1 /*
2 * Copyright (c) 1997, 2017, 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_VM_MACROASSEMBLER_X86_HPP
26 #define CPU_X86_VM_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 protected:
42
43 Address as_Address(AddressLiteral adr);
44 Address as_Address(ArrayAddress adr);
45
46 // Support for VM calls
47 //
48 // This is the base routine called by the different versions of call_VM_leaf. The interpreter
49 // may customize this version by overriding it for its purposes (e.g., to save/restore
50 // additional registers when doing a VM call).
51
52 virtual void call_VM_leaf_base(
53 address entry_point, // the entry point
54 int number_of_arguments // the number of arguments to pop after the call
55 );
56
57 // This is the base routine called by the different versions of call_VM. The interpreter
58 // may customize this version by overriding it for its purposes (e.g., to save/restore
59 // additional registers when doing a VM call).
60 //
61 // If no java_thread register is specified (noreg) than rdi will be used instead. call_VM_base
62 // returns the register which contains the thread upon return. If a thread register has been
63 // specified, the return value will correspond to that register. If no last_java_sp is specified
64 // (noreg) than rsp will be used instead.
65 virtual void call_VM_base( // returns the register containing the thread upon return
66 Register oop_result, // where an oop-result ends up if any; use noreg otherwise
67 Register java_thread, // the thread if computed before ; use noreg otherwise
68 Register last_java_sp, // to set up last_Java_frame in stubs; use noreg otherwise
69 address entry_point, // the entry point
70 int number_of_arguments, // the number of arguments (w/o thread) to pop after the call
71 bool check_exceptions // whether to check for pending exceptions after return
72 );
73
74 // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code.
75 // The implementation is only non-empty for the InterpreterMacroAssembler,
76 // as only the interpreter handles PopFrame and ForceEarlyReturn requests.
77 virtual void check_and_handle_popframe(Register java_thread);
78 virtual void check_and_handle_earlyret(Register java_thread);
79
80 void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true);
81
82 // helpers for FPU flag access
83 // tmp is a temporary register, if none is available use noreg
84 void save_rax (Register tmp);
85 void restore_rax(Register tmp);
86
87 public:
88 MacroAssembler(CodeBuffer* code) : Assembler(code) {}
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
100 // Required platform-specific helpers for Label::patch_instructions.
101 // They _shadow_ the declarations in AbstractAssembler, which are undefined.
102 void pd_patch_instruction(address branch, address target) {
103 unsigned char op = branch[0];
104 assert(op == 0xE8 /* call */ ||
105 op == 0xE9 /* jmp */ ||
106 op == 0xEB /* short jmp */ ||
107 (op & 0xF0) == 0x70 /* short jcc */ ||
108 op == 0x0F && (branch[1] & 0xF0) == 0x80 /* jcc */ ||
109 op == 0xC7 && branch[1] == 0xF8 /* xbegin */,
110 "Invalid opcode at patch point");
111
112 if (op == 0xEB || (op & 0xF0) == 0x70) {
113 // short offset operators (jmp and jcc)
114 char* disp = (char*) &branch[1];
115 int imm8 = target - (address) &disp[1];
116 guarantee(this->is8bit(imm8), "Short forward jump exceeds 8-bit offset");
117 *disp = imm8;
118 } else {
119 int* disp = (int*) &branch[(op == 0x0F || op == 0xC7)? 2: 1];
120 int imm32 = target - (address) &disp[1];
121 *disp = imm32;
122 }
123 }
124
125 // The following 4 methods return the offset of the appropriate move instruction
126
127 // Support for fast byte/short loading with zero extension (depending on particular CPU)
128 int load_unsigned_byte(Register dst, Address src);
129 int load_unsigned_short(Register dst, Address src);
130
131 // Support for fast byte/short loading with sign extension (depending on particular CPU)
132 int load_signed_byte(Register dst, Address src);
133 int load_signed_short(Register dst, Address src);
134
135 // Support for sign-extension (hi:lo = extend_sign(lo))
136 void extend_sign(Register hi, Register lo);
137
138 // Load and store values by size and signed-ness
139 void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2 = noreg);
140 void store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2 = noreg);
141
142 // Support for inc/dec with optimal instruction selection depending on value
143
144 void increment(Register reg, int value = 1) { LP64_ONLY(incrementq(reg, value)) NOT_LP64(incrementl(reg, value)) ; }
145 void decrement(Register reg, int value = 1) { LP64_ONLY(decrementq(reg, value)) NOT_LP64(decrementl(reg, value)) ; }
146
147 void decrementl(Address dst, int value = 1);
148 void decrementl(Register reg, int value = 1);
149
150 void decrementq(Register reg, int value = 1);
151 void decrementq(Address dst, int value = 1);
152
153 void incrementl(Address dst, int value = 1);
154 void incrementl(Register reg, int value = 1);
155
156 void incrementq(Register reg, int value = 1);
157 void incrementq(Address dst, int value = 1);
158
159 // special instructions for EVEX
160 void setvectmask(Register dst, Register src);
161 void restorevectmask();
162
163 // Support optimal SSE move instructions.
164 void movflt(XMMRegister dst, XMMRegister src) {
165 if (UseXmmRegToRegMoveAll) { movaps(dst, src); return; }
166 else { movss (dst, src); return; }
167 }
168 void movflt(XMMRegister dst, Address src) { movss(dst, src); }
169 void movflt(XMMRegister dst, AddressLiteral src);
170 void movflt(Address dst, XMMRegister src) { movss(dst, src); }
171
172 void movdbl(XMMRegister dst, XMMRegister src) {
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 // Stores
297 void store_check(Register obj); // store check for obj - register is destroyed afterwards
298 void store_check(Register obj, Address dst); // same as above, dst is exact store location (reg. is destroyed)
299
300 void resolve_jobject(Register value, Register thread, Register tmp);
301
302 #if INCLUDE_ALL_GCS
303
304 void g1_write_barrier_pre(Register obj,
305 Register pre_val,
306 Register thread,
307 Register tmp,
308 bool tosca_live,
309 bool expand_call);
310
311 void g1_write_barrier_post(Register store_addr,
312 Register new_val,
313 Register thread,
314 Register tmp,
315 Register tmp2);
316
317 #endif // INCLUDE_ALL_GCS
318
319 // C 'boolean' to Java boolean: x == 0 ? 0 : 1
320 void c2bool(Register x);
321
322 // C++ bool manipulation
323
324 void movbool(Register dst, Address src);
325 void movbool(Address dst, bool boolconst);
326 void movbool(Address dst, Register src);
327 void testbool(Register dst);
328
329 void load_mirror(Register mirror, Register method);
330
331 // oop manipulations
332 void load_klass(Register dst, Register src);
333 void store_klass(Register dst, Register src);
334
335 void load_heap_oop(Register dst, Address src);
336 void load_heap_oop_not_null(Register dst, Address src);
337 void store_heap_oop(Address dst, Register src);
338 void cmp_heap_oop(Register src1, Address src2, Register tmp = noreg);
339
340 // Used for storing NULL. All other oop constants should be
341 // stored using routines that take a jobject.
342 void store_heap_oop_null(Address dst);
343
344 void load_prototype_header(Register dst, Register src);
345
346 #ifdef _LP64
347 void store_klass_gap(Register dst, Register src);
348
349 // This dummy is to prevent a call to store_heap_oop from
350 // converting a zero (like NULL) into a Register by giving
351 // the compiler two choices it can't resolve
352
353 void store_heap_oop(Address dst, void* dummy);
354
355 void encode_heap_oop(Register r);
356 void decode_heap_oop(Register r);
357 void encode_heap_oop_not_null(Register r);
358 void decode_heap_oop_not_null(Register r);
359 void encode_heap_oop_not_null(Register dst, Register src);
360 void decode_heap_oop_not_null(Register dst, Register src);
361
362 void set_narrow_oop(Register dst, jobject obj);
363 void set_narrow_oop(Address dst, jobject obj);
364 void cmp_narrow_oop(Register dst, jobject obj);
365 void cmp_narrow_oop(Address dst, jobject obj);
366
367 void encode_klass_not_null(Register r);
368 void decode_klass_not_null(Register r);
369 void encode_klass_not_null(Register dst, Register src);
370 void decode_klass_not_null(Register dst, Register src);
371 void set_narrow_klass(Register dst, Klass* k);
372 void set_narrow_klass(Address dst, Klass* k);
373 void cmp_narrow_klass(Register dst, Klass* k);
374 void cmp_narrow_klass(Address dst, Klass* k);
375
376 // Returns the byte size of the instructions generated by decode_klass_not_null()
377 // when compressed klass pointers are being used.
378 static int instr_size_for_decode_klass_not_null();
379
380 // if heap base register is used - reinit it with the correct value
381 void reinit_heapbase();
382
383 DEBUG_ONLY(void verify_heapbase(const char* msg);)
384
385 #endif // _LP64
386
387 // Int division/remainder for Java
388 // (as idivl, but checks for special case as described in JVM spec.)
389 // returns idivl instruction offset for implicit exception handling
390 int corrected_idivl(Register reg);
391
392 // Long division/remainder for Java
393 // (as idivq, but checks for special case as described in JVM spec.)
394 // returns idivq instruction offset for implicit exception handling
395 int corrected_idivq(Register reg);
396
397 void int3();
398
399 // Long operation macros for a 32bit cpu
400 // Long negation for Java
401 void lneg(Register hi, Register lo);
402
403 // Long multiplication for Java
404 // (destroys contents of eax, ebx, ecx and edx)
405 void lmul(int x_rsp_offset, int y_rsp_offset); // rdx:rax = x * y
406
407 // Long shifts for Java
408 // (semantics as described in JVM spec.)
409 void lshl(Register hi, Register lo); // hi:lo << (rcx & 0x3f)
410 void lshr(Register hi, Register lo, bool sign_extension = false); // hi:lo >> (rcx & 0x3f)
411
412 // Long compare for Java
413 // (semantics as described in JVM spec.)
414 void lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo); // x_hi = lcmp(x, y)
415
416
417 // misc
418
419 // Sign extension
420 void sign_extend_short(Register reg);
421 void sign_extend_byte(Register reg);
422
423 // Division by power of 2, rounding towards 0
424 void division_with_shift(Register reg, int shift_value);
425
426 // Compares the top-most stack entries on the FPU stack and sets the eflags as follows:
427 //
428 // CF (corresponds to C0) if x < y
429 // PF (corresponds to C2) if unordered
430 // ZF (corresponds to C3) if x = y
431 //
432 // The arguments are in reversed order on the stack (i.e., top of stack is first argument).
433 // tmp is a temporary register, if none is available use noreg (only matters for non-P6 code)
434 void fcmp(Register tmp);
435 // Variant of the above which allows y to be further down the stack
436 // and which only pops x and y if specified. If pop_right is
437 // specified then pop_left must also be specified.
438 void fcmp(Register tmp, int index, bool pop_left, bool pop_right);
439
440 // Floating-point comparison for Java
441 // Compares the top-most stack entries on the FPU stack and stores the result in dst.
442 // The arguments are in reversed order on the stack (i.e., top of stack is first argument).
443 // (semantics as described in JVM spec.)
444 void fcmp2int(Register dst, bool unordered_is_less);
445 // Variant of the above which allows y to be further down the stack
446 // and which only pops x and y if specified. If pop_right is
447 // specified then pop_left must also be specified.
448 void fcmp2int(Register dst, bool unordered_is_less, int index, bool pop_left, bool pop_right);
449
450 // Floating-point remainder for Java (ST0 = ST0 fremr ST1, ST1 is empty afterwards)
451 // tmp is a temporary register, if none is available use noreg
452 void fremr(Register tmp);
453
454 // dst = c = a * b + c
455 void fmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c);
456 void fmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c);
457
458
459 // same as fcmp2int, but using SSE2
460 void cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less);
461 void cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less);
462
463 // branch to L if FPU flag C2 is set/not set
464 // tmp is a temporary register, if none is available use noreg
465 void jC2 (Register tmp, Label& L);
466 void jnC2(Register tmp, Label& L);
467
468 // Pop ST (ffree & fincstp combined)
469 void fpop();
470
471 // Load float value from 'address'. If UseSSE >= 1, the value is loaded into
472 // register xmm0. Otherwise, the value is loaded onto the FPU stack.
473 void load_float(Address src);
474
475 // Store float value to 'address'. If UseSSE >= 1, the value is stored
476 // from register xmm0. Otherwise, the value is stored from the FPU stack.
477 void store_float(Address dst);
478
479 // Load double value from 'address'. If UseSSE >= 2, the value is loaded into
480 // register xmm0. Otherwise, the value is loaded onto the FPU stack.
481 void load_double(Address src);
482
483 // Store double value to 'address'. If UseSSE >= 2, the value is stored
484 // from register xmm0. Otherwise, the value is stored from the FPU stack.
485 void store_double(Address dst);
486
487 // pushes double TOS element of FPU stack on CPU stack; pops from FPU stack
488 void push_fTOS();
489
490 // pops double TOS element from CPU stack and pushes on FPU stack
491 void pop_fTOS();
492
493 void empty_FPU_stack();
494
495 void push_IU_state();
496 void pop_IU_state();
497
498 void push_FPU_state();
499 void pop_FPU_state();
500
501 void push_CPU_state();
502 void pop_CPU_state();
503
504 // Round up to a power of two
505 void round_to(Register reg, int modulus);
506
507 // Callee saved registers handling
508 void push_callee_saved_registers();
509 void pop_callee_saved_registers();
510
511 // allocation
512 void eden_allocate(
513 Register obj, // result: pointer to object after successful allocation
514 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise
515 int con_size_in_bytes, // object size in bytes if known at compile time
516 Register t1, // temp register
517 Label& slow_case // continuation point if fast allocation fails
518 );
519 void tlab_allocate(
520 Register obj, // result: pointer to object after successful allocation
521 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise
522 int con_size_in_bytes, // object size in bytes if known at compile time
523 Register t1, // temp register
524 Register t2, // temp register
525 Label& slow_case // continuation point if fast allocation fails
526 );
527 Register tlab_refill(Label& retry_tlab, Label& try_eden, Label& slow_case); // returns TLS address
528 void zero_memory(Register address, Register length_in_bytes, int offset_in_bytes, Register temp);
529
530 void incr_allocated_bytes(Register thread,
531 Register var_size_in_bytes, int con_size_in_bytes,
532 Register t1 = noreg);
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
542 // virtual method calling
543 void lookup_virtual_method(Register recv_klass,
544 RegisterOrConstant vtable_index,
545 Register method_result);
546
547 // Test sub_klass against super_klass, with fast and slow paths.
548
549 // The fast path produces a tri-state answer: yes / no / maybe-slow.
550 // One of the three labels can be NULL, meaning take the fall-through.
551 // If super_check_offset is -1, the value is loaded up from super_klass.
552 // No registers are killed, except temp_reg.
553 void check_klass_subtype_fast_path(Register sub_klass,
554 Register super_klass,
555 Register temp_reg,
556 Label* L_success,
557 Label* L_failure,
558 Label* L_slow_path,
559 RegisterOrConstant super_check_offset = RegisterOrConstant(-1));
560
561 // The rest of the type check; must be wired to a corresponding fast path.
562 // It does not repeat the fast path logic, so don't use it standalone.
563 // The temp_reg and temp2_reg can be noreg, if no temps are available.
564 // Updates the sub's secondary super cache as necessary.
565 // If set_cond_codes, condition codes will be Z on success, NZ on failure.
566 void check_klass_subtype_slow_path(Register sub_klass,
567 Register super_klass,
568 Register temp_reg,
569 Register temp2_reg,
570 Label* L_success,
571 Label* L_failure,
572 bool set_cond_codes = false);
573
574 // Simplified, combined version, good for typical uses.
575 // Falls through on failure.
576 void check_klass_subtype(Register sub_klass,
577 Register super_klass,
578 Register temp_reg,
579 Label& L_success);
580
581 // method handles (JSR 292)
582 Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0);
583
584 //----
585 void set_word_if_not_zero(Register reg); // sets reg to 1 if not zero, otherwise 0
586
587 // Debugging
588
589 // only if +VerifyOops
590 // TODO: Make these macros with file and line like sparc version!
591 void verify_oop(Register reg, const char* s = "broken oop");
592 void verify_oop_addr(Address addr, const char * s = "broken oop addr");
593
594 // TODO: verify method and klass metadata (compare against vptr?)
595 void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
596 void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line){}
597
598 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
599 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
600
601 // only if +VerifyFPU
602 void verify_FPU(int stack_depth, const char* s = "illegal FPU state");
603
604 // Verify or restore cpu control state after JNI call
605 void restore_cpu_control_state_after_jni();
606
607 // prints msg, dumps registers and stops execution
608 void stop(const char* msg);
609
610 // prints msg and continues
611 void warn(const char* msg);
612
613 // dumps registers and other state
614 void print_state();
615
616 static void debug32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip, char* msg);
617 static void debug64(char* msg, int64_t pc, int64_t regs[]);
618 static void print_state32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip);
619 static void print_state64(int64_t pc, int64_t regs[]);
620
621 void os_breakpoint();
622
623 void untested() { stop("untested"); }
624
625 void unimplemented(const char* what = "") { char* b = new char[1024]; jio_snprintf(b, 1024, "unimplemented: %s", what); stop(b); }
626
627 void should_not_reach_here() { stop("should not reach here"); }
628
629 void print_CPU_state();
630
631 // Stack overflow checking
632 void bang_stack_with_offset(int offset) {
633 // stack grows down, caller passes positive offset
634 assert(offset > 0, "must bang with negative offset");
635 movl(Address(rsp, (-offset)), rax);
636 }
637
638 // Writes to stack successive pages until offset reached to check for
639 // stack overflow + shadow pages. Also, clobbers tmp
640 void bang_stack_size(Register size, Register tmp);
641
642 // Check for reserved stack access in method being exited (for JIT)
643 void reserved_stack_check();
644
645 virtual RegisterOrConstant delayed_value_impl(intptr_t* delayed_value_addr,
646 Register tmp,
647 int offset);
648
649 // Support for serializing memory accesses between threads
650 void serialize_memory(Register thread, Register tmp);
651
652 void verify_tlab();
653
654 // Biased locking support
655 // lock_reg and obj_reg must be loaded up with the appropriate values.
656 // swap_reg must be rax, and is killed.
657 // tmp_reg is optional. If it is supplied (i.e., != noreg) it will
658 // be killed; if not supplied, push/pop will be used internally to
659 // allocate a temporary (inefficient, avoid if possible).
660 // Optional slow case is for implementations (interpreter and C1) which branch to
661 // slow case directly. Leaves condition codes set for C2's Fast_Lock node.
662 // Returns offset of first potentially-faulting instruction for null
663 // check info (currently consumed only by C1). If
664 // swap_reg_contains_mark is true then returns -1 as it is assumed
665 // the calling code has already passed any potential faults.
666 int biased_locking_enter(Register lock_reg, Register obj_reg,
667 Register swap_reg, Register tmp_reg,
668 bool swap_reg_contains_mark,
669 Label& done, Label* slow_case = NULL,
670 BiasedLockingCounters* counters = NULL);
671 void biased_locking_exit (Register obj_reg, Register temp_reg, Label& done);
672 #ifdef COMPILER2
673 // Code used by cmpFastLock and cmpFastUnlock mach instructions in .ad file.
674 // See full desription in macroAssembler_x86.cpp.
675 void fast_lock(Register obj, Register box, Register tmp,
676 Register scr, Register cx1, Register cx2,
677 BiasedLockingCounters* counters,
678 RTMLockingCounters* rtm_counters,
679 RTMLockingCounters* stack_rtm_counters,
680 Metadata* method_data,
681 bool use_rtm, bool profile_rtm);
682 void fast_unlock(Register obj, Register box, Register tmp, bool use_rtm);
683 #if INCLUDE_RTM_OPT
684 void rtm_counters_update(Register abort_status, Register rtm_counters);
685 void branch_on_random_using_rdtsc(Register tmp, Register scr, int count, Label& brLabel);
686 void rtm_abort_ratio_calculation(Register tmp, Register rtm_counters_reg,
687 RTMLockingCounters* rtm_counters,
688 Metadata* method_data);
689 void rtm_profiling(Register abort_status_Reg, Register rtm_counters_Reg,
690 RTMLockingCounters* rtm_counters, Metadata* method_data, bool profile_rtm);
691 void rtm_retry_lock_on_abort(Register retry_count, Register abort_status, Label& retryLabel);
692 void rtm_retry_lock_on_busy(Register retry_count, Register box, Register tmp, Register scr, Label& retryLabel);
693 void rtm_stack_locking(Register obj, Register tmp, Register scr,
694 Register retry_on_abort_count,
695 RTMLockingCounters* stack_rtm_counters,
696 Metadata* method_data, bool profile_rtm,
697 Label& DONE_LABEL, Label& IsInflated);
698 void rtm_inflated_locking(Register obj, Register box, Register tmp,
699 Register scr, Register retry_on_busy_count,
700 Register retry_on_abort_count,
701 RTMLockingCounters* rtm_counters,
702 Metadata* method_data, bool profile_rtm,
703 Label& DONE_LABEL);
704 #endif
705 #endif
706
707 Condition negate_condition(Condition cond);
708
709 // Instructions that use AddressLiteral operands. These instruction can handle 32bit/64bit
710 // operands. In general the names are modified to avoid hiding the instruction in Assembler
711 // so that we don't need to implement all the varieties in the Assembler with trivial wrappers
712 // here in MacroAssembler. The major exception to this rule is call
713
714 // Arithmetics
715
716
717 void addptr(Address dst, int32_t src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)) ; }
718 void addptr(Address dst, Register src);
719
720 void addptr(Register dst, Address src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)); }
721 void addptr(Register dst, int32_t src);
722 void addptr(Register dst, Register src);
723 void addptr(Register dst, RegisterOrConstant src) {
724 if (src.is_constant()) addptr(dst, (int) src.as_constant());
725 else addptr(dst, src.as_register());
726 }
727
728 void andptr(Register dst, int32_t src);
729 void andptr(Register src1, Register src2) { LP64_ONLY(andq(src1, src2)) NOT_LP64(andl(src1, src2)) ; }
730
731 void cmp8(AddressLiteral src1, int imm);
732
733 // renamed to drag out the casting of address to int32_t/intptr_t
734 void cmp32(Register src1, int32_t imm);
735
736 void cmp32(AddressLiteral src1, int32_t imm);
737 // compare reg - mem, or reg - &mem
738 void cmp32(Register src1, AddressLiteral src2);
739
740 void cmp32(Register src1, Address src2);
741
742 #ifndef _LP64
743 void cmpklass(Address dst, Metadata* obj);
744 void cmpklass(Register dst, Metadata* obj);
745 void cmpoop(Address dst, jobject obj);
746 void cmpoop(Register dst, jobject obj);
747 #endif // _LP64
748
749 // NOTE src2 must be the lval. This is NOT an mem-mem compare
750 void cmpptr(Address src1, AddressLiteral src2);
751
752 void cmpptr(Register src1, AddressLiteral src2);
753
754 void cmpptr(Register src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
755 void cmpptr(Register src1, Address src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
756 // void cmpptr(Address src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
757
758 void cmpptr(Register src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
759 void cmpptr(Address src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
760
761 // cmp64 to avoild hiding cmpq
762 void cmp64(Register src1, AddressLiteral src);
763
764 void cmpxchgptr(Register reg, Address adr);
765
766 void locked_cmpxchgptr(Register reg, AddressLiteral adr);
767
768
769 void imulptr(Register dst, Register src) { LP64_ONLY(imulq(dst, src)) NOT_LP64(imull(dst, src)); }
770 void imulptr(Register dst, Register src, int imm32) { LP64_ONLY(imulq(dst, src, imm32)) NOT_LP64(imull(dst, src, imm32)); }
771
772
773 void negptr(Register dst) { LP64_ONLY(negq(dst)) NOT_LP64(negl(dst)); }
774
775 void notptr(Register dst) { LP64_ONLY(notq(dst)) NOT_LP64(notl(dst)); }
776
777 void shlptr(Register dst, int32_t shift);
778 void shlptr(Register dst) { LP64_ONLY(shlq(dst)) NOT_LP64(shll(dst)); }
779
780 void shrptr(Register dst, int32_t shift);
781 void shrptr(Register dst) { LP64_ONLY(shrq(dst)) NOT_LP64(shrl(dst)); }
782
783 void sarptr(Register dst) { LP64_ONLY(sarq(dst)) NOT_LP64(sarl(dst)); }
784 void sarptr(Register dst, int32_t src) { LP64_ONLY(sarq(dst, src)) NOT_LP64(sarl(dst, src)); }
785
786 void subptr(Address dst, int32_t src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); }
787
788 void subptr(Register dst, Address src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); }
789 void subptr(Register dst, int32_t src);
790 // Force generation of a 4 byte immediate value even if it fits into 8bit
791 void subptr_imm32(Register dst, int32_t src);
792 void subptr(Register dst, Register src);
793 void subptr(Register dst, RegisterOrConstant src) {
794 if (src.is_constant()) subptr(dst, (int) src.as_constant());
795 else subptr(dst, src.as_register());
796 }
797
798 void sbbptr(Address dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); }
799 void sbbptr(Register dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); }
800
801 void xchgptr(Register src1, Register src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; }
802 void xchgptr(Register src1, Address src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; }
803
804 void xaddptr(Address src1, Register src2) { LP64_ONLY(xaddq(src1, src2)) NOT_LP64(xaddl(src1, src2)) ; }
805
806
807
808 // Helper functions for statistics gathering.
809 // Conditionally (atomically, on MPs) increments passed counter address, preserving condition codes.
810 void cond_inc32(Condition cond, AddressLiteral counter_addr);
811 // Unconditional atomic increment.
812 void atomic_incl(Address counter_addr);
813 void atomic_incl(AddressLiteral counter_addr, Register scr = rscratch1);
814 #ifdef _LP64
815 void atomic_incq(Address counter_addr);
816 void atomic_incq(AddressLiteral counter_addr, Register scr = rscratch1);
817 #endif
818 void atomic_incptr(AddressLiteral counter_addr, Register scr = rscratch1) { LP64_ONLY(atomic_incq(counter_addr, scr)) NOT_LP64(atomic_incl(counter_addr, scr)) ; }
819 void atomic_incptr(Address counter_addr) { LP64_ONLY(atomic_incq(counter_addr)) NOT_LP64(atomic_incl(counter_addr)) ; }
820
821 void lea(Register dst, AddressLiteral adr);
822 void lea(Address dst, AddressLiteral adr);
823 void lea(Register dst, Address adr) { Assembler::lea(dst, adr); }
824
825 void leal32(Register dst, Address src) { leal(dst, src); }
826
827 // Import other testl() methods from the parent class or else
828 // they will be hidden by the following overriding declaration.
829 using Assembler::testl;
830 void testl(Register dst, AddressLiteral src);
831
832 void orptr(Register dst, Address src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); }
833 void orptr(Register dst, Register src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); }
834 void orptr(Register dst, int32_t src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); }
835 void orptr(Address dst, int32_t imm32) { LP64_ONLY(orq(dst, imm32)) NOT_LP64(orl(dst, imm32)); }
836
837 void testptr(Register src, int32_t imm32) { LP64_ONLY(testq(src, imm32)) NOT_LP64(testl(src, imm32)); }
838 void testptr(Register src1, Register src2);
839
840 void xorptr(Register dst, Register src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); }
841 void xorptr(Register dst, Address src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); }
842
843 // Calls
844
845 void call(Label& L, relocInfo::relocType rtype);
846 void call(Register entry);
847
848 // NOTE: this call transfers to the effective address of entry NOT
849 // the address contained by entry. This is because this is more natural
850 // for jumps/calls.
851 void call(AddressLiteral entry);
852
853 // Emit the CompiledIC call idiom
854 void ic_call(address entry, jint method_index = 0);
855
856 // Jumps
857
858 // NOTE: these jumps tranfer to the effective address of dst NOT
859 // the address contained by dst. This is because this is more natural
860 // for jumps/calls.
861 void jump(AddressLiteral dst);
862 void jump_cc(Condition cc, AddressLiteral dst);
863
864 // 32bit can do a case table jump in one instruction but we no longer allow the base
865 // to be installed in the Address class. This jump will tranfers to the address
866 // contained in the location described by entry (not the address of entry)
867 void jump(ArrayAddress entry);
868
869 // Floating
870
871 void andpd(XMMRegister dst, Address src) { Assembler::andpd(dst, src); }
872 void andpd(XMMRegister dst, AddressLiteral src);
873 void andpd(XMMRegister dst, XMMRegister src) { Assembler::andpd(dst, src); }
874
875 void andps(XMMRegister dst, XMMRegister src) { Assembler::andps(dst, src); }
876 void andps(XMMRegister dst, Address src) { Assembler::andps(dst, src); }
877 void andps(XMMRegister dst, AddressLiteral src);
878
879 void comiss(XMMRegister dst, XMMRegister src) { Assembler::comiss(dst, src); }
880 void comiss(XMMRegister dst, Address src) { Assembler::comiss(dst, src); }
881 void comiss(XMMRegister dst, AddressLiteral src);
882
883 void comisd(XMMRegister dst, XMMRegister src) { Assembler::comisd(dst, src); }
884 void comisd(XMMRegister dst, Address src) { Assembler::comisd(dst, src); }
885 void comisd(XMMRegister dst, AddressLiteral src);
886
887 void fadd_s(Address src) { Assembler::fadd_s(src); }
888 void fadd_s(AddressLiteral src) { Assembler::fadd_s(as_Address(src)); }
889
890 void fldcw(Address src) { Assembler::fldcw(src); }
891 void fldcw(AddressLiteral src);
892
893 void fld_s(int index) { Assembler::fld_s(index); }
894 void fld_s(Address src) { Assembler::fld_s(src); }
895 void fld_s(AddressLiteral src);
896
897 void fld_d(Address src) { Assembler::fld_d(src); }
898 void fld_d(AddressLiteral src);
899
900 void fld_x(Address src) { Assembler::fld_x(src); }
901 void fld_x(AddressLiteral src);
902
903 void fmul_s(Address src) { Assembler::fmul_s(src); }
904 void fmul_s(AddressLiteral src) { Assembler::fmul_s(as_Address(src)); }
905
906 void ldmxcsr(Address src) { Assembler::ldmxcsr(src); }
907 void ldmxcsr(AddressLiteral src);
908
909 #ifdef _LP64
910 private:
911 void sha256_AVX2_one_round_compute(
912 Register reg_old_h,
913 Register reg_a,
914 Register reg_b,
915 Register reg_c,
916 Register reg_d,
917 Register reg_e,
918 Register reg_f,
919 Register reg_g,
920 Register reg_h,
921 int iter);
922 void sha256_AVX2_four_rounds_compute_first(int start);
923 void sha256_AVX2_four_rounds_compute_last(int start);
924 void sha256_AVX2_one_round_and_sched(
925 XMMRegister xmm_0, /* == ymm4 on 0, 1, 2, 3 iterations, then rotate 4 registers left on 4, 8, 12 iterations */
926 XMMRegister xmm_1, /* ymm5 */ /* full cycle is 16 iterations */
927 XMMRegister xmm_2, /* ymm6 */
928 XMMRegister xmm_3, /* ymm7 */
929 Register reg_a, /* == eax on 0 iteration, then rotate 8 register right on each next iteration */
930 Register reg_b, /* ebx */ /* full cycle is 8 iterations */
931 Register reg_c, /* edi */
932 Register reg_d, /* esi */
933 Register reg_e, /* r8d */
934 Register reg_f, /* r9d */
935 Register reg_g, /* r10d */
936 Register reg_h, /* r11d */
937 int iter);
938
939 void addm(int disp, Register r1, Register r2);
940
941 public:
942 void sha256_AVX2(XMMRegister msg, XMMRegister state0, XMMRegister state1, XMMRegister msgtmp0,
943 XMMRegister msgtmp1, XMMRegister msgtmp2, XMMRegister msgtmp3, XMMRegister msgtmp4,
944 Register buf, Register state, Register ofs, Register limit, Register rsp,
945 bool multi_block, XMMRegister shuf_mask);
946 #endif
947
948 #ifdef _LP64
949 private:
950 void sha512_AVX2_one_round_compute(Register old_h, Register a, Register b, Register c, Register d,
951 Register e, Register f, Register g, Register h, int iteration);
952
953 void sha512_AVX2_one_round_and_schedule(XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
954 Register a, Register b, Register c, Register d, Register e, Register f,
955 Register g, Register h, int iteration);
956
957 void addmq(int disp, Register r1, Register r2);
958 public:
959 void sha512_AVX2(XMMRegister msg, XMMRegister state0, XMMRegister state1, XMMRegister msgtmp0,
960 XMMRegister msgtmp1, XMMRegister msgtmp2, XMMRegister msgtmp3, XMMRegister msgtmp4,
961 Register buf, Register state, Register ofs, Register limit, Register rsp, bool multi_block,
962 XMMRegister shuf_mask);
963 #endif
964
965 void fast_sha1(XMMRegister abcd, XMMRegister e0, XMMRegister e1, XMMRegister msg0,
966 XMMRegister msg1, XMMRegister msg2, XMMRegister msg3, XMMRegister shuf_mask,
967 Register buf, Register state, Register ofs, Register limit, Register rsp,
968 bool multi_block);
969
970 #ifdef _LP64
971 void fast_sha256(XMMRegister msg, XMMRegister state0, XMMRegister state1, XMMRegister msgtmp0,
972 XMMRegister msgtmp1, XMMRegister msgtmp2, XMMRegister msgtmp3, XMMRegister msgtmp4,
973 Register buf, Register state, Register ofs, Register limit, Register rsp,
974 bool multi_block, XMMRegister shuf_mask);
975 #else
976 void fast_sha256(XMMRegister msg, XMMRegister state0, XMMRegister state1, XMMRegister msgtmp0,
977 XMMRegister msgtmp1, XMMRegister msgtmp2, XMMRegister msgtmp3, XMMRegister msgtmp4,
978 Register buf, Register state, Register ofs, Register limit, Register rsp,
979 bool multi_block);
980 #endif
981
982 void fast_exp(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
983 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
984 Register rax, Register rcx, Register rdx, Register tmp);
985
986 #ifdef _LP64
987 void fast_log(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
988 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
989 Register rax, Register rcx, Register rdx, Register tmp1, Register tmp2);
990
991 void fast_log10(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
992 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
993 Register rax, Register rcx, Register rdx, Register r11);
994
995 void fast_pow(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, XMMRegister xmm4,
996 XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, Register rax, Register rcx,
997 Register rdx, Register tmp1, Register tmp2, Register tmp3, Register tmp4);
998
999 void fast_sin(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1000 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1001 Register rax, Register rbx, Register rcx, Register rdx, Register tmp1, Register tmp2,
1002 Register tmp3, Register tmp4);
1003
1004 void fast_cos(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1005 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1006 Register rax, Register rcx, Register rdx, Register tmp1,
1007 Register tmp2, Register tmp3, Register tmp4);
1008 void fast_tan(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1009 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1010 Register rax, Register rcx, Register rdx, Register tmp1,
1011 Register tmp2, Register tmp3, Register tmp4);
1012 #else
1013 void fast_log(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1014 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1015 Register rax, Register rcx, Register rdx, Register tmp1);
1016
1017 void fast_log10(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1018 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1019 Register rax, Register rcx, Register rdx, Register tmp);
1020
1021 void fast_pow(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, XMMRegister xmm4,
1022 XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, Register rax, Register rcx,
1023 Register rdx, Register tmp);
1024
1025 void fast_sin(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1026 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1027 Register rax, Register rbx, Register rdx);
1028
1029 void fast_cos(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1030 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1031 Register rax, Register rcx, Register rdx, Register tmp);
1032
1033 void libm_sincos_huge(XMMRegister xmm0, XMMRegister xmm1, Register eax, Register ecx,
1034 Register edx, Register ebx, Register esi, Register edi,
1035 Register ebp, Register esp);
1036
1037 void libm_reduce_pi04l(Register eax, Register ecx, Register edx, Register ebx,
1038 Register esi, Register edi, Register ebp, Register esp);
1039
1040 void libm_tancot_huge(XMMRegister xmm0, XMMRegister xmm1, Register eax, Register ecx,
1041 Register edx, Register ebx, Register esi, Register edi,
1042 Register ebp, Register esp);
1043
1044 void fast_tan(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1045 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1046 Register rax, Register rcx, Register rdx, Register tmp);
1047 #endif
1048
1049 void increase_precision();
1050 void restore_precision();
1051
1052 private:
1053
1054 // these are private because users should be doing movflt/movdbl
1055
1056 void movss(Address dst, XMMRegister src) { Assembler::movss(dst, src); }
1057 void movss(XMMRegister dst, XMMRegister src) { Assembler::movss(dst, src); }
1058 void movss(XMMRegister dst, Address src) { Assembler::movss(dst, src); }
1059 void movss(XMMRegister dst, AddressLiteral src);
1060
1061 void movlpd(XMMRegister dst, Address src) {Assembler::movlpd(dst, src); }
1062 void movlpd(XMMRegister dst, AddressLiteral src);
1063
1064 public:
1065
1066 void addsd(XMMRegister dst, XMMRegister src) { Assembler::addsd(dst, src); }
1067 void addsd(XMMRegister dst, Address src) { Assembler::addsd(dst, src); }
1068 void addsd(XMMRegister dst, AddressLiteral src);
1069
1070 void addss(XMMRegister dst, XMMRegister src) { Assembler::addss(dst, src); }
1071 void addss(XMMRegister dst, Address src) { Assembler::addss(dst, src); }
1072 void addss(XMMRegister dst, AddressLiteral src);
1073
1074 void addpd(XMMRegister dst, XMMRegister src) { Assembler::addpd(dst, src); }
1075 void addpd(XMMRegister dst, Address src) { Assembler::addpd(dst, src); }
1076 void addpd(XMMRegister dst, AddressLiteral src);
1077
1078 void divsd(XMMRegister dst, XMMRegister src) { Assembler::divsd(dst, src); }
1079 void divsd(XMMRegister dst, Address src) { Assembler::divsd(dst, src); }
1080 void divsd(XMMRegister dst, AddressLiteral src);
1081
1082 void divss(XMMRegister dst, XMMRegister src) { Assembler::divss(dst, src); }
1083 void divss(XMMRegister dst, Address src) { Assembler::divss(dst, src); }
1084 void divss(XMMRegister dst, AddressLiteral src);
1085
1086 // Move Unaligned Double Quadword
1087 void movdqu(Address dst, XMMRegister src);
1088 void movdqu(XMMRegister dst, Address src);
1089 void movdqu(XMMRegister dst, XMMRegister src);
1090 void movdqu(XMMRegister dst, AddressLiteral src, Register scratchReg = rscratch1);
1091 // AVX Unaligned forms
1092 void vmovdqu(Address dst, XMMRegister src);
1093 void vmovdqu(XMMRegister dst, Address src);
1094 void vmovdqu(XMMRegister dst, XMMRegister src);
1095 void vmovdqu(XMMRegister dst, AddressLiteral src);
1096
1097 // Move Aligned Double Quadword
1098 void movdqa(XMMRegister dst, Address src) { Assembler::movdqa(dst, src); }
1099 void movdqa(XMMRegister dst, XMMRegister src) { Assembler::movdqa(dst, src); }
1100 void movdqa(XMMRegister dst, AddressLiteral src);
1101
1102 void movsd(XMMRegister dst, XMMRegister src) { Assembler::movsd(dst, src); }
1103 void movsd(Address dst, XMMRegister src) { Assembler::movsd(dst, src); }
1104 void movsd(XMMRegister dst, Address src) { Assembler::movsd(dst, src); }
1105 void movsd(XMMRegister dst, AddressLiteral src);
1106
1107 void mulpd(XMMRegister dst, XMMRegister src) { Assembler::mulpd(dst, src); }
1108 void mulpd(XMMRegister dst, Address src) { Assembler::mulpd(dst, src); }
1109 void mulpd(XMMRegister dst, AddressLiteral src);
1110
1111 void mulsd(XMMRegister dst, XMMRegister src) { Assembler::mulsd(dst, src); }
1112 void mulsd(XMMRegister dst, Address src) { Assembler::mulsd(dst, src); }
1113 void mulsd(XMMRegister dst, AddressLiteral src);
1114
1115 void mulss(XMMRegister dst, XMMRegister src) { Assembler::mulss(dst, src); }
1116 void mulss(XMMRegister dst, Address src) { Assembler::mulss(dst, src); }
1117 void mulss(XMMRegister dst, AddressLiteral src);
1118
1119 // Carry-Less Multiplication Quadword
1120 void pclmulldq(XMMRegister dst, XMMRegister src) {
1121 // 0x00 - multiply lower 64 bits [0:63]
1122 Assembler::pclmulqdq(dst, src, 0x00);
1123 }
1124 void pclmulhdq(XMMRegister dst, XMMRegister src) {
1125 // 0x11 - multiply upper 64 bits [64:127]
1126 Assembler::pclmulqdq(dst, src, 0x11);
1127 }
1128
1129 void pcmpeqb(XMMRegister dst, XMMRegister src);
1130 void pcmpeqw(XMMRegister dst, XMMRegister src);
1131
1132 void pcmpestri(XMMRegister dst, Address src, int imm8);
1133 void pcmpestri(XMMRegister dst, XMMRegister src, int imm8);
1134
1135 void pmovzxbw(XMMRegister dst, XMMRegister src);
1136 void pmovzxbw(XMMRegister dst, Address src);
1137
1138 void pmovmskb(Register dst, XMMRegister src);
1139
1140 void ptest(XMMRegister dst, XMMRegister src);
1141
1142 void sqrtsd(XMMRegister dst, XMMRegister src) { Assembler::sqrtsd(dst, src); }
1143 void sqrtsd(XMMRegister dst, Address src) { Assembler::sqrtsd(dst, src); }
1144 void sqrtsd(XMMRegister dst, AddressLiteral src);
1145
1146 void sqrtss(XMMRegister dst, XMMRegister src) { Assembler::sqrtss(dst, src); }
1147 void sqrtss(XMMRegister dst, Address src) { Assembler::sqrtss(dst, src); }
1148 void sqrtss(XMMRegister dst, AddressLiteral src);
1149
1150 void subsd(XMMRegister dst, XMMRegister src) { Assembler::subsd(dst, src); }
1151 void subsd(XMMRegister dst, Address src) { Assembler::subsd(dst, src); }
1152 void subsd(XMMRegister dst, AddressLiteral src);
1153
1154 void subss(XMMRegister dst, XMMRegister src) { Assembler::subss(dst, src); }
1155 void subss(XMMRegister dst, Address src) { Assembler::subss(dst, src); }
1156 void subss(XMMRegister dst, AddressLiteral src);
1157
1158 void ucomiss(XMMRegister dst, XMMRegister src) { Assembler::ucomiss(dst, src); }
1159 void ucomiss(XMMRegister dst, Address src) { Assembler::ucomiss(dst, src); }
1160 void ucomiss(XMMRegister dst, AddressLiteral src);
1161
1162 void ucomisd(XMMRegister dst, XMMRegister src) { Assembler::ucomisd(dst, src); }
1163 void ucomisd(XMMRegister dst, Address src) { Assembler::ucomisd(dst, src); }
1164 void ucomisd(XMMRegister dst, AddressLiteral src);
1165
1166 // Bitwise Logical XOR of Packed Double-Precision Floating-Point Values
1167 void xorpd(XMMRegister dst, XMMRegister src);
1168 void xorpd(XMMRegister dst, Address src) { Assembler::xorpd(dst, src); }
1169 void xorpd(XMMRegister dst, AddressLiteral src);
1170
1171 // Bitwise Logical XOR of Packed Single-Precision Floating-Point Values
1172 void xorps(XMMRegister dst, XMMRegister src);
1173 void xorps(XMMRegister dst, Address src) { Assembler::xorps(dst, src); }
1174 void xorps(XMMRegister dst, AddressLiteral src);
1175
1176 // Shuffle Bytes
1177 void pshufb(XMMRegister dst, XMMRegister src) { Assembler::pshufb(dst, src); }
1178 void pshufb(XMMRegister dst, Address src) { Assembler::pshufb(dst, src); }
1179 void pshufb(XMMRegister dst, AddressLiteral src);
1180 // AVX 3-operands instructions
1181
1182 void vaddsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vaddsd(dst, nds, src); }
1183 void vaddsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vaddsd(dst, nds, src); }
1184 void vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1185
1186 void vaddss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vaddss(dst, nds, src); }
1187 void vaddss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vaddss(dst, nds, src); }
1188 void vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1189
1190 void vabsss(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len);
1191 void vabssd(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len);
1192
1193 void vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1194 void vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1195
1196 void vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1197 void vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1198
1199 void vpand(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vpand(dst, nds, src, vector_len); }
1200 void vpand(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vpand(dst, nds, src, vector_len); }
1201 void vpand(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len);
1202
1203 void vpbroadcastw(XMMRegister dst, XMMRegister src);
1204
1205 void vpcmpeqb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1206 void vpcmpeqw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1207
1208 void vpmovzxbw(XMMRegister dst, Address src, int vector_len);
1209 void vpmovmskb(Register dst, XMMRegister src);
1210
1211 void vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1212 void vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1213
1214 void vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1215 void vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1216
1217 void vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1218 void vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1219
1220 void vpsraw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len);
1221 void vpsraw(XMMRegister dst, XMMRegister nds, int shift, int vector_len);
1222
1223 void vpsrlw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len);
1224 void vpsrlw(XMMRegister dst, XMMRegister nds, int shift, int vector_len);
1225
1226 void vpsllw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len);
1227 void vpsllw(XMMRegister dst, XMMRegister nds, int shift, int vector_len);
1228
1229 void vptest(XMMRegister dst, XMMRegister src);
1230
1231 void punpcklbw(XMMRegister dst, XMMRegister src);
1232 void punpcklbw(XMMRegister dst, Address src) { Assembler::punpcklbw(dst, src); }
1233
1234 void pshuflw(XMMRegister dst, XMMRegister src, int mode);
1235 void pshuflw(XMMRegister dst, Address src, int mode) { Assembler::pshuflw(dst, src, mode); }
1236
1237 void vandpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vandpd(dst, nds, src, vector_len); }
1238 void vandpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vandpd(dst, nds, src, vector_len); }
1239 void vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len);
1240
1241 void vandps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vandps(dst, nds, src, vector_len); }
1242 void vandps(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vandps(dst, nds, src, vector_len); }
1243 void vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len);
1244
1245 void vdivsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vdivsd(dst, nds, src); }
1246 void vdivsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vdivsd(dst, nds, src); }
1247 void vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1248
1249 void vdivss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vdivss(dst, nds, src); }
1250 void vdivss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vdivss(dst, nds, src); }
1251 void vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1252
1253 void vmulsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vmulsd(dst, nds, src); }
1254 void vmulsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vmulsd(dst, nds, src); }
1255 void vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1256
1257 void vmulss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vmulss(dst, nds, src); }
1258 void vmulss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vmulss(dst, nds, src); }
1259 void vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1260
1261 void vsubsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vsubsd(dst, nds, src); }
1262 void vsubsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vsubsd(dst, nds, src); }
1263 void vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1264
1265 void vsubss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vsubss(dst, nds, src); }
1266 void vsubss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vsubss(dst, nds, src); }
1267 void vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1268
1269 void vnegatess(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1270 void vnegatesd(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1271
1272 // AVX Vector instructions
1273
1274 void vxorpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vxorpd(dst, nds, src, vector_len); }
1275 void vxorpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vxorpd(dst, nds, src, vector_len); }
1276 void vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len);
1277
1278 void vxorps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vxorps(dst, nds, src, vector_len); }
1279 void vxorps(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vxorps(dst, nds, src, vector_len); }
1280 void vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len);
1281
1282 void vpxor(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
1283 if (UseAVX > 1 || (vector_len < 1)) // vpxor 256 bit is available only in AVX2
1284 Assembler::vpxor(dst, nds, src, vector_len);
1285 else
1286 Assembler::vxorpd(dst, nds, src, vector_len);
1287 }
1288 void vpxor(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
1289 if (UseAVX > 1 || (vector_len < 1)) // vpxor 256 bit is available only in AVX2
1290 Assembler::vpxor(dst, nds, src, vector_len);
1291 else
1292 Assembler::vxorpd(dst, nds, src, vector_len);
1293 }
1294
1295 // Simple version for AVX2 256bit vectors
1296 void vpxor(XMMRegister dst, XMMRegister src) { Assembler::vpxor(dst, dst, src, true); }
1297 void vpxor(XMMRegister dst, Address src) { Assembler::vpxor(dst, dst, src, true); }
1298
1299 void vinserti128(XMMRegister dst, XMMRegister nds, XMMRegister src, uint8_t imm8) {
1300 if (UseAVX > 2) {
1301 Assembler::vinserti32x4(dst, dst, src, imm8);
1302 } else if (UseAVX > 1) {
1303 // vinserti128 is available only in AVX2
1304 Assembler::vinserti128(dst, nds, src, imm8);
1305 } else {
1306 Assembler::vinsertf128(dst, nds, src, imm8);
1307 }
1308 }
1309
1310 void vinserti128(XMMRegister dst, XMMRegister nds, Address src, uint8_t imm8) {
1311 if (UseAVX > 2) {
1312 Assembler::vinserti32x4(dst, dst, src, imm8);
1313 } else if (UseAVX > 1) {
1314 // vinserti128 is available only in AVX2
1315 Assembler::vinserti128(dst, nds, src, imm8);
1316 } else {
1317 Assembler::vinsertf128(dst, nds, src, imm8);
1318 }
1319 }
1320
1321 void vextracti128(XMMRegister dst, XMMRegister src, uint8_t imm8) {
1322 if (UseAVX > 2) {
1323 Assembler::vextracti32x4(dst, src, imm8);
1324 } else if (UseAVX > 1) {
1325 // vextracti128 is available only in AVX2
1326 Assembler::vextracti128(dst, src, imm8);
1327 } else {
1328 Assembler::vextractf128(dst, src, imm8);
1329 }
1330 }
1331
1332 void vextracti128(Address dst, XMMRegister src, uint8_t imm8) {
1333 if (UseAVX > 2) {
1334 Assembler::vextracti32x4(dst, src, imm8);
1335 } else if (UseAVX > 1) {
1336 // vextracti128 is available only in AVX2
1337 Assembler::vextracti128(dst, src, imm8);
1338 } else {
1339 Assembler::vextractf128(dst, src, imm8);
1340 }
1341 }
1342
1343 // 128bit copy to/from high 128 bits of 256bit (YMM) vector registers
1344 void vinserti128_high(XMMRegister dst, XMMRegister src) {
1345 vinserti128(dst, dst, src, 1);
1346 }
1347 void vinserti128_high(XMMRegister dst, Address src) {
1348 vinserti128(dst, dst, src, 1);
1349 }
1350 void vextracti128_high(XMMRegister dst, XMMRegister src) {
1351 vextracti128(dst, src, 1);
1352 }
1353 void vextracti128_high(Address dst, XMMRegister src) {
1354 vextracti128(dst, src, 1);
1355 }
1356
1357 void vinsertf128_high(XMMRegister dst, XMMRegister src) {
1358 if (UseAVX > 2) {
1359 Assembler::vinsertf32x4(dst, dst, src, 1);
1360 } else {
1361 Assembler::vinsertf128(dst, dst, src, 1);
1362 }
1363 }
1364
1365 void vinsertf128_high(XMMRegister dst, Address src) {
1366 if (UseAVX > 2) {
1367 Assembler::vinsertf32x4(dst, dst, src, 1);
1368 } else {
1369 Assembler::vinsertf128(dst, dst, src, 1);
1370 }
1371 }
1372
1373 void vextractf128_high(XMMRegister dst, XMMRegister src) {
1374 if (UseAVX > 2) {
1375 Assembler::vextractf32x4(dst, src, 1);
1376 } else {
1377 Assembler::vextractf128(dst, src, 1);
1378 }
1379 }
1380
1381 void vextractf128_high(Address dst, XMMRegister src) {
1382 if (UseAVX > 2) {
1383 Assembler::vextractf32x4(dst, src, 1);
1384 } else {
1385 Assembler::vextractf128(dst, src, 1);
1386 }
1387 }
1388
1389 // 256bit copy to/from high 256 bits of 512bit (ZMM) vector registers
1390 void vinserti64x4_high(XMMRegister dst, XMMRegister src) {
1391 Assembler::vinserti64x4(dst, dst, src, 1);
1392 }
1393 void vinsertf64x4_high(XMMRegister dst, XMMRegister src) {
1394 Assembler::vinsertf64x4(dst, dst, src, 1);
1395 }
1396 void vextracti64x4_high(XMMRegister dst, XMMRegister src) {
1397 Assembler::vextracti64x4(dst, src, 1);
1398 }
1399 void vextractf64x4_high(XMMRegister dst, XMMRegister src) {
1400 Assembler::vextractf64x4(dst, src, 1);
1401 }
1402 void vextractf64x4_high(Address dst, XMMRegister src) {
1403 Assembler::vextractf64x4(dst, src, 1);
1404 }
1405 void vinsertf64x4_high(XMMRegister dst, Address src) {
1406 Assembler::vinsertf64x4(dst, dst, src, 1);
1407 }
1408
1409 // 128bit copy to/from low 128 bits of 256bit (YMM) vector registers
1410 void vinserti128_low(XMMRegister dst, XMMRegister src) {
1411 vinserti128(dst, dst, src, 0);
1412 }
1413 void vinserti128_low(XMMRegister dst, Address src) {
1414 vinserti128(dst, dst, src, 0);
1415 }
1416 void vextracti128_low(XMMRegister dst, XMMRegister src) {
1417 vextracti128(dst, src, 0);
1418 }
1419 void vextracti128_low(Address dst, XMMRegister src) {
1420 vextracti128(dst, src, 0);
1421 }
1422
1423 void vinsertf128_low(XMMRegister dst, XMMRegister src) {
1424 if (UseAVX > 2) {
1425 Assembler::vinsertf32x4(dst, dst, src, 0);
1426 } else {
1427 Assembler::vinsertf128(dst, dst, src, 0);
1428 }
1429 }
1430
1431 void vinsertf128_low(XMMRegister dst, Address src) {
1432 if (UseAVX > 2) {
1433 Assembler::vinsertf32x4(dst, dst, src, 0);
1434 } else {
1435 Assembler::vinsertf128(dst, dst, src, 0);
1436 }
1437 }
1438
1439 void vextractf128_low(XMMRegister dst, XMMRegister src) {
1440 if (UseAVX > 2) {
1441 Assembler::vextractf32x4(dst, src, 0);
1442 } else {
1443 Assembler::vextractf128(dst, src, 0);
1444 }
1445 }
1446
1447 void vextractf128_low(Address dst, XMMRegister src) {
1448 if (UseAVX > 2) {
1449 Assembler::vextractf32x4(dst, src, 0);
1450 } else {
1451 Assembler::vextractf128(dst, src, 0);
1452 }
1453 }
1454
1455 // 256bit copy to/from low 256 bits of 512bit (ZMM) vector registers
1456 void vinserti64x4_low(XMMRegister dst, XMMRegister src) {
1457 Assembler::vinserti64x4(dst, dst, src, 0);
1458 }
1459 void vinsertf64x4_low(XMMRegister dst, XMMRegister src) {
1460 Assembler::vinsertf64x4(dst, dst, src, 0);
1461 }
1462 void vextracti64x4_low(XMMRegister dst, XMMRegister src) {
1463 Assembler::vextracti64x4(dst, src, 0);
1464 }
1465 void vextractf64x4_low(XMMRegister dst, XMMRegister src) {
1466 Assembler::vextractf64x4(dst, src, 0);
1467 }
1468 void vextractf64x4_low(Address dst, XMMRegister src) {
1469 Assembler::vextractf64x4(dst, src, 0);
1470 }
1471 void vinsertf64x4_low(XMMRegister dst, Address src) {
1472 Assembler::vinsertf64x4(dst, dst, src, 0);
1473 }
1474
1475 // Carry-Less Multiplication Quadword
1476 void vpclmulldq(XMMRegister dst, XMMRegister nds, XMMRegister src) {
1477 // 0x00 - multiply lower 64 bits [0:63]
1478 Assembler::vpclmulqdq(dst, nds, src, 0x00);
1479 }
1480 void vpclmulhdq(XMMRegister dst, XMMRegister nds, XMMRegister src) {
1481 // 0x11 - multiply upper 64 bits [64:127]
1482 Assembler::vpclmulqdq(dst, nds, src, 0x11);
1483 }
1484
1485 // Data
1486
1487 void cmov32( Condition cc, Register dst, Address src);
1488 void cmov32( Condition cc, Register dst, Register src);
1489
1490 void cmov( Condition cc, Register dst, Register src) { cmovptr(cc, dst, src); }
1491
1492 void cmovptr(Condition cc, Register dst, Address src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmov32(cc, dst, src)); }
1493 void cmovptr(Condition cc, Register dst, Register src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmov32(cc, dst, src)); }
1494
1495 void movoop(Register dst, jobject obj);
1496 void movoop(Address dst, jobject obj);
1497
1498 void mov_metadata(Register dst, Metadata* obj);
1499 void mov_metadata(Address dst, Metadata* obj);
1500
1501 void movptr(ArrayAddress dst, Register src);
1502 // can this do an lea?
1503 void movptr(Register dst, ArrayAddress src);
1504
1505 void movptr(Register dst, Address src);
1506
1507 #ifdef _LP64
1508 void movptr(Register dst, AddressLiteral src, Register scratch=rscratch1);
1509 #else
1510 void movptr(Register dst, AddressLiteral src, Register scratch=noreg); // Scratch reg is ignored in 32-bit
1511 #endif
1512
1513 void movptr(Register dst, intptr_t src);
1514 void movptr(Register dst, Register src);
1515 void movptr(Address dst, intptr_t src);
1516
1517 void movptr(Address dst, Register src);
1518
1519 void movptr(Register dst, RegisterOrConstant src) {
1520 if (src.is_constant()) movptr(dst, src.as_constant());
1521 else movptr(dst, src.as_register());
1522 }
1523
1524 #ifdef _LP64
1525 // Generally the next two are only used for moving NULL
1526 // Although there are situations in initializing the mark word where
1527 // they could be used. They are dangerous.
1528
1529 // They only exist on LP64 so that int32_t and intptr_t are not the same
1530 // and we have ambiguous declarations.
1531
1532 void movptr(Address dst, int32_t imm32);
1533 void movptr(Register dst, int32_t imm32);
1534 #endif // _LP64
1535
1536 // to avoid hiding movl
1537 void mov32(AddressLiteral dst, Register src);
1538 void mov32(Register dst, AddressLiteral src);
1539
1540 // to avoid hiding movb
1541 void movbyte(ArrayAddress dst, int src);
1542
1543 // Import other mov() methods from the parent class or else
1544 // they will be hidden by the following overriding declaration.
1545 using Assembler::movdl;
1546 using Assembler::movq;
1547 void movdl(XMMRegister dst, AddressLiteral src);
1548 void movq(XMMRegister dst, AddressLiteral src);
1549
1550 // Can push value or effective address
1551 void pushptr(AddressLiteral src);
1552
1553 void pushptr(Address src) { LP64_ONLY(pushq(src)) NOT_LP64(pushl(src)); }
1554 void popptr(Address src) { LP64_ONLY(popq(src)) NOT_LP64(popl(src)); }
1555
1556 void pushoop(jobject obj);
1557 void pushklass(Metadata* obj);
1558
1559 // sign extend as need a l to ptr sized element
1560 void movl2ptr(Register dst, Address src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(movl(dst, src)); }
1561 void movl2ptr(Register dst, Register src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(if (dst != src) movl(dst, src)); }
1562
1563 // C2 compiled method's prolog code.
1564 void verified_entry(int framesize, int stack_bang_size, bool fp_mode_24b);
1565
1566 // clear memory of size 'cnt' qwords, starting at 'base';
1567 // if 'is_large' is set, do not try to produce short loop
1568 void clear_mem(Register base, Register cnt, Register rtmp, bool is_large);
1569
1570 #ifdef COMPILER2
1571 void string_indexof_char(Register str1, Register cnt1, Register ch, Register result,
1572 XMMRegister vec1, XMMRegister vec2, XMMRegister vec3, Register tmp);
1573
1574 // IndexOf strings.
1575 // Small strings are loaded through stack if they cross page boundary.
1576 void string_indexof(Register str1, Register str2,
1577 Register cnt1, Register cnt2,
1578 int int_cnt2, Register result,
1579 XMMRegister vec, Register tmp,
1580 int ae);
1581
1582 // IndexOf for constant substrings with size >= 8 elements
1583 // which don't need to be loaded through stack.
1584 void string_indexofC8(Register str1, Register str2,
1585 Register cnt1, Register cnt2,
1586 int int_cnt2, Register result,
1587 XMMRegister vec, Register tmp,
1588 int ae);
1589
1590 // Smallest code: we don't need to load through stack,
1591 // check string tail.
1592
1593 // helper function for string_compare
1594 void load_next_elements(Register elem1, Register elem2, Register str1, Register str2,
1595 Address::ScaleFactor scale, Address::ScaleFactor scale1,
1596 Address::ScaleFactor scale2, Register index, int ae);
1597 // Compare strings.
1598 void string_compare(Register str1, Register str2,
1599 Register cnt1, Register cnt2, Register result,
1600 XMMRegister vec1, int ae);
1601
1602 // Search for Non-ASCII character (Negative byte value) in a byte array,
1603 // return true if it has any and false otherwise.
1604 void has_negatives(Register ary1, Register len,
1605 Register result, Register tmp1,
1606 XMMRegister vec1, XMMRegister vec2);
1607
1608 // Compare char[] or byte[] arrays.
1609 void arrays_equals(bool is_array_equ, Register ary1, Register ary2,
1610 Register limit, Register result, Register chr,
1611 XMMRegister vec1, XMMRegister vec2, bool is_char);
1612
1613 #endif
1614
1615 // Fill primitive arrays
1616 void generate_fill(BasicType t, bool aligned,
1617 Register to, Register value, Register count,
1618 Register rtmp, XMMRegister xtmp);
1619
1620 void encode_iso_array(Register src, Register dst, Register len,
1621 XMMRegister tmp1, XMMRegister tmp2, XMMRegister tmp3,
1622 XMMRegister tmp4, Register tmp5, Register result);
1623
1624 #ifdef _LP64
1625 void add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2);
1626 void multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
1627 Register y, Register y_idx, Register z,
1628 Register carry, Register product,
1629 Register idx, Register kdx);
1630 void multiply_add_128_x_128(Register x_xstart, Register y, Register z,
1631 Register yz_idx, Register idx,
1632 Register carry, Register product, int offset);
1633 void multiply_128_x_128_bmi2_loop(Register y, Register z,
1634 Register carry, Register carry2,
1635 Register idx, Register jdx,
1636 Register yz_idx1, Register yz_idx2,
1637 Register tmp, Register tmp3, Register tmp4);
1638 void multiply_128_x_128_loop(Register x_xstart, Register y, Register z,
1639 Register yz_idx, Register idx, Register jdx,
1640 Register carry, Register product,
1641 Register carry2);
1642 void multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register zlen,
1643 Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5);
1644 void square_rshift(Register x, Register len, Register z, Register tmp1, Register tmp3,
1645 Register tmp4, Register tmp5, Register rdxReg, Register raxReg);
1646 void multiply_add_64_bmi2(Register sum, Register op1, Register op2, Register carry,
1647 Register tmp2);
1648 void multiply_add_64(Register sum, Register op1, Register op2, Register carry,
1649 Register rdxReg, Register raxReg);
1650 void add_one_64(Register z, Register zlen, Register carry, Register tmp1);
1651 void lshift_by_1(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2,
1652 Register tmp3, Register tmp4);
1653 void square_to_len(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2,
1654 Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg);
1655
1656 void mul_add_128_x_32_loop(Register out, Register in, Register offset, Register len, Register tmp1,
1657 Register tmp2, Register tmp3, Register tmp4, Register tmp5, Register rdxReg,
1658 Register raxReg);
1659 void mul_add(Register out, Register in, Register offset, Register len, Register k, Register tmp1,
1660 Register tmp2, Register tmp3, Register tmp4, Register tmp5, Register rdxReg,
1661 Register raxReg);
1662 void vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale,
1663 Register result, Register tmp1, Register tmp2,
1664 XMMRegister vec1, XMMRegister vec2, XMMRegister vec3);
1665 #endif
1666
1667 // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
1668 void update_byte_crc32(Register crc, Register val, Register table);
1669 void kernel_crc32(Register crc, Register buf, Register len, Register table, Register tmp);
1670 // CRC32C code for java.util.zip.CRC32C::updateBytes() intrinsic
1671 // Note on a naming convention:
1672 // Prefix w = register only used on a Westmere+ architecture
1673 // Prefix n = register only used on a Nehalem architecture
1674 #ifdef _LP64
1675 void crc32c_ipl_alg4(Register in_out, uint32_t n,
1676 Register tmp1, Register tmp2, Register tmp3);
1677 #else
1678 void crc32c_ipl_alg4(Register in_out, uint32_t n,
1679 Register tmp1, Register tmp2, Register tmp3,
1680 XMMRegister xtmp1, XMMRegister xtmp2);
1681 #endif
1682 void crc32c_pclmulqdq(XMMRegister w_xtmp1,
1683 Register in_out,
1684 uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported,
1685 XMMRegister w_xtmp2,
1686 Register tmp1,
1687 Register n_tmp2, Register n_tmp3);
1688 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,
1689 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
1690 Register tmp1, Register tmp2,
1691 Register n_tmp3);
1692 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,
1693 Register in_out1, Register in_out2, Register in_out3,
1694 Register tmp1, Register tmp2, Register tmp3,
1695 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
1696 Register tmp4, Register tmp5,
1697 Register n_tmp6);
1698 void crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2,
1699 Register tmp1, Register tmp2, Register tmp3,
1700 Register tmp4, Register tmp5, Register tmp6,
1701 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
1702 bool is_pclmulqdq_supported);
1703 // Fold 128-bit data chunk
1704 void fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset);
1705 void fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, XMMRegister xbuf);
1706 // Fold 8-bit data
1707 void fold_8bit_crc32(Register crc, Register table, Register tmp);
1708 void fold_8bit_crc32(XMMRegister crc, Register table, XMMRegister xtmp, Register tmp);
1709
1710 // Compress char[] array to byte[].
1711 void char_array_compress(Register src, Register dst, Register len,
1712 XMMRegister tmp1, XMMRegister tmp2, XMMRegister tmp3,
1713 XMMRegister tmp4, Register tmp5, Register result);
1714
1715 // Inflate byte[] array to char[].
1716 void byte_array_inflate(Register src, Register dst, Register len,
1717 XMMRegister tmp1, Register tmp2);
1718
1719 };
1720
1721 /**
1722 * class SkipIfEqual:
1723 *
1724 * Instantiating this class will result in assembly code being output that will
1725 * jump around any code emitted between the creation of the instance and it's
1726 * automatic destruction at the end of a scope block, depending on the value of
1727 * the flag passed to the constructor, which will be checked at run-time.
1728 */
1729 class SkipIfEqual {
1730 private:
1731 MacroAssembler* _masm;
1732 Label _label;
1733
1734 public:
1735 SkipIfEqual(MacroAssembler*, const bool* flag_addr, bool value);
1736 ~SkipIfEqual();
1737 };
1738
1739 #endif // CPU_X86_VM_MACROASSEMBLER_X86_HPP