• file
• revisions
• annotate
• diff
• comparison
• raw

src/cpu/x86/vm/stubGenerator_x86_64.cpp@9ee9cf798b59 (annotated)

src/cpu/x86/vm/stubGenerator_x86_64.cpp

Thu, 02 Oct 2008 19:58:19 -0700

author

xdono

date

Thu, 02 Oct 2008 19:58:19 -0700

changeset 772

9ee9cf798b59

parent 739

dc7f315e41f7

child 797

f8199438385b

permissions

-rw-r--r--

6754988: Update copyright year
Summary: Update for files that have been modified starting July 2008
Reviewed-by: ohair, tbell

duke@435	1	/*
xdono@631	2	* Copyright 2003-2008 Sun Microsystems, Inc. All Rights Reserved.
duke@435	3	* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
duke@435	4	*
duke@435	5	* This code is free software; you can redistribute it and/or modify it
duke@435	6	* under the terms of the GNU General Public License version 2 only, as
duke@435	7	* published by the Free Software Foundation.
duke@435	8	*
duke@435	9	* This code is distributed in the hope that it will be useful, but WITHOUT
duke@435	10	* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
duke@435	11	* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
duke@435	12	* version 2 for more details (a copy is included in the LICENSE file that
duke@435	13	* accompanied this code).
duke@435	14	*
duke@435	15	* You should have received a copy of the GNU General Public License version
duke@435	16	* 2 along with this work; if not, write to the Free Software Foundation,
duke@435	17	* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
duke@435	18	*
duke@435	19	* Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
duke@435	20	* CA 95054 USA or visit www.sun.com if you need additional information or
duke@435	21	* have any questions.
duke@435	22	*
duke@435	23	*/
duke@435	24
duke@435	25	#include "incls/_precompiled.incl"
duke@435	26	#include "incls/_stubGenerator_x86_64.cpp.incl"
duke@435	27
duke@435	28	// Declaration and definition of StubGenerator (no .hpp file).
duke@435	29	// For a more detailed description of the stub routine structure
duke@435	30	// see the comment in stubRoutines.hpp
duke@435	31
duke@435	32	#define __ _masm->
coleenp@548	33	#define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8)
never@739	34	#define a__ ((Assembler*)_masm)->
duke@435	35
duke@435	36	#ifdef PRODUCT
duke@435	37	#define BLOCK_COMMENT(str) /* nothing */
duke@435	38	#else
duke@435	39	#define BLOCK_COMMENT(str) __ block_comment(str)
duke@435	40	#endif
duke@435	41
duke@435	42	#define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
duke@435	43	const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
duke@435	44
duke@435	45	// Stub Code definitions
duke@435	46
duke@435	47	static address handle_unsafe_access() {
duke@435	48	JavaThread* thread = JavaThread::current();
duke@435	49	address pc = thread->saved_exception_pc();
duke@435	50	// pc is the instruction which we must emulate
duke@435	51	// doing a no-op is fine: return garbage from the load
duke@435	52	// therefore, compute npc
duke@435	53	address npc = Assembler::locate_next_instruction(pc);
duke@435	54
duke@435	55	// request an async exception
duke@435	56	thread->set_pending_unsafe_access_error();
duke@435	57
duke@435	58	// return address of next instruction to execute
duke@435	59	return npc;
duke@435	60	}
duke@435	61
duke@435	62	class StubGenerator: public StubCodeGenerator {
duke@435	63	private:
duke@435	64
duke@435	65	#ifdef PRODUCT
duke@435	66	#define inc_counter_np(counter) (0)
duke@435	67	#else
duke@435	68	void inc_counter_np_(int& counter) {
duke@435	69	__ incrementl(ExternalAddress((address)&counter));
duke@435	70	}
duke@435	71	#define inc_counter_np(counter) \
duke@435	72	BLOCK_COMMENT("inc_counter " #counter); \
duke@435	73	inc_counter_np_(counter);
duke@435	74	#endif
duke@435	75
duke@435	76	// Call stubs are used to call Java from C
duke@435	77	//
duke@435	78	// Linux Arguments:
duke@435	79	// c_rarg0: call wrapper address address
duke@435	80	// c_rarg1: result address
duke@435	81	// c_rarg2: result type BasicType
duke@435	82	// c_rarg3: method methodOop
duke@435	83	// c_rarg4: (interpreter) entry point address
duke@435	84	// c_rarg5: parameters intptr_t*
duke@435	85	// 16(rbp): parameter size (in words) int
duke@435	86	// 24(rbp): thread Thread*
duke@435	87	//
duke@435	88	// [ return_from_Java ] <--- rsp
duke@435	89	// [ argument word n ]
duke@435	90	// ...
duke@435	91	// -12 [ argument word 1 ]
duke@435	92	// -11 [ saved r15 ] <--- rsp_after_call
duke@435	93	// -10 [ saved r14 ]
duke@435	94	// -9 [ saved r13 ]
duke@435	95	// -8 [ saved r12 ]
duke@435	96	// -7 [ saved rbx ]
duke@435	97	// -6 [ call wrapper ]
duke@435	98	// -5 [ result ]
duke@435	99	// -4 [ result type ]
duke@435	100	// -3 [ method ]
duke@435	101	// -2 [ entry point ]
duke@435	102	// -1 [ parameters ]
duke@435	103	// 0 [ saved rbp ] <--- rbp
duke@435	104	// 1 [ return address ]
duke@435	105	// 2 [ parameter size ]
duke@435	106	// 3 [ thread ]
duke@435	107	//
duke@435	108	// Windows Arguments:
duke@435	109	// c_rarg0: call wrapper address address
duke@435	110	// c_rarg1: result address
duke@435	111	// c_rarg2: result type BasicType
duke@435	112	// c_rarg3: method methodOop
duke@435	113	// 48(rbp): (interpreter) entry point address
duke@435	114	// 56(rbp): parameters intptr_t*
duke@435	115	// 64(rbp): parameter size (in words) int
duke@435	116	// 72(rbp): thread Thread*
duke@435	117	//
duke@435	118	// [ return_from_Java ] <--- rsp
duke@435	119	// [ argument word n ]
duke@435	120	// ...
duke@435	121	// -8 [ argument word 1 ]
duke@435	122	// -7 [ saved r15 ] <--- rsp_after_call
duke@435	123	// -6 [ saved r14 ]
duke@435	124	// -5 [ saved r13 ]
duke@435	125	// -4 [ saved r12 ]
duke@435	126	// -3 [ saved rdi ]
duke@435	127	// -2 [ saved rsi ]
duke@435	128	// -1 [ saved rbx ]
duke@435	129	// 0 [ saved rbp ] <--- rbp
duke@435	130	// 1 [ return address ]
duke@435	131	// 2 [ call wrapper ]
duke@435	132	// 3 [ result ]
duke@435	133	// 4 [ result type ]
duke@435	134	// 5 [ method ]
duke@435	135	// 6 [ entry point ]
duke@435	136	// 7 [ parameters ]
duke@435	137	// 8 [ parameter size ]
duke@435	138	// 9 [ thread ]
duke@435	139	//
duke@435	140	// Windows reserves the callers stack space for arguments 1-4.
duke@435	141	// We spill c_rarg0-c_rarg3 to this space.
duke@435	142
duke@435	143	// Call stub stack layout word offsets from rbp
duke@435	144	enum call_stub_layout {
duke@435	145	#ifdef _WIN64
duke@435	146	rsp_after_call_off = -7,
duke@435	147	r15_off = rsp_after_call_off,
duke@435	148	r14_off = -6,
duke@435	149	r13_off = -5,
duke@435	150	r12_off = -4,
duke@435	151	rdi_off = -3,
duke@435	152	rsi_off = -2,
duke@435	153	rbx_off = -1,
duke@435	154	rbp_off = 0,
duke@435	155	retaddr_off = 1,
duke@435	156	call_wrapper_off = 2,
duke@435	157	result_off = 3,
duke@435	158	result_type_off = 4,
duke@435	159	method_off = 5,
duke@435	160	entry_point_off = 6,
duke@435	161	parameters_off = 7,
duke@435	162	parameter_size_off = 8,
duke@435	163	thread_off = 9
duke@435	164	#else
duke@435	165	rsp_after_call_off = -12,
duke@435	166	mxcsr_off = rsp_after_call_off,
duke@435	167	r15_off = -11,
duke@435	168	r14_off = -10,
duke@435	169	r13_off = -9,
duke@435	170	r12_off = -8,
duke@435	171	rbx_off = -7,
duke@435	172	call_wrapper_off = -6,
duke@435	173	result_off = -5,
duke@435	174	result_type_off = -4,
duke@435	175	method_off = -3,
duke@435	176	entry_point_off = -2,
duke@435	177	parameters_off = -1,
duke@435	178	rbp_off = 0,
duke@435	179	retaddr_off = 1,
duke@435	180	parameter_size_off = 2,
duke@435	181	thread_off = 3
duke@435	182	#endif
duke@435	183	};
duke@435	184
duke@435	185	address generate_call_stub(address& return_address) {
duke@435	186	assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
duke@435	187	(int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
duke@435	188	"adjust this code");
duke@435	189	StubCodeMark mark(this, "StubRoutines", "call_stub");
duke@435	190	address start = __ pc();
duke@435	191
duke@435	192	// same as in generate_catch_exception()!
duke@435	193	const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
duke@435	194
duke@435	195	const Address call_wrapper (rbp, call_wrapper_off * wordSize);
duke@435	196	const Address result (rbp, result_off * wordSize);
duke@435	197	const Address result_type (rbp, result_type_off * wordSize);
duke@435	198	const Address method (rbp, method_off * wordSize);
duke@435	199	const Address entry_point (rbp, entry_point_off * wordSize);
duke@435	200	const Address parameters (rbp, parameters_off * wordSize);
duke@435	201	const Address parameter_size(rbp, parameter_size_off * wordSize);
duke@435	202
duke@435	203	// same as in generate_catch_exception()!
duke@435	204	const Address thread (rbp, thread_off * wordSize);
duke@435	205
duke@435	206	const Address r15_save(rbp, r15_off * wordSize);
duke@435	207	const Address r14_save(rbp, r14_off * wordSize);
duke@435	208	const Address r13_save(rbp, r13_off * wordSize);
duke@435	209	const Address r12_save(rbp, r12_off * wordSize);
duke@435	210	const Address rbx_save(rbp, rbx_off * wordSize);
duke@435	211
duke@435	212	// stub code
duke@435	213	__ enter();
never@739	214	__ subptr(rsp, -rsp_after_call_off * wordSize);
duke@435	215
duke@435	216	// save register parameters
duke@435	217	#ifndef _WIN64
never@739	218	__ movptr(parameters, c_rarg5); // parameters
never@739	219	__ movptr(entry_point, c_rarg4); // entry_point
duke@435	220	#endif
duke@435	221
never@739	222	__ movptr(method, c_rarg3); // method
never@739	223	__ movl(result_type, c_rarg2); // result type
never@739	224	__ movptr(result, c_rarg1); // result
never@739	225	__ movptr(call_wrapper, c_rarg0); // call wrapper
duke@435	226
duke@435	227	// save regs belonging to calling function
never@739	228	__ movptr(rbx_save, rbx);
never@739	229	__ movptr(r12_save, r12);
never@739	230	__ movptr(r13_save, r13);
never@739	231	__ movptr(r14_save, r14);
never@739	232	__ movptr(r15_save, r15);
duke@435	233
duke@435	234	#ifdef _WIN64
duke@435	235	const Address rdi_save(rbp, rdi_off * wordSize);
duke@435	236	const Address rsi_save(rbp, rsi_off * wordSize);
duke@435	237
never@739	238	__ movptr(rsi_save, rsi);
never@739	239	__ movptr(rdi_save, rdi);
duke@435	240	#else
duke@435	241	const Address mxcsr_save(rbp, mxcsr_off * wordSize);
duke@435	242	{
duke@435	243	Label skip_ldmx;
duke@435	244	__ stmxcsr(mxcsr_save);
duke@435	245	__ movl(rax, mxcsr_save);
duke@435	246	__ andl(rax, MXCSR_MASK); // Only check control and mask bits
never@739	247	ExternalAddress mxcsr_std(StubRoutines::x86::mxcsr_std());
duke@435	248	__ cmp32(rax, mxcsr_std);
duke@435	249	__ jcc(Assembler::equal, skip_ldmx);
duke@435	250	__ ldmxcsr(mxcsr_std);
duke@435	251	__ bind(skip_ldmx);
duke@435	252	}
duke@435	253	#endif
duke@435	254
duke@435	255	// Load up thread register
never@739	256	__ movptr(r15_thread, thread);
coleenp@548	257	__ reinit_heapbase();
duke@435	258
duke@435	259	#ifdef ASSERT
duke@435	260	// make sure we have no pending exceptions
duke@435	261	{
duke@435	262	Label L;
never@739	263	__ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
duke@435	264	__ jcc(Assembler::equal, L);
duke@435	265	__ stop("StubRoutines::call_stub: entered with pending exception");
duke@435	266	__ bind(L);
duke@435	267	}
duke@435	268	#endif
duke@435	269
duke@435	270	// pass parameters if any
duke@435	271	BLOCK_COMMENT("pass parameters if any");
duke@435	272	Label parameters_done;
duke@435	273	__ movl(c_rarg3, parameter_size);
duke@435	274	__ testl(c_rarg3, c_rarg3);
duke@435	275	__ jcc(Assembler::zero, parameters_done);
duke@435	276
duke@435	277	Label loop;
never@739	278	__ movptr(c_rarg2, parameters); // parameter pointer
never@739	279	__ movl(c_rarg1, c_rarg3); // parameter counter is in c_rarg1
duke@435	280	__ BIND(loop);
duke@435	281	if (TaggedStackInterpreter) {
never@739	282	__ movl(rax, Address(c_rarg2, 0)); // get tag
never@739	283	__ addptr(c_rarg2, wordSize); // advance to next tag
never@739	284	__ push(rax); // pass tag
duke@435	285	}
never@739	286	__ movptr(rax, Address(c_rarg2, 0));// get parameter
never@739	287	__ addptr(c_rarg2, wordSize); // advance to next parameter
never@739	288	__ decrementl(c_rarg1); // decrement counter
never@739	289	__ push(rax); // pass parameter
duke@435	290	__ jcc(Assembler::notZero, loop);
duke@435	291
duke@435	292	// call Java function
duke@435	293	__ BIND(parameters_done);
never@739	294	__ movptr(rbx, method); // get methodOop
never@739	295	__ movptr(c_rarg1, entry_point); // get entry_point
never@739	296	__ mov(r13, rsp); // set sender sp
duke@435	297	BLOCK_COMMENT("call Java function");
duke@435	298	__ call(c_rarg1);
duke@435	299
duke@435	300	BLOCK_COMMENT("call_stub_return_address:");
duke@435	301	return_address = __ pc();
duke@435	302
duke@435	303	// store result depending on type (everything that is not
duke@435	304	// T_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
never@739	305	__ movptr(c_rarg0, result);
duke@435	306	Label is_long, is_float, is_double, exit;
duke@435	307	__ movl(c_rarg1, result_type);
duke@435	308	__ cmpl(c_rarg1, T_OBJECT);
duke@435	309	__ jcc(Assembler::equal, is_long);
duke@435	310	__ cmpl(c_rarg1, T_LONG);
duke@435	311	__ jcc(Assembler::equal, is_long);
duke@435	312	__ cmpl(c_rarg1, T_FLOAT);
duke@435	313	__ jcc(Assembler::equal, is_float);
duke@435	314	__ cmpl(c_rarg1, T_DOUBLE);
duke@435	315	__ jcc(Assembler::equal, is_double);
duke@435	316
duke@435	317	// handle T_INT case
duke@435	318	__ movl(Address(c_rarg0, 0), rax);
duke@435	319
duke@435	320	__ BIND(exit);
duke@435	321
duke@435	322	// pop parameters
never@739	323	__ lea(rsp, rsp_after_call);
duke@435	324
duke@435	325	#ifdef ASSERT
duke@435	326	// verify that threads correspond
duke@435	327	{
duke@435	328	Label L, S;
never@739	329	__ cmpptr(r15_thread, thread);
duke@435	330	__ jcc(Assembler::notEqual, S);
duke@435	331	__ get_thread(rbx);
never@739	332	__ cmpptr(r15_thread, rbx);
duke@435	333	__ jcc(Assembler::equal, L);
duke@435	334	__ bind(S);
duke@435	335	__ jcc(Assembler::equal, L);
duke@435	336	__ stop("StubRoutines::call_stub: threads must correspond");
duke@435	337	__ bind(L);
duke@435	338	}
duke@435	339	#endif
duke@435	340
duke@435	341	// restore regs belonging to calling function
never@739	342	__ movptr(r15, r15_save);
never@739	343	__ movptr(r14, r14_save);
never@739	344	__ movptr(r13, r13_save);
never@739	345	__ movptr(r12, r12_save);
never@739	346	__ movptr(rbx, rbx_save);
duke@435	347
duke@435	348	#ifdef _WIN64
never@739	349	__ movptr(rdi, rdi_save);
never@739	350	__ movptr(rsi, rsi_save);
duke@435	351	#else
duke@435	352	__ ldmxcsr(mxcsr_save);
duke@435	353	#endif
duke@435	354
duke@435	355	// restore rsp
never@739	356	__ addptr(rsp, -rsp_after_call_off * wordSize);
duke@435	357
duke@435	358	// return
never@739	359	__ pop(rbp);
duke@435	360	__ ret(0);
duke@435	361
duke@435	362	// handle return types different from T_INT
duke@435	363	__ BIND(is_long);
duke@435	364	__ movq(Address(c_rarg0, 0), rax);
duke@435	365	__ jmp(exit);
duke@435	366
duke@435	367	__ BIND(is_float);
duke@435	368	__ movflt(Address(c_rarg0, 0), xmm0);
duke@435	369	__ jmp(exit);
duke@435	370
duke@435	371	__ BIND(is_double);
duke@435	372	__ movdbl(Address(c_rarg0, 0), xmm0);
duke@435	373	__ jmp(exit);
duke@435	374
duke@435	375	return start;
duke@435	376	}
duke@435	377
duke@435	378	// Return point for a Java call if there's an exception thrown in
duke@435	379	// Java code. The exception is caught and transformed into a
duke@435	380	// pending exception stored in JavaThread that can be tested from
duke@435	381	// within the VM.
duke@435	382	//
duke@435	383	// Note: Usually the parameters are removed by the callee. In case
duke@435	384	// of an exception crossing an activation frame boundary, that is
duke@435	385	// not the case if the callee is compiled code => need to setup the
duke@435	386	// rsp.
duke@435	387	//
duke@435	388	// rax: exception oop
duke@435	389
duke@435	390	address generate_catch_exception() {
duke@435	391	StubCodeMark mark(this, "StubRoutines", "catch_exception");
duke@435	392	address start = __ pc();
duke@435	393
duke@435	394	// same as in generate_call_stub():
duke@435	395	const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
duke@435	396	const Address thread (rbp, thread_off * wordSize);
duke@435	397
duke@435	398	#ifdef ASSERT
duke@435	399	// verify that threads correspond
duke@435	400	{
duke@435	401	Label L, S;
never@739	402	__ cmpptr(r15_thread, thread);
duke@435	403	__ jcc(Assembler::notEqual, S);
duke@435	404	__ get_thread(rbx);
never@739	405	__ cmpptr(r15_thread, rbx);
duke@435	406	__ jcc(Assembler::equal, L);
duke@435	407	__ bind(S);
duke@435	408	__ stop("StubRoutines::catch_exception: threads must correspond");
duke@435	409	__ bind(L);
duke@435	410	}
duke@435	411	#endif
duke@435	412
duke@435	413	// set pending exception
duke@435	414	__ verify_oop(rax);
duke@435	415
never@739	416	__ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
duke@435	417	__ lea(rscratch1, ExternalAddress((address)__FILE__));
never@739	418	__ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
duke@435	419	__ movl(Address(r15_thread, Thread::exception_line_offset()), (int) __LINE__);
duke@435	420
duke@435	421	// complete return to VM
duke@435	422	assert(StubRoutines::_call_stub_return_address != NULL,
duke@435	423	"_call_stub_return_address must have been generated before");
duke@435	424	__ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
duke@435	425
duke@435	426	return start;
duke@435	427	}
duke@435	428
duke@435	429	// Continuation point for runtime calls returning with a pending
duke@435	430	// exception. The pending exception check happened in the runtime
duke@435	431	// or native call stub. The pending exception in Thread is
duke@435	432	// converted into a Java-level exception.
duke@435	433	//
duke@435	434	// Contract with Java-level exception handlers:
duke@435	435	// rax: exception
duke@435	436	// rdx: throwing pc
duke@435	437	//
duke@435	438	// NOTE: At entry of this stub, exception-pc must be on stack !!
duke@435	439
duke@435	440	address generate_forward_exception() {
duke@435	441	StubCodeMark mark(this, "StubRoutines", "forward exception");
duke@435	442	address start = __ pc();
duke@435	443
duke@435	444	// Upon entry, the sp points to the return address returning into
duke@435	445	// Java (interpreted or compiled) code; i.e., the return address
duke@435	446	// becomes the throwing pc.
duke@435	447	//
duke@435	448	// Arguments pushed before the runtime call are still on the stack
duke@435	449	// but the exception handler will reset the stack pointer ->
duke@435	450	// ignore them. A potential result in registers can be ignored as
duke@435	451	// well.
duke@435	452
duke@435	453	#ifdef ASSERT
duke@435	454	// make sure this code is only executed if there is a pending exception
duke@435	455	{
duke@435	456	Label L;
never@739	457	__ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL);
duke@435	458	__ jcc(Assembler::notEqual, L);
duke@435	459	__ stop("StubRoutines::forward exception: no pending exception (1)");
duke@435	460	__ bind(L);
duke@435	461	}
duke@435	462	#endif
duke@435	463
duke@435	464	// compute exception handler into rbx
never@739	465	__ movptr(c_rarg0, Address(rsp, 0));
duke@435	466	BLOCK_COMMENT("call exception_handler_for_return_address");
duke@435	467	__ call_VM_leaf(CAST_FROM_FN_PTR(address,
duke@435	468	SharedRuntime::exception_handler_for_return_address),
duke@435	469	c_rarg0);
never@739	470	__ mov(rbx, rax);
duke@435	471
duke@435	472	// setup rax & rdx, remove return address & clear pending exception
never@739	473	__ pop(rdx);
never@739	474	__ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
duke@435	475	__ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int)NULL_WORD);
duke@435	476
duke@435	477	#ifdef ASSERT
duke@435	478	// make sure exception is set
duke@435	479	{
duke@435	480	Label L;
never@739	481	__ testptr(rax, rax);
duke@435	482	__ jcc(Assembler::notEqual, L);
duke@435	483	__ stop("StubRoutines::forward exception: no pending exception (2)");
duke@435	484	__ bind(L);
duke@435	485	}
duke@435	486	#endif
duke@435	487
duke@435	488	// continue at exception handler (return address removed)
duke@435	489	// rax: exception
duke@435	490	// rbx: exception handler
duke@435	491	// rdx: throwing pc
duke@435	492	__ verify_oop(rax);
duke@435	493	__ jmp(rbx);
duke@435	494
duke@435	495	return start;
duke@435	496	}
duke@435	497
duke@435	498	// Support for jint atomic::xchg(jint exchange_value, volatile jint* dest)
duke@435	499	//
duke@435	500	// Arguments :
duke@435	501	// c_rarg0: exchange_value
duke@435	502	// c_rarg0: dest
duke@435	503	//
duke@435	504	// Result:
duke@435	505	// *dest <- ex, return (orig *dest)
duke@435	506	address generate_atomic_xchg() {
duke@435	507	StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
duke@435	508	address start = __ pc();
duke@435	509
duke@435	510	__ movl(rax, c_rarg0); // Copy to eax we need a return value anyhow
duke@435	511	__ xchgl(rax, Address(c_rarg1, 0)); // automatic LOCK
duke@435	512	__ ret(0);
duke@435	513
duke@435	514	return start;
duke@435	515	}
duke@435	516
duke@435	517	// Support for intptr_t atomic::xchg_ptr(intptr_t exchange_value, volatile intptr_t* dest)
duke@435	518	//
duke@435	519	// Arguments :
duke@435	520	// c_rarg0: exchange_value
duke@435	521	// c_rarg1: dest
duke@435	522	//
duke@435	523	// Result:
duke@435	524	// *dest <- ex, return (orig *dest)
duke@435	525	address generate_atomic_xchg_ptr() {
duke@435	526	StubCodeMark mark(this, "StubRoutines", "atomic_xchg_ptr");
duke@435	527	address start = __ pc();
duke@435	528
never@739	529	__ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
never@739	530	__ xchgptr(rax, Address(c_rarg1, 0)); // automatic LOCK
duke@435	531	__ ret(0);
duke@435	532
duke@435	533	return start;
duke@435	534	}
duke@435	535
duke@435	536	// Support for jint atomic::atomic_cmpxchg(jint exchange_value, volatile jint* dest,
duke@435	537	// jint compare_value)
duke@435	538	//
duke@435	539	// Arguments :
duke@435	540	// c_rarg0: exchange_value
duke@435	541	// c_rarg1: dest
duke@435	542	// c_rarg2: compare_value
duke@435	543	//
duke@435	544	// Result:
duke@435	545	// if ( compare_value == *dest ) {
duke@435	546	// *dest = exchange_value
duke@435	547	// return compare_value;
duke@435	548	// else
duke@435	549	// return *dest;
duke@435	550	address generate_atomic_cmpxchg() {
duke@435	551	StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg");
duke@435	552	address start = __ pc();
duke@435	553
duke@435	554	__ movl(rax, c_rarg2);
duke@435	555	if ( os::is_MP() ) __ lock();
duke@435	556	__ cmpxchgl(c_rarg0, Address(c_rarg1, 0));
duke@435	557	__ ret(0);
duke@435	558
duke@435	559	return start;
duke@435	560	}
duke@435	561
duke@435	562	// Support for jint atomic::atomic_cmpxchg_long(jlong exchange_value,
duke@435	563	// volatile jlong* dest,
duke@435	564	// jlong compare_value)
duke@435	565	// Arguments :
duke@435	566	// c_rarg0: exchange_value
duke@435	567	// c_rarg1: dest
duke@435	568	// c_rarg2: compare_value
duke@435	569	//
duke@435	570	// Result:
duke@435	571	// if ( compare_value == *dest ) {
duke@435	572	// *dest = exchange_value
duke@435	573	// return compare_value;
duke@435	574	// else
duke@435	575	// return *dest;
duke@435	576	address generate_atomic_cmpxchg_long() {
duke@435	577	StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long");
duke@435	578	address start = __ pc();
duke@435	579
duke@435	580	__ movq(rax, c_rarg2);
duke@435	581	if ( os::is_MP() ) __ lock();
duke@435	582	__ cmpxchgq(c_rarg0, Address(c_rarg1, 0));
duke@435	583	__ ret(0);
duke@435	584
duke@435	585	return start;
duke@435	586	}
duke@435	587
duke@435	588	// Support for jint atomic::add(jint add_value, volatile jint* dest)
duke@435	589	//
duke@435	590	// Arguments :
duke@435	591	// c_rarg0: add_value
duke@435	592	// c_rarg1: dest
duke@435	593	//
duke@435	594	// Result:
duke@435	595	// *dest += add_value
duke@435	596	// return *dest;
duke@435	597	address generate_atomic_add() {
duke@435	598	StubCodeMark mark(this, "StubRoutines", "atomic_add");
duke@435	599	address start = __ pc();
duke@435	600
duke@435	601	__ movl(rax, c_rarg0);
duke@435	602	if ( os::is_MP() ) __ lock();
duke@435	603	__ xaddl(Address(c_rarg1, 0), c_rarg0);
duke@435	604	__ addl(rax, c_rarg0);
duke@435	605	__ ret(0);
duke@435	606
duke@435	607	return start;
duke@435	608	}
duke@435	609
duke@435	610	// Support for intptr_t atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest)
duke@435	611	//
duke@435	612	// Arguments :
duke@435	613	// c_rarg0: add_value
duke@435	614	// c_rarg1: dest
duke@435	615	//
duke@435	616	// Result:
duke@435	617	// *dest += add_value
duke@435	618	// return *dest;
duke@435	619	address generate_atomic_add_ptr() {
duke@435	620	StubCodeMark mark(this, "StubRoutines", "atomic_add_ptr");
duke@435	621	address start = __ pc();
duke@435	622
never@739	623	__ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
duke@435	624	if ( os::is_MP() ) __ lock();
never@739	625	__ xaddptr(Address(c_rarg1, 0), c_rarg0);
never@739	626	__ addptr(rax, c_rarg0);
duke@435	627	__ ret(0);
duke@435	628
duke@435	629	return start;
duke@435	630	}
duke@435	631
duke@435	632	// Support for intptr_t OrderAccess::fence()
duke@435	633	//
duke@435	634	// Arguments :
duke@435	635	//
duke@435	636	// Result:
duke@435	637	address generate_orderaccess_fence() {
duke@435	638	StubCodeMark mark(this, "StubRoutines", "orderaccess_fence");
duke@435	639	address start = __ pc();
duke@435	640	__ mfence();
duke@435	641	__ ret(0);
duke@435	642
duke@435	643	return start;
duke@435	644	}
duke@435	645
duke@435	646	// Support for intptr_t get_previous_fp()
duke@435	647	//
duke@435	648	// This routine is used to find the previous frame pointer for the
duke@435	649	// caller (current_frame_guess). This is used as part of debugging
duke@435	650	// ps() is seemingly lost trying to find frames.
duke@435	651	// This code assumes that caller current_frame_guess) has a frame.
duke@435	652	address generate_get_previous_fp() {
duke@435	653	StubCodeMark mark(this, "StubRoutines", "get_previous_fp");
duke@435	654	const Address old_fp(rbp, 0);
duke@435	655	const Address older_fp(rax, 0);
duke@435	656	address start = __ pc();
duke@435	657
duke@435	658	__ enter();
never@739	659	__ movptr(rax, old_fp); // callers fp
never@739	660	__ movptr(rax, older_fp); // the frame for ps()
never@739	661	__ pop(rbp);
duke@435	662	__ ret(0);
duke@435	663
duke@435	664	return start;
duke@435	665	}
duke@435	666
duke@435	667	//----------------------------------------------------------------------------------------------------
duke@435	668	// Support for void verify_mxcsr()
duke@435	669	//
duke@435	670	// This routine is used with -Xcheck:jni to verify that native
duke@435	671	// JNI code does not return to Java code without restoring the
duke@435	672	// MXCSR register to our expected state.
duke@435	673
duke@435	674	address generate_verify_mxcsr() {
duke@435	675	StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
duke@435	676	address start = __ pc();
duke@435	677
duke@435	678	const Address mxcsr_save(rsp, 0);
duke@435	679
duke@435	680	if (CheckJNICalls) {
duke@435	681	Label ok_ret;
never@739	682	__ push(rax);
never@739	683	__ subptr(rsp, wordSize); // allocate a temp location
duke@435	684	__ stmxcsr(mxcsr_save);
duke@435	685	__ movl(rax, mxcsr_save);
duke@435	686	__ andl(rax, MXCSR_MASK); // Only check control and mask bits
never@739	687	__ cmpl(rax, *(int *)(StubRoutines::x86::mxcsr_std()));
duke@435	688	__ jcc(Assembler::equal, ok_ret);
duke@435	689
duke@435	690	__ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
duke@435	691
never@739	692	__ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std()));
duke@435	693
duke@435	694	__ bind(ok_ret);
never@739	695	__ addptr(rsp, wordSize);
never@739	696	__ pop(rax);
duke@435	697	}
duke@435	698
duke@435	699	__ ret(0);
duke@435	700
duke@435	701	return start;
duke@435	702	}
duke@435	703
duke@435	704	address generate_f2i_fixup() {
duke@435	705	StubCodeMark mark(this, "StubRoutines", "f2i_fixup");
duke@435	706	Address inout(rsp, 5 * wordSize); // return address + 4 saves
duke@435	707
duke@435	708	address start = __ pc();
duke@435	709
duke@435	710	Label L;
duke@435	711
never@739	712	__ push(rax);
never@739	713	__ push(c_rarg3);
never@739	714	__ push(c_rarg2);
never@739	715	__ push(c_rarg1);
duke@435	716
duke@435	717	__ movl(rax, 0x7f800000);
duke@435	718	__ xorl(c_rarg3, c_rarg3);
duke@435	719	__ movl(c_rarg2, inout);
duke@435	720	__ movl(c_rarg1, c_rarg2);
duke@435	721	__ andl(c_rarg1, 0x7fffffff);
duke@435	722	__ cmpl(rax, c_rarg1); // NaN? -> 0
duke@435	723	__ jcc(Assembler::negative, L);
duke@435	724	__ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
duke@435	725	__ movl(c_rarg3, 0x80000000);
duke@435	726	__ movl(rax, 0x7fffffff);
duke@435	727	__ cmovl(Assembler::positive, c_rarg3, rax);
duke@435	728
duke@435	729	__ bind(L);
never@739	730	__ movptr(inout, c_rarg3);
never@739	731
never@739	732	__ pop(c_rarg1);
never@739	733	__ pop(c_rarg2);
never@739	734	__ pop(c_rarg3);
never@739	735	__ pop(rax);
duke@435	736
duke@435	737	__ ret(0);
duke@435	738
duke@435	739	return start;
duke@435	740	}
duke@435	741
duke@435	742	address generate_f2l_fixup() {
duke@435	743	StubCodeMark mark(this, "StubRoutines", "f2l_fixup");
duke@435	744	Address inout(rsp, 5 * wordSize); // return address + 4 saves
duke@435	745	address start = __ pc();
duke@435	746
duke@435	747	Label L;
duke@435	748
never@739	749	__ push(rax);
never@739	750	__ push(c_rarg3);
never@739	751	__ push(c_rarg2);
never@739	752	__ push(c_rarg1);
duke@435	753
duke@435	754	__ movl(rax, 0x7f800000);
duke@435	755	__ xorl(c_rarg3, c_rarg3);
duke@435	756	__ movl(c_rarg2, inout);
duke@435	757	__ movl(c_rarg1, c_rarg2);
duke@435	758	__ andl(c_rarg1, 0x7fffffff);
duke@435	759	__ cmpl(rax, c_rarg1); // NaN? -> 0
duke@435	760	__ jcc(Assembler::negative, L);
duke@435	761	__ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
duke@435	762	__ mov64(c_rarg3, 0x8000000000000000);
duke@435	763	__ mov64(rax, 0x7fffffffffffffff);
never@739	764	__ cmov(Assembler::positive, c_rarg3, rax);
duke@435	765
duke@435	766	__ bind(L);
never@739	767	__ movptr(inout, c_rarg3);
never@739	768
never@739	769	__ pop(c_rarg1);
never@739	770	__ pop(c_rarg2);
never@739	771	__ pop(c_rarg3);
never@739	772	__ pop(rax);
duke@435	773
duke@435	774	__ ret(0);
duke@435	775
duke@435	776	return start;
duke@435	777	}
duke@435	778
duke@435	779	address generate_d2i_fixup() {
duke@435	780	StubCodeMark mark(this, "StubRoutines", "d2i_fixup");
duke@435	781	Address inout(rsp, 6 * wordSize); // return address + 5 saves
duke@435	782
duke@435	783	address start = __ pc();
duke@435	784
duke@435	785	Label L;
duke@435	786
never@739	787	__ push(rax);
never@739	788	__ push(c_rarg3);
never@739	789	__ push(c_rarg2);
never@739	790	__ push(c_rarg1);
never@739	791	__ push(c_rarg0);
duke@435	792
duke@435	793	__ movl(rax, 0x7ff00000);
duke@435	794	__ movq(c_rarg2, inout);
duke@435	795	__ movl(c_rarg3, c_rarg2);
never@739	796	__ mov(c_rarg1, c_rarg2);
never@739	797	__ mov(c_rarg0, c_rarg2);
duke@435	798	__ negl(c_rarg3);
never@739	799	__ shrptr(c_rarg1, 0x20);
duke@435	800	__ orl(c_rarg3, c_rarg2);
duke@435	801	__ andl(c_rarg1, 0x7fffffff);
duke@435	802	__ xorl(c_rarg2, c_rarg2);
duke@435	803	__ shrl(c_rarg3, 0x1f);
duke@435	804	__ orl(c_rarg1, c_rarg3);
duke@435	805	__ cmpl(rax, c_rarg1);
duke@435	806	__ jcc(Assembler::negative, L); // NaN -> 0
never@739	807	__ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
duke@435	808	__ movl(c_rarg2, 0x80000000);
duke@435	809	__ movl(rax, 0x7fffffff);
never@739	810	__ cmov(Assembler::positive, c_rarg2, rax);
duke@435	811
duke@435	812	__ bind(L);
never@739	813	__ movptr(inout, c_rarg2);
never@739	814
never@739	815	__ pop(c_rarg0);
never@739	816	__ pop(c_rarg1);
never@739	817	__ pop(c_rarg2);
never@739	818	__ pop(c_rarg3);
never@739	819	__ pop(rax);
duke@435	820
duke@435	821	__ ret(0);
duke@435	822
duke@435	823	return start;
duke@435	824	}
duke@435	825
duke@435	826	address generate_d2l_fixup() {
duke@435	827	StubCodeMark mark(this, "StubRoutines", "d2l_fixup");
duke@435	828	Address inout(rsp, 6 * wordSize); // return address + 5 saves
duke@435	829
duke@435	830	address start = __ pc();
duke@435	831
duke@435	832	Label L;
duke@435	833
never@739	834	__ push(rax);
never@739	835	__ push(c_rarg3);
never@739	836	__ push(c_rarg2);
never@739	837	__ push(c_rarg1);
never@739	838	__ push(c_rarg0);
duke@435	839
duke@435	840	__ movl(rax, 0x7ff00000);
duke@435	841	__ movq(c_rarg2, inout);
duke@435	842	__ movl(c_rarg3, c_rarg2);
never@739	843	__ mov(c_rarg1, c_rarg2);
never@739	844	__ mov(c_rarg0, c_rarg2);
duke@435	845	__ negl(c_rarg3);
never@739	846	__ shrptr(c_rarg1, 0x20);
duke@435	847	__ orl(c_rarg3, c_rarg2);
duke@435	848	__ andl(c_rarg1, 0x7fffffff);
duke@435	849	__ xorl(c_rarg2, c_rarg2);
duke@435	850	__ shrl(c_rarg3, 0x1f);
duke@435	851	__ orl(c_rarg1, c_rarg3);
duke@435	852	__ cmpl(rax, c_rarg1);
duke@435	853	__ jcc(Assembler::negative, L); // NaN -> 0
duke@435	854	__ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
duke@435	855	__ mov64(c_rarg2, 0x8000000000000000);
duke@435	856	__ mov64(rax, 0x7fffffffffffffff);
duke@435	857	__ cmovq(Assembler::positive, c_rarg2, rax);
duke@435	858
duke@435	859	__ bind(L);
duke@435	860	__ movq(inout, c_rarg2);
duke@435	861
never@739	862	__ pop(c_rarg0);
never@739	863	__ pop(c_rarg1);
never@739	864	__ pop(c_rarg2);
never@739	865	__ pop(c_rarg3);
never@739	866	__ pop(rax);
duke@435	867
duke@435	868	__ ret(0);
duke@435	869
duke@435	870	return start;
duke@435	871	}
duke@435	872
duke@435	873	address generate_fp_mask(const char *stub_name, int64_t mask) {
duke@435	874	StubCodeMark mark(this, "StubRoutines", stub_name);
duke@435	875
duke@435	876	__ align(16);
duke@435	877	address start = __ pc();
duke@435	878
duke@435	879	__ emit_data64( mask, relocInfo::none );
duke@435	880	__ emit_data64( mask, relocInfo::none );
duke@435	881
duke@435	882	return start;
duke@435	883	}
duke@435	884
duke@435	885	// The following routine generates a subroutine to throw an
duke@435	886	// asynchronous UnknownError when an unsafe access gets a fault that
duke@435	887	// could not be reasonably prevented by the programmer. (Example:
duke@435	888	// SIGBUS/OBJERR.)
duke@435	889	address generate_handler_for_unsafe_access() {
duke@435	890	StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
duke@435	891	address start = __ pc();
duke@435	892
never@739	893	__ push(0); // hole for return address-to-be
never@739	894	__ pusha(); // push registers
duke@435	895	Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
duke@435	896
never@739	897	__ subptr(rsp, frame::arg_reg_save_area_bytes);
duke@435	898	BLOCK_COMMENT("call handle_unsafe_access");
duke@435	899	__ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
never@739	900	__ addptr(rsp, frame::arg_reg_save_area_bytes);
never@739	901
never@739	902	__ movptr(next_pc, rax); // stuff next address
never@739	903	__ popa();
duke@435	904	__ ret(0); // jump to next address
duke@435	905
duke@435	906	return start;
duke@435	907	}
duke@435	908
duke@435	909	// Non-destructive plausibility checks for oops
duke@435	910	//
duke@435	911	// Arguments:
duke@435	912	// all args on stack!
duke@435	913	//
duke@435	914	// Stack after saving c_rarg3:
duke@435	915	// [tos + 0]: saved c_rarg3
duke@435	916	// [tos + 1]: saved c_rarg2
kvn@559	917	// [tos + 2]: saved r12 (several TemplateTable methods use it)
kvn@559	918	// [tos + 3]: saved flags
kvn@559	919	// [tos + 4]: return address
kvn@559	920	// * [tos + 5]: error message (char*)
kvn@559	921	// * [tos + 6]: object to verify (oop)
kvn@559	922	// * [tos + 7]: saved rax - saved by caller and bashed
duke@435	923	// * = popped on exit
duke@435	924	address generate_verify_oop() {
duke@435	925	StubCodeMark mark(this, "StubRoutines", "verify_oop");
duke@435	926	address start = __ pc();
duke@435	927
duke@435	928	Label exit, error;
duke@435	929
never@739	930	__ pushf();
duke@435	931	__ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
duke@435	932
never@739	933	__ push(r12);
kvn@559	934
duke@435	935	// save c_rarg2 and c_rarg3
never@739	936	__ push(c_rarg2);
never@739	937	__ push(c_rarg3);
duke@435	938
kvn@559	939	enum {
kvn@559	940	// After previous pushes.
kvn@559	941	oop_to_verify = 6 * wordSize,
kvn@559	942	saved_rax = 7 * wordSize,
kvn@559	943
kvn@559	944	// Before the call to MacroAssembler::debug(), see below.
kvn@559	945	return_addr = 16 * wordSize,
kvn@559	946	error_msg = 17 * wordSize
kvn@559	947	};
kvn@559	948
duke@435	949	// get object
never@739	950	__ movptr(rax, Address(rsp, oop_to_verify));
duke@435	951
duke@435	952	// make sure object is 'reasonable'
never@739	953	__ testptr(rax, rax);
duke@435	954	__ jcc(Assembler::zero, exit); // if obj is NULL it is OK
duke@435	955	// Check if the oop is in the right area of memory
never@739	956	__ movptr(c_rarg2, rax);
duke@435	957	__ movptr(c_rarg3, (int64_t) Universe::verify_oop_mask());
never@739	958	__ andptr(c_rarg2, c_rarg3);
duke@435	959	__ movptr(c_rarg3, (int64_t) Universe::verify_oop_bits());
never@739	960	__ cmpptr(c_rarg2, c_rarg3);
duke@435	961	__ jcc(Assembler::notZero, error);
duke@435	962
kvn@559	963	// set r12 to heapbase for load_klass()
kvn@559	964	__ reinit_heapbase();
kvn@559	965
duke@435	966	// make sure klass is 'reasonable'
coleenp@548	967	__ load_klass(rax, rax); // get klass
never@739	968	__ testptr(rax, rax);
duke@435	969	__ jcc(Assembler::zero, error); // if klass is NULL it is broken
duke@435	970	// Check if the klass is in the right area of memory
never@739	971	__ mov(c_rarg2, rax);
duke@435	972	__ movptr(c_rarg3, (int64_t) Universe::verify_klass_mask());
never@739	973	__ andptr(c_rarg2, c_rarg3);
duke@435	974	__ movptr(c_rarg3, (int64_t) Universe::verify_klass_bits());
never@739	975	__ cmpptr(c_rarg2, c_rarg3);
duke@435	976	__ jcc(Assembler::notZero, error);
duke@435	977
duke@435	978	// make sure klass' klass is 'reasonable'
coleenp@548	979	__ load_klass(rax, rax);
never@739	980	__ testptr(rax, rax);
duke@435	981	__ jcc(Assembler::zero, error); // if klass' klass is NULL it is broken
duke@435	982	// Check if the klass' klass is in the right area of memory
duke@435	983	__ movptr(c_rarg3, (int64_t) Universe::verify_klass_mask());
never@739	984	__ andptr(rax, c_rarg3);
duke@435	985	__ movptr(c_rarg3, (int64_t) Universe::verify_klass_bits());
never@739	986	__ cmpptr(rax, c_rarg3);
duke@435	987	__ jcc(Assembler::notZero, error);
duke@435	988
duke@435	989	// return if everything seems ok
duke@435	990	__ bind(exit);
never@739	991	__ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
never@739	992	__ pop(c_rarg3); // restore c_rarg3
never@739	993	__ pop(c_rarg2); // restore c_rarg2
never@739	994	__ pop(r12); // restore r12
never@739	995	__ popf(); // restore flags
duke@435	996	__ ret(3 * wordSize); // pop caller saved stuff
duke@435	997
duke@435	998	// handle errors
duke@435	999	__ bind(error);
never@739	1000	__ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
never@739	1001	__ pop(c_rarg3); // get saved c_rarg3 back
never@739	1002	__ pop(c_rarg2); // get saved c_rarg2 back
never@739	1003	__ pop(r12); // get saved r12 back
never@739	1004	__ popf(); // get saved flags off stack --
duke@435	1005	// will be ignored
duke@435	1006
never@739	1007	__ pusha(); // push registers
duke@435	1008	// (rip is already
duke@435	1009	// already pushed)
kvn@559	1010	// debug(char* msg, int64_t pc, int64_t regs[])
duke@435	1011	// We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
duke@435	1012	// pushed all the registers, so now the stack looks like:
duke@435	1013	// [tos + 0] 16 saved registers
duke@435	1014	// [tos + 16] return address
kvn@559	1015	// * [tos + 17] error message (char*)
kvn@559	1016	// * [tos + 18] object to verify (oop)
kvn@559	1017	// * [tos + 19] saved rax - saved by caller and bashed
kvn@559	1018	// * = popped on exit
kvn@559	1019
never@739	1020	__ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message
never@739	1021	__ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address
never@739	1022	__ movq(c_rarg2, rsp); // pass address of regs on stack
never@739	1023	__ mov(r12, rsp); // remember rsp
never@739	1024	__ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
never@739	1025	__ andptr(rsp, -16); // align stack as required by ABI
duke@435	1026	BLOCK_COMMENT("call MacroAssembler::debug");
never@739	1027	__ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
never@739	1028	__ mov(rsp, r12); // restore rsp
never@739	1029	__ popa(); // pop registers (includes r12)
never@739	1030	__ ret(3 * wordSize); // pop caller saved stuff
duke@435	1031
duke@435	1032	return start;
duke@435	1033	}
duke@435	1034
duke@435	1035	static address disjoint_byte_copy_entry;
duke@435	1036	static address disjoint_short_copy_entry;
duke@435	1037	static address disjoint_int_copy_entry;
duke@435	1038	static address disjoint_long_copy_entry;
duke@435	1039	static address disjoint_oop_copy_entry;
duke@435	1040
duke@435	1041	static address byte_copy_entry;
duke@435	1042	static address short_copy_entry;
duke@435	1043	static address int_copy_entry;
duke@435	1044	static address long_copy_entry;
duke@435	1045	static address oop_copy_entry;
duke@435	1046
duke@435	1047	static address checkcast_copy_entry;
duke@435	1048
duke@435	1049	//
duke@435	1050	// Verify that a register contains clean 32-bits positive value
duke@435	1051	// (high 32-bits are 0) so it could be used in 64-bits shifts.
duke@435	1052	//
duke@435	1053	// Input:
duke@435	1054	// Rint - 32-bits value
duke@435	1055	// Rtmp - scratch
duke@435	1056	//
duke@435	1057	void assert_clean_int(Register Rint, Register Rtmp) {
duke@435	1058	#ifdef ASSERT
duke@435	1059	Label L;
duke@435	1060	assert_different_registers(Rtmp, Rint);
duke@435	1061	__ movslq(Rtmp, Rint);
duke@435	1062	__ cmpq(Rtmp, Rint);
kvn@559	1063	__ jcc(Assembler::equal, L);
duke@435	1064	__ stop("high 32-bits of int value are not 0");
duke@435	1065	__ bind(L);
duke@435	1066	#endif
duke@435	1067	}
duke@435	1068
duke@435	1069	// Generate overlap test for array copy stubs
duke@435	1070	//
duke@435	1071	// Input:
duke@435	1072	// c_rarg0 - from
duke@435	1073	// c_rarg1 - to
duke@435	1074	// c_rarg2 - element count
duke@435	1075	//
duke@435	1076	// Output:
duke@435	1077	// rax - &from[element count - 1]
duke@435	1078	//
duke@435	1079	void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) {
duke@435	1080	assert(no_overlap_target != NULL, "must be generated");
duke@435	1081	array_overlap_test(no_overlap_target, NULL, sf);
duke@435	1082	}
duke@435	1083	void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) {
duke@435	1084	array_overlap_test(NULL, &L_no_overlap, sf);
duke@435	1085	}
duke@435	1086	void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) {
duke@435	1087	const Register from = c_rarg0;
duke@435	1088	const Register to = c_rarg1;
duke@435	1089	const Register count = c_rarg2;
duke@435	1090	const Register end_from = rax;
duke@435	1091
never@739	1092	__ cmpptr(to, from);
never@739	1093	__ lea(end_from, Address(from, count, sf, 0));
duke@435	1094	if (NOLp == NULL) {
duke@435	1095	ExternalAddress no_overlap(no_overlap_target);
duke@435	1096	__ jump_cc(Assembler::belowEqual, no_overlap);
never@739	1097	__ cmpptr(to, end_from);
duke@435	1098	__ jump_cc(Assembler::aboveEqual, no_overlap);
duke@435	1099	} else {
duke@435	1100	__ jcc(Assembler::belowEqual, (*NOLp));
never@739	1101	__ cmpptr(to, end_from);
duke@435	1102	__ jcc(Assembler::aboveEqual, (*NOLp));
duke@435	1103	}
duke@435	1104	}
duke@435	1105
duke@435	1106	// Shuffle first three arg regs on Windows into Linux/Solaris locations.
duke@435	1107	//
duke@435	1108	// Outputs:
duke@435	1109	// rdi - rcx
duke@435	1110	// rsi - rdx
duke@435	1111	// rdx - r8
duke@435	1112	// rcx - r9
duke@435	1113	//
duke@435	1114	// Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
duke@435	1115	// are non-volatile. r9 and r10 should not be used by the caller.
duke@435	1116	//
duke@435	1117	void setup_arg_regs(int nargs = 3) {
duke@435	1118	const Register saved_rdi = r9;
duke@435	1119	const Register saved_rsi = r10;
duke@435	1120	assert(nargs == 3 || nargs == 4, "else fix");
duke@435	1121	#ifdef _WIN64
duke@435	1122	assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
duke@435	1123	"unexpected argument registers");
duke@435	1124	if (nargs >= 4)
never@739	1125	__ mov(rax, r9); // r9 is also saved_rdi
never@739	1126	__ movptr(saved_rdi, rdi);
never@739	1127	__ movptr(saved_rsi, rsi);
never@739	1128	__ mov(rdi, rcx); // c_rarg0
never@739	1129	__ mov(rsi, rdx); // c_rarg1
never@739	1130	__ mov(rdx, r8); // c_rarg2
duke@435	1131	if (nargs >= 4)
never@739	1132	__ mov(rcx, rax); // c_rarg3 (via rax)
duke@435	1133	#else
duke@435	1134	assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
duke@435	1135	"unexpected argument registers");
duke@435	1136	#endif
duke@435	1137	}
duke@435	1138
duke@435	1139	void restore_arg_regs() {
duke@435	1140	const Register saved_rdi = r9;
duke@435	1141	const Register saved_rsi = r10;
duke@435	1142	#ifdef _WIN64
never@739	1143	__ movptr(rdi, saved_rdi);
never@739	1144	__ movptr(rsi, saved_rsi);
duke@435	1145	#endif
duke@435	1146	}
duke@435	1147
duke@435	1148	// Generate code for an array write pre barrier
duke@435	1149	//
duke@435	1150	// addr - starting address
duke@435	1151	// count - element count
duke@435	1152	//
duke@435	1153	// Destroy no registers!
duke@435	1154	//
duke@435	1155	void gen_write_ref_array_pre_barrier(Register addr, Register count) {
duke@435	1156	#if 0 // G1 - only
duke@435	1157	assert_different_registers(addr, c_rarg1);
duke@435	1158	assert_different_registers(count, c_rarg0);
duke@435	1159	BarrierSet* bs = Universe::heap()->barrier_set();
duke@435	1160	switch (bs->kind()) {
duke@435	1161	case BarrierSet::G1SATBCT:
duke@435	1162	case BarrierSet::G1SATBCTLogging:
duke@435	1163	{
never@739	1164	__ pusha(); // push registers
never@739	1165	__ movptr(c_rarg0, addr);
never@739	1166	__ movptr(c_rarg1, count);
duke@435	1167	__ call(RuntimeAddress(BarrierSet::static_write_ref_array_pre));
never@739	1168	__ popa();
duke@435	1169	}
duke@435	1170	break;
duke@435	1171	case BarrierSet::CardTableModRef:
duke@435	1172	case BarrierSet::CardTableExtension:
duke@435	1173	case BarrierSet::ModRef:
duke@435	1174	break;
duke@435	1175	default :
duke@435	1176	ShouldNotReachHere();
duke@435	1177
duke@435	1178	}
duke@435	1179	#endif // 0 G1 - only
duke@435	1180	}
duke@435	1181
duke@435	1182	//
duke@435	1183	// Generate code for an array write post barrier
duke@435	1184	//
duke@435	1185	// Input:
duke@435	1186	// start - register containing starting address of destination array
duke@435	1187	// end - register containing ending address of destination array
duke@435	1188	// scratch - scratch register
duke@435	1189	//
duke@435	1190	// The input registers are overwritten.
duke@435	1191	// The ending address is inclusive.
duke@435	1192	void gen_write_ref_array_post_barrier(Register start, Register end, Register scratch) {
duke@435	1193	assert_different_registers(start, end, scratch);
duke@435	1194	BarrierSet* bs = Universe::heap()->barrier_set();
duke@435	1195	switch (bs->kind()) {
duke@435	1196	#if 0 // G1 - only
duke@435	1197	case BarrierSet::G1SATBCT:
duke@435	1198	case BarrierSet::G1SATBCTLogging:
duke@435	1199
duke@435	1200	{
never@739	1201	__ pusha(); // push registers (overkill)
duke@435	1202	// must compute element count unless barrier set interface is changed (other platforms supply count)
duke@435	1203	assert_different_registers(start, end, scratch);
never@739	1204	__ lea(scratch, Address(end, wordSize));
never@739	1205	__ subptr(scratch, start);
never@739	1206	__ shrptr(scratch, LogBytesPerWord);
never@739	1207	__ mov(c_rarg0, start);
never@739	1208	__ mov(c_rarg1, scratch);
duke@435	1209	__ call(RuntimeAddress(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post));
never@739	1210	__ popa();
duke@435	1211	}
duke@435	1212	break;
duke@435	1213	#endif // 0 G1 - only
duke@435	1214	case BarrierSet::CardTableModRef:
duke@435	1215	case BarrierSet::CardTableExtension:
duke@435	1216	{
duke@435	1217	CardTableModRefBS* ct = (CardTableModRefBS*)bs;
duke@435	1218	assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
duke@435	1219
duke@435	1220	Label L_loop;
duke@435	1221
never@739	1222	__ shrptr(start, CardTableModRefBS::card_shift);
never@739	1223	__ shrptr(end, CardTableModRefBS::card_shift);
never@739	1224	__ subptr(end, start); // number of bytes to copy
duke@435	1225
never@684	1226	intptr_t disp = (intptr_t) ct->byte_map_base;
never@684	1227	if (__ is_simm32(disp)) {
never@684	1228	Address cardtable(noreg, noreg, Address::no_scale, disp);
never@684	1229	__ lea(scratch, cardtable);
never@684	1230	} else {
never@684	1231	ExternalAddress cardtable((address)disp);
never@684	1232	__ lea(scratch, cardtable);
never@684	1233	}
never@684	1234
duke@435	1235	const Register count = end; // 'end' register contains bytes count now
never@739	1236	__ addptr(start, scratch);
duke@435	1237	__ BIND(L_loop);
duke@435	1238	__ movb(Address(start, count, Address::times_1), 0);
never@739	1239	__ decrement(count);
duke@435	1240	__ jcc(Assembler::greaterEqual, L_loop);
duke@435	1241	}
duke@435	1242	}
duke@435	1243	}
duke@435	1244
duke@435	1245	// Copy big chunks forward
duke@435	1246	//
duke@435	1247	// Inputs:
duke@435	1248	// end_from - source arrays end address
duke@435	1249	// end_to - destination array end address
duke@435	1250	// qword_count - 64-bits element count, negative
duke@435	1251	// to - scratch
duke@435	1252	// L_copy_32_bytes - entry label
duke@435	1253	// L_copy_8_bytes - exit label
duke@435	1254	//
duke@435	1255	void copy_32_bytes_forward(Register end_from, Register end_to,
duke@435	1256	Register qword_count, Register to,
duke@435	1257	Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
duke@435	1258	DEBUG_ONLY(__ stop("enter at entry label, not here"));
duke@435	1259	Label L_loop;
duke@435	1260	__ align(16);
duke@435	1261	__ BIND(L_loop);
duke@435	1262	__ movq(to, Address(end_from, qword_count, Address::times_8, -24));
duke@435	1263	__ movq(Address(end_to, qword_count, Address::times_8, -24), to);
duke@435	1264	__ movq(to, Address(end_from, qword_count, Address::times_8, -16));
duke@435	1265	__ movq(Address(end_to, qword_count, Address::times_8, -16), to);
duke@435	1266	__ movq(to, Address(end_from, qword_count, Address::times_8, - 8));
duke@435	1267	__ movq(Address(end_to, qword_count, Address::times_8, - 8), to);
duke@435	1268	__ movq(to, Address(end_from, qword_count, Address::times_8, - 0));
duke@435	1269	__ movq(Address(end_to, qword_count, Address::times_8, - 0), to);
duke@435	1270	__ BIND(L_copy_32_bytes);
never@739	1271	__ addptr(qword_count, 4);
duke@435	1272	__ jcc(Assembler::lessEqual, L_loop);
never@739	1273	__ subptr(qword_count, 4);
duke@435	1274	__ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords
duke@435	1275	}
duke@435	1276
duke@435	1277
duke@435	1278	// Copy big chunks backward
duke@435	1279	//
duke@435	1280	// Inputs:
duke@435	1281	// from - source arrays address
duke@435	1282	// dest - destination array address
duke@435	1283	// qword_count - 64-bits element count
duke@435	1284	// to - scratch
duke@435	1285	// L_copy_32_bytes - entry label
duke@435	1286	// L_copy_8_bytes - exit label
duke@435	1287	//
duke@435	1288	void copy_32_bytes_backward(Register from, Register dest,
duke@435	1289	Register qword_count, Register to,
duke@435	1290	Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
duke@435	1291	DEBUG_ONLY(__ stop("enter at entry label, not here"));
duke@435	1292	Label L_loop;
duke@435	1293	__ align(16);
duke@435	1294	__ BIND(L_loop);
duke@435	1295	__ movq(to, Address(from, qword_count, Address::times_8, 24));
duke@435	1296	__ movq(Address(dest, qword_count, Address::times_8, 24), to);
duke@435	1297	__ movq(to, Address(from, qword_count, Address::times_8, 16));
duke@435	1298	__ movq(Address(dest, qword_count, Address::times_8, 16), to);
duke@435	1299	__ movq(to, Address(from, qword_count, Address::times_8, 8));
duke@435	1300	__ movq(Address(dest, qword_count, Address::times_8, 8), to);
duke@435	1301	__ movq(to, Address(from, qword_count, Address::times_8, 0));
duke@435	1302	__ movq(Address(dest, qword_count, Address::times_8, 0), to);
duke@435	1303	__ BIND(L_copy_32_bytes);
never@739	1304	__ subptr(qword_count, 4);
duke@435	1305	__ jcc(Assembler::greaterEqual, L_loop);
never@739	1306	__ addptr(qword_count, 4);
duke@435	1307	__ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords
duke@435	1308	}
duke@435	1309
duke@435	1310
duke@435	1311	// Arguments:
duke@435	1312	// aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
duke@435	1313	// ignored
duke@435	1314	// name - stub name string
duke@435	1315	//
duke@435	1316	// Inputs:
duke@435	1317	// c_rarg0 - source array address
duke@435	1318	// c_rarg1 - destination array address
duke@435	1319	// c_rarg2 - element count, treated as ssize_t, can be zero
duke@435	1320	//
duke@435	1321	// If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
duke@435	1322	// we let the hardware handle it. The one to eight bytes within words,
duke@435	1323	// dwords or qwords that span cache line boundaries will still be loaded
duke@435	1324	// and stored atomically.
duke@435	1325	//
duke@435	1326	// Side Effects:
duke@435	1327	// disjoint_byte_copy_entry is set to the no-overlap entry point
duke@435	1328	// used by generate_conjoint_byte_copy().
duke@435	1329	//
duke@435	1330	address generate_disjoint_byte_copy(bool aligned, const char *name) {
duke@435	1331	__ align(CodeEntryAlignment);
duke@435	1332	StubCodeMark mark(this, "StubRoutines", name);
duke@435	1333	address start = __ pc();
duke@435	1334
duke@435	1335	Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
duke@435	1336	Label L_copy_byte, L_exit;
duke@435	1337	const Register from = rdi; // source array address
duke@435	1338	const Register to = rsi; // destination array address
duke@435	1339	const Register count = rdx; // elements count
duke@435	1340	const Register byte_count = rcx;
duke@435	1341	const Register qword_count = count;
duke@435	1342	const Register end_from = from; // source array end address
duke@435	1343	const Register end_to = to; // destination array end address
duke@435	1344	// End pointers are inclusive, and if count is not zero they point
duke@435	1345	// to the last unit copied: end_to[0] := end_from[0]
duke@435	1346
duke@435	1347	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	1348	assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
duke@435	1349
duke@435	1350	disjoint_byte_copy_entry = __ pc();
duke@435	1351	BLOCK_COMMENT("Entry:");
duke@435	1352	// caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
duke@435	1353
duke@435	1354	setup_arg_regs(); // from => rdi, to => rsi, count => rdx
duke@435	1355	// r9 and r10 may be used to save non-volatile registers
duke@435	1356
duke@435	1357	// 'from', 'to' and 'count' are now valid
never@739	1358	__ movptr(byte_count, count);
never@739	1359	__ shrptr(count, 3); // count => qword_count
duke@435	1360
duke@435	1361	// Copy from low to high addresses. Use 'to' as scratch.
never@739	1362	__ lea(end_from, Address(from, qword_count, Address::times_8, -8));
never@739	1363	__ lea(end_to, Address(to, qword_count, Address::times_8, -8));
never@739	1364	__ negptr(qword_count); // make the count negative
duke@435	1365	__ jmp(L_copy_32_bytes);
duke@435	1366
duke@435	1367	// Copy trailing qwords
duke@435	1368	__ BIND(L_copy_8_bytes);
duke@435	1369	__ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
duke@435	1370	__ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
never@739	1371	__ increment(qword_count);
duke@435	1372	__ jcc(Assembler::notZero, L_copy_8_bytes);
duke@435	1373
duke@435	1374	// Check for and copy trailing dword
duke@435	1375	__ BIND(L_copy_4_bytes);
never@739	1376	__ testl(byte_count, 4);
duke@435	1377	__ jccb(Assembler::zero, L_copy_2_bytes);
duke@435	1378	__ movl(rax, Address(end_from, 8));
duke@435	1379	__ movl(Address(end_to, 8), rax);
duke@435	1380
never@739	1381	__ addptr(end_from, 4);
never@739	1382	__ addptr(end_to, 4);
duke@435	1383
duke@435	1384	// Check for and copy trailing word
duke@435	1385	__ BIND(L_copy_2_bytes);
never@739	1386	__ testl(byte_count, 2);
duke@435	1387	__ jccb(Assembler::zero, L_copy_byte);
duke@435	1388	__ movw(rax, Address(end_from, 8));
duke@435	1389	__ movw(Address(end_to, 8), rax);
duke@435	1390
never@739	1391	__ addptr(end_from, 2);
never@739	1392	__ addptr(end_to, 2);
duke@435	1393
duke@435	1394	// Check for and copy trailing byte
duke@435	1395	__ BIND(L_copy_byte);
never@739	1396	__ testl(byte_count, 1);
duke@435	1397	__ jccb(Assembler::zero, L_exit);
duke@435	1398	__ movb(rax, Address(end_from, 8));
duke@435	1399	__ movb(Address(end_to, 8), rax);
duke@435	1400
duke@435	1401	__ BIND(L_exit);
duke@435	1402	inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
duke@435	1403	restore_arg_regs();
never@739	1404	__ xorptr(rax, rax); // return 0
duke@435	1405	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	1406	__ ret(0);
duke@435	1407
duke@435	1408	// Copy in 32-bytes chunks
duke@435	1409	copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
duke@435	1410	__ jmp(L_copy_4_bytes);
duke@435	1411
duke@435	1412	return start;
duke@435	1413	}
duke@435	1414
duke@435	1415	// Arguments:
duke@435	1416	// aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
duke@435	1417	// ignored
duke@435	1418	// name - stub name string
duke@435	1419	//
duke@435	1420	// Inputs:
duke@435	1421	// c_rarg0 - source array address
duke@435	1422	// c_rarg1 - destination array address
duke@435	1423	// c_rarg2 - element count, treated as ssize_t, can be zero
duke@435	1424	//
duke@435	1425	// If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
duke@435	1426	// we let the hardware handle it. The one to eight bytes within words,
duke@435	1427	// dwords or qwords that span cache line boundaries will still be loaded
duke@435	1428	// and stored atomically.
duke@435	1429	//
duke@435	1430	address generate_conjoint_byte_copy(bool aligned, const char *name) {
duke@435	1431	__ align(CodeEntryAlignment);
duke@435	1432	StubCodeMark mark(this, "StubRoutines", name);
duke@435	1433	address start = __ pc();
duke@435	1434
duke@435	1435	Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
duke@435	1436	const Register from = rdi; // source array address
duke@435	1437	const Register to = rsi; // destination array address
duke@435	1438	const Register count = rdx; // elements count
duke@435	1439	const Register byte_count = rcx;
duke@435	1440	const Register qword_count = count;
duke@435	1441
duke@435	1442	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	1443	assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
duke@435	1444
duke@435	1445	byte_copy_entry = __ pc();
duke@435	1446	BLOCK_COMMENT("Entry:");
duke@435	1447	// caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
duke@435	1448
duke@435	1449	array_overlap_test(disjoint_byte_copy_entry, Address::times_1);
duke@435	1450	setup_arg_regs(); // from => rdi, to => rsi, count => rdx
duke@435	1451	// r9 and r10 may be used to save non-volatile registers
duke@435	1452
duke@435	1453	// 'from', 'to' and 'count' are now valid
never@739	1454	__ movptr(byte_count, count);
never@739	1455	__ shrptr(count, 3); // count => qword_count
duke@435	1456
duke@435	1457	// Copy from high to low addresses.
duke@435	1458
duke@435	1459	// Check for and copy trailing byte
never@739	1460	__ testl(byte_count, 1);
duke@435	1461	__ jcc(Assembler::zero, L_copy_2_bytes);
duke@435	1462	__ movb(rax, Address(from, byte_count, Address::times_1, -1));
duke@435	1463	__ movb(Address(to, byte_count, Address::times_1, -1), rax);
never@739	1464	__ decrement(byte_count); // Adjust for possible trailing word
duke@435	1465
duke@435	1466	// Check for and copy trailing word
duke@435	1467	__ BIND(L_copy_2_bytes);
never@739	1468	__ testl(byte_count, 2);
duke@435	1469	__ jcc(Assembler::zero, L_copy_4_bytes);
duke@435	1470	__ movw(rax, Address(from, byte_count, Address::times_1, -2));
duke@435	1471	__ movw(Address(to, byte_count, Address::times_1, -2), rax);
duke@435	1472
duke@435	1473	// Check for and copy trailing dword
duke@435	1474	__ BIND(L_copy_4_bytes);
never@739	1475	__ testl(byte_count, 4);
duke@435	1476	__ jcc(Assembler::zero, L_copy_32_bytes);
duke@435	1477	__ movl(rax, Address(from, qword_count, Address::times_8));
duke@435	1478	__ movl(Address(to, qword_count, Address::times_8), rax);
duke@435	1479	__ jmp(L_copy_32_bytes);
duke@435	1480
duke@435	1481	// Copy trailing qwords
duke@435	1482	__ BIND(L_copy_8_bytes);
duke@435	1483	__ movq(rax, Address(from, qword_count, Address::times_8, -8));
duke@435	1484	__ movq(Address(to, qword_count, Address::times_8, -8), rax);
never@739	1485	__ decrement(qword_count);
duke@435	1486	__ jcc(Assembler::notZero, L_copy_8_bytes);
duke@435	1487
duke@435	1488	inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
duke@435	1489	restore_arg_regs();
never@739	1490	__ xorptr(rax, rax); // return 0
duke@435	1491	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	1492	__ ret(0);
duke@435	1493
duke@435	1494	// Copy in 32-bytes chunks
duke@435	1495	copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
duke@435	1496
duke@435	1497	inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr);
duke@435	1498	restore_arg_regs();
never@739	1499	__ xorptr(rax, rax); // return 0
duke@435	1500	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	1501	__ ret(0);
duke@435	1502
duke@435	1503	return start;
duke@435	1504	}
duke@435	1505
duke@435	1506	// Arguments:
duke@435	1507	// aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
duke@435	1508	// ignored
duke@435	1509	// name - stub name string
duke@435	1510	//
duke@435	1511	// Inputs:
duke@435	1512	// c_rarg0 - source array address
duke@435	1513	// c_rarg1 - destination array address
duke@435	1514	// c_rarg2 - element count, treated as ssize_t, can be zero
duke@435	1515	//
duke@435	1516	// If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
duke@435	1517	// let the hardware handle it. The two or four words within dwords
duke@435	1518	// or qwords that span cache line boundaries will still be loaded
duke@435	1519	// and stored atomically.
duke@435	1520	//
duke@435	1521	// Side Effects:
duke@435	1522	// disjoint_short_copy_entry is set to the no-overlap entry point
duke@435	1523	// used by generate_conjoint_short_copy().
duke@435	1524	//
duke@435	1525	address generate_disjoint_short_copy(bool aligned, const char *name) {
duke@435	1526	__ align(CodeEntryAlignment);
duke@435	1527	StubCodeMark mark(this, "StubRoutines", name);
duke@435	1528	address start = __ pc();
duke@435	1529
duke@435	1530	Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit;
duke@435	1531	const Register from = rdi; // source array address
duke@435	1532	const Register to = rsi; // destination array address
duke@435	1533	const Register count = rdx; // elements count
duke@435	1534	const Register word_count = rcx;
duke@435	1535	const Register qword_count = count;
duke@435	1536	const Register end_from = from; // source array end address
duke@435	1537	const Register end_to = to; // destination array end address
duke@435	1538	// End pointers are inclusive, and if count is not zero they point
duke@435	1539	// to the last unit copied: end_to[0] := end_from[0]
duke@435	1540
duke@435	1541	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	1542	assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
duke@435	1543
duke@435	1544	disjoint_short_copy_entry = __ pc();
duke@435	1545	BLOCK_COMMENT("Entry:");
duke@435	1546	// caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
duke@435	1547
duke@435	1548	setup_arg_regs(); // from => rdi, to => rsi, count => rdx
duke@435	1549	// r9 and r10 may be used to save non-volatile registers
duke@435	1550
duke@435	1551	// 'from', 'to' and 'count' are now valid
never@739	1552	__ movptr(word_count, count);
never@739	1553	__ shrptr(count, 2); // count => qword_count
duke@435	1554
duke@435	1555	// Copy from low to high addresses. Use 'to' as scratch.
never@739	1556	__ lea(end_from, Address(from, qword_count, Address::times_8, -8));
never@739	1557	__ lea(end_to, Address(to, qword_count, Address::times_8, -8));
never@739	1558	__ negptr(qword_count);
duke@435	1559	__ jmp(L_copy_32_bytes);
duke@435	1560
duke@435	1561	// Copy trailing qwords
duke@435	1562	__ BIND(L_copy_8_bytes);
duke@435	1563	__ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
duke@435	1564	__ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
never@739	1565	__ increment(qword_count);
duke@435	1566	__ jcc(Assembler::notZero, L_copy_8_bytes);
duke@435	1567
duke@435	1568	// Original 'dest' is trashed, so we can't use it as a
duke@435	1569	// base register for a possible trailing word copy
duke@435	1570
duke@435	1571	// Check for and copy trailing dword
duke@435	1572	__ BIND(L_copy_4_bytes);
never@739	1573	__ testl(word_count, 2);
duke@435	1574	__ jccb(Assembler::zero, L_copy_2_bytes);
duke@435	1575	__ movl(rax, Address(end_from, 8));
duke@435	1576	__ movl(Address(end_to, 8), rax);
duke@435	1577
never@739	1578	__ addptr(end_from, 4);
never@739	1579	__ addptr(end_to, 4);
duke@435	1580
duke@435	1581	// Check for and copy trailing word
duke@435	1582	__ BIND(L_copy_2_bytes);
never@739	1583	__ testl(word_count, 1);
duke@435	1584	__ jccb(Assembler::zero, L_exit);
duke@435	1585	__ movw(rax, Address(end_from, 8));
duke@435	1586	__ movw(Address(end_to, 8), rax);
duke@435	1587
duke@435	1588	__ BIND(L_exit);
duke@435	1589	inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
duke@435	1590	restore_arg_regs();
never@739	1591	__ xorptr(rax, rax); // return 0
duke@435	1592	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	1593	__ ret(0);
duke@435	1594
duke@435	1595	// Copy in 32-bytes chunks
duke@435	1596	copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
duke@435	1597	__ jmp(L_copy_4_bytes);
duke@435	1598
duke@435	1599	return start;
duke@435	1600	}
duke@435	1601
duke@435	1602	// Arguments:
duke@435	1603	// aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
duke@435	1604	// ignored
duke@435	1605	// name - stub name string
duke@435	1606	//
duke@435	1607	// Inputs:
duke@435	1608	// c_rarg0 - source array address
duke@435	1609	// c_rarg1 - destination array address
duke@435	1610	// c_rarg2 - element count, treated as ssize_t, can be zero
duke@435	1611	//
duke@435	1612	// If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
duke@435	1613	// let the hardware handle it. The two or four words within dwords
duke@435	1614	// or qwords that span cache line boundaries will still be loaded
duke@435	1615	// and stored atomically.
duke@435	1616	//
duke@435	1617	address generate_conjoint_short_copy(bool aligned, const char *name) {
duke@435	1618	__ align(CodeEntryAlignment);
duke@435	1619	StubCodeMark mark(this, "StubRoutines", name);
duke@435	1620	address start = __ pc();
duke@435	1621
duke@435	1622	Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes;
duke@435	1623	const Register from = rdi; // source array address
duke@435	1624	const Register to = rsi; // destination array address
duke@435	1625	const Register count = rdx; // elements count
duke@435	1626	const Register word_count = rcx;
duke@435	1627	const Register qword_count = count;
duke@435	1628
duke@435	1629	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	1630	assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
duke@435	1631
duke@435	1632	short_copy_entry = __ pc();
duke@435	1633	BLOCK_COMMENT("Entry:");
duke@435	1634	// caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
duke@435	1635
duke@435	1636	array_overlap_test(disjoint_short_copy_entry, Address::times_2);
duke@435	1637	setup_arg_regs(); // from => rdi, to => rsi, count => rdx
duke@435	1638	// r9 and r10 may be used to save non-volatile registers
duke@435	1639
duke@435	1640	// 'from', 'to' and 'count' are now valid
never@739	1641	__ movptr(word_count, count);
never@739	1642	__ shrptr(count, 2); // count => qword_count
duke@435	1643
duke@435	1644	// Copy from high to low addresses. Use 'to' as scratch.
duke@435	1645
duke@435	1646	// Check for and copy trailing word
never@739	1647	__ testl(word_count, 1);
duke@435	1648	__ jccb(Assembler::zero, L_copy_4_bytes);
duke@435	1649	__ movw(rax, Address(from, word_count, Address::times_2, -2));
duke@435	1650	__ movw(Address(to, word_count, Address::times_2, -2), rax);
duke@435	1651
duke@435	1652	// Check for and copy trailing dword
duke@435	1653	__ BIND(L_copy_4_bytes);
never@739	1654	__ testl(word_count, 2);
duke@435	1655	__ jcc(Assembler::zero, L_copy_32_bytes);
duke@435	1656	__ movl(rax, Address(from, qword_count, Address::times_8));
duke@435	1657	__ movl(Address(to, qword_count, Address::times_8), rax);
duke@435	1658	__ jmp(L_copy_32_bytes);
duke@435	1659
duke@435	1660	// Copy trailing qwords
duke@435	1661	__ BIND(L_copy_8_bytes);
duke@435	1662	__ movq(rax, Address(from, qword_count, Address::times_8, -8));
duke@435	1663	__ movq(Address(to, qword_count, Address::times_8, -8), rax);
never@739	1664	__ decrement(qword_count);
duke@435	1665	__ jcc(Assembler::notZero, L_copy_8_bytes);
duke@435	1666
duke@435	1667	inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
duke@435	1668	restore_arg_regs();
never@739	1669	__ xorptr(rax, rax); // return 0
duke@435	1670	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	1671	__ ret(0);
duke@435	1672
duke@435	1673	// Copy in 32-bytes chunks
duke@435	1674	copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
duke@435	1675
duke@435	1676	inc_counter_np(SharedRuntime::_jshort_array_copy_ctr);
duke@435	1677	restore_arg_regs();
never@739	1678	__ xorptr(rax, rax); // return 0
duke@435	1679	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	1680	__ ret(0);
duke@435	1681
duke@435	1682	return start;
duke@435	1683	}
duke@435	1684
duke@435	1685	// Arguments:
duke@435	1686	// aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
duke@435	1687	// ignored
coleenp@548	1688	// is_oop - true => oop array, so generate store check code
duke@435	1689	// name - stub name string
duke@435	1690	//
duke@435	1691	// Inputs:
duke@435	1692	// c_rarg0 - source array address
duke@435	1693	// c_rarg1 - destination array address
duke@435	1694	// c_rarg2 - element count, treated as ssize_t, can be zero
duke@435	1695	//
duke@435	1696	// If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
duke@435	1697	// the hardware handle it. The two dwords within qwords that span
duke@435	1698	// cache line boundaries will still be loaded and stored atomicly.
duke@435	1699	//
duke@435	1700	// Side Effects:
duke@435	1701	// disjoint_int_copy_entry is set to the no-overlap entry point
coleenp@548	1702	// used by generate_conjoint_int_oop_copy().
duke@435	1703	//
coleenp@548	1704	address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, const char *name) {
duke@435	1705	__ align(CodeEntryAlignment);
duke@435	1706	StubCodeMark mark(this, "StubRoutines", name);
duke@435	1707	address start = __ pc();
duke@435	1708
duke@435	1709	Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit;
duke@435	1710	const Register from = rdi; // source array address
duke@435	1711	const Register to = rsi; // destination array address
duke@435	1712	const Register count = rdx; // elements count
duke@435	1713	const Register dword_count = rcx;
duke@435	1714	const Register qword_count = count;
duke@435	1715	const Register end_from = from; // source array end address
duke@435	1716	const Register end_to = to; // destination array end address
coleenp@548	1717	const Register saved_to = r11; // saved destination array address
duke@435	1718	// End pointers are inclusive, and if count is not zero they point
duke@435	1719	// to the last unit copied: end_to[0] := end_from[0]
duke@435	1720
duke@435	1721	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	1722	assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
duke@435	1723
coleenp@548	1724	(is_oop ? disjoint_oop_copy_entry : disjoint_int_copy_entry) = __ pc();
coleenp@548	1725
coleenp@548	1726	if (is_oop) {
coleenp@548	1727	// no registers are destroyed by this call
coleenp@548	1728	gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2);
coleenp@548	1729	}
coleenp@548	1730
duke@435	1731	BLOCK_COMMENT("Entry:");
duke@435	1732	// caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
duke@435	1733
duke@435	1734	setup_arg_regs(); // from => rdi, to => rsi, count => rdx
duke@435	1735	// r9 and r10 may be used to save non-volatile registers
duke@435	1736
coleenp@548	1737	if (is_oop) {
coleenp@548	1738	__ movq(saved_to, to);
coleenp@548	1739	}
coleenp@548	1740
duke@435	1741	// 'from', 'to' and 'count' are now valid
never@739	1742	__ movptr(dword_count, count);
never@739	1743	__ shrptr(count, 1); // count => qword_count
duke@435	1744
duke@435	1745	// Copy from low to high addresses. Use 'to' as scratch.
never@739	1746	__ lea(end_from, Address(from, qword_count, Address::times_8, -8));
never@739	1747	__ lea(end_to, Address(to, qword_count, Address::times_8, -8));
never@739	1748	__ negptr(qword_count);
duke@435	1749	__ jmp(L_copy_32_bytes);
duke@435	1750
duke@435	1751	// Copy trailing qwords
duke@435	1752	__ BIND(L_copy_8_bytes);
duke@435	1753	__ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
duke@435	1754	__ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
never@739	1755	__ increment(qword_count);
duke@435	1756	__ jcc(Assembler::notZero, L_copy_8_bytes);
duke@435	1757
duke@435	1758	// Check for and copy trailing dword
duke@435	1759	__ BIND(L_copy_4_bytes);
never@739	1760	__ testl(dword_count, 1); // Only byte test since the value is 0 or 1
duke@435	1761	__ jccb(Assembler::zero, L_exit);
duke@435	1762	__ movl(rax, Address(end_from, 8));
duke@435	1763	__ movl(Address(end_to, 8), rax);
duke@435	1764
duke@435	1765	__ BIND(L_exit);
coleenp@548	1766	if (is_oop) {
coleenp@548	1767	__ leaq(end_to, Address(saved_to, dword_count, Address::times_4, -4));
coleenp@548	1768	gen_write_ref_array_post_barrier(saved_to, end_to, rax);
coleenp@548	1769	}
duke@435	1770	inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
duke@435	1771	restore_arg_regs();
never@739	1772	__ xorptr(rax, rax); // return 0
duke@435	1773	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	1774	__ ret(0);
duke@435	1775
duke@435	1776	// Copy 32-bytes chunks
duke@435	1777	copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
duke@435	1778	__ jmp(L_copy_4_bytes);
duke@435	1779
duke@435	1780	return start;
duke@435	1781	}
duke@435	1782
duke@435	1783	// Arguments:
duke@435	1784	// aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
duke@435	1785	// ignored
coleenp@548	1786	// is_oop - true => oop array, so generate store check code
duke@435	1787	// name - stub name string
duke@435	1788	//
duke@435	1789	// Inputs:
duke@435	1790	// c_rarg0 - source array address
duke@435	1791	// c_rarg1 - destination array address
duke@435	1792	// c_rarg2 - element count, treated as ssize_t, can be zero
duke@435	1793	//
duke@435	1794	// If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
duke@435	1795	// the hardware handle it. The two dwords within qwords that span
duke@435	1796	// cache line boundaries will still be loaded and stored atomicly.
duke@435	1797	//
coleenp@548	1798	address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, const char *name) {
duke@435	1799	__ align(CodeEntryAlignment);
duke@435	1800	StubCodeMark mark(this, "StubRoutines", name);
duke@435	1801	address start = __ pc();
duke@435	1802
coleenp@548	1803	Label L_copy_32_bytes, L_copy_8_bytes, L_copy_2_bytes, L_exit;
duke@435	1804	const Register from = rdi; // source array address
duke@435	1805	const Register to = rsi; // destination array address
duke@435	1806	const Register count = rdx; // elements count
duke@435	1807	const Register dword_count = rcx;
duke@435	1808	const Register qword_count = count;
duke@435	1809
duke@435	1810	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	1811	assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
duke@435	1812
coleenp@548	1813	if (is_oop) {
coleenp@548	1814	// no registers are destroyed by this call
coleenp@548	1815	gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2);
coleenp@548	1816	}
coleenp@548	1817
coleenp@548	1818	(is_oop ? oop_copy_entry : int_copy_entry) = __ pc();
duke@435	1819	BLOCK_COMMENT("Entry:");
duke@435	1820	// caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
duke@435	1821
coleenp@548	1822	array_overlap_test(is_oop ? disjoint_oop_copy_entry : disjoint_int_copy_entry,
coleenp@548	1823	Address::times_4);
duke@435	1824	setup_arg_regs(); // from => rdi, to => rsi, count => rdx
duke@435	1825	// r9 and r10 may be used to save non-volatile registers
duke@435	1826
coleenp@548	1827	assert_clean_int(count, rax); // Make sure 'count' is clean int.
duke@435	1828	// 'from', 'to' and 'count' are now valid
never@739	1829	__ movptr(dword_count, count);
never@739	1830	__ shrptr(count, 1); // count => qword_count
duke@435	1831
duke@435	1832	// Copy from high to low addresses. Use 'to' as scratch.
duke@435	1833
duke@435	1834	// Check for and copy trailing dword
never@739	1835	__ testl(dword_count, 1);
duke@435	1836	__ jcc(Assembler::zero, L_copy_32_bytes);
duke@435	1837	__ movl(rax, Address(from, dword_count, Address::times_4, -4));
duke@435	1838	__ movl(Address(to, dword_count, Address::times_4, -4), rax);
duke@435	1839	__ jmp(L_copy_32_bytes);
duke@435	1840
duke@435	1841	// Copy trailing qwords
duke@435	1842	__ BIND(L_copy_8_bytes);
duke@435	1843	__ movq(rax, Address(from, qword_count, Address::times_8, -8));
duke@435	1844	__ movq(Address(to, qword_count, Address::times_8, -8), rax);
never@739	1845	__ decrement(qword_count);
duke@435	1846	__ jcc(Assembler::notZero, L_copy_8_bytes);
duke@435	1847
duke@435	1848	inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
coleenp@548	1849	if (is_oop) {
coleenp@548	1850	__ jmp(L_exit);
coleenp@548	1851	}
duke@435	1852	restore_arg_regs();
never@739	1853	__ xorptr(rax, rax); // return 0
duke@435	1854	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	1855	__ ret(0);
duke@435	1856
duke@435	1857	// Copy in 32-bytes chunks
duke@435	1858	copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
duke@435	1859
coleenp@548	1860	inc_counter_np(SharedRuntime::_jint_array_copy_ctr);
coleenp@548	1861	__ bind(L_exit);
coleenp@548	1862	if (is_oop) {
coleenp@548	1863	Register end_to = rdx;
coleenp@548	1864	__ leaq(end_to, Address(to, dword_count, Address::times_4, -4));
coleenp@548	1865	gen_write_ref_array_post_barrier(to, end_to, rax);
coleenp@548	1866	}
duke@435	1867	restore_arg_regs();
never@739	1868	__ xorptr(rax, rax); // return 0
duke@435	1869	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	1870	__ ret(0);
duke@435	1871
duke@435	1872	return start;
duke@435	1873	}
duke@435	1874
duke@435	1875	// Arguments:
duke@435	1876	// aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
duke@435	1877	// ignored
duke@435	1878	// is_oop - true => oop array, so generate store check code
duke@435	1879	// name - stub name string
duke@435	1880	//
duke@435	1881	// Inputs:
duke@435	1882	// c_rarg0 - source array address
duke@435	1883	// c_rarg1 - destination array address
duke@435	1884	// c_rarg2 - element count, treated as ssize_t, can be zero
duke@435	1885	//
coleenp@548	1886	// Side Effects:
duke@435	1887	// disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the
duke@435	1888	// no-overlap entry point used by generate_conjoint_long_oop_copy().
duke@435	1889	//
duke@435	1890	address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, const char *name) {
duke@435	1891	__ align(CodeEntryAlignment);
duke@435	1892	StubCodeMark mark(this, "StubRoutines", name);
duke@435	1893	address start = __ pc();
duke@435	1894
duke@435	1895	Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
duke@435	1896	const Register from = rdi; // source array address
duke@435	1897	const Register to = rsi; // destination array address
duke@435	1898	const Register qword_count = rdx; // elements count
duke@435	1899	const Register end_from = from; // source array end address
duke@435	1900	const Register end_to = rcx; // destination array end address
duke@435	1901	const Register saved_to = to;
duke@435	1902	// End pointers are inclusive, and if count is not zero they point
duke@435	1903	// to the last unit copied: end_to[0] := end_from[0]
duke@435	1904
duke@435	1905	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	1906	// Save no-overlap entry point for generate_conjoint_long_oop_copy()
duke@435	1907	assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
duke@435	1908
duke@435	1909	if (is_oop) {
duke@435	1910	disjoint_oop_copy_entry = __ pc();
duke@435	1911	// no registers are destroyed by this call
duke@435	1912	gen_write_ref_array_pre_barrier(/* dest */ c_rarg1, /* count */ c_rarg2);
duke@435	1913	} else {
duke@435	1914	disjoint_long_copy_entry = __ pc();
duke@435	1915	}
duke@435	1916	BLOCK_COMMENT("Entry:");
duke@435	1917	// caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
duke@435	1918
duke@435	1919	setup_arg_regs(); // from => rdi, to => rsi, count => rdx
duke@435	1920	// r9 and r10 may be used to save non-volatile registers
duke@435	1921
duke@435	1922	// 'from', 'to' and 'qword_count' are now valid
duke@435	1923
duke@435	1924	// Copy from low to high addresses. Use 'to' as scratch.
never@739	1925	__ lea(end_from, Address(from, qword_count, Address::times_8, -8));
never@739	1926	__ lea(end_to, Address(to, qword_count, Address::times_8, -8));
never@739	1927	__ negptr(qword_count);
duke@435	1928	__ jmp(L_copy_32_bytes);
duke@435	1929
duke@435	1930	// Copy trailing qwords
duke@435	1931	__ BIND(L_copy_8_bytes);
duke@435	1932	__ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
duke@435	1933	__ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
never@739	1934	__ increment(qword_count);
duke@435	1935	__ jcc(Assembler::notZero, L_copy_8_bytes);
duke@435	1936
duke@435	1937	if (is_oop) {
duke@435	1938	__ jmp(L_exit);
duke@435	1939	} else {
duke@435	1940	inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
duke@435	1941	restore_arg_regs();
never@739	1942	__ xorptr(rax, rax); // return 0
duke@435	1943	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	1944	__ ret(0);
duke@435	1945	}
duke@435	1946
duke@435	1947	// Copy 64-byte chunks
duke@435	1948	copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
duke@435	1949
duke@435	1950	if (is_oop) {
duke@435	1951	__ BIND(L_exit);
duke@435	1952	gen_write_ref_array_post_barrier(saved_to, end_to, rax);
duke@435	1953	inc_counter_np(SharedRuntime::_oop_array_copy_ctr);
duke@435	1954	} else {
duke@435	1955	inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
duke@435	1956	}
duke@435	1957	restore_arg_regs();
never@739	1958	__ xorptr(rax, rax); // return 0
duke@435	1959	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	1960	__ ret(0);
duke@435	1961
duke@435	1962	return start;
duke@435	1963	}
duke@435	1964
duke@435	1965	// Arguments:
duke@435	1966	// aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
duke@435	1967	// ignored
duke@435	1968	// is_oop - true => oop array, so generate store check code
duke@435	1969	// name - stub name string
duke@435	1970	//
duke@435	1971	// Inputs:
duke@435	1972	// c_rarg0 - source array address
duke@435	1973	// c_rarg1 - destination array address
duke@435	1974	// c_rarg2 - element count, treated as ssize_t, can be zero
duke@435	1975	//
duke@435	1976	address generate_conjoint_long_oop_copy(bool aligned, bool is_oop, const char *name) {
duke@435	1977	__ align(CodeEntryAlignment);
duke@435	1978	StubCodeMark mark(this, "StubRoutines", name);
duke@435	1979	address start = __ pc();
duke@435	1980
duke@435	1981	Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
duke@435	1982	const Register from = rdi; // source array address
duke@435	1983	const Register to = rsi; // destination array address
duke@435	1984	const Register qword_count = rdx; // elements count
duke@435	1985	const Register saved_count = rcx;
duke@435	1986
duke@435	1987	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	1988	assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
duke@435	1989
duke@435	1990	address disjoint_copy_entry = NULL;
duke@435	1991	if (is_oop) {
coleenp@548	1992	assert(!UseCompressedOops, "shouldn't be called for compressed oops");
duke@435	1993	disjoint_copy_entry = disjoint_oop_copy_entry;
duke@435	1994	oop_copy_entry = __ pc();
coleenp@548	1995	array_overlap_test(disjoint_oop_copy_entry, Address::times_8);
duke@435	1996	} else {
duke@435	1997	disjoint_copy_entry = disjoint_long_copy_entry;
duke@435	1998	long_copy_entry = __ pc();
coleenp@548	1999	array_overlap_test(disjoint_long_copy_entry, Address::times_8);
duke@435	2000	}
duke@435	2001	BLOCK_COMMENT("Entry:");
duke@435	2002	// caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
duke@435	2003
duke@435	2004	array_overlap_test(disjoint_copy_entry, Address::times_8);
duke@435	2005	setup_arg_regs(); // from => rdi, to => rsi, count => rdx
duke@435	2006	// r9 and r10 may be used to save non-volatile registers
duke@435	2007
duke@435	2008	// 'from', 'to' and 'qword_count' are now valid
duke@435	2009
duke@435	2010	if (is_oop) {
duke@435	2011	// Save to and count for store barrier
never@739	2012	__ movptr(saved_count, qword_count);
duke@435	2013	// No registers are destroyed by this call
duke@435	2014	gen_write_ref_array_pre_barrier(to, saved_count);
duke@435	2015	}
duke@435	2016
duke@435	2017	__ jmp(L_copy_32_bytes);
duke@435	2018
duke@435	2019	// Copy trailing qwords
duke@435	2020	__ BIND(L_copy_8_bytes);
duke@435	2021	__ movq(rax, Address(from, qword_count, Address::times_8, -8));
duke@435	2022	__ movq(Address(to, qword_count, Address::times_8, -8), rax);
never@739	2023	__ decrement(qword_count);
duke@435	2024	__ jcc(Assembler::notZero, L_copy_8_bytes);
duke@435	2025
duke@435	2026	if (is_oop) {
duke@435	2027	__ jmp(L_exit);
duke@435	2028	} else {
duke@435	2029	inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
duke@435	2030	restore_arg_regs();
never@739	2031	__ xorptr(rax, rax); // return 0
duke@435	2032	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	2033	__ ret(0);
duke@435	2034	}
duke@435	2035
duke@435	2036	// Copy in 32-bytes chunks
duke@435	2037	copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
duke@435	2038
duke@435	2039	if (is_oop) {
duke@435	2040	__ BIND(L_exit);
never@739	2041	__ lea(rcx, Address(to, saved_count, Address::times_8, -8));
duke@435	2042	gen_write_ref_array_post_barrier(to, rcx, rax);
duke@435	2043	inc_counter_np(SharedRuntime::_oop_array_copy_ctr);
duke@435	2044	} else {
duke@435	2045	inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
duke@435	2046	}
duke@435	2047	restore_arg_regs();
never@739	2048	__ xorptr(rax, rax); // return 0
duke@435	2049	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	2050	__ ret(0);
duke@435	2051
duke@435	2052	return start;
duke@435	2053	}
duke@435	2054
duke@435	2055
duke@435	2056	// Helper for generating a dynamic type check.
duke@435	2057	// Smashes no registers.
duke@435	2058	void generate_type_check(Register sub_klass,
duke@435	2059	Register super_check_offset,
duke@435	2060	Register super_klass,
duke@435	2061	Label& L_success) {
duke@435	2062	assert_different_registers(sub_klass, super_check_offset, super_klass);
duke@435	2063
duke@435	2064	BLOCK_COMMENT("type_check:");
duke@435	2065
duke@435	2066	Label L_miss;
duke@435	2067
duke@435	2068	// a couple of useful fields in sub_klass:
duke@435	2069	int ss_offset = (klassOopDesc::header_size() * HeapWordSize +
duke@435	2070	Klass::secondary_supers_offset_in_bytes());
duke@435	2071	int sc_offset = (klassOopDesc::header_size() * HeapWordSize +
duke@435	2072	Klass::secondary_super_cache_offset_in_bytes());
duke@435	2073	Address secondary_supers_addr(sub_klass, ss_offset);
duke@435	2074	Address super_cache_addr( sub_klass, sc_offset);
duke@435	2075
duke@435	2076	// if the pointers are equal, we are done (e.g., String[] elements)
never@739	2077	__ cmpptr(super_klass, sub_klass);
duke@435	2078	__ jcc(Assembler::equal, L_success);
duke@435	2079
duke@435	2080	// check the supertype display:
duke@435	2081	Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0);
never@739	2082	__ cmpptr(super_klass, super_check_addr); // test the super type
duke@435	2083	__ jcc(Assembler::equal, L_success);
duke@435	2084
duke@435	2085	// if it was a primary super, we can just fail immediately
duke@435	2086	__ cmpl(super_check_offset, sc_offset);
duke@435	2087	__ jcc(Assembler::notEqual, L_miss);
duke@435	2088
duke@435	2089	// Now do a linear scan of the secondary super-klass chain.
duke@435	2090	// The repne_scan instruction uses fixed registers, which we must spill.
duke@435	2091	// (We need a couple more temps in any case.)
duke@435	2092	// This code is rarely used, so simplicity is a virtue here.
duke@435	2093	inc_counter_np(SharedRuntime::_partial_subtype_ctr);
duke@435	2094	{
never@739	2095	__ push(rax);
never@739	2096	__ push(rcx);
never@739	2097	__ push(rdi);
duke@435	2098	assert_different_registers(sub_klass, super_klass, rax, rcx, rdi);
duke@435	2099
never@739	2100	__ movptr(rdi, secondary_supers_addr);
duke@435	2101	// Load the array length.
duke@435	2102	__ movl(rcx, Address(rdi, arrayOopDesc::length_offset_in_bytes()));
duke@435	2103	// Skip to start of data.
never@739	2104	__ addptr(rdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
duke@435	2105	// Scan rcx words at [rdi] for occurance of rax
duke@435	2106	// Set NZ/Z based on last compare
never@739	2107	__ movptr(rax, super_klass);
coleenp@548	2108	if (UseCompressedOops) {
coleenp@548	2109	// Compare against compressed form. Don't need to uncompress because
coleenp@548	2110	// looks like orig rax is restored in popq below.
coleenp@548	2111	__ encode_heap_oop(rax);
coleenp@548	2112	__ repne_scanl();
coleenp@548	2113	} else {
never@739	2114	__ repne_scan();
coleenp@548	2115	}
duke@435	2116
duke@435	2117	// Unspill the temp. registers:
never@739	2118	__ pop(rdi);
never@739	2119	__ pop(rcx);
never@739	2120	__ pop(rax);
duke@435	2121
duke@435	2122	__ jcc(Assembler::notEqual, L_miss);
duke@435	2123	}
duke@435	2124
duke@435	2125	// Success. Cache the super we found and proceed in triumph.
never@739	2126	__ movptr(super_cache_addr, super_klass); // note: rax is dead
duke@435	2127	__ jmp(L_success);
duke@435	2128
duke@435	2129	// Fall through on failure!
duke@435	2130	__ BIND(L_miss);
duke@435	2131	}
duke@435	2132
duke@435	2133	//
duke@435	2134	// Generate checkcasting array copy stub
duke@435	2135	//
duke@435	2136	// Input:
duke@435	2137	// c_rarg0 - source array address
duke@435	2138	// c_rarg1 - destination array address
duke@435	2139	// c_rarg2 - element count, treated as ssize_t, can be zero
duke@435	2140	// c_rarg3 - size_t ckoff (super_check_offset)
duke@435	2141	// not Win64
duke@435	2142	// c_rarg4 - oop ckval (super_klass)
duke@435	2143	// Win64
duke@435	2144	// rsp+40 - oop ckval (super_klass)
duke@435	2145	//
duke@435	2146	// Output:
duke@435	2147	// rax == 0 - success
duke@435	2148	// rax == -1^K - failure, where K is partial transfer count
duke@435	2149	//
duke@435	2150	address generate_checkcast_copy(const char *name) {
duke@435	2151
duke@435	2152	Label L_load_element, L_store_element, L_do_card_marks, L_done;
duke@435	2153
duke@435	2154	// Input registers (after setup_arg_regs)
duke@435	2155	const Register from = rdi; // source array address
duke@435	2156	const Register to = rsi; // destination array address
duke@435	2157	const Register length = rdx; // elements count
duke@435	2158	const Register ckoff = rcx; // super_check_offset
duke@435	2159	const Register ckval = r8; // super_klass
duke@435	2160
duke@435	2161	// Registers used as temps (r13, r14 are save-on-entry)
duke@435	2162	const Register end_from = from; // source array end address
duke@435	2163	const Register end_to = r13; // destination array end address
duke@435	2164	const Register count = rdx; // -(count_remaining)
duke@435	2165	const Register r14_length = r14; // saved copy of length
duke@435	2166	// End pointers are inclusive, and if length is not zero they point
duke@435	2167	// to the last unit copied: end_to[0] := end_from[0]
duke@435	2168
duke@435	2169	const Register rax_oop = rax; // actual oop copied
duke@435	2170	const Register r11_klass = r11; // oop._klass
duke@435	2171
duke@435	2172	//---------------------------------------------------------------
duke@435	2173	// Assembler stub will be used for this call to arraycopy
duke@435	2174	// if the two arrays are subtypes of Object[] but the
duke@435	2175	// destination array type is not equal to or a supertype
duke@435	2176	// of the source type. Each element must be separately
duke@435	2177	// checked.
duke@435	2178
duke@435	2179	__ align(CodeEntryAlignment);
duke@435	2180	StubCodeMark mark(this, "StubRoutines", name);
duke@435	2181	address start = __ pc();
duke@435	2182
duke@435	2183	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	2184
duke@435	2185	checkcast_copy_entry = __ pc();
duke@435	2186	BLOCK_COMMENT("Entry:");
duke@435	2187
duke@435	2188	#ifdef ASSERT
duke@435	2189	// caller guarantees that the arrays really are different
duke@435	2190	// otherwise, we would have to make conjoint checks
duke@435	2191	{ Label L;
coleenp@548	2192	array_overlap_test(L, TIMES_OOP);
duke@435	2193	__ stop("checkcast_copy within a single array");
duke@435	2194	__ bind(L);
duke@435	2195	}
duke@435	2196	#endif //ASSERT
duke@435	2197
duke@435	2198	// allocate spill slots for r13, r14
duke@435	2199	enum {
duke@435	2200	saved_r13_offset,
duke@435	2201	saved_r14_offset,
duke@435	2202	saved_rbp_offset,
duke@435	2203	saved_rip_offset,
duke@435	2204	saved_rarg0_offset
duke@435	2205	};
never@739	2206	__ subptr(rsp, saved_rbp_offset * wordSize);
never@739	2207	__ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
never@739	2208	__ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
duke@435	2209	setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
duke@435	2210	// ckoff => rcx, ckval => r8
duke@435	2211	// r9 and r10 may be used to save non-volatile registers
duke@435	2212	#ifdef _WIN64
duke@435	2213	// last argument (#4) is on stack on Win64
duke@435	2214	const int ckval_offset = saved_rarg0_offset + 4;
never@739	2215	__ movptr(ckval, Address(rsp, ckval_offset * wordSize));
duke@435	2216	#endif
duke@435	2217
duke@435	2218	// check that int operands are properly extended to size_t
duke@435	2219	assert_clean_int(length, rax);
duke@435	2220	assert_clean_int(ckoff, rax);
duke@435	2221
duke@435	2222	#ifdef ASSERT
duke@435	2223	BLOCK_COMMENT("assert consistent ckoff/ckval");
duke@435	2224	// The ckoff and ckval must be mutually consistent,
duke@435	2225	// even though caller generates both.
duke@435	2226	{ Label L;
duke@435	2227	int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
duke@435	2228	Klass::super_check_offset_offset_in_bytes());
duke@435	2229	__ cmpl(ckoff, Address(ckval, sco_offset));
duke@435	2230	__ jcc(Assembler::equal, L);
duke@435	2231	__ stop("super_check_offset inconsistent");
duke@435	2232	__ bind(L);
duke@435	2233	}
duke@435	2234	#endif //ASSERT
duke@435	2235
duke@435	2236	// Loop-invariant addresses. They are exclusive end pointers.
coleenp@548	2237	Address end_from_addr(from, length, TIMES_OOP, 0);
coleenp@548	2238	Address end_to_addr(to, length, TIMES_OOP, 0);
duke@435	2239	// Loop-variant addresses. They assume post-incremented count < 0.
coleenp@548	2240	Address from_element_addr(end_from, count, TIMES_OOP, 0);
coleenp@548	2241	Address to_element_addr(end_to, count, TIMES_OOP, 0);
duke@435	2242
duke@435	2243	gen_write_ref_array_pre_barrier(to, count);
duke@435	2244
duke@435	2245	// Copy from low to high addresses, indexed from the end of each array.
never@739	2246	__ lea(end_from, end_from_addr);
never@739	2247	__ lea(end_to, end_to_addr);
never@739	2248	__ movptr(r14_length, length); // save a copy of the length
never@739	2249	assert(length == count, ""); // else fix next line:
never@739	2250	__ negptr(count); // negate and test the length
duke@435	2251	__ jcc(Assembler::notZero, L_load_element);
duke@435	2252
duke@435	2253	// Empty array: Nothing to do.
never@739	2254	__ xorptr(rax, rax); // return 0 on (trivial) success
duke@435	2255	__ jmp(L_done);
duke@435	2256
duke@435	2257	// ======== begin loop ========
duke@435	2258	// (Loop is rotated; its entry is L_load_element.)
duke@435	2259	// Loop control:
duke@435	2260	// for (count = -count; count != 0; count++)
duke@435	2261	// Base pointers src, dst are biased by 8*(count-1),to last element.
duke@435	2262	__ align(16);
duke@435	2263
duke@435	2264	__ BIND(L_store_element);
coleenp@548	2265	__ store_heap_oop(to_element_addr, rax_oop); // store the oop
never@739	2266	__ increment(count); // increment the count toward zero
duke@435	2267	__ jcc(Assembler::zero, L_do_card_marks);
duke@435	2268
duke@435	2269	// ======== loop entry is here ========
duke@435	2270	__ BIND(L_load_element);
coleenp@548	2271	__ load_heap_oop(rax_oop, from_element_addr); // load the oop
never@739	2272	__ testptr(rax_oop, rax_oop);
duke@435	2273	__ jcc(Assembler::zero, L_store_element);
duke@435	2274
coleenp@548	2275	__ load_klass(r11_klass, rax_oop);// query the object klass
duke@435	2276	generate_type_check(r11_klass, ckoff, ckval, L_store_element);
duke@435	2277	// ======== end loop ========
duke@435	2278
duke@435	2279	// It was a real error; we must depend on the caller to finish the job.
duke@435	2280	// Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
duke@435	2281	// Emit GC store barriers for the oops we have copied (r14 + rdx),
duke@435	2282	// and report their number to the caller.
duke@435	2283	assert_different_registers(rax, r14_length, count, to, end_to, rcx);
never@739	2284	__ lea(end_to, to_element_addr);
duke@435	2285	gen_write_ref_array_post_barrier(to, end_to, rcx);
never@739	2286	__ movptr(rax, r14_length); // original oops
never@739	2287	__ addptr(rax, count); // K = (original - remaining) oops
never@739	2288	__ notptr(rax); // report (-1^K) to caller
duke@435	2289	__ jmp(L_done);
duke@435	2290
duke@435	2291	// Come here on success only.
duke@435	2292	__ BIND(L_do_card_marks);
never@739	2293	__ addptr(end_to, -wordSize); // make an inclusive end pointer
duke@435	2294	gen_write_ref_array_post_barrier(to, end_to, rcx);
never@739	2295	__ xorptr(rax, rax); // return 0 on success
duke@435	2296
duke@435	2297	// Common exit point (success or failure).
duke@435	2298	__ BIND(L_done);
never@739	2299	__ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
never@739	2300	__ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
duke@435	2301	inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
duke@435	2302	restore_arg_regs();
duke@435	2303	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	2304	__ ret(0);
duke@435	2305
duke@435	2306	return start;
duke@435	2307	}
duke@435	2308
duke@435	2309	//
duke@435	2310	// Generate 'unsafe' array copy stub
duke@435	2311	// Though just as safe as the other stubs, it takes an unscaled
duke@435	2312	// size_t argument instead of an element count.
duke@435	2313	//
duke@435	2314	// Input:
duke@435	2315	// c_rarg0 - source array address
duke@435	2316	// c_rarg1 - destination array address
duke@435	2317	// c_rarg2 - byte count, treated as ssize_t, can be zero
duke@435	2318	//
duke@435	2319	// Examines the alignment of the operands and dispatches
duke@435	2320	// to a long, int, short, or byte copy loop.
duke@435	2321	//
duke@435	2322	address generate_unsafe_copy(const char *name) {
duke@435	2323
duke@435	2324	Label L_long_aligned, L_int_aligned, L_short_aligned;
duke@435	2325
duke@435	2326	// Input registers (before setup_arg_regs)
duke@435	2327	const Register from = c_rarg0; // source array address
duke@435	2328	const Register to = c_rarg1; // destination array address
duke@435	2329	const Register size = c_rarg2; // byte count (size_t)
duke@435	2330
duke@435	2331	// Register used as a temp
duke@435	2332	const Register bits = rax; // test copy of low bits
duke@435	2333
duke@435	2334	__ align(CodeEntryAlignment);
duke@435	2335	StubCodeMark mark(this, "StubRoutines", name);
duke@435	2336	address start = __ pc();
duke@435	2337
duke@435	2338	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	2339
duke@435	2340	// bump this on entry, not on exit:
duke@435	2341	inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
duke@435	2342
never@739	2343	__ mov(bits, from);
never@739	2344	__ orptr(bits, to);
never@739	2345	__ orptr(bits, size);
duke@435	2346
duke@435	2347	__ testb(bits, BytesPerLong-1);
duke@435	2348	__ jccb(Assembler::zero, L_long_aligned);
duke@435	2349
duke@435	2350	__ testb(bits, BytesPerInt-1);
duke@435	2351	__ jccb(Assembler::zero, L_int_aligned);
duke@435	2352
duke@435	2353	__ testb(bits, BytesPerShort-1);
duke@435	2354	__ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
duke@435	2355
duke@435	2356	__ BIND(L_short_aligned);
never@739	2357	__ shrptr(size, LogBytesPerShort); // size => short_count
duke@435	2358	__ jump(RuntimeAddress(short_copy_entry));
duke@435	2359
duke@435	2360	__ BIND(L_int_aligned);
never@739	2361	__ shrptr(size, LogBytesPerInt); // size => int_count
duke@435	2362	__ jump(RuntimeAddress(int_copy_entry));
duke@435	2363
duke@435	2364	__ BIND(L_long_aligned);
never@739	2365	__ shrptr(size, LogBytesPerLong); // size => qword_count
duke@435	2366	__ jump(RuntimeAddress(long_copy_entry));
duke@435	2367
duke@435	2368	return start;
duke@435	2369	}
duke@435	2370
duke@435	2371	// Perform range checks on the proposed arraycopy.
duke@435	2372	// Kills temp, but nothing else.
duke@435	2373	// Also, clean the sign bits of src_pos and dst_pos.
duke@435	2374	void arraycopy_range_checks(Register src, // source array oop (c_rarg0)
duke@435	2375	Register src_pos, // source position (c_rarg1)
duke@435	2376	Register dst, // destination array oo (c_rarg2)
duke@435	2377	Register dst_pos, // destination position (c_rarg3)
duke@435	2378	Register length,
duke@435	2379	Register temp,
duke@435	2380	Label& L_failed) {
duke@435	2381	BLOCK_COMMENT("arraycopy_range_checks:");
duke@435	2382
duke@435	2383	// if (src_pos + length > arrayOop(src)->length()) FAIL;
duke@435	2384	__ movl(temp, length);
duke@435	2385	__ addl(temp, src_pos); // src_pos + length
duke@435	2386	__ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
duke@435	2387	__ jcc(Assembler::above, L_failed);
duke@435	2388
duke@435	2389	// if (dst_pos + length > arrayOop(dst)->length()) FAIL;
duke@435	2390	__ movl(temp, length);
duke@435	2391	__ addl(temp, dst_pos); // dst_pos + length
duke@435	2392	__ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
duke@435	2393	__ jcc(Assembler::above, L_failed);
duke@435	2394
duke@435	2395	// Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
duke@435	2396	// Move with sign extension can be used since they are positive.
duke@435	2397	__ movslq(src_pos, src_pos);
duke@435	2398	__ movslq(dst_pos, dst_pos);
duke@435	2399
duke@435	2400	BLOCK_COMMENT("arraycopy_range_checks done");
duke@435	2401	}
duke@435	2402
duke@435	2403	//
duke@435	2404	// Generate generic array copy stubs
duke@435	2405	//
duke@435	2406	// Input:
duke@435	2407	// c_rarg0 - src oop
duke@435	2408	// c_rarg1 - src_pos (32-bits)
duke@435	2409	// c_rarg2 - dst oop
duke@435	2410	// c_rarg3 - dst_pos (32-bits)
duke@435	2411	// not Win64
duke@435	2412	// c_rarg4 - element count (32-bits)
duke@435	2413	// Win64
duke@435	2414	// rsp+40 - element count (32-bits)
duke@435	2415	//
duke@435	2416	// Output:
duke@435	2417	// rax == 0 - success
duke@435	2418	// rax == -1^K - failure, where K is partial transfer count
duke@435	2419	//
duke@435	2420	address generate_generic_copy(const char *name) {
duke@435	2421
duke@435	2422	Label L_failed, L_failed_0, L_objArray;
duke@435	2423	Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
duke@435	2424
duke@435	2425	// Input registers
duke@435	2426	const Register src = c_rarg0; // source array oop
duke@435	2427	const Register src_pos = c_rarg1; // source position
duke@435	2428	const Register dst = c_rarg2; // destination array oop
duke@435	2429	const Register dst_pos = c_rarg3; // destination position
duke@435	2430	// elements count is on stack on Win64
duke@435	2431	#ifdef _WIN64
duke@435	2432	#define C_RARG4 Address(rsp, 6 * wordSize)
duke@435	2433	#else
duke@435	2434	#define C_RARG4 c_rarg4
duke@435	2435	#endif
duke@435	2436
duke@435	2437	{ int modulus = CodeEntryAlignment;
duke@435	2438	int target = modulus - 5; // 5 = sizeof jmp(L_failed)
duke@435	2439	int advance = target - (__ offset() % modulus);
duke@435	2440	if (advance < 0) advance += modulus;
duke@435	2441	if (advance > 0) __ nop(advance);
duke@435	2442	}
duke@435	2443	StubCodeMark mark(this, "StubRoutines", name);
duke@435	2444
duke@435	2445	// Short-hop target to L_failed. Makes for denser prologue code.
duke@435	2446	__ BIND(L_failed_0);
duke@435	2447	__ jmp(L_failed);
duke@435	2448	assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
duke@435	2449
duke@435	2450	__ align(CodeEntryAlignment);
duke@435	2451	address start = __ pc();
duke@435	2452
duke@435	2453	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	2454
duke@435	2455	// bump this on entry, not on exit:
duke@435	2456	inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
duke@435	2457
duke@435	2458	//-----------------------------------------------------------------------
duke@435	2459	// Assembler stub will be used for this call to arraycopy
duke@435	2460	// if the following conditions are met:
duke@435	2461	//
duke@435	2462	// (1) src and dst must not be null.
duke@435	2463	// (2) src_pos must not be negative.
duke@435	2464	// (3) dst_pos must not be negative.
duke@435	2465	// (4) length must not be negative.
duke@435	2466	// (5) src klass and dst klass should be the same and not NULL.
duke@435	2467	// (6) src and dst should be arrays.
duke@435	2468	// (7) src_pos + length must not exceed length of src.
duke@435	2469	// (8) dst_pos + length must not exceed length of dst.
duke@435	2470	//
duke@435	2471
duke@435	2472	// if (src == NULL) return -1;
never@739	2473	__ testptr(src, src); // src oop
duke@435	2474	size_t j1off = __ offset();
duke@435	2475	__ jccb(Assembler::zero, L_failed_0);
duke@435	2476
duke@435	2477	// if (src_pos < 0) return -1;
duke@435	2478	__ testl(src_pos, src_pos); // src_pos (32-bits)
duke@435	2479	__ jccb(Assembler::negative, L_failed_0);
duke@435	2480
duke@435	2481	// if (dst == NULL) return -1;
never@739	2482	__ testptr(dst, dst); // dst oop
duke@435	2483	__ jccb(Assembler::zero, L_failed_0);
duke@435	2484
duke@435	2485	// if (dst_pos < 0) return -1;
duke@435	2486	__ testl(dst_pos, dst_pos); // dst_pos (32-bits)
duke@435	2487	size_t j4off = __ offset();
duke@435	2488	__ jccb(Assembler::negative, L_failed_0);
duke@435	2489
duke@435	2490	// The first four tests are very dense code,
duke@435	2491	// but not quite dense enough to put four
duke@435	2492	// jumps in a 16-byte instruction fetch buffer.
duke@435	2493	// That's good, because some branch predicters
duke@435	2494	// do not like jumps so close together.
duke@435	2495	// Make sure of this.
duke@435	2496	guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
duke@435	2497
duke@435	2498	// registers used as temp
duke@435	2499	const Register r11_length = r11; // elements count to copy
duke@435	2500	const Register r10_src_klass = r10; // array klass
coleenp@548	2501	const Register r9_dst_klass = r9; // dest array klass
duke@435	2502
duke@435	2503	// if (length < 0) return -1;
duke@435	2504	__ movl(r11_length, C_RARG4); // length (elements count, 32-bits value)
duke@435	2505	__ testl(r11_length, r11_length);
duke@435	2506	__ jccb(Assembler::negative, L_failed_0);
duke@435	2507
coleenp@548	2508	__ load_klass(r10_src_klass, src);
duke@435	2509	#ifdef ASSERT
duke@435	2510	// assert(src->klass() != NULL);
duke@435	2511	BLOCK_COMMENT("assert klasses not null");
duke@435	2512	{ Label L1, L2;
never@739	2513	__ testptr(r10_src_klass, r10_src_klass);
duke@435	2514	__ jcc(Assembler::notZero, L2); // it is broken if klass is NULL
duke@435	2515	__ bind(L1);
duke@435	2516	__ stop("broken null klass");
duke@435	2517	__ bind(L2);
coleenp@548	2518	__ load_klass(r9_dst_klass, dst);
coleenp@548	2519	__ cmpq(r9_dst_klass, 0);
duke@435	2520	__ jcc(Assembler::equal, L1); // this would be broken also
duke@435	2521	BLOCK_COMMENT("assert done");
duke@435	2522	}
duke@435	2523	#endif
duke@435	2524
duke@435	2525	// Load layout helper (32-bits)
duke@435	2526	//
duke@435	2527	// |array_tag| | header_size | element_type | |log2_element_size|
duke@435	2528	// 32 30 24 16 8 2 0
duke@435	2529	//
duke@435	2530	// array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
duke@435	2531	//
duke@435	2532
duke@435	2533	int lh_offset = klassOopDesc::header_size() * HeapWordSize +
duke@435	2534	Klass::layout_helper_offset_in_bytes();
duke@435	2535
duke@435	2536	const Register rax_lh = rax; // layout helper
duke@435	2537
duke@435	2538	__ movl(rax_lh, Address(r10_src_klass, lh_offset));
duke@435	2539
duke@435	2540	// Handle objArrays completely differently...
duke@435	2541	jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
duke@435	2542	__ cmpl(rax_lh, objArray_lh);
duke@435	2543	__ jcc(Assembler::equal, L_objArray);
duke@435	2544
duke@435	2545	// if (src->klass() != dst->klass()) return -1;
coleenp@548	2546	__ load_klass(r9_dst_klass, dst);
coleenp@548	2547	__ cmpq(r10_src_klass, r9_dst_klass);
duke@435	2548	__ jcc(Assembler::notEqual, L_failed);
duke@435	2549
duke@435	2550	// if (!src->is_Array()) return -1;
duke@435	2551	__ cmpl(rax_lh, Klass::_lh_neutral_value);
duke@435	2552	__ jcc(Assembler::greaterEqual, L_failed);
duke@435	2553
duke@435	2554	// At this point, it is known to be a typeArray (array_tag 0x3).
duke@435	2555	#ifdef ASSERT
duke@435	2556	{ Label L;
duke@435	2557	__ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
duke@435	2558	__ jcc(Assembler::greaterEqual, L);
duke@435	2559	__ stop("must be a primitive array");
duke@435	2560	__ bind(L);
duke@435	2561	}
duke@435	2562	#endif
duke@435	2563
duke@435	2564	arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
duke@435	2565	r10, L_failed);
duke@435	2566
duke@435	2567	// typeArrayKlass
duke@435	2568	//
duke@435	2569	// src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
duke@435	2570	// dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
duke@435	2571	//
duke@435	2572
duke@435	2573	const Register r10_offset = r10; // array offset
duke@435	2574	const Register rax_elsize = rax_lh; // element size
duke@435	2575
duke@435	2576	__ movl(r10_offset, rax_lh);
duke@435	2577	__ shrl(r10_offset, Klass::_lh_header_size_shift);
never@739	2578	__ andptr(r10_offset, Klass::_lh_header_size_mask); // array_offset
never@739	2579	__ addptr(src, r10_offset); // src array offset
never@739	2580	__ addptr(dst, r10_offset); // dst array offset
duke@435	2581	BLOCK_COMMENT("choose copy loop based on element size");
duke@435	2582	__ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize
duke@435	2583
duke@435	2584	// next registers should be set before the jump to corresponding stub
duke@435	2585	const Register from = c_rarg0; // source array address
duke@435	2586	const Register to = c_rarg1; // destination array address
duke@435	2587	const Register count = c_rarg2; // elements count
duke@435	2588
duke@435	2589	// 'from', 'to', 'count' registers should be set in such order
duke@435	2590	// since they are the same as 'src', 'src_pos', 'dst'.
duke@435	2591
duke@435	2592	__ BIND(L_copy_bytes);
duke@435	2593	__ cmpl(rax_elsize, 0);
duke@435	2594	__ jccb(Assembler::notEqual, L_copy_shorts);
never@739	2595	__ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr
never@739	2596	__ lea(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr
never@739	2597	__ movl2ptr(count, r11_length); // length
duke@435	2598	__ jump(RuntimeAddress(byte_copy_entry));
duke@435	2599
duke@435	2600	__ BIND(L_copy_shorts);
duke@435	2601	__ cmpl(rax_elsize, LogBytesPerShort);
duke@435	2602	__ jccb(Assembler::notEqual, L_copy_ints);
never@739	2603	__ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr
never@739	2604	__ lea(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr
never@739	2605	__ movl2ptr(count, r11_length); // length
duke@435	2606	__ jump(RuntimeAddress(short_copy_entry));
duke@435	2607
duke@435	2608	__ BIND(L_copy_ints);
duke@435	2609	__ cmpl(rax_elsize, LogBytesPerInt);
duke@435	2610	__ jccb(Assembler::notEqual, L_copy_longs);
never@739	2611	__ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr
never@739	2612	__ lea(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr
never@739	2613	__ movl2ptr(count, r11_length); // length
duke@435	2614	__ jump(RuntimeAddress(int_copy_entry));
duke@435	2615
duke@435	2616	__ BIND(L_copy_longs);
duke@435	2617	#ifdef ASSERT
duke@435	2618	{ Label L;
duke@435	2619	__ cmpl(rax_elsize, LogBytesPerLong);
duke@435	2620	__ jcc(Assembler::equal, L);
duke@435	2621	__ stop("must be long copy, but elsize is wrong");
duke@435	2622	__ bind(L);
duke@435	2623	}
duke@435	2624	#endif
never@739	2625	__ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr
never@739	2626	__ lea(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr
never@739	2627	__ movl2ptr(count, r11_length); // length
duke@435	2628	__ jump(RuntimeAddress(long_copy_entry));
duke@435	2629
duke@435	2630	// objArrayKlass
duke@435	2631	__ BIND(L_objArray);
duke@435	2632	// live at this point: r10_src_klass, src[_pos], dst[_pos]
duke@435	2633
duke@435	2634	Label L_plain_copy, L_checkcast_copy;
duke@435	2635	// test array classes for subtyping
coleenp@548	2636	__ load_klass(r9_dst_klass, dst);
coleenp@548	2637	__ cmpq(r10_src_klass, r9_dst_klass); // usual case is exact equality
duke@435	2638	__ jcc(Assembler::notEqual, L_checkcast_copy);
duke@435	2639
duke@435	2640	// Identically typed arrays can be copied without element-wise checks.
duke@435	2641	arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
duke@435	2642	r10, L_failed);
duke@435	2643
never@739	2644	__ lea(from, Address(src, src_pos, TIMES_OOP,
duke@435	2645	arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
never@739	2646	__ lea(to, Address(dst, dst_pos, TIMES_OOP,
never@739	2647	arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
never@739	2648	__ movl2ptr(count, r11_length); // length
duke@435	2649	__ BIND(L_plain_copy);
duke@435	2650	__ jump(RuntimeAddress(oop_copy_entry));
duke@435	2651
duke@435	2652	__ BIND(L_checkcast_copy);
duke@435	2653	// live at this point: r10_src_klass, !r11_length
duke@435	2654	{
duke@435	2655	// assert(r11_length == C_RARG4); // will reload from here
duke@435	2656	Register r11_dst_klass = r11;
coleenp@548	2657	__ load_klass(r11_dst_klass, dst);
duke@435	2658
duke@435	2659	// Before looking at dst.length, make sure dst is also an objArray.
duke@435	2660	__ cmpl(Address(r11_dst_klass, lh_offset), objArray_lh);
duke@435	2661	__ jcc(Assembler::notEqual, L_failed);
duke@435	2662
duke@435	2663	// It is safe to examine both src.length and dst.length.
duke@435	2664	#ifndef _WIN64
duke@435	2665	arraycopy_range_checks(src, src_pos, dst, dst_pos, C_RARG4,
duke@435	2666	rax, L_failed);
duke@435	2667	#else
duke@435	2668	__ movl(r11_length, C_RARG4); // reload
duke@435	2669	arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
duke@435	2670	rax, L_failed);
coleenp@548	2671	__ load_klass(r11_dst_klass, dst); // reload
duke@435	2672	#endif
duke@435	2673
duke@435	2674	// Marshal the base address arguments now, freeing registers.
never@739	2675	__ lea(from, Address(src, src_pos, TIMES_OOP,
duke@435	2676	arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
never@739	2677	__ lea(to, Address(dst, dst_pos, TIMES_OOP,
duke@435	2678	arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
duke@435	2679	__ movl(count, C_RARG4); // length (reloaded)
duke@435	2680	Register sco_temp = c_rarg3; // this register is free now
duke@435	2681	assert_different_registers(from, to, count, sco_temp,
duke@435	2682	r11_dst_klass, r10_src_klass);
duke@435	2683	assert_clean_int(count, sco_temp);
duke@435	2684
duke@435	2685	// Generate the type check.
duke@435	2686	int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
duke@435	2687	Klass::super_check_offset_offset_in_bytes());
duke@435	2688	__ movl(sco_temp, Address(r11_dst_klass, sco_offset));
duke@435	2689	assert_clean_int(sco_temp, rax);
duke@435	2690	generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
duke@435	2691
duke@435	2692	// Fetch destination element klass from the objArrayKlass header.
duke@435	2693	int ek_offset = (klassOopDesc::header_size() * HeapWordSize +
duke@435	2694	objArrayKlass::element_klass_offset_in_bytes());
never@739	2695	__ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset));
duke@435	2696	__ movl(sco_temp, Address(r11_dst_klass, sco_offset));
duke@435	2697	assert_clean_int(sco_temp, rax);
duke@435	2698
duke@435	2699	// the checkcast_copy loop needs two extra arguments:
duke@435	2700	assert(c_rarg3 == sco_temp, "#3 already in place");
never@739	2701	__ movptr(C_RARG4, r11_dst_klass); // dst.klass.element_klass
duke@435	2702	__ jump(RuntimeAddress(checkcast_copy_entry));
duke@435	2703	}
duke@435	2704
duke@435	2705	__ BIND(L_failed);
never@739	2706	__ xorptr(rax, rax);
never@739	2707	__ notptr(rax); // return -1
duke@435	2708	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	2709	__ ret(0);
duke@435	2710
duke@435	2711	return start;
duke@435	2712	}
duke@435	2713
duke@435	2714	#undef length_arg
duke@435	2715
duke@435	2716	void generate_arraycopy_stubs() {
duke@435	2717	// Call the conjoint generation methods immediately after
duke@435	2718	// the disjoint ones so that short branches from the former
duke@435	2719	// to the latter can be generated.
duke@435	2720	StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, "jbyte_disjoint_arraycopy");
duke@435	2721	StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, "jbyte_arraycopy");
duke@435	2722
duke@435	2723	StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, "jshort_disjoint_arraycopy");
duke@435	2724	StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, "jshort_arraycopy");
duke@435	2725
coleenp@548	2726	StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, false, "jint_disjoint_arraycopy");
coleenp@548	2727	StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, "jint_arraycopy");
duke@435	2728
duke@435	2729	StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, false, "jlong_disjoint_arraycopy");
duke@435	2730	StubRoutines::_jlong_arraycopy = generate_conjoint_long_oop_copy(false, false, "jlong_arraycopy");
duke@435	2731
coleenp@548	2732
coleenp@548	2733	if (UseCompressedOops) {
coleenp@548	2734	StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, true, "oop_disjoint_arraycopy");
coleenp@548	2735	StubRoutines::_oop_arraycopy = generate_conjoint_int_oop_copy(false, true, "oop_arraycopy");
coleenp@548	2736	} else {
coleenp@548	2737	StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, true, "oop_disjoint_arraycopy");
coleenp@548	2738	StubRoutines::_oop_arraycopy = generate_conjoint_long_oop_copy(false, true, "oop_arraycopy");
coleenp@548	2739	}
duke@435	2740
duke@435	2741	StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy");
duke@435	2742	StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy");
duke@435	2743	StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy");
duke@435	2744
duke@435	2745	// We don't generate specialized code for HeapWord-aligned source
duke@435	2746	// arrays, so just use the code we've already generated
duke@435	2747	StubRoutines::_arrayof_jbyte_disjoint_arraycopy = StubRoutines::_jbyte_disjoint_arraycopy;
duke@435	2748	StubRoutines::_arrayof_jbyte_arraycopy = StubRoutines::_jbyte_arraycopy;
duke@435	2749
duke@435	2750	StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy;
duke@435	2751	StubRoutines::_arrayof_jshort_arraycopy = StubRoutines::_jshort_arraycopy;
duke@435	2752
duke@435	2753	StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
duke@435	2754	StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
duke@435	2755
duke@435	2756	StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
duke@435	2757	StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
duke@435	2758
duke@435	2759	StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
duke@435	2760	StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
duke@435	2761	}
duke@435	2762
duke@435	2763	#undef __
duke@435	2764	#define __ masm->
duke@435	2765
duke@435	2766	// Continuation point for throwing of implicit exceptions that are
duke@435	2767	// not handled in the current activation. Fabricates an exception
duke@435	2768	// oop and initiates normal exception dispatching in this
duke@435	2769	// frame. Since we need to preserve callee-saved values (currently
duke@435	2770	// only for C2, but done for C1 as well) we need a callee-saved oop
duke@435	2771	// map and therefore have to make these stubs into RuntimeStubs
duke@435	2772	// rather than BufferBlobs. If the compiler needs all registers to
duke@435	2773	// be preserved between the fault point and the exception handler
duke@435	2774	// then it must assume responsibility for that in
duke@435	2775	// AbstractCompiler::continuation_for_implicit_null_exception or
duke@435	2776	// continuation_for_implicit_division_by_zero_exception. All other
duke@435	2777	// implicit exceptions (e.g., NullPointerException or
duke@435	2778	// AbstractMethodError on entry) are either at call sites or
duke@435	2779	// otherwise assume that stack unwinding will be initiated, so
duke@435	2780	// caller saved registers were assumed volatile in the compiler.
duke@435	2781	address generate_throw_exception(const char* name,
duke@435	2782	address runtime_entry,
duke@435	2783	bool restore_saved_exception_pc) {
duke@435	2784	// Information about frame layout at time of blocking runtime call.
duke@435	2785	// Note that we only have to preserve callee-saved registers since
duke@435	2786	// the compilers are responsible for supplying a continuation point
duke@435	2787	// if they expect all registers to be preserved.
duke@435	2788	enum layout {
duke@435	2789	rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
duke@435	2790	rbp_off2,
duke@435	2791	return_off,
duke@435	2792	return_off2,
duke@435	2793	framesize // inclusive of return address
duke@435	2794	};
duke@435	2795
duke@435	2796	int insts_size = 512;
duke@435	2797	int locs_size = 64;
duke@435	2798
duke@435	2799	CodeBuffer code(name, insts_size, locs_size);
duke@435	2800	OopMapSet* oop_maps = new OopMapSet();
duke@435	2801	MacroAssembler* masm = new MacroAssembler(&code);
duke@435	2802
duke@435	2803	address start = __ pc();
duke@435	2804
duke@435	2805	// This is an inlined and slightly modified version of call_VM
duke@435	2806	// which has the ability to fetch the return PC out of
duke@435	2807	// thread-local storage and also sets up last_Java_sp slightly
duke@435	2808	// differently than the real call_VM
duke@435	2809	if (restore_saved_exception_pc) {
never@739	2810	__ movptr(rax,
never@739	2811	Address(r15_thread,
never@739	2812	in_bytes(JavaThread::saved_exception_pc_offset())));
never@739	2813	__ push(rax);
duke@435	2814	}
duke@435	2815
duke@435	2816	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	2817
duke@435	2818	assert(is_even(framesize/2), "sp not 16-byte aligned");
duke@435	2819
duke@435	2820	// return address and rbp are already in place
never@739	2821	__ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
duke@435	2822
duke@435	2823	int frame_complete = __ pc() - start;
duke@435	2824
duke@435	2825	// Set up last_Java_sp and last_Java_fp
duke@435	2826	__ set_last_Java_frame(rsp, rbp, NULL);
duke@435	2827
duke@435	2828	// Call runtime
never@739	2829	__ movptr(c_rarg0, r15_thread);
duke@435	2830	BLOCK_COMMENT("call runtime_entry");
duke@435	2831	__ call(RuntimeAddress(runtime_entry));
duke@435	2832
duke@435	2833	// Generate oop map
duke@435	2834	OopMap* map = new OopMap(framesize, 0);
duke@435	2835
duke@435	2836	oop_maps->add_gc_map(__ pc() - start, map);
duke@435	2837
duke@435	2838	__ reset_last_Java_frame(true, false);
duke@435	2839
duke@435	2840	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	2841
duke@435	2842	// check for pending exceptions
duke@435	2843	#ifdef ASSERT
duke@435	2844	Label L;
never@739	2845	__ cmpptr(Address(r15_thread, Thread::pending_exception_offset()),
never@739	2846	(int32_t) NULL_WORD);
duke@435	2847	__ jcc(Assembler::notEqual, L);
duke@435	2848	__ should_not_reach_here();
duke@435	2849	__ bind(L);
duke@435	2850	#endif // ASSERT
duke@435	2851	__ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
duke@435	2852
duke@435	2853
duke@435	2854	// codeBlob framesize is in words (not VMRegImpl::slot_size)
duke@435	2855	RuntimeStub* stub =
duke@435	2856	RuntimeStub::new_runtime_stub(name,
duke@435	2857	&code,
duke@435	2858	frame_complete,
duke@435	2859	(framesize >> (LogBytesPerWord - LogBytesPerInt)),
duke@435	2860	oop_maps, false);
duke@435	2861	return stub->entry_point();
duke@435	2862	}
duke@435	2863
duke@435	2864	// Initialization
duke@435	2865	void generate_initial() {
duke@435	2866	// Generates all stubs and initializes the entry points
duke@435	2867
duke@435	2868	// This platform-specific stub is needed by generate_call_stub()
never@739	2869	StubRoutines::x86::_mxcsr_std = generate_fp_mask("mxcsr_std", 0x0000000000001F80);
duke@435	2870
duke@435	2871	// entry points that exist in all platforms Note: This is code
duke@435	2872	// that could be shared among different platforms - however the
duke@435	2873	// benefit seems to be smaller than the disadvantage of having a
duke@435	2874	// much more complicated generator structure. See also comment in
duke@435	2875	// stubRoutines.hpp.
duke@435	2876
duke@435	2877	StubRoutines::_forward_exception_entry = generate_forward_exception();
duke@435	2878
duke@435	2879	StubRoutines::_call_stub_entry =
duke@435	2880	generate_call_stub(StubRoutines::_call_stub_return_address);
duke@435	2881
duke@435	2882	// is referenced by megamorphic call
duke@435	2883	StubRoutines::_catch_exception_entry = generate_catch_exception();
duke@435	2884
duke@435	2885	// atomic calls
duke@435	2886	StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
duke@435	2887	StubRoutines::_atomic_xchg_ptr_entry = generate_atomic_xchg_ptr();
duke@435	2888	StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg();
duke@435	2889	StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
duke@435	2890	StubRoutines::_atomic_add_entry = generate_atomic_add();
duke@435	2891	StubRoutines::_atomic_add_ptr_entry = generate_atomic_add_ptr();
duke@435	2892	StubRoutines::_fence_entry = generate_orderaccess_fence();
duke@435	2893
duke@435	2894	StubRoutines::_handler_for_unsafe_access_entry =
duke@435	2895	generate_handler_for_unsafe_access();
duke@435	2896
duke@435	2897	// platform dependent
never@739	2898	StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp();
never@739	2899
never@739	2900	StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
duke@435	2901	}
duke@435	2902
duke@435	2903	void generate_all() {
duke@435	2904	// Generates all stubs and initializes the entry points
duke@435	2905
duke@435	2906	// These entry points require SharedInfo::stack0 to be set up in
duke@435	2907	// non-core builds and need to be relocatable, so they each
duke@435	2908	// fabricate a RuntimeStub internally.
duke@435	2909	StubRoutines::_throw_AbstractMethodError_entry =
duke@435	2910	generate_throw_exception("AbstractMethodError throw_exception",
duke@435	2911	CAST_FROM_FN_PTR(address,
duke@435	2912	SharedRuntime::
duke@435	2913	throw_AbstractMethodError),
duke@435	2914	false);
duke@435	2915
dcubed@451	2916	StubRoutines::_throw_IncompatibleClassChangeError_entry =
dcubed@451	2917	generate_throw_exception("IncompatibleClassChangeError throw_exception",
dcubed@451	2918	CAST_FROM_FN_PTR(address,
dcubed@451	2919	SharedRuntime::
dcubed@451	2920	throw_IncompatibleClassChangeError),
dcubed@451	2921	false);
dcubed@451	2922
duke@435	2923	StubRoutines::_throw_ArithmeticException_entry =
duke@435	2924	generate_throw_exception("ArithmeticException throw_exception",
duke@435	2925	CAST_FROM_FN_PTR(address,
duke@435	2926	SharedRuntime::
duke@435	2927	throw_ArithmeticException),
duke@435	2928	true);
duke@435	2929
duke@435	2930	StubRoutines::_throw_NullPointerException_entry =
duke@435	2931	generate_throw_exception("NullPointerException throw_exception",
duke@435	2932	CAST_FROM_FN_PTR(address,
duke@435	2933	SharedRuntime::
duke@435	2934	throw_NullPointerException),
duke@435	2935	true);
duke@435	2936
duke@435	2937	StubRoutines::_throw_NullPointerException_at_call_entry =
duke@435	2938	generate_throw_exception("NullPointerException at call throw_exception",
duke@435	2939	CAST_FROM_FN_PTR(address,
duke@435	2940	SharedRuntime::
duke@435	2941	throw_NullPointerException_at_call),
duke@435	2942	false);
duke@435	2943
duke@435	2944	StubRoutines::_throw_StackOverflowError_entry =
duke@435	2945	generate_throw_exception("StackOverflowError throw_exception",
duke@435	2946	CAST_FROM_FN_PTR(address,
duke@435	2947	SharedRuntime::
duke@435	2948	throw_StackOverflowError),
duke@435	2949	false);
duke@435	2950
duke@435	2951	// entry points that are platform specific
never@739	2952	StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
never@739	2953	StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
never@739	2954	StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
never@739	2955	StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
never@739	2956
never@739	2957	StubRoutines::x86::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF);
never@739	2958	StubRoutines::x86::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000);
never@739	2959	StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
never@739	2960	StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
duke@435	2961
duke@435	2962	// support for verify_oop (must happen after universe_init)
duke@435	2963	StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
duke@435	2964
duke@435	2965	// arraycopy stubs used by compilers
duke@435	2966	generate_arraycopy_stubs();
duke@435	2967	}
duke@435	2968
duke@435	2969	public:
duke@435	2970	StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
duke@435	2971	if (all) {
duke@435	2972	generate_all();
duke@435	2973	} else {
duke@435	2974	generate_initial();
duke@435	2975	}
duke@435	2976	}
duke@435	2977	}; // end class declaration
duke@435	2978
duke@435	2979	address StubGenerator::disjoint_byte_copy_entry = NULL;
duke@435	2980	address StubGenerator::disjoint_short_copy_entry = NULL;
duke@435	2981	address StubGenerator::disjoint_int_copy_entry = NULL;
duke@435	2982	address StubGenerator::disjoint_long_copy_entry = NULL;
duke@435	2983	address StubGenerator::disjoint_oop_copy_entry = NULL;
duke@435	2984
duke@435	2985	address StubGenerator::byte_copy_entry = NULL;
duke@435	2986	address StubGenerator::short_copy_entry = NULL;
duke@435	2987	address StubGenerator::int_copy_entry = NULL;
duke@435	2988	address StubGenerator::long_copy_entry = NULL;
duke@435	2989	address StubGenerator::oop_copy_entry = NULL;
duke@435	2990
duke@435	2991	address StubGenerator::checkcast_copy_entry = NULL;
duke@435	2992
duke@435	2993	void StubGenerator_generate(CodeBuffer* code, bool all) {
duke@435	2994	StubGenerator g(code, all);
duke@435	2995	}