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

src/cpu/x86/vm/interpreter_x86_64.cpp@ba764ed4b6f2 (annotated)

src/cpu/x86/vm/interpreter_x86_64.cpp

Sun, 13 Apr 2008 17:43:42 -0400

author

coleenp

date

Sun, 13 Apr 2008 17:43:42 -0400

changeset 548

ba764ed4b6f2

parent 435

a61af66fc99e

child 631

d1605aabd0a1

permissions

-rw-r--r--

6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
Summary: Compressed oops in instances, arrays, and headers. Code contributors are coleenp, phh, never, swamyv
Reviewed-by: jmasa, kamg, acorn, tbell, kvn, rasbold

duke@435	1	/*
duke@435	2	* Copyright 2003-2007 Sun Microsystems, Inc. All Rights Reserved.
duke@435	3	* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
duke@435	4	*
duke@435	5	* This code is free software; you can redistribute it and/or modify it
duke@435	6	* under the terms of the GNU General Public License version 2 only, as
duke@435	7	* published by the Free Software Foundation.
duke@435	8	*
duke@435	9	* This code is distributed in the hope that it will be useful, but WITHOUT
duke@435	10	* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
duke@435	11	* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
duke@435	12	* version 2 for more details (a copy is included in the LICENSE file that
duke@435	13	* accompanied this code).
duke@435	14	*
duke@435	15	* You should have received a copy of the GNU General Public License version
duke@435	16	* 2 along with this work; if not, write to the Free Software Foundation,
duke@435	17	* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
duke@435	18	*
duke@435	19	* Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
duke@435	20	* CA 95054 USA or visit www.sun.com if you need additional information or
duke@435	21	* have any questions.
duke@435	22	*
duke@435	23	*/
duke@435	24
duke@435	25	#include "incls/_precompiled.incl"
duke@435	26	#include "incls/_interpreter_x86_64.cpp.incl"
duke@435	27
duke@435	28	#define __ _masm->
duke@435	29
duke@435	30
duke@435	31	#ifdef _WIN64
duke@435	32	address AbstractInterpreterGenerator::generate_slow_signature_handler() {
duke@435	33	address entry = __ pc();
duke@435	34
duke@435	35	// rbx: method
duke@435	36	// r14: pointer to locals
duke@435	37	// c_rarg3: first stack arg - wordSize
duke@435	38	__ movq(c_rarg3, rsp);
duke@435	39	// adjust rsp
duke@435	40	__ subq(rsp, 4 * wordSize);
duke@435	41	__ call_VM(noreg,
duke@435	42	CAST_FROM_FN_PTR(address,
duke@435	43	InterpreterRuntime::slow_signature_handler),
duke@435	44	rbx, r14, c_rarg3);
duke@435	45
duke@435	46	// rax: result handler
duke@435	47
duke@435	48	// Stack layout:
duke@435	49	// rsp: 3 integer or float args (if static first is unused)
duke@435	50	// 1 float/double identifiers
duke@435	51	// return address
duke@435	52	// stack args
duke@435	53	// garbage
duke@435	54	// expression stack bottom
duke@435	55	// bcp (NULL)
duke@435	56	// ...
duke@435	57
duke@435	58	// Do FP first so we can use c_rarg3 as temp
duke@435	59	__ movl(c_rarg3, Address(rsp, 3 * wordSize)); // float/double identifiers
duke@435	60
duke@435	61	for ( int i= 0; i < Argument::n_int_register_parameters_c-1; i++ ) {
duke@435	62	XMMRegister floatreg = as_XMMRegister(i+1);
duke@435	63	Label isfloatordouble, isdouble, next;
duke@435	64
duke@435	65	__ testl(c_rarg3, 1 << (i*2)); // Float or Double?
duke@435	66	__ jcc(Assembler::notZero, isfloatordouble);
duke@435	67
duke@435	68	// Do Int register here
duke@435	69	switch ( i ) {
duke@435	70	case 0:
duke@435	71	__ movl(rscratch1, Address(rbx, methodOopDesc::access_flags_offset()));
duke@435	72	__ testl(rscratch1, JVM_ACC_STATIC);
duke@435	73	__ cmovq(Assembler::zero, c_rarg1, Address(rsp, 0));
duke@435	74	break;
duke@435	75	case 1:
duke@435	76	__ movq(c_rarg2, Address(rsp, wordSize));
duke@435	77	break;
duke@435	78	case 2:
duke@435	79	__ movq(c_rarg3, Address(rsp, 2 * wordSize));
duke@435	80	break;
duke@435	81	default:
duke@435	82	break;
duke@435	83	}
duke@435	84
duke@435	85	__ jmp (next);
duke@435	86
duke@435	87	__ bind(isfloatordouble);
duke@435	88	__ testl(c_rarg3, 1 << ((i*2)+1)); // Double?
duke@435	89	__ jcc(Assembler::notZero, isdouble);
duke@435	90
duke@435	91	// Do Float Here
duke@435	92	__ movflt(floatreg, Address(rsp, i * wordSize));
duke@435	93	__ jmp(next);
duke@435	94
duke@435	95	// Do Double here
duke@435	96	__ bind(isdouble);
duke@435	97	__ movdbl(floatreg, Address(rsp, i * wordSize));
duke@435	98
duke@435	99	__ bind(next);
duke@435	100	}
duke@435	101
duke@435	102
duke@435	103	// restore rsp
duke@435	104	__ addq(rsp, 4 * wordSize);
duke@435	105
duke@435	106	__ ret(0);
duke@435	107
duke@435	108	return entry;
duke@435	109	}
duke@435	110	#else
duke@435	111	address AbstractInterpreterGenerator::generate_slow_signature_handler() {
duke@435	112	address entry = __ pc();
duke@435	113
duke@435	114	// rbx: method
duke@435	115	// r14: pointer to locals
duke@435	116	// c_rarg3: first stack arg - wordSize
duke@435	117	__ movq(c_rarg3, rsp);
duke@435	118	// adjust rsp
duke@435	119	__ subq(rsp, 14 * wordSize);
duke@435	120	__ call_VM(noreg,
duke@435	121	CAST_FROM_FN_PTR(address,
duke@435	122	InterpreterRuntime::slow_signature_handler),
duke@435	123	rbx, r14, c_rarg3);
duke@435	124
duke@435	125	// rax: result handler
duke@435	126
duke@435	127	// Stack layout:
duke@435	128	// rsp: 5 integer args (if static first is unused)
duke@435	129	// 1 float/double identifiers
duke@435	130	// 8 double args
duke@435	131	// return address
duke@435	132	// stack args
duke@435	133	// garbage
duke@435	134	// expression stack bottom
duke@435	135	// bcp (NULL)
duke@435	136	// ...
duke@435	137
duke@435	138	// Do FP first so we can use c_rarg3 as temp
duke@435	139	__ movl(c_rarg3, Address(rsp, 5 * wordSize)); // float/double identifiers
duke@435	140
duke@435	141	for (int i = 0; i < Argument::n_float_register_parameters_c; i++) {
duke@435	142	const XMMRegister r = as_XMMRegister(i);
duke@435	143
duke@435	144	Label d, done;
duke@435	145
duke@435	146	__ testl(c_rarg3, 1 << i);
duke@435	147	__ jcc(Assembler::notZero, d);
duke@435	148	__ movflt(r, Address(rsp, (6 + i) * wordSize));
duke@435	149	__ jmp(done);
duke@435	150	__ bind(d);
duke@435	151	__ movdbl(r, Address(rsp, (6 + i) * wordSize));
duke@435	152	__ bind(done);
duke@435	153	}
duke@435	154
duke@435	155	// Now handle integrals. Only do c_rarg1 if not static.
duke@435	156	__ movl(c_rarg3, Address(rbx, methodOopDesc::access_flags_offset()));
duke@435	157	__ testl(c_rarg3, JVM_ACC_STATIC);
duke@435	158	__ cmovq(Assembler::zero, c_rarg1, Address(rsp, 0));
duke@435	159
duke@435	160	__ movq(c_rarg2, Address(rsp, wordSize));
duke@435	161	__ movq(c_rarg3, Address(rsp, 2 * wordSize));
duke@435	162	__ movq(c_rarg4, Address(rsp, 3 * wordSize));
duke@435	163	__ movq(c_rarg5, Address(rsp, 4 * wordSize));
duke@435	164
duke@435	165	// restore rsp
duke@435	166	__ addq(rsp, 14 * wordSize);
duke@435	167
duke@435	168	__ ret(0);
duke@435	169
duke@435	170	return entry;
duke@435	171	}
duke@435	172	#endif
duke@435	173
duke@435	174
duke@435	175	//
duke@435	176	// Various method entries
duke@435	177	//
duke@435	178
duke@435	179	address InterpreterGenerator::generate_math_entry(
duke@435	180	AbstractInterpreter::MethodKind kind) {
duke@435	181	// rbx: methodOop
duke@435	182
duke@435	183	if (!InlineIntrinsics) return NULL; // Generate a vanilla entry
duke@435	184
duke@435	185	assert(kind == Interpreter::java_lang_math_sqrt,
duke@435	186	"Other intrinsics are not special");
duke@435	187
duke@435	188	address entry_point = __ pc();
duke@435	189
duke@435	190	// These don't need a safepoint check because they aren't virtually
duke@435	191	// callable. We won't enter these intrinsics from compiled code.
duke@435	192	// If in the future we added an intrinsic which was virtually callable
duke@435	193	// we'd have to worry about how to safepoint so that this code is used.
duke@435	194
duke@435	195	// mathematical functions inlined by compiler
duke@435	196	// (interpreter must provide identical implementation
duke@435	197	// in order to avoid monotonicity bugs when switching
duke@435	198	// from interpreter to compiler in the middle of some
duke@435	199	// computation)
duke@435	200
duke@435	201	// Note: For JDK 1.2 StrictMath doesn't exist and Math.sin/cos/sqrt are
duke@435	202	// native methods. Interpreter::method_kind(...) does a check for
duke@435	203	// native methods first before checking for intrinsic methods and
duke@435	204	// thus will never select this entry point. Make sure it is not
duke@435	205	// called accidentally since the SharedRuntime entry points will
duke@435	206	// not work for JDK 1.2.
duke@435	207	//
duke@435	208	// We no longer need to check for JDK 1.2 since it's EOL'ed.
duke@435	209	// The following check existed in pre 1.6 implementation,
duke@435	210	// if (Universe::is_jdk12x_version()) {
duke@435	211	// __ should_not_reach_here();
duke@435	212	// }
duke@435	213	// Universe::is_jdk12x_version() always returns false since
duke@435	214	// the JDK version is not yet determined when this method is called.
duke@435	215	// This method is called during interpreter_init() whereas
duke@435	216	// JDK version is only determined when universe2_init() is called.
duke@435	217
duke@435	218	// Note: For JDK 1.3 StrictMath exists and Math.sin/cos/sqrt are
duke@435	219	// java methods. Interpreter::method_kind(...) will select
duke@435	220	// this entry point for the corresponding methods in JDK 1.3.
duke@435	221	__ sqrtsd(xmm0, Address(rsp, wordSize));
duke@435	222
duke@435	223	__ popq(rax);
duke@435	224	__ movq(rsp, r13);
duke@435	225	__ jmp(rax);
duke@435	226
duke@435	227	return entry_point;
duke@435	228	}
duke@435	229
duke@435	230
duke@435	231	// Abstract method entry
duke@435	232	// Attempt to execute abstract method. Throw exception
duke@435	233	address InterpreterGenerator::generate_abstract_entry(void) {
duke@435	234	// rbx: methodOop
duke@435	235	// r13: sender SP
duke@435	236
duke@435	237	address entry_point = __ pc();
duke@435	238
duke@435	239	// abstract method entry
duke@435	240	// remove return address. Not really needed, since exception
duke@435	241	// handling throws away expression stack
duke@435	242	__ popq(rbx);
duke@435	243
duke@435	244	// adjust stack to what a normal return would do
duke@435	245	__ movq(rsp, r13);
duke@435	246
duke@435	247	// throw exception
duke@435	248	__ call_VM(noreg, CAST_FROM_FN_PTR(address,
duke@435	249	InterpreterRuntime::throw_AbstractMethodError));
duke@435	250	// the call_VM checks for exception, so we should never return here.
duke@435	251	__ should_not_reach_here();
duke@435	252
duke@435	253	return entry_point;
duke@435	254	}
duke@435	255
duke@435	256
duke@435	257	// Empty method, generate a very fast return.
duke@435	258
duke@435	259	address InterpreterGenerator::generate_empty_entry(void) {
duke@435	260	// rbx: methodOop
duke@435	261	// r13: sender sp must set sp to this value on return
duke@435	262
duke@435	263	if (!UseFastEmptyMethods) {
duke@435	264	return NULL;
duke@435	265	}
duke@435	266
duke@435	267	address entry_point = __ pc();
duke@435	268
duke@435	269	// If we need a safepoint check, generate full interpreter entry.
duke@435	270	Label slow_path;
duke@435	271	__ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
duke@435	272	SafepointSynchronize::_not_synchronized);
duke@435	273	__ jcc(Assembler::notEqual, slow_path);
duke@435	274
duke@435	275	// do nothing for empty methods (do not even increment invocation counter)
duke@435	276	// Code: _return
duke@435	277	// _return
duke@435	278	// return w/o popping parameters
duke@435	279	__ popq(rax);
duke@435	280	__ movq(rsp, r13);
duke@435	281	__ jmp(rax);
duke@435	282
duke@435	283	__ bind(slow_path);
duke@435	284	(void) generate_normal_entry(false);
duke@435	285	return entry_point;
duke@435	286
duke@435	287	}
duke@435	288
duke@435	289	// Call an accessor method (assuming it is resolved, otherwise drop
duke@435	290	// into vanilla (slow path) entry
duke@435	291	address InterpreterGenerator::generate_accessor_entry(void) {
duke@435	292	// rbx: methodOop
duke@435	293
duke@435	294	// r13: senderSP must preserver for slow path, set SP to it on fast path
duke@435	295
duke@435	296	address entry_point = __ pc();
duke@435	297	Label xreturn_path;
duke@435	298
duke@435	299	// do fastpath for resolved accessor methods
duke@435	300	if (UseFastAccessorMethods) {
duke@435	301	// Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites
duke@435	302	// thereof; parameter size = 1
duke@435	303	// Note: We can only use this code if the getfield has been resolved
duke@435	304	// and if we don't have a null-pointer exception => check for
duke@435	305	// these conditions first and use slow path if necessary.
duke@435	306	Label slow_path;
duke@435	307	// If we need a safepoint check, generate full interpreter entry.
duke@435	308	__ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
duke@435	309	SafepointSynchronize::_not_synchronized);
duke@435	310
duke@435	311	__ jcc(Assembler::notEqual, slow_path);
duke@435	312	// rbx: method
duke@435	313	__ movq(rax, Address(rsp, wordSize));
duke@435	314
duke@435	315	// check if local 0 != NULL and read field
duke@435	316	__ testq(rax, rax);
duke@435	317	__ jcc(Assembler::zero, slow_path);
duke@435	318
duke@435	319	__ movq(rdi, Address(rbx, methodOopDesc::constants_offset()));
duke@435	320	// read first instruction word and extract bytecode @ 1 and index @ 2
duke@435	321	__ movq(rdx, Address(rbx, methodOopDesc::const_offset()));
duke@435	322	__ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset()));
duke@435	323	// Shift codes right to get the index on the right.
duke@435	324	// The bytecode fetched looks like <index><0xb4><0x2a>
duke@435	325	__ shrl(rdx, 2 * BitsPerByte);
duke@435	326	__ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
duke@435	327	__ movq(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes()));
duke@435	328
duke@435	329	// rax: local 0
duke@435	330	// rbx: method
duke@435	331	// rdx: constant pool cache index
duke@435	332	// rdi: constant pool cache
duke@435	333
duke@435	334	// check if getfield has been resolved and read constant pool cache entry
duke@435	335	// check the validity of the cache entry by testing whether _indices field
duke@435	336	// contains Bytecode::_getfield in b1 byte.
duke@435	337	assert(in_words(ConstantPoolCacheEntry::size()) == 4,
duke@435	338	"adjust shift below");
duke@435	339	__ movl(rcx,
duke@435	340	Address(rdi,
duke@435	341	rdx,
duke@435	342	Address::times_8,
duke@435	343	constantPoolCacheOopDesc::base_offset() +
duke@435	344	ConstantPoolCacheEntry::indices_offset()));
duke@435	345	__ shrl(rcx, 2 * BitsPerByte);
duke@435	346	__ andl(rcx, 0xFF);
duke@435	347	__ cmpl(rcx, Bytecodes::_getfield);
duke@435	348	__ jcc(Assembler::notEqual, slow_path);
duke@435	349
duke@435	350	// Note: constant pool entry is not valid before bytecode is resolved
duke@435	351	__ movq(rcx,
duke@435	352	Address(rdi,
duke@435	353	rdx,
duke@435	354	Address::times_8,
duke@435	355	constantPoolCacheOopDesc::base_offset() +
duke@435	356	ConstantPoolCacheEntry::f2_offset()));
duke@435	357	// edx: flags
duke@435	358	__ movl(rdx,
duke@435	359	Address(rdi,
duke@435	360	rdx,
duke@435	361	Address::times_8,
duke@435	362	constantPoolCacheOopDesc::base_offset() +
duke@435	363	ConstantPoolCacheEntry::flags_offset()));
duke@435	364
duke@435	365	Label notObj, notInt, notByte, notShort;
duke@435	366	const Address field_address(rax, rcx, Address::times_1);
duke@435	367
duke@435	368	// Need to differentiate between igetfield, agetfield, bgetfield etc.
duke@435	369	// because they are different sizes.
duke@435	370	// Use the type from the constant pool cache
duke@435	371	__ shrl(rdx, ConstantPoolCacheEntry::tosBits);
duke@435	372	// Make sure we don't need to mask edx for tosBits after the above shift
duke@435	373	ConstantPoolCacheEntry::verify_tosBits();
duke@435	374
duke@435	375	__ cmpl(rdx, atos);
duke@435	376	__ jcc(Assembler::notEqual, notObj);
duke@435	377	// atos
coleenp@548	378	__ load_heap_oop(rax, field_address);
duke@435	379	__ jmp(xreturn_path);
duke@435	380
duke@435	381	__ bind(notObj);
duke@435	382	__ cmpl(rdx, itos);
duke@435	383	__ jcc(Assembler::notEqual, notInt);
duke@435	384	// itos
duke@435	385	__ movl(rax, field_address);
duke@435	386	__ jmp(xreturn_path);
duke@435	387
duke@435	388	__ bind(notInt);
duke@435	389	__ cmpl(rdx, btos);
duke@435	390	__ jcc(Assembler::notEqual, notByte);
duke@435	391	// btos
duke@435	392	__ load_signed_byte(rax, field_address);
duke@435	393	__ jmp(xreturn_path);
duke@435	394
duke@435	395	__ bind(notByte);
duke@435	396	__ cmpl(rdx, stos);
duke@435	397	__ jcc(Assembler::notEqual, notShort);
duke@435	398	// stos
duke@435	399	__ load_signed_word(rax, field_address);
duke@435	400	__ jmp(xreturn_path);
duke@435	401
duke@435	402	__ bind(notShort);
duke@435	403	#ifdef ASSERT
duke@435	404	Label okay;
duke@435	405	__ cmpl(rdx, ctos);
duke@435	406	__ jcc(Assembler::equal, okay);
duke@435	407	__ stop("what type is this?");
duke@435	408	__ bind(okay);
duke@435	409	#endif
duke@435	410	// ctos
duke@435	411	__ load_unsigned_word(rax, field_address);
duke@435	412
duke@435	413	__ bind(xreturn_path);
duke@435	414
duke@435	415	// _ireturn/_areturn
duke@435	416	__ popq(rdi);
duke@435	417	__ movq(rsp, r13);
duke@435	418	__ jmp(rdi);
duke@435	419	__ ret(0);
duke@435	420
duke@435	421	// generate a vanilla interpreter entry as the slow path
duke@435	422	__ bind(slow_path);
duke@435	423	(void) generate_normal_entry(false);
duke@435	424	} else {
duke@435	425	(void) generate_normal_entry(false);
duke@435	426	}
duke@435	427
duke@435	428	return entry_point;
duke@435	429	}
duke@435	430
duke@435	431	// This method tells the deoptimizer how big an interpreted frame must be:
duke@435	432	int AbstractInterpreter::size_activation(methodOop method,
duke@435	433	int tempcount,
duke@435	434	int popframe_extra_args,
duke@435	435	int moncount,
duke@435	436	int callee_param_count,
duke@435	437	int callee_locals,
duke@435	438	bool is_top_frame) {
duke@435	439	return layout_activation(method,
duke@435	440	tempcount, popframe_extra_args, moncount,
duke@435	441	callee_param_count, callee_locals,
duke@435	442	(frame*) NULL, (frame*) NULL, is_top_frame);
duke@435	443	}
duke@435	444
duke@435	445	void Deoptimization::unwind_callee_save_values(frame* f, vframeArray* vframe_array) {
duke@435	446
duke@435	447	// This code is sort of the equivalent of C2IAdapter::setup_stack_frame back in
duke@435	448	// the days we had adapter frames. When we deoptimize a situation where a
duke@435	449	// compiled caller calls a compiled caller will have registers it expects
duke@435	450	// to survive the call to the callee. If we deoptimize the callee the only
duke@435	451	// way we can restore these registers is to have the oldest interpreter
duke@435	452	// frame that we create restore these values. That is what this routine
duke@435	453	// will accomplish.
duke@435	454
duke@435	455	// At the moment we have modified c2 to not have any callee save registers
duke@435	456	// so this problem does not exist and this routine is just a place holder.
duke@435	457
duke@435	458	assert(f->is_interpreted_frame(), "must be interpreted");
duke@435	459	}