Thu, 15 Aug 2013 20:04:10 -0400
8003424: Enable Class Data Sharing for CompressedOops
8016729: ObjectAlignmentInBytes=16 now forces the use of heap based compressed oops
8005933: The -Xshare:auto option is ignored for -server
Summary: Move klass metaspace above the heap and support CDS with compressed klass ptrs.
Reviewed-by: coleenp, kvn, mgerdin, tschatzl, stefank
1 /*
2 * Copyright (c) 2003, 2012, 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/macroAssembler.hpp"
27 #include "interpreter/bytecodeHistogram.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "interpreter/interpreterGenerator.hpp"
30 #include "interpreter/interpreterRuntime.hpp"
31 #include "interpreter/templateTable.hpp"
32 #include "oops/arrayOop.hpp"
33 #include "oops/methodData.hpp"
34 #include "oops/method.hpp"
35 #include "oops/oop.inline.hpp"
36 #include "prims/jvmtiExport.hpp"
37 #include "prims/jvmtiThreadState.hpp"
38 #include "prims/methodHandles.hpp"
39 #include "runtime/arguments.hpp"
40 #include "runtime/deoptimization.hpp"
41 #include "runtime/frame.inline.hpp"
42 #include "runtime/sharedRuntime.hpp"
43 #include "runtime/stubRoutines.hpp"
44 #include "runtime/synchronizer.hpp"
45 #include "runtime/timer.hpp"
46 #include "runtime/vframeArray.hpp"
47 #include "utilities/debug.hpp"
48 #ifdef COMPILER1
49 #include "c1/c1_Runtime1.hpp"
50 #endif
52 #define __ _masm->
55 #ifdef _WIN64
56 address AbstractInterpreterGenerator::generate_slow_signature_handler() {
57 address entry = __ pc();
59 // rbx: method
60 // r14: pointer to locals
61 // c_rarg3: first stack arg - wordSize
62 __ mov(c_rarg3, rsp);
63 // adjust rsp
64 __ subptr(rsp, 4 * wordSize);
65 __ call_VM(noreg,
66 CAST_FROM_FN_PTR(address,
67 InterpreterRuntime::slow_signature_handler),
68 rbx, r14, c_rarg3);
70 // rax: result handler
72 // Stack layout:
73 // rsp: 3 integer or float args (if static first is unused)
74 // 1 float/double identifiers
75 // return address
76 // stack args
77 // garbage
78 // expression stack bottom
79 // bcp (NULL)
80 // ...
82 // Do FP first so we can use c_rarg3 as temp
83 __ movl(c_rarg3, Address(rsp, 3 * wordSize)); // float/double identifiers
85 for ( int i= 0; i < Argument::n_int_register_parameters_c-1; i++ ) {
86 XMMRegister floatreg = as_XMMRegister(i+1);
87 Label isfloatordouble, isdouble, next;
89 __ testl(c_rarg3, 1 << (i*2)); // Float or Double?
90 __ jcc(Assembler::notZero, isfloatordouble);
92 // Do Int register here
93 switch ( i ) {
94 case 0:
95 __ movl(rscratch1, Address(rbx, Method::access_flags_offset()));
96 __ testl(rscratch1, JVM_ACC_STATIC);
97 __ cmovptr(Assembler::zero, c_rarg1, Address(rsp, 0));
98 break;
99 case 1:
100 __ movptr(c_rarg2, Address(rsp, wordSize));
101 break;
102 case 2:
103 __ movptr(c_rarg3, Address(rsp, 2 * wordSize));
104 break;
105 default:
106 break;
107 }
109 __ jmp (next);
111 __ bind(isfloatordouble);
112 __ testl(c_rarg3, 1 << ((i*2)+1)); // Double?
113 __ jcc(Assembler::notZero, isdouble);
115 // Do Float Here
116 __ movflt(floatreg, Address(rsp, i * wordSize));
117 __ jmp(next);
119 // Do Double here
120 __ bind(isdouble);
121 __ movdbl(floatreg, Address(rsp, i * wordSize));
123 __ bind(next);
124 }
127 // restore rsp
128 __ addptr(rsp, 4 * wordSize);
130 __ ret(0);
132 return entry;
133 }
134 #else
135 address AbstractInterpreterGenerator::generate_slow_signature_handler() {
136 address entry = __ pc();
138 // rbx: method
139 // r14: pointer to locals
140 // c_rarg3: first stack arg - wordSize
141 __ mov(c_rarg3, rsp);
142 // adjust rsp
143 __ subptr(rsp, 14 * wordSize);
144 __ call_VM(noreg,
145 CAST_FROM_FN_PTR(address,
146 InterpreterRuntime::slow_signature_handler),
147 rbx, r14, c_rarg3);
149 // rax: result handler
151 // Stack layout:
152 // rsp: 5 integer args (if static first is unused)
153 // 1 float/double identifiers
154 // 8 double args
155 // return address
156 // stack args
157 // garbage
158 // expression stack bottom
159 // bcp (NULL)
160 // ...
162 // Do FP first so we can use c_rarg3 as temp
163 __ movl(c_rarg3, Address(rsp, 5 * wordSize)); // float/double identifiers
165 for (int i = 0; i < Argument::n_float_register_parameters_c; i++) {
166 const XMMRegister r = as_XMMRegister(i);
168 Label d, done;
170 __ testl(c_rarg3, 1 << i);
171 __ jcc(Assembler::notZero, d);
172 __ movflt(r, Address(rsp, (6 + i) * wordSize));
173 __ jmp(done);
174 __ bind(d);
175 __ movdbl(r, Address(rsp, (6 + i) * wordSize));
176 __ bind(done);
177 }
179 // Now handle integrals. Only do c_rarg1 if not static.
180 __ movl(c_rarg3, Address(rbx, Method::access_flags_offset()));
181 __ testl(c_rarg3, JVM_ACC_STATIC);
182 __ cmovptr(Assembler::zero, c_rarg1, Address(rsp, 0));
184 __ movptr(c_rarg2, Address(rsp, wordSize));
185 __ movptr(c_rarg3, Address(rsp, 2 * wordSize));
186 __ movptr(c_rarg4, Address(rsp, 3 * wordSize));
187 __ movptr(c_rarg5, Address(rsp, 4 * wordSize));
189 // restore rsp
190 __ addptr(rsp, 14 * wordSize);
192 __ ret(0);
194 return entry;
195 }
196 #endif
199 //
200 // Various method entries
201 //
203 address InterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
205 // rbx,: Method*
206 // rcx: scratrch
207 // r13: sender sp
209 if (!InlineIntrinsics) return NULL; // Generate a vanilla entry
211 address entry_point = __ pc();
213 // These don't need a safepoint check because they aren't virtually
214 // callable. We won't enter these intrinsics from compiled code.
215 // If in the future we added an intrinsic which was virtually callable
216 // we'd have to worry about how to safepoint so that this code is used.
218 // mathematical functions inlined by compiler
219 // (interpreter must provide identical implementation
220 // in order to avoid monotonicity bugs when switching
221 // from interpreter to compiler in the middle of some
222 // computation)
223 //
224 // stack: [ ret adr ] <-- rsp
225 // [ lo(arg) ]
226 // [ hi(arg) ]
227 //
229 // Note: For JDK 1.2 StrictMath doesn't exist and Math.sin/cos/sqrt are
230 // native methods. Interpreter::method_kind(...) does a check for
231 // native methods first before checking for intrinsic methods and
232 // thus will never select this entry point. Make sure it is not
233 // called accidentally since the SharedRuntime entry points will
234 // not work for JDK 1.2.
235 //
236 // We no longer need to check for JDK 1.2 since it's EOL'ed.
237 // The following check existed in pre 1.6 implementation,
238 // if (Universe::is_jdk12x_version()) {
239 // __ should_not_reach_here();
240 // }
241 // Universe::is_jdk12x_version() always returns false since
242 // the JDK version is not yet determined when this method is called.
243 // This method is called during interpreter_init() whereas
244 // JDK version is only determined when universe2_init() is called.
246 // Note: For JDK 1.3 StrictMath exists and Math.sin/cos/sqrt are
247 // java methods. Interpreter::method_kind(...) will select
248 // this entry point for the corresponding methods in JDK 1.3.
249 // get argument
251 if (kind == Interpreter::java_lang_math_sqrt) {
252 __ sqrtsd(xmm0, Address(rsp, wordSize));
253 } else {
254 __ fld_d(Address(rsp, wordSize));
255 switch (kind) {
256 case Interpreter::java_lang_math_sin :
257 __ trigfunc('s');
258 break;
259 case Interpreter::java_lang_math_cos :
260 __ trigfunc('c');
261 break;
262 case Interpreter::java_lang_math_tan :
263 __ trigfunc('t');
264 break;
265 case Interpreter::java_lang_math_abs:
266 __ fabs();
267 break;
268 case Interpreter::java_lang_math_log:
269 __ flog();
270 break;
271 case Interpreter::java_lang_math_log10:
272 __ flog10();
273 break;
274 case Interpreter::java_lang_math_pow:
275 __ fld_d(Address(rsp, 3*wordSize)); // second argument (one
276 // empty stack slot)
277 __ pow_with_fallback(0);
278 break;
279 case Interpreter::java_lang_math_exp:
280 __ exp_with_fallback(0);
281 break;
282 default :
283 ShouldNotReachHere();
284 }
286 // return double result in xmm0 for interpreter and compilers.
287 __ subptr(rsp, 2*wordSize);
288 // Round to 64bit precision
289 __ fstp_d(Address(rsp, 0));
290 __ movdbl(xmm0, Address(rsp, 0));
291 __ addptr(rsp, 2*wordSize);
292 }
295 __ pop(rax);
296 __ mov(rsp, r13);
297 __ jmp(rax);
299 return entry_point;
300 }
303 // Abstract method entry
304 // Attempt to execute abstract method. Throw exception
305 address InterpreterGenerator::generate_abstract_entry(void) {
306 // rbx: Method*
307 // r13: sender SP
309 address entry_point = __ pc();
311 // abstract method entry
313 // pop return address, reset last_sp to NULL
314 __ empty_expression_stack();
315 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
316 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
318 // throw exception
319 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
320 InterpreterRuntime::throw_AbstractMethodError));
321 // the call_VM checks for exception, so we should never return here.
322 __ should_not_reach_here();
324 return entry_point;
325 }
328 // Empty method, generate a very fast return.
330 address InterpreterGenerator::generate_empty_entry(void) {
331 // rbx: Method*
332 // r13: sender sp must set sp to this value on return
334 if (!UseFastEmptyMethods) {
335 return NULL;
336 }
338 address entry_point = __ pc();
340 // If we need a safepoint check, generate full interpreter entry.
341 Label slow_path;
342 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
343 SafepointSynchronize::_not_synchronized);
344 __ jcc(Assembler::notEqual, slow_path);
346 // do nothing for empty methods (do not even increment invocation counter)
347 // Code: _return
348 // _return
349 // return w/o popping parameters
350 __ pop(rax);
351 __ mov(rsp, r13);
352 __ jmp(rax);
354 __ bind(slow_path);
355 (void) generate_normal_entry(false);
356 return entry_point;
358 }
360 void Deoptimization::unwind_callee_save_values(frame* f, vframeArray* vframe_array) {
362 // This code is sort of the equivalent of C2IAdapter::setup_stack_frame back in
363 // the days we had adapter frames. When we deoptimize a situation where a
364 // compiled caller calls a compiled caller will have registers it expects
365 // to survive the call to the callee. If we deoptimize the callee the only
366 // way we can restore these registers is to have the oldest interpreter
367 // frame that we create restore these values. That is what this routine
368 // will accomplish.
370 // At the moment we have modified c2 to not have any callee save registers
371 // so this problem does not exist and this routine is just a place holder.
373 assert(f->is_interpreted_frame(), "must be interpreted");
374 }