Mercurial > hg > graal-compiler
annotate src/cpu/x86/vm/interpreter_x86_64.cpp @ 144:e3729351c946
6697534: Premature GC and invalid lgrp selection with NUMA-aware allocator.
Summary: Don't move tops of the chunks in ensure_parsibility(). Handle the situation with Solaris when a machine has a locality group with no memory.
Reviewed-by: apetrusenko, jcoomes, ysr
author | iveresov |
---|---|
date | Fri, 09 May 2008 16:34:08 +0400 |
parents | ba764ed4b6f2 |
children | d1605aabd0a1 |
rev | line source |
---|---|
0 | 1 /* |
2 * Copyright 2003-2007 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
20 * CA 95054 USA or visit www.sun.com if you need additional information or | |
21 * have any questions. | |
22 * | |
23 */ | |
24 | |
25 #include "incls/_precompiled.incl" | |
26 #include "incls/_interpreter_x86_64.cpp.incl" | |
27 | |
28 #define __ _masm-> | |
29 | |
30 | |
31 #ifdef _WIN64 | |
32 address AbstractInterpreterGenerator::generate_slow_signature_handler() { | |
33 address entry = __ pc(); | |
34 | |
35 // rbx: method | |
36 // r14: pointer to locals | |
37 // c_rarg3: first stack arg - wordSize | |
38 __ movq(c_rarg3, rsp); | |
39 // adjust rsp | |
40 __ subq(rsp, 4 * wordSize); | |
41 __ call_VM(noreg, | |
42 CAST_FROM_FN_PTR(address, | |
43 InterpreterRuntime::slow_signature_handler), | |
44 rbx, r14, c_rarg3); | |
45 | |
46 // rax: result handler | |
47 | |
48 // Stack layout: | |
49 // rsp: 3 integer or float args (if static first is unused) | |
50 // 1 float/double identifiers | |
51 // return address | |
52 // stack args | |
53 // garbage | |
54 // expression stack bottom | |
55 // bcp (NULL) | |
56 // ... | |
57 | |
58 // Do FP first so we can use c_rarg3 as temp | |
59 __ movl(c_rarg3, Address(rsp, 3 * wordSize)); // float/double identifiers | |
60 | |
61 for ( int i= 0; i < Argument::n_int_register_parameters_c-1; i++ ) { | |
62 XMMRegister floatreg = as_XMMRegister(i+1); | |
63 Label isfloatordouble, isdouble, next; | |
64 | |
65 __ testl(c_rarg3, 1 << (i*2)); // Float or Double? | |
66 __ jcc(Assembler::notZero, isfloatordouble); | |
67 | |
68 // Do Int register here | |
69 switch ( i ) { | |
70 case 0: | |
71 __ movl(rscratch1, Address(rbx, methodOopDesc::access_flags_offset())); | |
72 __ testl(rscratch1, JVM_ACC_STATIC); | |
73 __ cmovq(Assembler::zero, c_rarg1, Address(rsp, 0)); | |
74 break; | |
75 case 1: | |
76 __ movq(c_rarg2, Address(rsp, wordSize)); | |
77 break; | |
78 case 2: | |
79 __ movq(c_rarg3, Address(rsp, 2 * wordSize)); | |
80 break; | |
81 default: | |
82 break; | |
83 } | |
84 | |
85 __ jmp (next); | |
86 | |
87 __ bind(isfloatordouble); | |
88 __ testl(c_rarg3, 1 << ((i*2)+1)); // Double? | |
89 __ jcc(Assembler::notZero, isdouble); | |
90 | |
91 // Do Float Here | |
92 __ movflt(floatreg, Address(rsp, i * wordSize)); | |
93 __ jmp(next); | |
94 | |
95 // Do Double here | |
96 __ bind(isdouble); | |
97 __ movdbl(floatreg, Address(rsp, i * wordSize)); | |
98 | |
99 __ bind(next); | |
100 } | |
101 | |
102 | |
103 // restore rsp | |
104 __ addq(rsp, 4 * wordSize); | |
105 | |
106 __ ret(0); | |
107 | |
108 return entry; | |
109 } | |
110 #else | |
111 address AbstractInterpreterGenerator::generate_slow_signature_handler() { | |
112 address entry = __ pc(); | |
113 | |
114 // rbx: method | |
115 // r14: pointer to locals | |
116 // c_rarg3: first stack arg - wordSize | |
117 __ movq(c_rarg3, rsp); | |
118 // adjust rsp | |
119 __ subq(rsp, 14 * wordSize); | |
120 __ call_VM(noreg, | |
121 CAST_FROM_FN_PTR(address, | |
122 InterpreterRuntime::slow_signature_handler), | |
123 rbx, r14, c_rarg3); | |
124 | |
125 // rax: result handler | |
126 | |
127 // Stack layout: | |
128 // rsp: 5 integer args (if static first is unused) | |
129 // 1 float/double identifiers | |
130 // 8 double args | |
131 // return address | |
132 // stack args | |
133 // garbage | |
134 // expression stack bottom | |
135 // bcp (NULL) | |
136 // ... | |
137 | |
138 // Do FP first so we can use c_rarg3 as temp | |
139 __ movl(c_rarg3, Address(rsp, 5 * wordSize)); // float/double identifiers | |
140 | |
141 for (int i = 0; i < Argument::n_float_register_parameters_c; i++) { | |
142 const XMMRegister r = as_XMMRegister(i); | |
143 | |
144 Label d, done; | |
145 | |
146 __ testl(c_rarg3, 1 << i); | |
147 __ jcc(Assembler::notZero, d); | |
148 __ movflt(r, Address(rsp, (6 + i) * wordSize)); | |
149 __ jmp(done); | |
150 __ bind(d); | |
151 __ movdbl(r, Address(rsp, (6 + i) * wordSize)); | |
152 __ bind(done); | |
153 } | |
154 | |
155 // Now handle integrals. Only do c_rarg1 if not static. | |
156 __ movl(c_rarg3, Address(rbx, methodOopDesc::access_flags_offset())); | |
157 __ testl(c_rarg3, JVM_ACC_STATIC); | |
158 __ cmovq(Assembler::zero, c_rarg1, Address(rsp, 0)); | |
159 | |
160 __ movq(c_rarg2, Address(rsp, wordSize)); | |
161 __ movq(c_rarg3, Address(rsp, 2 * wordSize)); | |
162 __ movq(c_rarg4, Address(rsp, 3 * wordSize)); | |
163 __ movq(c_rarg5, Address(rsp, 4 * wordSize)); | |
164 | |
165 // restore rsp | |
166 __ addq(rsp, 14 * wordSize); | |
167 | |
168 __ ret(0); | |
169 | |
170 return entry; | |
171 } | |
172 #endif | |
173 | |
174 | |
175 // | |
176 // Various method entries | |
177 // | |
178 | |
179 address InterpreterGenerator::generate_math_entry( | |
180 AbstractInterpreter::MethodKind kind) { | |
181 // rbx: methodOop | |
182 | |
183 if (!InlineIntrinsics) return NULL; // Generate a vanilla entry | |
184 | |
185 assert(kind == Interpreter::java_lang_math_sqrt, | |
186 "Other intrinsics are not special"); | |
187 | |
188 address entry_point = __ pc(); | |
189 | |
190 // These don't need a safepoint check because they aren't virtually | |
191 // callable. We won't enter these intrinsics from compiled code. | |
192 // If in the future we added an intrinsic which was virtually callable | |
193 // we'd have to worry about how to safepoint so that this code is used. | |
194 | |
195 // mathematical functions inlined by compiler | |
196 // (interpreter must provide identical implementation | |
197 // in order to avoid monotonicity bugs when switching | |
198 // from interpreter to compiler in the middle of some | |
199 // computation) | |
200 | |
201 // Note: For JDK 1.2 StrictMath doesn't exist and Math.sin/cos/sqrt are | |
202 // native methods. Interpreter::method_kind(...) does a check for | |
203 // native methods first before checking for intrinsic methods and | |
204 // thus will never select this entry point. Make sure it is not | |
205 // called accidentally since the SharedRuntime entry points will | |
206 // not work for JDK 1.2. | |
207 // | |
208 // We no longer need to check for JDK 1.2 since it's EOL'ed. | |
209 // The following check existed in pre 1.6 implementation, | |
210 // if (Universe::is_jdk12x_version()) { | |
211 // __ should_not_reach_here(); | |
212 // } | |
213 // Universe::is_jdk12x_version() always returns false since | |
214 // the JDK version is not yet determined when this method is called. | |
215 // This method is called during interpreter_init() whereas | |
216 // JDK version is only determined when universe2_init() is called. | |
217 | |
218 // Note: For JDK 1.3 StrictMath exists and Math.sin/cos/sqrt are | |
219 // java methods. Interpreter::method_kind(...) will select | |
220 // this entry point for the corresponding methods in JDK 1.3. | |
221 __ sqrtsd(xmm0, Address(rsp, wordSize)); | |
222 | |
223 __ popq(rax); | |
224 __ movq(rsp, r13); | |
225 __ jmp(rax); | |
226 | |
227 return entry_point; | |
228 } | |
229 | |
230 | |
231 // Abstract method entry | |
232 // Attempt to execute abstract method. Throw exception | |
233 address InterpreterGenerator::generate_abstract_entry(void) { | |
234 // rbx: methodOop | |
235 // r13: sender SP | |
236 | |
237 address entry_point = __ pc(); | |
238 | |
239 // abstract method entry | |
240 // remove return address. Not really needed, since exception | |
241 // handling throws away expression stack | |
242 __ popq(rbx); | |
243 | |
244 // adjust stack to what a normal return would do | |
245 __ movq(rsp, r13); | |
246 | |
247 // throw exception | |
248 __ call_VM(noreg, CAST_FROM_FN_PTR(address, | |
249 InterpreterRuntime::throw_AbstractMethodError)); | |
250 // the call_VM checks for exception, so we should never return here. | |
251 __ should_not_reach_here(); | |
252 | |
253 return entry_point; | |
254 } | |
255 | |
256 | |
257 // Empty method, generate a very fast return. | |
258 | |
259 address InterpreterGenerator::generate_empty_entry(void) { | |
260 // rbx: methodOop | |
261 // r13: sender sp must set sp to this value on return | |
262 | |
263 if (!UseFastEmptyMethods) { | |
264 return NULL; | |
265 } | |
266 | |
267 address entry_point = __ pc(); | |
268 | |
269 // If we need a safepoint check, generate full interpreter entry. | |
270 Label slow_path; | |
271 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), | |
272 SafepointSynchronize::_not_synchronized); | |
273 __ jcc(Assembler::notEqual, slow_path); | |
274 | |
275 // do nothing for empty methods (do not even increment invocation counter) | |
276 // Code: _return | |
277 // _return | |
278 // return w/o popping parameters | |
279 __ popq(rax); | |
280 __ movq(rsp, r13); | |
281 __ jmp(rax); | |
282 | |
283 __ bind(slow_path); | |
284 (void) generate_normal_entry(false); | |
285 return entry_point; | |
286 | |
287 } | |
288 | |
289 // Call an accessor method (assuming it is resolved, otherwise drop | |
290 // into vanilla (slow path) entry | |
291 address InterpreterGenerator::generate_accessor_entry(void) { | |
292 // rbx: methodOop | |
293 | |
294 // r13: senderSP must preserver for slow path, set SP to it on fast path | |
295 | |
296 address entry_point = __ pc(); | |
297 Label xreturn_path; | |
298 | |
299 // do fastpath for resolved accessor methods | |
300 if (UseFastAccessorMethods) { | |
301 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites | |
302 // thereof; parameter size = 1 | |
303 // Note: We can only use this code if the getfield has been resolved | |
304 // and if we don't have a null-pointer exception => check for | |
305 // these conditions first and use slow path if necessary. | |
306 Label slow_path; | |
307 // If we need a safepoint check, generate full interpreter entry. | |
308 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), | |
309 SafepointSynchronize::_not_synchronized); | |
310 | |
311 __ jcc(Assembler::notEqual, slow_path); | |
312 // rbx: method | |
313 __ movq(rax, Address(rsp, wordSize)); | |
314 | |
315 // check if local 0 != NULL and read field | |
316 __ testq(rax, rax); | |
317 __ jcc(Assembler::zero, slow_path); | |
318 | |
319 __ movq(rdi, Address(rbx, methodOopDesc::constants_offset())); | |
320 // read first instruction word and extract bytecode @ 1 and index @ 2 | |
321 __ movq(rdx, Address(rbx, methodOopDesc::const_offset())); | |
322 __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset())); | |
323 // Shift codes right to get the index on the right. | |
324 // The bytecode fetched looks like <index><0xb4><0x2a> | |
325 __ shrl(rdx, 2 * BitsPerByte); | |
326 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size()))); | |
327 __ movq(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes())); | |
328 | |
329 // rax: local 0 | |
330 // rbx: method | |
331 // rdx: constant pool cache index | |
332 // rdi: constant pool cache | |
333 | |
334 // check if getfield has been resolved and read constant pool cache entry | |
335 // check the validity of the cache entry by testing whether _indices field | |
336 // contains Bytecode::_getfield in b1 byte. | |
337 assert(in_words(ConstantPoolCacheEntry::size()) == 4, | |
338 "adjust shift below"); | |
339 __ movl(rcx, | |
340 Address(rdi, | |
341 rdx, | |
342 Address::times_8, | |
343 constantPoolCacheOopDesc::base_offset() + | |
344 ConstantPoolCacheEntry::indices_offset())); | |
345 __ shrl(rcx, 2 * BitsPerByte); | |
346 __ andl(rcx, 0xFF); | |
347 __ cmpl(rcx, Bytecodes::_getfield); | |
348 __ jcc(Assembler::notEqual, slow_path); | |
349 | |
350 // Note: constant pool entry is not valid before bytecode is resolved | |
351 __ movq(rcx, | |
352 Address(rdi, | |
353 rdx, | |
354 Address::times_8, | |
355 constantPoolCacheOopDesc::base_offset() + | |
356 ConstantPoolCacheEntry::f2_offset())); | |
357 // edx: flags | |
358 __ movl(rdx, | |
359 Address(rdi, | |
360 rdx, | |
361 Address::times_8, | |
362 constantPoolCacheOopDesc::base_offset() + | |
363 ConstantPoolCacheEntry::flags_offset())); | |
364 | |
365 Label notObj, notInt, notByte, notShort; | |
366 const Address field_address(rax, rcx, Address::times_1); | |
367 | |
368 // Need to differentiate between igetfield, agetfield, bgetfield etc. | |
369 // because they are different sizes. | |
370 // Use the type from the constant pool cache | |
371 __ shrl(rdx, ConstantPoolCacheEntry::tosBits); | |
372 // Make sure we don't need to mask edx for tosBits after the above shift | |
373 ConstantPoolCacheEntry::verify_tosBits(); | |
374 | |
375 __ cmpl(rdx, atos); | |
376 __ jcc(Assembler::notEqual, notObj); | |
377 // atos | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
378 __ load_heap_oop(rax, field_address); |
0 | 379 __ jmp(xreturn_path); |
380 | |
381 __ bind(notObj); | |
382 __ cmpl(rdx, itos); | |
383 __ jcc(Assembler::notEqual, notInt); | |
384 // itos | |
385 __ movl(rax, field_address); | |
386 __ jmp(xreturn_path); | |
387 | |
388 __ bind(notInt); | |
389 __ cmpl(rdx, btos); | |
390 __ jcc(Assembler::notEqual, notByte); | |
391 // btos | |
392 __ load_signed_byte(rax, field_address); | |
393 __ jmp(xreturn_path); | |
394 | |
395 __ bind(notByte); | |
396 __ cmpl(rdx, stos); | |
397 __ jcc(Assembler::notEqual, notShort); | |
398 // stos | |
399 __ load_signed_word(rax, field_address); | |
400 __ jmp(xreturn_path); | |
401 | |
402 __ bind(notShort); | |
403 #ifdef ASSERT | |
404 Label okay; | |
405 __ cmpl(rdx, ctos); | |
406 __ jcc(Assembler::equal, okay); | |
407 __ stop("what type is this?"); | |
408 __ bind(okay); | |
409 #endif | |
410 // ctos | |
411 __ load_unsigned_word(rax, field_address); | |
412 | |
413 __ bind(xreturn_path); | |
414 | |
415 // _ireturn/_areturn | |
416 __ popq(rdi); | |
417 __ movq(rsp, r13); | |
418 __ jmp(rdi); | |
419 __ ret(0); | |
420 | |
421 // generate a vanilla interpreter entry as the slow path | |
422 __ bind(slow_path); | |
423 (void) generate_normal_entry(false); | |
424 } else { | |
425 (void) generate_normal_entry(false); | |
426 } | |
427 | |
428 return entry_point; | |
429 } | |
430 | |
431 // This method tells the deoptimizer how big an interpreted frame must be: | |
432 int AbstractInterpreter::size_activation(methodOop method, | |
433 int tempcount, | |
434 int popframe_extra_args, | |
435 int moncount, | |
436 int callee_param_count, | |
437 int callee_locals, | |
438 bool is_top_frame) { | |
439 return layout_activation(method, | |
440 tempcount, popframe_extra_args, moncount, | |
441 callee_param_count, callee_locals, | |
442 (frame*) NULL, (frame*) NULL, is_top_frame); | |
443 } | |
444 | |
445 void Deoptimization::unwind_callee_save_values(frame* f, vframeArray* vframe_array) { | |
446 | |
447 // This code is sort of the equivalent of C2IAdapter::setup_stack_frame back in | |
448 // the days we had adapter frames. When we deoptimize a situation where a | |
449 // compiled caller calls a compiled caller will have registers it expects | |
450 // to survive the call to the callee. If we deoptimize the callee the only | |
451 // way we can restore these registers is to have the oldest interpreter | |
452 // frame that we create restore these values. That is what this routine | |
453 // will accomplish. | |
454 | |
455 // At the moment we have modified c2 to not have any callee save registers | |
456 // so this problem does not exist and this routine is just a place holder. | |
457 | |
458 assert(f->is_interpreted_frame(), "must be interpreted"); | |
459 } |