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

src/cpu/x86/vm/c1_CodeStubs_x86.cpp@f3de1255b035 (annotated)

src/cpu/x86/vm/c1_CodeStubs_x86.cpp

Wed, 07 May 2008 08:06:46 -0700

author

rasbold

date

Wed, 07 May 2008 08:06:46 -0700

changeset 580

f3de1255b035

parent 435

a61af66fc99e

child 739

dc7f315e41f7

child 777

37f87013dfd8

permissions

-rw-r--r--

6603011: RFE: Optimize long division
Summary: Transform long division by constant into multiply
Reviewed-by: never, kvn

duke@435	1	/*
duke@435	2	* Copyright 1999-2006 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/_c1_CodeStubs_x86.cpp.incl"
duke@435	27
duke@435	28
duke@435	29	#define __ ce->masm()->
duke@435	30
duke@435	31	float ConversionStub::float_zero = 0.0;
duke@435	32	double ConversionStub::double_zero = 0.0;
duke@435	33
duke@435	34	void ConversionStub::emit_code(LIR_Assembler* ce) {
duke@435	35	__ bind(_entry);
duke@435	36	assert(bytecode() == Bytecodes::_f2i || bytecode() == Bytecodes::_d2i, "other conversions do not require stub");
duke@435	37
duke@435	38
duke@435	39	if (input()->is_single_xmm()) {
duke@435	40	__ comiss(input()->as_xmm_float_reg(),
duke@435	41	ExternalAddress((address)&float_zero));
duke@435	42	} else if (input()->is_double_xmm()) {
duke@435	43	__ comisd(input()->as_xmm_double_reg(),
duke@435	44	ExternalAddress((address)&double_zero));
duke@435	45	} else {
duke@435	46	__ pushl(rax);
duke@435	47	__ ftst();
duke@435	48	__ fnstsw_ax();
duke@435	49	__ sahf();
duke@435	50	__ popl(rax);
duke@435	51	}
duke@435	52
duke@435	53	Label NaN, do_return;
duke@435	54	__ jccb(Assembler::parity, NaN);
duke@435	55	__ jccb(Assembler::below, do_return);
duke@435	56
duke@435	57	// input is > 0 -> return maxInt
duke@435	58	// result register already contains 0x80000000, so subtracting 1 gives 0x7fffffff
duke@435	59	__ decrement(result()->as_register());
duke@435	60	__ jmpb(do_return);
duke@435	61
duke@435	62	// input is NaN -> return 0
duke@435	63	__ bind(NaN);
duke@435	64	__ xorl(result()->as_register(), result()->as_register());
duke@435	65
duke@435	66	__ bind(do_return);
duke@435	67	__ jmp(_continuation);
duke@435	68	}
duke@435	69
duke@435	70	#ifdef TIERED
duke@435	71	void CounterOverflowStub::emit_code(LIR_Assembler* ce) {
duke@435	72	__ bind(_entry);
duke@435	73	ce->store_parameter(_bci, 0);
duke@435	74	__ call(RuntimeAddress(Runtime1::entry_for(Runtime1::counter_overflow_id)));
duke@435	75	ce->add_call_info_here(_info);
duke@435	76	ce->verify_oop_map(_info);
duke@435	77
duke@435	78	__ jmp(_continuation);
duke@435	79	}
duke@435	80	#endif // TIERED
duke@435	81
duke@435	82
duke@435	83
duke@435	84	RangeCheckStub::RangeCheckStub(CodeEmitInfo* info, LIR_Opr index,
duke@435	85	bool throw_index_out_of_bounds_exception)
duke@435	86	: _throw_index_out_of_bounds_exception(throw_index_out_of_bounds_exception)
duke@435	87	, _index(index)
duke@435	88	{
duke@435	89	_info = info == NULL ? NULL : new CodeEmitInfo(info);
duke@435	90	}
duke@435	91
duke@435	92
duke@435	93	void RangeCheckStub::emit_code(LIR_Assembler* ce) {
duke@435	94	__ bind(_entry);
duke@435	95	// pass the array index on stack because all registers must be preserved
duke@435	96	if (_index->is_cpu_register()) {
duke@435	97	ce->store_parameter(_index->as_register(), 0);
duke@435	98	} else {
duke@435	99	ce->store_parameter(_index->as_jint(), 0);
duke@435	100	}
duke@435	101	Runtime1::StubID stub_id;
duke@435	102	if (_throw_index_out_of_bounds_exception) {
duke@435	103	stub_id = Runtime1::throw_index_exception_id;
duke@435	104	} else {
duke@435	105	stub_id = Runtime1::throw_range_check_failed_id;
duke@435	106	}
duke@435	107	__ call(RuntimeAddress(Runtime1::entry_for(stub_id)));
duke@435	108	ce->add_call_info_here(_info);
duke@435	109	debug_only(__ should_not_reach_here());
duke@435	110	}
duke@435	111
duke@435	112
duke@435	113	void DivByZeroStub::emit_code(LIR_Assembler* ce) {
duke@435	114	if (_offset != -1) {
duke@435	115	ce->compilation()->implicit_exception_table()->append(_offset, __ offset());
duke@435	116	}
duke@435	117	__ bind(_entry);
duke@435	118	__ call(RuntimeAddress(Runtime1::entry_for(Runtime1::throw_div0_exception_id)));
duke@435	119	ce->add_call_info_here(_info);
duke@435	120	debug_only(__ should_not_reach_here());
duke@435	121	}
duke@435	122
duke@435	123
duke@435	124	// Implementation of NewInstanceStub
duke@435	125
duke@435	126	NewInstanceStub::NewInstanceStub(LIR_Opr klass_reg, LIR_Opr result, ciInstanceKlass* klass, CodeEmitInfo* info, Runtime1::StubID stub_id) {
duke@435	127	_result = result;
duke@435	128	_klass = klass;
duke@435	129	_klass_reg = klass_reg;
duke@435	130	_info = new CodeEmitInfo(info);
duke@435	131	assert(stub_id == Runtime1::new_instance_id ||
duke@435	132	stub_id == Runtime1::fast_new_instance_id ||
duke@435	133	stub_id == Runtime1::fast_new_instance_init_check_id,
duke@435	134	"need new_instance id");
duke@435	135	_stub_id = stub_id;
duke@435	136	}
duke@435	137
duke@435	138
duke@435	139	void NewInstanceStub::emit_code(LIR_Assembler* ce) {
duke@435	140	assert(__ rsp_offset() == 0, "frame size should be fixed");
duke@435	141	__ bind(_entry);
duke@435	142	__ movl(rdx, _klass_reg->as_register());
duke@435	143	__ call(RuntimeAddress(Runtime1::entry_for(_stub_id)));
duke@435	144	ce->add_call_info_here(_info);
duke@435	145	ce->verify_oop_map(_info);
duke@435	146	assert(_result->as_register() == rax, "result must in rax,");
duke@435	147	__ jmp(_continuation);
duke@435	148	}
duke@435	149
duke@435	150
duke@435	151	// Implementation of NewTypeArrayStub
duke@435	152
duke@435	153	NewTypeArrayStub::NewTypeArrayStub(LIR_Opr klass_reg, LIR_Opr length, LIR_Opr result, CodeEmitInfo* info) {
duke@435	154	_klass_reg = klass_reg;
duke@435	155	_length = length;
duke@435	156	_result = result;
duke@435	157	_info = new CodeEmitInfo(info);
duke@435	158	}
duke@435	159
duke@435	160
duke@435	161	void NewTypeArrayStub::emit_code(LIR_Assembler* ce) {
duke@435	162	assert(__ rsp_offset() == 0, "frame size should be fixed");
duke@435	163	__ bind(_entry);
duke@435	164	assert(_length->as_register() == rbx, "length must in rbx,");
duke@435	165	assert(_klass_reg->as_register() == rdx, "klass_reg must in rdx");
duke@435	166	__ call(RuntimeAddress(Runtime1::entry_for(Runtime1::new_type_array_id)));
duke@435	167	ce->add_call_info_here(_info);
duke@435	168	ce->verify_oop_map(_info);
duke@435	169	assert(_result->as_register() == rax, "result must in rax,");
duke@435	170	__ jmp(_continuation);
duke@435	171	}
duke@435	172
duke@435	173
duke@435	174	// Implementation of NewObjectArrayStub
duke@435	175
duke@435	176	NewObjectArrayStub::NewObjectArrayStub(LIR_Opr klass_reg, LIR_Opr length, LIR_Opr result, CodeEmitInfo* info) {
duke@435	177	_klass_reg = klass_reg;
duke@435	178	_result = result;
duke@435	179	_length = length;
duke@435	180	_info = new CodeEmitInfo(info);
duke@435	181	}
duke@435	182
duke@435	183
duke@435	184	void NewObjectArrayStub::emit_code(LIR_Assembler* ce) {
duke@435	185	assert(__ rsp_offset() == 0, "frame size should be fixed");
duke@435	186	__ bind(_entry);
duke@435	187	assert(_length->as_register() == rbx, "length must in rbx,");
duke@435	188	assert(_klass_reg->as_register() == rdx, "klass_reg must in rdx");
duke@435	189	__ call(RuntimeAddress(Runtime1::entry_for(Runtime1::new_object_array_id)));
duke@435	190	ce->add_call_info_here(_info);
duke@435	191	ce->verify_oop_map(_info);
duke@435	192	assert(_result->as_register() == rax, "result must in rax,");
duke@435	193	__ jmp(_continuation);
duke@435	194	}
duke@435	195
duke@435	196
duke@435	197	// Implementation of MonitorAccessStubs
duke@435	198
duke@435	199	MonitorEnterStub::MonitorEnterStub(LIR_Opr obj_reg, LIR_Opr lock_reg, CodeEmitInfo* info)
duke@435	200	: MonitorAccessStub(obj_reg, lock_reg)
duke@435	201	{
duke@435	202	_info = new CodeEmitInfo(info);
duke@435	203	}
duke@435	204
duke@435	205
duke@435	206	void MonitorEnterStub::emit_code(LIR_Assembler* ce) {
duke@435	207	assert(__ rsp_offset() == 0, "frame size should be fixed");
duke@435	208	__ bind(_entry);
duke@435	209	ce->store_parameter(_obj_reg->as_register(), 1);
duke@435	210	ce->store_parameter(_lock_reg->as_register(), 0);
duke@435	211	Runtime1::StubID enter_id;
duke@435	212	if (ce->compilation()->has_fpu_code()) {
duke@435	213	enter_id = Runtime1::monitorenter_id;
duke@435	214	} else {
duke@435	215	enter_id = Runtime1::monitorenter_nofpu_id;
duke@435	216	}
duke@435	217	__ call(RuntimeAddress(Runtime1::entry_for(enter_id)));
duke@435	218	ce->add_call_info_here(_info);
duke@435	219	ce->verify_oop_map(_info);
duke@435	220	__ jmp(_continuation);
duke@435	221	}
duke@435	222
duke@435	223
duke@435	224	void MonitorExitStub::emit_code(LIR_Assembler* ce) {
duke@435	225	__ bind(_entry);
duke@435	226	if (_compute_lock) {
duke@435	227	// lock_reg was destroyed by fast unlocking attempt => recompute it
duke@435	228	ce->monitor_address(_monitor_ix, _lock_reg);
duke@435	229	}
duke@435	230	ce->store_parameter(_lock_reg->as_register(), 0);
duke@435	231	// note: non-blocking leaf routine => no call info needed
duke@435	232	Runtime1::StubID exit_id;
duke@435	233	if (ce->compilation()->has_fpu_code()) {
duke@435	234	exit_id = Runtime1::monitorexit_id;
duke@435	235	} else {
duke@435	236	exit_id = Runtime1::monitorexit_nofpu_id;
duke@435	237	}
duke@435	238	__ call(RuntimeAddress(Runtime1::entry_for(exit_id)));
duke@435	239	__ jmp(_continuation);
duke@435	240	}
duke@435	241
duke@435	242
duke@435	243	// Implementation of patching:
duke@435	244	// - Copy the code at given offset to an inlined buffer (first the bytes, then the number of bytes)
duke@435	245	// - Replace original code with a call to the stub
duke@435	246	// At Runtime:
duke@435	247	// - call to stub, jump to runtime
duke@435	248	// - in runtime: preserve all registers (rspecially objects, i.e., source and destination object)
duke@435	249	// - in runtime: after initializing class, restore original code, reexecute instruction
duke@435	250
duke@435	251	int PatchingStub::_patch_info_offset = -NativeGeneralJump::instruction_size;
duke@435	252
duke@435	253	void PatchingStub::align_patch_site(MacroAssembler* masm) {
duke@435	254	// We're patching a 5-7 byte instruction on intel and we need to
duke@435	255	// make sure that we don't see a piece of the instruction. It
duke@435	256	// appears mostly impossible on Intel to simply invalidate other
duke@435	257	// processors caches and since they may do aggressive prefetch it's
duke@435	258	// very hard to make a guess about what code might be in the icache.
duke@435	259	// Force the instruction to be double word aligned so that it
duke@435	260	// doesn't span a cache line.
duke@435	261	masm->align(round_to(NativeGeneralJump::instruction_size, wordSize));
duke@435	262	}
duke@435	263
duke@435	264	void PatchingStub::emit_code(LIR_Assembler* ce) {
duke@435	265	assert(NativeCall::instruction_size <= _bytes_to_copy && _bytes_to_copy <= 0xFF, "not enough room for call");
duke@435	266
duke@435	267	Label call_patch;
duke@435	268
duke@435	269	// static field accesses have special semantics while the class
duke@435	270	// initializer is being run so we emit a test which can be used to
duke@435	271	// check that this code is being executed by the initializing
duke@435	272	// thread.
duke@435	273	address being_initialized_entry = __ pc();
duke@435	274	if (CommentedAssembly) {
duke@435	275	__ block_comment(" patch template");
duke@435	276	}
duke@435	277	if (_id == load_klass_id) {
duke@435	278	// produce a copy of the load klass instruction for use by the being initialized case
duke@435	279	address start = __ pc();
duke@435	280	jobject o = NULL;
duke@435	281	__ movoop(_obj, o);
duke@435	282	#ifdef ASSERT
duke@435	283	for (int i = 0; i < _bytes_to_copy; i++) {
duke@435	284	address ptr = (address)(_pc_start + i);
duke@435	285	int a_byte = (*ptr) & 0xFF;
duke@435	286	assert(a_byte == *start++, "should be the same code");
duke@435	287	}
duke@435	288	#endif
duke@435	289	} else {
duke@435	290	// make a copy the code which is going to be patched.
duke@435	291	for ( int i = 0; i < _bytes_to_copy; i++) {
duke@435	292	address ptr = (address)(_pc_start + i);
duke@435	293	int a_byte = (*ptr) & 0xFF;
duke@435	294	__ a_byte (a_byte);
duke@435	295	*ptr = 0x90; // make the site look like a nop
duke@435	296	}
duke@435	297	}
duke@435	298
duke@435	299	address end_of_patch = __ pc();
duke@435	300	int bytes_to_skip = 0;
duke@435	301	if (_id == load_klass_id) {
duke@435	302	int offset = __ offset();
duke@435	303	if (CommentedAssembly) {
duke@435	304	__ block_comment(" being_initialized check");
duke@435	305	}
duke@435	306	assert(_obj != noreg, "must be a valid register");
duke@435	307	Register tmp = rax;
duke@435	308	if (_obj == tmp) tmp = rbx;
duke@435	309	__ pushl(tmp);
duke@435	310	__ get_thread(tmp);
duke@435	311	__ cmpl(tmp, Address(_obj, instanceKlass::init_thread_offset_in_bytes() + sizeof(klassOopDesc)));
duke@435	312	__ popl(tmp);
duke@435	313	__ jcc(Assembler::notEqual, call_patch);
duke@435	314
duke@435	315	// access_field patches may execute the patched code before it's
duke@435	316	// copied back into place so we need to jump back into the main
duke@435	317	// code of the nmethod to continue execution.
duke@435	318	__ jmp(_patch_site_continuation);
duke@435	319
duke@435	320	// make sure this extra code gets skipped
duke@435	321	bytes_to_skip += __ offset() - offset;
duke@435	322	}
duke@435	323	if (CommentedAssembly) {
duke@435	324	__ block_comment("patch data encoded as movl");
duke@435	325	}
duke@435	326	// Now emit the patch record telling the runtime how to find the
duke@435	327	// pieces of the patch. We only need 3 bytes but for readability of
duke@435	328	// the disassembly we make the data look like a movl reg, imm32,
duke@435	329	// which requires 5 bytes
duke@435	330	int sizeof_patch_record = 5;
duke@435	331	bytes_to_skip += sizeof_patch_record;
duke@435	332
duke@435	333	// emit the offsets needed to find the code to patch
duke@435	334	int being_initialized_entry_offset = __ pc() - being_initialized_entry + sizeof_patch_record;
duke@435	335
duke@435	336	__ a_byte(0xB8);
duke@435	337	__ a_byte(0);
duke@435	338	__ a_byte(being_initialized_entry_offset);
duke@435	339	__ a_byte(bytes_to_skip);
duke@435	340	__ a_byte(_bytes_to_copy);
duke@435	341	address patch_info_pc = __ pc();
duke@435	342	assert(patch_info_pc - end_of_patch == bytes_to_skip, "incorrect patch info");
duke@435	343
duke@435	344	address entry = __ pc();
duke@435	345	NativeGeneralJump::insert_unconditional((address)_pc_start, entry);
duke@435	346	address target = NULL;
duke@435	347	switch (_id) {
duke@435	348	case access_field_id: target = Runtime1::entry_for(Runtime1::access_field_patching_id); break;
duke@435	349	case load_klass_id: target = Runtime1::entry_for(Runtime1::load_klass_patching_id); break;
duke@435	350	default: ShouldNotReachHere();
duke@435	351	}
duke@435	352	__ bind(call_patch);
duke@435	353
duke@435	354	if (CommentedAssembly) {
duke@435	355	__ block_comment("patch entry point");
duke@435	356	}
duke@435	357	__ call(RuntimeAddress(target));
duke@435	358	assert(_patch_info_offset == (patch_info_pc - __ pc()), "must not change");
duke@435	359	ce->add_call_info_here(_info);
duke@435	360	int jmp_off = __ offset();
duke@435	361	__ jmp(_patch_site_entry);
duke@435	362	// Add enough nops so deoptimization can overwrite the jmp above with a call
duke@435	363	// and not destroy the world.
duke@435	364	for (int j = __ offset() ; j < jmp_off + 5 ; j++ ) {
duke@435	365	__ nop();
duke@435	366	}
duke@435	367	if (_id == load_klass_id) {
duke@435	368	CodeSection* cs = __ code_section();
duke@435	369	RelocIterator iter(cs, (address)_pc_start, (address)(_pc_start + 1));
duke@435	370	relocInfo::change_reloc_info_for_address(&iter, (address) _pc_start, relocInfo::oop_type, relocInfo::none);
duke@435	371	}
duke@435	372	}
duke@435	373
duke@435	374
duke@435	375	void ImplicitNullCheckStub::emit_code(LIR_Assembler* ce) {
duke@435	376	ce->compilation()->implicit_exception_table()->append(_offset, __ offset());
duke@435	377	__ bind(_entry);
duke@435	378	__ call(RuntimeAddress(Runtime1::entry_for(Runtime1::throw_null_pointer_exception_id)));
duke@435	379	ce->add_call_info_here(_info);
duke@435	380	debug_only(__ should_not_reach_here());
duke@435	381	}
duke@435	382
duke@435	383
duke@435	384	void SimpleExceptionStub::emit_code(LIR_Assembler* ce) {
duke@435	385	assert(__ rsp_offset() == 0, "frame size should be fixed");
duke@435	386
duke@435	387	__ bind(_entry);
duke@435	388	// pass the object on stack because all registers must be preserved
duke@435	389	if (_obj->is_cpu_register()) {
duke@435	390	ce->store_parameter(_obj->as_register(), 0);
duke@435	391	}
duke@435	392	__ call(RuntimeAddress(Runtime1::entry_for(_stub)));
duke@435	393	ce->add_call_info_here(_info);
duke@435	394	debug_only(__ should_not_reach_here());
duke@435	395	}
duke@435	396
duke@435	397
duke@435	398	ArrayStoreExceptionStub::ArrayStoreExceptionStub(CodeEmitInfo* info):
duke@435	399	_info(info) {
duke@435	400	}
duke@435	401
duke@435	402
duke@435	403	void ArrayStoreExceptionStub::emit_code(LIR_Assembler* ce) {
duke@435	404	assert(__ rsp_offset() == 0, "frame size should be fixed");
duke@435	405	__ bind(_entry);
duke@435	406	__ call(RuntimeAddress(Runtime1::entry_for(Runtime1::throw_array_store_exception_id)));
duke@435	407	ce->add_call_info_here(_info);
duke@435	408	debug_only(__ should_not_reach_here());
duke@435	409	}
duke@435	410
duke@435	411
duke@435	412	void ArrayCopyStub::emit_code(LIR_Assembler* ce) {
duke@435	413	//---------------slow case: call to native-----------------
duke@435	414	__ bind(_entry);
duke@435	415	// Figure out where the args should go
duke@435	416	// This should really convert the IntrinsicID to the methodOop and signature
duke@435	417	// but I don't know how to do that.
duke@435	418	//
duke@435	419	VMRegPair args[5];
duke@435	420	BasicType signature[5] = { T_OBJECT, T_INT, T_OBJECT, T_INT, T_INT};
duke@435	421	SharedRuntime::java_calling_convention(signature, args, 5, true);
duke@435	422
duke@435	423	// push parameters
duke@435	424	// (src, src_pos, dest, destPos, length)
duke@435	425	Register r[5];
duke@435	426	r[0] = src()->as_register();
duke@435	427	r[1] = src_pos()->as_register();
duke@435	428	r[2] = dst()->as_register();
duke@435	429	r[3] = dst_pos()->as_register();
duke@435	430	r[4] = length()->as_register();
duke@435	431
duke@435	432	// next registers will get stored on the stack
duke@435	433	for (int i = 0; i < 5 ; i++ ) {
duke@435	434	VMReg r_1 = args[i].first();
duke@435	435	if (r_1->is_stack()) {
duke@435	436	int st_off = r_1->reg2stack() * wordSize;
duke@435	437	__ movl (Address(rsp, st_off), r[i]);
duke@435	438	} else {
duke@435	439	assert(r[i] == args[i].first()->as_Register(), "Wrong register for arg ");
duke@435	440	}
duke@435	441	}
duke@435	442
duke@435	443	ce->align_call(lir_static_call);
duke@435	444
duke@435	445	ce->emit_static_call_stub();
duke@435	446	AddressLiteral resolve(SharedRuntime::get_resolve_static_call_stub(),
duke@435	447	relocInfo::static_call_type);
duke@435	448	__ call(resolve);
duke@435	449	ce->add_call_info_here(info());
duke@435	450
duke@435	451	#ifndef PRODUCT
duke@435	452	__ increment(ExternalAddress((address)&Runtime1::_arraycopy_slowcase_cnt));
duke@435	453	#endif
duke@435	454
duke@435	455	__ jmp(_continuation);
duke@435	456	}
duke@435	457
duke@435	458
duke@435	459	#undef __