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
33 // MacroAssembler extends Assembler by frequently used macros.
34 //
35 // Instructions for which a 'better' code sequence exists depending
36 // on arguments should also go in here.
37
38 class MacroAssembler: public Assembler {
39 friend class LIR_Assembler;
40 friend class Runtime1; // as_Address()
41
42 protected:
43
44 Address as_Address(AddressLiteral adr);
45 Address as_Address(ArrayAddress adr);
46
47 // Support for VM calls
48 //
49 // This is the base routine called by the different versions of call_VM_leaf. The interpreter
50 // may customize this version by overriding it for its purposes (e.g., to save/restore
51 // additional registers when doing a VM call).
52 #ifdef CC_INTERP
53 // c++ interpreter never wants to use interp_masm version of call_VM
54 #define VIRTUAL
55 #else
56 #define VIRTUAL virtual
57 #endif
58
59 VIRTUAL void call_VM_leaf_base(
60 address entry_point, // the entry point
61 int number_of_arguments // the number of arguments to pop after the call
62 );
63
64 // This is the base routine called by the different versions of call_VM. The interpreter
65 // may customize this version by overriding it for its purposes (e.g., to save/restore
66 // additional registers when doing a VM call).
67 //
68 // If no java_thread register is specified (noreg) than rdi will be used instead. call_VM_base
69 // returns the register which contains the thread upon return. If a thread register has been
70 // specified, the return value will correspond to that register. If no last_java_sp is specified
71 // (noreg) than rsp will be used instead.
72 VIRTUAL void call_VM_base( // returns the register containing the thread upon return
73 Register oop_result, // where an oop-result ends up if any; use noreg otherwise
74 Register java_thread, // the thread if computed before ; use noreg otherwise
75 Register last_java_sp, // to set up last_Java_frame in stubs; use noreg otherwise
76 address entry_point, // the entry point
77 int number_of_arguments, // the number of arguments (w/o thread) to pop after the call
78 bool check_exceptions // whether to check for pending exceptions after return
79 );
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 void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true);
88
89 // helpers for FPU flag access
90 // tmp is a temporary register, if none is available use noreg
91 void save_rax (Register tmp);
92 void restore_rax(Register tmp);
93
94 public:
95 MacroAssembler(CodeBuffer* code) : Assembler(code) {}
96
97 // Support for NULL-checks
98 //
99 // Generates code that causes a NULL OS exception if the content of reg is NULL.
100 // If the accessed location is M[reg + offset] and the offset is known, provide the
101 // offset. No explicit code generation is needed if the offset is within a certain
102 // range (0 <= offset <= page_size).
103
104 void null_check(Register reg, int offset = -1);
105 static bool needs_explicit_null_check(intptr_t offset);
106
107 // Required platform-specific helpers for Label::patch_instructions.
108 // They _shadow_ the declarations in AbstractAssembler, which are undefined.
109 void pd_patch_instruction(address branch, address target) {
110 unsigned char op = branch[0];
111 assert(op == 0xE8 /* call */ ||
112 op == 0xE9 /* jmp */ ||
113 op == 0xEB /* short jmp */ ||
114 (op & 0xF0) == 0x70 /* short jcc */ ||
115 op == 0x0F && (branch[1] & 0xF0) == 0x80 /* jcc */ ||
116 op == 0xC7 && branch[1] == 0xF8 /* xbegin */,
117 "Invalid opcode at patch point");
118
119 if (op == 0xEB || (op & 0xF0) == 0x70) {
120 // short offset operators (jmp and jcc)
121 char* disp = (char*) &branch[1];
122 int imm8 = target - (address) &disp[1];
123 guarantee(this->is8bit(imm8), "Short forward jump exceeds 8-bit offset");
124 *disp = imm8;
125 } else {
126 int* disp = (int*) &branch[(op == 0x0F || op == 0xC7)? 2: 1];
127 int imm32 = target - (address) &disp[1];
128 *disp = imm32;
129 }
130 }
131
132 // The following 4 methods return the offset of the appropriate move instruction
133
134 // Support for fast byte/short loading with zero extension (depending on particular CPU)
135 int load_unsigned_byte(Register dst, Address src);
136 int load_unsigned_short(Register dst, Address src);
137
138 // Support for fast byte/short loading with sign extension (depending on particular CPU)
139 int load_signed_byte(Register dst, Address src);
140 int load_signed_short(Register dst, Address src);
141
142 // Support for sign-extension (hi:lo = extend_sign(lo))
143 void extend_sign(Register hi, Register lo);
144
145 // Load and store values by size and signed-ness
146 void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2 = noreg);
147 void store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2 = noreg);
148
149 // Support for inc/dec with optimal instruction selection depending on value
150
151 void increment(Register reg, int value = 1) { LP64_ONLY(incrementq(reg, value)) NOT_LP64(incrementl(reg, value)) ; }
152 void decrement(Register reg, int value = 1) { LP64_ONLY(decrementq(reg, value)) NOT_LP64(decrementl(reg, value)) ; }
153
154 void decrementl(Address dst, int value = 1);
155 void decrementl(Register reg, int value = 1);
156
157 void decrementq(Register reg, int value = 1);
158 void decrementq(Address dst, int value = 1);
159
160 void incrementl(Address dst, int value = 1);
161 void incrementl(Register reg, int value = 1);
162
163 void incrementq(Register reg, int value = 1);
164 void incrementq(Address dst, int value = 1);
165
166 // Support optimal SSE move instructions.
167 void movflt(XMMRegister dst, XMMRegister src) {
168 if (UseXmmRegToRegMoveAll) { movaps(dst, src); return; }
169 else { movss (dst, src); return; }
170 }
171 void movflt(XMMRegister dst, Address src) { movss(dst, src); }
172 void movflt(XMMRegister dst, AddressLiteral src);
173 void movflt(Address dst, XMMRegister src) { movss(dst, src); }
174
175 void movdbl(XMMRegister dst, XMMRegister src) {
176 if (UseXmmRegToRegMoveAll) { movapd(dst, src); return; }
177 else { movsd (dst, src); return; }
178 }
179
180 void movdbl(XMMRegister dst, AddressLiteral src);
181
182 void movdbl(XMMRegister dst, Address src) {
183 if (UseXmmLoadAndClearUpper) { movsd (dst, src); return; }
184 else { movlpd(dst, src); return; }
185 }
186 void movdbl(Address dst, XMMRegister src) { movsd(dst, src); }
187
188 void incrementl(AddressLiteral dst);
189 void incrementl(ArrayAddress dst);
190
191 void incrementq(AddressLiteral dst);
192
193 // Alignment
194 void align(int modulus);
195
196 // A 5 byte nop that is safe for patching (see patch_verified_entry)
197 void fat_nop();
198
199 // Stack frame creation/removal
200 void enter();
201 void leave();
202
203 // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information)
204 // The pointer will be loaded into the thread register.
205 void get_thread(Register thread);
206
207
208 // Support for VM calls
209 //
210 // It is imperative that all calls into the VM are handled via the call_VM macros.
211 // They make sure that the stack linkage is setup correctly. call_VM's correspond
212 // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points.
213
214
215 void call_VM(Register oop_result,
216 address entry_point,
217 bool check_exceptions = true);
218 void call_VM(Register oop_result,
219 address entry_point,
220 Register arg_1,
221 bool check_exceptions = true);
222 void call_VM(Register oop_result,
223 address entry_point,
224 Register arg_1, Register arg_2,
225 bool check_exceptions = true);
226 void call_VM(Register oop_result,
227 address entry_point,
228 Register arg_1, Register arg_2, Register arg_3,
229 bool check_exceptions = true);
230
231 // Overloadings with last_Java_sp
232 void call_VM(Register oop_result,
233 Register last_java_sp,
234 address entry_point,
235 int number_of_arguments = 0,
236 bool check_exceptions = true);
237 void call_VM(Register oop_result,
238 Register last_java_sp,
239 address entry_point,
240 Register arg_1, bool
241 check_exceptions = true);
242 void call_VM(Register oop_result,
243 Register last_java_sp,
244 address entry_point,
245 Register arg_1, Register arg_2,
246 bool check_exceptions = true);
247 void call_VM(Register oop_result,
248 Register last_java_sp,
249 address entry_point,
250 Register arg_1, Register arg_2, Register arg_3,
251 bool check_exceptions = true);
252
253 void get_vm_result (Register oop_result, Register thread);
254 void get_vm_result_2(Register metadata_result, Register thread);
255
256 // These always tightly bind to MacroAssembler::call_VM_base
257 // bypassing the virtual implementation
258 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
259 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, 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, bool check_exceptions = true);
261 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);
262 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);
263
264 void call_VM_leaf(address entry_point,
265 int number_of_arguments = 0);
266 void call_VM_leaf(address entry_point,
267 Register arg_1);
268 void call_VM_leaf(address entry_point,
269 Register arg_1, Register arg_2);
270 void call_VM_leaf(address entry_point,
271 Register arg_1, Register arg_2, Register arg_3);
272
273 // These always tightly bind to MacroAssembler::call_VM_leaf_base
274 // bypassing the virtual implementation
275 void super_call_VM_leaf(address entry_point);
276 void super_call_VM_leaf(address entry_point, Register arg_1);
277 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2);
278 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3);
279 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4);
280
281 // last Java Frame (fills frame anchor)
282 void set_last_Java_frame(Register thread,
283 Register last_java_sp,
284 Register last_java_fp,
285 address last_java_pc);
286
287 // thread in the default location (r15_thread on 64bit)
288 void set_last_Java_frame(Register last_java_sp,
289 Register last_java_fp,
290 address last_java_pc);
291
292 void reset_last_Java_frame(Register thread, bool clear_fp);
293
294 // thread in the default location (r15_thread on 64bit)
295 void reset_last_Java_frame(bool clear_fp);
296
297 // Stores
298 void store_check(Register obj); // store check for obj - register is destroyed afterwards
299 void store_check(Register obj, Address dst); // same as above, dst is exact store location (reg. is destroyed)
300
301 #if INCLUDE_ALL_GCS
302
303 void g1_write_barrier_pre(Register obj,
304 Register pre_val,
305 Register thread,
306 Register tmp,
307 bool tosca_live,
308 bool expand_call);
309
310 void g1_write_barrier_post(Register store_addr,
311 Register new_val,
312 Register thread,
313 Register tmp,
314 Register tmp2);
315
316 #endif // INCLUDE_ALL_GCS
317
318 // split store_check(Register obj) to enhance instruction interleaving
319 void store_check_part_1(Register obj);
320 void store_check_part_2(Register obj);
321
322 // C 'boolean' to Java boolean: x == 0 ? 0 : 1
323 void c2bool(Register x);
324
325 // C++ bool manipulation
326
327 void movbool(Register dst, Address src);
328 void movbool(Address dst, bool boolconst);
329 void movbool(Address dst, Register src);
330 void testbool(Register dst);
331
332 // oop manipulations
333 void load_klass(Register dst, Register src);
334 void store_klass(Register dst, Register src);
335
336 void load_heap_oop(Register dst, Address src);
337 void load_heap_oop_not_null(Register dst, Address src);
338 void store_heap_oop(Address dst, Register src);
339 void cmp_heap_oop(Register src1, Address src2, Register tmp = noreg);
340
341 // Used for storing NULL. All other oop constants should be
342 // stored using routines that take a jobject.
343 void store_heap_oop_null(Address dst);
344
345 void load_prototype_header(Register dst, Register src);
346
347 #ifdef _LP64
348 void store_klass_gap(Register dst, Register src);
349
350 // This dummy is to prevent a call to store_heap_oop from
351 // converting a zero (like NULL) into a Register by giving
352 // the compiler two choices it can't resolve
353
354 void store_heap_oop(Address dst, void* dummy);
355
356 void encode_heap_oop(Register r);
357 void decode_heap_oop(Register r);
358 void encode_heap_oop_not_null(Register r);
359 void decode_heap_oop_not_null(Register r);
360 void encode_heap_oop_not_null(Register dst, Register src);
361 void decode_heap_oop_not_null(Register dst, Register src);
362
363 void set_narrow_oop(Register dst, jobject obj);
364 void set_narrow_oop(Address dst, jobject obj);
365 void cmp_narrow_oop(Register dst, jobject obj);
366 void cmp_narrow_oop(Address dst, jobject obj);
367
368 void encode_klass_not_null(Register r);
369 void decode_klass_not_null(Register r);
370 void encode_klass_not_null(Register dst, Register src);
371 void decode_klass_not_null(Register dst, Register src);
372 void set_narrow_klass(Register dst, Klass* k);
373 void set_narrow_klass(Address dst, Klass* k);
374 void cmp_narrow_klass(Register dst, Klass* k);
375 void cmp_narrow_klass(Address dst, Klass* k);
376
377 // Returns the byte size of the instructions generated by decode_klass_not_null()
378 // when compressed klass pointers are being used.
379 static int instr_size_for_decode_klass_not_null();
380
381 // if heap base register is used - reinit it with the correct value
382 void reinit_heapbase();
383
384 DEBUG_ONLY(void verify_heapbase(const char* msg);)
385
386 #endif // _LP64
387
388 // Int division/remainder for Java
389 // (as idivl, but checks for special case as described in JVM spec.)
390 // returns idivl instruction offset for implicit exception handling
391 int corrected_idivl(Register reg);
392
393 // Long division/remainder for Java
394 // (as idivq, but checks for special case as described in JVM spec.)
395 // returns idivq instruction offset for implicit exception handling
396 int corrected_idivq(Register reg);
397
398 void int3();
399
400 // Long operation macros for a 32bit cpu
401 // Long negation for Java
402 void lneg(Register hi, Register lo);
403
404 // Long multiplication for Java
405 // (destroys contents of eax, ebx, ecx and edx)
406 void lmul(int x_rsp_offset, int y_rsp_offset); // rdx:rax = x * y
407
408 // Long shifts for Java
409 // (semantics as described in JVM spec.)
410 void lshl(Register hi, Register lo); // hi:lo << (rcx & 0x3f)
411 void lshr(Register hi, Register lo, bool sign_extension = false); // hi:lo >> (rcx & 0x3f)
412
413 // Long compare for Java
414 // (semantics as described in JVM spec.)
415 void lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo); // x_hi = lcmp(x, y)
416
417
418 // misc
419
420 // Sign extension
421 void sign_extend_short(Register reg);
422 void sign_extend_byte(Register reg);
423
424 // Division by power of 2, rounding towards 0
425 void division_with_shift(Register reg, int shift_value);
426
427 // Compares the top-most stack entries on the FPU stack and sets the eflags as follows:
428 //
429 // CF (corresponds to C0) if x < y
430 // PF (corresponds to C2) if unordered
431 // ZF (corresponds to C3) if x = y
432 //
433 // The arguments are in reversed order on the stack (i.e., top of stack is first argument).
434 // tmp is a temporary register, if none is available use noreg (only matters for non-P6 code)
435 void fcmp(Register tmp);
436 // Variant of the above which allows y to be further down the stack
437 // and which only pops x and y if specified. If pop_right is
438 // specified then pop_left must also be specified.
439 void fcmp(Register tmp, int index, bool pop_left, bool pop_right);
440
441 // Floating-point comparison for Java
442 // Compares the top-most stack entries on the FPU stack and stores the result in dst.
443 // The arguments are in reversed order on the stack (i.e., top of stack is first argument).
444 // (semantics as described in JVM spec.)
445 void fcmp2int(Register dst, bool unordered_is_less);
446 // Variant of the above which allows y to be further down the stack
447 // and which only pops x and y if specified. If pop_right is
448 // specified then pop_left must also be specified.
449 void fcmp2int(Register dst, bool unordered_is_less, int index, bool pop_left, bool pop_right);
450
451 // Floating-point remainder for Java (ST0 = ST0 fremr ST1, ST1 is empty afterwards)
452 // tmp is a temporary register, if none is available use noreg
453 void fremr(Register tmp);
454
455
456 // same as fcmp2int, but using SSE2
457 void cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less);
458 void cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less);
459
460 // Inlined sin/cos generator for Java; must not use CPU instruction
461 // directly on Intel as it does not have high enough precision
462 // outside of the range [-pi/4, pi/4]. Extra argument indicate the
463 // number of FPU stack slots in use; all but the topmost will
464 // require saving if a slow case is necessary. Assumes argument is
465 // on FP TOS; result is on FP TOS. No cpu registers are changed by
466 // this code.
467 void trigfunc(char trig, int num_fpu_regs_in_use = 1);
468
469 // branch to L if FPU flag C2 is set/not set
470 // tmp is a temporary register, if none is available use noreg
471 void jC2 (Register tmp, Label& L);
472 void jnC2(Register tmp, Label& L);
473
474 // Pop ST (ffree & fincstp combined)
475 void fpop();
476
477 // pushes double TOS element of FPU stack on CPU stack; pops from FPU stack
478 void push_fTOS();
479
480 // pops double TOS element from CPU stack and pushes on FPU stack
481 void pop_fTOS();
482
483 void empty_FPU_stack();
484
485 void push_IU_state();
486 void pop_IU_state();
487
488 void push_FPU_state();
489 void pop_FPU_state();
490
491 void push_CPU_state();
492 void pop_CPU_state();
493
494 // Round up to a power of two
495 void round_to(Register reg, int modulus);
496
497 // Callee saved registers handling
498 void push_callee_saved_registers();
499 void pop_callee_saved_registers();
500
501 // allocation
502 void eden_allocate(
503 Register obj, // result: pointer to object after successful allocation
504 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise
505 int con_size_in_bytes, // object size in bytes if known at compile time
506 Register t1, // temp register
507 Label& slow_case // continuation point if fast allocation fails
508 );
509 void tlab_allocate(
510 Register obj, // result: pointer to object after successful allocation
511 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise
512 int con_size_in_bytes, // object size in bytes if known at compile time
513 Register t1, // temp register
514 Register t2, // temp register
515 Label& slow_case // continuation point if fast allocation fails
516 );
517 Register tlab_refill(Label& retry_tlab, Label& try_eden, Label& slow_case); // returns TLS address
518 void incr_allocated_bytes(Register thread,
519 Register var_size_in_bytes, int con_size_in_bytes,
520 Register t1 = noreg);
521
522 // interface method calling
523 void lookup_interface_method(Register recv_klass,
524 Register intf_klass,
525 RegisterOrConstant itable_index,
526 Register method_result,
527 Register scan_temp,
528 Label& no_such_interface,
529 bool return_method = true);
530
531 // virtual method calling
532 void lookup_virtual_method(Register recv_klass,
533 RegisterOrConstant vtable_index,
534 Register method_result);
535
536 // Test sub_klass against super_klass, with fast and slow paths.
537
538 // The fast path produces a tri-state answer: yes / no / maybe-slow.
539 // One of the three labels can be NULL, meaning take the fall-through.
540 // If super_check_offset is -1, the value is loaded up from super_klass.
541 // No registers are killed, except temp_reg.
542 void check_klass_subtype_fast_path(Register sub_klass,
543 Register super_klass,
544 Register temp_reg,
545 Label* L_success,
546 Label* L_failure,
547 Label* L_slow_path,
548 RegisterOrConstant super_check_offset = RegisterOrConstant(-1));
549
550 // The rest of the type check; must be wired to a corresponding fast path.
551 // It does not repeat the fast path logic, so don't use it standalone.
552 // The temp_reg and temp2_reg can be noreg, if no temps are available.
553 // Updates the sub's secondary super cache as necessary.
554 // If set_cond_codes, condition codes will be Z on success, NZ on failure.
555 void check_klass_subtype_slow_path(Register sub_klass,
556 Register super_klass,
557 Register temp_reg,
558 Register temp2_reg,
559 Label* L_success,
560 Label* L_failure,
561 bool set_cond_codes = false);
562
563 // Simplified, combined version, good for typical uses.
564 // Falls through on failure.
565 void check_klass_subtype(Register sub_klass,
566 Register super_klass,
567 Register temp_reg,
568 Label& L_success);
569
570 // method handles (JSR 292)
571 Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0);
572
573 //----
574 void set_word_if_not_zero(Register reg); // sets reg to 1 if not zero, otherwise 0
575
576 // Debugging
577
578 // only if +VerifyOops
579 // TODO: Make these macros with file and line like sparc version!
580 void verify_oop(Register reg, const char* s = "broken oop");
581 void verify_oop_addr(Address addr, const char * s = "broken oop addr");
582
583 // TODO: verify method and klass metadata (compare against vptr?)
584 void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
585 void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line){}
586
587 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
588 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
589
590 // only if +VerifyFPU
591 void verify_FPU(int stack_depth, const char* s = "illegal FPU state");
592
593 // Verify or restore cpu control state after JNI call
594 void restore_cpu_control_state_after_jni();
595
596 // prints msg, dumps registers and stops execution
597 void stop(const char* msg);
598
599 // prints msg and continues
600 void warn(const char* msg);
601
602 // dumps registers and other state
603 void print_state();
604
605 static void debug32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip, char* msg);
606 static void debug64(char* msg, int64_t pc, int64_t regs[]);
607 static void print_state32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip);
608 static void print_state64(int64_t pc, int64_t regs[]);
609
610 void os_breakpoint();
611
612 void untested() { stop("untested"); }
613
614 void unimplemented(const char* what = "") { char* b = new char[1024]; jio_snprintf(b, 1024, "unimplemented: %s", what); stop(b); }
615
616 void should_not_reach_here() { stop("should not reach here"); }
617
618 void print_CPU_state();
619
620 // Stack overflow checking
621 void bang_stack_with_offset(int offset) {
622 // stack grows down, caller passes positive offset
623 assert(offset > 0, "must bang with negative offset");
624 movl(Address(rsp, (-offset)), rax);
625 }
626
627 // Writes to stack successive pages until offset reached to check for
628 // stack overflow + shadow pages. Also, clobbers tmp
629 void bang_stack_size(Register size, Register tmp);
630
631 virtual RegisterOrConstant delayed_value_impl(intptr_t* delayed_value_addr,
632 Register tmp,
633 int offset);
634
635 // Support for serializing memory accesses between threads
636 void serialize_memory(Register thread, Register tmp);
637
638 void verify_tlab();
639
640 // Biased locking support
641 // lock_reg and obj_reg must be loaded up with the appropriate values.
642 // swap_reg must be rax, and is killed.
643 // tmp_reg is optional. If it is supplied (i.e., != noreg) it will
644 // be killed; if not supplied, push/pop will be used internally to
645 // allocate a temporary (inefficient, avoid if possible).
646 // Optional slow case is for implementations (interpreter and C1) which branch to
647 // slow case directly. Leaves condition codes set for C2's Fast_Lock node.
648 // Returns offset of first potentially-faulting instruction for null
649 // check info (currently consumed only by C1). If
650 // swap_reg_contains_mark is true then returns -1 as it is assumed
651 // the calling code has already passed any potential faults.
652 int biased_locking_enter(Register lock_reg, Register obj_reg,
653 Register swap_reg, Register tmp_reg,
654 bool swap_reg_contains_mark,
655 Label& done, Label* slow_case = NULL,
656 BiasedLockingCounters* counters = NULL);
657 void biased_locking_exit (Register obj_reg, Register temp_reg, Label& done);
658 #ifdef COMPILER2
659 // Code used by cmpFastLock and cmpFastUnlock mach instructions in .ad file.
660 // See full desription in macroAssembler_x86.cpp.
661 void fast_lock(Register obj, Register box, Register tmp,
662 Register scr, Register cx1, Register cx2,
663 BiasedLockingCounters* counters,
664 RTMLockingCounters* rtm_counters,
665 RTMLockingCounters* stack_rtm_counters,
666 Metadata* method_data,
667 bool use_rtm, bool profile_rtm);
668 void fast_unlock(Register obj, Register box, Register tmp, bool use_rtm);
669 #if INCLUDE_RTM_OPT
670 void rtm_counters_update(Register abort_status, Register rtm_counters);
671 void branch_on_random_using_rdtsc(Register tmp, Register scr, int count, Label& brLabel);
672 void rtm_abort_ratio_calculation(Register tmp, Register rtm_counters_reg,
673 RTMLockingCounters* rtm_counters,
674 Metadata* method_data);
675 void rtm_profiling(Register abort_status_Reg, Register rtm_counters_Reg,
676 RTMLockingCounters* rtm_counters, Metadata* method_data, bool profile_rtm);
677 void rtm_retry_lock_on_abort(Register retry_count, Register abort_status, Label& retryLabel);
678 void rtm_retry_lock_on_busy(Register retry_count, Register box, Register tmp, Register scr, Label& retryLabel);
679 void rtm_stack_locking(Register obj, Register tmp, Register scr,
680 Register retry_on_abort_count,
681 RTMLockingCounters* stack_rtm_counters,
682 Metadata* method_data, bool profile_rtm,
683 Label& DONE_LABEL, Label& IsInflated);
684 void rtm_inflated_locking(Register obj, Register box, Register tmp,
685 Register scr, Register retry_on_busy_count,
686 Register retry_on_abort_count,
687 RTMLockingCounters* rtm_counters,
688 Metadata* method_data, bool profile_rtm,
689 Label& DONE_LABEL);
690 #endif
691 #endif
692
693 Condition negate_condition(Condition cond);
694
695 // Instructions that use AddressLiteral operands. These instruction can handle 32bit/64bit
696 // operands. In general the names are modified to avoid hiding the instruction in Assembler
697 // so that we don't need to implement all the varieties in the Assembler with trivial wrappers
698 // here in MacroAssembler. The major exception to this rule is call
699
700 // Arithmetics
701
702
703 void addptr(Address dst, int32_t src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)) ; }
704 void addptr(Address dst, Register src);
705
706 void addptr(Register dst, Address src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)); }
707 void addptr(Register dst, int32_t src);
708 void addptr(Register dst, Register src);
709 void addptr(Register dst, RegisterOrConstant src) {
710 if (src.is_constant()) addptr(dst, (int) src.as_constant());
711 else addptr(dst, src.as_register());
712 }
713
714 void andptr(Register dst, int32_t src);
715 void andptr(Register src1, Register src2) { LP64_ONLY(andq(src1, src2)) NOT_LP64(andl(src1, src2)) ; }
716
717 void cmp8(AddressLiteral src1, int imm);
718
719 // renamed to drag out the casting of address to int32_t/intptr_t
720 void cmp32(Register src1, int32_t imm);
721
722 void cmp32(AddressLiteral src1, int32_t imm);
723 // compare reg - mem, or reg - &mem
724 void cmp32(Register src1, AddressLiteral src2);
725
726 void cmp32(Register src1, Address src2);
727
728 #ifndef _LP64
729 void cmpklass(Address dst, Metadata* obj);
730 void cmpklass(Register dst, Metadata* obj);
731 void cmpoop(Address dst, jobject obj);
732 void cmpoop(Register dst, jobject obj);
733 #endif // _LP64
734
735 // NOTE src2 must be the lval. This is NOT an mem-mem compare
736 void cmpptr(Address src1, AddressLiteral src2);
737
738 void cmpptr(Register src1, AddressLiteral src2);
739
740 void cmpptr(Register src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
741 void cmpptr(Register src1, Address src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
742 // void cmpptr(Address src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
743
744 void cmpptr(Register src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
745 void cmpptr(Address src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
746
747 // cmp64 to avoild hiding cmpq
748 void cmp64(Register src1, AddressLiteral src);
749
750 void cmpxchgptr(Register reg, Address adr);
751
752 void locked_cmpxchgptr(Register reg, AddressLiteral adr);
753
754
755 void imulptr(Register dst, Register src) { LP64_ONLY(imulq(dst, src)) NOT_LP64(imull(dst, src)); }
756 void imulptr(Register dst, Register src, int imm32) { LP64_ONLY(imulq(dst, src, imm32)) NOT_LP64(imull(dst, src, imm32)); }
757
758
759 void negptr(Register dst) { LP64_ONLY(negq(dst)) NOT_LP64(negl(dst)); }
760
761 void notptr(Register dst) { LP64_ONLY(notq(dst)) NOT_LP64(notl(dst)); }
762
763 void shlptr(Register dst, int32_t shift);
764 void shlptr(Register dst) { LP64_ONLY(shlq(dst)) NOT_LP64(shll(dst)); }
765
766 void shrptr(Register dst, int32_t shift);
767 void shrptr(Register dst) { LP64_ONLY(shrq(dst)) NOT_LP64(shrl(dst)); }
768
769 void sarptr(Register dst) { LP64_ONLY(sarq(dst)) NOT_LP64(sarl(dst)); }
770 void sarptr(Register dst, int32_t src) { LP64_ONLY(sarq(dst, src)) NOT_LP64(sarl(dst, src)); }
771
772 void subptr(Address dst, int32_t src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); }
773
774 void subptr(Register dst, Address src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); }
775 void subptr(Register dst, int32_t src);
776 // Force generation of a 4 byte immediate value even if it fits into 8bit
777 void subptr_imm32(Register dst, int32_t src);
778 void subptr(Register dst, Register src);
779 void subptr(Register dst, RegisterOrConstant src) {
780 if (src.is_constant()) subptr(dst, (int) src.as_constant());
781 else subptr(dst, src.as_register());
782 }
783
784 void sbbptr(Address dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); }
785 void sbbptr(Register dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); }
786
787 void xchgptr(Register src1, Register src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; }
788 void xchgptr(Register src1, Address src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; }
789
790 void xaddptr(Address src1, Register src2) { LP64_ONLY(xaddq(src1, src2)) NOT_LP64(xaddl(src1, src2)) ; }
791
792
793
794 // Helper functions for statistics gathering.
795 // Conditionally (atomically, on MPs) increments passed counter address, preserving condition codes.
796 void cond_inc32(Condition cond, AddressLiteral counter_addr);
797 // Unconditional atomic increment.
798 void atomic_incl(Address counter_addr);
799 void atomic_incl(AddressLiteral counter_addr, Register scr = rscratch1);
800 #ifdef _LP64
801 void atomic_incq(Address counter_addr);
802 void atomic_incq(AddressLiteral counter_addr, Register scr = rscratch1);
803 #endif
804 void atomic_incptr(AddressLiteral counter_addr, Register scr = rscratch1) { LP64_ONLY(atomic_incq(counter_addr, scr)) NOT_LP64(atomic_incl(counter_addr, scr)) ; }
805 void atomic_incptr(Address counter_addr) { LP64_ONLY(atomic_incq(counter_addr)) NOT_LP64(atomic_incl(counter_addr)) ; }
806
807 void lea(Register dst, AddressLiteral adr);
808 void lea(Address dst, AddressLiteral adr);
809 void lea(Register dst, Address adr) { Assembler::lea(dst, adr); }
810
811 void leal32(Register dst, Address src) { leal(dst, src); }
812
813 // Import other testl() methods from the parent class or else
814 // they will be hidden by the following overriding declaration.
815 using Assembler::testl;
816 void testl(Register dst, AddressLiteral src);
817
818 void orptr(Register dst, Address src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); }
819 void orptr(Register dst, Register src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); }
820 void orptr(Register dst, int32_t src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); }
821 void orptr(Address dst, int32_t imm32) { LP64_ONLY(orq(dst, imm32)) NOT_LP64(orl(dst, imm32)); }
822
823 void testptr(Register src, int32_t imm32) { LP64_ONLY(testq(src, imm32)) NOT_LP64(testl(src, imm32)); }
824 void testptr(Register src1, Register src2);
825
826 void xorptr(Register dst, Register src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); }
827 void xorptr(Register dst, Address src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); }
828
829 // Calls
830
831 void call(Label& L, relocInfo::relocType rtype);
832 void call(Register entry);
833
834 // NOTE: this call tranfers to the effective address of entry NOT
835 // the address contained by entry. This is because this is more natural
836 // for jumps/calls.
837 void call(AddressLiteral entry);
838
839 // Emit the CompiledIC call idiom
840 void ic_call(address entry);
841
842 // Jumps
843
844 // NOTE: these jumps tranfer to the effective address of dst NOT
845 // the address contained by dst. This is because this is more natural
846 // for jumps/calls.
847 void jump(AddressLiteral dst);
848 void jump_cc(Condition cc, AddressLiteral dst);
849
850 // 32bit can do a case table jump in one instruction but we no longer allow the base
851 // to be installed in the Address class. This jump will tranfers to the address
852 // contained in the location described by entry (not the address of entry)
853 void jump(ArrayAddress entry);
854
855 // Floating
856
857 void andpd(XMMRegister dst, Address src) { Assembler::andpd(dst, src); }
858 void andpd(XMMRegister dst, AddressLiteral src);
859
860 void andps(XMMRegister dst, XMMRegister src) { Assembler::andps(dst, src); }
861 void andps(XMMRegister dst, Address src) { Assembler::andps(dst, src); }
862 void andps(XMMRegister dst, AddressLiteral src);
863
864 void comiss(XMMRegister dst, XMMRegister src) { Assembler::comiss(dst, src); }
865 void comiss(XMMRegister dst, Address src) { Assembler::comiss(dst, src); }
866 void comiss(XMMRegister dst, AddressLiteral src);
867
868 void comisd(XMMRegister dst, XMMRegister src) { Assembler::comisd(dst, src); }
869 void comisd(XMMRegister dst, Address src) { Assembler::comisd(dst, src); }
870 void comisd(XMMRegister dst, AddressLiteral src);
871
872 void fadd_s(Address src) { Assembler::fadd_s(src); }
873 void fadd_s(AddressLiteral src) { Assembler::fadd_s(as_Address(src)); }
874
875 void fldcw(Address src) { Assembler::fldcw(src); }
876 void fldcw(AddressLiteral src);
877
878 void fld_s(int index) { Assembler::fld_s(index); }
879 void fld_s(Address src) { Assembler::fld_s(src); }
880 void fld_s(AddressLiteral src);
881
882 void fld_d(Address src) { Assembler::fld_d(src); }
883 void fld_d(AddressLiteral src);
884
885 void fld_x(Address src) { Assembler::fld_x(src); }
886 void fld_x(AddressLiteral src);
887
888 void fmul_s(Address src) { Assembler::fmul_s(src); }
889 void fmul_s(AddressLiteral src) { Assembler::fmul_s(as_Address(src)); }
890
891 void ldmxcsr(Address src) { Assembler::ldmxcsr(src); }
892 void ldmxcsr(AddressLiteral src);
893
894 // compute pow(x,y) and exp(x) with x86 instructions. Don't cover
895 // all corner cases and may result in NaN and require fallback to a
896 // runtime call.
897 void fast_pow();
898 void fast_exp();
899 void increase_precision();
900 void restore_precision();
901
902 // computes exp(x). Fallback to runtime call included.
903 void exp_with_fallback(int num_fpu_regs_in_use) { pow_or_exp(true, num_fpu_regs_in_use); }
904 // computes pow(x,y). Fallback to runtime call included.
905 void pow_with_fallback(int num_fpu_regs_in_use) { pow_or_exp(false, num_fpu_regs_in_use); }
906
907 private:
908
909 // call runtime as a fallback for trig functions and pow/exp.
910 void fp_runtime_fallback(address runtime_entry, int nb_args, int num_fpu_regs_in_use);
911
912 // computes 2^(Ylog2X); Ylog2X in ST(0)
913 void pow_exp_core_encoding();
914
915 // computes pow(x,y) or exp(x). Fallback to runtime call included.
916 void pow_or_exp(bool is_exp, int num_fpu_regs_in_use);
917
918 // these are private because users should be doing movflt/movdbl
919
920 void movss(Address dst, XMMRegister src) { Assembler::movss(dst, src); }
921 void movss(XMMRegister dst, XMMRegister src) { Assembler::movss(dst, src); }
922 void movss(XMMRegister dst, Address src) { Assembler::movss(dst, src); }
923 void movss(XMMRegister dst, AddressLiteral src);
924
925 void movlpd(XMMRegister dst, Address src) {Assembler::movlpd(dst, src); }
926 void movlpd(XMMRegister dst, AddressLiteral src);
927
928 public:
929
930 void addsd(XMMRegister dst, XMMRegister src) { Assembler::addsd(dst, src); }
931 void addsd(XMMRegister dst, Address src) { Assembler::addsd(dst, src); }
932 void addsd(XMMRegister dst, AddressLiteral src);
933
934 void addss(XMMRegister dst, XMMRegister src) { Assembler::addss(dst, src); }
935 void addss(XMMRegister dst, Address src) { Assembler::addss(dst, src); }
936 void addss(XMMRegister dst, AddressLiteral src);
937
938 void divsd(XMMRegister dst, XMMRegister src) { Assembler::divsd(dst, src); }
939 void divsd(XMMRegister dst, Address src) { Assembler::divsd(dst, src); }
940 void divsd(XMMRegister dst, AddressLiteral src);
941
942 void divss(XMMRegister dst, XMMRegister src) { Assembler::divss(dst, src); }
943 void divss(XMMRegister dst, Address src) { Assembler::divss(dst, src); }
944 void divss(XMMRegister dst, AddressLiteral src);
945
946 // Move Unaligned Double Quadword
947 void movdqu(Address dst, XMMRegister src) { Assembler::movdqu(dst, src); }
948 void movdqu(XMMRegister dst, Address src) { Assembler::movdqu(dst, src); }
949 void movdqu(XMMRegister dst, XMMRegister src) { Assembler::movdqu(dst, src); }
950 void movdqu(XMMRegister dst, AddressLiteral src);
951
952 // Move Aligned Double Quadword
953 void movdqa(XMMRegister dst, Address src) { Assembler::movdqa(dst, src); }
954 void movdqa(XMMRegister dst, XMMRegister src) { Assembler::movdqa(dst, src); }
955 void movdqa(XMMRegister dst, AddressLiteral src);
956
957 void movsd(XMMRegister dst, XMMRegister src) { Assembler::movsd(dst, src); }
958 void movsd(Address dst, XMMRegister src) { Assembler::movsd(dst, src); }
959 void movsd(XMMRegister dst, Address src) { Assembler::movsd(dst, src); }
960 void movsd(XMMRegister dst, AddressLiteral src);
961
962 void mulsd(XMMRegister dst, XMMRegister src) { Assembler::mulsd(dst, src); }
963 void mulsd(XMMRegister dst, Address src) { Assembler::mulsd(dst, src); }
964 void mulsd(XMMRegister dst, AddressLiteral src);
965
966 void mulss(XMMRegister dst, XMMRegister src) { Assembler::mulss(dst, src); }
967 void mulss(XMMRegister dst, Address src) { Assembler::mulss(dst, src); }
968 void mulss(XMMRegister dst, AddressLiteral src);
969
970 // Carry-Less Multiplication Quadword
971 void pclmulldq(XMMRegister dst, XMMRegister src) {
972 // 0x00 - multiply lower 64 bits [0:63]
973 Assembler::pclmulqdq(dst, src, 0x00);
974 }
975 void pclmulhdq(XMMRegister dst, XMMRegister src) {
976 // 0x11 - multiply upper 64 bits [64:127]
977 Assembler::pclmulqdq(dst, src, 0x11);
978 }
979
980 void sqrtsd(XMMRegister dst, XMMRegister src) { Assembler::sqrtsd(dst, src); }
981 void sqrtsd(XMMRegister dst, Address src) { Assembler::sqrtsd(dst, src); }
982 void sqrtsd(XMMRegister dst, AddressLiteral src);
983
984 void sqrtss(XMMRegister dst, XMMRegister src) { Assembler::sqrtss(dst, src); }
985 void sqrtss(XMMRegister dst, Address src) { Assembler::sqrtss(dst, src); }
986 void sqrtss(XMMRegister dst, AddressLiteral src);
987
988 void subsd(XMMRegister dst, XMMRegister src) { Assembler::subsd(dst, src); }
989 void subsd(XMMRegister dst, Address src) { Assembler::subsd(dst, src); }
990 void subsd(XMMRegister dst, AddressLiteral src);
991
992 void subss(XMMRegister dst, XMMRegister src) { Assembler::subss(dst, src); }
993 void subss(XMMRegister dst, Address src) { Assembler::subss(dst, src); }
994 void subss(XMMRegister dst, AddressLiteral src);
995
996 void ucomiss(XMMRegister dst, XMMRegister src) { Assembler::ucomiss(dst, src); }
997 void ucomiss(XMMRegister dst, Address src) { Assembler::ucomiss(dst, src); }
998 void ucomiss(XMMRegister dst, AddressLiteral src);
999
1000 void ucomisd(XMMRegister dst, XMMRegister src) { Assembler::ucomisd(dst, src); }
1001 void ucomisd(XMMRegister dst, Address src) { Assembler::ucomisd(dst, src); }
1002 void ucomisd(XMMRegister dst, AddressLiteral src);
1003
1004 // Bitwise Logical XOR of Packed Double-Precision Floating-Point Values
1005 void xorpd(XMMRegister dst, XMMRegister src) { Assembler::xorpd(dst, src); }
1006 void xorpd(XMMRegister dst, Address src) { Assembler::xorpd(dst, src); }
1007 void xorpd(XMMRegister dst, AddressLiteral src);
1008
1009 // Bitwise Logical XOR of Packed Single-Precision Floating-Point Values
1010 void xorps(XMMRegister dst, XMMRegister src) { Assembler::xorps(dst, src); }
1011 void xorps(XMMRegister dst, Address src) { Assembler::xorps(dst, src); }
1012 void xorps(XMMRegister dst, AddressLiteral src);
1013
1014 // Shuffle Bytes
1015 void pshufb(XMMRegister dst, XMMRegister src) { Assembler::pshufb(dst, src); }
1016 void pshufb(XMMRegister dst, Address src) { Assembler::pshufb(dst, src); }
1017 void pshufb(XMMRegister dst, AddressLiteral src);
1018 // AVX 3-operands instructions
1019
1020 void vaddsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vaddsd(dst, nds, src); }
1021 void vaddsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vaddsd(dst, nds, src); }
1022 void vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1023
1024 void vaddss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vaddss(dst, nds, src); }
1025 void vaddss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vaddss(dst, nds, src); }
1026 void vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1027
1028 void vandpd(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256) { Assembler::vandpd(dst, nds, src, vector256); }
1029 void vandpd(XMMRegister dst, XMMRegister nds, Address src, bool vector256) { Assembler::vandpd(dst, nds, src, vector256); }
1030 void vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, bool vector256);
1031
1032 void vandps(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256) { Assembler::vandps(dst, nds, src, vector256); }
1033 void vandps(XMMRegister dst, XMMRegister nds, Address src, bool vector256) { Assembler::vandps(dst, nds, src, vector256); }
1034 void vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, bool vector256);
1035
1036 void vdivsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vdivsd(dst, nds, src); }
1037 void vdivsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vdivsd(dst, nds, src); }
1038 void vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1039
1040 void vdivss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vdivss(dst, nds, src); }
1041 void vdivss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vdivss(dst, nds, src); }
1042 void vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1043
1044 void vmulsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vmulsd(dst, nds, src); }
1045 void vmulsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vmulsd(dst, nds, src); }
1046 void vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1047
1048 void vmulss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vmulss(dst, nds, src); }
1049 void vmulss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vmulss(dst, nds, src); }
1050 void vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1051
1052 void vsubsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vsubsd(dst, nds, src); }
1053 void vsubsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vsubsd(dst, nds, src); }
1054 void vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1055
1056 void vsubss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vsubss(dst, nds, src); }
1057 void vsubss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vsubss(dst, nds, src); }
1058 void vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1059
1060 // AVX Vector instructions
1061
1062 void vxorpd(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256) { Assembler::vxorpd(dst, nds, src, vector256); }
1063 void vxorpd(XMMRegister dst, XMMRegister nds, Address src, bool vector256) { Assembler::vxorpd(dst, nds, src, vector256); }
1064 void vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, bool vector256);
1065
1066 void vxorps(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256) { Assembler::vxorps(dst, nds, src, vector256); }
1067 void vxorps(XMMRegister dst, XMMRegister nds, Address src, bool vector256) { Assembler::vxorps(dst, nds, src, vector256); }
1068 void vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, bool vector256);
1069
1070 void vpxor(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256) {
1071 if (UseAVX > 1 || !vector256) // vpxor 256 bit is available only in AVX2
1072 Assembler::vpxor(dst, nds, src, vector256);
1073 else
1074 Assembler::vxorpd(dst, nds, src, vector256);
1075 }
1076 void vpxor(XMMRegister dst, XMMRegister nds, Address src, bool vector256) {
1077 if (UseAVX > 1 || !vector256) // vpxor 256 bit is available only in AVX2
1078 Assembler::vpxor(dst, nds, src, vector256);
1079 else
1080 Assembler::vxorpd(dst, nds, src, vector256);
1081 }
1082
1083 // Simple version for AVX2 256bit vectors
1084 void vpxor(XMMRegister dst, XMMRegister src) { Assembler::vpxor(dst, dst, src, true); }
1085 void vpxor(XMMRegister dst, Address src) { Assembler::vpxor(dst, dst, src, true); }
1086
1087 // Move packed integer values from low 128 bit to hign 128 bit in 256 bit vector.
1088 void vinserti128h(XMMRegister dst, XMMRegister nds, XMMRegister src) {
1089 if (UseAVX > 1) // vinserti128h is available only in AVX2
1090 Assembler::vinserti128h(dst, nds, src);
1091 else
1092 Assembler::vinsertf128h(dst, nds, src);
1093 }
1094
1095 // Carry-Less Multiplication Quadword
1096 void vpclmulldq(XMMRegister dst, XMMRegister nds, XMMRegister src) {
1097 // 0x00 - multiply lower 64 bits [0:63]
1098 Assembler::vpclmulqdq(dst, nds, src, 0x00);
1099 }
1100 void vpclmulhdq(XMMRegister dst, XMMRegister nds, XMMRegister src) {
1101 // 0x11 - multiply upper 64 bits [64:127]
1102 Assembler::vpclmulqdq(dst, nds, src, 0x11);
1103 }
1104
1105 // Data
1106
1107 void cmov32( Condition cc, Register dst, Address src);
1108 void cmov32( Condition cc, Register dst, Register src);
1109
1110 void cmov( Condition cc, Register dst, Register src) { cmovptr(cc, dst, src); }
1111
1112 void cmovptr(Condition cc, Register dst, Address src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmov32(cc, dst, src)); }
1113 void cmovptr(Condition cc, Register dst, Register src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmov32(cc, dst, src)); }
1114
1115 void movoop(Register dst, jobject obj);
1116 void movoop(Address dst, jobject obj);
1117
1118 void mov_metadata(Register dst, Metadata* obj);
1119 void mov_metadata(Address dst, Metadata* obj);
1120
1121 void movptr(ArrayAddress dst, Register src);
1122 // can this do an lea?
1123 void movptr(Register dst, ArrayAddress src);
1124
1125 void movptr(Register dst, Address src);
1126
1127 #ifdef _LP64
1128 void movptr(Register dst, AddressLiteral src, Register scratch=rscratch1);
1129 #else
1130 void movptr(Register dst, AddressLiteral src, Register scratch=noreg); // Scratch reg is ignored in 32-bit
1131 #endif
1132
1133 void movptr(Register dst, intptr_t src);
1134 void movptr(Register dst, Register src);
1135 void movptr(Address dst, intptr_t src);
1136
1137 void movptr(Address dst, Register src);
1138
1139 void movptr(Register dst, RegisterOrConstant src) {
1140 if (src.is_constant()) movptr(dst, src.as_constant());
1141 else movptr(dst, src.as_register());
1142 }
1143
1144 #ifdef _LP64
1145 // Generally the next two are only used for moving NULL
1146 // Although there are situations in initializing the mark word where
1147 // they could be used. They are dangerous.
1148
1149 // They only exist on LP64 so that int32_t and intptr_t are not the same
1150 // and we have ambiguous declarations.
1151
1152 void movptr(Address dst, int32_t imm32);
1153 void movptr(Register dst, int32_t imm32);
1154 #endif // _LP64
1155
1156 // to avoid hiding movl
1157 void mov32(AddressLiteral dst, Register src);
1158 void mov32(Register dst, AddressLiteral src);
1159
1160 // to avoid hiding movb
1161 void movbyte(ArrayAddress dst, int src);
1162
1163 // Import other mov() methods from the parent class or else
1164 // they will be hidden by the following overriding declaration.
1165 using Assembler::movdl;
1166 using Assembler::movq;
1167 void movdl(XMMRegister dst, AddressLiteral src);
1168 void movq(XMMRegister dst, AddressLiteral src);
1169
1170 // Can push value or effective address
1171 void pushptr(AddressLiteral src);
1172
1173 void pushptr(Address src) { LP64_ONLY(pushq(src)) NOT_LP64(pushl(src)); }
1174 void popptr(Address src) { LP64_ONLY(popq(src)) NOT_LP64(popl(src)); }
1175
1176 void pushoop(jobject obj);
1177 void pushklass(Metadata* obj);
1178
1179 // sign extend as need a l to ptr sized element
1180 void movl2ptr(Register dst, Address src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(movl(dst, src)); }
1181 void movl2ptr(Register dst, Register src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(if (dst != src) movl(dst, src)); }
1182
1183 // C2 compiled method's prolog code.
1184 void verified_entry(int framesize, int stack_bang_size, bool fp_mode_24b);
1185
1186 // clear memory of size 'cnt' qwords, starting at 'base'.
1187 void clear_mem(Register base, Register cnt, Register rtmp);
1188
1189 // IndexOf strings.
1190 // Small strings are loaded through stack if they cross page boundary.
1191 void string_indexof(Register str1, Register str2,
1192 Register cnt1, Register cnt2,
1193 int int_cnt2, Register result,
1194 XMMRegister vec, Register tmp);
1195
1196 // IndexOf for constant substrings with size >= 8 elements
1197 // which don't need to be loaded through stack.
1198 void string_indexofC8(Register str1, Register str2,
1199 Register cnt1, Register cnt2,
1200 int int_cnt2, Register result,
1201 XMMRegister vec, Register tmp);
1202
1203 // Smallest code: we don't need to load through stack,
1204 // check string tail.
1205
1206 // Compare strings.
1207 void string_compare(Register str1, Register str2,
1208 Register cnt1, Register cnt2, Register result,
1209 XMMRegister vec1);
1210
1211 // Compare char[] arrays.
1212 void char_arrays_equals(bool is_array_equ, Register ary1, Register ary2,
1213 Register limit, Register result, Register chr,
1214 XMMRegister vec1, XMMRegister vec2);
1215
1216 // Fill primitive arrays
1217 void generate_fill(BasicType t, bool aligned,
1218 Register to, Register value, Register count,
1219 Register rtmp, XMMRegister xtmp);
1220
1221 void encode_iso_array(Register src, Register dst, Register len,
1222 XMMRegister tmp1, XMMRegister tmp2, XMMRegister tmp3,
1223 XMMRegister tmp4, Register tmp5, Register result);
1224
1225 #ifdef _LP64
1226 void add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2);
1227 void multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
1228 Register y, Register y_idx, Register z,
1229 Register carry, Register product,
1230 Register idx, Register kdx);
1231 void multiply_add_128_x_128(Register x_xstart, Register y, Register z,
1232 Register yz_idx, Register idx,
1233 Register carry, Register product, int offset);
1234 void multiply_128_x_128_bmi2_loop(Register y, Register z,
1235 Register carry, Register carry2,
1236 Register idx, Register jdx,
1237 Register yz_idx1, Register yz_idx2,
1238 Register tmp, Register tmp3, Register tmp4);
1239 void multiply_128_x_128_loop(Register x_xstart, Register y, Register z,
1240 Register yz_idx, Register idx, Register jdx,
1241 Register carry, Register product,
1242 Register carry2);
1243 void multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register zlen,
1244 Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5);
1245
1246 void square_rshift(Register x, Register len, Register z, Register tmp1, Register tmp3,
1247 Register tmp4, Register tmp5, Register rdxReg, Register raxReg);
1248 void multiply_add_64_bmi2(Register sum, Register op1, Register op2, Register carry,
1249 Register tmp2);
1250 void multiply_add_64(Register sum, Register op1, Register op2, Register carry,
1251 Register rdxReg, Register raxReg);
1252 void add_one_64(Register z, Register zlen, Register carry, Register tmp1);
1253 void lshift_by_1(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2,
1254 Register tmp3, Register tmp4);
1255 void square_to_len(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2,
1256 Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg);
1257
1258 void mul_add_128_x_32_loop(Register out, Register in, Register offset, Register len, Register tmp1,
1259 Register tmp2, Register tmp3, Register tmp4, Register tmp5, Register rdxReg,
1260 Register raxReg);
1261 void mul_add(Register out, Register in, Register offset, Register len, Register k, Register tmp1,
1262 Register tmp2, Register tmp3, Register tmp4, Register tmp5, Register rdxReg,
1263 Register raxReg);
1264 #endif
1265
1266 // CRC32 code for java.util.zip.CRC32::updateBytes() instrinsic.
1267 void update_byte_crc32(Register crc, Register val, Register table);
1268 void kernel_crc32(Register crc, Register buf, Register len, Register table, Register tmp);
1269 // Fold 128-bit data chunk
1270 void fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset);
1271 void fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, XMMRegister xbuf);
1272 // Fold 8-bit data
1273 void fold_8bit_crc32(Register crc, Register table, Register tmp);
1274 void fold_8bit_crc32(XMMRegister crc, Register table, XMMRegister xtmp, Register tmp);
1275
1276 #undef VIRTUAL
1277
1278 };
1279
1280 /**
1281 * class SkipIfEqual:
1282 *
1283 * Instantiating this class will result in assembly code being output that will
1284 * jump around any code emitted between the creation of the instance and it's
1285 * automatic destruction at the end of a scope block, depending on the value of
1286 * the flag passed to the constructor, which will be checked at run-time.
1287 */
1288 class SkipIfEqual {
1289 private:
1290 MacroAssembler* _masm;
1291 Label _label;
1292
1293 public:
1294 SkipIfEqual(MacroAssembler*, const bool* flag_addr, bool value);
1295 ~SkipIfEqual();
1296 };
1297
1298 #endif // CPU_X86_VM_MACROASSEMBLER_X86_HPP