Wed, 17 Feb 2016 13:40:12 +0300
8081778: Use Intel x64 CPU instructions for RSA acceleration
Summary: Add intrinsics for BigInteger squareToLen and mulAdd methods.
Reviewed-by: kvn, jrose
1 /*
2 * Copyright (c) 1998, 2015, 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 "classfile/systemDictionary.hpp"
27 #include "classfile/vmSymbols.hpp"
28 #include "code/compiledIC.hpp"
29 #include "code/icBuffer.hpp"
30 #include "code/nmethod.hpp"
31 #include "code/pcDesc.hpp"
32 #include "code/scopeDesc.hpp"
33 #include "code/vtableStubs.hpp"
34 #include "compiler/compileBroker.hpp"
35 #include "compiler/compilerOracle.hpp"
36 #include "compiler/oopMap.hpp"
37 #include "gc_implementation/g1/g1SATBCardTableModRefBS.hpp"
38 #include "gc_implementation/g1/heapRegion.hpp"
39 #include "gc_interface/collectedHeap.hpp"
40 #include "interpreter/bytecode.hpp"
41 #include "interpreter/interpreter.hpp"
42 #include "interpreter/linkResolver.hpp"
43 #include "memory/barrierSet.hpp"
44 #include "memory/gcLocker.inline.hpp"
45 #include "memory/oopFactory.hpp"
46 #include "oops/objArrayKlass.hpp"
47 #include "oops/oop.inline.hpp"
48 #include "opto/addnode.hpp"
49 #include "opto/callnode.hpp"
50 #include "opto/cfgnode.hpp"
51 #include "opto/connode.hpp"
52 #include "opto/graphKit.hpp"
53 #include "opto/machnode.hpp"
54 #include "opto/matcher.hpp"
55 #include "opto/memnode.hpp"
56 #include "opto/mulnode.hpp"
57 #include "opto/runtime.hpp"
58 #include "opto/subnode.hpp"
59 #include "runtime/fprofiler.hpp"
60 #include "runtime/handles.inline.hpp"
61 #include "runtime/interfaceSupport.hpp"
62 #include "runtime/javaCalls.hpp"
63 #include "runtime/sharedRuntime.hpp"
64 #include "runtime/signature.hpp"
65 #include "runtime/threadCritical.hpp"
66 #include "runtime/vframe.hpp"
67 #include "runtime/vframeArray.hpp"
68 #include "runtime/vframe_hp.hpp"
69 #include "utilities/copy.hpp"
70 #include "utilities/preserveException.hpp"
71 #if defined AD_MD_HPP
72 # include AD_MD_HPP
73 #elif defined TARGET_ARCH_MODEL_x86_32
74 # include "adfiles/ad_x86_32.hpp"
75 #elif defined TARGET_ARCH_MODEL_x86_64
76 # include "adfiles/ad_x86_64.hpp"
77 #elif defined TARGET_ARCH_MODEL_sparc
78 # include "adfiles/ad_sparc.hpp"
79 #elif defined TARGET_ARCH_MODEL_zero
80 # include "adfiles/ad_zero.hpp"
81 #elif defined TARGET_ARCH_MODEL_ppc_64
82 # include "adfiles/ad_ppc_64.hpp"
83 #endif
86 // For debugging purposes:
87 // To force FullGCALot inside a runtime function, add the following two lines
88 //
89 // Universe::release_fullgc_alot_dummy();
90 // MarkSweep::invoke(0, "Debugging");
91 //
92 // At command line specify the parameters: -XX:+FullGCALot -XX:FullGCALotStart=100000000
97 // Compiled code entry points
98 address OptoRuntime::_new_instance_Java = NULL;
99 address OptoRuntime::_new_array_Java = NULL;
100 address OptoRuntime::_new_array_nozero_Java = NULL;
101 address OptoRuntime::_multianewarray2_Java = NULL;
102 address OptoRuntime::_multianewarray3_Java = NULL;
103 address OptoRuntime::_multianewarray4_Java = NULL;
104 address OptoRuntime::_multianewarray5_Java = NULL;
105 address OptoRuntime::_multianewarrayN_Java = NULL;
106 address OptoRuntime::_g1_wb_pre_Java = NULL;
107 address OptoRuntime::_g1_wb_post_Java = NULL;
108 address OptoRuntime::_vtable_must_compile_Java = NULL;
109 address OptoRuntime::_complete_monitor_locking_Java = NULL;
110 address OptoRuntime::_rethrow_Java = NULL;
112 address OptoRuntime::_slow_arraycopy_Java = NULL;
113 address OptoRuntime::_register_finalizer_Java = NULL;
115 # ifdef ENABLE_ZAP_DEAD_LOCALS
116 address OptoRuntime::_zap_dead_Java_locals_Java = NULL;
117 address OptoRuntime::_zap_dead_native_locals_Java = NULL;
118 # endif
120 ExceptionBlob* OptoRuntime::_exception_blob;
122 // This should be called in an assertion at the start of OptoRuntime routines
123 // which are entered from compiled code (all of them)
124 #ifdef ASSERT
125 static bool check_compiled_frame(JavaThread* thread) {
126 assert(thread->last_frame().is_runtime_frame(), "cannot call runtime directly from compiled code");
127 RegisterMap map(thread, false);
128 frame caller = thread->last_frame().sender(&map);
129 assert(caller.is_compiled_frame(), "not being called from compiled like code");
130 return true;
131 }
132 #endif // ASSERT
135 #define gen(env, var, type_func_gen, c_func, fancy_jump, pass_tls, save_arg_regs, return_pc) \
136 var = generate_stub(env, type_func_gen, CAST_FROM_FN_PTR(address, c_func), #var, fancy_jump, pass_tls, save_arg_regs, return_pc); \
137 if (var == NULL) { return false; }
139 bool OptoRuntime::generate(ciEnv* env) {
141 generate_exception_blob();
143 // Note: tls: Means fetching the return oop out of the thread-local storage
144 //
145 // variable/name type-function-gen , runtime method ,fncy_jp, tls,save_args,retpc
146 // -------------------------------------------------------------------------------------------------------------------------------
147 gen(env, _new_instance_Java , new_instance_Type , new_instance_C , 0 , true , false, false);
148 gen(env, _new_array_Java , new_array_Type , new_array_C , 0 , true , false, false);
149 gen(env, _new_array_nozero_Java , new_array_Type , new_array_nozero_C , 0 , true , false, false);
150 gen(env, _multianewarray2_Java , multianewarray2_Type , multianewarray2_C , 0 , true , false, false);
151 gen(env, _multianewarray3_Java , multianewarray3_Type , multianewarray3_C , 0 , true , false, false);
152 gen(env, _multianewarray4_Java , multianewarray4_Type , multianewarray4_C , 0 , true , false, false);
153 gen(env, _multianewarray5_Java , multianewarray5_Type , multianewarray5_C , 0 , true , false, false);
154 gen(env, _multianewarrayN_Java , multianewarrayN_Type , multianewarrayN_C , 0 , true , false, false);
155 gen(env, _g1_wb_pre_Java , g1_wb_pre_Type , SharedRuntime::g1_wb_pre , 0 , false, false, false);
156 gen(env, _g1_wb_post_Java , g1_wb_post_Type , SharedRuntime::g1_wb_post , 0 , false, false, false);
157 gen(env, _complete_monitor_locking_Java , complete_monitor_enter_Type , SharedRuntime::complete_monitor_locking_C, 0, false, false, false);
158 gen(env, _rethrow_Java , rethrow_Type , rethrow_C , 2 , true , false, true );
160 gen(env, _slow_arraycopy_Java , slow_arraycopy_Type , SharedRuntime::slow_arraycopy_C , 0 , false, false, false);
161 gen(env, _register_finalizer_Java , register_finalizer_Type , register_finalizer , 0 , false, false, false);
163 # ifdef ENABLE_ZAP_DEAD_LOCALS
164 gen(env, _zap_dead_Java_locals_Java , zap_dead_locals_Type , zap_dead_Java_locals_C , 0 , false, true , false );
165 gen(env, _zap_dead_native_locals_Java , zap_dead_locals_Type , zap_dead_native_locals_C , 0 , false, true , false );
166 # endif
167 return true;
168 }
170 #undef gen
173 // Helper method to do generation of RunTimeStub's
174 address OptoRuntime::generate_stub( ciEnv* env,
175 TypeFunc_generator gen, address C_function,
176 const char *name, int is_fancy_jump,
177 bool pass_tls,
178 bool save_argument_registers,
179 bool return_pc ) {
180 ResourceMark rm;
181 Compile C( env, gen, C_function, name, is_fancy_jump, pass_tls, save_argument_registers, return_pc );
182 return C.stub_entry_point();
183 }
185 const char* OptoRuntime::stub_name(address entry) {
186 #ifndef PRODUCT
187 CodeBlob* cb = CodeCache::find_blob(entry);
188 RuntimeStub* rs =(RuntimeStub *)cb;
189 assert(rs != NULL && rs->is_runtime_stub(), "not a runtime stub");
190 return rs->name();
191 #else
192 // Fast implementation for product mode (maybe it should be inlined too)
193 return "runtime stub";
194 #endif
195 }
198 //=============================================================================
199 // Opto compiler runtime routines
200 //=============================================================================
203 //=============================allocation======================================
204 // We failed the fast-path allocation. Now we need to do a scavenge or GC
205 // and try allocation again.
207 void OptoRuntime::new_store_pre_barrier(JavaThread* thread) {
208 // After any safepoint, just before going back to compiled code,
209 // we inform the GC that we will be doing initializing writes to
210 // this object in the future without emitting card-marks, so
211 // GC may take any compensating steps.
212 // NOTE: Keep this code consistent with GraphKit::store_barrier.
214 oop new_obj = thread->vm_result();
215 if (new_obj == NULL) return;
217 assert(Universe::heap()->can_elide_tlab_store_barriers(),
218 "compiler must check this first");
219 // GC may decide to give back a safer copy of new_obj.
220 new_obj = Universe::heap()->new_store_pre_barrier(thread, new_obj);
221 thread->set_vm_result(new_obj);
222 }
224 // object allocation
225 JRT_BLOCK_ENTRY(void, OptoRuntime::new_instance_C(Klass* klass, JavaThread* thread))
226 JRT_BLOCK;
227 #ifndef PRODUCT
228 SharedRuntime::_new_instance_ctr++; // new instance requires GC
229 #endif
230 assert(check_compiled_frame(thread), "incorrect caller");
232 // These checks are cheap to make and support reflective allocation.
233 int lh = klass->layout_helper();
234 if (Klass::layout_helper_needs_slow_path(lh)
235 || !InstanceKlass::cast(klass)->is_initialized()) {
236 KlassHandle kh(THREAD, klass);
237 kh->check_valid_for_instantiation(false, THREAD);
238 if (!HAS_PENDING_EXCEPTION) {
239 InstanceKlass::cast(kh())->initialize(THREAD);
240 }
241 if (!HAS_PENDING_EXCEPTION) {
242 klass = kh();
243 } else {
244 klass = NULL;
245 }
246 }
248 if (klass != NULL) {
249 // Scavenge and allocate an instance.
250 oop result = InstanceKlass::cast(klass)->allocate_instance(THREAD);
251 thread->set_vm_result(result);
253 // Pass oops back through thread local storage. Our apparent type to Java
254 // is that we return an oop, but we can block on exit from this routine and
255 // a GC can trash the oop in C's return register. The generated stub will
256 // fetch the oop from TLS after any possible GC.
257 }
259 deoptimize_caller_frame(thread, HAS_PENDING_EXCEPTION);
260 JRT_BLOCK_END;
262 if (GraphKit::use_ReduceInitialCardMarks()) {
263 // inform GC that we won't do card marks for initializing writes.
264 new_store_pre_barrier(thread);
265 }
266 JRT_END
269 // array allocation
270 JRT_BLOCK_ENTRY(void, OptoRuntime::new_array_C(Klass* array_type, int len, JavaThread *thread))
271 JRT_BLOCK;
272 #ifndef PRODUCT
273 SharedRuntime::_new_array_ctr++; // new array requires GC
274 #endif
275 assert(check_compiled_frame(thread), "incorrect caller");
277 // Scavenge and allocate an instance.
278 oop result;
280 if (array_type->oop_is_typeArray()) {
281 // The oopFactory likes to work with the element type.
282 // (We could bypass the oopFactory, since it doesn't add much value.)
283 BasicType elem_type = TypeArrayKlass::cast(array_type)->element_type();
284 result = oopFactory::new_typeArray(elem_type, len, THREAD);
285 } else {
286 // Although the oopFactory likes to work with the elem_type,
287 // the compiler prefers the array_type, since it must already have
288 // that latter value in hand for the fast path.
289 Klass* elem_type = ObjArrayKlass::cast(array_type)->element_klass();
290 result = oopFactory::new_objArray(elem_type, len, THREAD);
291 }
293 // Pass oops back through thread local storage. Our apparent type to Java
294 // is that we return an oop, but we can block on exit from this routine and
295 // a GC can trash the oop in C's return register. The generated stub will
296 // fetch the oop from TLS after any possible GC.
297 deoptimize_caller_frame(thread, HAS_PENDING_EXCEPTION);
298 thread->set_vm_result(result);
299 JRT_BLOCK_END;
301 if (GraphKit::use_ReduceInitialCardMarks()) {
302 // inform GC that we won't do card marks for initializing writes.
303 new_store_pre_barrier(thread);
304 }
305 JRT_END
307 // array allocation without zeroing
308 JRT_BLOCK_ENTRY(void, OptoRuntime::new_array_nozero_C(Klass* array_type, int len, JavaThread *thread))
309 JRT_BLOCK;
310 #ifndef PRODUCT
311 SharedRuntime::_new_array_ctr++; // new array requires GC
312 #endif
313 assert(check_compiled_frame(thread), "incorrect caller");
315 // Scavenge and allocate an instance.
316 oop result;
318 assert(array_type->oop_is_typeArray(), "should be called only for type array");
319 // The oopFactory likes to work with the element type.
320 BasicType elem_type = TypeArrayKlass::cast(array_type)->element_type();
321 result = oopFactory::new_typeArray_nozero(elem_type, len, THREAD);
323 // Pass oops back through thread local storage. Our apparent type to Java
324 // is that we return an oop, but we can block on exit from this routine and
325 // a GC can trash the oop in C's return register. The generated stub will
326 // fetch the oop from TLS after any possible GC.
327 deoptimize_caller_frame(thread, HAS_PENDING_EXCEPTION);
328 thread->set_vm_result(result);
329 JRT_BLOCK_END;
331 if (GraphKit::use_ReduceInitialCardMarks()) {
332 // inform GC that we won't do card marks for initializing writes.
333 new_store_pre_barrier(thread);
334 }
336 oop result = thread->vm_result();
337 if ((len > 0) && (result != NULL) &&
338 is_deoptimized_caller_frame(thread)) {
339 // Zero array here if the caller is deoptimized.
340 int size = ((typeArrayOop)result)->object_size();
341 BasicType elem_type = TypeArrayKlass::cast(array_type)->element_type();
342 const size_t hs = arrayOopDesc::header_size(elem_type);
343 // Align to next 8 bytes to avoid trashing arrays's length.
344 const size_t aligned_hs = align_object_offset(hs);
345 HeapWord* obj = (HeapWord*)result;
346 if (aligned_hs > hs) {
347 Copy::zero_to_words(obj+hs, aligned_hs-hs);
348 }
349 // Optimized zeroing.
350 Copy::fill_to_aligned_words(obj+aligned_hs, size-aligned_hs);
351 }
353 JRT_END
355 // Note: multianewarray for one dimension is handled inline by GraphKit::new_array.
357 // multianewarray for 2 dimensions
358 JRT_ENTRY(void, OptoRuntime::multianewarray2_C(Klass* elem_type, int len1, int len2, JavaThread *thread))
359 #ifndef PRODUCT
360 SharedRuntime::_multi2_ctr++; // multianewarray for 1 dimension
361 #endif
362 assert(check_compiled_frame(thread), "incorrect caller");
363 assert(elem_type->is_klass(), "not a class");
364 jint dims[2];
365 dims[0] = len1;
366 dims[1] = len2;
367 oop obj = ArrayKlass::cast(elem_type)->multi_allocate(2, dims, THREAD);
368 deoptimize_caller_frame(thread, HAS_PENDING_EXCEPTION);
369 thread->set_vm_result(obj);
370 JRT_END
372 // multianewarray for 3 dimensions
373 JRT_ENTRY(void, OptoRuntime::multianewarray3_C(Klass* elem_type, int len1, int len2, int len3, JavaThread *thread))
374 #ifndef PRODUCT
375 SharedRuntime::_multi3_ctr++; // multianewarray for 1 dimension
376 #endif
377 assert(check_compiled_frame(thread), "incorrect caller");
378 assert(elem_type->is_klass(), "not a class");
379 jint dims[3];
380 dims[0] = len1;
381 dims[1] = len2;
382 dims[2] = len3;
383 oop obj = ArrayKlass::cast(elem_type)->multi_allocate(3, dims, THREAD);
384 deoptimize_caller_frame(thread, HAS_PENDING_EXCEPTION);
385 thread->set_vm_result(obj);
386 JRT_END
388 // multianewarray for 4 dimensions
389 JRT_ENTRY(void, OptoRuntime::multianewarray4_C(Klass* elem_type, int len1, int len2, int len3, int len4, JavaThread *thread))
390 #ifndef PRODUCT
391 SharedRuntime::_multi4_ctr++; // multianewarray for 1 dimension
392 #endif
393 assert(check_compiled_frame(thread), "incorrect caller");
394 assert(elem_type->is_klass(), "not a class");
395 jint dims[4];
396 dims[0] = len1;
397 dims[1] = len2;
398 dims[2] = len3;
399 dims[3] = len4;
400 oop obj = ArrayKlass::cast(elem_type)->multi_allocate(4, dims, THREAD);
401 deoptimize_caller_frame(thread, HAS_PENDING_EXCEPTION);
402 thread->set_vm_result(obj);
403 JRT_END
405 // multianewarray for 5 dimensions
406 JRT_ENTRY(void, OptoRuntime::multianewarray5_C(Klass* elem_type, int len1, int len2, int len3, int len4, int len5, JavaThread *thread))
407 #ifndef PRODUCT
408 SharedRuntime::_multi5_ctr++; // multianewarray for 1 dimension
409 #endif
410 assert(check_compiled_frame(thread), "incorrect caller");
411 assert(elem_type->is_klass(), "not a class");
412 jint dims[5];
413 dims[0] = len1;
414 dims[1] = len2;
415 dims[2] = len3;
416 dims[3] = len4;
417 dims[4] = len5;
418 oop obj = ArrayKlass::cast(elem_type)->multi_allocate(5, dims, THREAD);
419 deoptimize_caller_frame(thread, HAS_PENDING_EXCEPTION);
420 thread->set_vm_result(obj);
421 JRT_END
423 JRT_ENTRY(void, OptoRuntime::multianewarrayN_C(Klass* elem_type, arrayOopDesc* dims, JavaThread *thread))
424 assert(check_compiled_frame(thread), "incorrect caller");
425 assert(elem_type->is_klass(), "not a class");
426 assert(oop(dims)->is_typeArray(), "not an array");
428 ResourceMark rm;
429 jint len = dims->length();
430 assert(len > 0, "Dimensions array should contain data");
431 jint *j_dims = typeArrayOop(dims)->int_at_addr(0);
432 jint *c_dims = NEW_RESOURCE_ARRAY(jint, len);
433 Copy::conjoint_jints_atomic(j_dims, c_dims, len);
435 oop obj = ArrayKlass::cast(elem_type)->multi_allocate(len, c_dims, THREAD);
436 deoptimize_caller_frame(thread, HAS_PENDING_EXCEPTION);
437 thread->set_vm_result(obj);
438 JRT_END
441 const TypeFunc *OptoRuntime::new_instance_Type() {
442 // create input type (domain)
443 const Type **fields = TypeTuple::fields(1);
444 fields[TypeFunc::Parms+0] = TypeInstPtr::NOTNULL; // Klass to be allocated
445 const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+1, fields);
447 // create result type (range)
448 fields = TypeTuple::fields(1);
449 fields[TypeFunc::Parms+0] = TypeRawPtr::NOTNULL; // Returned oop
451 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms+1, fields);
453 return TypeFunc::make(domain, range);
454 }
457 const TypeFunc *OptoRuntime::athrow_Type() {
458 // create input type (domain)
459 const Type **fields = TypeTuple::fields(1);
460 fields[TypeFunc::Parms+0] = TypeInstPtr::NOTNULL; // Klass to be allocated
461 const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+1, fields);
463 // create result type (range)
464 fields = TypeTuple::fields(0);
466 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms+0, fields);
468 return TypeFunc::make(domain, range);
469 }
472 const TypeFunc *OptoRuntime::new_array_Type() {
473 // create input type (domain)
474 const Type **fields = TypeTuple::fields(2);
475 fields[TypeFunc::Parms+0] = TypeInstPtr::NOTNULL; // element klass
476 fields[TypeFunc::Parms+1] = TypeInt::INT; // array size
477 const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+2, fields);
479 // create result type (range)
480 fields = TypeTuple::fields(1);
481 fields[TypeFunc::Parms+0] = TypeRawPtr::NOTNULL; // Returned oop
483 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms+1, fields);
485 return TypeFunc::make(domain, range);
486 }
488 const TypeFunc *OptoRuntime::multianewarray_Type(int ndim) {
489 // create input type (domain)
490 const int nargs = ndim + 1;
491 const Type **fields = TypeTuple::fields(nargs);
492 fields[TypeFunc::Parms+0] = TypeInstPtr::NOTNULL; // element klass
493 for( int i = 1; i < nargs; i++ )
494 fields[TypeFunc::Parms + i] = TypeInt::INT; // array size
495 const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+nargs, fields);
497 // create result type (range)
498 fields = TypeTuple::fields(1);
499 fields[TypeFunc::Parms+0] = TypeRawPtr::NOTNULL; // Returned oop
500 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms+1, fields);
502 return TypeFunc::make(domain, range);
503 }
505 const TypeFunc *OptoRuntime::multianewarray2_Type() {
506 return multianewarray_Type(2);
507 }
509 const TypeFunc *OptoRuntime::multianewarray3_Type() {
510 return multianewarray_Type(3);
511 }
513 const TypeFunc *OptoRuntime::multianewarray4_Type() {
514 return multianewarray_Type(4);
515 }
517 const TypeFunc *OptoRuntime::multianewarray5_Type() {
518 return multianewarray_Type(5);
519 }
521 const TypeFunc *OptoRuntime::multianewarrayN_Type() {
522 // create input type (domain)
523 const Type **fields = TypeTuple::fields(2);
524 fields[TypeFunc::Parms+0] = TypeInstPtr::NOTNULL; // element klass
525 fields[TypeFunc::Parms+1] = TypeInstPtr::NOTNULL; // array of dim sizes
526 const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+2, fields);
528 // create result type (range)
529 fields = TypeTuple::fields(1);
530 fields[TypeFunc::Parms+0] = TypeRawPtr::NOTNULL; // Returned oop
531 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms+1, fields);
533 return TypeFunc::make(domain, range);
534 }
536 const TypeFunc *OptoRuntime::g1_wb_pre_Type() {
537 const Type **fields = TypeTuple::fields(2);
538 fields[TypeFunc::Parms+0] = TypeInstPtr::NOTNULL; // original field value
539 fields[TypeFunc::Parms+1] = TypeRawPtr::NOTNULL; // thread
540 const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+2, fields);
542 // create result type (range)
543 fields = TypeTuple::fields(0);
544 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms+0, fields);
546 return TypeFunc::make(domain, range);
547 }
549 const TypeFunc *OptoRuntime::g1_wb_post_Type() {
551 const Type **fields = TypeTuple::fields(2);
552 fields[TypeFunc::Parms+0] = TypeRawPtr::NOTNULL; // Card addr
553 fields[TypeFunc::Parms+1] = TypeRawPtr::NOTNULL; // thread
554 const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+2, fields);
556 // create result type (range)
557 fields = TypeTuple::fields(0);
558 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms, fields);
560 return TypeFunc::make(domain, range);
561 }
563 const TypeFunc *OptoRuntime::uncommon_trap_Type() {
564 // create input type (domain)
565 const Type **fields = TypeTuple::fields(1);
566 // Symbol* name of class to be loaded
567 fields[TypeFunc::Parms+0] = TypeInt::INT;
568 const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+1, fields);
570 // create result type (range)
571 fields = TypeTuple::fields(0);
572 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms+0, fields);
574 return TypeFunc::make(domain, range);
575 }
577 # ifdef ENABLE_ZAP_DEAD_LOCALS
578 // Type used for stub generation for zap_dead_locals.
579 // No inputs or outputs
580 const TypeFunc *OptoRuntime::zap_dead_locals_Type() {
581 // create input type (domain)
582 const Type **fields = TypeTuple::fields(0);
583 const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms,fields);
585 // create result type (range)
586 fields = TypeTuple::fields(0);
587 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms,fields);
589 return TypeFunc::make(domain,range);
590 }
591 # endif
594 //-----------------------------------------------------------------------------
595 // Monitor Handling
596 const TypeFunc *OptoRuntime::complete_monitor_enter_Type() {
597 // create input type (domain)
598 const Type **fields = TypeTuple::fields(2);
599 fields[TypeFunc::Parms+0] = TypeInstPtr::NOTNULL; // Object to be Locked
600 fields[TypeFunc::Parms+1] = TypeRawPtr::BOTTOM; // Address of stack location for lock
601 const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+2,fields);
603 // create result type (range)
604 fields = TypeTuple::fields(0);
606 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms+0,fields);
608 return TypeFunc::make(domain,range);
609 }
612 //-----------------------------------------------------------------------------
613 const TypeFunc *OptoRuntime::complete_monitor_exit_Type() {
614 // create input type (domain)
615 const Type **fields = TypeTuple::fields(2);
616 fields[TypeFunc::Parms+0] = TypeInstPtr::NOTNULL; // Object to be Locked
617 fields[TypeFunc::Parms+1] = TypeRawPtr::BOTTOM; // Address of stack location for lock
618 const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+2,fields);
620 // create result type (range)
621 fields = TypeTuple::fields(0);
623 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms+0,fields);
625 return TypeFunc::make(domain,range);
626 }
628 const TypeFunc* OptoRuntime::flush_windows_Type() {
629 // create input type (domain)
630 const Type** fields = TypeTuple::fields(1);
631 fields[TypeFunc::Parms+0] = NULL; // void
632 const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms, fields);
634 // create result type
635 fields = TypeTuple::fields(1);
636 fields[TypeFunc::Parms+0] = NULL; // void
637 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms, fields);
639 return TypeFunc::make(domain, range);
640 }
642 const TypeFunc* OptoRuntime::l2f_Type() {
643 // create input type (domain)
644 const Type **fields = TypeTuple::fields(2);
645 fields[TypeFunc::Parms+0] = TypeLong::LONG;
646 fields[TypeFunc::Parms+1] = Type::HALF;
647 const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+2, fields);
649 // create result type (range)
650 fields = TypeTuple::fields(1);
651 fields[TypeFunc::Parms+0] = Type::FLOAT;
652 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms+1, fields);
654 return TypeFunc::make(domain, range);
655 }
657 const TypeFunc* OptoRuntime::modf_Type() {
658 const Type **fields = TypeTuple::fields(2);
659 fields[TypeFunc::Parms+0] = Type::FLOAT;
660 fields[TypeFunc::Parms+1] = Type::FLOAT;
661 const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+2, fields);
663 // create result type (range)
664 fields = TypeTuple::fields(1);
665 fields[TypeFunc::Parms+0] = Type::FLOAT;
667 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms+1, fields);
669 return TypeFunc::make(domain, range);
670 }
672 const TypeFunc *OptoRuntime::Math_D_D_Type() {
673 // create input type (domain)
674 const Type **fields = TypeTuple::fields(2);
675 // Symbol* name of class to be loaded
676 fields[TypeFunc::Parms+0] = Type::DOUBLE;
677 fields[TypeFunc::Parms+1] = Type::HALF;
678 const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+2, fields);
680 // create result type (range)
681 fields = TypeTuple::fields(2);
682 fields[TypeFunc::Parms+0] = Type::DOUBLE;
683 fields[TypeFunc::Parms+1] = Type::HALF;
684 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms+2, fields);
686 return TypeFunc::make(domain, range);
687 }
689 const TypeFunc* OptoRuntime::Math_DD_D_Type() {
690 const Type **fields = TypeTuple::fields(4);
691 fields[TypeFunc::Parms+0] = Type::DOUBLE;
692 fields[TypeFunc::Parms+1] = Type::HALF;
693 fields[TypeFunc::Parms+2] = Type::DOUBLE;
694 fields[TypeFunc::Parms+3] = Type::HALF;
695 const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+4, fields);
697 // create result type (range)
698 fields = TypeTuple::fields(2);
699 fields[TypeFunc::Parms+0] = Type::DOUBLE;
700 fields[TypeFunc::Parms+1] = Type::HALF;
701 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms+2, fields);
703 return TypeFunc::make(domain, range);
704 }
706 //-------------- currentTimeMillis, currentTimeNanos, etc
708 const TypeFunc* OptoRuntime::void_long_Type() {
709 // create input type (domain)
710 const Type **fields = TypeTuple::fields(0);
711 const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+0, fields);
713 // create result type (range)
714 fields = TypeTuple::fields(2);
715 fields[TypeFunc::Parms+0] = TypeLong::LONG;
716 fields[TypeFunc::Parms+1] = Type::HALF;
717 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms+2, fields);
719 return TypeFunc::make(domain, range);
720 }
722 // arraycopy stub variations:
723 enum ArrayCopyType {
724 ac_fast, // void(ptr, ptr, size_t)
725 ac_checkcast, // int(ptr, ptr, size_t, size_t, ptr)
726 ac_slow, // void(ptr, int, ptr, int, int)
727 ac_generic // int(ptr, int, ptr, int, int)
728 };
730 static const TypeFunc* make_arraycopy_Type(ArrayCopyType act) {
731 // create input type (domain)
732 int num_args = (act == ac_fast ? 3 : 5);
733 int num_size_args = (act == ac_fast ? 1 : act == ac_checkcast ? 2 : 0);
734 int argcnt = num_args;
735 LP64_ONLY(argcnt += num_size_args); // halfwords for lengths
736 const Type** fields = TypeTuple::fields(argcnt);
737 int argp = TypeFunc::Parms;
738 fields[argp++] = TypePtr::NOTNULL; // src
739 if (num_size_args == 0) {
740 fields[argp++] = TypeInt::INT; // src_pos
741 }
742 fields[argp++] = TypePtr::NOTNULL; // dest
743 if (num_size_args == 0) {
744 fields[argp++] = TypeInt::INT; // dest_pos
745 fields[argp++] = TypeInt::INT; // length
746 }
747 while (num_size_args-- > 0) {
748 fields[argp++] = TypeX_X; // size in whatevers (size_t)
749 LP64_ONLY(fields[argp++] = Type::HALF); // other half of long length
750 }
751 if (act == ac_checkcast) {
752 fields[argp++] = TypePtr::NOTNULL; // super_klass
753 }
754 assert(argp == TypeFunc::Parms+argcnt, "correct decoding of act");
755 const TypeTuple* domain = TypeTuple::make(TypeFunc::Parms+argcnt, fields);
757 // create result type if needed
758 int retcnt = (act == ac_checkcast || act == ac_generic ? 1 : 0);
759 fields = TypeTuple::fields(1);
760 if (retcnt == 0)
761 fields[TypeFunc::Parms+0] = NULL; // void
762 else
763 fields[TypeFunc::Parms+0] = TypeInt::INT; // status result, if needed
764 const TypeTuple* range = TypeTuple::make(TypeFunc::Parms+retcnt, fields);
765 return TypeFunc::make(domain, range);
766 }
768 const TypeFunc* OptoRuntime::fast_arraycopy_Type() {
769 // This signature is simple: Two base pointers and a size_t.
770 return make_arraycopy_Type(ac_fast);
771 }
773 const TypeFunc* OptoRuntime::checkcast_arraycopy_Type() {
774 // An extension of fast_arraycopy_Type which adds type checking.
775 return make_arraycopy_Type(ac_checkcast);
776 }
778 const TypeFunc* OptoRuntime::slow_arraycopy_Type() {
779 // This signature is exactly the same as System.arraycopy.
780 // There are no intptr_t (int/long) arguments.
781 return make_arraycopy_Type(ac_slow);
782 }
784 const TypeFunc* OptoRuntime::generic_arraycopy_Type() {
785 // This signature is like System.arraycopy, except that it returns status.
786 return make_arraycopy_Type(ac_generic);
787 }
790 const TypeFunc* OptoRuntime::array_fill_Type() {
791 const Type** fields;
792 int argp = TypeFunc::Parms;
793 if (CCallingConventionRequiresIntsAsLongs) {
794 // create input type (domain): pointer, int, size_t
795 fields = TypeTuple::fields(3 LP64_ONLY( + 2));
796 fields[argp++] = TypePtr::NOTNULL;
797 fields[argp++] = TypeLong::LONG;
798 fields[argp++] = Type::HALF;
799 } else {
800 // create input type (domain): pointer, int, size_t
801 fields = TypeTuple::fields(3 LP64_ONLY( + 1));
802 fields[argp++] = TypePtr::NOTNULL;
803 fields[argp++] = TypeInt::INT;
804 }
805 fields[argp++] = TypeX_X; // size in whatevers (size_t)
806 LP64_ONLY(fields[argp++] = Type::HALF); // other half of long length
807 const TypeTuple *domain = TypeTuple::make(argp, fields);
809 // create result type
810 fields = TypeTuple::fields(1);
811 fields[TypeFunc::Parms+0] = NULL; // void
812 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms, fields);
814 return TypeFunc::make(domain, range);
815 }
817 // for aescrypt encrypt/decrypt operations, just three pointers returning void (length is constant)
818 const TypeFunc* OptoRuntime::aescrypt_block_Type() {
819 // create input type (domain)
820 int num_args = 3;
821 if (Matcher::pass_original_key_for_aes()) {
822 num_args = 4;
823 }
824 int argcnt = num_args;
825 const Type** fields = TypeTuple::fields(argcnt);
826 int argp = TypeFunc::Parms;
827 fields[argp++] = TypePtr::NOTNULL; // src
828 fields[argp++] = TypePtr::NOTNULL; // dest
829 fields[argp++] = TypePtr::NOTNULL; // k array
830 if (Matcher::pass_original_key_for_aes()) {
831 fields[argp++] = TypePtr::NOTNULL; // original k array
832 }
833 assert(argp == TypeFunc::Parms+argcnt, "correct decoding");
834 const TypeTuple* domain = TypeTuple::make(TypeFunc::Parms+argcnt, fields);
836 // no result type needed
837 fields = TypeTuple::fields(1);
838 fields[TypeFunc::Parms+0] = NULL; // void
839 const TypeTuple* range = TypeTuple::make(TypeFunc::Parms, fields);
840 return TypeFunc::make(domain, range);
841 }
843 /**
844 * int updateBytesCRC32(int crc, byte* b, int len)
845 */
846 const TypeFunc* OptoRuntime::updateBytesCRC32_Type() {
847 // create input type (domain)
848 int num_args = 3;
849 int argcnt = num_args;
850 const Type** fields = TypeTuple::fields(argcnt);
851 int argp = TypeFunc::Parms;
852 fields[argp++] = TypeInt::INT; // crc
853 fields[argp++] = TypePtr::NOTNULL; // src
854 fields[argp++] = TypeInt::INT; // len
855 assert(argp == TypeFunc::Parms+argcnt, "correct decoding");
856 const TypeTuple* domain = TypeTuple::make(TypeFunc::Parms+argcnt, fields);
858 // result type needed
859 fields = TypeTuple::fields(1);
860 fields[TypeFunc::Parms+0] = TypeInt::INT; // crc result
861 const TypeTuple* range = TypeTuple::make(TypeFunc::Parms+1, fields);
862 return TypeFunc::make(domain, range);
863 }
865 // for cipherBlockChaining calls of aescrypt encrypt/decrypt, four pointers and a length, returning int
866 const TypeFunc* OptoRuntime::cipherBlockChaining_aescrypt_Type() {
867 // create input type (domain)
868 int num_args = 5;
869 if (Matcher::pass_original_key_for_aes()) {
870 num_args = 6;
871 }
872 int argcnt = num_args;
873 const Type** fields = TypeTuple::fields(argcnt);
874 int argp = TypeFunc::Parms;
875 fields[argp++] = TypePtr::NOTNULL; // src
876 fields[argp++] = TypePtr::NOTNULL; // dest
877 fields[argp++] = TypePtr::NOTNULL; // k array
878 fields[argp++] = TypePtr::NOTNULL; // r array
879 fields[argp++] = TypeInt::INT; // src len
880 if (Matcher::pass_original_key_for_aes()) {
881 fields[argp++] = TypePtr::NOTNULL; // original k array
882 }
883 assert(argp == TypeFunc::Parms+argcnt, "correct decoding");
884 const TypeTuple* domain = TypeTuple::make(TypeFunc::Parms+argcnt, fields);
886 // returning cipher len (int)
887 fields = TypeTuple::fields(1);
888 fields[TypeFunc::Parms+0] = TypeInt::INT;
889 const TypeTuple* range = TypeTuple::make(TypeFunc::Parms+1, fields);
890 return TypeFunc::make(domain, range);
891 }
893 /*
894 * void implCompress(byte[] buf, int ofs)
895 */
896 const TypeFunc* OptoRuntime::sha_implCompress_Type() {
897 // create input type (domain)
898 int num_args = 2;
899 int argcnt = num_args;
900 const Type** fields = TypeTuple::fields(argcnt);
901 int argp = TypeFunc::Parms;
902 fields[argp++] = TypePtr::NOTNULL; // buf
903 fields[argp++] = TypePtr::NOTNULL; // state
904 assert(argp == TypeFunc::Parms+argcnt, "correct decoding");
905 const TypeTuple* domain = TypeTuple::make(TypeFunc::Parms+argcnt, fields);
907 // no result type needed
908 fields = TypeTuple::fields(1);
909 fields[TypeFunc::Parms+0] = NULL; // void
910 const TypeTuple* range = TypeTuple::make(TypeFunc::Parms, fields);
911 return TypeFunc::make(domain, range);
912 }
914 /*
915 * int implCompressMultiBlock(byte[] b, int ofs, int limit)
916 */
917 const TypeFunc* OptoRuntime::digestBase_implCompressMB_Type() {
918 // create input type (domain)
919 int num_args = 4;
920 int argcnt = num_args;
921 const Type** fields = TypeTuple::fields(argcnt);
922 int argp = TypeFunc::Parms;
923 fields[argp++] = TypePtr::NOTNULL; // buf
924 fields[argp++] = TypePtr::NOTNULL; // state
925 fields[argp++] = TypeInt::INT; // ofs
926 fields[argp++] = TypeInt::INT; // limit
927 assert(argp == TypeFunc::Parms+argcnt, "correct decoding");
928 const TypeTuple* domain = TypeTuple::make(TypeFunc::Parms+argcnt, fields);
930 // returning ofs (int)
931 fields = TypeTuple::fields(1);
932 fields[TypeFunc::Parms+0] = TypeInt::INT; // ofs
933 const TypeTuple* range = TypeTuple::make(TypeFunc::Parms+1, fields);
934 return TypeFunc::make(domain, range);
935 }
937 const TypeFunc* OptoRuntime::multiplyToLen_Type() {
938 // create input type (domain)
939 int num_args = 6;
940 int argcnt = num_args;
941 const Type** fields = TypeTuple::fields(argcnt);
942 int argp = TypeFunc::Parms;
943 fields[argp++] = TypePtr::NOTNULL; // x
944 fields[argp++] = TypeInt::INT; // xlen
945 fields[argp++] = TypePtr::NOTNULL; // y
946 fields[argp++] = TypeInt::INT; // ylen
947 fields[argp++] = TypePtr::NOTNULL; // z
948 fields[argp++] = TypeInt::INT; // zlen
949 assert(argp == TypeFunc::Parms+argcnt, "correct decoding");
950 const TypeTuple* domain = TypeTuple::make(TypeFunc::Parms+argcnt, fields);
952 // no result type needed
953 fields = TypeTuple::fields(1);
954 fields[TypeFunc::Parms+0] = NULL;
955 const TypeTuple* range = TypeTuple::make(TypeFunc::Parms, fields);
956 return TypeFunc::make(domain, range);
957 }
959 const TypeFunc* OptoRuntime::squareToLen_Type() {
960 // create input type (domain)
961 int num_args = 4;
962 int argcnt = num_args;
963 const Type** fields = TypeTuple::fields(argcnt);
964 int argp = TypeFunc::Parms;
965 fields[argp++] = TypePtr::NOTNULL; // x
966 fields[argp++] = TypeInt::INT; // len
967 fields[argp++] = TypePtr::NOTNULL; // z
968 fields[argp++] = TypeInt::INT; // zlen
969 assert(argp == TypeFunc::Parms+argcnt, "correct decoding");
970 const TypeTuple* domain = TypeTuple::make(TypeFunc::Parms+argcnt, fields);
972 // no result type needed
973 fields = TypeTuple::fields(1);
974 fields[TypeFunc::Parms+0] = NULL;
975 const TypeTuple* range = TypeTuple::make(TypeFunc::Parms, fields);
976 return TypeFunc::make(domain, range);
977 }
979 // for mulAdd calls, 2 pointers and 3 ints, returning int
980 const TypeFunc* OptoRuntime::mulAdd_Type() {
981 // create input type (domain)
982 int num_args = 5;
983 int argcnt = num_args;
984 const Type** fields = TypeTuple::fields(argcnt);
985 int argp = TypeFunc::Parms;
986 fields[argp++] = TypePtr::NOTNULL; // out
987 fields[argp++] = TypePtr::NOTNULL; // in
988 fields[argp++] = TypeInt::INT; // offset
989 fields[argp++] = TypeInt::INT; // len
990 fields[argp++] = TypeInt::INT; // k
991 assert(argp == TypeFunc::Parms+argcnt, "correct decoding");
992 const TypeTuple* domain = TypeTuple::make(TypeFunc::Parms+argcnt, fields);
994 // returning carry (int)
995 fields = TypeTuple::fields(1);
996 fields[TypeFunc::Parms+0] = TypeInt::INT;
997 const TypeTuple* range = TypeTuple::make(TypeFunc::Parms+1, fields);
998 return TypeFunc::make(domain, range);
999 }
1003 //------------- Interpreter state access for on stack replacement
1004 const TypeFunc* OptoRuntime::osr_end_Type() {
1005 // create input type (domain)
1006 const Type **fields = TypeTuple::fields(1);
1007 fields[TypeFunc::Parms+0] = TypeRawPtr::BOTTOM; // OSR temp buf
1008 const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+1, fields);
1010 // create result type
1011 fields = TypeTuple::fields(1);
1012 // fields[TypeFunc::Parms+0] = TypeInstPtr::NOTNULL; // locked oop
1013 fields[TypeFunc::Parms+0] = NULL; // void
1014 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms, fields);
1015 return TypeFunc::make(domain, range);
1016 }
1018 //-------------- methodData update helpers
1020 const TypeFunc* OptoRuntime::profile_receiver_type_Type() {
1021 // create input type (domain)
1022 const Type **fields = TypeTuple::fields(2);
1023 fields[TypeFunc::Parms+0] = TypeAryPtr::NOTNULL; // methodData pointer
1024 fields[TypeFunc::Parms+1] = TypeInstPtr::BOTTOM; // receiver oop
1025 const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+2, fields);
1027 // create result type
1028 fields = TypeTuple::fields(1);
1029 fields[TypeFunc::Parms+0] = NULL; // void
1030 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms, fields);
1031 return TypeFunc::make(domain,range);
1032 }
1034 JRT_LEAF(void, OptoRuntime::profile_receiver_type_C(DataLayout* data, oopDesc* receiver))
1035 if (receiver == NULL) return;
1036 Klass* receiver_klass = receiver->klass();
1038 intptr_t* mdp = ((intptr_t*)(data)) + DataLayout::header_size_in_cells();
1039 int empty_row = -1; // free row, if any is encountered
1041 // ReceiverTypeData* vc = new ReceiverTypeData(mdp);
1042 for (uint row = 0; row < ReceiverTypeData::row_limit(); row++) {
1043 // if (vc->receiver(row) == receiver_klass)
1044 int receiver_off = ReceiverTypeData::receiver_cell_index(row);
1045 intptr_t row_recv = *(mdp + receiver_off);
1046 if (row_recv == (intptr_t) receiver_klass) {
1047 // vc->set_receiver_count(row, vc->receiver_count(row) + DataLayout::counter_increment);
1048 int count_off = ReceiverTypeData::receiver_count_cell_index(row);
1049 *(mdp + count_off) += DataLayout::counter_increment;
1050 return;
1051 } else if (row_recv == 0) {
1052 // else if (vc->receiver(row) == NULL)
1053 empty_row = (int) row;
1054 }
1055 }
1057 if (empty_row != -1) {
1058 int receiver_off = ReceiverTypeData::receiver_cell_index(empty_row);
1059 // vc->set_receiver(empty_row, receiver_klass);
1060 *(mdp + receiver_off) = (intptr_t) receiver_klass;
1061 // vc->set_receiver_count(empty_row, DataLayout::counter_increment);
1062 int count_off = ReceiverTypeData::receiver_count_cell_index(empty_row);
1063 *(mdp + count_off) = DataLayout::counter_increment;
1064 } else {
1065 // Receiver did not match any saved receiver and there is no empty row for it.
1066 // Increment total counter to indicate polymorphic case.
1067 intptr_t* count_p = (intptr_t*)(((byte*)(data)) + in_bytes(CounterData::count_offset()));
1068 *count_p += DataLayout::counter_increment;
1069 }
1070 JRT_END
1072 //-------------------------------------------------------------------------------------
1073 // register policy
1075 bool OptoRuntime::is_callee_saved_register(MachRegisterNumbers reg) {
1076 assert(reg >= 0 && reg < _last_Mach_Reg, "must be a machine register");
1077 switch (register_save_policy[reg]) {
1078 case 'C': return false; //SOC
1079 case 'E': return true ; //SOE
1080 case 'N': return false; //NS
1081 case 'A': return false; //AS
1082 }
1083 ShouldNotReachHere();
1084 return false;
1085 }
1087 //-----------------------------------------------------------------------
1088 // Exceptions
1089 //
1091 static void trace_exception(oop exception_oop, address exception_pc, const char* msg) PRODUCT_RETURN;
1093 // The method is an entry that is always called by a C++ method not
1094 // directly from compiled code. Compiled code will call the C++ method following.
1095 // We can't allow async exception to be installed during exception processing.
1096 JRT_ENTRY_NO_ASYNC(address, OptoRuntime::handle_exception_C_helper(JavaThread* thread, nmethod* &nm))
1098 // Do not confuse exception_oop with pending_exception. The exception_oop
1099 // is only used to pass arguments into the method. Not for general
1100 // exception handling. DO NOT CHANGE IT to use pending_exception, since
1101 // the runtime stubs checks this on exit.
1102 assert(thread->exception_oop() != NULL, "exception oop is found");
1103 address handler_address = NULL;
1105 Handle exception(thread, thread->exception_oop());
1106 address pc = thread->exception_pc();
1108 // Clear out the exception oop and pc since looking up an
1109 // exception handler can cause class loading, which might throw an
1110 // exception and those fields are expected to be clear during
1111 // normal bytecode execution.
1112 thread->clear_exception_oop_and_pc();
1114 if (TraceExceptions) {
1115 trace_exception(exception(), pc, "");
1116 }
1118 // for AbortVMOnException flag
1119 NOT_PRODUCT(Exceptions::debug_check_abort(exception));
1121 #ifdef ASSERT
1122 if (!(exception->is_a(SystemDictionary::Throwable_klass()))) {
1123 // should throw an exception here
1124 ShouldNotReachHere();
1125 }
1126 #endif
1128 // new exception handling: this method is entered only from adapters
1129 // exceptions from compiled java methods are handled in compiled code
1130 // using rethrow node
1132 nm = CodeCache::find_nmethod(pc);
1133 assert(nm != NULL, "No NMethod found");
1134 if (nm->is_native_method()) {
1135 fatal("Native method should not have path to exception handling");
1136 } else {
1137 // we are switching to old paradigm: search for exception handler in caller_frame
1138 // instead in exception handler of caller_frame.sender()
1140 if (JvmtiExport::can_post_on_exceptions()) {
1141 // "Full-speed catching" is not necessary here,
1142 // since we're notifying the VM on every catch.
1143 // Force deoptimization and the rest of the lookup
1144 // will be fine.
1145 deoptimize_caller_frame(thread);
1146 }
1148 // Check the stack guard pages. If enabled, look for handler in this frame;
1149 // otherwise, forcibly unwind the frame.
1150 //
1151 // 4826555: use default current sp for reguard_stack instead of &nm: it's more accurate.
1152 bool force_unwind = !thread->reguard_stack();
1153 bool deopting = false;
1154 if (nm->is_deopt_pc(pc)) {
1155 deopting = true;
1156 RegisterMap map(thread, false);
1157 frame deoptee = thread->last_frame().sender(&map);
1158 assert(deoptee.is_deoptimized_frame(), "must be deopted");
1159 // Adjust the pc back to the original throwing pc
1160 pc = deoptee.pc();
1161 }
1163 // If we are forcing an unwind because of stack overflow then deopt is
1164 // irrelevant since we are throwing the frame away anyway.
1166 if (deopting && !force_unwind) {
1167 handler_address = SharedRuntime::deopt_blob()->unpack_with_exception();
1168 } else {
1170 handler_address =
1171 force_unwind ? NULL : nm->handler_for_exception_and_pc(exception, pc);
1173 if (handler_address == NULL) {
1174 Handle original_exception(thread, exception());
1175 handler_address = SharedRuntime::compute_compiled_exc_handler(nm, pc, exception, force_unwind, true);
1176 assert (handler_address != NULL, "must have compiled handler");
1177 // Update the exception cache only when the unwind was not forced
1178 // and there didn't happen another exception during the computation of the
1179 // compiled exception handler.
1180 if (!force_unwind && original_exception() == exception()) {
1181 nm->add_handler_for_exception_and_pc(exception,pc,handler_address);
1182 }
1183 } else {
1184 assert(handler_address == SharedRuntime::compute_compiled_exc_handler(nm, pc, exception, force_unwind, true), "Must be the same");
1185 }
1186 }
1188 thread->set_exception_pc(pc);
1189 thread->set_exception_handler_pc(handler_address);
1191 // Check if the exception PC is a MethodHandle call site.
1192 thread->set_is_method_handle_return(nm->is_method_handle_return(pc));
1193 }
1195 // Restore correct return pc. Was saved above.
1196 thread->set_exception_oop(exception());
1197 return handler_address;
1199 JRT_END
1201 // We are entering here from exception_blob
1202 // If there is a compiled exception handler in this method, we will continue there;
1203 // otherwise we will unwind the stack and continue at the caller of top frame method
1204 // Note we enter without the usual JRT wrapper. We will call a helper routine that
1205 // will do the normal VM entry. We do it this way so that we can see if the nmethod
1206 // we looked up the handler for has been deoptimized in the meantime. If it has been
1207 // we must not use the handler and instead return the deopt blob.
1208 address OptoRuntime::handle_exception_C(JavaThread* thread) {
1209 //
1210 // We are in Java not VM and in debug mode we have a NoHandleMark
1211 //
1212 #ifndef PRODUCT
1213 SharedRuntime::_find_handler_ctr++; // find exception handler
1214 #endif
1215 debug_only(NoHandleMark __hm;)
1216 nmethod* nm = NULL;
1217 address handler_address = NULL;
1218 {
1219 // Enter the VM
1221 ResetNoHandleMark rnhm;
1222 handler_address = handle_exception_C_helper(thread, nm);
1223 }
1225 // Back in java: Use no oops, DON'T safepoint
1227 // Now check to see if the handler we are returning is in a now
1228 // deoptimized frame
1230 if (nm != NULL) {
1231 RegisterMap map(thread, false);
1232 frame caller = thread->last_frame().sender(&map);
1233 #ifdef ASSERT
1234 assert(caller.is_compiled_frame(), "must be");
1235 #endif // ASSERT
1236 if (caller.is_deoptimized_frame()) {
1237 handler_address = SharedRuntime::deopt_blob()->unpack_with_exception();
1238 }
1239 }
1240 return handler_address;
1241 }
1243 //------------------------------rethrow----------------------------------------
1244 // We get here after compiled code has executed a 'RethrowNode'. The callee
1245 // is either throwing or rethrowing an exception. The callee-save registers
1246 // have been restored, synchronized objects have been unlocked and the callee
1247 // stack frame has been removed. The return address was passed in.
1248 // Exception oop is passed as the 1st argument. This routine is then called
1249 // from the stub. On exit, we know where to jump in the caller's code.
1250 // After this C code exits, the stub will pop his frame and end in a jump
1251 // (instead of a return). We enter the caller's default handler.
1252 //
1253 // This must be JRT_LEAF:
1254 // - caller will not change its state as we cannot block on exit,
1255 // therefore raw_exception_handler_for_return_address is all it takes
1256 // to handle deoptimized blobs
1257 //
1258 // However, there needs to be a safepoint check in the middle! So compiled
1259 // safepoints are completely watertight.
1260 //
1261 // Thus, it cannot be a leaf since it contains the No_GC_Verifier.
1262 //
1263 // *THIS IS NOT RECOMMENDED PROGRAMMING STYLE*
1264 //
1265 address OptoRuntime::rethrow_C(oopDesc* exception, JavaThread* thread, address ret_pc) {
1266 #ifndef PRODUCT
1267 SharedRuntime::_rethrow_ctr++; // count rethrows
1268 #endif
1269 assert (exception != NULL, "should have thrown a NULLPointerException");
1270 #ifdef ASSERT
1271 if (!(exception->is_a(SystemDictionary::Throwable_klass()))) {
1272 // should throw an exception here
1273 ShouldNotReachHere();
1274 }
1275 #endif
1277 thread->set_vm_result(exception);
1278 // Frame not compiled (handles deoptimization blob)
1279 return SharedRuntime::raw_exception_handler_for_return_address(thread, ret_pc);
1280 }
1283 const TypeFunc *OptoRuntime::rethrow_Type() {
1284 // create input type (domain)
1285 const Type **fields = TypeTuple::fields(1);
1286 fields[TypeFunc::Parms+0] = TypeInstPtr::NOTNULL; // Exception oop
1287 const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+1,fields);
1289 // create result type (range)
1290 fields = TypeTuple::fields(1);
1291 fields[TypeFunc::Parms+0] = TypeInstPtr::NOTNULL; // Exception oop
1292 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms+1, fields);
1294 return TypeFunc::make(domain, range);
1295 }
1298 void OptoRuntime::deoptimize_caller_frame(JavaThread *thread, bool doit) {
1299 // Deoptimize the caller before continuing, as the compiled
1300 // exception handler table may not be valid.
1301 if (!StressCompiledExceptionHandlers && doit) {
1302 deoptimize_caller_frame(thread);
1303 }
1304 }
1306 void OptoRuntime::deoptimize_caller_frame(JavaThread *thread) {
1307 // Called from within the owner thread, so no need for safepoint
1308 RegisterMap reg_map(thread);
1309 frame stub_frame = thread->last_frame();
1310 assert(stub_frame.is_runtime_frame() || exception_blob()->contains(stub_frame.pc()), "sanity check");
1311 frame caller_frame = stub_frame.sender(®_map);
1313 // Deoptimize the caller frame.
1314 Deoptimization::deoptimize_frame(thread, caller_frame.id());
1315 }
1318 bool OptoRuntime::is_deoptimized_caller_frame(JavaThread *thread) {
1319 // Called from within the owner thread, so no need for safepoint
1320 RegisterMap reg_map(thread);
1321 frame stub_frame = thread->last_frame();
1322 assert(stub_frame.is_runtime_frame() || exception_blob()->contains(stub_frame.pc()), "sanity check");
1323 frame caller_frame = stub_frame.sender(®_map);
1324 return caller_frame.is_deoptimized_frame();
1325 }
1328 const TypeFunc *OptoRuntime::register_finalizer_Type() {
1329 // create input type (domain)
1330 const Type **fields = TypeTuple::fields(1);
1331 fields[TypeFunc::Parms+0] = TypeInstPtr::NOTNULL; // oop; Receiver
1332 // // The JavaThread* is passed to each routine as the last argument
1333 // fields[TypeFunc::Parms+1] = TypeRawPtr::NOTNULL; // JavaThread *; Executing thread
1334 const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+1,fields);
1336 // create result type (range)
1337 fields = TypeTuple::fields(0);
1339 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms+0,fields);
1341 return TypeFunc::make(domain,range);
1342 }
1345 //-----------------------------------------------------------------------------
1346 // Dtrace support. entry and exit probes have the same signature
1347 const TypeFunc *OptoRuntime::dtrace_method_entry_exit_Type() {
1348 // create input type (domain)
1349 const Type **fields = TypeTuple::fields(2);
1350 fields[TypeFunc::Parms+0] = TypeRawPtr::BOTTOM; // Thread-local storage
1351 fields[TypeFunc::Parms+1] = TypeMetadataPtr::BOTTOM; // Method*; Method we are entering
1352 const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+2,fields);
1354 // create result type (range)
1355 fields = TypeTuple::fields(0);
1357 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms+0,fields);
1359 return TypeFunc::make(domain,range);
1360 }
1362 const TypeFunc *OptoRuntime::dtrace_object_alloc_Type() {
1363 // create input type (domain)
1364 const Type **fields = TypeTuple::fields(2);
1365 fields[TypeFunc::Parms+0] = TypeRawPtr::BOTTOM; // Thread-local storage
1366 fields[TypeFunc::Parms+1] = TypeInstPtr::NOTNULL; // oop; newly allocated object
1368 const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+2,fields);
1370 // create result type (range)
1371 fields = TypeTuple::fields(0);
1373 const TypeTuple *range = TypeTuple::make(TypeFunc::Parms+0,fields);
1375 return TypeFunc::make(domain,range);
1376 }
1379 JRT_ENTRY_NO_ASYNC(void, OptoRuntime::register_finalizer(oopDesc* obj, JavaThread* thread))
1380 assert(obj->is_oop(), "must be a valid oop");
1381 assert(obj->klass()->has_finalizer(), "shouldn't be here otherwise");
1382 InstanceKlass::register_finalizer(instanceOop(obj), CHECK);
1383 JRT_END
1385 //-----------------------------------------------------------------------------
1387 NamedCounter * volatile OptoRuntime::_named_counters = NULL;
1389 //
1390 // dump the collected NamedCounters.
1391 //
1392 void OptoRuntime::print_named_counters() {
1393 int total_lock_count = 0;
1394 int eliminated_lock_count = 0;
1396 NamedCounter* c = _named_counters;
1397 while (c) {
1398 if (c->tag() == NamedCounter::LockCounter || c->tag() == NamedCounter::EliminatedLockCounter) {
1399 int count = c->count();
1400 if (count > 0) {
1401 bool eliminated = c->tag() == NamedCounter::EliminatedLockCounter;
1402 if (Verbose) {
1403 tty->print_cr("%d %s%s", count, c->name(), eliminated ? " (eliminated)" : "");
1404 }
1405 total_lock_count += count;
1406 if (eliminated) {
1407 eliminated_lock_count += count;
1408 }
1409 }
1410 } else if (c->tag() == NamedCounter::BiasedLockingCounter) {
1411 BiasedLockingCounters* blc = ((BiasedLockingNamedCounter*)c)->counters();
1412 if (blc->nonzero()) {
1413 tty->print_cr("%s", c->name());
1414 blc->print_on(tty);
1415 }
1416 #if INCLUDE_RTM_OPT
1417 } else if (c->tag() == NamedCounter::RTMLockingCounter) {
1418 RTMLockingCounters* rlc = ((RTMLockingNamedCounter*)c)->counters();
1419 if (rlc->nonzero()) {
1420 tty->print_cr("%s", c->name());
1421 rlc->print_on(tty);
1422 }
1423 #endif
1424 }
1425 c = c->next();
1426 }
1427 if (total_lock_count > 0) {
1428 tty->print_cr("dynamic locks: %d", total_lock_count);
1429 if (eliminated_lock_count) {
1430 tty->print_cr("eliminated locks: %d (%d%%)", eliminated_lock_count,
1431 (int)(eliminated_lock_count * 100.0 / total_lock_count));
1432 }
1433 }
1434 }
1436 //
1437 // Allocate a new NamedCounter. The JVMState is used to generate the
1438 // name which consists of method@line for the inlining tree.
1439 //
1441 NamedCounter* OptoRuntime::new_named_counter(JVMState* youngest_jvms, NamedCounter::CounterTag tag) {
1442 int max_depth = youngest_jvms->depth();
1444 // Visit scopes from youngest to oldest.
1445 bool first = true;
1446 stringStream st;
1447 for (int depth = max_depth; depth >= 1; depth--) {
1448 JVMState* jvms = youngest_jvms->of_depth(depth);
1449 ciMethod* m = jvms->has_method() ? jvms->method() : NULL;
1450 if (!first) {
1451 st.print(" ");
1452 } else {
1453 first = false;
1454 }
1455 int bci = jvms->bci();
1456 if (bci < 0) bci = 0;
1457 st.print("%s.%s@%d", m->holder()->name()->as_utf8(), m->name()->as_utf8(), bci);
1458 // To print linenumbers instead of bci use: m->line_number_from_bci(bci)
1459 }
1460 NamedCounter* c;
1461 if (tag == NamedCounter::BiasedLockingCounter) {
1462 c = new BiasedLockingNamedCounter(strdup(st.as_string()));
1463 } else if (tag == NamedCounter::RTMLockingCounter) {
1464 c = new RTMLockingNamedCounter(strdup(st.as_string()));
1465 } else {
1466 c = new NamedCounter(strdup(st.as_string()), tag);
1467 }
1469 // atomically add the new counter to the head of the list. We only
1470 // add counters so this is safe.
1471 NamedCounter* head;
1472 do {
1473 c->set_next(NULL);
1474 head = _named_counters;
1475 c->set_next(head);
1476 } while (Atomic::cmpxchg_ptr(c, &_named_counters, head) != head);
1477 return c;
1478 }
1480 //-----------------------------------------------------------------------------
1481 // Non-product code
1482 #ifndef PRODUCT
1484 int trace_exception_counter = 0;
1485 static void trace_exception(oop exception_oop, address exception_pc, const char* msg) {
1486 ttyLocker ttyl;
1487 trace_exception_counter++;
1488 tty->print("%d [Exception (%s): ", trace_exception_counter, msg);
1489 exception_oop->print_value();
1490 tty->print(" in ");
1491 CodeBlob* blob = CodeCache::find_blob(exception_pc);
1492 if (blob->is_nmethod()) {
1493 nmethod* nm = blob->as_nmethod_or_null();
1494 nm->method()->print_value();
1495 } else if (blob->is_runtime_stub()) {
1496 tty->print("<runtime-stub>");
1497 } else {
1498 tty->print("<unknown>");
1499 }
1500 tty->print(" at " INTPTR_FORMAT, p2i(exception_pc));
1501 tty->print_cr("]");
1502 }
1504 #endif // PRODUCT
1507 # ifdef ENABLE_ZAP_DEAD_LOCALS
1508 // Called from call sites in compiled code with oop maps (actually safepoints)
1509 // Zaps dead locals in first java frame.
1510 // Is entry because may need to lock to generate oop maps
1511 // Currently, only used for compiler frames, but someday may be used
1512 // for interpreter frames, too.
1514 int OptoRuntime::ZapDeadCompiledLocals_count = 0;
1516 // avoid pointers to member funcs with these helpers
1517 static bool is_java_frame( frame* f) { return f->is_java_frame(); }
1518 static bool is_native_frame(frame* f) { return f->is_native_frame(); }
1521 void OptoRuntime::zap_dead_java_or_native_locals(JavaThread* thread,
1522 bool (*is_this_the_right_frame_to_zap)(frame*)) {
1523 assert(JavaThread::current() == thread, "is this needed?");
1525 if ( !ZapDeadCompiledLocals ) return;
1527 bool skip = false;
1529 if ( ZapDeadCompiledLocalsFirst == 0 ) ; // nothing special
1530 else if ( ZapDeadCompiledLocalsFirst > ZapDeadCompiledLocals_count ) skip = true;
1531 else if ( ZapDeadCompiledLocalsFirst == ZapDeadCompiledLocals_count )
1532 warning("starting zapping after skipping");
1534 if ( ZapDeadCompiledLocalsLast == -1 ) ; // nothing special
1535 else if ( ZapDeadCompiledLocalsLast < ZapDeadCompiledLocals_count ) skip = true;
1536 else if ( ZapDeadCompiledLocalsLast == ZapDeadCompiledLocals_count )
1537 warning("about to zap last zap");
1539 ++ZapDeadCompiledLocals_count; // counts skipped zaps, too
1541 if ( skip ) return;
1543 // find java frame and zap it
1545 for (StackFrameStream sfs(thread); !sfs.is_done(); sfs.next()) {
1546 if (is_this_the_right_frame_to_zap(sfs.current()) ) {
1547 sfs.current()->zap_dead_locals(thread, sfs.register_map());
1548 return;
1549 }
1550 }
1551 warning("no frame found to zap in zap_dead_Java_locals_C");
1552 }
1554 JRT_LEAF(void, OptoRuntime::zap_dead_Java_locals_C(JavaThread* thread))
1555 zap_dead_java_or_native_locals(thread, is_java_frame);
1556 JRT_END
1558 // The following does not work because for one thing, the
1559 // thread state is wrong; it expects java, but it is native.
1560 // Also, the invariants in a native stub are different and
1561 // I'm not sure it is safe to have a MachCalRuntimeDirectNode
1562 // in there.
1563 // So for now, we do not zap in native stubs.
1565 JRT_LEAF(void, OptoRuntime::zap_dead_native_locals_C(JavaThread* thread))
1566 zap_dead_java_or_native_locals(thread, is_native_frame);
1567 JRT_END
1569 # endif