Mon, 11 Oct 2010 04:18:58 -0700
6829194: JSR 292 needs to support compressed oops
Reviewed-by: kvn, jrose
1 /*
2 * Copyright (c) 1997, 2009, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 #include "incls/_precompiled.incl"
26 #include "incls/_vtableStubs_sparc.cpp.incl"
28 // machine-dependent part of VtableStubs: create vtableStub of correct size and
29 // initialize its code
31 #define __ masm->
34 #ifndef PRODUCT
35 extern "C" void bad_compiled_vtable_index(JavaThread* thread, oopDesc* receiver, int index);
36 #endif
39 // Used by compiler only; may use only caller saved, non-argument registers
40 // NOTE: %%%% if any change is made to this stub make sure that the function
41 // pd_code_size_limit is changed to ensure the correct size for VtableStub
42 VtableStub* VtableStubs::create_vtable_stub(int vtable_index) {
43 const int sparc_code_length = VtableStub::pd_code_size_limit(true);
44 VtableStub* s = new(sparc_code_length) VtableStub(true, vtable_index);
45 ResourceMark rm;
46 CodeBuffer cb(s->entry_point(), sparc_code_length);
47 MacroAssembler* masm = new MacroAssembler(&cb);
49 #ifndef PRODUCT
50 if (CountCompiledCalls) {
51 __ inc_counter(SharedRuntime::nof_megamorphic_calls_addr(), G5, G3_scratch);
52 }
53 #endif /* PRODUCT */
55 assert(VtableStub::receiver_location() == O0->as_VMReg(), "receiver expected in O0");
57 // get receiver klass
58 address npe_addr = __ pc();
59 __ load_klass(O0, G3_scratch);
61 // set methodOop (in case of interpreted method), and destination address
62 int entry_offset = instanceKlass::vtable_start_offset() + vtable_index*vtableEntry::size();
63 #ifndef PRODUCT
64 if (DebugVtables) {
65 Label L;
66 // check offset vs vtable length
67 __ ld(G3_scratch, instanceKlass::vtable_length_offset()*wordSize, G5);
68 __ cmp(G5, vtable_index*vtableEntry::size());
69 __ br(Assembler::greaterUnsigned, false, Assembler::pt, L);
70 __ delayed()->nop();
71 __ set(vtable_index, O2);
72 __ call_VM(noreg, CAST_FROM_FN_PTR(address, bad_compiled_vtable_index), O0, O2);
73 __ bind(L);
74 }
75 #endif
76 int v_off = entry_offset*wordSize + vtableEntry::method_offset_in_bytes();
77 if( __ is_simm13(v_off) ) {
78 __ ld_ptr(G3, v_off, G5_method);
79 } else {
80 __ set(v_off,G5);
81 __ ld_ptr(G3, G5, G5_method);
82 }
84 #ifndef PRODUCT
85 if (DebugVtables) {
86 Label L;
87 __ br_notnull(G5_method, false, Assembler::pt, L);
88 __ delayed()->nop();
89 __ stop("Vtable entry is ZERO");
90 __ bind(L);
91 }
92 #endif
94 address ame_addr = __ pc(); // if the vtable entry is null, the method is abstract
95 // NOTE: for vtable dispatches, the vtable entry will never be null.
97 __ ld_ptr(G5_method, in_bytes(methodOopDesc::from_compiled_offset()), G3_scratch);
99 // jump to target (either compiled code or c2iadapter)
100 __ JMP(G3_scratch, 0);
101 // load methodOop (in case we call c2iadapter)
102 __ delayed()->nop();
104 masm->flush();
106 if (PrintMiscellaneous && (WizardMode || Verbose)) {
107 tty->print_cr("vtable #%d at "PTR_FORMAT"[%d] left over: %d",
108 vtable_index, s->entry_point(),
109 (int)(s->code_end() - s->entry_point()),
110 (int)(s->code_end() - __ pc()));
111 }
112 guarantee(__ pc() <= s->code_end(), "overflowed buffer");
113 // shut the door on sizing bugs
114 int slop = 2*BytesPerInstWord; // 32-bit offset is this much larger than a 13-bit one
115 assert(vtable_index > 10 || __ pc() + slop <= s->code_end(), "room for sethi;add");
117 s->set_exception_points(npe_addr, ame_addr);
118 return s;
119 }
122 // NOTE: %%%% if any change is made to this stub make sure that the function
123 // pd_code_size_limit is changed to ensure the correct size for VtableStub
124 VtableStub* VtableStubs::create_itable_stub(int itable_index) {
125 const int sparc_code_length = VtableStub::pd_code_size_limit(false);
126 VtableStub* s = new(sparc_code_length) VtableStub(false, itable_index);
127 ResourceMark rm;
128 CodeBuffer cb(s->entry_point(), sparc_code_length);
129 MacroAssembler* masm = new MacroAssembler(&cb);
131 Register G3_klassOop = G3_scratch;
132 Register G5_interface = G5; // Passed in as an argument
133 Label search;
135 // Entry arguments:
136 // G5_interface: Interface
137 // O0: Receiver
138 assert(VtableStub::receiver_location() == O0->as_VMReg(), "receiver expected in O0");
140 // get receiver klass (also an implicit null-check)
141 address npe_addr = __ pc();
142 __ load_klass(O0, G3_klassOop);
143 __ verify_oop(G3_klassOop);
145 // Push a new window to get some temp registers. This chops the head of all
146 // my 64-bit %o registers in the LION build, but this is OK because no longs
147 // are passed in the %o registers. Instead, longs are passed in G1 and G4
148 // and so those registers are not available here.
149 __ save(SP,-frame::register_save_words*wordSize,SP);
151 #ifndef PRODUCT
152 if (CountCompiledCalls) {
153 __ inc_counter(SharedRuntime::nof_megamorphic_calls_addr(), L0, L1);
154 }
155 #endif /* PRODUCT */
157 Label throw_icce;
159 Register L5_method = L5;
160 __ lookup_interface_method(// inputs: rec. class, interface, itable index
161 G3_klassOop, G5_interface, itable_index,
162 // outputs: method, scan temp. reg
163 L5_method, L2, L3,
164 throw_icce);
166 #ifndef PRODUCT
167 if (DebugVtables) {
168 Label L01;
169 __ bpr(Assembler::rc_nz, false, Assembler::pt, L5_method, L01);
170 __ delayed()->nop();
171 __ stop("methodOop is null");
172 __ bind(L01);
173 __ verify_oop(L5_method);
174 }
175 #endif
177 // If the following load is through a NULL pointer, we'll take an OS
178 // exception that should translate into an AbstractMethodError. We need the
179 // window count to be correct at that time.
180 __ restore(L5_method, 0, G5_method);
181 // Restore registers *before* the AME point.
183 address ame_addr = __ pc(); // if the vtable entry is null, the method is abstract
184 __ ld_ptr(G5_method, in_bytes(methodOopDesc::from_compiled_offset()), G3_scratch);
186 // G5_method: methodOop
187 // O0: Receiver
188 // G3_scratch: entry point
189 __ JMP(G3_scratch, 0);
190 __ delayed()->nop();
192 __ bind(throw_icce);
193 AddressLiteral icce(StubRoutines::throw_IncompatibleClassChangeError_entry());
194 __ jump_to(icce, G3_scratch);
195 __ delayed()->restore();
197 masm->flush();
199 if (PrintMiscellaneous && (WizardMode || Verbose)) {
200 tty->print_cr("itable #%d at "PTR_FORMAT"[%d] left over: %d",
201 itable_index, s->entry_point(),
202 (int)(s->code_end() - s->entry_point()),
203 (int)(s->code_end() - __ pc()));
204 }
205 guarantee(__ pc() <= s->code_end(), "overflowed buffer");
206 // shut the door on sizing bugs
207 int slop = 2*BytesPerInstWord; // 32-bit offset is this much larger than a 13-bit one
208 assert(itable_index > 10 || __ pc() + slop <= s->code_end(), "room for sethi;add");
210 s->set_exception_points(npe_addr, ame_addr);
211 return s;
212 }
215 int VtableStub::pd_code_size_limit(bool is_vtable_stub) {
216 if (TraceJumps || DebugVtables || CountCompiledCalls || VerifyOops) return 1000;
217 else {
218 const int slop = 2*BytesPerInstWord; // sethi;add (needed for long offsets)
219 if (is_vtable_stub) {
220 // ld;ld;ld,jmp,nop
221 const int basic = 5*BytesPerInstWord +
222 // shift;add for load_klass (only shift with zero heap based)
223 (UseCompressedOops ?
224 ((Universe::narrow_oop_base() == NULL) ? BytesPerInstWord : 2*BytesPerInstWord) : 0);
225 return basic + slop;
226 } else {
227 const int basic = (28 LP64_ONLY(+ 6)) * BytesPerInstWord +
228 // shift;add for load_klass (only shift with zero heap based)
229 (UseCompressedOops ?
230 ((Universe::narrow_oop_base() == NULL) ? BytesPerInstWord : 2*BytesPerInstWord) : 0);
231 return (basic + slop);
232 }
233 }
235 // In order to tune these parameters, run the JVM with VM options
236 // +PrintMiscellaneous and +WizardMode to see information about
237 // actual itable stubs. Look for lines like this:
238 // itable #1 at 0x5551212[116] left over: 8
239 // Reduce the constants so that the "left over" number is 8
240 // Do not aim at a left-over number of zero, because a very
241 // large vtable or itable offset (> 4K) will require an extra
242 // sethi/or pair of instructions.
243 //
244 // The JVM98 app. _202_jess has a megamorphic interface call.
245 // The itable code looks like this:
246 // Decoding VtableStub itbl[1]@16
247 // ld [ %o0 + 4 ], %g3
248 // save %sp, -64, %sp
249 // ld [ %g3 + 0xe8 ], %l2
250 // sll %l2, 2, %l2
251 // add %l2, 0x134, %l2
252 // and %l2, -8, %l2 ! NOT_LP64 only
253 // add %g3, %l2, %l2
254 // add %g3, 4, %g3
255 // ld [ %l2 ], %l5
256 // brz,pn %l5, throw_icce
257 // cmp %l5, %g5
258 // be %icc, success
259 // add %l2, 8, %l2
260 // loop:
261 // ld [ %l2 ], %l5
262 // brz,pn %l5, throw_icce
263 // cmp %l5, %g5
264 // bne,pn %icc, loop
265 // add %l2, 8, %l2
266 // success:
267 // ld [ %l2 + -4 ], %l2
268 // ld [ %g3 + %l2 ], %l5
269 // restore %l5, 0, %g5
270 // ld [ %g5 + 0x44 ], %g3
271 // jmp %g3
272 // nop
273 // throw_icce:
274 // sethi %hi(throw_ICCE_entry), %g3
275 // ! 5 more instructions here, LP64_ONLY
276 // jmp %g3 + %lo(throw_ICCE_entry)
277 // restore
278 }
281 int VtableStub::pd_code_alignment() {
282 // UltraSPARC cache line size is 8 instructions:
283 const unsigned int icache_line_size = 32;
284 return icache_line_size;
285 }