Tue, 09 Mar 2010 20:16:19 +0100
6919934: JSR 292 needs to support x86 C1
Summary: This implements JSR 292 support for C1 x86.
Reviewed-by: never, jrose, kvn
1 /*
2 * Copyright 1999-2010 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
20 * CA 95054 USA or visit www.sun.com if you need additional information or
21 * have any questions.
22 *
23 */
25 #include "incls/_precompiled.incl"
26 #include "incls/_stubGenerator_x86_32.cpp.incl"
28 // Declaration and definition of StubGenerator (no .hpp file).
29 // For a more detailed description of the stub routine structure
30 // see the comment in stubRoutines.hpp
32 #define __ _masm->
33 #define a__ ((Assembler*)_masm)->
35 #ifdef PRODUCT
36 #define BLOCK_COMMENT(str) /* nothing */
37 #else
38 #define BLOCK_COMMENT(str) __ block_comment(str)
39 #endif
41 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
43 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
44 const int FPU_CNTRL_WRD_MASK = 0xFFFF;
46 // -------------------------------------------------------------------------------------------------------------------------
47 // Stub Code definitions
49 static address handle_unsafe_access() {
50 JavaThread* thread = JavaThread::current();
51 address pc = thread->saved_exception_pc();
52 // pc is the instruction which we must emulate
53 // doing a no-op is fine: return garbage from the load
54 // therefore, compute npc
55 address npc = Assembler::locate_next_instruction(pc);
57 // request an async exception
58 thread->set_pending_unsafe_access_error();
60 // return address of next instruction to execute
61 return npc;
62 }
64 class StubGenerator: public StubCodeGenerator {
65 private:
67 #ifdef PRODUCT
68 #define inc_counter_np(counter) (0)
69 #else
70 void inc_counter_np_(int& counter) {
71 __ incrementl(ExternalAddress((address)&counter));
72 }
73 #define inc_counter_np(counter) \
74 BLOCK_COMMENT("inc_counter " #counter); \
75 inc_counter_np_(counter);
76 #endif //PRODUCT
78 void inc_copy_counter_np(BasicType t) {
79 #ifndef PRODUCT
80 switch (t) {
81 case T_BYTE: inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); return;
82 case T_SHORT: inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); return;
83 case T_INT: inc_counter_np(SharedRuntime::_jint_array_copy_ctr); return;
84 case T_LONG: inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); return;
85 case T_OBJECT: inc_counter_np(SharedRuntime::_oop_array_copy_ctr); return;
86 }
87 ShouldNotReachHere();
88 #endif //PRODUCT
89 }
91 //------------------------------------------------------------------------------------------------------------------------
92 // Call stubs are used to call Java from C
93 //
94 // [ return_from_Java ] <--- rsp
95 // [ argument word n ]
96 // ...
97 // -N [ argument word 1 ]
98 // -7 [ Possible padding for stack alignment ]
99 // -6 [ Possible padding for stack alignment ]
100 // -5 [ Possible padding for stack alignment ]
101 // -4 [ mxcsr save ] <--- rsp_after_call
102 // -3 [ saved rbx, ]
103 // -2 [ saved rsi ]
104 // -1 [ saved rdi ]
105 // 0 [ saved rbp, ] <--- rbp,
106 // 1 [ return address ]
107 // 2 [ ptr. to call wrapper ]
108 // 3 [ result ]
109 // 4 [ result_type ]
110 // 5 [ method ]
111 // 6 [ entry_point ]
112 // 7 [ parameters ]
113 // 8 [ parameter_size ]
114 // 9 [ thread ]
117 address generate_call_stub(address& return_address) {
118 StubCodeMark mark(this, "StubRoutines", "call_stub");
119 address start = __ pc();
121 // stub code parameters / addresses
122 assert(frame::entry_frame_call_wrapper_offset == 2, "adjust this code");
123 bool sse_save = false;
124 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_catch_exception()!
125 const int locals_count_in_bytes (4*wordSize);
126 const Address mxcsr_save (rbp, -4 * wordSize);
127 const Address saved_rbx (rbp, -3 * wordSize);
128 const Address saved_rsi (rbp, -2 * wordSize);
129 const Address saved_rdi (rbp, -1 * wordSize);
130 const Address result (rbp, 3 * wordSize);
131 const Address result_type (rbp, 4 * wordSize);
132 const Address method (rbp, 5 * wordSize);
133 const Address entry_point (rbp, 6 * wordSize);
134 const Address parameters (rbp, 7 * wordSize);
135 const Address parameter_size(rbp, 8 * wordSize);
136 const Address thread (rbp, 9 * wordSize); // same as in generate_catch_exception()!
137 sse_save = UseSSE > 0;
139 // stub code
140 __ enter();
141 __ movptr(rcx, parameter_size); // parameter counter
142 __ shlptr(rcx, Interpreter::logStackElementSize()); // convert parameter count to bytes
143 __ addptr(rcx, locals_count_in_bytes); // reserve space for register saves
144 __ subptr(rsp, rcx);
145 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack
147 // save rdi, rsi, & rbx, according to C calling conventions
148 __ movptr(saved_rdi, rdi);
149 __ movptr(saved_rsi, rsi);
150 __ movptr(saved_rbx, rbx);
151 // save and initialize %mxcsr
152 if (sse_save) {
153 Label skip_ldmx;
154 __ stmxcsr(mxcsr_save);
155 __ movl(rax, mxcsr_save);
156 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
157 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
158 __ cmp32(rax, mxcsr_std);
159 __ jcc(Assembler::equal, skip_ldmx);
160 __ ldmxcsr(mxcsr_std);
161 __ bind(skip_ldmx);
162 }
164 // make sure the control word is correct.
165 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
167 #ifdef ASSERT
168 // make sure we have no pending exceptions
169 { Label L;
170 __ movptr(rcx, thread);
171 __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
172 __ jcc(Assembler::equal, L);
173 __ stop("StubRoutines::call_stub: entered with pending exception");
174 __ bind(L);
175 }
176 #endif
178 // pass parameters if any
179 BLOCK_COMMENT("pass parameters if any");
180 Label parameters_done;
181 __ movl(rcx, parameter_size); // parameter counter
182 __ testl(rcx, rcx);
183 __ jcc(Assembler::zero, parameters_done);
185 // parameter passing loop
187 Label loop;
188 // Copy Java parameters in reverse order (receiver last)
189 // Note that the argument order is inverted in the process
190 // source is rdx[rcx: N-1..0]
191 // dest is rsp[rbx: 0..N-1]
193 __ movptr(rdx, parameters); // parameter pointer
194 __ xorptr(rbx, rbx);
196 __ BIND(loop);
197 if (TaggedStackInterpreter) {
198 __ movptr(rax, Address(rdx, rcx, Interpreter::stackElementScale(),
199 -2*wordSize)); // get tag
200 __ movptr(Address(rsp, rbx, Interpreter::stackElementScale(),
201 Interpreter::expr_tag_offset_in_bytes(0)), rax); // store tag
202 }
204 // get parameter
205 __ movptr(rax, Address(rdx, rcx, Interpreter::stackElementScale(), -wordSize));
206 __ movptr(Address(rsp, rbx, Interpreter::stackElementScale(),
207 Interpreter::expr_offset_in_bytes(0)), rax); // store parameter
208 __ increment(rbx);
209 __ decrement(rcx);
210 __ jcc(Assembler::notZero, loop);
212 // call Java function
213 __ BIND(parameters_done);
214 __ movptr(rbx, method); // get methodOop
215 __ movptr(rax, entry_point); // get entry_point
216 __ mov(rsi, rsp); // set sender sp
217 BLOCK_COMMENT("call Java function");
218 __ call(rax);
220 BLOCK_COMMENT("call_stub_return_address:");
221 return_address = __ pc();
223 Label common_return;
225 __ BIND(common_return);
227 // store result depending on type
228 // (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
229 __ movptr(rdi, result);
230 Label is_long, is_float, is_double, exit;
231 __ movl(rsi, result_type);
232 __ cmpl(rsi, T_LONG);
233 __ jcc(Assembler::equal, is_long);
234 __ cmpl(rsi, T_FLOAT);
235 __ jcc(Assembler::equal, is_float);
236 __ cmpl(rsi, T_DOUBLE);
237 __ jcc(Assembler::equal, is_double);
239 // handle T_INT case
240 __ movl(Address(rdi, 0), rax);
241 __ BIND(exit);
243 // check that FPU stack is empty
244 __ verify_FPU(0, "generate_call_stub");
246 // pop parameters
247 __ lea(rsp, rsp_after_call);
249 // restore %mxcsr
250 if (sse_save) {
251 __ ldmxcsr(mxcsr_save);
252 }
254 // restore rdi, rsi and rbx,
255 __ movptr(rbx, saved_rbx);
256 __ movptr(rsi, saved_rsi);
257 __ movptr(rdi, saved_rdi);
258 __ addptr(rsp, 4*wordSize);
260 // return
261 __ pop(rbp);
262 __ ret(0);
264 // handle return types different from T_INT
265 __ BIND(is_long);
266 __ movl(Address(rdi, 0 * wordSize), rax);
267 __ movl(Address(rdi, 1 * wordSize), rdx);
268 __ jmp(exit);
270 __ BIND(is_float);
271 // interpreter uses xmm0 for return values
272 if (UseSSE >= 1) {
273 __ movflt(Address(rdi, 0), xmm0);
274 } else {
275 __ fstp_s(Address(rdi, 0));
276 }
277 __ jmp(exit);
279 __ BIND(is_double);
280 // interpreter uses xmm0 for return values
281 if (UseSSE >= 2) {
282 __ movdbl(Address(rdi, 0), xmm0);
283 } else {
284 __ fstp_d(Address(rdi, 0));
285 }
286 __ jmp(exit);
288 // If we call compiled code directly from the call stub we will
289 // need to adjust the return back to the call stub to a specialized
290 // piece of code that can handle compiled results and cleaning the fpu
291 // stack. compiled code will be set to return here instead of the
292 // return above that handles interpreter returns.
294 BLOCK_COMMENT("call_stub_compiled_return:");
295 StubRoutines::x86::set_call_stub_compiled_return( __ pc());
297 #ifdef COMPILER2
298 if (UseSSE >= 2) {
299 __ verify_FPU(0, "call_stub_compiled_return");
300 } else {
301 for (int i = 1; i < 8; i++) {
302 __ ffree(i);
303 }
305 // UseSSE <= 1 so double result should be left on TOS
306 __ movl(rsi, result_type);
307 __ cmpl(rsi, T_DOUBLE);
308 __ jcc(Assembler::equal, common_return);
309 if (UseSSE == 0) {
310 // UseSSE == 0 so float result should be left on TOS
311 __ cmpl(rsi, T_FLOAT);
312 __ jcc(Assembler::equal, common_return);
313 }
314 __ ffree(0);
315 }
316 #endif /* COMPILER2 */
317 __ jmp(common_return);
319 return start;
320 }
323 //------------------------------------------------------------------------------------------------------------------------
324 // Return point for a Java call if there's an exception thrown in Java code.
325 // The exception is caught and transformed into a pending exception stored in
326 // JavaThread that can be tested from within the VM.
327 //
328 // Note: Usually the parameters are removed by the callee. In case of an exception
329 // crossing an activation frame boundary, that is not the case if the callee
330 // is compiled code => need to setup the rsp.
331 //
332 // rax,: exception oop
334 address generate_catch_exception() {
335 StubCodeMark mark(this, "StubRoutines", "catch_exception");
336 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_call_stub()!
337 const Address thread (rbp, 9 * wordSize); // same as in generate_call_stub()!
338 address start = __ pc();
340 // get thread directly
341 __ movptr(rcx, thread);
342 #ifdef ASSERT
343 // verify that threads correspond
344 { Label L;
345 __ get_thread(rbx);
346 __ cmpptr(rbx, rcx);
347 __ jcc(Assembler::equal, L);
348 __ stop("StubRoutines::catch_exception: threads must correspond");
349 __ bind(L);
350 }
351 #endif
352 // set pending exception
353 __ verify_oop(rax);
354 __ movptr(Address(rcx, Thread::pending_exception_offset()), rax );
355 __ lea(Address(rcx, Thread::exception_file_offset ()),
356 ExternalAddress((address)__FILE__));
357 __ movl(Address(rcx, Thread::exception_line_offset ()), __LINE__ );
358 // complete return to VM
359 assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before");
360 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
362 return start;
363 }
366 //------------------------------------------------------------------------------------------------------------------------
367 // Continuation point for runtime calls returning with a pending exception.
368 // The pending exception check happened in the runtime or native call stub.
369 // The pending exception in Thread is converted into a Java-level exception.
370 //
371 // Contract with Java-level exception handlers:
372 // rax: exception
373 // rdx: throwing pc
374 //
375 // NOTE: At entry of this stub, exception-pc must be on stack !!
377 address generate_forward_exception() {
378 StubCodeMark mark(this, "StubRoutines", "forward exception");
379 address start = __ pc();
380 const Register thread = rcx;
382 // other registers used in this stub
383 const Register exception_oop = rax;
384 const Register handler_addr = rbx;
385 const Register exception_pc = rdx;
387 // Upon entry, the sp points to the return address returning into Java
388 // (interpreted or compiled) code; i.e., the return address becomes the
389 // throwing pc.
390 //
391 // Arguments pushed before the runtime call are still on the stack but
392 // the exception handler will reset the stack pointer -> ignore them.
393 // A potential result in registers can be ignored as well.
395 #ifdef ASSERT
396 // make sure this code is only executed if there is a pending exception
397 { Label L;
398 __ get_thread(thread);
399 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
400 __ jcc(Assembler::notEqual, L);
401 __ stop("StubRoutines::forward exception: no pending exception (1)");
402 __ bind(L);
403 }
404 #endif
406 // compute exception handler into rbx,
407 __ get_thread(thread);
408 __ movptr(exception_pc, Address(rsp, 0));
409 BLOCK_COMMENT("call exception_handler_for_return_address");
410 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, exception_pc);
411 __ mov(handler_addr, rax);
413 // setup rax & rdx, remove return address & clear pending exception
414 __ get_thread(thread);
415 __ pop(exception_pc);
416 __ movptr(exception_oop, Address(thread, Thread::pending_exception_offset()));
417 __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
419 #ifdef ASSERT
420 // make sure exception is set
421 { Label L;
422 __ testptr(exception_oop, exception_oop);
423 __ jcc(Assembler::notEqual, L);
424 __ stop("StubRoutines::forward exception: no pending exception (2)");
425 __ bind(L);
426 }
427 #endif
429 // Verify that there is really a valid exception in RAX.
430 __ verify_oop(exception_oop);
432 // Restore SP from BP if the exception PC is a MethodHandle call site.
433 __ cmpl(Address(thread, JavaThread::is_method_handle_exception_offset()), 0);
434 __ cmovptr(Assembler::notEqual, rsp, rbp);
436 // continue at exception handler (return address removed)
437 // rax: exception
438 // rbx: exception handler
439 // rdx: throwing pc
440 __ jmp(handler_addr);
442 return start;
443 }
446 //----------------------------------------------------------------------------------------------------
447 // Support for jint Atomic::xchg(jint exchange_value, volatile jint* dest)
448 //
449 // xchg exists as far back as 8086, lock needed for MP only
450 // Stack layout immediately after call:
451 //
452 // 0 [ret addr ] <--- rsp
453 // 1 [ ex ]
454 // 2 [ dest ]
455 //
456 // Result: *dest <- ex, return (old *dest)
457 //
458 // Note: win32 does not currently use this code
460 address generate_atomic_xchg() {
461 StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
462 address start = __ pc();
464 __ push(rdx);
465 Address exchange(rsp, 2 * wordSize);
466 Address dest_addr(rsp, 3 * wordSize);
467 __ movl(rax, exchange);
468 __ movptr(rdx, dest_addr);
469 __ xchgl(rax, Address(rdx, 0));
470 __ pop(rdx);
471 __ ret(0);
473 return start;
474 }
476 //----------------------------------------------------------------------------------------------------
477 // Support for void verify_mxcsr()
478 //
479 // This routine is used with -Xcheck:jni to verify that native
480 // JNI code does not return to Java code without restoring the
481 // MXCSR register to our expected state.
484 address generate_verify_mxcsr() {
485 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
486 address start = __ pc();
488 const Address mxcsr_save(rsp, 0);
490 if (CheckJNICalls && UseSSE > 0 ) {
491 Label ok_ret;
492 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
493 __ push(rax);
494 __ subptr(rsp, wordSize); // allocate a temp location
495 __ stmxcsr(mxcsr_save);
496 __ movl(rax, mxcsr_save);
497 __ andl(rax, MXCSR_MASK);
498 __ cmp32(rax, mxcsr_std);
499 __ jcc(Assembler::equal, ok_ret);
501 __ warn("MXCSR changed by native JNI code.");
503 __ ldmxcsr(mxcsr_std);
505 __ bind(ok_ret);
506 __ addptr(rsp, wordSize);
507 __ pop(rax);
508 }
510 __ ret(0);
512 return start;
513 }
516 //---------------------------------------------------------------------------
517 // Support for void verify_fpu_cntrl_wrd()
518 //
519 // This routine is used with -Xcheck:jni to verify that native
520 // JNI code does not return to Java code without restoring the
521 // FP control word to our expected state.
523 address generate_verify_fpu_cntrl_wrd() {
524 StubCodeMark mark(this, "StubRoutines", "verify_spcw");
525 address start = __ pc();
527 const Address fpu_cntrl_wrd_save(rsp, 0);
529 if (CheckJNICalls) {
530 Label ok_ret;
531 __ push(rax);
532 __ subptr(rsp, wordSize); // allocate a temp location
533 __ fnstcw(fpu_cntrl_wrd_save);
534 __ movl(rax, fpu_cntrl_wrd_save);
535 __ andl(rax, FPU_CNTRL_WRD_MASK);
536 ExternalAddress fpu_std(StubRoutines::addr_fpu_cntrl_wrd_std());
537 __ cmp32(rax, fpu_std);
538 __ jcc(Assembler::equal, ok_ret);
540 __ warn("Floating point control word changed by native JNI code.");
542 __ fldcw(fpu_std);
544 __ bind(ok_ret);
545 __ addptr(rsp, wordSize);
546 __ pop(rax);
547 }
549 __ ret(0);
551 return start;
552 }
554 //---------------------------------------------------------------------------
555 // Wrapper for slow-case handling of double-to-integer conversion
556 // d2i or f2i fast case failed either because it is nan or because
557 // of under/overflow.
558 // Input: FPU TOS: float value
559 // Output: rax, (rdx): integer (long) result
561 address generate_d2i_wrapper(BasicType t, address fcn) {
562 StubCodeMark mark(this, "StubRoutines", "d2i_wrapper");
563 address start = __ pc();
565 // Capture info about frame layout
566 enum layout { FPUState_off = 0,
567 rbp_off = FPUStateSizeInWords,
568 rdi_off,
569 rsi_off,
570 rcx_off,
571 rbx_off,
572 saved_argument_off,
573 saved_argument_off2, // 2nd half of double
574 framesize
575 };
577 assert(FPUStateSizeInWords == 27, "update stack layout");
579 // Save outgoing argument to stack across push_FPU_state()
580 __ subptr(rsp, wordSize * 2);
581 __ fstp_d(Address(rsp, 0));
583 // Save CPU & FPU state
584 __ push(rbx);
585 __ push(rcx);
586 __ push(rsi);
587 __ push(rdi);
588 __ push(rbp);
589 __ push_FPU_state();
591 // push_FPU_state() resets the FP top of stack
592 // Load original double into FP top of stack
593 __ fld_d(Address(rsp, saved_argument_off * wordSize));
594 // Store double into stack as outgoing argument
595 __ subptr(rsp, wordSize*2);
596 __ fst_d(Address(rsp, 0));
598 // Prepare FPU for doing math in C-land
599 __ empty_FPU_stack();
600 // Call the C code to massage the double. Result in EAX
601 if (t == T_INT)
602 { BLOCK_COMMENT("SharedRuntime::d2i"); }
603 else if (t == T_LONG)
604 { BLOCK_COMMENT("SharedRuntime::d2l"); }
605 __ call_VM_leaf( fcn, 2 );
607 // Restore CPU & FPU state
608 __ pop_FPU_state();
609 __ pop(rbp);
610 __ pop(rdi);
611 __ pop(rsi);
612 __ pop(rcx);
613 __ pop(rbx);
614 __ addptr(rsp, wordSize * 2);
616 __ ret(0);
618 return start;
619 }
622 //---------------------------------------------------------------------------
623 // The following routine generates a subroutine to throw an asynchronous
624 // UnknownError when an unsafe access gets a fault that could not be
625 // reasonably prevented by the programmer. (Example: SIGBUS/OBJERR.)
626 address generate_handler_for_unsafe_access() {
627 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
628 address start = __ pc();
630 __ push(0); // hole for return address-to-be
631 __ pusha(); // push registers
632 Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
633 BLOCK_COMMENT("call handle_unsafe_access");
634 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
635 __ movptr(next_pc, rax); // stuff next address
636 __ popa();
637 __ ret(0); // jump to next address
639 return start;
640 }
643 //----------------------------------------------------------------------------------------------------
644 // Non-destructive plausibility checks for oops
646 address generate_verify_oop() {
647 StubCodeMark mark(this, "StubRoutines", "verify_oop");
648 address start = __ pc();
650 // Incoming arguments on stack after saving rax,:
651 //
652 // [tos ]: saved rdx
653 // [tos + 1]: saved EFLAGS
654 // [tos + 2]: return address
655 // [tos + 3]: char* error message
656 // [tos + 4]: oop object to verify
657 // [tos + 5]: saved rax, - saved by caller and bashed
659 Label exit, error;
660 __ pushf();
661 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
662 __ push(rdx); // save rdx
663 // make sure object is 'reasonable'
664 __ movptr(rax, Address(rsp, 4 * wordSize)); // get object
665 __ testptr(rax, rax);
666 __ jcc(Assembler::zero, exit); // if obj is NULL it is ok
668 // Check if the oop is in the right area of memory
669 const int oop_mask = Universe::verify_oop_mask();
670 const int oop_bits = Universe::verify_oop_bits();
671 __ mov(rdx, rax);
672 __ andptr(rdx, oop_mask);
673 __ cmpptr(rdx, oop_bits);
674 __ jcc(Assembler::notZero, error);
676 // make sure klass is 'reasonable'
677 __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass
678 __ testptr(rax, rax);
679 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
681 // Check if the klass is in the right area of memory
682 const int klass_mask = Universe::verify_klass_mask();
683 const int klass_bits = Universe::verify_klass_bits();
684 __ mov(rdx, rax);
685 __ andptr(rdx, klass_mask);
686 __ cmpptr(rdx, klass_bits);
687 __ jcc(Assembler::notZero, error);
689 // make sure klass' klass is 'reasonable'
690 __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass' klass
691 __ testptr(rax, rax);
692 __ jcc(Assembler::zero, error); // if klass' klass is NULL it is broken
694 __ mov(rdx, rax);
695 __ andptr(rdx, klass_mask);
696 __ cmpptr(rdx, klass_bits);
697 __ jcc(Assembler::notZero, error); // if klass not in right area
698 // of memory it is broken too.
700 // return if everything seems ok
701 __ bind(exit);
702 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
703 __ pop(rdx); // restore rdx
704 __ popf(); // restore EFLAGS
705 __ ret(3 * wordSize); // pop arguments
707 // handle errors
708 __ bind(error);
709 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
710 __ pop(rdx); // get saved rdx back
711 __ popf(); // get saved EFLAGS off stack -- will be ignored
712 __ pusha(); // push registers (eip = return address & msg are already pushed)
713 BLOCK_COMMENT("call MacroAssembler::debug");
714 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32)));
715 __ popa();
716 __ ret(3 * wordSize); // pop arguments
717 return start;
718 }
720 //
721 // Generate pre-barrier for array stores
722 //
723 // Input:
724 // start - starting address
725 // count - element count
726 void gen_write_ref_array_pre_barrier(Register start, Register count) {
727 assert_different_registers(start, count);
728 BarrierSet* bs = Universe::heap()->barrier_set();
729 switch (bs->kind()) {
730 case BarrierSet::G1SATBCT:
731 case BarrierSet::G1SATBCTLogging:
732 {
733 __ pusha(); // push registers
734 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre),
735 start, count);
736 __ popa();
737 }
738 break;
739 case BarrierSet::CardTableModRef:
740 case BarrierSet::CardTableExtension:
741 case BarrierSet::ModRef:
742 break;
743 default :
744 ShouldNotReachHere();
746 }
747 }
750 //
751 // Generate a post-barrier for an array store
752 //
753 // start - starting address
754 // count - element count
755 //
756 // The two input registers are overwritten.
757 //
758 void gen_write_ref_array_post_barrier(Register start, Register count) {
759 BarrierSet* bs = Universe::heap()->barrier_set();
760 assert_different_registers(start, count);
761 switch (bs->kind()) {
762 case BarrierSet::G1SATBCT:
763 case BarrierSet::G1SATBCTLogging:
764 {
765 __ pusha(); // push registers
766 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post),
767 start, count);
768 __ popa();
769 }
770 break;
772 case BarrierSet::CardTableModRef:
773 case BarrierSet::CardTableExtension:
774 {
775 CardTableModRefBS* ct = (CardTableModRefBS*)bs;
776 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
778 Label L_loop;
779 const Register end = count; // elements count; end == start+count-1
780 assert_different_registers(start, end);
782 __ lea(end, Address(start, count, Address::times_ptr, -wordSize));
783 __ shrptr(start, CardTableModRefBS::card_shift);
784 __ shrptr(end, CardTableModRefBS::card_shift);
785 __ subptr(end, start); // end --> count
786 __ BIND(L_loop);
787 intptr_t disp = (intptr_t) ct->byte_map_base;
788 Address cardtable(start, count, Address::times_1, disp);
789 __ movb(cardtable, 0);
790 __ decrement(count);
791 __ jcc(Assembler::greaterEqual, L_loop);
792 }
793 break;
794 case BarrierSet::ModRef:
795 break;
796 default :
797 ShouldNotReachHere();
799 }
800 }
803 // Copy 64 bytes chunks
804 //
805 // Inputs:
806 // from - source array address
807 // to_from - destination array address - from
808 // qword_count - 8-bytes element count, negative
809 //
810 void xmm_copy_forward(Register from, Register to_from, Register qword_count) {
811 assert( UseSSE >= 2, "supported cpu only" );
812 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
813 // Copy 64-byte chunks
814 __ jmpb(L_copy_64_bytes);
815 __ align(16);
816 __ BIND(L_copy_64_bytes_loop);
818 if(UseUnalignedLoadStores) {
819 __ movdqu(xmm0, Address(from, 0));
820 __ movdqu(Address(from, to_from, Address::times_1, 0), xmm0);
821 __ movdqu(xmm1, Address(from, 16));
822 __ movdqu(Address(from, to_from, Address::times_1, 16), xmm1);
823 __ movdqu(xmm2, Address(from, 32));
824 __ movdqu(Address(from, to_from, Address::times_1, 32), xmm2);
825 __ movdqu(xmm3, Address(from, 48));
826 __ movdqu(Address(from, to_from, Address::times_1, 48), xmm3);
828 } else {
829 __ movq(xmm0, Address(from, 0));
830 __ movq(Address(from, to_from, Address::times_1, 0), xmm0);
831 __ movq(xmm1, Address(from, 8));
832 __ movq(Address(from, to_from, Address::times_1, 8), xmm1);
833 __ movq(xmm2, Address(from, 16));
834 __ movq(Address(from, to_from, Address::times_1, 16), xmm2);
835 __ movq(xmm3, Address(from, 24));
836 __ movq(Address(from, to_from, Address::times_1, 24), xmm3);
837 __ movq(xmm4, Address(from, 32));
838 __ movq(Address(from, to_from, Address::times_1, 32), xmm4);
839 __ movq(xmm5, Address(from, 40));
840 __ movq(Address(from, to_from, Address::times_1, 40), xmm5);
841 __ movq(xmm6, Address(from, 48));
842 __ movq(Address(from, to_from, Address::times_1, 48), xmm6);
843 __ movq(xmm7, Address(from, 56));
844 __ movq(Address(from, to_from, Address::times_1, 56), xmm7);
845 }
847 __ addl(from, 64);
848 __ BIND(L_copy_64_bytes);
849 __ subl(qword_count, 8);
850 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
851 __ addl(qword_count, 8);
852 __ jccb(Assembler::zero, L_exit);
853 //
854 // length is too short, just copy qwords
855 //
856 __ BIND(L_copy_8_bytes);
857 __ movq(xmm0, Address(from, 0));
858 __ movq(Address(from, to_from, Address::times_1), xmm0);
859 __ addl(from, 8);
860 __ decrement(qword_count);
861 __ jcc(Assembler::greater, L_copy_8_bytes);
862 __ BIND(L_exit);
863 }
865 // Copy 64 bytes chunks
866 //
867 // Inputs:
868 // from - source array address
869 // to_from - destination array address - from
870 // qword_count - 8-bytes element count, negative
871 //
872 void mmx_copy_forward(Register from, Register to_from, Register qword_count) {
873 assert( VM_Version::supports_mmx(), "supported cpu only" );
874 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
875 // Copy 64-byte chunks
876 __ jmpb(L_copy_64_bytes);
877 __ align(16);
878 __ BIND(L_copy_64_bytes_loop);
879 __ movq(mmx0, Address(from, 0));
880 __ movq(mmx1, Address(from, 8));
881 __ movq(mmx2, Address(from, 16));
882 __ movq(Address(from, to_from, Address::times_1, 0), mmx0);
883 __ movq(mmx3, Address(from, 24));
884 __ movq(Address(from, to_from, Address::times_1, 8), mmx1);
885 __ movq(mmx4, Address(from, 32));
886 __ movq(Address(from, to_from, Address::times_1, 16), mmx2);
887 __ movq(mmx5, Address(from, 40));
888 __ movq(Address(from, to_from, Address::times_1, 24), mmx3);
889 __ movq(mmx6, Address(from, 48));
890 __ movq(Address(from, to_from, Address::times_1, 32), mmx4);
891 __ movq(mmx7, Address(from, 56));
892 __ movq(Address(from, to_from, Address::times_1, 40), mmx5);
893 __ movq(Address(from, to_from, Address::times_1, 48), mmx6);
894 __ movq(Address(from, to_from, Address::times_1, 56), mmx7);
895 __ addptr(from, 64);
896 __ BIND(L_copy_64_bytes);
897 __ subl(qword_count, 8);
898 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
899 __ addl(qword_count, 8);
900 __ jccb(Assembler::zero, L_exit);
901 //
902 // length is too short, just copy qwords
903 //
904 __ BIND(L_copy_8_bytes);
905 __ movq(mmx0, Address(from, 0));
906 __ movq(Address(from, to_from, Address::times_1), mmx0);
907 __ addptr(from, 8);
908 __ decrement(qword_count);
909 __ jcc(Assembler::greater, L_copy_8_bytes);
910 __ BIND(L_exit);
911 __ emms();
912 }
914 address generate_disjoint_copy(BasicType t, bool aligned,
915 Address::ScaleFactor sf,
916 address* entry, const char *name) {
917 __ align(CodeEntryAlignment);
918 StubCodeMark mark(this, "StubRoutines", name);
919 address start = __ pc();
921 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
922 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes;
924 int shift = Address::times_ptr - sf;
926 const Register from = rsi; // source array address
927 const Register to = rdi; // destination array address
928 const Register count = rcx; // elements count
929 const Register to_from = to; // (to - from)
930 const Register saved_to = rdx; // saved destination array address
932 __ enter(); // required for proper stackwalking of RuntimeStub frame
933 __ push(rsi);
934 __ push(rdi);
935 __ movptr(from , Address(rsp, 12+ 4));
936 __ movptr(to , Address(rsp, 12+ 8));
937 __ movl(count, Address(rsp, 12+ 12));
938 if (t == T_OBJECT) {
939 __ testl(count, count);
940 __ jcc(Assembler::zero, L_0_count);
941 gen_write_ref_array_pre_barrier(to, count);
942 __ mov(saved_to, to); // save 'to'
943 }
945 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
946 BLOCK_COMMENT("Entry:");
948 __ subptr(to, from); // to --> to_from
949 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
950 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
951 if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
952 // align source address at 4 bytes address boundary
953 if (t == T_BYTE) {
954 // One byte misalignment happens only for byte arrays
955 __ testl(from, 1);
956 __ jccb(Assembler::zero, L_skip_align1);
957 __ movb(rax, Address(from, 0));
958 __ movb(Address(from, to_from, Address::times_1, 0), rax);
959 __ increment(from);
960 __ decrement(count);
961 __ BIND(L_skip_align1);
962 }
963 // Two bytes misalignment happens only for byte and short (char) arrays
964 __ testl(from, 2);
965 __ jccb(Assembler::zero, L_skip_align2);
966 __ movw(rax, Address(from, 0));
967 __ movw(Address(from, to_from, Address::times_1, 0), rax);
968 __ addptr(from, 2);
969 __ subl(count, 1<<(shift-1));
970 __ BIND(L_skip_align2);
971 }
972 if (!VM_Version::supports_mmx()) {
973 __ mov(rax, count); // save 'count'
974 __ shrl(count, shift); // bytes count
975 __ addptr(to_from, from);// restore 'to'
976 __ rep_mov();
977 __ subptr(to_from, from);// restore 'to_from'
978 __ mov(count, rax); // restore 'count'
979 __ jmpb(L_copy_2_bytes); // all dwords were copied
980 } else {
981 if (!UseUnalignedLoadStores) {
982 // align to 8 bytes, we know we are 4 byte aligned to start
983 __ testptr(from, 4);
984 __ jccb(Assembler::zero, L_copy_64_bytes);
985 __ movl(rax, Address(from, 0));
986 __ movl(Address(from, to_from, Address::times_1, 0), rax);
987 __ addptr(from, 4);
988 __ subl(count, 1<<shift);
989 }
990 __ BIND(L_copy_64_bytes);
991 __ mov(rax, count);
992 __ shrl(rax, shift+1); // 8 bytes chunk count
993 //
994 // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop
995 //
996 if (UseXMMForArrayCopy) {
997 xmm_copy_forward(from, to_from, rax);
998 } else {
999 mmx_copy_forward(from, to_from, rax);
1000 }
1001 }
1002 // copy tailing dword
1003 __ BIND(L_copy_4_bytes);
1004 __ testl(count, 1<<shift);
1005 __ jccb(Assembler::zero, L_copy_2_bytes);
1006 __ movl(rax, Address(from, 0));
1007 __ movl(Address(from, to_from, Address::times_1, 0), rax);
1008 if (t == T_BYTE || t == T_SHORT) {
1009 __ addptr(from, 4);
1010 __ BIND(L_copy_2_bytes);
1011 // copy tailing word
1012 __ testl(count, 1<<(shift-1));
1013 __ jccb(Assembler::zero, L_copy_byte);
1014 __ movw(rax, Address(from, 0));
1015 __ movw(Address(from, to_from, Address::times_1, 0), rax);
1016 if (t == T_BYTE) {
1017 __ addptr(from, 2);
1018 __ BIND(L_copy_byte);
1019 // copy tailing byte
1020 __ testl(count, 1);
1021 __ jccb(Assembler::zero, L_exit);
1022 __ movb(rax, Address(from, 0));
1023 __ movb(Address(from, to_from, Address::times_1, 0), rax);
1024 __ BIND(L_exit);
1025 } else {
1026 __ BIND(L_copy_byte);
1027 }
1028 } else {
1029 __ BIND(L_copy_2_bytes);
1030 }
1032 if (t == T_OBJECT) {
1033 __ movl(count, Address(rsp, 12+12)); // reread 'count'
1034 __ mov(to, saved_to); // restore 'to'
1035 gen_write_ref_array_post_barrier(to, count);
1036 __ BIND(L_0_count);
1037 }
1038 inc_copy_counter_np(t);
1039 __ pop(rdi);
1040 __ pop(rsi);
1041 __ leave(); // required for proper stackwalking of RuntimeStub frame
1042 __ xorptr(rax, rax); // return 0
1043 __ ret(0);
1044 return start;
1045 }
1048 address generate_conjoint_copy(BasicType t, bool aligned,
1049 Address::ScaleFactor sf,
1050 address nooverlap_target,
1051 address* entry, const char *name) {
1052 __ align(CodeEntryAlignment);
1053 StubCodeMark mark(this, "StubRoutines", name);
1054 address start = __ pc();
1056 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
1057 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop;
1059 int shift = Address::times_ptr - sf;
1061 const Register src = rax; // source array address
1062 const Register dst = rdx; // destination array address
1063 const Register from = rsi; // source array address
1064 const Register to = rdi; // destination array address
1065 const Register count = rcx; // elements count
1066 const Register end = rax; // array end address
1068 __ enter(); // required for proper stackwalking of RuntimeStub frame
1069 __ push(rsi);
1070 __ push(rdi);
1071 __ movptr(src , Address(rsp, 12+ 4)); // from
1072 __ movptr(dst , Address(rsp, 12+ 8)); // to
1073 __ movl2ptr(count, Address(rsp, 12+12)); // count
1074 if (t == T_OBJECT) {
1075 gen_write_ref_array_pre_barrier(dst, count);
1076 }
1078 if (entry != NULL) {
1079 *entry = __ pc(); // Entry point from generic arraycopy stub.
1080 BLOCK_COMMENT("Entry:");
1081 }
1083 if (t == T_OBJECT) {
1084 __ testl(count, count);
1085 __ jcc(Assembler::zero, L_0_count);
1086 }
1087 __ mov(from, src);
1088 __ mov(to , dst);
1090 // arrays overlap test
1091 RuntimeAddress nooverlap(nooverlap_target);
1092 __ cmpptr(dst, src);
1093 __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size
1094 __ jump_cc(Assembler::belowEqual, nooverlap);
1095 __ cmpptr(dst, end);
1096 __ jump_cc(Assembler::aboveEqual, nooverlap);
1098 // copy from high to low
1099 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1100 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
1101 if (t == T_BYTE || t == T_SHORT) {
1102 // Align the end of destination array at 4 bytes address boundary
1103 __ lea(end, Address(dst, count, sf, 0));
1104 if (t == T_BYTE) {
1105 // One byte misalignment happens only for byte arrays
1106 __ testl(end, 1);
1107 __ jccb(Assembler::zero, L_skip_align1);
1108 __ decrement(count);
1109 __ movb(rdx, Address(from, count, sf, 0));
1110 __ movb(Address(to, count, sf, 0), rdx);
1111 __ BIND(L_skip_align1);
1112 }
1113 // Two bytes misalignment happens only for byte and short (char) arrays
1114 __ testl(end, 2);
1115 __ jccb(Assembler::zero, L_skip_align2);
1116 __ subptr(count, 1<<(shift-1));
1117 __ movw(rdx, Address(from, count, sf, 0));
1118 __ movw(Address(to, count, sf, 0), rdx);
1119 __ BIND(L_skip_align2);
1120 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1121 __ jcc(Assembler::below, L_copy_4_bytes);
1122 }
1124 if (!VM_Version::supports_mmx()) {
1125 __ std();
1126 __ mov(rax, count); // Save 'count'
1127 __ mov(rdx, to); // Save 'to'
1128 __ lea(rsi, Address(from, count, sf, -4));
1129 __ lea(rdi, Address(to , count, sf, -4));
1130 __ shrptr(count, shift); // bytes count
1131 __ rep_mov();
1132 __ cld();
1133 __ mov(count, rax); // restore 'count'
1134 __ andl(count, (1<<shift)-1); // mask the number of rest elements
1135 __ movptr(from, Address(rsp, 12+4)); // reread 'from'
1136 __ mov(to, rdx); // restore 'to'
1137 __ jmpb(L_copy_2_bytes); // all dword were copied
1138 } else {
1139 // Align to 8 bytes the end of array. It is aligned to 4 bytes already.
1140 __ testptr(end, 4);
1141 __ jccb(Assembler::zero, L_copy_8_bytes);
1142 __ subl(count, 1<<shift);
1143 __ movl(rdx, Address(from, count, sf, 0));
1144 __ movl(Address(to, count, sf, 0), rdx);
1145 __ jmpb(L_copy_8_bytes);
1147 __ align(16);
1148 // Move 8 bytes
1149 __ BIND(L_copy_8_bytes_loop);
1150 if (UseXMMForArrayCopy) {
1151 __ movq(xmm0, Address(from, count, sf, 0));
1152 __ movq(Address(to, count, sf, 0), xmm0);
1153 } else {
1154 __ movq(mmx0, Address(from, count, sf, 0));
1155 __ movq(Address(to, count, sf, 0), mmx0);
1156 }
1157 __ BIND(L_copy_8_bytes);
1158 __ subl(count, 2<<shift);
1159 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1160 __ addl(count, 2<<shift);
1161 if (!UseXMMForArrayCopy) {
1162 __ emms();
1163 }
1164 }
1165 __ BIND(L_copy_4_bytes);
1166 // copy prefix qword
1167 __ testl(count, 1<<shift);
1168 __ jccb(Assembler::zero, L_copy_2_bytes);
1169 __ movl(rdx, Address(from, count, sf, -4));
1170 __ movl(Address(to, count, sf, -4), rdx);
1172 if (t == T_BYTE || t == T_SHORT) {
1173 __ subl(count, (1<<shift));
1174 __ BIND(L_copy_2_bytes);
1175 // copy prefix dword
1176 __ testl(count, 1<<(shift-1));
1177 __ jccb(Assembler::zero, L_copy_byte);
1178 __ movw(rdx, Address(from, count, sf, -2));
1179 __ movw(Address(to, count, sf, -2), rdx);
1180 if (t == T_BYTE) {
1181 __ subl(count, 1<<(shift-1));
1182 __ BIND(L_copy_byte);
1183 // copy prefix byte
1184 __ testl(count, 1);
1185 __ jccb(Assembler::zero, L_exit);
1186 __ movb(rdx, Address(from, 0));
1187 __ movb(Address(to, 0), rdx);
1188 __ BIND(L_exit);
1189 } else {
1190 __ BIND(L_copy_byte);
1191 }
1192 } else {
1193 __ BIND(L_copy_2_bytes);
1194 }
1195 if (t == T_OBJECT) {
1196 __ movl2ptr(count, Address(rsp, 12+12)); // reread count
1197 gen_write_ref_array_post_barrier(to, count);
1198 __ BIND(L_0_count);
1199 }
1200 inc_copy_counter_np(t);
1201 __ pop(rdi);
1202 __ pop(rsi);
1203 __ leave(); // required for proper stackwalking of RuntimeStub frame
1204 __ xorptr(rax, rax); // return 0
1205 __ ret(0);
1206 return start;
1207 }
1210 address generate_disjoint_long_copy(address* entry, const char *name) {
1211 __ align(CodeEntryAlignment);
1212 StubCodeMark mark(this, "StubRoutines", name);
1213 address start = __ pc();
1215 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1216 const Register from = rax; // source array address
1217 const Register to = rdx; // destination array address
1218 const Register count = rcx; // elements count
1219 const Register to_from = rdx; // (to - from)
1221 __ enter(); // required for proper stackwalking of RuntimeStub frame
1222 __ movptr(from , Address(rsp, 8+0)); // from
1223 __ movptr(to , Address(rsp, 8+4)); // to
1224 __ movl2ptr(count, Address(rsp, 8+8)); // count
1226 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
1227 BLOCK_COMMENT("Entry:");
1229 __ subptr(to, from); // to --> to_from
1230 if (VM_Version::supports_mmx()) {
1231 if (UseXMMForArrayCopy) {
1232 xmm_copy_forward(from, to_from, count);
1233 } else {
1234 mmx_copy_forward(from, to_from, count);
1235 }
1236 } else {
1237 __ jmpb(L_copy_8_bytes);
1238 __ align(16);
1239 __ BIND(L_copy_8_bytes_loop);
1240 __ fild_d(Address(from, 0));
1241 __ fistp_d(Address(from, to_from, Address::times_1));
1242 __ addptr(from, 8);
1243 __ BIND(L_copy_8_bytes);
1244 __ decrement(count);
1245 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1246 }
1247 inc_copy_counter_np(T_LONG);
1248 __ leave(); // required for proper stackwalking of RuntimeStub frame
1249 __ xorptr(rax, rax); // return 0
1250 __ ret(0);
1251 return start;
1252 }
1254 address generate_conjoint_long_copy(address nooverlap_target,
1255 address* entry, const char *name) {
1256 __ align(CodeEntryAlignment);
1257 StubCodeMark mark(this, "StubRoutines", name);
1258 address start = __ pc();
1260 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1261 const Register from = rax; // source array address
1262 const Register to = rdx; // destination array address
1263 const Register count = rcx; // elements count
1264 const Register end_from = rax; // source array end address
1266 __ enter(); // required for proper stackwalking of RuntimeStub frame
1267 __ movptr(from , Address(rsp, 8+0)); // from
1268 __ movptr(to , Address(rsp, 8+4)); // to
1269 __ movl2ptr(count, Address(rsp, 8+8)); // count
1271 *entry = __ pc(); // Entry point from generic arraycopy stub.
1272 BLOCK_COMMENT("Entry:");
1274 // arrays overlap test
1275 __ cmpptr(to, from);
1276 RuntimeAddress nooverlap(nooverlap_target);
1277 __ jump_cc(Assembler::belowEqual, nooverlap);
1278 __ lea(end_from, Address(from, count, Address::times_8, 0));
1279 __ cmpptr(to, end_from);
1280 __ movptr(from, Address(rsp, 8)); // from
1281 __ jump_cc(Assembler::aboveEqual, nooverlap);
1283 __ jmpb(L_copy_8_bytes);
1285 __ align(16);
1286 __ BIND(L_copy_8_bytes_loop);
1287 if (VM_Version::supports_mmx()) {
1288 if (UseXMMForArrayCopy) {
1289 __ movq(xmm0, Address(from, count, Address::times_8));
1290 __ movq(Address(to, count, Address::times_8), xmm0);
1291 } else {
1292 __ movq(mmx0, Address(from, count, Address::times_8));
1293 __ movq(Address(to, count, Address::times_8), mmx0);
1294 }
1295 } else {
1296 __ fild_d(Address(from, count, Address::times_8));
1297 __ fistp_d(Address(to, count, Address::times_8));
1298 }
1299 __ BIND(L_copy_8_bytes);
1300 __ decrement(count);
1301 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1303 if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
1304 __ emms();
1305 }
1306 inc_copy_counter_np(T_LONG);
1307 __ leave(); // required for proper stackwalking of RuntimeStub frame
1308 __ xorptr(rax, rax); // return 0
1309 __ ret(0);
1310 return start;
1311 }
1314 // Helper for generating a dynamic type check.
1315 // The sub_klass must be one of {rbx, rdx, rsi}.
1316 // The temp is killed.
1317 void generate_type_check(Register sub_klass,
1318 Address& super_check_offset_addr,
1319 Address& super_klass_addr,
1320 Register temp,
1321 Label* L_success, Label* L_failure) {
1322 BLOCK_COMMENT("type_check:");
1324 Label L_fallthrough;
1325 #define LOCAL_JCC(assembler_con, label_ptr) \
1326 if (label_ptr != NULL) __ jcc(assembler_con, *(label_ptr)); \
1327 else __ jcc(assembler_con, L_fallthrough) /*omit semi*/
1329 // The following is a strange variation of the fast path which requires
1330 // one less register, because needed values are on the argument stack.
1331 // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp,
1332 // L_success, L_failure, NULL);
1333 assert_different_registers(sub_klass, temp);
1335 int sc_offset = (klassOopDesc::header_size() * HeapWordSize +
1336 Klass::secondary_super_cache_offset_in_bytes());
1338 // if the pointers are equal, we are done (e.g., String[] elements)
1339 __ cmpptr(sub_klass, super_klass_addr);
1340 LOCAL_JCC(Assembler::equal, L_success);
1342 // check the supertype display:
1343 __ movl2ptr(temp, super_check_offset_addr);
1344 Address super_check_addr(sub_klass, temp, Address::times_1, 0);
1345 __ movptr(temp, super_check_addr); // load displayed supertype
1346 __ cmpptr(temp, super_klass_addr); // test the super type
1347 LOCAL_JCC(Assembler::equal, L_success);
1349 // if it was a primary super, we can just fail immediately
1350 __ cmpl(super_check_offset_addr, sc_offset);
1351 LOCAL_JCC(Assembler::notEqual, L_failure);
1353 // The repne_scan instruction uses fixed registers, which will get spilled.
1354 // We happen to know this works best when super_klass is in rax.
1355 Register super_klass = temp;
1356 __ movptr(super_klass, super_klass_addr);
1357 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg,
1358 L_success, L_failure);
1360 __ bind(L_fallthrough);
1362 if (L_success == NULL) { BLOCK_COMMENT("L_success:"); }
1363 if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); }
1365 #undef LOCAL_JCC
1366 }
1368 //
1369 // Generate checkcasting array copy stub
1370 //
1371 // Input:
1372 // 4(rsp) - source array address
1373 // 8(rsp) - destination array address
1374 // 12(rsp) - element count, can be zero
1375 // 16(rsp) - size_t ckoff (super_check_offset)
1376 // 20(rsp) - oop ckval (super_klass)
1377 //
1378 // Output:
1379 // rax, == 0 - success
1380 // rax, == -1^K - failure, where K is partial transfer count
1381 //
1382 address generate_checkcast_copy(const char *name, address* entry) {
1383 __ align(CodeEntryAlignment);
1384 StubCodeMark mark(this, "StubRoutines", name);
1385 address start = __ pc();
1387 Label L_load_element, L_store_element, L_do_card_marks, L_done;
1389 // register use:
1390 // rax, rdx, rcx -- loop control (end_from, end_to, count)
1391 // rdi, rsi -- element access (oop, klass)
1392 // rbx, -- temp
1393 const Register from = rax; // source array address
1394 const Register to = rdx; // destination array address
1395 const Register length = rcx; // elements count
1396 const Register elem = rdi; // each oop copied
1397 const Register elem_klass = rsi; // each elem._klass (sub_klass)
1398 const Register temp = rbx; // lone remaining temp
1400 __ enter(); // required for proper stackwalking of RuntimeStub frame
1402 __ push(rsi);
1403 __ push(rdi);
1404 __ push(rbx);
1406 Address from_arg(rsp, 16+ 4); // from
1407 Address to_arg(rsp, 16+ 8); // to
1408 Address length_arg(rsp, 16+12); // elements count
1409 Address ckoff_arg(rsp, 16+16); // super_check_offset
1410 Address ckval_arg(rsp, 16+20); // super_klass
1412 // Load up:
1413 __ movptr(from, from_arg);
1414 __ movptr(to, to_arg);
1415 __ movl2ptr(length, length_arg);
1417 *entry = __ pc(); // Entry point from generic arraycopy stub.
1418 BLOCK_COMMENT("Entry:");
1420 //---------------------------------------------------------------
1421 // Assembler stub will be used for this call to arraycopy
1422 // if the two arrays are subtypes of Object[] but the
1423 // destination array type is not equal to or a supertype
1424 // of the source type. Each element must be separately
1425 // checked.
1427 // Loop-invariant addresses. They are exclusive end pointers.
1428 Address end_from_addr(from, length, Address::times_ptr, 0);
1429 Address end_to_addr(to, length, Address::times_ptr, 0);
1431 Register end_from = from; // re-use
1432 Register end_to = to; // re-use
1433 Register count = length; // re-use
1435 // Loop-variant addresses. They assume post-incremented count < 0.
1436 Address from_element_addr(end_from, count, Address::times_ptr, 0);
1437 Address to_element_addr(end_to, count, Address::times_ptr, 0);
1438 Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes());
1440 // Copy from low to high addresses, indexed from the end of each array.
1441 gen_write_ref_array_pre_barrier(to, count);
1442 __ lea(end_from, end_from_addr);
1443 __ lea(end_to, end_to_addr);
1444 assert(length == count, ""); // else fix next line:
1445 __ negptr(count); // negate and test the length
1446 __ jccb(Assembler::notZero, L_load_element);
1448 // Empty array: Nothing to do.
1449 __ xorptr(rax, rax); // return 0 on (trivial) success
1450 __ jmp(L_done);
1452 // ======== begin loop ========
1453 // (Loop is rotated; its entry is L_load_element.)
1454 // Loop control:
1455 // for (count = -count; count != 0; count++)
1456 // Base pointers src, dst are biased by 8*count,to last element.
1457 __ align(16);
1459 __ BIND(L_store_element);
1460 __ movptr(to_element_addr, elem); // store the oop
1461 __ increment(count); // increment the count toward zero
1462 __ jccb(Assembler::zero, L_do_card_marks);
1464 // ======== loop entry is here ========
1465 __ BIND(L_load_element);
1466 __ movptr(elem, from_element_addr); // load the oop
1467 __ testptr(elem, elem);
1468 __ jccb(Assembler::zero, L_store_element);
1470 // (Could do a trick here: Remember last successful non-null
1471 // element stored and make a quick oop equality check on it.)
1473 __ movptr(elem_klass, elem_klass_addr); // query the object klass
1474 generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp,
1475 &L_store_element, NULL);
1476 // (On fall-through, we have failed the element type check.)
1477 // ======== end loop ========
1479 // It was a real error; we must depend on the caller to finish the job.
1480 // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops.
1481 // Emit GC store barriers for the oops we have copied (length_arg + count),
1482 // and report their number to the caller.
1483 __ addl(count, length_arg); // transfers = (length - remaining)
1484 __ movl2ptr(rax, count); // save the value
1485 __ notptr(rax); // report (-1^K) to caller
1486 __ movptr(to, to_arg); // reload
1487 assert_different_registers(to, count, rax);
1488 gen_write_ref_array_post_barrier(to, count);
1489 __ jmpb(L_done);
1491 // Come here on success only.
1492 __ BIND(L_do_card_marks);
1493 __ movl2ptr(count, length_arg);
1494 __ movptr(to, to_arg); // reload
1495 gen_write_ref_array_post_barrier(to, count);
1496 __ xorptr(rax, rax); // return 0 on success
1498 // Common exit point (success or failure).
1499 __ BIND(L_done);
1500 __ pop(rbx);
1501 __ pop(rdi);
1502 __ pop(rsi);
1503 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
1504 __ leave(); // required for proper stackwalking of RuntimeStub frame
1505 __ ret(0);
1507 return start;
1508 }
1510 //
1511 // Generate 'unsafe' array copy stub
1512 // Though just as safe as the other stubs, it takes an unscaled
1513 // size_t argument instead of an element count.
1514 //
1515 // Input:
1516 // 4(rsp) - source array address
1517 // 8(rsp) - destination array address
1518 // 12(rsp) - byte count, can be zero
1519 //
1520 // Output:
1521 // rax, == 0 - success
1522 // rax, == -1 - need to call System.arraycopy
1523 //
1524 // Examines the alignment of the operands and dispatches
1525 // to a long, int, short, or byte copy loop.
1526 //
1527 address generate_unsafe_copy(const char *name,
1528 address byte_copy_entry,
1529 address short_copy_entry,
1530 address int_copy_entry,
1531 address long_copy_entry) {
1533 Label L_long_aligned, L_int_aligned, L_short_aligned;
1535 __ align(CodeEntryAlignment);
1536 StubCodeMark mark(this, "StubRoutines", name);
1537 address start = __ pc();
1539 const Register from = rax; // source array address
1540 const Register to = rdx; // destination array address
1541 const Register count = rcx; // elements count
1543 __ enter(); // required for proper stackwalking of RuntimeStub frame
1544 __ push(rsi);
1545 __ push(rdi);
1546 Address from_arg(rsp, 12+ 4); // from
1547 Address to_arg(rsp, 12+ 8); // to
1548 Address count_arg(rsp, 12+12); // byte count
1550 // Load up:
1551 __ movptr(from , from_arg);
1552 __ movptr(to , to_arg);
1553 __ movl2ptr(count, count_arg);
1555 // bump this on entry, not on exit:
1556 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
1558 const Register bits = rsi;
1559 __ mov(bits, from);
1560 __ orptr(bits, to);
1561 __ orptr(bits, count);
1563 __ testl(bits, BytesPerLong-1);
1564 __ jccb(Assembler::zero, L_long_aligned);
1566 __ testl(bits, BytesPerInt-1);
1567 __ jccb(Assembler::zero, L_int_aligned);
1569 __ testl(bits, BytesPerShort-1);
1570 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
1572 __ BIND(L_short_aligned);
1573 __ shrptr(count, LogBytesPerShort); // size => short_count
1574 __ movl(count_arg, count); // update 'count'
1575 __ jump(RuntimeAddress(short_copy_entry));
1577 __ BIND(L_int_aligned);
1578 __ shrptr(count, LogBytesPerInt); // size => int_count
1579 __ movl(count_arg, count); // update 'count'
1580 __ jump(RuntimeAddress(int_copy_entry));
1582 __ BIND(L_long_aligned);
1583 __ shrptr(count, LogBytesPerLong); // size => qword_count
1584 __ movl(count_arg, count); // update 'count'
1585 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1586 __ pop(rsi);
1587 __ jump(RuntimeAddress(long_copy_entry));
1589 return start;
1590 }
1593 // Perform range checks on the proposed arraycopy.
1594 // Smashes src_pos and dst_pos. (Uses them up for temps.)
1595 void arraycopy_range_checks(Register src,
1596 Register src_pos,
1597 Register dst,
1598 Register dst_pos,
1599 Address& length,
1600 Label& L_failed) {
1601 BLOCK_COMMENT("arraycopy_range_checks:");
1602 const Register src_end = src_pos; // source array end position
1603 const Register dst_end = dst_pos; // destination array end position
1604 __ addl(src_end, length); // src_pos + length
1605 __ addl(dst_end, length); // dst_pos + length
1607 // if (src_pos + length > arrayOop(src)->length() ) FAIL;
1608 __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes()));
1609 __ jcc(Assembler::above, L_failed);
1611 // if (dst_pos + length > arrayOop(dst)->length() ) FAIL;
1612 __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes()));
1613 __ jcc(Assembler::above, L_failed);
1615 BLOCK_COMMENT("arraycopy_range_checks done");
1616 }
1619 //
1620 // Generate generic array copy stubs
1621 //
1622 // Input:
1623 // 4(rsp) - src oop
1624 // 8(rsp) - src_pos
1625 // 12(rsp) - dst oop
1626 // 16(rsp) - dst_pos
1627 // 20(rsp) - element count
1628 //
1629 // Output:
1630 // rax, == 0 - success
1631 // rax, == -1^K - failure, where K is partial transfer count
1632 //
1633 address generate_generic_copy(const char *name,
1634 address entry_jbyte_arraycopy,
1635 address entry_jshort_arraycopy,
1636 address entry_jint_arraycopy,
1637 address entry_oop_arraycopy,
1638 address entry_jlong_arraycopy,
1639 address entry_checkcast_arraycopy) {
1640 Label L_failed, L_failed_0, L_objArray;
1642 { int modulus = CodeEntryAlignment;
1643 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
1644 int advance = target - (__ offset() % modulus);
1645 if (advance < 0) advance += modulus;
1646 if (advance > 0) __ nop(advance);
1647 }
1648 StubCodeMark mark(this, "StubRoutines", name);
1650 // Short-hop target to L_failed. Makes for denser prologue code.
1651 __ BIND(L_failed_0);
1652 __ jmp(L_failed);
1653 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
1655 __ align(CodeEntryAlignment);
1656 address start = __ pc();
1658 __ enter(); // required for proper stackwalking of RuntimeStub frame
1659 __ push(rsi);
1660 __ push(rdi);
1662 // bump this on entry, not on exit:
1663 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
1665 // Input values
1666 Address SRC (rsp, 12+ 4);
1667 Address SRC_POS (rsp, 12+ 8);
1668 Address DST (rsp, 12+12);
1669 Address DST_POS (rsp, 12+16);
1670 Address LENGTH (rsp, 12+20);
1672 //-----------------------------------------------------------------------
1673 // Assembler stub will be used for this call to arraycopy
1674 // if the following conditions are met:
1675 //
1676 // (1) src and dst must not be null.
1677 // (2) src_pos must not be negative.
1678 // (3) dst_pos must not be negative.
1679 // (4) length must not be negative.
1680 // (5) src klass and dst klass should be the same and not NULL.
1681 // (6) src and dst should be arrays.
1682 // (7) src_pos + length must not exceed length of src.
1683 // (8) dst_pos + length must not exceed length of dst.
1684 //
1686 const Register src = rax; // source array oop
1687 const Register src_pos = rsi;
1688 const Register dst = rdx; // destination array oop
1689 const Register dst_pos = rdi;
1690 const Register length = rcx; // transfer count
1692 // if (src == NULL) return -1;
1693 __ movptr(src, SRC); // src oop
1694 __ testptr(src, src);
1695 __ jccb(Assembler::zero, L_failed_0);
1697 // if (src_pos < 0) return -1;
1698 __ movl2ptr(src_pos, SRC_POS); // src_pos
1699 __ testl(src_pos, src_pos);
1700 __ jccb(Assembler::negative, L_failed_0);
1702 // if (dst == NULL) return -1;
1703 __ movptr(dst, DST); // dst oop
1704 __ testptr(dst, dst);
1705 __ jccb(Assembler::zero, L_failed_0);
1707 // if (dst_pos < 0) return -1;
1708 __ movl2ptr(dst_pos, DST_POS); // dst_pos
1709 __ testl(dst_pos, dst_pos);
1710 __ jccb(Assembler::negative, L_failed_0);
1712 // if (length < 0) return -1;
1713 __ movl2ptr(length, LENGTH); // length
1714 __ testl(length, length);
1715 __ jccb(Assembler::negative, L_failed_0);
1717 // if (src->klass() == NULL) return -1;
1718 Address src_klass_addr(src, oopDesc::klass_offset_in_bytes());
1719 Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes());
1720 const Register rcx_src_klass = rcx; // array klass
1721 __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes()));
1723 #ifdef ASSERT
1724 // assert(src->klass() != NULL);
1725 BLOCK_COMMENT("assert klasses not null");
1726 { Label L1, L2;
1727 __ testptr(rcx_src_klass, rcx_src_klass);
1728 __ jccb(Assembler::notZero, L2); // it is broken if klass is NULL
1729 __ bind(L1);
1730 __ stop("broken null klass");
1731 __ bind(L2);
1732 __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD);
1733 __ jccb(Assembler::equal, L1); // this would be broken also
1734 BLOCK_COMMENT("assert done");
1735 }
1736 #endif //ASSERT
1738 // Load layout helper (32-bits)
1739 //
1740 // |array_tag| | header_size | element_type | |log2_element_size|
1741 // 32 30 24 16 8 2 0
1742 //
1743 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
1744 //
1746 int lh_offset = klassOopDesc::header_size() * HeapWordSize +
1747 Klass::layout_helper_offset_in_bytes();
1748 Address src_klass_lh_addr(rcx_src_klass, lh_offset);
1750 // Handle objArrays completely differently...
1751 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
1752 __ cmpl(src_klass_lh_addr, objArray_lh);
1753 __ jcc(Assembler::equal, L_objArray);
1755 // if (src->klass() != dst->klass()) return -1;
1756 __ cmpptr(rcx_src_klass, dst_klass_addr);
1757 __ jccb(Assembler::notEqual, L_failed_0);
1759 const Register rcx_lh = rcx; // layout helper
1760 assert(rcx_lh == rcx_src_klass, "known alias");
1761 __ movl(rcx_lh, src_klass_lh_addr);
1763 // if (!src->is_Array()) return -1;
1764 __ cmpl(rcx_lh, Klass::_lh_neutral_value);
1765 __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp
1767 // At this point, it is known to be a typeArray (array_tag 0x3).
1768 #ifdef ASSERT
1769 { Label L;
1770 __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
1771 __ jcc(Assembler::greaterEqual, L); // signed cmp
1772 __ stop("must be a primitive array");
1773 __ bind(L);
1774 }
1775 #endif
1777 assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh);
1778 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1780 // typeArrayKlass
1781 //
1782 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
1783 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
1784 //
1785 const Register rsi_offset = rsi; // array offset
1786 const Register src_array = src; // src array offset
1787 const Register dst_array = dst; // dst array offset
1788 const Register rdi_elsize = rdi; // log2 element size
1790 __ mov(rsi_offset, rcx_lh);
1791 __ shrptr(rsi_offset, Klass::_lh_header_size_shift);
1792 __ andptr(rsi_offset, Klass::_lh_header_size_mask); // array_offset
1793 __ addptr(src_array, rsi_offset); // src array offset
1794 __ addptr(dst_array, rsi_offset); // dst array offset
1795 __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize
1797 // next registers should be set before the jump to corresponding stub
1798 const Register from = src; // source array address
1799 const Register to = dst; // destination array address
1800 const Register count = rcx; // elements count
1801 // some of them should be duplicated on stack
1802 #define FROM Address(rsp, 12+ 4)
1803 #define TO Address(rsp, 12+ 8) // Not used now
1804 #define COUNT Address(rsp, 12+12) // Only for oop arraycopy
1806 BLOCK_COMMENT("scale indexes to element size");
1807 __ movl2ptr(rsi, SRC_POS); // src_pos
1808 __ shlptr(rsi); // src_pos << rcx (log2 elsize)
1809 assert(src_array == from, "");
1810 __ addptr(from, rsi); // from = src_array + SRC_POS << log2 elsize
1811 __ movl2ptr(rdi, DST_POS); // dst_pos
1812 __ shlptr(rdi); // dst_pos << rcx (log2 elsize)
1813 assert(dst_array == to, "");
1814 __ addptr(to, rdi); // to = dst_array + DST_POS << log2 elsize
1815 __ movptr(FROM, from); // src_addr
1816 __ mov(rdi_elsize, rcx_lh); // log2 elsize
1817 __ movl2ptr(count, LENGTH); // elements count
1819 BLOCK_COMMENT("choose copy loop based on element size");
1820 __ cmpl(rdi_elsize, 0);
1822 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy));
1823 __ cmpl(rdi_elsize, LogBytesPerShort);
1824 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy));
1825 __ cmpl(rdi_elsize, LogBytesPerInt);
1826 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy));
1827 #ifdef ASSERT
1828 __ cmpl(rdi_elsize, LogBytesPerLong);
1829 __ jccb(Assembler::notEqual, L_failed);
1830 #endif
1831 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1832 __ pop(rsi);
1833 __ jump(RuntimeAddress(entry_jlong_arraycopy));
1835 __ BIND(L_failed);
1836 __ xorptr(rax, rax);
1837 __ notptr(rax); // return -1
1838 __ pop(rdi);
1839 __ pop(rsi);
1840 __ leave(); // required for proper stackwalking of RuntimeStub frame
1841 __ ret(0);
1843 // objArrayKlass
1844 __ BIND(L_objArray);
1845 // live at this point: rcx_src_klass, src[_pos], dst[_pos]
1847 Label L_plain_copy, L_checkcast_copy;
1848 // test array classes for subtyping
1849 __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality
1850 __ jccb(Assembler::notEqual, L_checkcast_copy);
1852 // Identically typed arrays can be copied without element-wise checks.
1853 assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass);
1854 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1856 __ BIND(L_plain_copy);
1857 __ movl2ptr(count, LENGTH); // elements count
1858 __ movl2ptr(src_pos, SRC_POS); // reload src_pos
1859 __ lea(from, Address(src, src_pos, Address::times_ptr,
1860 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
1861 __ movl2ptr(dst_pos, DST_POS); // reload dst_pos
1862 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1863 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
1864 __ movptr(FROM, from); // src_addr
1865 __ movptr(TO, to); // dst_addr
1866 __ movl(COUNT, count); // count
1867 __ jump(RuntimeAddress(entry_oop_arraycopy));
1869 __ BIND(L_checkcast_copy);
1870 // live at this point: rcx_src_klass, dst[_pos], src[_pos]
1871 {
1872 // Handy offsets:
1873 int ek_offset = (klassOopDesc::header_size() * HeapWordSize +
1874 objArrayKlass::element_klass_offset_in_bytes());
1875 int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
1876 Klass::super_check_offset_offset_in_bytes());
1878 Register rsi_dst_klass = rsi;
1879 Register rdi_temp = rdi;
1880 assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos");
1881 assert(rdi_temp == dst_pos, "expected alias w/ dst_pos");
1882 Address dst_klass_lh_addr(rsi_dst_klass, lh_offset);
1884 // Before looking at dst.length, make sure dst is also an objArray.
1885 __ movptr(rsi_dst_klass, dst_klass_addr);
1886 __ cmpl(dst_klass_lh_addr, objArray_lh);
1887 __ jccb(Assembler::notEqual, L_failed);
1889 // It is safe to examine both src.length and dst.length.
1890 __ movl2ptr(src_pos, SRC_POS); // reload rsi
1891 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1892 // (Now src_pos and dst_pos are killed, but not src and dst.)
1894 // We'll need this temp (don't forget to pop it after the type check).
1895 __ push(rbx);
1896 Register rbx_src_klass = rbx;
1898 __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx
1899 __ movptr(rsi_dst_klass, dst_klass_addr);
1900 Address super_check_offset_addr(rsi_dst_klass, sco_offset);
1901 Label L_fail_array_check;
1902 generate_type_check(rbx_src_klass,
1903 super_check_offset_addr, dst_klass_addr,
1904 rdi_temp, NULL, &L_fail_array_check);
1905 // (On fall-through, we have passed the array type check.)
1906 __ pop(rbx);
1907 __ jmp(L_plain_copy);
1909 __ BIND(L_fail_array_check);
1910 // Reshuffle arguments so we can call checkcast_arraycopy:
1912 // match initial saves for checkcast_arraycopy
1913 // push(rsi); // already done; see above
1914 // push(rdi); // already done; see above
1915 // push(rbx); // already done; see above
1917 // Marshal outgoing arguments now, freeing registers.
1918 Address from_arg(rsp, 16+ 4); // from
1919 Address to_arg(rsp, 16+ 8); // to
1920 Address length_arg(rsp, 16+12); // elements count
1921 Address ckoff_arg(rsp, 16+16); // super_check_offset
1922 Address ckval_arg(rsp, 16+20); // super_klass
1924 Address SRC_POS_arg(rsp, 16+ 8);
1925 Address DST_POS_arg(rsp, 16+16);
1926 Address LENGTH_arg(rsp, 16+20);
1927 // push rbx, changed the incoming offsets (why not just use rbp,??)
1928 // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, "");
1930 __ movptr(rbx, Address(rsi_dst_klass, ek_offset));
1931 __ movl2ptr(length, LENGTH_arg); // reload elements count
1932 __ movl2ptr(src_pos, SRC_POS_arg); // reload src_pos
1933 __ movl2ptr(dst_pos, DST_POS_arg); // reload dst_pos
1935 __ movptr(ckval_arg, rbx); // destination element type
1936 __ movl(rbx, Address(rbx, sco_offset));
1937 __ movl(ckoff_arg, rbx); // corresponding class check offset
1939 __ movl(length_arg, length); // outgoing length argument
1941 __ lea(from, Address(src, src_pos, Address::times_ptr,
1942 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1943 __ movptr(from_arg, from);
1945 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1946 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1947 __ movptr(to_arg, to);
1948 __ jump(RuntimeAddress(entry_checkcast_arraycopy));
1949 }
1951 return start;
1952 }
1954 void generate_arraycopy_stubs() {
1955 address entry;
1956 address entry_jbyte_arraycopy;
1957 address entry_jshort_arraycopy;
1958 address entry_jint_arraycopy;
1959 address entry_oop_arraycopy;
1960 address entry_jlong_arraycopy;
1961 address entry_checkcast_arraycopy;
1963 StubRoutines::_arrayof_jbyte_disjoint_arraycopy =
1964 generate_disjoint_copy(T_BYTE, true, Address::times_1, &entry,
1965 "arrayof_jbyte_disjoint_arraycopy");
1966 StubRoutines::_arrayof_jbyte_arraycopy =
1967 generate_conjoint_copy(T_BYTE, true, Address::times_1, entry,
1968 NULL, "arrayof_jbyte_arraycopy");
1969 StubRoutines::_jbyte_disjoint_arraycopy =
1970 generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry,
1971 "jbyte_disjoint_arraycopy");
1972 StubRoutines::_jbyte_arraycopy =
1973 generate_conjoint_copy(T_BYTE, false, Address::times_1, entry,
1974 &entry_jbyte_arraycopy, "jbyte_arraycopy");
1976 StubRoutines::_arrayof_jshort_disjoint_arraycopy =
1977 generate_disjoint_copy(T_SHORT, true, Address::times_2, &entry,
1978 "arrayof_jshort_disjoint_arraycopy");
1979 StubRoutines::_arrayof_jshort_arraycopy =
1980 generate_conjoint_copy(T_SHORT, true, Address::times_2, entry,
1981 NULL, "arrayof_jshort_arraycopy");
1982 StubRoutines::_jshort_disjoint_arraycopy =
1983 generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry,
1984 "jshort_disjoint_arraycopy");
1985 StubRoutines::_jshort_arraycopy =
1986 generate_conjoint_copy(T_SHORT, false, Address::times_2, entry,
1987 &entry_jshort_arraycopy, "jshort_arraycopy");
1989 // Next arrays are always aligned on 4 bytes at least.
1990 StubRoutines::_jint_disjoint_arraycopy =
1991 generate_disjoint_copy(T_INT, true, Address::times_4, &entry,
1992 "jint_disjoint_arraycopy");
1993 StubRoutines::_jint_arraycopy =
1994 generate_conjoint_copy(T_INT, true, Address::times_4, entry,
1995 &entry_jint_arraycopy, "jint_arraycopy");
1997 StubRoutines::_oop_disjoint_arraycopy =
1998 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
1999 "oop_disjoint_arraycopy");
2000 StubRoutines::_oop_arraycopy =
2001 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
2002 &entry_oop_arraycopy, "oop_arraycopy");
2004 StubRoutines::_jlong_disjoint_arraycopy =
2005 generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy");
2006 StubRoutines::_jlong_arraycopy =
2007 generate_conjoint_long_copy(entry, &entry_jlong_arraycopy,
2008 "jlong_arraycopy");
2010 StubRoutines::_arrayof_jint_disjoint_arraycopy =
2011 StubRoutines::_jint_disjoint_arraycopy;
2012 StubRoutines::_arrayof_oop_disjoint_arraycopy =
2013 StubRoutines::_oop_disjoint_arraycopy;
2014 StubRoutines::_arrayof_jlong_disjoint_arraycopy =
2015 StubRoutines::_jlong_disjoint_arraycopy;
2017 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2018 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2019 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2021 StubRoutines::_checkcast_arraycopy =
2022 generate_checkcast_copy("checkcast_arraycopy",
2023 &entry_checkcast_arraycopy);
2025 StubRoutines::_unsafe_arraycopy =
2026 generate_unsafe_copy("unsafe_arraycopy",
2027 entry_jbyte_arraycopy,
2028 entry_jshort_arraycopy,
2029 entry_jint_arraycopy,
2030 entry_jlong_arraycopy);
2032 StubRoutines::_generic_arraycopy =
2033 generate_generic_copy("generic_arraycopy",
2034 entry_jbyte_arraycopy,
2035 entry_jshort_arraycopy,
2036 entry_jint_arraycopy,
2037 entry_oop_arraycopy,
2038 entry_jlong_arraycopy,
2039 entry_checkcast_arraycopy);
2040 }
2042 void generate_math_stubs() {
2043 {
2044 StubCodeMark mark(this, "StubRoutines", "log");
2045 StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
2047 __ fld_d(Address(rsp, 4));
2048 __ flog();
2049 __ ret(0);
2050 }
2051 {
2052 StubCodeMark mark(this, "StubRoutines", "log10");
2053 StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
2055 __ fld_d(Address(rsp, 4));
2056 __ flog10();
2057 __ ret(0);
2058 }
2059 {
2060 StubCodeMark mark(this, "StubRoutines", "sin");
2061 StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc();
2063 __ fld_d(Address(rsp, 4));
2064 __ trigfunc('s');
2065 __ ret(0);
2066 }
2067 {
2068 StubCodeMark mark(this, "StubRoutines", "cos");
2069 StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
2071 __ fld_d(Address(rsp, 4));
2072 __ trigfunc('c');
2073 __ ret(0);
2074 }
2075 {
2076 StubCodeMark mark(this, "StubRoutines", "tan");
2077 StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
2079 __ fld_d(Address(rsp, 4));
2080 __ trigfunc('t');
2081 __ ret(0);
2082 }
2084 // The intrinsic version of these seem to return the same value as
2085 // the strict version.
2086 StubRoutines::_intrinsic_exp = SharedRuntime::dexp;
2087 StubRoutines::_intrinsic_pow = SharedRuntime::dpow;
2088 }
2090 public:
2091 // Information about frame layout at time of blocking runtime call.
2092 // Note that we only have to preserve callee-saved registers since
2093 // the compilers are responsible for supplying a continuation point
2094 // if they expect all registers to be preserved.
2095 enum layout {
2096 thread_off, // last_java_sp
2097 rbp_off, // callee saved register
2098 ret_pc,
2099 framesize
2100 };
2102 private:
2104 #undef __
2105 #define __ masm->
2107 //------------------------------------------------------------------------------------------------------------------------
2108 // Continuation point for throwing of implicit exceptions that are not handled in
2109 // the current activation. Fabricates an exception oop and initiates normal
2110 // exception dispatching in this frame.
2111 //
2112 // Previously the compiler (c2) allowed for callee save registers on Java calls.
2113 // This is no longer true after adapter frames were removed but could possibly
2114 // be brought back in the future if the interpreter code was reworked and it
2115 // was deemed worthwhile. The comment below was left to describe what must
2116 // happen here if callee saves were resurrected. As it stands now this stub
2117 // could actually be a vanilla BufferBlob and have now oopMap at all.
2118 // Since it doesn't make much difference we've chosen to leave it the
2119 // way it was in the callee save days and keep the comment.
2121 // If we need to preserve callee-saved values we need a callee-saved oop map and
2122 // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs.
2123 // If the compiler needs all registers to be preserved between the fault
2124 // point and the exception handler then it must assume responsibility for that in
2125 // AbstractCompiler::continuation_for_implicit_null_exception or
2126 // continuation_for_implicit_division_by_zero_exception. All other implicit
2127 // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are
2128 // either at call sites or otherwise assume that stack unwinding will be initiated,
2129 // so caller saved registers were assumed volatile in the compiler.
2130 address generate_throw_exception(const char* name, address runtime_entry,
2131 bool restore_saved_exception_pc) {
2133 int insts_size = 256;
2134 int locs_size = 32;
2136 CodeBuffer code(name, insts_size, locs_size);
2137 OopMapSet* oop_maps = new OopMapSet();
2138 MacroAssembler* masm = new MacroAssembler(&code);
2140 address start = __ pc();
2142 // This is an inlined and slightly modified version of call_VM
2143 // which has the ability to fetch the return PC out of
2144 // thread-local storage and also sets up last_Java_sp slightly
2145 // differently than the real call_VM
2146 Register java_thread = rbx;
2147 __ get_thread(java_thread);
2148 if (restore_saved_exception_pc) {
2149 __ movptr(rax, Address(java_thread, in_bytes(JavaThread::saved_exception_pc_offset())));
2150 __ push(rax);
2151 }
2153 __ enter(); // required for proper stackwalking of RuntimeStub frame
2155 // pc and rbp, already pushed
2156 __ subptr(rsp, (framesize-2) * wordSize); // prolog
2158 // Frame is now completed as far as size and linkage.
2160 int frame_complete = __ pc() - start;
2162 // push java thread (becomes first argument of C function)
2163 __ movptr(Address(rsp, thread_off * wordSize), java_thread);
2165 // Set up last_Java_sp and last_Java_fp
2166 __ set_last_Java_frame(java_thread, rsp, rbp, NULL);
2168 // Call runtime
2169 BLOCK_COMMENT("call runtime_entry");
2170 __ call(RuntimeAddress(runtime_entry));
2171 // Generate oop map
2172 OopMap* map = new OopMap(framesize, 0);
2173 oop_maps->add_gc_map(__ pc() - start, map);
2175 // restore the thread (cannot use the pushed argument since arguments
2176 // may be overwritten by C code generated by an optimizing compiler);
2177 // however can use the register value directly if it is callee saved.
2178 __ get_thread(java_thread);
2180 __ reset_last_Java_frame(java_thread, true, false);
2182 __ leave(); // required for proper stackwalking of RuntimeStub frame
2184 // check for pending exceptions
2185 #ifdef ASSERT
2186 Label L;
2187 __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
2188 __ jcc(Assembler::notEqual, L);
2189 __ should_not_reach_here();
2190 __ bind(L);
2191 #endif /* ASSERT */
2192 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2195 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false);
2196 return stub->entry_point();
2197 }
2200 void create_control_words() {
2201 // Round to nearest, 53-bit mode, exceptions masked
2202 StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
2203 // Round to zero, 53-bit mode, exception mased
2204 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
2205 // Round to nearest, 24-bit mode, exceptions masked
2206 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
2207 // Round to nearest, 64-bit mode, exceptions masked
2208 StubRoutines::_fpu_cntrl_wrd_64 = 0x037F;
2209 // Round to nearest, 64-bit mode, exceptions masked
2210 StubRoutines::_mxcsr_std = 0x1F80;
2211 // Note: the following two constants are 80-bit values
2212 // layout is critical for correct loading by FPU.
2213 // Bias for strict fp multiply/divide
2214 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
2215 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
2216 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
2217 // Un-Bias for strict fp multiply/divide
2218 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
2219 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
2220 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
2221 }
2223 //---------------------------------------------------------------------------
2224 // Initialization
2226 void generate_initial() {
2227 // Generates all stubs and initializes the entry points
2229 //------------------------------------------------------------------------------------------------------------------------
2230 // entry points that exist in all platforms
2231 // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than
2232 // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp.
2233 StubRoutines::_forward_exception_entry = generate_forward_exception();
2235 StubRoutines::_call_stub_entry =
2236 generate_call_stub(StubRoutines::_call_stub_return_address);
2237 // is referenced by megamorphic call
2238 StubRoutines::_catch_exception_entry = generate_catch_exception();
2240 // These are currently used by Solaris/Intel
2241 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
2243 StubRoutines::_handler_for_unsafe_access_entry =
2244 generate_handler_for_unsafe_access();
2246 // platform dependent
2247 create_control_words();
2249 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
2250 StubRoutines::x86::_verify_fpu_cntrl_wrd_entry = generate_verify_fpu_cntrl_wrd();
2251 StubRoutines::_d2i_wrapper = generate_d2i_wrapper(T_INT,
2252 CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
2253 StubRoutines::_d2l_wrapper = generate_d2i_wrapper(T_LONG,
2254 CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
2255 }
2258 void generate_all() {
2259 // Generates all stubs and initializes the entry points
2261 // These entry points require SharedInfo::stack0 to be set up in non-core builds
2262 // and need to be relocatable, so they each fabricate a RuntimeStub internally.
2263 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError), false);
2264 StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError), false);
2265 StubRoutines::_throw_ArithmeticException_entry = generate_throw_exception("ArithmeticException throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_ArithmeticException), true);
2266 StubRoutines::_throw_NullPointerException_entry = generate_throw_exception("NullPointerException throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException), true);
2267 StubRoutines::_throw_NullPointerException_at_call_entry= generate_throw_exception("NullPointerException at call throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException_at_call), false);
2268 StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError), false);
2270 //------------------------------------------------------------------------------------------------------------------------
2271 // entry points that are platform specific
2273 // support for verify_oop (must happen after universe_init)
2274 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
2276 // arraycopy stubs used by compilers
2277 generate_arraycopy_stubs();
2279 // generic method handle stubs
2280 if (EnableMethodHandles && SystemDictionary::MethodHandle_klass() != NULL) {
2281 for (MethodHandles::EntryKind ek = MethodHandles::_EK_FIRST;
2282 ek < MethodHandles::_EK_LIMIT;
2283 ek = MethodHandles::EntryKind(1 + (int)ek)) {
2284 StubCodeMark mark(this, "MethodHandle", MethodHandles::entry_name(ek));
2285 MethodHandles::generate_method_handle_stub(_masm, ek);
2286 }
2287 }
2289 generate_math_stubs();
2290 }
2293 public:
2294 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
2295 if (all) {
2296 generate_all();
2297 } else {
2298 generate_initial();
2299 }
2300 }
2301 }; // end class declaration
2304 void StubGenerator_generate(CodeBuffer* code, bool all) {
2305 StubGenerator g(code, all);
2306 }