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

src/cpu/x86/vm/stubGenerator_x86_64.cpp@e961c11b85fe (annotated)

src/cpu/x86/vm/stubGenerator_x86_64.cpp

Wed, 03 Apr 2013 11:12:57 -0700

author

kvn

date

Wed, 03 Apr 2013 11:12:57 -0700

changeset 4873

e961c11b85fe

parent 4411

e2e6bf86682c

child 5156

3f281b313240

permissions

-rw-r--r--

8011102: Clear AVX registers after return from JNI call
Summary: Execute vzeroupper instruction after JNI call and on exits in jit compiled code which use 256bit vectors.
Reviewed-by: roland

duke@435	1	/*
coleenp@4037	2	* Copyright (c) 2003, 2012, Oracle and/or its affiliates. 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	*
trims@1907	19	* Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
trims@1907	20	* or visit www.oracle.com if you need additional information or have any
trims@1907	21	* questions.
duke@435	22	*
duke@435	23	*/
duke@435	24
stefank@2314	25	#include "precompiled.hpp"
twisti@4318	26	#include "asm/macroAssembler.hpp"
twisti@4318	27	#include "asm/macroAssembler.inline.hpp"
stefank@2314	28	#include "interpreter/interpreter.hpp"
stefank@2314	29	#include "nativeInst_x86.hpp"
stefank@2314	30	#include "oops/instanceOop.hpp"
coleenp@4037	31	#include "oops/method.hpp"
stefank@2314	32	#include "oops/objArrayKlass.hpp"
stefank@2314	33	#include "oops/oop.inline.hpp"
stefank@2314	34	#include "prims/methodHandles.hpp"
stefank@2314	35	#include "runtime/frame.inline.hpp"
stefank@2314	36	#include "runtime/handles.inline.hpp"
stefank@2314	37	#include "runtime/sharedRuntime.hpp"
stefank@2314	38	#include "runtime/stubCodeGenerator.hpp"
stefank@2314	39	#include "runtime/stubRoutines.hpp"
stefank@4299	40	#include "runtime/thread.inline.hpp"
stefank@2314	41	#include "utilities/top.hpp"
stefank@2314	42	#ifdef COMPILER2
stefank@2314	43	#include "opto/runtime.hpp"
stefank@2314	44	#endif
duke@435	45
duke@435	46	// Declaration and definition of StubGenerator (no .hpp file).
duke@435	47	// For a more detailed description of the stub routine structure
duke@435	48	// see the comment in stubRoutines.hpp
duke@435	49
duke@435	50	#define __ _masm->
coleenp@548	51	#define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8)
never@739	52	#define a__ ((Assembler*)_masm)->
duke@435	53
duke@435	54	#ifdef PRODUCT
duke@435	55	#define BLOCK_COMMENT(str) /* nothing */
duke@435	56	#else
duke@435	57	#define BLOCK_COMMENT(str) __ block_comment(str)
duke@435	58	#endif
duke@435	59
duke@435	60	#define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
duke@435	61	const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
duke@435	62
duke@435	63	// Stub Code definitions
duke@435	64
duke@435	65	static address handle_unsafe_access() {
duke@435	66	JavaThread* thread = JavaThread::current();
duke@435	67	address pc = thread->saved_exception_pc();
duke@435	68	// pc is the instruction which we must emulate
duke@435	69	// doing a no-op is fine: return garbage from the load
duke@435	70	// therefore, compute npc
duke@435	71	address npc = Assembler::locate_next_instruction(pc);
duke@435	72
duke@435	73	// request an async exception
duke@435	74	thread->set_pending_unsafe_access_error();
duke@435	75
duke@435	76	// return address of next instruction to execute
duke@435	77	return npc;
duke@435	78	}
duke@435	79
duke@435	80	class StubGenerator: public StubCodeGenerator {
duke@435	81	private:
duke@435	82
duke@435	83	#ifdef PRODUCT
duke@435	84	#define inc_counter_np(counter) (0)
duke@435	85	#else
duke@435	86	void inc_counter_np_(int& counter) {
never@3314	87	// This can destroy rscratch1 if counter is far from the code cache
duke@435	88	__ incrementl(ExternalAddress((address)&counter));
duke@435	89	}
duke@435	90	#define inc_counter_np(counter) \
duke@435	91	BLOCK_COMMENT("inc_counter " #counter); \
duke@435	92	inc_counter_np_(counter);
duke@435	93	#endif
duke@435	94
duke@435	95	// Call stubs are used to call Java from C
duke@435	96	//
duke@435	97	// Linux Arguments:
duke@435	98	// c_rarg0: call wrapper address address
duke@435	99	// c_rarg1: result address
duke@435	100	// c_rarg2: result type BasicType
coleenp@4037	101	// c_rarg3: method Method*
duke@435	102	// c_rarg4: (interpreter) entry point address
duke@435	103	// c_rarg5: parameters intptr_t*
duke@435	104	// 16(rbp): parameter size (in words) int
duke@435	105	// 24(rbp): thread Thread*
duke@435	106	//
duke@435	107	// [ return_from_Java ] <--- rsp
duke@435	108	// [ argument word n ]
duke@435	109	// ...
duke@435	110	// -12 [ argument word 1 ]
duke@435	111	// -11 [ saved r15 ] <--- rsp_after_call
duke@435	112	// -10 [ saved r14 ]
duke@435	113	// -9 [ saved r13 ]
duke@435	114	// -8 [ saved r12 ]
duke@435	115	// -7 [ saved rbx ]
duke@435	116	// -6 [ call wrapper ]
duke@435	117	// -5 [ result ]
duke@435	118	// -4 [ result type ]
duke@435	119	// -3 [ method ]
duke@435	120	// -2 [ entry point ]
duke@435	121	// -1 [ parameters ]
duke@435	122	// 0 [ saved rbp ] <--- rbp
duke@435	123	// 1 [ return address ]
duke@435	124	// 2 [ parameter size ]
duke@435	125	// 3 [ thread ]
duke@435	126	//
duke@435	127	// Windows Arguments:
duke@435	128	// c_rarg0: call wrapper address address
duke@435	129	// c_rarg1: result address
duke@435	130	// c_rarg2: result type BasicType
coleenp@4037	131	// c_rarg3: method Method*
duke@435	132	// 48(rbp): (interpreter) entry point address
duke@435	133	// 56(rbp): parameters intptr_t*
duke@435	134	// 64(rbp): parameter size (in words) int
duke@435	135	// 72(rbp): thread Thread*
duke@435	136	//
duke@435	137	// [ return_from_Java ] <--- rsp
duke@435	138	// [ argument word n ]
duke@435	139	// ...
iveresov@2689	140	// -28 [ argument word 1 ]
iveresov@2689	141	// -27 [ saved xmm15 ] <--- rsp_after_call
iveresov@2689	142	// [ saved xmm7-xmm14 ]
iveresov@2689	143	// -9 [ saved xmm6 ] (each xmm register takes 2 slots)
iveresov@2689	144	// -7 [ saved r15 ]
duke@435	145	// -6 [ saved r14 ]
duke@435	146	// -5 [ saved r13 ]
duke@435	147	// -4 [ saved r12 ]
duke@435	148	// -3 [ saved rdi ]
duke@435	149	// -2 [ saved rsi ]
duke@435	150	// -1 [ saved rbx ]
duke@435	151	// 0 [ saved rbp ] <--- rbp
duke@435	152	// 1 [ return address ]
duke@435	153	// 2 [ call wrapper ]
duke@435	154	// 3 [ result ]
duke@435	155	// 4 [ result type ]
duke@435	156	// 5 [ method ]
duke@435	157	// 6 [ entry point ]
duke@435	158	// 7 [ parameters ]
duke@435	159	// 8 [ parameter size ]
duke@435	160	// 9 [ thread ]
duke@435	161	//
duke@435	162	// Windows reserves the callers stack space for arguments 1-4.
duke@435	163	// We spill c_rarg0-c_rarg3 to this space.
duke@435	164
duke@435	165	// Call stub stack layout word offsets from rbp
duke@435	166	enum call_stub_layout {
duke@435	167	#ifdef _WIN64
iveresov@2689	168	xmm_save_first = 6, // save from xmm6
iveresov@2689	169	xmm_save_last = 15, // to xmm15
iveresov@2689	170	xmm_save_base = -9,
iveresov@2689	171	rsp_after_call_off = xmm_save_base - 2 * (xmm_save_last - xmm_save_first), // -27
iveresov@2689	172	r15_off = -7,
duke@435	173	r14_off = -6,
duke@435	174	r13_off = -5,
duke@435	175	r12_off = -4,
duke@435	176	rdi_off = -3,
duke@435	177	rsi_off = -2,
duke@435	178	rbx_off = -1,
duke@435	179	rbp_off = 0,
duke@435	180	retaddr_off = 1,
duke@435	181	call_wrapper_off = 2,
duke@435	182	result_off = 3,
duke@435	183	result_type_off = 4,
duke@435	184	method_off = 5,
duke@435	185	entry_point_off = 6,
duke@435	186	parameters_off = 7,
duke@435	187	parameter_size_off = 8,
duke@435	188	thread_off = 9
duke@435	189	#else
duke@435	190	rsp_after_call_off = -12,
duke@435	191	mxcsr_off = rsp_after_call_off,
duke@435	192	r15_off = -11,
duke@435	193	r14_off = -10,
duke@435	194	r13_off = -9,
duke@435	195	r12_off = -8,
duke@435	196	rbx_off = -7,
duke@435	197	call_wrapper_off = -6,
duke@435	198	result_off = -5,
duke@435	199	result_type_off = -4,
duke@435	200	method_off = -3,
duke@435	201	entry_point_off = -2,
duke@435	202	parameters_off = -1,
duke@435	203	rbp_off = 0,
duke@435	204	retaddr_off = 1,
duke@435	205	parameter_size_off = 2,
duke@435	206	thread_off = 3
duke@435	207	#endif
duke@435	208	};
duke@435	209
iveresov@2689	210	#ifdef _WIN64
iveresov@2689	211	Address xmm_save(int reg) {
iveresov@2689	212	assert(reg >= xmm_save_first && reg <= xmm_save_last, "XMM register number out of range");
iveresov@2689	213	return Address(rbp, (xmm_save_base - (reg - xmm_save_first) * 2) * wordSize);
iveresov@2689	214	}
iveresov@2689	215	#endif
iveresov@2689	216
duke@435	217	address generate_call_stub(address& return_address) {
duke@435	218	assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
duke@435	219	(int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
duke@435	220	"adjust this code");
duke@435	221	StubCodeMark mark(this, "StubRoutines", "call_stub");
duke@435	222	address start = __ pc();
duke@435	223
duke@435	224	// same as in generate_catch_exception()!
duke@435	225	const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
duke@435	226
duke@435	227	const Address call_wrapper (rbp, call_wrapper_off * wordSize);
duke@435	228	const Address result (rbp, result_off * wordSize);
duke@435	229	const Address result_type (rbp, result_type_off * wordSize);
duke@435	230	const Address method (rbp, method_off * wordSize);
duke@435	231	const Address entry_point (rbp, entry_point_off * wordSize);
duke@435	232	const Address parameters (rbp, parameters_off * wordSize);
duke@435	233	const Address parameter_size(rbp, parameter_size_off * wordSize);
duke@435	234
duke@435	235	// same as in generate_catch_exception()!
duke@435	236	const Address thread (rbp, thread_off * wordSize);
duke@435	237
duke@435	238	const Address r15_save(rbp, r15_off * wordSize);
duke@435	239	const Address r14_save(rbp, r14_off * wordSize);
duke@435	240	const Address r13_save(rbp, r13_off * wordSize);
duke@435	241	const Address r12_save(rbp, r12_off * wordSize);
duke@435	242	const Address rbx_save(rbp, rbx_off * wordSize);
duke@435	243
duke@435	244	// stub code
duke@435	245	__ enter();
never@739	246	__ subptr(rsp, -rsp_after_call_off * wordSize);
duke@435	247
duke@435	248	// save register parameters
duke@435	249	#ifndef _WIN64
never@739	250	__ movptr(parameters, c_rarg5); // parameters
never@739	251	__ movptr(entry_point, c_rarg4); // entry_point
duke@435	252	#endif
duke@435	253
never@739	254	__ movptr(method, c_rarg3); // method
never@739	255	__ movl(result_type, c_rarg2); // result type
never@739	256	__ movptr(result, c_rarg1); // result
never@739	257	__ movptr(call_wrapper, c_rarg0); // call wrapper
duke@435	258
duke@435	259	// save regs belonging to calling function
never@739	260	__ movptr(rbx_save, rbx);
never@739	261	__ movptr(r12_save, r12);
never@739	262	__ movptr(r13_save, r13);
never@739	263	__ movptr(r14_save, r14);
never@739	264	__ movptr(r15_save, r15);
duke@435	265	#ifdef _WIN64
iveresov@2689	266	for (int i = 6; i <= 15; i++) {
iveresov@2689	267	__ movdqu(xmm_save(i), as_XMMRegister(i));
iveresov@2689	268	}
iveresov@2689	269
duke@435	270	const Address rdi_save(rbp, rdi_off * wordSize);
duke@435	271	const Address rsi_save(rbp, rsi_off * wordSize);
duke@435	272
never@739	273	__ movptr(rsi_save, rsi);
never@739	274	__ movptr(rdi_save, rdi);
duke@435	275	#else
duke@435	276	const Address mxcsr_save(rbp, mxcsr_off * wordSize);
duke@435	277	{
duke@435	278	Label skip_ldmx;
duke@435	279	__ stmxcsr(mxcsr_save);
duke@435	280	__ movl(rax, mxcsr_save);
duke@435	281	__ andl(rax, MXCSR_MASK); // Only check control and mask bits
never@739	282	ExternalAddress mxcsr_std(StubRoutines::x86::mxcsr_std());
duke@435	283	__ cmp32(rax, mxcsr_std);
duke@435	284	__ jcc(Assembler::equal, skip_ldmx);
duke@435	285	__ ldmxcsr(mxcsr_std);
duke@435	286	__ bind(skip_ldmx);
duke@435	287	}
duke@435	288	#endif
duke@435	289
duke@435	290	// Load up thread register
never@739	291	__ movptr(r15_thread, thread);
coleenp@548	292	__ reinit_heapbase();
duke@435	293
duke@435	294	#ifdef ASSERT
duke@435	295	// make sure we have no pending exceptions
duke@435	296	{
duke@435	297	Label L;
never@739	298	__ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
duke@435	299	__ jcc(Assembler::equal, L);
duke@435	300	__ stop("StubRoutines::call_stub: entered with pending exception");
duke@435	301	__ bind(L);
duke@435	302	}
duke@435	303	#endif
duke@435	304
duke@435	305	// pass parameters if any
duke@435	306	BLOCK_COMMENT("pass parameters if any");
duke@435	307	Label parameters_done;
duke@435	308	__ movl(c_rarg3, parameter_size);
duke@435	309	__ testl(c_rarg3, c_rarg3);
duke@435	310	__ jcc(Assembler::zero, parameters_done);
duke@435	311
duke@435	312	Label loop;
never@739	313	__ movptr(c_rarg2, parameters); // parameter pointer
never@739	314	__ movl(c_rarg1, c_rarg3); // parameter counter is in c_rarg1
duke@435	315	__ BIND(loop);
never@739	316	__ movptr(rax, Address(c_rarg2, 0));// get parameter
never@739	317	__ addptr(c_rarg2, wordSize); // advance to next parameter
never@739	318	__ decrementl(c_rarg1); // decrement counter
never@739	319	__ push(rax); // pass parameter
duke@435	320	__ jcc(Assembler::notZero, loop);
duke@435	321
duke@435	322	// call Java function
duke@435	323	__ BIND(parameters_done);
coleenp@4037	324	__ movptr(rbx, method); // get Method*
never@739	325	__ movptr(c_rarg1, entry_point); // get entry_point
never@739	326	__ mov(r13, rsp); // set sender sp
duke@435	327	BLOCK_COMMENT("call Java function");
duke@435	328	__ call(c_rarg1);
duke@435	329
duke@435	330	BLOCK_COMMENT("call_stub_return_address:");
duke@435	331	return_address = __ pc();
duke@435	332
duke@435	333	// store result depending on type (everything that is not
duke@435	334	// T_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
never@739	335	__ movptr(c_rarg0, result);
duke@435	336	Label is_long, is_float, is_double, exit;
duke@435	337	__ movl(c_rarg1, result_type);
duke@435	338	__ cmpl(c_rarg1, T_OBJECT);
duke@435	339	__ jcc(Assembler::equal, is_long);
duke@435	340	__ cmpl(c_rarg1, T_LONG);
duke@435	341	__ jcc(Assembler::equal, is_long);
duke@435	342	__ cmpl(c_rarg1, T_FLOAT);
duke@435	343	__ jcc(Assembler::equal, is_float);
duke@435	344	__ cmpl(c_rarg1, T_DOUBLE);
duke@435	345	__ jcc(Assembler::equal, is_double);
duke@435	346
duke@435	347	// handle T_INT case
duke@435	348	__ movl(Address(c_rarg0, 0), rax);
duke@435	349
duke@435	350	__ BIND(exit);
duke@435	351
duke@435	352	// pop parameters
never@739	353	__ lea(rsp, rsp_after_call);
duke@435	354
duke@435	355	#ifdef ASSERT
duke@435	356	// verify that threads correspond
duke@435	357	{
duke@435	358	Label L, S;
never@739	359	__ cmpptr(r15_thread, thread);
duke@435	360	__ jcc(Assembler::notEqual, S);
duke@435	361	__ get_thread(rbx);
never@739	362	__ cmpptr(r15_thread, rbx);
duke@435	363	__ jcc(Assembler::equal, L);
duke@435	364	__ bind(S);
duke@435	365	__ jcc(Assembler::equal, L);
duke@435	366	__ stop("StubRoutines::call_stub: threads must correspond");
duke@435	367	__ bind(L);
duke@435	368	}
duke@435	369	#endif
duke@435	370
duke@435	371	// restore regs belonging to calling function
iveresov@2689	372	#ifdef _WIN64
iveresov@2689	373	for (int i = 15; i >= 6; i--) {
iveresov@2689	374	__ movdqu(as_XMMRegister(i), xmm_save(i));
iveresov@2689	375	}
iveresov@2689	376	#endif
never@739	377	__ movptr(r15, r15_save);
never@739	378	__ movptr(r14, r14_save);
never@739	379	__ movptr(r13, r13_save);
never@739	380	__ movptr(r12, r12_save);
never@739	381	__ movptr(rbx, rbx_save);
duke@435	382
duke@435	383	#ifdef _WIN64
never@739	384	__ movptr(rdi, rdi_save);
never@739	385	__ movptr(rsi, rsi_save);
duke@435	386	#else
duke@435	387	__ ldmxcsr(mxcsr_save);
duke@435	388	#endif
duke@435	389
duke@435	390	// restore rsp
never@739	391	__ addptr(rsp, -rsp_after_call_off * wordSize);
duke@435	392
duke@435	393	// return
never@739	394	__ pop(rbp);
duke@435	395	__ ret(0);
duke@435	396
duke@435	397	// handle return types different from T_INT
duke@435	398	__ BIND(is_long);
duke@435	399	__ movq(Address(c_rarg0, 0), rax);
duke@435	400	__ jmp(exit);
duke@435	401
duke@435	402	__ BIND(is_float);
duke@435	403	__ movflt(Address(c_rarg0, 0), xmm0);
duke@435	404	__ jmp(exit);
duke@435	405
duke@435	406	__ BIND(is_double);
duke@435	407	__ movdbl(Address(c_rarg0, 0), xmm0);
duke@435	408	__ jmp(exit);
duke@435	409
duke@435	410	return start;
duke@435	411	}
duke@435	412
duke@435	413	// Return point for a Java call if there's an exception thrown in
duke@435	414	// Java code. The exception is caught and transformed into a
duke@435	415	// pending exception stored in JavaThread that can be tested from
duke@435	416	// within the VM.
duke@435	417	//
duke@435	418	// Note: Usually the parameters are removed by the callee. In case
duke@435	419	// of an exception crossing an activation frame boundary, that is
duke@435	420	// not the case if the callee is compiled code => need to setup the
duke@435	421	// rsp.
duke@435	422	//
duke@435	423	// rax: exception oop
duke@435	424
duke@435	425	address generate_catch_exception() {
duke@435	426	StubCodeMark mark(this, "StubRoutines", "catch_exception");
duke@435	427	address start = __ pc();
duke@435	428
duke@435	429	// same as in generate_call_stub():
duke@435	430	const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
duke@435	431	const Address thread (rbp, thread_off * wordSize);
duke@435	432
duke@435	433	#ifdef ASSERT
duke@435	434	// verify that threads correspond
duke@435	435	{
duke@435	436	Label L, S;
never@739	437	__ cmpptr(r15_thread, thread);
duke@435	438	__ jcc(Assembler::notEqual, S);
duke@435	439	__ get_thread(rbx);
never@739	440	__ cmpptr(r15_thread, rbx);
duke@435	441	__ jcc(Assembler::equal, L);
duke@435	442	__ bind(S);
duke@435	443	__ stop("StubRoutines::catch_exception: threads must correspond");
duke@435	444	__ bind(L);
duke@435	445	}
duke@435	446	#endif
duke@435	447
duke@435	448	// set pending exception
duke@435	449	__ verify_oop(rax);
duke@435	450
never@739	451	__ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
duke@435	452	__ lea(rscratch1, ExternalAddress((address)__FILE__));
never@739	453	__ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
duke@435	454	__ movl(Address(r15_thread, Thread::exception_line_offset()), (int) __LINE__);
duke@435	455
duke@435	456	// complete return to VM
duke@435	457	assert(StubRoutines::_call_stub_return_address != NULL,
duke@435	458	"_call_stub_return_address must have been generated before");
duke@435	459	__ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
duke@435	460
duke@435	461	return start;
duke@435	462	}
duke@435	463
duke@435	464	// Continuation point for runtime calls returning with a pending
duke@435	465	// exception. The pending exception check happened in the runtime
duke@435	466	// or native call stub. The pending exception in Thread is
duke@435	467	// converted into a Java-level exception.
duke@435	468	//
duke@435	469	// Contract with Java-level exception handlers:
duke@435	470	// rax: exception
duke@435	471	// rdx: throwing pc
duke@435	472	//
duke@435	473	// NOTE: At entry of this stub, exception-pc must be on stack !!
duke@435	474
duke@435	475	address generate_forward_exception() {
duke@435	476	StubCodeMark mark(this, "StubRoutines", "forward exception");
duke@435	477	address start = __ pc();
duke@435	478
duke@435	479	// Upon entry, the sp points to the return address returning into
duke@435	480	// Java (interpreted or compiled) code; i.e., the return address
duke@435	481	// becomes the throwing pc.
duke@435	482	//
duke@435	483	// Arguments pushed before the runtime call are still on the stack
duke@435	484	// but the exception handler will reset the stack pointer ->
duke@435	485	// ignore them. A potential result in registers can be ignored as
duke@435	486	// well.
duke@435	487
duke@435	488	#ifdef ASSERT
duke@435	489	// make sure this code is only executed if there is a pending exception
duke@435	490	{
duke@435	491	Label L;
never@739	492	__ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL);
duke@435	493	__ jcc(Assembler::notEqual, L);
duke@435	494	__ stop("StubRoutines::forward exception: no pending exception (1)");
duke@435	495	__ bind(L);
duke@435	496	}
duke@435	497	#endif
duke@435	498
duke@435	499	// compute exception handler into rbx
never@739	500	__ movptr(c_rarg0, Address(rsp, 0));
duke@435	501	BLOCK_COMMENT("call exception_handler_for_return_address");
duke@435	502	__ call_VM_leaf(CAST_FROM_FN_PTR(address,
duke@435	503	SharedRuntime::exception_handler_for_return_address),
twisti@1730	504	r15_thread, c_rarg0);
never@739	505	__ mov(rbx, rax);
duke@435	506
duke@435	507	// setup rax & rdx, remove return address & clear pending exception
never@739	508	__ pop(rdx);
never@739	509	__ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
xlu@947	510	__ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
duke@435	511
duke@435	512	#ifdef ASSERT
duke@435	513	// make sure exception is set
duke@435	514	{
duke@435	515	Label L;
never@739	516	__ testptr(rax, rax);
duke@435	517	__ jcc(Assembler::notEqual, L);
duke@435	518	__ stop("StubRoutines::forward exception: no pending exception (2)");
duke@435	519	__ bind(L);
duke@435	520	}
duke@435	521	#endif
duke@435	522
duke@435	523	// continue at exception handler (return address removed)
duke@435	524	// rax: exception
duke@435	525	// rbx: exception handler
duke@435	526	// rdx: throwing pc
duke@435	527	__ verify_oop(rax);
duke@435	528	__ jmp(rbx);
duke@435	529
duke@435	530	return start;
duke@435	531	}
duke@435	532
duke@435	533	// Support for jint atomic::xchg(jint exchange_value, volatile jint* dest)
duke@435	534	//
duke@435	535	// Arguments :
duke@435	536	// c_rarg0: exchange_value
duke@435	537	// c_rarg0: dest
duke@435	538	//
duke@435	539	// Result:
duke@435	540	// *dest <- ex, return (orig *dest)
duke@435	541	address generate_atomic_xchg() {
duke@435	542	StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
duke@435	543	address start = __ pc();
duke@435	544
duke@435	545	__ movl(rax, c_rarg0); // Copy to eax we need a return value anyhow
duke@435	546	__ xchgl(rax, Address(c_rarg1, 0)); // automatic LOCK
duke@435	547	__ ret(0);
duke@435	548
duke@435	549	return start;
duke@435	550	}
duke@435	551
duke@435	552	// Support for intptr_t atomic::xchg_ptr(intptr_t exchange_value, volatile intptr_t* dest)
duke@435	553	//
duke@435	554	// Arguments :
duke@435	555	// c_rarg0: exchange_value
duke@435	556	// c_rarg1: dest
duke@435	557	//
duke@435	558	// Result:
duke@435	559	// *dest <- ex, return (orig *dest)
duke@435	560	address generate_atomic_xchg_ptr() {
duke@435	561	StubCodeMark mark(this, "StubRoutines", "atomic_xchg_ptr");
duke@435	562	address start = __ pc();
duke@435	563
never@739	564	__ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
never@739	565	__ xchgptr(rax, Address(c_rarg1, 0)); // automatic LOCK
duke@435	566	__ ret(0);
duke@435	567
duke@435	568	return start;
duke@435	569	}
duke@435	570
duke@435	571	// Support for jint atomic::atomic_cmpxchg(jint exchange_value, volatile jint* dest,
duke@435	572	// jint compare_value)
duke@435	573	//
duke@435	574	// Arguments :
duke@435	575	// c_rarg0: exchange_value
duke@435	576	// c_rarg1: dest
duke@435	577	// c_rarg2: compare_value
duke@435	578	//
duke@435	579	// Result:
duke@435	580	// if ( compare_value == *dest ) {
duke@435	581	// *dest = exchange_value
duke@435	582	// return compare_value;
duke@435	583	// else
duke@435	584	// return *dest;
duke@435	585	address generate_atomic_cmpxchg() {
duke@435	586	StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg");
duke@435	587	address start = __ pc();
duke@435	588
duke@435	589	__ movl(rax, c_rarg2);
duke@435	590	if ( os::is_MP() ) __ lock();
duke@435	591	__ cmpxchgl(c_rarg0, Address(c_rarg1, 0));
duke@435	592	__ ret(0);
duke@435	593
duke@435	594	return start;
duke@435	595	}
duke@435	596
duke@435	597	// Support for jint atomic::atomic_cmpxchg_long(jlong exchange_value,
duke@435	598	// volatile jlong* dest,
duke@435	599	// jlong compare_value)
duke@435	600	// Arguments :
duke@435	601	// c_rarg0: exchange_value
duke@435	602	// c_rarg1: dest
duke@435	603	// c_rarg2: compare_value
duke@435	604	//
duke@435	605	// Result:
duke@435	606	// if ( compare_value == *dest ) {
duke@435	607	// *dest = exchange_value
duke@435	608	// return compare_value;
duke@435	609	// else
duke@435	610	// return *dest;
duke@435	611	address generate_atomic_cmpxchg_long() {
duke@435	612	StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long");
duke@435	613	address start = __ pc();
duke@435	614
duke@435	615	__ movq(rax, c_rarg2);
duke@435	616	if ( os::is_MP() ) __ lock();
duke@435	617	__ cmpxchgq(c_rarg0, Address(c_rarg1, 0));
duke@435	618	__ ret(0);
duke@435	619
duke@435	620	return start;
duke@435	621	}
duke@435	622
duke@435	623	// Support for jint atomic::add(jint add_value, volatile jint* dest)
duke@435	624	//
duke@435	625	// Arguments :
duke@435	626	// c_rarg0: add_value
duke@435	627	// c_rarg1: dest
duke@435	628	//
duke@435	629	// Result:
duke@435	630	// *dest += add_value
duke@435	631	// return *dest;
duke@435	632	address generate_atomic_add() {
duke@435	633	StubCodeMark mark(this, "StubRoutines", "atomic_add");
duke@435	634	address start = __ pc();
duke@435	635
duke@435	636	__ movl(rax, c_rarg0);
duke@435	637	if ( os::is_MP() ) __ lock();
duke@435	638	__ xaddl(Address(c_rarg1, 0), c_rarg0);
duke@435	639	__ addl(rax, c_rarg0);
duke@435	640	__ ret(0);
duke@435	641
duke@435	642	return start;
duke@435	643	}
duke@435	644
duke@435	645	// Support for intptr_t atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest)
duke@435	646	//
duke@435	647	// Arguments :
duke@435	648	// c_rarg0: add_value
duke@435	649	// c_rarg1: dest
duke@435	650	//
duke@435	651	// Result:
duke@435	652	// *dest += add_value
duke@435	653	// return *dest;
duke@435	654	address generate_atomic_add_ptr() {
duke@435	655	StubCodeMark mark(this, "StubRoutines", "atomic_add_ptr");
duke@435	656	address start = __ pc();
duke@435	657
never@739	658	__ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
duke@435	659	if ( os::is_MP() ) __ lock();
never@739	660	__ xaddptr(Address(c_rarg1, 0), c_rarg0);
never@739	661	__ addptr(rax, c_rarg0);
duke@435	662	__ ret(0);
duke@435	663
duke@435	664	return start;
duke@435	665	}
duke@435	666
duke@435	667	// Support for intptr_t OrderAccess::fence()
duke@435	668	//
duke@435	669	// Arguments :
duke@435	670	//
duke@435	671	// Result:
duke@435	672	address generate_orderaccess_fence() {
duke@435	673	StubCodeMark mark(this, "StubRoutines", "orderaccess_fence");
duke@435	674	address start = __ pc();
never@1106	675	__ membar(Assembler::StoreLoad);
duke@435	676	__ ret(0);
duke@435	677
duke@435	678	return start;
duke@435	679	}
duke@435	680
duke@435	681	// Support for intptr_t get_previous_fp()
duke@435	682	//
duke@435	683	// This routine is used to find the previous frame pointer for the
duke@435	684	// caller (current_frame_guess). This is used as part of debugging
duke@435	685	// ps() is seemingly lost trying to find frames.
duke@435	686	// This code assumes that caller current_frame_guess) has a frame.
duke@435	687	address generate_get_previous_fp() {
duke@435	688	StubCodeMark mark(this, "StubRoutines", "get_previous_fp");
duke@435	689	const Address old_fp(rbp, 0);
duke@435	690	const Address older_fp(rax, 0);
duke@435	691	address start = __ pc();
duke@435	692
duke@435	693	__ enter();
never@739	694	__ movptr(rax, old_fp); // callers fp
never@739	695	__ movptr(rax, older_fp); // the frame for ps()
never@739	696	__ pop(rbp);
duke@435	697	__ ret(0);
duke@435	698
duke@435	699	return start;
duke@435	700	}
duke@435	701
roland@3606	702	// Support for intptr_t get_previous_sp()
roland@3606	703	//
roland@3606	704	// This routine is used to find the previous stack pointer for the
roland@3606	705	// caller.
roland@3606	706	address generate_get_previous_sp() {
roland@3606	707	StubCodeMark mark(this, "StubRoutines", "get_previous_sp");
roland@3606	708	address start = __ pc();
roland@3606	709
roland@3606	710	__ movptr(rax, rsp);
roland@3606	711	__ addptr(rax, 8); // return address is at the top of the stack.
roland@3606	712	__ ret(0);
roland@3606	713
roland@3606	714	return start;
roland@3606	715	}
roland@3606	716
duke@435	717	//----------------------------------------------------------------------------------------------------
duke@435	718	// Support for void verify_mxcsr()
duke@435	719	//
duke@435	720	// This routine is used with -Xcheck:jni to verify that native
duke@435	721	// JNI code does not return to Java code without restoring the
duke@435	722	// MXCSR register to our expected state.
duke@435	723
duke@435	724	address generate_verify_mxcsr() {
duke@435	725	StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
duke@435	726	address start = __ pc();
duke@435	727
duke@435	728	const Address mxcsr_save(rsp, 0);
duke@435	729
duke@435	730	if (CheckJNICalls) {
duke@435	731	Label ok_ret;
never@739	732	__ push(rax);
never@739	733	__ subptr(rsp, wordSize); // allocate a temp location
duke@435	734	__ stmxcsr(mxcsr_save);
duke@435	735	__ movl(rax, mxcsr_save);
duke@435	736	__ andl(rax, MXCSR_MASK); // Only check control and mask bits
never@739	737	__ cmpl(rax, *(int *)(StubRoutines::x86::mxcsr_std()));
duke@435	738	__ jcc(Assembler::equal, ok_ret);
duke@435	739
duke@435	740	__ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
duke@435	741
never@739	742	__ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std()));
duke@435	743
duke@435	744	__ bind(ok_ret);
never@739	745	__ addptr(rsp, wordSize);
never@739	746	__ pop(rax);
duke@435	747	}
duke@435	748
duke@435	749	__ ret(0);
duke@435	750
duke@435	751	return start;
duke@435	752	}
duke@435	753
duke@435	754	address generate_f2i_fixup() {
duke@435	755	StubCodeMark mark(this, "StubRoutines", "f2i_fixup");
duke@435	756	Address inout(rsp, 5 * wordSize); // return address + 4 saves
duke@435	757
duke@435	758	address start = __ pc();
duke@435	759
duke@435	760	Label L;
duke@435	761
never@739	762	__ push(rax);
never@739	763	__ push(c_rarg3);
never@739	764	__ push(c_rarg2);
never@739	765	__ push(c_rarg1);
duke@435	766
duke@435	767	__ movl(rax, 0x7f800000);
duke@435	768	__ xorl(c_rarg3, c_rarg3);
duke@435	769	__ movl(c_rarg2, inout);
duke@435	770	__ movl(c_rarg1, c_rarg2);
duke@435	771	__ andl(c_rarg1, 0x7fffffff);
duke@435	772	__ cmpl(rax, c_rarg1); // NaN? -> 0
duke@435	773	__ jcc(Assembler::negative, L);
duke@435	774	__ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
duke@435	775	__ movl(c_rarg3, 0x80000000);
duke@435	776	__ movl(rax, 0x7fffffff);
duke@435	777	__ cmovl(Assembler::positive, c_rarg3, rax);
duke@435	778
duke@435	779	__ bind(L);
never@739	780	__ movptr(inout, c_rarg3);
never@739	781
never@739	782	__ pop(c_rarg1);
never@739	783	__ pop(c_rarg2);
never@739	784	__ pop(c_rarg3);
never@739	785	__ pop(rax);
duke@435	786
duke@435	787	__ ret(0);
duke@435	788
duke@435	789	return start;
duke@435	790	}
duke@435	791
duke@435	792	address generate_f2l_fixup() {
duke@435	793	StubCodeMark mark(this, "StubRoutines", "f2l_fixup");
duke@435	794	Address inout(rsp, 5 * wordSize); // return address + 4 saves
duke@435	795	address start = __ pc();
duke@435	796
duke@435	797	Label L;
duke@435	798
never@739	799	__ push(rax);
never@739	800	__ push(c_rarg3);
never@739	801	__ push(c_rarg2);
never@739	802	__ push(c_rarg1);
duke@435	803
duke@435	804	__ movl(rax, 0x7f800000);
duke@435	805	__ xorl(c_rarg3, c_rarg3);
duke@435	806	__ movl(c_rarg2, inout);
duke@435	807	__ movl(c_rarg1, c_rarg2);
duke@435	808	__ andl(c_rarg1, 0x7fffffff);
duke@435	809	__ cmpl(rax, c_rarg1); // NaN? -> 0
duke@435	810	__ jcc(Assembler::negative, L);
duke@435	811	__ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
duke@435	812	__ mov64(c_rarg3, 0x8000000000000000);
duke@435	813	__ mov64(rax, 0x7fffffffffffffff);
never@739	814	__ cmov(Assembler::positive, c_rarg3, rax);
duke@435	815
duke@435	816	__ bind(L);
never@739	817	__ movptr(inout, c_rarg3);
never@739	818
never@739	819	__ pop(c_rarg1);
never@739	820	__ pop(c_rarg2);
never@739	821	__ pop(c_rarg3);
never@739	822	__ pop(rax);
duke@435	823
duke@435	824	__ ret(0);
duke@435	825
duke@435	826	return start;
duke@435	827	}
duke@435	828
duke@435	829	address generate_d2i_fixup() {
duke@435	830	StubCodeMark mark(this, "StubRoutines", "d2i_fixup");
duke@435	831	Address inout(rsp, 6 * wordSize); // return address + 5 saves
duke@435	832
duke@435	833	address start = __ pc();
duke@435	834
duke@435	835	Label L;
duke@435	836
never@739	837	__ push(rax);
never@739	838	__ push(c_rarg3);
never@739	839	__ push(c_rarg2);
never@739	840	__ push(c_rarg1);
never@739	841	__ push(c_rarg0);
duke@435	842
duke@435	843	__ movl(rax, 0x7ff00000);
duke@435	844	__ movq(c_rarg2, inout);
duke@435	845	__ movl(c_rarg3, c_rarg2);
never@739	846	__ mov(c_rarg1, c_rarg2);
never@739	847	__ mov(c_rarg0, c_rarg2);
duke@435	848	__ negl(c_rarg3);
never@739	849	__ shrptr(c_rarg1, 0x20);
duke@435	850	__ orl(c_rarg3, c_rarg2);
duke@435	851	__ andl(c_rarg1, 0x7fffffff);
duke@435	852	__ xorl(c_rarg2, c_rarg2);
duke@435	853	__ shrl(c_rarg3, 0x1f);
duke@435	854	__ orl(c_rarg1, c_rarg3);
duke@435	855	__ cmpl(rax, c_rarg1);
duke@435	856	__ jcc(Assembler::negative, L); // NaN -> 0
never@739	857	__ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
duke@435	858	__ movl(c_rarg2, 0x80000000);
duke@435	859	__ movl(rax, 0x7fffffff);
never@739	860	__ cmov(Assembler::positive, c_rarg2, rax);
duke@435	861
duke@435	862	__ bind(L);
never@739	863	__ movptr(inout, c_rarg2);
never@739	864
never@739	865	__ pop(c_rarg0);
never@739	866	__ pop(c_rarg1);
never@739	867	__ pop(c_rarg2);
never@739	868	__ pop(c_rarg3);
never@739	869	__ pop(rax);
duke@435	870
duke@435	871	__ ret(0);
duke@435	872
duke@435	873	return start;
duke@435	874	}
duke@435	875
duke@435	876	address generate_d2l_fixup() {
duke@435	877	StubCodeMark mark(this, "StubRoutines", "d2l_fixup");
duke@435	878	Address inout(rsp, 6 * wordSize); // return address + 5 saves
duke@435	879
duke@435	880	address start = __ pc();
duke@435	881
duke@435	882	Label L;
duke@435	883
never@739	884	__ push(rax);
never@739	885	__ push(c_rarg3);
never@739	886	__ push(c_rarg2);
never@739	887	__ push(c_rarg1);
never@739	888	__ push(c_rarg0);
duke@435	889
duke@435	890	__ movl(rax, 0x7ff00000);
duke@435	891	__ movq(c_rarg2, inout);
duke@435	892	__ movl(c_rarg3, c_rarg2);
never@739	893	__ mov(c_rarg1, c_rarg2);
never@739	894	__ mov(c_rarg0, c_rarg2);
duke@435	895	__ negl(c_rarg3);
never@739	896	__ shrptr(c_rarg1, 0x20);
duke@435	897	__ orl(c_rarg3, c_rarg2);
duke@435	898	__ andl(c_rarg1, 0x7fffffff);
duke@435	899	__ xorl(c_rarg2, c_rarg2);
duke@435	900	__ shrl(c_rarg3, 0x1f);
duke@435	901	__ orl(c_rarg1, c_rarg3);
duke@435	902	__ cmpl(rax, c_rarg1);
duke@435	903	__ jcc(Assembler::negative, L); // NaN -> 0
duke@435	904	__ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
duke@435	905	__ mov64(c_rarg2, 0x8000000000000000);
duke@435	906	__ mov64(rax, 0x7fffffffffffffff);
duke@435	907	__ cmovq(Assembler::positive, c_rarg2, rax);
duke@435	908
duke@435	909	__ bind(L);
duke@435	910	__ movq(inout, c_rarg2);
duke@435	911
never@739	912	__ pop(c_rarg0);
never@739	913	__ pop(c_rarg1);
never@739	914	__ pop(c_rarg2);
never@739	915	__ pop(c_rarg3);
never@739	916	__ pop(rax);
duke@435	917
duke@435	918	__ ret(0);
duke@435	919
duke@435	920	return start;
duke@435	921	}
duke@435	922
duke@435	923	address generate_fp_mask(const char *stub_name, int64_t mask) {
kvn@1800	924	__ align(CodeEntryAlignment);
duke@435	925	StubCodeMark mark(this, "StubRoutines", stub_name);
duke@435	926	address start = __ pc();
duke@435	927
duke@435	928	__ emit_data64( mask, relocInfo::none );
duke@435	929	__ emit_data64( mask, relocInfo::none );
duke@435	930
duke@435	931	return start;
duke@435	932	}
duke@435	933
duke@435	934	// The following routine generates a subroutine to throw an
duke@435	935	// asynchronous UnknownError when an unsafe access gets a fault that
duke@435	936	// could not be reasonably prevented by the programmer. (Example:
duke@435	937	// SIGBUS/OBJERR.)
duke@435	938	address generate_handler_for_unsafe_access() {
duke@435	939	StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
duke@435	940	address start = __ pc();
duke@435	941
never@739	942	__ push(0); // hole for return address-to-be
never@739	943	__ pusha(); // push registers
duke@435	944	Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
duke@435	945
never@3136	946	// FIXME: this probably needs alignment logic
never@3136	947
never@739	948	__ subptr(rsp, frame::arg_reg_save_area_bytes);
duke@435	949	BLOCK_COMMENT("call handle_unsafe_access");
duke@435	950	__ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
never@739	951	__ addptr(rsp, frame::arg_reg_save_area_bytes);
never@739	952
never@739	953	__ movptr(next_pc, rax); // stuff next address
never@739	954	__ popa();
duke@435	955	__ ret(0); // jump to next address
duke@435	956
duke@435	957	return start;
duke@435	958	}
duke@435	959
duke@435	960	// Non-destructive plausibility checks for oops
duke@435	961	//
duke@435	962	// Arguments:
duke@435	963	// all args on stack!
duke@435	964	//
duke@435	965	// Stack after saving c_rarg3:
duke@435	966	// [tos + 0]: saved c_rarg3
duke@435	967	// [tos + 1]: saved c_rarg2
kvn@559	968	// [tos + 2]: saved r12 (several TemplateTable methods use it)
kvn@559	969	// [tos + 3]: saved flags
kvn@559	970	// [tos + 4]: return address
kvn@559	971	// * [tos + 5]: error message (char*)
kvn@559	972	// * [tos + 6]: object to verify (oop)
kvn@559	973	// * [tos + 7]: saved rax - saved by caller and bashed
kvn@1938	974	// * [tos + 8]: saved r10 (rscratch1) - saved by caller
duke@435	975	// * = popped on exit
duke@435	976	address generate_verify_oop() {
duke@435	977	StubCodeMark mark(this, "StubRoutines", "verify_oop");
duke@435	978	address start = __ pc();
duke@435	979
duke@435	980	Label exit, error;
duke@435	981
never@739	982	__ pushf();
duke@435	983	__ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
duke@435	984
never@739	985	__ push(r12);
kvn@559	986
duke@435	987	// save c_rarg2 and c_rarg3
never@739	988	__ push(c_rarg2);
never@739	989	__ push(c_rarg3);
duke@435	990
kvn@559	991	enum {
kvn@559	992	// After previous pushes.
kvn@559	993	oop_to_verify = 6 * wordSize,
kvn@559	994	saved_rax = 7 * wordSize,
kvn@1938	995	saved_r10 = 8 * wordSize,
kvn@559	996
kvn@559	997	// Before the call to MacroAssembler::debug(), see below.
kvn@559	998	return_addr = 16 * wordSize,
kvn@559	999	error_msg = 17 * wordSize
kvn@559	1000	};
kvn@559	1001
duke@435	1002	// get object
never@739	1003	__ movptr(rax, Address(rsp, oop_to_verify));
duke@435	1004
duke@435	1005	// make sure object is 'reasonable'
never@739	1006	__ testptr(rax, rax);
duke@435	1007	__ jcc(Assembler::zero, exit); // if obj is NULL it is OK
duke@435	1008	// Check if the oop is in the right area of memory
never@739	1009	__ movptr(c_rarg2, rax);
xlu@947	1010	__ movptr(c_rarg3, (intptr_t) Universe::verify_oop_mask());
never@739	1011	__ andptr(c_rarg2, c_rarg3);
xlu@947	1012	__ movptr(c_rarg3, (intptr_t) Universe::verify_oop_bits());
never@739	1013	__ cmpptr(c_rarg2, c_rarg3);
duke@435	1014	__ jcc(Assembler::notZero, error);
duke@435	1015
kvn@559	1016	// set r12 to heapbase for load_klass()
kvn@559	1017	__ reinit_heapbase();
kvn@559	1018
coleenp@4037	1019	// make sure klass is 'reasonable', which is not zero.
coleenp@548	1020	__ load_klass(rax, rax); // get klass
never@739	1021	__ testptr(rax, rax);
duke@435	1022	__ jcc(Assembler::zero, error); // if klass is NULL it is broken
coleenp@4037	1023	// TODO: Future assert that klass is lower 4g memory for UseCompressedKlassPointers
duke@435	1024
duke@435	1025	// return if everything seems ok
duke@435	1026	__ bind(exit);
never@739	1027	__ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
kvn@1938	1028	__ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
never@739	1029	__ pop(c_rarg3); // restore c_rarg3
never@739	1030	__ pop(c_rarg2); // restore c_rarg2
never@739	1031	__ pop(r12); // restore r12
never@739	1032	__ popf(); // restore flags
kvn@1938	1033	__ ret(4 * wordSize); // pop caller saved stuff
duke@435	1034
duke@435	1035	// handle errors
duke@435	1036	__ bind(error);
never@739	1037	__ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
kvn@1938	1038	__ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
never@739	1039	__ pop(c_rarg3); // get saved c_rarg3 back
never@739	1040	__ pop(c_rarg2); // get saved c_rarg2 back
never@739	1041	__ pop(r12); // get saved r12 back
never@739	1042	__ popf(); // get saved flags off stack --
duke@435	1043	// will be ignored
duke@435	1044
never@739	1045	__ pusha(); // push registers
duke@435	1046	// (rip is already
duke@435	1047	// already pushed)
kvn@559	1048	// debug(char* msg, int64_t pc, int64_t regs[])
duke@435	1049	// We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
duke@435	1050	// pushed all the registers, so now the stack looks like:
duke@435	1051	// [tos + 0] 16 saved registers
duke@435	1052	// [tos + 16] return address
kvn@559	1053	// * [tos + 17] error message (char*)
kvn@559	1054	// * [tos + 18] object to verify (oop)
kvn@559	1055	// * [tos + 19] saved rax - saved by caller and bashed
kvn@1938	1056	// * [tos + 20] saved r10 (rscratch1) - saved by caller
kvn@559	1057	// * = popped on exit
kvn@559	1058
never@739	1059	__ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message
never@739	1060	__ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address
never@739	1061	__ movq(c_rarg2, rsp); // pass address of regs on stack
never@739	1062	__ mov(r12, rsp); // remember rsp
never@739	1063	__ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
never@739	1064	__ andptr(rsp, -16); // align stack as required by ABI
duke@435	1065	BLOCK_COMMENT("call MacroAssembler::debug");
never@739	1066	__ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
never@739	1067	__ mov(rsp, r12); // restore rsp
never@739	1068	__ popa(); // pop registers (includes r12)
kvn@1938	1069	__ ret(4 * wordSize); // pop caller saved stuff
duke@435	1070
duke@435	1071	return start;
duke@435	1072	}
duke@435	1073
duke@435	1074	//
duke@435	1075	// Verify that a register contains clean 32-bits positive value
duke@435	1076	// (high 32-bits are 0) so it could be used in 64-bits shifts.
duke@435	1077	//
duke@435	1078	// Input:
duke@435	1079	// Rint - 32-bits value
duke@435	1080	// Rtmp - scratch
duke@435	1081	//
duke@435	1082	void assert_clean_int(Register Rint, Register Rtmp) {
duke@435	1083	#ifdef ASSERT
duke@435	1084	Label L;
duke@435	1085	assert_different_registers(Rtmp, Rint);
duke@435	1086	__ movslq(Rtmp, Rint);
duke@435	1087	__ cmpq(Rtmp, Rint);
kvn@559	1088	__ jcc(Assembler::equal, L);
duke@435	1089	__ stop("high 32-bits of int value are not 0");
duke@435	1090	__ bind(L);
duke@435	1091	#endif
duke@435	1092	}
duke@435	1093
duke@435	1094	// Generate overlap test for array copy stubs
duke@435	1095	//
duke@435	1096	// Input:
duke@435	1097	// c_rarg0 - from
duke@435	1098	// c_rarg1 - to
duke@435	1099	// c_rarg2 - element count
duke@435	1100	//
duke@435	1101	// Output:
duke@435	1102	// rax - &from[element count - 1]
duke@435	1103	//
duke@435	1104	void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) {
duke@435	1105	assert(no_overlap_target != NULL, "must be generated");
duke@435	1106	array_overlap_test(no_overlap_target, NULL, sf);
duke@435	1107	}
duke@435	1108	void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) {
duke@435	1109	array_overlap_test(NULL, &L_no_overlap, sf);
duke@435	1110	}
duke@435	1111	void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) {
duke@435	1112	const Register from = c_rarg0;
duke@435	1113	const Register to = c_rarg1;
duke@435	1114	const Register count = c_rarg2;
duke@435	1115	const Register end_from = rax;
duke@435	1116
never@739	1117	__ cmpptr(to, from);
never@739	1118	__ lea(end_from, Address(from, count, sf, 0));
duke@435	1119	if (NOLp == NULL) {
duke@435	1120	ExternalAddress no_overlap(no_overlap_target);
duke@435	1121	__ jump_cc(Assembler::belowEqual, no_overlap);
never@739	1122	__ cmpptr(to, end_from);
duke@435	1123	__ jump_cc(Assembler::aboveEqual, no_overlap);
duke@435	1124	} else {
duke@435	1125	__ jcc(Assembler::belowEqual, (*NOLp));
never@739	1126	__ cmpptr(to, end_from);
duke@435	1127	__ jcc(Assembler::aboveEqual, (*NOLp));
duke@435	1128	}
duke@435	1129	}
duke@435	1130
duke@435	1131	// Shuffle first three arg regs on Windows into Linux/Solaris locations.
duke@435	1132	//
duke@435	1133	// Outputs:
duke@435	1134	// rdi - rcx
duke@435	1135	// rsi - rdx
duke@435	1136	// rdx - r8
duke@435	1137	// rcx - r9
duke@435	1138	//
duke@435	1139	// Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
duke@435	1140	// are non-volatile. r9 and r10 should not be used by the caller.
duke@435	1141	//
duke@435	1142	void setup_arg_regs(int nargs = 3) {
duke@435	1143	const Register saved_rdi = r9;
duke@435	1144	const Register saved_rsi = r10;
duke@435	1145	assert(nargs == 3 || nargs == 4, "else fix");
duke@435	1146	#ifdef _WIN64
duke@435	1147	assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
duke@435	1148	"unexpected argument registers");
duke@435	1149	if (nargs >= 4)
never@739	1150	__ mov(rax, r9); // r9 is also saved_rdi
never@739	1151	__ movptr(saved_rdi, rdi);
never@739	1152	__ movptr(saved_rsi, rsi);
never@739	1153	__ mov(rdi, rcx); // c_rarg0
never@739	1154	__ mov(rsi, rdx); // c_rarg1
never@739	1155	__ mov(rdx, r8); // c_rarg2
duke@435	1156	if (nargs >= 4)
never@739	1157	__ mov(rcx, rax); // c_rarg3 (via rax)
duke@435	1158	#else
duke@435	1159	assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
duke@435	1160	"unexpected argument registers");
duke@435	1161	#endif
duke@435	1162	}
duke@435	1163
duke@435	1164	void restore_arg_regs() {
duke@435	1165	const Register saved_rdi = r9;
duke@435	1166	const Register saved_rsi = r10;
duke@435	1167	#ifdef _WIN64
never@739	1168	__ movptr(rdi, saved_rdi);
never@739	1169	__ movptr(rsi, saved_rsi);
duke@435	1170	#endif
duke@435	1171	}
duke@435	1172
duke@435	1173	// Generate code for an array write pre barrier
duke@435	1174	//
duke@435	1175	// addr - starting address
iveresov@2606	1176	// count - element count
iveresov@2606	1177	// tmp - scratch register
duke@435	1178	//
duke@435	1179	// Destroy no registers!
duke@435	1180	//
iveresov@2606	1181	void gen_write_ref_array_pre_barrier(Register addr, Register count, bool dest_uninitialized) {
duke@435	1182	BarrierSet* bs = Universe::heap()->barrier_set();
duke@435	1183	switch (bs->kind()) {
duke@435	1184	case BarrierSet::G1SATBCT:
duke@435	1185	case BarrierSet::G1SATBCTLogging:
iveresov@2606	1186	// With G1, don't generate the call if we statically know that the target in uninitialized
iveresov@2606	1187	if (!dest_uninitialized) {
iveresov@2606	1188	__ pusha(); // push registers
iveresov@2606	1189	if (count == c_rarg0) {
iveresov@2606	1190	if (addr == c_rarg1) {
iveresov@2606	1191	// exactly backwards!!
iveresov@2606	1192	__ xchgptr(c_rarg1, c_rarg0);
iveresov@2606	1193	} else {
iveresov@2606	1194	__ movptr(c_rarg1, count);
iveresov@2606	1195	__ movptr(c_rarg0, addr);
iveresov@2606	1196	}
iveresov@2606	1197	} else {
iveresov@2606	1198	__ movptr(c_rarg0, addr);
iveresov@2606	1199	__ movptr(c_rarg1, count);
iveresov@2606	1200	}
iveresov@2606	1201	__ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre), 2);
iveresov@2606	1202	__ popa();
duke@435	1203	}
iveresov@2606	1204	break;
duke@435	1205	case BarrierSet::CardTableModRef:
duke@435	1206	case BarrierSet::CardTableExtension:
duke@435	1207	case BarrierSet::ModRef:
duke@435	1208	break;
ysr@777	1209	default:
duke@435	1210	ShouldNotReachHere();
duke@435	1211
duke@435	1212	}
duke@435	1213	}
duke@435	1214
duke@435	1215	//
duke@435	1216	// Generate code for an array write post barrier
duke@435	1217	//
duke@435	1218	// Input:
duke@435	1219	// start - register containing starting address of destination array
duke@435	1220	// end - register containing ending address of destination array
duke@435	1221	// scratch - scratch register
duke@435	1222	//
duke@435	1223	// The input registers are overwritten.
duke@435	1224	// The ending address is inclusive.
duke@435	1225	void gen_write_ref_array_post_barrier(Register start, Register end, Register scratch) {
duke@435	1226	assert_different_registers(start, end, scratch);
duke@435	1227	BarrierSet* bs = Universe::heap()->barrier_set();
duke@435	1228	switch (bs->kind()) {
duke@435	1229	case BarrierSet::G1SATBCT:
duke@435	1230	case BarrierSet::G1SATBCTLogging:
duke@435	1231
duke@435	1232	{
never@739	1233	__ pusha(); // push registers (overkill)
duke@435	1234	// must compute element count unless barrier set interface is changed (other platforms supply count)
duke@435	1235	assert_different_registers(start, end, scratch);
ysr@1280	1236	__ lea(scratch, Address(end, BytesPerHeapOop));
ysr@1280	1237	__ subptr(scratch, start); // subtract start to get #bytes
ysr@1280	1238	__ shrptr(scratch, LogBytesPerHeapOop); // convert to element count
never@739	1239	__ mov(c_rarg0, start);
never@739	1240	__ mov(c_rarg1, scratch);
apetrusenko@1627	1241	__ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post), 2);
never@739	1242	__ popa();
duke@435	1243	}
duke@435	1244	break;
duke@435	1245	case BarrierSet::CardTableModRef:
duke@435	1246	case BarrierSet::CardTableExtension:
duke@435	1247	{
duke@435	1248	CardTableModRefBS* ct = (CardTableModRefBS*)bs;
duke@435	1249	assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
duke@435	1250
duke@435	1251	Label L_loop;
duke@435	1252
never@739	1253	__ shrptr(start, CardTableModRefBS::card_shift);
ysr@1280	1254	__ addptr(end, BytesPerHeapOop);
never@739	1255	__ shrptr(end, CardTableModRefBS::card_shift);
never@739	1256	__ subptr(end, start); // number of bytes to copy
duke@435	1257
never@684	1258	intptr_t disp = (intptr_t) ct->byte_map_base;
twisti@3310	1259	if (Assembler::is_simm32(disp)) {
never@684	1260	Address cardtable(noreg, noreg, Address::no_scale, disp);
never@684	1261	__ lea(scratch, cardtable);
never@684	1262	} else {
never@684	1263	ExternalAddress cardtable((address)disp);
never@684	1264	__ lea(scratch, cardtable);
never@684	1265	}
never@684	1266
duke@435	1267	const Register count = end; // 'end' register contains bytes count now
never@739	1268	__ addptr(start, scratch);
duke@435	1269	__ BIND(L_loop);
duke@435	1270	__ movb(Address(start, count, Address::times_1), 0);
never@739	1271	__ decrement(count);
duke@435	1272	__ jcc(Assembler::greaterEqual, L_loop);
duke@435	1273	}
ysr@777	1274	break;
ysr@777	1275	default:
ysr@777	1276	ShouldNotReachHere();
ysr@777	1277
ysr@777	1278	}
ysr@777	1279	}
duke@435	1280
kvn@840	1281
duke@435	1282	// Copy big chunks forward
duke@435	1283	//
duke@435	1284	// Inputs:
duke@435	1285	// end_from - source arrays end address
duke@435	1286	// end_to - destination array end address
duke@435	1287	// qword_count - 64-bits element count, negative
duke@435	1288	// to - scratch
kvn@4411	1289	// L_copy_bytes - entry label
duke@435	1290	// L_copy_8_bytes - exit label
duke@435	1291	//
kvn@4411	1292	void copy_bytes_forward(Register end_from, Register end_to,
duke@435	1293	Register qword_count, Register to,
kvn@4411	1294	Label& L_copy_bytes, Label& L_copy_8_bytes) {
duke@435	1295	DEBUG_ONLY(__ stop("enter at entry label, not here"));
duke@435	1296	Label L_loop;
kvn@1800	1297	__ align(OptoLoopAlignment);
kvn@4411	1298	if (UseUnalignedLoadStores) {
kvn@4411	1299	Label L_end;
kvn@4411	1300	// Copy 64-bytes per iteration
kvn@4411	1301	__ BIND(L_loop);
kvn@4411	1302	if (UseAVX >= 2) {
kvn@4411	1303	__ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
kvn@4411	1304	__ vmovdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
kvn@4411	1305	__ vmovdqu(xmm1, Address(end_from, qword_count, Address::times_8, -24));
kvn@4411	1306	__ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm1);
kvn@4411	1307	} else {
kvn@4411	1308	__ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
kvn@4411	1309	__ movdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
kvn@4411	1310	__ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, -40));
kvn@4411	1311	__ movdqu(Address(end_to, qword_count, Address::times_8, -40), xmm1);
kvn@4411	1312	__ movdqu(xmm2, Address(end_from, qword_count, Address::times_8, -24));
kvn@4411	1313	__ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm2);
kvn@4411	1314	__ movdqu(xmm3, Address(end_from, qword_count, Address::times_8, - 8));
kvn@4411	1315	__ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm3);
kvn@4411	1316	}
kvn@4411	1317	__ BIND(L_copy_bytes);
kvn@4411	1318	__ addptr(qword_count, 8);
kvn@4411	1319	__ jcc(Assembler::lessEqual, L_loop);
kvn@4411	1320	__ subptr(qword_count, 4); // sub(8) and add(4)
kvn@4411	1321	__ jccb(Assembler::greater, L_end);
kvn@4411	1322	// Copy trailing 32 bytes
kvn@4411	1323	if (UseAVX >= 2) {
kvn@4411	1324	__ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
kvn@4411	1325	__ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
kvn@4411	1326	} else {
kvn@4411	1327	__ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
kvn@4411	1328	__ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
kvn@4411	1329	__ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, - 8));
kvn@4411	1330	__ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm1);
kvn@4411	1331	}
kvn@4411	1332	__ addptr(qword_count, 4);
kvn@4411	1333	__ BIND(L_end);
kvn@4873	1334	if (UseAVX >= 2) {
kvn@4873	1335	// clean upper bits of YMM registers
kvn@4873	1336	__ vzeroupper();
kvn@4873	1337	}
kvn@840	1338	} else {
kvn@4411	1339	// Copy 32-bytes per iteration
kvn@4411	1340	__ BIND(L_loop);
kvn@840	1341	__ movq(to, Address(end_from, qword_count, Address::times_8, -24));
kvn@840	1342	__ movq(Address(end_to, qword_count, Address::times_8, -24), to);
kvn@840	1343	__ movq(to, Address(end_from, qword_count, Address::times_8, -16));
kvn@840	1344	__ movq(Address(end_to, qword_count, Address::times_8, -16), to);
kvn@840	1345	__ movq(to, Address(end_from, qword_count, Address::times_8, - 8));
kvn@840	1346	__ movq(Address(end_to, qword_count, Address::times_8, - 8), to);
kvn@840	1347	__ movq(to, Address(end_from, qword_count, Address::times_8, - 0));
kvn@840	1348	__ movq(Address(end_to, qword_count, Address::times_8, - 0), to);
kvn@4411	1349
kvn@4411	1350	__ BIND(L_copy_bytes);
kvn@4411	1351	__ addptr(qword_count, 4);
kvn@4411	1352	__ jcc(Assembler::lessEqual, L_loop);
kvn@840	1353	}
never@739	1354	__ subptr(qword_count, 4);
duke@435	1355	__ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords
duke@435	1356	}
duke@435	1357
duke@435	1358	// Copy big chunks backward
duke@435	1359	//
duke@435	1360	// Inputs:
duke@435	1361	// from - source arrays address
duke@435	1362	// dest - destination array address
duke@435	1363	// qword_count - 64-bits element count
duke@435	1364	// to - scratch
kvn@4411	1365	// L_copy_bytes - entry label
duke@435	1366	// L_copy_8_bytes - exit label
duke@435	1367	//
kvn@4411	1368	void copy_bytes_backward(Register from, Register dest,
duke@435	1369	Register qword_count, Register to,
kvn@4411	1370	Label& L_copy_bytes, Label& L_copy_8_bytes) {
duke@435	1371	DEBUG_ONLY(__ stop("enter at entry label, not here"));
duke@435	1372	Label L_loop;
kvn@1800	1373	__ align(OptoLoopAlignment);
kvn@4411	1374	if (UseUnalignedLoadStores) {
kvn@4411	1375	Label L_end;
kvn@4411	1376	// Copy 64-bytes per iteration
kvn@4411	1377	__ BIND(L_loop);
kvn@4411	1378	if (UseAVX >= 2) {
kvn@4411	1379	__ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 32));
kvn@4411	1380	__ vmovdqu(Address(dest, qword_count, Address::times_8, 32), xmm0);
kvn@4411	1381	__ vmovdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
kvn@4411	1382	__ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
kvn@4411	1383	} else {
kvn@4411	1384	__ movdqu(xmm0, Address(from, qword_count, Address::times_8, 48));
kvn@4411	1385	__ movdqu(Address(dest, qword_count, Address::times_8, 48), xmm0);
kvn@4411	1386	__ movdqu(xmm1, Address(from, qword_count, Address::times_8, 32));
kvn@4411	1387	__ movdqu(Address(dest, qword_count, Address::times_8, 32), xmm1);
kvn@4411	1388	__ movdqu(xmm2, Address(from, qword_count, Address::times_8, 16));
kvn@4411	1389	__ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm2);
kvn@4411	1390	__ movdqu(xmm3, Address(from, qword_count, Address::times_8, 0));
kvn@4411	1391	__ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm3);
kvn@4411	1392	}
kvn@4411	1393	__ BIND(L_copy_bytes);
kvn@4411	1394	__ subptr(qword_count, 8);
kvn@4411	1395	__ jcc(Assembler::greaterEqual, L_loop);
kvn@4411	1396
kvn@4411	1397	__ addptr(qword_count, 4); // add(8) and sub(4)
kvn@4411	1398	__ jccb(Assembler::less, L_end);
kvn@4411	1399	// Copy trailing 32 bytes
kvn@4411	1400	if (UseAVX >= 2) {
kvn@4411	1401	__ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 0));
kvn@4411	1402	__ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm0);
kvn@4411	1403	} else {
kvn@4411	1404	__ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16));
kvn@4411	1405	__ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm0);
kvn@4411	1406	__ movdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
kvn@4411	1407	__ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
kvn@4411	1408	}
kvn@4411	1409	__ subptr(qword_count, 4);
kvn@4411	1410	__ BIND(L_end);
kvn@4873	1411	if (UseAVX >= 2) {
kvn@4873	1412	// clean upper bits of YMM registers
kvn@4873	1413	__ vzeroupper();
kvn@4873	1414	}
kvn@840	1415	} else {
kvn@4411	1416	// Copy 32-bytes per iteration
kvn@4411	1417	__ BIND(L_loop);
kvn@840	1418	__ movq(to, Address(from, qword_count, Address::times_8, 24));
kvn@840	1419	__ movq(Address(dest, qword_count, Address::times_8, 24), to);
kvn@840	1420	__ movq(to, Address(from, qword_count, Address::times_8, 16));
kvn@840	1421	__ movq(Address(dest, qword_count, Address::times_8, 16), to);
kvn@840	1422	__ movq(to, Address(from, qword_count, Address::times_8, 8));
kvn@840	1423	__ movq(Address(dest, qword_count, Address::times_8, 8), to);
kvn@840	1424	__ movq(to, Address(from, qword_count, Address::times_8, 0));
kvn@840	1425	__ movq(Address(dest, qword_count, Address::times_8, 0), to);
kvn@4411	1426
kvn@4411	1427	__ BIND(L_copy_bytes);
kvn@4411	1428	__ subptr(qword_count, 4);
kvn@4411	1429	__ jcc(Assembler::greaterEqual, L_loop);
kvn@840	1430	}
never@739	1431	__ addptr(qword_count, 4);
duke@435	1432	__ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords
duke@435	1433	}
duke@435	1434
duke@435	1435
duke@435	1436	// Arguments:
duke@435	1437	// aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
duke@435	1438	// ignored
duke@435	1439	// name - stub name string
duke@435	1440	//
duke@435	1441	// Inputs:
duke@435	1442	// c_rarg0 - source array address
duke@435	1443	// c_rarg1 - destination array address
duke@435	1444	// c_rarg2 - element count, treated as ssize_t, can be zero
duke@435	1445	//
duke@435	1446	// If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
duke@435	1447	// we let the hardware handle it. The one to eight bytes within words,
duke@435	1448	// dwords or qwords that span cache line boundaries will still be loaded
duke@435	1449	// and stored atomically.
duke@435	1450	//
duke@435	1451	// Side Effects:
duke@435	1452	// disjoint_byte_copy_entry is set to the no-overlap entry point
duke@435	1453	// used by generate_conjoint_byte_copy().
duke@435	1454	//
iveresov@2595	1455	address generate_disjoint_byte_copy(bool aligned, address* entry, const char *name) {
duke@435	1456	__ align(CodeEntryAlignment);
duke@435	1457	StubCodeMark mark(this, "StubRoutines", name);
duke@435	1458	address start = __ pc();
duke@435	1459
kvn@4411	1460	Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
duke@435	1461	Label L_copy_byte, L_exit;
duke@435	1462	const Register from = rdi; // source array address
duke@435	1463	const Register to = rsi; // destination array address
duke@435	1464	const Register count = rdx; // elements count
duke@435	1465	const Register byte_count = rcx;
duke@435	1466	const Register qword_count = count;
duke@435	1467	const Register end_from = from; // source array end address
duke@435	1468	const Register end_to = to; // destination array end address
duke@435	1469	// End pointers are inclusive, and if count is not zero they point
duke@435	1470	// to the last unit copied: end_to[0] := end_from[0]
duke@435	1471
duke@435	1472	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	1473	assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
duke@435	1474
iveresov@2595	1475	if (entry != NULL) {
iveresov@2595	1476	*entry = __ pc();
iveresov@2595	1477	// caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
iveresov@2595	1478	BLOCK_COMMENT("Entry:");
iveresov@2595	1479	}
duke@435	1480
duke@435	1481	setup_arg_regs(); // from => rdi, to => rsi, count => rdx
duke@435	1482	// r9 and r10 may be used to save non-volatile registers
duke@435	1483
duke@435	1484	// 'from', 'to' and 'count' are now valid
never@739	1485	__ movptr(byte_count, count);
never@739	1486	__ shrptr(count, 3); // count => qword_count
duke@435	1487
duke@435	1488	// Copy from low to high addresses. Use 'to' as scratch.
never@739	1489	__ lea(end_from, Address(from, qword_count, Address::times_8, -8));
never@739	1490	__ lea(end_to, Address(to, qword_count, Address::times_8, -8));
never@739	1491	__ negptr(qword_count); // make the count negative
kvn@4411	1492	__ jmp(L_copy_bytes);
duke@435	1493
duke@435	1494	// Copy trailing qwords
duke@435	1495	__ BIND(L_copy_8_bytes);
duke@435	1496	__ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
duke@435	1497	__ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
never@739	1498	__ increment(qword_count);
duke@435	1499	__ jcc(Assembler::notZero, L_copy_8_bytes);
duke@435	1500
duke@435	1501	// Check for and copy trailing dword
duke@435	1502	__ BIND(L_copy_4_bytes);
never@739	1503	__ testl(byte_count, 4);
duke@435	1504	__ jccb(Assembler::zero, L_copy_2_bytes);
duke@435	1505	__ movl(rax, Address(end_from, 8));
duke@435	1506	__ movl(Address(end_to, 8), rax);
duke@435	1507
never@739	1508	__ addptr(end_from, 4);
never@739	1509	__ addptr(end_to, 4);
duke@435	1510
duke@435	1511	// Check for and copy trailing word
duke@435	1512	__ BIND(L_copy_2_bytes);
never@739	1513	__ testl(byte_count, 2);
duke@435	1514	__ jccb(Assembler::zero, L_copy_byte);
duke@435	1515	__ movw(rax, Address(end_from, 8));
duke@435	1516	__ movw(Address(end_to, 8), rax);
duke@435	1517
never@739	1518	__ addptr(end_from, 2);
never@739	1519	__ addptr(end_to, 2);
duke@435	1520
duke@435	1521	// Check for and copy trailing byte
duke@435	1522	__ BIND(L_copy_byte);
never@739	1523	__ testl(byte_count, 1);
duke@435	1524	__ jccb(Assembler::zero, L_exit);
duke@435	1525	__ movb(rax, Address(end_from, 8));
duke@435	1526	__ movb(Address(end_to, 8), rax);
duke@435	1527
duke@435	1528	__ BIND(L_exit);
duke@435	1529	restore_arg_regs();
never@3314	1530	inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
never@739	1531	__ xorptr(rax, rax); // return 0
duke@435	1532	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	1533	__ ret(0);
duke@435	1534
kvn@4411	1535	// Copy in multi-bytes chunks
kvn@4411	1536	copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
duke@435	1537	__ jmp(L_copy_4_bytes);
duke@435	1538
duke@435	1539	return start;
duke@435	1540	}
duke@435	1541
duke@435	1542	// Arguments:
duke@435	1543	// aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
duke@435	1544	// ignored
duke@435	1545	// name - stub name string
duke@435	1546	//
duke@435	1547	// Inputs:
duke@435	1548	// c_rarg0 - source array address
duke@435	1549	// c_rarg1 - destination array address
duke@435	1550	// c_rarg2 - element count, treated as ssize_t, can be zero
duke@435	1551	//
duke@435	1552	// If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
duke@435	1553	// we let the hardware handle it. The one to eight bytes within words,
duke@435	1554	// dwords or qwords that span cache line boundaries will still be loaded
duke@435	1555	// and stored atomically.
duke@435	1556	//
iveresov@2595	1557	address generate_conjoint_byte_copy(bool aligned, address nooverlap_target,
iveresov@2595	1558	address* entry, const char *name) {
duke@435	1559	__ align(CodeEntryAlignment);
duke@435	1560	StubCodeMark mark(this, "StubRoutines", name);
duke@435	1561	address start = __ pc();
duke@435	1562
kvn@4411	1563	Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
duke@435	1564	const Register from = rdi; // source array address
duke@435	1565	const Register to = rsi; // destination array address
duke@435	1566	const Register count = rdx; // elements count
duke@435	1567	const Register byte_count = rcx;
duke@435	1568	const Register qword_count = count;
duke@435	1569
duke@435	1570	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	1571	assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
duke@435	1572
iveresov@2595	1573	if (entry != NULL) {
iveresov@2595	1574	*entry = __ pc();
iveresov@2595	1575	// caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
iveresov@2595	1576	BLOCK_COMMENT("Entry:");
iveresov@2595	1577	}
iveresov@2595	1578
iveresov@2595	1579	array_overlap_test(nooverlap_target, Address::times_1);
duke@435	1580	setup_arg_regs(); // from => rdi, to => rsi, count => rdx
duke@435	1581	// r9 and r10 may be used to save non-volatile registers
duke@435	1582
duke@435	1583	// 'from', 'to' and 'count' are now valid
never@739	1584	__ movptr(byte_count, count);
never@739	1585	__ shrptr(count, 3); // count => qword_count
duke@435	1586
duke@435	1587	// Copy from high to low addresses.
duke@435	1588
duke@435	1589	// Check for and copy trailing byte
never@739	1590	__ testl(byte_count, 1);
duke@435	1591	__ jcc(Assembler::zero, L_copy_2_bytes);
duke@435	1592	__ movb(rax, Address(from, byte_count, Address::times_1, -1));
duke@435	1593	__ movb(Address(to, byte_count, Address::times_1, -1), rax);
never@739	1594	__ decrement(byte_count); // Adjust for possible trailing word
duke@435	1595
duke@435	1596	// Check for and copy trailing word
duke@435	1597	__ BIND(L_copy_2_bytes);
never@739	1598	__ testl(byte_count, 2);
duke@435	1599	__ jcc(Assembler::zero, L_copy_4_bytes);
duke@435	1600	__ movw(rax, Address(from, byte_count, Address::times_1, -2));
duke@435	1601	__ movw(Address(to, byte_count, Address::times_1, -2), rax);
duke@435	1602
duke@435	1603	// Check for and copy trailing dword
duke@435	1604	__ BIND(L_copy_4_bytes);
never@739	1605	__ testl(byte_count, 4);
kvn@4411	1606	__ jcc(Assembler::zero, L_copy_bytes);
duke@435	1607	__ movl(rax, Address(from, qword_count, Address::times_8));
duke@435	1608	__ movl(Address(to, qword_count, Address::times_8), rax);
kvn@4411	1609	__ jmp(L_copy_bytes);
duke@435	1610
duke@435	1611	// Copy trailing qwords
duke@435	1612	__ BIND(L_copy_8_bytes);
duke@435	1613	__ movq(rax, Address(from, qword_count, Address::times_8, -8));
duke@435	1614	__ movq(Address(to, qword_count, Address::times_8, -8), rax);
never@739	1615	__ decrement(qword_count);
duke@435	1616	__ jcc(Assembler::notZero, L_copy_8_bytes);
duke@435	1617
duke@435	1618	restore_arg_regs();
never@3314	1619	inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
never@739	1620	__ xorptr(rax, rax); // return 0
duke@435	1621	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	1622	__ ret(0);
duke@435	1623
kvn@4411	1624	// Copy in multi-bytes chunks
kvn@4411	1625	copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
duke@435	1626
duke@435	1627	restore_arg_regs();
never@3314	1628	inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
never@739	1629	__ xorptr(rax, rax); // return 0
duke@435	1630	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	1631	__ ret(0);
duke@435	1632
duke@435	1633	return start;
duke@435	1634	}
duke@435	1635
duke@435	1636	// Arguments:
duke@435	1637	// aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
duke@435	1638	// ignored
duke@435	1639	// name - stub name string
duke@435	1640	//
duke@435	1641	// Inputs:
duke@435	1642	// c_rarg0 - source array address
duke@435	1643	// c_rarg1 - destination array address
duke@435	1644	// c_rarg2 - element count, treated as ssize_t, can be zero
duke@435	1645	//
duke@435	1646	// If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
duke@435	1647	// let the hardware handle it. The two or four words within dwords
duke@435	1648	// or qwords that span cache line boundaries will still be loaded
duke@435	1649	// and stored atomically.
duke@435	1650	//
duke@435	1651	// Side Effects:
duke@435	1652	// disjoint_short_copy_entry is set to the no-overlap entry point
duke@435	1653	// used by generate_conjoint_short_copy().
duke@435	1654	//
iveresov@2595	1655	address generate_disjoint_short_copy(bool aligned, address *entry, const char *name) {
duke@435	1656	__ align(CodeEntryAlignment);
duke@435	1657	StubCodeMark mark(this, "StubRoutines", name);
duke@435	1658	address start = __ pc();
duke@435	1659
kvn@4411	1660	Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit;
duke@435	1661	const Register from = rdi; // source array address
duke@435	1662	const Register to = rsi; // destination array address
duke@435	1663	const Register count = rdx; // elements count
duke@435	1664	const Register word_count = rcx;
duke@435	1665	const Register qword_count = count;
duke@435	1666	const Register end_from = from; // source array end address
duke@435	1667	const Register end_to = to; // destination array end address
duke@435	1668	// End pointers are inclusive, and if count is not zero they point
duke@435	1669	// to the last unit copied: end_to[0] := end_from[0]
duke@435	1670
duke@435	1671	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	1672	assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
duke@435	1673
iveresov@2595	1674	if (entry != NULL) {
iveresov@2595	1675	*entry = __ pc();
iveresov@2595	1676	// caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
iveresov@2595	1677	BLOCK_COMMENT("Entry:");
iveresov@2595	1678	}
duke@435	1679
duke@435	1680	setup_arg_regs(); // from => rdi, to => rsi, count => rdx
duke@435	1681	// r9 and r10 may be used to save non-volatile registers
duke@435	1682
duke@435	1683	// 'from', 'to' and 'count' are now valid
never@739	1684	__ movptr(word_count, count);
never@739	1685	__ shrptr(count, 2); // count => qword_count
duke@435	1686
duke@435	1687	// Copy from low to high addresses. Use 'to' as scratch.
never@739	1688	__ lea(end_from, Address(from, qword_count, Address::times_8, -8));
never@739	1689	__ lea(end_to, Address(to, qword_count, Address::times_8, -8));
never@739	1690	__ negptr(qword_count);
kvn@4411	1691	__ jmp(L_copy_bytes);
duke@435	1692
duke@435	1693	// Copy trailing qwords
duke@435	1694	__ BIND(L_copy_8_bytes);
duke@435	1695	__ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
duke@435	1696	__ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
never@739	1697	__ increment(qword_count);
duke@435	1698	__ jcc(Assembler::notZero, L_copy_8_bytes);
duke@435	1699
duke@435	1700	// Original 'dest' is trashed, so we can't use it as a
duke@435	1701	// base register for a possible trailing word copy
duke@435	1702
duke@435	1703	// Check for and copy trailing dword
duke@435	1704	__ BIND(L_copy_4_bytes);
never@739	1705	__ testl(word_count, 2);
duke@435	1706	__ jccb(Assembler::zero, L_copy_2_bytes);
duke@435	1707	__ movl(rax, Address(end_from, 8));
duke@435	1708	__ movl(Address(end_to, 8), rax);
duke@435	1709
never@739	1710	__ addptr(end_from, 4);
never@739	1711	__ addptr(end_to, 4);
duke@435	1712
duke@435	1713	// Check for and copy trailing word
duke@435	1714	__ BIND(L_copy_2_bytes);
never@739	1715	__ testl(word_count, 1);
duke@435	1716	__ jccb(Assembler::zero, L_exit);
duke@435	1717	__ movw(rax, Address(end_from, 8));
duke@435	1718	__ movw(Address(end_to, 8), rax);
duke@435	1719
duke@435	1720	__ BIND(L_exit);
duke@435	1721	restore_arg_regs();
never@3314	1722	inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
never@739	1723	__ xorptr(rax, rax); // return 0
duke@435	1724	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	1725	__ ret(0);
duke@435	1726
kvn@4411	1727	// Copy in multi-bytes chunks
kvn@4411	1728	copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
duke@435	1729	__ jmp(L_copy_4_bytes);
duke@435	1730
duke@435	1731	return start;
duke@435	1732	}
duke@435	1733
never@2118	1734	address generate_fill(BasicType t, bool aligned, const char *name) {
never@2118	1735	__ align(CodeEntryAlignment);
never@2118	1736	StubCodeMark mark(this, "StubRoutines", name);
never@2118	1737	address start = __ pc();
never@2118	1738
never@2118	1739	BLOCK_COMMENT("Entry:");
never@2118	1740
never@2118	1741	const Register to = c_rarg0; // source array address
never@2118	1742	const Register value = c_rarg1; // value
never@2118	1743	const Register count = c_rarg2; // elements count
never@2118	1744
never@2118	1745	__ enter(); // required for proper stackwalking of RuntimeStub frame
never@2118	1746
never@2118	1747	__ generate_fill(t, aligned, to, value, count, rax, xmm0);
never@2118	1748
never@2118	1749	__ leave(); // required for proper stackwalking of RuntimeStub frame
never@2118	1750	__ ret(0);
never@2118	1751	return start;
never@2118	1752	}
never@2118	1753
duke@435	1754	// Arguments:
duke@435	1755	// aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
duke@435	1756	// ignored
duke@435	1757	// name - stub name string
duke@435	1758	//
duke@435	1759	// Inputs:
duke@435	1760	// c_rarg0 - source array address
duke@435	1761	// c_rarg1 - destination array address
duke@435	1762	// c_rarg2 - element count, treated as ssize_t, can be zero
duke@435	1763	//
duke@435	1764	// If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
duke@435	1765	// let the hardware handle it. The two or four words within dwords
duke@435	1766	// or qwords that span cache line boundaries will still be loaded
duke@435	1767	// and stored atomically.
duke@435	1768	//
iveresov@2595	1769	address generate_conjoint_short_copy(bool aligned, address nooverlap_target,
iveresov@2595	1770	address *entry, const char *name) {
duke@435	1771	__ align(CodeEntryAlignment);
duke@435	1772	StubCodeMark mark(this, "StubRoutines", name);
duke@435	1773	address start = __ pc();
duke@435	1774
kvn@4411	1775	Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes;
duke@435	1776	const Register from = rdi; // source array address
duke@435	1777	const Register to = rsi; // destination array address
duke@435	1778	const Register count = rdx; // elements count
duke@435	1779	const Register word_count = rcx;
duke@435	1780	const Register qword_count = count;
duke@435	1781
duke@435	1782	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	1783	assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
duke@435	1784
iveresov@2595	1785	if (entry != NULL) {
iveresov@2595	1786	*entry = __ pc();
iveresov@2595	1787	// caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
iveresov@2595	1788	BLOCK_COMMENT("Entry:");
iveresov@2595	1789	}
iveresov@2595	1790
iveresov@2595	1791	array_overlap_test(nooverlap_target, Address::times_2);
duke@435	1792	setup_arg_regs(); // from => rdi, to => rsi, count => rdx
duke@435	1793	// r9 and r10 may be used to save non-volatile registers
duke@435	1794
duke@435	1795	// 'from', 'to' and 'count' are now valid
never@739	1796	__ movptr(word_count, count);
never@739	1797	__ shrptr(count, 2); // count => qword_count
duke@435	1798
duke@435	1799	// Copy from high to low addresses. Use 'to' as scratch.
duke@435	1800
duke@435	1801	// Check for and copy trailing word
never@739	1802	__ testl(word_count, 1);
duke@435	1803	__ jccb(Assembler::zero, L_copy_4_bytes);
duke@435	1804	__ movw(rax, Address(from, word_count, Address::times_2, -2));
duke@435	1805	__ movw(Address(to, word_count, Address::times_2, -2), rax);
duke@435	1806
duke@435	1807	// Check for and copy trailing dword
duke@435	1808	__ BIND(L_copy_4_bytes);
never@739	1809	__ testl(word_count, 2);
kvn@4411	1810	__ jcc(Assembler::zero, L_copy_bytes);
duke@435	1811	__ movl(rax, Address(from, qword_count, Address::times_8));
duke@435	1812	__ movl(Address(to, qword_count, Address::times_8), rax);
kvn@4411	1813	__ jmp(L_copy_bytes);
duke@435	1814
duke@435	1815	// Copy trailing qwords
duke@435	1816	__ BIND(L_copy_8_bytes);
duke@435	1817	__ movq(rax, Address(from, qword_count, Address::times_8, -8));
duke@435	1818	__ movq(Address(to, qword_count, Address::times_8, -8), rax);
never@739	1819	__ decrement(qword_count);
duke@435	1820	__ jcc(Assembler::notZero, L_copy_8_bytes);
duke@435	1821
duke@435	1822	restore_arg_regs();
never@3314	1823	inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
never@739	1824	__ xorptr(rax, rax); // return 0
duke@435	1825	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	1826	__ ret(0);
duke@435	1827
kvn@4411	1828	// Copy in multi-bytes chunks
kvn@4411	1829	copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
duke@435	1830
duke@435	1831	restore_arg_regs();
never@3314	1832	inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
never@739	1833	__ xorptr(rax, rax); // return 0
duke@435	1834	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	1835	__ ret(0);
duke@435	1836
duke@435	1837	return start;
duke@435	1838	}
duke@435	1839
duke@435	1840	// Arguments:
duke@435	1841	// aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
duke@435	1842	// ignored
coleenp@548	1843	// is_oop - true => oop array, so generate store check code
duke@435	1844	// name - stub name string
duke@435	1845	//
duke@435	1846	// Inputs:
duke@435	1847	// c_rarg0 - source array address
duke@435	1848	// c_rarg1 - destination array address
duke@435	1849	// c_rarg2 - element count, treated as ssize_t, can be zero
duke@435	1850	//
duke@435	1851	// If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
duke@435	1852	// the hardware handle it. The two dwords within qwords that span
duke@435	1853	// cache line boundaries will still be loaded and stored atomicly.
duke@435	1854	//
duke@435	1855	// Side Effects:
duke@435	1856	// disjoint_int_copy_entry is set to the no-overlap entry point
coleenp@548	1857	// used by generate_conjoint_int_oop_copy().
duke@435	1858	//
iveresov@2606	1859	address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, address* entry,
iveresov@2606	1860	const char *name, bool dest_uninitialized = false) {
duke@435	1861	__ align(CodeEntryAlignment);
duke@435	1862	StubCodeMark mark(this, "StubRoutines", name);
duke@435	1863	address start = __ pc();
duke@435	1864
kvn@4411	1865	Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit;
duke@435	1866	const Register from = rdi; // source array address
duke@435	1867	const Register to = rsi; // destination array address
duke@435	1868	const Register count = rdx; // elements count
duke@435	1869	const Register dword_count = rcx;
duke@435	1870	const Register qword_count = count;
duke@435	1871	const Register end_from = from; // source array end address
duke@435	1872	const Register end_to = to; // destination array end address
coleenp@548	1873	const Register saved_to = r11; // saved destination array address
duke@435	1874	// End pointers are inclusive, and if count is not zero they point
duke@435	1875	// to the last unit copied: end_to[0] := end_from[0]
duke@435	1876
duke@435	1877	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	1878	assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
duke@435	1879
iveresov@2595	1880	if (entry != NULL) {
iveresov@2595	1881	*entry = __ pc();
iveresov@2595	1882	// caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
iveresov@2595	1883	BLOCK_COMMENT("Entry:");
coleenp@548	1884	}
coleenp@548	1885
duke@435	1886	setup_arg_regs(); // from => rdi, to => rsi, count => rdx
duke@435	1887	// r9 and r10 may be used to save non-volatile registers
coleenp@548	1888	if (is_oop) {
coleenp@548	1889	__ movq(saved_to, to);
iveresov@2606	1890	gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
coleenp@548	1891	}
coleenp@548	1892
duke@435	1893	// 'from', 'to' and 'count' are now valid
never@739	1894	__ movptr(dword_count, count);
never@739	1895	__ shrptr(count, 1); // count => qword_count
duke@435	1896
duke@435	1897	// Copy from low to high addresses. Use 'to' as scratch.
never@739	1898	__ lea(end_from, Address(from, qword_count, Address::times_8, -8));
never@739	1899	__ lea(end_to, Address(to, qword_count, Address::times_8, -8));
never@739	1900	__ negptr(qword_count);
kvn@4411	1901	__ jmp(L_copy_bytes);
duke@435	1902
duke@435	1903	// Copy trailing qwords
duke@435	1904	__ BIND(L_copy_8_bytes);
duke@435	1905	__ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
duke@435	1906	__ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
never@739	1907	__ increment(qword_count);
duke@435	1908	__ jcc(Assembler::notZero, L_copy_8_bytes);
duke@435	1909
duke@435	1910	// Check for and copy trailing dword
duke@435	1911	__ BIND(L_copy_4_bytes);
never@739	1912	__ testl(dword_count, 1); // Only byte test since the value is 0 or 1
duke@435	1913	__ jccb(Assembler::zero, L_exit);
duke@435	1914	__ movl(rax, Address(end_from, 8));
duke@435	1915	__ movl(Address(end_to, 8), rax);
duke@435	1916
duke@435	1917	__ BIND(L_exit);
coleenp@548	1918	if (is_oop) {
coleenp@548	1919	__ leaq(end_to, Address(saved_to, dword_count, Address::times_4, -4));
coleenp@548	1920	gen_write_ref_array_post_barrier(saved_to, end_to, rax);
coleenp@548	1921	}
duke@435	1922	restore_arg_regs();
never@3314	1923	inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
never@739	1924	__ xorptr(rax, rax); // return 0
duke@435	1925	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	1926	__ ret(0);
duke@435	1927
kvn@4411	1928	// Copy in multi-bytes chunks
kvn@4411	1929	copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
duke@435	1930	__ jmp(L_copy_4_bytes);
duke@435	1931
duke@435	1932	return start;
duke@435	1933	}
duke@435	1934
duke@435	1935	// Arguments:
duke@435	1936	// aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
duke@435	1937	// ignored
coleenp@548	1938	// is_oop - true => oop array, so generate store check code
duke@435	1939	// name - stub name string
duke@435	1940	//
duke@435	1941	// Inputs:
duke@435	1942	// c_rarg0 - source array address
duke@435	1943	// c_rarg1 - destination array address
duke@435	1944	// c_rarg2 - element count, treated as ssize_t, can be zero
duke@435	1945	//
duke@435	1946	// If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
duke@435	1947	// the hardware handle it. The two dwords within qwords that span
duke@435	1948	// cache line boundaries will still be loaded and stored atomicly.
duke@435	1949	//
iveresov@2595	1950	address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, address nooverlap_target,
iveresov@2606	1951	address *entry, const char *name,
iveresov@2606	1952	bool dest_uninitialized = false) {
duke@435	1953	__ align(CodeEntryAlignment);
duke@435	1954	StubCodeMark mark(this, "StubRoutines", name);
duke@435	1955	address start = __ pc();
duke@435	1956
kvn@4411	1957	Label L_copy_bytes, L_copy_8_bytes, L_copy_2_bytes, L_exit;
duke@435	1958	const Register from = rdi; // source array address
duke@435	1959	const Register to = rsi; // destination array address
duke@435	1960	const Register count = rdx; // elements count
duke@435	1961	const Register dword_count = rcx;
duke@435	1962	const Register qword_count = count;
duke@435	1963
duke@435	1964	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	1965	assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
duke@435	1966
iveresov@2595	1967	if (entry != NULL) {
iveresov@2595	1968	*entry = __ pc();
iveresov@2595	1969	// caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
iveresov@2595	1970	BLOCK_COMMENT("Entry:");
iveresov@2595	1971	}
iveresov@2595	1972
iveresov@2595	1973	array_overlap_test(nooverlap_target, Address::times_4);
iveresov@2595	1974	setup_arg_regs(); // from => rdi, to => rsi, count => rdx
iveresov@2595	1975	// r9 and r10 may be used to save non-volatile registers
iveresov@2595	1976
coleenp@548	1977	if (is_oop) {
coleenp@548	1978	// no registers are destroyed by this call
iveresov@2606	1979	gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
coleenp@548	1980	}
coleenp@548	1981
coleenp@548	1982	assert_clean_int(count, rax); // Make sure 'count' is clean int.
duke@435	1983	// 'from', 'to' and 'count' are now valid
never@739	1984	__ movptr(dword_count, count);
never@739	1985	__ shrptr(count, 1); // count => qword_count
duke@435	1986
duke@435	1987	// Copy from high to low addresses. Use 'to' as scratch.
duke@435	1988
duke@435	1989	// Check for and copy trailing dword
never@739	1990	__ testl(dword_count, 1);
kvn@4411	1991	__ jcc(Assembler::zero, L_copy_bytes);
duke@435	1992	__ movl(rax, Address(from, dword_count, Address::times_4, -4));
duke@435	1993	__ movl(Address(to, dword_count, Address::times_4, -4), rax);
kvn@4411	1994	__ jmp(L_copy_bytes);
duke@435	1995
duke@435	1996	// Copy trailing qwords
duke@435	1997	__ BIND(L_copy_8_bytes);
duke@435	1998	__ movq(rax, Address(from, qword_count, Address::times_8, -8));
duke@435	1999	__ movq(Address(to, qword_count, Address::times_8, -8), rax);
never@739	2000	__ decrement(qword_count);
duke@435	2001	__ jcc(Assembler::notZero, L_copy_8_bytes);
duke@435	2002
coleenp@548	2003	if (is_oop) {
coleenp@548	2004	__ jmp(L_exit);
coleenp@548	2005	}
duke@435	2006	restore_arg_regs();
never@3314	2007	inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
never@739	2008	__ xorptr(rax, rax); // return 0
duke@435	2009	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	2010	__ ret(0);
duke@435	2011
kvn@4411	2012	// Copy in multi-bytes chunks
kvn@4411	2013	copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
duke@435	2014
coleenp@548	2015	__ bind(L_exit);
coleenp@548	2016	if (is_oop) {
coleenp@548	2017	Register end_to = rdx;
coleenp@548	2018	__ leaq(end_to, Address(to, dword_count, Address::times_4, -4));
coleenp@548	2019	gen_write_ref_array_post_barrier(to, end_to, rax);
coleenp@548	2020	}
duke@435	2021	restore_arg_regs();
never@3314	2022	inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
never@739	2023	__ xorptr(rax, rax); // return 0
duke@435	2024	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	2025	__ ret(0);
duke@435	2026
duke@435	2027	return start;
duke@435	2028	}
duke@435	2029
duke@435	2030	// Arguments:
duke@435	2031	// aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
duke@435	2032	// ignored
duke@435	2033	// is_oop - true => oop array, so generate store check code
duke@435	2034	// name - stub name string
duke@435	2035	//
duke@435	2036	// Inputs:
duke@435	2037	// c_rarg0 - source array address
duke@435	2038	// c_rarg1 - destination array address
duke@435	2039	// c_rarg2 - element count, treated as ssize_t, can be zero
duke@435	2040	//
coleenp@548	2041	// Side Effects:
duke@435	2042	// disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the
duke@435	2043	// no-overlap entry point used by generate_conjoint_long_oop_copy().
duke@435	2044	//
iveresov@2606	2045	address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, address *entry,
iveresov@2606	2046	const char *name, bool dest_uninitialized = false) {
duke@435	2047	__ align(CodeEntryAlignment);
duke@435	2048	StubCodeMark mark(this, "StubRoutines", name);
duke@435	2049	address start = __ pc();
duke@435	2050
kvn@4411	2051	Label L_copy_bytes, L_copy_8_bytes, L_exit;
duke@435	2052	const Register from = rdi; // source array address
duke@435	2053	const Register to = rsi; // destination array address
duke@435	2054	const Register qword_count = rdx; // elements count
duke@435	2055	const Register end_from = from; // source array end address
duke@435	2056	const Register end_to = rcx; // destination array end address
duke@435	2057	const Register saved_to = to;
duke@435	2058	// End pointers are inclusive, and if count is not zero they point
duke@435	2059	// to the last unit copied: end_to[0] := end_from[0]
duke@435	2060
duke@435	2061	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	2062	// Save no-overlap entry point for generate_conjoint_long_oop_copy()
duke@435	2063	assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
duke@435	2064
iveresov@2595	2065	if (entry != NULL) {
iveresov@2595	2066	*entry = __ pc();
iveresov@2595	2067	// caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
iveresov@2595	2068	BLOCK_COMMENT("Entry:");
duke@435	2069	}
duke@435	2070
duke@435	2071	setup_arg_regs(); // from => rdi, to => rsi, count => rdx
duke@435	2072	// r9 and r10 may be used to save non-volatile registers
duke@435	2073	// 'from', 'to' and 'qword_count' are now valid
iveresov@2595	2074	if (is_oop) {
iveresov@2595	2075	// no registers are destroyed by this call
iveresov@2606	2076	gen_write_ref_array_pre_barrier(to, qword_count, dest_uninitialized);
iveresov@2595	2077	}
duke@435	2078
duke@435	2079	// Copy from low to high addresses. Use 'to' as scratch.
never@739	2080	__ lea(end_from, Address(from, qword_count, Address::times_8, -8));
never@739	2081	__ lea(end_to, Address(to, qword_count, Address::times_8, -8));
never@739	2082	__ negptr(qword_count);
kvn@4411	2083	__ jmp(L_copy_bytes);
duke@435	2084
duke@435	2085	// Copy trailing qwords
duke@435	2086	__ BIND(L_copy_8_bytes);
duke@435	2087	__ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
duke@435	2088	__ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
never@739	2089	__ increment(qword_count);
duke@435	2090	__ jcc(Assembler::notZero, L_copy_8_bytes);
duke@435	2091
duke@435	2092	if (is_oop) {
duke@435	2093	__ jmp(L_exit);
duke@435	2094	} else {
duke@435	2095	restore_arg_regs();
never@3314	2096	inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
never@739	2097	__ xorptr(rax, rax); // return 0
duke@435	2098	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	2099	__ ret(0);
duke@435	2100	}
duke@435	2101
kvn@4411	2102	// Copy in multi-bytes chunks
kvn@4411	2103	copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
duke@435	2104
duke@435	2105	if (is_oop) {
duke@435	2106	__ BIND(L_exit);
duke@435	2107	gen_write_ref_array_post_barrier(saved_to, end_to, rax);
duke@435	2108	}
duke@435	2109	restore_arg_regs();
never@3314	2110	if (is_oop) {
never@3314	2111	inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
never@3314	2112	} else {
never@3314	2113	inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
never@3314	2114	}
never@739	2115	__ xorptr(rax, rax); // return 0
duke@435	2116	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	2117	__ ret(0);
duke@435	2118
duke@435	2119	return start;
duke@435	2120	}
duke@435	2121
duke@435	2122	// Arguments:
duke@435	2123	// aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
duke@435	2124	// ignored
duke@435	2125	// is_oop - true => oop array, so generate store check code
duke@435	2126	// name - stub name string
duke@435	2127	//
duke@435	2128	// Inputs:
duke@435	2129	// c_rarg0 - source array address
duke@435	2130	// c_rarg1 - destination array address
duke@435	2131	// c_rarg2 - element count, treated as ssize_t, can be zero
duke@435	2132	//
iveresov@2606	2133	address generate_conjoint_long_oop_copy(bool aligned, bool is_oop,
iveresov@2606	2134	address nooverlap_target, address *entry,
iveresov@2606	2135	const char *name, bool dest_uninitialized = false) {
duke@435	2136	__ align(CodeEntryAlignment);
duke@435	2137	StubCodeMark mark(this, "StubRoutines", name);
duke@435	2138	address start = __ pc();
duke@435	2139
kvn@4411	2140	Label L_copy_bytes, L_copy_8_bytes, L_exit;
duke@435	2141	const Register from = rdi; // source array address
duke@435	2142	const Register to = rsi; // destination array address
duke@435	2143	const Register qword_count = rdx; // elements count
duke@435	2144	const Register saved_count = rcx;
duke@435	2145
duke@435	2146	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	2147	assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
duke@435	2148
iveresov@2595	2149	if (entry != NULL) {
iveresov@2595	2150	*entry = __ pc();
iveresov@2595	2151	// caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
iveresov@2595	2152	BLOCK_COMMENT("Entry:");
duke@435	2153	}
iveresov@2595	2154
iveresov@2595	2155	array_overlap_test(nooverlap_target, Address::times_8);
duke@435	2156	setup_arg_regs(); // from => rdi, to => rsi, count => rdx
duke@435	2157	// r9 and r10 may be used to save non-volatile registers
duke@435	2158	// 'from', 'to' and 'qword_count' are now valid
duke@435	2159	if (is_oop) {
duke@435	2160	// Save to and count for store barrier
never@739	2161	__ movptr(saved_count, qword_count);
duke@435	2162	// No registers are destroyed by this call
iveresov@2606	2163	gen_write_ref_array_pre_barrier(to, saved_count, dest_uninitialized);
duke@435	2164	}
duke@435	2165
kvn@4411	2166	__ jmp(L_copy_bytes);
duke@435	2167
duke@435	2168	// Copy trailing qwords
duke@435	2169	__ BIND(L_copy_8_bytes);
duke@435	2170	__ movq(rax, Address(from, qword_count, Address::times_8, -8));
duke@435	2171	__ movq(Address(to, qword_count, Address::times_8, -8), rax);
never@739	2172	__ decrement(qword_count);
duke@435	2173	__ jcc(Assembler::notZero, L_copy_8_bytes);
duke@435	2174
duke@435	2175	if (is_oop) {
duke@435	2176	__ jmp(L_exit);
duke@435	2177	} else {
duke@435	2178	restore_arg_regs();
never@3314	2179	inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
never@739	2180	__ xorptr(rax, rax); // return 0
duke@435	2181	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	2182	__ ret(0);
duke@435	2183	}
duke@435	2184
kvn@4411	2185	// Copy in multi-bytes chunks
kvn@4411	2186	copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
duke@435	2187
duke@435	2188	if (is_oop) {
duke@435	2189	__ BIND(L_exit);
never@739	2190	__ lea(rcx, Address(to, saved_count, Address::times_8, -8));
duke@435	2191	gen_write_ref_array_post_barrier(to, rcx, rax);
duke@435	2192	}
duke@435	2193	restore_arg_regs();
never@3314	2194	if (is_oop) {
never@3314	2195	inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
never@3314	2196	} else {
never@3314	2197	inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
never@3314	2198	}
never@739	2199	__ xorptr(rax, rax); // return 0
duke@435	2200	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	2201	__ ret(0);
duke@435	2202
duke@435	2203	return start;
duke@435	2204	}
duke@435	2205
duke@435	2206
duke@435	2207	// Helper for generating a dynamic type check.
duke@435	2208	// Smashes no registers.
duke@435	2209	void generate_type_check(Register sub_klass,
duke@435	2210	Register super_check_offset,
duke@435	2211	Register super_klass,
duke@435	2212	Label& L_success) {
duke@435	2213	assert_different_registers(sub_klass, super_check_offset, super_klass);
duke@435	2214
duke@435	2215	BLOCK_COMMENT("type_check:");
duke@435	2216
duke@435	2217	Label L_miss;
duke@435	2218
jrose@1079	2219	__ check_klass_subtype_fast_path(sub_klass, super_klass, noreg, &L_success, &L_miss, NULL,
jrose@1079	2220	super_check_offset);
jrose@1079	2221	__ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL);
duke@435	2222
duke@435	2223	// Fall through on failure!
duke@435	2224	__ BIND(L_miss);
duke@435	2225	}
duke@435	2226
duke@435	2227	//
duke@435	2228	// Generate checkcasting array copy stub
duke@435	2229	//
duke@435	2230	// Input:
duke@435	2231	// c_rarg0 - source array address
duke@435	2232	// c_rarg1 - destination array address
duke@435	2233	// c_rarg2 - element count, treated as ssize_t, can be zero
duke@435	2234	// c_rarg3 - size_t ckoff (super_check_offset)
duke@435	2235	// not Win64
duke@435	2236	// c_rarg4 - oop ckval (super_klass)
duke@435	2237	// Win64
duke@435	2238	// rsp+40 - oop ckval (super_klass)
duke@435	2239	//
duke@435	2240	// Output:
duke@435	2241	// rax == 0 - success
duke@435	2242	// rax == -1^K - failure, where K is partial transfer count
duke@435	2243	//
iveresov@2606	2244	address generate_checkcast_copy(const char *name, address *entry,
iveresov@2606	2245	bool dest_uninitialized = false) {
duke@435	2246
duke@435	2247	Label L_load_element, L_store_element, L_do_card_marks, L_done;
duke@435	2248
duke@435	2249	// Input registers (after setup_arg_regs)
duke@435	2250	const Register from = rdi; // source array address
duke@435	2251	const Register to = rsi; // destination array address
duke@435	2252	const Register length = rdx; // elements count
duke@435	2253	const Register ckoff = rcx; // super_check_offset
duke@435	2254	const Register ckval = r8; // super_klass
duke@435	2255
duke@435	2256	// Registers used as temps (r13, r14 are save-on-entry)
duke@435	2257	const Register end_from = from; // source array end address
duke@435	2258	const Register end_to = r13; // destination array end address
duke@435	2259	const Register count = rdx; // -(count_remaining)
duke@435	2260	const Register r14_length = r14; // saved copy of length
duke@435	2261	// End pointers are inclusive, and if length is not zero they point
duke@435	2262	// to the last unit copied: end_to[0] := end_from[0]
duke@435	2263
duke@435	2264	const Register rax_oop = rax; // actual oop copied
duke@435	2265	const Register r11_klass = r11; // oop._klass
duke@435	2266
duke@435	2267	//---------------------------------------------------------------
duke@435	2268	// Assembler stub will be used for this call to arraycopy
duke@435	2269	// if the two arrays are subtypes of Object[] but the
duke@435	2270	// destination array type is not equal to or a supertype
duke@435	2271	// of the source type. Each element must be separately
duke@435	2272	// checked.
duke@435	2273
duke@435	2274	__ align(CodeEntryAlignment);
duke@435	2275	StubCodeMark mark(this, "StubRoutines", name);
duke@435	2276	address start = __ pc();
duke@435	2277
duke@435	2278	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	2279
duke@435	2280	#ifdef ASSERT
duke@435	2281	// caller guarantees that the arrays really are different
duke@435	2282	// otherwise, we would have to make conjoint checks
duke@435	2283	{ Label L;
coleenp@548	2284	array_overlap_test(L, TIMES_OOP);
duke@435	2285	__ stop("checkcast_copy within a single array");
duke@435	2286	__ bind(L);
duke@435	2287	}
duke@435	2288	#endif //ASSERT
duke@435	2289
duke@435	2290	setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
duke@435	2291	// ckoff => rcx, ckval => r8
duke@435	2292	// r9 and r10 may be used to save non-volatile registers
duke@435	2293	#ifdef _WIN64
duke@435	2294	// last argument (#4) is on stack on Win64
twisti@2348	2295	__ movptr(ckval, Address(rsp, 6 * wordSize));
duke@435	2296	#endif
duke@435	2297
twisti@2348	2298	// Caller of this entry point must set up the argument registers.
iveresov@2595	2299	if (entry != NULL) {
iveresov@2595	2300	*entry = __ pc();
iveresov@2595	2301	BLOCK_COMMENT("Entry:");
iveresov@2595	2302	}
twisti@2348	2303
twisti@2348	2304	// allocate spill slots for r13, r14
twisti@2348	2305	enum {
twisti@2348	2306	saved_r13_offset,
twisti@2348	2307	saved_r14_offset,
twisti@2348	2308	saved_rbp_offset
twisti@2348	2309	};
twisti@2348	2310	__ subptr(rsp, saved_rbp_offset * wordSize);
twisti@2348	2311	__ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
twisti@2348	2312	__ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
twisti@2348	2313
duke@435	2314	// check that int operands are properly extended to size_t
duke@435	2315	assert_clean_int(length, rax);
duke@435	2316	assert_clean_int(ckoff, rax);
duke@435	2317
duke@435	2318	#ifdef ASSERT
duke@435	2319	BLOCK_COMMENT("assert consistent ckoff/ckval");
duke@435	2320	// The ckoff and ckval must be mutually consistent,
duke@435	2321	// even though caller generates both.
duke@435	2322	{ Label L;
stefank@3391	2323	int sco_offset = in_bytes(Klass::super_check_offset_offset());
duke@435	2324	__ cmpl(ckoff, Address(ckval, sco_offset));
duke@435	2325	__ jcc(Assembler::equal, L);
duke@435	2326	__ stop("super_check_offset inconsistent");
duke@435	2327	__ bind(L);
duke@435	2328	}
duke@435	2329	#endif //ASSERT
duke@435	2330
duke@435	2331	// Loop-invariant addresses. They are exclusive end pointers.
coleenp@548	2332	Address end_from_addr(from, length, TIMES_OOP, 0);
coleenp@548	2333	Address end_to_addr(to, length, TIMES_OOP, 0);
duke@435	2334	// Loop-variant addresses. They assume post-incremented count < 0.
coleenp@548	2335	Address from_element_addr(end_from, count, TIMES_OOP, 0);
coleenp@548	2336	Address to_element_addr(end_to, count, TIMES_OOP, 0);
duke@435	2337
iveresov@2606	2338	gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
duke@435	2339
duke@435	2340	// Copy from low to high addresses, indexed from the end of each array.
never@739	2341	__ lea(end_from, end_from_addr);
never@739	2342	__ lea(end_to, end_to_addr);
never@739	2343	__ movptr(r14_length, length); // save a copy of the length
never@739	2344	assert(length == count, ""); // else fix next line:
never@739	2345	__ negptr(count); // negate and test the length
duke@435	2346	__ jcc(Assembler::notZero, L_load_element);
duke@435	2347
duke@435	2348	// Empty array: Nothing to do.
never@739	2349	__ xorptr(rax, rax); // return 0 on (trivial) success
duke@435	2350	__ jmp(L_done);
duke@435	2351
duke@435	2352	// ======== begin loop ========
duke@435	2353	// (Loop is rotated; its entry is L_load_element.)
duke@435	2354	// Loop control:
duke@435	2355	// for (count = -count; count != 0; count++)
duke@435	2356	// Base pointers src, dst are biased by 8*(count-1),to last element.
kvn@1800	2357	__ align(OptoLoopAlignment);
duke@435	2358
duke@435	2359	__ BIND(L_store_element);
coleenp@548	2360	__ store_heap_oop(to_element_addr, rax_oop); // store the oop
never@739	2361	__ increment(count); // increment the count toward zero
duke@435	2362	__ jcc(Assembler::zero, L_do_card_marks);
duke@435	2363
duke@435	2364	// ======== loop entry is here ========
duke@435	2365	__ BIND(L_load_element);
coleenp@548	2366	__ load_heap_oop(rax_oop, from_element_addr); // load the oop
never@739	2367	__ testptr(rax_oop, rax_oop);
duke@435	2368	__ jcc(Assembler::zero, L_store_element);
duke@435	2369
coleenp@548	2370	__ load_klass(r11_klass, rax_oop);// query the object klass
duke@435	2371	generate_type_check(r11_klass, ckoff, ckval, L_store_element);
duke@435	2372	// ======== end loop ========
duke@435	2373
duke@435	2374	// It was a real error; we must depend on the caller to finish the job.
duke@435	2375	// Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
duke@435	2376	// Emit GC store barriers for the oops we have copied (r14 + rdx),
duke@435	2377	// and report their number to the caller.
duke@435	2378	assert_different_registers(rax, r14_length, count, to, end_to, rcx);
never@739	2379	__ lea(end_to, to_element_addr);
ysr@1280	2380	__ addptr(end_to, -heapOopSize); // make an inclusive end pointer
apetrusenko@797	2381	gen_write_ref_array_post_barrier(to, end_to, rscratch1);
never@739	2382	__ movptr(rax, r14_length); // original oops
never@739	2383	__ addptr(rax, count); // K = (original - remaining) oops
never@739	2384	__ notptr(rax); // report (-1^K) to caller
duke@435	2385	__ jmp(L_done);
duke@435	2386
duke@435	2387	// Come here on success only.
duke@435	2388	__ BIND(L_do_card_marks);
ysr@1280	2389	__ addptr(end_to, -heapOopSize); // make an inclusive end pointer
apetrusenko@797	2390	gen_write_ref_array_post_barrier(to, end_to, rscratch1);
never@739	2391	__ xorptr(rax, rax); // return 0 on success
duke@435	2392
duke@435	2393	// Common exit point (success or failure).
duke@435	2394	__ BIND(L_done);
never@739	2395	__ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
never@739	2396	__ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
duke@435	2397	restore_arg_regs();
never@3314	2398	inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); // Update counter after rscratch1 is free
duke@435	2399	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	2400	__ ret(0);
duke@435	2401
duke@435	2402	return start;
duke@435	2403	}
duke@435	2404
duke@435	2405	//
duke@435	2406	// Generate 'unsafe' array copy stub
duke@435	2407	// Though just as safe as the other stubs, it takes an unscaled
duke@435	2408	// size_t argument instead of an element count.
duke@435	2409	//
duke@435	2410	// Input:
duke@435	2411	// c_rarg0 - source array address
duke@435	2412	// c_rarg1 - destination array address
duke@435	2413	// c_rarg2 - byte count, treated as ssize_t, can be zero
duke@435	2414	//
duke@435	2415	// Examines the alignment of the operands and dispatches
duke@435	2416	// to a long, int, short, or byte copy loop.
duke@435	2417	//
iveresov@2595	2418	address generate_unsafe_copy(const char *name,
iveresov@2595	2419	address byte_copy_entry, address short_copy_entry,
iveresov@2595	2420	address int_copy_entry, address long_copy_entry) {
duke@435	2421
duke@435	2422	Label L_long_aligned, L_int_aligned, L_short_aligned;
duke@435	2423
duke@435	2424	// Input registers (before setup_arg_regs)
duke@435	2425	const Register from = c_rarg0; // source array address
duke@435	2426	const Register to = c_rarg1; // destination array address
duke@435	2427	const Register size = c_rarg2; // byte count (size_t)
duke@435	2428
duke@435	2429	// Register used as a temp
duke@435	2430	const Register bits = rax; // test copy of low bits
duke@435	2431
duke@435	2432	__ align(CodeEntryAlignment);
duke@435	2433	StubCodeMark mark(this, "StubRoutines", name);
duke@435	2434	address start = __ pc();
duke@435	2435
duke@435	2436	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	2437
duke@435	2438	// bump this on entry, not on exit:
duke@435	2439	inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
duke@435	2440
never@739	2441	__ mov(bits, from);
never@739	2442	__ orptr(bits, to);
never@739	2443	__ orptr(bits, size);
duke@435	2444
duke@435	2445	__ testb(bits, BytesPerLong-1);
duke@435	2446	__ jccb(Assembler::zero, L_long_aligned);
duke@435	2447
duke@435	2448	__ testb(bits, BytesPerInt-1);
duke@435	2449	__ jccb(Assembler::zero, L_int_aligned);
duke@435	2450
duke@435	2451	__ testb(bits, BytesPerShort-1);
duke@435	2452	__ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
duke@435	2453
duke@435	2454	__ BIND(L_short_aligned);
never@739	2455	__ shrptr(size, LogBytesPerShort); // size => short_count
duke@435	2456	__ jump(RuntimeAddress(short_copy_entry));
duke@435	2457
duke@435	2458	__ BIND(L_int_aligned);
never@739	2459	__ shrptr(size, LogBytesPerInt); // size => int_count
duke@435	2460	__ jump(RuntimeAddress(int_copy_entry));
duke@435	2461
duke@435	2462	__ BIND(L_long_aligned);
never@739	2463	__ shrptr(size, LogBytesPerLong); // size => qword_count
duke@435	2464	__ jump(RuntimeAddress(long_copy_entry));
duke@435	2465
duke@435	2466	return start;
duke@435	2467	}
duke@435	2468
duke@435	2469	// Perform range checks on the proposed arraycopy.
duke@435	2470	// Kills temp, but nothing else.
duke@435	2471	// Also, clean the sign bits of src_pos and dst_pos.
duke@435	2472	void arraycopy_range_checks(Register src, // source array oop (c_rarg0)
duke@435	2473	Register src_pos, // source position (c_rarg1)
duke@435	2474	Register dst, // destination array oo (c_rarg2)
duke@435	2475	Register dst_pos, // destination position (c_rarg3)
duke@435	2476	Register length,
duke@435	2477	Register temp,
duke@435	2478	Label& L_failed) {
duke@435	2479	BLOCK_COMMENT("arraycopy_range_checks:");
duke@435	2480
duke@435	2481	// if (src_pos + length > arrayOop(src)->length()) FAIL;
duke@435	2482	__ movl(temp, length);
duke@435	2483	__ addl(temp, src_pos); // src_pos + length
duke@435	2484	__ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
duke@435	2485	__ jcc(Assembler::above, L_failed);
duke@435	2486
duke@435	2487	// if (dst_pos + length > arrayOop(dst)->length()) FAIL;
duke@435	2488	__ movl(temp, length);
duke@435	2489	__ addl(temp, dst_pos); // dst_pos + length
duke@435	2490	__ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
duke@435	2491	__ jcc(Assembler::above, L_failed);
duke@435	2492
duke@435	2493	// Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
duke@435	2494	// Move with sign extension can be used since they are positive.
duke@435	2495	__ movslq(src_pos, src_pos);
duke@435	2496	__ movslq(dst_pos, dst_pos);
duke@435	2497
duke@435	2498	BLOCK_COMMENT("arraycopy_range_checks done");
duke@435	2499	}
duke@435	2500
duke@435	2501	//
duke@435	2502	// Generate generic array copy stubs
duke@435	2503	//
duke@435	2504	// Input:
duke@435	2505	// c_rarg0 - src oop
duke@435	2506	// c_rarg1 - src_pos (32-bits)
duke@435	2507	// c_rarg2 - dst oop
duke@435	2508	// c_rarg3 - dst_pos (32-bits)
duke@435	2509	// not Win64
duke@435	2510	// c_rarg4 - element count (32-bits)
duke@435	2511	// Win64
duke@435	2512	// rsp+40 - element count (32-bits)
duke@435	2513	//
duke@435	2514	// Output:
duke@435	2515	// rax == 0 - success
duke@435	2516	// rax == -1^K - failure, where K is partial transfer count
duke@435	2517	//
iveresov@2595	2518	address generate_generic_copy(const char *name,
iveresov@2595	2519	address byte_copy_entry, address short_copy_entry,
iveresov@2691	2520	address int_copy_entry, address oop_copy_entry,
iveresov@2691	2521	address long_copy_entry, address checkcast_copy_entry) {
duke@435	2522
duke@435	2523	Label L_failed, L_failed_0, L_objArray;
duke@435	2524	Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
duke@435	2525
duke@435	2526	// Input registers
duke@435	2527	const Register src = c_rarg0; // source array oop
duke@435	2528	const Register src_pos = c_rarg1; // source position
duke@435	2529	const Register dst = c_rarg2; // destination array oop
duke@435	2530	const Register dst_pos = c_rarg3; // destination position
twisti@2348	2531	#ifndef _WIN64
twisti@2348	2532	const Register length = c_rarg4;
duke@435	2533	#else
twisti@2348	2534	const Address length(rsp, 6 * wordSize); // elements count is on stack on Win64
duke@435	2535	#endif
duke@435	2536
duke@435	2537	{ int modulus = CodeEntryAlignment;
duke@435	2538	int target = modulus - 5; // 5 = sizeof jmp(L_failed)
duke@435	2539	int advance = target - (__ offset() % modulus);
duke@435	2540	if (advance < 0) advance += modulus;
duke@435	2541	if (advance > 0) __ nop(advance);
duke@435	2542	}
duke@435	2543	StubCodeMark mark(this, "StubRoutines", name);
duke@435	2544
duke@435	2545	// Short-hop target to L_failed. Makes for denser prologue code.
duke@435	2546	__ BIND(L_failed_0);
duke@435	2547	__ jmp(L_failed);
duke@435	2548	assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
duke@435	2549
duke@435	2550	__ align(CodeEntryAlignment);
duke@435	2551	address start = __ pc();
duke@435	2552
duke@435	2553	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	2554
duke@435	2555	// bump this on entry, not on exit:
duke@435	2556	inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
duke@435	2557
duke@435	2558	//-----------------------------------------------------------------------
duke@435	2559	// Assembler stub will be used for this call to arraycopy
duke@435	2560	// if the following conditions are met:
duke@435	2561	//
duke@435	2562	// (1) src and dst must not be null.
duke@435	2563	// (2) src_pos must not be negative.
duke@435	2564	// (3) dst_pos must not be negative.
duke@435	2565	// (4) length must not be negative.
duke@435	2566	// (5) src klass and dst klass should be the same and not NULL.
duke@435	2567	// (6) src and dst should be arrays.
duke@435	2568	// (7) src_pos + length must not exceed length of src.
duke@435	2569	// (8) dst_pos + length must not exceed length of dst.
duke@435	2570	//
duke@435	2571
duke@435	2572	// if (src == NULL) return -1;
never@739	2573	__ testptr(src, src); // src oop
duke@435	2574	size_t j1off = __ offset();
duke@435	2575	__ jccb(Assembler::zero, L_failed_0);
duke@435	2576
duke@435	2577	// if (src_pos < 0) return -1;
duke@435	2578	__ testl(src_pos, src_pos); // src_pos (32-bits)
duke@435	2579	__ jccb(Assembler::negative, L_failed_0);
duke@435	2580
duke@435	2581	// if (dst == NULL) return -1;
never@739	2582	__ testptr(dst, dst); // dst oop
duke@435	2583	__ jccb(Assembler::zero, L_failed_0);
duke@435	2584
duke@435	2585	// if (dst_pos < 0) return -1;
duke@435	2586	__ testl(dst_pos, dst_pos); // dst_pos (32-bits)
duke@435	2587	size_t j4off = __ offset();
duke@435	2588	__ jccb(Assembler::negative, L_failed_0);
duke@435	2589
duke@435	2590	// The first four tests are very dense code,
duke@435	2591	// but not quite dense enough to put four
duke@435	2592	// jumps in a 16-byte instruction fetch buffer.
duke@435	2593	// That's good, because some branch predicters
duke@435	2594	// do not like jumps so close together.
duke@435	2595	// Make sure of this.
duke@435	2596	guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
duke@435	2597
duke@435	2598	// registers used as temp
duke@435	2599	const Register r11_length = r11; // elements count to copy
duke@435	2600	const Register r10_src_klass = r10; // array klass
duke@435	2601
duke@435	2602	// if (length < 0) return -1;
twisti@2348	2603	__ movl(r11_length, length); // length (elements count, 32-bits value)
duke@435	2604	__ testl(r11_length, r11_length);
duke@435	2605	__ jccb(Assembler::negative, L_failed_0);
duke@435	2606
coleenp@548	2607	__ load_klass(r10_src_klass, src);
duke@435	2608	#ifdef ASSERT
duke@435	2609	// assert(src->klass() != NULL);
twisti@2348	2610	{
twisti@2348	2611	BLOCK_COMMENT("assert klasses not null {");
twisti@2348	2612	Label L1, L2;
never@739	2613	__ testptr(r10_src_klass, r10_src_klass);
duke@435	2614	__ jcc(Assembler::notZero, L2); // it is broken if klass is NULL
duke@435	2615	__ bind(L1);
duke@435	2616	__ stop("broken null klass");
duke@435	2617	__ bind(L2);
twisti@2348	2618	__ load_klass(rax, dst);
twisti@2348	2619	__ cmpq(rax, 0);
duke@435	2620	__ jcc(Assembler::equal, L1); // this would be broken also
twisti@2348	2621	BLOCK_COMMENT("} assert klasses not null done");
duke@435	2622	}
duke@435	2623	#endif
duke@435	2624
duke@435	2625	// Load layout helper (32-bits)
duke@435	2626	//
duke@435	2627	// |array_tag| | header_size | element_type | |log2_element_size|
duke@435	2628	// 32 30 24 16 8 2 0
duke@435	2629	//
duke@435	2630	// array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
duke@435	2631	//
duke@435	2632
stefank@3391	2633	const int lh_offset = in_bytes(Klass::layout_helper_offset());
twisti@2348	2634
twisti@2348	2635	// Handle objArrays completely differently...
twisti@2348	2636	const jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
twisti@2348	2637	__ cmpl(Address(r10_src_klass, lh_offset), objArray_lh);
twisti@2348	2638	__ jcc(Assembler::equal, L_objArray);
twisti@2348	2639
twisti@2348	2640	// if (src->klass() != dst->klass()) return -1;
twisti@2348	2641	__ load_klass(rax, dst);
twisti@2348	2642	__ cmpq(r10_src_klass, rax);
twisti@2348	2643	__ jcc(Assembler::notEqual, L_failed);
duke@435	2644
duke@435	2645	const Register rax_lh = rax; // layout helper
duke@435	2646	__ movl(rax_lh, Address(r10_src_klass, lh_offset));
duke@435	2647
duke@435	2648	// if (!src->is_Array()) return -1;
duke@435	2649	__ cmpl(rax_lh, Klass::_lh_neutral_value);
duke@435	2650	__ jcc(Assembler::greaterEqual, L_failed);
duke@435	2651
duke@435	2652	// At this point, it is known to be a typeArray (array_tag 0x3).
duke@435	2653	#ifdef ASSERT
twisti@2348	2654	{
twisti@2348	2655	BLOCK_COMMENT("assert primitive array {");
twisti@2348	2656	Label L;
duke@435	2657	__ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
duke@435	2658	__ jcc(Assembler::greaterEqual, L);
duke@435	2659	__ stop("must be a primitive array");
duke@435	2660	__ bind(L);
twisti@2348	2661	BLOCK_COMMENT("} assert primitive array done");
duke@435	2662	}
duke@435	2663	#endif
duke@435	2664
duke@435	2665	arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
duke@435	2666	r10, L_failed);
duke@435	2667
coleenp@4142	2668	// TypeArrayKlass
duke@435	2669	//
duke@435	2670	// src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
duke@435	2671	// dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
duke@435	2672	//
duke@435	2673
duke@435	2674	const Register r10_offset = r10; // array offset
duke@435	2675	const Register rax_elsize = rax_lh; // element size
duke@435	2676
duke@435	2677	__ movl(r10_offset, rax_lh);
duke@435	2678	__ shrl(r10_offset, Klass::_lh_header_size_shift);
never@739	2679	__ andptr(r10_offset, Klass::_lh_header_size_mask); // array_offset
never@739	2680	__ addptr(src, r10_offset); // src array offset
never@739	2681	__ addptr(dst, r10_offset); // dst array offset
duke@435	2682	BLOCK_COMMENT("choose copy loop based on element size");
duke@435	2683	__ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize
duke@435	2684
duke@435	2685	// next registers should be set before the jump to corresponding stub
duke@435	2686	const Register from = c_rarg0; // source array address
duke@435	2687	const Register to = c_rarg1; // destination array address
duke@435	2688	const Register count = c_rarg2; // elements count
duke@435	2689
duke@435	2690	// 'from', 'to', 'count' registers should be set in such order
duke@435	2691	// since they are the same as 'src', 'src_pos', 'dst'.
duke@435	2692
duke@435	2693	__ BIND(L_copy_bytes);
duke@435	2694	__ cmpl(rax_elsize, 0);
duke@435	2695	__ jccb(Assembler::notEqual, L_copy_shorts);
never@739	2696	__ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr
never@739	2697	__ lea(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr
never@739	2698	__ movl2ptr(count, r11_length); // length
duke@435	2699	__ jump(RuntimeAddress(byte_copy_entry));
duke@435	2700
duke@435	2701	__ BIND(L_copy_shorts);
duke@435	2702	__ cmpl(rax_elsize, LogBytesPerShort);
duke@435	2703	__ jccb(Assembler::notEqual, L_copy_ints);
never@739	2704	__ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr
never@739	2705	__ lea(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr
never@739	2706	__ movl2ptr(count, r11_length); // length
duke@435	2707	__ jump(RuntimeAddress(short_copy_entry));
duke@435	2708
duke@435	2709	__ BIND(L_copy_ints);
duke@435	2710	__ cmpl(rax_elsize, LogBytesPerInt);
duke@435	2711	__ jccb(Assembler::notEqual, L_copy_longs);
never@739	2712	__ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr
never@739	2713	__ lea(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr
never@739	2714	__ movl2ptr(count, r11_length); // length
duke@435	2715	__ jump(RuntimeAddress(int_copy_entry));
duke@435	2716
duke@435	2717	__ BIND(L_copy_longs);
duke@435	2718	#ifdef ASSERT
twisti@2348	2719	{
twisti@2348	2720	BLOCK_COMMENT("assert long copy {");
twisti@2348	2721	Label L;
duke@435	2722	__ cmpl(rax_elsize, LogBytesPerLong);
duke@435	2723	__ jcc(Assembler::equal, L);
duke@435	2724	__ stop("must be long copy, but elsize is wrong");
duke@435	2725	__ bind(L);
twisti@2348	2726	BLOCK_COMMENT("} assert long copy done");
duke@435	2727	}
duke@435	2728	#endif
never@739	2729	__ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr
never@739	2730	__ lea(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr
never@739	2731	__ movl2ptr(count, r11_length); // length
duke@435	2732	__ jump(RuntimeAddress(long_copy_entry));
duke@435	2733
coleenp@4142	2734	// ObjArrayKlass
duke@435	2735	__ BIND(L_objArray);
twisti@2348	2736	// live at this point: r10_src_klass, r11_length, src[_pos], dst[_pos]
duke@435	2737
duke@435	2738	Label L_plain_copy, L_checkcast_copy;
duke@435	2739	// test array classes for subtyping
twisti@2348	2740	__ load_klass(rax, dst);
twisti@2348	2741	__ cmpq(r10_src_klass, rax); // usual case is exact equality
duke@435	2742	__ jcc(Assembler::notEqual, L_checkcast_copy);
duke@435	2743
duke@435	2744	// Identically typed arrays can be copied without element-wise checks.
duke@435	2745	arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
duke@435	2746	r10, L_failed);
duke@435	2747
never@739	2748	__ lea(from, Address(src, src_pos, TIMES_OOP,
duke@435	2749	arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
never@739	2750	__ lea(to, Address(dst, dst_pos, TIMES_OOP,
never@739	2751	arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
never@739	2752	__ movl2ptr(count, r11_length); // length
duke@435	2753	__ BIND(L_plain_copy);
duke@435	2754	__ jump(RuntimeAddress(oop_copy_entry));
duke@435	2755
duke@435	2756	__ BIND(L_checkcast_copy);
twisti@2348	2757	// live at this point: r10_src_klass, r11_length, rax (dst_klass)
duke@435	2758	{
duke@435	2759	// Before looking at dst.length, make sure dst is also an objArray.
twisti@2348	2760	__ cmpl(Address(rax, lh_offset), objArray_lh);
duke@435	2761	__ jcc(Assembler::notEqual, L_failed);
duke@435	2762
duke@435	2763	// It is safe to examine both src.length and dst.length.
duke@435	2764	arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
duke@435	2765	rax, L_failed);
twisti@2348	2766
twisti@2348	2767	const Register r11_dst_klass = r11;
coleenp@548	2768	__ load_klass(r11_dst_klass, dst); // reload
duke@435	2769
duke@435	2770	// Marshal the base address arguments now, freeing registers.
never@739	2771	__ lea(from, Address(src, src_pos, TIMES_OOP,
duke@435	2772	arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
never@739	2773	__ lea(to, Address(dst, dst_pos, TIMES_OOP,
duke@435	2774	arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
twisti@2348	2775	__ movl(count, length); // length (reloaded)
duke@435	2776	Register sco_temp = c_rarg3; // this register is free now
duke@435	2777	assert_different_registers(from, to, count, sco_temp,
duke@435	2778	r11_dst_klass, r10_src_klass);
duke@435	2779	assert_clean_int(count, sco_temp);
duke@435	2780
duke@435	2781	// Generate the type check.
stefank@3391	2782	const int sco_offset = in_bytes(Klass::super_check_offset_offset());
duke@435	2783	__ movl(sco_temp, Address(r11_dst_klass, sco_offset));
duke@435	2784	assert_clean_int(sco_temp, rax);
duke@435	2785	generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
duke@435	2786
coleenp@4142	2787	// Fetch destination element klass from the ObjArrayKlass header.
coleenp@4142	2788	int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
never@739	2789	__ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset));
twisti@2348	2790	__ movl( sco_temp, Address(r11_dst_klass, sco_offset));
duke@435	2791	assert_clean_int(sco_temp, rax);
duke@435	2792
duke@435	2793	// the checkcast_copy loop needs two extra arguments:
duke@435	2794	assert(c_rarg3 == sco_temp, "#3 already in place");
twisti@2348	2795	// Set up arguments for checkcast_copy_entry.
twisti@2348	2796	setup_arg_regs(4);
twisti@2348	2797	__ movptr(r8, r11_dst_klass); // dst.klass.element_klass, r8 is c_rarg4 on Linux/Solaris
duke@435	2798	__ jump(RuntimeAddress(checkcast_copy_entry));
duke@435	2799	}
duke@435	2800
duke@435	2801	__ BIND(L_failed);
never@739	2802	__ xorptr(rax, rax);
never@739	2803	__ notptr(rax); // return -1
duke@435	2804	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	2805	__ ret(0);
duke@435	2806
duke@435	2807	return start;
duke@435	2808	}
duke@435	2809
duke@435	2810	void generate_arraycopy_stubs() {
iveresov@2595	2811	address entry;
iveresov@2595	2812	address entry_jbyte_arraycopy;
iveresov@2595	2813	address entry_jshort_arraycopy;
iveresov@2595	2814	address entry_jint_arraycopy;
iveresov@2595	2815	address entry_oop_arraycopy;
iveresov@2595	2816	address entry_jlong_arraycopy;
iveresov@2595	2817	address entry_checkcast_arraycopy;
iveresov@2595	2818
iveresov@2595	2819	StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, &entry,
iveresov@2595	2820	"jbyte_disjoint_arraycopy");
iveresov@2595	2821	StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, entry, &entry_jbyte_arraycopy,
iveresov@2595	2822	"jbyte_arraycopy");
iveresov@2595	2823
iveresov@2595	2824	StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, &entry,
iveresov@2595	2825	"jshort_disjoint_arraycopy");
iveresov@2595	2826	StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, entry, &entry_jshort_arraycopy,
iveresov@2595	2827	"jshort_arraycopy");
iveresov@2595	2828
iveresov@2595	2829	StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, false, &entry,
iveresov@2595	2830	"jint_disjoint_arraycopy");
iveresov@2595	2831	StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, entry,
iveresov@2595	2832	&entry_jint_arraycopy, "jint_arraycopy");
iveresov@2595	2833
iveresov@2595	2834	StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, false, &entry,
iveresov@2595	2835	"jlong_disjoint_arraycopy");
iveresov@2595	2836	StubRoutines::_jlong_arraycopy = generate_conjoint_long_oop_copy(false, false, entry,
iveresov@2595	2837	&entry_jlong_arraycopy, "jlong_arraycopy");
duke@435	2838
coleenp@548	2839
coleenp@548	2840	if (UseCompressedOops) {
iveresov@2595	2841	StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, true, &entry,
iveresov@2595	2842	"oop_disjoint_arraycopy");
iveresov@2595	2843	StubRoutines::_oop_arraycopy = generate_conjoint_int_oop_copy(false, true, entry,
iveresov@2595	2844	&entry_oop_arraycopy, "oop_arraycopy");
iveresov@2606	2845	StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_int_oop_copy(false, true, &entry,
iveresov@2606	2846	"oop_disjoint_arraycopy_uninit",
iveresov@2606	2847	/*dest_uninitialized*/true);
iveresov@2606	2848	StubRoutines::_oop_arraycopy_uninit = generate_conjoint_int_oop_copy(false, true, entry,
iveresov@2606	2849	NULL, "oop_arraycopy_uninit",
iveresov@2606	2850	/*dest_uninitialized*/true);
coleenp@548	2851	} else {
iveresov@2595	2852	StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, true, &entry,
iveresov@2595	2853	"oop_disjoint_arraycopy");
iveresov@2595	2854	StubRoutines::_oop_arraycopy = generate_conjoint_long_oop_copy(false, true, entry,
iveresov@2595	2855	&entry_oop_arraycopy, "oop_arraycopy");
iveresov@2606	2856	StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_long_oop_copy(false, true, &entry,
iveresov@2606	2857	"oop_disjoint_arraycopy_uninit",
iveresov@2606	2858	/*dest_uninitialized*/true);
iveresov@2606	2859	StubRoutines::_oop_arraycopy_uninit = generate_conjoint_long_oop_copy(false, true, entry,
iveresov@2606	2860	NULL, "oop_arraycopy_uninit",
iveresov@2606	2861	/*dest_uninitialized*/true);
coleenp@548	2862	}
duke@435	2863
iveresov@2606	2864	StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
iveresov@2606	2865	StubRoutines::_checkcast_arraycopy_uninit = generate_checkcast_copy("checkcast_arraycopy_uninit", NULL,
iveresov@2606	2866	/*dest_uninitialized*/true);
iveresov@2606	2867
iveresov@2595	2868	StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy",
iveresov@2595	2869	entry_jbyte_arraycopy,
iveresov@2595	2870	entry_jshort_arraycopy,
iveresov@2595	2871	entry_jint_arraycopy,
iveresov@2595	2872	entry_jlong_arraycopy);
iveresov@2595	2873	StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy",
iveresov@2595	2874	entry_jbyte_arraycopy,
iveresov@2595	2875	entry_jshort_arraycopy,
iveresov@2595	2876	entry_jint_arraycopy,
iveresov@2595	2877	entry_oop_arraycopy,
iveresov@2595	2878	entry_jlong_arraycopy,
iveresov@2595	2879	entry_checkcast_arraycopy);
duke@435	2880
never@2118	2881	StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
never@2118	2882	StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
never@2118	2883	StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
never@2118	2884	StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
never@2118	2885	StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
never@2118	2886	StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
never@2118	2887
duke@435	2888	// We don't generate specialized code for HeapWord-aligned source
duke@435	2889	// arrays, so just use the code we've already generated
duke@435	2890	StubRoutines::_arrayof_jbyte_disjoint_arraycopy = StubRoutines::_jbyte_disjoint_arraycopy;
duke@435	2891	StubRoutines::_arrayof_jbyte_arraycopy = StubRoutines::_jbyte_arraycopy;
duke@435	2892
duke@435	2893	StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy;
duke@435	2894	StubRoutines::_arrayof_jshort_arraycopy = StubRoutines::_jshort_arraycopy;
duke@435	2895
duke@435	2896	StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
duke@435	2897	StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
duke@435	2898
duke@435	2899	StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
duke@435	2900	StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
duke@435	2901
duke@435	2902	StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
duke@435	2903	StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
iveresov@2606	2904
iveresov@2606	2905	StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
iveresov@2606	2906	StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
duke@435	2907	}
duke@435	2908
never@1609	2909	void generate_math_stubs() {
never@1609	2910	{
never@1609	2911	StubCodeMark mark(this, "StubRoutines", "log");
never@1609	2912	StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
never@1609	2913
never@1609	2914	__ subq(rsp, 8);
never@1609	2915	__ movdbl(Address(rsp, 0), xmm0);
never@1609	2916	__ fld_d(Address(rsp, 0));
never@1609	2917	__ flog();
never@1609	2918	__ fstp_d(Address(rsp, 0));
never@1609	2919	__ movdbl(xmm0, Address(rsp, 0));
never@1609	2920	__ addq(rsp, 8);
never@1609	2921	__ ret(0);
never@1609	2922	}
never@1609	2923	{
never@1609	2924	StubCodeMark mark(this, "StubRoutines", "log10");
never@1609	2925	StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
never@1609	2926
never@1609	2927	__ subq(rsp, 8);
never@1609	2928	__ movdbl(Address(rsp, 0), xmm0);
never@1609	2929	__ fld_d(Address(rsp, 0));
never@1609	2930	__ flog10();
never@1609	2931	__ fstp_d(Address(rsp, 0));
never@1609	2932	__ movdbl(xmm0, Address(rsp, 0));
never@1609	2933	__ addq(rsp, 8);
never@1609	2934	__ ret(0);
never@1609	2935	}
never@1609	2936	{
never@1609	2937	StubCodeMark mark(this, "StubRoutines", "sin");
never@1609	2938	StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc();
never@1609	2939
never@1609	2940	__ subq(rsp, 8);
never@1609	2941	__ movdbl(Address(rsp, 0), xmm0);
never@1609	2942	__ fld_d(Address(rsp, 0));
never@1609	2943	__ trigfunc('s');
never@1609	2944	__ fstp_d(Address(rsp, 0));
never@1609	2945	__ movdbl(xmm0, Address(rsp, 0));
never@1609	2946	__ addq(rsp, 8);
never@1609	2947	__ ret(0);
never@1609	2948	}
never@1609	2949	{
never@1609	2950	StubCodeMark mark(this, "StubRoutines", "cos");
never@1609	2951	StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
never@1609	2952
never@1609	2953	__ subq(rsp, 8);
never@1609	2954	__ movdbl(Address(rsp, 0), xmm0);
never@1609	2955	__ fld_d(Address(rsp, 0));
never@1609	2956	__ trigfunc('c');
never@1609	2957	__ fstp_d(Address(rsp, 0));
never@1609	2958	__ movdbl(xmm0, Address(rsp, 0));
never@1609	2959	__ addq(rsp, 8);
never@1609	2960	__ ret(0);
never@1609	2961	}
never@1609	2962	{
never@1609	2963	StubCodeMark mark(this, "StubRoutines", "tan");
never@1609	2964	StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
never@1609	2965
never@1609	2966	__ subq(rsp, 8);
never@1609	2967	__ movdbl(Address(rsp, 0), xmm0);
never@1609	2968	__ fld_d(Address(rsp, 0));
never@1609	2969	__ trigfunc('t');
never@1609	2970	__ fstp_d(Address(rsp, 0));
never@1609	2971	__ movdbl(xmm0, Address(rsp, 0));
never@1609	2972	__ addq(rsp, 8);
never@1609	2973	__ ret(0);
never@1609	2974	}
roland@3787	2975	{
roland@3787	2976	StubCodeMark mark(this, "StubRoutines", "exp");
roland@3787	2977	StubRoutines::_intrinsic_exp = (double (*)(double)) __ pc();
roland@3787	2978
roland@3787	2979	__ subq(rsp, 8);
roland@3787	2980	__ movdbl(Address(rsp, 0), xmm0);
roland@3787	2981	__ fld_d(Address(rsp, 0));
roland@3787	2982	__ exp_with_fallback(0);
roland@3787	2983	__ fstp_d(Address(rsp, 0));
roland@3787	2984	__ movdbl(xmm0, Address(rsp, 0));
roland@3787	2985	__ addq(rsp, 8);
roland@3787	2986	__ ret(0);
roland@3787	2987	}
roland@3787	2988	{
roland@3787	2989	StubCodeMark mark(this, "StubRoutines", "pow");
roland@3787	2990	StubRoutines::_intrinsic_pow = (double (*)(double,double)) __ pc();
roland@3787	2991
roland@3787	2992	__ subq(rsp, 8);
roland@3787	2993	__ movdbl(Address(rsp, 0), xmm1);
roland@3787	2994	__ fld_d(Address(rsp, 0));
roland@3787	2995	__ movdbl(Address(rsp, 0), xmm0);
roland@3787	2996	__ fld_d(Address(rsp, 0));
roland@3787	2997	__ pow_with_fallback(0);
roland@3787	2998	__ fstp_d(Address(rsp, 0));
roland@3787	2999	__ movdbl(xmm0, Address(rsp, 0));
roland@3787	3000	__ addq(rsp, 8);
roland@3787	3001	__ ret(0);
roland@3787	3002	}
never@1609	3003	}
never@1609	3004
kvn@4205	3005	// AES intrinsic stubs
kvn@4205	3006	enum {AESBlockSize = 16};
kvn@4205	3007
kvn@4205	3008	address generate_key_shuffle_mask() {
kvn@4205	3009	__ align(16);
kvn@4205	3010	StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
kvn@4205	3011	address start = __ pc();
kvn@4205	3012	__ emit_data64( 0x0405060700010203, relocInfo::none );
kvn@4205	3013	__ emit_data64( 0x0c0d0e0f08090a0b, relocInfo::none );
kvn@4205	3014	return start;
kvn@4205	3015	}
kvn@4205	3016
kvn@4205	3017	// Utility routine for loading a 128-bit key word in little endian format
kvn@4205	3018	// can optionally specify that the shuffle mask is already in an xmmregister
kvn@4205	3019	void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
kvn@4205	3020	__ movdqu(xmmdst, Address(key, offset));
kvn@4205	3021	if (xmm_shuf_mask != NULL) {
kvn@4205	3022	__ pshufb(xmmdst, xmm_shuf_mask);
kvn@4205	3023	} else {
kvn@4205	3024	__ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
kvn@4205	3025	}
kvn@4205	3026	}
kvn@4205	3027
kvn@4205	3028	// Arguments:
kvn@4205	3029	//
kvn@4205	3030	// Inputs:
kvn@4205	3031	// c_rarg0 - source byte array address
kvn@4205	3032	// c_rarg1 - destination byte array address
kvn@4205	3033	// c_rarg2 - K (key) in little endian int array
kvn@4205	3034	//
kvn@4205	3035	address generate_aescrypt_encryptBlock() {
kvn@4363	3036	assert(UseAES, "need AES instructions and misaligned SSE support");
kvn@4205	3037	__ align(CodeEntryAlignment);
kvn@4205	3038	StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
kvn@4205	3039	Label L_doLast;
kvn@4205	3040	address start = __ pc();
kvn@4205	3041
kvn@4205	3042	const Register from = c_rarg0; // source array address
kvn@4205	3043	const Register to = c_rarg1; // destination array address
kvn@4205	3044	const Register key = c_rarg2; // key array address
kvn@4205	3045	const Register keylen = rax;
kvn@4205	3046
kvn@4205	3047	const XMMRegister xmm_result = xmm0;
kvn@4363	3048	const XMMRegister xmm_key_shuf_mask = xmm1;
kvn@4363	3049	// On win64 xmm6-xmm15 must be preserved so don't use them.
kvn@4363	3050	const XMMRegister xmm_temp1 = xmm2;
kvn@4363	3051	const XMMRegister xmm_temp2 = xmm3;
kvn@4363	3052	const XMMRegister xmm_temp3 = xmm4;
kvn@4363	3053	const XMMRegister xmm_temp4 = xmm5;
kvn@4205	3054
kvn@4205	3055	__ enter(); // required for proper stackwalking of RuntimeStub frame
kvn@4205	3056
kvn@4363	3057	// keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
kvn@4205	3058	__ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
kvn@4205	3059
kvn@4205	3060	__ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
kvn@4205	3061	__ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input
kvn@4205	3062
kvn@4205	3063	// For encryption, the java expanded key ordering is just what we need
kvn@4205	3064	// we don't know if the key is aligned, hence not using load-execute form
kvn@4205	3065
kvn@4363	3066	load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
kvn@4363	3067	__ pxor(xmm_result, xmm_temp1);
kvn@4363	3068
kvn@4363	3069	load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
kvn@4363	3070	load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
kvn@4363	3071	load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
kvn@4363	3072	load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
kvn@4363	3073
kvn@4363	3074	__ aesenc(xmm_result, xmm_temp1);
kvn@4363	3075	__ aesenc(xmm_result, xmm_temp2);
kvn@4363	3076	__ aesenc(xmm_result, xmm_temp3);
kvn@4363	3077	__ aesenc(xmm_result, xmm_temp4);
kvn@4363	3078
kvn@4363	3079	load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
kvn@4363	3080	load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
kvn@4363	3081	load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
kvn@4363	3082	load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
kvn@4363	3083
kvn@4363	3084	__ aesenc(xmm_result, xmm_temp1);
kvn@4363	3085	__ aesenc(xmm_result, xmm_temp2);
kvn@4363	3086	__ aesenc(xmm_result, xmm_temp3);
kvn@4363	3087	__ aesenc(xmm_result, xmm_temp4);
kvn@4363	3088
kvn@4363	3089	load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
kvn@4363	3090	load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
kvn@4363	3091
kvn@4363	3092	__ cmpl(keylen, 44);
kvn@4363	3093	__ jccb(Assembler::equal, L_doLast);
kvn@4363	3094
kvn@4363	3095	__ aesenc(xmm_result, xmm_temp1);
kvn@4363	3096	__ aesenc(xmm_result, xmm_temp2);
kvn@4363	3097
kvn@4363	3098	load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
kvn@4363	3099	load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
kvn@4363	3100
kvn@4363	3101	__ cmpl(keylen, 52);
kvn@4363	3102	__ jccb(Assembler::equal, L_doLast);
kvn@4363	3103
kvn@4363	3104	__ aesenc(xmm_result, xmm_temp1);
kvn@4363	3105	__ aesenc(xmm_result, xmm_temp2);
kvn@4363	3106
kvn@4363	3107	load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
kvn@4363	3108	load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
kvn@4205	3109
kvn@4205	3110	__ BIND(L_doLast);
kvn@4363	3111	__ aesenc(xmm_result, xmm_temp1);
kvn@4363	3112	__ aesenclast(xmm_result, xmm_temp2);
kvn@4205	3113	__ movdqu(Address(to, 0), xmm_result); // store the result
kvn@4205	3114	__ xorptr(rax, rax); // return 0
kvn@4205	3115	__ leave(); // required for proper stackwalking of RuntimeStub frame
kvn@4205	3116	__ ret(0);
kvn@4205	3117
kvn@4205	3118	return start;
kvn@4205	3119	}
kvn@4205	3120
kvn@4205	3121
kvn@4205	3122	// Arguments:
kvn@4205	3123	//
kvn@4205	3124	// Inputs:
kvn@4205	3125	// c_rarg0 - source byte array address
kvn@4205	3126	// c_rarg1 - destination byte array address
kvn@4205	3127	// c_rarg2 - K (key) in little endian int array
kvn@4205	3128	//
kvn@4205	3129	address generate_aescrypt_decryptBlock() {
kvn@4363	3130	assert(UseAES, "need AES instructions and misaligned SSE support");
kvn@4205	3131	__ align(CodeEntryAlignment);
kvn@4205	3132	StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
kvn@4205	3133	Label L_doLast;
kvn@4205	3134	address start = __ pc();
kvn@4205	3135
kvn@4205	3136	const Register from = c_rarg0; // source array address
kvn@4205	3137	const Register to = c_rarg1; // destination array address
kvn@4205	3138	const Register key = c_rarg2; // key array address
kvn@4205	3139	const Register keylen = rax;
kvn@4205	3140
kvn@4205	3141	const XMMRegister xmm_result = xmm0;
kvn@4363	3142	const XMMRegister xmm_key_shuf_mask = xmm1;
kvn@4363	3143	// On win64 xmm6-xmm15 must be preserved so don't use them.
kvn@4363	3144	const XMMRegister xmm_temp1 = xmm2;
kvn@4363	3145	const XMMRegister xmm_temp2 = xmm3;
kvn@4363	3146	const XMMRegister xmm_temp3 = xmm4;
kvn@4363	3147	const XMMRegister xmm_temp4 = xmm5;
kvn@4205	3148
kvn@4205	3149	__ enter(); // required for proper stackwalking of RuntimeStub frame
kvn@4205	3150
kvn@4363	3151	// keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
kvn@4205	3152	__ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
kvn@4205	3153
kvn@4205	3154	__ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
kvn@4205	3155	__ movdqu(xmm_result, Address(from, 0));
kvn@4205	3156
kvn@4205	3157	// for decryption java expanded key ordering is rotated one position from what we want
kvn@4205	3158	// so we start from 0x10 here and hit 0x00 last
kvn@4205	3159	// we don't know if the key is aligned, hence not using load-execute form
kvn@4363	3160	load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
kvn@4363	3161	load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
kvn@4363	3162	load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
kvn@4363	3163	load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
kvn@4363	3164
kvn@4363	3165	__ pxor (xmm_result, xmm_temp1);
kvn@4363	3166	__ aesdec(xmm_result, xmm_temp2);
kvn@4363	3167	__ aesdec(xmm_result, xmm_temp3);
kvn@4363	3168	__ aesdec(xmm_result, xmm_temp4);
kvn@4363	3169
kvn@4363	3170	load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
kvn@4363	3171	load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
kvn@4363	3172	load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
kvn@4363	3173	load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
kvn@4363	3174
kvn@4363	3175	__ aesdec(xmm_result, xmm_temp1);
kvn@4363	3176	__ aesdec(xmm_result, xmm_temp2);
kvn@4363	3177	__ aesdec(xmm_result, xmm_temp3);
kvn@4363	3178	__ aesdec(xmm_result, xmm_temp4);
kvn@4363	3179
kvn@4363	3180	load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
kvn@4363	3181	load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
kvn@4363	3182	load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
kvn@4363	3183
kvn@4363	3184	__ cmpl(keylen, 44);
kvn@4363	3185	__ jccb(Assembler::equal, L_doLast);
kvn@4363	3186
kvn@4363	3187	__ aesdec(xmm_result, xmm_temp1);
kvn@4363	3188	__ aesdec(xmm_result, xmm_temp2);
kvn@4363	3189
kvn@4363	3190	load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
kvn@4363	3191	load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
kvn@4363	3192
kvn@4363	3193	__ cmpl(keylen, 52);
kvn@4363	3194	__ jccb(Assembler::equal, L_doLast);
kvn@4363	3195
kvn@4363	3196	__ aesdec(xmm_result, xmm_temp1);
kvn@4363	3197	__ aesdec(xmm_result, xmm_temp2);
kvn@4363	3198
kvn@4363	3199	load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
kvn@4363	3200	load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
kvn@4205	3201
kvn@4205	3202	__ BIND(L_doLast);
kvn@4363	3203	__ aesdec(xmm_result, xmm_temp1);
kvn@4363	3204	__ aesdec(xmm_result, xmm_temp2);
kvn@4363	3205
kvn@4205	3206	// for decryption the aesdeclast operation is always on key+0x00
kvn@4363	3207	__ aesdeclast(xmm_result, xmm_temp3);
kvn@4205	3208	__ movdqu(Address(to, 0), xmm_result); // store the result
kvn@4205	3209	__ xorptr(rax, rax); // return 0
kvn@4205	3210	__ leave(); // required for proper stackwalking of RuntimeStub frame
kvn@4205	3211	__ ret(0);
kvn@4205	3212
kvn@4205	3213	return start;
kvn@4205	3214	}
kvn@4205	3215
kvn@4205	3216
kvn@4205	3217	// Arguments:
kvn@4205	3218	//
kvn@4205	3219	// Inputs:
kvn@4205	3220	// c_rarg0 - source byte array address
kvn@4205	3221	// c_rarg1 - destination byte array address
kvn@4205	3222	// c_rarg2 - K (key) in little endian int array
kvn@4205	3223	// c_rarg3 - r vector byte array address
kvn@4205	3224	// c_rarg4 - input length
kvn@4205	3225	//
kvn@4205	3226	address generate_cipherBlockChaining_encryptAESCrypt() {
kvn@4363	3227	assert(UseAES, "need AES instructions and misaligned SSE support");
kvn@4205	3228	__ align(CodeEntryAlignment);
kvn@4205	3229	StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
kvn@4205	3230	address start = __ pc();
kvn@4205	3231
kvn@4205	3232	Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
kvn@4205	3233	const Register from = c_rarg0; // source array address
kvn@4205	3234	const Register to = c_rarg1; // destination array address
kvn@4205	3235	const Register key = c_rarg2; // key array address
kvn@4205	3236	const Register rvec = c_rarg3; // r byte array initialized from initvector array address
kvn@4205	3237	// and left with the results of the last encryption block
kvn@4205	3238	#ifndef _WIN64
kvn@4205	3239	const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
kvn@4205	3240	#else
kvn@4205	3241	const Address len_mem(rsp, 6 * wordSize); // length is on stack on Win64
kvn@4205	3242	const Register len_reg = r10; // pick the first volatile windows register
kvn@4205	3243	#endif
kvn@4205	3244	const Register pos = rax;
kvn@4205	3245
kvn@4205	3246	// xmm register assignments for the loops below
kvn@4205	3247	const XMMRegister xmm_result = xmm0;
kvn@4205	3248	const XMMRegister xmm_temp = xmm1;
kvn@4205	3249	// keys 0-10 preloaded into xmm2-xmm12
kvn@4205	3250	const int XMM_REG_NUM_KEY_FIRST = 2;
kvn@4363	3251	const int XMM_REG_NUM_KEY_LAST = 15;
kvn@4205	3252	const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
kvn@4363	3253	const XMMRegister xmm_key10 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+10);
kvn@4363	3254	const XMMRegister xmm_key11 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+11);
kvn@4363	3255	const XMMRegister xmm_key12 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+12);
kvn@4363	3256	const XMMRegister xmm_key13 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+13);
kvn@4205	3257
kvn@4205	3258	__ enter(); // required for proper stackwalking of RuntimeStub frame
kvn@4205	3259
kvn@4205	3260	#ifdef _WIN64
kvn@4205	3261	// on win64, fill len_reg from stack position
kvn@4205	3262	__ movl(len_reg, len_mem);
kvn@4363	3263	// save the xmm registers which must be preserved 6-15
kvn@4205	3264	__ subptr(rsp, -rsp_after_call_off * wordSize);
kvn@4205	3265	for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) {
kvn@4205	3266	__ movdqu(xmm_save(i), as_XMMRegister(i));
kvn@4205	3267	}
kvn@4205	3268	#endif
kvn@4205	3269
kvn@4205	3270	const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front
kvn@4205	3271	__ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
kvn@4363	3272	// load up xmm regs xmm2 thru xmm12 with key 0x00 - 0xa0
kvn@4363	3273	for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_FIRST+10; rnum++) {
kvn@4205	3274	load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
kvn@4205	3275	offset += 0x10;
kvn@4205	3276	}
kvn@4205	3277	__ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec
kvn@4205	3278
kvn@4205	3279	// now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
kvn@4205	3280	__ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
kvn@4205	3281	__ cmpl(rax, 44);
kvn@4205	3282	__ jcc(Assembler::notEqual, L_key_192_256);
kvn@4205	3283
kvn@4205	3284	// 128 bit code follows here
kvn@4205	3285	__ movptr(pos, 0);
kvn@4205	3286	__ align(OptoLoopAlignment);
kvn@4363	3287
kvn@4205	3288	__ BIND(L_loopTop_128);
kvn@4205	3289	__ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
kvn@4205	3290	__ pxor (xmm_result, xmm_temp); // xor with the current r vector
kvn@4205	3291	__ pxor (xmm_result, xmm_key0); // do the aes rounds
kvn@4363	3292	for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 9; rnum++) {
kvn@4205	3293	__ aesenc(xmm_result, as_XMMRegister(rnum));
kvn@4205	3294	}
kvn@4205	3295	__ aesenclast(xmm_result, xmm_key10);
kvn@4205	3296	__ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
kvn@4205	3297	// no need to store r to memory until we exit
kvn@4205	3298	__ addptr(pos, AESBlockSize);
kvn@4205	3299	__ subptr(len_reg, AESBlockSize);
kvn@4205	3300	__ jcc(Assembler::notEqual, L_loopTop_128);
kvn@4205	3301
kvn@4205	3302	__ BIND(L_exit);
kvn@4205	3303	__ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object
kvn@4205	3304
kvn@4205	3305	#ifdef _WIN64
kvn@4205	3306	// restore xmm regs belonging to calling function
kvn@4205	3307	for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) {
kvn@4205	3308	__ movdqu(as_XMMRegister(i), xmm_save(i));
kvn@4205	3309	}
kvn@4205	3310	#endif
kvn@4205	3311	__ movl(rax, 0); // return 0 (why?)
kvn@4205	3312	__ leave(); // required for proper stackwalking of RuntimeStub frame
kvn@4205	3313	__ ret(0);
kvn@4205	3314
kvn@4205	3315	__ BIND(L_key_192_256);
kvn@4205	3316	// here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
kvn@4363	3317	load_key(xmm_key11, key, 0xb0, xmm_key_shuf_mask);
kvn@4363	3318	load_key(xmm_key12, key, 0xc0, xmm_key_shuf_mask);
kvn@4205	3319	__ cmpl(rax, 52);
kvn@4205	3320	__ jcc(Assembler::notEqual, L_key_256);
kvn@4205	3321
kvn@4205	3322	// 192-bit code follows here (could be changed to use more xmm registers)
kvn@4205	3323	__ movptr(pos, 0);
kvn@4205	3324	__ align(OptoLoopAlignment);
kvn@4363	3325
kvn@4205	3326	__ BIND(L_loopTop_192);
kvn@4205	3327	__ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
kvn@4205	3328	__ pxor (xmm_result, xmm_temp); // xor with the current r vector
kvn@4205	3329	__ pxor (xmm_result, xmm_key0); // do the aes rounds
kvn@4363	3330	for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 11; rnum++) {
kvn@4205	3331	__ aesenc(xmm_result, as_XMMRegister(rnum));
kvn@4205	3332	}
kvn@4363	3333	__ aesenclast(xmm_result, xmm_key12);
kvn@4205	3334	__ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
kvn@4205	3335	// no need to store r to memory until we exit
kvn@4205	3336	__ addptr(pos, AESBlockSize);
kvn@4205	3337	__ subptr(len_reg, AESBlockSize);
kvn@4205	3338	__ jcc(Assembler::notEqual, L_loopTop_192);
kvn@4205	3339	__ jmp(L_exit);
kvn@4205	3340
kvn@4205	3341	__ BIND(L_key_256);
kvn@4205	3342	// 256-bit code follows here (could be changed to use more xmm registers)
kvn@4363	3343	load_key(xmm_key13, key, 0xd0, xmm_key_shuf_mask);
kvn@4205	3344	__ movptr(pos, 0);
kvn@4205	3345	__ align(OptoLoopAlignment);
kvn@4363	3346
kvn@4205	3347	__ BIND(L_loopTop_256);
kvn@4205	3348	__ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
kvn@4205	3349	__ pxor (xmm_result, xmm_temp); // xor with the current r vector
kvn@4205	3350	__ pxor (xmm_result, xmm_key0); // do the aes rounds
kvn@4363	3351	for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 13; rnum++) {
kvn@4205	3352	__ aesenc(xmm_result, as_XMMRegister(rnum));
kvn@4205	3353	}
kvn@4205	3354	load_key(xmm_temp, key, 0xe0);
kvn@4205	3355	__ aesenclast(xmm_result, xmm_temp);
kvn@4205	3356	__ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
kvn@4205	3357	// no need to store r to memory until we exit
kvn@4205	3358	__ addptr(pos, AESBlockSize);
kvn@4205	3359	__ subptr(len_reg, AESBlockSize);
kvn@4205	3360	__ jcc(Assembler::notEqual, L_loopTop_256);
kvn@4205	3361	__ jmp(L_exit);
kvn@4205	3362
kvn@4205	3363	return start;
kvn@4205	3364	}
kvn@4205	3365
kvn@4205	3366
kvn@4205	3367
kvn@4205	3368	// This is a version of CBC/AES Decrypt which does 4 blocks in a loop at a time
kvn@4205	3369	// to hide instruction latency
kvn@4205	3370	//
kvn@4205	3371	// Arguments:
kvn@4205	3372	//
kvn@4205	3373	// Inputs:
kvn@4205	3374	// c_rarg0 - source byte array address
kvn@4205	3375	// c_rarg1 - destination byte array address
kvn@4205	3376	// c_rarg2 - K (key) in little endian int array
kvn@4205	3377	// c_rarg3 - r vector byte array address
kvn@4205	3378	// c_rarg4 - input length
kvn@4205	3379	//
kvn@4205	3380
kvn@4205	3381	address generate_cipherBlockChaining_decryptAESCrypt_Parallel() {
kvn@4363	3382	assert(UseAES, "need AES instructions and misaligned SSE support");
kvn@4205	3383	__ align(CodeEntryAlignment);
kvn@4205	3384	StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
kvn@4205	3385	address start = __ pc();
kvn@4205	3386
kvn@4205	3387	Label L_exit, L_key_192_256, L_key_256;
kvn@4205	3388	Label L_singleBlock_loopTop_128, L_multiBlock_loopTop_128;
kvn@4205	3389	Label L_singleBlock_loopTop_192, L_singleBlock_loopTop_256;
kvn@4205	3390	const Register from = c_rarg0; // source array address
kvn@4205	3391	const Register to = c_rarg1; // destination array address
kvn@4205	3392	const Register key = c_rarg2; // key array address
kvn@4205	3393	const Register rvec = c_rarg3; // r byte array initialized from initvector array address
kvn@4205	3394	// and left with the results of the last encryption block
kvn@4205	3395	#ifndef _WIN64
kvn@4205	3396	const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
kvn@4205	3397	#else
kvn@4205	3398	const Address len_mem(rsp, 6 * wordSize); // length is on stack on Win64
kvn@4205	3399	const Register len_reg = r10; // pick the first volatile windows register
kvn@4205	3400	#endif
kvn@4205	3401	const Register pos = rax;
kvn@4205	3402
kvn@4205	3403	// keys 0-10 preloaded into xmm2-xmm12
kvn@4205	3404	const int XMM_REG_NUM_KEY_FIRST = 5;
kvn@4205	3405	const int XMM_REG_NUM_KEY_LAST = 15;
kvn@4363	3406	const XMMRegister xmm_key_first = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
kvn@4205	3407	const XMMRegister xmm_key_last = as_XMMRegister(XMM_REG_NUM_KEY_LAST);
kvn@4205	3408
kvn@4205	3409	__ enter(); // required for proper stackwalking of RuntimeStub frame
kvn@4205	3410
kvn@4205	3411	#ifdef _WIN64
kvn@4205	3412	// on win64, fill len_reg from stack position
kvn@4205	3413	__ movl(len_reg, len_mem);
kvn@4205	3414	// save the xmm registers which must be preserved 6-15
kvn@4205	3415	__ subptr(rsp, -rsp_after_call_off * wordSize);
kvn@4205	3416	for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) {
kvn@4205	3417	__ movdqu(xmm_save(i), as_XMMRegister(i));
kvn@4205	3418	}
kvn@4205	3419	#endif
kvn@4205	3420	// the java expanded key ordering is rotated one position from what we want
kvn@4205	3421	// so we start from 0x10 here and hit 0x00 last
kvn@4205	3422	const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front
kvn@4205	3423	__ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
kvn@4205	3424	// load up xmm regs 5 thru 15 with key 0x10 - 0xa0 - 0x00
kvn@4363	3425	for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum < XMM_REG_NUM_KEY_LAST; rnum++) {
kvn@4205	3426	load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
kvn@4205	3427	offset += 0x10;
kvn@4205	3428	}
kvn@4363	3429	load_key(xmm_key_last, key, 0x00, xmm_key_shuf_mask);
kvn@4205	3430
kvn@4205	3431	const XMMRegister xmm_prev_block_cipher = xmm1; // holds cipher of previous block
kvn@4363	3432
kvn@4205	3433	// registers holding the four results in the parallelized loop
kvn@4205	3434	const XMMRegister xmm_result0 = xmm0;
kvn@4205	3435	const XMMRegister xmm_result1 = xmm2;
kvn@4205	3436	const XMMRegister xmm_result2 = xmm3;
kvn@4205	3437	const XMMRegister xmm_result3 = xmm4;
kvn@4205	3438
kvn@4205	3439	__ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // initialize with initial rvec
kvn@4205	3440
kvn@4205	3441	// now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
kvn@4205	3442	__ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
kvn@4205	3443	__ cmpl(rax, 44);
kvn@4205	3444	__ jcc(Assembler::notEqual, L_key_192_256);
kvn@4205	3445
kvn@4205	3446
kvn@4205	3447	// 128-bit code follows here, parallelized
kvn@4205	3448	__ movptr(pos, 0);
kvn@4205	3449	__ align(OptoLoopAlignment);
kvn@4205	3450	__ BIND(L_multiBlock_loopTop_128);
kvn@4205	3451	__ cmpptr(len_reg, 4*AESBlockSize); // see if at least 4 blocks left
kvn@4205	3452	__ jcc(Assembler::less, L_singleBlock_loopTop_128);
kvn@4205	3453
kvn@4205	3454	__ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0*AESBlockSize)); // get next 4 blocks into xmmresult registers
kvn@4205	3455	__ movdqu(xmm_result1, Address(from, pos, Address::times_1, 1*AESBlockSize));
kvn@4205	3456	__ movdqu(xmm_result2, Address(from, pos, Address::times_1, 2*AESBlockSize));
kvn@4205	3457	__ movdqu(xmm_result3, Address(from, pos, Address::times_1, 3*AESBlockSize));
kvn@4205	3458
kvn@4205	3459	#define DoFour(opc, src_reg) \
kvn@4205	3460	__ opc(xmm_result0, src_reg); \
kvn@4205	3461	__ opc(xmm_result1, src_reg); \
kvn@4205	3462	__ opc(xmm_result2, src_reg); \
kvn@4205	3463	__ opc(xmm_result3, src_reg);
kvn@4205	3464
kvn@4205	3465	DoFour(pxor, xmm_key_first);
kvn@4205	3466	for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) {
kvn@4205	3467	DoFour(aesdec, as_XMMRegister(rnum));
kvn@4205	3468	}
kvn@4205	3469	DoFour(aesdeclast, xmm_key_last);
kvn@4205	3470	// for each result, xor with the r vector of previous cipher block
kvn@4205	3471	__ pxor(xmm_result0, xmm_prev_block_cipher);
kvn@4205	3472	__ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 0*AESBlockSize));
kvn@4205	3473	__ pxor(xmm_result1, xmm_prev_block_cipher);
kvn@4205	3474	__ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 1*AESBlockSize));
kvn@4205	3475	__ pxor(xmm_result2, xmm_prev_block_cipher);
kvn@4205	3476	__ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 2*AESBlockSize));
kvn@4205	3477	__ pxor(xmm_result3, xmm_prev_block_cipher);
kvn@4205	3478	__ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 3*AESBlockSize)); // this will carry over to next set of blocks
kvn@4205	3479
kvn@4205	3480	__ movdqu(Address(to, pos, Address::times_1, 0*AESBlockSize), xmm_result0); // store 4 results into the next 64 bytes of output
kvn@4205	3481	__ movdqu(Address(to, pos, Address::times_1, 1*AESBlockSize), xmm_result1);
kvn@4205	3482	__ movdqu(Address(to, pos, Address::times_1, 2*AESBlockSize), xmm_result2);
kvn@4205	3483	__ movdqu(Address(to, pos, Address::times_1, 3*AESBlockSize), xmm_result3);
kvn@4205	3484
kvn@4205	3485	__ addptr(pos, 4*AESBlockSize);
kvn@4205	3486	__ subptr(len_reg, 4*AESBlockSize);
kvn@4205	3487	__ jmp(L_multiBlock_loopTop_128);
kvn@4205	3488
kvn@4205	3489	// registers used in the non-parallelized loops
kvn@4363	3490	// xmm register assignments for the loops below
kvn@4363	3491	const XMMRegister xmm_result = xmm0;
kvn@4205	3492	const XMMRegister xmm_prev_block_cipher_save = xmm2;
kvn@4363	3493	const XMMRegister xmm_key11 = xmm3;
kvn@4363	3494	const XMMRegister xmm_key12 = xmm4;
kvn@4363	3495	const XMMRegister xmm_temp = xmm4;
kvn@4205	3496
kvn@4205	3497	__ align(OptoLoopAlignment);
kvn@4205	3498	__ BIND(L_singleBlock_loopTop_128);
kvn@4205	3499	__ cmpptr(len_reg, 0); // any blocks left??
kvn@4205	3500	__ jcc(Assembler::equal, L_exit);
kvn@4205	3501	__ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
kvn@4205	3502	__ movdqa(xmm_prev_block_cipher_save, xmm_result); // save for next r vector
kvn@4205	3503	__ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
kvn@4205	3504	for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) {
kvn@4205	3505	__ aesdec(xmm_result, as_XMMRegister(rnum));
kvn@4205	3506	}
kvn@4205	3507	__ aesdeclast(xmm_result, xmm_key_last);
kvn@4205	3508	__ pxor (xmm_result, xmm_prev_block_cipher); // xor with the current r vector
kvn@4205	3509	__ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
kvn@4205	3510	// no need to store r to memory until we exit
kvn@4205	3511	__ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save); // set up next r vector with cipher input from this block
kvn@4205	3512
kvn@4205	3513	__ addptr(pos, AESBlockSize);
kvn@4205	3514	__ subptr(len_reg, AESBlockSize);
kvn@4205	3515	__ jmp(L_singleBlock_loopTop_128);
kvn@4205	3516
kvn@4205	3517
kvn@4205	3518	__ BIND(L_exit);
kvn@4205	3519	__ movdqu(Address(rvec, 0), xmm_prev_block_cipher); // final value of r stored in rvec of CipherBlockChaining object
kvn@4205	3520	#ifdef _WIN64
kvn@4205	3521	// restore regs belonging to calling function
kvn@4205	3522	for (int i = 6; i <= XMM_REG_NUM_KEY_LAST; i++) {
kvn@4205	3523	__ movdqu(as_XMMRegister(i), xmm_save(i));
kvn@4205	3524	}
kvn@4205	3525	#endif
kvn@4205	3526	__ movl(rax, 0); // return 0 (why?)
kvn@4205	3527	__ leave(); // required for proper stackwalking of RuntimeStub frame
kvn@4205	3528	__ ret(0);
kvn@4205	3529
kvn@4205	3530
kvn@4205	3531	__ BIND(L_key_192_256);
kvn@4205	3532	// here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
kvn@4363	3533	load_key(xmm_key11, key, 0xb0);
kvn@4205	3534	__ cmpl(rax, 52);
kvn@4205	3535	__ jcc(Assembler::notEqual, L_key_256);
kvn@4205	3536
kvn@4205	3537	// 192-bit code follows here (could be optimized to use parallelism)
kvn@4363	3538	load_key(xmm_key12, key, 0xc0); // 192-bit key goes up to c0
kvn@4205	3539	__ movptr(pos, 0);
kvn@4205	3540	__ align(OptoLoopAlignment);
kvn@4363	3541
kvn@4205	3542	__ BIND(L_singleBlock_loopTop_192);
kvn@4205	3543	__ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
kvn@4205	3544	__ movdqa(xmm_prev_block_cipher_save, xmm_result); // save for next r vector
kvn@4205	3545	__ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
kvn@4205	3546	for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) {
kvn@4205	3547	__ aesdec(xmm_result, as_XMMRegister(rnum));
kvn@4205	3548	}
kvn@4363	3549	__ aesdec(xmm_result, xmm_key11);
kvn@4363	3550	__ aesdec(xmm_result, xmm_key12);
kvn@4205	3551	__ aesdeclast(xmm_result, xmm_key_last); // xmm15 always came from key+0
kvn@4205	3552	__ pxor (xmm_result, xmm_prev_block_cipher); // xor with the current r vector
kvn@4363	3553	__ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
kvn@4205	3554	// no need to store r to memory until we exit
kvn@4363	3555	__ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save); // set up next r vector with cipher input from this block
kvn@4205	3556	__ addptr(pos, AESBlockSize);
kvn@4205	3557	__ subptr(len_reg, AESBlockSize);
kvn@4205	3558	__ jcc(Assembler::notEqual,L_singleBlock_loopTop_192);
kvn@4205	3559	__ jmp(L_exit);
kvn@4205	3560
kvn@4205	3561	__ BIND(L_key_256);
kvn@4205	3562	// 256-bit code follows here (could be optimized to use parallelism)
kvn@4205	3563	__ movptr(pos, 0);
kvn@4205	3564	__ align(OptoLoopAlignment);
kvn@4363	3565
kvn@4205	3566	__ BIND(L_singleBlock_loopTop_256);
kvn@4363	3567	__ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
kvn@4205	3568	__ movdqa(xmm_prev_block_cipher_save, xmm_result); // save for next r vector
kvn@4205	3569	__ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
kvn@4205	3570	for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST - 1; rnum++) {
kvn@4205	3571	__ aesdec(xmm_result, as_XMMRegister(rnum));
kvn@4205	3572	}
kvn@4363	3573	__ aesdec(xmm_result, xmm_key11);
kvn@4363	3574	load_key(xmm_temp, key, 0xc0);
kvn@4363	3575	__ aesdec(xmm_result, xmm_temp);
kvn@4363	3576	load_key(xmm_temp, key, 0xd0);
kvn@4363	3577	__ aesdec(xmm_result, xmm_temp);
kvn@4363	3578	load_key(xmm_temp, key, 0xe0); // 256-bit key goes up to e0
kvn@4363	3579	__ aesdec(xmm_result, xmm_temp);
kvn@4363	3580	__ aesdeclast(xmm_result, xmm_key_last); // xmm15 came from key+0
kvn@4205	3581	__ pxor (xmm_result, xmm_prev_block_cipher); // xor with the current r vector
kvn@4363	3582	__ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
kvn@4205	3583	// no need to store r to memory until we exit
kvn@4363	3584	__ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save); // set up next r vector with cipher input from this block
kvn@4205	3585	__ addptr(pos, AESBlockSize);
kvn@4205	3586	__ subptr(len_reg, AESBlockSize);
kvn@4205	3587	__ jcc(Assembler::notEqual,L_singleBlock_loopTop_256);
kvn@4205	3588	__ jmp(L_exit);
kvn@4205	3589
kvn@4205	3590	return start;
kvn@4205	3591	}
kvn@4205	3592
kvn@4205	3593
kvn@4205	3594
duke@435	3595	#undef __
duke@435	3596	#define __ masm->
duke@435	3597
duke@435	3598	// Continuation point for throwing of implicit exceptions that are
duke@435	3599	// not handled in the current activation. Fabricates an exception
duke@435	3600	// oop and initiates normal exception dispatching in this
duke@435	3601	// frame. Since we need to preserve callee-saved values (currently
duke@435	3602	// only for C2, but done for C1 as well) we need a callee-saved oop
duke@435	3603	// map and therefore have to make these stubs into RuntimeStubs
duke@435	3604	// rather than BufferBlobs. If the compiler needs all registers to
duke@435	3605	// be preserved between the fault point and the exception handler
duke@435	3606	// then it must assume responsibility for that in
duke@435	3607	// AbstractCompiler::continuation_for_implicit_null_exception or
duke@435	3608	// continuation_for_implicit_division_by_zero_exception. All other
duke@435	3609	// implicit exceptions (e.g., NullPointerException or
duke@435	3610	// AbstractMethodError on entry) are either at call sites or
duke@435	3611	// otherwise assume that stack unwinding will be initiated, so
duke@435	3612	// caller saved registers were assumed volatile in the compiler.
duke@435	3613	address generate_throw_exception(const char* name,
duke@435	3614	address runtime_entry,
never@2978	3615	Register arg1 = noreg,
never@2978	3616	Register arg2 = noreg) {
duke@435	3617	// Information about frame layout at time of blocking runtime call.
duke@435	3618	// Note that we only have to preserve callee-saved registers since
duke@435	3619	// the compilers are responsible for supplying a continuation point
duke@435	3620	// if they expect all registers to be preserved.
duke@435	3621	enum layout {
duke@435	3622	rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
duke@435	3623	rbp_off2,
duke@435	3624	return_off,
duke@435	3625	return_off2,
duke@435	3626	framesize // inclusive of return address
duke@435	3627	};
duke@435	3628
duke@435	3629	int insts_size = 512;
duke@435	3630	int locs_size = 64;
duke@435	3631
duke@435	3632	CodeBuffer code(name, insts_size, locs_size);
duke@435	3633	OopMapSet* oop_maps = new OopMapSet();
duke@435	3634	MacroAssembler* masm = new MacroAssembler(&code);
duke@435	3635
duke@435	3636	address start = __ pc();
duke@435	3637
duke@435	3638	// This is an inlined and slightly modified version of call_VM
duke@435	3639	// which has the ability to fetch the return PC out of
duke@435	3640	// thread-local storage and also sets up last_Java_sp slightly
duke@435	3641	// differently than the real call_VM
duke@435	3642
duke@435	3643	__ enter(); // required for proper stackwalking of RuntimeStub frame
duke@435	3644
duke@435	3645	assert(is_even(framesize/2), "sp not 16-byte aligned");
duke@435	3646
duke@435	3647	// return address and rbp are already in place
never@739	3648	__ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
duke@435	3649
duke@435	3650	int frame_complete = __ pc() - start;
duke@435	3651
duke@435	3652	// Set up last_Java_sp and last_Java_fp
roland@3522	3653	address the_pc = __ pc();
roland@3522	3654	__ set_last_Java_frame(rsp, rbp, the_pc);
roland@3522	3655	__ andptr(rsp, -(StackAlignmentInBytes)); // Align stack
duke@435	3656
duke@435	3657	// Call runtime
never@2978	3658	if (arg1 != noreg) {
never@2978	3659	assert(arg2 != c_rarg1, "clobbered");
never@2978	3660	__ movptr(c_rarg1, arg1);
never@2978	3661	}
never@2978	3662	if (arg2 != noreg) {
never@2978	3663	__ movptr(c_rarg2, arg2);
never@2978	3664	}
never@739	3665	__ movptr(c_rarg0, r15_thread);
duke@435	3666	BLOCK_COMMENT("call runtime_entry");
duke@435	3667	__ call(RuntimeAddress(runtime_entry));
duke@435	3668
duke@435	3669	// Generate oop map
duke@435	3670	OopMap* map = new OopMap(framesize, 0);
duke@435	3671
roland@3568	3672	oop_maps->add_gc_map(the_pc - start, map);
duke@435	3673
roland@3522	3674	__ reset_last_Java_frame(true, true);
duke@435	3675
duke@435	3676	__ leave(); // required for proper stackwalking of RuntimeStub frame
duke@435	3677
duke@435	3678	// check for pending exceptions
duke@435	3679	#ifdef ASSERT
duke@435	3680	Label L;
never@739	3681	__ cmpptr(Address(r15_thread, Thread::pending_exception_offset()),
never@739	3682	(int32_t) NULL_WORD);
duke@435	3683	__ jcc(Assembler::notEqual, L);
duke@435	3684	__ should_not_reach_here();
duke@435	3685	__ bind(L);
duke@435	3686	#endif // ASSERT
duke@435	3687	__ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
duke@435	3688
duke@435	3689
duke@435	3690	// codeBlob framesize is in words (not VMRegImpl::slot_size)
duke@435	3691	RuntimeStub* stub =
duke@435	3692	RuntimeStub::new_runtime_stub(name,
duke@435	3693	&code,
duke@435	3694	frame_complete,
duke@435	3695	(framesize >> (LogBytesPerWord - LogBytesPerInt)),
duke@435	3696	oop_maps, false);
duke@435	3697	return stub->entry_point();
duke@435	3698	}
duke@435	3699
duke@435	3700	// Initialization
duke@435	3701	void generate_initial() {
duke@435	3702	// Generates all stubs and initializes the entry points
duke@435	3703
duke@435	3704	// This platform-specific stub is needed by generate_call_stub()
never@739	3705	StubRoutines::x86::_mxcsr_std = generate_fp_mask("mxcsr_std", 0x0000000000001F80);
duke@435	3706
duke@435	3707	// entry points that exist in all platforms Note: This is code
duke@435	3708	// that could be shared among different platforms - however the
duke@435	3709	// benefit seems to be smaller than the disadvantage of having a
duke@435	3710	// much more complicated generator structure. See also comment in
duke@435	3711	// stubRoutines.hpp.
duke@435	3712
duke@435	3713	StubRoutines::_forward_exception_entry = generate_forward_exception();
duke@435	3714
duke@435	3715	StubRoutines::_call_stub_entry =
duke@435	3716	generate_call_stub(StubRoutines::_call_stub_return_address);
duke@435	3717
duke@435	3718	// is referenced by megamorphic call
duke@435	3719	StubRoutines::_catch_exception_entry = generate_catch_exception();
duke@435	3720
duke@435	3721	// atomic calls
duke@435	3722	StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
duke@435	3723	StubRoutines::_atomic_xchg_ptr_entry = generate_atomic_xchg_ptr();
duke@435	3724	StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg();
duke@435	3725	StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
duke@435	3726	StubRoutines::_atomic_add_entry = generate_atomic_add();
duke@435	3727	StubRoutines::_atomic_add_ptr_entry = generate_atomic_add_ptr();
duke@435	3728	StubRoutines::_fence_entry = generate_orderaccess_fence();
duke@435	3729
duke@435	3730	StubRoutines::_handler_for_unsafe_access_entry =
duke@435	3731	generate_handler_for_unsafe_access();
duke@435	3732
duke@435	3733	// platform dependent
never@739	3734	StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp();
roland@3606	3735	StubRoutines::x86::_get_previous_sp_entry = generate_get_previous_sp();
never@739	3736
never@739	3737	StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
never@2978	3738
bdelsart@3372	3739	// Build this early so it's available for the interpreter.
bdelsart@3372	3740	StubRoutines::_throw_StackOverflowError_entry =
bdelsart@3372	3741	generate_throw_exception("StackOverflowError throw_exception",
bdelsart@3372	3742	CAST_FROM_FN_PTR(address,
bdelsart@3372	3743	SharedRuntime::
bdelsart@3372	3744	throw_StackOverflowError));
duke@435	3745	}
duke@435	3746
duke@435	3747	void generate_all() {
duke@435	3748	// Generates all stubs and initializes the entry points
duke@435	3749
duke@435	3750	// These entry points require SharedInfo::stack0 to be set up in
duke@435	3751	// non-core builds and need to be relocatable, so they each
duke@435	3752	// fabricate a RuntimeStub internally.
duke@435	3753	StubRoutines::_throw_AbstractMethodError_entry =
duke@435	3754	generate_throw_exception("AbstractMethodError throw_exception",
duke@435	3755	CAST_FROM_FN_PTR(address,
duke@435	3756	SharedRuntime::
never@3136	3757	throw_AbstractMethodError));
duke@435	3758
dcubed@451	3759	StubRoutines::_throw_IncompatibleClassChangeError_entry =
dcubed@451	3760	generate_throw_exception("IncompatibleClassChangeError throw_exception",
dcubed@451	3761	CAST_FROM_FN_PTR(address,
dcubed@451	3762	SharedRuntime::
never@3136	3763	throw_IncompatibleClassChangeError));
duke@435	3764
duke@435	3765	StubRoutines::_throw_NullPointerException_at_call_entry =
duke@435	3766	generate_throw_exception("NullPointerException at call throw_exception",
duke@435	3767	CAST_FROM_FN_PTR(address,
duke@435	3768	SharedRuntime::
never@3136	3769	throw_NullPointerException_at_call));
duke@435	3770
duke@435	3771	// entry points that are platform specific
never@739	3772	StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
never@739	3773	StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
never@739	3774	StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
never@739	3775	StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
never@739	3776
never@739	3777	StubRoutines::x86::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF);
never@739	3778	StubRoutines::x86::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000);
never@739	3779	StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
never@739	3780	StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
duke@435	3781
duke@435	3782	// support for verify_oop (must happen after universe_init)
duke@435	3783	StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
duke@435	3784
duke@435	3785	// arraycopy stubs used by compilers
duke@435	3786	generate_arraycopy_stubs();
twisti@1543	3787
never@1609	3788	generate_math_stubs();
kvn@4205	3789
kvn@4205	3790	// don't bother generating these AES intrinsic stubs unless global flag is set
kvn@4205	3791	if (UseAESIntrinsics) {
kvn@4205	3792	StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // needed by the others
kvn@4205	3793
kvn@4205	3794	StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
kvn@4205	3795	StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
kvn@4205	3796	StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
kvn@4205	3797	StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt_Parallel();
kvn@4205	3798	}
duke@435	3799	}
duke@435	3800
duke@435	3801	public:
duke@435	3802	StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
duke@435	3803	if (all) {
duke@435	3804	generate_all();
duke@435	3805	} else {
duke@435	3806	generate_initial();
duke@435	3807	}
duke@435	3808	}
duke@435	3809	}; // end class declaration
duke@435	3810
duke@435	3811	void StubGenerator_generate(CodeBuffer* code, bool all) {
duke@435	3812	StubGenerator g(code, all);
duke@435	3813	}