1 /*
2 * Copyright (c) 1997, 2020, 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 #ifndef CPU_X86_MACROASSEMBLER_X86_HPP
26 #define CPU_X86_MACROASSEMBLER_X86_HPP
27
28 #include "asm/assembler.hpp"
29 #include "utilities/macros.hpp"
30 #include "runtime/rtmLocking.hpp"
31
32 // MacroAssembler extends Assembler by frequently used macros.
33 //
34 // Instructions for which a 'better' code sequence exists depending
35 // on arguments should also go in here.
36
37 class MacroAssembler: public Assembler {
38 friend class LIR_Assembler;
39 friend class Runtime1; // as_Address()
40
41 public:
42 // Support for VM calls
43 //
44 // This is the base routine called by the different versions of call_VM_leaf. The interpreter
45 // may customize this version by overriding it for its purposes (e.g., to save/restore
46 // additional registers when doing a VM call).
47
48 virtual void call_VM_leaf_base(
49 address entry_point, // the entry point
50 int number_of_arguments // the number of arguments to pop after the call
51 );
52
53 protected:
54 // This is the base routine called by the different versions of call_VM. The interpreter
55 // may customize this version by overriding it for its purposes (e.g., to save/restore
56 // additional registers when doing a VM call).
57 //
58 // If no java_thread register is specified (noreg) than rdi will be used instead. call_VM_base
59 // returns the register which contains the thread upon return. If a thread register has been
60 // specified, the return value will correspond to that register. If no last_java_sp is specified
61 // (noreg) than rsp will be used instead.
62 virtual void call_VM_base( // returns the register containing the thread upon return
63 Register oop_result, // where an oop-result ends up if any; use noreg otherwise
64 Register java_thread, // the thread if computed before ; use noreg otherwise
65 Register last_java_sp, // to set up last_Java_frame in stubs; use noreg otherwise
66 address entry_point, // the entry point
67 int number_of_arguments, // the number of arguments (w/o thread) to pop after the call
68 bool check_exceptions // whether to check for pending exceptions after return
69 );
70
71 void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true);
72
73 // helpers for FPU flag access
74 // tmp is a temporary register, if none is available use noreg
75 void save_rax (Register tmp);
76 void restore_rax(Register tmp);
77
78 public:
79 MacroAssembler(CodeBuffer* code) : Assembler(code) {}
80
81 // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code.
82 // The implementation is only non-empty for the InterpreterMacroAssembler,
83 // as only the interpreter handles PopFrame and ForceEarlyReturn requests.
84 virtual void check_and_handle_popframe(Register java_thread);
85 virtual void check_and_handle_earlyret(Register java_thread);
86
87 Address as_Address(AddressLiteral adr);
88 Address as_Address(ArrayAddress adr);
89
90 // Support for NULL-checks
91 //
92 // Generates code that causes a NULL OS exception if the content of reg is NULL.
93 // If the accessed location is M[reg + offset] and the offset is known, provide the
94 // offset. No explicit code generation is needed if the offset is within a certain
95 // range (0 <= offset <= page_size).
96
97 void null_check(Register reg, int offset = -1);
98 static bool needs_explicit_null_check(intptr_t offset);
99 static bool uses_implicit_null_check(void* address);
100
101 // Required platform-specific helpers for Label::patch_instructions.
102 // They _shadow_ the declarations in AbstractAssembler, which are undefined.
103 void pd_patch_instruction(address branch, address target, const char* file, int line) {
104 unsigned char op = branch[0];
105 assert(op == 0xE8 /* call */ ||
106 op == 0xE9 /* jmp */ ||
107 op == 0xEB /* short jmp */ ||
108 (op & 0xF0) == 0x70 /* short jcc */ ||
109 op == 0x0F && (branch[1] & 0xF0) == 0x80 /* jcc */ ||
110 op == 0xC7 && branch[1] == 0xF8 /* xbegin */,
111 "Invalid opcode at patch point");
112
113 if (op == 0xEB || (op & 0xF0) == 0x70) {
114 // short offset operators (jmp and jcc)
115 char* disp = (char*) &branch[1];
116 int imm8 = target - (address) &disp[1];
117 guarantee(this->is8bit(imm8), "Short forward jump exceeds 8-bit offset at %s:%d",
118 file == NULL ? "<NULL>" : file, line);
119 *disp = imm8;
120 } else {
121 int* disp = (int*) &branch[(op == 0x0F || op == 0xC7)? 2: 1];
122 int imm32 = target - (address) &disp[1];
123 *disp = imm32;
124 }
125 }
126
127 // The following 4 methods return the offset of the appropriate move instruction
128
129 // Support for fast byte/short loading with zero extension (depending on particular CPU)
130 int load_unsigned_byte(Register dst, Address src);
131 int load_unsigned_short(Register dst, Address src);
132
133 // Support for fast byte/short loading with sign extension (depending on particular CPU)
134 int load_signed_byte(Register dst, Address src);
135 int load_signed_short(Register dst, Address src);
136
137 // Support for sign-extension (hi:lo = extend_sign(lo))
138 void extend_sign(Register hi, Register lo);
139
140 // Load and store values by size and signed-ness
141 void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2 = noreg);
142 void store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2 = noreg);
143
144 // Support for inc/dec with optimal instruction selection depending on value
145
146 void increment(Register reg, int value = 1) { LP64_ONLY(incrementq(reg, value)) NOT_LP64(incrementl(reg, value)) ; }
147 void decrement(Register reg, int value = 1) { LP64_ONLY(decrementq(reg, value)) NOT_LP64(decrementl(reg, value)) ; }
148
149 void decrementl(Address dst, int value = 1);
150 void decrementl(Register reg, int value = 1);
151
152 void decrementq(Register reg, int value = 1);
153 void decrementq(Address dst, int value = 1);
154
155 void incrementl(Address dst, int value = 1);
156 void incrementl(Register reg, int value = 1);
157
158 void incrementq(Register reg, int value = 1);
159 void incrementq(Address dst, int value = 1);
160
161 // Support optimal SSE move instructions.
162 void movflt(XMMRegister dst, XMMRegister src) {
163 if (dst-> encoding() == src->encoding()) return;
164 if (UseXmmRegToRegMoveAll) { movaps(dst, src); return; }
165 else { movss (dst, src); return; }
166 }
167 void movflt(XMMRegister dst, Address src) { movss(dst, src); }
168 void movflt(XMMRegister dst, AddressLiteral src);
169 void movflt(Address dst, XMMRegister src) { movss(dst, src); }
170
171 void movdbl(XMMRegister dst, XMMRegister src) {
172 if (dst-> encoding() == src->encoding()) return;
173 if (UseXmmRegToRegMoveAll) { movapd(dst, src); return; }
174 else { movsd (dst, src); return; }
175 }
176
177 void movdbl(XMMRegister dst, AddressLiteral src);
178
179 void movdbl(XMMRegister dst, Address src) {
180 if (UseXmmLoadAndClearUpper) { movsd (dst, src); return; }
181 else { movlpd(dst, src); return; }
182 }
183 void movdbl(Address dst, XMMRegister src) { movsd(dst, src); }
184
185 void incrementl(AddressLiteral dst);
186 void incrementl(ArrayAddress dst);
187
188 void incrementq(AddressLiteral dst);
189
190 // Alignment
191 void align(int modulus);
192 void align(int modulus, int target);
193
194 // A 5 byte nop that is safe for patching (see patch_verified_entry)
195 void fat_nop();
196
197 // Stack frame creation/removal
198 void enter();
199 void leave();
200
201 // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information)
202 // The pointer will be loaded into the thread register.
203 void get_thread(Register thread);
204
205
206 // Support for VM calls
207 //
208 // It is imperative that all calls into the VM are handled via the call_VM macros.
209 // They make sure that the stack linkage is setup correctly. call_VM's correspond
210 // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points.
211
212
213 void call_VM(Register oop_result,
214 address entry_point,
215 bool check_exceptions = true);
216 void call_VM(Register oop_result,
217 address entry_point,
218 Register arg_1,
219 bool check_exceptions = true);
220 void call_VM(Register oop_result,
221 address entry_point,
222 Register arg_1, Register arg_2,
223 bool check_exceptions = true);
224 void call_VM(Register oop_result,
225 address entry_point,
226 Register arg_1, Register arg_2, Register arg_3,
227 bool check_exceptions = true);
228
229 // Overloadings with last_Java_sp
230 void call_VM(Register oop_result,
231 Register last_java_sp,
232 address entry_point,
233 int number_of_arguments = 0,
234 bool check_exceptions = true);
235 void call_VM(Register oop_result,
236 Register last_java_sp,
237 address entry_point,
238 Register arg_1, bool
239 check_exceptions = true);
240 void call_VM(Register oop_result,
241 Register last_java_sp,
242 address entry_point,
243 Register arg_1, Register arg_2,
244 bool check_exceptions = true);
245 void call_VM(Register oop_result,
246 Register last_java_sp,
247 address entry_point,
248 Register arg_1, Register arg_2, Register arg_3,
249 bool check_exceptions = true);
250
251 void get_vm_result (Register oop_result, Register thread);
252 void get_vm_result_2(Register metadata_result, Register thread);
253
254 // These always tightly bind to MacroAssembler::call_VM_base
255 // bypassing the virtual implementation
256 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
257 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true);
258 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
259 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions = true);
260 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4, bool check_exceptions = true);
261
262 void call_VM_leaf0(address entry_point);
263 void call_VM_leaf(address entry_point,
264 int number_of_arguments = 0);
265 void call_VM_leaf(address entry_point,
266 Register arg_1);
267 void call_VM_leaf(address entry_point,
268 Register arg_1, Register arg_2);
269 void call_VM_leaf(address entry_point,
270 Register arg_1, Register arg_2, Register arg_3);
271
272 // These always tightly bind to MacroAssembler::call_VM_leaf_base
273 // bypassing the virtual implementation
274 void super_call_VM_leaf(address entry_point);
275 void super_call_VM_leaf(address entry_point, Register arg_1);
276 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2);
277 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3);
278 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4);
279
280 // last Java Frame (fills frame anchor)
281 void set_last_Java_frame(Register thread,
282 Register last_java_sp,
283 Register last_java_fp,
284 address last_java_pc);
285
286 // thread in the default location (r15_thread on 64bit)
287 void set_last_Java_frame(Register last_java_sp,
288 Register last_java_fp,
289 address last_java_pc);
290
291 void reset_last_Java_frame(Register thread, bool clear_fp);
292
293 // thread in the default location (r15_thread on 64bit)
294 void reset_last_Java_frame(bool clear_fp);
295
296 // jobjects
297 void clear_jweak_tag(Register possibly_jweak);
298 void resolve_jobject(Register value, Register thread, Register tmp);
299
300 // C 'boolean' to Java boolean: x == 0 ? 0 : 1
301 void c2bool(Register x);
302
303 // C++ bool manipulation
304
305 void movbool(Register dst, Address src);
306 void movbool(Address dst, bool boolconst);
307 void movbool(Address dst, Register src);
308 void testbool(Register dst);
309
310 void resolve_oop_handle(Register result, Register tmp = rscratch2);
311 void resolve_weak_handle(Register result, Register tmp);
312 void load_mirror(Register mirror, Register method, Register tmp = rscratch2);
313 void load_method_holder_cld(Register rresult, Register rmethod);
314
315 void load_method_holder(Register holder, Register method);
316
317 // oop manipulations
318 void load_klass(Register dst, Register src);
319 void store_klass(Register dst, Register src);
320
321 void access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
322 Register tmp1, Register thread_tmp);
323 void access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register src,
324 Register tmp1, Register tmp2);
325
326 // Resolves obj access. Result is placed in the same register.
327 // All other registers are preserved.
328 void resolve(DecoratorSet decorators, Register obj);
329
330 void load_heap_oop(Register dst, Address src, Register tmp1 = noreg,
331 Register thread_tmp = noreg, DecoratorSet decorators = 0);
332 void load_heap_oop_not_null(Register dst, Address src, Register tmp1 = noreg,
333 Register thread_tmp = noreg, DecoratorSet decorators = 0);
334 void store_heap_oop(Address dst, Register src, Register tmp1 = noreg,
335 Register tmp2 = noreg, DecoratorSet decorators = 0);
336
337 // Used for storing NULL. All other oop constants should be
338 // stored using routines that take a jobject.
339 void store_heap_oop_null(Address dst);
340
341 void load_prototype_header(Register dst, Register src);
342
343 #ifdef _LP64
344 void store_klass_gap(Register dst, Register src);
345
346 // This dummy is to prevent a call to store_heap_oop from
347 // converting a zero (like NULL) into a Register by giving
348 // the compiler two choices it can't resolve
349
350 void store_heap_oop(Address dst, void* dummy);
351
352 void encode_heap_oop(Register r);
353 void decode_heap_oop(Register r);
354 void encode_heap_oop_not_null(Register r);
355 void decode_heap_oop_not_null(Register r);
356 void encode_heap_oop_not_null(Register dst, Register src);
357 void decode_heap_oop_not_null(Register dst, Register src);
358
359 void set_narrow_oop(Register dst, jobject obj);
360 void set_narrow_oop(Address dst, jobject obj);
361 void cmp_narrow_oop(Register dst, jobject obj);
362 void cmp_narrow_oop(Address dst, jobject obj);
363
364 void encode_klass_not_null(Register r);
365 void decode_klass_not_null(Register r);
366 void encode_klass_not_null(Register dst, Register src);
367 void decode_klass_not_null(Register dst, Register src);
368 void set_narrow_klass(Register dst, Klass* k);
369 void set_narrow_klass(Address dst, Klass* k);
370 void cmp_narrow_klass(Register dst, Klass* k);
371 void cmp_narrow_klass(Address dst, Klass* k);
372
373 // Returns the byte size of the instructions generated by decode_klass_not_null()
374 // when compressed klass pointers are being used.
375 static int instr_size_for_decode_klass_not_null();
376
377 // if heap base register is used - reinit it with the correct value
378 void reinit_heapbase();
379
380 DEBUG_ONLY(void verify_heapbase(const char* msg);)
381
382 #endif // _LP64
383
384 // Int division/remainder for Java
385 // (as idivl, but checks for special case as described in JVM spec.)
386 // returns idivl instruction offset for implicit exception handling
387 int corrected_idivl(Register reg);
388
389 // Long division/remainder for Java
390 // (as idivq, but checks for special case as described in JVM spec.)
391 // returns idivq instruction offset for implicit exception handling
392 int corrected_idivq(Register reg);
393
394 void int3();
395
396 // Long operation macros for a 32bit cpu
397 // Long negation for Java
398 void lneg(Register hi, Register lo);
399
400 // Long multiplication for Java
401 // (destroys contents of eax, ebx, ecx and edx)
402 void lmul(int x_rsp_offset, int y_rsp_offset); // rdx:rax = x * y
403
404 // Long shifts for Java
405 // (semantics as described in JVM spec.)
406 void lshl(Register hi, Register lo); // hi:lo << (rcx & 0x3f)
407 void lshr(Register hi, Register lo, bool sign_extension = false); // hi:lo >> (rcx & 0x3f)
408
409 // Long compare for Java
410 // (semantics as described in JVM spec.)
411 void lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo); // x_hi = lcmp(x, y)
412
413
414 // misc
415
416 // Sign extension
417 void sign_extend_short(Register reg);
418 void sign_extend_byte(Register reg);
419
420 // Division by power of 2, rounding towards 0
421 void division_with_shift(Register reg, int shift_value);
422
423 #ifndef _LP64
424 // Compares the top-most stack entries on the FPU stack and sets the eflags as follows:
425 //
426 // CF (corresponds to C0) if x < y
427 // PF (corresponds to C2) if unordered
428 // ZF (corresponds to C3) if x = y
429 //
430 // The arguments are in reversed order on the stack (i.e., top of stack is first argument).
431 // tmp is a temporary register, if none is available use noreg (only matters for non-P6 code)
432 void fcmp(Register tmp);
433 // Variant of the above which allows y to be further down the stack
434 // and which only pops x and y if specified. If pop_right is
435 // specified then pop_left must also be specified.
436 void fcmp(Register tmp, int index, bool pop_left, bool pop_right);
437
438 // Floating-point comparison for Java
439 // Compares the top-most stack entries on the FPU stack and stores the result in dst.
440 // The arguments are in reversed order on the stack (i.e., top of stack is first argument).
441 // (semantics as described in JVM spec.)
442 void fcmp2int(Register dst, bool unordered_is_less);
443 // Variant of the above which allows y to be further down the stack
444 // and which only pops x and y if specified. If pop_right is
445 // specified then pop_left must also be specified.
446 void fcmp2int(Register dst, bool unordered_is_less, int index, bool pop_left, bool pop_right);
447
448 // Floating-point remainder for Java (ST0 = ST0 fremr ST1, ST1 is empty afterwards)
449 // tmp is a temporary register, if none is available use noreg
450 void fremr(Register tmp);
451
452 // only if +VerifyFPU
453 void verify_FPU(int stack_depth, const char* s = "illegal FPU state");
454 #endif // !LP64
455
456 // dst = c = a * b + c
457 void fmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c);
458 void fmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c);
459
460 void vfmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len);
461 void vfmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len);
462 void vfmad(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len);
463 void vfmaf(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len);
464
465
466 // same as fcmp2int, but using SSE2
467 void cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less);
468 void cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less);
469
470 // branch to L if FPU flag C2 is set/not set
471 // tmp is a temporary register, if none is available use noreg
472 void jC2 (Register tmp, Label& L);
473 void jnC2(Register tmp, Label& L);
474
475 // Load float value from 'address'. If UseSSE >= 1, the value is loaded into
476 // register xmm0. Otherwise, the value is loaded onto the FPU stack.
477 void load_float(Address src);
478
479 // Store float value to 'address'. If UseSSE >= 1, the value is stored
480 // from register xmm0. Otherwise, the value is stored from the FPU stack.
481 void store_float(Address dst);
482
483 // Load double value from 'address'. If UseSSE >= 2, the value is loaded into
484 // register xmm0. Otherwise, the value is loaded onto the FPU stack.
485 void load_double(Address src);
486
487 // Store double value to 'address'. If UseSSE >= 2, the value is stored
488 // from register xmm0. Otherwise, the value is stored from the FPU stack.
489 void store_double(Address dst);
490
491 #ifndef _LP64
492 // Pop ST (ffree & fincstp combined)
493 void fpop();
494
495 void empty_FPU_stack();
496 #endif // !_LP64
497
498 void push_IU_state();
499 void pop_IU_state();
500
501 void push_FPU_state();
502 void pop_FPU_state();
503
504 void push_CPU_state();
505 void pop_CPU_state();
506
507 // Round up to a power of two
508 void round_to(Register reg, int modulus);
509
510 // Callee saved registers handling
511 void push_callee_saved_registers();
512 void pop_callee_saved_registers();
513
514 // allocation
515 void eden_allocate(
516 Register thread, // Current thread
517 Register obj, // result: pointer to object after successful allocation
518 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise
519 int con_size_in_bytes, // object size in bytes if known at compile time
520 Register t1, // temp register
521 Label& slow_case // continuation point if fast allocation fails
522 );
523 void tlab_allocate(
524 Register thread, // Current thread
525 Register obj, // result: pointer to object after successful allocation
526 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise
527 int con_size_in_bytes, // object size in bytes if known at compile time
528 Register t1, // temp register
529 Register t2, // temp register
530 Label& slow_case // continuation point if fast allocation fails
531 );
532 void zero_memory(Register address, Register length_in_bytes, int offset_in_bytes, Register temp);
533
534 // interface method calling
535 void lookup_interface_method(Register recv_klass,
536 Register intf_klass,
537 RegisterOrConstant itable_index,
538 Register method_result,
539 Register scan_temp,
540 Label& no_such_interface,
541 bool return_method = true);
542
543 // virtual method calling
544 void lookup_virtual_method(Register recv_klass,
545 RegisterOrConstant vtable_index,
546 Register method_result);
547
548 // Test sub_klass against super_klass, with fast and slow paths.
549
550 // The fast path produces a tri-state answer: yes / no / maybe-slow.
551 // One of the three labels can be NULL, meaning take the fall-through.
552 // If super_check_offset is -1, the value is loaded up from super_klass.
553 // No registers are killed, except temp_reg.
554 void check_klass_subtype_fast_path(Register sub_klass,
555 Register super_klass,
556 Register temp_reg,
557 Label* L_success,
558 Label* L_failure,
559 Label* L_slow_path,
560 RegisterOrConstant super_check_offset = RegisterOrConstant(-1));
561
562 // The rest of the type check; must be wired to a corresponding fast path.
563 // It does not repeat the fast path logic, so don't use it standalone.
564 // The temp_reg and temp2_reg can be noreg, if no temps are available.
565 // Updates the sub's secondary super cache as necessary.
566 // If set_cond_codes, condition codes will be Z on success, NZ on failure.
567 void check_klass_subtype_slow_path(Register sub_klass,
568 Register super_klass,
569 Register temp_reg,
570 Register temp2_reg,
571 Label* L_success,
572 Label* L_failure,
573 bool set_cond_codes = false);
574
575 // Simplified, combined version, good for typical uses.
576 // Falls through on failure.
577 void check_klass_subtype(Register sub_klass,
578 Register super_klass,
579 Register temp_reg,
580 Label& L_success);
581
582 void clinit_barrier(Register klass,
583 Register thread,
584 Label* L_fast_path = NULL,
585 Label* L_slow_path = NULL);
586
587 // method handles (JSR 292)
588 Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0);
589
590 //----
591 void set_word_if_not_zero(Register reg); // sets reg to 1 if not zero, otherwise 0
592
593 // Debugging
594
595 // only if +VerifyOops
596 void _verify_oop(Register reg, const char* s, const char* file, int line);
597 void _verify_oop_addr(Address addr, const char* s, const char* file, int line);
598
599 // TODO: verify method and klass metadata (compare against vptr?)
600 void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
601 void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line){}
602
603 #define verify_oop(reg) _verify_oop(reg, "broken oop " #reg, __FILE__, __LINE__)
604 #define verify_oop_msg(reg, msg) _verify_oop(reg, "broken oop " #reg ", " #msg, __FILE__, __LINE__)
605 #define verify_oop_addr(addr) _verify_oop_addr(addr, "broken oop addr " #addr, __FILE__, __LINE__)
606 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
607 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
608
609 // Verify or restore cpu control state after JNI call
610 void restore_cpu_control_state_after_jni();
611
612 // prints msg, dumps registers and stops execution
613 void stop(const char* msg);
614
615 // prints msg and continues
616 void warn(const char* msg);
617
618 // dumps registers and other state
619 void print_state();
620
621 static void debug32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip, char* msg);
622 static void debug64(char* msg, int64_t pc, int64_t regs[]);
623 static void print_state32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip);
624 static void print_state64(int64_t pc, int64_t regs[]);
625
626 void os_breakpoint();
627
628 void untested() { stop("untested"); }
629
630 void unimplemented(const char* what = "");
631
632 void should_not_reach_here() { stop("should not reach here"); }
633
634 void print_CPU_state();
635
636 // Stack overflow checking
637 void bang_stack_with_offset(int offset) {
638 // stack grows down, caller passes positive offset
639 assert(offset > 0, "must bang with negative offset");
640 movl(Address(rsp, (-offset)), rax);
641 }
642
643 // Writes to stack successive pages until offset reached to check for
644 // stack overflow + shadow pages. Also, clobbers tmp
645 void bang_stack_size(Register size, Register tmp);
646
647 // Check for reserved stack access in method being exited (for JIT)
648 void reserved_stack_check();
649
650 virtual RegisterOrConstant delayed_value_impl(intptr_t* delayed_value_addr,
651 Register tmp,
652 int offset);
653
654 // If thread_reg is != noreg the code assumes the register passed contains
655 // the thread (required on 64 bit).
656 void safepoint_poll(Label& slow_path, Register thread_reg, Register temp_reg);
657
658 void verify_tlab();
659
660 // Biased locking support
661 // lock_reg and obj_reg must be loaded up with the appropriate values.
662 // swap_reg must be rax, and is killed.
663 // tmp_reg is optional. If it is supplied (i.e., != noreg) it will
664 // be killed; if not supplied, push/pop will be used internally to
665 // allocate a temporary (inefficient, avoid if possible).
666 // Optional slow case is for implementations (interpreter and C1) which branch to
667 // slow case directly. Leaves condition codes set for C2's Fast_Lock node.
668 // Returns offset of first potentially-faulting instruction for null
669 // check info (currently consumed only by C1). If
670 // swap_reg_contains_mark is true then returns -1 as it is assumed
671 // the calling code has already passed any potential faults.
672 int biased_locking_enter(Register lock_reg, Register obj_reg,
673 Register swap_reg, Register tmp_reg,
674 bool swap_reg_contains_mark,
675 Label& done, Label* slow_case = NULL,
676 BiasedLockingCounters* counters = NULL);
677 void biased_locking_exit (Register obj_reg, Register temp_reg, Label& done);
678
679 Condition negate_condition(Condition cond);
680
681 // Instructions that use AddressLiteral operands. These instruction can handle 32bit/64bit
682 // operands. In general the names are modified to avoid hiding the instruction in Assembler
683 // so that we don't need to implement all the varieties in the Assembler with trivial wrappers
684 // here in MacroAssembler. The major exception to this rule is call
685
686 // Arithmetics
687
688
689 void addptr(Address dst, int32_t src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)) ; }
690 void addptr(Address dst, Register src);
691
692 void addptr(Register dst, Address src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)); }
693 void addptr(Register dst, int32_t src);
694 void addptr(Register dst, Register src);
695 void addptr(Register dst, RegisterOrConstant src) {
696 if (src.is_constant()) addptr(dst, (int) src.as_constant());
697 else addptr(dst, src.as_register());
698 }
699
700 void andptr(Register dst, int32_t src);
701 void andptr(Register src1, Register src2) { LP64_ONLY(andq(src1, src2)) NOT_LP64(andl(src1, src2)) ; }
702
703 void cmp8(AddressLiteral src1, int imm);
704
705 // renamed to drag out the casting of address to int32_t/intptr_t
706 void cmp32(Register src1, int32_t imm);
707
708 void cmp32(AddressLiteral src1, int32_t imm);
709 // compare reg - mem, or reg - &mem
710 void cmp32(Register src1, AddressLiteral src2);
711
712 void cmp32(Register src1, Address src2);
713
714 #ifndef _LP64
715 void cmpklass(Address dst, Metadata* obj);
716 void cmpklass(Register dst, Metadata* obj);
717 void cmpoop(Address dst, jobject obj);
718 void cmpoop_raw(Address dst, jobject obj);
719 #endif // _LP64
720
721 void cmpoop(Register src1, Register src2);
722 void cmpoop(Register src1, Address src2);
723 void cmpoop(Register dst, jobject obj);
724 void cmpoop_raw(Register dst, jobject obj);
725
726 // NOTE src2 must be the lval. This is NOT an mem-mem compare
727 void cmpptr(Address src1, AddressLiteral src2);
728
729 void cmpptr(Register src1, AddressLiteral src2);
730
731 void cmpptr(Register src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
732 void cmpptr(Register src1, Address src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
733 // void cmpptr(Address src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
734
735 void cmpptr(Register src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
736 void cmpptr(Address src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
737
738 // cmp64 to avoild hiding cmpq
739 void cmp64(Register src1, AddressLiteral src);
740
741 void cmpxchgptr(Register reg, Address adr);
742
743 void locked_cmpxchgptr(Register reg, AddressLiteral adr);
744
745
746 void imulptr(Register dst, Register src) { LP64_ONLY(imulq(dst, src)) NOT_LP64(imull(dst, src)); }
747 void imulptr(Register dst, Register src, int imm32) { LP64_ONLY(imulq(dst, src, imm32)) NOT_LP64(imull(dst, src, imm32)); }
748
749
750 void negptr(Register dst) { LP64_ONLY(negq(dst)) NOT_LP64(negl(dst)); }
751
752 void notptr(Register dst) { LP64_ONLY(notq(dst)) NOT_LP64(notl(dst)); }
753
754 void shlptr(Register dst, int32_t shift);
755 void shlptr(Register dst) { LP64_ONLY(shlq(dst)) NOT_LP64(shll(dst)); }
756
757 void shrptr(Register dst, int32_t shift);
758 void shrptr(Register dst) { LP64_ONLY(shrq(dst)) NOT_LP64(shrl(dst)); }
759
760 void sarptr(Register dst) { LP64_ONLY(sarq(dst)) NOT_LP64(sarl(dst)); }
761 void sarptr(Register dst, int32_t src) { LP64_ONLY(sarq(dst, src)) NOT_LP64(sarl(dst, src)); }
762
763 void subptr(Address dst, int32_t src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); }
764
765 void subptr(Register dst, Address src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); }
766 void subptr(Register dst, int32_t src);
767 // Force generation of a 4 byte immediate value even if it fits into 8bit
768 void subptr_imm32(Register dst, int32_t src);
769 void subptr(Register dst, Register src);
770 void subptr(Register dst, RegisterOrConstant src) {
771 if (src.is_constant()) subptr(dst, (int) src.as_constant());
772 else subptr(dst, src.as_register());
773 }
774
775 void sbbptr(Address dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); }
776 void sbbptr(Register dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); }
777
778 void xchgptr(Register src1, Register src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; }
779 void xchgptr(Register src1, Address src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; }
780
781 void xaddptr(Address src1, Register src2) { LP64_ONLY(xaddq(src1, src2)) NOT_LP64(xaddl(src1, src2)) ; }
782
783
784
785 // Helper functions for statistics gathering.
786 // Conditionally (atomically, on MPs) increments passed counter address, preserving condition codes.
787 void cond_inc32(Condition cond, AddressLiteral counter_addr);
788 // Unconditional atomic increment.
789 void atomic_incl(Address counter_addr);
790 void atomic_incl(AddressLiteral counter_addr, Register scr = rscratch1);
791 #ifdef _LP64
792 void atomic_incq(Address counter_addr);
793 void atomic_incq(AddressLiteral counter_addr, Register scr = rscratch1);
794 #endif
795 void atomic_incptr(AddressLiteral counter_addr, Register scr = rscratch1) { LP64_ONLY(atomic_incq(counter_addr, scr)) NOT_LP64(atomic_incl(counter_addr, scr)) ; }
796 void atomic_incptr(Address counter_addr) { LP64_ONLY(atomic_incq(counter_addr)) NOT_LP64(atomic_incl(counter_addr)) ; }
797
798 void lea(Register dst, AddressLiteral adr);
799 void lea(Address dst, AddressLiteral adr);
800 void lea(Register dst, Address adr) { Assembler::lea(dst, adr); }
801
802 void leal32(Register dst, Address src) { leal(dst, src); }
803
804 // Import other testl() methods from the parent class or else
805 // they will be hidden by the following overriding declaration.
806 using Assembler::testl;
807 void testl(Register dst, AddressLiteral src);
808
809 void orptr(Register dst, Address src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); }
810 void orptr(Register dst, Register src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); }
811 void orptr(Register dst, int32_t src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); }
812 void orptr(Address dst, int32_t imm32) { LP64_ONLY(orq(dst, imm32)) NOT_LP64(orl(dst, imm32)); }
813
814 void testptr(Register src, int32_t imm32) { LP64_ONLY(testq(src, imm32)) NOT_LP64(testl(src, imm32)); }
815 void testptr(Register src1, Address src2) { LP64_ONLY(testq(src1, src2)) NOT_LP64(testl(src1, src2)); }
816 void testptr(Register src1, Register src2);
817
818 void xorptr(Register dst, Register src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); }
819 void xorptr(Register dst, Address src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); }
820
821 // Calls
822
823 void call(Label& L, relocInfo::relocType rtype);
824 void call(Register entry);
825
826 // NOTE: this call transfers to the effective address of entry NOT
827 // the address contained by entry. This is because this is more natural
828 // for jumps/calls.
829 void call(AddressLiteral entry);
830
831 // Emit the CompiledIC call idiom
832 void ic_call(address entry, jint method_index = 0);
833
834 // Jumps
835
836 // NOTE: these jumps tranfer to the effective address of dst NOT
837 // the address contained by dst. This is because this is more natural
838 // for jumps/calls.
839 void jump(AddressLiteral dst);
840 void jump_cc(Condition cc, AddressLiteral dst);
841
842 // 32bit can do a case table jump in one instruction but we no longer allow the base
843 // to be installed in the Address class. This jump will tranfers to the address
844 // contained in the location described by entry (not the address of entry)
845 void jump(ArrayAddress entry);
846
847 // Floating
848
849 void andpd(XMMRegister dst, Address src) { Assembler::andpd(dst, src); }
850 void andpd(XMMRegister dst, AddressLiteral src, Register scratch_reg = rscratch1);
851 void andpd(XMMRegister dst, XMMRegister src) { Assembler::andpd(dst, src); }
852
853 void andps(XMMRegister dst, XMMRegister src) { Assembler::andps(dst, src); }
854 void andps(XMMRegister dst, Address src) { Assembler::andps(dst, src); }
855 void andps(XMMRegister dst, AddressLiteral src, Register scratch_reg = rscratch1);
856
857 void comiss(XMMRegister dst, XMMRegister src) { Assembler::comiss(dst, src); }
858 void comiss(XMMRegister dst, Address src) { Assembler::comiss(dst, src); }
859 void comiss(XMMRegister dst, AddressLiteral src);
860
861 void comisd(XMMRegister dst, XMMRegister src) { Assembler::comisd(dst, src); }
862 void comisd(XMMRegister dst, Address src) { Assembler::comisd(dst, src); }
863 void comisd(XMMRegister dst, AddressLiteral src);
864
865 #ifndef _LP64
866 void fadd_s(Address src) { Assembler::fadd_s(src); }
867 void fadd_s(AddressLiteral src) { Assembler::fadd_s(as_Address(src)); }
868
869 void fldcw(Address src) { Assembler::fldcw(src); }
870 void fldcw(AddressLiteral src);
871
872 void fld_s(int index) { Assembler::fld_s(index); }
873 void fld_s(Address src) { Assembler::fld_s(src); }
874 void fld_s(AddressLiteral src);
875
876 void fld_d(Address src) { Assembler::fld_d(src); }
877 void fld_d(AddressLiteral src);
878
879 void fld_x(Address src) { Assembler::fld_x(src); }
880 void fld_x(AddressLiteral src);
881
882 void fmul_s(Address src) { Assembler::fmul_s(src); }
883 void fmul_s(AddressLiteral src) { Assembler::fmul_s(as_Address(src)); }
884 #endif // _LP64
885
886 void ldmxcsr(Address src) { Assembler::ldmxcsr(src); }
887 void ldmxcsr(AddressLiteral src);
888
889 #ifdef _LP64
890 private:
891 void sha256_AVX2_one_round_compute(
892 Register reg_old_h,
893 Register reg_a,
894 Register reg_b,
895 Register reg_c,
896 Register reg_d,
897 Register reg_e,
898 Register reg_f,
899 Register reg_g,
900 Register reg_h,
901 int iter);
902 void sha256_AVX2_four_rounds_compute_first(int start);
903 void sha256_AVX2_four_rounds_compute_last(int start);
904 void sha256_AVX2_one_round_and_sched(
905 XMMRegister xmm_0, /* == ymm4 on 0, 1, 2, 3 iterations, then rotate 4 registers left on 4, 8, 12 iterations */
906 XMMRegister xmm_1, /* ymm5 */ /* full cycle is 16 iterations */
907 XMMRegister xmm_2, /* ymm6 */
908 XMMRegister xmm_3, /* ymm7 */
909 Register reg_a, /* == eax on 0 iteration, then rotate 8 register right on each next iteration */
910 Register reg_b, /* ebx */ /* full cycle is 8 iterations */
911 Register reg_c, /* edi */
912 Register reg_d, /* esi */
913 Register reg_e, /* r8d */
914 Register reg_f, /* r9d */
915 Register reg_g, /* r10d */
916 Register reg_h, /* r11d */
917 int iter);
918
919 void addm(int disp, Register r1, Register r2);
920 void gfmul(XMMRegister tmp0, XMMRegister t);
921 void schoolbookAAD(int i, Register subkeyH, XMMRegister data, XMMRegister tmp0,
922 XMMRegister tmp1, XMMRegister tmp2, XMMRegister tmp3);
923 void generateHtbl_one_block(Register htbl);
924 void generateHtbl_eight_blocks(Register htbl);
925 public:
926 void sha256_AVX2(XMMRegister msg, XMMRegister state0, XMMRegister state1, XMMRegister msgtmp0,
927 XMMRegister msgtmp1, XMMRegister msgtmp2, XMMRegister msgtmp3, XMMRegister msgtmp4,
928 Register buf, Register state, Register ofs, Register limit, Register rsp,
929 bool multi_block, XMMRegister shuf_mask);
930 void avx_ghash(Register state, Register htbl, Register data, Register blocks);
931 #endif
932
933 #ifdef _LP64
934 private:
935 void sha512_AVX2_one_round_compute(Register old_h, Register a, Register b, Register c, Register d,
936 Register e, Register f, Register g, Register h, int iteration);
937
938 void sha512_AVX2_one_round_and_schedule(XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
939 Register a, Register b, Register c, Register d, Register e, Register f,
940 Register g, Register h, int iteration);
941
942 void addmq(int disp, Register r1, Register r2);
943 public:
944 void sha512_AVX2(XMMRegister msg, XMMRegister state0, XMMRegister state1, XMMRegister msgtmp0,
945 XMMRegister msgtmp1, XMMRegister msgtmp2, XMMRegister msgtmp3, XMMRegister msgtmp4,
946 Register buf, Register state, Register ofs, Register limit, Register rsp, bool multi_block,
947 XMMRegister shuf_mask);
948 private:
949 void roundEnc(XMMRegister key, int rnum);
950 void lastroundEnc(XMMRegister key, int rnum);
951 void roundDec(XMMRegister key, int rnum);
952 void lastroundDec(XMMRegister key, int rnum);
953 void ev_load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask);
954
955 public:
956 void aesecb_encrypt(Register source_addr, Register dest_addr, Register key, Register len);
957 void aesecb_decrypt(Register source_addr, Register dest_addr, Register key, Register len);
958 void aesctr_encrypt(Register src_addr, Register dest_addr, Register key, Register counter,
959 Register len_reg, Register used, Register used_addr, Register saved_encCounter_start);
960
961 #endif
962
963 void fast_sha1(XMMRegister abcd, XMMRegister e0, XMMRegister e1, XMMRegister msg0,
964 XMMRegister msg1, XMMRegister msg2, XMMRegister msg3, XMMRegister shuf_mask,
965 Register buf, Register state, Register ofs, Register limit, Register rsp,
966 bool multi_block);
967
968 #ifdef _LP64
969 void fast_sha256(XMMRegister msg, XMMRegister state0, XMMRegister state1, XMMRegister msgtmp0,
970 XMMRegister msgtmp1, XMMRegister msgtmp2, XMMRegister msgtmp3, XMMRegister msgtmp4,
971 Register buf, Register state, Register ofs, Register limit, Register rsp,
972 bool multi_block, XMMRegister shuf_mask);
973 #else
974 void fast_sha256(XMMRegister msg, XMMRegister state0, XMMRegister state1, XMMRegister msgtmp0,
975 XMMRegister msgtmp1, XMMRegister msgtmp2, XMMRegister msgtmp3, XMMRegister msgtmp4,
976 Register buf, Register state, Register ofs, Register limit, Register rsp,
977 bool multi_block);
978 #endif
979
980 void fast_exp(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
981 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
982 Register rax, Register rcx, Register rdx, Register tmp);
983
984 #ifdef _LP64
985 void fast_log(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
986 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
987 Register rax, Register rcx, Register rdx, Register tmp1, Register tmp2);
988
989 void fast_log10(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
990 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
991 Register rax, Register rcx, Register rdx, Register r11);
992
993 void fast_pow(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, XMMRegister xmm4,
994 XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, Register rax, Register rcx,
995 Register rdx, Register tmp1, Register tmp2, Register tmp3, Register tmp4);
996
997 void fast_sin(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
998 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
999 Register rax, Register rbx, Register rcx, Register rdx, Register tmp1, Register tmp2,
1000 Register tmp3, Register tmp4);
1001
1002 void fast_cos(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1003 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1004 Register rax, Register rcx, Register rdx, Register tmp1,
1005 Register tmp2, Register tmp3, Register tmp4);
1006 void fast_tan(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1007 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1008 Register rax, Register rcx, Register rdx, Register tmp1,
1009 Register tmp2, Register tmp3, Register tmp4);
1010 #else
1011 void fast_log(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1012 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1013 Register rax, Register rcx, Register rdx, Register tmp1);
1014
1015 void fast_log10(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1016 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1017 Register rax, Register rcx, Register rdx, Register tmp);
1018
1019 void fast_pow(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, XMMRegister xmm4,
1020 XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, Register rax, Register rcx,
1021 Register rdx, Register tmp);
1022
1023 void fast_sin(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1024 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1025 Register rax, Register rbx, Register rdx);
1026
1027 void fast_cos(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1028 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1029 Register rax, Register rcx, Register rdx, Register tmp);
1030
1031 void libm_sincos_huge(XMMRegister xmm0, XMMRegister xmm1, Register eax, Register ecx,
1032 Register edx, Register ebx, Register esi, Register edi,
1033 Register ebp, Register esp);
1034
1035 void libm_reduce_pi04l(Register eax, Register ecx, Register edx, Register ebx,
1036 Register esi, Register edi, Register ebp, Register esp);
1037
1038 void libm_tancot_huge(XMMRegister xmm0, XMMRegister xmm1, Register eax, Register ecx,
1039 Register edx, Register ebx, Register esi, Register edi,
1040 Register ebp, Register esp);
1041
1042 void fast_tan(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1043 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1044 Register rax, Register rcx, Register rdx, Register tmp);
1045 #endif
1046
1047 private:
1048
1049 // these are private because users should be doing movflt/movdbl
1050
1051 void movss(XMMRegister dst, XMMRegister src) { Assembler::movss(dst, src); }
1052 void movss(Address dst, XMMRegister src) { Assembler::movss(dst, src); }
1053 void movss(XMMRegister dst, Address src) { Assembler::movss(dst, src); }
1054 void movss(XMMRegister dst, AddressLiteral src);
1055
1056 void movlpd(XMMRegister dst, Address src) {Assembler::movlpd(dst, src); }
1057 void movlpd(XMMRegister dst, AddressLiteral src);
1058
1059 public:
1060
1061 void addsd(XMMRegister dst, XMMRegister src) { Assembler::addsd(dst, src); }
1062 void addsd(XMMRegister dst, Address src) { Assembler::addsd(dst, src); }
1063 void addsd(XMMRegister dst, AddressLiteral src);
1064
1065 void addss(XMMRegister dst, XMMRegister src) { Assembler::addss(dst, src); }
1066 void addss(XMMRegister dst, Address src) { Assembler::addss(dst, src); }
1067 void addss(XMMRegister dst, AddressLiteral src);
1068
1069 void addpd(XMMRegister dst, XMMRegister src) { Assembler::addpd(dst, src); }
1070 void addpd(XMMRegister dst, Address src) { Assembler::addpd(dst, src); }
1071 void addpd(XMMRegister dst, AddressLiteral src);
1072
1073 void divsd(XMMRegister dst, XMMRegister src) { Assembler::divsd(dst, src); }
1074 void divsd(XMMRegister dst, Address src) { Assembler::divsd(dst, src); }
1075 void divsd(XMMRegister dst, AddressLiteral src);
1076
1077 void divss(XMMRegister dst, XMMRegister src) { Assembler::divss(dst, src); }
1078 void divss(XMMRegister dst, Address src) { Assembler::divss(dst, src); }
1079 void divss(XMMRegister dst, AddressLiteral src);
1080
1081 // Move Unaligned Double Quadword
1082 void movdqu(Address dst, XMMRegister src);
1083 void movdqu(XMMRegister dst, Address src);
1084 void movdqu(XMMRegister dst, XMMRegister src);
1085 void movdqu(XMMRegister dst, AddressLiteral src, Register scratchReg = rscratch1);
1086
1087 void kmovwl(KRegister dst, Register src) { Assembler::kmovwl(dst, src); }
1088 void kmovwl(Register dst, KRegister src) { Assembler::kmovwl(dst, src); }
1089 void kmovwl(KRegister dst, Address src) { Assembler::kmovwl(dst, src); }
1090 void kmovwl(KRegister dst, AddressLiteral src, Register scratch_reg = rscratch1);
1091
1092 // AVX Unaligned forms
1093 void vmovdqu(Address dst, XMMRegister src);
1094 void vmovdqu(XMMRegister dst, Address src);
1095 void vmovdqu(XMMRegister dst, XMMRegister src);
1096 void vmovdqu(XMMRegister dst, AddressLiteral src, Register scratch_reg = rscratch1);
1097
1098 // AVX512 Unaligned
1099 void evmovdqub(Address dst, XMMRegister src, bool merge, int vector_len) { Assembler::evmovdqub(dst, src, merge, vector_len); }
1100 void evmovdqub(XMMRegister dst, Address src, bool merge, int vector_len) { Assembler::evmovdqub(dst, src, merge, vector_len); }
1101 void evmovdqub(XMMRegister dst, XMMRegister src, bool merge, int vector_len) { Assembler::evmovdqub(dst, src, merge, vector_len); }
1102 void evmovdqub(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len) { Assembler::evmovdqub(dst, mask, src, merge, vector_len); }
1103 void evmovdqub(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register scratch_reg);
1104
1105 void evmovdquw(Address dst, XMMRegister src, bool merge, int vector_len) { Assembler::evmovdquw(dst, src, merge, vector_len); }
1106 void evmovdquw(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len) { Assembler::evmovdquw(dst, mask, src, merge, vector_len); }
1107 void evmovdquw(XMMRegister dst, Address src, bool merge, int vector_len) { Assembler::evmovdquw(dst, src, merge, vector_len); }
1108 void evmovdquw(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len) { Assembler::evmovdquw(dst, mask, src, merge, vector_len); }
1109 void evmovdquw(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register scratch_reg);
1110
1111 void evmovdqul(Address dst, XMMRegister src, int vector_len) { Assembler::evmovdqul(dst, src, vector_len); }
1112 void evmovdqul(XMMRegister dst, Address src, int vector_len) { Assembler::evmovdqul(dst, src, vector_len); }
1113 void evmovdqul(XMMRegister dst, XMMRegister src, int vector_len) {
1114 if (dst->encoding() == src->encoding()) return;
1115 Assembler::evmovdqul(dst, src, vector_len);
1116 }
1117 void evmovdqul(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len) { Assembler::evmovdqul(dst, mask, src, merge, vector_len); }
1118 void evmovdqul(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len) { Assembler::evmovdqul(dst, mask, src, merge, vector_len); }
1119 void evmovdqul(XMMRegister dst, KRegister mask, XMMRegister src, bool merge, int vector_len) {
1120 if (dst->encoding() == src->encoding() && mask == k0) return;
1121 Assembler::evmovdqul(dst, mask, src, merge, vector_len);
1122 }
1123 void evmovdqul(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register scratch_reg);
1124
1125 void evmovdquq(XMMRegister dst, Address src, int vector_len) { Assembler::evmovdquq(dst, src, vector_len); }
1126 void evmovdquq(Address dst, XMMRegister src, int vector_len) { Assembler::evmovdquq(dst, src, vector_len); }
1127 void evmovdquq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch);
1128 void evmovdquq(XMMRegister dst, XMMRegister src, int vector_len) {
1129 if (dst->encoding() == src->encoding()) return;
1130 Assembler::evmovdquq(dst, src, vector_len);
1131 }
1132 void evmovdquq(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len) { Assembler::evmovdquq(dst, mask, src, merge, vector_len); }
1133 void evmovdquq(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len) { Assembler::evmovdquq(dst, mask, src, merge, vector_len); }
1134 void evmovdquq(XMMRegister dst, KRegister mask, XMMRegister src, bool merge, int vector_len) {
1135 if (dst->encoding() == src->encoding() && mask == k0) return;
1136 Assembler::evmovdquq(dst, mask, src, merge, vector_len);
1137 }
1138 void evmovdquq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register scratch_reg);
1139
1140 // Move Aligned Double Quadword
1141 void movdqa(XMMRegister dst, Address src) { Assembler::movdqa(dst, src); }
1142 void movdqa(XMMRegister dst, XMMRegister src) { Assembler::movdqa(dst, src); }
1143 void movdqa(XMMRegister dst, AddressLiteral src);
1144
1145 void movsd(XMMRegister dst, XMMRegister src) { Assembler::movsd(dst, src); }
1146 void movsd(Address dst, XMMRegister src) { Assembler::movsd(dst, src); }
1147 void movsd(XMMRegister dst, Address src) { Assembler::movsd(dst, src); }
1148 void movsd(XMMRegister dst, AddressLiteral src);
1149
1150 void mulpd(XMMRegister dst, XMMRegister src) { Assembler::mulpd(dst, src); }
1151 void mulpd(XMMRegister dst, Address src) { Assembler::mulpd(dst, src); }
1152 void mulpd(XMMRegister dst, AddressLiteral src);
1153
1154 void mulsd(XMMRegister dst, XMMRegister src) { Assembler::mulsd(dst, src); }
1155 void mulsd(XMMRegister dst, Address src) { Assembler::mulsd(dst, src); }
1156 void mulsd(XMMRegister dst, AddressLiteral src);
1157
1158 void mulss(XMMRegister dst, XMMRegister src) { Assembler::mulss(dst, src); }
1159 void mulss(XMMRegister dst, Address src) { Assembler::mulss(dst, src); }
1160 void mulss(XMMRegister dst, AddressLiteral src);
1161
1162 // Carry-Less Multiplication Quadword
1163 void pclmulldq(XMMRegister dst, XMMRegister src) {
1164 // 0x00 - multiply lower 64 bits [0:63]
1165 Assembler::pclmulqdq(dst, src, 0x00);
1166 }
1167 void pclmulhdq(XMMRegister dst, XMMRegister src) {
1168 // 0x11 - multiply upper 64 bits [64:127]
1169 Assembler::pclmulqdq(dst, src, 0x11);
1170 }
1171
1172 void pcmpeqb(XMMRegister dst, XMMRegister src);
1173 void pcmpeqw(XMMRegister dst, XMMRegister src);
1174
1175 void pcmpestri(XMMRegister dst, Address src, int imm8);
1176 void pcmpestri(XMMRegister dst, XMMRegister src, int imm8);
1177
1178 void pmovzxbw(XMMRegister dst, XMMRegister src);
1179 void pmovzxbw(XMMRegister dst, Address src);
1180
1181 void pmovmskb(Register dst, XMMRegister src);
1182
1183 void ptest(XMMRegister dst, XMMRegister src);
1184
1185 void sqrtsd(XMMRegister dst, XMMRegister src) { Assembler::sqrtsd(dst, src); }
1186 void sqrtsd(XMMRegister dst, Address src) { Assembler::sqrtsd(dst, src); }
1187 void sqrtsd(XMMRegister dst, AddressLiteral src);
1188
1189 void roundsd(XMMRegister dst, XMMRegister src, int32_t rmode) { Assembler::roundsd(dst, src, rmode); }
1190 void roundsd(XMMRegister dst, Address src, int32_t rmode) { Assembler::roundsd(dst, src, rmode); }
1191 void roundsd(XMMRegister dst, AddressLiteral src, int32_t rmode, Register scratch_reg);
1192
1193 void sqrtss(XMMRegister dst, XMMRegister src) { Assembler::sqrtss(dst, src); }
1194 void sqrtss(XMMRegister dst, Address src) { Assembler::sqrtss(dst, src); }
1195 void sqrtss(XMMRegister dst, AddressLiteral src);
1196
1197 void subsd(XMMRegister dst, XMMRegister src) { Assembler::subsd(dst, src); }
1198 void subsd(XMMRegister dst, Address src) { Assembler::subsd(dst, src); }
1199 void subsd(XMMRegister dst, AddressLiteral src);
1200
1201 void subss(XMMRegister dst, XMMRegister src) { Assembler::subss(dst, src); }
1202 void subss(XMMRegister dst, Address src) { Assembler::subss(dst, src); }
1203 void subss(XMMRegister dst, AddressLiteral src);
1204
1205 void ucomiss(XMMRegister dst, XMMRegister src) { Assembler::ucomiss(dst, src); }
1206 void ucomiss(XMMRegister dst, Address src) { Assembler::ucomiss(dst, src); }
1207 void ucomiss(XMMRegister dst, AddressLiteral src);
1208
1209 void ucomisd(XMMRegister dst, XMMRegister src) { Assembler::ucomisd(dst, src); }
1210 void ucomisd(XMMRegister dst, Address src) { Assembler::ucomisd(dst, src); }
1211 void ucomisd(XMMRegister dst, AddressLiteral src);
1212
1213 // Bitwise Logical XOR of Packed Double-Precision Floating-Point Values
1214 void xorpd(XMMRegister dst, XMMRegister src);
1215 void xorpd(XMMRegister dst, Address src) { Assembler::xorpd(dst, src); }
1216 void xorpd(XMMRegister dst, AddressLiteral src, Register scratch_reg = rscratch1);
1217
1218 // Bitwise Logical XOR of Packed Single-Precision Floating-Point Values
1219 void xorps(XMMRegister dst, XMMRegister src);
1220 void xorps(XMMRegister dst, Address src) { Assembler::xorps(dst, src); }
1221 void xorps(XMMRegister dst, AddressLiteral src, Register scratch_reg = rscratch1);
1222
1223 // Shuffle Bytes
1224 void pshufb(XMMRegister dst, XMMRegister src) { Assembler::pshufb(dst, src); }
1225 void pshufb(XMMRegister dst, Address src) { Assembler::pshufb(dst, src); }
1226 void pshufb(XMMRegister dst, AddressLiteral src);
1227 // AVX 3-operands instructions
1228
1229 void vaddsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vaddsd(dst, nds, src); }
1230 void vaddsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vaddsd(dst, nds, src); }
1231 void vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1232
1233 void vaddss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vaddss(dst, nds, src); }
1234 void vaddss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vaddss(dst, nds, src); }
1235 void vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1236
1237 void vabsss(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len);
1238 void vabssd(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len);
1239
1240 void vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1241 void vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1242
1243 void vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1244 void vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1245
1246 void vpaddd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vpaddd(dst, nds, src, vector_len); }
1247 void vpaddd(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vpaddd(dst, nds, src, vector_len); }
1248 void vpaddd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch);
1249
1250 void vpand(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vpand(dst, nds, src, vector_len); }
1251 void vpand(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vpand(dst, nds, src, vector_len); }
1252 void vpand(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register scratch_reg = rscratch1);
1253
1254 void vpbroadcastw(XMMRegister dst, XMMRegister src, int vector_len);
1255 void vpbroadcastw(XMMRegister dst, Address src, int vector_len) { Assembler::vpbroadcastw(dst, src, vector_len); }
1256
1257 void vpcmpeqb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1258
1259 void vpcmpeqw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1260 void evpcmpeqd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src, int vector_len, Register scratch_reg);
1261
1262 // Vector compares
1263 void evpcmpd(KRegister kdst, KRegister mask, XMMRegister nds, XMMRegister src,
1264 int comparison, int vector_len) { Assembler::evpcmpd(kdst, mask, nds, src, comparison, vector_len); }
1265 void evpcmpd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
1266 int comparison, int vector_len, Register scratch_reg);
1267 void evpcmpq(KRegister kdst, KRegister mask, XMMRegister nds, XMMRegister src,
1268 int comparison, int vector_len) { Assembler::evpcmpq(kdst, mask, nds, src, comparison, vector_len); }
1269 void evpcmpq(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
1270 int comparison, int vector_len, Register scratch_reg);
1271 void evpcmpb(KRegister kdst, KRegister mask, XMMRegister nds, XMMRegister src,
1272 int comparison, int vector_len) { Assembler::evpcmpb(kdst, mask, nds, src, comparison, vector_len); }
1273 void evpcmpb(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
1274 int comparison, int vector_len, Register scratch_reg);
1275 void evpcmpw(KRegister kdst, KRegister mask, XMMRegister nds, XMMRegister src,
1276 int comparison, int vector_len) { Assembler::evpcmpw(kdst, mask, nds, src, comparison, vector_len); }
1277 void evpcmpw(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src,
1278 int comparison, int vector_len, Register scratch_reg);
1279
1280
1281 // Emit comparison instruction for the specified comparison predicate.
1282 void vpcmpCCW(XMMRegister dst, XMMRegister nds, XMMRegister src, ComparisonPredicate cond, Width width, int vector_len, Register scratch_reg);
1283 void vpcmpCC(XMMRegister dst, XMMRegister nds, XMMRegister src, int cond_encoding, Width width, int vector_len);
1284
1285 void vpmovzxbw(XMMRegister dst, Address src, int vector_len);
1286 void vpmovzxbw(XMMRegister dst, XMMRegister src, int vector_len) { Assembler::vpmovzxbw(dst, src, vector_len); }
1287
1288 void vpmovmskb(Register dst, XMMRegister src);
1289
1290 void vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1291 void vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1292
1293 void vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1294 void vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1295
1296 void vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1297 void vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1298
1299 void vpsraw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len);
1300 void vpsraw(XMMRegister dst, XMMRegister nds, int shift, int vector_len);
1301
1302 void evpsraq(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len);
1303 void evpsraq(XMMRegister dst, XMMRegister nds, int shift, int vector_len);
1304
1305 void vpsrlw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len);
1306 void vpsrlw(XMMRegister dst, XMMRegister nds, int shift, int vector_len);
1307
1308 void vpsllw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len);
1309 void vpsllw(XMMRegister dst, XMMRegister nds, int shift, int vector_len);
1310
1311 void vptest(XMMRegister dst, XMMRegister src);
1312 void vptest(XMMRegister dst, XMMRegister src, int vector_len) { Assembler::vptest(dst, src, vector_len); }
1313
1314 void punpcklbw(XMMRegister dst, XMMRegister src);
1315 void punpcklbw(XMMRegister dst, Address src) { Assembler::punpcklbw(dst, src); }
1316
1317 void pshufd(XMMRegister dst, Address src, int mode);
1318 void pshufd(XMMRegister dst, XMMRegister src, int mode) { Assembler::pshufd(dst, src, mode); }
1319
1320 void pshuflw(XMMRegister dst, XMMRegister src, int mode);
1321 void pshuflw(XMMRegister dst, Address src, int mode) { Assembler::pshuflw(dst, src, mode); }
1322
1323 void vandpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vandpd(dst, nds, src, vector_len); }
1324 void vandpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vandpd(dst, nds, src, vector_len); }
1325 void vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register scratch_reg = rscratch1);
1326
1327 void vandps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vandps(dst, nds, src, vector_len); }
1328 void vandps(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vandps(dst, nds, src, vector_len); }
1329 void vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register scratch_reg = rscratch1);
1330
1331 void evpord(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src, bool merge, int vector_len, Register scratch_reg);
1332
1333 void vdivsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vdivsd(dst, nds, src); }
1334 void vdivsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vdivsd(dst, nds, src); }
1335 void vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1336
1337 void vdivss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vdivss(dst, nds, src); }
1338 void vdivss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vdivss(dst, nds, src); }
1339 void vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1340
1341 void vmulsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vmulsd(dst, nds, src); }
1342 void vmulsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vmulsd(dst, nds, src); }
1343 void vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1344
1345 void vmulss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vmulss(dst, nds, src); }
1346 void vmulss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vmulss(dst, nds, src); }
1347 void vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1348
1349 void vsubsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vsubsd(dst, nds, src); }
1350 void vsubsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vsubsd(dst, nds, src); }
1351 void vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1352
1353 void vsubss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vsubss(dst, nds, src); }
1354 void vsubss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vsubss(dst, nds, src); }
1355 void vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1356
1357 void vnegatess(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1358 void vnegatesd(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1359
1360 // AVX Vector instructions
1361
1362 void vxorpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vxorpd(dst, nds, src, vector_len); }
1363 void vxorpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vxorpd(dst, nds, src, vector_len); }
1364 void vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register scratch_reg = rscratch1);
1365
1366 void vxorps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vxorps(dst, nds, src, vector_len); }
1367 void vxorps(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vxorps(dst, nds, src, vector_len); }
1368 void vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register scratch_reg = rscratch1);
1369
1370 void vpxor(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
1371 if (UseAVX > 1 || (vector_len < 1)) // vpxor 256 bit is available only in AVX2
1372 Assembler::vpxor(dst, nds, src, vector_len);
1373 else
1374 Assembler::vxorpd(dst, nds, src, vector_len);
1375 }
1376 void vpxor(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
1377 if (UseAVX > 1 || (vector_len < 1)) // vpxor 256 bit is available only in AVX2
1378 Assembler::vpxor(dst, nds, src, vector_len);
1379 else
1380 Assembler::vxorpd(dst, nds, src, vector_len);
1381 }
1382 void vpxor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register scratch_reg = rscratch1);
1383
1384 // Simple version for AVX2 256bit vectors
1385 void vpxor(XMMRegister dst, XMMRegister src) { Assembler::vpxor(dst, dst, src, true); }
1386 void vpxor(XMMRegister dst, Address src) { Assembler::vpxor(dst, dst, src, true); }
1387
1388 void vpermd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vpermd(dst, nds, src, vector_len); }
1389 void vpermd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register scratch_reg);
1390
1391 void vinserti128(XMMRegister dst, XMMRegister nds, XMMRegister src, uint8_t imm8) {
1392 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1393 Assembler::vinserti32x4(dst, dst, src, imm8);
1394 } else if (UseAVX > 1) {
1395 // vinserti128 is available only in AVX2
1396 Assembler::vinserti128(dst, nds, src, imm8);
1397 } else {
1398 Assembler::vinsertf128(dst, nds, src, imm8);
1399 }
1400 }
1401
1402 void vinserti128(XMMRegister dst, XMMRegister nds, Address src, uint8_t imm8) {
1403 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1404 Assembler::vinserti32x4(dst, dst, src, imm8);
1405 } else if (UseAVX > 1) {
1406 // vinserti128 is available only in AVX2
1407 Assembler::vinserti128(dst, nds, src, imm8);
1408 } else {
1409 Assembler::vinsertf128(dst, nds, src, imm8);
1410 }
1411 }
1412
1413 void vextracti128(XMMRegister dst, XMMRegister src, uint8_t imm8) {
1414 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1415 Assembler::vextracti32x4(dst, src, imm8);
1416 } else if (UseAVX > 1) {
1417 // vextracti128 is available only in AVX2
1418 Assembler::vextracti128(dst, src, imm8);
1419 } else {
1420 Assembler::vextractf128(dst, src, imm8);
1421 }
1422 }
1423
1424 void vextracti128(Address dst, XMMRegister src, uint8_t imm8) {
1425 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1426 Assembler::vextracti32x4(dst, src, imm8);
1427 } else if (UseAVX > 1) {
1428 // vextracti128 is available only in AVX2
1429 Assembler::vextracti128(dst, src, imm8);
1430 } else {
1431 Assembler::vextractf128(dst, src, imm8);
1432 }
1433 }
1434
1435 // 128bit copy to/from high 128 bits of 256bit (YMM) vector registers
1436 void vinserti128_high(XMMRegister dst, XMMRegister src) {
1437 vinserti128(dst, dst, src, 1);
1438 }
1439 void vinserti128_high(XMMRegister dst, Address src) {
1440 vinserti128(dst, dst, src, 1);
1441 }
1442 void vextracti128_high(XMMRegister dst, XMMRegister src) {
1443 vextracti128(dst, src, 1);
1444 }
1445 void vextracti128_high(Address dst, XMMRegister src) {
1446 vextracti128(dst, src, 1);
1447 }
1448
1449 void vinsertf128_high(XMMRegister dst, XMMRegister src) {
1450 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1451 Assembler::vinsertf32x4(dst, dst, src, 1);
1452 } else {
1453 Assembler::vinsertf128(dst, dst, src, 1);
1454 }
1455 }
1456
1457 void vinsertf128_high(XMMRegister dst, Address src) {
1458 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1459 Assembler::vinsertf32x4(dst, dst, src, 1);
1460 } else {
1461 Assembler::vinsertf128(dst, dst, src, 1);
1462 }
1463 }
1464
1465 void vextractf128_high(XMMRegister dst, XMMRegister src) {
1466 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1467 Assembler::vextractf32x4(dst, src, 1);
1468 } else {
1469 Assembler::vextractf128(dst, src, 1);
1470 }
1471 }
1472
1473 void vextractf128_high(Address dst, XMMRegister src) {
1474 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1475 Assembler::vextractf32x4(dst, src, 1);
1476 } else {
1477 Assembler::vextractf128(dst, src, 1);
1478 }
1479 }
1480
1481 // 256bit copy to/from high 256 bits of 512bit (ZMM) vector registers
1482 void vinserti64x4_high(XMMRegister dst, XMMRegister src) {
1483 Assembler::vinserti64x4(dst, dst, src, 1);
1484 }
1485 void vinsertf64x4_high(XMMRegister dst, XMMRegister src) {
1486 Assembler::vinsertf64x4(dst, dst, src, 1);
1487 }
1488 void vextracti64x4_high(XMMRegister dst, XMMRegister src) {
1489 Assembler::vextracti64x4(dst, src, 1);
1490 }
1491 void vextractf64x4_high(XMMRegister dst, XMMRegister src) {
1492 Assembler::vextractf64x4(dst, src, 1);
1493 }
1494 void vextractf64x4_high(Address dst, XMMRegister src) {
1495 Assembler::vextractf64x4(dst, src, 1);
1496 }
1497 void vinsertf64x4_high(XMMRegister dst, Address src) {
1498 Assembler::vinsertf64x4(dst, dst, src, 1);
1499 }
1500
1501 // 128bit copy to/from low 128 bits of 256bit (YMM) vector registers
1502 void vinserti128_low(XMMRegister dst, XMMRegister src) {
1503 vinserti128(dst, dst, src, 0);
1504 }
1505 void vinserti128_low(XMMRegister dst, Address src) {
1506 vinserti128(dst, dst, src, 0);
1507 }
1508 void vextracti128_low(XMMRegister dst, XMMRegister src) {
1509 vextracti128(dst, src, 0);
1510 }
1511 void vextracti128_low(Address dst, XMMRegister src) {
1512 vextracti128(dst, src, 0);
1513 }
1514
1515 void vinsertf128_low(XMMRegister dst, XMMRegister src) {
1516 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1517 Assembler::vinsertf32x4(dst, dst, src, 0);
1518 } else {
1519 Assembler::vinsertf128(dst, dst, src, 0);
1520 }
1521 }
1522
1523 void vinsertf128_low(XMMRegister dst, Address src) {
1524 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1525 Assembler::vinsertf32x4(dst, dst, src, 0);
1526 } else {
1527 Assembler::vinsertf128(dst, dst, src, 0);
1528 }
1529 }
1530
1531 void vextractf128_low(XMMRegister dst, XMMRegister src) {
1532 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1533 Assembler::vextractf32x4(dst, src, 0);
1534 } else {
1535 Assembler::vextractf128(dst, src, 0);
1536 }
1537 }
1538
1539 void vextractf128_low(Address dst, XMMRegister src) {
1540 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1541 Assembler::vextractf32x4(dst, src, 0);
1542 } else {
1543 Assembler::vextractf128(dst, src, 0);
1544 }
1545 }
1546
1547 // 256bit copy to/from low 256 bits of 512bit (ZMM) vector registers
1548 void vinserti64x4_low(XMMRegister dst, XMMRegister src) {
1549 Assembler::vinserti64x4(dst, dst, src, 0);
1550 }
1551 void vinsertf64x4_low(XMMRegister dst, XMMRegister src) {
1552 Assembler::vinsertf64x4(dst, dst, src, 0);
1553 }
1554 void vextracti64x4_low(XMMRegister dst, XMMRegister src) {
1555 Assembler::vextracti64x4(dst, src, 0);
1556 }
1557 void vextractf64x4_low(XMMRegister dst, XMMRegister src) {
1558 Assembler::vextractf64x4(dst, src, 0);
1559 }
1560 void vextractf64x4_low(Address dst, XMMRegister src) {
1561 Assembler::vextractf64x4(dst, src, 0);
1562 }
1563 void vinsertf64x4_low(XMMRegister dst, Address src) {
1564 Assembler::vinsertf64x4(dst, dst, src, 0);
1565 }
1566
1567 // Carry-Less Multiplication Quadword
1568 void vpclmulldq(XMMRegister dst, XMMRegister nds, XMMRegister src) {
1569 // 0x00 - multiply lower 64 bits [0:63]
1570 Assembler::vpclmulqdq(dst, nds, src, 0x00);
1571 }
1572 void vpclmulhdq(XMMRegister dst, XMMRegister nds, XMMRegister src) {
1573 // 0x11 - multiply upper 64 bits [64:127]
1574 Assembler::vpclmulqdq(dst, nds, src, 0x11);
1575 }
1576 void vpclmullqhqdq(XMMRegister dst, XMMRegister nds, XMMRegister src) {
1577 // 0x10 - multiply nds[0:63] and src[64:127]
1578 Assembler::vpclmulqdq(dst, nds, src, 0x10);
1579 }
1580 void vpclmulhqlqdq(XMMRegister dst, XMMRegister nds, XMMRegister src) {
1581 //0x01 - multiply nds[64:127] and src[0:63]
1582 Assembler::vpclmulqdq(dst, nds, src, 0x01);
1583 }
1584
1585 void evpclmulldq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
1586 // 0x00 - multiply lower 64 bits [0:63]
1587 Assembler::evpclmulqdq(dst, nds, src, 0x00, vector_len);
1588 }
1589 void evpclmulhdq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
1590 // 0x11 - multiply upper 64 bits [64:127]
1591 Assembler::evpclmulqdq(dst, nds, src, 0x11, vector_len);
1592 }
1593
1594 // Data
1595
1596 void cmov32( Condition cc, Register dst, Address src);
1597 void cmov32( Condition cc, Register dst, Register src);
1598
1599 void cmov( Condition cc, Register dst, Register src) { cmovptr(cc, dst, src); }
1600
1601 void cmovptr(Condition cc, Register dst, Address src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmov32(cc, dst, src)); }
1602 void cmovptr(Condition cc, Register dst, Register src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmov32(cc, dst, src)); }
1603
1604 void movoop(Register dst, jobject obj);
1605 void movoop(Address dst, jobject obj);
1606
1607 void mov_metadata(Register dst, Metadata* obj);
1608 void mov_metadata(Address dst, Metadata* obj);
1609
1610 void movptr(ArrayAddress dst, Register src);
1611 // can this do an lea?
1612 void movptr(Register dst, ArrayAddress src);
1613
1614 void movptr(Register dst, Address src);
1615
1616 #ifdef _LP64
1617 void movptr(Register dst, AddressLiteral src, Register scratch=rscratch1);
1618 #else
1619 void movptr(Register dst, AddressLiteral src, Register scratch=noreg); // Scratch reg is ignored in 32-bit
1620 #endif
1621
1622 void movptr(Register dst, intptr_t src);
1623 void movptr(Register dst, Register src);
1624 void movptr(Address dst, intptr_t src);
1625
1626 void movptr(Address dst, Register src);
1627
1628 void movptr(Register dst, RegisterOrConstant src) {
1629 if (src.is_constant()) movptr(dst, src.as_constant());
1630 else movptr(dst, src.as_register());
1631 }
1632
1633 #ifdef _LP64
1634 // Generally the next two are only used for moving NULL
1635 // Although there are situations in initializing the mark word where
1636 // they could be used. They are dangerous.
1637
1638 // They only exist on LP64 so that int32_t and intptr_t are not the same
1639 // and we have ambiguous declarations.
1640
1641 void movptr(Address dst, int32_t imm32);
1642 void movptr(Register dst, int32_t imm32);
1643 #endif // _LP64
1644
1645 // to avoid hiding movl
1646 void mov32(AddressLiteral dst, Register src);
1647 void mov32(Register dst, AddressLiteral src);
1648
1649 // to avoid hiding movb
1650 void movbyte(ArrayAddress dst, int src);
1651
1652 // Import other mov() methods from the parent class or else
1653 // they will be hidden by the following overriding declaration.
1654 using Assembler::movdl;
1655 using Assembler::movq;
1656 void movdl(XMMRegister dst, AddressLiteral src);
1657 void movq(XMMRegister dst, AddressLiteral src);
1658
1659 // Can push value or effective address
1660 void pushptr(AddressLiteral src);
1661
1662 void pushptr(Address src) { LP64_ONLY(pushq(src)) NOT_LP64(pushl(src)); }
1663 void popptr(Address src) { LP64_ONLY(popq(src)) NOT_LP64(popl(src)); }
1664
1665 void pushoop(jobject obj);
1666 void pushklass(Metadata* obj);
1667
1668 // sign extend as need a l to ptr sized element
1669 void movl2ptr(Register dst, Address src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(movl(dst, src)); }
1670 void movl2ptr(Register dst, Register src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(if (dst != src) movl(dst, src)); }
1671
1672
1673 public:
1674 // C2 compiled method's prolog code.
1675 void verified_entry(int framesize, int stack_bang_size, bool fp_mode_24b, bool is_stub);
1676
1677 // clear memory of size 'cnt' qwords, starting at 'base';
1678 // if 'is_large' is set, do not try to produce short loop
1679 void clear_mem(Register base, Register cnt, Register rtmp, XMMRegister xtmp, bool is_large);
1680
1681 // clear memory of size 'cnt' qwords, starting at 'base' using XMM/YMM registers
1682 void xmm_clear_mem(Register base, Register cnt, XMMRegister xtmp);
1683
1684 // Fill primitive arrays
1685 void generate_fill(BasicType t, bool aligned,
1686 Register to, Register value, Register count,
1687 Register rtmp, XMMRegister xtmp);
1688
1689 void encode_iso_array(Register src, Register dst, Register len,
1690 XMMRegister tmp1, XMMRegister tmp2, XMMRegister tmp3,
1691 XMMRegister tmp4, Register tmp5, Register result);
1692
1693 #ifdef _LP64
1694 void add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2);
1695 void multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
1696 Register y, Register y_idx, Register z,
1697 Register carry, Register product,
1698 Register idx, Register kdx);
1699 void multiply_add_128_x_128(Register x_xstart, Register y, Register z,
1700 Register yz_idx, Register idx,
1701 Register carry, Register product, int offset);
1702 void multiply_128_x_128_bmi2_loop(Register y, Register z,
1703 Register carry, Register carry2,
1704 Register idx, Register jdx,
1705 Register yz_idx1, Register yz_idx2,
1706 Register tmp, Register tmp3, Register tmp4);
1707 void multiply_128_x_128_loop(Register x_xstart, Register y, Register z,
1708 Register yz_idx, Register idx, Register jdx,
1709 Register carry, Register product,
1710 Register carry2);
1711 void multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register zlen,
1712 Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5);
1713 void square_rshift(Register x, Register len, Register z, Register tmp1, Register tmp3,
1714 Register tmp4, Register tmp5, Register rdxReg, Register raxReg);
1715 void multiply_add_64_bmi2(Register sum, Register op1, Register op2, Register carry,
1716 Register tmp2);
1717 void multiply_add_64(Register sum, Register op1, Register op2, Register carry,
1718 Register rdxReg, Register raxReg);
1719 void add_one_64(Register z, Register zlen, Register carry, Register tmp1);
1720 void lshift_by_1(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2,
1721 Register tmp3, Register tmp4);
1722 void square_to_len(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2,
1723 Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg);
1724
1725 void mul_add_128_x_32_loop(Register out, Register in, Register offset, Register len, Register tmp1,
1726 Register tmp2, Register tmp3, Register tmp4, Register tmp5, Register rdxReg,
1727 Register raxReg);
1728 void mul_add(Register out, Register in, Register offset, Register len, Register k, Register tmp1,
1729 Register tmp2, Register tmp3, Register tmp4, Register tmp5, Register rdxReg,
1730 Register raxReg);
1731 void vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale,
1732 Register result, Register tmp1, Register tmp2,
1733 XMMRegister vec1, XMMRegister vec2, XMMRegister vec3);
1734 #endif
1735
1736 // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
1737 void update_byte_crc32(Register crc, Register val, Register table);
1738 void kernel_crc32(Register crc, Register buf, Register len, Register table, Register tmp);
1739 // CRC32C code for java.util.zip.CRC32C::updateBytes() intrinsic
1740 // Note on a naming convention:
1741 // Prefix w = register only used on a Westmere+ architecture
1742 // Prefix n = register only used on a Nehalem architecture
1743 #ifdef _LP64
1744 void crc32c_ipl_alg4(Register in_out, uint32_t n,
1745 Register tmp1, Register tmp2, Register tmp3);
1746 #else
1747 void crc32c_ipl_alg4(Register in_out, uint32_t n,
1748 Register tmp1, Register tmp2, Register tmp3,
1749 XMMRegister xtmp1, XMMRegister xtmp2);
1750 #endif
1751 void crc32c_pclmulqdq(XMMRegister w_xtmp1,
1752 Register in_out,
1753 uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported,
1754 XMMRegister w_xtmp2,
1755 Register tmp1,
1756 Register n_tmp2, Register n_tmp3);
1757 void 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,
1758 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
1759 Register tmp1, Register tmp2,
1760 Register n_tmp3);
1761 void 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,
1762 Register in_out1, Register in_out2, Register in_out3,
1763 Register tmp1, Register tmp2, Register tmp3,
1764 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
1765 Register tmp4, Register tmp5,
1766 Register n_tmp6);
1767 void crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2,
1768 Register tmp1, Register tmp2, Register tmp3,
1769 Register tmp4, Register tmp5, Register tmp6,
1770 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
1771 bool is_pclmulqdq_supported);
1772 // Fold 128-bit data chunk
1773 void fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset);
1774 void fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, XMMRegister xbuf);
1775 // Fold 8-bit data
1776 void fold_8bit_crc32(Register crc, Register table, Register tmp);
1777 void fold_8bit_crc32(XMMRegister crc, Register table, XMMRegister xtmp, Register tmp);
1778 void fold_128bit_crc32_avx512(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset);
1779
1780 // Compress char[] array to byte[].
1781 void char_array_compress(Register src, Register dst, Register len,
1782 XMMRegister tmp1, XMMRegister tmp2, XMMRegister tmp3,
1783 XMMRegister tmp4, Register tmp5, Register result);
1784
1785 // Inflate byte[] array to char[].
1786 void byte_array_inflate(Register src, Register dst, Register len,
1787 XMMRegister tmp1, Register tmp2);
1788
1789 #ifdef _LP64
1790 void convert_f2i(Register dst, XMMRegister src);
1791 void convert_d2i(Register dst, XMMRegister src);
1792 void convert_f2l(Register dst, XMMRegister src);
1793 void convert_d2l(Register dst, XMMRegister src);
1794
1795 void cache_wb(Address line);
1796 void cache_wbsync(bool is_pre);
1797 #endif // _LP64
1798 };
1799
1800 /**
1801 * class SkipIfEqual:
1802 *
1803 * Instantiating this class will result in assembly code being output that will
1804 * jump around any code emitted between the creation of the instance and it's
1805 * automatic destruction at the end of a scope block, depending on the value of
1806 * the flag passed to the constructor, which will be checked at run-time.
1807 */
1808 class SkipIfEqual {
1809 private:
1810 MacroAssembler* _masm;
1811 Label _label;
1812
1813 public:
1814 SkipIfEqual(MacroAssembler*, const bool* flag_addr, bool value);
1815 ~SkipIfEqual();
1816 };
1817
1818 #endif // CPU_X86_MACROASSEMBLER_X86_HPP