Tue, 24 Jul 2012 10:51:00 -0700
7023639: JSR 292 method handle invocation needs a fast path for compiled code
6984705: JSR 292 method handle creation should not go through JNI
Summary: remove assembly code for JDK 7 chained method handles
Reviewed-by: jrose, twisti, kvn, mhaupt
Contributed-by: John Rose <john.r.rose@oracle.com>, Christian Thalinger <christian.thalinger@oracle.com>, Michael Haupt <michael.haupt@oracle.com>
1 /*
2 * Copyright (c) 1997, 2011, 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 "precompiled.hpp"
26 #include "asm/assembler.hpp"
27 #include "assembler_sparc.inline.hpp"
28 #include "code/vtableStubs.hpp"
29 #include "interp_masm_sparc.hpp"
30 #include "memory/resourceArea.hpp"
31 #include "oops/instanceKlass.hpp"
32 #include "oops/klassVtable.hpp"
33 #include "runtime/sharedRuntime.hpp"
34 #include "vmreg_sparc.inline.hpp"
35 #ifdef COMPILER2
36 #include "opto/runtime.hpp"
37 #endif
39 // machine-dependent part of VtableStubs: create vtableStub of correct size and
40 // initialize its code
42 #define __ masm->
45 #ifndef PRODUCT
46 extern "C" void bad_compiled_vtable_index(JavaThread* thread, oopDesc* receiver, int index);
47 #endif
50 // Used by compiler only; may use only caller saved, non-argument registers
51 // NOTE: %%%% if any change is made to this stub make sure that the function
52 // pd_code_size_limit is changed to ensure the correct size for VtableStub
53 VtableStub* VtableStubs::create_vtable_stub(int vtable_index) {
54 const int sparc_code_length = VtableStub::pd_code_size_limit(true);
55 VtableStub* s = new(sparc_code_length) VtableStub(true, vtable_index);
56 ResourceMark rm;
57 CodeBuffer cb(s->entry_point(), sparc_code_length);
58 MacroAssembler* masm = new MacroAssembler(&cb);
60 #ifndef PRODUCT
61 if (CountCompiledCalls) {
62 __ inc_counter(SharedRuntime::nof_megamorphic_calls_addr(), G5, G3_scratch);
63 }
64 #endif /* PRODUCT */
66 assert(VtableStub::receiver_location() == O0->as_VMReg(), "receiver expected in O0");
68 // get receiver klass
69 address npe_addr = __ pc();
70 __ load_klass(O0, G3_scratch);
72 // set methodOop (in case of interpreted method), and destination address
73 #ifndef PRODUCT
74 if (DebugVtables) {
75 Label L;
76 // check offset vs vtable length
77 __ ld(G3_scratch, instanceKlass::vtable_length_offset()*wordSize, G5);
78 __ cmp_and_br_short(G5, vtable_index*vtableEntry::size(), Assembler::greaterUnsigned, Assembler::pt, L);
79 __ set(vtable_index, O2);
80 __ call_VM(noreg, CAST_FROM_FN_PTR(address, bad_compiled_vtable_index), O0, O2);
81 __ bind(L);
82 }
83 #endif
85 __ lookup_virtual_method(G3_scratch, vtable_index, G5_method);
87 #ifndef PRODUCT
88 if (DebugVtables) {
89 Label L;
90 __ br_notnull_short(G5_method, Assembler::pt, L);
91 __ stop("Vtable entry is ZERO");
92 __ bind(L);
93 }
94 #endif
96 address ame_addr = __ pc(); // if the vtable entry is null, the method is abstract
97 // NOTE: for vtable dispatches, the vtable entry will never be null.
99 __ ld_ptr(G5_method, in_bytes(methodOopDesc::from_compiled_offset()), G3_scratch);
101 // jump to target (either compiled code or c2iadapter)
102 __ JMP(G3_scratch, 0);
103 // load methodOop (in case we call c2iadapter)
104 __ delayed()->nop();
106 masm->flush();
108 if (PrintMiscellaneous && (WizardMode || Verbose)) {
109 tty->print_cr("vtable #%d at "PTR_FORMAT"[%d] left over: %d",
110 vtable_index, s->entry_point(),
111 (int)(s->code_end() - s->entry_point()),
112 (int)(s->code_end() - __ pc()));
113 }
114 guarantee(__ pc() <= s->code_end(), "overflowed buffer");
115 // shut the door on sizing bugs
116 int slop = 2*BytesPerInstWord; // 32-bit offset is this much larger than a 13-bit one
117 assert(vtable_index > 10 || __ pc() + slop <= s->code_end(), "room for sethi;add");
119 s->set_exception_points(npe_addr, ame_addr);
120 return s;
121 }
124 // NOTE: %%%% if any change is made to this stub make sure that the function
125 // pd_code_size_limit is changed to ensure the correct size for VtableStub
126 VtableStub* VtableStubs::create_itable_stub(int itable_index) {
127 const int sparc_code_length = VtableStub::pd_code_size_limit(false);
128 VtableStub* s = new(sparc_code_length) VtableStub(false, itable_index);
129 ResourceMark rm;
130 CodeBuffer cb(s->entry_point(), sparc_code_length);
131 MacroAssembler* masm = new MacroAssembler(&cb);
133 Register G3_klassOop = G3_scratch;
134 Register G5_interface = G5; // Passed in as an argument
135 Label search;
137 // Entry arguments:
138 // G5_interface: Interface
139 // O0: Receiver
140 assert(VtableStub::receiver_location() == O0->as_VMReg(), "receiver expected in O0");
142 // get receiver klass (also an implicit null-check)
143 address npe_addr = __ pc();
144 __ load_klass(O0, G3_klassOop);
145 __ verify_oop(G3_klassOop);
147 // Push a new window to get some temp registers. This chops the head of all
148 // my 64-bit %o registers in the LION build, but this is OK because no longs
149 // are passed in the %o registers. Instead, longs are passed in G1 and G4
150 // and so those registers are not available here.
151 __ save(SP,-frame::register_save_words*wordSize,SP);
153 #ifndef PRODUCT
154 if (CountCompiledCalls) {
155 __ inc_counter(SharedRuntime::nof_megamorphic_calls_addr(), L0, L1);
156 }
157 #endif /* PRODUCT */
159 Label throw_icce;
161 Register L5_method = L5;
162 __ lookup_interface_method(// inputs: rec. class, interface, itable index
163 G3_klassOop, G5_interface, itable_index,
164 // outputs: method, scan temp. reg
165 L5_method, L2, L3,
166 throw_icce);
168 #ifndef PRODUCT
169 if (DebugVtables) {
170 Label L01;
171 __ br_notnull_short(L5_method, Assembler::pt, L01);
172 __ stop("methodOop is null");
173 __ bind(L01);
174 __ verify_oop(L5_method);
175 }
176 #endif
178 // If the following load is through a NULL pointer, we'll take an OS
179 // exception that should translate into an AbstractMethodError. We need the
180 // window count to be correct at that time.
181 __ restore(L5_method, 0, G5_method);
182 // Restore registers *before* the AME point.
184 address ame_addr = __ pc(); // if the vtable entry is null, the method is abstract
185 __ ld_ptr(G5_method, in_bytes(methodOopDesc::from_compiled_offset()), G3_scratch);
187 // G5_method: methodOop
188 // O0: Receiver
189 // G3_scratch: entry point
190 __ JMP(G3_scratch, 0);
191 __ delayed()->nop();
193 __ bind(throw_icce);
194 AddressLiteral icce(StubRoutines::throw_IncompatibleClassChangeError_entry());
195 __ jump_to(icce, G3_scratch);
196 __ delayed()->restore();
198 masm->flush();
200 if (PrintMiscellaneous && (WizardMode || Verbose)) {
201 tty->print_cr("itable #%d at "PTR_FORMAT"[%d] left over: %d",
202 itable_index, s->entry_point(),
203 (int)(s->code_end() - s->entry_point()),
204 (int)(s->code_end() - __ pc()));
205 }
206 guarantee(__ pc() <= s->code_end(), "overflowed buffer");
207 // shut the door on sizing bugs
208 int slop = 2*BytesPerInstWord; // 32-bit offset is this much larger than a 13-bit one
209 assert(itable_index > 10 || __ pc() + slop <= s->code_end(), "room for sethi;add");
211 s->set_exception_points(npe_addr, ame_addr);
212 return s;
213 }
216 int VtableStub::pd_code_size_limit(bool is_vtable_stub) {
217 if (TraceJumps || DebugVtables || CountCompiledCalls || VerifyOops) return 1000;
218 else {
219 const int slop = 2*BytesPerInstWord; // sethi;add (needed for long offsets)
220 if (is_vtable_stub) {
221 // ld;ld;ld,jmp,nop
222 const int basic = 5*BytesPerInstWord +
223 // shift;add for load_klass (only shift with zero heap based)
224 (UseCompressedOops ?
225 ((Universe::narrow_oop_base() == NULL) ? BytesPerInstWord : 2*BytesPerInstWord) : 0);
226 return basic + slop;
227 } else {
228 const int basic = (28 LP64_ONLY(+ 6)) * BytesPerInstWord +
229 // shift;add for load_klass (only shift with zero heap based)
230 (UseCompressedOops ?
231 ((Universe::narrow_oop_base() == NULL) ? BytesPerInstWord : 2*BytesPerInstWord) : 0);
232 return (basic + slop);
233 }
234 }
236 // In order to tune these parameters, run the JVM with VM options
237 // +PrintMiscellaneous and +WizardMode to see information about
238 // actual itable stubs. Look for lines like this:
239 // itable #1 at 0x5551212[116] left over: 8
240 // Reduce the constants so that the "left over" number is 8
241 // Do not aim at a left-over number of zero, because a very
242 // large vtable or itable offset (> 4K) will require an extra
243 // sethi/or pair of instructions.
244 //
245 // The JVM98 app. _202_jess has a megamorphic interface call.
246 // The itable code looks like this:
247 // Decoding VtableStub itbl[1]@16
248 // ld [ %o0 + 4 ], %g3
249 // save %sp, -64, %sp
250 // ld [ %g3 + 0xe8 ], %l2
251 // sll %l2, 2, %l2
252 // add %l2, 0x134, %l2
253 // and %l2, -8, %l2 ! NOT_LP64 only
254 // add %g3, %l2, %l2
255 // add %g3, 4, %g3
256 // ld [ %l2 ], %l5
257 // brz,pn %l5, throw_icce
258 // cmp %l5, %g5
259 // be %icc, success
260 // add %l2, 8, %l2
261 // loop:
262 // ld [ %l2 ], %l5
263 // brz,pn %l5, throw_icce
264 // cmp %l5, %g5
265 // bne,pn %icc, loop
266 // add %l2, 8, %l2
267 // success:
268 // ld [ %l2 + -4 ], %l2
269 // ld [ %g3 + %l2 ], %l5
270 // restore %l5, 0, %g5
271 // ld [ %g5 + 0x44 ], %g3
272 // jmp %g3
273 // nop
274 // throw_icce:
275 // sethi %hi(throw_ICCE_entry), %g3
276 // ! 5 more instructions here, LP64_ONLY
277 // jmp %g3 + %lo(throw_ICCE_entry)
278 // restore
279 }
282 int VtableStub::pd_code_alignment() {
283 // UltraSPARC cache line size is 8 instructions:
284 const unsigned int icache_line_size = 32;
285 return icache_line_size;
286 }