xtmp, xtmp, 0);
5907
5908 subptr(count, 8 << shift);
5909 jcc(Assembler::less, L_check_fill_8_bytes);
5910 align(16);
5911
5912 BIND(L_fill_32_bytes_loop);
5913
5914 if (UseUnalignedLoadStores) {
5915 movdqu(Address(to, 0), xtmp);
5916 movdqu(Address(to, 16), xtmp);
5917 } else {
5918 movq(Address(to, 0), xtmp);
5919 movq(Address(to, 8), xtmp);
5920 movq(Address(to, 16), xtmp);
5921 movq(Address(to, 24), xtmp);
5922 }
5923
5924 addptr(to, 32);
5925 subptr(count, 8 << shift);
5926 jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
5927
5928 BIND(L_check_fill_8_bytes);
5929 }
5930 addptr(count, 8 << shift);
5931 jccb(Assembler::zero, L_exit);
5932 jmpb(L_fill_8_bytes);
5933
5934 //
5935 // length is too short, just fill qwords
5936 //
5937 BIND(L_fill_8_bytes_loop);
5938 movq(Address(to, 0), xtmp);
5939 addptr(to, 8);
5940 BIND(L_fill_8_bytes);
5941 subptr(count, 1 << (shift + 1));
5942 jcc(Assembler::greaterEqual, L_fill_8_bytes_loop);
5943 }
5944 }
5945 // fill trailing 4 bytes
5946 BIND(L_fill_4_bytes);
5947 testl(count, 1<<shift);
5948 jccb(Assembler::zero, L_fill_2_bytes);
5949 movl(Address(to, 0), value);
5950 if (t == T_BYTE || t == T_SHORT) {
5951 Label L_fill_byte;
5952 addptr(to, 4);
5953 BIND(L_fill_2_bytes);
5954 // fill trailing 2 bytes
5955 testl(count, 1<<(shift-1));
5956 jccb(Assembler::zero, L_fill_byte);
5957 movw(Address(to, 0), value);
5958 if (t == T_BYTE) {
5959 addptr(to, 2);
5960 BIND(L_fill_byte);
5961 // fill trailing byte
5962 testl(count, 1);
5963 jccb(Assembler::zero, L_exit);
5964 movb(Address(to, 0), value);
5965 } else {
5966 BIND(L_fill_byte);
5967 }
5968 } else {
5969 BIND(L_fill_2_bytes);
5970 }
5971 BIND(L_exit);
5972 }
5973
5974 void MacroAssembler::evpbroadcast(BasicType type, XMMRegister dst, Register src, int vector_len) {
5975 switch(type) {
5976 case T_BYTE:
5977 case T_BOOLEAN:
5978 evpbroadcastb(dst, src, vector_len);
5979 break;
5980 case T_SHORT:
5981 case T_CHAR:
5982 evpbroadcastw(dst, src, vector_len);
5983 break;
5984 case T_INT:
5985 case T_FLOAT:
5986 evpbroadcastd(dst, src, vector_len);
5987 break;
5988 case T_LONG:
5989 case T_DOUBLE:
5990 evpbroadcastq(dst, src, vector_len);
5991 break;
5992 default:
5993 fatal("Unhandled type : %s", type2name(type));
5994 break;
5995 }
5996 }
5997
5998 // encode char[] to byte[] in ISO_8859_1 or ASCII
5999 //@IntrinsicCandidate
6000 //private static int implEncodeISOArray(byte[] sa, int sp,
6001 //byte[] da, int dp, int len) {
6002 // int i = 0;
6003 // for (; i < len; i++) {
6004 // char c = StringUTF16.getChar(sa, sp++);
6005 // if (c > '\u00FF')
6006 // break;
6007 // da[dp++] = (byte)c;
6008 // }
6009 // return i;
6010 //}
6011 //
6012 //@IntrinsicCandidate
6013 //private static int implEncodeAsciiArray(char[] sa, int sp,
6014 // byte[] da, int dp, int len) {
6015 // int i = 0;
6016 // for (; i < len; i++) {
6017 // char c = sa[sp++];
6018 // if (c >= '\u0080')
6019 // break;
6020 // da[dp++] = (byte)c;
6021 // }
6022 // return i;
6023 //}
6024 void MacroAssembler::encode_iso_array(Register src, Register dst, Register len,
6025 XMMRegister tmp1Reg, XMMRegister tmp2Reg,
6026 XMMRegister tmp3Reg, XMMRegister tmp4Reg,
6027 Register tmp5, Register result, bool ascii) {
6028
6029 // rsi: src
6030 // rdi: dst
6031 // rdx: len
6032 // rcx: tmp5
6033 // rax: result
6034 ShortBranchVerifier sbv(this);
6035 assert_different_registers(src, dst, len, tmp5, result);
6036 Label L_done, L_copy_1_char, L_copy_1_char_exit;
6037
6038 int mask = ascii ? 0xff80ff80 : 0xff00ff00;
6039 int short_mask = ascii ? 0xff80 : 0xff00;
6040
6041 // set result
6042 xorl(result, result);
6043 // check for zero length
6044 testl(len, len);
6045 jcc(Assembler::zero, L_done);
6046
6047 movl(result, len);
6048
6049 // Setup pointers
6050 lea(src, Address(src, len, Address::times_2)); // char[]
6051 lea(dst, Address(dst, len, Address::times_1)); // byte[]
6052 negptr(len);
6053
6054 if (UseSSE42Intrinsics || UseAVX >= 2) {
6055 Label L_copy_8_chars, L_copy_8_chars_exit;
6056 Label L_chars_16_check, L_copy_16_chars, L_copy_16_chars_exit;
6057
6058 if (UseAVX >= 2) {
6059 Label L_chars_32_check, L_copy_32_chars, L_copy_32_chars_exit;
6060 movl(tmp5, mask); // create mask to test for Unicode or non-ASCII chars in vector
6061 movdl(tmp1Reg, tmp5);
6062 vpbroadcastd(tmp1Reg, tmp1Reg, Assembler::AVX_256bit);
6063 jmp(L_chars_32_check);
6064
6065 bind(L_copy_32_chars);
6066 vmovdqu(tmp3Reg, Address(src, len, Address::times_2, -64));
6067 vmovdqu(tmp4Reg, Address(src, len, Address::times_2, -32));
6068 vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
6069 vptest(tmp2Reg, tmp1Reg); // check for Unicode or non-ASCII chars in vector
6070 jccb(Assembler::notZero, L_copy_32_chars_exit);
6071 vpackuswb(tmp3Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
6072 vpermq(tmp4Reg, tmp3Reg, 0xD8, /* vector_len */ 1);
6073 vmovdqu(Address(dst, len, Address::times_1, -32), tmp4Reg);
6074
6075 bind(L_chars_32_check);
6076 addptr(len, 32);
6077 jcc(Assembler::lessEqual, L_copy_32_chars);
6078
6079 bind(L_copy_32_chars_exit);
6080 subptr(len, 16);
6081 jccb(Assembler::greater, L_copy_16_chars_exit);
6082
6083 } else if (UseSSE42Intrinsics) {
6084 movl(tmp5, mask); // create mask to test for Unicode or non-ASCII chars in vector
6085 movdl(tmp1Reg, tmp5);
6086 pshufd(tmp1Reg, tmp1Reg, 0);
6087 jmpb(L_chars_16_check);
6088 }
6089
6090 bind(L_copy_16_chars);
6091 if (UseAVX >= 2) {
6092 vmovdqu(tmp2Reg, Address(src, len, Address::times_2, -32));
6093 vptest(tmp2Reg, tmp1Reg);
6094 jcc(Assembler::notZero, L_copy_16_chars_exit);
6095 vpackuswb(tmp2Reg, tmp2Reg, tmp1Reg, /* vector_len */ 1);
6096 vpermq(tmp3Reg, tmp2Reg, 0xD8, /* vector_len */ 1);
6097 } else {
6098 if (UseAVX > 0) {
6099 movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
6100 movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
6101 vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 0);
6102 } else {
6103 movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
6104 por(tmp2Reg, tmp3Reg);
6105 movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
6106 por(tmp2Reg, tmp4Reg);
6107 }
6108 ptest(tmp2Reg, tmp1Reg); // check for Unicode or non-ASCII chars in vector
6109 jccb(Assembler::notZero, L_copy_16_chars_exit);
6110 packuswb(tmp3Reg, tmp4Reg);
6111 }
6112 movdqu(Address(dst, len, Address::times_1, -16), tmp3Reg);
6113
6114 bind(L_chars_16_check);
6115 addptr(len, 16);
6116 jcc(Assembler::lessEqual, L_copy_16_chars);
6117
6118 bind(L_copy_16_chars_exit);
6119 if (UseAVX >= 2) {
6120 // clean upper bits of YMM registers
6121 vpxor(tmp2Reg, tmp2Reg);
6122 vpxor(tmp3Reg, tmp3Reg);
6123 vpxor(tmp4Reg, tmp4Reg);
6124 movdl(tmp1Reg, tmp5);
6125 pshufd(tmp1Reg, tmp1Reg, 0);
6126 }
6127 subptr(len, 8);
6128 jccb(Assembler::greater, L_copy_8_chars_exit);
6129
6130 bind(L_copy_8_chars);
6131 movdqu(tmp3Reg, Address(src, len, Address::times_2, -16));
6132 ptest(tmp3Reg, tmp1Reg);
6133 jccb(Assembler::notZero, L_copy_8_chars_exit);
6134 packuswb(tmp3Reg, tmp1Reg);
6135 movq(Address(dst, len, Address::times_1, -8), tmp3Reg);
6136 addptr(len, 8);
6137 jccb(Assembler::lessEqual, L_copy_8_chars);
6138
6139 bind(L_copy_8_chars_exit);
6140 subptr(len, 8);
6141 jccb(Assembler::zero, L_done);
6142 }
6143
6144 bind(L_copy_1_char);
6145 load_unsigned_short(tmp5, Address(src, len, Address::times_2, 0));
6146 testl(tmp5, short_mask); // check if Unicode or non-ASCII char
6147 jccb(Assembler::notZero, L_copy_1_char_exit);
6148 movb(Address(dst, len, Address::times_1, 0), tmp5);
6149 addptr(len, 1);
6150 jccb(Assembler::less, L_copy_1_char);
6151
6152 bind(L_copy_1_char_exit);
6153 addptr(result, len); // len is negative count of not processed elements
6154
6155 bind(L_done);
6156 }
6157
6158 /**
6159 * Helper for multiply_to_len().
6160 */
6161 void MacroAssembler::add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
6162 addq(dest_lo, src1);
6163 adcq(dest_hi, 0);
6164 addq(dest_lo, src2);
6165 adcq(dest_hi, 0);
6166 }
6167
6168 /**
6169 * Multiply 64 bit by 64 bit first loop.
6170 */
6171 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
6172 Register y, Register y_idx, Register z,
6173 Register carry, Register product,
6174 Register idx, Register kdx) {
6175 //
6176 // jlong carry, x[], y[], z[];
6177 // for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
6178 // huge_128 product = y[idx] * x[xstart] + carry;
6179 // z[kdx] = (jlong)product;
6180 // carry = (jlong)(product >>> 64);
6181 // }
6182 // z[xstart] = carry;
6183 //
6184
6185 Label L_first_loop, L_first_loop_exit;
6186 Label L_one_x, L_one_y, L_multiply;
6187
6188 decrementl(xstart);
6189 jcc(Assembler::negative, L_one_x);
6190
6191 movq(x_xstart, Address(x, xstart, Address::times_4, 0));
6192 rorq(x_xstart, 32); // convert big-endian to little-endian
6193
6194 bind(L_first_loop);
6195 decrementl(idx);
6196 jcc(Assembler::negative, L_first_loop_exit);
6197 decrementl(idx);
6198 jcc(Assembler::negative, L_one_y);
6199 movq(y_idx, Address(y, idx, Address::times_4, 0));
6200 rorq(y_idx, 32); // convert big-endian to little-endian
6201 bind(L_multiply);
6202 movq(product, x_xstart);
6203 mulq(y_idx); // product(rax) * y_idx -> rdx:rax
6204 addq(product, carry);
6205 adcq(rdx, 0);
6206 subl(kdx, 2);
6207 movl(Address(z, kdx, Address::times_4, 4), product);
6208 shrq(product, 32);
6209 movl(Address(z, kdx, Address::times_4, 0), product);
6210 movq(carry, rdx);
6211 jmp(L_first_loop);
6212
6213 bind(L_one_y);
6214 movl(y_idx, Address(y, 0));
6215 jmp(L_multiply);
6216
6217 bind(L_one_x);
6218 movl(x_xstart, Address(x, 0));
6219 jmp(L_first_loop);
6220
6221 bind(L_first_loop_exit);
6222 }
6223
6224 /**
6225 * Multiply 64 bit by 64 bit and add 128 bit.
6226 */
6227 void MacroAssembler::multiply_add_128_x_128(Register x_xstart, Register y, Register z,
6228 Register yz_idx, Register idx,
6229 Register carry, Register product, int offset) {
6230 // huge_128 product = (y[idx] * x_xstart) + z[kdx] + carry;
6231 // z[kdx] = (jlong)product;
6232
6233 movq(yz_idx, Address(y, idx, Address::times_4, offset));
6234 rorq(yz_idx, 32); // convert big-endian to little-endian
6235 movq(product, x_xstart);
6236 mulq(yz_idx); // product(rax) * yz_idx -> rdx:product(rax)
6237 movq(yz_idx, Address(z, idx, Address::times_4, offset));
6238 rorq(yz_idx, 32); // convert big-endian to little-endian
6239
6240 add2_with_carry(rdx, product, carry, yz_idx);
6241
6242 movl(Address(z, idx, Address::times_4, offset+4), product);
6243 shrq(product, 32);
6244 movl(Address(z, idx, Address::times_4, offset), product);
6245
6246 }
6247
6248 /**
6249 * Multiply 128 bit by 128 bit. Unrolled inner loop.
6250 */
6251 void MacroAssembler::multiply_128_x_128_loop(Register x_xstart, Register y, Register z,
6252 Register yz_idx, Register idx, Register jdx,
6253 Register carry, Register product,
6254 Register carry2) {
6255 // jlong carry, x[], y[], z[];
6256 // int kdx = ystart+1;
6257 // for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
6258 // huge_128 product = (y[idx+1] * x_xstart) + z[kdx+idx+1] + carry;
6259 // z[kdx+idx+1] = (jlong)product;
6260 // jlong carry2 = (jlong)(product >>> 64);
6261 // product = (y[idx] * x_xstart) + z[kdx+idx] + carry2;
6262 // z[kdx+idx] = (jlong)product;
6263 // carry = (jlong)(product >>> 64);
6264 // }
6265 // idx += 2;
6266 // if (idx > 0) {
6267 // product = (y[idx] * x_xstart) + z[kdx+idx] + carry;
6268 // z[kdx+idx] = (jlong)product;
6269 // carry = (jlong)(product >>> 64);
6270 // }
6271 //
6272
6273 Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
6274
6275 movl(jdx, idx);
6276 andl(jdx, 0xFFFFFFFC);
6277 shrl(jdx, 2);
6278
6279 bind(L_third_loop);
6280 subl(jdx, 1);
6281 jcc(Assembler::negative, L_third_loop_exit);
6282 subl(idx, 4);
6283
6284 multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 8);
6285 movq(carry2, rdx);
6286
6287 multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry2, product, 0);
6288 movq(carry, rdx);
6289 jmp(L_third_loop);
6290
6291 bind (L_third_loop_exit);
6292
6293 andl (idx, 0x3);
6294 jcc(Assembler::zero, L_post_third_loop_done);
6295
6296 Label L_check_1;
6297 subl(idx, 2);
6298 jcc(Assembler::negative, L_check_1);
6299
6300 multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 0);
6301 movq(carry, rdx);
6302
6303 bind (L_check_1);
6304 addl (idx, 0x2);
6305 andl (idx, 0x1);
6306 subl(idx, 1);
6307 jcc(Assembler::negative, L_post_third_loop_done);
6308
6309 movl(yz_idx, Address(y, idx, Address::times_4, 0));
6310 movq(product, x_xstart);
6311 mulq(yz_idx); // product(rax) * yz_idx -> rdx:product(rax)
6312 movl(yz_idx, Address(z, idx, Address::times_4, 0));
6313
6314 add2_with_carry(rdx, product, yz_idx, carry);
6315
6316 movl(Address(z, idx, Address::times_4, 0), product);
6317 shrq(product, 32);
6318
6319 shlq(rdx, 32);
6320 orq(product, rdx);
6321 movq(carry, product);
6322
6323 bind(L_post_third_loop_done);
6324 }
6325
6326 /**
6327 * Multiply 128 bit by 128 bit using BMI2. Unrolled inner loop.
6328 *
6329 */
6330 void MacroAssembler::multiply_128_x_128_bmi2_loop(Register y, Register z,
6331 Register carry, Register carry2,
6332 Register idx, Register jdx,
6333 Register yz_idx1, Register yz_idx2,
6334 Register tmp, Register tmp3, Register tmp4) {
6335 assert(UseBMI2Instructions, "should be used only when BMI2 is available");
6336
6337 // jlong carry, x[], y[], z[];
6338 // int kdx = ystart+1;
6339 // for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
6340 // huge_128 tmp3 = (y[idx+1] * rdx) + z[kdx+idx+1] + carry;
6341 // jlong carry2 = (jlong)(tmp3 >>> 64);
6342 // huge_128 tmp4 = (y[idx] * rdx) + z[kdx+idx] + carry2;
6343 // carry = (jlong)(tmp4 >>> 64);
6344 // z[kdx+idx+1] = (jlong)tmp3;
6345 // z[kdx+idx] = (jlong)tmp4;
6346 // }
6347 // idx += 2;
6348 // if (idx > 0) {
6349 // yz_idx1 = (y[idx] * rdx) + z[kdx+idx] + carry;
6350 // z[kdx+idx] = (jlong)yz_idx1;
6351 // carry = (jlong)(yz_idx1 >>> 64);
6352 // }
6353 //
6354
6355 Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
6356
6357 movl(jdx, idx);
6358 andl(jdx, 0xFFFFFFFC);
6359 shrl(jdx, 2);
6360
6361 bind(L_third_loop);
6362 subl(jdx, 1);
6363 jcc(Assembler::negative, L_third_loop_exit);
6364 subl(idx, 4);
6365
6366 movq(yz_idx1, Address(y, idx, Address::times_4, 8));
6367 rorxq(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
6368 movq(yz_idx2, Address(y, idx, Address::times_4, 0));
6369 rorxq(yz_idx2, yz_idx2, 32);
6370
6371 mulxq(tmp4, tmp3, yz_idx1); // yz_idx1 * rdx -> tmp4:tmp3
6372 mulxq(carry2, tmp, yz_idx2); // yz_idx2 * rdx -> carry2:tmp
6373
6374 movq(yz_idx1, Address(z, idx, Address::times_4, 8));
6375 rorxq(yz_idx1, yz_idx1, 32);
6376 movq(yz_idx2, Address(z, idx, Address::times_4, 0));
6377 rorxq(yz_idx2, yz_idx2, 32);
6378
6379 if (VM_Version::supports_adx()) {
6380 adcxq(tmp3, carry);
6381 adoxq(tmp3, yz_idx1);
6382
6383 adcxq(tmp4, tmp);
6384 adoxq(tmp4, yz_idx2);
6385
6386 movl(carry, 0); // does not affect flags
6387 adcxq(carry2, carry);
6388 adoxq(carry2, carry);
6389 } else {
6390 add2_with_carry(tmp4, tmp3, carry, yz_idx1);
6391 add2_with_carry(carry2, tmp4, tmp, yz_idx2);
6392 }
6393 movq(carry, carry2);
6394
6395 movl(Address(z, idx, Address::times_4, 12), tmp3);
6396 shrq(tmp3, 32);
6397 movl(Address(z, idx, Address::times_4, 8), tmp3);
6398
6399 movl(Address(z, idx, Address::times_4, 4), tmp4);
6400 shrq(tmp4, 32);
6401 movl(Address(z, idx, Address::times_4, 0), tmp4);
6402
6403 jmp(L_third_loop);
6404
6405 bind (L_third_loop_exit);
6406
6407 andl (idx, 0x3);
6408 jcc(Assembler::zero, L_post_third_loop_done);
6409
6410 Label L_check_1;
6411 subl(idx, 2);
6412 jcc(Assembler::negative, L_check_1);
6413
6414 movq(yz_idx1, Address(y, idx, Address::times_4, 0));
6415 rorxq(yz_idx1, yz_idx1, 32);
6416 mulxq(tmp4, tmp3, yz_idx1); // yz_idx1 * rdx -> tmp4:tmp3
6417 movq(yz_idx2, Address(z, idx, Address::times_4, 0));
6418 rorxq(yz_idx2, yz_idx2, 32);
6419
6420 add2_with_carry(tmp4, tmp3, carry, yz_idx2);
6421
6422 movl(Address(z, idx, Address::times_4, 4), tmp3);
6423 shrq(tmp3, 32);
6424 movl(Address(z, idx, Address::times_4, 0), tmp3);
6425 movq(carry, tmp4);
6426
6427 bind (L_check_1);
6428 addl (idx, 0x2);
6429 andl (idx, 0x1);
6430 subl(idx, 1);
6431 jcc(Assembler::negative, L_post_third_loop_done);
6432 movl(tmp4, Address(y, idx, Address::times_4, 0));
6433 mulxq(carry2, tmp3, tmp4); // tmp4 * rdx -> carry2:tmp3
6434 movl(tmp4, Address(z, idx, Address::times_4, 0));
6435
6436 add2_with_carry(carry2, tmp3, tmp4, carry);
6437
6438 movl(Address(z, idx, Address::times_4, 0), tmp3);
6439 shrq(tmp3, 32);
6440
6441 shlq(carry2, 32);
6442 orq(tmp3, carry2);
6443 movq(carry, tmp3);
6444
6445 bind(L_post_third_loop_done);
6446 }
6447
6448 /**
6449 * Code for BigInteger::multiplyToLen() intrinsic.
6450 *
6451 * rdi: x
6452 * rax: xlen
6453 * rsi: y
6454 * rcx: ylen
6455 * r8: z
6456 * r11: tmp0
6457 * r12: tmp1
6458 * r13: tmp2
6459 * r14: tmp3
6460 * r15: tmp4
6461 * rbx: tmp5
6462 *
6463 */
6464 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register tmp0,
6465 Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5) {
6466 ShortBranchVerifier sbv(this);
6467 assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, rdx);
6468
6469 push(tmp0);
6470 push(tmp1);
6471 push(tmp2);
6472 push(tmp3);
6473 push(tmp4);
6474 push(tmp5);
6475
6476 push(xlen);
6477
6478 const Register idx = tmp1;
6479 const Register kdx = tmp2;
6480 const Register xstart = tmp3;
6481
6482 const Register y_idx = tmp4;
6483 const Register carry = tmp5;
6484 const Register product = xlen;
6485 const Register x_xstart = tmp0;
6486
6487 // First Loop.
6488 //
6489 // final static long LONG_MASK = 0xffffffffL;
6490 // int xstart = xlen - 1;
6491 // int ystart = ylen - 1;
6492 // long carry = 0;
6493 // for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
6494 // long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
6495 // z[kdx] = (int)product;
6496 // carry = product >>> 32;
6497 // }
6498 // z[xstart] = (int)carry;
6499 //
6500
6501 movl(idx, ylen); // idx = ylen;
6502 lea(kdx, Address(xlen, ylen)); // kdx = xlen+ylen;
6503 xorq(carry, carry); // carry = 0;
6504
6505 Label L_done;
6506
6507 movl(xstart, xlen);
6508 decrementl(xstart);
6509 jcc(Assembler::negative, L_done);
6510
6511 multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
6512
6513 Label L_second_loop;
6514 testl(kdx, kdx);
6515 jcc(Assembler::zero, L_second_loop);
6516
6517 Label L_carry;
6518 subl(kdx, 1);
6519 jcc(Assembler::zero, L_carry);
6520
6521 movl(Address(z, kdx, Address::times_4, 0), carry);
6522 shrq(carry, 32);
6523 subl(kdx, 1);
6524
6525 bind(L_carry);
6526 movl(Address(z, kdx, Address::times_4, 0), carry);
6527
6528 // Second and third (nested) loops.
6529 //
6530 // for (int i = xstart-1; i >= 0; i--) { // Second loop
6531 // carry = 0;
6532 // for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
6533 // long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
6534 // (z[k] & LONG_MASK) + carry;
6535 // z[k] = (int)product;
6536 // carry = product >>> 32;
6537 // }
6538 // z[i] = (int)carry;
6539 // }
6540 //
6541 // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = rdx
6542
6543 const Register jdx = tmp1;
6544
6545 bind(L_second_loop);
6546 xorl(carry, carry); // carry = 0;
6547 movl(jdx, ylen); // j = ystart+1
6548
6549 subl(xstart, 1); // i = xstart-1;
6550 jcc(Assembler::negative, L_done);
6551
6552 push (z);
6553
6554 Label L_last_x;
6555 lea(z, Address(z, xstart, Address::times_4, 4)); // z = z + k - j
6556 subl(xstart, 1); // i = xstart-1;
6557 jcc(Assembler::negative, L_last_x);
6558
6559 if (UseBMI2Instructions) {
6560 movq(rdx, Address(x, xstart, Address::times_4, 0));
6561 rorxq(rdx, rdx, 32); // convert big-endian to little-endian
6562 } else {
6563 movq(x_xstart, Address(x, xstart, Address::times_4, 0));
6564 rorq(x_xstart, 32); // convert big-endian to little-endian
6565 }
6566
6567 Label L_third_loop_prologue;
6568 bind(L_third_loop_prologue);
6569
6570 push (x);
6571 push (xstart);
6572 push (ylen);
6573
6574
6575 if (UseBMI2Instructions) {
6576 multiply_128_x_128_bmi2_loop(y, z, carry, x, jdx, ylen, product, tmp2, x_xstart, tmp3, tmp4);
6577 } else { // !UseBMI2Instructions
6578 multiply_128_x_128_loop(x_xstart, y, z, y_idx, jdx, ylen, carry, product, x);
6579 }
6580
6581 pop(ylen);
6582 pop(xlen);
6583 pop(x);
6584 pop(z);
6585
6586 movl(tmp3, xlen);
6587 addl(tmp3, 1);
6588 movl(Address(z, tmp3, Address::times_4, 0), carry);
6589 subl(tmp3, 1);
6590 jccb(Assembler::negative, L_done);
6591
6592 shrq(carry, 32);
6593 movl(Address(z, tmp3, Address::times_4, 0), carry);
6594 jmp(L_second_loop);
6595
6596 // Next infrequent code is moved outside loops.
6597 bind(L_last_x);
6598 if (UseBMI2Instructions) {
6599 movl(rdx, Address(x, 0));
6600 } else {
6601 movl(x_xstart, Address(x, 0));
6602 }
6603 jmp(L_third_loop_prologue);
6604
6605 bind(L_done);
6606
6607 pop(xlen);
6608
6609 pop(tmp5);
6610 pop(tmp4);
6611 pop(tmp3);
6612 pop(tmp2);
6613 pop(tmp1);
6614 pop(tmp0);
6615 }
6616
6617 void MacroAssembler::vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale,
6618 Register result, Register tmp1, Register tmp2, XMMRegister rymm0, XMMRegister rymm1, XMMRegister rymm2){
6619 assert(UseSSE42Intrinsics, "SSE4.2 must be enabled.");
6620 Label VECTOR16_LOOP, VECTOR8_LOOP, VECTOR4_LOOP;
6621 Label VECTOR8_TAIL, VECTOR4_TAIL;
6622 Label VECTOR32_NOT_EQUAL, VECTOR16_NOT_EQUAL, VECTOR8_NOT_EQUAL, VECTOR4_NOT_EQUAL;
6623 Label SAME_TILL_END, DONE;
6624 Label BYTES_LOOP, BYTES_TAIL, BYTES_NOT_EQUAL;
6625
6626 //scale is in rcx in both Win64 and Unix
6627 ShortBranchVerifier sbv(this);
6628
6629 shlq(length);
6630 xorq(result, result);
6631
6632 if ((AVX3Threshold == 0) && (UseAVX > 2) &&
6633 VM_Version::supports_avx512vlbw()) {
6634 Label VECTOR64_LOOP, VECTOR64_NOT_EQUAL, VECTOR32_TAIL;
6635
6636 cmpq(length, 64);
6637 jcc(Assembler::less, VECTOR32_TAIL);
6638
6639 movq(tmp1, length);
6640 andq(tmp1, 0x3F); // tail count
6641 andq(length, ~(0x3F)); //vector count
6642
6643 bind(VECTOR64_LOOP);
6644 // AVX512 code to compare 64 byte vectors.
6645 evmovdqub(rymm0, Address(obja, result), Assembler::AVX_512bit);
6646 evpcmpeqb(k7, rymm0, Address(objb, result), Assembler::AVX_512bit);
6647 kortestql(k7, k7);
6648 jcc(Assembler::aboveEqual, VECTOR64_NOT_EQUAL); // mismatch
6649 addq(result, 64);
6650 subq(length, 64);
6651 jccb(Assembler::notZero, VECTOR64_LOOP);
6652
6653 //bind(VECTOR64_TAIL);
6654 testq(tmp1, tmp1);
6655 jcc(Assembler::zero, SAME_TILL_END);
6656
6657 //bind(VECTOR64_TAIL);
6658 // AVX512 code to compare up to 63 byte vectors.
6659 mov64(tmp2, 0xFFFFFFFFFFFFFFFF);
6660 shlxq(tmp2, tmp2, tmp1);
6661 notq(tmp2);
6662 kmovql(k3, tmp2);
6663
6664 evmovdqub(rymm0, k3, Address(obja, result), false, Assembler::AVX_512bit);
6665 evpcmpeqb(k7, k3, rymm0, Address(objb, result), Assembler::AVX_512bit);
6666
6667 ktestql(k7, k3);
6668 jcc(Assembler::below, SAME_TILL_END); // not mismatch
6669
6670 bind(VECTOR64_NOT_EQUAL);
6671 kmovql(tmp1, k7);
6672 notq(tmp1);
6673 tzcntq(tmp1, tmp1);
6674 addq(result, tmp1);
6675 shrq(result);
6676 jmp(DONE);
6677 bind(VECTOR32_TAIL);
6678 }
6679
6680 cmpq(length, 8);
6681 jcc(Assembler::equal, VECTOR8_LOOP);
6682 jcc(Assembler::less, VECTOR4_TAIL);
6683
6684 if (UseAVX >= 2) {
6685 Label VECTOR16_TAIL, VECTOR32_LOOP;
6686
6687 cmpq(length, 16);
6688 jcc(Assembler::equal, VECTOR16_LOOP);
6689 jcc(Assembler::less, VECTOR8_LOOP);
6690
6691 cmpq(length, 32);
6692 jccb(Assembler::less, VECTOR16_TAIL);
6693
6694 subq(length, 32);
6695 bind(VECTOR32_LOOP);
6696 vmovdqu(rymm0, Address(obja, result));
6697 vmovdqu(rymm1, Address(objb, result));
6698 vpxor(rymm2, rymm0, rymm1, Assembler::AVX_256bit);
6699 vptest(rymm2, rymm2);
6700 jcc(Assembler::notZero, VECTOR32_NOT_EQUAL);//mismatch found
6701 addq(result, 32);
6702 subq(length, 32);
6703 jcc(Assembler::greaterEqual, VECTOR32_LOOP);
6704 addq(length, 32);
6705 jcc(Assembler::equal, SAME_TILL_END);
6706 //falling through if less than 32 bytes left //close the branch here.
6707
6708 bind(VECTOR16_TAIL);
6709 cmpq(length, 16);
6710 jccb(Assembler::less, VECTOR8_TAIL);
6711 bind(VECTOR16_LOOP);
6712 movdqu(rymm0, Address(obja, result));
6713 movdqu(rymm1, Address(objb, result));
6714 vpxor(rymm2, rymm0, rymm1, Assembler::AVX_128bit);
6715 ptest(rymm2, rymm2);
6716 jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
6717 addq(result, 16);
6718 subq(length, 16);
6719 jcc(Assembler::equal, SAME_TILL_END);
6720 //falling through if less than 16 bytes left
6721 } else {//regular intrinsics
6722
6723 cmpq(length, 16);
6724 jccb(Assembler::less, VECTOR8_TAIL);
6725
6726 subq(length, 16);
6727 bind(VECTOR16_LOOP);
6728 movdqu(rymm0, Address(obja, result));
6729 movdqu(rymm1, Address(objb, result));
6730 pxor(rymm0, rymm1);
6731 ptest(rymm0, rymm0);
6732 jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
6733 addq(result, 16);
6734 subq(length, 16);
6735 jccb(Assembler::greaterEqual, VECTOR16_LOOP);
6736 addq(length, 16);
6737 jcc(Assembler::equal, SAME_TILL_END);
6738 //falling through if less than 16 bytes left
6739 }
6740
6741 bind(VECTOR8_TAIL);
6742 cmpq(length, 8);
6743 jccb(Assembler::less, VECTOR4_TAIL);
6744 bind(VECTOR8_LOOP);
6745 movq(tmp1, Address(obja, result));
6746 movq(tmp2, Address(objb, result));
6747 xorq(tmp1, tmp2);
6748 testq(tmp1, tmp1);
6749 jcc(Assembler::notZero, VECTOR8_NOT_EQUAL);//mismatch found
6750 addq(result, 8);
6751 subq(length, 8);
6752 jcc(Assembler::equal, SAME_TILL_END);
6753 //falling through if less than 8 bytes left
6754
6755 bind(VECTOR4_TAIL);
6756 cmpq(length, 4);
6757 jccb(Assembler::less, BYTES_TAIL);
6758 bind(VECTOR4_LOOP);
6759 movl(tmp1, Address(obja, result));
6760 xorl(tmp1, Address(objb, result));
6761 testl(tmp1, tmp1);
6762 jcc(Assembler::notZero, VECTOR4_NOT_EQUAL);//mismatch found
6763 addq(result, 4);
6764 subq(length, 4);
6765 jcc(Assembler::equal, SAME_TILL_END);
6766 //falling through if less than 4 bytes left
6767
6768 bind(BYTES_TAIL);
6769 bind(BYTES_LOOP);
6770 load_unsigned_byte(tmp1, Address(obja, result));
6771 load_unsigned_byte(tmp2, Address(objb, result));
6772 xorl(tmp1, tmp2);
6773 testl(tmp1, tmp1);
6774 jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
6775 decq(length);
6776 jcc(Assembler::zero, SAME_TILL_END);
6777 incq(result);
6778 load_unsigned_byte(tmp1, Address(obja, result));
6779 load_unsigned_byte(tmp2, Address(objb, result));
6780 xorl(tmp1, tmp2);
6781 testl(tmp1, tmp1);
6782 jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
6783 decq(length);
6784 jcc(Assembler::zero, SAME_TILL_END);
6785 incq(result);
6786 load_unsigned_byte(tmp1, Address(obja, result));
6787 load_unsigned_byte(tmp2, Address(objb, result));
6788 xorl(tmp1, tmp2);
6789 testl(tmp1, tmp1);
6790 jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
6791 jmp(SAME_TILL_END);
6792
6793 if (UseAVX >= 2) {
6794 bind(VECTOR32_NOT_EQUAL);
6795 vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_256bit);
6796 vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_256bit);
6797 vpxor(rymm0, rymm0, rymm2, Assembler::AVX_256bit);
6798 vpmovmskb(tmp1, rymm0);
6799 bsfq(tmp1, tmp1);
6800 addq(result, tmp1);
6801 shrq(result);
6802 jmp(DONE);
6803 }
6804
6805 bind(VECTOR16_NOT_EQUAL);
6806 if (UseAVX >= 2) {
6807 vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_128bit);
6808 vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_128bit);
6809 pxor(rymm0, rymm2);
6810 } else {
6811 pcmpeqb(rymm2, rymm2);
6812 pxor(rymm0, rymm1);
6813 pcmpeqb(rymm0, rymm1);
6814 pxor(rymm0, rymm2);
6815 }
6816 pmovmskb(tmp1, rymm0);
6817 bsfq(tmp1, tmp1);
6818 addq(result, tmp1);
6819 shrq(result);
6820 jmpb(DONE);
6821
6822 bind(VECTOR8_NOT_EQUAL);
6823 bind(VECTOR4_NOT_EQUAL);
6824 bsfq(tmp1, tmp1);
6825 shrq(tmp1, 3);
6826 addq(result, tmp1);
6827 bind(BYTES_NOT_EQUAL);
6828 shrq(result);
6829 jmpb(DONE);
6830
6831 bind(SAME_TILL_END);
6832 mov64(result, -1);
6833
6834 bind(DONE);
6835 }
6836
6837 //Helper functions for square_to_len()
6838
6839 /**
6840 * Store the squares of x[], right shifted one bit (divided by 2) into z[]
6841 * Preserves x and z and modifies rest of the registers.
6842 */
6843 void MacroAssembler::square_rshift(Register x, Register xlen, Register z, Register tmp1, Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
6844 // Perform square and right shift by 1
6845 // Handle odd xlen case first, then for even xlen do the following
6846 // jlong carry = 0;
6847 // for (int j=0, i=0; j < xlen; j+=2, i+=4) {
6848 // huge_128 product = x[j:j+1] * x[j:j+1];
6849 // z[i:i+1] = (carry << 63) | (jlong)(product >>> 65);
6850 // z[i+2:i+3] = (jlong)(product >>> 1);
6851 // carry = (jlong)product;
6852 // }
6853
6854 xorq(tmp5, tmp5); // carry
6855 xorq(rdxReg, rdxReg);
6856 xorl(tmp1, tmp1); // index for x
6857 xorl(tmp4, tmp4); // index for z
6858
6859 Label L_first_loop, L_first_loop_exit;
6860
6861 testl(xlen, 1);
6862 jccb(Assembler::zero, L_first_loop); //jump if xlen is even
6863
6864 // Square and right shift by 1 the odd element using 32 bit multiply
6865 movl(raxReg, Address(x, tmp1, Address::times_4, 0));
6866 imulq(raxReg, raxReg);
6867 shrq(raxReg, 1);
6868 adcq(tmp5, 0);
6869 movq(Address(z, tmp4, Address::times_4, 0), raxReg);
6870 incrementl(tmp1);
6871 addl(tmp4, 2);
6872
6873 // Square and right shift by 1 the rest using 64 bit multiply
6874 bind(L_first_loop);
6875 cmpptr(tmp1, xlen);
6876 jccb(Assembler::equal, L_first_loop_exit);
6877
6878 // Square
6879 movq(raxReg, Address(x, tmp1, Address::times_4, 0));
6880 rorq(raxReg, 32); // convert big-endian to little-endian
6881 mulq(raxReg); // 64-bit multiply rax * rax -> rdx:rax
6882
6883 // Right shift by 1 and save carry
6884 shrq(tmp5, 1); // rdx:rax:tmp5 = (tmp5:rdx:rax) >>> 1
6885 rcrq(rdxReg, 1);
6886 rcrq(raxReg, 1);
6887 adcq(tmp5, 0);
6888
6889 // Store result in z
6890 movq(Address(z, tmp4, Address::times_4, 0), rdxReg);
6891 movq(Address(z, tmp4, Address::times_4, 8), raxReg);
6892
6893 // Update indices for x and z
6894 addl(tmp1, 2);
6895 addl(tmp4, 4);
6896 jmp(L_first_loop);
6897
6898 bind(L_first_loop_exit);
6899 }
6900
6901
6902 /**
6903 * Perform the following multiply add operation using BMI2 instructions
6904 * carry:sum = sum + op1*op2 + carry
6905 * op2 should be in rdx
6906 * op2 is preserved, all other registers are modified
6907 */
6908 void MacroAssembler::multiply_add_64_bmi2(Register sum, Register op1, Register op2, Register carry, Register tmp2) {
6909 // assert op2 is rdx
6910 mulxq(tmp2, op1, op1); // op1 * op2 -> tmp2:op1
6911 addq(sum, carry);
6912 adcq(tmp2, 0);
6913 addq(sum, op1);
6914 adcq(tmp2, 0);
6915 movq(carry, tmp2);
6916 }
6917
6918 /**
6919 * Perform the following multiply add operation:
6920 * carry:sum = sum + op1*op2 + carry
6921 * Preserves op1, op2 and modifies rest of registers
6922 */
6923 void MacroAssembler::multiply_add_64(Register sum, Register op1, Register op2, Register carry, Register rdxReg, Register raxReg) {
6924 // rdx:rax = op1 * op2
6925 movq(raxReg, op2);
6926 mulq(op1);
6927
6928 // rdx:rax = sum + carry + rdx:rax
6929 addq(sum, carry);
6930 adcq(rdxReg, 0);
6931 addq(sum, raxReg);
6932 adcq(rdxReg, 0);
6933
6934 // carry:sum = rdx:sum
6935 movq(carry, rdxReg);
6936 }
6937
6938 /**
6939 * Add 64 bit long carry into z[] with carry propagation.
6940 * Preserves z and carry register values and modifies rest of registers.
6941 *
6942 */
6943 void MacroAssembler::add_one_64(Register z, Register zlen, Register carry, Register tmp1) {
6944 Label L_fourth_loop, L_fourth_loop_exit;
6945
6946 movl(tmp1, 1);
6947 subl(zlen, 2);
6948 addq(Address(z, zlen, Address::times_4, 0), carry);
6949
6950 bind(L_fourth_loop);
6951 jccb(Assembler::carryClear, L_fourth_loop_exit);
6952 subl(zlen, 2);
6953 jccb(Assembler::negative, L_fourth_loop_exit);
6954 addq(Address(z, zlen, Address::times_4, 0), tmp1);
6955 jmp(L_fourth_loop);
6956 bind(L_fourth_loop_exit);
6957 }
6958
6959 /**
6960 * Shift z[] left by 1 bit.
6961 * Preserves x, len, z and zlen registers and modifies rest of the registers.
6962 *
6963 */
6964 void MacroAssembler::lshift_by_1(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
6965
6966 Label L_fifth_loop, L_fifth_loop_exit;
6967
6968 // Fifth loop
6969 // Perform primitiveLeftShift(z, zlen, 1)
6970
6971 const Register prev_carry = tmp1;
6972 const Register new_carry = tmp4;
6973 const Register value = tmp2;
6974 const Register zidx = tmp3;
6975
6976 // int zidx, carry;
6977 // long value;
6978 // carry = 0;
6979 // for (zidx = zlen-2; zidx >=0; zidx -= 2) {
6980 // (carry:value) = (z[i] << 1) | carry ;
6981 // z[i] = value;
6982 // }
6983
6984 movl(zidx, zlen);
6985 xorl(prev_carry, prev_carry); // clear carry flag and prev_carry register
6986
6987 bind(L_fifth_loop);
6988 decl(zidx); // Use decl to preserve carry flag
6989 decl(zidx);
6990 jccb(Assembler::negative, L_fifth_loop_exit);
6991
6992 if (UseBMI2Instructions) {
6993 movq(value, Address(z, zidx, Address::times_4, 0));
6994 rclq(value, 1);
6995 rorxq(value, value, 32);
6996 movq(Address(z, zidx, Address::times_4, 0), value); // Store back in big endian form
6997 }
6998 else {
6999 // clear new_carry
7000 xorl(new_carry, new_carry);
7001
7002 // Shift z[i] by 1, or in previous carry and save new carry
7003 movq(value, Address(z, zidx, Address::times_4, 0));
7004 shlq(value, 1);
7005 adcl(new_carry, 0);
7006
7007 orq(value, prev_carry);
7008 rorq(value, 0x20);
7009 movq(Address(z, zidx, Address::times_4, 0), value); // Store back in big endian form
7010
7011 // Set previous carry = new carry
7012 movl(prev_carry, new_carry);
7013 }
7014 jmp(L_fifth_loop);
7015
7016 bind(L_fifth_loop_exit);
7017 }
7018
7019
7020 /**
7021 * Code for BigInteger::squareToLen() intrinsic
7022 *
7023 * rdi: x
7024 * rsi: len
7025 * r8: z
7026 * rcx: zlen
7027 * r12: tmp1
7028 * r13: tmp2
7029 * r14: tmp3
7030 * r15: tmp4
7031 * rbx: tmp5
7032 *
7033 */
7034 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) {
7035
7036 Label L_second_loop, L_second_loop_exit, L_third_loop, L_third_loop_exit, L_last_x, L_multiply;
7037 push(tmp1);
7038 push(tmp2);
7039 push(tmp3);
7040 push(tmp4);
7041 push(tmp5);
7042
7043 // First loop
7044 // Store the squares, right shifted one bit (i.e., divided by 2).
7045 square_rshift(x, len, z, tmp1, tmp3, tmp4, tmp5, rdxReg, raxReg);
7046
7047 // Add in off-diagonal sums.
7048 //
7049 // Second, third (nested) and fourth loops.
7050 // zlen +=2;
7051 // for (int xidx=len-2,zidx=zlen-4; xidx > 0; xidx-=2,zidx-=4) {
7052 // carry = 0;
7053 // long op2 = x[xidx:xidx+1];
7054 // for (int j=xidx-2,k=zidx; j >= 0; j-=2) {
7055 // k -= 2;
7056 // long op1 = x[j:j+1];
7057 // long sum = z[k:k+1];
7058 // carry:sum = multiply_add_64(sum, op1, op2, carry, tmp_regs);
7059 // z[k:k+1] = sum;
7060 // }
7061 // add_one_64(z, k, carry, tmp_regs);
7062 // }
7063
7064 const Register carry = tmp5;
7065 const Register sum = tmp3;
7066 const Register op1 = tmp4;
7067 Register op2 = tmp2;
7068
7069 push(zlen);
7070 push(len);
7071 addl(zlen,2);
7072 bind(L_second_loop);
7073 xorq(carry, carry);
7074 subl(zlen, 4);
7075 subl(len, 2);
7076 push(zlen);
7077 push(len);
7078 cmpl(len, 0);
7079 jccb(Assembler::lessEqual, L_second_loop_exit);
7080
7081 // Multiply an array by one 64 bit long.
7082 if (UseBMI2Instructions) {
7083 op2 = rdxReg;
7084 movq(op2, Address(x, len, Address::times_4, 0));
7085 rorxq(op2, op2, 32);
7086 }
7087 else {
7088 movq(op2, Address(x, len, Address::times_4, 0));
7089 rorq(op2, 32);
7090 }
7091
7092 bind(L_third_loop);
7093 decrementl(len);
7094 jccb(Assembler::negative, L_third_loop_exit);
7095 decrementl(len);
7096 jccb(Assembler::negative, L_last_x);
7097
7098 movq(op1, Address(x, len, Address::times_4, 0));
7099 rorq(op1, 32);
7100
7101 bind(L_multiply);
7102 subl(zlen, 2);
7103 movq(sum, Address(z, zlen, Address::times_4, 0));
7104
7105 // Multiply 64 bit by 64 bit and add 64 bits lower half and upper 64 bits as carry.
7106 if (UseBMI2Instructions) {
7107 multiply_add_64_bmi2(sum, op1, op2, carry, tmp2);
7108 }
7109 else {
7110 multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
7111 }
7112
7113 movq(Address(z, zlen, Address::times_4, 0), sum);
7114
7115 jmp(L_third_loop);
7116 bind(L_third_loop_exit);
7117
7118 // Fourth loop
7119 // Add 64 bit long carry into z with carry propagation.
7120 // Uses offsetted zlen.
7121 add_one_64(z, zlen, carry, tmp1);
7122
7123 pop(len);
7124 pop(zlen);
7125 jmp(L_second_loop);
7126
7127 // Next infrequent code is moved outside loops.
7128 bind(L_last_x);
7129 movl(op1, Address(x, 0));
7130 jmp(L_multiply);
7131
7132 bind(L_second_loop_exit);
7133 pop(len);
7134 pop(zlen);
7135 pop(len);
7136 pop(zlen);
7137
7138 // Fifth loop
7139 // Shift z left 1 bit.
7140 lshift_by_1(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4);
7141
7142 // z[zlen-1] |= x[len-1] & 1;
7143 movl(tmp3, Address(x, len, Address::times_4, -4));
7144 andl(tmp3, 1);
7145 orl(Address(z, zlen, Address::times_4, -4), tmp3);
7146
7147 pop(tmp5);
7148 pop(tmp4);
7149 pop(tmp3);
7150 pop(tmp2);
7151 pop(tmp1);
7152 }
7153
7154 /**
7155 * Helper function for mul_add()
7156 * Multiply the in[] by int k and add to out[] starting at offset offs using
7157 * 128 bit by 32 bit multiply and return the carry in tmp5.
7158 * Only quad int aligned length of in[] is operated on in this function.
7159 * k is in rdxReg for BMI2Instructions, for others it is in tmp2.
7160 * This function preserves out, in and k registers.
7161 * len and offset point to the appropriate index in "in" & "out" correspondingly
7162 * tmp5 has the carry.
7163 * other registers are temporary and are modified.
7164 *
7165 */
7166 void MacroAssembler::mul_add_128_x_32_loop(Register out, Register in,
7167 Register offset, Register len, Register tmp1, Register tmp2, Register tmp3,
7168 Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
7169
7170 Label L_first_loop, L_first_loop_exit;
7171
7172 movl(tmp1, len);
7173 shrl(tmp1, 2);
7174
7175 bind(L_first_loop);
7176 subl(tmp1, 1);
7177 jccb(Assembler::negative, L_first_loop_exit);
7178
7179 subl(len, 4);
7180 subl(offset, 4);
7181
7182 Register op2 = tmp2;
7183 const Register sum = tmp3;
7184 const Register op1 = tmp4;
7185 const Register carry = tmp5;
7186
7187 if (UseBMI2Instructions) {
7188 op2 = rdxReg;
7189 }
7190
7191 movq(op1, Address(in, len, Address::times_4, 8));
7192 rorq(op1, 32);
7193 movq(sum, Address(out, offset, Address::times_4, 8));
7194 rorq(sum, 32);
7195 if (UseBMI2Instructions) {
7196 multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
7197 }
7198 else {
7199 multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
7200 }
7201 // Store back in big endian from little endian
7202 rorq(sum, 0x20);
7203 movq(Address(out, offset, Address::times_4, 8), sum);
7204
7205 movq(op1, Address(in, len, Address::times_4, 0));
7206 rorq(op1, 32);
7207 movq(sum, Address(out, offset, Address::times_4, 0));
7208 rorq(sum, 32);
7209 if (UseBMI2Instructions) {
7210 multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
7211 }
7212 else {
7213 multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
7214 }
7215 // Store back in big endian from little endian
7216 rorq(sum, 0x20);
7217 movq(Address(out, offset, Address::times_4, 0), sum);
7218
7219 jmp(L_first_loop);
7220 bind(L_first_loop_exit);
7221 }
7222
7223 /**
7224 * Code for BigInteger::mulAdd() intrinsic
7225 *
7226 * rdi: out
7227 * rsi: in
7228 * r11: offs (out.length - offset)
7229 * rcx: len
7230 * r8: k
7231 * r12: tmp1
7232 * r13: tmp2
7233 * r14: tmp3
7234 * r15: tmp4
7235 * rbx: tmp5
7236 * Multiply the in[] by word k and add to out[], return the carry in rax
7237 */
7238 void MacroAssembler::mul_add(Register out, Register in, Register offs,
7239 Register len, Register k, Register tmp1, Register tmp2, Register tmp3,
7240 Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
7241
7242 Label L_carry, L_last_in, L_done;
7243
7244 // carry = 0;
7245 // for (int j=len-1; j >= 0; j--) {
7246 // long product = (in[j] & LONG_MASK) * kLong +
7247 // (out[offs] & LONG_MASK) + carry;
7248 // out[offs--] = (int)product;
7249 // carry = product >>> 32;
7250 // }
7251 //
7252 push(tmp1);
7253 push(tmp2);
7254 push(tmp3);
7255 push(tmp4);
7256 push(tmp5);
7257
7258 Register op2 = tmp2;
7259 const Register sum = tmp3;
7260 const Register op1 = tmp4;
7261 const Register carry = tmp5;
7262
7263 if (UseBMI2Instructions) {
7264 op2 = rdxReg;
7265 movl(op2, k);
7266 }
7267 else {
7268 movl(op2, k);
7269 }
7270
7271 xorq(carry, carry);
7272
7273 //First loop
7274
7275 //Multiply in[] by k in a 4 way unrolled loop using 128 bit by 32 bit multiply
7276 //The carry is in tmp5
7277 mul_add_128_x_32_loop(out, in, offs, len, tmp1, tmp2, tmp3, tmp4, tmp5, rdxReg, raxReg);
7278
7279 //Multiply the trailing in[] entry using 64 bit by 32 bit, if any
7280 decrementl(len);
7281 jccb(Assembler::negative, L_carry);
7282 decrementl(len);
7283 jccb(Assembler::negative, L_last_in);
7284
7285 movq(op1, Address(in, len, Address::times_4, 0));
7286 rorq(op1, 32);
7287
7288 subl(offs, 2);
7289 movq(sum, Address(out, offs, Address::times_4, 0));
7290 rorq(sum, 32);
7291
7292 if (UseBMI2Instructions) {
7293 multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
7294 }
7295 else {
7296 multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
7297 }
7298
7299 // Store back in big endian from little endian
7300 rorq(sum, 0x20);
7301 movq(Address(out, offs, Address::times_4, 0), sum);
7302
7303 testl(len, len);
7304 jccb(Assembler::zero, L_carry);
7305
7306 //Multiply the last in[] entry, if any
7307 bind(L_last_in);
7308 movl(op1, Address(in, 0));
7309 movl(sum, Address(out, offs, Address::times_4, -4));
7310
7311 movl(raxReg, k);
7312 mull(op1); //tmp4 * eax -> edx:eax
7313 addl(sum, carry);
7314 adcl(rdxReg, 0);
7315 addl(sum, raxReg);
7316 adcl(rdxReg, 0);
7317 movl(carry, rdxReg);
7318
7319 movl(Address(out, offs, Address::times_4, -4), sum);
7320
7321 bind(L_carry);
7322 //return tmp5/carry as carry in rax
7323 movl(rax, carry);
7324
7325 bind(L_done);
7326 pop(tmp5);
7327 pop(tmp4);
7328 pop(tmp3);
7329 pop(tmp2);
7330 pop(tmp1);
7331 }
7332
7333 /**
7334 * Emits code to update CRC-32 with a byte value according to constants in table
7335 *
7336 * @param [in,out]crc Register containing the crc.
7337 * @param [in]val Register containing the byte to fold into the CRC.
7338 * @param [in]table Register containing the table of crc constants.
7339 *
7340 * uint32_t crc;
7341 * val = crc_table[(val ^ crc) & 0xFF];
7342 * crc = val ^ (crc >> 8);
7343 *
7344 */
7345 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
7346 xorl(val, crc);
7347 andl(val, 0xFF);
7348 shrl(crc, 8); // unsigned shift
7349 xorl(crc, Address(table, val, Address::times_4, 0));
7350 }
7351
7352 /**
7353 * Fold 128-bit data chunk
7354 */
7355 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset) {
7356 if (UseAVX > 0) {
7357 vpclmulhdq(xtmp, xK, xcrc); // [123:64]
7358 vpclmulldq(xcrc, xK, xcrc); // [63:0]
7359 vpxor(xcrc, xcrc, Address(buf, offset), 0 /* vector_len */);
7360 pxor(xcrc, xtmp);
7361 } else {
7362 movdqa(xtmp, xcrc);
7363 pclmulhdq(xtmp, xK); // [123:64]
7364 pclmulldq(xcrc, xK); // [63:0]
7365 pxor(xcrc, xtmp);
7366 movdqu(xtmp, Address(buf, offset));
7367 pxor(xcrc, xtmp);
7368 }
7369 }
7370
7371 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, XMMRegister xbuf) {
7372 if (UseAVX > 0) {
7373 vpclmulhdq(xtmp, xK, xcrc);
7374 vpclmulldq(xcrc, xK, xcrc);
7375 pxor(xcrc, xbuf);
7376 pxor(xcrc, xtmp);
7377 } else {
7378 movdqa(xtmp, xcrc);
7379 pclmulhdq(xtmp, xK);
7380 pclmulldq(xcrc, xK);
7381 pxor(xcrc, xbuf);
7382 pxor(xcrc, xtmp);
7383 }
7384 }
7385
7386 /**
7387 * 8-bit folds to compute 32-bit CRC
7388 *
7389 * uint64_t xcrc;
7390 * timesXtoThe32[xcrc & 0xFF] ^ (xcrc >> 8);
7391 */
7392 void MacroAssembler::fold_8bit_crc32(XMMRegister xcrc, Register table, XMMRegister xtmp, Register tmp) {
7393 movdl(tmp, xcrc);
7394 andl(tmp, 0xFF);
7395 movdl(xtmp, Address(table, tmp, Address::times_4, 0));
7396 psrldq(xcrc, 1); // unsigned shift one byte
7397 pxor(xcrc, xtmp);
7398 }
7399
7400 /**
7401 * uint32_t crc;
7402 * timesXtoThe32[crc & 0xFF] ^ (crc >> 8);
7403 */
7404 void MacroAssembler::fold_8bit_crc32(Register crc, Register table, Register tmp) {
7405 movl(tmp, crc);
7406 andl(tmp, 0xFF);
7407 shrl(crc, 8);
7408 xorl(crc, Address(table, tmp, Address::times_4, 0));
7409 }
7410
7411 /**
7412 * @param crc register containing existing CRC (32-bit)
7413 * @param buf register pointing to input byte buffer (byte*)
7414 * @param len register containing number of bytes
7415 * @param table register that will contain address of CRC table
7416 * @param tmp scratch register
7417 */
7418 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len, Register table, Register tmp) {
7419 assert_different_registers(crc, buf, len, table, tmp, rax);
7420
7421 Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
7422 Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
7423
7424 // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
7425 // context for the registers used, where all instructions below are using 128-bit mode
7426 // On EVEX without VL and BW, these instructions will all be AVX.
7427 lea(table, ExternalAddress(StubRoutines::crc_table_addr()));
7428 notl(crc); // ~crc
7429 cmpl(len, 16);
7430 jcc(Assembler::less, L_tail);
7431
7432 // Align buffer to 16 bytes
7433 movl(tmp, buf);
7434 andl(tmp, 0xF);
7435 jccb(Assembler::zero, L_aligned);
7436 subl(tmp, 16);
7437 addl(len, tmp);
7438
7439 align(4);
7440 BIND(L_align_loop);
7441 movsbl(rax, Address(buf, 0)); // load byte with sign extension
7442 update_byte_crc32(crc, rax, table);
7443 increment(buf);
7444 incrementl(tmp);
7445 jccb(Assembler::less, L_align_loop);
7446
7447 BIND(L_aligned);
7448 movl(tmp, len); // save
7449 shrl(len, 4);
7450 jcc(Assembler::zero, L_tail_restore);
7451
7452 // Fold crc into first bytes of vector
7453 movdqa(xmm1, Address(buf, 0));
7454 movdl(rax, xmm1);
7455 xorl(crc, rax);
7456 if (VM_Version::supports_sse4_1()) {
7457 pinsrd(xmm1, crc, 0);
7458 } else {
7459 pinsrw(xmm1, crc, 0);
7460 shrl(crc, 16);
7461 pinsrw(xmm1, crc, 1);
7462 }
7463 addptr(buf, 16);
7464 subl(len, 4); // len > 0
7465 jcc(Assembler::less, L_fold_tail);
7466
7467 movdqa(xmm2, Address(buf, 0));
7468 movdqa(xmm3, Address(buf, 16));
7469 movdqa(xmm4, Address(buf, 32));
7470 addptr(buf, 48);
7471 subl(len, 3);
7472 jcc(Assembler::lessEqual, L_fold_512b);
7473
7474 // Fold total 512 bits of polynomial on each iteration,
7475 // 128 bits per each of 4 parallel streams.
7476 movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 32), rscratch1);
7477
7478 align32();
7479 BIND(L_fold_512b_loop);
7480 fold_128bit_crc32(xmm1, xmm0, xmm5, buf, 0);
7481 fold_128bit_crc32(xmm2, xmm0, xmm5, buf, 16);
7482 fold_128bit_crc32(xmm3, xmm0, xmm5, buf, 32);
7483 fold_128bit_crc32(xmm4, xmm0, xmm5, buf, 48);
7484 addptr(buf, 64);
7485 subl(len, 4);
7486 jcc(Assembler::greater, L_fold_512b_loop);
7487
7488 // Fold 512 bits to 128 bits.
7489 BIND(L_fold_512b);
7490 movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
7491 fold_128bit_crc32(xmm1, xmm0, xmm5, xmm2);
7492 fold_128bit_crc32(xmm1, xmm0, xmm5, xmm3);
7493 fold_128bit_crc32(xmm1, xmm0, xmm5, xmm4);
7494
7495 // Fold the rest of 128 bits data chunks
7496 BIND(L_fold_tail);
7497 addl(len, 3);
7498 jccb(Assembler::lessEqual, L_fold_128b);
7499 movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
7500
7501 BIND(L_fold_tail_loop);
7502 fold_128bit_crc32(xmm1, xmm0, xmm5, buf, 0);
7503 addptr(buf, 16);
7504 decrementl(len);
7505 jccb(Assembler::greater, L_fold_tail_loop);
7506
7507 // Fold 128 bits in xmm1 down into 32 bits in crc register.
7508 BIND(L_fold_128b);
7509 movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr()), rscratch1);
7510 if (UseAVX > 0) {
7511 vpclmulqdq(xmm2, xmm0, xmm1, 0x1);
7512 vpand(xmm3, xmm0, xmm2, 0 /* vector_len */);
7513 vpclmulqdq(xmm0, xmm0, xmm3, 0x1);
7514 } else {
7515 movdqa(xmm2, xmm0);
7516 pclmulqdq(xmm2, xmm1, 0x1);
7517 movdqa(xmm3, xmm0);
7518 pand(xmm3, xmm2);
7519 pclmulqdq(xmm0, xmm3, 0x1);
7520 }
7521 psrldq(xmm1, 8);
7522 psrldq(xmm2, 4);
7523 pxor(xmm0, xmm1);
7524 pxor(xmm0, xmm2);
7525
7526 // 8 8-bit folds to compute 32-bit CRC.
7527 for (int j = 0; j < 4; j++) {
7528 fold_8bit_crc32(xmm0, table, xmm1, rax);
7529 }
7530 movdl(crc, xmm0); // mov 32 bits to general register
7531 for (int j = 0; j < 4; j++) {
7532 fold_8bit_crc32(crc, table, rax);
7533 }
7534
7535 BIND(L_tail_restore);
7536 movl(len, tmp); // restore
7537 BIND(L_tail);
7538 andl(len, 0xf);
7539 jccb(Assembler::zero, L_exit);
7540
7541 // Fold the rest of bytes
7542 align(4);
7543 BIND(L_tail_loop);
7544 movsbl(rax, Address(buf, 0)); // load byte with sign extension
7545 update_byte_crc32(crc, rax, table);
7546 increment(buf);
7547 decrementl(len);
7548 jccb(Assembler::greater, L_tail_loop);
7549
7550 BIND(L_exit);
7551 notl(crc); // ~c
7552 }
7553
7554 // Helper function for AVX 512 CRC32
7555 // Fold 512-bit data chunks
7556 void MacroAssembler::fold512bit_crc32_avx512(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf,
7557 Register pos, int offset) {
7558 evmovdquq(xmm3, Address(buf, pos, Address::times_1, offset), Assembler::AVX_512bit);
7559 evpclmulqdq(xtmp, xcrc, xK, 0x10, Assembler::AVX_512bit); // [123:64]
7560 evpclmulqdq(xmm2, xcrc, xK, 0x01, Assembler::AVX_512bit); // [63:0]
7561 evpxorq(xcrc, xtmp, xmm2, Assembler::AVX_512bit /* vector_len */);
7562 evpxorq(xcrc, xcrc, xmm3, Assembler::AVX_512bit /* vector_len */);
7563 }
7564
7565 // Helper function for AVX 512 CRC32
7566 // Compute CRC32 for < 256B buffers
7567 void MacroAssembler::kernel_crc32_avx512_256B(Register crc, Register buf, Register len, Register table, Register pos,
7568 Register tmp1, Register tmp2, Label& L_barrett, Label& L_16B_reduction_loop,
7569 Label& L_get_last_two_xmms, Label& L_128_done, Label& L_cleanup) {
7570
7571 Label L_less_than_32, L_exact_16_left, L_less_than_16_left;
7572 Label L_less_than_8_left, L_less_than_4_left, L_less_than_2_left, L_zero_left;
7573 Label L_only_less_than_4, L_only_less_than_3, L_only_less_than_2;
7574
7575 // check if there is enough buffer to be able to fold 16B at a time
7576 cmpl(len, 32);
7577 jcc(Assembler::less, L_less_than_32);
7578
7579 // if there is, load the constants
7580 movdqu(xmm10, Address(table, 1 * 16)); //rk1 and rk2 in xmm10
7581 movdl(xmm0, crc); // get the initial crc value
7582 movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
7583 pxor(xmm7, xmm0);
7584
7585 // update the buffer pointer
7586 addl(pos, 16);
7587 //update the counter.subtract 32 instead of 16 to save one instruction from the loop
7588 subl(len, 32);
7589 jmp(L_16B_reduction_loop);
7590
7591 bind(L_less_than_32);
7592 //mov initial crc to the return value. this is necessary for zero - length buffers.
7593 movl(rax, crc);
7594 testl(len, len);
7595 jcc(Assembler::equal, L_cleanup);
7596
7597 movdl(xmm0, crc); //get the initial crc value
7598
7599 cmpl(len, 16);
7600 jcc(Assembler::equal, L_exact_16_left);
7601 jcc(Assembler::less, L_less_than_16_left);
7602
7603 movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
7604 pxor(xmm7, xmm0); //xor the initial crc value
7605 addl(pos, 16);
7606 subl(len, 16);
7607 movdqu(xmm10, Address(table, 1 * 16)); // rk1 and rk2 in xmm10
7608 jmp(L_get_last_two_xmms);
7609
7610 bind(L_less_than_16_left);
7611 //use stack space to load data less than 16 bytes, zero - out the 16B in memory first.
7612 pxor(xmm1, xmm1);
7613 movptr(tmp1, rsp);
7614 movdqu(Address(tmp1, 0 * 16), xmm1);
7615
7616 cmpl(len, 4);
7617 jcc(Assembler::less, L_only_less_than_4);
7618
7619 //backup the counter value
7620 movl(tmp2, len);
7621 cmpl(len, 8);
7622 jcc(Assembler::less, L_less_than_8_left);
7623
7624 //load 8 Bytes
7625 movq(rax, Address(buf, pos, Address::times_1, 0 * 16));
7626 movq(Address(tmp1, 0 * 16), rax);
7627 addptr(tmp1, 8);
7628 subl(len, 8);
7629 addl(pos, 8);
7630
7631 bind(L_less_than_8_left);
7632 cmpl(len, 4);
7633 jcc(Assembler::less, L_less_than_4_left);
7634
7635 //load 4 Bytes
7636 movl(rax, Address(buf, pos, Address::times_1, 0));
7637 movl(Address(tmp1, 0 * 16), rax);
7638 addptr(tmp1, 4);
7639 subl(len, 4);
7640 addl(pos, 4);
7641
7642 bind(L_less_than_4_left);
7643 cmpl(len, 2);
7644 jcc(Assembler::less, L_less_than_2_left);
7645
7646 // load 2 Bytes
7647 movw(rax, Address(buf, pos, Address::times_1, 0));
7648 movl(Address(tmp1, 0 * 16), rax);
7649 addptr(tmp1, 2);
7650 subl(len, 2);
7651 addl(pos, 2);
7652
7653 bind(L_less_than_2_left);
7654 cmpl(len, 1);
7655 jcc(Assembler::less, L_zero_left);
7656
7657 // load 1 Byte
7658 movb(rax, Address(buf, pos, Address::times_1, 0));
7659 movb(Address(tmp1, 0 * 16), rax);
7660
7661 bind(L_zero_left);
7662 movdqu(xmm7, Address(rsp, 0));
7663 pxor(xmm7, xmm0); //xor the initial crc value
7664
7665 lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
7666 movdqu(xmm0, Address(rax, tmp2));
7667 pshufb(xmm7, xmm0);
7668 jmp(L_128_done);
7669
7670 bind(L_exact_16_left);
7671 movdqu(xmm7, Address(buf, pos, Address::times_1, 0));
7672 pxor(xmm7, xmm0); //xor the initial crc value
7673 jmp(L_128_done);
7674
7675 bind(L_only_less_than_4);
7676 cmpl(len, 3);
7677 jcc(Assembler::less, L_only_less_than_3);
7678
7679 // load 3 Bytes
7680 movb(rax, Address(buf, pos, Address::times_1, 0));
7681 movb(Address(tmp1, 0), rax);
7682
7683 movb(rax, Address(buf, pos, Address::times_1, 1));
7684 movb(Address(tmp1, 1), rax);
7685
7686 movb(rax, Address(buf, pos, Address::times_1, 2));
7687 movb(Address(tmp1, 2), rax);
7688
7689 movdqu(xmm7, Address(rsp, 0));
7690 pxor(xmm7, xmm0); //xor the initial crc value
7691
7692 pslldq(xmm7, 0x5);
7693 jmp(L_barrett);
7694 bind(L_only_less_than_3);
7695 cmpl(len, 2);
7696 jcc(Assembler::less, L_only_less_than_2);
7697
7698 // load 2 Bytes
7699 movb(rax, Address(buf, pos, Address::times_1, 0));
7700 movb(Address(tmp1, 0), rax);
7701
7702 movb(rax, Address(buf, pos, Address::times_1, 1));
7703 movb(Address(tmp1, 1), rax);
7704
7705 movdqu(xmm7, Address(rsp, 0));
7706 pxor(xmm7, xmm0); //xor the initial crc value
7707
7708 pslldq(xmm7, 0x6);
7709 jmp(L_barrett);
7710
7711 bind(L_only_less_than_2);
7712 //load 1 Byte
7713 movb(rax, Address(buf, pos, Address::times_1, 0));
7714 movb(Address(tmp1, 0), rax);
7715
7716 movdqu(xmm7, Address(rsp, 0));
7717 pxor(xmm7, xmm0); //xor the initial crc value
7718
7719 pslldq(xmm7, 0x7);
7720 }
7721
7722 /**
7723 * Compute CRC32 using AVX512 instructions
7724 * param crc register containing existing CRC (32-bit)
7725 * param buf register pointing to input byte buffer (byte*)
7726 * param len register containing number of bytes
7727 * param table address of crc or crc32c table
7728 * param tmp1 scratch register
7729 * param tmp2 scratch register
7730 * return rax result register
7731 *
7732 * This routine is identical for crc32c with the exception of the precomputed constant
7733 * table which will be passed as the table argument. The calculation steps are
7734 * the same for both variants.
7735 */
7736 void MacroAssembler::kernel_crc32_avx512(Register crc, Register buf, Register len, Register table, Register tmp1, Register tmp2) {
7737 assert_different_registers(crc, buf, len, table, tmp1, tmp2, rax, r12);
7738
7739 Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
7740 Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
7741 Label L_less_than_256, L_fold_128_B_loop, L_fold_256_B_loop;
7742 Label L_fold_128_B_register, L_final_reduction_for_128, L_16B_reduction_loop;
7743 Label L_128_done, L_get_last_two_xmms, L_barrett, L_cleanup;
7744
7745 const Register pos = r12;
7746 push(r12);
7747 subptr(rsp, 16 * 2 + 8);
7748
7749 // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
7750 // context for the registers used, where all instructions below are using 128-bit mode
7751 // On EVEX without VL and BW, these instructions will all be AVX.
7752 movl(pos, 0);
7753
7754 // check if smaller than 256B
7755 cmpl(len, 256);
7756 jcc(Assembler::less, L_less_than_256);
7757
7758 // load the initial crc value
7759 movdl(xmm10, crc);
7760
7761 // receive the initial 64B data, xor the initial crc value
7762 evmovdquq(xmm0, Address(buf, pos, Address::times_1, 0 * 64), Assembler::AVX_512bit);
7763 evmovdquq(xmm4, Address(buf, pos, Address::times_1, 1 * 64), Assembler::AVX_512bit);
7764 evpxorq(xmm0, xmm0, xmm10, Assembler::AVX_512bit);
7765 evbroadcasti32x4(xmm10, Address(table, 2 * 16), Assembler::AVX_512bit); //zmm10 has rk3 and rk4
7766
7767 subl(len, 256);
7768 cmpl(len, 256);
7769 jcc(Assembler::less, L_fold_128_B_loop);
7770
7771 evmovdquq(xmm7, Address(buf, pos, Address::times_1, 2 * 64), Assembler::AVX_512bit);
7772 evmovdquq(xmm8, Address(buf, pos, Address::times_1, 3 * 64), Assembler::AVX_512bit);
7773 evbroadcasti32x4(xmm16, Address(table, 0 * 16), Assembler::AVX_512bit); //zmm16 has rk-1 and rk-2
7774 subl(len, 256);
7775
7776 bind(L_fold_256_B_loop);
7777 addl(pos, 256);
7778 fold512bit_crc32_avx512(xmm0, xmm16, xmm1, buf, pos, 0 * 64);
7779 fold512bit_crc32_avx512(xmm4, xmm16, xmm1, buf, pos, 1 * 64);
7780 fold512bit_crc32_avx512(xmm7, xmm16, xmm1, buf, pos, 2 * 64);
7781 fold512bit_crc32_avx512(xmm8, xmm16, xmm1, buf, pos, 3 * 64);
7782
7783 subl(len, 256);
7784 jcc(Assembler::greaterEqual, L_fold_256_B_loop);
7785
7786 // Fold 256 into 128
7787 addl(pos, 256);
7788 evpclmulqdq(xmm1, xmm0, xmm10, 0x01, Assembler::AVX_512bit);
7789 evpclmulqdq(xmm2, xmm0, xmm10, 0x10, Assembler::AVX_512bit);
7790 vpternlogq(xmm7, 0x96, xmm1, xmm2, Assembler::AVX_512bit); // xor ABC
7791
7792 evpclmulqdq(xmm5, xmm4, xmm10, 0x01, Assembler::AVX_512bit);
7793 evpclmulqdq(xmm6, xmm4, xmm10, 0x10, Assembler::AVX_512bit);
7794 vpternlogq(xmm8, 0x96, xmm5, xmm6, Assembler::AVX_512bit); // xor ABC
7795
7796 evmovdquq(xmm0, xmm7, Assembler::AVX_512bit);
7797 evmovdquq(xmm4, xmm8, Assembler::AVX_512bit);
7798
7799 addl(len, 128);
7800 jmp(L_fold_128_B_register);
7801
7802 // at this section of the code, there is 128 * x + y(0 <= y<128) bytes of buffer.The fold_128_B_loop
7803 // loop will fold 128B at a time until we have 128 + y Bytes of buffer
7804
7805 // fold 128B at a time.This section of the code folds 8 xmm registers in parallel
7806 bind(L_fold_128_B_loop);
7807 addl(pos, 128);
7808 fold512bit_crc32_avx512(xmm0, xmm10, xmm1, buf, pos, 0 * 64);
7809 fold512bit_crc32_avx512(xmm4, xmm10, xmm1, buf, pos, 1 * 64);
7810
7811 subl(len, 128);
7812 jcc(Assembler::greaterEqual, L_fold_128_B_loop);
7813
7814 addl(pos, 128);
7815
7816 // at this point, the buffer pointer is pointing at the last y Bytes of the buffer, where 0 <= y < 128
7817 // the 128B of folded data is in 8 of the xmm registers : xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7
7818 bind(L_fold_128_B_register);
7819 evmovdquq(xmm16, Address(table, 5 * 16), Assembler::AVX_512bit); // multiply by rk9-rk16
7820 evmovdquq(xmm11, Address(table, 9 * 16), Assembler::AVX_512bit); // multiply by rk17-rk20, rk1,rk2, 0,0
7821 evpclmulqdq(xmm1, xmm0, xmm16, 0x01, Assembler::AVX_512bit);
7822 evpclmulqdq(xmm2, xmm0, xmm16, 0x10, Assembler::AVX_512bit);
7823 // save last that has no multiplicand
7824 vextracti64x2(xmm7, xmm4, 3);
7825
7826 evpclmulqdq(xmm5, xmm4, xmm11, 0x01, Assembler::AVX_512bit);
7827 evpclmulqdq(xmm6, xmm4, xmm11, 0x10, Assembler::AVX_512bit);
7828 // Needed later in reduction loop
7829 movdqu(xmm10, Address(table, 1 * 16));
7830 vpternlogq(xmm1, 0x96, xmm2, xmm5, Assembler::AVX_512bit); // xor ABC
7831 vpternlogq(xmm1, 0x96, xmm6, xmm7, Assembler::AVX_512bit); // xor ABC
7832
7833 // Swap 1,0,3,2 - 01 00 11 10
7834 evshufi64x2(xmm8, xmm1, xmm1, 0x4e, Assembler::AVX_512bit);
7835 evpxorq(xmm8, xmm8, xmm1, Assembler::AVX_256bit);
7836 vextracti128(xmm5, xmm8, 1);
7837 evpxorq(xmm7, xmm5, xmm8, Assembler::AVX_128bit);
7838
7839 // instead of 128, we add 128 - 16 to the loop counter to save 1 instruction from the loop
7840 // instead of a cmp instruction, we use the negative flag with the jl instruction
7841 addl(len, 128 - 16);
7842 jcc(Assembler::less, L_final_reduction_for_128);
7843
7844 bind(L_16B_reduction_loop);
7845 vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
7846 vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
7847 vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
7848 movdqu(xmm0, Address(buf, pos, Address::times_1, 0 * 16));
7849 vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
7850 addl(pos, 16);
7851 subl(len, 16);
7852 jcc(Assembler::greaterEqual, L_16B_reduction_loop);
7853
7854 bind(L_final_reduction_for_128);
7855 addl(len, 16);
7856 jcc(Assembler::equal, L_128_done);
7857
7858 bind(L_get_last_two_xmms);
7859 movdqu(xmm2, xmm7);
7860 addl(pos, len);
7861 movdqu(xmm1, Address(buf, pos, Address::times_1, -16));
7862 subl(pos, len);
7863
7864 // get rid of the extra data that was loaded before
7865 // load the shift constant
7866 lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
7867 movdqu(xmm0, Address(rax, len));
7868 addl(rax, len);
7869
7870 vpshufb(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
7871 //Change mask to 512
7872 vpxor(xmm0, xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 2 * 16), Assembler::AVX_128bit, tmp2);
7873 vpshufb(xmm2, xmm2, xmm0, Assembler::AVX_128bit);
7874
7875 blendvpb(xmm2, xmm2, xmm1, xmm0, Assembler::AVX_128bit);
7876 vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
7877 vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
7878 vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
7879 vpxor(xmm7, xmm7, xmm2, Assembler::AVX_128bit);
7880
7881 bind(L_128_done);
7882 // compute crc of a 128-bit value
7883 movdqu(xmm10, Address(table, 3 * 16));
7884 movdqu(xmm0, xmm7);
7885
7886 // 64b fold
7887 vpclmulqdq(xmm7, xmm7, xmm10, 0x0);
7888 vpsrldq(xmm0, xmm0, 0x8, Assembler::AVX_128bit);
7889 vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
7890
7891 // 32b fold
7892 movdqu(xmm0, xmm7);
7893 vpslldq(xmm7, xmm7, 0x4, Assembler::AVX_128bit);
7894 vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
7895 vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
7896 jmp(L_barrett);
7897
7898 bind(L_less_than_256);
7899 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);
7900
7901 //barrett reduction
7902 bind(L_barrett);
7903 vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 1 * 16), Assembler::AVX_128bit, tmp2);
7904 movdqu(xmm1, xmm7);
7905 movdqu(xmm2, xmm7);
7906 movdqu(xmm10, Address(table, 4 * 16));
7907
7908 pclmulqdq(xmm7, xmm10, 0x0);
7909 pxor(xmm7, xmm2);
7910 vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr()), Assembler::AVX_128bit, tmp2);
7911 movdqu(xmm2, xmm7);
7912 pclmulqdq(xmm7, xmm10, 0x10);
7913 pxor(xmm7, xmm2);
7914 pxor(xmm7, xmm1);
7915 pextrd(crc, xmm7, 2);
7916
7917 bind(L_cleanup);
7918 addptr(rsp, 16 * 2 + 8);
7919 pop(r12);
7920 }
7921
7922 // S. Gueron / Information Processing Letters 112 (2012) 184
7923 // Algorithm 4: Computing carry-less multiplication using a precomputed lookup table.
7924 // Input: A 32 bit value B = [byte3, byte2, byte1, byte0].
7925 // Output: the 64-bit carry-less product of B * CONST
7926 void MacroAssembler::crc32c_ipl_alg4(Register in, uint32_t n,
7927 Register tmp1, Register tmp2, Register tmp3) {
7928 lea(tmp3, ExternalAddress(StubRoutines::crc32c_table_addr()));
7929 if (n > 0) {
7930 addq(tmp3, n * 256 * 8);
7931 }
7932 // Q1 = TABLEExt[n][B & 0xFF];
7933 movl(tmp1, in);
7934 andl(tmp1, 0x000000FF);
7935 shll(tmp1, 3);
7936 addq(tmp1, tmp3);
7937 movq(tmp1, Address(tmp1, 0));
7938
7939 // Q2 = TABLEExt[n][B >> 8 & 0xFF];
7940 movl(tmp2, in);
7941 shrl(tmp2, 8);
7942 andl(tmp2, 0x000000FF);
7943 shll(tmp2, 3);
7944 addq(tmp2, tmp3);
7945 movq(tmp2, Address(tmp2, 0));
7946
7947 shlq(tmp2, 8);
7948 xorq(tmp1, tmp2);
7949
7950 // Q3 = TABLEExt[n][B >> 16 & 0xFF];
7951 movl(tmp2, in);
7952 shrl(tmp2, 16);
7953 andl(tmp2, 0x000000FF);
7954 shll(tmp2, 3);
7955 addq(tmp2, tmp3);
7956 movq(tmp2, Address(tmp2, 0));
7957
7958 shlq(tmp2, 16);
7959 xorq(tmp1, tmp2);
7960
7961 // Q4 = TABLEExt[n][B >> 24 & 0xFF];
7962 shrl(in, 24);
7963 andl(in, 0x000000FF);
7964 shll(in, 3);
7965 addq(in, tmp3);
7966 movq(in, Address(in, 0));
7967
7968 shlq(in, 24);
7969 xorq(in, tmp1);
7970 // return Q1 ^ Q2 << 8 ^ Q3 << 16 ^ Q4 << 24;
7971 }
7972
7973 void MacroAssembler::crc32c_pclmulqdq(XMMRegister w_xtmp1,
7974 Register in_out,
7975 uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported,
7976 XMMRegister w_xtmp2,
7977 Register tmp1,
7978 Register n_tmp2, Register n_tmp3) {
7979 if (is_pclmulqdq_supported) {
7980 movdl(w_xtmp1, in_out); // modified blindly
7981
7982 movl(tmp1, const_or_pre_comp_const_index);
7983 movdl(w_xtmp2, tmp1);
7984 pclmulqdq(w_xtmp1, w_xtmp2, 0);
7985
7986 movdq(in_out, w_xtmp1);
7987 } else {
7988 crc32c_ipl_alg4(in_out, const_or_pre_comp_const_index, tmp1, n_tmp2, n_tmp3);
7989 }
7990 }
7991
7992 // Recombination Alternative 2: No bit-reflections
7993 // T1 = (CRC_A * U1) << 1
7994 // T2 = (CRC_B * U2) << 1
7995 // C1 = T1 >> 32
7996 // C2 = T2 >> 32
7997 // T1 = T1 & 0xFFFFFFFF
7998 // T2 = T2 & 0xFFFFFFFF
7999 // T1 = CRC32(0, T1)
8000 // T2 = CRC32(0, T2)
8001 // C1 = C1 ^ T1
8002 // C2 = C2 ^ T2
8003 // CRC = C1 ^ C2 ^ CRC_C
8004 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,
8005 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
8006 Register tmp1, Register tmp2,
8007 Register n_tmp3) {
8008 crc32c_pclmulqdq(w_xtmp1, in_out, const_or_pre_comp_const_index_u1, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
8009 crc32c_pclmulqdq(w_xtmp2, in1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
8010 shlq(in_out, 1);
8011 movl(tmp1, in_out);
8012 shrq(in_out, 32);
8013 xorl(tmp2, tmp2);
8014 crc32(tmp2, tmp1, 4);
8015 xorl(in_out, tmp2); // we don't care about upper 32 bit contents here
8016 shlq(in1, 1);
8017 movl(tmp1, in1);
8018 shrq(in1, 32);
8019 xorl(tmp2, tmp2);
8020 crc32(tmp2, tmp1, 4);
8021 xorl(in1, tmp2);
8022 xorl(in_out, in1);
8023 xorl(in_out, in2);
8024 }
8025
8026 // Set N to predefined value
8027 // Subtract from a length of a buffer
8028 // execute in a loop:
8029 // CRC_A = 0xFFFFFFFF, CRC_B = 0, CRC_C = 0
8030 // for i = 1 to N do
8031 // CRC_A = CRC32(CRC_A, A[i])
8032 // CRC_B = CRC32(CRC_B, B[i])
8033 // CRC_C = CRC32(CRC_C, C[i])
8034 // end for
8035 // Recombine
8036 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,
8037 Register in_out1, Register in_out2, Register in_out3,
8038 Register tmp1, Register tmp2, Register tmp3,
8039 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
8040 Register tmp4, Register tmp5,
8041 Register n_tmp6) {
8042 Label L_processPartitions;
8043 Label L_processPartition;
8044 Label L_exit;
8045
8046 bind(L_processPartitions);
8047 cmpl(in_out1, 3 * size);
8048 jcc(Assembler::less, L_exit);
8049 xorl(tmp1, tmp1);
8050 xorl(tmp2, tmp2);
8051 movq(tmp3, in_out2);
8052 addq(tmp3, size);
8053
8054 bind(L_processPartition);
8055 crc32(in_out3, Address(in_out2, 0), 8);
8056 crc32(tmp1, Address(in_out2, size), 8);
8057 crc32(tmp2, Address(in_out2, size * 2), 8);
8058 addq(in_out2, 8);
8059 cmpq(in_out2, tmp3);
8060 jcc(Assembler::less, L_processPartition);
8061 crc32c_rec_alt2(const_or_pre_comp_const_index_u1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, in_out3, tmp1, tmp2,
8062 w_xtmp1, w_xtmp2, w_xtmp3,
8063 tmp4, tmp5,
8064 n_tmp6);
8065 addq(in_out2, 2 * size);
8066 subl(in_out1, 3 * size);
8067 jmp(L_processPartitions);
8068
8069 bind(L_exit);
8070 }
8071
8072 // Algorithm 2: Pipelined usage of the CRC32 instruction.
8073 // Input: A buffer I of L bytes.
8074 // Output: the CRC32C value of the buffer.
8075 // Notations:
8076 // Write L = 24N + r, with N = floor (L/24).
8077 // r = L mod 24 (0 <= r < 24).
8078 // Consider I as the concatenation of A|B|C|R, where A, B, C, each,
8079 // N quadwords, and R consists of r bytes.
8080 // A[j] = I [8j+7:8j], j= 0, 1, ..., N-1
8081 // B[j] = I [N + 8j+7:N + 8j], j= 0, 1, ..., N-1
8082 // C[j] = I [2N + 8j+7:2N + 8j], j= 0, 1, ..., N-1
8083 // if r > 0 R[j] = I [3N +j], j= 0, 1, ...,r-1
8084 void MacroAssembler::crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2,
8085 Register tmp1, Register tmp2, Register tmp3,
8086 Register tmp4, Register tmp5, Register tmp6,
8087 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
8088 bool is_pclmulqdq_supported) {
8089 uint32_t const_or_pre_comp_const_index[CRC32C_NUM_PRECOMPUTED_CONSTANTS];
8090 Label L_wordByWord;
8091 Label L_byteByByteProlog;
8092 Label L_byteByByte;
8093 Label L_exit;
8094
8095 if (is_pclmulqdq_supported ) {
8096 const_or_pre_comp_const_index[1] = *(uint32_t *)StubRoutines::crc32c_table_addr();
8097 const_or_pre_comp_const_index[0] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 1);
8098
8099 const_or_pre_comp_const_index[3] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 2);
8100 const_or_pre_comp_const_index[2] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 3);
8101
8102 const_or_pre_comp_const_index[5] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 4);
8103 const_or_pre_comp_const_index[4] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 5);
8104 assert((CRC32C_NUM_PRECOMPUTED_CONSTANTS - 1 ) == 5, "Checking whether you declared all of the constants based on the number of \"chunks\"");
8105 } else {
8106 const_or_pre_comp_const_index[0] = 1;
8107 const_or_pre_comp_const_index[1] = 0;
8108
8109 const_or_pre_comp_const_index[2] = 3;
8110 const_or_pre_comp_const_index[3] = 2;
8111
8112 const_or_pre_comp_const_index[4] = 5;
8113 const_or_pre_comp_const_index[5] = 4;
8114 }
8115 crc32c_proc_chunk(CRC32C_HIGH, const_or_pre_comp_const_index[0], const_or_pre_comp_const_index[1], is_pclmulqdq_supported,
8116 in2, in1, in_out,
8117 tmp1, tmp2, tmp3,
8118 w_xtmp1, w_xtmp2, w_xtmp3,
8119 tmp4, tmp5,
8120 tmp6);
8121 crc32c_proc_chunk(CRC32C_MIDDLE, const_or_pre_comp_const_index[2], const_or_pre_comp_const_index[3], is_pclmulqdq_supported,
8122 in2, in1, in_out,
8123 tmp1, tmp2, tmp3,
8124 w_xtmp1, w_xtmp2, w_xtmp3,
8125 tmp4, tmp5,
8126 tmp6);
8127 crc32c_proc_chunk(CRC32C_LOW, const_or_pre_comp_const_index[4], const_or_pre_comp_const_index[5], is_pclmulqdq_supported,
8128 in2, in1, in_out,
8129 tmp1, tmp2, tmp3,
8130 w_xtmp1, w_xtmp2, w_xtmp3,
8131 tmp4, tmp5,
8132 tmp6);
8133 movl(tmp1, in2);
8134 andl(tmp1, 0x00000007);
8135 negl(tmp1);
8136 addl(tmp1, in2);
8137 addq(tmp1, in1);
8138
8139 cmpq(in1, tmp1);
8140 jccb(Assembler::greaterEqual, L_byteByByteProlog);
8141 align(16);
8142 BIND(L_wordByWord);
8143 crc32(in_out, Address(in1, 0), 8);
8144 addq(in1, 8);
8145 cmpq(in1, tmp1);
8146 jcc(Assembler::less, L_wordByWord);
8147
8148 BIND(L_byteByByteProlog);
8149 andl(in2, 0x00000007);
8150 movl(tmp2, 1);
8151
8152 cmpl(tmp2, in2);
8153 jccb(Assembler::greater, L_exit);
8154 BIND(L_byteByByte);
8155 crc32(in_out, Address(in1, 0), 1);
8156 incq(in1);
8157 incl(tmp2);
8158 cmpl(tmp2, in2);
8159 jcc(Assembler::lessEqual, L_byteByByte);
8160
8161 BIND(L_exit);
8162 }
8163 #undef BIND
8164 #undef BLOCK_COMMENT
8165
8166 // Compress char[] array to byte[].
8167 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
8168 // Return the array length if every element in array can be encoded,
8169 // otherwise, the index of first non-latin1 (> 0xff) character.
8170 // @IntrinsicCandidate
8171 // public static int compress(char[] src, int srcOff, byte[] dst, int dstOff, int len) {
8172 // for (int i = 0; i < len; i++) {
8173 // char c = src[srcOff];
8174 // if (c > 0xff) {
8175 // return i; // return index of non-latin1 char
8176 // }
8177 // dst[dstOff] = (byte)c;
8178 // srcOff++;
8179 // dstOff++;
8180 // }
8181 // return len;
8182 // }
8183 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
8184 XMMRegister tmp1Reg, XMMRegister tmp2Reg,
8185 XMMRegister tmp3Reg, XMMRegister tmp4Reg,
8186 Register tmp5, Register result, KRegister mask1, KRegister mask2) {
8187 Label copy_chars_loop, done, reset_sp, copy_tail;
8188
8189 // rsi: src
8190 // rdi: dst
8191 // rdx: len
8192 // rcx: tmp5
8193 // rax: result
8194
8195 // rsi holds start addr of source char[] to be compressed
8196 // rdi holds start addr of destination byte[]
8197 // rdx holds length
8198
8199 assert(len != result, "");
8200
8201 // save length for return
8202 movl(result, len);
8203
8204 if ((AVX3Threshold == 0) && (UseAVX > 2) && // AVX512
8205 VM_Version::supports_avx512vlbw() &&
8206 VM_Version::supports_bmi2()) {
8207
8208 Label copy_32_loop, copy_loop_tail, below_threshold, reset_for_copy_tail;
8209
8210 // alignment
8211 Label post_alignment;
8212
8213 // if length of the string is less than 32, handle it the old fashioned way
8214 testl(len, -32);
8215 jcc(Assembler::zero, below_threshold);
8216
8217 // First check whether a character is compressible ( <= 0xFF).
8218 // Create mask to test for Unicode chars inside zmm vector
8219 movl(tmp5, 0x00FF);
8220 evpbroadcastw(tmp2Reg, tmp5, Assembler::AVX_512bit);
8221
8222 testl(len, -64);
8223 jccb(Assembler::zero, post_alignment);
8224
8225 movl(tmp5, dst);
8226 andl(tmp5, (32 - 1));
8227 negl(tmp5);
8228 andl(tmp5, (32 - 1));
8229
8230 // bail out when there is nothing to be done
8231 testl(tmp5, 0xFFFFFFFF);
8232 jccb(Assembler::zero, post_alignment);
8233
8234 // ~(~0 << len), where len is the # of remaining elements to process
8235 movl(len, 0xFFFFFFFF);
8236 shlxl(len, len, tmp5);
8237 notl(len);
8238 kmovdl(mask2, len);
8239 movl(len, result);
8240
8241 evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
8242 evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
8243 ktestd(mask1, mask2);
8244 jcc(Assembler::carryClear, copy_tail);
8245
8246 evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
8247
8248 addptr(src, tmp5);
8249 addptr(src, tmp5);
8250 addptr(dst, tmp5);
8251 subl(len, tmp5);
8252
8253 bind(post_alignment);
8254 // end of alignment
8255
8256 movl(tmp5, len);
8257 andl(tmp5, (32 - 1)); // tail count (in chars)
8258 andl(len, ~(32 - 1)); // vector count (in chars)
8259 jccb(Assembler::zero, copy_loop_tail);
8260
8261 lea(src, Address(src, len, Address::times_2));
8262 lea(dst, Address(dst, len, Address::times_1));
8263 negptr(len);
8264
8265 bind(copy_32_loop);
8266 evmovdquw(tmp1Reg, Address(src, len, Address::times_2), Assembler::AVX_512bit);
8267 evpcmpuw(mask1, tmp1Reg, tmp2Reg, Assembler::le, Assembler::AVX_512bit);
8268 kortestdl(mask1, mask1);
8269 jccb(Assembler::carryClear, reset_for_copy_tail);
8270
8271 // All elements in current processed chunk are valid candidates for
8272 // compression. Write a truncated byte elements to the memory.
8273 evpmovwb(Address(dst, len, Address::times_1), tmp1Reg, Assembler::AVX_512bit);
8274 addptr(len, 32);
8275 jccb(Assembler::notZero, copy_32_loop);
8276
8277 bind(copy_loop_tail);
8278 // bail out when there is nothing to be done
8279 testl(tmp5, 0xFFFFFFFF);
8280 jcc(Assembler::zero, done);
8281
8282 movl(len, tmp5);
8283
8284 // ~(~0 << len), where len is the # of remaining elements to process
8285 movl(tmp5, 0xFFFFFFFF);
8286 shlxl(tmp5, tmp5, len);
8287 notl(tmp5);
8288
8289 kmovdl(mask2, tmp5);
8290
8291 evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
8292 evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
8293 ktestd(mask1, mask2);
8294 jcc(Assembler::carryClear, copy_tail);
8295
8296 evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
8297 jmp(done);
8298
8299 bind(reset_for_copy_tail);
8300 lea(src, Address(src, tmp5, Address::times_2));
8301 lea(dst, Address(dst, tmp5, Address::times_1));
8302 subptr(len, tmp5);
8303 jmp(copy_chars_loop);
8304
8305 bind(below_threshold);
8306 }
8307
8308 if (UseSSE42Intrinsics) {
8309 Label copy_32_loop, copy_16, copy_tail_sse, reset_for_copy_tail;
8310
8311 // vectored compression
8312 testl(len, 0xfffffff8);
8313 jcc(Assembler::zero, copy_tail);
8314
8315 movl(tmp5, 0xff00ff00); // create mask to test for Unicode chars in vectors
8316 movdl(tmp1Reg, tmp5);
8317 pshufd(tmp1Reg, tmp1Reg, 0); // store Unicode mask in tmp1Reg
8318
8319 andl(len, 0xfffffff0);
8320 jccb(Assembler::zero, copy_16);
8321
8322 // compress 16 chars per iter
8323 pxor(tmp4Reg, tmp4Reg);
8324
8325 lea(src, Address(src, len, Address::times_2));
8326 lea(dst, Address(dst, len, Address::times_1));
8327 negptr(len);
8328
8329 bind(copy_32_loop);
8330 movdqu(tmp2Reg, Address(src, len, Address::times_2)); // load 1st 8 characters
8331 por(tmp4Reg, tmp2Reg);
8332 movdqu(tmp3Reg, Address(src, len, Address::times_2, 16)); // load next 8 characters
8333 por(tmp4Reg, tmp3Reg);
8334 ptest(tmp4Reg, tmp1Reg); // check for Unicode chars in next vector
8335 jccb(Assembler::notZero, reset_for_copy_tail);
8336 packuswb(tmp2Reg, tmp3Reg); // only ASCII chars; compress each to 1 byte
8337 movdqu(Address(dst, len, Address::times_1), tmp2Reg);
8338 addptr(len, 16);
8339 jccb(Assembler::notZero, copy_32_loop);
8340
8341 // compress next vector of 8 chars (if any)
8342 bind(copy_16);
8343 // len = 0
8344 testl(result, 0x00000008); // check if there's a block of 8 chars to compress
8345 jccb(Assembler::zero, copy_tail_sse);
8346
8347 pxor(tmp3Reg, tmp3Reg);
8348
8349 movdqu(tmp2Reg, Address(src, 0));
8350 ptest(tmp2Reg, tmp1Reg); // check for Unicode chars in vector
8351 jccb(Assembler::notZero, reset_for_copy_tail);
8352 packuswb(tmp2Reg, tmp3Reg); // only LATIN1 chars; compress each to 1 byte
8353 movq(Address(dst, 0), tmp2Reg);
8354 addptr(src, 16);
8355 addptr(dst, 8);
8356 jmpb(copy_tail_sse);
8357
8358 bind(reset_for_copy_tail);
8359 movl(tmp5, result);
8360 andl(tmp5, 0x0000000f);
8361 lea(src, Address(src, tmp5, Address::times_2));
8362 lea(dst, Address(dst, tmp5, Address::times_1));
8363 subptr(len, tmp5);
8364 jmpb(copy_chars_loop);
8365
8366 bind(copy_tail_sse);
8367 movl(len, result);
8368 andl(len, 0x00000007); // tail count (in chars)
8369 }
8370 // compress 1 char per iter
8371 bind(copy_tail);
8372 testl(len, len);
8373 jccb(Assembler::zero, done);
8374 lea(src, Address(src, len, Address::times_2));
8375 lea(dst, Address(dst, len, Address::times_1));
8376 negptr(len);
8377
8378 bind(copy_chars_loop);
8379 load_unsigned_short(tmp5, Address(src, len, Address::times_2));
8380 testl(tmp5, 0xff00); // check if Unicode char
8381 jccb(Assembler::notZero, reset_sp);
8382 movb(Address(dst, len, Address::times_1), tmp5); // ASCII char; compress to 1 byte
8383 increment(len);
8384 jccb(Assembler::notZero, copy_chars_loop);
8385
8386 // add len then return (len will be zero if compress succeeded, otherwise negative)
8387 bind(reset_sp);
8388 addl(result, len);
8389
8390 bind(done);
8391 }
8392
8393 // Inflate byte[] array to char[].
8394 // ..\jdk\src\java.base\share\classes\java\lang\StringLatin1.java
8395 // @IntrinsicCandidate
8396 // private static void inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len) {
8397 // for (int i = 0; i < len; i++) {
8398 // dst[dstOff++] = (char)(src[srcOff++] & 0xff);
8399 // }
8400 // }
8401 void MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
8402 XMMRegister tmp1, Register tmp2, KRegister mask) {
8403 Label copy_chars_loop, done, below_threshold, avx3_threshold;
8404 // rsi: src
8405 // rdi: dst
8406 // rdx: len
8407 // rcx: tmp2
8408
8409 // rsi holds start addr of source byte[] to be inflated
8410 // rdi holds start addr of destination char[]
8411 // rdx holds length
8412 assert_different_registers(src, dst, len, tmp2);
8413 movl(tmp2, len);
8414 if ((UseAVX > 2) && // AVX512
8415 VM_Version::supports_avx512vlbw() &&
8416 VM_Version::supports_bmi2()) {
8417
8418 Label copy_32_loop, copy_tail;
8419 Register tmp3_aliased = len;
8420
8421 // if length of the string is less than 16, handle it in an old fashioned way
8422 testl(len, -16);
8423 jcc(Assembler::zero, below_threshold);
8424
8425 testl(len, -1 * AVX3Threshold);
8426 jcc(Assembler::zero, avx3_threshold);
8427
8428 // In order to use only one arithmetic operation for the main loop we use
8429 // this pre-calculation
8430 andl(tmp2, (32 - 1)); // tail count (in chars), 32 element wide loop
8431 andl(len, -32); // vector count
8432 jccb(Assembler::zero, copy_tail);
8433
8434 lea(src, Address(src, len, Address::times_1));
8435 lea(dst, Address(dst, len, Address::times_2));
8436 negptr(len);
8437
8438
8439 // inflate 32 chars per iter
8440 bind(copy_32_loop);
8441 vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_512bit);
8442 evmovdquw(Address(dst, len, Address::times_2), tmp1, Assembler::AVX_512bit);
8443 addptr(len, 32);
8444 jcc(Assembler::notZero, copy_32_loop);
8445
8446 bind(copy_tail);
8447 // bail out when there is nothing to be done
8448 testl(tmp2, -1); // we don't destroy the contents of tmp2 here
8449 jcc(Assembler::zero, done);
8450
8451 // ~(~0 << length), where length is the # of remaining elements to process
8452 movl(tmp3_aliased, -1);
8453 shlxl(tmp3_aliased, tmp3_aliased, tmp2);
8454 notl(tmp3_aliased);
8455 kmovdl(mask, tmp3_aliased);
8456 evpmovzxbw(tmp1, mask, Address(src, 0), Assembler::AVX_512bit);
8457 evmovdquw(Address(dst, 0), mask, tmp1, /*merge*/ true, Assembler::AVX_512bit);
8458
8459 jmp(done);
8460 bind(avx3_threshold);
8461 }
8462 if (UseSSE42Intrinsics) {
8463 Label copy_16_loop, copy_8_loop, copy_bytes, copy_new_tail, copy_tail;
8464
8465 if (UseAVX > 1) {
8466 andl(tmp2, (16 - 1));
8467 andl(len, -16);
8468 jccb(Assembler::zero, copy_new_tail);
8469 } else {
8470 andl(tmp2, 0x00000007); // tail count (in chars)
8471 andl(len, 0xfffffff8); // vector count (in chars)
8472 jccb(Assembler::zero, copy_tail);
8473 }
8474
8475 // vectored inflation
8476 lea(src, Address(src, len, Address::times_1));
8477 lea(dst, Address(dst, len, Address::times_2));
8478 negptr(len);
8479
8480 if (UseAVX > 1) {
8481 bind(copy_16_loop);
8482 vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_256bit);
8483 vmovdqu(Address(dst, len, Address::times_2), tmp1);
8484 addptr(len, 16);
8485 jcc(Assembler::notZero, copy_16_loop);
8486
8487 bind(below_threshold);
8488 bind(copy_new_tail);
8489 movl(len, tmp2);
8490 andl(tmp2, 0x00000007);
8491 andl(len, 0xFFFFFFF8);
8492 jccb(Assembler::zero, copy_tail);
8493
8494 pmovzxbw(tmp1, Address(src, 0));
8495 movdqu(Address(dst, 0), tmp1);
8496 addptr(src, 8);
8497 addptr(dst, 2 * 8);
8498
8499 jmp(copy_tail, true);
8500 }
8501
8502 // inflate 8 chars per iter
8503 bind(copy_8_loop);
8504 pmovzxbw(tmp1, Address(src, len, Address::times_1)); // unpack to 8 words
8505 movdqu(Address(dst, len, Address::times_2), tmp1);
8506 addptr(len, 8);
8507 jcc(Assembler::notZero, copy_8_loop);
8508
8509 bind(copy_tail);
8510 movl(len, tmp2);
8511
8512 cmpl(len, 4);
8513 jccb(Assembler::less, copy_bytes);
8514
8515 movdl(tmp1, Address(src, 0)); // load 4 byte chars
8516 pmovzxbw(tmp1, tmp1);
8517 movq(Address(dst, 0), tmp1);
8518 subptr(len, 4);
8519 addptr(src, 4);
8520 addptr(dst, 8);
8521
8522 bind(copy_bytes);
8523 } else {
8524 bind(below_threshold);
8525 }
8526
8527 testl(len, len);
8528 jccb(Assembler::zero, done);
8529 lea(src, Address(src, len, Address::times_1));
8530 lea(dst, Address(dst, len, Address::times_2));
8531 negptr(len);
8532
8533 // inflate 1 char per iter
8534 bind(copy_chars_loop);
8535 load_unsigned_byte(tmp2, Address(src, len, Address::times_1)); // load byte char
8536 movw(Address(dst, len, Address::times_2), tmp2); // inflate byte char to word
8537 increment(len);
8538 jcc(Assembler::notZero, copy_chars_loop);
8539
8540 bind(done);
8541 }
8542
8543 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, XMMRegister src, bool merge, int vector_len) {
8544 switch(type) {
8545 case T_BYTE:
8546 case T_BOOLEAN:
8547 evmovdqub(dst, kmask, src, merge, vector_len);
8548 break;
8549 case T_CHAR:
8550 case T_SHORT:
8551 evmovdquw(dst, kmask, src, merge, vector_len);
8552 break;
8553 case T_INT:
8554 case T_FLOAT:
8555 evmovdqul(dst, kmask, src, merge, vector_len);
8556 break;
8557 case T_LONG:
8558 case T_DOUBLE:
8559 evmovdquq(dst, kmask, src, merge, vector_len);
8560 break;
8561 default:
8562 fatal("Unexpected type argument %s", type2name(type));
8563 break;
8564 }
8565 }
8566
8567
8568 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address src, bool merge, int vector_len) {
8569 switch(type) {
8570 case T_BYTE:
8571 case T_BOOLEAN:
8572 evmovdqub(dst, kmask, src, merge, vector_len);
8573 break;
8574 case T_CHAR:
8575 case T_SHORT:
8576 evmovdquw(dst, kmask, src, merge, vector_len);
8577 break;
8578 case T_INT:
8579 case T_FLOAT:
8580 evmovdqul(dst, kmask, src, merge, vector_len);
8581 break;
8582 case T_LONG:
8583 case T_DOUBLE:
8584 evmovdquq(dst, kmask, src, merge, vector_len);
8585 break;
8586 default:
8587 fatal("Unexpected type argument %s", type2name(type));
8588 break;
8589 }
8590 }
8591
8592 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister src, bool merge, int vector_len) {
8593 switch(type) {
8594 case T_BYTE:
8595 case T_BOOLEAN:
8596 evmovdqub(dst, kmask, src, merge, vector_len);
8597 break;
8598 case T_CHAR:
8599 case T_SHORT:
8600 evmovdquw(dst, kmask, src, merge, vector_len);
8601 break;
8602 case T_INT:
8603 case T_FLOAT:
8604 evmovdqul(dst, kmask, src, merge, vector_len);
8605 break;
8606 case T_LONG:
8607 case T_DOUBLE:
8608 evmovdquq(dst, kmask, src, merge, vector_len);
8609 break;
8610 default:
8611 fatal("Unexpected type argument %s", type2name(type));
8612 break;
8613 }
8614 }
8615
8616 void MacroAssembler::knot(uint masklen, KRegister dst, KRegister src, KRegister ktmp, Register rtmp) {
8617 switch(masklen) {
8618 case 2:
8619 knotbl(dst, src);
8620 movl(rtmp, 3);
8621 kmovbl(ktmp, rtmp);
8622 kandbl(dst, ktmp, dst);
8623 break;
8624 case 4:
8625 knotbl(dst, src);
8626 movl(rtmp, 15);
8627 kmovbl(ktmp, rtmp);
8628 kandbl(dst, ktmp, dst);
8629 break;
8630 case 8:
8631 knotbl(dst, src);
8632 break;
8633 case 16:
8634 knotwl(dst, src);
8635 break;
8636 case 32:
8637 knotdl(dst, src);
8638 break;
8639 case 64:
8640 knotql(dst, src);
8641 break;
8642 default:
8643 fatal("Unexpected vector length %d", masklen);
8644 break;
8645 }
8646 }
8647
8648 void MacroAssembler::kand(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
8649 switch(type) {
8650 case T_BOOLEAN:
8651 case T_BYTE:
8652 kandbl(dst, src1, src2);
8653 break;
8654 case T_CHAR:
8655 case T_SHORT:
8656 kandwl(dst, src1, src2);
8657 break;
8658 case T_INT:
8659 case T_FLOAT:
8660 kanddl(dst, src1, src2);
8661 break;
8662 case T_LONG:
8663 case T_DOUBLE:
8664 kandql(dst, src1, src2);
8665 break;
8666 default:
8667 fatal("Unexpected type argument %s", type2name(type));
8668 break;
8669 }
8670 }
8671
8672 void MacroAssembler::kor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
8673 switch(type) {
8674 case T_BOOLEAN:
8675 case T_BYTE:
8676 korbl(dst, src1, src2);
8677 break;
8678 case T_CHAR:
8679 case T_SHORT:
8680 korwl(dst, src1, src2);
8681 break;
8682 case T_INT:
8683 case T_FLOAT:
8684 kordl(dst, src1, src2);
8685 break;
8686 case T_LONG:
8687 case T_DOUBLE:
8688 korql(dst, src1, src2);
8689 break;
8690 default:
8691 fatal("Unexpected type argument %s", type2name(type));
8692 break;
8693 }
8694 }
8695
8696 void MacroAssembler::kxor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
8697 switch(type) {
8698 case T_BOOLEAN:
8699 case T_BYTE:
8700 kxorbl(dst, src1, src2);
8701 break;
8702 case T_CHAR:
8703 case T_SHORT:
8704 kxorwl(dst, src1, src2);
8705 break;
8706 case T_INT:
8707 case T_FLOAT:
8708 kxordl(dst, src1, src2);
8709 break;
8710 case T_LONG:
8711 case T_DOUBLE:
8712 kxorql(dst, src1, src2);
8713 break;
8714 default:
8715 fatal("Unexpected type argument %s", type2name(type));
8716 break;
8717 }
8718 }
8719
8720 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
8721 switch(type) {
8722 case T_BOOLEAN:
8723 case T_BYTE:
8724 evpermb(dst, mask, nds, src, merge, vector_len); break;
8725 case T_CHAR:
8726 case T_SHORT:
8727 evpermw(dst, mask, nds, src, merge, vector_len); break;
8728 case T_INT:
8729 case T_FLOAT:
8730 evpermd(dst, mask, nds, src, merge, vector_len); break;
8731 case T_LONG:
8732 case T_DOUBLE:
8733 evpermq(dst, mask, nds, src, merge, vector_len); break;
8734 default:
8735 fatal("Unexpected type argument %s", type2name(type)); break;
8736 }
8737 }
8738
8739 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
8740 switch(type) {
8741 case T_BOOLEAN:
8742 case T_BYTE:
8743 evpermb(dst, mask, nds, src, merge, vector_len); break;
8744 case T_CHAR:
8745 case T_SHORT:
8746 evpermw(dst, mask, nds, src, merge, vector_len); break;
8747 case T_INT:
8748 case T_FLOAT:
8749 evpermd(dst, mask, nds, src, merge, vector_len); break;
8750 case T_LONG:
8751 case T_DOUBLE:
8752 evpermq(dst, mask, nds, src, merge, vector_len); break;
8753 default:
8754 fatal("Unexpected type argument %s", type2name(type)); break;
8755 }
8756 }
8757
8758 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
8759 switch(type) {
8760 case T_BYTE:
8761 evpminub(dst, mask, nds, src, merge, vector_len); break;
8762 case T_SHORT:
8763 evpminuw(dst, mask, nds, src, merge, vector_len); break;
8764 case T_INT:
8765 evpminud(dst, mask, nds, src, merge, vector_len); break;
8766 case T_LONG:
8767 evpminuq(dst, mask, nds, src, merge, vector_len); break;
8768 default:
8769 fatal("Unexpected type argument %s", type2name(type)); break;
8770 }
8771 }
8772
8773 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
8774 switch(type) {
8775 case T_BYTE:
8776 evpmaxub(dst, mask, nds, src, merge, vector_len); break;
8777 case T_SHORT:
8778 evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
8779 case T_INT:
8780 evpmaxud(dst, mask, nds, src, merge, vector_len); break;
8781 case T_LONG:
8782 evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
8783 default:
8784 fatal("Unexpected type argument %s", type2name(type)); break;
8785 }
8786 }
8787
8788 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
8789 switch(type) {
8790 case T_BYTE:
8791 evpminub(dst, mask, nds, src, merge, vector_len); break;
8792 case T_SHORT:
8793 evpminuw(dst, mask, nds, src, merge, vector_len); break;
8794 case T_INT:
8795 evpminud(dst, mask, nds, src, merge, vector_len); break;
8796 case T_LONG:
8797 evpminuq(dst, mask, nds, src, merge, vector_len); break;
8798 default:
8799 fatal("Unexpected type argument %s", type2name(type)); break;
8800 }
8801 }
8802
8803 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
8804 switch(type) {
8805 case T_BYTE:
8806 evpmaxub(dst, mask, nds, src, merge, vector_len); break;
8807 case T_SHORT:
8808 evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
8809 case T_INT:
8810 evpmaxud(dst, mask, nds, src, merge, vector_len); break;
8811 case T_LONG:
8812 evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
8813 default:
8814 fatal("Unexpected type argument %s", type2name(type)); break;
8815 }
8816 }
8817
8818 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
8819 switch(type) {
8820 case T_BYTE:
8821 evpminsb(dst, mask, nds, src, merge, vector_len); break;
8822 case T_SHORT:
8823 evpminsw(dst, mask, nds, src, merge, vector_len); break;
8824 case T_INT:
8825 evpminsd(dst, mask, nds, src, merge, vector_len); break;
8826 case T_LONG:
8827 evpminsq(dst, mask, nds, src, merge, vector_len); break;
8828 default:
8829 fatal("Unexpected type argument %s", type2name(type)); break;
8830 }
8831 }
8832
8833 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
8834 switch(type) {
8835 case T_BYTE:
8836 evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
8837 case T_SHORT:
8838 evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
8839 case T_INT:
8840 evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
8841 case T_LONG:
8842 evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
8843 default:
8844 fatal("Unexpected type argument %s", type2name(type)); break;
8845 }
8846 }
8847
8848 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
8849 switch(type) {
8850 case T_BYTE:
8851 evpminsb(dst, mask, nds, src, merge, vector_len); break;
8852 case T_SHORT:
8853 evpminsw(dst, mask, nds, src, merge, vector_len); break;
8854 case T_INT:
8855 evpminsd(dst, mask, nds, src, merge, vector_len); break;
8856 case T_LONG:
8857 evpminsq(dst, mask, nds, src, merge, vector_len); break;
8858 default:
8859 fatal("Unexpected type argument %s", type2name(type)); break;
8860 }
8861 }
8862
8863 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
8864 switch(type) {
8865 case T_BYTE:
8866 evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
8867 case T_SHORT:
8868 evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
8869 case T_INT:
8870 evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
8871 case T_LONG:
8872 evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
8873 default:
8874 fatal("Unexpected type argument %s", type2name(type)); break;
8875 }
8876 }
8877
8878 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
8879 switch(type) {
8880 case T_INT:
8881 evpxord(dst, mask, nds, src, merge, vector_len); break;
8882 case T_LONG:
8883 evpxorq(dst, mask, nds, src, merge, vector_len); break;
8884 default:
8885 fatal("Unexpected type argument %s", type2name(type)); break;
8886 }
8887 }
8888
8889 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
8890 switch(type) {
8891 case T_INT:
8892 evpxord(dst, mask, nds, src, merge, vector_len); break;
8893 case T_LONG:
8894 evpxorq(dst, mask, nds, src, merge, vector_len); break;
8895 default:
8896 fatal("Unexpected type argument %s", type2name(type)); break;
8897 }
8898 }
8899
8900 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
8901 switch(type) {
8902 case T_INT:
8903 Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
8904 case T_LONG:
8905 evporq(dst, mask, nds, src, merge, vector_len); break;
8906 default:
8907 fatal("Unexpected type argument %s", type2name(type)); break;
8908 }
8909 }
8910
8911 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
8912 switch(type) {
8913 case T_INT:
8914 Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
8915 case T_LONG:
8916 evporq(dst, mask, nds, src, merge, vector_len); break;
8917 default:
8918 fatal("Unexpected type argument %s", type2name(type)); break;
8919 }
8920 }
8921
8922 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
8923 switch(type) {
8924 case T_INT:
8925 evpandd(dst, mask, nds, src, merge, vector_len); break;
8926 case T_LONG:
8927 evpandq(dst, mask, nds, src, merge, vector_len); break;
8928 default:
8929 fatal("Unexpected type argument %s", type2name(type)); break;
8930 }
8931 }
8932
8933 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
8934 switch(type) {
8935 case T_INT:
8936 evpandd(dst, mask, nds, src, merge, vector_len); break;
8937 case T_LONG:
8938 evpandq(dst, mask, nds, src, merge, vector_len); break;
8939 default:
8940 fatal("Unexpected type argument %s", type2name(type)); break;
8941 }
8942 }
8943
8944 void MacroAssembler::kortest(uint masklen, KRegister src1, KRegister src2) {
8945 switch(masklen) {
8946 case 8:
8947 kortestbl(src1, src2);
8948 break;
8949 case 16:
8950 kortestwl(src1, src2);
8951 break;
8952 case 32:
8953 kortestdl(src1, src2);
8954 break;
8955 case 64:
8956 kortestql(src1, src2);
8957 break;
8958 default:
8959 fatal("Unexpected mask length %d", masklen);
8960 break;
8961 }
8962 }
8963
8964
8965 void MacroAssembler::ktest(uint masklen, KRegister src1, KRegister src2) {
8966 switch(masklen) {
8967 case 8:
8968 ktestbl(src1, src2);
8969 break;
8970 case 16:
8971 ktestwl(src1, src2);
8972 break;
8973 case 32:
8974 ktestdl(src1, src2);
8975 break;
8976 case 64:
8977 ktestql(src1, src2);
8978 break;
8979 default:
8980 fatal("Unexpected mask length %d", masklen);
8981 break;
8982 }
8983 }
8984
8985 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
8986 switch(type) {
8987 case T_INT:
8988 evprold(dst, mask, src, shift, merge, vlen_enc); break;
8989 case T_LONG:
8990 evprolq(dst, mask, src, shift, merge, vlen_enc); break;
8991 default:
8992 fatal("Unexpected type argument %s", type2name(type)); break;
8993 break;
8994 }
8995 }
8996
8997 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
8998 switch(type) {
8999 case T_INT:
9000 evprord(dst, mask, src, shift, merge, vlen_enc); break;
9001 case T_LONG:
9002 evprorq(dst, mask, src, shift, merge, vlen_enc); break;
9003 default:
9004 fatal("Unexpected type argument %s", type2name(type)); break;
9005 }
9006 }
9007
9008 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
9009 switch(type) {
9010 case T_INT:
9011 evprolvd(dst, mask, src1, src2, merge, vlen_enc); break;
9012 case T_LONG:
9013 evprolvq(dst, mask, src1, src2, merge, vlen_enc); break;
9014 default:
9015 fatal("Unexpected type argument %s", type2name(type)); break;
9016 }
9017 }
9018
9019 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
9020 switch(type) {
9021 case T_INT:
9022 evprorvd(dst, mask, src1, src2, merge, vlen_enc); break;
9023 case T_LONG:
9024 evprorvq(dst, mask, src1, src2, merge, vlen_enc); break;
9025 default:
9026 fatal("Unexpected type argument %s", type2name(type)); break;
9027 }
9028 }
9029
9030 void MacroAssembler::evpandq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
9031 assert(rscratch != noreg || always_reachable(src), "missing");
9032
9033 if (reachable(src)) {
9034 evpandq(dst, nds, as_Address(src), vector_len);
9035 } else {
9036 lea(rscratch, src);
9037 evpandq(dst, nds, Address(rscratch, 0), vector_len);
9038 }
9039 }
9040
9041 void MacroAssembler::evpaddq(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
9042 assert(rscratch != noreg || always_reachable(src), "missing");
9043
9044 if (reachable(src)) {
9045 Assembler::evpaddq(dst, mask, nds, as_Address(src), merge, vector_len);
9046 } else {
9047 lea(rscratch, src);
9048 Assembler::evpaddq(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
9049 }
9050 }
9051
9052 void MacroAssembler::evporq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
9053 assert(rscratch != noreg || always_reachable(src), "missing");
9054
9055 if (reachable(src)) {
9056 evporq(dst, nds, as_Address(src), vector_len);
9057 } else {
9058 lea(rscratch, src);
9059 evporq(dst, nds, Address(rscratch, 0), vector_len);
9060 }
9061 }
9062
9063 void MacroAssembler::vpshufb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
9064 assert(rscratch != noreg || always_reachable(src), "missing");
9065
9066 if (reachable(src)) {
9067 vpshufb(dst, nds, as_Address(src), vector_len);
9068 } else {
9069 lea(rscratch, src);
9070 vpshufb(dst, nds, Address(rscratch, 0), vector_len);
9071 }
9072 }
9073
9074 void MacroAssembler::vpor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
9075 assert(rscratch != noreg || always_reachable(src), "missing");
9076
9077 if (reachable(src)) {
9078 Assembler::vpor(dst, nds, as_Address(src), vector_len);
9079 } else {
9080 lea(rscratch, src);
9081 Assembler::vpor(dst, nds, Address(rscratch, 0), vector_len);
9082 }
9083 }
9084
9085 void MacroAssembler::vpternlogq(XMMRegister dst, int imm8, XMMRegister src2, AddressLiteral src3, int vector_len, Register rscratch) {
9086 assert(rscratch != noreg || always_reachable(src3), "missing");
9087
9088 if (reachable(src3)) {
9089 vpternlogq(dst, imm8, src2, as_Address(src3), vector_len);
9090 } else {
9091 lea(rscratch, src3);
9092 vpternlogq(dst, imm8, src2, Address(rscratch, 0), vector_len);
9093 }
9094 }
9095
9096 #if COMPILER2_OR_JVMCI
9097
9098 void MacroAssembler::fill_masked(BasicType bt, Address dst, XMMRegister xmm, KRegister mask,
9099 Register length, Register temp, int vec_enc) {
9100 // Computing mask for predicated vector store.
9101 movptr(temp, -1);
9102 bzhiq(temp, temp, length);
9103 kmov(mask, temp);
9104 evmovdqu(bt, mask, dst, xmm, true, vec_enc);
9105 }
9106
9107 // Set memory operation for length "less than" 64 bytes.
9108 void MacroAssembler::fill64_masked(uint shift, Register dst, int disp,
9109 XMMRegister xmm, KRegister mask, Register length,
9110 Register temp, bool use64byteVector) {
9111 assert(MaxVectorSize >= 32, "vector length should be >= 32");
9112 const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
9113 if (!use64byteVector) {
9114 fill32(dst, disp, xmm);
9115 subptr(length, 32 >> shift);
9116 fill32_masked(shift, dst, disp + 32, xmm, mask, length, temp);
9117 } else {
9118 assert(MaxVectorSize == 64, "vector length != 64");
9119 fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_512bit);
9120 }
9121 }
9122
9123
9124 void MacroAssembler::fill32_masked(uint shift, Register dst, int disp,
9125 XMMRegister xmm, KRegister mask, Register length,
9126 Register temp) {
9127 assert(MaxVectorSize >= 32, "vector length should be >= 32");
9128 const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
9129 fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_256bit);
9130 }
9131
9132
9133 void MacroAssembler::fill32(Address dst, XMMRegister xmm) {
9134 assert(MaxVectorSize >= 32, "vector length should be >= 32");
9135 vmovdqu(dst, xmm);
9136 }
9137
9138 void MacroAssembler::fill32(Register dst, int disp, XMMRegister xmm) {
9139 fill32(Address(dst, disp), xmm);
9140 }
9141
9142 void MacroAssembler::fill64(Address dst, XMMRegister xmm, bool use64byteVector) {
9143 assert(MaxVectorSize >= 32, "vector length should be >= 32");
9144 if (!use64byteVector) {
9145 fill32(dst, xmm);
9146 fill32(dst.plus_disp(32), xmm);
9147 } else {
9148 evmovdquq(dst, xmm, Assembler::AVX_512bit);
9149 }
9150 }
9151
9152 void MacroAssembler::fill64(Register dst, int disp, XMMRegister xmm, bool use64byteVector) {
9153 fill64(Address(dst, disp), xmm, use64byteVector);
9154 }
9155
9156 void MacroAssembler::generate_fill_avx3(BasicType type, Register to, Register value,
9157 Register count, Register rtmp, XMMRegister xtmp) {
9158 Label L_exit;
9159 Label L_fill_start;
9160 Label L_fill_64_bytes;
9161 Label L_fill_96_bytes;
9162 Label L_fill_128_bytes;
9163 Label L_fill_128_bytes_loop;
9164 Label L_fill_128_loop_header;
9165 Label L_fill_128_bytes_loop_header;
9166 Label L_fill_128_bytes_loop_pre_header;
9167 Label L_fill_zmm_sequence;
9168
9169 int shift = -1;
9170 int avx3threshold = VM_Version::avx3_threshold();
9171 switch(type) {
9172 case T_BYTE: shift = 0;
9173 break;
9174 case T_SHORT: shift = 1;
9175 break;
9176 case T_INT: shift = 2;
9177 break;
9178 /* Uncomment when LONG fill stubs are supported.
9179 case T_LONG: shift = 3;
9180 break;
9181 */
9182 default:
9183 fatal("Unhandled type: %s\n", type2name(type));
9184 }
9185
9186 if ((avx3threshold != 0) || (MaxVectorSize == 32)) {
9187
9188 if (MaxVectorSize == 64) {
9189 cmpq(count, avx3threshold >> shift);
9190 jcc(Assembler::greater, L_fill_zmm_sequence);
9191 }
9192
9193 evpbroadcast(type, xtmp, value, Assembler::AVX_256bit);
9194
9195 bind(L_fill_start);
9196
9197 cmpq(count, 32 >> shift);
9198 jccb(Assembler::greater, L_fill_64_bytes);
9199 fill32_masked(shift, to, 0, xtmp, k2, count, rtmp);
9200 jmp(L_exit);
9201
9202 bind(L_fill_64_bytes);
9203 cmpq(count, 64 >> shift);
9204 jccb(Assembler::greater, L_fill_96_bytes);
9205 fill64_masked(shift, to, 0, xtmp, k2, count, rtmp);
9206 jmp(L_exit);
9207
9208 bind(L_fill_96_bytes);
9209 cmpq(count, 96 >> shift);
9210 jccb(Assembler::greater, L_fill_128_bytes);
9211 fill64(to, 0, xtmp);
9212 subq(count, 64 >> shift);
9213 fill32_masked(shift, to, 64, xtmp, k2, count, rtmp);
9214 jmp(L_exit);
9215
9216 bind(L_fill_128_bytes);
9217 cmpq(count, 128 >> shift);
9218 jccb(Assembler::greater, L_fill_128_bytes_loop_pre_header);
9219 fill64(to, 0, xtmp);
9220 fill32(to, 64, xtmp);
9221 subq(count, 96 >> shift);
9222 fill32_masked(shift, to, 96, xtmp, k2, count, rtmp);
9223 jmp(L_exit);
9224
9225 bind(L_fill_128_bytes_loop_pre_header);
9226 {
9227 mov(rtmp, to);
9228 andq(rtmp, 31);
9229 jccb(Assembler::zero, L_fill_128_bytes_loop_header);
9230 negq(rtmp);
9231 addq(rtmp, 32);
9232 mov64(r8, -1L);
9233 bzhiq(r8, r8, rtmp);
9234 kmovql(k2, r8);
9235 evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_256bit);
9236 addq(to, rtmp);
9237 shrq(rtmp, shift);
9238 subq(count, rtmp);
9239 }
9240
9241 cmpq(count, 128 >> shift);
9242 jcc(Assembler::less, L_fill_start);
9243
9244 bind(L_fill_128_bytes_loop_header);
9245 subq(count, 128 >> shift);
9246
9247 align32();
9248 bind(L_fill_128_bytes_loop);
9249 fill64(to, 0, xtmp);
9250 fill64(to, 64, xtmp);
9251 addq(to, 128);
9252 subq(count, 128 >> shift);
9253 jccb(Assembler::greaterEqual, L_fill_128_bytes_loop);
9254
9255 addq(count, 128 >> shift);
9256 jcc(Assembler::zero, L_exit);
9257 jmp(L_fill_start);
9258 }
9259
9260 if (MaxVectorSize == 64) {
9261 // Sequence using 64 byte ZMM register.
9262 Label L_fill_128_bytes_zmm;
9263 Label L_fill_192_bytes_zmm;
9264 Label L_fill_192_bytes_loop_zmm;
9265 Label L_fill_192_bytes_loop_header_zmm;
9266 Label L_fill_192_bytes_loop_pre_header_zmm;
9267 Label L_fill_start_zmm_sequence;
9268
9269 bind(L_fill_zmm_sequence);
9270 evpbroadcast(type, xtmp, value, Assembler::AVX_512bit);
9271
9272 bind(L_fill_start_zmm_sequence);
9273 cmpq(count, 64 >> shift);
9274 jccb(Assembler::greater, L_fill_128_bytes_zmm);
9275 fill64_masked(shift, to, 0, xtmp, k2, count, rtmp, true);
9276 jmp(L_exit);
9277
9278 bind(L_fill_128_bytes_zmm);
9279 cmpq(count, 128 >> shift);
9280 jccb(Assembler::greater, L_fill_192_bytes_zmm);
9281 fill64(to, 0, xtmp, true);
9282 subq(count, 64 >> shift);
9283 fill64_masked(shift, to, 64, xtmp, k2, count, rtmp, true);
9284 jmp(L_exit);
9285
9286 bind(L_fill_192_bytes_zmm);
9287 cmpq(count, 192 >> shift);
9288 jccb(Assembler::greater, L_fill_192_bytes_loop_pre_header_zmm);
9289 fill64(to, 0, xtmp, true);
9290 fill64(to, 64, xtmp, true);
9291 subq(count, 128 >> shift);
9292 fill64_masked(shift, to, 128, xtmp, k2, count, rtmp, true);
9293 jmp(L_exit);
9294
9295 bind(L_fill_192_bytes_loop_pre_header_zmm);
9296 {
9297 movq(rtmp, to);
9298 andq(rtmp, 63);
9299 jccb(Assembler::zero, L_fill_192_bytes_loop_header_zmm);
9300 negq(rtmp);
9301 addq(rtmp, 64);
9302 mov64(r8, -1L);
9303 bzhiq(r8, r8, rtmp);
9304 kmovql(k2, r8);
9305 evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_512bit);
9306 addq(to, rtmp);
9307 shrq(rtmp, shift);
9308 subq(count, rtmp);
9309 }
9310
9311 cmpq(count, 192 >> shift);
9312 jcc(Assembler::less, L_fill_start_zmm_sequence);
9313
9314 bind(L_fill_192_bytes_loop_header_zmm);
9315 subq(count, 192 >> shift);
9316
9317 align32();
9318 bind(L_fill_192_bytes_loop_zmm);
9319 fill64(to, 0, xtmp, true);
9320 fill64(to, 64, xtmp, true);
9321 fill64(to, 128, xtmp, true);
9322 addq(to, 192);
9323 subq(count, 192 >> shift);
9324 jccb(Assembler::greaterEqual, L_fill_192_bytes_loop_zmm);
9325
9326 addq(count, 192 >> shift);
9327 jcc(Assembler::zero, L_exit);
9328 jmp(L_fill_start_zmm_sequence);
9329 }
9330 bind(L_exit);
9331 }
9332 #endif //COMPILER2_OR_JVMCI
9333
9334
9335 void MacroAssembler::convert_f2i(Register dst, XMMRegister src) {
9336 Label done;
9337 cvttss2sil(dst, src);
9338 // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
9339 cmpl(dst, 0x80000000); // float_sign_flip
9340 jccb(Assembler::notEqual, done);
9341 subptr(rsp, 8);
9342 movflt(Address(rsp, 0), src);
9343 call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2i_fixup())));
9344 pop(dst);
9345 bind(done);
9346 }
9347
9348 void MacroAssembler::convert_d2i(Register dst, XMMRegister src) {
9349 Label done;
9350 cvttsd2sil(dst, src);
9351 // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
9352 cmpl(dst, 0x80000000); // float_sign_flip
9353 jccb(Assembler::notEqual, done);
9354 subptr(rsp, 8);
9355 movdbl(Address(rsp, 0), src);
9356 call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2i_fixup())));
9357 pop(dst);
9358 bind(done);
9359 }
9360
9361 void MacroAssembler::convert_f2l(Register dst, XMMRegister src) {
9362 Label done;
9363 cvttss2siq(dst, src);
9364 cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
9365 jccb(Assembler::notEqual, done);
9366 subptr(rsp, 8);
9367 movflt(Address(rsp, 0), src);
9368 call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2l_fixup())));
9369 pop(dst);
9370 bind(done);
9371 }
9372
9373 void MacroAssembler::round_float(Register dst, XMMRegister src, Register rtmp, Register rcx) {
9374 // Following code is line by line assembly translation rounding algorithm.
9375 // Please refer to java.lang.Math.round(float) algorithm for details.
9376 const int32_t FloatConsts_EXP_BIT_MASK = 0x7F800000;
9377 const int32_t FloatConsts_SIGNIFICAND_WIDTH = 24;
9378 const int32_t FloatConsts_EXP_BIAS = 127;
9379 const int32_t FloatConsts_SIGNIF_BIT_MASK = 0x007FFFFF;
9380 const int32_t MINUS_32 = 0xFFFFFFE0;
9381 Label L_special_case, L_block1, L_exit;
9382 movl(rtmp, FloatConsts_EXP_BIT_MASK);
9383 movdl(dst, src);
9384 andl(dst, rtmp);
9385 sarl(dst, FloatConsts_SIGNIFICAND_WIDTH - 1);
9386 movl(rtmp, FloatConsts_SIGNIFICAND_WIDTH - 2 + FloatConsts_EXP_BIAS);
9387 subl(rtmp, dst);
9388 movl(rcx, rtmp);
9389 movl(dst, MINUS_32);
9390 testl(rtmp, dst);
9391 jccb(Assembler::notEqual, L_special_case);
9392 movdl(dst, src);
9393 andl(dst, FloatConsts_SIGNIF_BIT_MASK);
9394 orl(dst, FloatConsts_SIGNIF_BIT_MASK + 1);
9395 movdl(rtmp, src);
9396 testl(rtmp, rtmp);
9397 jccb(Assembler::greaterEqual, L_block1);
9398 negl(dst);
9399 bind(L_block1);
9400 sarl(dst);
9401 addl(dst, 0x1);
9402 sarl(dst, 0x1);
9403 jmp(L_exit);
9404 bind(L_special_case);
9405 convert_f2i(dst, src);
9406 bind(L_exit);
9407 }
9408
9409 void MacroAssembler::round_double(Register dst, XMMRegister src, Register rtmp, Register rcx) {
9410 // Following code is line by line assembly translation rounding algorithm.
9411 // Please refer to java.lang.Math.round(double) algorithm for details.
9412 const int64_t DoubleConsts_EXP_BIT_MASK = 0x7FF0000000000000L;
9413 const int64_t DoubleConsts_SIGNIFICAND_WIDTH = 53;
9414 const int64_t DoubleConsts_EXP_BIAS = 1023;
9415 const int64_t DoubleConsts_SIGNIF_BIT_MASK = 0x000FFFFFFFFFFFFFL;
9416 const int64_t MINUS_64 = 0xFFFFFFFFFFFFFFC0L;
9417 Label L_special_case, L_block1, L_exit;
9418 mov64(rtmp, DoubleConsts_EXP_BIT_MASK);
9419 movq(dst, src);
9420 andq(dst, rtmp);
9421 sarq(dst, DoubleConsts_SIGNIFICAND_WIDTH - 1);
9422 mov64(rtmp, DoubleConsts_SIGNIFICAND_WIDTH - 2 + DoubleConsts_EXP_BIAS);
9423 subq(rtmp, dst);
9424 movq(rcx, rtmp);
9425 mov64(dst, MINUS_64);
9426 testq(rtmp, dst);
9427 jccb(Assembler::notEqual, L_special_case);
9428 movq(dst, src);
9429 mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK);
9430 andq(dst, rtmp);
9431 mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK + 1);
9432 orq(dst, rtmp);
9433 movq(rtmp, src);
9434 testq(rtmp, rtmp);
9435 jccb(Assembler::greaterEqual, L_block1);
9436 negq(dst);
9437 bind(L_block1);
9438 sarq(dst);
9439 addq(dst, 0x1);
9440 sarq(dst, 0x1);
9441 jmp(L_exit);
9442 bind(L_special_case);
9443 convert_d2l(dst, src);
9444 bind(L_exit);
9445 }
9446
9447 void MacroAssembler::convert_d2l(Register dst, XMMRegister src) {
9448 Label done;
9449 cvttsd2siq(dst, src);
9450 cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
9451 jccb(Assembler::notEqual, done);
9452 subptr(rsp, 8);
9453 movdbl(Address(rsp, 0), src);
9454 call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2l_fixup())));
9455 pop(dst);
9456 bind(done);
9457 }
9458
9459 void MacroAssembler::cache_wb(Address line)
9460 {
9461 // 64 bit cpus always support clflush
9462 assert(VM_Version::supports_clflush(), "clflush should be available");
9463 bool optimized = VM_Version::supports_clflushopt();
9464 bool no_evict = VM_Version::supports_clwb();
9465
9466 // prefer clwb (writeback without evict) otherwise
9467 // prefer clflushopt (potentially parallel writeback with evict)
9468 // otherwise fallback on clflush (serial writeback with evict)
9469
9470 if (optimized) {
9471 if (no_evict) {
9472 clwb(line);
9473 } else {
9474 clflushopt(line);
9475 }
9476 } else {
9477 // no need for fence when using CLFLUSH
9478 clflush(line);
9479 }
9480 }
9481
9482 void MacroAssembler::cache_wbsync(bool is_pre)
9483 {
9484 assert(VM_Version::supports_clflush(), "clflush should be available");
9485 bool optimized = VM_Version::supports_clflushopt();
9486 bool no_evict = VM_Version::supports_clwb();
9487
9488 // pick the correct implementation
9489
9490 if (!is_pre && (optimized || no_evict)) {
9491 // need an sfence for post flush when using clflushopt or clwb
9492 // otherwise no no need for any synchroniaztion
9493
9494 sfence();
9495 }
9496 }
9497
9498 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) {
9499 switch (cond) {
9500 // Note some conditions are synonyms for others
9501 case Assembler::zero: return Assembler::notZero;
9502 case Assembler::notZero: return Assembler::zero;
9503 case Assembler::less: return Assembler::greaterEqual;
9504 case Assembler::lessEqual: return Assembler::greater;
9505 case Assembler::greater: return Assembler::lessEqual;
9506 case Assembler::greaterEqual: return Assembler::less;
9507 case Assembler::below: return Assembler::aboveEqual;
9508 case Assembler::belowEqual: return Assembler::above;
9509 case Assembler::above: return Assembler::belowEqual;
9510 case Assembler::aboveEqual: return Assembler::below;
9511 case Assembler::overflow: return Assembler::noOverflow;
9512 case Assembler::noOverflow: return Assembler::overflow;
9513 case Assembler::negative: return Assembler::positive;
9514 case Assembler::positive: return Assembler::negative;
9515 case Assembler::parity: return Assembler::noParity;
9516 case Assembler::noParity: return Assembler::parity;
9517 }
9518 ShouldNotReachHere(); return Assembler::overflow;
9519 }
9520
9521 // This is simply a call to Thread::current()
9522 void MacroAssembler::get_thread_slow(Register thread) {
9523 if (thread != rax) {
9524 push(rax);
9525 }
9526 push(rdi);
9527 push(rsi);
9528 push(rdx);
9529 push(rcx);
9530 push(r8);
9531 push(r9);
9532 push(r10);
9533 push(r11);
9534
9535 MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, Thread::current), 0);
9536
9537 pop(r11);
9538 pop(r10);
9539 pop(r9);
9540 pop(r8);
9541 pop(rcx);
9542 pop(rdx);
9543 pop(rsi);
9544 pop(rdi);
9545 if (thread != rax) {
9546 mov(thread, rax);
9547 pop(rax);
9548 }
9549 }
9550
9551 void MacroAssembler::check_stack_alignment(Register sp, const char* msg, unsigned bias, Register tmp) {
9552 Label L_stack_ok;
9553 if (bias == 0) {
9554 testptr(sp, 2 * wordSize - 1);
9555 } else {
9556 // lea(tmp, Address(rsp, bias);
9557 mov(tmp, sp);
9558 addptr(tmp, bias);
9559 testptr(tmp, 2 * wordSize - 1);
9560 }
9561 jcc(Assembler::equal, L_stack_ok);
9562 block_comment(msg);
9563 stop(msg);
9564 bind(L_stack_ok);
9565 }
9566
9567 // Implements lightweight-locking.
9568 //
9569 // obj: the object to be locked
9570 // reg_rax: rax
9571 // thread: the thread which attempts to lock obj
9572 // tmp: a temporary register
9573 void MacroAssembler::lightweight_lock(Register basic_lock, Register obj, Register reg_rax, Register tmp, Label& slow) {
9574 Register thread = r15_thread;
9575
9576 assert(reg_rax == rax, "");
9577 assert_different_registers(basic_lock, obj, reg_rax, thread, tmp);
9578
9579 Label push;
9580 const Register top = tmp;
9581
9582 // Preload the markWord. It is important that this is the first
9583 // instruction emitted as it is part of C1's null check semantics.
9584 movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
9585
9586 if (UseObjectMonitorTable) {
9587 // Clear cache in case fast locking succeeds or we need to take the slow-path.
9588 movptr(Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))), 0);
9589 }
9590
9591 if (DiagnoseSyncOnValueBasedClasses != 0) {
9592 load_klass(tmp, obj, rscratch1);
9593 testb(Address(tmp, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
9594 jcc(Assembler::notZero, slow);
9595 }
9596
9597 // Load top.
9598 movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
9599
9600 // Check if the lock-stack is full.
9601 cmpl(top, LockStack::end_offset());
9602 jcc(Assembler::greaterEqual, slow);
9603
9604 // Check for recursion.
9605 cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
9606 jcc(Assembler::equal, push);
9607
9608 // Check header for monitor (0b10).
9609 testptr(reg_rax, markWord::monitor_value);
9610 jcc(Assembler::notZero, slow);
9611
9612 // Try to lock. Transition lock bits 0b01 => 0b00
9613 movptr(tmp, reg_rax);
9614 andptr(tmp, ~(int32_t)markWord::unlocked_value);
9615 orptr(reg_rax, markWord::unlocked_value);
9616 lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
9617 jcc(Assembler::notEqual, slow);
9618
9619 // Restore top, CAS clobbers register.
9620 movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
9621
9622 bind(push);
9623 // After successful lock, push object on lock-stack.
9624 movptr(Address(thread, top), obj);
9625 incrementl(top, oopSize);
9626 movl(Address(thread, JavaThread::lock_stack_top_offset()), top);
9627 }
9628
9629 // Implements lightweight-unlocking.
9630 //
9631 // obj: the object to be unlocked
9632 // reg_rax: rax
9633 // thread: the thread
9634 // tmp: a temporary register
9635 void MacroAssembler::lightweight_unlock(Register obj, Register reg_rax, Register tmp, Label& slow) {
9636 Register thread = r15_thread;
9637
9638 assert(reg_rax == rax, "");
9639 assert_different_registers(obj, reg_rax, thread, tmp);
9640
9641 Label unlocked, push_and_slow;
9642 const Register top = tmp;
9643
9644 // Check if obj is top of lock-stack.
9645 movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
9646 cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
9647 jcc(Assembler::notEqual, slow);
9648
9649 // Pop lock-stack.
9650 DEBUG_ONLY(movptr(Address(thread, top, Address::times_1, -oopSize), 0);)
9651 subl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
9652
9653 // Check if recursive.
9654 cmpptr(obj, Address(thread, top, Address::times_1, -2 * oopSize));
9655 jcc(Assembler::equal, unlocked);
9656
9657 // Not recursive. Check header for monitor (0b10).
9658 movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
9659 testptr(reg_rax, markWord::monitor_value);
9660 jcc(Assembler::notZero, push_and_slow);
9661
9662 #ifdef ASSERT
9663 // Check header not unlocked (0b01).
9664 Label not_unlocked;
9665 testptr(reg_rax, markWord::unlocked_value);
9666 jcc(Assembler::zero, not_unlocked);
9667 stop("lightweight_unlock already unlocked");
9668 bind(not_unlocked);
9669 #endif
9670
9671 // Try to unlock. Transition lock bits 0b00 => 0b01
9672 movptr(tmp, reg_rax);
9673 orptr(tmp, markWord::unlocked_value);
9674 lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
9675 jcc(Assembler::equal, unlocked);
9676
9677 bind(push_and_slow);
9678 // Restore lock-stack and handle the unlock in runtime.
9679 #ifdef ASSERT
9680 movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
9681 movptr(Address(thread, top), obj);
9682 #endif
9683 addl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
9684 jmp(slow);
9685
9686 bind(unlocked);
9687 }
9688
9689 // Saves legacy GPRs state on stack.
9690 void MacroAssembler::save_legacy_gprs() {
9691 subq(rsp, 16 * wordSize);
9692 movq(Address(rsp, 15 * wordSize), rax);
9693 movq(Address(rsp, 14 * wordSize), rcx);
9694 movq(Address(rsp, 13 * wordSize), rdx);
9695 movq(Address(rsp, 12 * wordSize), rbx);
9696 movq(Address(rsp, 10 * wordSize), rbp);
9697 movq(Address(rsp, 9 * wordSize), rsi);
9698 movq(Address(rsp, 8 * wordSize), rdi);
9699 movq(Address(rsp, 7 * wordSize), r8);
9700 movq(Address(rsp, 6 * wordSize), r9);
9701 movq(Address(rsp, 5 * wordSize), r10);
9702 movq(Address(rsp, 4 * wordSize), r11);
9703 movq(Address(rsp, 3 * wordSize), r12);
9704 movq(Address(rsp, 2 * wordSize), r13);
9705 movq(Address(rsp, wordSize), r14);
9706 movq(Address(rsp, 0), r15);
9707 }
9708
9709 // Resotres back legacy GPRs state from stack.
9710 void MacroAssembler::restore_legacy_gprs() {
9711 movq(r15, Address(rsp, 0));
9712 movq(r14, Address(rsp, wordSize));
9713 movq(r13, Address(rsp, 2 * wordSize));
9714 movq(r12, Address(rsp, 3 * wordSize));
9715 movq(r11, Address(rsp, 4 * wordSize));
9716 movq(r10, Address(rsp, 5 * wordSize));
9717 movq(r9, Address(rsp, 6 * wordSize));
9718 movq(r8, Address(rsp, 7 * wordSize));
9719 movq(rdi, Address(rsp, 8 * wordSize));
9720 movq(rsi, Address(rsp, 9 * wordSize));
9721 movq(rbp, Address(rsp, 10 * wordSize));
9722 movq(rbx, Address(rsp, 12 * wordSize));
9723 movq(rdx, Address(rsp, 13 * wordSize));
9724 movq(rcx, Address(rsp, 14 * wordSize));
9725 movq(rax, Address(rsp, 15 * wordSize));
9726 addq(rsp, 16 * wordSize);
9727 }
9728
9729 void MacroAssembler::setcc(Assembler::Condition comparison, Register dst) {
9730 if (VM_Version::supports_apx_f()) {
9731 esetzucc(comparison, dst);
9732 } else {
9733 setb(comparison, dst);
9734 movzbl(dst, dst);
9735 }
9736 }