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/aotCodeCache.hpp"
28 #include "code/compiledIC.hpp"
29 #include "compiler/compiler_globals.hpp"
30 #include "compiler/disassembler.hpp"
31 #include "crc32c.h"
32 #include "gc/shared/barrierSet.hpp"
33 #include "gc/shared/barrierSetAssembler.hpp"
34 #include "gc/shared/collectedHeap.inline.hpp"
35 #include "gc/shared/tlab_globals.hpp"
36 #include "interpreter/bytecodeHistogram.hpp"
37 #include "interpreter/interpreter.hpp"
38 #include "interpreter/interpreterRuntime.hpp"
39 #include "jvm.h"
40 #include "memory/resourceArea.hpp"
41 #include "memory/universe.hpp"
42 #include "oops/accessDecorators.hpp"
43 #include "oops/compressedKlass.inline.hpp"
44 #include "oops/compressedOops.inline.hpp"
45 #include "oops/klass.inline.hpp"
46 #include "prims/methodHandles.hpp"
47 #include "runtime/continuation.hpp"
48 #include "runtime/interfaceSupport.inline.hpp"
49 #include "runtime/javaThread.hpp"
50 #include "runtime/jniHandles.hpp"
51 #include "runtime/objectMonitor.hpp"
52 #include "runtime/os.hpp"
53 #include "runtime/safepoint.hpp"
54 #include "runtime/safepointMechanism.hpp"
55 #include "runtime/sharedRuntime.hpp"
56 #include "runtime/stubRoutines.hpp"
57 #include "utilities/checkedCast.hpp"
58 #include "utilities/macros.hpp"
59
60 #ifdef PRODUCT
61 #define BLOCK_COMMENT(str) /* nothing */
62 #define STOP(error) stop(error)
63 #else
64 #define BLOCK_COMMENT(str) block_comment(str)
65 #define STOP(error) block_comment(error); stop(error)
66 #endif
67
68 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
69
70 #ifdef ASSERT
71 bool AbstractAssembler::pd_check_instruction_mark() { return true; }
72 #endif
73
74 static const Assembler::Condition reverse[] = {
75 Assembler::noOverflow /* overflow = 0x0 */ ,
76 Assembler::overflow /* noOverflow = 0x1 */ ,
77 Assembler::aboveEqual /* carrySet = 0x2, below = 0x2 */ ,
78 Assembler::below /* aboveEqual = 0x3, carryClear = 0x3 */ ,
79 Assembler::notZero /* zero = 0x4, equal = 0x4 */ ,
80 Assembler::zero /* notZero = 0x5, notEqual = 0x5 */ ,
81 Assembler::above /* belowEqual = 0x6 */ ,
82 Assembler::belowEqual /* above = 0x7 */ ,
83 Assembler::positive /* negative = 0x8 */ ,
84 Assembler::negative /* positive = 0x9 */ ,
85 Assembler::noParity /* parity = 0xa */ ,
86 Assembler::parity /* noParity = 0xb */ ,
87 Assembler::greaterEqual /* less = 0xc */ ,
88 Assembler::less /* greaterEqual = 0xd */ ,
89 Assembler::greater /* lessEqual = 0xe */ ,
90 Assembler::lessEqual /* greater = 0xf, */
91
92 };
93
94
95 // Implementation of MacroAssembler
96
97 Address MacroAssembler::as_Address(AddressLiteral adr) {
98 // amd64 always does this as a pc-rel
99 // we can be absolute or disp based on the instruction type
100 // jmp/call are displacements others are absolute
101 assert(!adr.is_lval(), "must be rval");
102 assert(reachable(adr), "must be");
103 return Address(checked_cast<int32_t>(adr.target() - pc()), adr.target(), adr.reloc());
104
105 }
106
107 Address MacroAssembler::as_Address(ArrayAddress adr, Register rscratch) {
108 AddressLiteral base = adr.base();
109 lea(rscratch, base);
110 Address index = adr.index();
111 assert(index._disp == 0, "must not have disp"); // maybe it can?
112 Address array(rscratch, index._index, index._scale, index._disp);
113 return array;
114 }
115
116 void MacroAssembler::call_VM_leaf_base(address entry_point, int num_args) {
117 Label L, E;
118
119 #ifdef _WIN64
120 // Windows always allocates space for it's register args
121 assert(num_args <= 4, "only register arguments supported");
122 subq(rsp, frame::arg_reg_save_area_bytes);
123 #endif
124
125 // Align stack if necessary
126 testl(rsp, 15);
127 jcc(Assembler::zero, L);
128
129 subq(rsp, 8);
130 call(RuntimeAddress(entry_point));
131 addq(rsp, 8);
132 jmp(E);
133
134 bind(L);
135 call(RuntimeAddress(entry_point));
136
137 bind(E);
138
139 #ifdef _WIN64
140 // restore stack pointer
141 addq(rsp, frame::arg_reg_save_area_bytes);
142 #endif
143 }
144
145 void MacroAssembler::cmp64(Register src1, AddressLiteral src2, Register rscratch) {
146 assert(!src2.is_lval(), "should use cmpptr");
147 assert(rscratch != noreg || always_reachable(src2), "missing");
148
149 if (reachable(src2)) {
150 cmpq(src1, as_Address(src2));
151 } else {
152 lea(rscratch, src2);
153 Assembler::cmpq(src1, Address(rscratch, 0));
154 }
155 }
156
157 int MacroAssembler::corrected_idivq(Register reg) {
158 // Full implementation of Java ldiv and lrem; checks for special
159 // case as described in JVM spec., p.243 & p.271. The function
160 // returns the (pc) offset of the idivl instruction - may be needed
161 // for implicit exceptions.
162 //
163 // normal case special case
164 //
165 // input : rax: dividend min_long
166 // reg: divisor (may not be eax/edx) -1
167 //
168 // output: rax: quotient (= rax idiv reg) min_long
169 // rdx: remainder (= rax irem reg) 0
170 assert(reg != rax && reg != rdx, "reg cannot be rax or rdx register");
171 static const int64_t min_long = 0x8000000000000000;
172 Label normal_case, special_case;
173
174 // check for special case
175 cmp64(rax, ExternalAddress((address) &min_long), rdx /*rscratch*/);
176 jcc(Assembler::notEqual, normal_case);
177 xorl(rdx, rdx); // prepare rdx for possible special case (where
178 // remainder = 0)
179 cmpq(reg, -1);
180 jcc(Assembler::equal, special_case);
181
182 // handle normal case
183 bind(normal_case);
184 cdqq();
185 int idivq_offset = offset();
186 idivq(reg);
187
188 // normal and special case exit
189 bind(special_case);
190
191 return idivq_offset;
192 }
193
194 void MacroAssembler::decrementq(Register reg, int value) {
195 if (value == min_jint) { subq(reg, value); return; }
196 if (value < 0) { incrementq(reg, -value); return; }
197 if (value == 0) { ; return; }
198 if (value == 1 && UseIncDec) { decq(reg) ; return; }
199 /* else */ { subq(reg, value) ; return; }
200 }
201
202 void MacroAssembler::decrementq(Address dst, int value) {
203 if (value == min_jint) { subq(dst, value); return; }
204 if (value < 0) { incrementq(dst, -value); return; }
205 if (value == 0) { ; return; }
206 if (value == 1 && UseIncDec) { decq(dst) ; return; }
207 /* else */ { subq(dst, value) ; return; }
208 }
209
210 void MacroAssembler::incrementq(AddressLiteral dst, Register rscratch) {
211 assert(rscratch != noreg || always_reachable(dst), "missing");
212
213 if (reachable(dst)) {
214 incrementq(as_Address(dst));
215 } else {
216 lea(rscratch, dst);
217 incrementq(Address(rscratch, 0));
218 }
219 }
220
221 void MacroAssembler::incrementq(Register reg, int value) {
222 if (value == min_jint) { addq(reg, value); return; }
223 if (value < 0) { decrementq(reg, -value); return; }
224 if (value == 0) { ; return; }
225 if (value == 1 && UseIncDec) { incq(reg) ; return; }
226 /* else */ { addq(reg, value) ; return; }
227 }
228
229 void MacroAssembler::incrementq(Address dst, int value) {
230 if (value == min_jint) { addq(dst, value); return; }
231 if (value < 0) { decrementq(dst, -value); return; }
232 if (value == 0) { ; return; }
233 if (value == 1 && UseIncDec) { incq(dst) ; return; }
234 /* else */ { addq(dst, value) ; return; }
235 }
236
237 // 32bit can do a case table jump in one instruction but we no longer allow the base
238 // to be installed in the Address class
239 void MacroAssembler::jump(ArrayAddress entry, Register rscratch) {
240 lea(rscratch, entry.base());
241 Address dispatch = entry.index();
242 assert(dispatch._base == noreg, "must be");
243 dispatch._base = rscratch;
244 jmp(dispatch);
245 }
246
247 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) {
248 ShouldNotReachHere(); // 64bit doesn't use two regs
249 cmpq(x_lo, y_lo);
250 }
251
252 void MacroAssembler::lea(Register dst, AddressLiteral src) {
253 mov_literal64(dst, (intptr_t)src.target(), src.rspec());
254 }
255
256 void MacroAssembler::lea(Address dst, AddressLiteral adr, Register rscratch) {
257 lea(rscratch, adr);
258 movptr(dst, rscratch);
259 }
260
261 void MacroAssembler::leave() {
262 // %%% is this really better? Why not on 32bit too?
263 emit_int8((unsigned char)0xC9); // LEAVE
264 }
265
266 void MacroAssembler::lneg(Register hi, Register lo) {
267 ShouldNotReachHere(); // 64bit doesn't use two regs
268 negq(lo);
269 }
270
271 void MacroAssembler::movoop(Register dst, jobject obj) {
272 mov_literal64(dst, (intptr_t)obj, oop_Relocation::spec_for_immediate());
273 }
274
275 void MacroAssembler::movoop(Address dst, jobject obj, Register rscratch) {
276 mov_literal64(rscratch, (intptr_t)obj, oop_Relocation::spec_for_immediate());
277 movq(dst, rscratch);
278 }
279
280 void MacroAssembler::mov_metadata(Register dst, Metadata* obj) {
281 mov_literal64(dst, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
282 }
283
284 void MacroAssembler::mov_metadata(Address dst, Metadata* obj, Register rscratch) {
285 mov_literal64(rscratch, (intptr_t)obj, metadata_Relocation::spec_for_immediate());
286 movq(dst, rscratch);
287 }
288
289 void MacroAssembler::movptr(Register dst, AddressLiteral src) {
290 if (src.is_lval()) {
291 mov_literal64(dst, (intptr_t)src.target(), src.rspec());
292 } else {
293 if (reachable(src)) {
294 movq(dst, as_Address(src));
295 } else {
296 lea(dst, src);
297 movq(dst, Address(dst, 0));
298 }
299 }
300 }
301
302 void MacroAssembler::movptr(ArrayAddress dst, Register src, Register rscratch) {
303 movq(as_Address(dst, rscratch), src);
304 }
305
306 void MacroAssembler::movptr(Register dst, ArrayAddress src) {
307 movq(dst, as_Address(src, dst /*rscratch*/));
308 }
309
310 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
311 void MacroAssembler::movptr(Address dst, intptr_t src, Register rscratch) {
312 if (is_simm32(src)) {
313 movptr(dst, checked_cast<int32_t>(src));
314 } else {
315 mov64(rscratch, src);
316 movq(dst, rscratch);
317 }
318 }
319
320 void MacroAssembler::pushoop(jobject obj, Register rscratch) {
321 movoop(rscratch, obj);
322 push(rscratch);
323 }
324
325 void MacroAssembler::pushklass(Metadata* obj, Register rscratch) {
326 mov_metadata(rscratch, obj);
327 push(rscratch);
328 }
329
330 void MacroAssembler::pushptr(AddressLiteral src, Register rscratch) {
331 lea(rscratch, src);
332 if (src.is_lval()) {
333 push(rscratch);
334 } else {
335 pushq(Address(rscratch, 0));
336 }
337 }
338
339 static void pass_arg0(MacroAssembler* masm, Register arg) {
340 if (c_rarg0 != arg ) {
341 masm->mov(c_rarg0, arg);
342 }
343 }
344
345 static void pass_arg1(MacroAssembler* masm, Register arg) {
346 if (c_rarg1 != arg ) {
347 masm->mov(c_rarg1, arg);
348 }
349 }
350
351 static void pass_arg2(MacroAssembler* masm, Register arg) {
352 if (c_rarg2 != arg ) {
353 masm->mov(c_rarg2, arg);
354 }
355 }
356
357 static void pass_arg3(MacroAssembler* masm, Register arg) {
358 if (c_rarg3 != arg ) {
359 masm->mov(c_rarg3, arg);
360 }
361 }
362
363 void MacroAssembler::stop(const char* msg) {
364 if (ShowMessageBoxOnError) {
365 address rip = pc();
366 pusha(); // get regs on stack
367 lea(c_rarg1, InternalAddress(rip));
368 movq(c_rarg2, rsp); // pass pointer to regs array
369 }
370 // Skip AOT caching C strings in scratch buffer.
371 const char* str = (code_section()->scratch_emit()) ? msg : AOTCodeCache::add_C_string(msg);
372 lea(c_rarg0, ExternalAddress((address) str));
373 andq(rsp, -16); // align stack as required by ABI
374 call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
375 hlt();
376 }
377
378 void MacroAssembler::warn(const char* msg) {
379 push(rbp);
380 movq(rbp, rsp);
381 andq(rsp, -16); // align stack as required by push_CPU_state and call
382 push_CPU_state(); // keeps alignment at 16 bytes
383
384 #ifdef _WIN64
385 // Windows always allocates space for its register args
386 subq(rsp, frame::arg_reg_save_area_bytes);
387 #endif
388 lea(c_rarg0, ExternalAddress((address) msg));
389 call(RuntimeAddress(CAST_FROM_FN_PTR(address, warning)));
390
391 #ifdef _WIN64
392 // restore stack pointer
393 addq(rsp, frame::arg_reg_save_area_bytes);
394 #endif
395 pop_CPU_state();
396 mov(rsp, rbp);
397 pop(rbp);
398 }
399
400 void MacroAssembler::print_state() {
401 address rip = pc();
402 pusha(); // get regs on stack
403 push(rbp);
404 movq(rbp, rsp);
405 andq(rsp, -16); // align stack as required by push_CPU_state and call
406 push_CPU_state(); // keeps alignment at 16 bytes
407
408 lea(c_rarg0, InternalAddress(rip));
409 lea(c_rarg1, Address(rbp, wordSize)); // pass pointer to regs array
410 call_VM_leaf(CAST_FROM_FN_PTR(address, MacroAssembler::print_state64), c_rarg0, c_rarg1);
411
412 pop_CPU_state();
413 mov(rsp, rbp);
414 pop(rbp);
415 popa();
416 }
417
418 #ifndef PRODUCT
419 extern "C" void findpc(intptr_t x);
420 #endif
421
422 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[]) {
423 // In order to get locks to work, we need to fake a in_VM state
424 if (ShowMessageBoxOnError) {
425 JavaThread* thread = JavaThread::current();
426 JavaThreadState saved_state = thread->thread_state();
427 thread->set_thread_state(_thread_in_vm);
428 #ifndef PRODUCT
429 if (CountBytecodes || TraceBytecodes || StopInterpreterAt) {
430 ttyLocker ttyl;
431 BytecodeCounter::print();
432 }
433 #endif
434 // To see where a verify_oop failed, get $ebx+40/X for this frame.
435 // XXX correct this offset for amd64
436 // This is the value of eip which points to where verify_oop will return.
437 if (os::message_box(msg, "Execution stopped, print registers?")) {
438 print_state64(pc, regs);
439 BREAKPOINT;
440 }
441 }
442 fatal("DEBUG MESSAGE: %s", msg);
443 }
444
445 void MacroAssembler::print_state64(int64_t pc, int64_t regs[]) {
446 ttyLocker ttyl;
447 DebuggingContext debugging{};
448 tty->print_cr("rip = 0x%016lx", (intptr_t)pc);
449 #ifndef PRODUCT
450 tty->cr();
451 findpc(pc);
452 tty->cr();
453 #endif
454 #define PRINT_REG(rax, value) \
455 { tty->print("%s = ", #rax); os::print_location(tty, value); }
456 PRINT_REG(rax, regs[15]);
457 PRINT_REG(rbx, regs[12]);
458 PRINT_REG(rcx, regs[14]);
459 PRINT_REG(rdx, regs[13]);
460 PRINT_REG(rdi, regs[8]);
461 PRINT_REG(rsi, regs[9]);
462 PRINT_REG(rbp, regs[10]);
463 // rsp is actually not stored by pusha(), compute the old rsp from regs (rsp after pusha): regs + 16 = old rsp
464 PRINT_REG(rsp, (intptr_t)(®s[16]));
465 PRINT_REG(r8 , regs[7]);
466 PRINT_REG(r9 , regs[6]);
467 PRINT_REG(r10, regs[5]);
468 PRINT_REG(r11, regs[4]);
469 PRINT_REG(r12, regs[3]);
470 PRINT_REG(r13, regs[2]);
471 PRINT_REG(r14, regs[1]);
472 PRINT_REG(r15, regs[0]);
473 #undef PRINT_REG
474 // Print some words near the top of the stack.
475 int64_t* rsp = ®s[16];
476 int64_t* dump_sp = rsp;
477 for (int col1 = 0; col1 < 8; col1++) {
478 tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
479 os::print_location(tty, *dump_sp++);
480 }
481 for (int row = 0; row < 25; row++) {
482 tty->print("(rsp+0x%03x) 0x%016lx: ", (int)((intptr_t)dump_sp - (intptr_t)rsp), (intptr_t)dump_sp);
483 for (int col = 0; col < 4; col++) {
484 tty->print(" 0x%016lx", (intptr_t)*dump_sp++);
485 }
486 tty->cr();
487 }
488 // Print some instructions around pc:
489 Disassembler::decode((address)pc-64, (address)pc);
490 tty->print_cr("--------");
491 Disassembler::decode((address)pc, (address)pc+32);
492 }
493
494 // The java_calling_convention describes stack locations as ideal slots on
495 // a frame with no abi restrictions. Since we must observe abi restrictions
496 // (like the placement of the register window) the slots must be biased by
497 // the following value.
498 static int reg2offset_in(VMReg r) {
499 // Account for saved rbp and return address
500 // This should really be in_preserve_stack_slots
501 return (r->reg2stack() + 4) * VMRegImpl::stack_slot_size;
502 }
503
504 static int reg2offset_out(VMReg r) {
505 return (r->reg2stack() + SharedRuntime::out_preserve_stack_slots()) * VMRegImpl::stack_slot_size;
506 }
507
508 // A long move
509 void MacroAssembler::long_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
510
511 // The calling conventions assures us that each VMregpair is either
512 // all really one physical register or adjacent stack slots.
513
514 if (src.is_single_phys_reg() ) {
515 if (dst.is_single_phys_reg()) {
516 if (dst.first() != src.first()) {
517 mov(dst.first()->as_Register(), src.first()->as_Register());
518 }
519 } else {
520 assert(dst.is_single_reg(), "not a stack pair: (%s, %s), (%s, %s)",
521 src.first()->name(), src.second()->name(), dst.first()->name(), dst.second()->name());
522 movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
523 }
524 } else if (dst.is_single_phys_reg()) {
525 assert(src.is_single_reg(), "not a stack pair");
526 movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
527 } else {
528 assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
529 movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
530 movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
531 }
532 }
533
534 // A double move
535 void MacroAssembler::double_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
536
537 // The calling conventions assures us that each VMregpair is either
538 // all really one physical register or adjacent stack slots.
539
540 if (src.is_single_phys_reg() ) {
541 if (dst.is_single_phys_reg()) {
542 // In theory these overlap but the ordering is such that this is likely a nop
543 if ( src.first() != dst.first()) {
544 movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister());
545 }
546 } else {
547 assert(dst.is_single_reg(), "not a stack pair");
548 movdbl(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
549 }
550 } else if (dst.is_single_phys_reg()) {
551 assert(src.is_single_reg(), "not a stack pair");
552 movdbl(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
553 } else {
554 assert(src.is_single_reg() && dst.is_single_reg(), "not stack pairs");
555 movq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
556 movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
557 }
558 }
559
560
561 // A float arg may have to do float reg int reg conversion
562 void MacroAssembler::float_move(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
563 assert(!src.second()->is_valid() && !dst.second()->is_valid(), "bad float_move");
564
565 // The calling conventions assures us that each VMregpair is either
566 // all really one physical register or adjacent stack slots.
567
568 if (src.first()->is_stack()) {
569 if (dst.first()->is_stack()) {
570 movl(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
571 movptr(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
572 } else {
573 // stack to reg
574 assert(dst.first()->is_XMMRegister(), "only expect xmm registers as parameters");
575 movflt(dst.first()->as_XMMRegister(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
576 }
577 } else if (dst.first()->is_stack()) {
578 // reg to stack
579 assert(src.first()->is_XMMRegister(), "only expect xmm registers as parameters");
580 movflt(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_XMMRegister());
581 } else {
582 // reg to reg
583 // In theory these overlap but the ordering is such that this is likely a nop
584 if ( src.first() != dst.first()) {
585 movdbl(dst.first()->as_XMMRegister(), src.first()->as_XMMRegister());
586 }
587 }
588 }
589
590 // On 64 bit we will store integer like items to the stack as
591 // 64 bits items (x86_32/64 abi) even though java would only store
592 // 32bits for a parameter. On 32bit it will simply be 32 bits
593 // So this routine will do 32->32 on 32bit and 32->64 on 64bit
594 void MacroAssembler::move32_64(VMRegPair src, VMRegPair dst, Register tmp, int in_stk_bias, int out_stk_bias) {
595 if (src.first()->is_stack()) {
596 if (dst.first()->is_stack()) {
597 // stack to stack
598 movslq(tmp, Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
599 movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), tmp);
600 } else {
601 // stack to reg
602 movslq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first()) + in_stk_bias));
603 }
604 } else if (dst.first()->is_stack()) {
605 // reg to stack
606 // Do we really have to sign extend???
607 // __ movslq(src.first()->as_Register(), src.first()->as_Register());
608 movq(Address(rsp, reg2offset_out(dst.first()) + out_stk_bias), src.first()->as_Register());
609 } else {
610 // Do we really have to sign extend???
611 // __ movslq(dst.first()->as_Register(), src.first()->as_Register());
612 if (dst.first() != src.first()) {
613 movq(dst.first()->as_Register(), src.first()->as_Register());
614 }
615 }
616 }
617
618 void MacroAssembler::move_ptr(VMRegPair src, VMRegPair dst) {
619 if (src.first()->is_stack()) {
620 if (dst.first()->is_stack()) {
621 // stack to stack
622 movq(rax, Address(rbp, reg2offset_in(src.first())));
623 movq(Address(rsp, reg2offset_out(dst.first())), rax);
624 } else {
625 // stack to reg
626 movq(dst.first()->as_Register(), Address(rbp, reg2offset_in(src.first())));
627 }
628 } else if (dst.first()->is_stack()) {
629 // reg to stack
630 movq(Address(rsp, reg2offset_out(dst.first())), src.first()->as_Register());
631 } else {
632 if (dst.first() != src.first()) {
633 movq(dst.first()->as_Register(), src.first()->as_Register());
634 }
635 }
636 }
637
638 // An oop arg. Must pass a handle not the oop itself
639 void MacroAssembler::object_move(OopMap* map,
640 int oop_handle_offset,
641 int framesize_in_slots,
642 VMRegPair src,
643 VMRegPair dst,
644 bool is_receiver,
645 int* receiver_offset) {
646
647 // must pass a handle. First figure out the location we use as a handle
648
649 Register rHandle = dst.first()->is_stack() ? rax : dst.first()->as_Register();
650
651 // See if oop is null if it is we need no handle
652
653 if (src.first()->is_stack()) {
654
655 // Oop is already on the stack as an argument
656 int offset_in_older_frame = src.first()->reg2stack() + SharedRuntime::out_preserve_stack_slots();
657 map->set_oop(VMRegImpl::stack2reg(offset_in_older_frame + framesize_in_slots));
658 if (is_receiver) {
659 *receiver_offset = (offset_in_older_frame + framesize_in_slots) * VMRegImpl::stack_slot_size;
660 }
661
662 cmpptr(Address(rbp, reg2offset_in(src.first())), NULL_WORD);
663 lea(rHandle, Address(rbp, reg2offset_in(src.first())));
664 // conditionally move a null
665 cmovptr(Assembler::equal, rHandle, Address(rbp, reg2offset_in(src.first())));
666 } else {
667
668 // Oop is in a register we must store it to the space we reserve
669 // on the stack for oop_handles and pass a handle if oop is non-null
670
671 const Register rOop = src.first()->as_Register();
672 int oop_slot;
673 if (rOop == j_rarg0)
674 oop_slot = 0;
675 else if (rOop == j_rarg1)
676 oop_slot = 1;
677 else if (rOop == j_rarg2)
678 oop_slot = 2;
679 else if (rOop == j_rarg3)
680 oop_slot = 3;
681 else if (rOop == j_rarg4)
682 oop_slot = 4;
683 else {
684 assert(rOop == j_rarg5, "wrong register");
685 oop_slot = 5;
686 }
687
688 oop_slot = oop_slot * VMRegImpl::slots_per_word + oop_handle_offset;
689 int offset = oop_slot*VMRegImpl::stack_slot_size;
690
691 map->set_oop(VMRegImpl::stack2reg(oop_slot));
692 // Store oop in handle area, may be null
693 movptr(Address(rsp, offset), rOop);
694 if (is_receiver) {
695 *receiver_offset = offset;
696 }
697
698 cmpptr(rOop, NULL_WORD);
699 lea(rHandle, Address(rsp, offset));
700 // conditionally move a null from the handle area where it was just stored
701 cmovptr(Assembler::equal, rHandle, Address(rsp, offset));
702 }
703
704 // If arg is on the stack then place it otherwise it is already in correct reg.
705 if (dst.first()->is_stack()) {
706 movptr(Address(rsp, reg2offset_out(dst.first())), rHandle);
707 }
708 }
709
710 void MacroAssembler::addptr(Register dst, int32_t imm32) {
711 addq(dst, imm32);
712 }
713
714 void MacroAssembler::addptr(Register dst, Register src) {
715 addq(dst, src);
716 }
717
718 void MacroAssembler::addptr(Address dst, Register src) {
719 addq(dst, src);
720 }
721
722 void MacroAssembler::addsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
723 assert(rscratch != noreg || always_reachable(src), "missing");
724
725 if (reachable(src)) {
726 Assembler::addsd(dst, as_Address(src));
727 } else {
728 lea(rscratch, src);
729 Assembler::addsd(dst, Address(rscratch, 0));
730 }
731 }
732
733 void MacroAssembler::addss(XMMRegister dst, AddressLiteral src, Register rscratch) {
734 assert(rscratch != noreg || always_reachable(src), "missing");
735
736 if (reachable(src)) {
737 addss(dst, as_Address(src));
738 } else {
739 lea(rscratch, src);
740 addss(dst, Address(rscratch, 0));
741 }
742 }
743
744 void MacroAssembler::addpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
745 assert(rscratch != noreg || always_reachable(src), "missing");
746
747 if (reachable(src)) {
748 Assembler::addpd(dst, as_Address(src));
749 } else {
750 lea(rscratch, src);
751 Assembler::addpd(dst, Address(rscratch, 0));
752 }
753 }
754
755 // See 8273459. Function for ensuring 64-byte alignment, intended for stubs only.
756 // Stub code is generated once and never copied.
757 // NMethods can't use this because they get copied and we can't force alignment > 32 bytes.
758 void MacroAssembler::align64() {
759 align(64, (uint)(uintptr_t)pc());
760 }
761
762 void MacroAssembler::align32() {
763 align(32, (uint)(uintptr_t)pc());
764 }
765
766 void MacroAssembler::align(uint modulus) {
767 // 8273459: Ensure alignment is possible with current segment alignment
768 assert(modulus <= (uintx)CodeEntryAlignment, "Alignment must be <= CodeEntryAlignment");
769 align(modulus, offset());
770 }
771
772 void MacroAssembler::align(uint modulus, uint target) {
773 if (target % modulus != 0) {
774 nop(modulus - (target % modulus));
775 }
776 }
777
778 void MacroAssembler::push_f(XMMRegister r) {
779 subptr(rsp, wordSize);
780 movflt(Address(rsp, 0), r);
781 }
782
783 void MacroAssembler::pop_f(XMMRegister r) {
784 movflt(r, Address(rsp, 0));
785 addptr(rsp, wordSize);
786 }
787
788 void MacroAssembler::push_d(XMMRegister r) {
789 subptr(rsp, 2 * wordSize);
790 movdbl(Address(rsp, 0), r);
791 }
792
793 void MacroAssembler::pop_d(XMMRegister r) {
794 movdbl(r, Address(rsp, 0));
795 addptr(rsp, 2 * Interpreter::stackElementSize);
796 }
797
798 void MacroAssembler::andpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
799 // Used in sign-masking with aligned address.
800 assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
801 assert(rscratch != noreg || always_reachable(src), "missing");
802
803 if (UseAVX > 2 &&
804 (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
805 (dst->encoding() >= 16)) {
806 vpand(dst, dst, src, AVX_512bit, rscratch);
807 } else if (reachable(src)) {
808 Assembler::andpd(dst, as_Address(src));
809 } else {
810 lea(rscratch, src);
811 Assembler::andpd(dst, Address(rscratch, 0));
812 }
813 }
814
815 void MacroAssembler::andps(XMMRegister dst, AddressLiteral src, Register rscratch) {
816 // Used in sign-masking with aligned address.
817 assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
818 assert(rscratch != noreg || always_reachable(src), "missing");
819
820 if (reachable(src)) {
821 Assembler::andps(dst, as_Address(src));
822 } else {
823 lea(rscratch, src);
824 Assembler::andps(dst, Address(rscratch, 0));
825 }
826 }
827
828 void MacroAssembler::andptr(Register dst, int32_t imm32) {
829 andq(dst, imm32);
830 }
831
832 void MacroAssembler::andq(Register dst, AddressLiteral src, Register rscratch) {
833 assert(rscratch != noreg || always_reachable(src), "missing");
834
835 if (reachable(src)) {
836 andq(dst, as_Address(src));
837 } else {
838 lea(rscratch, src);
839 andq(dst, Address(rscratch, 0));
840 }
841 }
842
843 void MacroAssembler::atomic_incl(Address counter_addr) {
844 lock();
845 incrementl(counter_addr);
846 }
847
848 void MacroAssembler::atomic_incl(AddressLiteral counter_addr, Register rscratch) {
849 assert(rscratch != noreg || always_reachable(counter_addr), "missing");
850
851 if (reachable(counter_addr)) {
852 atomic_incl(as_Address(counter_addr));
853 } else {
854 lea(rscratch, counter_addr);
855 atomic_incl(Address(rscratch, 0));
856 }
857 }
858
859 void MacroAssembler::atomic_incq(Address counter_addr) {
860 lock();
861 incrementq(counter_addr);
862 }
863
864 void MacroAssembler::atomic_incq(AddressLiteral counter_addr, Register rscratch) {
865 assert(rscratch != noreg || always_reachable(counter_addr), "missing");
866
867 if (reachable(counter_addr)) {
868 atomic_incq(as_Address(counter_addr));
869 } else {
870 lea(rscratch, counter_addr);
871 atomic_incq(Address(rscratch, 0));
872 }
873 }
874
875 // Writes to stack successive pages until offset reached to check for
876 // stack overflow + shadow pages. This clobbers tmp.
877 void MacroAssembler::bang_stack_size(Register size, Register tmp) {
878 movptr(tmp, rsp);
879 // Bang stack for total size given plus shadow page size.
880 // Bang one page at a time because large size can bang beyond yellow and
881 // red zones.
882 Label loop;
883 bind(loop);
884 movl(Address(tmp, (-(int)os::vm_page_size())), size );
885 subptr(tmp, (int)os::vm_page_size());
886 subl(size, (int)os::vm_page_size());
887 jcc(Assembler::greater, loop);
888
889 // Bang down shadow pages too.
890 // At this point, (tmp-0) is the last address touched, so don't
891 // touch it again. (It was touched as (tmp-pagesize) but then tmp
892 // was post-decremented.) Skip this address by starting at i=1, and
893 // touch a few more pages below. N.B. It is important to touch all
894 // the way down including all pages in the shadow zone.
895 for (int i = 1; i < ((int)StackOverflow::stack_shadow_zone_size() / (int)os::vm_page_size()); i++) {
896 // this could be any sized move but this is can be a debugging crumb
897 // so the bigger the better.
898 movptr(Address(tmp, (-i*(int)os::vm_page_size())), size );
899 }
900 }
901
902 void MacroAssembler::reserved_stack_check() {
903 // testing if reserved zone needs to be enabled
904 Label no_reserved_zone_enabling;
905
906 cmpptr(rsp, Address(r15_thread, JavaThread::reserved_stack_activation_offset()));
907 jcc(Assembler::below, no_reserved_zone_enabling);
908
909 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), r15_thread);
910 jump(RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
911 should_not_reach_here();
912
913 bind(no_reserved_zone_enabling);
914 }
915
916 void MacroAssembler::c2bool(Register x) {
917 // implements x == 0 ? 0 : 1
918 // note: must only look at least-significant byte of x
919 // since C-style booleans are stored in one byte
920 // only! (was bug)
921 andl(x, 0xFF);
922 setb(Assembler::notZero, x);
923 }
924
925 // Wouldn't need if AddressLiteral version had new name
926 void MacroAssembler::call(Label& L, relocInfo::relocType rtype) {
927 Assembler::call(L, rtype);
928 }
929
930 void MacroAssembler::call(Register entry) {
931 Assembler::call(entry);
932 }
933
934 void MacroAssembler::call(AddressLiteral entry, Register rscratch) {
935 assert(rscratch != noreg || always_reachable(entry), "missing");
936
937 if (reachable(entry)) {
938 Assembler::call_literal(entry.target(), entry.rspec());
939 } else {
940 lea(rscratch, entry);
941 Assembler::call(rscratch);
942 }
943 }
944
945 void MacroAssembler::ic_call(address entry, jint method_index) {
946 RelocationHolder rh = virtual_call_Relocation::spec(pc(), method_index);
947 // Needs full 64-bit immediate for later patching.
948 mov64(rax, (int64_t)Universe::non_oop_word());
949 call(AddressLiteral(entry, rh));
950 }
951
952 int MacroAssembler::ic_check_size() {
953 return UseCompactObjectHeaders ? 17 : 14;
954 }
955
956 int MacroAssembler::ic_check(int end_alignment) {
957 Register receiver = j_rarg0;
958 Register data = rax;
959 Register temp = rscratch1;
960
961 // The UEP of a code blob ensures that the VEP is padded. However, the padding of the UEP is placed
962 // before the inline cache check, so we don't have to execute any nop instructions when dispatching
963 // through the UEP, yet we can ensure that the VEP is aligned appropriately. That's why we align
964 // before the inline cache check here, and not after
965 align(end_alignment, offset() + ic_check_size());
966
967 int uep_offset = offset();
968
969 if (UseCompactObjectHeaders) {
970 load_narrow_klass_compact(temp, receiver);
971 cmpl(temp, Address(data, CompiledICData::speculated_klass_offset()));
972 } else if (UseCompressedClassPointers) {
973 movl(temp, Address(receiver, oopDesc::klass_offset_in_bytes()));
974 cmpl(temp, Address(data, CompiledICData::speculated_klass_offset()));
975 } else {
976 movptr(temp, Address(receiver, oopDesc::klass_offset_in_bytes()));
977 cmpptr(temp, Address(data, CompiledICData::speculated_klass_offset()));
978 }
979
980 // if inline cache check fails, then jump to runtime routine
981 jump_cc(Assembler::notEqual, RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
982 assert((offset() % end_alignment) == 0, "Misaligned verified entry point (%d, %d, %d)", uep_offset, offset(), end_alignment);
983
984 return uep_offset;
985 }
986
987 void MacroAssembler::emit_static_call_stub() {
988 // Static stub relocation also tags the Method* in the code-stream.
989 mov_metadata(rbx, (Metadata*) nullptr); // Method is zapped till fixup time.
990 // This is recognized as unresolved by relocs/nativeinst/ic code.
991 jump(RuntimeAddress(pc()));
992 }
993
994 // Implementation of call_VM versions
995
996 void MacroAssembler::call_VM(Register oop_result,
997 address entry_point,
998 bool check_exceptions) {
999 Label C, E;
1000 call(C, relocInfo::none);
1001 jmp(E);
1002
1003 bind(C);
1004 call_VM_helper(oop_result, entry_point, 0, check_exceptions);
1005 ret(0);
1006
1007 bind(E);
1008 }
1009
1010 void MacroAssembler::call_VM(Register oop_result,
1011 address entry_point,
1012 Register arg_1,
1013 bool check_exceptions) {
1014 Label C, E;
1015 call(C, relocInfo::none);
1016 jmp(E);
1017
1018 bind(C);
1019 pass_arg1(this, arg_1);
1020 call_VM_helper(oop_result, entry_point, 1, check_exceptions);
1021 ret(0);
1022
1023 bind(E);
1024 }
1025
1026 void MacroAssembler::call_VM(Register oop_result,
1027 address entry_point,
1028 Register arg_1,
1029 Register arg_2,
1030 bool check_exceptions) {
1031 Label C, E;
1032 call(C, relocInfo::none);
1033 jmp(E);
1034
1035 bind(C);
1036
1037 assert_different_registers(arg_1, c_rarg2);
1038
1039 pass_arg2(this, arg_2);
1040 pass_arg1(this, arg_1);
1041 call_VM_helper(oop_result, entry_point, 2, check_exceptions);
1042 ret(0);
1043
1044 bind(E);
1045 }
1046
1047 void MacroAssembler::call_VM(Register oop_result,
1048 address entry_point,
1049 Register arg_1,
1050 Register arg_2,
1051 Register arg_3,
1052 bool check_exceptions) {
1053 Label C, E;
1054 call(C, relocInfo::none);
1055 jmp(E);
1056
1057 bind(C);
1058
1059 assert_different_registers(arg_1, c_rarg2, c_rarg3);
1060 assert_different_registers(arg_2, c_rarg3);
1061 pass_arg3(this, arg_3);
1062 pass_arg2(this, arg_2);
1063 pass_arg1(this, arg_1);
1064 call_VM_helper(oop_result, entry_point, 3, check_exceptions);
1065 ret(0);
1066
1067 bind(E);
1068 }
1069
1070 void MacroAssembler::call_VM(Register oop_result,
1071 Register last_java_sp,
1072 address entry_point,
1073 int number_of_arguments,
1074 bool check_exceptions) {
1075 call_VM_base(oop_result, last_java_sp, entry_point, number_of_arguments, check_exceptions);
1076 }
1077
1078 void MacroAssembler::call_VM(Register oop_result,
1079 Register last_java_sp,
1080 address entry_point,
1081 Register arg_1,
1082 bool check_exceptions) {
1083 pass_arg1(this, arg_1);
1084 call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
1085 }
1086
1087 void MacroAssembler::call_VM(Register oop_result,
1088 Register last_java_sp,
1089 address entry_point,
1090 Register arg_1,
1091 Register arg_2,
1092 bool check_exceptions) {
1093
1094 assert_different_registers(arg_1, c_rarg2);
1095 pass_arg2(this, arg_2);
1096 pass_arg1(this, arg_1);
1097 call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
1098 }
1099
1100 void MacroAssembler::call_VM(Register oop_result,
1101 Register last_java_sp,
1102 address entry_point,
1103 Register arg_1,
1104 Register arg_2,
1105 Register arg_3,
1106 bool check_exceptions) {
1107 assert_different_registers(arg_1, c_rarg2, c_rarg3);
1108 assert_different_registers(arg_2, c_rarg3);
1109 pass_arg3(this, arg_3);
1110 pass_arg2(this, arg_2);
1111 pass_arg1(this, arg_1);
1112 call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
1113 }
1114
1115 void MacroAssembler::super_call_VM(Register oop_result,
1116 Register last_java_sp,
1117 address entry_point,
1118 int number_of_arguments,
1119 bool check_exceptions) {
1120 MacroAssembler::call_VM_base(oop_result, last_java_sp, entry_point, number_of_arguments, check_exceptions);
1121 }
1122
1123 void MacroAssembler::super_call_VM(Register oop_result,
1124 Register last_java_sp,
1125 address entry_point,
1126 Register arg_1,
1127 bool check_exceptions) {
1128 pass_arg1(this, arg_1);
1129 super_call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions);
1130 }
1131
1132 void MacroAssembler::super_call_VM(Register oop_result,
1133 Register last_java_sp,
1134 address entry_point,
1135 Register arg_1,
1136 Register arg_2,
1137 bool check_exceptions) {
1138
1139 assert_different_registers(arg_1, c_rarg2);
1140 pass_arg2(this, arg_2);
1141 pass_arg1(this, arg_1);
1142 super_call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions);
1143 }
1144
1145 void MacroAssembler::super_call_VM(Register oop_result,
1146 Register last_java_sp,
1147 address entry_point,
1148 Register arg_1,
1149 Register arg_2,
1150 Register arg_3,
1151 bool check_exceptions) {
1152 assert_different_registers(arg_1, c_rarg2, c_rarg3);
1153 assert_different_registers(arg_2, c_rarg3);
1154 pass_arg3(this, arg_3);
1155 pass_arg2(this, arg_2);
1156 pass_arg1(this, arg_1);
1157 super_call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions);
1158 }
1159
1160 void MacroAssembler::call_VM_base(Register oop_result,
1161 Register last_java_sp,
1162 address entry_point,
1163 int number_of_arguments,
1164 bool check_exceptions) {
1165 Register java_thread = r15_thread;
1166
1167 // determine last_java_sp register
1168 if (!last_java_sp->is_valid()) {
1169 last_java_sp = rsp;
1170 }
1171 // debugging support
1172 assert(number_of_arguments >= 0 , "cannot have negative number of arguments");
1173 #ifdef ASSERT
1174 // TraceBytecodes does not use r12 but saves it over the call, so don't verify
1175 // r12 is the heapbase.
1176 if (UseCompressedOops && !TraceBytecodes) verify_heapbase("call_VM_base: heap base corrupted?");
1177 #endif // ASSERT
1178
1179 assert(java_thread != oop_result , "cannot use the same register for java_thread & oop_result");
1180 assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp");
1181
1182 // push java thread (becomes first argument of C function)
1183
1184 mov(c_rarg0, r15_thread);
1185
1186 // set last Java frame before call
1187 assert(last_java_sp != rbp, "can't use ebp/rbp");
1188
1189 // Only interpreter should have to set fp
1190 set_last_Java_frame(last_java_sp, rbp, nullptr, rscratch1);
1191
1192 // do the call, remove parameters
1193 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1194
1195 #ifdef ASSERT
1196 // Check that thread register is not clobbered.
1197 guarantee(java_thread != rax, "change this code");
1198 push(rax);
1199 { Label L;
1200 get_thread_slow(rax);
1201 cmpptr(java_thread, rax);
1202 jcc(Assembler::equal, L);
1203 STOP("MacroAssembler::call_VM_base: java_thread not callee saved?");
1204 bind(L);
1205 }
1206 pop(rax);
1207 #endif
1208
1209 // reset last Java frame
1210 // Only interpreter should have to clear fp
1211 reset_last_Java_frame(true);
1212
1213 // C++ interp handles this in the interpreter
1214 check_and_handle_popframe();
1215 check_and_handle_earlyret();
1216
1217 if (check_exceptions) {
1218 // check for pending exceptions (java_thread is set upon return)
1219 cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
1220 // This used to conditionally jump to forward_exception however it is
1221 // possible if we relocate that the branch will not reach. So we must jump
1222 // around so we can always reach
1223
1224 Label ok;
1225 jcc(Assembler::equal, ok);
1226 jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
1227 bind(ok);
1228 }
1229
1230 // get oop result if there is one and reset the value in the thread
1231 if (oop_result->is_valid()) {
1232 get_vm_result_oop(oop_result);
1233 }
1234 }
1235
1236 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
1237 // Calculate the value for last_Java_sp somewhat subtle.
1238 // call_VM does an intermediate call which places a return address on
1239 // the stack just under the stack pointer as the user finished with it.
1240 // This allows use to retrieve last_Java_pc from last_Java_sp[-1].
1241
1242 // We've pushed one address, correct last_Java_sp
1243 lea(rax, Address(rsp, wordSize));
1244
1245 call_VM_base(oop_result, rax, entry_point, number_of_arguments, check_exceptions);
1246 }
1247
1248 // Use this method when MacroAssembler version of call_VM_leaf_base() should be called from Interpreter.
1249 void MacroAssembler::call_VM_leaf0(address entry_point) {
1250 MacroAssembler::call_VM_leaf_base(entry_point, 0);
1251 }
1252
1253 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
1254 call_VM_leaf_base(entry_point, number_of_arguments);
1255 }
1256
1257 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) {
1258 pass_arg0(this, arg_0);
1259 call_VM_leaf(entry_point, 1);
1260 }
1261
1262 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
1263
1264 assert_different_registers(arg_0, c_rarg1);
1265 pass_arg1(this, arg_1);
1266 pass_arg0(this, arg_0);
1267 call_VM_leaf(entry_point, 2);
1268 }
1269
1270 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
1271 assert_different_registers(arg_0, c_rarg1, c_rarg2);
1272 assert_different_registers(arg_1, c_rarg2);
1273 pass_arg2(this, arg_2);
1274 pass_arg1(this, arg_1);
1275 pass_arg0(this, arg_0);
1276 call_VM_leaf(entry_point, 3);
1277 }
1278
1279 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
1280 assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
1281 assert_different_registers(arg_1, c_rarg2, c_rarg3);
1282 assert_different_registers(arg_2, c_rarg3);
1283 pass_arg3(this, arg_3);
1284 pass_arg2(this, arg_2);
1285 pass_arg1(this, arg_1);
1286 pass_arg0(this, arg_0);
1287 call_VM_leaf(entry_point, 3);
1288 }
1289
1290 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
1291 pass_arg0(this, arg_0);
1292 MacroAssembler::call_VM_leaf_base(entry_point, 1);
1293 }
1294
1295 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
1296 assert_different_registers(arg_0, c_rarg1);
1297 pass_arg1(this, arg_1);
1298 pass_arg0(this, arg_0);
1299 MacroAssembler::call_VM_leaf_base(entry_point, 2);
1300 }
1301
1302 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
1303 assert_different_registers(arg_0, c_rarg1, c_rarg2);
1304 assert_different_registers(arg_1, c_rarg2);
1305 pass_arg2(this, arg_2);
1306 pass_arg1(this, arg_1);
1307 pass_arg0(this, arg_0);
1308 MacroAssembler::call_VM_leaf_base(entry_point, 3);
1309 }
1310
1311 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
1312 assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
1313 assert_different_registers(arg_1, c_rarg2, c_rarg3);
1314 assert_different_registers(arg_2, c_rarg3);
1315 pass_arg3(this, arg_3);
1316 pass_arg2(this, arg_2);
1317 pass_arg1(this, arg_1);
1318 pass_arg0(this, arg_0);
1319 MacroAssembler::call_VM_leaf_base(entry_point, 4);
1320 }
1321
1322 void MacroAssembler::get_vm_result_oop(Register oop_result) {
1323 movptr(oop_result, Address(r15_thread, JavaThread::vm_result_oop_offset()));
1324 movptr(Address(r15_thread, JavaThread::vm_result_oop_offset()), NULL_WORD);
1325 verify_oop_msg(oop_result, "broken oop in call_VM_base");
1326 }
1327
1328 void MacroAssembler::get_vm_result_metadata(Register metadata_result) {
1329 movptr(metadata_result, Address(r15_thread, JavaThread::vm_result_metadata_offset()));
1330 movptr(Address(r15_thread, JavaThread::vm_result_metadata_offset()), NULL_WORD);
1331 }
1332
1333 void MacroAssembler::check_and_handle_earlyret() {
1334 }
1335
1336 void MacroAssembler::check_and_handle_popframe() {
1337 }
1338
1339 void MacroAssembler::cmp32(AddressLiteral src1, int32_t imm, Register rscratch) {
1340 assert(rscratch != noreg || always_reachable(src1), "missing");
1341
1342 if (reachable(src1)) {
1343 cmpl(as_Address(src1), imm);
1344 } else {
1345 lea(rscratch, src1);
1346 cmpl(Address(rscratch, 0), imm);
1347 }
1348 }
1349
1350 void MacroAssembler::cmp32(Register src1, AddressLiteral src2, Register rscratch) {
1351 assert(!src2.is_lval(), "use cmpptr");
1352 assert(rscratch != noreg || always_reachable(src2), "missing");
1353
1354 if (reachable(src2)) {
1355 cmpl(src1, as_Address(src2));
1356 } else {
1357 lea(rscratch, src2);
1358 cmpl(src1, Address(rscratch, 0));
1359 }
1360 }
1361
1362 void MacroAssembler::cmp32(Register src1, int32_t imm) {
1363 Assembler::cmpl(src1, imm);
1364 }
1365
1366 void MacroAssembler::cmp32(Register src1, Address src2) {
1367 Assembler::cmpl(src1, src2);
1368 }
1369
1370 void MacroAssembler::cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
1371 ucomisd(opr1, opr2);
1372
1373 Label L;
1374 if (unordered_is_less) {
1375 movl(dst, -1);
1376 jcc(Assembler::parity, L);
1377 jcc(Assembler::below , L);
1378 movl(dst, 0);
1379 jcc(Assembler::equal , L);
1380 increment(dst);
1381 } else { // unordered is greater
1382 movl(dst, 1);
1383 jcc(Assembler::parity, L);
1384 jcc(Assembler::above , L);
1385 movl(dst, 0);
1386 jcc(Assembler::equal , L);
1387 decrementl(dst);
1388 }
1389 bind(L);
1390 }
1391
1392 void MacroAssembler::cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) {
1393 ucomiss(opr1, opr2);
1394
1395 Label L;
1396 if (unordered_is_less) {
1397 movl(dst, -1);
1398 jcc(Assembler::parity, L);
1399 jcc(Assembler::below , L);
1400 movl(dst, 0);
1401 jcc(Assembler::equal , L);
1402 increment(dst);
1403 } else { // unordered is greater
1404 movl(dst, 1);
1405 jcc(Assembler::parity, L);
1406 jcc(Assembler::above , L);
1407 movl(dst, 0);
1408 jcc(Assembler::equal , L);
1409 decrementl(dst);
1410 }
1411 bind(L);
1412 }
1413
1414
1415 void MacroAssembler::cmp8(AddressLiteral src1, int imm, Register rscratch) {
1416 assert(rscratch != noreg || always_reachable(src1), "missing");
1417
1418 if (reachable(src1)) {
1419 cmpb(as_Address(src1), imm);
1420 } else {
1421 lea(rscratch, src1);
1422 cmpb(Address(rscratch, 0), imm);
1423 }
1424 }
1425
1426 void MacroAssembler::cmpptr(Register src1, AddressLiteral src2, Register rscratch) {
1427 assert(rscratch != noreg || always_reachable(src2), "missing");
1428
1429 if (src2.is_lval()) {
1430 movptr(rscratch, src2);
1431 Assembler::cmpq(src1, rscratch);
1432 } else if (reachable(src2)) {
1433 cmpq(src1, as_Address(src2));
1434 } else {
1435 lea(rscratch, src2);
1436 Assembler::cmpq(src1, Address(rscratch, 0));
1437 }
1438 }
1439
1440 void MacroAssembler::cmpptr(Address src1, AddressLiteral src2, Register rscratch) {
1441 assert(src2.is_lval(), "not a mem-mem compare");
1442 // moves src2's literal address
1443 movptr(rscratch, src2);
1444 Assembler::cmpq(src1, rscratch);
1445 }
1446
1447 void MacroAssembler::cmpoop(Register src1, Register src2) {
1448 cmpptr(src1, src2);
1449 }
1450
1451 void MacroAssembler::cmpoop(Register src1, Address src2) {
1452 cmpptr(src1, src2);
1453 }
1454
1455 void MacroAssembler::cmpoop(Register src1, jobject src2, Register rscratch) {
1456 movoop(rscratch, src2);
1457 cmpptr(src1, rscratch);
1458 }
1459
1460 void MacroAssembler::locked_cmpxchgptr(Register reg, AddressLiteral adr, Register rscratch) {
1461 assert(rscratch != noreg || always_reachable(adr), "missing");
1462
1463 if (reachable(adr)) {
1464 lock();
1465 cmpxchgptr(reg, as_Address(adr));
1466 } else {
1467 lea(rscratch, adr);
1468 lock();
1469 cmpxchgptr(reg, Address(rscratch, 0));
1470 }
1471 }
1472
1473 void MacroAssembler::cmpxchgptr(Register reg, Address adr) {
1474 cmpxchgq(reg, adr);
1475 }
1476
1477 void MacroAssembler::comisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
1478 assert(rscratch != noreg || always_reachable(src), "missing");
1479
1480 if (reachable(src)) {
1481 Assembler::comisd(dst, as_Address(src));
1482 } else {
1483 lea(rscratch, src);
1484 Assembler::comisd(dst, Address(rscratch, 0));
1485 }
1486 }
1487
1488 void MacroAssembler::comiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
1489 assert(rscratch != noreg || always_reachable(src), "missing");
1490
1491 if (reachable(src)) {
1492 Assembler::comiss(dst, as_Address(src));
1493 } else {
1494 lea(rscratch, src);
1495 Assembler::comiss(dst, Address(rscratch, 0));
1496 }
1497 }
1498
1499
1500 void MacroAssembler::cond_inc32(Condition cond, AddressLiteral counter_addr, Register rscratch) {
1501 assert(rscratch != noreg || always_reachable(counter_addr), "missing");
1502
1503 Condition negated_cond = negate_condition(cond);
1504 Label L;
1505 jcc(negated_cond, L);
1506 pushf(); // Preserve flags
1507 atomic_incl(counter_addr, rscratch);
1508 popf();
1509 bind(L);
1510 }
1511
1512 int MacroAssembler::corrected_idivl(Register reg) {
1513 // Full implementation of Java idiv and irem; checks for
1514 // special case as described in JVM spec., p.243 & p.271.
1515 // The function returns the (pc) offset of the idivl
1516 // instruction - may be needed for implicit exceptions.
1517 //
1518 // normal case special case
1519 //
1520 // input : rax,: dividend min_int
1521 // reg: divisor (may not be rax,/rdx) -1
1522 //
1523 // output: rax,: quotient (= rax, idiv reg) min_int
1524 // rdx: remainder (= rax, irem reg) 0
1525 assert(reg != rax && reg != rdx, "reg cannot be rax, or rdx register");
1526 const int min_int = 0x80000000;
1527 Label normal_case, special_case;
1528
1529 // check for special case
1530 cmpl(rax, min_int);
1531 jcc(Assembler::notEqual, normal_case);
1532 xorl(rdx, rdx); // prepare rdx for possible special case (where remainder = 0)
1533 cmpl(reg, -1);
1534 jcc(Assembler::equal, special_case);
1535
1536 // handle normal case
1537 bind(normal_case);
1538 cdql();
1539 int idivl_offset = offset();
1540 idivl(reg);
1541
1542 // normal and special case exit
1543 bind(special_case);
1544
1545 return idivl_offset;
1546 }
1547
1548
1549
1550 void MacroAssembler::decrementl(Register reg, int value) {
1551 if (value == min_jint) {subl(reg, value) ; return; }
1552 if (value < 0) { incrementl(reg, -value); return; }
1553 if (value == 0) { ; return; }
1554 if (value == 1 && UseIncDec) { decl(reg) ; return; }
1555 /* else */ { subl(reg, value) ; return; }
1556 }
1557
1558 void MacroAssembler::decrementl(Address dst, int value) {
1559 if (value == min_jint) {subl(dst, value) ; return; }
1560 if (value < 0) { incrementl(dst, -value); return; }
1561 if (value == 0) { ; return; }
1562 if (value == 1 && UseIncDec) { decl(dst) ; return; }
1563 /* else */ { subl(dst, value) ; return; }
1564 }
1565
1566 void MacroAssembler::division_with_shift (Register reg, int shift_value) {
1567 assert(shift_value > 0, "illegal shift value");
1568 Label _is_positive;
1569 testl (reg, reg);
1570 jcc (Assembler::positive, _is_positive);
1571 int offset = (1 << shift_value) - 1 ;
1572
1573 if (offset == 1) {
1574 incrementl(reg);
1575 } else {
1576 addl(reg, offset);
1577 }
1578
1579 bind (_is_positive);
1580 sarl(reg, shift_value);
1581 }
1582
1583 void MacroAssembler::divsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
1584 assert(rscratch != noreg || always_reachable(src), "missing");
1585
1586 if (reachable(src)) {
1587 Assembler::divsd(dst, as_Address(src));
1588 } else {
1589 lea(rscratch, src);
1590 Assembler::divsd(dst, Address(rscratch, 0));
1591 }
1592 }
1593
1594 void MacroAssembler::divss(XMMRegister dst, AddressLiteral src, Register rscratch) {
1595 assert(rscratch != noreg || always_reachable(src), "missing");
1596
1597 if (reachable(src)) {
1598 Assembler::divss(dst, as_Address(src));
1599 } else {
1600 lea(rscratch, src);
1601 Assembler::divss(dst, Address(rscratch, 0));
1602 }
1603 }
1604
1605 void MacroAssembler::enter() {
1606 push(rbp);
1607 mov(rbp, rsp);
1608 }
1609
1610 void MacroAssembler::post_call_nop() {
1611 if (!Continuations::enabled()) {
1612 return;
1613 }
1614 InstructionMark im(this);
1615 relocate(post_call_nop_Relocation::spec());
1616 InlineSkippedInstructionsCounter skipCounter(this);
1617 emit_int8((uint8_t)0x0f);
1618 emit_int8((uint8_t)0x1f);
1619 emit_int8((uint8_t)0x84);
1620 emit_int8((uint8_t)0x00);
1621 emit_int32(0x00);
1622 }
1623
1624 // A 5 byte nop that is safe for patching (see patch_verified_entry)
1625 void MacroAssembler::fat_nop() {
1626 if (UseAddressNop) {
1627 addr_nop_5();
1628 } else {
1629 emit_int8((uint8_t)0x26); // es:
1630 emit_int8((uint8_t)0x2e); // cs:
1631 emit_int8((uint8_t)0x64); // fs:
1632 emit_int8((uint8_t)0x65); // gs:
1633 emit_int8((uint8_t)0x90);
1634 }
1635 }
1636
1637 void MacroAssembler::mulpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
1638 assert(rscratch != noreg || always_reachable(src), "missing");
1639 if (reachable(src)) {
1640 Assembler::mulpd(dst, as_Address(src));
1641 } else {
1642 lea(rscratch, src);
1643 Assembler::mulpd(dst, Address(rscratch, 0));
1644 }
1645 }
1646
1647 // dst = c = a * b + c
1648 void MacroAssembler::fmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
1649 Assembler::vfmadd231sd(c, a, b);
1650 if (dst != c) {
1651 movdbl(dst, c);
1652 }
1653 }
1654
1655 // dst = c = a * b + c
1656 void MacroAssembler::fmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c) {
1657 Assembler::vfmadd231ss(c, a, b);
1658 if (dst != c) {
1659 movflt(dst, c);
1660 }
1661 }
1662
1663 // dst = c = a * b + c
1664 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
1665 Assembler::vfmadd231pd(c, a, b, vector_len);
1666 if (dst != c) {
1667 vmovdqu(dst, c);
1668 }
1669 }
1670
1671 // dst = c = a * b + c
1672 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len) {
1673 Assembler::vfmadd231ps(c, a, b, vector_len);
1674 if (dst != c) {
1675 vmovdqu(dst, c);
1676 }
1677 }
1678
1679 // dst = c = a * b + c
1680 void MacroAssembler::vfmad(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
1681 Assembler::vfmadd231pd(c, a, b, vector_len);
1682 if (dst != c) {
1683 vmovdqu(dst, c);
1684 }
1685 }
1686
1687 // dst = c = a * b + c
1688 void MacroAssembler::vfmaf(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len) {
1689 Assembler::vfmadd231ps(c, a, b, vector_len);
1690 if (dst != c) {
1691 vmovdqu(dst, c);
1692 }
1693 }
1694
1695 void MacroAssembler::incrementl(AddressLiteral dst, Register rscratch) {
1696 assert(rscratch != noreg || always_reachable(dst), "missing");
1697
1698 if (reachable(dst)) {
1699 incrementl(as_Address(dst));
1700 } else {
1701 lea(rscratch, dst);
1702 incrementl(Address(rscratch, 0));
1703 }
1704 }
1705
1706 void MacroAssembler::incrementl(ArrayAddress dst, Register rscratch) {
1707 incrementl(as_Address(dst, rscratch));
1708 }
1709
1710 void MacroAssembler::incrementl(Register reg, int value) {
1711 if (value == min_jint) {addl(reg, value) ; return; }
1712 if (value < 0) { decrementl(reg, -value); return; }
1713 if (value == 0) { ; return; }
1714 if (value == 1 && UseIncDec) { incl(reg) ; return; }
1715 /* else */ { addl(reg, value) ; return; }
1716 }
1717
1718 void MacroAssembler::incrementl(Address dst, int value) {
1719 if (value == min_jint) {addl(dst, value) ; return; }
1720 if (value < 0) { decrementl(dst, -value); return; }
1721 if (value == 0) { ; return; }
1722 if (value == 1 && UseIncDec) { incl(dst) ; return; }
1723 /* else */ { addl(dst, value) ; return; }
1724 }
1725
1726 void MacroAssembler::jump(AddressLiteral dst, Register rscratch) {
1727 assert(rscratch != noreg || always_reachable(dst), "missing");
1728 assert(!dst.rspec().reloc()->is_data(), "should not use ExternalAddress for jump");
1729 if (reachable(dst)) {
1730 jmp_literal(dst.target(), dst.rspec());
1731 } else {
1732 lea(rscratch, dst);
1733 jmp(rscratch);
1734 }
1735 }
1736
1737 void MacroAssembler::jump_cc(Condition cc, AddressLiteral dst, Register rscratch) {
1738 assert(rscratch != noreg || always_reachable(dst), "missing");
1739 assert(!dst.rspec().reloc()->is_data(), "should not use ExternalAddress for jump_cc");
1740 if (reachable(dst)) {
1741 InstructionMark im(this);
1742 relocate(dst.reloc());
1743 const int short_size = 2;
1744 const int long_size = 6;
1745 int offs = (intptr_t)dst.target() - ((intptr_t)pc());
1746 if (dst.reloc() == relocInfo::none && is8bit(offs - short_size)) {
1747 // 0111 tttn #8-bit disp
1748 emit_int8(0x70 | cc);
1749 emit_int8((offs - short_size) & 0xFF);
1750 } else {
1751 // 0000 1111 1000 tttn #32-bit disp
1752 emit_int8(0x0F);
1753 emit_int8((unsigned char)(0x80 | cc));
1754 emit_int32(offs - long_size);
1755 }
1756 } else {
1757 #ifdef ASSERT
1758 warning("reversing conditional branch");
1759 #endif /* ASSERT */
1760 Label skip;
1761 jccb(reverse[cc], skip);
1762 lea(rscratch, dst);
1763 Assembler::jmp(rscratch);
1764 bind(skip);
1765 }
1766 }
1767
1768 void MacroAssembler::cmp32_mxcsr_std(Address mxcsr_save, Register tmp, Register rscratch) {
1769 ExternalAddress mxcsr_std(StubRoutines::x86::addr_mxcsr_std());
1770 assert(rscratch != noreg || always_reachable(mxcsr_std), "missing");
1771
1772 stmxcsr(mxcsr_save);
1773 movl(tmp, mxcsr_save);
1774 if (EnableX86ECoreOpts) {
1775 // The mxcsr_std has status bits set for performance on ECore
1776 orl(tmp, 0x003f);
1777 } else {
1778 // Mask out status bits (only check control and mask bits)
1779 andl(tmp, 0xFFC0);
1780 }
1781 cmp32(tmp, mxcsr_std, rscratch);
1782 }
1783
1784 void MacroAssembler::ldmxcsr(AddressLiteral src, Register rscratch) {
1785 assert(rscratch != noreg || always_reachable(src), "missing");
1786
1787 if (reachable(src)) {
1788 Assembler::ldmxcsr(as_Address(src));
1789 } else {
1790 lea(rscratch, src);
1791 Assembler::ldmxcsr(Address(rscratch, 0));
1792 }
1793 }
1794
1795 int MacroAssembler::load_signed_byte(Register dst, Address src) {
1796 int off = offset();
1797 movsbl(dst, src); // movsxb
1798 return off;
1799 }
1800
1801 // Note: load_signed_short used to be called load_signed_word.
1802 // Although the 'w' in x86 opcodes refers to the term "word" in the assembler
1803 // manual, which means 16 bits, that usage is found nowhere in HotSpot code.
1804 // The term "word" in HotSpot means a 32- or 64-bit machine word.
1805 int MacroAssembler::load_signed_short(Register dst, Address src) {
1806 // This is dubious to me since it seems safe to do a signed 16 => 64 bit
1807 // version but this is what 64bit has always done. This seems to imply
1808 // that users are only using 32bits worth.
1809 int off = offset();
1810 movswl(dst, src); // movsxw
1811 return off;
1812 }
1813
1814 int MacroAssembler::load_unsigned_byte(Register dst, Address src) {
1815 // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
1816 // and "3.9 Partial Register Penalties", p. 22).
1817 int off = offset();
1818 movzbl(dst, src); // movzxb
1819 return off;
1820 }
1821
1822 // Note: load_unsigned_short used to be called load_unsigned_word.
1823 int MacroAssembler::load_unsigned_short(Register dst, Address src) {
1824 // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16,
1825 // and "3.9 Partial Register Penalties", p. 22).
1826 int off = offset();
1827 movzwl(dst, src); // movzxw
1828 return off;
1829 }
1830
1831 void MacroAssembler::load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2) {
1832 switch (size_in_bytes) {
1833 case 8: movq(dst, src); break;
1834 case 4: movl(dst, src); break;
1835 case 2: is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break;
1836 case 1: is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break;
1837 default: ShouldNotReachHere();
1838 }
1839 }
1840
1841 void MacroAssembler::store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2) {
1842 switch (size_in_bytes) {
1843 case 8: movq(dst, src); break;
1844 case 4: movl(dst, src); break;
1845 case 2: movw(dst, src); break;
1846 case 1: movb(dst, src); break;
1847 default: ShouldNotReachHere();
1848 }
1849 }
1850
1851 void MacroAssembler::mov32(AddressLiteral dst, Register src, Register rscratch) {
1852 assert(rscratch != noreg || always_reachable(dst), "missing");
1853
1854 if (reachable(dst)) {
1855 movl(as_Address(dst), src);
1856 } else {
1857 lea(rscratch, dst);
1858 movl(Address(rscratch, 0), src);
1859 }
1860 }
1861
1862 void MacroAssembler::mov32(Register dst, AddressLiteral src) {
1863 if (reachable(src)) {
1864 movl(dst, as_Address(src));
1865 } else {
1866 lea(dst, src);
1867 movl(dst, Address(dst, 0));
1868 }
1869 }
1870
1871 // C++ bool manipulation
1872
1873 void MacroAssembler::movbool(Register dst, Address src) {
1874 if(sizeof(bool) == 1)
1875 movb(dst, src);
1876 else if(sizeof(bool) == 2)
1877 movw(dst, src);
1878 else if(sizeof(bool) == 4)
1879 movl(dst, src);
1880 else
1881 // unsupported
1882 ShouldNotReachHere();
1883 }
1884
1885 void MacroAssembler::movbool(Address dst, bool boolconst) {
1886 if(sizeof(bool) == 1)
1887 movb(dst, (int) boolconst);
1888 else if(sizeof(bool) == 2)
1889 movw(dst, (int) boolconst);
1890 else if(sizeof(bool) == 4)
1891 movl(dst, (int) boolconst);
1892 else
1893 // unsupported
1894 ShouldNotReachHere();
1895 }
1896
1897 void MacroAssembler::movbool(Address dst, Register src) {
1898 if(sizeof(bool) == 1)
1899 movb(dst, src);
1900 else if(sizeof(bool) == 2)
1901 movw(dst, src);
1902 else if(sizeof(bool) == 4)
1903 movl(dst, src);
1904 else
1905 // unsupported
1906 ShouldNotReachHere();
1907 }
1908
1909 void MacroAssembler::movdl(XMMRegister dst, AddressLiteral src, Register rscratch) {
1910 assert(rscratch != noreg || always_reachable(src), "missing");
1911
1912 if (reachable(src)) {
1913 movdl(dst, as_Address(src));
1914 } else {
1915 lea(rscratch, src);
1916 movdl(dst, Address(rscratch, 0));
1917 }
1918 }
1919
1920 void MacroAssembler::movq(XMMRegister dst, AddressLiteral src, Register rscratch) {
1921 assert(rscratch != noreg || always_reachable(src), "missing");
1922
1923 if (reachable(src)) {
1924 movq(dst, as_Address(src));
1925 } else {
1926 lea(rscratch, src);
1927 movq(dst, Address(rscratch, 0));
1928 }
1929 }
1930
1931 void MacroAssembler::movdbl(XMMRegister dst, AddressLiteral src, Register rscratch) {
1932 assert(rscratch != noreg || always_reachable(src), "missing");
1933
1934 if (reachable(src)) {
1935 if (UseXmmLoadAndClearUpper) {
1936 movsd (dst, as_Address(src));
1937 } else {
1938 movlpd(dst, as_Address(src));
1939 }
1940 } else {
1941 lea(rscratch, src);
1942 if (UseXmmLoadAndClearUpper) {
1943 movsd (dst, Address(rscratch, 0));
1944 } else {
1945 movlpd(dst, Address(rscratch, 0));
1946 }
1947 }
1948 }
1949
1950 void MacroAssembler::movflt(XMMRegister dst, AddressLiteral src, Register rscratch) {
1951 assert(rscratch != noreg || always_reachable(src), "missing");
1952
1953 if (reachable(src)) {
1954 movss(dst, as_Address(src));
1955 } else {
1956 lea(rscratch, src);
1957 movss(dst, Address(rscratch, 0));
1958 }
1959 }
1960
1961 void MacroAssembler::movptr(Register dst, Register src) {
1962 movq(dst, src);
1963 }
1964
1965 void MacroAssembler::movptr(Register dst, Address src) {
1966 movq(dst, src);
1967 }
1968
1969 // src should NEVER be a real pointer. Use AddressLiteral for true pointers
1970 void MacroAssembler::movptr(Register dst, intptr_t src) {
1971 if (is_uimm32(src)) {
1972 movl(dst, checked_cast<uint32_t>(src));
1973 } else if (is_simm32(src)) {
1974 movq(dst, checked_cast<int32_t>(src));
1975 } else {
1976 mov64(dst, src);
1977 }
1978 }
1979
1980 void MacroAssembler::movptr(Address dst, Register src) {
1981 movq(dst, src);
1982 }
1983
1984 void MacroAssembler::movptr(Address dst, int32_t src) {
1985 movslq(dst, src);
1986 }
1987
1988 void MacroAssembler::movdqu(Address dst, XMMRegister src) {
1989 assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
1990 Assembler::movdqu(dst, src);
1991 }
1992
1993 void MacroAssembler::movdqu(XMMRegister dst, Address src) {
1994 assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
1995 Assembler::movdqu(dst, src);
1996 }
1997
1998 void MacroAssembler::movdqu(XMMRegister dst, XMMRegister src) {
1999 assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
2000 Assembler::movdqu(dst, src);
2001 }
2002
2003 void MacroAssembler::movdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
2004 assert(rscratch != noreg || always_reachable(src), "missing");
2005
2006 if (reachable(src)) {
2007 movdqu(dst, as_Address(src));
2008 } else {
2009 lea(rscratch, src);
2010 movdqu(dst, Address(rscratch, 0));
2011 }
2012 }
2013
2014 void MacroAssembler::vmovdqu(Address dst, XMMRegister src) {
2015 assert(((src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
2016 Assembler::vmovdqu(dst, src);
2017 }
2018
2019 void MacroAssembler::vmovdqu(XMMRegister dst, Address src) {
2020 assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
2021 Assembler::vmovdqu(dst, src);
2022 }
2023
2024 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src) {
2025 assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
2026 Assembler::vmovdqu(dst, src);
2027 }
2028
2029 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, Register rscratch) {
2030 assert(rscratch != noreg || always_reachable(src), "missing");
2031
2032 if (reachable(src)) {
2033 vmovdqu(dst, as_Address(src));
2034 }
2035 else {
2036 lea(rscratch, src);
2037 vmovdqu(dst, Address(rscratch, 0));
2038 }
2039 }
2040
2041 void MacroAssembler::vmovdqu(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
2042 assert(rscratch != noreg || always_reachable(src), "missing");
2043
2044 if (vector_len == AVX_512bit) {
2045 evmovdquq(dst, src, AVX_512bit, rscratch);
2046 } else if (vector_len == AVX_256bit) {
2047 vmovdqu(dst, src, rscratch);
2048 } else {
2049 movdqu(dst, src, rscratch);
2050 }
2051 }
2052
2053 void MacroAssembler::vmovdqu(XMMRegister dst, XMMRegister src, int vector_len) {
2054 if (vector_len == AVX_512bit) {
2055 evmovdquq(dst, src, AVX_512bit);
2056 } else if (vector_len == AVX_256bit) {
2057 vmovdqu(dst, src);
2058 } else {
2059 movdqu(dst, src);
2060 }
2061 }
2062
2063 void MacroAssembler::vmovdqu(Address dst, XMMRegister src, int vector_len) {
2064 if (vector_len == AVX_512bit) {
2065 evmovdquq(dst, src, AVX_512bit);
2066 } else if (vector_len == AVX_256bit) {
2067 vmovdqu(dst, src);
2068 } else {
2069 movdqu(dst, src);
2070 }
2071 }
2072
2073 void MacroAssembler::vmovdqu(XMMRegister dst, Address src, int vector_len) {
2074 if (vector_len == AVX_512bit) {
2075 evmovdquq(dst, src, AVX_512bit);
2076 } else if (vector_len == AVX_256bit) {
2077 vmovdqu(dst, src);
2078 } else {
2079 movdqu(dst, src);
2080 }
2081 }
2082
2083 void MacroAssembler::vmovdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
2084 assert(rscratch != noreg || always_reachable(src), "missing");
2085
2086 if (reachable(src)) {
2087 vmovdqa(dst, as_Address(src));
2088 }
2089 else {
2090 lea(rscratch, src);
2091 vmovdqa(dst, Address(rscratch, 0));
2092 }
2093 }
2094
2095 void MacroAssembler::vmovdqa(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
2096 assert(rscratch != noreg || always_reachable(src), "missing");
2097
2098 if (vector_len == AVX_512bit) {
2099 evmovdqaq(dst, src, AVX_512bit, rscratch);
2100 } else if (vector_len == AVX_256bit) {
2101 vmovdqa(dst, src, rscratch);
2102 } else {
2103 movdqa(dst, src, rscratch);
2104 }
2105 }
2106
2107 void MacroAssembler::kmov(KRegister dst, Address src) {
2108 if (VM_Version::supports_avx512bw()) {
2109 kmovql(dst, src);
2110 } else {
2111 assert(VM_Version::supports_evex(), "");
2112 kmovwl(dst, src);
2113 }
2114 }
2115
2116 void MacroAssembler::kmov(Address dst, KRegister src) {
2117 if (VM_Version::supports_avx512bw()) {
2118 kmovql(dst, src);
2119 } else {
2120 assert(VM_Version::supports_evex(), "");
2121 kmovwl(dst, src);
2122 }
2123 }
2124
2125 void MacroAssembler::kmov(KRegister dst, KRegister src) {
2126 if (VM_Version::supports_avx512bw()) {
2127 kmovql(dst, src);
2128 } else {
2129 assert(VM_Version::supports_evex(), "");
2130 kmovwl(dst, src);
2131 }
2132 }
2133
2134 void MacroAssembler::kmov(Register dst, KRegister src) {
2135 if (VM_Version::supports_avx512bw()) {
2136 kmovql(dst, src);
2137 } else {
2138 assert(VM_Version::supports_evex(), "");
2139 kmovwl(dst, src);
2140 }
2141 }
2142
2143 void MacroAssembler::kmov(KRegister dst, Register src) {
2144 if (VM_Version::supports_avx512bw()) {
2145 kmovql(dst, src);
2146 } else {
2147 assert(VM_Version::supports_evex(), "");
2148 kmovwl(dst, src);
2149 }
2150 }
2151
2152 void MacroAssembler::kmovql(KRegister dst, AddressLiteral src, Register rscratch) {
2153 assert(rscratch != noreg || always_reachable(src), "missing");
2154
2155 if (reachable(src)) {
2156 kmovql(dst, as_Address(src));
2157 } else {
2158 lea(rscratch, src);
2159 kmovql(dst, Address(rscratch, 0));
2160 }
2161 }
2162
2163 void MacroAssembler::kmovwl(KRegister dst, AddressLiteral src, Register rscratch) {
2164 assert(rscratch != noreg || always_reachable(src), "missing");
2165
2166 if (reachable(src)) {
2167 kmovwl(dst, as_Address(src));
2168 } else {
2169 lea(rscratch, src);
2170 kmovwl(dst, Address(rscratch, 0));
2171 }
2172 }
2173
2174 void MacroAssembler::evmovdqub(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
2175 int vector_len, Register rscratch) {
2176 assert(rscratch != noreg || always_reachable(src), "missing");
2177
2178 if (reachable(src)) {
2179 Assembler::evmovdqub(dst, mask, as_Address(src), merge, vector_len);
2180 } else {
2181 lea(rscratch, src);
2182 Assembler::evmovdqub(dst, mask, Address(rscratch, 0), merge, vector_len);
2183 }
2184 }
2185
2186 void MacroAssembler::evmovdquw(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
2187 int vector_len, Register rscratch) {
2188 assert(rscratch != noreg || always_reachable(src), "missing");
2189
2190 if (reachable(src)) {
2191 Assembler::evmovdquw(dst, mask, as_Address(src), merge, vector_len);
2192 } else {
2193 lea(rscratch, src);
2194 Assembler::evmovdquw(dst, mask, Address(rscratch, 0), merge, vector_len);
2195 }
2196 }
2197
2198 void MacroAssembler::evmovdqul(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
2199 assert(rscratch != noreg || always_reachable(src), "missing");
2200
2201 if (reachable(src)) {
2202 Assembler::evmovdqul(dst, mask, as_Address(src), merge, vector_len);
2203 } else {
2204 lea(rscratch, src);
2205 Assembler::evmovdqul(dst, mask, Address(rscratch, 0), merge, vector_len);
2206 }
2207 }
2208
2209 void MacroAssembler::evmovdquq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
2210 assert(rscratch != noreg || always_reachable(src), "missing");
2211
2212 if (reachable(src)) {
2213 Assembler::evmovdquq(dst, mask, as_Address(src), merge, vector_len);
2214 } else {
2215 lea(rscratch, src);
2216 Assembler::evmovdquq(dst, mask, Address(rscratch, 0), merge, vector_len);
2217 }
2218 }
2219
2220 void MacroAssembler::evmovdquq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
2221 assert(rscratch != noreg || always_reachable(src), "missing");
2222
2223 if (reachable(src)) {
2224 Assembler::evmovdquq(dst, as_Address(src), vector_len);
2225 } else {
2226 lea(rscratch, src);
2227 Assembler::evmovdquq(dst, Address(rscratch, 0), vector_len);
2228 }
2229 }
2230
2231 void MacroAssembler::evmovdqaq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
2232 assert(rscratch != noreg || always_reachable(src), "missing");
2233
2234 if (reachable(src)) {
2235 Assembler::evmovdqaq(dst, mask, as_Address(src), merge, vector_len);
2236 } else {
2237 lea(rscratch, src);
2238 Assembler::evmovdqaq(dst, mask, Address(rscratch, 0), merge, vector_len);
2239 }
2240 }
2241
2242 void MacroAssembler::evmovdqaq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
2243 assert(rscratch != noreg || always_reachable(src), "missing");
2244
2245 if (reachable(src)) {
2246 Assembler::evmovdqaq(dst, as_Address(src), vector_len);
2247 } else {
2248 lea(rscratch, src);
2249 Assembler::evmovdqaq(dst, Address(rscratch, 0), vector_len);
2250 }
2251 }
2252
2253 void MacroAssembler::movapd(XMMRegister dst, AddressLiteral src, Register rscratch) {
2254 assert(rscratch != noreg || always_reachable(src), "missing");
2255
2256 if (reachable(src)) {
2257 Assembler::movapd(dst, as_Address(src));
2258 } else {
2259 lea(rscratch, src);
2260 Assembler::movapd(dst, Address(rscratch, 0));
2261 }
2262 }
2263
2264 void MacroAssembler::movdqa(XMMRegister dst, AddressLiteral src, Register rscratch) {
2265 assert(rscratch != noreg || always_reachable(src), "missing");
2266
2267 if (reachable(src)) {
2268 Assembler::movdqa(dst, as_Address(src));
2269 } else {
2270 lea(rscratch, src);
2271 Assembler::movdqa(dst, Address(rscratch, 0));
2272 }
2273 }
2274
2275 void MacroAssembler::movsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
2276 assert(rscratch != noreg || always_reachable(src), "missing");
2277
2278 if (reachable(src)) {
2279 Assembler::movsd(dst, as_Address(src));
2280 } else {
2281 lea(rscratch, src);
2282 Assembler::movsd(dst, Address(rscratch, 0));
2283 }
2284 }
2285
2286 void MacroAssembler::movss(XMMRegister dst, AddressLiteral src, Register rscratch) {
2287 assert(rscratch != noreg || always_reachable(src), "missing");
2288
2289 if (reachable(src)) {
2290 Assembler::movss(dst, as_Address(src));
2291 } else {
2292 lea(rscratch, src);
2293 Assembler::movss(dst, Address(rscratch, 0));
2294 }
2295 }
2296
2297 void MacroAssembler::movddup(XMMRegister dst, AddressLiteral src, Register rscratch) {
2298 assert(rscratch != noreg || always_reachable(src), "missing");
2299
2300 if (reachable(src)) {
2301 Assembler::movddup(dst, as_Address(src));
2302 } else {
2303 lea(rscratch, src);
2304 Assembler::movddup(dst, Address(rscratch, 0));
2305 }
2306 }
2307
2308 void MacroAssembler::vmovddup(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
2309 assert(rscratch != noreg || always_reachable(src), "missing");
2310
2311 if (reachable(src)) {
2312 Assembler::vmovddup(dst, as_Address(src), vector_len);
2313 } else {
2314 lea(rscratch, src);
2315 Assembler::vmovddup(dst, Address(rscratch, 0), vector_len);
2316 }
2317 }
2318
2319 void MacroAssembler::mulsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
2320 assert(rscratch != noreg || always_reachable(src), "missing");
2321
2322 if (reachable(src)) {
2323 Assembler::mulsd(dst, as_Address(src));
2324 } else {
2325 lea(rscratch, src);
2326 Assembler::mulsd(dst, Address(rscratch, 0));
2327 }
2328 }
2329
2330 void MacroAssembler::mulss(XMMRegister dst, AddressLiteral src, Register rscratch) {
2331 assert(rscratch != noreg || always_reachable(src), "missing");
2332
2333 if (reachable(src)) {
2334 Assembler::mulss(dst, as_Address(src));
2335 } else {
2336 lea(rscratch, src);
2337 Assembler::mulss(dst, Address(rscratch, 0));
2338 }
2339 }
2340
2341 void MacroAssembler::null_check(Register reg, int offset) {
2342 if (needs_explicit_null_check(offset)) {
2343 // provoke OS null exception if reg is null by
2344 // accessing M[reg] w/o changing any (non-CC) registers
2345 // NOTE: cmpl is plenty here to provoke a segv
2346 cmpptr(rax, Address(reg, 0));
2347 // Note: should probably use testl(rax, Address(reg, 0));
2348 // may be shorter code (however, this version of
2349 // testl needs to be implemented first)
2350 } else {
2351 // nothing to do, (later) access of M[reg + offset]
2352 // will provoke OS null exception if reg is null
2353 }
2354 }
2355
2356 void MacroAssembler::os_breakpoint() {
2357 // instead of directly emitting a breakpoint, call os:breakpoint for better debugability
2358 // (e.g., MSVC can't call ps() otherwise)
2359 call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint)));
2360 }
2361
2362 void MacroAssembler::unimplemented(const char* what) {
2363 const char* buf = nullptr;
2364 {
2365 ResourceMark rm;
2366 stringStream ss;
2367 ss.print("unimplemented: %s", what);
2368 buf = code_string(ss.as_string());
2369 }
2370 stop(buf);
2371 }
2372
2373 #define XSTATE_BV 0x200
2374
2375 void MacroAssembler::pop_CPU_state() {
2376 pop_FPU_state();
2377 pop_IU_state();
2378 }
2379
2380 void MacroAssembler::pop_FPU_state() {
2381 fxrstor(Address(rsp, 0));
2382 addptr(rsp, FPUStateSizeInWords * wordSize);
2383 }
2384
2385 void MacroAssembler::pop_IU_state() {
2386 popa();
2387 addq(rsp, 8);
2388 popf();
2389 }
2390
2391 // Save Integer and Float state
2392 // Warning: Stack must be 16 byte aligned (64bit)
2393 void MacroAssembler::push_CPU_state() {
2394 push_IU_state();
2395 push_FPU_state();
2396 }
2397
2398 void MacroAssembler::push_FPU_state() {
2399 subptr(rsp, FPUStateSizeInWords * wordSize);
2400 fxsave(Address(rsp, 0));
2401 }
2402
2403 void MacroAssembler::push_IU_state() {
2404 // Push flags first because pusha kills them
2405 pushf();
2406 // Make sure rsp stays 16-byte aligned
2407 subq(rsp, 8);
2408 pusha();
2409 }
2410
2411 void MacroAssembler::push_cont_fastpath() {
2412 if (!Continuations::enabled()) return;
2413
2414 Label L_done;
2415 cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
2416 jccb(Assembler::belowEqual, L_done);
2417 movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), rsp);
2418 bind(L_done);
2419 }
2420
2421 void MacroAssembler::pop_cont_fastpath() {
2422 if (!Continuations::enabled()) return;
2423
2424 Label L_done;
2425 cmpptr(rsp, Address(r15_thread, JavaThread::cont_fastpath_offset()));
2426 jccb(Assembler::below, L_done);
2427 movptr(Address(r15_thread, JavaThread::cont_fastpath_offset()), 0);
2428 bind(L_done);
2429 }
2430
2431 void MacroAssembler::inc_held_monitor_count() {
2432 incrementq(Address(r15_thread, JavaThread::held_monitor_count_offset()));
2433 }
2434
2435 void MacroAssembler::dec_held_monitor_count() {
2436 decrementq(Address(r15_thread, JavaThread::held_monitor_count_offset()));
2437 }
2438
2439 #ifdef ASSERT
2440 void MacroAssembler::stop_if_in_cont(Register cont, const char* name) {
2441 Label no_cont;
2442 movptr(cont, Address(r15_thread, JavaThread::cont_entry_offset()));
2443 testl(cont, cont);
2444 jcc(Assembler::zero, no_cont);
2445 stop(name);
2446 bind(no_cont);
2447 }
2448 #endif
2449
2450 void MacroAssembler::reset_last_Java_frame(bool clear_fp) { // determine java_thread register
2451 // we must set sp to zero to clear frame
2452 movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), NULL_WORD);
2453 // must clear fp, so that compiled frames are not confused; it is
2454 // possible that we need it only for debugging
2455 if (clear_fp) {
2456 movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), NULL_WORD);
2457 }
2458 // Always clear the pc because it could have been set by make_walkable()
2459 movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), NULL_WORD);
2460 vzeroupper();
2461 }
2462
2463 void MacroAssembler::round_to(Register reg, int modulus) {
2464 addptr(reg, modulus - 1);
2465 andptr(reg, -modulus);
2466 }
2467
2468 void MacroAssembler::safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod) {
2469 if (at_return) {
2470 // Note that when in_nmethod is set, the stack pointer is incremented before the poll. Therefore,
2471 // we may safely use rsp instead to perform the stack watermark check.
2472 cmpptr(in_nmethod ? rsp : rbp, Address(r15_thread, JavaThread::polling_word_offset()));
2473 jcc(Assembler::above, slow_path);
2474 return;
2475 }
2476 testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
2477 jcc(Assembler::notZero, slow_path); // handshake bit set implies poll
2478 }
2479
2480 // Calls to C land
2481 //
2482 // When entering C land, the rbp, & rsp of the last Java frame have to be recorded
2483 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp
2484 // has to be reset to 0. This is required to allow proper stack traversal.
2485 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
2486 Register last_java_fp,
2487 address last_java_pc,
2488 Register rscratch) {
2489 vzeroupper();
2490 // determine last_java_sp register
2491 if (!last_java_sp->is_valid()) {
2492 last_java_sp = rsp;
2493 }
2494 // last_java_fp is optional
2495 if (last_java_fp->is_valid()) {
2496 movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), last_java_fp);
2497 }
2498 // last_java_pc is optional
2499 if (last_java_pc != nullptr) {
2500 Address java_pc(r15_thread,
2501 JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset());
2502 lea(java_pc, InternalAddress(last_java_pc), rscratch);
2503 }
2504 movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), last_java_sp);
2505 }
2506
2507 void MacroAssembler::set_last_Java_frame(Register last_java_sp,
2508 Register last_java_fp,
2509 Label &L,
2510 Register scratch) {
2511 lea(scratch, L);
2512 movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), scratch);
2513 set_last_Java_frame(last_java_sp, last_java_fp, nullptr, scratch);
2514 }
2515
2516 void MacroAssembler::shlptr(Register dst, int imm8) {
2517 shlq(dst, imm8);
2518 }
2519
2520 void MacroAssembler::shrptr(Register dst, int imm8) {
2521 shrq(dst, imm8);
2522 }
2523
2524 void MacroAssembler::sign_extend_byte(Register reg) {
2525 movsbl(reg, reg); // movsxb
2526 }
2527
2528 void MacroAssembler::sign_extend_short(Register reg) {
2529 movswl(reg, reg); // movsxw
2530 }
2531
2532 void MacroAssembler::testl(Address dst, int32_t imm32) {
2533 if (imm32 >= 0 && is8bit(imm32)) {
2534 testb(dst, imm32);
2535 } else {
2536 Assembler::testl(dst, imm32);
2537 }
2538 }
2539
2540 void MacroAssembler::testl(Register dst, int32_t imm32) {
2541 if (imm32 >= 0 && is8bit(imm32) && dst->has_byte_register()) {
2542 testb(dst, imm32);
2543 } else {
2544 Assembler::testl(dst, imm32);
2545 }
2546 }
2547
2548 void MacroAssembler::testl(Register dst, AddressLiteral src) {
2549 assert(always_reachable(src), "Address should be reachable");
2550 testl(dst, as_Address(src));
2551 }
2552
2553 void MacroAssembler::testq(Address dst, int32_t imm32) {
2554 if (imm32 >= 0) {
2555 testl(dst, imm32);
2556 } else {
2557 Assembler::testq(dst, imm32);
2558 }
2559 }
2560
2561 void MacroAssembler::testq(Register dst, int32_t imm32) {
2562 if (imm32 >= 0) {
2563 testl(dst, imm32);
2564 } else {
2565 Assembler::testq(dst, imm32);
2566 }
2567 }
2568
2569 void MacroAssembler::pcmpeqb(XMMRegister dst, XMMRegister src) {
2570 assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2571 Assembler::pcmpeqb(dst, src);
2572 }
2573
2574 void MacroAssembler::pcmpeqw(XMMRegister dst, XMMRegister src) {
2575 assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2576 Assembler::pcmpeqw(dst, src);
2577 }
2578
2579 void MacroAssembler::pcmpestri(XMMRegister dst, Address src, int imm8) {
2580 assert((dst->encoding() < 16),"XMM register should be 0-15");
2581 Assembler::pcmpestri(dst, src, imm8);
2582 }
2583
2584 void MacroAssembler::pcmpestri(XMMRegister dst, XMMRegister src, int imm8) {
2585 assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
2586 Assembler::pcmpestri(dst, src, imm8);
2587 }
2588
2589 void MacroAssembler::pmovzxbw(XMMRegister dst, XMMRegister src) {
2590 assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2591 Assembler::pmovzxbw(dst, src);
2592 }
2593
2594 void MacroAssembler::pmovzxbw(XMMRegister dst, Address src) {
2595 assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2596 Assembler::pmovzxbw(dst, src);
2597 }
2598
2599 void MacroAssembler::pmovmskb(Register dst, XMMRegister src) {
2600 assert((src->encoding() < 16),"XMM register should be 0-15");
2601 Assembler::pmovmskb(dst, src);
2602 }
2603
2604 void MacroAssembler::ptest(XMMRegister dst, XMMRegister src) {
2605 assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
2606 Assembler::ptest(dst, src);
2607 }
2608
2609 void MacroAssembler::sqrtss(XMMRegister dst, AddressLiteral src, Register rscratch) {
2610 assert(rscratch != noreg || always_reachable(src), "missing");
2611
2612 if (reachable(src)) {
2613 Assembler::sqrtss(dst, as_Address(src));
2614 } else {
2615 lea(rscratch, src);
2616 Assembler::sqrtss(dst, Address(rscratch, 0));
2617 }
2618 }
2619
2620 void MacroAssembler::subsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
2621 assert(rscratch != noreg || always_reachable(src), "missing");
2622
2623 if (reachable(src)) {
2624 Assembler::subsd(dst, as_Address(src));
2625 } else {
2626 lea(rscratch, src);
2627 Assembler::subsd(dst, Address(rscratch, 0));
2628 }
2629 }
2630
2631 void MacroAssembler::roundsd(XMMRegister dst, AddressLiteral src, int32_t rmode, Register rscratch) {
2632 assert(rscratch != noreg || always_reachable(src), "missing");
2633
2634 if (reachable(src)) {
2635 Assembler::roundsd(dst, as_Address(src), rmode);
2636 } else {
2637 lea(rscratch, src);
2638 Assembler::roundsd(dst, Address(rscratch, 0), rmode);
2639 }
2640 }
2641
2642 void MacroAssembler::subss(XMMRegister dst, AddressLiteral src, Register rscratch) {
2643 assert(rscratch != noreg || always_reachable(src), "missing");
2644
2645 if (reachable(src)) {
2646 Assembler::subss(dst, as_Address(src));
2647 } else {
2648 lea(rscratch, src);
2649 Assembler::subss(dst, Address(rscratch, 0));
2650 }
2651 }
2652
2653 void MacroAssembler::ucomisd(XMMRegister dst, AddressLiteral src, Register rscratch) {
2654 assert(rscratch != noreg || always_reachable(src), "missing");
2655
2656 if (reachable(src)) {
2657 Assembler::ucomisd(dst, as_Address(src));
2658 } else {
2659 lea(rscratch, src);
2660 Assembler::ucomisd(dst, Address(rscratch, 0));
2661 }
2662 }
2663
2664 void MacroAssembler::ucomiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
2665 assert(rscratch != noreg || always_reachable(src), "missing");
2666
2667 if (reachable(src)) {
2668 Assembler::ucomiss(dst, as_Address(src));
2669 } else {
2670 lea(rscratch, src);
2671 Assembler::ucomiss(dst, Address(rscratch, 0));
2672 }
2673 }
2674
2675 void MacroAssembler::xorpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
2676 assert(rscratch != noreg || always_reachable(src), "missing");
2677
2678 // Used in sign-bit flipping with aligned address.
2679 assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
2680
2681 if (UseAVX > 2 &&
2682 (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
2683 (dst->encoding() >= 16)) {
2684 vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
2685 } else if (reachable(src)) {
2686 Assembler::xorpd(dst, as_Address(src));
2687 } else {
2688 lea(rscratch, src);
2689 Assembler::xorpd(dst, Address(rscratch, 0));
2690 }
2691 }
2692
2693 void MacroAssembler::xorpd(XMMRegister dst, XMMRegister src) {
2694 if (UseAVX > 2 &&
2695 (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
2696 ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
2697 Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
2698 } else {
2699 Assembler::xorpd(dst, src);
2700 }
2701 }
2702
2703 void MacroAssembler::xorps(XMMRegister dst, XMMRegister src) {
2704 if (UseAVX > 2 &&
2705 (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
2706 ((dst->encoding() >= 16) || (src->encoding() >= 16))) {
2707 Assembler::vpxor(dst, dst, src, Assembler::AVX_512bit);
2708 } else {
2709 Assembler::xorps(dst, src);
2710 }
2711 }
2712
2713 void MacroAssembler::xorps(XMMRegister dst, AddressLiteral src, Register rscratch) {
2714 assert(rscratch != noreg || always_reachable(src), "missing");
2715
2716 // Used in sign-bit flipping with aligned address.
2717 assert((UseAVX > 0) || (((intptr_t)src.target() & 15) == 0), "SSE mode requires address alignment 16 bytes");
2718
2719 if (UseAVX > 2 &&
2720 (!VM_Version::supports_avx512dq() || !VM_Version::supports_avx512vl()) &&
2721 (dst->encoding() >= 16)) {
2722 vpxor(dst, dst, src, Assembler::AVX_512bit, rscratch);
2723 } else if (reachable(src)) {
2724 Assembler::xorps(dst, as_Address(src));
2725 } else {
2726 lea(rscratch, src);
2727 Assembler::xorps(dst, Address(rscratch, 0));
2728 }
2729 }
2730
2731 void MacroAssembler::pshufb(XMMRegister dst, AddressLiteral src, Register rscratch) {
2732 assert(rscratch != noreg || always_reachable(src), "missing");
2733
2734 // Used in sign-bit flipping with aligned address.
2735 bool aligned_adr = (((intptr_t)src.target() & 15) == 0);
2736 assert((UseAVX > 0) || aligned_adr, "SSE mode requires address alignment 16 bytes");
2737 if (reachable(src)) {
2738 Assembler::pshufb(dst, as_Address(src));
2739 } else {
2740 lea(rscratch, src);
2741 Assembler::pshufb(dst, Address(rscratch, 0));
2742 }
2743 }
2744
2745 // AVX 3-operands instructions
2746
2747 void MacroAssembler::vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
2748 assert(rscratch != noreg || always_reachable(src), "missing");
2749
2750 if (reachable(src)) {
2751 vaddsd(dst, nds, as_Address(src));
2752 } else {
2753 lea(rscratch, src);
2754 vaddsd(dst, nds, Address(rscratch, 0));
2755 }
2756 }
2757
2758 void MacroAssembler::vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
2759 assert(rscratch != noreg || always_reachable(src), "missing");
2760
2761 if (reachable(src)) {
2762 vaddss(dst, nds, as_Address(src));
2763 } else {
2764 lea(rscratch, src);
2765 vaddss(dst, nds, Address(rscratch, 0));
2766 }
2767 }
2768
2769 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
2770 assert(UseAVX > 0, "requires some form of AVX");
2771 assert(rscratch != noreg || always_reachable(src), "missing");
2772
2773 if (reachable(src)) {
2774 Assembler::vpaddb(dst, nds, as_Address(src), vector_len);
2775 } else {
2776 lea(rscratch, src);
2777 Assembler::vpaddb(dst, nds, Address(rscratch, 0), vector_len);
2778 }
2779 }
2780
2781 void MacroAssembler::vpaddd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
2782 assert(UseAVX > 0, "requires some form of AVX");
2783 assert(rscratch != noreg || always_reachable(src), "missing");
2784
2785 if (reachable(src)) {
2786 Assembler::vpaddd(dst, nds, as_Address(src), vector_len);
2787 } else {
2788 lea(rscratch, src);
2789 Assembler::vpaddd(dst, nds, Address(rscratch, 0), vector_len);
2790 }
2791 }
2792
2793 void MacroAssembler::vabsss(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
2794 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
2795 assert(rscratch != noreg || always_reachable(negate_field), "missing");
2796
2797 vandps(dst, nds, negate_field, vector_len, rscratch);
2798 }
2799
2800 void MacroAssembler::vabssd(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch) {
2801 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
2802 assert(rscratch != noreg || always_reachable(negate_field), "missing");
2803
2804 vandpd(dst, nds, negate_field, vector_len, rscratch);
2805 }
2806
2807 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
2808 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2809 Assembler::vpaddb(dst, nds, src, vector_len);
2810 }
2811
2812 void MacroAssembler::vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
2813 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2814 Assembler::vpaddb(dst, nds, src, vector_len);
2815 }
2816
2817 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
2818 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2819 Assembler::vpaddw(dst, nds, src, vector_len);
2820 }
2821
2822 void MacroAssembler::vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
2823 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2824 Assembler::vpaddw(dst, nds, src, vector_len);
2825 }
2826
2827 void MacroAssembler::vpand(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
2828 assert(rscratch != noreg || always_reachable(src), "missing");
2829
2830 if (reachable(src)) {
2831 Assembler::vpand(dst, nds, as_Address(src), vector_len);
2832 } else {
2833 lea(rscratch, src);
2834 Assembler::vpand(dst, nds, Address(rscratch, 0), vector_len);
2835 }
2836 }
2837
2838 void MacroAssembler::vpbroadcastd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
2839 assert(rscratch != noreg || always_reachable(src), "missing");
2840
2841 if (reachable(src)) {
2842 Assembler::vpbroadcastd(dst, as_Address(src), vector_len);
2843 } else {
2844 lea(rscratch, src);
2845 Assembler::vpbroadcastd(dst, Address(rscratch, 0), vector_len);
2846 }
2847 }
2848
2849 void MacroAssembler::vbroadcasti128(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
2850 assert(rscratch != noreg || always_reachable(src), "missing");
2851
2852 if (reachable(src)) {
2853 Assembler::vbroadcasti128(dst, as_Address(src), vector_len);
2854 } else {
2855 lea(rscratch, src);
2856 Assembler::vbroadcasti128(dst, Address(rscratch, 0), vector_len);
2857 }
2858 }
2859
2860 void MacroAssembler::vpbroadcastq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
2861 assert(rscratch != noreg || always_reachable(src), "missing");
2862
2863 if (reachable(src)) {
2864 Assembler::vpbroadcastq(dst, as_Address(src), vector_len);
2865 } else {
2866 lea(rscratch, src);
2867 Assembler::vpbroadcastq(dst, Address(rscratch, 0), vector_len);
2868 }
2869 }
2870
2871 void MacroAssembler::vbroadcastsd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
2872 assert(rscratch != noreg || always_reachable(src), "missing");
2873
2874 if (reachable(src)) {
2875 Assembler::vbroadcastsd(dst, as_Address(src), vector_len);
2876 } else {
2877 lea(rscratch, src);
2878 Assembler::vbroadcastsd(dst, Address(rscratch, 0), vector_len);
2879 }
2880 }
2881
2882 void MacroAssembler::vbroadcastss(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch) {
2883 assert(rscratch != noreg || always_reachable(src), "missing");
2884
2885 if (reachable(src)) {
2886 Assembler::vbroadcastss(dst, as_Address(src), vector_len);
2887 } else {
2888 lea(rscratch, src);
2889 Assembler::vbroadcastss(dst, Address(rscratch, 0), vector_len);
2890 }
2891 }
2892
2893 // Vector float blend
2894 // vblendvps(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
2895 void MacroAssembler::vblendvps(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
2896 // WARN: Allow dst == (src1|src2), mask == scratch
2897 bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1;
2898 bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst;
2899 bool dst_available = dst != mask && (dst != src1 || dst != src2);
2900 if (blend_emulation && scratch_available && dst_available) {
2901 if (compute_mask) {
2902 vpsrad(scratch, mask, 32, vector_len);
2903 mask = scratch;
2904 }
2905 if (dst == src1) {
2906 vpandn(dst, mask, src1, vector_len); // if mask == 0, src1
2907 vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
2908 } else {
2909 vpand (dst, mask, src2, vector_len); // if mask == 1, src2
2910 vpandn(scratch, mask, src1, vector_len); // if mask == 0, src1
2911 }
2912 vpor(dst, dst, scratch, vector_len);
2913 } else {
2914 Assembler::vblendvps(dst, src1, src2, mask, vector_len);
2915 }
2916 }
2917
2918 // vblendvpd(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg)
2919 void MacroAssembler::vblendvpd(XMMRegister dst, XMMRegister src1, XMMRegister src2, XMMRegister mask, int vector_len, bool compute_mask, XMMRegister scratch) {
2920 // WARN: Allow dst == (src1|src2), mask == scratch
2921 bool blend_emulation = EnableX86ECoreOpts && UseAVX > 1;
2922 bool scratch_available = scratch != xnoreg && scratch != src1 && scratch != src2 && scratch != dst && (!compute_mask || scratch != mask);
2923 bool dst_available = dst != mask && (dst != src1 || dst != src2);
2924 if (blend_emulation && scratch_available && dst_available) {
2925 if (compute_mask) {
2926 vpxor(scratch, scratch, scratch, vector_len);
2927 vpcmpgtq(scratch, scratch, mask, vector_len);
2928 mask = scratch;
2929 }
2930 if (dst == src1) {
2931 vpandn(dst, mask, src1, vector_len); // if mask == 0, src
2932 vpand (scratch, mask, src2, vector_len); // if mask == 1, src2
2933 } else {
2934 vpand (dst, mask, src2, vector_len); // if mask == 1, src2
2935 vpandn(scratch, mask, src1, vector_len); // if mask == 0, src
2936 }
2937 vpor(dst, dst, scratch, vector_len);
2938 } else {
2939 Assembler::vblendvpd(dst, src1, src2, mask, vector_len);
2940 }
2941 }
2942
2943 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
2944 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2945 Assembler::vpcmpeqb(dst, nds, src, vector_len);
2946 }
2947
2948 void MacroAssembler::vpcmpeqb(XMMRegister dst, XMMRegister src1, Address src2, int vector_len) {
2949 assert(((dst->encoding() < 16 && src1->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2950 Assembler::vpcmpeqb(dst, src1, src2, vector_len);
2951 }
2952
2953 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
2954 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2955 Assembler::vpcmpeqw(dst, nds, src, vector_len);
2956 }
2957
2958 void MacroAssembler::vpcmpeqw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
2959 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
2960 Assembler::vpcmpeqw(dst, nds, src, vector_len);
2961 }
2962
2963 void MacroAssembler::evpcmpeqd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
2964 assert(rscratch != noreg || always_reachable(src), "missing");
2965
2966 if (reachable(src)) {
2967 Assembler::evpcmpeqd(kdst, mask, nds, as_Address(src), vector_len);
2968 } else {
2969 lea(rscratch, src);
2970 Assembler::evpcmpeqd(kdst, mask, nds, Address(rscratch, 0), vector_len);
2971 }
2972 }
2973
2974 void MacroAssembler::evpcmpd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
2975 int comparison, bool is_signed, int vector_len, Register rscratch) {
2976 assert(rscratch != noreg || always_reachable(src), "missing");
2977
2978 if (reachable(src)) {
2979 Assembler::evpcmpd(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
2980 } else {
2981 lea(rscratch, src);
2982 Assembler::evpcmpd(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
2983 }
2984 }
2985
2986 void MacroAssembler::evpcmpq(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
2987 int comparison, bool is_signed, int vector_len, Register rscratch) {
2988 assert(rscratch != noreg || always_reachable(src), "missing");
2989
2990 if (reachable(src)) {
2991 Assembler::evpcmpq(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
2992 } else {
2993 lea(rscratch, src);
2994 Assembler::evpcmpq(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
2995 }
2996 }
2997
2998 void MacroAssembler::evpcmpb(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
2999 int comparison, bool is_signed, int vector_len, Register rscratch) {
3000 assert(rscratch != noreg || always_reachable(src), "missing");
3001
3002 if (reachable(src)) {
3003 Assembler::evpcmpb(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
3004 } else {
3005 lea(rscratch, src);
3006 Assembler::evpcmpb(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
3007 }
3008 }
3009
3010 void MacroAssembler::evpcmpw(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
3011 int comparison, bool is_signed, int vector_len, Register rscratch) {
3012 assert(rscratch != noreg || always_reachable(src), "missing");
3013
3014 if (reachable(src)) {
3015 Assembler::evpcmpw(kdst, mask, nds, as_Address(src), comparison, is_signed, vector_len);
3016 } else {
3017 lea(rscratch, src);
3018 Assembler::evpcmpw(kdst, mask, nds, Address(rscratch, 0), comparison, is_signed, vector_len);
3019 }
3020 }
3021
3022 void MacroAssembler::vpcmpCC(XMMRegister dst, XMMRegister nds, XMMRegister src, int cond_encoding, Width width, int vector_len) {
3023 if (width == Assembler::Q) {
3024 Assembler::vpcmpCCq(dst, nds, src, cond_encoding, vector_len);
3025 } else {
3026 Assembler::vpcmpCCbwd(dst, nds, src, cond_encoding, vector_len);
3027 }
3028 }
3029
3030 void MacroAssembler::vpcmpCCW(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister xtmp, ComparisonPredicate cond, Width width, int vector_len) {
3031 int eq_cond_enc = 0x29;
3032 int gt_cond_enc = 0x37;
3033 if (width != Assembler::Q) {
3034 eq_cond_enc = 0x74 + width;
3035 gt_cond_enc = 0x64 + width;
3036 }
3037 switch (cond) {
3038 case eq:
3039 vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
3040 break;
3041 case neq:
3042 vpcmpCC(dst, nds, src, eq_cond_enc, width, vector_len);
3043 vallones(xtmp, vector_len);
3044 vpxor(dst, xtmp, dst, vector_len);
3045 break;
3046 case le:
3047 vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
3048 vallones(xtmp, vector_len);
3049 vpxor(dst, xtmp, dst, vector_len);
3050 break;
3051 case nlt:
3052 vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
3053 vallones(xtmp, vector_len);
3054 vpxor(dst, xtmp, dst, vector_len);
3055 break;
3056 case lt:
3057 vpcmpCC(dst, src, nds, gt_cond_enc, width, vector_len);
3058 break;
3059 case nle:
3060 vpcmpCC(dst, nds, src, gt_cond_enc, width, vector_len);
3061 break;
3062 default:
3063 assert(false, "Should not reach here");
3064 }
3065 }
3066
3067 void MacroAssembler::vpmovzxbw(XMMRegister dst, Address src, int vector_len) {
3068 assert(((dst->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3069 Assembler::vpmovzxbw(dst, src, vector_len);
3070 }
3071
3072 void MacroAssembler::vpmovmskb(Register dst, XMMRegister src, int vector_len) {
3073 assert((src->encoding() < 16),"XMM register should be 0-15");
3074 Assembler::vpmovmskb(dst, src, vector_len);
3075 }
3076
3077 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
3078 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3079 Assembler::vpmullw(dst, nds, src, vector_len);
3080 }
3081
3082 void MacroAssembler::vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
3083 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3084 Assembler::vpmullw(dst, nds, src, vector_len);
3085 }
3086
3087 void MacroAssembler::vpmulld(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3088 assert((UseAVX > 0), "AVX support is needed");
3089 assert(rscratch != noreg || always_reachable(src), "missing");
3090
3091 if (reachable(src)) {
3092 Assembler::vpmulld(dst, nds, as_Address(src), vector_len);
3093 } else {
3094 lea(rscratch, src);
3095 Assembler::vpmulld(dst, nds, Address(rscratch, 0), vector_len);
3096 }
3097 }
3098
3099 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
3100 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3101 Assembler::vpsubb(dst, nds, src, vector_len);
3102 }
3103
3104 void MacroAssembler::vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
3105 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3106 Assembler::vpsubb(dst, nds, src, vector_len);
3107 }
3108
3109 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
3110 assert(((dst->encoding() < 16 && src->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3111 Assembler::vpsubw(dst, nds, src, vector_len);
3112 }
3113
3114 void MacroAssembler::vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
3115 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3116 Assembler::vpsubw(dst, nds, src, vector_len);
3117 }
3118
3119 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
3120 assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3121 Assembler::vpsraw(dst, nds, shift, vector_len);
3122 }
3123
3124 void MacroAssembler::vpsraw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
3125 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3126 Assembler::vpsraw(dst, nds, shift, vector_len);
3127 }
3128
3129 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
3130 assert(UseAVX > 2,"");
3131 if (!VM_Version::supports_avx512vl() && vector_len < 2) {
3132 vector_len = 2;
3133 }
3134 Assembler::evpsraq(dst, nds, shift, vector_len);
3135 }
3136
3137 void MacroAssembler::evpsraq(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
3138 assert(UseAVX > 2,"");
3139 if (!VM_Version::supports_avx512vl() && vector_len < 2) {
3140 vector_len = 2;
3141 }
3142 Assembler::evpsraq(dst, nds, shift, vector_len);
3143 }
3144
3145 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
3146 assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3147 Assembler::vpsrlw(dst, nds, shift, vector_len);
3148 }
3149
3150 void MacroAssembler::vpsrlw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
3151 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3152 Assembler::vpsrlw(dst, nds, shift, vector_len);
3153 }
3154
3155 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len) {
3156 assert(((dst->encoding() < 16 && shift->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3157 Assembler::vpsllw(dst, nds, shift, vector_len);
3158 }
3159
3160 void MacroAssembler::vpsllw(XMMRegister dst, XMMRegister nds, int shift, int vector_len) {
3161 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3162 Assembler::vpsllw(dst, nds, shift, vector_len);
3163 }
3164
3165 void MacroAssembler::vptest(XMMRegister dst, XMMRegister src) {
3166 assert((dst->encoding() < 16 && src->encoding() < 16),"XMM register should be 0-15");
3167 Assembler::vptest(dst, src);
3168 }
3169
3170 void MacroAssembler::punpcklbw(XMMRegister dst, XMMRegister src) {
3171 assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3172 Assembler::punpcklbw(dst, src);
3173 }
3174
3175 void MacroAssembler::pshufd(XMMRegister dst, Address src, int mode) {
3176 assert(((dst->encoding() < 16) || VM_Version::supports_avx512vl()),"XMM register should be 0-15");
3177 Assembler::pshufd(dst, src, mode);
3178 }
3179
3180 void MacroAssembler::pshuflw(XMMRegister dst, XMMRegister src, int mode) {
3181 assert(((dst->encoding() < 16 && src->encoding() < 16) || VM_Version::supports_avx512vlbw()),"XMM register should be 0-15");
3182 Assembler::pshuflw(dst, src, mode);
3183 }
3184
3185 void MacroAssembler::vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3186 assert(rscratch != noreg || always_reachable(src), "missing");
3187
3188 if (reachable(src)) {
3189 vandpd(dst, nds, as_Address(src), vector_len);
3190 } else {
3191 lea(rscratch, src);
3192 vandpd(dst, nds, Address(rscratch, 0), vector_len);
3193 }
3194 }
3195
3196 void MacroAssembler::vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3197 assert(rscratch != noreg || always_reachable(src), "missing");
3198
3199 if (reachable(src)) {
3200 vandps(dst, nds, as_Address(src), vector_len);
3201 } else {
3202 lea(rscratch, src);
3203 vandps(dst, nds, Address(rscratch, 0), vector_len);
3204 }
3205 }
3206
3207 void MacroAssembler::evpord(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src,
3208 bool merge, int vector_len, Register rscratch) {
3209 assert(rscratch != noreg || always_reachable(src), "missing");
3210
3211 if (reachable(src)) {
3212 Assembler::evpord(dst, mask, nds, as_Address(src), merge, vector_len);
3213 } else {
3214 lea(rscratch, src);
3215 Assembler::evpord(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
3216 }
3217 }
3218
3219 void MacroAssembler::vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3220 assert(rscratch != noreg || always_reachable(src), "missing");
3221
3222 if (reachable(src)) {
3223 vdivsd(dst, nds, as_Address(src));
3224 } else {
3225 lea(rscratch, src);
3226 vdivsd(dst, nds, Address(rscratch, 0));
3227 }
3228 }
3229
3230 void MacroAssembler::vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3231 assert(rscratch != noreg || always_reachable(src), "missing");
3232
3233 if (reachable(src)) {
3234 vdivss(dst, nds, as_Address(src));
3235 } else {
3236 lea(rscratch, src);
3237 vdivss(dst, nds, Address(rscratch, 0));
3238 }
3239 }
3240
3241 void MacroAssembler::vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3242 assert(rscratch != noreg || always_reachable(src), "missing");
3243
3244 if (reachable(src)) {
3245 vmulsd(dst, nds, as_Address(src));
3246 } else {
3247 lea(rscratch, src);
3248 vmulsd(dst, nds, Address(rscratch, 0));
3249 }
3250 }
3251
3252 void MacroAssembler::vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3253 assert(rscratch != noreg || always_reachable(src), "missing");
3254
3255 if (reachable(src)) {
3256 vmulss(dst, nds, as_Address(src));
3257 } else {
3258 lea(rscratch, src);
3259 vmulss(dst, nds, Address(rscratch, 0));
3260 }
3261 }
3262
3263 void MacroAssembler::vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3264 assert(rscratch != noreg || always_reachable(src), "missing");
3265
3266 if (reachable(src)) {
3267 vsubsd(dst, nds, as_Address(src));
3268 } else {
3269 lea(rscratch, src);
3270 vsubsd(dst, nds, Address(rscratch, 0));
3271 }
3272 }
3273
3274 void MacroAssembler::vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3275 assert(rscratch != noreg || always_reachable(src), "missing");
3276
3277 if (reachable(src)) {
3278 vsubss(dst, nds, as_Address(src));
3279 } else {
3280 lea(rscratch, src);
3281 vsubss(dst, nds, Address(rscratch, 0));
3282 }
3283 }
3284
3285 void MacroAssembler::vnegatess(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3286 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
3287 assert(rscratch != noreg || always_reachable(src), "missing");
3288
3289 vxorps(dst, nds, src, Assembler::AVX_128bit, rscratch);
3290 }
3291
3292 void MacroAssembler::vnegatesd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch) {
3293 assert(((dst->encoding() < 16 && nds->encoding() < 16) || VM_Version::supports_avx512vldq()),"XMM register should be 0-15");
3294 assert(rscratch != noreg || always_reachable(src), "missing");
3295
3296 vxorpd(dst, nds, src, Assembler::AVX_128bit, rscratch);
3297 }
3298
3299 void MacroAssembler::vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3300 assert(rscratch != noreg || always_reachable(src), "missing");
3301
3302 if (reachable(src)) {
3303 vxorpd(dst, nds, as_Address(src), vector_len);
3304 } else {
3305 lea(rscratch, src);
3306 vxorpd(dst, nds, Address(rscratch, 0), vector_len);
3307 }
3308 }
3309
3310 void MacroAssembler::vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3311 assert(rscratch != noreg || always_reachable(src), "missing");
3312
3313 if (reachable(src)) {
3314 vxorps(dst, nds, as_Address(src), vector_len);
3315 } else {
3316 lea(rscratch, src);
3317 vxorps(dst, nds, Address(rscratch, 0), vector_len);
3318 }
3319 }
3320
3321 void MacroAssembler::vpxor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3322 assert(rscratch != noreg || always_reachable(src), "missing");
3323
3324 if (UseAVX > 1 || (vector_len < 1)) {
3325 if (reachable(src)) {
3326 Assembler::vpxor(dst, nds, as_Address(src), vector_len);
3327 } else {
3328 lea(rscratch, src);
3329 Assembler::vpxor(dst, nds, Address(rscratch, 0), vector_len);
3330 }
3331 } else {
3332 MacroAssembler::vxorpd(dst, nds, src, vector_len, rscratch);
3333 }
3334 }
3335
3336 void MacroAssembler::vpermd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
3337 assert(rscratch != noreg || always_reachable(src), "missing");
3338
3339 if (reachable(src)) {
3340 Assembler::vpermd(dst, nds, as_Address(src), vector_len);
3341 } else {
3342 lea(rscratch, src);
3343 Assembler::vpermd(dst, nds, Address(rscratch, 0), vector_len);
3344 }
3345 }
3346
3347 void MacroAssembler::clear_jobject_tag(Register possibly_non_local) {
3348 const int32_t inverted_mask = ~static_cast<int32_t>(JNIHandles::tag_mask);
3349 STATIC_ASSERT(inverted_mask == -4); // otherwise check this code
3350 // The inverted mask is sign-extended
3351 andptr(possibly_non_local, inverted_mask);
3352 }
3353
3354 void MacroAssembler::resolve_jobject(Register value,
3355 Register tmp) {
3356 Register thread = r15_thread;
3357 assert_different_registers(value, thread, tmp);
3358 Label done, tagged, weak_tagged;
3359 testptr(value, value);
3360 jcc(Assembler::zero, done); // Use null as-is.
3361 testptr(value, JNIHandles::tag_mask); // Test for tag.
3362 jcc(Assembler::notZero, tagged);
3363
3364 // Resolve local handle
3365 access_load_at(T_OBJECT, IN_NATIVE | AS_RAW, value, Address(value, 0), tmp);
3366 verify_oop(value);
3367 jmp(done);
3368
3369 bind(tagged);
3370 testptr(value, JNIHandles::TypeTag::weak_global); // Test for weak tag.
3371 jcc(Assembler::notZero, weak_tagged);
3372
3373 // Resolve global handle
3374 access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
3375 verify_oop(value);
3376 jmp(done);
3377
3378 bind(weak_tagged);
3379 // Resolve jweak.
3380 access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
3381 value, Address(value, -JNIHandles::TypeTag::weak_global), tmp);
3382 verify_oop(value);
3383
3384 bind(done);
3385 }
3386
3387 void MacroAssembler::resolve_global_jobject(Register value,
3388 Register tmp) {
3389 Register thread = r15_thread;
3390 assert_different_registers(value, thread, tmp);
3391 Label done;
3392
3393 testptr(value, value);
3394 jcc(Assembler::zero, done); // Use null as-is.
3395
3396 #ifdef ASSERT
3397 {
3398 Label valid_global_tag;
3399 testptr(value, JNIHandles::TypeTag::global); // Test for global tag.
3400 jcc(Assembler::notZero, valid_global_tag);
3401 stop("non global jobject using resolve_global_jobject");
3402 bind(valid_global_tag);
3403 }
3404 #endif
3405
3406 // Resolve global handle
3407 access_load_at(T_OBJECT, IN_NATIVE, value, Address(value, -JNIHandles::TypeTag::global), tmp);
3408 verify_oop(value);
3409
3410 bind(done);
3411 }
3412
3413 void MacroAssembler::subptr(Register dst, int32_t imm32) {
3414 subq(dst, imm32);
3415 }
3416
3417 // Force generation of a 4 byte immediate value even if it fits into 8bit
3418 void MacroAssembler::subptr_imm32(Register dst, int32_t imm32) {
3419 subq_imm32(dst, imm32);
3420 }
3421
3422 void MacroAssembler::subptr(Register dst, Register src) {
3423 subq(dst, src);
3424 }
3425
3426 // C++ bool manipulation
3427 void MacroAssembler::testbool(Register dst) {
3428 if(sizeof(bool) == 1)
3429 testb(dst, 0xff);
3430 else if(sizeof(bool) == 2) {
3431 // testw implementation needed for two byte bools
3432 ShouldNotReachHere();
3433 } else if(sizeof(bool) == 4)
3434 testl(dst, dst);
3435 else
3436 // unsupported
3437 ShouldNotReachHere();
3438 }
3439
3440 void MacroAssembler::testptr(Register dst, Register src) {
3441 testq(dst, src);
3442 }
3443
3444 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
3445 void MacroAssembler::tlab_allocate(Register obj,
3446 Register var_size_in_bytes,
3447 int con_size_in_bytes,
3448 Register t1,
3449 Register t2,
3450 Label& slow_case) {
3451 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
3452 bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
3453 }
3454
3455 RegSet MacroAssembler::call_clobbered_gp_registers() {
3456 RegSet regs;
3457 regs += RegSet::of(rax, rcx, rdx);
3458 #ifndef _WINDOWS
3459 regs += RegSet::of(rsi, rdi);
3460 #endif
3461 regs += RegSet::range(r8, r11);
3462 if (UseAPX) {
3463 regs += RegSet::range(r16, as_Register(Register::number_of_registers - 1));
3464 }
3465 return regs;
3466 }
3467
3468 XMMRegSet MacroAssembler::call_clobbered_xmm_registers() {
3469 int num_xmm_registers = XMMRegister::available_xmm_registers();
3470 #if defined(_WINDOWS)
3471 XMMRegSet result = XMMRegSet::range(xmm0, xmm5);
3472 if (num_xmm_registers > 16) {
3473 result += XMMRegSet::range(xmm16, as_XMMRegister(num_xmm_registers - 1));
3474 }
3475 return result;
3476 #else
3477 return XMMRegSet::range(xmm0, as_XMMRegister(num_xmm_registers - 1));
3478 #endif
3479 }
3480
3481 // C1 only ever uses the first double/float of the XMM register.
3482 static int xmm_save_size() { return sizeof(double); }
3483
3484 static void save_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
3485 masm->movdbl(Address(rsp, offset), reg);
3486 }
3487
3488 static void restore_xmm_register(MacroAssembler* masm, int offset, XMMRegister reg) {
3489 masm->movdbl(reg, Address(rsp, offset));
3490 }
3491
3492 static int register_section_sizes(RegSet gp_registers, XMMRegSet xmm_registers,
3493 bool save_fpu, int& gp_area_size, int& xmm_area_size) {
3494
3495 gp_area_size = align_up(gp_registers.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size,
3496 StackAlignmentInBytes);
3497 xmm_area_size = save_fpu ? xmm_registers.size() * xmm_save_size() : 0;
3498
3499 return gp_area_size + xmm_area_size;
3500 }
3501
3502 void MacroAssembler::push_call_clobbered_registers_except(RegSet exclude, bool save_fpu) {
3503 block_comment("push_call_clobbered_registers start");
3504 // Regular registers
3505 RegSet gp_registers_to_push = call_clobbered_gp_registers() - exclude;
3506
3507 int gp_area_size;
3508 int xmm_area_size;
3509 int total_save_size = register_section_sizes(gp_registers_to_push, call_clobbered_xmm_registers(), save_fpu,
3510 gp_area_size, xmm_area_size);
3511 subptr(rsp, total_save_size);
3512
3513 push_set(gp_registers_to_push, 0);
3514
3515 if (save_fpu) {
3516 push_set(call_clobbered_xmm_registers(), gp_area_size);
3517 }
3518
3519 block_comment("push_call_clobbered_registers end");
3520 }
3521
3522 void MacroAssembler::pop_call_clobbered_registers_except(RegSet exclude, bool restore_fpu) {
3523 block_comment("pop_call_clobbered_registers start");
3524
3525 RegSet gp_registers_to_pop = call_clobbered_gp_registers() - exclude;
3526
3527 int gp_area_size;
3528 int xmm_area_size;
3529 int total_save_size = register_section_sizes(gp_registers_to_pop, call_clobbered_xmm_registers(), restore_fpu,
3530 gp_area_size, xmm_area_size);
3531
3532 if (restore_fpu) {
3533 pop_set(call_clobbered_xmm_registers(), gp_area_size);
3534 }
3535
3536 pop_set(gp_registers_to_pop, 0);
3537
3538 addptr(rsp, total_save_size);
3539
3540 vzeroupper();
3541
3542 block_comment("pop_call_clobbered_registers end");
3543 }
3544
3545 void MacroAssembler::push_set(XMMRegSet set, int offset) {
3546 assert(is_aligned(set.size() * xmm_save_size(), StackAlignmentInBytes), "must be");
3547 int spill_offset = offset;
3548
3549 for (RegSetIterator<XMMRegister> it = set.begin(); *it != xnoreg; ++it) {
3550 save_xmm_register(this, spill_offset, *it);
3551 spill_offset += xmm_save_size();
3552 }
3553 }
3554
3555 void MacroAssembler::pop_set(XMMRegSet set, int offset) {
3556 int restore_size = set.size() * xmm_save_size();
3557 assert(is_aligned(restore_size, StackAlignmentInBytes), "must be");
3558
3559 int restore_offset = offset + restore_size - xmm_save_size();
3560
3561 for (ReverseRegSetIterator<XMMRegister> it = set.rbegin(); *it != xnoreg; ++it) {
3562 restore_xmm_register(this, restore_offset, *it);
3563 restore_offset -= xmm_save_size();
3564 }
3565 }
3566
3567 void MacroAssembler::push_set(RegSet set, int offset) {
3568 int spill_offset;
3569 if (offset == -1) {
3570 int register_push_size = set.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
3571 int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
3572 subptr(rsp, aligned_size);
3573 spill_offset = 0;
3574 } else {
3575 spill_offset = offset;
3576 }
3577
3578 for (RegSetIterator<Register> it = set.begin(); *it != noreg; ++it) {
3579 movptr(Address(rsp, spill_offset), *it);
3580 spill_offset += Register::max_slots_per_register * VMRegImpl::stack_slot_size;
3581 }
3582 }
3583
3584 void MacroAssembler::pop_set(RegSet set, int offset) {
3585
3586 int gp_reg_size = Register::max_slots_per_register * VMRegImpl::stack_slot_size;
3587 int restore_size = set.size() * gp_reg_size;
3588 int aligned_size = align_up(restore_size, StackAlignmentInBytes);
3589
3590 int restore_offset;
3591 if (offset == -1) {
3592 restore_offset = restore_size - gp_reg_size;
3593 } else {
3594 restore_offset = offset + restore_size - gp_reg_size;
3595 }
3596 for (ReverseRegSetIterator<Register> it = set.rbegin(); *it != noreg; ++it) {
3597 movptr(*it, Address(rsp, restore_offset));
3598 restore_offset -= gp_reg_size;
3599 }
3600
3601 if (offset == -1) {
3602 addptr(rsp, aligned_size);
3603 }
3604 }
3605
3606 // Preserves the contents of address, destroys the contents length_in_bytes and temp.
3607 void MacroAssembler::zero_memory(Register address, Register length_in_bytes, int offset_in_bytes, Register temp) {
3608 assert(address != length_in_bytes && address != temp && temp != length_in_bytes, "registers must be different");
3609 assert((offset_in_bytes & (BytesPerWord - 1)) == 0, "offset must be a multiple of BytesPerWord");
3610 Label done;
3611
3612 testptr(length_in_bytes, length_in_bytes);
3613 jcc(Assembler::zero, done);
3614
3615 // initialize topmost word, divide index by 2, check if odd and test if zero
3616 // note: for the remaining code to work, index must be a multiple of BytesPerWord
3617 #ifdef ASSERT
3618 {
3619 Label L;
3620 testptr(length_in_bytes, BytesPerWord - 1);
3621 jcc(Assembler::zero, L);
3622 stop("length must be a multiple of BytesPerWord");
3623 bind(L);
3624 }
3625 #endif
3626 Register index = length_in_bytes;
3627 xorptr(temp, temp); // use _zero reg to clear memory (shorter code)
3628 if (UseIncDec) {
3629 shrptr(index, 3); // divide by 8/16 and set carry flag if bit 2 was set
3630 } else {
3631 shrptr(index, 2); // use 2 instructions to avoid partial flag stall
3632 shrptr(index, 1);
3633 }
3634
3635 // initialize remaining object fields: index is a multiple of 2 now
3636 {
3637 Label loop;
3638 bind(loop);
3639 movptr(Address(address, index, Address::times_8, offset_in_bytes - 1*BytesPerWord), temp);
3640 decrement(index);
3641 jcc(Assembler::notZero, loop);
3642 }
3643
3644 bind(done);
3645 }
3646
3647 // Look up the method for a megamorphic invokeinterface call.
3648 // The target method is determined by <intf_klass, itable_index>.
3649 // The receiver klass is in recv_klass.
3650 // On success, the result will be in method_result, and execution falls through.
3651 // On failure, execution transfers to the given label.
3652 void MacroAssembler::lookup_interface_method(Register recv_klass,
3653 Register intf_klass,
3654 RegisterOrConstant itable_index,
3655 Register method_result,
3656 Register scan_temp,
3657 Label& L_no_such_interface,
3658 bool return_method) {
3659 assert_different_registers(recv_klass, intf_klass, scan_temp);
3660 assert_different_registers(method_result, intf_klass, scan_temp);
3661 assert(recv_klass != method_result || !return_method,
3662 "recv_klass can be destroyed when method isn't needed");
3663
3664 assert(itable_index.is_constant() || itable_index.as_register() == method_result,
3665 "caller must use same register for non-constant itable index as for method");
3666
3667 // Compute start of first itableOffsetEntry (which is at the end of the vtable)
3668 int vtable_base = in_bytes(Klass::vtable_start_offset());
3669 int itentry_off = in_bytes(itableMethodEntry::method_offset());
3670 int scan_step = itableOffsetEntry::size() * wordSize;
3671 int vte_size = vtableEntry::size_in_bytes();
3672 Address::ScaleFactor times_vte_scale = Address::times_ptr;
3673 assert(vte_size == wordSize, "else adjust times_vte_scale");
3674
3675 movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
3676
3677 // Could store the aligned, prescaled offset in the klass.
3678 lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base));
3679
3680 if (return_method) {
3681 // Adjust recv_klass by scaled itable_index, so we can free itable_index.
3682 assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
3683 lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off));
3684 }
3685
3686 // for (scan = klass->itable(); scan->interface() != nullptr; scan += scan_step) {
3687 // if (scan->interface() == intf) {
3688 // result = (klass + scan->offset() + itable_index);
3689 // }
3690 // }
3691 Label search, found_method;
3692
3693 for (int peel = 1; peel >= 0; peel--) {
3694 movptr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset()));
3695 cmpptr(intf_klass, method_result);
3696
3697 if (peel) {
3698 jccb(Assembler::equal, found_method);
3699 } else {
3700 jccb(Assembler::notEqual, search);
3701 // (invert the test to fall through to found_method...)
3702 }
3703
3704 if (!peel) break;
3705
3706 bind(search);
3707
3708 // Check that the previous entry is non-null. A null entry means that
3709 // the receiver class doesn't implement the interface, and wasn't the
3710 // same as when the caller was compiled.
3711 testptr(method_result, method_result);
3712 jcc(Assembler::zero, L_no_such_interface);
3713 addptr(scan_temp, scan_step);
3714 }
3715
3716 bind(found_method);
3717
3718 if (return_method) {
3719 // Got a hit.
3720 movl(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset()));
3721 movptr(method_result, Address(recv_klass, scan_temp, Address::times_1));
3722 }
3723 }
3724
3725 // Look up the method for a megamorphic invokeinterface call in a single pass over itable:
3726 // - check recv_klass (actual object class) is a subtype of resolved_klass from CompiledICData
3727 // - find a holder_klass (class that implements the method) vtable offset and get the method from vtable by index
3728 // The target method is determined by <holder_klass, itable_index>.
3729 // The receiver klass is in recv_klass.
3730 // On success, the result will be in method_result, and execution falls through.
3731 // On failure, execution transfers to the given label.
3732 void MacroAssembler::lookup_interface_method_stub(Register recv_klass,
3733 Register holder_klass,
3734 Register resolved_klass,
3735 Register method_result,
3736 Register scan_temp,
3737 Register temp_reg2,
3738 Register receiver,
3739 int itable_index,
3740 Label& L_no_such_interface) {
3741 assert_different_registers(recv_klass, method_result, holder_klass, resolved_klass, scan_temp, temp_reg2, receiver);
3742 Register temp_itbl_klass = method_result;
3743 Register temp_reg = (temp_reg2 == noreg ? recv_klass : temp_reg2); // reuse recv_klass register on 32-bit x86 impl
3744
3745 int vtable_base = in_bytes(Klass::vtable_start_offset());
3746 int itentry_off = in_bytes(itableMethodEntry::method_offset());
3747 int scan_step = itableOffsetEntry::size() * wordSize;
3748 int vte_size = vtableEntry::size_in_bytes();
3749 int ioffset = in_bytes(itableOffsetEntry::interface_offset());
3750 int ooffset = in_bytes(itableOffsetEntry::offset_offset());
3751 Address::ScaleFactor times_vte_scale = Address::times_ptr;
3752 assert(vte_size == wordSize, "adjust times_vte_scale");
3753
3754 Label L_loop_scan_resolved_entry, L_resolved_found, L_holder_found;
3755
3756 // temp_itbl_klass = recv_klass.itable[0]
3757 // scan_temp = &recv_klass.itable[0] + step
3758 movl(scan_temp, Address(recv_klass, Klass::vtable_length_offset()));
3759 movptr(temp_itbl_klass, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset));
3760 lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base + ioffset + scan_step));
3761 xorptr(temp_reg, temp_reg);
3762
3763 // Initial checks:
3764 // - if (holder_klass != resolved_klass), go to "scan for resolved"
3765 // - if (itable[0] == 0), no such interface
3766 // - if (itable[0] == holder_klass), shortcut to "holder found"
3767 cmpptr(holder_klass, resolved_klass);
3768 jccb(Assembler::notEqual, L_loop_scan_resolved_entry);
3769 testptr(temp_itbl_klass, temp_itbl_klass);
3770 jccb(Assembler::zero, L_no_such_interface);
3771 cmpptr(holder_klass, temp_itbl_klass);
3772 jccb(Assembler::equal, L_holder_found);
3773
3774 // Loop: Look for holder_klass record in itable
3775 // do {
3776 // tmp = itable[index];
3777 // index += step;
3778 // if (tmp == holder_klass) {
3779 // goto L_holder_found; // Found!
3780 // }
3781 // } while (tmp != 0);
3782 // goto L_no_such_interface // Not found.
3783 Label L_scan_holder;
3784 bind(L_scan_holder);
3785 movptr(temp_itbl_klass, Address(scan_temp, 0));
3786 addptr(scan_temp, scan_step);
3787 cmpptr(holder_klass, temp_itbl_klass);
3788 jccb(Assembler::equal, L_holder_found);
3789 testptr(temp_itbl_klass, temp_itbl_klass);
3790 jccb(Assembler::notZero, L_scan_holder);
3791
3792 jmpb(L_no_such_interface);
3793
3794 // Loop: Look for resolved_class record in itable
3795 // do {
3796 // tmp = itable[index];
3797 // index += step;
3798 // if (tmp == holder_klass) {
3799 // // Also check if we have met a holder klass
3800 // holder_tmp = itable[index-step-ioffset];
3801 // }
3802 // if (tmp == resolved_klass) {
3803 // goto L_resolved_found; // Found!
3804 // }
3805 // } while (tmp != 0);
3806 // goto L_no_such_interface // Not found.
3807 //
3808 Label L_loop_scan_resolved;
3809 bind(L_loop_scan_resolved);
3810 movptr(temp_itbl_klass, Address(scan_temp, 0));
3811 addptr(scan_temp, scan_step);
3812 bind(L_loop_scan_resolved_entry);
3813 cmpptr(holder_klass, temp_itbl_klass);
3814 cmovl(Assembler::equal, temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
3815 cmpptr(resolved_klass, temp_itbl_klass);
3816 jccb(Assembler::equal, L_resolved_found);
3817 testptr(temp_itbl_klass, temp_itbl_klass);
3818 jccb(Assembler::notZero, L_loop_scan_resolved);
3819
3820 jmpb(L_no_such_interface);
3821
3822 Label L_ready;
3823
3824 // See if we already have a holder klass. If not, go and scan for it.
3825 bind(L_resolved_found);
3826 testptr(temp_reg, temp_reg);
3827 jccb(Assembler::zero, L_scan_holder);
3828 jmpb(L_ready);
3829
3830 bind(L_holder_found);
3831 movl(temp_reg, Address(scan_temp, ooffset - ioffset - scan_step));
3832
3833 // Finally, temp_reg contains holder_klass vtable offset
3834 bind(L_ready);
3835 assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
3836 if (temp_reg2 == noreg) { // recv_klass register is clobbered for 32-bit x86 impl
3837 load_klass(scan_temp, receiver, noreg);
3838 movptr(method_result, Address(scan_temp, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
3839 } else {
3840 movptr(method_result, Address(recv_klass, temp_reg, Address::times_1, itable_index * wordSize + itentry_off));
3841 }
3842 }
3843
3844
3845 // virtual method calling
3846 void MacroAssembler::lookup_virtual_method(Register recv_klass,
3847 RegisterOrConstant vtable_index,
3848 Register method_result) {
3849 const ByteSize base = Klass::vtable_start_offset();
3850 assert(vtableEntry::size() * wordSize == wordSize, "else adjust the scaling in the code below");
3851 Address vtable_entry_addr(recv_klass,
3852 vtable_index, Address::times_ptr,
3853 base + vtableEntry::method_offset());
3854 movptr(method_result, vtable_entry_addr);
3855 }
3856
3857
3858 void MacroAssembler::check_klass_subtype(Register sub_klass,
3859 Register super_klass,
3860 Register temp_reg,
3861 Label& L_success) {
3862 Label L_failure;
3863 check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg, &L_success, &L_failure, nullptr);
3864 check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, nullptr);
3865 bind(L_failure);
3866 }
3867
3868
3869 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
3870 Register super_klass,
3871 Register temp_reg,
3872 Label* L_success,
3873 Label* L_failure,
3874 Label* L_slow_path,
3875 RegisterOrConstant super_check_offset) {
3876 assert_different_registers(sub_klass, super_klass, temp_reg);
3877 bool must_load_sco = (super_check_offset.constant_or_zero() == -1);
3878 if (super_check_offset.is_register()) {
3879 assert_different_registers(sub_klass, super_klass,
3880 super_check_offset.as_register());
3881 } else if (must_load_sco) {
3882 assert(temp_reg != noreg, "supply either a temp or a register offset");
3883 }
3884
3885 Label L_fallthrough;
3886 int label_nulls = 0;
3887 if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
3888 if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
3889 if (L_slow_path == nullptr) { L_slow_path = &L_fallthrough; label_nulls++; }
3890 assert(label_nulls <= 1, "at most one null in the batch");
3891
3892 int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
3893 int sco_offset = in_bytes(Klass::super_check_offset_offset());
3894 Address super_check_offset_addr(super_klass, sco_offset);
3895
3896 // Hacked jcc, which "knows" that L_fallthrough, at least, is in
3897 // range of a jccb. If this routine grows larger, reconsider at
3898 // least some of these.
3899 #define local_jcc(assembler_cond, label) \
3900 if (&(label) == &L_fallthrough) jccb(assembler_cond, label); \
3901 else jcc( assembler_cond, label) /*omit semi*/
3902
3903 // Hacked jmp, which may only be used just before L_fallthrough.
3904 #define final_jmp(label) \
3905 if (&(label) == &L_fallthrough) { /*do nothing*/ } \
3906 else jmp(label) /*omit semi*/
3907
3908 // If the pointers are equal, we are done (e.g., String[] elements).
3909 // This self-check enables sharing of secondary supertype arrays among
3910 // non-primary types such as array-of-interface. Otherwise, each such
3911 // type would need its own customized SSA.
3912 // We move this check to the front of the fast path because many
3913 // type checks are in fact trivially successful in this manner,
3914 // so we get a nicely predicted branch right at the start of the check.
3915 cmpptr(sub_klass, super_klass);
3916 local_jcc(Assembler::equal, *L_success);
3917
3918 // Check the supertype display:
3919 if (must_load_sco) {
3920 // Positive movl does right thing on LP64.
3921 movl(temp_reg, super_check_offset_addr);
3922 super_check_offset = RegisterOrConstant(temp_reg);
3923 }
3924 Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0);
3925 cmpptr(super_klass, super_check_addr); // load displayed supertype
3926
3927 // This check has worked decisively for primary supers.
3928 // Secondary supers are sought in the super_cache ('super_cache_addr').
3929 // (Secondary supers are interfaces and very deeply nested subtypes.)
3930 // This works in the same check above because of a tricky aliasing
3931 // between the super_cache and the primary super display elements.
3932 // (The 'super_check_addr' can address either, as the case requires.)
3933 // Note that the cache is updated below if it does not help us find
3934 // what we need immediately.
3935 // So if it was a primary super, we can just fail immediately.
3936 // Otherwise, it's the slow path for us (no success at this point).
3937
3938 if (super_check_offset.is_register()) {
3939 local_jcc(Assembler::equal, *L_success);
3940 cmpl(super_check_offset.as_register(), sc_offset);
3941 if (L_failure == &L_fallthrough) {
3942 local_jcc(Assembler::equal, *L_slow_path);
3943 } else {
3944 local_jcc(Assembler::notEqual, *L_failure);
3945 final_jmp(*L_slow_path);
3946 }
3947 } else if (super_check_offset.as_constant() == sc_offset) {
3948 // Need a slow path; fast failure is impossible.
3949 if (L_slow_path == &L_fallthrough) {
3950 local_jcc(Assembler::equal, *L_success);
3951 } else {
3952 local_jcc(Assembler::notEqual, *L_slow_path);
3953 final_jmp(*L_success);
3954 }
3955 } else {
3956 // No slow path; it's a fast decision.
3957 if (L_failure == &L_fallthrough) {
3958 local_jcc(Assembler::equal, *L_success);
3959 } else {
3960 local_jcc(Assembler::notEqual, *L_failure);
3961 final_jmp(*L_success);
3962 }
3963 }
3964
3965 bind(L_fallthrough);
3966
3967 #undef local_jcc
3968 #undef final_jmp
3969 }
3970
3971
3972 void MacroAssembler::check_klass_subtype_slow_path_linear(Register sub_klass,
3973 Register super_klass,
3974 Register temp_reg,
3975 Register temp2_reg,
3976 Label* L_success,
3977 Label* L_failure,
3978 bool set_cond_codes) {
3979 assert_different_registers(sub_klass, super_klass, temp_reg);
3980 if (temp2_reg != noreg)
3981 assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg);
3982 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg)
3983
3984 Label L_fallthrough;
3985 int label_nulls = 0;
3986 if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
3987 if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
3988 assert(label_nulls <= 1, "at most one null in the batch");
3989
3990 // a couple of useful fields in sub_klass:
3991 int ss_offset = in_bytes(Klass::secondary_supers_offset());
3992 int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
3993 Address secondary_supers_addr(sub_klass, ss_offset);
3994 Address super_cache_addr( sub_klass, sc_offset);
3995
3996 // Do a linear scan of the secondary super-klass chain.
3997 // This code is rarely used, so simplicity is a virtue here.
3998 // The repne_scan instruction uses fixed registers, which we must spill.
3999 // Don't worry too much about pre-existing connections with the input regs.
4000
4001 assert(sub_klass != rax, "killed reg"); // killed by mov(rax, super)
4002 assert(sub_klass != rcx, "killed reg"); // killed by lea(rcx, &pst_counter)
4003
4004 // Get super_klass value into rax (even if it was in rdi or rcx).
4005 bool pushed_rax = false, pushed_rcx = false, pushed_rdi = false;
4006 if (super_klass != rax) {
4007 if (!IS_A_TEMP(rax)) { push(rax); pushed_rax = true; }
4008 mov(rax, super_klass);
4009 }
4010 if (!IS_A_TEMP(rcx)) { push(rcx); pushed_rcx = true; }
4011 if (!IS_A_TEMP(rdi)) { push(rdi); pushed_rdi = true; }
4012
4013 #ifndef PRODUCT
4014 uint* pst_counter = &SharedRuntime::_partial_subtype_ctr;
4015 ExternalAddress pst_counter_addr((address) pst_counter);
4016 lea(rcx, pst_counter_addr);
4017 incrementl(Address(rcx, 0));
4018 #endif //PRODUCT
4019
4020 // We will consult the secondary-super array.
4021 movptr(rdi, secondary_supers_addr);
4022 // Load the array length. (Positive movl does right thing on LP64.)
4023 movl(rcx, Address(rdi, Array<Klass*>::length_offset_in_bytes()));
4024 // Skip to start of data.
4025 addptr(rdi, Array<Klass*>::base_offset_in_bytes());
4026
4027 // Scan RCX words at [RDI] for an occurrence of RAX.
4028 // Set NZ/Z based on last compare.
4029 // Z flag value will not be set by 'repne' if RCX == 0 since 'repne' does
4030 // not change flags (only scas instruction which is repeated sets flags).
4031 // Set Z = 0 (not equal) before 'repne' to indicate that class was not found.
4032
4033 testptr(rax,rax); // Set Z = 0
4034 repne_scan();
4035
4036 // Unspill the temp. registers:
4037 if (pushed_rdi) pop(rdi);
4038 if (pushed_rcx) pop(rcx);
4039 if (pushed_rax) pop(rax);
4040
4041 if (set_cond_codes) {
4042 // Special hack for the AD files: rdi is guaranteed non-zero.
4043 assert(!pushed_rdi, "rdi must be left non-null");
4044 // Also, the condition codes are properly set Z/NZ on succeed/failure.
4045 }
4046
4047 if (L_failure == &L_fallthrough)
4048 jccb(Assembler::notEqual, *L_failure);
4049 else jcc(Assembler::notEqual, *L_failure);
4050
4051 // Success. Cache the super we found and proceed in triumph.
4052 movptr(super_cache_addr, super_klass);
4053
4054 if (L_success != &L_fallthrough) {
4055 jmp(*L_success);
4056 }
4057
4058 #undef IS_A_TEMP
4059
4060 bind(L_fallthrough);
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 Label* L_success,
4068 Label* L_failure,
4069 bool set_cond_codes) {
4070 assert(set_cond_codes == false, "must be false on 64-bit x86");
4071 check_klass_subtype_slow_path
4072 (sub_klass, super_klass, temp_reg, temp2_reg, noreg, noreg,
4073 L_success, L_failure);
4074 }
4075
4076 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass,
4077 Register super_klass,
4078 Register temp_reg,
4079 Register temp2_reg,
4080 Register temp3_reg,
4081 Register temp4_reg,
4082 Label* L_success,
4083 Label* L_failure) {
4084 if (UseSecondarySupersTable) {
4085 check_klass_subtype_slow_path_table
4086 (sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, temp4_reg,
4087 L_success, L_failure);
4088 } else {
4089 check_klass_subtype_slow_path_linear
4090 (sub_klass, super_klass, temp_reg, temp2_reg, L_success, L_failure, /*set_cond_codes*/false);
4091 }
4092 }
4093
4094 Register MacroAssembler::allocate_if_noreg(Register r,
4095 RegSetIterator<Register> &available_regs,
4096 RegSet ®s_to_push) {
4097 if (!r->is_valid()) {
4098 r = *available_regs++;
4099 regs_to_push += r;
4100 }
4101 return r;
4102 }
4103
4104 void MacroAssembler::check_klass_subtype_slow_path_table(Register sub_klass,
4105 Register super_klass,
4106 Register temp_reg,
4107 Register temp2_reg,
4108 Register temp3_reg,
4109 Register result_reg,
4110 Label* L_success,
4111 Label* L_failure) {
4112 // NB! Callers may assume that, when temp2_reg is a valid register,
4113 // this code sets it to a nonzero value.
4114 bool temp2_reg_was_valid = temp2_reg->is_valid();
4115
4116 RegSet temps = RegSet::of(temp_reg, temp2_reg, temp3_reg);
4117
4118 Label L_fallthrough;
4119 int label_nulls = 0;
4120 if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
4121 if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
4122 assert(label_nulls <= 1, "at most one null in the batch");
4123
4124 BLOCK_COMMENT("check_klass_subtype_slow_path_table");
4125
4126 RegSetIterator<Register> available_regs
4127 = (RegSet::of(rax, rcx, rdx, r8) + r9 + r10 + r11 + r12 - temps - sub_klass - super_klass).begin();
4128
4129 RegSet pushed_regs;
4130
4131 temp_reg = allocate_if_noreg(temp_reg, available_regs, pushed_regs);
4132 temp2_reg = allocate_if_noreg(temp2_reg, available_regs, pushed_regs);
4133 temp3_reg = allocate_if_noreg(temp3_reg, available_regs, pushed_regs);
4134 result_reg = allocate_if_noreg(result_reg, available_regs, pushed_regs);
4135 Register temp4_reg = allocate_if_noreg(noreg, available_regs, pushed_regs);
4136
4137 assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg, temp3_reg, result_reg);
4138
4139 {
4140
4141 int register_push_size = pushed_regs.size() * Register::max_slots_per_register * VMRegImpl::stack_slot_size;
4142 int aligned_size = align_up(register_push_size, StackAlignmentInBytes);
4143 subptr(rsp, aligned_size);
4144 push_set(pushed_regs, 0);
4145
4146 lookup_secondary_supers_table_var(sub_klass,
4147 super_klass,
4148 temp_reg, temp2_reg, temp3_reg, temp4_reg, result_reg);
4149 cmpq(result_reg, 0);
4150
4151 // Unspill the temp. registers:
4152 pop_set(pushed_regs, 0);
4153 // Increment SP but do not clobber flags.
4154 lea(rsp, Address(rsp, aligned_size));
4155 }
4156
4157 if (temp2_reg_was_valid) {
4158 movq(temp2_reg, 1);
4159 }
4160
4161 jcc(Assembler::notEqual, *L_failure);
4162
4163 if (L_success != &L_fallthrough) {
4164 jmp(*L_success);
4165 }
4166
4167 bind(L_fallthrough);
4168 }
4169
4170 // population_count variant for running without the POPCNT
4171 // instruction, which was introduced with SSE4.2 in 2008.
4172 void MacroAssembler::population_count(Register dst, Register src,
4173 Register scratch1, Register scratch2) {
4174 assert_different_registers(src, scratch1, scratch2);
4175 if (UsePopCountInstruction) {
4176 Assembler::popcntq(dst, src);
4177 } else {
4178 assert_different_registers(src, scratch1, scratch2);
4179 assert_different_registers(dst, scratch1, scratch2);
4180 Label loop, done;
4181
4182 mov(scratch1, src);
4183 // dst = 0;
4184 // while(scratch1 != 0) {
4185 // dst++;
4186 // scratch1 &= (scratch1 - 1);
4187 // }
4188 xorl(dst, dst);
4189 testq(scratch1, scratch1);
4190 jccb(Assembler::equal, done);
4191 {
4192 bind(loop);
4193 incq(dst);
4194 movq(scratch2, scratch1);
4195 decq(scratch2);
4196 andq(scratch1, scratch2);
4197 jccb(Assembler::notEqual, loop);
4198 }
4199 bind(done);
4200 }
4201 #ifdef ASSERT
4202 mov64(scratch1, 0xCafeBabeDeadBeef);
4203 movq(scratch2, scratch1);
4204 #endif
4205 }
4206
4207 // Ensure that the inline code and the stub are using the same registers.
4208 #define LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS \
4209 do { \
4210 assert(r_super_klass == rax, "mismatch"); \
4211 assert(r_array_base == rbx, "mismatch"); \
4212 assert(r_array_length == rcx, "mismatch"); \
4213 assert(r_array_index == rdx, "mismatch"); \
4214 assert(r_sub_klass == rsi || r_sub_klass == noreg, "mismatch"); \
4215 assert(r_bitmap == r11 || r_bitmap == noreg, "mismatch"); \
4216 assert(result == rdi || result == noreg, "mismatch"); \
4217 } while(0)
4218
4219 // Versions of salq and rorq that don't need count to be in rcx
4220
4221 void MacroAssembler::salq(Register dest, Register count) {
4222 if (count == rcx) {
4223 Assembler::salq(dest);
4224 } else {
4225 assert_different_registers(rcx, dest);
4226 xchgq(rcx, count);
4227 Assembler::salq(dest);
4228 xchgq(rcx, count);
4229 }
4230 }
4231
4232 void MacroAssembler::rorq(Register dest, Register count) {
4233 if (count == rcx) {
4234 Assembler::rorq(dest);
4235 } else {
4236 assert_different_registers(rcx, dest);
4237 xchgq(rcx, count);
4238 Assembler::rorq(dest);
4239 xchgq(rcx, count);
4240 }
4241 }
4242
4243 // Return true: we succeeded in generating this code
4244 //
4245 // At runtime, return 0 in result if r_super_klass is a superclass of
4246 // r_sub_klass, otherwise return nonzero. Use this if you know the
4247 // super_klass_slot of the class you're looking for. This is always
4248 // the case for instanceof and checkcast.
4249 void MacroAssembler::lookup_secondary_supers_table_const(Register r_sub_klass,
4250 Register r_super_klass,
4251 Register temp1,
4252 Register temp2,
4253 Register temp3,
4254 Register temp4,
4255 Register result,
4256 u1 super_klass_slot) {
4257 assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
4258
4259 Label L_fallthrough, L_success, L_failure;
4260
4261 BLOCK_COMMENT("lookup_secondary_supers_table {");
4262
4263 const Register
4264 r_array_index = temp1,
4265 r_array_length = temp2,
4266 r_array_base = temp3,
4267 r_bitmap = temp4;
4268
4269 LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS;
4270
4271 xorq(result, result); // = 0
4272
4273 movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
4274 movq(r_array_index, r_bitmap);
4275
4276 // First check the bitmap to see if super_klass might be present. If
4277 // the bit is zero, we are certain that super_klass is not one of
4278 // the secondary supers.
4279 u1 bit = super_klass_slot;
4280 {
4281 // NB: If the count in a x86 shift instruction is 0, the flags are
4282 // not affected, so we do a testq instead.
4283 int shift_count = Klass::SECONDARY_SUPERS_TABLE_MASK - bit;
4284 if (shift_count != 0) {
4285 salq(r_array_index, shift_count);
4286 } else {
4287 testq(r_array_index, r_array_index);
4288 }
4289 }
4290 // We test the MSB of r_array_index, i.e. its sign bit
4291 jcc(Assembler::positive, L_failure);
4292
4293 // Get the first array index that can contain super_klass into r_array_index.
4294 if (bit != 0) {
4295 population_count(r_array_index, r_array_index, temp2, temp3);
4296 } else {
4297 movl(r_array_index, 1);
4298 }
4299 // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
4300
4301 // We will consult the secondary-super array.
4302 movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
4303
4304 // We're asserting that the first word in an Array<Klass*> is the
4305 // length, and the second word is the first word of the data. If
4306 // that ever changes, r_array_base will have to be adjusted here.
4307 assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
4308 assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
4309
4310 cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
4311 jccb(Assembler::equal, L_success);
4312
4313 // Is there another entry to check? Consult the bitmap.
4314 btq(r_bitmap, (bit + 1) & Klass::SECONDARY_SUPERS_TABLE_MASK);
4315 jccb(Assembler::carryClear, L_failure);
4316
4317 // Linear probe. Rotate the bitmap so that the next bit to test is
4318 // in Bit 1.
4319 if (bit != 0) {
4320 rorq(r_bitmap, bit);
4321 }
4322
4323 // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
4324 // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
4325 // Kills: r_array_length.
4326 // Returns: result.
4327 call(RuntimeAddress(StubRoutines::lookup_secondary_supers_table_slow_path_stub()));
4328 // Result (0/1) is in rdi
4329 jmpb(L_fallthrough);
4330
4331 bind(L_failure);
4332 incq(result); // 0 => 1
4333
4334 bind(L_success);
4335 // result = 0;
4336
4337 bind(L_fallthrough);
4338 BLOCK_COMMENT("} lookup_secondary_supers_table");
4339
4340 if (VerifySecondarySupers) {
4341 verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
4342 temp1, temp2, temp3);
4343 }
4344 }
4345
4346 // At runtime, return 0 in result if r_super_klass is a superclass of
4347 // r_sub_klass, otherwise return nonzero. Use this version of
4348 // lookup_secondary_supers_table() if you don't know ahead of time
4349 // which superclass will be searched for. Used by interpreter and
4350 // runtime stubs. It is larger and has somewhat greater latency than
4351 // the version above, which takes a constant super_klass_slot.
4352 void MacroAssembler::lookup_secondary_supers_table_var(Register r_sub_klass,
4353 Register r_super_klass,
4354 Register temp1,
4355 Register temp2,
4356 Register temp3,
4357 Register temp4,
4358 Register result) {
4359 assert_different_registers(r_sub_klass, r_super_klass, temp1, temp2, temp3, temp4, result);
4360 assert_different_registers(r_sub_klass, r_super_klass, rcx);
4361 RegSet temps = RegSet::of(temp1, temp2, temp3, temp4);
4362
4363 Label L_fallthrough, L_success, L_failure;
4364
4365 BLOCK_COMMENT("lookup_secondary_supers_table {");
4366
4367 RegSetIterator<Register> available_regs = (temps - rcx).begin();
4368
4369 // FIXME. Once we are sure that all paths reaching this point really
4370 // do pass rcx as one of our temps we can get rid of the following
4371 // workaround.
4372 assert(temps.contains(rcx), "fix this code");
4373
4374 // We prefer to have our shift count in rcx. If rcx is one of our
4375 // temps, use it for slot. If not, pick any of our temps.
4376 Register slot;
4377 if (!temps.contains(rcx)) {
4378 slot = *available_regs++;
4379 } else {
4380 slot = rcx;
4381 }
4382
4383 const Register r_array_index = *available_regs++;
4384 const Register r_bitmap = *available_regs++;
4385
4386 // The logic above guarantees this property, but we state it here.
4387 assert_different_registers(r_array_index, r_bitmap, rcx);
4388
4389 movq(r_bitmap, Address(r_sub_klass, Klass::secondary_supers_bitmap_offset()));
4390 movq(r_array_index, r_bitmap);
4391
4392 // First check the bitmap to see if super_klass might be present. If
4393 // the bit is zero, we are certain that super_klass is not one of
4394 // the secondary supers.
4395 movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
4396 xorl(slot, (u1)(Klass::SECONDARY_SUPERS_TABLE_SIZE - 1)); // slot ^ 63 === 63 - slot (mod 64)
4397 salq(r_array_index, slot);
4398
4399 testq(r_array_index, r_array_index);
4400 // We test the MSB of r_array_index, i.e. its sign bit
4401 jcc(Assembler::positive, L_failure);
4402
4403 const Register r_array_base = *available_regs++;
4404
4405 // Get the first array index that can contain super_klass into r_array_index.
4406 // Note: Clobbers r_array_base and slot.
4407 population_count(r_array_index, r_array_index, /*temp2*/r_array_base, /*temp3*/slot);
4408
4409 // NB! r_array_index is off by 1. It is compensated by keeping r_array_base off by 1 word.
4410
4411 // We will consult the secondary-super array.
4412 movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
4413
4414 // We're asserting that the first word in an Array<Klass*> is the
4415 // length, and the second word is the first word of the data. If
4416 // that ever changes, r_array_base will have to be adjusted here.
4417 assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "Adjust this code");
4418 assert(Array<Klass*>::length_offset_in_bytes() == 0, "Adjust this code");
4419
4420 cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
4421 jccb(Assembler::equal, L_success);
4422
4423 // Restore slot to its true value
4424 movb(slot, Address(r_super_klass, Klass::hash_slot_offset()));
4425
4426 // Linear probe. Rotate the bitmap so that the next bit to test is
4427 // in Bit 1.
4428 rorq(r_bitmap, slot);
4429
4430 // Is there another entry to check? Consult the bitmap.
4431 btq(r_bitmap, 1);
4432 jccb(Assembler::carryClear, L_failure);
4433
4434 // Calls into the stub generated by lookup_secondary_supers_table_slow_path.
4435 // Arguments: r_super_klass, r_array_base, r_array_index, r_bitmap.
4436 // Kills: r_array_length.
4437 // Returns: result.
4438 lookup_secondary_supers_table_slow_path(r_super_klass,
4439 r_array_base,
4440 r_array_index,
4441 r_bitmap,
4442 /*temp1*/result,
4443 /*temp2*/slot,
4444 &L_success,
4445 nullptr);
4446
4447 bind(L_failure);
4448 movq(result, 1);
4449 jmpb(L_fallthrough);
4450
4451 bind(L_success);
4452 xorq(result, result); // = 0
4453
4454 bind(L_fallthrough);
4455 BLOCK_COMMENT("} lookup_secondary_supers_table");
4456
4457 if (VerifySecondarySupers) {
4458 verify_secondary_supers_table(r_sub_klass, r_super_klass, result,
4459 temp1, temp2, temp3);
4460 }
4461 }
4462
4463 void MacroAssembler::repne_scanq(Register addr, Register value, Register count, Register limit,
4464 Label* L_success, Label* L_failure) {
4465 Label L_loop, L_fallthrough;
4466 {
4467 int label_nulls = 0;
4468 if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
4469 if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
4470 assert(label_nulls <= 1, "at most one null in the batch");
4471 }
4472 bind(L_loop);
4473 cmpq(value, Address(addr, count, Address::times_8));
4474 jcc(Assembler::equal, *L_success);
4475 addl(count, 1);
4476 cmpl(count, limit);
4477 jcc(Assembler::less, L_loop);
4478
4479 if (&L_fallthrough != L_failure) {
4480 jmp(*L_failure);
4481 }
4482 bind(L_fallthrough);
4483 }
4484
4485 // Called by code generated by check_klass_subtype_slow_path
4486 // above. This is called when there is a collision in the hashed
4487 // lookup in the secondary supers array.
4488 void MacroAssembler::lookup_secondary_supers_table_slow_path(Register r_super_klass,
4489 Register r_array_base,
4490 Register r_array_index,
4491 Register r_bitmap,
4492 Register temp1,
4493 Register temp2,
4494 Label* L_success,
4495 Label* L_failure) {
4496 assert_different_registers(r_super_klass, r_array_base, r_array_index, r_bitmap, temp1, temp2);
4497
4498 const Register
4499 r_array_length = temp1,
4500 r_sub_klass = noreg,
4501 result = noreg;
4502
4503 Label L_fallthrough;
4504 int label_nulls = 0;
4505 if (L_success == nullptr) { L_success = &L_fallthrough; label_nulls++; }
4506 if (L_failure == nullptr) { L_failure = &L_fallthrough; label_nulls++; }
4507 assert(label_nulls <= 1, "at most one null in the batch");
4508
4509 // Load the array length.
4510 movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
4511 // And adjust the array base to point to the data.
4512 // NB! Effectively increments current slot index by 1.
4513 assert(Array<Klass*>::base_offset_in_bytes() == wordSize, "");
4514 addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
4515
4516 // Linear probe
4517 Label L_huge;
4518
4519 // The bitmap is full to bursting.
4520 // Implicit invariant: BITMAP_FULL implies (length > 0)
4521 cmpl(r_array_length, (int32_t)Klass::SECONDARY_SUPERS_TABLE_SIZE - 2);
4522 jcc(Assembler::greater, L_huge);
4523
4524 // NB! Our caller has checked bits 0 and 1 in the bitmap. The
4525 // current slot (at secondary_supers[r_array_index]) has not yet
4526 // been inspected, and r_array_index may be out of bounds if we
4527 // wrapped around the end of the array.
4528
4529 { // This is conventional linear probing, but instead of terminating
4530 // when a null entry is found in the table, we maintain a bitmap
4531 // in which a 0 indicates missing entries.
4532 // The check above guarantees there are 0s in the bitmap, so the loop
4533 // eventually terminates.
4534
4535 xorl(temp2, temp2); // = 0;
4536
4537 Label L_again;
4538 bind(L_again);
4539
4540 // Check for array wraparound.
4541 cmpl(r_array_index, r_array_length);
4542 cmovl(Assembler::greaterEqual, r_array_index, temp2);
4543
4544 cmpq(r_super_klass, Address(r_array_base, r_array_index, Address::times_8));
4545 jcc(Assembler::equal, *L_success);
4546
4547 // If the next bit in bitmap is zero, we're done.
4548 btq(r_bitmap, 2); // look-ahead check (Bit 2); Bits 0 and 1 are tested by now
4549 jcc(Assembler::carryClear, *L_failure);
4550
4551 rorq(r_bitmap, 1); // Bits 1/2 => 0/1
4552 addl(r_array_index, 1);
4553
4554 jmp(L_again);
4555 }
4556
4557 { // Degenerate case: more than 64 secondary supers.
4558 // FIXME: We could do something smarter here, maybe a vectorized
4559 // comparison or a binary search, but is that worth any added
4560 // complexity?
4561 bind(L_huge);
4562 xorl(r_array_index, r_array_index); // = 0
4563 repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length,
4564 L_success,
4565 (&L_fallthrough != L_failure ? L_failure : nullptr));
4566
4567 bind(L_fallthrough);
4568 }
4569 }
4570
4571 struct VerifyHelperArguments {
4572 Klass* _super;
4573 Klass* _sub;
4574 intptr_t _linear_result;
4575 intptr_t _table_result;
4576 };
4577
4578 static void verify_secondary_supers_table_helper(const char* msg, VerifyHelperArguments* args) {
4579 Klass::on_secondary_supers_verification_failure(args->_super,
4580 args->_sub,
4581 args->_linear_result,
4582 args->_table_result,
4583 msg);
4584 }
4585
4586 // Make sure that the hashed lookup and a linear scan agree.
4587 void MacroAssembler::verify_secondary_supers_table(Register r_sub_klass,
4588 Register r_super_klass,
4589 Register result,
4590 Register temp1,
4591 Register temp2,
4592 Register temp3) {
4593 const Register
4594 r_array_index = temp1,
4595 r_array_length = temp2,
4596 r_array_base = temp3,
4597 r_bitmap = noreg;
4598
4599 BLOCK_COMMENT("verify_secondary_supers_table {");
4600
4601 Label L_success, L_failure, L_check, L_done;
4602
4603 movptr(r_array_base, Address(r_sub_klass, in_bytes(Klass::secondary_supers_offset())));
4604 movl(r_array_length, Address(r_array_base, Array<Klass*>::length_offset_in_bytes()));
4605 // And adjust the array base to point to the data.
4606 addptr(r_array_base, Array<Klass*>::base_offset_in_bytes());
4607
4608 testl(r_array_length, r_array_length); // array_length == 0?
4609 jcc(Assembler::zero, L_failure);
4610
4611 movl(r_array_index, 0);
4612 repne_scanq(r_array_base, r_super_klass, r_array_index, r_array_length, &L_success);
4613 // fall through to L_failure
4614
4615 const Register linear_result = r_array_index; // reuse temp1
4616
4617 bind(L_failure); // not present
4618 movl(linear_result, 1);
4619 jmp(L_check);
4620
4621 bind(L_success); // present
4622 movl(linear_result, 0);
4623
4624 bind(L_check);
4625 cmpl(linear_result, result);
4626 jcc(Assembler::equal, L_done);
4627
4628 { // To avoid calling convention issues, build a record on the stack
4629 // and pass the pointer to that instead.
4630 push(result);
4631 push(linear_result);
4632 push(r_sub_klass);
4633 push(r_super_klass);
4634 movptr(c_rarg1, rsp);
4635 movptr(c_rarg0, (uintptr_t) "mismatch");
4636 call(RuntimeAddress(CAST_FROM_FN_PTR(address, verify_secondary_supers_table_helper)));
4637 should_not_reach_here();
4638 }
4639 bind(L_done);
4640
4641 BLOCK_COMMENT("} verify_secondary_supers_table");
4642 }
4643
4644 #undef LOOKUP_SECONDARY_SUPERS_TABLE_REGISTERS
4645
4646 void MacroAssembler::clinit_barrier(Register klass, Label* L_fast_path, Label* L_slow_path) {
4647 assert(L_fast_path != nullptr || L_slow_path != nullptr, "at least one is required");
4648
4649 Label L_fallthrough;
4650 if (L_fast_path == nullptr) {
4651 L_fast_path = &L_fallthrough;
4652 } else if (L_slow_path == nullptr) {
4653 L_slow_path = &L_fallthrough;
4654 }
4655
4656 // Fast path check: class is fully initialized.
4657 // init_state needs acquire, but x86 is TSO, and so we are already good.
4658 cmpb(Address(klass, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
4659 jcc(Assembler::equal, *L_fast_path);
4660
4661 // Fast path check: current thread is initializer thread
4662 cmpptr(r15_thread, Address(klass, InstanceKlass::init_thread_offset()));
4663 if (L_slow_path == &L_fallthrough) {
4664 jcc(Assembler::equal, *L_fast_path);
4665 bind(*L_slow_path);
4666 } else if (L_fast_path == &L_fallthrough) {
4667 jcc(Assembler::notEqual, *L_slow_path);
4668 bind(*L_fast_path);
4669 } else {
4670 Unimplemented();
4671 }
4672 }
4673
4674 void MacroAssembler::cmov32(Condition cc, Register dst, Address src) {
4675 if (VM_Version::supports_cmov()) {
4676 cmovl(cc, dst, src);
4677 } else {
4678 Label L;
4679 jccb(negate_condition(cc), L);
4680 movl(dst, src);
4681 bind(L);
4682 }
4683 }
4684
4685 void MacroAssembler::cmov32(Condition cc, Register dst, Register src) {
4686 if (VM_Version::supports_cmov()) {
4687 cmovl(cc, dst, src);
4688 } else {
4689 Label L;
4690 jccb(negate_condition(cc), L);
4691 movl(dst, src);
4692 bind(L);
4693 }
4694 }
4695
4696 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
4697 if (!VerifyOops) return;
4698
4699 BLOCK_COMMENT("verify_oop {");
4700 push(rscratch1);
4701 push(rax); // save rax
4702 push(reg); // pass register argument
4703
4704 // Pass register number to verify_oop_subroutine
4705 const char* b = nullptr;
4706 {
4707 ResourceMark rm;
4708 stringStream ss;
4709 ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
4710 b = code_string(ss.as_string());
4711 }
4712 AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
4713 pushptr(buffer.addr(), rscratch1);
4714
4715 // call indirectly to solve generation ordering problem
4716 movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
4717 call(rax);
4718 // Caller pops the arguments (oop, message) and restores rax, r10
4719 BLOCK_COMMENT("} verify_oop");
4720 }
4721
4722 void MacroAssembler::vallones(XMMRegister dst, int vector_len) {
4723 if (UseAVX > 2 && (vector_len == Assembler::AVX_512bit || VM_Version::supports_avx512vl())) {
4724 // Only pcmpeq has dependency breaking treatment (i.e the execution can begin without
4725 // waiting for the previous result on dst), not vpcmpeqd, so just use vpternlog
4726 vpternlogd(dst, 0xFF, dst, dst, vector_len);
4727 } else if (VM_Version::supports_avx()) {
4728 vpcmpeqd(dst, dst, dst, vector_len);
4729 } else {
4730 pcmpeqd(dst, dst);
4731 }
4732 }
4733
4734 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
4735 int extra_slot_offset) {
4736 // cf. TemplateTable::prepare_invoke(), if (load_receiver).
4737 int stackElementSize = Interpreter::stackElementSize;
4738 int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0);
4739 #ifdef ASSERT
4740 int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1);
4741 assert(offset1 - offset == stackElementSize, "correct arithmetic");
4742 #endif
4743 Register scale_reg = noreg;
4744 Address::ScaleFactor scale_factor = Address::no_scale;
4745 if (arg_slot.is_constant()) {
4746 offset += arg_slot.as_constant() * stackElementSize;
4747 } else {
4748 scale_reg = arg_slot.as_register();
4749 scale_factor = Address::times(stackElementSize);
4750 }
4751 offset += wordSize; // return PC is on stack
4752 return Address(rsp, scale_reg, scale_factor, offset);
4753 }
4754
4755 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
4756 if (!VerifyOops) return;
4757
4758 push(rscratch1);
4759 push(rax); // save rax,
4760 // addr may contain rsp so we will have to adjust it based on the push
4761 // we just did (and on 64 bit we do two pushes)
4762 // NOTE: 64bit seemed to have had a bug in that it did movq(addr, rax); which
4763 // stores rax into addr which is backwards of what was intended.
4764 if (addr.uses(rsp)) {
4765 lea(rax, addr);
4766 pushptr(Address(rax, 2 * BytesPerWord));
4767 } else {
4768 pushptr(addr);
4769 }
4770
4771 // Pass register number to verify_oop_subroutine
4772 const char* b = nullptr;
4773 {
4774 ResourceMark rm;
4775 stringStream ss;
4776 ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);
4777 b = code_string(ss.as_string());
4778 }
4779 AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
4780 pushptr(buffer.addr(), rscratch1);
4781
4782 // call indirectly to solve generation ordering problem
4783 movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
4784 call(rax);
4785 // Caller pops the arguments (addr, message) and restores rax, r10.
4786 }
4787
4788 void MacroAssembler::verify_tlab() {
4789 #ifdef ASSERT
4790 if (UseTLAB && VerifyOops) {
4791 Label next, ok;
4792 Register t1 = rsi;
4793
4794 push(t1);
4795
4796 movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
4797 cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_start_offset())));
4798 jcc(Assembler::aboveEqual, next);
4799 STOP("assert(top >= start)");
4800 should_not_reach_here();
4801
4802 bind(next);
4803 movptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_end_offset())));
4804 cmpptr(t1, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
4805 jcc(Assembler::aboveEqual, ok);
4806 STOP("assert(top <= end)");
4807 should_not_reach_here();
4808
4809 bind(ok);
4810 pop(t1);
4811 }
4812 #endif
4813 }
4814
4815 class ControlWord {
4816 public:
4817 int32_t _value;
4818
4819 int rounding_control() const { return (_value >> 10) & 3 ; }
4820 int precision_control() const { return (_value >> 8) & 3 ; }
4821 bool precision() const { return ((_value >> 5) & 1) != 0; }
4822 bool underflow() const { return ((_value >> 4) & 1) != 0; }
4823 bool overflow() const { return ((_value >> 3) & 1) != 0; }
4824 bool zero_divide() const { return ((_value >> 2) & 1) != 0; }
4825 bool denormalized() const { return ((_value >> 1) & 1) != 0; }
4826 bool invalid() const { return ((_value >> 0) & 1) != 0; }
4827
4828 void print() const {
4829 // rounding control
4830 const char* rc;
4831 switch (rounding_control()) {
4832 case 0: rc = "round near"; break;
4833 case 1: rc = "round down"; break;
4834 case 2: rc = "round up "; break;
4835 case 3: rc = "chop "; break;
4836 default:
4837 rc = nullptr; // silence compiler warnings
4838 fatal("Unknown rounding control: %d", rounding_control());
4839 };
4840 // precision control
4841 const char* pc;
4842 switch (precision_control()) {
4843 case 0: pc = "24 bits "; break;
4844 case 1: pc = "reserved"; break;
4845 case 2: pc = "53 bits "; break;
4846 case 3: pc = "64 bits "; break;
4847 default:
4848 pc = nullptr; // silence compiler warnings
4849 fatal("Unknown precision control: %d", precision_control());
4850 };
4851 // flags
4852 char f[9];
4853 f[0] = ' ';
4854 f[1] = ' ';
4855 f[2] = (precision ()) ? 'P' : 'p';
4856 f[3] = (underflow ()) ? 'U' : 'u';
4857 f[4] = (overflow ()) ? 'O' : 'o';
4858 f[5] = (zero_divide ()) ? 'Z' : 'z';
4859 f[6] = (denormalized()) ? 'D' : 'd';
4860 f[7] = (invalid ()) ? 'I' : 'i';
4861 f[8] = '\x0';
4862 // output
4863 printf("%04x masks = %s, %s, %s", _value & 0xFFFF, f, rc, pc);
4864 }
4865
4866 };
4867
4868 class StatusWord {
4869 public:
4870 int32_t _value;
4871
4872 bool busy() const { return ((_value >> 15) & 1) != 0; }
4873 bool C3() const { return ((_value >> 14) & 1) != 0; }
4874 bool C2() const { return ((_value >> 10) & 1) != 0; }
4875 bool C1() const { return ((_value >> 9) & 1) != 0; }
4876 bool C0() const { return ((_value >> 8) & 1) != 0; }
4877 int top() const { return (_value >> 11) & 7 ; }
4878 bool error_status() const { return ((_value >> 7) & 1) != 0; }
4879 bool stack_fault() const { return ((_value >> 6) & 1) != 0; }
4880 bool precision() const { return ((_value >> 5) & 1) != 0; }
4881 bool underflow() const { return ((_value >> 4) & 1) != 0; }
4882 bool overflow() const { return ((_value >> 3) & 1) != 0; }
4883 bool zero_divide() const { return ((_value >> 2) & 1) != 0; }
4884 bool denormalized() const { return ((_value >> 1) & 1) != 0; }
4885 bool invalid() const { return ((_value >> 0) & 1) != 0; }
4886
4887 void print() const {
4888 // condition codes
4889 char c[5];
4890 c[0] = (C3()) ? '3' : '-';
4891 c[1] = (C2()) ? '2' : '-';
4892 c[2] = (C1()) ? '1' : '-';
4893 c[3] = (C0()) ? '0' : '-';
4894 c[4] = '\x0';
4895 // flags
4896 char f[9];
4897 f[0] = (error_status()) ? 'E' : '-';
4898 f[1] = (stack_fault ()) ? 'S' : '-';
4899 f[2] = (precision ()) ? 'P' : '-';
4900 f[3] = (underflow ()) ? 'U' : '-';
4901 f[4] = (overflow ()) ? 'O' : '-';
4902 f[5] = (zero_divide ()) ? 'Z' : '-';
4903 f[6] = (denormalized()) ? 'D' : '-';
4904 f[7] = (invalid ()) ? 'I' : '-';
4905 f[8] = '\x0';
4906 // output
4907 printf("%04x flags = %s, cc = %s, top = %d", _value & 0xFFFF, f, c, top());
4908 }
4909
4910 };
4911
4912 class TagWord {
4913 public:
4914 int32_t _value;
4915
4916 int tag_at(int i) const { return (_value >> (i*2)) & 3; }
4917
4918 void print() const {
4919 printf("%04x", _value & 0xFFFF);
4920 }
4921
4922 };
4923
4924 class FPU_Register {
4925 public:
4926 int32_t _m0;
4927 int32_t _m1;
4928 int16_t _ex;
4929
4930 bool is_indefinite() const {
4931 return _ex == -1 && _m1 == (int32_t)0xC0000000 && _m0 == 0;
4932 }
4933
4934 void print() const {
4935 char sign = (_ex < 0) ? '-' : '+';
4936 const char* kind = (_ex == 0x7FFF || _ex == (int16_t)-1) ? "NaN" : " ";
4937 printf("%c%04hx.%08x%08x %s", sign, _ex, _m1, _m0, kind);
4938 };
4939
4940 };
4941
4942 class FPU_State {
4943 public:
4944 enum {
4945 register_size = 10,
4946 number_of_registers = 8,
4947 register_mask = 7
4948 };
4949
4950 ControlWord _control_word;
4951 StatusWord _status_word;
4952 TagWord _tag_word;
4953 int32_t _error_offset;
4954 int32_t _error_selector;
4955 int32_t _data_offset;
4956 int32_t _data_selector;
4957 int8_t _register[register_size * number_of_registers];
4958
4959 int tag_for_st(int i) const { return _tag_word.tag_at((_status_word.top() + i) & register_mask); }
4960 FPU_Register* st(int i) const { return (FPU_Register*)&_register[register_size * i]; }
4961
4962 const char* tag_as_string(int tag) const {
4963 switch (tag) {
4964 case 0: return "valid";
4965 case 1: return "zero";
4966 case 2: return "special";
4967 case 3: return "empty";
4968 }
4969 ShouldNotReachHere();
4970 return nullptr;
4971 }
4972
4973 void print() const {
4974 // print computation registers
4975 { int t = _status_word.top();
4976 for (int i = 0; i < number_of_registers; i++) {
4977 int j = (i - t) & register_mask;
4978 printf("%c r%d = ST%d = ", (j == 0 ? '*' : ' '), i, j);
4979 st(j)->print();
4980 printf(" %s\n", tag_as_string(_tag_word.tag_at(i)));
4981 }
4982 }
4983 printf("\n");
4984 // print control registers
4985 printf("ctrl = "); _control_word.print(); printf("\n");
4986 printf("stat = "); _status_word .print(); printf("\n");
4987 printf("tags = "); _tag_word .print(); printf("\n");
4988 }
4989
4990 };
4991
4992 class Flag_Register {
4993 public:
4994 int32_t _value;
4995
4996 bool overflow() const { return ((_value >> 11) & 1) != 0; }
4997 bool direction() const { return ((_value >> 10) & 1) != 0; }
4998 bool sign() const { return ((_value >> 7) & 1) != 0; }
4999 bool zero() const { return ((_value >> 6) & 1) != 0; }
5000 bool auxiliary_carry() const { return ((_value >> 4) & 1) != 0; }
5001 bool parity() const { return ((_value >> 2) & 1) != 0; }
5002 bool carry() const { return ((_value >> 0) & 1) != 0; }
5003
5004 void print() const {
5005 // flags
5006 char f[8];
5007 f[0] = (overflow ()) ? 'O' : '-';
5008 f[1] = (direction ()) ? 'D' : '-';
5009 f[2] = (sign ()) ? 'S' : '-';
5010 f[3] = (zero ()) ? 'Z' : '-';
5011 f[4] = (auxiliary_carry()) ? 'A' : '-';
5012 f[5] = (parity ()) ? 'P' : '-';
5013 f[6] = (carry ()) ? 'C' : '-';
5014 f[7] = '\x0';
5015 // output
5016 printf("%08x flags = %s", _value, f);
5017 }
5018
5019 };
5020
5021 class IU_Register {
5022 public:
5023 int32_t _value;
5024
5025 void print() const {
5026 printf("%08x %11d", _value, _value);
5027 }
5028
5029 };
5030
5031 class IU_State {
5032 public:
5033 Flag_Register _eflags;
5034 IU_Register _rdi;
5035 IU_Register _rsi;
5036 IU_Register _rbp;
5037 IU_Register _rsp;
5038 IU_Register _rbx;
5039 IU_Register _rdx;
5040 IU_Register _rcx;
5041 IU_Register _rax;
5042
5043 void print() const {
5044 // computation registers
5045 printf("rax, = "); _rax.print(); printf("\n");
5046 printf("rbx, = "); _rbx.print(); printf("\n");
5047 printf("rcx = "); _rcx.print(); printf("\n");
5048 printf("rdx = "); _rdx.print(); printf("\n");
5049 printf("rdi = "); _rdi.print(); printf("\n");
5050 printf("rsi = "); _rsi.print(); printf("\n");
5051 printf("rbp, = "); _rbp.print(); printf("\n");
5052 printf("rsp = "); _rsp.print(); printf("\n");
5053 printf("\n");
5054 // control registers
5055 printf("flgs = "); _eflags.print(); printf("\n");
5056 }
5057 };
5058
5059
5060 class CPU_State {
5061 public:
5062 FPU_State _fpu_state;
5063 IU_State _iu_state;
5064
5065 void print() const {
5066 printf("--------------------------------------------------\n");
5067 _iu_state .print();
5068 printf("\n");
5069 _fpu_state.print();
5070 printf("--------------------------------------------------\n");
5071 }
5072
5073 };
5074
5075
5076 static void _print_CPU_state(CPU_State* state) {
5077 state->print();
5078 };
5079
5080
5081 void MacroAssembler::print_CPU_state() {
5082 push_CPU_state();
5083 push(rsp); // pass CPU state
5084 call(RuntimeAddress(CAST_FROM_FN_PTR(address, _print_CPU_state)));
5085 addptr(rsp, wordSize); // discard argument
5086 pop_CPU_state();
5087 }
5088
5089 void MacroAssembler::restore_cpu_control_state_after_jni(Register rscratch) {
5090 // Either restore the MXCSR register after returning from the JNI Call
5091 // or verify that it wasn't changed (with -Xcheck:jni flag).
5092 if (VM_Version::supports_sse()) {
5093 if (RestoreMXCSROnJNICalls) {
5094 ldmxcsr(ExternalAddress(StubRoutines::x86::addr_mxcsr_std()), rscratch);
5095 } else if (CheckJNICalls) {
5096 call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
5097 }
5098 }
5099 // Clear upper bits of YMM registers to avoid SSE <-> AVX transition penalty.
5100 vzeroupper();
5101 }
5102
5103 // ((OopHandle)result).resolve();
5104 void MacroAssembler::resolve_oop_handle(Register result, Register tmp) {
5105 assert_different_registers(result, tmp);
5106
5107 // Only 64 bit platforms support GCs that require a tmp register
5108 // Only IN_HEAP loads require a thread_tmp register
5109 // OopHandle::resolve is an indirection like jobject.
5110 access_load_at(T_OBJECT, IN_NATIVE,
5111 result, Address(result, 0), tmp);
5112 }
5113
5114 // ((WeakHandle)result).resolve();
5115 void MacroAssembler::resolve_weak_handle(Register rresult, Register rtmp) {
5116 assert_different_registers(rresult, rtmp);
5117 Label resolved;
5118
5119 // A null weak handle resolves to null.
5120 cmpptr(rresult, 0);
5121 jcc(Assembler::equal, resolved);
5122
5123 // Only 64 bit platforms support GCs that require a tmp register
5124 // Only IN_HEAP loads require a thread_tmp register
5125 // WeakHandle::resolve is an indirection like jweak.
5126 access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
5127 rresult, Address(rresult, 0), rtmp);
5128 bind(resolved);
5129 }
5130
5131 void MacroAssembler::load_mirror(Register mirror, Register method, Register tmp) {
5132 // get mirror
5133 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
5134 load_method_holder(mirror, method);
5135 movptr(mirror, Address(mirror, mirror_offset));
5136 resolve_oop_handle(mirror, tmp);
5137 }
5138
5139 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
5140 load_method_holder(rresult, rmethod);
5141 movptr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
5142 }
5143
5144 void MacroAssembler::load_method_holder(Register holder, Register method) {
5145 movptr(holder, Address(method, Method::const_offset())); // ConstMethod*
5146 movptr(holder, Address(holder, ConstMethod::constants_offset())); // ConstantPool*
5147 movptr(holder, Address(holder, ConstantPool::pool_holder_offset())); // InstanceKlass*
5148 }
5149
5150 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
5151 assert(UseCompactObjectHeaders, "expect compact object headers");
5152 movq(dst, Address(src, oopDesc::mark_offset_in_bytes()));
5153 shrq(dst, markWord::klass_shift);
5154 }
5155
5156 void MacroAssembler::load_klass(Register dst, Register src, Register tmp) {
5157 assert_different_registers(src, tmp);
5158 assert_different_registers(dst, tmp);
5159
5160 if (UseCompactObjectHeaders) {
5161 load_narrow_klass_compact(dst, src);
5162 decode_klass_not_null(dst, tmp);
5163 } else if (UseCompressedClassPointers) {
5164 movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5165 decode_klass_not_null(dst, tmp);
5166 } else {
5167 movptr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
5168 }
5169 }
5170
5171 void MacroAssembler::store_klass(Register dst, Register src, Register tmp) {
5172 assert(!UseCompactObjectHeaders, "not with compact headers");
5173 assert_different_registers(src, tmp);
5174 assert_different_registers(dst, tmp);
5175 if (UseCompressedClassPointers) {
5176 encode_klass_not_null(src, tmp);
5177 movl(Address(dst, oopDesc::klass_offset_in_bytes()), src);
5178 } else {
5179 movptr(Address(dst, oopDesc::klass_offset_in_bytes()), src);
5180 }
5181 }
5182
5183 void MacroAssembler::cmp_klass(Register klass, Register obj, Register tmp) {
5184 if (UseCompactObjectHeaders) {
5185 assert(tmp != noreg, "need tmp");
5186 assert_different_registers(klass, obj, tmp);
5187 load_narrow_klass_compact(tmp, obj);
5188 cmpl(klass, tmp);
5189 } else if (UseCompressedClassPointers) {
5190 cmpl(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
5191 } else {
5192 cmpptr(klass, Address(obj, oopDesc::klass_offset_in_bytes()));
5193 }
5194 }
5195
5196 void MacroAssembler::cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2) {
5197 if (UseCompactObjectHeaders) {
5198 assert(tmp2 != noreg, "need tmp2");
5199 assert_different_registers(obj1, obj2, tmp1, tmp2);
5200 load_narrow_klass_compact(tmp1, obj1);
5201 load_narrow_klass_compact(tmp2, obj2);
5202 cmpl(tmp1, tmp2);
5203 } else if (UseCompressedClassPointers) {
5204 movl(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5205 cmpl(tmp1, Address(obj2, oopDesc::klass_offset_in_bytes()));
5206 } else {
5207 movptr(tmp1, Address(obj1, oopDesc::klass_offset_in_bytes()));
5208 cmpptr(tmp1, Address(obj2, oopDesc::klass_offset_in_bytes()));
5209 }
5210 }
5211
5212 void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
5213 Register tmp1) {
5214 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
5215 decorators = AccessInternal::decorator_fixup(decorators, type);
5216 bool as_raw = (decorators & AS_RAW) != 0;
5217 if (as_raw) {
5218 bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1);
5219 } else {
5220 bs->load_at(this, decorators, type, dst, src, tmp1);
5221 }
5222 }
5223
5224 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
5225 Register tmp1, Register tmp2, Register tmp3) {
5226 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
5227 decorators = AccessInternal::decorator_fixup(decorators, type);
5228 bool as_raw = (decorators & AS_RAW) != 0;
5229 if (as_raw) {
5230 bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5231 } else {
5232 bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
5233 }
5234 }
5235
5236 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
5237 access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1);
5238 }
5239
5240 // Doesn't do verification, generates fixed size code
5241 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
5242 access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1);
5243 }
5244
5245 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
5246 Register tmp2, Register tmp3, DecoratorSet decorators) {
5247 access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
5248 }
5249
5250 // Used for storing nulls.
5251 void MacroAssembler::store_heap_oop_null(Address dst) {
5252 access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
5253 }
5254
5255 void MacroAssembler::store_klass_gap(Register dst, Register src) {
5256 assert(!UseCompactObjectHeaders, "Don't use with compact headers");
5257 if (UseCompressedClassPointers) {
5258 // Store to klass gap in destination
5259 movl(Address(dst, oopDesc::klass_gap_offset_in_bytes()), src);
5260 }
5261 }
5262
5263 #ifdef ASSERT
5264 void MacroAssembler::verify_heapbase(const char* msg) {
5265 assert (UseCompressedOops, "should be compressed");
5266 assert (Universe::heap() != nullptr, "java heap should be initialized");
5267 if (CheckCompressedOops) {
5268 Label ok;
5269 ExternalAddress src2(CompressedOops::base_addr());
5270 const bool is_src2_reachable = reachable(src2);
5271 if (!is_src2_reachable) {
5272 push(rscratch1); // cmpptr trashes rscratch1
5273 }
5274 cmpptr(r12_heapbase, src2, rscratch1);
5275 jcc(Assembler::equal, ok);
5276 STOP(msg);
5277 bind(ok);
5278 if (!is_src2_reachable) {
5279 pop(rscratch1);
5280 }
5281 }
5282 }
5283 #endif
5284
5285 // Algorithm must match oop.inline.hpp encode_heap_oop.
5286 void MacroAssembler::encode_heap_oop(Register r) {
5287 #ifdef ASSERT
5288 verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?");
5289 #endif
5290 verify_oop_msg(r, "broken oop in encode_heap_oop");
5291 if (CompressedOops::base() == nullptr) {
5292 if (CompressedOops::shift() != 0) {
5293 assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5294 shrq(r, LogMinObjAlignmentInBytes);
5295 }
5296 return;
5297 }
5298 testq(r, r);
5299 cmovq(Assembler::equal, r, r12_heapbase);
5300 subq(r, r12_heapbase);
5301 shrq(r, LogMinObjAlignmentInBytes);
5302 }
5303
5304 void MacroAssembler::encode_heap_oop_not_null(Register r) {
5305 #ifdef ASSERT
5306 verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?");
5307 if (CheckCompressedOops) {
5308 Label ok;
5309 testq(r, r);
5310 jcc(Assembler::notEqual, ok);
5311 STOP("null oop passed to encode_heap_oop_not_null");
5312 bind(ok);
5313 }
5314 #endif
5315 verify_oop_msg(r, "broken oop in encode_heap_oop_not_null");
5316 if (CompressedOops::base() != nullptr) {
5317 subq(r, r12_heapbase);
5318 }
5319 if (CompressedOops::shift() != 0) {
5320 assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5321 shrq(r, LogMinObjAlignmentInBytes);
5322 }
5323 }
5324
5325 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) {
5326 #ifdef ASSERT
5327 verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?");
5328 if (CheckCompressedOops) {
5329 Label ok;
5330 testq(src, src);
5331 jcc(Assembler::notEqual, ok);
5332 STOP("null oop passed to encode_heap_oop_not_null2");
5333 bind(ok);
5334 }
5335 #endif
5336 verify_oop_msg(src, "broken oop in encode_heap_oop_not_null2");
5337 if (dst != src) {
5338 movq(dst, src);
5339 }
5340 if (CompressedOops::base() != nullptr) {
5341 subq(dst, r12_heapbase);
5342 }
5343 if (CompressedOops::shift() != 0) {
5344 assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5345 shrq(dst, LogMinObjAlignmentInBytes);
5346 }
5347 }
5348
5349 void MacroAssembler::decode_heap_oop(Register r) {
5350 #ifdef ASSERT
5351 verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?");
5352 #endif
5353 if (CompressedOops::base() == nullptr) {
5354 if (CompressedOops::shift() != 0) {
5355 assert (LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5356 shlq(r, LogMinObjAlignmentInBytes);
5357 }
5358 } else {
5359 Label done;
5360 shlq(r, LogMinObjAlignmentInBytes);
5361 jccb(Assembler::equal, done);
5362 addq(r, r12_heapbase);
5363 bind(done);
5364 }
5365 verify_oop_msg(r, "broken oop in decode_heap_oop");
5366 }
5367
5368 void MacroAssembler::decode_heap_oop_not_null(Register r) {
5369 // Note: it will change flags
5370 assert (UseCompressedOops, "should only be used for compressed headers");
5371 assert (Universe::heap() != nullptr, "java heap should be initialized");
5372 // Cannot assert, unverified entry point counts instructions (see .ad file)
5373 // vtableStubs also counts instructions in pd_code_size_limit.
5374 // Also do not verify_oop as this is called by verify_oop.
5375 if (CompressedOops::shift() != 0) {
5376 assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5377 shlq(r, LogMinObjAlignmentInBytes);
5378 if (CompressedOops::base() != nullptr) {
5379 addq(r, r12_heapbase);
5380 }
5381 } else {
5382 assert (CompressedOops::base() == nullptr, "sanity");
5383 }
5384 }
5385
5386 void MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) {
5387 // Note: it will change flags
5388 assert (UseCompressedOops, "should only be used for compressed headers");
5389 assert (Universe::heap() != nullptr, "java heap should be initialized");
5390 // Cannot assert, unverified entry point counts instructions (see .ad file)
5391 // vtableStubs also counts instructions in pd_code_size_limit.
5392 // Also do not verify_oop as this is called by verify_oop.
5393 if (CompressedOops::shift() != 0) {
5394 assert(LogMinObjAlignmentInBytes == CompressedOops::shift(), "decode alg wrong");
5395 if (LogMinObjAlignmentInBytes == Address::times_8) {
5396 leaq(dst, Address(r12_heapbase, src, Address::times_8, 0));
5397 } else {
5398 if (dst != src) {
5399 movq(dst, src);
5400 }
5401 shlq(dst, LogMinObjAlignmentInBytes);
5402 if (CompressedOops::base() != nullptr) {
5403 addq(dst, r12_heapbase);
5404 }
5405 }
5406 } else {
5407 assert (CompressedOops::base() == nullptr, "sanity");
5408 if (dst != src) {
5409 movq(dst, src);
5410 }
5411 }
5412 }
5413
5414 void MacroAssembler::encode_klass_not_null(Register r, Register tmp) {
5415 BLOCK_COMMENT("encode_klass_not_null {");
5416 assert_different_registers(r, tmp);
5417 if (CompressedKlassPointers::base() != nullptr) {
5418 if (AOTCodeCache::is_on_for_dump()) {
5419 movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5420 } else {
5421 movptr(tmp, (intptr_t)CompressedKlassPointers::base());
5422 }
5423 subq(r, tmp);
5424 }
5425 if (CompressedKlassPointers::shift() != 0) {
5426 shrq(r, CompressedKlassPointers::shift());
5427 }
5428 BLOCK_COMMENT("} encode_klass_not_null");
5429 }
5430
5431 void MacroAssembler::encode_and_move_klass_not_null(Register dst, Register src) {
5432 BLOCK_COMMENT("encode_and_move_klass_not_null {");
5433 assert_different_registers(src, dst);
5434 if (CompressedKlassPointers::base() != nullptr) {
5435 if (AOTCodeCache::is_on_for_dump()) {
5436 movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5437 negl(dst);
5438 } else {
5439 movptr(dst, -(intptr_t)CompressedKlassPointers::base());
5440 }
5441 addq(dst, src);
5442 } else {
5443 movptr(dst, src);
5444 }
5445 if (CompressedKlassPointers::shift() != 0) {
5446 shrq(dst, CompressedKlassPointers::shift());
5447 }
5448 BLOCK_COMMENT("} encode_and_move_klass_not_null");
5449 }
5450
5451 void MacroAssembler::decode_klass_not_null(Register r, Register tmp) {
5452 BLOCK_COMMENT("decode_klass_not_null {");
5453 assert_different_registers(r, tmp);
5454 // Note: it will change flags
5455 assert(UseCompressedClassPointers, "should only be used for compressed headers");
5456 // Cannot assert, unverified entry point counts instructions (see .ad file)
5457 // vtableStubs also counts instructions in pd_code_size_limit.
5458 // Also do not verify_oop as this is called by verify_oop.
5459 if (CompressedKlassPointers::shift() != 0) {
5460 shlq(r, CompressedKlassPointers::shift());
5461 }
5462 if (CompressedKlassPointers::base() != nullptr) {
5463 if (AOTCodeCache::is_on_for_dump()) {
5464 movptr(tmp, ExternalAddress(CompressedKlassPointers::base_addr()));
5465 } else {
5466 movptr(tmp, (intptr_t)CompressedKlassPointers::base());
5467 }
5468 addq(r, tmp);
5469 }
5470 BLOCK_COMMENT("} decode_klass_not_null");
5471 }
5472
5473 void MacroAssembler::decode_and_move_klass_not_null(Register dst, Register src) {
5474 BLOCK_COMMENT("decode_and_move_klass_not_null {");
5475 assert_different_registers(src, dst);
5476 // Note: it will change flags
5477 assert (UseCompressedClassPointers, "should only be used for compressed headers");
5478 // Cannot assert, unverified entry point counts instructions (see .ad file)
5479 // vtableStubs also counts instructions in pd_code_size_limit.
5480 // Also do not verify_oop as this is called by verify_oop.
5481
5482 if (CompressedKlassPointers::base() == nullptr &&
5483 CompressedKlassPointers::shift() == 0) {
5484 // The best case scenario is that there is no base or shift. Then it is already
5485 // a pointer that needs nothing but a register rename.
5486 movl(dst, src);
5487 } else {
5488 if (CompressedKlassPointers::shift() <= Address::times_8) {
5489 if (CompressedKlassPointers::base() != nullptr) {
5490 if (AOTCodeCache::is_on_for_dump()) {
5491 movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5492 } else {
5493 movptr(dst, (intptr_t)CompressedKlassPointers::base());
5494 }
5495 } else {
5496 xorq(dst, dst);
5497 }
5498 if (CompressedKlassPointers::shift() != 0) {
5499 assert(CompressedKlassPointers::shift() == Address::times_8, "klass not aligned on 64bits?");
5500 leaq(dst, Address(dst, src, Address::times_8, 0));
5501 } else {
5502 addq(dst, src);
5503 }
5504 } else {
5505 if (CompressedKlassPointers::base() != nullptr) {
5506 if (AOTCodeCache::is_on_for_dump()) {
5507 movptr(dst, ExternalAddress(CompressedKlassPointers::base_addr()));
5508 shrq(dst, CompressedKlassPointers::shift());
5509 } else {
5510 const intptr_t base_right_shifted =
5511 (intptr_t)CompressedKlassPointers::base() >> CompressedKlassPointers::shift();
5512 movptr(dst, base_right_shifted);
5513 }
5514 } else {
5515 xorq(dst, dst);
5516 }
5517 addq(dst, src);
5518 shlq(dst, CompressedKlassPointers::shift());
5519 }
5520 }
5521 BLOCK_COMMENT("} decode_and_move_klass_not_null");
5522 }
5523
5524 void MacroAssembler::set_narrow_oop(Register 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 mov_narrow_oop(dst, oop_index, rspec);
5531 }
5532
5533 void MacroAssembler::set_narrow_oop(Address dst, jobject obj) {
5534 assert (UseCompressedOops, "should only be used for compressed headers");
5535 assert (Universe::heap() != nullptr, "java heap should be initialized");
5536 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5537 int oop_index = oop_recorder()->find_index(obj);
5538 RelocationHolder rspec = oop_Relocation::spec(oop_index);
5539 mov_narrow_oop(dst, oop_index, rspec);
5540 }
5541
5542 void MacroAssembler::set_narrow_klass(Register dst, Klass* k) {
5543 assert (UseCompressedClassPointers, "should only be used for compressed headers");
5544 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5545 int klass_index = oop_recorder()->find_index(k);
5546 RelocationHolder rspec = metadata_Relocation::spec(klass_index);
5547 mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
5548 }
5549
5550 void MacroAssembler::set_narrow_klass(Address dst, Klass* k) {
5551 assert (UseCompressedClassPointers, "should only be used for compressed headers");
5552 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5553 int klass_index = oop_recorder()->find_index(k);
5554 RelocationHolder rspec = metadata_Relocation::spec(klass_index);
5555 mov_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
5556 }
5557
5558 void MacroAssembler::cmp_narrow_oop(Register dst, jobject obj) {
5559 assert (UseCompressedOops, "should only be used for compressed headers");
5560 assert (Universe::heap() != nullptr, "java heap should be initialized");
5561 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5562 int oop_index = oop_recorder()->find_index(obj);
5563 RelocationHolder rspec = oop_Relocation::spec(oop_index);
5564 Assembler::cmp_narrow_oop(dst, oop_index, rspec);
5565 }
5566
5567 void MacroAssembler::cmp_narrow_oop(Address dst, jobject obj) {
5568 assert (UseCompressedOops, "should only be used for compressed headers");
5569 assert (Universe::heap() != nullptr, "java heap should be initialized");
5570 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5571 int oop_index = oop_recorder()->find_index(obj);
5572 RelocationHolder rspec = oop_Relocation::spec(oop_index);
5573 Assembler::cmp_narrow_oop(dst, oop_index, rspec);
5574 }
5575
5576 void MacroAssembler::cmp_narrow_klass(Register dst, Klass* k) {
5577 assert (UseCompressedClassPointers, "should only be used for compressed headers");
5578 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5579 int klass_index = oop_recorder()->find_index(k);
5580 RelocationHolder rspec = metadata_Relocation::spec(klass_index);
5581 Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
5582 }
5583
5584 void MacroAssembler::cmp_narrow_klass(Address dst, Klass* k) {
5585 assert (UseCompressedClassPointers, "should only be used for compressed headers");
5586 assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
5587 int klass_index = oop_recorder()->find_index(k);
5588 RelocationHolder rspec = metadata_Relocation::spec(klass_index);
5589 Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
5590 }
5591
5592 void MacroAssembler::reinit_heapbase() {
5593 if (UseCompressedOops) {
5594 if (Universe::heap() != nullptr) {
5595 if (CompressedOops::base() == nullptr) {
5596 MacroAssembler::xorptr(r12_heapbase, r12_heapbase);
5597 } else {
5598 mov64(r12_heapbase, (int64_t)CompressedOops::base());
5599 }
5600 } else {
5601 movptr(r12_heapbase, ExternalAddress(CompressedOops::base_addr()));
5602 }
5603 }
5604 }
5605
5606 #if COMPILER2_OR_JVMCI
5607
5608 // clear memory of size 'cnt' qwords, starting at 'base' using XMM/YMM/ZMM registers
5609 void MacroAssembler::xmm_clear_mem(Register base, Register cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
5610 // cnt - number of qwords (8-byte words).
5611 // base - start address, qword aligned.
5612 Label L_zero_64_bytes, L_loop, L_sloop, L_tail, L_end;
5613 bool use64byteVector = (MaxVectorSize == 64) && (VM_Version::avx3_threshold() == 0);
5614 if (use64byteVector) {
5615 vpxor(xtmp, xtmp, xtmp, AVX_512bit);
5616 } else if (MaxVectorSize >= 32) {
5617 vpxor(xtmp, xtmp, xtmp, AVX_256bit);
5618 } else {
5619 pxor(xtmp, xtmp);
5620 }
5621 jmp(L_zero_64_bytes);
5622
5623 BIND(L_loop);
5624 if (MaxVectorSize >= 32) {
5625 fill64(base, 0, xtmp, use64byteVector);
5626 } else {
5627 movdqu(Address(base, 0), xtmp);
5628 movdqu(Address(base, 16), xtmp);
5629 movdqu(Address(base, 32), xtmp);
5630 movdqu(Address(base, 48), xtmp);
5631 }
5632 addptr(base, 64);
5633
5634 BIND(L_zero_64_bytes);
5635 subptr(cnt, 8);
5636 jccb(Assembler::greaterEqual, L_loop);
5637
5638 // Copy trailing 64 bytes
5639 if (use64byteVector) {
5640 addptr(cnt, 8);
5641 jccb(Assembler::equal, L_end);
5642 fill64_masked(3, base, 0, xtmp, mask, cnt, rtmp, true);
5643 jmp(L_end);
5644 } else {
5645 addptr(cnt, 4);
5646 jccb(Assembler::less, L_tail);
5647 if (MaxVectorSize >= 32) {
5648 vmovdqu(Address(base, 0), xtmp);
5649 } else {
5650 movdqu(Address(base, 0), xtmp);
5651 movdqu(Address(base, 16), xtmp);
5652 }
5653 }
5654 addptr(base, 32);
5655 subptr(cnt, 4);
5656
5657 BIND(L_tail);
5658 addptr(cnt, 4);
5659 jccb(Assembler::lessEqual, L_end);
5660 if (UseAVX > 2 && MaxVectorSize >= 32 && VM_Version::supports_avx512vl()) {
5661 fill32_masked(3, base, 0, xtmp, mask, cnt, rtmp);
5662 } else {
5663 decrement(cnt);
5664
5665 BIND(L_sloop);
5666 movq(Address(base, 0), xtmp);
5667 addptr(base, 8);
5668 decrement(cnt);
5669 jccb(Assembler::greaterEqual, L_sloop);
5670 }
5671 BIND(L_end);
5672 }
5673
5674 // Clearing constant sized memory using YMM/ZMM registers.
5675 void MacroAssembler::clear_mem(Register base, int cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
5676 assert(UseAVX > 2 && VM_Version::supports_avx512vl(), "");
5677 bool use64byteVector = (MaxVectorSize > 32) && (VM_Version::avx3_threshold() == 0);
5678
5679 int vector64_count = (cnt & (~0x7)) >> 3;
5680 cnt = cnt & 0x7;
5681 const int fill64_per_loop = 4;
5682 const int max_unrolled_fill64 = 8;
5683
5684 // 64 byte initialization loop.
5685 vpxor(xtmp, xtmp, xtmp, use64byteVector ? AVX_512bit : AVX_256bit);
5686 int start64 = 0;
5687 if (vector64_count > max_unrolled_fill64) {
5688 Label LOOP;
5689 Register index = rtmp;
5690
5691 start64 = vector64_count - (vector64_count % fill64_per_loop);
5692
5693 movl(index, 0);
5694 BIND(LOOP);
5695 for (int i = 0; i < fill64_per_loop; i++) {
5696 fill64(Address(base, index, Address::times_1, i * 64), xtmp, use64byteVector);
5697 }
5698 addl(index, fill64_per_loop * 64);
5699 cmpl(index, start64 * 64);
5700 jccb(Assembler::less, LOOP);
5701 }
5702 for (int i = start64; i < vector64_count; i++) {
5703 fill64(base, i * 64, xtmp, use64byteVector);
5704 }
5705
5706 // Clear remaining 64 byte tail.
5707 int disp = vector64_count * 64;
5708 if (cnt) {
5709 switch (cnt) {
5710 case 1:
5711 movq(Address(base, disp), xtmp);
5712 break;
5713 case 2:
5714 evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_128bit);
5715 break;
5716 case 3:
5717 movl(rtmp, 0x7);
5718 kmovwl(mask, rtmp);
5719 evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_256bit);
5720 break;
5721 case 4:
5722 evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
5723 break;
5724 case 5:
5725 if (use64byteVector) {
5726 movl(rtmp, 0x1F);
5727 kmovwl(mask, rtmp);
5728 evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
5729 } else {
5730 evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
5731 movq(Address(base, disp + 32), xtmp);
5732 }
5733 break;
5734 case 6:
5735 if (use64byteVector) {
5736 movl(rtmp, 0x3F);
5737 kmovwl(mask, rtmp);
5738 evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
5739 } else {
5740 evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
5741 evmovdqu(T_LONG, k0, Address(base, disp + 32), xtmp, false, Assembler::AVX_128bit);
5742 }
5743 break;
5744 case 7:
5745 if (use64byteVector) {
5746 movl(rtmp, 0x7F);
5747 kmovwl(mask, rtmp);
5748 evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
5749 } else {
5750 evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
5751 movl(rtmp, 0x7);
5752 kmovwl(mask, rtmp);
5753 evmovdqu(T_LONG, mask, Address(base, disp + 32), xtmp, true, Assembler::AVX_256bit);
5754 }
5755 break;
5756 default:
5757 fatal("Unexpected length : %d\n",cnt);
5758 break;
5759 }
5760 }
5761 }
5762
5763 void MacroAssembler::clear_mem(Register base, Register cnt, Register tmp, XMMRegister xtmp,
5764 bool is_large, KRegister mask) {
5765 // cnt - number of qwords (8-byte words).
5766 // base - start address, qword aligned.
5767 // is_large - if optimizers know cnt is larger than InitArrayShortSize
5768 assert(base==rdi, "base register must be edi for rep stos");
5769 assert(tmp==rax, "tmp register must be eax for rep stos");
5770 assert(cnt==rcx, "cnt register must be ecx for rep stos");
5771 assert(InitArrayShortSize % BytesPerLong == 0,
5772 "InitArrayShortSize should be the multiple of BytesPerLong");
5773
5774 Label DONE;
5775 if (!is_large || !UseXMMForObjInit) {
5776 xorptr(tmp, tmp);
5777 }
5778
5779 if (!is_large) {
5780 Label LOOP, LONG;
5781 cmpptr(cnt, InitArrayShortSize/BytesPerLong);
5782 jccb(Assembler::greater, LONG);
5783
5784 decrement(cnt);
5785 jccb(Assembler::negative, DONE); // Zero length
5786
5787 // Use individual pointer-sized stores for small counts:
5788 BIND(LOOP);
5789 movptr(Address(base, cnt, Address::times_ptr), tmp);
5790 decrement(cnt);
5791 jccb(Assembler::greaterEqual, LOOP);
5792 jmpb(DONE);
5793
5794 BIND(LONG);
5795 }
5796
5797 // Use longer rep-prefixed ops for non-small counts:
5798 if (UseFastStosb) {
5799 shlptr(cnt, 3); // convert to number of bytes
5800 rep_stosb();
5801 } else if (UseXMMForObjInit) {
5802 xmm_clear_mem(base, cnt, tmp, xtmp, mask);
5803 } else {
5804 rep_stos();
5805 }
5806
5807 BIND(DONE);
5808 }
5809
5810 #endif //COMPILER2_OR_JVMCI
5811
5812
5813 void MacroAssembler::generate_fill(BasicType t, bool aligned,
5814 Register to, Register value, Register count,
5815 Register rtmp, XMMRegister xtmp) {
5816 ShortBranchVerifier sbv(this);
5817 assert_different_registers(to, value, count, rtmp);
5818 Label L_exit;
5819 Label L_fill_2_bytes, L_fill_4_bytes;
5820
5821 #if defined(COMPILER2)
5822 if(MaxVectorSize >=32 &&
5823 VM_Version::supports_avx512vlbw() &&
5824 VM_Version::supports_bmi2()) {
5825 generate_fill_avx3(t, to, value, count, rtmp, xtmp);
5826 return;
5827 }
5828 #endif
5829
5830 int shift = -1;
5831 switch (t) {
5832 case T_BYTE:
5833 shift = 2;
5834 break;
5835 case T_SHORT:
5836 shift = 1;
5837 break;
5838 case T_INT:
5839 shift = 0;
5840 break;
5841 default: ShouldNotReachHere();
5842 }
5843
5844 if (t == T_BYTE) {
5845 andl(value, 0xff);
5846 movl(rtmp, value);
5847 shll(rtmp, 8);
5848 orl(value, rtmp);
5849 }
5850 if (t == T_SHORT) {
5851 andl(value, 0xffff);
5852 }
5853 if (t == T_BYTE || t == T_SHORT) {
5854 movl(rtmp, value);
5855 shll(rtmp, 16);
5856 orl(value, rtmp);
5857 }
5858
5859 cmpptr(count, 2<<shift); // Short arrays (< 8 bytes) fill by element
5860 jcc(Assembler::below, L_fill_4_bytes); // use unsigned cmp
5861 if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
5862 Label L_skip_align2;
5863 // align source address at 4 bytes address boundary
5864 if (t == T_BYTE) {
5865 Label L_skip_align1;
5866 // One byte misalignment happens only for byte arrays
5867 testptr(to, 1);
5868 jccb(Assembler::zero, L_skip_align1);
5869 movb(Address(to, 0), value);
5870 increment(to);
5871 decrement(count);
5872 BIND(L_skip_align1);
5873 }
5874 // Two bytes misalignment happens only for byte and short (char) arrays
5875 testptr(to, 2);
5876 jccb(Assembler::zero, L_skip_align2);
5877 movw(Address(to, 0), value);
5878 addptr(to, 2);
5879 subptr(count, 1<<(shift-1));
5880 BIND(L_skip_align2);
5881 }
5882 {
5883 Label L_fill_32_bytes;
5884 if (!UseUnalignedLoadStores) {
5885 // align to 8 bytes, we know we are 4 byte aligned to start
5886 testptr(to, 4);
5887 jccb(Assembler::zero, L_fill_32_bytes);
5888 movl(Address(to, 0), value);
5889 addptr(to, 4);
5890 subptr(count, 1<<shift);
5891 }
5892 BIND(L_fill_32_bytes);
5893 {
5894 Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
5895 movdl(xtmp, value);
5896 if (UseAVX >= 2 && UseUnalignedLoadStores) {
5897 Label L_check_fill_32_bytes;
5898 if (UseAVX > 2) {
5899 // Fill 64-byte chunks
5900 Label L_fill_64_bytes_loop_avx3, L_check_fill_64_bytes_avx2;
5901
5902 // If number of bytes to fill < VM_Version::avx3_threshold(), perform fill using AVX2
5903 cmpptr(count, VM_Version::avx3_threshold());
5904 jccb(Assembler::below, L_check_fill_64_bytes_avx2);
5905
5906 vpbroadcastd(xtmp, xtmp, Assembler::AVX_512bit);
5907
5908 subptr(count, 16 << shift);
5909 jccb(Assembler::less, L_check_fill_32_bytes);
5910 align(16);
5911
5912 BIND(L_fill_64_bytes_loop_avx3);
5913 evmovdqul(Address(to, 0), xtmp, Assembler::AVX_512bit);
5914 addptr(to, 64);
5915 subptr(count, 16 << shift);
5916 jcc(Assembler::greaterEqual, L_fill_64_bytes_loop_avx3);
5917 jmpb(L_check_fill_32_bytes);
5918
5919 BIND(L_check_fill_64_bytes_avx2);
5920 }
5921 // Fill 64-byte chunks
5922 Label L_fill_64_bytes_loop;
5923 vpbroadcastd(xtmp, xtmp, Assembler::AVX_256bit);
5924
5925 subptr(count, 16 << shift);
5926 jcc(Assembler::less, L_check_fill_32_bytes);
5927 align(16);
5928
5929 BIND(L_fill_64_bytes_loop);
5930 vmovdqu(Address(to, 0), xtmp);
5931 vmovdqu(Address(to, 32), xtmp);
5932 addptr(to, 64);
5933 subptr(count, 16 << shift);
5934 jcc(Assembler::greaterEqual, L_fill_64_bytes_loop);
5935
5936 BIND(L_check_fill_32_bytes);
5937 addptr(count, 8 << shift);
5938 jccb(Assembler::less, L_check_fill_8_bytes);
5939 vmovdqu(Address(to, 0), xtmp);
5940 addptr(to, 32);
5941 subptr(count, 8 << shift);
5942
5943 BIND(L_check_fill_8_bytes);
5944 // clean upper bits of YMM registers
5945 movdl(xtmp, value);
5946 pshufd(xtmp, xtmp, 0);
5947 } else {
5948 // Fill 32-byte chunks
5949 pshufd(xtmp, xtmp, 0);
5950
5951 subptr(count, 8 << shift);
5952 jcc(Assembler::less, L_check_fill_8_bytes);
5953 align(16);
5954
5955 BIND(L_fill_32_bytes_loop);
5956
5957 if (UseUnalignedLoadStores) {
5958 movdqu(Address(to, 0), xtmp);
5959 movdqu(Address(to, 16), xtmp);
5960 } else {
5961 movq(Address(to, 0), xtmp);
5962 movq(Address(to, 8), xtmp);
5963 movq(Address(to, 16), xtmp);
5964 movq(Address(to, 24), xtmp);
5965 }
5966
5967 addptr(to, 32);
5968 subptr(count, 8 << shift);
5969 jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
5970
5971 BIND(L_check_fill_8_bytes);
5972 }
5973 addptr(count, 8 << shift);
5974 jccb(Assembler::zero, L_exit);
5975 jmpb(L_fill_8_bytes);
5976
5977 //
5978 // length is too short, just fill qwords
5979 //
5980 BIND(L_fill_8_bytes_loop);
5981 movq(Address(to, 0), xtmp);
5982 addptr(to, 8);
5983 BIND(L_fill_8_bytes);
5984 subptr(count, 1 << (shift + 1));
5985 jcc(Assembler::greaterEqual, L_fill_8_bytes_loop);
5986 }
5987 }
5988 // fill trailing 4 bytes
5989 BIND(L_fill_4_bytes);
5990 testl(count, 1<<shift);
5991 jccb(Assembler::zero, L_fill_2_bytes);
5992 movl(Address(to, 0), value);
5993 if (t == T_BYTE || t == T_SHORT) {
5994 Label L_fill_byte;
5995 addptr(to, 4);
5996 BIND(L_fill_2_bytes);
5997 // fill trailing 2 bytes
5998 testl(count, 1<<(shift-1));
5999 jccb(Assembler::zero, L_fill_byte);
6000 movw(Address(to, 0), value);
6001 if (t == T_BYTE) {
6002 addptr(to, 2);
6003 BIND(L_fill_byte);
6004 // fill trailing byte
6005 testl(count, 1);
6006 jccb(Assembler::zero, L_exit);
6007 movb(Address(to, 0), value);
6008 } else {
6009 BIND(L_fill_byte);
6010 }
6011 } else {
6012 BIND(L_fill_2_bytes);
6013 }
6014 BIND(L_exit);
6015 }
6016
6017 void MacroAssembler::evpbroadcast(BasicType type, XMMRegister dst, Register src, int vector_len) {
6018 switch(type) {
6019 case T_BYTE:
6020 case T_BOOLEAN:
6021 evpbroadcastb(dst, src, vector_len);
6022 break;
6023 case T_SHORT:
6024 case T_CHAR:
6025 evpbroadcastw(dst, src, vector_len);
6026 break;
6027 case T_INT:
6028 case T_FLOAT:
6029 evpbroadcastd(dst, src, vector_len);
6030 break;
6031 case T_LONG:
6032 case T_DOUBLE:
6033 evpbroadcastq(dst, src, vector_len);
6034 break;
6035 default:
6036 fatal("Unhandled type : %s", type2name(type));
6037 break;
6038 }
6039 }
6040
6041 // encode char[] to byte[] in ISO_8859_1 or ASCII
6042 //@IntrinsicCandidate
6043 //private static int implEncodeISOArray(byte[] sa, int sp,
6044 //byte[] da, int dp, int len) {
6045 // int i = 0;
6046 // for (; i < len; i++) {
6047 // char c = StringUTF16.getChar(sa, sp++);
6048 // if (c > '\u00FF')
6049 // break;
6050 // da[dp++] = (byte)c;
6051 // }
6052 // return i;
6053 //}
6054 //
6055 //@IntrinsicCandidate
6056 //private static int implEncodeAsciiArray(char[] sa, int sp,
6057 // byte[] da, int dp, int len) {
6058 // int i = 0;
6059 // for (; i < len; i++) {
6060 // char c = sa[sp++];
6061 // if (c >= '\u0080')
6062 // break;
6063 // da[dp++] = (byte)c;
6064 // }
6065 // return i;
6066 //}
6067 void MacroAssembler::encode_iso_array(Register src, Register dst, Register len,
6068 XMMRegister tmp1Reg, XMMRegister tmp2Reg,
6069 XMMRegister tmp3Reg, XMMRegister tmp4Reg,
6070 Register tmp5, Register result, bool ascii) {
6071
6072 // rsi: src
6073 // rdi: dst
6074 // rdx: len
6075 // rcx: tmp5
6076 // rax: result
6077 ShortBranchVerifier sbv(this);
6078 assert_different_registers(src, dst, len, tmp5, result);
6079 Label L_done, L_copy_1_char, L_copy_1_char_exit;
6080
6081 int mask = ascii ? 0xff80ff80 : 0xff00ff00;
6082 int short_mask = ascii ? 0xff80 : 0xff00;
6083
6084 // set result
6085 xorl(result, result);
6086 // check for zero length
6087 testl(len, len);
6088 jcc(Assembler::zero, L_done);
6089
6090 movl(result, len);
6091
6092 // Setup pointers
6093 lea(src, Address(src, len, Address::times_2)); // char[]
6094 lea(dst, Address(dst, len, Address::times_1)); // byte[]
6095 negptr(len);
6096
6097 if (UseSSE42Intrinsics || UseAVX >= 2) {
6098 Label L_copy_8_chars, L_copy_8_chars_exit;
6099 Label L_chars_16_check, L_copy_16_chars, L_copy_16_chars_exit;
6100
6101 if (UseAVX >= 2) {
6102 Label L_chars_32_check, L_copy_32_chars, L_copy_32_chars_exit;
6103 movl(tmp5, mask); // create mask to test for Unicode or non-ASCII chars in vector
6104 movdl(tmp1Reg, tmp5);
6105 vpbroadcastd(tmp1Reg, tmp1Reg, Assembler::AVX_256bit);
6106 jmp(L_chars_32_check);
6107
6108 bind(L_copy_32_chars);
6109 vmovdqu(tmp3Reg, Address(src, len, Address::times_2, -64));
6110 vmovdqu(tmp4Reg, Address(src, len, Address::times_2, -32));
6111 vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
6112 vptest(tmp2Reg, tmp1Reg); // check for Unicode or non-ASCII chars in vector
6113 jccb(Assembler::notZero, L_copy_32_chars_exit);
6114 vpackuswb(tmp3Reg, tmp3Reg, tmp4Reg, /* vector_len */ 1);
6115 vpermq(tmp4Reg, tmp3Reg, 0xD8, /* vector_len */ 1);
6116 vmovdqu(Address(dst, len, Address::times_1, -32), tmp4Reg);
6117
6118 bind(L_chars_32_check);
6119 addptr(len, 32);
6120 jcc(Assembler::lessEqual, L_copy_32_chars);
6121
6122 bind(L_copy_32_chars_exit);
6123 subptr(len, 16);
6124 jccb(Assembler::greater, L_copy_16_chars_exit);
6125
6126 } else if (UseSSE42Intrinsics) {
6127 movl(tmp5, mask); // create mask to test for Unicode or non-ASCII chars in vector
6128 movdl(tmp1Reg, tmp5);
6129 pshufd(tmp1Reg, tmp1Reg, 0);
6130 jmpb(L_chars_16_check);
6131 }
6132
6133 bind(L_copy_16_chars);
6134 if (UseAVX >= 2) {
6135 vmovdqu(tmp2Reg, Address(src, len, Address::times_2, -32));
6136 vptest(tmp2Reg, tmp1Reg);
6137 jcc(Assembler::notZero, L_copy_16_chars_exit);
6138 vpackuswb(tmp2Reg, tmp2Reg, tmp1Reg, /* vector_len */ 1);
6139 vpermq(tmp3Reg, tmp2Reg, 0xD8, /* vector_len */ 1);
6140 } else {
6141 if (UseAVX > 0) {
6142 movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
6143 movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
6144 vpor(tmp2Reg, tmp3Reg, tmp4Reg, /* vector_len */ 0);
6145 } else {
6146 movdqu(tmp3Reg, Address(src, len, Address::times_2, -32));
6147 por(tmp2Reg, tmp3Reg);
6148 movdqu(tmp4Reg, Address(src, len, Address::times_2, -16));
6149 por(tmp2Reg, tmp4Reg);
6150 }
6151 ptest(tmp2Reg, tmp1Reg); // check for Unicode or non-ASCII chars in vector
6152 jccb(Assembler::notZero, L_copy_16_chars_exit);
6153 packuswb(tmp3Reg, tmp4Reg);
6154 }
6155 movdqu(Address(dst, len, Address::times_1, -16), tmp3Reg);
6156
6157 bind(L_chars_16_check);
6158 addptr(len, 16);
6159 jcc(Assembler::lessEqual, L_copy_16_chars);
6160
6161 bind(L_copy_16_chars_exit);
6162 if (UseAVX >= 2) {
6163 // clean upper bits of YMM registers
6164 vpxor(tmp2Reg, tmp2Reg);
6165 vpxor(tmp3Reg, tmp3Reg);
6166 vpxor(tmp4Reg, tmp4Reg);
6167 movdl(tmp1Reg, tmp5);
6168 pshufd(tmp1Reg, tmp1Reg, 0);
6169 }
6170 subptr(len, 8);
6171 jccb(Assembler::greater, L_copy_8_chars_exit);
6172
6173 bind(L_copy_8_chars);
6174 movdqu(tmp3Reg, Address(src, len, Address::times_2, -16));
6175 ptest(tmp3Reg, tmp1Reg);
6176 jccb(Assembler::notZero, L_copy_8_chars_exit);
6177 packuswb(tmp3Reg, tmp1Reg);
6178 movq(Address(dst, len, Address::times_1, -8), tmp3Reg);
6179 addptr(len, 8);
6180 jccb(Assembler::lessEqual, L_copy_8_chars);
6181
6182 bind(L_copy_8_chars_exit);
6183 subptr(len, 8);
6184 jccb(Assembler::zero, L_done);
6185 }
6186
6187 bind(L_copy_1_char);
6188 load_unsigned_short(tmp5, Address(src, len, Address::times_2, 0));
6189 testl(tmp5, short_mask); // check if Unicode or non-ASCII char
6190 jccb(Assembler::notZero, L_copy_1_char_exit);
6191 movb(Address(dst, len, Address::times_1, 0), tmp5);
6192 addptr(len, 1);
6193 jccb(Assembler::less, L_copy_1_char);
6194
6195 bind(L_copy_1_char_exit);
6196 addptr(result, len); // len is negative count of not processed elements
6197
6198 bind(L_done);
6199 }
6200
6201 /**
6202 * Helper for multiply_to_len().
6203 */
6204 void MacroAssembler::add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
6205 addq(dest_lo, src1);
6206 adcq(dest_hi, 0);
6207 addq(dest_lo, src2);
6208 adcq(dest_hi, 0);
6209 }
6210
6211 /**
6212 * Multiply 64 bit by 64 bit first loop.
6213 */
6214 void MacroAssembler::multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
6215 Register y, Register y_idx, Register z,
6216 Register carry, Register product,
6217 Register idx, Register kdx) {
6218 //
6219 // jlong carry, x[], y[], z[];
6220 // for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
6221 // huge_128 product = y[idx] * x[xstart] + carry;
6222 // z[kdx] = (jlong)product;
6223 // carry = (jlong)(product >>> 64);
6224 // }
6225 // z[xstart] = carry;
6226 //
6227
6228 Label L_first_loop, L_first_loop_exit;
6229 Label L_one_x, L_one_y, L_multiply;
6230
6231 decrementl(xstart);
6232 jcc(Assembler::negative, L_one_x);
6233
6234 movq(x_xstart, Address(x, xstart, Address::times_4, 0));
6235 rorq(x_xstart, 32); // convert big-endian to little-endian
6236
6237 bind(L_first_loop);
6238 decrementl(idx);
6239 jcc(Assembler::negative, L_first_loop_exit);
6240 decrementl(idx);
6241 jcc(Assembler::negative, L_one_y);
6242 movq(y_idx, Address(y, idx, Address::times_4, 0));
6243 rorq(y_idx, 32); // convert big-endian to little-endian
6244 bind(L_multiply);
6245 movq(product, x_xstart);
6246 mulq(y_idx); // product(rax) * y_idx -> rdx:rax
6247 addq(product, carry);
6248 adcq(rdx, 0);
6249 subl(kdx, 2);
6250 movl(Address(z, kdx, Address::times_4, 4), product);
6251 shrq(product, 32);
6252 movl(Address(z, kdx, Address::times_4, 0), product);
6253 movq(carry, rdx);
6254 jmp(L_first_loop);
6255
6256 bind(L_one_y);
6257 movl(y_idx, Address(y, 0));
6258 jmp(L_multiply);
6259
6260 bind(L_one_x);
6261 movl(x_xstart, Address(x, 0));
6262 jmp(L_first_loop);
6263
6264 bind(L_first_loop_exit);
6265 }
6266
6267 /**
6268 * Multiply 64 bit by 64 bit and add 128 bit.
6269 */
6270 void MacroAssembler::multiply_add_128_x_128(Register x_xstart, Register y, Register z,
6271 Register yz_idx, Register idx,
6272 Register carry, Register product, int offset) {
6273 // huge_128 product = (y[idx] * x_xstart) + z[kdx] + carry;
6274 // z[kdx] = (jlong)product;
6275
6276 movq(yz_idx, Address(y, idx, Address::times_4, offset));
6277 rorq(yz_idx, 32); // convert big-endian to little-endian
6278 movq(product, x_xstart);
6279 mulq(yz_idx); // product(rax) * yz_idx -> rdx:product(rax)
6280 movq(yz_idx, Address(z, idx, Address::times_4, offset));
6281 rorq(yz_idx, 32); // convert big-endian to little-endian
6282
6283 add2_with_carry(rdx, product, carry, yz_idx);
6284
6285 movl(Address(z, idx, Address::times_4, offset+4), product);
6286 shrq(product, 32);
6287 movl(Address(z, idx, Address::times_4, offset), product);
6288
6289 }
6290
6291 /**
6292 * Multiply 128 bit by 128 bit. Unrolled inner loop.
6293 */
6294 void MacroAssembler::multiply_128_x_128_loop(Register x_xstart, Register y, Register z,
6295 Register yz_idx, Register idx, Register jdx,
6296 Register carry, Register product,
6297 Register carry2) {
6298 // jlong carry, x[], y[], z[];
6299 // int kdx = ystart+1;
6300 // for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
6301 // huge_128 product = (y[idx+1] * x_xstart) + z[kdx+idx+1] + carry;
6302 // z[kdx+idx+1] = (jlong)product;
6303 // jlong carry2 = (jlong)(product >>> 64);
6304 // product = (y[idx] * x_xstart) + z[kdx+idx] + carry2;
6305 // z[kdx+idx] = (jlong)product;
6306 // carry = (jlong)(product >>> 64);
6307 // }
6308 // idx += 2;
6309 // if (idx > 0) {
6310 // product = (y[idx] * x_xstart) + z[kdx+idx] + carry;
6311 // z[kdx+idx] = (jlong)product;
6312 // carry = (jlong)(product >>> 64);
6313 // }
6314 //
6315
6316 Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
6317
6318 movl(jdx, idx);
6319 andl(jdx, 0xFFFFFFFC);
6320 shrl(jdx, 2);
6321
6322 bind(L_third_loop);
6323 subl(jdx, 1);
6324 jcc(Assembler::negative, L_third_loop_exit);
6325 subl(idx, 4);
6326
6327 multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 8);
6328 movq(carry2, rdx);
6329
6330 multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry2, product, 0);
6331 movq(carry, rdx);
6332 jmp(L_third_loop);
6333
6334 bind (L_third_loop_exit);
6335
6336 andl (idx, 0x3);
6337 jcc(Assembler::zero, L_post_third_loop_done);
6338
6339 Label L_check_1;
6340 subl(idx, 2);
6341 jcc(Assembler::negative, L_check_1);
6342
6343 multiply_add_128_x_128(x_xstart, y, z, yz_idx, idx, carry, product, 0);
6344 movq(carry, rdx);
6345
6346 bind (L_check_1);
6347 addl (idx, 0x2);
6348 andl (idx, 0x1);
6349 subl(idx, 1);
6350 jcc(Assembler::negative, L_post_third_loop_done);
6351
6352 movl(yz_idx, Address(y, idx, Address::times_4, 0));
6353 movq(product, x_xstart);
6354 mulq(yz_idx); // product(rax) * yz_idx -> rdx:product(rax)
6355 movl(yz_idx, Address(z, idx, Address::times_4, 0));
6356
6357 add2_with_carry(rdx, product, yz_idx, carry);
6358
6359 movl(Address(z, idx, Address::times_4, 0), product);
6360 shrq(product, 32);
6361
6362 shlq(rdx, 32);
6363 orq(product, rdx);
6364 movq(carry, product);
6365
6366 bind(L_post_third_loop_done);
6367 }
6368
6369 /**
6370 * Multiply 128 bit by 128 bit using BMI2. Unrolled inner loop.
6371 *
6372 */
6373 void MacroAssembler::multiply_128_x_128_bmi2_loop(Register y, Register z,
6374 Register carry, Register carry2,
6375 Register idx, Register jdx,
6376 Register yz_idx1, Register yz_idx2,
6377 Register tmp, Register tmp3, Register tmp4) {
6378 assert(UseBMI2Instructions, "should be used only when BMI2 is available");
6379
6380 // jlong carry, x[], y[], z[];
6381 // int kdx = ystart+1;
6382 // for (int idx=ystart-2; idx >= 0; idx -= 2) { // Third loop
6383 // huge_128 tmp3 = (y[idx+1] * rdx) + z[kdx+idx+1] + carry;
6384 // jlong carry2 = (jlong)(tmp3 >>> 64);
6385 // huge_128 tmp4 = (y[idx] * rdx) + z[kdx+idx] + carry2;
6386 // carry = (jlong)(tmp4 >>> 64);
6387 // z[kdx+idx+1] = (jlong)tmp3;
6388 // z[kdx+idx] = (jlong)tmp4;
6389 // }
6390 // idx += 2;
6391 // if (idx > 0) {
6392 // yz_idx1 = (y[idx] * rdx) + z[kdx+idx] + carry;
6393 // z[kdx+idx] = (jlong)yz_idx1;
6394 // carry = (jlong)(yz_idx1 >>> 64);
6395 // }
6396 //
6397
6398 Label L_third_loop, L_third_loop_exit, L_post_third_loop_done;
6399
6400 movl(jdx, idx);
6401 andl(jdx, 0xFFFFFFFC);
6402 shrl(jdx, 2);
6403
6404 bind(L_third_loop);
6405 subl(jdx, 1);
6406 jcc(Assembler::negative, L_third_loop_exit);
6407 subl(idx, 4);
6408
6409 movq(yz_idx1, Address(y, idx, Address::times_4, 8));
6410 rorxq(yz_idx1, yz_idx1, 32); // convert big-endian to little-endian
6411 movq(yz_idx2, Address(y, idx, Address::times_4, 0));
6412 rorxq(yz_idx2, yz_idx2, 32);
6413
6414 mulxq(tmp4, tmp3, yz_idx1); // yz_idx1 * rdx -> tmp4:tmp3
6415 mulxq(carry2, tmp, yz_idx2); // yz_idx2 * rdx -> carry2:tmp
6416
6417 movq(yz_idx1, Address(z, idx, Address::times_4, 8));
6418 rorxq(yz_idx1, yz_idx1, 32);
6419 movq(yz_idx2, Address(z, idx, Address::times_4, 0));
6420 rorxq(yz_idx2, yz_idx2, 32);
6421
6422 if (VM_Version::supports_adx()) {
6423 adcxq(tmp3, carry);
6424 adoxq(tmp3, yz_idx1);
6425
6426 adcxq(tmp4, tmp);
6427 adoxq(tmp4, yz_idx2);
6428
6429 movl(carry, 0); // does not affect flags
6430 adcxq(carry2, carry);
6431 adoxq(carry2, carry);
6432 } else {
6433 add2_with_carry(tmp4, tmp3, carry, yz_idx1);
6434 add2_with_carry(carry2, tmp4, tmp, yz_idx2);
6435 }
6436 movq(carry, carry2);
6437
6438 movl(Address(z, idx, Address::times_4, 12), tmp3);
6439 shrq(tmp3, 32);
6440 movl(Address(z, idx, Address::times_4, 8), tmp3);
6441
6442 movl(Address(z, idx, Address::times_4, 4), tmp4);
6443 shrq(tmp4, 32);
6444 movl(Address(z, idx, Address::times_4, 0), tmp4);
6445
6446 jmp(L_third_loop);
6447
6448 bind (L_third_loop_exit);
6449
6450 andl (idx, 0x3);
6451 jcc(Assembler::zero, L_post_third_loop_done);
6452
6453 Label L_check_1;
6454 subl(idx, 2);
6455 jcc(Assembler::negative, L_check_1);
6456
6457 movq(yz_idx1, Address(y, idx, Address::times_4, 0));
6458 rorxq(yz_idx1, yz_idx1, 32);
6459 mulxq(tmp4, tmp3, yz_idx1); // yz_idx1 * rdx -> tmp4:tmp3
6460 movq(yz_idx2, Address(z, idx, Address::times_4, 0));
6461 rorxq(yz_idx2, yz_idx2, 32);
6462
6463 add2_with_carry(tmp4, tmp3, carry, yz_idx2);
6464
6465 movl(Address(z, idx, Address::times_4, 4), tmp3);
6466 shrq(tmp3, 32);
6467 movl(Address(z, idx, Address::times_4, 0), tmp3);
6468 movq(carry, tmp4);
6469
6470 bind (L_check_1);
6471 addl (idx, 0x2);
6472 andl (idx, 0x1);
6473 subl(idx, 1);
6474 jcc(Assembler::negative, L_post_third_loop_done);
6475 movl(tmp4, Address(y, idx, Address::times_4, 0));
6476 mulxq(carry2, tmp3, tmp4); // tmp4 * rdx -> carry2:tmp3
6477 movl(tmp4, Address(z, idx, Address::times_4, 0));
6478
6479 add2_with_carry(carry2, tmp3, tmp4, carry);
6480
6481 movl(Address(z, idx, Address::times_4, 0), tmp3);
6482 shrq(tmp3, 32);
6483
6484 shlq(carry2, 32);
6485 orq(tmp3, carry2);
6486 movq(carry, tmp3);
6487
6488 bind(L_post_third_loop_done);
6489 }
6490
6491 /**
6492 * Code for BigInteger::multiplyToLen() intrinsic.
6493 *
6494 * rdi: x
6495 * rax: xlen
6496 * rsi: y
6497 * rcx: ylen
6498 * r8: z
6499 * r11: tmp0
6500 * r12: tmp1
6501 * r13: tmp2
6502 * r14: tmp3
6503 * r15: tmp4
6504 * rbx: tmp5
6505 *
6506 */
6507 void MacroAssembler::multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register tmp0,
6508 Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5) {
6509 ShortBranchVerifier sbv(this);
6510 assert_different_registers(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, rdx);
6511
6512 push(tmp0);
6513 push(tmp1);
6514 push(tmp2);
6515 push(tmp3);
6516 push(tmp4);
6517 push(tmp5);
6518
6519 push(xlen);
6520
6521 const Register idx = tmp1;
6522 const Register kdx = tmp2;
6523 const Register xstart = tmp3;
6524
6525 const Register y_idx = tmp4;
6526 const Register carry = tmp5;
6527 const Register product = xlen;
6528 const Register x_xstart = tmp0;
6529
6530 // First Loop.
6531 //
6532 // final static long LONG_MASK = 0xffffffffL;
6533 // int xstart = xlen - 1;
6534 // int ystart = ylen - 1;
6535 // long carry = 0;
6536 // for (int idx=ystart, kdx=ystart+1+xstart; idx >= 0; idx-, kdx--) {
6537 // long product = (y[idx] & LONG_MASK) * (x[xstart] & LONG_MASK) + carry;
6538 // z[kdx] = (int)product;
6539 // carry = product >>> 32;
6540 // }
6541 // z[xstart] = (int)carry;
6542 //
6543
6544 movl(idx, ylen); // idx = ylen;
6545 lea(kdx, Address(xlen, ylen)); // kdx = xlen+ylen;
6546 xorq(carry, carry); // carry = 0;
6547
6548 Label L_done;
6549
6550 movl(xstart, xlen);
6551 decrementl(xstart);
6552 jcc(Assembler::negative, L_done);
6553
6554 multiply_64_x_64_loop(x, xstart, x_xstart, y, y_idx, z, carry, product, idx, kdx);
6555
6556 Label L_second_loop;
6557 testl(kdx, kdx);
6558 jcc(Assembler::zero, L_second_loop);
6559
6560 Label L_carry;
6561 subl(kdx, 1);
6562 jcc(Assembler::zero, L_carry);
6563
6564 movl(Address(z, kdx, Address::times_4, 0), carry);
6565 shrq(carry, 32);
6566 subl(kdx, 1);
6567
6568 bind(L_carry);
6569 movl(Address(z, kdx, Address::times_4, 0), carry);
6570
6571 // Second and third (nested) loops.
6572 //
6573 // for (int i = xstart-1; i >= 0; i--) { // Second loop
6574 // carry = 0;
6575 // for (int jdx=ystart, k=ystart+1+i; jdx >= 0; jdx--, k--) { // Third loop
6576 // long product = (y[jdx] & LONG_MASK) * (x[i] & LONG_MASK) +
6577 // (z[k] & LONG_MASK) + carry;
6578 // z[k] = (int)product;
6579 // carry = product >>> 32;
6580 // }
6581 // z[i] = (int)carry;
6582 // }
6583 //
6584 // i = xlen, j = tmp1, k = tmp2, carry = tmp5, x[i] = rdx
6585
6586 const Register jdx = tmp1;
6587
6588 bind(L_second_loop);
6589 xorl(carry, carry); // carry = 0;
6590 movl(jdx, ylen); // j = ystart+1
6591
6592 subl(xstart, 1); // i = xstart-1;
6593 jcc(Assembler::negative, L_done);
6594
6595 push (z);
6596
6597 Label L_last_x;
6598 lea(z, Address(z, xstart, Address::times_4, 4)); // z = z + k - j
6599 subl(xstart, 1); // i = xstart-1;
6600 jcc(Assembler::negative, L_last_x);
6601
6602 if (UseBMI2Instructions) {
6603 movq(rdx, Address(x, xstart, Address::times_4, 0));
6604 rorxq(rdx, rdx, 32); // convert big-endian to little-endian
6605 } else {
6606 movq(x_xstart, Address(x, xstart, Address::times_4, 0));
6607 rorq(x_xstart, 32); // convert big-endian to little-endian
6608 }
6609
6610 Label L_third_loop_prologue;
6611 bind(L_third_loop_prologue);
6612
6613 push (x);
6614 push (xstart);
6615 push (ylen);
6616
6617
6618 if (UseBMI2Instructions) {
6619 multiply_128_x_128_bmi2_loop(y, z, carry, x, jdx, ylen, product, tmp2, x_xstart, tmp3, tmp4);
6620 } else { // !UseBMI2Instructions
6621 multiply_128_x_128_loop(x_xstart, y, z, y_idx, jdx, ylen, carry, product, x);
6622 }
6623
6624 pop(ylen);
6625 pop(xlen);
6626 pop(x);
6627 pop(z);
6628
6629 movl(tmp3, xlen);
6630 addl(tmp3, 1);
6631 movl(Address(z, tmp3, Address::times_4, 0), carry);
6632 subl(tmp3, 1);
6633 jccb(Assembler::negative, L_done);
6634
6635 shrq(carry, 32);
6636 movl(Address(z, tmp3, Address::times_4, 0), carry);
6637 jmp(L_second_loop);
6638
6639 // Next infrequent code is moved outside loops.
6640 bind(L_last_x);
6641 if (UseBMI2Instructions) {
6642 movl(rdx, Address(x, 0));
6643 } else {
6644 movl(x_xstart, Address(x, 0));
6645 }
6646 jmp(L_third_loop_prologue);
6647
6648 bind(L_done);
6649
6650 pop(xlen);
6651
6652 pop(tmp5);
6653 pop(tmp4);
6654 pop(tmp3);
6655 pop(tmp2);
6656 pop(tmp1);
6657 pop(tmp0);
6658 }
6659
6660 void MacroAssembler::vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale,
6661 Register result, Register tmp1, Register tmp2, XMMRegister rymm0, XMMRegister rymm1, XMMRegister rymm2){
6662 assert(UseSSE42Intrinsics, "SSE4.2 must be enabled.");
6663 Label VECTOR16_LOOP, VECTOR8_LOOP, VECTOR4_LOOP;
6664 Label VECTOR8_TAIL, VECTOR4_TAIL;
6665 Label VECTOR32_NOT_EQUAL, VECTOR16_NOT_EQUAL, VECTOR8_NOT_EQUAL, VECTOR4_NOT_EQUAL;
6666 Label SAME_TILL_END, DONE;
6667 Label BYTES_LOOP, BYTES_TAIL, BYTES_NOT_EQUAL;
6668
6669 //scale is in rcx in both Win64 and Unix
6670 ShortBranchVerifier sbv(this);
6671
6672 shlq(length);
6673 xorq(result, result);
6674
6675 if ((AVX3Threshold == 0) && (UseAVX > 2) &&
6676 VM_Version::supports_avx512vlbw()) {
6677 Label VECTOR64_LOOP, VECTOR64_NOT_EQUAL, VECTOR32_TAIL;
6678
6679 cmpq(length, 64);
6680 jcc(Assembler::less, VECTOR32_TAIL);
6681
6682 movq(tmp1, length);
6683 andq(tmp1, 0x3F); // tail count
6684 andq(length, ~(0x3F)); //vector count
6685
6686 bind(VECTOR64_LOOP);
6687 // AVX512 code to compare 64 byte vectors.
6688 evmovdqub(rymm0, Address(obja, result), Assembler::AVX_512bit);
6689 evpcmpeqb(k7, rymm0, Address(objb, result), Assembler::AVX_512bit);
6690 kortestql(k7, k7);
6691 jcc(Assembler::aboveEqual, VECTOR64_NOT_EQUAL); // mismatch
6692 addq(result, 64);
6693 subq(length, 64);
6694 jccb(Assembler::notZero, VECTOR64_LOOP);
6695
6696 //bind(VECTOR64_TAIL);
6697 testq(tmp1, tmp1);
6698 jcc(Assembler::zero, SAME_TILL_END);
6699
6700 //bind(VECTOR64_TAIL);
6701 // AVX512 code to compare up to 63 byte vectors.
6702 mov64(tmp2, 0xFFFFFFFFFFFFFFFF);
6703 shlxq(tmp2, tmp2, tmp1);
6704 notq(tmp2);
6705 kmovql(k3, tmp2);
6706
6707 evmovdqub(rymm0, k3, Address(obja, result), false, Assembler::AVX_512bit);
6708 evpcmpeqb(k7, k3, rymm0, Address(objb, result), Assembler::AVX_512bit);
6709
6710 ktestql(k7, k3);
6711 jcc(Assembler::below, SAME_TILL_END); // not mismatch
6712
6713 bind(VECTOR64_NOT_EQUAL);
6714 kmovql(tmp1, k7);
6715 notq(tmp1);
6716 tzcntq(tmp1, tmp1);
6717 addq(result, tmp1);
6718 shrq(result);
6719 jmp(DONE);
6720 bind(VECTOR32_TAIL);
6721 }
6722
6723 cmpq(length, 8);
6724 jcc(Assembler::equal, VECTOR8_LOOP);
6725 jcc(Assembler::less, VECTOR4_TAIL);
6726
6727 if (UseAVX >= 2) {
6728 Label VECTOR16_TAIL, VECTOR32_LOOP;
6729
6730 cmpq(length, 16);
6731 jcc(Assembler::equal, VECTOR16_LOOP);
6732 jcc(Assembler::less, VECTOR8_LOOP);
6733
6734 cmpq(length, 32);
6735 jccb(Assembler::less, VECTOR16_TAIL);
6736
6737 subq(length, 32);
6738 bind(VECTOR32_LOOP);
6739 vmovdqu(rymm0, Address(obja, result));
6740 vmovdqu(rymm1, Address(objb, result));
6741 vpxor(rymm2, rymm0, rymm1, Assembler::AVX_256bit);
6742 vptest(rymm2, rymm2);
6743 jcc(Assembler::notZero, VECTOR32_NOT_EQUAL);//mismatch found
6744 addq(result, 32);
6745 subq(length, 32);
6746 jcc(Assembler::greaterEqual, VECTOR32_LOOP);
6747 addq(length, 32);
6748 jcc(Assembler::equal, SAME_TILL_END);
6749 //falling through if less than 32 bytes left //close the branch here.
6750
6751 bind(VECTOR16_TAIL);
6752 cmpq(length, 16);
6753 jccb(Assembler::less, VECTOR8_TAIL);
6754 bind(VECTOR16_LOOP);
6755 movdqu(rymm0, Address(obja, result));
6756 movdqu(rymm1, Address(objb, result));
6757 vpxor(rymm2, rymm0, rymm1, Assembler::AVX_128bit);
6758 ptest(rymm2, rymm2);
6759 jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
6760 addq(result, 16);
6761 subq(length, 16);
6762 jcc(Assembler::equal, SAME_TILL_END);
6763 //falling through if less than 16 bytes left
6764 } else {//regular intrinsics
6765
6766 cmpq(length, 16);
6767 jccb(Assembler::less, VECTOR8_TAIL);
6768
6769 subq(length, 16);
6770 bind(VECTOR16_LOOP);
6771 movdqu(rymm0, Address(obja, result));
6772 movdqu(rymm1, Address(objb, result));
6773 pxor(rymm0, rymm1);
6774 ptest(rymm0, rymm0);
6775 jcc(Assembler::notZero, VECTOR16_NOT_EQUAL);//mismatch found
6776 addq(result, 16);
6777 subq(length, 16);
6778 jccb(Assembler::greaterEqual, VECTOR16_LOOP);
6779 addq(length, 16);
6780 jcc(Assembler::equal, SAME_TILL_END);
6781 //falling through if less than 16 bytes left
6782 }
6783
6784 bind(VECTOR8_TAIL);
6785 cmpq(length, 8);
6786 jccb(Assembler::less, VECTOR4_TAIL);
6787 bind(VECTOR8_LOOP);
6788 movq(tmp1, Address(obja, result));
6789 movq(tmp2, Address(objb, result));
6790 xorq(tmp1, tmp2);
6791 testq(tmp1, tmp1);
6792 jcc(Assembler::notZero, VECTOR8_NOT_EQUAL);//mismatch found
6793 addq(result, 8);
6794 subq(length, 8);
6795 jcc(Assembler::equal, SAME_TILL_END);
6796 //falling through if less than 8 bytes left
6797
6798 bind(VECTOR4_TAIL);
6799 cmpq(length, 4);
6800 jccb(Assembler::less, BYTES_TAIL);
6801 bind(VECTOR4_LOOP);
6802 movl(tmp1, Address(obja, result));
6803 xorl(tmp1, Address(objb, result));
6804 testl(tmp1, tmp1);
6805 jcc(Assembler::notZero, VECTOR4_NOT_EQUAL);//mismatch found
6806 addq(result, 4);
6807 subq(length, 4);
6808 jcc(Assembler::equal, SAME_TILL_END);
6809 //falling through if less than 4 bytes left
6810
6811 bind(BYTES_TAIL);
6812 bind(BYTES_LOOP);
6813 load_unsigned_byte(tmp1, Address(obja, result));
6814 load_unsigned_byte(tmp2, Address(objb, result));
6815 xorl(tmp1, tmp2);
6816 testl(tmp1, tmp1);
6817 jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
6818 decq(length);
6819 jcc(Assembler::zero, SAME_TILL_END);
6820 incq(result);
6821 load_unsigned_byte(tmp1, Address(obja, result));
6822 load_unsigned_byte(tmp2, Address(objb, result));
6823 xorl(tmp1, tmp2);
6824 testl(tmp1, tmp1);
6825 jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
6826 decq(length);
6827 jcc(Assembler::zero, SAME_TILL_END);
6828 incq(result);
6829 load_unsigned_byte(tmp1, Address(obja, result));
6830 load_unsigned_byte(tmp2, Address(objb, result));
6831 xorl(tmp1, tmp2);
6832 testl(tmp1, tmp1);
6833 jcc(Assembler::notZero, BYTES_NOT_EQUAL);//mismatch found
6834 jmp(SAME_TILL_END);
6835
6836 if (UseAVX >= 2) {
6837 bind(VECTOR32_NOT_EQUAL);
6838 vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_256bit);
6839 vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_256bit);
6840 vpxor(rymm0, rymm0, rymm2, Assembler::AVX_256bit);
6841 vpmovmskb(tmp1, rymm0);
6842 bsfq(tmp1, tmp1);
6843 addq(result, tmp1);
6844 shrq(result);
6845 jmp(DONE);
6846 }
6847
6848 bind(VECTOR16_NOT_EQUAL);
6849 if (UseAVX >= 2) {
6850 vpcmpeqb(rymm2, rymm2, rymm2, Assembler::AVX_128bit);
6851 vpcmpeqb(rymm0, rymm0, rymm1, Assembler::AVX_128bit);
6852 pxor(rymm0, rymm2);
6853 } else {
6854 pcmpeqb(rymm2, rymm2);
6855 pxor(rymm0, rymm1);
6856 pcmpeqb(rymm0, rymm1);
6857 pxor(rymm0, rymm2);
6858 }
6859 pmovmskb(tmp1, rymm0);
6860 bsfq(tmp1, tmp1);
6861 addq(result, tmp1);
6862 shrq(result);
6863 jmpb(DONE);
6864
6865 bind(VECTOR8_NOT_EQUAL);
6866 bind(VECTOR4_NOT_EQUAL);
6867 bsfq(tmp1, tmp1);
6868 shrq(tmp1, 3);
6869 addq(result, tmp1);
6870 bind(BYTES_NOT_EQUAL);
6871 shrq(result);
6872 jmpb(DONE);
6873
6874 bind(SAME_TILL_END);
6875 mov64(result, -1);
6876
6877 bind(DONE);
6878 }
6879
6880 //Helper functions for square_to_len()
6881
6882 /**
6883 * Store the squares of x[], right shifted one bit (divided by 2) into z[]
6884 * Preserves x and z and modifies rest of the registers.
6885 */
6886 void MacroAssembler::square_rshift(Register x, Register xlen, Register z, Register tmp1, Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
6887 // Perform square and right shift by 1
6888 // Handle odd xlen case first, then for even xlen do the following
6889 // jlong carry = 0;
6890 // for (int j=0, i=0; j < xlen; j+=2, i+=4) {
6891 // huge_128 product = x[j:j+1] * x[j:j+1];
6892 // z[i:i+1] = (carry << 63) | (jlong)(product >>> 65);
6893 // z[i+2:i+3] = (jlong)(product >>> 1);
6894 // carry = (jlong)product;
6895 // }
6896
6897 xorq(tmp5, tmp5); // carry
6898 xorq(rdxReg, rdxReg);
6899 xorl(tmp1, tmp1); // index for x
6900 xorl(tmp4, tmp4); // index for z
6901
6902 Label L_first_loop, L_first_loop_exit;
6903
6904 testl(xlen, 1);
6905 jccb(Assembler::zero, L_first_loop); //jump if xlen is even
6906
6907 // Square and right shift by 1 the odd element using 32 bit multiply
6908 movl(raxReg, Address(x, tmp1, Address::times_4, 0));
6909 imulq(raxReg, raxReg);
6910 shrq(raxReg, 1);
6911 adcq(tmp5, 0);
6912 movq(Address(z, tmp4, Address::times_4, 0), raxReg);
6913 incrementl(tmp1);
6914 addl(tmp4, 2);
6915
6916 // Square and right shift by 1 the rest using 64 bit multiply
6917 bind(L_first_loop);
6918 cmpptr(tmp1, xlen);
6919 jccb(Assembler::equal, L_first_loop_exit);
6920
6921 // Square
6922 movq(raxReg, Address(x, tmp1, Address::times_4, 0));
6923 rorq(raxReg, 32); // convert big-endian to little-endian
6924 mulq(raxReg); // 64-bit multiply rax * rax -> rdx:rax
6925
6926 // Right shift by 1 and save carry
6927 shrq(tmp5, 1); // rdx:rax:tmp5 = (tmp5:rdx:rax) >>> 1
6928 rcrq(rdxReg, 1);
6929 rcrq(raxReg, 1);
6930 adcq(tmp5, 0);
6931
6932 // Store result in z
6933 movq(Address(z, tmp4, Address::times_4, 0), rdxReg);
6934 movq(Address(z, tmp4, Address::times_4, 8), raxReg);
6935
6936 // Update indices for x and z
6937 addl(tmp1, 2);
6938 addl(tmp4, 4);
6939 jmp(L_first_loop);
6940
6941 bind(L_first_loop_exit);
6942 }
6943
6944
6945 /**
6946 * Perform the following multiply add operation using BMI2 instructions
6947 * carry:sum = sum + op1*op2 + carry
6948 * op2 should be in rdx
6949 * op2 is preserved, all other registers are modified
6950 */
6951 void MacroAssembler::multiply_add_64_bmi2(Register sum, Register op1, Register op2, Register carry, Register tmp2) {
6952 // assert op2 is rdx
6953 mulxq(tmp2, op1, op1); // op1 * op2 -> tmp2:op1
6954 addq(sum, carry);
6955 adcq(tmp2, 0);
6956 addq(sum, op1);
6957 adcq(tmp2, 0);
6958 movq(carry, tmp2);
6959 }
6960
6961 /**
6962 * Perform the following multiply add operation:
6963 * carry:sum = sum + op1*op2 + carry
6964 * Preserves op1, op2 and modifies rest of registers
6965 */
6966 void MacroAssembler::multiply_add_64(Register sum, Register op1, Register op2, Register carry, Register rdxReg, Register raxReg) {
6967 // rdx:rax = op1 * op2
6968 movq(raxReg, op2);
6969 mulq(op1);
6970
6971 // rdx:rax = sum + carry + rdx:rax
6972 addq(sum, carry);
6973 adcq(rdxReg, 0);
6974 addq(sum, raxReg);
6975 adcq(rdxReg, 0);
6976
6977 // carry:sum = rdx:sum
6978 movq(carry, rdxReg);
6979 }
6980
6981 /**
6982 * Add 64 bit long carry into z[] with carry propagation.
6983 * Preserves z and carry register values and modifies rest of registers.
6984 *
6985 */
6986 void MacroAssembler::add_one_64(Register z, Register zlen, Register carry, Register tmp1) {
6987 Label L_fourth_loop, L_fourth_loop_exit;
6988
6989 movl(tmp1, 1);
6990 subl(zlen, 2);
6991 addq(Address(z, zlen, Address::times_4, 0), carry);
6992
6993 bind(L_fourth_loop);
6994 jccb(Assembler::carryClear, L_fourth_loop_exit);
6995 subl(zlen, 2);
6996 jccb(Assembler::negative, L_fourth_loop_exit);
6997 addq(Address(z, zlen, Address::times_4, 0), tmp1);
6998 jmp(L_fourth_loop);
6999 bind(L_fourth_loop_exit);
7000 }
7001
7002 /**
7003 * Shift z[] left by 1 bit.
7004 * Preserves x, len, z and zlen registers and modifies rest of the registers.
7005 *
7006 */
7007 void MacroAssembler::lshift_by_1(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
7008
7009 Label L_fifth_loop, L_fifth_loop_exit;
7010
7011 // Fifth loop
7012 // Perform primitiveLeftShift(z, zlen, 1)
7013
7014 const Register prev_carry = tmp1;
7015 const Register new_carry = tmp4;
7016 const Register value = tmp2;
7017 const Register zidx = tmp3;
7018
7019 // int zidx, carry;
7020 // long value;
7021 // carry = 0;
7022 // for (zidx = zlen-2; zidx >=0; zidx -= 2) {
7023 // (carry:value) = (z[i] << 1) | carry ;
7024 // z[i] = value;
7025 // }
7026
7027 movl(zidx, zlen);
7028 xorl(prev_carry, prev_carry); // clear carry flag and prev_carry register
7029
7030 bind(L_fifth_loop);
7031 decl(zidx); // Use decl to preserve carry flag
7032 decl(zidx);
7033 jccb(Assembler::negative, L_fifth_loop_exit);
7034
7035 if (UseBMI2Instructions) {
7036 movq(value, Address(z, zidx, Address::times_4, 0));
7037 rclq(value, 1);
7038 rorxq(value, value, 32);
7039 movq(Address(z, zidx, Address::times_4, 0), value); // Store back in big endian form
7040 }
7041 else {
7042 // clear new_carry
7043 xorl(new_carry, new_carry);
7044
7045 // Shift z[i] by 1, or in previous carry and save new carry
7046 movq(value, Address(z, zidx, Address::times_4, 0));
7047 shlq(value, 1);
7048 adcl(new_carry, 0);
7049
7050 orq(value, prev_carry);
7051 rorq(value, 0x20);
7052 movq(Address(z, zidx, Address::times_4, 0), value); // Store back in big endian form
7053
7054 // Set previous carry = new carry
7055 movl(prev_carry, new_carry);
7056 }
7057 jmp(L_fifth_loop);
7058
7059 bind(L_fifth_loop_exit);
7060 }
7061
7062
7063 /**
7064 * Code for BigInteger::squareToLen() intrinsic
7065 *
7066 * rdi: x
7067 * rsi: len
7068 * r8: z
7069 * rcx: zlen
7070 * r12: tmp1
7071 * r13: tmp2
7072 * r14: tmp3
7073 * r15: tmp4
7074 * rbx: tmp5
7075 *
7076 */
7077 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) {
7078
7079 Label L_second_loop, L_second_loop_exit, L_third_loop, L_third_loop_exit, L_last_x, L_multiply;
7080 push(tmp1);
7081 push(tmp2);
7082 push(tmp3);
7083 push(tmp4);
7084 push(tmp5);
7085
7086 // First loop
7087 // Store the squares, right shifted one bit (i.e., divided by 2).
7088 square_rshift(x, len, z, tmp1, tmp3, tmp4, tmp5, rdxReg, raxReg);
7089
7090 // Add in off-diagonal sums.
7091 //
7092 // Second, third (nested) and fourth loops.
7093 // zlen +=2;
7094 // for (int xidx=len-2,zidx=zlen-4; xidx > 0; xidx-=2,zidx-=4) {
7095 // carry = 0;
7096 // long op2 = x[xidx:xidx+1];
7097 // for (int j=xidx-2,k=zidx; j >= 0; j-=2) {
7098 // k -= 2;
7099 // long op1 = x[j:j+1];
7100 // long sum = z[k:k+1];
7101 // carry:sum = multiply_add_64(sum, op1, op2, carry, tmp_regs);
7102 // z[k:k+1] = sum;
7103 // }
7104 // add_one_64(z, k, carry, tmp_regs);
7105 // }
7106
7107 const Register carry = tmp5;
7108 const Register sum = tmp3;
7109 const Register op1 = tmp4;
7110 Register op2 = tmp2;
7111
7112 push(zlen);
7113 push(len);
7114 addl(zlen,2);
7115 bind(L_second_loop);
7116 xorq(carry, carry);
7117 subl(zlen, 4);
7118 subl(len, 2);
7119 push(zlen);
7120 push(len);
7121 cmpl(len, 0);
7122 jccb(Assembler::lessEqual, L_second_loop_exit);
7123
7124 // Multiply an array by one 64 bit long.
7125 if (UseBMI2Instructions) {
7126 op2 = rdxReg;
7127 movq(op2, Address(x, len, Address::times_4, 0));
7128 rorxq(op2, op2, 32);
7129 }
7130 else {
7131 movq(op2, Address(x, len, Address::times_4, 0));
7132 rorq(op2, 32);
7133 }
7134
7135 bind(L_third_loop);
7136 decrementl(len);
7137 jccb(Assembler::negative, L_third_loop_exit);
7138 decrementl(len);
7139 jccb(Assembler::negative, L_last_x);
7140
7141 movq(op1, Address(x, len, Address::times_4, 0));
7142 rorq(op1, 32);
7143
7144 bind(L_multiply);
7145 subl(zlen, 2);
7146 movq(sum, Address(z, zlen, Address::times_4, 0));
7147
7148 // Multiply 64 bit by 64 bit and add 64 bits lower half and upper 64 bits as carry.
7149 if (UseBMI2Instructions) {
7150 multiply_add_64_bmi2(sum, op1, op2, carry, tmp2);
7151 }
7152 else {
7153 multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
7154 }
7155
7156 movq(Address(z, zlen, Address::times_4, 0), sum);
7157
7158 jmp(L_third_loop);
7159 bind(L_third_loop_exit);
7160
7161 // Fourth loop
7162 // Add 64 bit long carry into z with carry propagation.
7163 // Uses offsetted zlen.
7164 add_one_64(z, zlen, carry, tmp1);
7165
7166 pop(len);
7167 pop(zlen);
7168 jmp(L_second_loop);
7169
7170 // Next infrequent code is moved outside loops.
7171 bind(L_last_x);
7172 movl(op1, Address(x, 0));
7173 jmp(L_multiply);
7174
7175 bind(L_second_loop_exit);
7176 pop(len);
7177 pop(zlen);
7178 pop(len);
7179 pop(zlen);
7180
7181 // Fifth loop
7182 // Shift z left 1 bit.
7183 lshift_by_1(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4);
7184
7185 // z[zlen-1] |= x[len-1] & 1;
7186 movl(tmp3, Address(x, len, Address::times_4, -4));
7187 andl(tmp3, 1);
7188 orl(Address(z, zlen, Address::times_4, -4), tmp3);
7189
7190 pop(tmp5);
7191 pop(tmp4);
7192 pop(tmp3);
7193 pop(tmp2);
7194 pop(tmp1);
7195 }
7196
7197 /**
7198 * Helper function for mul_add()
7199 * Multiply the in[] by int k and add to out[] starting at offset offs using
7200 * 128 bit by 32 bit multiply and return the carry in tmp5.
7201 * Only quad int aligned length of in[] is operated on in this function.
7202 * k is in rdxReg for BMI2Instructions, for others it is in tmp2.
7203 * This function preserves out, in and k registers.
7204 * len and offset point to the appropriate index in "in" & "out" correspondingly
7205 * tmp5 has the carry.
7206 * other registers are temporary and are modified.
7207 *
7208 */
7209 void MacroAssembler::mul_add_128_x_32_loop(Register out, Register in,
7210 Register offset, Register len, Register tmp1, Register tmp2, Register tmp3,
7211 Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
7212
7213 Label L_first_loop, L_first_loop_exit;
7214
7215 movl(tmp1, len);
7216 shrl(tmp1, 2);
7217
7218 bind(L_first_loop);
7219 subl(tmp1, 1);
7220 jccb(Assembler::negative, L_first_loop_exit);
7221
7222 subl(len, 4);
7223 subl(offset, 4);
7224
7225 Register op2 = tmp2;
7226 const Register sum = tmp3;
7227 const Register op1 = tmp4;
7228 const Register carry = tmp5;
7229
7230 if (UseBMI2Instructions) {
7231 op2 = rdxReg;
7232 }
7233
7234 movq(op1, Address(in, len, Address::times_4, 8));
7235 rorq(op1, 32);
7236 movq(sum, Address(out, offset, Address::times_4, 8));
7237 rorq(sum, 32);
7238 if (UseBMI2Instructions) {
7239 multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
7240 }
7241 else {
7242 multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
7243 }
7244 // Store back in big endian from little endian
7245 rorq(sum, 0x20);
7246 movq(Address(out, offset, Address::times_4, 8), sum);
7247
7248 movq(op1, Address(in, len, Address::times_4, 0));
7249 rorq(op1, 32);
7250 movq(sum, Address(out, offset, Address::times_4, 0));
7251 rorq(sum, 32);
7252 if (UseBMI2Instructions) {
7253 multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
7254 }
7255 else {
7256 multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
7257 }
7258 // Store back in big endian from little endian
7259 rorq(sum, 0x20);
7260 movq(Address(out, offset, Address::times_4, 0), sum);
7261
7262 jmp(L_first_loop);
7263 bind(L_first_loop_exit);
7264 }
7265
7266 /**
7267 * Code for BigInteger::mulAdd() intrinsic
7268 *
7269 * rdi: out
7270 * rsi: in
7271 * r11: offs (out.length - offset)
7272 * rcx: len
7273 * r8: k
7274 * r12: tmp1
7275 * r13: tmp2
7276 * r14: tmp3
7277 * r15: tmp4
7278 * rbx: tmp5
7279 * Multiply the in[] by word k and add to out[], return the carry in rax
7280 */
7281 void MacroAssembler::mul_add(Register out, Register in, Register offs,
7282 Register len, Register k, Register tmp1, Register tmp2, Register tmp3,
7283 Register tmp4, Register tmp5, Register rdxReg, Register raxReg) {
7284
7285 Label L_carry, L_last_in, L_done;
7286
7287 // carry = 0;
7288 // for (int j=len-1; j >= 0; j--) {
7289 // long product = (in[j] & LONG_MASK) * kLong +
7290 // (out[offs] & LONG_MASK) + carry;
7291 // out[offs--] = (int)product;
7292 // carry = product >>> 32;
7293 // }
7294 //
7295 push(tmp1);
7296 push(tmp2);
7297 push(tmp3);
7298 push(tmp4);
7299 push(tmp5);
7300
7301 Register op2 = tmp2;
7302 const Register sum = tmp3;
7303 const Register op1 = tmp4;
7304 const Register carry = tmp5;
7305
7306 if (UseBMI2Instructions) {
7307 op2 = rdxReg;
7308 movl(op2, k);
7309 }
7310 else {
7311 movl(op2, k);
7312 }
7313
7314 xorq(carry, carry);
7315
7316 //First loop
7317
7318 //Multiply in[] by k in a 4 way unrolled loop using 128 bit by 32 bit multiply
7319 //The carry is in tmp5
7320 mul_add_128_x_32_loop(out, in, offs, len, tmp1, tmp2, tmp3, tmp4, tmp5, rdxReg, raxReg);
7321
7322 //Multiply the trailing in[] entry using 64 bit by 32 bit, if any
7323 decrementl(len);
7324 jccb(Assembler::negative, L_carry);
7325 decrementl(len);
7326 jccb(Assembler::negative, L_last_in);
7327
7328 movq(op1, Address(in, len, Address::times_4, 0));
7329 rorq(op1, 32);
7330
7331 subl(offs, 2);
7332 movq(sum, Address(out, offs, Address::times_4, 0));
7333 rorq(sum, 32);
7334
7335 if (UseBMI2Instructions) {
7336 multiply_add_64_bmi2(sum, op1, op2, carry, raxReg);
7337 }
7338 else {
7339 multiply_add_64(sum, op1, op2, carry, rdxReg, raxReg);
7340 }
7341
7342 // Store back in big endian from little endian
7343 rorq(sum, 0x20);
7344 movq(Address(out, offs, Address::times_4, 0), sum);
7345
7346 testl(len, len);
7347 jccb(Assembler::zero, L_carry);
7348
7349 //Multiply the last in[] entry, if any
7350 bind(L_last_in);
7351 movl(op1, Address(in, 0));
7352 movl(sum, Address(out, offs, Address::times_4, -4));
7353
7354 movl(raxReg, k);
7355 mull(op1); //tmp4 * eax -> edx:eax
7356 addl(sum, carry);
7357 adcl(rdxReg, 0);
7358 addl(sum, raxReg);
7359 adcl(rdxReg, 0);
7360 movl(carry, rdxReg);
7361
7362 movl(Address(out, offs, Address::times_4, -4), sum);
7363
7364 bind(L_carry);
7365 //return tmp5/carry as carry in rax
7366 movl(rax, carry);
7367
7368 bind(L_done);
7369 pop(tmp5);
7370 pop(tmp4);
7371 pop(tmp3);
7372 pop(tmp2);
7373 pop(tmp1);
7374 }
7375
7376 /**
7377 * Emits code to update CRC-32 with a byte value according to constants in table
7378 *
7379 * @param [in,out]crc Register containing the crc.
7380 * @param [in]val Register containing the byte to fold into the CRC.
7381 * @param [in]table Register containing the table of crc constants.
7382 *
7383 * uint32_t crc;
7384 * val = crc_table[(val ^ crc) & 0xFF];
7385 * crc = val ^ (crc >> 8);
7386 *
7387 */
7388 void MacroAssembler::update_byte_crc32(Register crc, Register val, Register table) {
7389 xorl(val, crc);
7390 andl(val, 0xFF);
7391 shrl(crc, 8); // unsigned shift
7392 xorl(crc, Address(table, val, Address::times_4, 0));
7393 }
7394
7395 /**
7396 * Fold 128-bit data chunk
7397 */
7398 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset) {
7399 if (UseAVX > 0) {
7400 vpclmulhdq(xtmp, xK, xcrc); // [123:64]
7401 vpclmulldq(xcrc, xK, xcrc); // [63:0]
7402 vpxor(xcrc, xcrc, Address(buf, offset), 0 /* vector_len */);
7403 pxor(xcrc, xtmp);
7404 } else {
7405 movdqa(xtmp, xcrc);
7406 pclmulhdq(xtmp, xK); // [123:64]
7407 pclmulldq(xcrc, xK); // [63:0]
7408 pxor(xcrc, xtmp);
7409 movdqu(xtmp, Address(buf, offset));
7410 pxor(xcrc, xtmp);
7411 }
7412 }
7413
7414 void MacroAssembler::fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, XMMRegister xbuf) {
7415 if (UseAVX > 0) {
7416 vpclmulhdq(xtmp, xK, xcrc);
7417 vpclmulldq(xcrc, xK, xcrc);
7418 pxor(xcrc, xbuf);
7419 pxor(xcrc, xtmp);
7420 } else {
7421 movdqa(xtmp, xcrc);
7422 pclmulhdq(xtmp, xK);
7423 pclmulldq(xcrc, xK);
7424 pxor(xcrc, xbuf);
7425 pxor(xcrc, xtmp);
7426 }
7427 }
7428
7429 /**
7430 * 8-bit folds to compute 32-bit CRC
7431 *
7432 * uint64_t xcrc;
7433 * timesXtoThe32[xcrc & 0xFF] ^ (xcrc >> 8);
7434 */
7435 void MacroAssembler::fold_8bit_crc32(XMMRegister xcrc, Register table, XMMRegister xtmp, Register tmp) {
7436 movdl(tmp, xcrc);
7437 andl(tmp, 0xFF);
7438 movdl(xtmp, Address(table, tmp, Address::times_4, 0));
7439 psrldq(xcrc, 1); // unsigned shift one byte
7440 pxor(xcrc, xtmp);
7441 }
7442
7443 /**
7444 * uint32_t crc;
7445 * timesXtoThe32[crc & 0xFF] ^ (crc >> 8);
7446 */
7447 void MacroAssembler::fold_8bit_crc32(Register crc, Register table, Register tmp) {
7448 movl(tmp, crc);
7449 andl(tmp, 0xFF);
7450 shrl(crc, 8);
7451 xorl(crc, Address(table, tmp, Address::times_4, 0));
7452 }
7453
7454 /**
7455 * @param crc register containing existing CRC (32-bit)
7456 * @param buf register pointing to input byte buffer (byte*)
7457 * @param len register containing number of bytes
7458 * @param table register that will contain address of CRC table
7459 * @param tmp scratch register
7460 */
7461 void MacroAssembler::kernel_crc32(Register crc, Register buf, Register len, Register table, Register tmp) {
7462 assert_different_registers(crc, buf, len, table, tmp, rax);
7463
7464 Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
7465 Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
7466
7467 // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
7468 // context for the registers used, where all instructions below are using 128-bit mode
7469 // On EVEX without VL and BW, these instructions will all be AVX.
7470 lea(table, ExternalAddress(StubRoutines::crc_table_addr()));
7471 notl(crc); // ~crc
7472 cmpl(len, 16);
7473 jcc(Assembler::less, L_tail);
7474
7475 // Align buffer to 16 bytes
7476 movl(tmp, buf);
7477 andl(tmp, 0xF);
7478 jccb(Assembler::zero, L_aligned);
7479 subl(tmp, 16);
7480 addl(len, tmp);
7481
7482 align(4);
7483 BIND(L_align_loop);
7484 movsbl(rax, Address(buf, 0)); // load byte with sign extension
7485 update_byte_crc32(crc, rax, table);
7486 increment(buf);
7487 incrementl(tmp);
7488 jccb(Assembler::less, L_align_loop);
7489
7490 BIND(L_aligned);
7491 movl(tmp, len); // save
7492 shrl(len, 4);
7493 jcc(Assembler::zero, L_tail_restore);
7494
7495 // Fold crc into first bytes of vector
7496 movdqa(xmm1, Address(buf, 0));
7497 movdl(rax, xmm1);
7498 xorl(crc, rax);
7499 if (VM_Version::supports_sse4_1()) {
7500 pinsrd(xmm1, crc, 0);
7501 } else {
7502 pinsrw(xmm1, crc, 0);
7503 shrl(crc, 16);
7504 pinsrw(xmm1, crc, 1);
7505 }
7506 addptr(buf, 16);
7507 subl(len, 4); // len > 0
7508 jcc(Assembler::less, L_fold_tail);
7509
7510 movdqa(xmm2, Address(buf, 0));
7511 movdqa(xmm3, Address(buf, 16));
7512 movdqa(xmm4, Address(buf, 32));
7513 addptr(buf, 48);
7514 subl(len, 3);
7515 jcc(Assembler::lessEqual, L_fold_512b);
7516
7517 // Fold total 512 bits of polynomial on each iteration,
7518 // 128 bits per each of 4 parallel streams.
7519 movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 32), rscratch1);
7520
7521 align32();
7522 BIND(L_fold_512b_loop);
7523 fold_128bit_crc32(xmm1, xmm0, xmm5, buf, 0);
7524 fold_128bit_crc32(xmm2, xmm0, xmm5, buf, 16);
7525 fold_128bit_crc32(xmm3, xmm0, xmm5, buf, 32);
7526 fold_128bit_crc32(xmm4, xmm0, xmm5, buf, 48);
7527 addptr(buf, 64);
7528 subl(len, 4);
7529 jcc(Assembler::greater, L_fold_512b_loop);
7530
7531 // Fold 512 bits to 128 bits.
7532 BIND(L_fold_512b);
7533 movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
7534 fold_128bit_crc32(xmm1, xmm0, xmm5, xmm2);
7535 fold_128bit_crc32(xmm1, xmm0, xmm5, xmm3);
7536 fold_128bit_crc32(xmm1, xmm0, xmm5, xmm4);
7537
7538 // Fold the rest of 128 bits data chunks
7539 BIND(L_fold_tail);
7540 addl(len, 3);
7541 jccb(Assembler::lessEqual, L_fold_128b);
7542 movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr() + 16), rscratch1);
7543
7544 BIND(L_fold_tail_loop);
7545 fold_128bit_crc32(xmm1, xmm0, xmm5, buf, 0);
7546 addptr(buf, 16);
7547 decrementl(len);
7548 jccb(Assembler::greater, L_fold_tail_loop);
7549
7550 // Fold 128 bits in xmm1 down into 32 bits in crc register.
7551 BIND(L_fold_128b);
7552 movdqu(xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_addr()), rscratch1);
7553 if (UseAVX > 0) {
7554 vpclmulqdq(xmm2, xmm0, xmm1, 0x1);
7555 vpand(xmm3, xmm0, xmm2, 0 /* vector_len */);
7556 vpclmulqdq(xmm0, xmm0, xmm3, 0x1);
7557 } else {
7558 movdqa(xmm2, xmm0);
7559 pclmulqdq(xmm2, xmm1, 0x1);
7560 movdqa(xmm3, xmm0);
7561 pand(xmm3, xmm2);
7562 pclmulqdq(xmm0, xmm3, 0x1);
7563 }
7564 psrldq(xmm1, 8);
7565 psrldq(xmm2, 4);
7566 pxor(xmm0, xmm1);
7567 pxor(xmm0, xmm2);
7568
7569 // 8 8-bit folds to compute 32-bit CRC.
7570 for (int j = 0; j < 4; j++) {
7571 fold_8bit_crc32(xmm0, table, xmm1, rax);
7572 }
7573 movdl(crc, xmm0); // mov 32 bits to general register
7574 for (int j = 0; j < 4; j++) {
7575 fold_8bit_crc32(crc, table, rax);
7576 }
7577
7578 BIND(L_tail_restore);
7579 movl(len, tmp); // restore
7580 BIND(L_tail);
7581 andl(len, 0xf);
7582 jccb(Assembler::zero, L_exit);
7583
7584 // Fold the rest of bytes
7585 align(4);
7586 BIND(L_tail_loop);
7587 movsbl(rax, Address(buf, 0)); // load byte with sign extension
7588 update_byte_crc32(crc, rax, table);
7589 increment(buf);
7590 decrementl(len);
7591 jccb(Assembler::greater, L_tail_loop);
7592
7593 BIND(L_exit);
7594 notl(crc); // ~c
7595 }
7596
7597 // Helper function for AVX 512 CRC32
7598 // Fold 512-bit data chunks
7599 void MacroAssembler::fold512bit_crc32_avx512(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf,
7600 Register pos, int offset) {
7601 evmovdquq(xmm3, Address(buf, pos, Address::times_1, offset), Assembler::AVX_512bit);
7602 evpclmulqdq(xtmp, xcrc, xK, 0x10, Assembler::AVX_512bit); // [123:64]
7603 evpclmulqdq(xmm2, xcrc, xK, 0x01, Assembler::AVX_512bit); // [63:0]
7604 evpxorq(xcrc, xtmp, xmm2, Assembler::AVX_512bit /* vector_len */);
7605 evpxorq(xcrc, xcrc, xmm3, Assembler::AVX_512bit /* vector_len */);
7606 }
7607
7608 // Helper function for AVX 512 CRC32
7609 // Compute CRC32 for < 256B buffers
7610 void MacroAssembler::kernel_crc32_avx512_256B(Register crc, Register buf, Register len, Register table, Register pos,
7611 Register tmp1, Register tmp2, Label& L_barrett, Label& L_16B_reduction_loop,
7612 Label& L_get_last_two_xmms, Label& L_128_done, Label& L_cleanup) {
7613
7614 Label L_less_than_32, L_exact_16_left, L_less_than_16_left;
7615 Label L_less_than_8_left, L_less_than_4_left, L_less_than_2_left, L_zero_left;
7616 Label L_only_less_than_4, L_only_less_than_3, L_only_less_than_2;
7617
7618 // check if there is enough buffer to be able to fold 16B at a time
7619 cmpl(len, 32);
7620 jcc(Assembler::less, L_less_than_32);
7621
7622 // if there is, load the constants
7623 movdqu(xmm10, Address(table, 1 * 16)); //rk1 and rk2 in xmm10
7624 movdl(xmm0, crc); // get the initial crc value
7625 movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
7626 pxor(xmm7, xmm0);
7627
7628 // update the buffer pointer
7629 addl(pos, 16);
7630 //update the counter.subtract 32 instead of 16 to save one instruction from the loop
7631 subl(len, 32);
7632 jmp(L_16B_reduction_loop);
7633
7634 bind(L_less_than_32);
7635 //mov initial crc to the return value. this is necessary for zero - length buffers.
7636 movl(rax, crc);
7637 testl(len, len);
7638 jcc(Assembler::equal, L_cleanup);
7639
7640 movdl(xmm0, crc); //get the initial crc value
7641
7642 cmpl(len, 16);
7643 jcc(Assembler::equal, L_exact_16_left);
7644 jcc(Assembler::less, L_less_than_16_left);
7645
7646 movdqu(xmm7, Address(buf, pos, Address::times_1, 0 * 16)); //load the plaintext
7647 pxor(xmm7, xmm0); //xor the initial crc value
7648 addl(pos, 16);
7649 subl(len, 16);
7650 movdqu(xmm10, Address(table, 1 * 16)); // rk1 and rk2 in xmm10
7651 jmp(L_get_last_two_xmms);
7652
7653 bind(L_less_than_16_left);
7654 //use stack space to load data less than 16 bytes, zero - out the 16B in memory first.
7655 pxor(xmm1, xmm1);
7656 movptr(tmp1, rsp);
7657 movdqu(Address(tmp1, 0 * 16), xmm1);
7658
7659 cmpl(len, 4);
7660 jcc(Assembler::less, L_only_less_than_4);
7661
7662 //backup the counter value
7663 movl(tmp2, len);
7664 cmpl(len, 8);
7665 jcc(Assembler::less, L_less_than_8_left);
7666
7667 //load 8 Bytes
7668 movq(rax, Address(buf, pos, Address::times_1, 0 * 16));
7669 movq(Address(tmp1, 0 * 16), rax);
7670 addptr(tmp1, 8);
7671 subl(len, 8);
7672 addl(pos, 8);
7673
7674 bind(L_less_than_8_left);
7675 cmpl(len, 4);
7676 jcc(Assembler::less, L_less_than_4_left);
7677
7678 //load 4 Bytes
7679 movl(rax, Address(buf, pos, Address::times_1, 0));
7680 movl(Address(tmp1, 0 * 16), rax);
7681 addptr(tmp1, 4);
7682 subl(len, 4);
7683 addl(pos, 4);
7684
7685 bind(L_less_than_4_left);
7686 cmpl(len, 2);
7687 jcc(Assembler::less, L_less_than_2_left);
7688
7689 // load 2 Bytes
7690 movw(rax, Address(buf, pos, Address::times_1, 0));
7691 movl(Address(tmp1, 0 * 16), rax);
7692 addptr(tmp1, 2);
7693 subl(len, 2);
7694 addl(pos, 2);
7695
7696 bind(L_less_than_2_left);
7697 cmpl(len, 1);
7698 jcc(Assembler::less, L_zero_left);
7699
7700 // load 1 Byte
7701 movb(rax, Address(buf, pos, Address::times_1, 0));
7702 movb(Address(tmp1, 0 * 16), rax);
7703
7704 bind(L_zero_left);
7705 movdqu(xmm7, Address(rsp, 0));
7706 pxor(xmm7, xmm0); //xor the initial crc value
7707
7708 lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
7709 movdqu(xmm0, Address(rax, tmp2));
7710 pshufb(xmm7, xmm0);
7711 jmp(L_128_done);
7712
7713 bind(L_exact_16_left);
7714 movdqu(xmm7, Address(buf, pos, Address::times_1, 0));
7715 pxor(xmm7, xmm0); //xor the initial crc value
7716 jmp(L_128_done);
7717
7718 bind(L_only_less_than_4);
7719 cmpl(len, 3);
7720 jcc(Assembler::less, L_only_less_than_3);
7721
7722 // load 3 Bytes
7723 movb(rax, Address(buf, pos, Address::times_1, 0));
7724 movb(Address(tmp1, 0), rax);
7725
7726 movb(rax, Address(buf, pos, Address::times_1, 1));
7727 movb(Address(tmp1, 1), rax);
7728
7729 movb(rax, Address(buf, pos, Address::times_1, 2));
7730 movb(Address(tmp1, 2), rax);
7731
7732 movdqu(xmm7, Address(rsp, 0));
7733 pxor(xmm7, xmm0); //xor the initial crc value
7734
7735 pslldq(xmm7, 0x5);
7736 jmp(L_barrett);
7737 bind(L_only_less_than_3);
7738 cmpl(len, 2);
7739 jcc(Assembler::less, L_only_less_than_2);
7740
7741 // load 2 Bytes
7742 movb(rax, Address(buf, pos, Address::times_1, 0));
7743 movb(Address(tmp1, 0), rax);
7744
7745 movb(rax, Address(buf, pos, Address::times_1, 1));
7746 movb(Address(tmp1, 1), rax);
7747
7748 movdqu(xmm7, Address(rsp, 0));
7749 pxor(xmm7, xmm0); //xor the initial crc value
7750
7751 pslldq(xmm7, 0x6);
7752 jmp(L_barrett);
7753
7754 bind(L_only_less_than_2);
7755 //load 1 Byte
7756 movb(rax, Address(buf, pos, Address::times_1, 0));
7757 movb(Address(tmp1, 0), rax);
7758
7759 movdqu(xmm7, Address(rsp, 0));
7760 pxor(xmm7, xmm0); //xor the initial crc value
7761
7762 pslldq(xmm7, 0x7);
7763 }
7764
7765 /**
7766 * Compute CRC32 using AVX512 instructions
7767 * param crc register containing existing CRC (32-bit)
7768 * param buf register pointing to input byte buffer (byte*)
7769 * param len register containing number of bytes
7770 * param table address of crc or crc32c table
7771 * param tmp1 scratch register
7772 * param tmp2 scratch register
7773 * return rax result register
7774 *
7775 * This routine is identical for crc32c with the exception of the precomputed constant
7776 * table which will be passed as the table argument. The calculation steps are
7777 * the same for both variants.
7778 */
7779 void MacroAssembler::kernel_crc32_avx512(Register crc, Register buf, Register len, Register table, Register tmp1, Register tmp2) {
7780 assert_different_registers(crc, buf, len, table, tmp1, tmp2, rax, r12);
7781
7782 Label L_tail, L_tail_restore, L_tail_loop, L_exit, L_align_loop, L_aligned;
7783 Label L_fold_tail, L_fold_128b, L_fold_512b, L_fold_512b_loop, L_fold_tail_loop;
7784 Label L_less_than_256, L_fold_128_B_loop, L_fold_256_B_loop;
7785 Label L_fold_128_B_register, L_final_reduction_for_128, L_16B_reduction_loop;
7786 Label L_128_done, L_get_last_two_xmms, L_barrett, L_cleanup;
7787
7788 const Register pos = r12;
7789 push(r12);
7790 subptr(rsp, 16 * 2 + 8);
7791
7792 // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
7793 // context for the registers used, where all instructions below are using 128-bit mode
7794 // On EVEX without VL and BW, these instructions will all be AVX.
7795 movl(pos, 0);
7796
7797 // check if smaller than 256B
7798 cmpl(len, 256);
7799 jcc(Assembler::less, L_less_than_256);
7800
7801 // load the initial crc value
7802 movdl(xmm10, crc);
7803
7804 // receive the initial 64B data, xor the initial crc value
7805 evmovdquq(xmm0, Address(buf, pos, Address::times_1, 0 * 64), Assembler::AVX_512bit);
7806 evmovdquq(xmm4, Address(buf, pos, Address::times_1, 1 * 64), Assembler::AVX_512bit);
7807 evpxorq(xmm0, xmm0, xmm10, Assembler::AVX_512bit);
7808 evbroadcasti32x4(xmm10, Address(table, 2 * 16), Assembler::AVX_512bit); //zmm10 has rk3 and rk4
7809
7810 subl(len, 256);
7811 cmpl(len, 256);
7812 jcc(Assembler::less, L_fold_128_B_loop);
7813
7814 evmovdquq(xmm7, Address(buf, pos, Address::times_1, 2 * 64), Assembler::AVX_512bit);
7815 evmovdquq(xmm8, Address(buf, pos, Address::times_1, 3 * 64), Assembler::AVX_512bit);
7816 evbroadcasti32x4(xmm16, Address(table, 0 * 16), Assembler::AVX_512bit); //zmm16 has rk-1 and rk-2
7817 subl(len, 256);
7818
7819 bind(L_fold_256_B_loop);
7820 addl(pos, 256);
7821 fold512bit_crc32_avx512(xmm0, xmm16, xmm1, buf, pos, 0 * 64);
7822 fold512bit_crc32_avx512(xmm4, xmm16, xmm1, buf, pos, 1 * 64);
7823 fold512bit_crc32_avx512(xmm7, xmm16, xmm1, buf, pos, 2 * 64);
7824 fold512bit_crc32_avx512(xmm8, xmm16, xmm1, buf, pos, 3 * 64);
7825
7826 subl(len, 256);
7827 jcc(Assembler::greaterEqual, L_fold_256_B_loop);
7828
7829 // Fold 256 into 128
7830 addl(pos, 256);
7831 evpclmulqdq(xmm1, xmm0, xmm10, 0x01, Assembler::AVX_512bit);
7832 evpclmulqdq(xmm2, xmm0, xmm10, 0x10, Assembler::AVX_512bit);
7833 vpternlogq(xmm7, 0x96, xmm1, xmm2, Assembler::AVX_512bit); // xor ABC
7834
7835 evpclmulqdq(xmm5, xmm4, xmm10, 0x01, Assembler::AVX_512bit);
7836 evpclmulqdq(xmm6, xmm4, xmm10, 0x10, Assembler::AVX_512bit);
7837 vpternlogq(xmm8, 0x96, xmm5, xmm6, Assembler::AVX_512bit); // xor ABC
7838
7839 evmovdquq(xmm0, xmm7, Assembler::AVX_512bit);
7840 evmovdquq(xmm4, xmm8, Assembler::AVX_512bit);
7841
7842 addl(len, 128);
7843 jmp(L_fold_128_B_register);
7844
7845 // at this section of the code, there is 128 * x + y(0 <= y<128) bytes of buffer.The fold_128_B_loop
7846 // loop will fold 128B at a time until we have 128 + y Bytes of buffer
7847
7848 // fold 128B at a time.This section of the code folds 8 xmm registers in parallel
7849 bind(L_fold_128_B_loop);
7850 addl(pos, 128);
7851 fold512bit_crc32_avx512(xmm0, xmm10, xmm1, buf, pos, 0 * 64);
7852 fold512bit_crc32_avx512(xmm4, xmm10, xmm1, buf, pos, 1 * 64);
7853
7854 subl(len, 128);
7855 jcc(Assembler::greaterEqual, L_fold_128_B_loop);
7856
7857 addl(pos, 128);
7858
7859 // at this point, the buffer pointer is pointing at the last y Bytes of the buffer, where 0 <= y < 128
7860 // the 128B of folded data is in 8 of the xmm registers : xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7
7861 bind(L_fold_128_B_register);
7862 evmovdquq(xmm16, Address(table, 5 * 16), Assembler::AVX_512bit); // multiply by rk9-rk16
7863 evmovdquq(xmm11, Address(table, 9 * 16), Assembler::AVX_512bit); // multiply by rk17-rk20, rk1,rk2, 0,0
7864 evpclmulqdq(xmm1, xmm0, xmm16, 0x01, Assembler::AVX_512bit);
7865 evpclmulqdq(xmm2, xmm0, xmm16, 0x10, Assembler::AVX_512bit);
7866 // save last that has no multiplicand
7867 vextracti64x2(xmm7, xmm4, 3);
7868
7869 evpclmulqdq(xmm5, xmm4, xmm11, 0x01, Assembler::AVX_512bit);
7870 evpclmulqdq(xmm6, xmm4, xmm11, 0x10, Assembler::AVX_512bit);
7871 // Needed later in reduction loop
7872 movdqu(xmm10, Address(table, 1 * 16));
7873 vpternlogq(xmm1, 0x96, xmm2, xmm5, Assembler::AVX_512bit); // xor ABC
7874 vpternlogq(xmm1, 0x96, xmm6, xmm7, Assembler::AVX_512bit); // xor ABC
7875
7876 // Swap 1,0,3,2 - 01 00 11 10
7877 evshufi64x2(xmm8, xmm1, xmm1, 0x4e, Assembler::AVX_512bit);
7878 evpxorq(xmm8, xmm8, xmm1, Assembler::AVX_256bit);
7879 vextracti128(xmm5, xmm8, 1);
7880 evpxorq(xmm7, xmm5, xmm8, Assembler::AVX_128bit);
7881
7882 // instead of 128, we add 128 - 16 to the loop counter to save 1 instruction from the loop
7883 // instead of a cmp instruction, we use the negative flag with the jl instruction
7884 addl(len, 128 - 16);
7885 jcc(Assembler::less, L_final_reduction_for_128);
7886
7887 bind(L_16B_reduction_loop);
7888 vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
7889 vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
7890 vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
7891 movdqu(xmm0, Address(buf, pos, Address::times_1, 0 * 16));
7892 vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
7893 addl(pos, 16);
7894 subl(len, 16);
7895 jcc(Assembler::greaterEqual, L_16B_reduction_loop);
7896
7897 bind(L_final_reduction_for_128);
7898 addl(len, 16);
7899 jcc(Assembler::equal, L_128_done);
7900
7901 bind(L_get_last_two_xmms);
7902 movdqu(xmm2, xmm7);
7903 addl(pos, len);
7904 movdqu(xmm1, Address(buf, pos, Address::times_1, -16));
7905 subl(pos, len);
7906
7907 // get rid of the extra data that was loaded before
7908 // load the shift constant
7909 lea(rax, ExternalAddress(StubRoutines::x86::shuf_table_crc32_avx512_addr()));
7910 movdqu(xmm0, Address(rax, len));
7911 addl(rax, len);
7912
7913 vpshufb(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
7914 //Change mask to 512
7915 vpxor(xmm0, xmm0, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 2 * 16), Assembler::AVX_128bit, tmp2);
7916 vpshufb(xmm2, xmm2, xmm0, Assembler::AVX_128bit);
7917
7918 blendvpb(xmm2, xmm2, xmm1, xmm0, Assembler::AVX_128bit);
7919 vpclmulqdq(xmm8, xmm7, xmm10, 0x01);
7920 vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
7921 vpxor(xmm7, xmm7, xmm8, Assembler::AVX_128bit);
7922 vpxor(xmm7, xmm7, xmm2, Assembler::AVX_128bit);
7923
7924 bind(L_128_done);
7925 // compute crc of a 128-bit value
7926 movdqu(xmm10, Address(table, 3 * 16));
7927 movdqu(xmm0, xmm7);
7928
7929 // 64b fold
7930 vpclmulqdq(xmm7, xmm7, xmm10, 0x0);
7931 vpsrldq(xmm0, xmm0, 0x8, Assembler::AVX_128bit);
7932 vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
7933
7934 // 32b fold
7935 movdqu(xmm0, xmm7);
7936 vpslldq(xmm7, xmm7, 0x4, Assembler::AVX_128bit);
7937 vpclmulqdq(xmm7, xmm7, xmm10, 0x10);
7938 vpxor(xmm7, xmm7, xmm0, Assembler::AVX_128bit);
7939 jmp(L_barrett);
7940
7941 bind(L_less_than_256);
7942 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);
7943
7944 //barrett reduction
7945 bind(L_barrett);
7946 vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr() + 1 * 16), Assembler::AVX_128bit, tmp2);
7947 movdqu(xmm1, xmm7);
7948 movdqu(xmm2, xmm7);
7949 movdqu(xmm10, Address(table, 4 * 16));
7950
7951 pclmulqdq(xmm7, xmm10, 0x0);
7952 pxor(xmm7, xmm2);
7953 vpand(xmm7, xmm7, ExternalAddress(StubRoutines::x86::crc_by128_masks_avx512_addr()), Assembler::AVX_128bit, tmp2);
7954 movdqu(xmm2, xmm7);
7955 pclmulqdq(xmm7, xmm10, 0x10);
7956 pxor(xmm7, xmm2);
7957 pxor(xmm7, xmm1);
7958 pextrd(crc, xmm7, 2);
7959
7960 bind(L_cleanup);
7961 addptr(rsp, 16 * 2 + 8);
7962 pop(r12);
7963 }
7964
7965 // S. Gueron / Information Processing Letters 112 (2012) 184
7966 // Algorithm 4: Computing carry-less multiplication using a precomputed lookup table.
7967 // Input: A 32 bit value B = [byte3, byte2, byte1, byte0].
7968 // Output: the 64-bit carry-less product of B * CONST
7969 void MacroAssembler::crc32c_ipl_alg4(Register in, uint32_t n,
7970 Register tmp1, Register tmp2, Register tmp3) {
7971 lea(tmp3, ExternalAddress(StubRoutines::crc32c_table_addr()));
7972 if (n > 0) {
7973 addq(tmp3, n * 256 * 8);
7974 }
7975 // Q1 = TABLEExt[n][B & 0xFF];
7976 movl(tmp1, in);
7977 andl(tmp1, 0x000000FF);
7978 shll(tmp1, 3);
7979 addq(tmp1, tmp3);
7980 movq(tmp1, Address(tmp1, 0));
7981
7982 // Q2 = TABLEExt[n][B >> 8 & 0xFF];
7983 movl(tmp2, in);
7984 shrl(tmp2, 8);
7985 andl(tmp2, 0x000000FF);
7986 shll(tmp2, 3);
7987 addq(tmp2, tmp3);
7988 movq(tmp2, Address(tmp2, 0));
7989
7990 shlq(tmp2, 8);
7991 xorq(tmp1, tmp2);
7992
7993 // Q3 = TABLEExt[n][B >> 16 & 0xFF];
7994 movl(tmp2, in);
7995 shrl(tmp2, 16);
7996 andl(tmp2, 0x000000FF);
7997 shll(tmp2, 3);
7998 addq(tmp2, tmp3);
7999 movq(tmp2, Address(tmp2, 0));
8000
8001 shlq(tmp2, 16);
8002 xorq(tmp1, tmp2);
8003
8004 // Q4 = TABLEExt[n][B >> 24 & 0xFF];
8005 shrl(in, 24);
8006 andl(in, 0x000000FF);
8007 shll(in, 3);
8008 addq(in, tmp3);
8009 movq(in, Address(in, 0));
8010
8011 shlq(in, 24);
8012 xorq(in, tmp1);
8013 // return Q1 ^ Q2 << 8 ^ Q3 << 16 ^ Q4 << 24;
8014 }
8015
8016 void MacroAssembler::crc32c_pclmulqdq(XMMRegister w_xtmp1,
8017 Register in_out,
8018 uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported,
8019 XMMRegister w_xtmp2,
8020 Register tmp1,
8021 Register n_tmp2, Register n_tmp3) {
8022 if (is_pclmulqdq_supported) {
8023 movdl(w_xtmp1, in_out); // modified blindly
8024
8025 movl(tmp1, const_or_pre_comp_const_index);
8026 movdl(w_xtmp2, tmp1);
8027 pclmulqdq(w_xtmp1, w_xtmp2, 0);
8028
8029 movdq(in_out, w_xtmp1);
8030 } else {
8031 crc32c_ipl_alg4(in_out, const_or_pre_comp_const_index, tmp1, n_tmp2, n_tmp3);
8032 }
8033 }
8034
8035 // Recombination Alternative 2: No bit-reflections
8036 // T1 = (CRC_A * U1) << 1
8037 // T2 = (CRC_B * U2) << 1
8038 // C1 = T1 >> 32
8039 // C2 = T2 >> 32
8040 // T1 = T1 & 0xFFFFFFFF
8041 // T2 = T2 & 0xFFFFFFFF
8042 // T1 = CRC32(0, T1)
8043 // T2 = CRC32(0, T2)
8044 // C1 = C1 ^ T1
8045 // C2 = C2 ^ T2
8046 // CRC = C1 ^ C2 ^ CRC_C
8047 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,
8048 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
8049 Register tmp1, Register tmp2,
8050 Register n_tmp3) {
8051 crc32c_pclmulqdq(w_xtmp1, in_out, const_or_pre_comp_const_index_u1, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
8052 crc32c_pclmulqdq(w_xtmp2, in1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, w_xtmp3, tmp1, tmp2, n_tmp3);
8053 shlq(in_out, 1);
8054 movl(tmp1, in_out);
8055 shrq(in_out, 32);
8056 xorl(tmp2, tmp2);
8057 crc32(tmp2, tmp1, 4);
8058 xorl(in_out, tmp2); // we don't care about upper 32 bit contents here
8059 shlq(in1, 1);
8060 movl(tmp1, in1);
8061 shrq(in1, 32);
8062 xorl(tmp2, tmp2);
8063 crc32(tmp2, tmp1, 4);
8064 xorl(in1, tmp2);
8065 xorl(in_out, in1);
8066 xorl(in_out, in2);
8067 }
8068
8069 // Set N to predefined value
8070 // Subtract from a length of a buffer
8071 // execute in a loop:
8072 // CRC_A = 0xFFFFFFFF, CRC_B = 0, CRC_C = 0
8073 // for i = 1 to N do
8074 // CRC_A = CRC32(CRC_A, A[i])
8075 // CRC_B = CRC32(CRC_B, B[i])
8076 // CRC_C = CRC32(CRC_C, C[i])
8077 // end for
8078 // Recombine
8079 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,
8080 Register in_out1, Register in_out2, Register in_out3,
8081 Register tmp1, Register tmp2, Register tmp3,
8082 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
8083 Register tmp4, Register tmp5,
8084 Register n_tmp6) {
8085 Label L_processPartitions;
8086 Label L_processPartition;
8087 Label L_exit;
8088
8089 bind(L_processPartitions);
8090 cmpl(in_out1, 3 * size);
8091 jcc(Assembler::less, L_exit);
8092 xorl(tmp1, tmp1);
8093 xorl(tmp2, tmp2);
8094 movq(tmp3, in_out2);
8095 addq(tmp3, size);
8096
8097 bind(L_processPartition);
8098 crc32(in_out3, Address(in_out2, 0), 8);
8099 crc32(tmp1, Address(in_out2, size), 8);
8100 crc32(tmp2, Address(in_out2, size * 2), 8);
8101 addq(in_out2, 8);
8102 cmpq(in_out2, tmp3);
8103 jcc(Assembler::less, L_processPartition);
8104 crc32c_rec_alt2(const_or_pre_comp_const_index_u1, const_or_pre_comp_const_index_u2, is_pclmulqdq_supported, in_out3, tmp1, tmp2,
8105 w_xtmp1, w_xtmp2, w_xtmp3,
8106 tmp4, tmp5,
8107 n_tmp6);
8108 addq(in_out2, 2 * size);
8109 subl(in_out1, 3 * size);
8110 jmp(L_processPartitions);
8111
8112 bind(L_exit);
8113 }
8114
8115 // Algorithm 2: Pipelined usage of the CRC32 instruction.
8116 // Input: A buffer I of L bytes.
8117 // Output: the CRC32C value of the buffer.
8118 // Notations:
8119 // Write L = 24N + r, with N = floor (L/24).
8120 // r = L mod 24 (0 <= r < 24).
8121 // Consider I as the concatenation of A|B|C|R, where A, B, C, each,
8122 // N quadwords, and R consists of r bytes.
8123 // A[j] = I [8j+7:8j], j= 0, 1, ..., N-1
8124 // B[j] = I [N + 8j+7:N + 8j], j= 0, 1, ..., N-1
8125 // C[j] = I [2N + 8j+7:2N + 8j], j= 0, 1, ..., N-1
8126 // if r > 0 R[j] = I [3N +j], j= 0, 1, ...,r-1
8127 void MacroAssembler::crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2,
8128 Register tmp1, Register tmp2, Register tmp3,
8129 Register tmp4, Register tmp5, Register tmp6,
8130 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
8131 bool is_pclmulqdq_supported) {
8132 uint32_t const_or_pre_comp_const_index[CRC32C_NUM_PRECOMPUTED_CONSTANTS];
8133 Label L_wordByWord;
8134 Label L_byteByByteProlog;
8135 Label L_byteByByte;
8136 Label L_exit;
8137
8138 if (is_pclmulqdq_supported ) {
8139 const_or_pre_comp_const_index[1] = *(uint32_t *)StubRoutines::crc32c_table_addr();
8140 const_or_pre_comp_const_index[0] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 1);
8141
8142 const_or_pre_comp_const_index[3] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 2);
8143 const_or_pre_comp_const_index[2] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 3);
8144
8145 const_or_pre_comp_const_index[5] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 4);
8146 const_or_pre_comp_const_index[4] = *((uint32_t *)StubRoutines::crc32c_table_addr() + 5);
8147 assert((CRC32C_NUM_PRECOMPUTED_CONSTANTS - 1 ) == 5, "Checking whether you declared all of the constants based on the number of \"chunks\"");
8148 } else {
8149 const_or_pre_comp_const_index[0] = 1;
8150 const_or_pre_comp_const_index[1] = 0;
8151
8152 const_or_pre_comp_const_index[2] = 3;
8153 const_or_pre_comp_const_index[3] = 2;
8154
8155 const_or_pre_comp_const_index[4] = 5;
8156 const_or_pre_comp_const_index[5] = 4;
8157 }
8158 crc32c_proc_chunk(CRC32C_HIGH, const_or_pre_comp_const_index[0], const_or_pre_comp_const_index[1], is_pclmulqdq_supported,
8159 in2, in1, in_out,
8160 tmp1, tmp2, tmp3,
8161 w_xtmp1, w_xtmp2, w_xtmp3,
8162 tmp4, tmp5,
8163 tmp6);
8164 crc32c_proc_chunk(CRC32C_MIDDLE, const_or_pre_comp_const_index[2], const_or_pre_comp_const_index[3], is_pclmulqdq_supported,
8165 in2, in1, in_out,
8166 tmp1, tmp2, tmp3,
8167 w_xtmp1, w_xtmp2, w_xtmp3,
8168 tmp4, tmp5,
8169 tmp6);
8170 crc32c_proc_chunk(CRC32C_LOW, const_or_pre_comp_const_index[4], const_or_pre_comp_const_index[5], is_pclmulqdq_supported,
8171 in2, in1, in_out,
8172 tmp1, tmp2, tmp3,
8173 w_xtmp1, w_xtmp2, w_xtmp3,
8174 tmp4, tmp5,
8175 tmp6);
8176 movl(tmp1, in2);
8177 andl(tmp1, 0x00000007);
8178 negl(tmp1);
8179 addl(tmp1, in2);
8180 addq(tmp1, in1);
8181
8182 cmpq(in1, tmp1);
8183 jccb(Assembler::greaterEqual, L_byteByByteProlog);
8184 align(16);
8185 BIND(L_wordByWord);
8186 crc32(in_out, Address(in1, 0), 8);
8187 addq(in1, 8);
8188 cmpq(in1, tmp1);
8189 jcc(Assembler::less, L_wordByWord);
8190
8191 BIND(L_byteByByteProlog);
8192 andl(in2, 0x00000007);
8193 movl(tmp2, 1);
8194
8195 cmpl(tmp2, in2);
8196 jccb(Assembler::greater, L_exit);
8197 BIND(L_byteByByte);
8198 crc32(in_out, Address(in1, 0), 1);
8199 incq(in1);
8200 incl(tmp2);
8201 cmpl(tmp2, in2);
8202 jcc(Assembler::lessEqual, L_byteByByte);
8203
8204 BIND(L_exit);
8205 }
8206 #undef BIND
8207 #undef BLOCK_COMMENT
8208
8209 // Compress char[] array to byte[].
8210 // Intrinsic for java.lang.StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
8211 // Return the array length if every element in array can be encoded,
8212 // otherwise, the index of first non-latin1 (> 0xff) character.
8213 // @IntrinsicCandidate
8214 // public static int compress(char[] src, int srcOff, byte[] dst, int dstOff, int len) {
8215 // for (int i = 0; i < len; i++) {
8216 // char c = src[srcOff];
8217 // if (c > 0xff) {
8218 // return i; // return index of non-latin1 char
8219 // }
8220 // dst[dstOff] = (byte)c;
8221 // srcOff++;
8222 // dstOff++;
8223 // }
8224 // return len;
8225 // }
8226 void MacroAssembler::char_array_compress(Register src, Register dst, Register len,
8227 XMMRegister tmp1Reg, XMMRegister tmp2Reg,
8228 XMMRegister tmp3Reg, XMMRegister tmp4Reg,
8229 Register tmp5, Register result, KRegister mask1, KRegister mask2) {
8230 Label copy_chars_loop, done, reset_sp, copy_tail;
8231
8232 // rsi: src
8233 // rdi: dst
8234 // rdx: len
8235 // rcx: tmp5
8236 // rax: result
8237
8238 // rsi holds start addr of source char[] to be compressed
8239 // rdi holds start addr of destination byte[]
8240 // rdx holds length
8241
8242 assert(len != result, "");
8243
8244 // save length for return
8245 movl(result, len);
8246
8247 if ((AVX3Threshold == 0) && (UseAVX > 2) && // AVX512
8248 VM_Version::supports_avx512vlbw() &&
8249 VM_Version::supports_bmi2()) {
8250
8251 Label copy_32_loop, copy_loop_tail, below_threshold, reset_for_copy_tail;
8252
8253 // alignment
8254 Label post_alignment;
8255
8256 // if length of the string is less than 32, handle it the old fashioned way
8257 testl(len, -32);
8258 jcc(Assembler::zero, below_threshold);
8259
8260 // First check whether a character is compressible ( <= 0xFF).
8261 // Create mask to test for Unicode chars inside zmm vector
8262 movl(tmp5, 0x00FF);
8263 evpbroadcastw(tmp2Reg, tmp5, Assembler::AVX_512bit);
8264
8265 testl(len, -64);
8266 jccb(Assembler::zero, post_alignment);
8267
8268 movl(tmp5, dst);
8269 andl(tmp5, (32 - 1));
8270 negl(tmp5);
8271 andl(tmp5, (32 - 1));
8272
8273 // bail out when there is nothing to be done
8274 testl(tmp5, 0xFFFFFFFF);
8275 jccb(Assembler::zero, post_alignment);
8276
8277 // ~(~0 << len), where len is the # of remaining elements to process
8278 movl(len, 0xFFFFFFFF);
8279 shlxl(len, len, tmp5);
8280 notl(len);
8281 kmovdl(mask2, len);
8282 movl(len, result);
8283
8284 evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
8285 evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
8286 ktestd(mask1, mask2);
8287 jcc(Assembler::carryClear, copy_tail);
8288
8289 evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
8290
8291 addptr(src, tmp5);
8292 addptr(src, tmp5);
8293 addptr(dst, tmp5);
8294 subl(len, tmp5);
8295
8296 bind(post_alignment);
8297 // end of alignment
8298
8299 movl(tmp5, len);
8300 andl(tmp5, (32 - 1)); // tail count (in chars)
8301 andl(len, ~(32 - 1)); // vector count (in chars)
8302 jccb(Assembler::zero, copy_loop_tail);
8303
8304 lea(src, Address(src, len, Address::times_2));
8305 lea(dst, Address(dst, len, Address::times_1));
8306 negptr(len);
8307
8308 bind(copy_32_loop);
8309 evmovdquw(tmp1Reg, Address(src, len, Address::times_2), Assembler::AVX_512bit);
8310 evpcmpuw(mask1, tmp1Reg, tmp2Reg, Assembler::le, Assembler::AVX_512bit);
8311 kortestdl(mask1, mask1);
8312 jccb(Assembler::carryClear, reset_for_copy_tail);
8313
8314 // All elements in current processed chunk are valid candidates for
8315 // compression. Write a truncated byte elements to the memory.
8316 evpmovwb(Address(dst, len, Address::times_1), tmp1Reg, Assembler::AVX_512bit);
8317 addptr(len, 32);
8318 jccb(Assembler::notZero, copy_32_loop);
8319
8320 bind(copy_loop_tail);
8321 // bail out when there is nothing to be done
8322 testl(tmp5, 0xFFFFFFFF);
8323 jcc(Assembler::zero, done);
8324
8325 movl(len, tmp5);
8326
8327 // ~(~0 << len), where len is the # of remaining elements to process
8328 movl(tmp5, 0xFFFFFFFF);
8329 shlxl(tmp5, tmp5, len);
8330 notl(tmp5);
8331
8332 kmovdl(mask2, tmp5);
8333
8334 evmovdquw(tmp1Reg, mask2, Address(src, 0), /*merge*/ false, Assembler::AVX_512bit);
8335 evpcmpw(mask1, mask2, tmp1Reg, tmp2Reg, Assembler::le, /*signed*/ false, Assembler::AVX_512bit);
8336 ktestd(mask1, mask2);
8337 jcc(Assembler::carryClear, copy_tail);
8338
8339 evpmovwb(Address(dst, 0), mask2, tmp1Reg, Assembler::AVX_512bit);
8340 jmp(done);
8341
8342 bind(reset_for_copy_tail);
8343 lea(src, Address(src, tmp5, Address::times_2));
8344 lea(dst, Address(dst, tmp5, Address::times_1));
8345 subptr(len, tmp5);
8346 jmp(copy_chars_loop);
8347
8348 bind(below_threshold);
8349 }
8350
8351 if (UseSSE42Intrinsics) {
8352 Label copy_32_loop, copy_16, copy_tail_sse, reset_for_copy_tail;
8353
8354 // vectored compression
8355 testl(len, 0xfffffff8);
8356 jcc(Assembler::zero, copy_tail);
8357
8358 movl(tmp5, 0xff00ff00); // create mask to test for Unicode chars in vectors
8359 movdl(tmp1Reg, tmp5);
8360 pshufd(tmp1Reg, tmp1Reg, 0); // store Unicode mask in tmp1Reg
8361
8362 andl(len, 0xfffffff0);
8363 jccb(Assembler::zero, copy_16);
8364
8365 // compress 16 chars per iter
8366 pxor(tmp4Reg, tmp4Reg);
8367
8368 lea(src, Address(src, len, Address::times_2));
8369 lea(dst, Address(dst, len, Address::times_1));
8370 negptr(len);
8371
8372 bind(copy_32_loop);
8373 movdqu(tmp2Reg, Address(src, len, Address::times_2)); // load 1st 8 characters
8374 por(tmp4Reg, tmp2Reg);
8375 movdqu(tmp3Reg, Address(src, len, Address::times_2, 16)); // load next 8 characters
8376 por(tmp4Reg, tmp3Reg);
8377 ptest(tmp4Reg, tmp1Reg); // check for Unicode chars in next vector
8378 jccb(Assembler::notZero, reset_for_copy_tail);
8379 packuswb(tmp2Reg, tmp3Reg); // only ASCII chars; compress each to 1 byte
8380 movdqu(Address(dst, len, Address::times_1), tmp2Reg);
8381 addptr(len, 16);
8382 jccb(Assembler::notZero, copy_32_loop);
8383
8384 // compress next vector of 8 chars (if any)
8385 bind(copy_16);
8386 // len = 0
8387 testl(result, 0x00000008); // check if there's a block of 8 chars to compress
8388 jccb(Assembler::zero, copy_tail_sse);
8389
8390 pxor(tmp3Reg, tmp3Reg);
8391
8392 movdqu(tmp2Reg, Address(src, 0));
8393 ptest(tmp2Reg, tmp1Reg); // check for Unicode chars in vector
8394 jccb(Assembler::notZero, reset_for_copy_tail);
8395 packuswb(tmp2Reg, tmp3Reg); // only LATIN1 chars; compress each to 1 byte
8396 movq(Address(dst, 0), tmp2Reg);
8397 addptr(src, 16);
8398 addptr(dst, 8);
8399 jmpb(copy_tail_sse);
8400
8401 bind(reset_for_copy_tail);
8402 movl(tmp5, result);
8403 andl(tmp5, 0x0000000f);
8404 lea(src, Address(src, tmp5, Address::times_2));
8405 lea(dst, Address(dst, tmp5, Address::times_1));
8406 subptr(len, tmp5);
8407 jmpb(copy_chars_loop);
8408
8409 bind(copy_tail_sse);
8410 movl(len, result);
8411 andl(len, 0x00000007); // tail count (in chars)
8412 }
8413 // compress 1 char per iter
8414 bind(copy_tail);
8415 testl(len, len);
8416 jccb(Assembler::zero, done);
8417 lea(src, Address(src, len, Address::times_2));
8418 lea(dst, Address(dst, len, Address::times_1));
8419 negptr(len);
8420
8421 bind(copy_chars_loop);
8422 load_unsigned_short(tmp5, Address(src, len, Address::times_2));
8423 testl(tmp5, 0xff00); // check if Unicode char
8424 jccb(Assembler::notZero, reset_sp);
8425 movb(Address(dst, len, Address::times_1), tmp5); // ASCII char; compress to 1 byte
8426 increment(len);
8427 jccb(Assembler::notZero, copy_chars_loop);
8428
8429 // add len then return (len will be zero if compress succeeded, otherwise negative)
8430 bind(reset_sp);
8431 addl(result, len);
8432
8433 bind(done);
8434 }
8435
8436 // Inflate byte[] array to char[].
8437 // ..\jdk\src\java.base\share\classes\java\lang\StringLatin1.java
8438 // @IntrinsicCandidate
8439 // private static void inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len) {
8440 // for (int i = 0; i < len; i++) {
8441 // dst[dstOff++] = (char)(src[srcOff++] & 0xff);
8442 // }
8443 // }
8444 void MacroAssembler::byte_array_inflate(Register src, Register dst, Register len,
8445 XMMRegister tmp1, Register tmp2, KRegister mask) {
8446 Label copy_chars_loop, done, below_threshold, avx3_threshold;
8447 // rsi: src
8448 // rdi: dst
8449 // rdx: len
8450 // rcx: tmp2
8451
8452 // rsi holds start addr of source byte[] to be inflated
8453 // rdi holds start addr of destination char[]
8454 // rdx holds length
8455 assert_different_registers(src, dst, len, tmp2);
8456 movl(tmp2, len);
8457 if ((UseAVX > 2) && // AVX512
8458 VM_Version::supports_avx512vlbw() &&
8459 VM_Version::supports_bmi2()) {
8460
8461 Label copy_32_loop, copy_tail;
8462 Register tmp3_aliased = len;
8463
8464 // if length of the string is less than 16, handle it in an old fashioned way
8465 testl(len, -16);
8466 jcc(Assembler::zero, below_threshold);
8467
8468 testl(len, -1 * AVX3Threshold);
8469 jcc(Assembler::zero, avx3_threshold);
8470
8471 // In order to use only one arithmetic operation for the main loop we use
8472 // this pre-calculation
8473 andl(tmp2, (32 - 1)); // tail count (in chars), 32 element wide loop
8474 andl(len, -32); // vector count
8475 jccb(Assembler::zero, copy_tail);
8476
8477 lea(src, Address(src, len, Address::times_1));
8478 lea(dst, Address(dst, len, Address::times_2));
8479 negptr(len);
8480
8481
8482 // inflate 32 chars per iter
8483 bind(copy_32_loop);
8484 vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_512bit);
8485 evmovdquw(Address(dst, len, Address::times_2), tmp1, Assembler::AVX_512bit);
8486 addptr(len, 32);
8487 jcc(Assembler::notZero, copy_32_loop);
8488
8489 bind(copy_tail);
8490 // bail out when there is nothing to be done
8491 testl(tmp2, -1); // we don't destroy the contents of tmp2 here
8492 jcc(Assembler::zero, done);
8493
8494 // ~(~0 << length), where length is the # of remaining elements to process
8495 movl(tmp3_aliased, -1);
8496 shlxl(tmp3_aliased, tmp3_aliased, tmp2);
8497 notl(tmp3_aliased);
8498 kmovdl(mask, tmp3_aliased);
8499 evpmovzxbw(tmp1, mask, Address(src, 0), Assembler::AVX_512bit);
8500 evmovdquw(Address(dst, 0), mask, tmp1, /*merge*/ true, Assembler::AVX_512bit);
8501
8502 jmp(done);
8503 bind(avx3_threshold);
8504 }
8505 if (UseSSE42Intrinsics) {
8506 Label copy_16_loop, copy_8_loop, copy_bytes, copy_new_tail, copy_tail;
8507
8508 if (UseAVX > 1) {
8509 andl(tmp2, (16 - 1));
8510 andl(len, -16);
8511 jccb(Assembler::zero, copy_new_tail);
8512 } else {
8513 andl(tmp2, 0x00000007); // tail count (in chars)
8514 andl(len, 0xfffffff8); // vector count (in chars)
8515 jccb(Assembler::zero, copy_tail);
8516 }
8517
8518 // vectored inflation
8519 lea(src, Address(src, len, Address::times_1));
8520 lea(dst, Address(dst, len, Address::times_2));
8521 negptr(len);
8522
8523 if (UseAVX > 1) {
8524 bind(copy_16_loop);
8525 vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_256bit);
8526 vmovdqu(Address(dst, len, Address::times_2), tmp1);
8527 addptr(len, 16);
8528 jcc(Assembler::notZero, copy_16_loop);
8529
8530 bind(below_threshold);
8531 bind(copy_new_tail);
8532 movl(len, tmp2);
8533 andl(tmp2, 0x00000007);
8534 andl(len, 0xFFFFFFF8);
8535 jccb(Assembler::zero, copy_tail);
8536
8537 pmovzxbw(tmp1, Address(src, 0));
8538 movdqu(Address(dst, 0), tmp1);
8539 addptr(src, 8);
8540 addptr(dst, 2 * 8);
8541
8542 jmp(copy_tail, true);
8543 }
8544
8545 // inflate 8 chars per iter
8546 bind(copy_8_loop);
8547 pmovzxbw(tmp1, Address(src, len, Address::times_1)); // unpack to 8 words
8548 movdqu(Address(dst, len, Address::times_2), tmp1);
8549 addptr(len, 8);
8550 jcc(Assembler::notZero, copy_8_loop);
8551
8552 bind(copy_tail);
8553 movl(len, tmp2);
8554
8555 cmpl(len, 4);
8556 jccb(Assembler::less, copy_bytes);
8557
8558 movdl(tmp1, Address(src, 0)); // load 4 byte chars
8559 pmovzxbw(tmp1, tmp1);
8560 movq(Address(dst, 0), tmp1);
8561 subptr(len, 4);
8562 addptr(src, 4);
8563 addptr(dst, 8);
8564
8565 bind(copy_bytes);
8566 } else {
8567 bind(below_threshold);
8568 }
8569
8570 testl(len, len);
8571 jccb(Assembler::zero, done);
8572 lea(src, Address(src, len, Address::times_1));
8573 lea(dst, Address(dst, len, Address::times_2));
8574 negptr(len);
8575
8576 // inflate 1 char per iter
8577 bind(copy_chars_loop);
8578 load_unsigned_byte(tmp2, Address(src, len, Address::times_1)); // load byte char
8579 movw(Address(dst, len, Address::times_2), tmp2); // inflate byte char to word
8580 increment(len);
8581 jcc(Assembler::notZero, copy_chars_loop);
8582
8583 bind(done);
8584 }
8585
8586 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, XMMRegister src, bool merge, int vector_len) {
8587 switch(type) {
8588 case T_BYTE:
8589 case T_BOOLEAN:
8590 evmovdqub(dst, kmask, src, merge, vector_len);
8591 break;
8592 case T_CHAR:
8593 case T_SHORT:
8594 evmovdquw(dst, kmask, src, merge, vector_len);
8595 break;
8596 case T_INT:
8597 case T_FLOAT:
8598 evmovdqul(dst, kmask, src, merge, vector_len);
8599 break;
8600 case T_LONG:
8601 case T_DOUBLE:
8602 evmovdquq(dst, kmask, src, merge, vector_len);
8603 break;
8604 default:
8605 fatal("Unexpected type argument %s", type2name(type));
8606 break;
8607 }
8608 }
8609
8610
8611 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address src, bool merge, int vector_len) {
8612 switch(type) {
8613 case T_BYTE:
8614 case T_BOOLEAN:
8615 evmovdqub(dst, kmask, src, merge, vector_len);
8616 break;
8617 case T_CHAR:
8618 case T_SHORT:
8619 evmovdquw(dst, kmask, src, merge, vector_len);
8620 break;
8621 case T_INT:
8622 case T_FLOAT:
8623 evmovdqul(dst, kmask, src, merge, vector_len);
8624 break;
8625 case T_LONG:
8626 case T_DOUBLE:
8627 evmovdquq(dst, kmask, src, merge, vector_len);
8628 break;
8629 default:
8630 fatal("Unexpected type argument %s", type2name(type));
8631 break;
8632 }
8633 }
8634
8635 void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister src, bool merge, int vector_len) {
8636 switch(type) {
8637 case T_BYTE:
8638 case T_BOOLEAN:
8639 evmovdqub(dst, kmask, src, merge, vector_len);
8640 break;
8641 case T_CHAR:
8642 case T_SHORT:
8643 evmovdquw(dst, kmask, src, merge, vector_len);
8644 break;
8645 case T_INT:
8646 case T_FLOAT:
8647 evmovdqul(dst, kmask, src, merge, vector_len);
8648 break;
8649 case T_LONG:
8650 case T_DOUBLE:
8651 evmovdquq(dst, kmask, src, merge, vector_len);
8652 break;
8653 default:
8654 fatal("Unexpected type argument %s", type2name(type));
8655 break;
8656 }
8657 }
8658
8659 void MacroAssembler::knot(uint masklen, KRegister dst, KRegister src, KRegister ktmp, Register rtmp) {
8660 switch(masklen) {
8661 case 2:
8662 knotbl(dst, src);
8663 movl(rtmp, 3);
8664 kmovbl(ktmp, rtmp);
8665 kandbl(dst, ktmp, dst);
8666 break;
8667 case 4:
8668 knotbl(dst, src);
8669 movl(rtmp, 15);
8670 kmovbl(ktmp, rtmp);
8671 kandbl(dst, ktmp, dst);
8672 break;
8673 case 8:
8674 knotbl(dst, src);
8675 break;
8676 case 16:
8677 knotwl(dst, src);
8678 break;
8679 case 32:
8680 knotdl(dst, src);
8681 break;
8682 case 64:
8683 knotql(dst, src);
8684 break;
8685 default:
8686 fatal("Unexpected vector length %d", masklen);
8687 break;
8688 }
8689 }
8690
8691 void MacroAssembler::kand(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
8692 switch(type) {
8693 case T_BOOLEAN:
8694 case T_BYTE:
8695 kandbl(dst, src1, src2);
8696 break;
8697 case T_CHAR:
8698 case T_SHORT:
8699 kandwl(dst, src1, src2);
8700 break;
8701 case T_INT:
8702 case T_FLOAT:
8703 kanddl(dst, src1, src2);
8704 break;
8705 case T_LONG:
8706 case T_DOUBLE:
8707 kandql(dst, src1, src2);
8708 break;
8709 default:
8710 fatal("Unexpected type argument %s", type2name(type));
8711 break;
8712 }
8713 }
8714
8715 void MacroAssembler::kor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
8716 switch(type) {
8717 case T_BOOLEAN:
8718 case T_BYTE:
8719 korbl(dst, src1, src2);
8720 break;
8721 case T_CHAR:
8722 case T_SHORT:
8723 korwl(dst, src1, src2);
8724 break;
8725 case T_INT:
8726 case T_FLOAT:
8727 kordl(dst, src1, src2);
8728 break;
8729 case T_LONG:
8730 case T_DOUBLE:
8731 korql(dst, src1, src2);
8732 break;
8733 default:
8734 fatal("Unexpected type argument %s", type2name(type));
8735 break;
8736 }
8737 }
8738
8739 void MacroAssembler::kxor(BasicType type, KRegister dst, KRegister src1, KRegister src2) {
8740 switch(type) {
8741 case T_BOOLEAN:
8742 case T_BYTE:
8743 kxorbl(dst, src1, src2);
8744 break;
8745 case T_CHAR:
8746 case T_SHORT:
8747 kxorwl(dst, src1, src2);
8748 break;
8749 case T_INT:
8750 case T_FLOAT:
8751 kxordl(dst, src1, src2);
8752 break;
8753 case T_LONG:
8754 case T_DOUBLE:
8755 kxorql(dst, src1, src2);
8756 break;
8757 default:
8758 fatal("Unexpected type argument %s", type2name(type));
8759 break;
8760 }
8761 }
8762
8763 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
8764 switch(type) {
8765 case T_BOOLEAN:
8766 case T_BYTE:
8767 evpermb(dst, mask, nds, src, merge, vector_len); break;
8768 case T_CHAR:
8769 case T_SHORT:
8770 evpermw(dst, mask, nds, src, merge, vector_len); break;
8771 case T_INT:
8772 case T_FLOAT:
8773 evpermd(dst, mask, nds, src, merge, vector_len); break;
8774 case T_LONG:
8775 case T_DOUBLE:
8776 evpermq(dst, mask, nds, src, merge, vector_len); break;
8777 default:
8778 fatal("Unexpected type argument %s", type2name(type)); break;
8779 }
8780 }
8781
8782 void MacroAssembler::evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
8783 switch(type) {
8784 case T_BOOLEAN:
8785 case T_BYTE:
8786 evpermb(dst, mask, nds, src, merge, vector_len); break;
8787 case T_CHAR:
8788 case T_SHORT:
8789 evpermw(dst, mask, nds, src, merge, vector_len); break;
8790 case T_INT:
8791 case T_FLOAT:
8792 evpermd(dst, mask, nds, src, merge, vector_len); break;
8793 case T_LONG:
8794 case T_DOUBLE:
8795 evpermq(dst, mask, nds, src, merge, vector_len); break;
8796 default:
8797 fatal("Unexpected type argument %s", type2name(type)); break;
8798 }
8799 }
8800
8801 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
8802 switch(type) {
8803 case T_BYTE:
8804 evpminub(dst, mask, nds, src, merge, vector_len); break;
8805 case T_SHORT:
8806 evpminuw(dst, mask, nds, src, merge, vector_len); break;
8807 case T_INT:
8808 evpminud(dst, mask, nds, src, merge, vector_len); break;
8809 case T_LONG:
8810 evpminuq(dst, mask, nds, src, merge, vector_len); break;
8811 default:
8812 fatal("Unexpected type argument %s", type2name(type)); break;
8813 }
8814 }
8815
8816 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
8817 switch(type) {
8818 case T_BYTE:
8819 evpmaxub(dst, mask, nds, src, merge, vector_len); break;
8820 case T_SHORT:
8821 evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
8822 case T_INT:
8823 evpmaxud(dst, mask, nds, src, merge, vector_len); break;
8824 case T_LONG:
8825 evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
8826 default:
8827 fatal("Unexpected type argument %s", type2name(type)); break;
8828 }
8829 }
8830
8831 void MacroAssembler::evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
8832 switch(type) {
8833 case T_BYTE:
8834 evpminub(dst, mask, nds, src, merge, vector_len); break;
8835 case T_SHORT:
8836 evpminuw(dst, mask, nds, src, merge, vector_len); break;
8837 case T_INT:
8838 evpminud(dst, mask, nds, src, merge, vector_len); break;
8839 case T_LONG:
8840 evpminuq(dst, mask, nds, src, merge, vector_len); break;
8841 default:
8842 fatal("Unexpected type argument %s", type2name(type)); break;
8843 }
8844 }
8845
8846 void MacroAssembler::evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
8847 switch(type) {
8848 case T_BYTE:
8849 evpmaxub(dst, mask, nds, src, merge, vector_len); break;
8850 case T_SHORT:
8851 evpmaxuw(dst, mask, nds, src, merge, vector_len); break;
8852 case T_INT:
8853 evpmaxud(dst, mask, nds, src, merge, vector_len); break;
8854 case T_LONG:
8855 evpmaxuq(dst, mask, nds, src, merge, vector_len); break;
8856 default:
8857 fatal("Unexpected type argument %s", type2name(type)); break;
8858 }
8859 }
8860
8861 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
8862 switch(type) {
8863 case T_BYTE:
8864 evpminsb(dst, mask, nds, src, merge, vector_len); break;
8865 case T_SHORT:
8866 evpminsw(dst, mask, nds, src, merge, vector_len); break;
8867 case T_INT:
8868 evpminsd(dst, mask, nds, src, merge, vector_len); break;
8869 case T_LONG:
8870 evpminsq(dst, mask, nds, src, merge, vector_len); break;
8871 default:
8872 fatal("Unexpected type argument %s", type2name(type)); break;
8873 }
8874 }
8875
8876 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
8877 switch(type) {
8878 case T_BYTE:
8879 evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
8880 case T_SHORT:
8881 evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
8882 case T_INT:
8883 evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
8884 case T_LONG:
8885 evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
8886 default:
8887 fatal("Unexpected type argument %s", type2name(type)); break;
8888 }
8889 }
8890
8891 void MacroAssembler::evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
8892 switch(type) {
8893 case T_BYTE:
8894 evpminsb(dst, mask, nds, src, merge, vector_len); break;
8895 case T_SHORT:
8896 evpminsw(dst, mask, nds, src, merge, vector_len); break;
8897 case T_INT:
8898 evpminsd(dst, mask, nds, src, merge, vector_len); break;
8899 case T_LONG:
8900 evpminsq(dst, mask, nds, src, merge, vector_len); break;
8901 default:
8902 fatal("Unexpected type argument %s", type2name(type)); break;
8903 }
8904 }
8905
8906 void MacroAssembler::evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
8907 switch(type) {
8908 case T_BYTE:
8909 evpmaxsb(dst, mask, nds, src, merge, vector_len); break;
8910 case T_SHORT:
8911 evpmaxsw(dst, mask, nds, src, merge, vector_len); break;
8912 case T_INT:
8913 evpmaxsd(dst, mask, nds, src, merge, vector_len); break;
8914 case T_LONG:
8915 evpmaxsq(dst, mask, nds, src, merge, vector_len); break;
8916 default:
8917 fatal("Unexpected type argument %s", type2name(type)); break;
8918 }
8919 }
8920
8921 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
8922 switch(type) {
8923 case T_INT:
8924 evpxord(dst, mask, nds, src, merge, vector_len); break;
8925 case T_LONG:
8926 evpxorq(dst, mask, nds, src, merge, vector_len); break;
8927 default:
8928 fatal("Unexpected type argument %s", type2name(type)); break;
8929 }
8930 }
8931
8932 void MacroAssembler::evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
8933 switch(type) {
8934 case T_INT:
8935 evpxord(dst, mask, nds, src, merge, vector_len); break;
8936 case T_LONG:
8937 evpxorq(dst, mask, nds, src, merge, vector_len); break;
8938 default:
8939 fatal("Unexpected type argument %s", type2name(type)); break;
8940 }
8941 }
8942
8943 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
8944 switch(type) {
8945 case T_INT:
8946 Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
8947 case T_LONG:
8948 evporq(dst, mask, nds, src, merge, vector_len); break;
8949 default:
8950 fatal("Unexpected type argument %s", type2name(type)); break;
8951 }
8952 }
8953
8954 void MacroAssembler::evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
8955 switch(type) {
8956 case T_INT:
8957 Assembler::evpord(dst, mask, nds, src, merge, vector_len); break;
8958 case T_LONG:
8959 evporq(dst, mask, nds, src, merge, vector_len); break;
8960 default:
8961 fatal("Unexpected type argument %s", type2name(type)); break;
8962 }
8963 }
8964
8965 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len) {
8966 switch(type) {
8967 case T_INT:
8968 evpandd(dst, mask, nds, src, merge, vector_len); break;
8969 case T_LONG:
8970 evpandq(dst, mask, nds, src, merge, vector_len); break;
8971 default:
8972 fatal("Unexpected type argument %s", type2name(type)); break;
8973 }
8974 }
8975
8976 void MacroAssembler::evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len) {
8977 switch(type) {
8978 case T_INT:
8979 evpandd(dst, mask, nds, src, merge, vector_len); break;
8980 case T_LONG:
8981 evpandq(dst, mask, nds, src, merge, vector_len); break;
8982 default:
8983 fatal("Unexpected type argument %s", type2name(type)); break;
8984 }
8985 }
8986
8987 void MacroAssembler::kortest(uint masklen, KRegister src1, KRegister src2) {
8988 switch(masklen) {
8989 case 8:
8990 kortestbl(src1, src2);
8991 break;
8992 case 16:
8993 kortestwl(src1, src2);
8994 break;
8995 case 32:
8996 kortestdl(src1, src2);
8997 break;
8998 case 64:
8999 kortestql(src1, src2);
9000 break;
9001 default:
9002 fatal("Unexpected mask length %d", masklen);
9003 break;
9004 }
9005 }
9006
9007
9008 void MacroAssembler::ktest(uint masklen, KRegister src1, KRegister src2) {
9009 switch(masklen) {
9010 case 8:
9011 ktestbl(src1, src2);
9012 break;
9013 case 16:
9014 ktestwl(src1, src2);
9015 break;
9016 case 32:
9017 ktestdl(src1, src2);
9018 break;
9019 case 64:
9020 ktestql(src1, src2);
9021 break;
9022 default:
9023 fatal("Unexpected mask length %d", masklen);
9024 break;
9025 }
9026 }
9027
9028 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
9029 switch(type) {
9030 case T_INT:
9031 evprold(dst, mask, src, shift, merge, vlen_enc); break;
9032 case T_LONG:
9033 evprolq(dst, mask, src, shift, merge, vlen_enc); break;
9034 default:
9035 fatal("Unexpected type argument %s", type2name(type)); break;
9036 break;
9037 }
9038 }
9039
9040 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc) {
9041 switch(type) {
9042 case T_INT:
9043 evprord(dst, mask, src, shift, merge, vlen_enc); break;
9044 case T_LONG:
9045 evprorq(dst, mask, src, shift, merge, vlen_enc); break;
9046 default:
9047 fatal("Unexpected type argument %s", type2name(type)); break;
9048 }
9049 }
9050
9051 void MacroAssembler::evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
9052 switch(type) {
9053 case T_INT:
9054 evprolvd(dst, mask, src1, src2, merge, vlen_enc); break;
9055 case T_LONG:
9056 evprolvq(dst, mask, src1, src2, merge, vlen_enc); break;
9057 default:
9058 fatal("Unexpected type argument %s", type2name(type)); break;
9059 }
9060 }
9061
9062 void MacroAssembler::evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc) {
9063 switch(type) {
9064 case T_INT:
9065 evprorvd(dst, mask, src1, src2, merge, vlen_enc); break;
9066 case T_LONG:
9067 evprorvq(dst, mask, src1, src2, merge, vlen_enc); break;
9068 default:
9069 fatal("Unexpected type argument %s", type2name(type)); break;
9070 }
9071 }
9072
9073 void MacroAssembler::evpandq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
9074 assert(rscratch != noreg || always_reachable(src), "missing");
9075
9076 if (reachable(src)) {
9077 evpandq(dst, nds, as_Address(src), vector_len);
9078 } else {
9079 lea(rscratch, src);
9080 evpandq(dst, nds, Address(rscratch, 0), vector_len);
9081 }
9082 }
9083
9084 void MacroAssembler::evpaddq(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src, bool merge, int vector_len, Register rscratch) {
9085 assert(rscratch != noreg || always_reachable(src), "missing");
9086
9087 if (reachable(src)) {
9088 Assembler::evpaddq(dst, mask, nds, as_Address(src), merge, vector_len);
9089 } else {
9090 lea(rscratch, src);
9091 Assembler::evpaddq(dst, mask, nds, Address(rscratch, 0), merge, vector_len);
9092 }
9093 }
9094
9095 void MacroAssembler::evporq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
9096 assert(rscratch != noreg || always_reachable(src), "missing");
9097
9098 if (reachable(src)) {
9099 evporq(dst, nds, as_Address(src), vector_len);
9100 } else {
9101 lea(rscratch, src);
9102 evporq(dst, nds, Address(rscratch, 0), vector_len);
9103 }
9104 }
9105
9106 void MacroAssembler::vpshufb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
9107 assert(rscratch != noreg || always_reachable(src), "missing");
9108
9109 if (reachable(src)) {
9110 vpshufb(dst, nds, as_Address(src), vector_len);
9111 } else {
9112 lea(rscratch, src);
9113 vpshufb(dst, nds, Address(rscratch, 0), vector_len);
9114 }
9115 }
9116
9117 void MacroAssembler::vpor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch) {
9118 assert(rscratch != noreg || always_reachable(src), "missing");
9119
9120 if (reachable(src)) {
9121 Assembler::vpor(dst, nds, as_Address(src), vector_len);
9122 } else {
9123 lea(rscratch, src);
9124 Assembler::vpor(dst, nds, Address(rscratch, 0), vector_len);
9125 }
9126 }
9127
9128 void MacroAssembler::vpternlogq(XMMRegister dst, int imm8, XMMRegister src2, AddressLiteral src3, int vector_len, Register rscratch) {
9129 assert(rscratch != noreg || always_reachable(src3), "missing");
9130
9131 if (reachable(src3)) {
9132 vpternlogq(dst, imm8, src2, as_Address(src3), vector_len);
9133 } else {
9134 lea(rscratch, src3);
9135 vpternlogq(dst, imm8, src2, Address(rscratch, 0), vector_len);
9136 }
9137 }
9138
9139 #if COMPILER2_OR_JVMCI
9140
9141 void MacroAssembler::fill_masked(BasicType bt, Address dst, XMMRegister xmm, KRegister mask,
9142 Register length, Register temp, int vec_enc) {
9143 // Computing mask for predicated vector store.
9144 movptr(temp, -1);
9145 bzhiq(temp, temp, length);
9146 kmov(mask, temp);
9147 evmovdqu(bt, mask, dst, xmm, true, vec_enc);
9148 }
9149
9150 // Set memory operation for length "less than" 64 bytes.
9151 void MacroAssembler::fill64_masked(uint shift, Register dst, int disp,
9152 XMMRegister xmm, KRegister mask, Register length,
9153 Register temp, bool use64byteVector) {
9154 assert(MaxVectorSize >= 32, "vector length should be >= 32");
9155 const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
9156 if (!use64byteVector) {
9157 fill32(dst, disp, xmm);
9158 subptr(length, 32 >> shift);
9159 fill32_masked(shift, dst, disp + 32, xmm, mask, length, temp);
9160 } else {
9161 assert(MaxVectorSize == 64, "vector length != 64");
9162 fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_512bit);
9163 }
9164 }
9165
9166
9167 void MacroAssembler::fill32_masked(uint shift, Register dst, int disp,
9168 XMMRegister xmm, KRegister mask, Register length,
9169 Register temp) {
9170 assert(MaxVectorSize >= 32, "vector length should be >= 32");
9171 const BasicType type[] = { T_BYTE, T_SHORT, T_INT, T_LONG};
9172 fill_masked(type[shift], Address(dst, disp), xmm, mask, length, temp, Assembler::AVX_256bit);
9173 }
9174
9175
9176 void MacroAssembler::fill32(Address dst, XMMRegister xmm) {
9177 assert(MaxVectorSize >= 32, "vector length should be >= 32");
9178 vmovdqu(dst, xmm);
9179 }
9180
9181 void MacroAssembler::fill32(Register dst, int disp, XMMRegister xmm) {
9182 fill32(Address(dst, disp), xmm);
9183 }
9184
9185 void MacroAssembler::fill64(Address dst, XMMRegister xmm, bool use64byteVector) {
9186 assert(MaxVectorSize >= 32, "vector length should be >= 32");
9187 if (!use64byteVector) {
9188 fill32(dst, xmm);
9189 fill32(dst.plus_disp(32), xmm);
9190 } else {
9191 evmovdquq(dst, xmm, Assembler::AVX_512bit);
9192 }
9193 }
9194
9195 void MacroAssembler::fill64(Register dst, int disp, XMMRegister xmm, bool use64byteVector) {
9196 fill64(Address(dst, disp), xmm, use64byteVector);
9197 }
9198
9199 void MacroAssembler::generate_fill_avx3(BasicType type, Register to, Register value,
9200 Register count, Register rtmp, XMMRegister xtmp) {
9201 Label L_exit;
9202 Label L_fill_start;
9203 Label L_fill_64_bytes;
9204 Label L_fill_96_bytes;
9205 Label L_fill_128_bytes;
9206 Label L_fill_128_bytes_loop;
9207 Label L_fill_128_loop_header;
9208 Label L_fill_128_bytes_loop_header;
9209 Label L_fill_128_bytes_loop_pre_header;
9210 Label L_fill_zmm_sequence;
9211
9212 int shift = -1;
9213 int avx3threshold = VM_Version::avx3_threshold();
9214 switch(type) {
9215 case T_BYTE: shift = 0;
9216 break;
9217 case T_SHORT: shift = 1;
9218 break;
9219 case T_INT: shift = 2;
9220 break;
9221 /* Uncomment when LONG fill stubs are supported.
9222 case T_LONG: shift = 3;
9223 break;
9224 */
9225 default:
9226 fatal("Unhandled type: %s\n", type2name(type));
9227 }
9228
9229 if ((avx3threshold != 0) || (MaxVectorSize == 32)) {
9230
9231 if (MaxVectorSize == 64) {
9232 cmpq(count, avx3threshold >> shift);
9233 jcc(Assembler::greater, L_fill_zmm_sequence);
9234 }
9235
9236 evpbroadcast(type, xtmp, value, Assembler::AVX_256bit);
9237
9238 bind(L_fill_start);
9239
9240 cmpq(count, 32 >> shift);
9241 jccb(Assembler::greater, L_fill_64_bytes);
9242 fill32_masked(shift, to, 0, xtmp, k2, count, rtmp);
9243 jmp(L_exit);
9244
9245 bind(L_fill_64_bytes);
9246 cmpq(count, 64 >> shift);
9247 jccb(Assembler::greater, L_fill_96_bytes);
9248 fill64_masked(shift, to, 0, xtmp, k2, count, rtmp);
9249 jmp(L_exit);
9250
9251 bind(L_fill_96_bytes);
9252 cmpq(count, 96 >> shift);
9253 jccb(Assembler::greater, L_fill_128_bytes);
9254 fill64(to, 0, xtmp);
9255 subq(count, 64 >> shift);
9256 fill32_masked(shift, to, 64, xtmp, k2, count, rtmp);
9257 jmp(L_exit);
9258
9259 bind(L_fill_128_bytes);
9260 cmpq(count, 128 >> shift);
9261 jccb(Assembler::greater, L_fill_128_bytes_loop_pre_header);
9262 fill64(to, 0, xtmp);
9263 fill32(to, 64, xtmp);
9264 subq(count, 96 >> shift);
9265 fill32_masked(shift, to, 96, xtmp, k2, count, rtmp);
9266 jmp(L_exit);
9267
9268 bind(L_fill_128_bytes_loop_pre_header);
9269 {
9270 mov(rtmp, to);
9271 andq(rtmp, 31);
9272 jccb(Assembler::zero, L_fill_128_bytes_loop_header);
9273 negq(rtmp);
9274 addq(rtmp, 32);
9275 mov64(r8, -1L);
9276 bzhiq(r8, r8, rtmp);
9277 kmovql(k2, r8);
9278 evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_256bit);
9279 addq(to, rtmp);
9280 shrq(rtmp, shift);
9281 subq(count, rtmp);
9282 }
9283
9284 cmpq(count, 128 >> shift);
9285 jcc(Assembler::less, L_fill_start);
9286
9287 bind(L_fill_128_bytes_loop_header);
9288 subq(count, 128 >> shift);
9289
9290 align32();
9291 bind(L_fill_128_bytes_loop);
9292 fill64(to, 0, xtmp);
9293 fill64(to, 64, xtmp);
9294 addq(to, 128);
9295 subq(count, 128 >> shift);
9296 jccb(Assembler::greaterEqual, L_fill_128_bytes_loop);
9297
9298 addq(count, 128 >> shift);
9299 jcc(Assembler::zero, L_exit);
9300 jmp(L_fill_start);
9301 }
9302
9303 if (MaxVectorSize == 64) {
9304 // Sequence using 64 byte ZMM register.
9305 Label L_fill_128_bytes_zmm;
9306 Label L_fill_192_bytes_zmm;
9307 Label L_fill_192_bytes_loop_zmm;
9308 Label L_fill_192_bytes_loop_header_zmm;
9309 Label L_fill_192_bytes_loop_pre_header_zmm;
9310 Label L_fill_start_zmm_sequence;
9311
9312 bind(L_fill_zmm_sequence);
9313 evpbroadcast(type, xtmp, value, Assembler::AVX_512bit);
9314
9315 bind(L_fill_start_zmm_sequence);
9316 cmpq(count, 64 >> shift);
9317 jccb(Assembler::greater, L_fill_128_bytes_zmm);
9318 fill64_masked(shift, to, 0, xtmp, k2, count, rtmp, true);
9319 jmp(L_exit);
9320
9321 bind(L_fill_128_bytes_zmm);
9322 cmpq(count, 128 >> shift);
9323 jccb(Assembler::greater, L_fill_192_bytes_zmm);
9324 fill64(to, 0, xtmp, true);
9325 subq(count, 64 >> shift);
9326 fill64_masked(shift, to, 64, xtmp, k2, count, rtmp, true);
9327 jmp(L_exit);
9328
9329 bind(L_fill_192_bytes_zmm);
9330 cmpq(count, 192 >> shift);
9331 jccb(Assembler::greater, L_fill_192_bytes_loop_pre_header_zmm);
9332 fill64(to, 0, xtmp, true);
9333 fill64(to, 64, xtmp, true);
9334 subq(count, 128 >> shift);
9335 fill64_masked(shift, to, 128, xtmp, k2, count, rtmp, true);
9336 jmp(L_exit);
9337
9338 bind(L_fill_192_bytes_loop_pre_header_zmm);
9339 {
9340 movq(rtmp, to);
9341 andq(rtmp, 63);
9342 jccb(Assembler::zero, L_fill_192_bytes_loop_header_zmm);
9343 negq(rtmp);
9344 addq(rtmp, 64);
9345 mov64(r8, -1L);
9346 bzhiq(r8, r8, rtmp);
9347 kmovql(k2, r8);
9348 evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_512bit);
9349 addq(to, rtmp);
9350 shrq(rtmp, shift);
9351 subq(count, rtmp);
9352 }
9353
9354 cmpq(count, 192 >> shift);
9355 jcc(Assembler::less, L_fill_start_zmm_sequence);
9356
9357 bind(L_fill_192_bytes_loop_header_zmm);
9358 subq(count, 192 >> shift);
9359
9360 align32();
9361 bind(L_fill_192_bytes_loop_zmm);
9362 fill64(to, 0, xtmp, true);
9363 fill64(to, 64, xtmp, true);
9364 fill64(to, 128, xtmp, true);
9365 addq(to, 192);
9366 subq(count, 192 >> shift);
9367 jccb(Assembler::greaterEqual, L_fill_192_bytes_loop_zmm);
9368
9369 addq(count, 192 >> shift);
9370 jcc(Assembler::zero, L_exit);
9371 jmp(L_fill_start_zmm_sequence);
9372 }
9373 bind(L_exit);
9374 }
9375 #endif //COMPILER2_OR_JVMCI
9376
9377
9378 void MacroAssembler::convert_f2i(Register dst, XMMRegister src) {
9379 Label done;
9380 cvttss2sil(dst, src);
9381 // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
9382 cmpl(dst, 0x80000000); // float_sign_flip
9383 jccb(Assembler::notEqual, done);
9384 subptr(rsp, 8);
9385 movflt(Address(rsp, 0), src);
9386 call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2i_fixup())));
9387 pop(dst);
9388 bind(done);
9389 }
9390
9391 void MacroAssembler::convert_d2i(Register dst, XMMRegister src) {
9392 Label done;
9393 cvttsd2sil(dst, src);
9394 // Conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
9395 cmpl(dst, 0x80000000); // float_sign_flip
9396 jccb(Assembler::notEqual, done);
9397 subptr(rsp, 8);
9398 movdbl(Address(rsp, 0), src);
9399 call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2i_fixup())));
9400 pop(dst);
9401 bind(done);
9402 }
9403
9404 void MacroAssembler::convert_f2l(Register dst, XMMRegister src) {
9405 Label done;
9406 cvttss2siq(dst, src);
9407 cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
9408 jccb(Assembler::notEqual, done);
9409 subptr(rsp, 8);
9410 movflt(Address(rsp, 0), src);
9411 call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::f2l_fixup())));
9412 pop(dst);
9413 bind(done);
9414 }
9415
9416 void MacroAssembler::round_float(Register dst, XMMRegister src, Register rtmp, Register rcx) {
9417 // Following code is line by line assembly translation rounding algorithm.
9418 // Please refer to java.lang.Math.round(float) algorithm for details.
9419 const int32_t FloatConsts_EXP_BIT_MASK = 0x7F800000;
9420 const int32_t FloatConsts_SIGNIFICAND_WIDTH = 24;
9421 const int32_t FloatConsts_EXP_BIAS = 127;
9422 const int32_t FloatConsts_SIGNIF_BIT_MASK = 0x007FFFFF;
9423 const int32_t MINUS_32 = 0xFFFFFFE0;
9424 Label L_special_case, L_block1, L_exit;
9425 movl(rtmp, FloatConsts_EXP_BIT_MASK);
9426 movdl(dst, src);
9427 andl(dst, rtmp);
9428 sarl(dst, FloatConsts_SIGNIFICAND_WIDTH - 1);
9429 movl(rtmp, FloatConsts_SIGNIFICAND_WIDTH - 2 + FloatConsts_EXP_BIAS);
9430 subl(rtmp, dst);
9431 movl(rcx, rtmp);
9432 movl(dst, MINUS_32);
9433 testl(rtmp, dst);
9434 jccb(Assembler::notEqual, L_special_case);
9435 movdl(dst, src);
9436 andl(dst, FloatConsts_SIGNIF_BIT_MASK);
9437 orl(dst, FloatConsts_SIGNIF_BIT_MASK + 1);
9438 movdl(rtmp, src);
9439 testl(rtmp, rtmp);
9440 jccb(Assembler::greaterEqual, L_block1);
9441 negl(dst);
9442 bind(L_block1);
9443 sarl(dst);
9444 addl(dst, 0x1);
9445 sarl(dst, 0x1);
9446 jmp(L_exit);
9447 bind(L_special_case);
9448 convert_f2i(dst, src);
9449 bind(L_exit);
9450 }
9451
9452 void MacroAssembler::round_double(Register dst, XMMRegister src, Register rtmp, Register rcx) {
9453 // Following code is line by line assembly translation rounding algorithm.
9454 // Please refer to java.lang.Math.round(double) algorithm for details.
9455 const int64_t DoubleConsts_EXP_BIT_MASK = 0x7FF0000000000000L;
9456 const int64_t DoubleConsts_SIGNIFICAND_WIDTH = 53;
9457 const int64_t DoubleConsts_EXP_BIAS = 1023;
9458 const int64_t DoubleConsts_SIGNIF_BIT_MASK = 0x000FFFFFFFFFFFFFL;
9459 const int64_t MINUS_64 = 0xFFFFFFFFFFFFFFC0L;
9460 Label L_special_case, L_block1, L_exit;
9461 mov64(rtmp, DoubleConsts_EXP_BIT_MASK);
9462 movq(dst, src);
9463 andq(dst, rtmp);
9464 sarq(dst, DoubleConsts_SIGNIFICAND_WIDTH - 1);
9465 mov64(rtmp, DoubleConsts_SIGNIFICAND_WIDTH - 2 + DoubleConsts_EXP_BIAS);
9466 subq(rtmp, dst);
9467 movq(rcx, rtmp);
9468 mov64(dst, MINUS_64);
9469 testq(rtmp, dst);
9470 jccb(Assembler::notEqual, L_special_case);
9471 movq(dst, src);
9472 mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK);
9473 andq(dst, rtmp);
9474 mov64(rtmp, DoubleConsts_SIGNIF_BIT_MASK + 1);
9475 orq(dst, rtmp);
9476 movq(rtmp, src);
9477 testq(rtmp, rtmp);
9478 jccb(Assembler::greaterEqual, L_block1);
9479 negq(dst);
9480 bind(L_block1);
9481 sarq(dst);
9482 addq(dst, 0x1);
9483 sarq(dst, 0x1);
9484 jmp(L_exit);
9485 bind(L_special_case);
9486 convert_d2l(dst, src);
9487 bind(L_exit);
9488 }
9489
9490 void MacroAssembler::convert_d2l(Register dst, XMMRegister src) {
9491 Label done;
9492 cvttsd2siq(dst, src);
9493 cmp64(dst, ExternalAddress((address) StubRoutines::x86::double_sign_flip()));
9494 jccb(Assembler::notEqual, done);
9495 subptr(rsp, 8);
9496 movdbl(Address(rsp, 0), src);
9497 call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::d2l_fixup())));
9498 pop(dst);
9499 bind(done);
9500 }
9501
9502 void MacroAssembler::cache_wb(Address line)
9503 {
9504 // 64 bit cpus always support clflush
9505 assert(VM_Version::supports_clflush(), "clflush should be available");
9506 bool optimized = VM_Version::supports_clflushopt();
9507 bool no_evict = VM_Version::supports_clwb();
9508
9509 // prefer clwb (writeback without evict) otherwise
9510 // prefer clflushopt (potentially parallel writeback with evict)
9511 // otherwise fallback on clflush (serial writeback with evict)
9512
9513 if (optimized) {
9514 if (no_evict) {
9515 clwb(line);
9516 } else {
9517 clflushopt(line);
9518 }
9519 } else {
9520 // no need for fence when using CLFLUSH
9521 clflush(line);
9522 }
9523 }
9524
9525 void MacroAssembler::cache_wbsync(bool is_pre)
9526 {
9527 assert(VM_Version::supports_clflush(), "clflush should be available");
9528 bool optimized = VM_Version::supports_clflushopt();
9529 bool no_evict = VM_Version::supports_clwb();
9530
9531 // pick the correct implementation
9532
9533 if (!is_pre && (optimized || no_evict)) {
9534 // need an sfence for post flush when using clflushopt or clwb
9535 // otherwise no no need for any synchroniaztion
9536
9537 sfence();
9538 }
9539 }
9540
9541 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) {
9542 switch (cond) {
9543 // Note some conditions are synonyms for others
9544 case Assembler::zero: return Assembler::notZero;
9545 case Assembler::notZero: return Assembler::zero;
9546 case Assembler::less: return Assembler::greaterEqual;
9547 case Assembler::lessEqual: return Assembler::greater;
9548 case Assembler::greater: return Assembler::lessEqual;
9549 case Assembler::greaterEqual: return Assembler::less;
9550 case Assembler::below: return Assembler::aboveEqual;
9551 case Assembler::belowEqual: return Assembler::above;
9552 case Assembler::above: return Assembler::belowEqual;
9553 case Assembler::aboveEqual: return Assembler::below;
9554 case Assembler::overflow: return Assembler::noOverflow;
9555 case Assembler::noOverflow: return Assembler::overflow;
9556 case Assembler::negative: return Assembler::positive;
9557 case Assembler::positive: return Assembler::negative;
9558 case Assembler::parity: return Assembler::noParity;
9559 case Assembler::noParity: return Assembler::parity;
9560 }
9561 ShouldNotReachHere(); return Assembler::overflow;
9562 }
9563
9564 // This is simply a call to Thread::current()
9565 void MacroAssembler::get_thread_slow(Register thread) {
9566 if (thread != rax) {
9567 push(rax);
9568 }
9569 push(rdi);
9570 push(rsi);
9571 push(rdx);
9572 push(rcx);
9573 push(r8);
9574 push(r9);
9575 push(r10);
9576 push(r11);
9577
9578 MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, Thread::current), 0);
9579
9580 pop(r11);
9581 pop(r10);
9582 pop(r9);
9583 pop(r8);
9584 pop(rcx);
9585 pop(rdx);
9586 pop(rsi);
9587 pop(rdi);
9588 if (thread != rax) {
9589 mov(thread, rax);
9590 pop(rax);
9591 }
9592 }
9593
9594 void MacroAssembler::check_stack_alignment(Register sp, const char* msg, unsigned bias, Register tmp) {
9595 Label L_stack_ok;
9596 if (bias == 0) {
9597 testptr(sp, 2 * wordSize - 1);
9598 } else {
9599 // lea(tmp, Address(rsp, bias);
9600 mov(tmp, sp);
9601 addptr(tmp, bias);
9602 testptr(tmp, 2 * wordSize - 1);
9603 }
9604 jcc(Assembler::equal, L_stack_ok);
9605 block_comment(msg);
9606 stop(msg);
9607 bind(L_stack_ok);
9608 }
9609
9610 // Implements lightweight-locking.
9611 //
9612 // obj: the object to be locked
9613 // reg_rax: rax
9614 // thread: the thread which attempts to lock obj
9615 // tmp: a temporary register
9616 void MacroAssembler::lightweight_lock(Register basic_lock, Register obj, Register reg_rax, Register tmp, Label& slow) {
9617 Register thread = r15_thread;
9618
9619 assert(reg_rax == rax, "");
9620 assert_different_registers(basic_lock, obj, reg_rax, thread, tmp);
9621
9622 Label push;
9623 const Register top = tmp;
9624
9625 // Preload the markWord. It is important that this is the first
9626 // instruction emitted as it is part of C1's null check semantics.
9627 movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
9628
9629 if (UseObjectMonitorTable) {
9630 // Clear cache in case fast locking succeeds or we need to take the slow-path.
9631 movptr(Address(basic_lock, BasicObjectLock::lock_offset() + in_ByteSize((BasicLock::object_monitor_cache_offset_in_bytes()))), 0);
9632 }
9633
9634 if (DiagnoseSyncOnValueBasedClasses != 0) {
9635 load_klass(tmp, obj, rscratch1);
9636 testb(Address(tmp, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
9637 jcc(Assembler::notZero, slow);
9638 }
9639
9640 // Load top.
9641 movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
9642
9643 // Check if the lock-stack is full.
9644 cmpl(top, LockStack::end_offset());
9645 jcc(Assembler::greaterEqual, slow);
9646
9647 // Check for recursion.
9648 cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
9649 jcc(Assembler::equal, push);
9650
9651 // Check header for monitor (0b10).
9652 testptr(reg_rax, markWord::monitor_value);
9653 jcc(Assembler::notZero, slow);
9654
9655 // Try to lock. Transition lock bits 0b01 => 0b00
9656 movptr(tmp, reg_rax);
9657 andptr(tmp, ~(int32_t)markWord::unlocked_value);
9658 orptr(reg_rax, markWord::unlocked_value);
9659 lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
9660 jcc(Assembler::notEqual, slow);
9661
9662 // Restore top, CAS clobbers register.
9663 movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
9664
9665 bind(push);
9666 // After successful lock, push object on lock-stack.
9667 movptr(Address(thread, top), obj);
9668 incrementl(top, oopSize);
9669 movl(Address(thread, JavaThread::lock_stack_top_offset()), top);
9670 }
9671
9672 // Implements lightweight-unlocking.
9673 //
9674 // obj: the object to be unlocked
9675 // reg_rax: rax
9676 // thread: the thread
9677 // tmp: a temporary register
9678 void MacroAssembler::lightweight_unlock(Register obj, Register reg_rax, Register tmp, Label& slow) {
9679 Register thread = r15_thread;
9680
9681 assert(reg_rax == rax, "");
9682 assert_different_registers(obj, reg_rax, thread, tmp);
9683
9684 Label unlocked, push_and_slow;
9685 const Register top = tmp;
9686
9687 // Check if obj is top of lock-stack.
9688 movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
9689 cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
9690 jcc(Assembler::notEqual, slow);
9691
9692 // Pop lock-stack.
9693 DEBUG_ONLY(movptr(Address(thread, top, Address::times_1, -oopSize), 0);)
9694 subl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
9695
9696 // Check if recursive.
9697 cmpptr(obj, Address(thread, top, Address::times_1, -2 * oopSize));
9698 jcc(Assembler::equal, unlocked);
9699
9700 // Not recursive. Check header for monitor (0b10).
9701 movptr(reg_rax, Address(obj, oopDesc::mark_offset_in_bytes()));
9702 testptr(reg_rax, markWord::monitor_value);
9703 jcc(Assembler::notZero, push_and_slow);
9704
9705 #ifdef ASSERT
9706 // Check header not unlocked (0b01).
9707 Label not_unlocked;
9708 testptr(reg_rax, markWord::unlocked_value);
9709 jcc(Assembler::zero, not_unlocked);
9710 stop("lightweight_unlock already unlocked");
9711 bind(not_unlocked);
9712 #endif
9713
9714 // Try to unlock. Transition lock bits 0b00 => 0b01
9715 movptr(tmp, reg_rax);
9716 orptr(tmp, markWord::unlocked_value);
9717 lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
9718 jcc(Assembler::equal, unlocked);
9719
9720 bind(push_and_slow);
9721 // Restore lock-stack and handle the unlock in runtime.
9722 #ifdef ASSERT
9723 movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
9724 movptr(Address(thread, top), obj);
9725 #endif
9726 addl(Address(thread, JavaThread::lock_stack_top_offset()), oopSize);
9727 jmp(slow);
9728
9729 bind(unlocked);
9730 }
9731
9732 // Saves legacy GPRs state on stack.
9733 void MacroAssembler::save_legacy_gprs() {
9734 subq(rsp, 16 * wordSize);
9735 movq(Address(rsp, 15 * wordSize), rax);
9736 movq(Address(rsp, 14 * wordSize), rcx);
9737 movq(Address(rsp, 13 * wordSize), rdx);
9738 movq(Address(rsp, 12 * wordSize), rbx);
9739 movq(Address(rsp, 10 * wordSize), rbp);
9740 movq(Address(rsp, 9 * wordSize), rsi);
9741 movq(Address(rsp, 8 * wordSize), rdi);
9742 movq(Address(rsp, 7 * wordSize), r8);
9743 movq(Address(rsp, 6 * wordSize), r9);
9744 movq(Address(rsp, 5 * wordSize), r10);
9745 movq(Address(rsp, 4 * wordSize), r11);
9746 movq(Address(rsp, 3 * wordSize), r12);
9747 movq(Address(rsp, 2 * wordSize), r13);
9748 movq(Address(rsp, wordSize), r14);
9749 movq(Address(rsp, 0), r15);
9750 }
9751
9752 // Resotres back legacy GPRs state from stack.
9753 void MacroAssembler::restore_legacy_gprs() {
9754 movq(r15, Address(rsp, 0));
9755 movq(r14, Address(rsp, wordSize));
9756 movq(r13, Address(rsp, 2 * wordSize));
9757 movq(r12, Address(rsp, 3 * wordSize));
9758 movq(r11, Address(rsp, 4 * wordSize));
9759 movq(r10, Address(rsp, 5 * wordSize));
9760 movq(r9, Address(rsp, 6 * wordSize));
9761 movq(r8, Address(rsp, 7 * wordSize));
9762 movq(rdi, Address(rsp, 8 * wordSize));
9763 movq(rsi, Address(rsp, 9 * wordSize));
9764 movq(rbp, Address(rsp, 10 * wordSize));
9765 movq(rbx, Address(rsp, 12 * wordSize));
9766 movq(rdx, Address(rsp, 13 * wordSize));
9767 movq(rcx, Address(rsp, 14 * wordSize));
9768 movq(rax, Address(rsp, 15 * wordSize));
9769 addq(rsp, 16 * wordSize);
9770 }
9771
9772 void MacroAssembler::load_aotrc_address(Register reg, address a) {
9773 #if INCLUDE_CDS
9774 assert(AOTRuntimeConstants::contains(a), "address out of range for data area");
9775 if (AOTCodeCache::is_on_for_dump()) {
9776 // all aotrc field addresses should be registered in the AOTCodeCache address table
9777 lea(reg, ExternalAddress(a));
9778 } else {
9779 mov64(reg, (uint64_t)a);
9780 }
9781 #else
9782 ShouldNotReachHere();
9783 #endif
9784 }
9785
9786 void MacroAssembler::setcc(Assembler::Condition comparison, Register dst) {
9787 if (VM_Version::supports_apx_f()) {
9788 esetzucc(comparison, dst);
9789 } else {
9790 setb(comparison, dst);
9791 movzbl(dst, dst);
9792 }
9793 }