rev 56556 : 8232151: Minimal VM build broken after JDK-8232050
Reviewed-by: dholmes, clanger, redestad
1 /*
2 * Copyright (c) 1999, 2019, 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 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/systemDictionary.hpp"
29 #include "classfile/vmSymbols.hpp"
30 #include "compiler/compileBroker.hpp"
31 #include "compiler/compileLog.hpp"
32 #include "gc/shared/barrierSet.hpp"
33 #include "jfr/support/jfrIntrinsics.hpp"
34 #include "memory/resourceArea.hpp"
35 #include "oops/klass.inline.hpp"
36 #include "oops/objArrayKlass.hpp"
37 #include "opto/addnode.hpp"
38 #include "opto/arraycopynode.hpp"
39 #include "opto/c2compiler.hpp"
40 #include "opto/callGenerator.hpp"
41 #include "opto/castnode.hpp"
42 #include "opto/cfgnode.hpp"
43 #include "opto/convertnode.hpp"
44 #include "opto/countbitsnode.hpp"
45 #include "opto/intrinsicnode.hpp"
46 #include "opto/idealKit.hpp"
47 #include "opto/mathexactnode.hpp"
48 #include "opto/movenode.hpp"
49 #include "opto/mulnode.hpp"
50 #include "opto/narrowptrnode.hpp"
51 #include "opto/opaquenode.hpp"
52 #include "opto/parse.hpp"
53 #include "opto/runtime.hpp"
54 #include "opto/rootnode.hpp"
55 #include "opto/subnode.hpp"
56 #include "prims/nativeLookup.hpp"
57 #include "prims/unsafe.hpp"
58 #include "runtime/objectMonitor.hpp"
59 #include "runtime/sharedRuntime.hpp"
60 #include "utilities/macros.hpp"
61
62
63 class LibraryIntrinsic : public InlineCallGenerator {
64 // Extend the set of intrinsics known to the runtime:
65 public:
66 private:
67 bool _is_virtual;
68 bool _does_virtual_dispatch;
69 int8_t _predicates_count; // Intrinsic is predicated by several conditions
70 int8_t _last_predicate; // Last generated predicate
71 vmIntrinsics::ID _intrinsic_id;
72
73 public:
74 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
75 : InlineCallGenerator(m),
76 _is_virtual(is_virtual),
77 _does_virtual_dispatch(does_virtual_dispatch),
78 _predicates_count((int8_t)predicates_count),
79 _last_predicate((int8_t)-1),
80 _intrinsic_id(id)
81 {
82 }
83 virtual bool is_intrinsic() const { return true; }
84 virtual bool is_virtual() const { return _is_virtual; }
85 virtual bool is_predicated() const { return _predicates_count > 0; }
86 virtual int predicates_count() const { return _predicates_count; }
87 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; }
88 virtual JVMState* generate(JVMState* jvms);
89 virtual Node* generate_predicate(JVMState* jvms, int predicate);
90 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
91 };
92
93
94 // Local helper class for LibraryIntrinsic:
95 class LibraryCallKit : public GraphKit {
96 private:
97 LibraryIntrinsic* _intrinsic; // the library intrinsic being called
98 Node* _result; // the result node, if any
99 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted
100
101 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type);
102
103 public:
104 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
105 : GraphKit(jvms),
106 _intrinsic(intrinsic),
107 _result(NULL)
108 {
109 // Check if this is a root compile. In that case we don't have a caller.
110 if (!jvms->has_method()) {
111 _reexecute_sp = sp();
112 } else {
113 // Find out how many arguments the interpreter needs when deoptimizing
114 // and save the stack pointer value so it can used by uncommon_trap.
115 // We find the argument count by looking at the declared signature.
116 bool ignored_will_link;
117 ciSignature* declared_signature = NULL;
118 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
119 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
120 _reexecute_sp = sp() + nargs; // "push" arguments back on stack
121 }
122 }
123
124 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
125
126 ciMethod* caller() const { return jvms()->method(); }
127 int bci() const { return jvms()->bci(); }
128 LibraryIntrinsic* intrinsic() const { return _intrinsic; }
129 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); }
130 ciMethod* callee() const { return _intrinsic->method(); }
131
132 bool try_to_inline(int predicate);
133 Node* try_to_predicate(int predicate);
134
135 void push_result() {
136 // Push the result onto the stack.
137 if (!stopped() && result() != NULL) {
138 BasicType bt = result()->bottom_type()->basic_type();
139 push_node(bt, result());
140 }
141 }
142
143 private:
144 void fatal_unexpected_iid(vmIntrinsics::ID iid) {
145 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid));
146 }
147
148 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
149 void set_result(RegionNode* region, PhiNode* value);
150 Node* result() { return _result; }
151
152 virtual int reexecute_sp() { return _reexecute_sp; }
153
154 // Helper functions to inline natives
155 Node* generate_guard(Node* test, RegionNode* region, float true_prob);
156 Node* generate_slow_guard(Node* test, RegionNode* region);
157 Node* generate_fair_guard(Node* test, RegionNode* region);
158 Node* generate_negative_guard(Node* index, RegionNode* region,
159 // resulting CastII of index:
160 Node* *pos_index = NULL);
161 Node* generate_limit_guard(Node* offset, Node* subseq_length,
162 Node* array_length,
163 RegionNode* region);
164 void generate_string_range_check(Node* array, Node* offset,
165 Node* length, bool char_count);
166 Node* generate_current_thread(Node* &tls_output);
167 Node* load_mirror_from_klass(Node* klass);
168 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
169 RegionNode* region, int null_path,
170 int offset);
171 Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
172 RegionNode* region, int null_path) {
173 int offset = java_lang_Class::klass_offset_in_bytes();
174 return load_klass_from_mirror_common(mirror, never_see_null,
175 region, null_path,
176 offset);
177 }
178 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
179 RegionNode* region, int null_path) {
180 int offset = java_lang_Class::array_klass_offset_in_bytes();
181 return load_klass_from_mirror_common(mirror, never_see_null,
182 region, null_path,
183 offset);
184 }
185 Node* generate_access_flags_guard(Node* kls,
186 int modifier_mask, int modifier_bits,
187 RegionNode* region);
188 Node* generate_interface_guard(Node* kls, RegionNode* region);
189 Node* generate_array_guard(Node* kls, RegionNode* region) {
190 return generate_array_guard_common(kls, region, false, false);
191 }
192 Node* generate_non_array_guard(Node* kls, RegionNode* region) {
193 return generate_array_guard_common(kls, region, false, true);
194 }
195 Node* generate_objArray_guard(Node* kls, RegionNode* region) {
196 return generate_array_guard_common(kls, region, true, false);
197 }
198 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
199 return generate_array_guard_common(kls, region, true, true);
200 }
201 Node* generate_array_guard_common(Node* kls, RegionNode* region,
202 bool obj_array, bool not_array);
203 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
204 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
205 bool is_virtual = false, bool is_static = false);
206 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
207 return generate_method_call(method_id, false, true);
208 }
209 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
210 return generate_method_call(method_id, true, false);
211 }
212 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
213 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
214
215 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
216 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
217 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
218 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
219 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
220 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
221 bool inline_string_indexOfChar();
222 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
223 bool inline_string_toBytesU();
224 bool inline_string_getCharsU();
225 bool inline_string_copy(bool compress);
226 bool inline_string_char_access(bool is_store);
227 Node* round_double_node(Node* n);
228 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
229 bool inline_math_native(vmIntrinsics::ID id);
230 bool inline_math(vmIntrinsics::ID id);
231 bool inline_double_math(vmIntrinsics::ID id);
232 template <typename OverflowOp>
233 bool inline_math_overflow(Node* arg1, Node* arg2);
234 void inline_math_mathExact(Node* math, Node* test);
235 bool inline_math_addExactI(bool is_increment);
236 bool inline_math_addExactL(bool is_increment);
237 bool inline_math_multiplyExactI();
238 bool inline_math_multiplyExactL();
239 bool inline_math_multiplyHigh();
240 bool inline_math_negateExactI();
241 bool inline_math_negateExactL();
242 bool inline_math_subtractExactI(bool is_decrement);
243 bool inline_math_subtractExactL(bool is_decrement);
244 bool inline_min_max(vmIntrinsics::ID id);
245 bool inline_notify(vmIntrinsics::ID id);
246 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
247 // This returns Type::AnyPtr, RawPtr, or OopPtr.
248 int classify_unsafe_addr(Node* &base, Node* &offset, BasicType type);
249 Node* make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type = T_ILLEGAL, bool can_cast = false);
250
251 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
252 DecoratorSet mo_decorator_for_access_kind(AccessKind kind);
253 bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
254 static bool klass_needs_init_guard(Node* kls);
255 bool inline_unsafe_allocate();
256 bool inline_unsafe_newArray(bool uninitialized);
257 bool inline_unsafe_writeback0();
258 bool inline_unsafe_writebackSync0(bool is_pre);
259 bool inline_unsafe_copyMemory();
260 bool inline_native_currentThread();
261
262 bool inline_native_time_funcs(address method, const char* funcName);
263 #ifdef JFR_HAVE_INTRINSICS
264 bool inline_native_classID();
265 bool inline_native_getEventWriter();
266 #endif
267 bool inline_native_isInterrupted();
268 bool inline_native_Class_query(vmIntrinsics::ID id);
269 bool inline_native_subtype_check();
270 bool inline_native_getLength();
271 bool inline_array_copyOf(bool is_copyOfRange);
272 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
273 bool inline_preconditions_checkIndex();
274 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array);
275 bool inline_native_clone(bool is_virtual);
276 bool inline_native_Reflection_getCallerClass();
277 // Helper function for inlining native object hash method
278 bool inline_native_hashcode(bool is_virtual, bool is_static);
279 bool inline_native_getClass();
280
281 // Helper functions for inlining arraycopy
282 bool inline_arraycopy();
283 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
284 RegionNode* slow_region);
285 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
286 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp,
287 uint new_idx);
288
289 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
290 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
291 bool inline_unsafe_fence(vmIntrinsics::ID id);
292 bool inline_onspinwait();
293 bool inline_fp_conversions(vmIntrinsics::ID id);
294 bool inline_number_methods(vmIntrinsics::ID id);
295 bool inline_reference_get();
296 bool inline_Class_cast();
297 bool inline_aescrypt_Block(vmIntrinsics::ID id);
298 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
299 bool inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id);
300 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
301 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
302 Node* inline_electronicCodeBook_AESCrypt_predicate(bool decrypting);
303 Node* inline_counterMode_AESCrypt_predicate();
304 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
305 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
306 bool inline_ghash_processBlocks();
307 bool inline_base64_encodeBlock();
308 bool inline_sha_implCompress(vmIntrinsics::ID id);
309 bool inline_digestBase_implCompressMB(int predicate);
310 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
311 bool long_state, address stubAddr, const char *stubName,
312 Node* src_start, Node* ofs, Node* limit);
313 Node* get_state_from_sha_object(Node *sha_object);
314 Node* get_state_from_sha5_object(Node *sha_object);
315 Node* inline_digestBase_implCompressMB_predicate(int predicate);
316 bool inline_encodeISOArray();
317 bool inline_updateCRC32();
318 bool inline_updateBytesCRC32();
319 bool inline_updateByteBufferCRC32();
320 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
321 bool inline_updateBytesCRC32C();
322 bool inline_updateDirectByteBufferCRC32C();
323 bool inline_updateBytesAdler32();
324 bool inline_updateByteBufferAdler32();
325 bool inline_multiplyToLen();
326 bool inline_hasNegatives();
327 bool inline_squareToLen();
328 bool inline_mulAdd();
329 bool inline_montgomeryMultiply();
330 bool inline_montgomerySquare();
331 bool inline_vectorizedMismatch();
332 bool inline_fma(vmIntrinsics::ID id);
333 bool inline_character_compare(vmIntrinsics::ID id);
334 bool inline_fp_min_max(vmIntrinsics::ID id);
335
336 bool inline_profileBoolean();
337 bool inline_isCompileConstant();
338 void clear_upper_avx() {
339 #ifdef X86
340 if (UseAVX >= 2) {
341 C->set_clear_upper_avx(true);
342 }
343 #endif
344 }
345 };
346
347 //---------------------------make_vm_intrinsic----------------------------
348 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
349 vmIntrinsics::ID id = m->intrinsic_id();
350 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
351
352 if (!m->is_loaded()) {
353 // Do not attempt to inline unloaded methods.
354 return NULL;
355 }
356
357 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
358 bool is_available = false;
359
360 {
361 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
362 // the compiler must transition to '_thread_in_vm' state because both
363 // methods access VM-internal data.
364 VM_ENTRY_MARK;
365 methodHandle mh(THREAD, m->get_Method());
366 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
367 !C->directive()->is_intrinsic_disabled(mh) &&
368 !vmIntrinsics::is_disabled_by_flags(mh);
369
370 }
371
372 if (is_available) {
373 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
374 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
375 return new LibraryIntrinsic(m, is_virtual,
376 vmIntrinsics::predicates_needed(id),
377 vmIntrinsics::does_virtual_dispatch(id),
378 (vmIntrinsics::ID) id);
379 } else {
380 return NULL;
381 }
382 }
383
384 //----------------------register_library_intrinsics-----------------------
385 // Initialize this file's data structures, for each Compile instance.
386 void Compile::register_library_intrinsics() {
387 // Nothing to do here.
388 }
389
390 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
391 LibraryCallKit kit(jvms, this);
392 Compile* C = kit.C;
393 int nodes = C->unique();
394 #ifndef PRODUCT
395 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
396 char buf[1000];
397 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
398 tty->print_cr("Intrinsic %s", str);
399 }
400 #endif
401 ciMethod* callee = kit.callee();
402 const int bci = kit.bci();
403
404 // Try to inline the intrinsic.
405 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
406 kit.try_to_inline(_last_predicate)) {
407 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
408 : "(intrinsic)";
409 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
410 if (C->print_intrinsics() || C->print_inlining()) {
411 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
412 }
413 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
414 if (C->log()) {
415 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
416 vmIntrinsics::name_at(intrinsic_id()),
417 (is_virtual() ? " virtual='1'" : ""),
418 C->unique() - nodes);
419 }
420 // Push the result from the inlined method onto the stack.
421 kit.push_result();
422 C->print_inlining_update(this);
423 return kit.transfer_exceptions_into_jvms();
424 }
425
426 // The intrinsic bailed out
427 if (jvms->has_method()) {
428 // Not a root compile.
429 const char* msg;
430 if (callee->intrinsic_candidate()) {
431 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
432 } else {
433 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
434 : "failed to inline (intrinsic), method not annotated";
435 }
436 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
437 if (C->print_intrinsics() || C->print_inlining()) {
438 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
439 }
440 } else {
441 // Root compile
442 ResourceMark rm;
443 stringStream msg_stream;
444 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
445 vmIntrinsics::name_at(intrinsic_id()),
446 is_virtual() ? " (virtual)" : "", bci);
447 const char *msg = msg_stream.as_string();
448 log_debug(jit, inlining)("%s", msg);
449 if (C->print_intrinsics() || C->print_inlining()) {
450 tty->print("%s", msg);
451 }
452 }
453 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
454 C->print_inlining_update(this);
455 return NULL;
456 }
457
458 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
459 LibraryCallKit kit(jvms, this);
460 Compile* C = kit.C;
461 int nodes = C->unique();
462 _last_predicate = predicate;
463 #ifndef PRODUCT
464 assert(is_predicated() && predicate < predicates_count(), "sanity");
465 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
466 char buf[1000];
467 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
468 tty->print_cr("Predicate for intrinsic %s", str);
469 }
470 #endif
471 ciMethod* callee = kit.callee();
472 const int bci = kit.bci();
473
474 Node* slow_ctl = kit.try_to_predicate(predicate);
475 if (!kit.failing()) {
476 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
477 : "(intrinsic, predicate)";
478 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
479 if (C->print_intrinsics() || C->print_inlining()) {
480 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
481 }
482 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
483 if (C->log()) {
484 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
485 vmIntrinsics::name_at(intrinsic_id()),
486 (is_virtual() ? " virtual='1'" : ""),
487 C->unique() - nodes);
488 }
489 return slow_ctl; // Could be NULL if the check folds.
490 }
491
492 // The intrinsic bailed out
493 if (jvms->has_method()) {
494 // Not a root compile.
495 const char* msg = "failed to generate predicate for intrinsic";
496 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
497 if (C->print_intrinsics() || C->print_inlining()) {
498 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
499 }
500 } else {
501 // Root compile
502 ResourceMark rm;
503 stringStream msg_stream;
504 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
505 vmIntrinsics::name_at(intrinsic_id()),
506 is_virtual() ? " (virtual)" : "", bci);
507 const char *msg = msg_stream.as_string();
508 log_debug(jit, inlining)("%s", msg);
509 if (C->print_intrinsics() || C->print_inlining()) {
510 C->print_inlining_stream()->print("%s", msg);
511 }
512 }
513 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
514 return NULL;
515 }
516
517 bool LibraryCallKit::try_to_inline(int predicate) {
518 // Handle symbolic names for otherwise undistinguished boolean switches:
519 const bool is_store = true;
520 const bool is_compress = true;
521 const bool is_static = true;
522 const bool is_volatile = true;
523
524 if (!jvms()->has_method()) {
525 // Root JVMState has a null method.
526 assert(map()->memory()->Opcode() == Op_Parm, "");
527 // Insert the memory aliasing node
528 set_all_memory(reset_memory());
529 }
530 assert(merged_memory(), "");
531
532
533 switch (intrinsic_id()) {
534 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
535 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
536 case vmIntrinsics::_getClass: return inline_native_getClass();
537
538 case vmIntrinsics::_ceil:
539 case vmIntrinsics::_floor:
540 case vmIntrinsics::_rint:
541 case vmIntrinsics::_dsin:
542 case vmIntrinsics::_dcos:
543 case vmIntrinsics::_dtan:
544 case vmIntrinsics::_dabs:
545 case vmIntrinsics::_fabs:
546 case vmIntrinsics::_iabs:
547 case vmIntrinsics::_labs:
548 case vmIntrinsics::_datan2:
549 case vmIntrinsics::_dsqrt:
550 case vmIntrinsics::_dexp:
551 case vmIntrinsics::_dlog:
552 case vmIntrinsics::_dlog10:
553 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id());
554
555 case vmIntrinsics::_min:
556 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
557
558 case vmIntrinsics::_notify:
559 case vmIntrinsics::_notifyAll:
560 return inline_notify(intrinsic_id());
561
562 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
563 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
564 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
565 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
566 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
567 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
568 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
569 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
570 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
571 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
572 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
573 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
574 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
575
576 case vmIntrinsics::_arraycopy: return inline_arraycopy();
577
578 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
579 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
580 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
581 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
582
583 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
584 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
585 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
586 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
587 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
588 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
589 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
590
591 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
592 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
593
594 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
595 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
596 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
597 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
598
599 case vmIntrinsics::_compressStringC:
600 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
601 case vmIntrinsics::_inflateStringC:
602 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
603
604 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
605 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
606 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
607 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
608 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
609 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
610 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
611 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
612 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
613
614 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
615 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
616 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
617 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
618 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
619 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
620 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
621 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
622 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
623
624 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
625 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
626 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
627 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
628 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
629 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
630 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
631 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
632 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
633
634 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
635 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
636 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
637 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
638 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
639 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
640 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
641 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
642 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
643
644 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
645 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
646 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
647 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
648
649 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
650 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
651 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
652 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
653
654 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
655 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
656 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
657 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
658 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
659 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
660 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
661 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
662 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
663
664 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
665 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
666 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
667 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
668 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
669 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
670 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
671 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
672 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
673
674 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
675 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
676 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
677 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
678 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
679 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
680 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
681 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
682 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
683
684 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
685 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
686 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
687 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
688 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
689 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
690 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
691 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
692 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
693
694 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
695 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
696 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
697 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
698 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
699
700 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
701 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
702 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
703 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
704 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
705 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
706 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
707 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
708 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
709 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
710 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
711 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
712 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
713 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
714 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
715 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
716 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
717 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
718 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
719 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
720
721 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
722 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
723 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
724 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
725 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
726 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
727 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
728 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
729 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
730 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
731 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
732 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
733 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
734 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
735 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
736
737 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
738 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
739 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
740 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
741
742 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
743 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
744 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
745 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
746 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
747
748 case vmIntrinsics::_loadFence:
749 case vmIntrinsics::_storeFence:
750 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
751
752 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
753
754 case vmIntrinsics::_currentThread: return inline_native_currentThread();
755 case vmIntrinsics::_isInterrupted: return inline_native_isInterrupted();
756
757 #ifdef JFR_HAVE_INTRINSICS
758 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime");
759 case vmIntrinsics::_getClassId: return inline_native_classID();
760 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
761 #endif
762 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
763 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
764 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
765 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
766 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
767 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
768 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
769 case vmIntrinsics::_getLength: return inline_native_getLength();
770 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
771 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
772 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
773 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
774 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
775 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
776
777 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
778 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
779
780 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
781
782 case vmIntrinsics::_isInstance:
783 case vmIntrinsics::_getModifiers:
784 case vmIntrinsics::_isInterface:
785 case vmIntrinsics::_isArray:
786 case vmIntrinsics::_isPrimitive:
787 case vmIntrinsics::_getSuperclass:
788 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
789
790 case vmIntrinsics::_floatToRawIntBits:
791 case vmIntrinsics::_floatToIntBits:
792 case vmIntrinsics::_intBitsToFloat:
793 case vmIntrinsics::_doubleToRawLongBits:
794 case vmIntrinsics::_doubleToLongBits:
795 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
796
797 case vmIntrinsics::_numberOfLeadingZeros_i:
798 case vmIntrinsics::_numberOfLeadingZeros_l:
799 case vmIntrinsics::_numberOfTrailingZeros_i:
800 case vmIntrinsics::_numberOfTrailingZeros_l:
801 case vmIntrinsics::_bitCount_i:
802 case vmIntrinsics::_bitCount_l:
803 case vmIntrinsics::_reverseBytes_i:
804 case vmIntrinsics::_reverseBytes_l:
805 case vmIntrinsics::_reverseBytes_s:
806 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
807
808 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
809
810 case vmIntrinsics::_Reference_get: return inline_reference_get();
811
812 case vmIntrinsics::_Class_cast: return inline_Class_cast();
813
814 case vmIntrinsics::_aescrypt_encryptBlock:
815 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
816
817 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
818 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
819 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
820
821 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
822 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
823 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
824
825 case vmIntrinsics::_counterMode_AESCrypt:
826 return inline_counterMode_AESCrypt(intrinsic_id());
827
828 case vmIntrinsics::_sha_implCompress:
829 case vmIntrinsics::_sha2_implCompress:
830 case vmIntrinsics::_sha5_implCompress:
831 return inline_sha_implCompress(intrinsic_id());
832
833 case vmIntrinsics::_digestBase_implCompressMB:
834 return inline_digestBase_implCompressMB(predicate);
835
836 case vmIntrinsics::_multiplyToLen:
837 return inline_multiplyToLen();
838
839 case vmIntrinsics::_squareToLen:
840 return inline_squareToLen();
841
842 case vmIntrinsics::_mulAdd:
843 return inline_mulAdd();
844
845 case vmIntrinsics::_montgomeryMultiply:
846 return inline_montgomeryMultiply();
847 case vmIntrinsics::_montgomerySquare:
848 return inline_montgomerySquare();
849
850 case vmIntrinsics::_vectorizedMismatch:
851 return inline_vectorizedMismatch();
852
853 case vmIntrinsics::_ghash_processBlocks:
854 return inline_ghash_processBlocks();
855 case vmIntrinsics::_base64_encodeBlock:
856 return inline_base64_encodeBlock();
857
858 case vmIntrinsics::_encodeISOArray:
859 case vmIntrinsics::_encodeByteISOArray:
860 return inline_encodeISOArray();
861
862 case vmIntrinsics::_updateCRC32:
863 return inline_updateCRC32();
864 case vmIntrinsics::_updateBytesCRC32:
865 return inline_updateBytesCRC32();
866 case vmIntrinsics::_updateByteBufferCRC32:
867 return inline_updateByteBufferCRC32();
868
869 case vmIntrinsics::_updateBytesCRC32C:
870 return inline_updateBytesCRC32C();
871 case vmIntrinsics::_updateDirectByteBufferCRC32C:
872 return inline_updateDirectByteBufferCRC32C();
873
874 case vmIntrinsics::_updateBytesAdler32:
875 return inline_updateBytesAdler32();
876 case vmIntrinsics::_updateByteBufferAdler32:
877 return inline_updateByteBufferAdler32();
878
879 case vmIntrinsics::_profileBoolean:
880 return inline_profileBoolean();
881 case vmIntrinsics::_isCompileConstant:
882 return inline_isCompileConstant();
883
884 case vmIntrinsics::_hasNegatives:
885 return inline_hasNegatives();
886
887 case vmIntrinsics::_fmaD:
888 case vmIntrinsics::_fmaF:
889 return inline_fma(intrinsic_id());
890
891 case vmIntrinsics::_isDigit:
892 case vmIntrinsics::_isLowerCase:
893 case vmIntrinsics::_isUpperCase:
894 case vmIntrinsics::_isWhitespace:
895 return inline_character_compare(intrinsic_id());
896
897 case vmIntrinsics::_maxF:
898 case vmIntrinsics::_minF:
899 case vmIntrinsics::_maxD:
900 case vmIntrinsics::_minD:
901 return inline_fp_min_max(intrinsic_id());
902
903 default:
904 // If you get here, it may be that someone has added a new intrinsic
905 // to the list in vmSymbols.hpp without implementing it here.
906 #ifndef PRODUCT
907 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
908 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
909 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
910 }
911 #endif
912 return false;
913 }
914 }
915
916 Node* LibraryCallKit::try_to_predicate(int predicate) {
917 if (!jvms()->has_method()) {
918 // Root JVMState has a null method.
919 assert(map()->memory()->Opcode() == Op_Parm, "");
920 // Insert the memory aliasing node
921 set_all_memory(reset_memory());
922 }
923 assert(merged_memory(), "");
924
925 switch (intrinsic_id()) {
926 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
927 return inline_cipherBlockChaining_AESCrypt_predicate(false);
928 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
929 return inline_cipherBlockChaining_AESCrypt_predicate(true);
930 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
931 return inline_electronicCodeBook_AESCrypt_predicate(false);
932 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
933 return inline_electronicCodeBook_AESCrypt_predicate(true);
934 case vmIntrinsics::_counterMode_AESCrypt:
935 return inline_counterMode_AESCrypt_predicate();
936 case vmIntrinsics::_digestBase_implCompressMB:
937 return inline_digestBase_implCompressMB_predicate(predicate);
938
939 default:
940 // If you get here, it may be that someone has added a new intrinsic
941 // to the list in vmSymbols.hpp without implementing it here.
942 #ifndef PRODUCT
943 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
944 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
945 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
946 }
947 #endif
948 Node* slow_ctl = control();
949 set_control(top()); // No fast path instrinsic
950 return slow_ctl;
951 }
952 }
953
954 //------------------------------set_result-------------------------------
955 // Helper function for finishing intrinsics.
956 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
957 record_for_igvn(region);
958 set_control(_gvn.transform(region));
959 set_result( _gvn.transform(value));
960 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
961 }
962
963 //------------------------------generate_guard---------------------------
964 // Helper function for generating guarded fast-slow graph structures.
965 // The given 'test', if true, guards a slow path. If the test fails
966 // then a fast path can be taken. (We generally hope it fails.)
967 // In all cases, GraphKit::control() is updated to the fast path.
968 // The returned value represents the control for the slow path.
969 // The return value is never 'top'; it is either a valid control
970 // or NULL if it is obvious that the slow path can never be taken.
971 // Also, if region and the slow control are not NULL, the slow edge
972 // is appended to the region.
973 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
974 if (stopped()) {
975 // Already short circuited.
976 return NULL;
977 }
978
979 // Build an if node and its projections.
980 // If test is true we take the slow path, which we assume is uncommon.
981 if (_gvn.type(test) == TypeInt::ZERO) {
982 // The slow branch is never taken. No need to build this guard.
983 return NULL;
984 }
985
986 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
987
988 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
989 if (if_slow == top()) {
990 // The slow branch is never taken. No need to build this guard.
991 return NULL;
992 }
993
994 if (region != NULL)
995 region->add_req(if_slow);
996
997 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
998 set_control(if_fast);
999
1000 return if_slow;
1001 }
1002
1003 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
1004 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
1005 }
1006 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
1007 return generate_guard(test, region, PROB_FAIR);
1008 }
1009
1010 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
1011 Node* *pos_index) {
1012 if (stopped())
1013 return NULL; // already stopped
1014 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
1015 return NULL; // index is already adequately typed
1016 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
1017 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1018 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
1019 if (is_neg != NULL && pos_index != NULL) {
1020 // Emulate effect of Parse::adjust_map_after_if.
1021 Node* ccast = new CastIINode(index, TypeInt::POS);
1022 ccast->set_req(0, control());
1023 (*pos_index) = _gvn.transform(ccast);
1024 }
1025 return is_neg;
1026 }
1027
1028 // Make sure that 'position' is a valid limit index, in [0..length].
1029 // There are two equivalent plans for checking this:
1030 // A. (offset + copyLength) unsigned<= arrayLength
1031 // B. offset <= (arrayLength - copyLength)
1032 // We require that all of the values above, except for the sum and
1033 // difference, are already known to be non-negative.
1034 // Plan A is robust in the face of overflow, if offset and copyLength
1035 // are both hugely positive.
1036 //
1037 // Plan B is less direct and intuitive, but it does not overflow at
1038 // all, since the difference of two non-negatives is always
1039 // representable. Whenever Java methods must perform the equivalent
1040 // check they generally use Plan B instead of Plan A.
1041 // For the moment we use Plan A.
1042 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
1043 Node* subseq_length,
1044 Node* array_length,
1045 RegionNode* region) {
1046 if (stopped())
1047 return NULL; // already stopped
1048 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
1049 if (zero_offset && subseq_length->eqv_uncast(array_length))
1050 return NULL; // common case of whole-array copy
1051 Node* last = subseq_length;
1052 if (!zero_offset) // last += offset
1053 last = _gvn.transform(new AddINode(last, offset));
1054 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
1055 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1056 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
1057 return is_over;
1058 }
1059
1060 // Emit range checks for the given String.value byte array
1061 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
1062 if (stopped()) {
1063 return; // already stopped
1064 }
1065 RegionNode* bailout = new RegionNode(1);
1066 record_for_igvn(bailout);
1067 if (char_count) {
1068 // Convert char count to byte count
1069 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1070 }
1071
1072 // Offset and count must not be negative
1073 generate_negative_guard(offset, bailout);
1074 generate_negative_guard(count, bailout);
1075 // Offset + count must not exceed length of array
1076 generate_limit_guard(offset, count, load_array_length(array), bailout);
1077
1078 if (bailout->req() > 1) {
1079 PreserveJVMState pjvms(this);
1080 set_control(_gvn.transform(bailout));
1081 uncommon_trap(Deoptimization::Reason_intrinsic,
1082 Deoptimization::Action_maybe_recompile);
1083 }
1084 }
1085
1086 //--------------------------generate_current_thread--------------------
1087 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1088 ciKlass* thread_klass = env()->Thread_klass();
1089 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1090 Node* thread = _gvn.transform(new ThreadLocalNode());
1091 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1092 Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered);
1093 tls_output = thread;
1094 return threadObj;
1095 }
1096
1097
1098 //------------------------------make_string_method_node------------------------
1099 // Helper method for String intrinsic functions. This version is called with
1100 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1101 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1102 // containing the lengths of str1 and str2.
1103 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1104 Node* result = NULL;
1105 switch (opcode) {
1106 case Op_StrIndexOf:
1107 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1108 str1_start, cnt1, str2_start, cnt2, ae);
1109 break;
1110 case Op_StrComp:
1111 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1112 str1_start, cnt1, str2_start, cnt2, ae);
1113 break;
1114 case Op_StrEquals:
1115 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1116 // Use the constant length if there is one because optimized match rule may exist.
1117 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1118 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1119 break;
1120 default:
1121 ShouldNotReachHere();
1122 return NULL;
1123 }
1124
1125 // All these intrinsics have checks.
1126 C->set_has_split_ifs(true); // Has chance for split-if optimization
1127 clear_upper_avx();
1128
1129 return _gvn.transform(result);
1130 }
1131
1132 //------------------------------inline_string_compareTo------------------------
1133 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1134 Node* arg1 = argument(0);
1135 Node* arg2 = argument(1);
1136
1137 arg1 = must_be_not_null(arg1, true);
1138 arg2 = must_be_not_null(arg2, true);
1139
1140 arg1 = access_resolve(arg1, ACCESS_READ);
1141 arg2 = access_resolve(arg2, ACCESS_READ);
1142
1143 // Get start addr and length of first argument
1144 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1145 Node* arg1_cnt = load_array_length(arg1);
1146
1147 // Get start addr and length of second argument
1148 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1149 Node* arg2_cnt = load_array_length(arg2);
1150
1151 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1152 set_result(result);
1153 return true;
1154 }
1155
1156 //------------------------------inline_string_equals------------------------
1157 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1158 Node* arg1 = argument(0);
1159 Node* arg2 = argument(1);
1160
1161 // paths (plus control) merge
1162 RegionNode* region = new RegionNode(3);
1163 Node* phi = new PhiNode(region, TypeInt::BOOL);
1164
1165 if (!stopped()) {
1166
1167 arg1 = must_be_not_null(arg1, true);
1168 arg2 = must_be_not_null(arg2, true);
1169
1170 arg1 = access_resolve(arg1, ACCESS_READ);
1171 arg2 = access_resolve(arg2, ACCESS_READ);
1172
1173 // Get start addr and length of first argument
1174 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1175 Node* arg1_cnt = load_array_length(arg1);
1176
1177 // Get start addr and length of second argument
1178 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1179 Node* arg2_cnt = load_array_length(arg2);
1180
1181 // Check for arg1_cnt != arg2_cnt
1182 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1183 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1184 Node* if_ne = generate_slow_guard(bol, NULL);
1185 if (if_ne != NULL) {
1186 phi->init_req(2, intcon(0));
1187 region->init_req(2, if_ne);
1188 }
1189
1190 // Check for count == 0 is done by assembler code for StrEquals.
1191
1192 if (!stopped()) {
1193 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1194 phi->init_req(1, equals);
1195 region->init_req(1, control());
1196 }
1197 }
1198
1199 // post merge
1200 set_control(_gvn.transform(region));
1201 record_for_igvn(region);
1202
1203 set_result(_gvn.transform(phi));
1204 return true;
1205 }
1206
1207 //------------------------------inline_array_equals----------------------------
1208 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1209 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1210 Node* arg1 = argument(0);
1211 Node* arg2 = argument(1);
1212
1213 arg1 = access_resolve(arg1, ACCESS_READ);
1214 arg2 = access_resolve(arg2, ACCESS_READ);
1215
1216 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1217 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1218 clear_upper_avx();
1219
1220 return true;
1221 }
1222
1223 //------------------------------inline_hasNegatives------------------------------
1224 bool LibraryCallKit::inline_hasNegatives() {
1225 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1226 return false;
1227 }
1228
1229 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1230 // no receiver since it is static method
1231 Node* ba = argument(0);
1232 Node* offset = argument(1);
1233 Node* len = argument(2);
1234
1235 ba = must_be_not_null(ba, true);
1236
1237 // Range checks
1238 generate_string_range_check(ba, offset, len, false);
1239 if (stopped()) {
1240 return true;
1241 }
1242 ba = access_resolve(ba, ACCESS_READ);
1243 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1244 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1245 set_result(_gvn.transform(result));
1246 return true;
1247 }
1248
1249 bool LibraryCallKit::inline_preconditions_checkIndex() {
1250 Node* index = argument(0);
1251 Node* length = argument(1);
1252 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1253 return false;
1254 }
1255
1256 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1257 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1258
1259 {
1260 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1261 uncommon_trap(Deoptimization::Reason_intrinsic,
1262 Deoptimization::Action_make_not_entrant);
1263 }
1264
1265 if (stopped()) {
1266 return false;
1267 }
1268
1269 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1270 BoolTest::mask btest = BoolTest::lt;
1271 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1272 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1273 _gvn.set_type(rc, rc->Value(&_gvn));
1274 if (!rc_bool->is_Con()) {
1275 record_for_igvn(rc);
1276 }
1277 set_control(_gvn.transform(new IfTrueNode(rc)));
1278 {
1279 PreserveJVMState pjvms(this);
1280 set_control(_gvn.transform(new IfFalseNode(rc)));
1281 uncommon_trap(Deoptimization::Reason_range_check,
1282 Deoptimization::Action_make_not_entrant);
1283 }
1284
1285 if (stopped()) {
1286 return false;
1287 }
1288
1289 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1290 result->set_req(0, control());
1291 result = _gvn.transform(result);
1292 set_result(result);
1293 replace_in_map(index, result);
1294 clear_upper_avx();
1295 return true;
1296 }
1297
1298 //------------------------------inline_string_indexOf------------------------
1299 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1300 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1301 return false;
1302 }
1303 Node* src = argument(0);
1304 Node* tgt = argument(1);
1305
1306 // Make the merge point
1307 RegionNode* result_rgn = new RegionNode(4);
1308 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1309
1310 src = must_be_not_null(src, true);
1311 tgt = must_be_not_null(tgt, true);
1312
1313 src = access_resolve(src, ACCESS_READ);
1314 tgt = access_resolve(tgt, ACCESS_READ);
1315
1316 // Get start addr and length of source string
1317 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1318 Node* src_count = load_array_length(src);
1319
1320 // Get start addr and length of substring
1321 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1322 Node* tgt_count = load_array_length(tgt);
1323
1324 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1325 // Divide src size by 2 if String is UTF16 encoded
1326 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1327 }
1328 if (ae == StrIntrinsicNode::UU) {
1329 // Divide substring size by 2 if String is UTF16 encoded
1330 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1331 }
1332
1333 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1334 if (result != NULL) {
1335 result_phi->init_req(3, result);
1336 result_rgn->init_req(3, control());
1337 }
1338 set_control(_gvn.transform(result_rgn));
1339 record_for_igvn(result_rgn);
1340 set_result(_gvn.transform(result_phi));
1341
1342 return true;
1343 }
1344
1345 //-----------------------------inline_string_indexOf-----------------------
1346 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1347 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1348 return false;
1349 }
1350 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1351 return false;
1352 }
1353 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1354 Node* src = argument(0); // byte[]
1355 Node* src_count = argument(1); // char count
1356 Node* tgt = argument(2); // byte[]
1357 Node* tgt_count = argument(3); // char count
1358 Node* from_index = argument(4); // char index
1359
1360 src = must_be_not_null(src, true);
1361 tgt = must_be_not_null(tgt, true);
1362
1363 src = access_resolve(src, ACCESS_READ);
1364 tgt = access_resolve(tgt, ACCESS_READ);
1365
1366 // Multiply byte array index by 2 if String is UTF16 encoded
1367 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1368 src_count = _gvn.transform(new SubINode(src_count, from_index));
1369 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1370 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1371
1372 // Range checks
1373 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1374 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1375 if (stopped()) {
1376 return true;
1377 }
1378
1379 RegionNode* region = new RegionNode(5);
1380 Node* phi = new PhiNode(region, TypeInt::INT);
1381
1382 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1383 if (result != NULL) {
1384 // The result is index relative to from_index if substring was found, -1 otherwise.
1385 // Generate code which will fold into cmove.
1386 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1387 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1388
1389 Node* if_lt = generate_slow_guard(bol, NULL);
1390 if (if_lt != NULL) {
1391 // result == -1
1392 phi->init_req(3, result);
1393 region->init_req(3, if_lt);
1394 }
1395 if (!stopped()) {
1396 result = _gvn.transform(new AddINode(result, from_index));
1397 phi->init_req(4, result);
1398 region->init_req(4, control());
1399 }
1400 }
1401
1402 set_control(_gvn.transform(region));
1403 record_for_igvn(region);
1404 set_result(_gvn.transform(phi));
1405 clear_upper_avx();
1406
1407 return true;
1408 }
1409
1410 // Create StrIndexOfNode with fast path checks
1411 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1412 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1413 // Check for substr count > string count
1414 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1415 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1416 Node* if_gt = generate_slow_guard(bol, NULL);
1417 if (if_gt != NULL) {
1418 phi->init_req(1, intcon(-1));
1419 region->init_req(1, if_gt);
1420 }
1421 if (!stopped()) {
1422 // Check for substr count == 0
1423 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1424 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1425 Node* if_zero = generate_slow_guard(bol, NULL);
1426 if (if_zero != NULL) {
1427 phi->init_req(2, intcon(0));
1428 region->init_req(2, if_zero);
1429 }
1430 }
1431 if (!stopped()) {
1432 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1433 }
1434 return NULL;
1435 }
1436
1437 //-----------------------------inline_string_indexOfChar-----------------------
1438 bool LibraryCallKit::inline_string_indexOfChar() {
1439 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1440 return false;
1441 }
1442 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1443 return false;
1444 }
1445 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1446 Node* src = argument(0); // byte[]
1447 Node* tgt = argument(1); // tgt is int ch
1448 Node* from_index = argument(2);
1449 Node* max = argument(3);
1450
1451 src = must_be_not_null(src, true);
1452 src = access_resolve(src, ACCESS_READ);
1453
1454 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1455 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1456 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1457
1458 // Range checks
1459 generate_string_range_check(src, src_offset, src_count, true);
1460 if (stopped()) {
1461 return true;
1462 }
1463
1464 RegionNode* region = new RegionNode(3);
1465 Node* phi = new PhiNode(region, TypeInt::INT);
1466
1467 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1468 C->set_has_split_ifs(true); // Has chance for split-if optimization
1469 _gvn.transform(result);
1470
1471 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1472 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1473
1474 Node* if_lt = generate_slow_guard(bol, NULL);
1475 if (if_lt != NULL) {
1476 // result == -1
1477 phi->init_req(2, result);
1478 region->init_req(2, if_lt);
1479 }
1480 if (!stopped()) {
1481 result = _gvn.transform(new AddINode(result, from_index));
1482 phi->init_req(1, result);
1483 region->init_req(1, control());
1484 }
1485 set_control(_gvn.transform(region));
1486 record_for_igvn(region);
1487 set_result(_gvn.transform(phi));
1488
1489 return true;
1490 }
1491 //---------------------------inline_string_copy---------------------
1492 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1493 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1494 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1495 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1496 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1497 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1498 bool LibraryCallKit::inline_string_copy(bool compress) {
1499 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1500 return false;
1501 }
1502 int nargs = 5; // 2 oops, 3 ints
1503 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1504
1505 Node* src = argument(0);
1506 Node* src_offset = argument(1);
1507 Node* dst = argument(2);
1508 Node* dst_offset = argument(3);
1509 Node* length = argument(4);
1510
1511 // Check for allocation before we add nodes that would confuse
1512 // tightly_coupled_allocation()
1513 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1514
1515 // Figure out the size and type of the elements we will be copying.
1516 const Type* src_type = src->Value(&_gvn);
1517 const Type* dst_type = dst->Value(&_gvn);
1518 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1519 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1520 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1521 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1522 "Unsupported array types for inline_string_copy");
1523
1524 src = must_be_not_null(src, true);
1525 dst = must_be_not_null(dst, true);
1526
1527 // Convert char[] offsets to byte[] offsets
1528 bool convert_src = (compress && src_elem == T_BYTE);
1529 bool convert_dst = (!compress && dst_elem == T_BYTE);
1530 if (convert_src) {
1531 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1532 } else if (convert_dst) {
1533 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1534 }
1535
1536 // Range checks
1537 generate_string_range_check(src, src_offset, length, convert_src);
1538 generate_string_range_check(dst, dst_offset, length, convert_dst);
1539 if (stopped()) {
1540 return true;
1541 }
1542
1543 src = access_resolve(src, ACCESS_READ);
1544 dst = access_resolve(dst, ACCESS_WRITE);
1545
1546 Node* src_start = array_element_address(src, src_offset, src_elem);
1547 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1548 // 'src_start' points to src array + scaled offset
1549 // 'dst_start' points to dst array + scaled offset
1550 Node* count = NULL;
1551 if (compress) {
1552 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1553 } else {
1554 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1555 }
1556
1557 if (alloc != NULL) {
1558 if (alloc->maybe_set_complete(&_gvn)) {
1559 // "You break it, you buy it."
1560 InitializeNode* init = alloc->initialization();
1561 assert(init->is_complete(), "we just did this");
1562 init->set_complete_with_arraycopy();
1563 assert(dst->is_CheckCastPP(), "sanity");
1564 assert(dst->in(0)->in(0) == init, "dest pinned");
1565 }
1566 // Do not let stores that initialize this object be reordered with
1567 // a subsequent store that would make this object accessible by
1568 // other threads.
1569 // Record what AllocateNode this StoreStore protects so that
1570 // escape analysis can go from the MemBarStoreStoreNode to the
1571 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1572 // based on the escape status of the AllocateNode.
1573 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1574 }
1575 if (compress) {
1576 set_result(_gvn.transform(count));
1577 }
1578 clear_upper_avx();
1579
1580 return true;
1581 }
1582
1583 #ifdef _LP64
1584 #define XTOP ,top() /*additional argument*/
1585 #else //_LP64
1586 #define XTOP /*no additional argument*/
1587 #endif //_LP64
1588
1589 //------------------------inline_string_toBytesU--------------------------
1590 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1591 bool LibraryCallKit::inline_string_toBytesU() {
1592 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1593 return false;
1594 }
1595 // Get the arguments.
1596 Node* value = argument(0);
1597 Node* offset = argument(1);
1598 Node* length = argument(2);
1599
1600 Node* newcopy = NULL;
1601
1602 // Set the original stack and the reexecute bit for the interpreter to reexecute
1603 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1604 { PreserveReexecuteState preexecs(this);
1605 jvms()->set_should_reexecute(true);
1606
1607 // Check if a null path was taken unconditionally.
1608 value = null_check(value);
1609
1610 RegionNode* bailout = new RegionNode(1);
1611 record_for_igvn(bailout);
1612
1613 // Range checks
1614 generate_negative_guard(offset, bailout);
1615 generate_negative_guard(length, bailout);
1616 generate_limit_guard(offset, length, load_array_length(value), bailout);
1617 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1618 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1619
1620 if (bailout->req() > 1) {
1621 PreserveJVMState pjvms(this);
1622 set_control(_gvn.transform(bailout));
1623 uncommon_trap(Deoptimization::Reason_intrinsic,
1624 Deoptimization::Action_maybe_recompile);
1625 }
1626 if (stopped()) {
1627 return true;
1628 }
1629
1630 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1631 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1632 newcopy = new_array(klass_node, size, 0); // no arguments to push
1633 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1634
1635 // Calculate starting addresses.
1636 value = access_resolve(value, ACCESS_READ);
1637 Node* src_start = array_element_address(value, offset, T_CHAR);
1638 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1639
1640 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1641 const TypeInt* toffset = gvn().type(offset)->is_int();
1642 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1643
1644 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1645 const char* copyfunc_name = "arraycopy";
1646 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1647 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1648 OptoRuntime::fast_arraycopy_Type(),
1649 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1650 src_start, dst_start, ConvI2X(length) XTOP);
1651 // Do not let reads from the cloned object float above the arraycopy.
1652 if (alloc != NULL) {
1653 if (alloc->maybe_set_complete(&_gvn)) {
1654 // "You break it, you buy it."
1655 InitializeNode* init = alloc->initialization();
1656 assert(init->is_complete(), "we just did this");
1657 init->set_complete_with_arraycopy();
1658 assert(newcopy->is_CheckCastPP(), "sanity");
1659 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1660 }
1661 // Do not let stores that initialize this object be reordered with
1662 // a subsequent store that would make this object accessible by
1663 // other threads.
1664 // Record what AllocateNode this StoreStore protects so that
1665 // escape analysis can go from the MemBarStoreStoreNode to the
1666 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1667 // based on the escape status of the AllocateNode.
1668 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1669 } else {
1670 insert_mem_bar(Op_MemBarCPUOrder);
1671 }
1672 } // original reexecute is set back here
1673
1674 C->set_has_split_ifs(true); // Has chance for split-if optimization
1675 if (!stopped()) {
1676 set_result(newcopy);
1677 }
1678 clear_upper_avx();
1679
1680 return true;
1681 }
1682
1683 //------------------------inline_string_getCharsU--------------------------
1684 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1685 bool LibraryCallKit::inline_string_getCharsU() {
1686 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1687 return false;
1688 }
1689
1690 // Get the arguments.
1691 Node* src = argument(0);
1692 Node* src_begin = argument(1);
1693 Node* src_end = argument(2); // exclusive offset (i < src_end)
1694 Node* dst = argument(3);
1695 Node* dst_begin = argument(4);
1696
1697 // Check for allocation before we add nodes that would confuse
1698 // tightly_coupled_allocation()
1699 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1700
1701 // Check if a null path was taken unconditionally.
1702 src = null_check(src);
1703 dst = null_check(dst);
1704 if (stopped()) {
1705 return true;
1706 }
1707
1708 // Get length and convert char[] offset to byte[] offset
1709 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1710 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1711
1712 // Range checks
1713 generate_string_range_check(src, src_begin, length, true);
1714 generate_string_range_check(dst, dst_begin, length, false);
1715 if (stopped()) {
1716 return true;
1717 }
1718
1719 if (!stopped()) {
1720 src = access_resolve(src, ACCESS_READ);
1721 dst = access_resolve(dst, ACCESS_WRITE);
1722
1723 // Calculate starting addresses.
1724 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1725 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1726
1727 // Check if array addresses are aligned to HeapWordSize
1728 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1729 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1730 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1731 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1732
1733 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1734 const char* copyfunc_name = "arraycopy";
1735 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1736 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1737 OptoRuntime::fast_arraycopy_Type(),
1738 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1739 src_start, dst_start, ConvI2X(length) XTOP);
1740 // Do not let reads from the cloned object float above the arraycopy.
1741 if (alloc != NULL) {
1742 if (alloc->maybe_set_complete(&_gvn)) {
1743 // "You break it, you buy it."
1744 InitializeNode* init = alloc->initialization();
1745 assert(init->is_complete(), "we just did this");
1746 init->set_complete_with_arraycopy();
1747 assert(dst->is_CheckCastPP(), "sanity");
1748 assert(dst->in(0)->in(0) == init, "dest pinned");
1749 }
1750 // Do not let stores that initialize this object be reordered with
1751 // a subsequent store that would make this object accessible by
1752 // other threads.
1753 // Record what AllocateNode this StoreStore protects so that
1754 // escape analysis can go from the MemBarStoreStoreNode to the
1755 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1756 // based on the escape status of the AllocateNode.
1757 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1758 } else {
1759 insert_mem_bar(Op_MemBarCPUOrder);
1760 }
1761 }
1762
1763 C->set_has_split_ifs(true); // Has chance for split-if optimization
1764 return true;
1765 }
1766
1767 //----------------------inline_string_char_access----------------------------
1768 // Store/Load char to/from byte[] array.
1769 // static void StringUTF16.putChar(byte[] val, int index, int c)
1770 // static char StringUTF16.getChar(byte[] val, int index)
1771 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1772 Node* value = argument(0);
1773 Node* index = argument(1);
1774 Node* ch = is_store ? argument(2) : NULL;
1775
1776 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1777 // correctly requires matched array shapes.
1778 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1779 "sanity: byte[] and char[] bases agree");
1780 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1781 "sanity: byte[] and char[] scales agree");
1782
1783 // Bail when getChar over constants is requested: constant folding would
1784 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1785 // Java method would constant fold nicely instead.
1786 if (!is_store && value->is_Con() && index->is_Con()) {
1787 return false;
1788 }
1789
1790 value = must_be_not_null(value, true);
1791 value = access_resolve(value, is_store ? ACCESS_WRITE : ACCESS_READ);
1792
1793 Node* adr = array_element_address(value, index, T_CHAR);
1794 if (adr->is_top()) {
1795 return false;
1796 }
1797 if (is_store) {
1798 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1799 } else {
1800 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
1801 set_result(ch);
1802 }
1803 return true;
1804 }
1805
1806 //--------------------------round_double_node--------------------------------
1807 // Round a double node if necessary.
1808 Node* LibraryCallKit::round_double_node(Node* n) {
1809 if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1810 n = _gvn.transform(new RoundDoubleNode(0, n));
1811 return n;
1812 }
1813
1814 //------------------------------inline_math-----------------------------------
1815 // public static double Math.abs(double)
1816 // public static double Math.sqrt(double)
1817 // public static double Math.log(double)
1818 // public static double Math.log10(double)
1819 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1820 Node* arg = round_double_node(argument(0));
1821 Node* n = NULL;
1822 switch (id) {
1823 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1824 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1825 case vmIntrinsics::_ceil: n = new RoundDoubleModeNode(arg, makecon(TypeInt::make(2))); break;
1826 case vmIntrinsics::_floor: n = new RoundDoubleModeNode(arg, makecon(TypeInt::make(1))); break;
1827 case vmIntrinsics::_rint: n = new RoundDoubleModeNode(arg, makecon(TypeInt::make(0))); break;
1828 default: fatal_unexpected_iid(id); break;
1829 }
1830 set_result(_gvn.transform(n));
1831 return true;
1832 }
1833
1834 //------------------------------inline_math-----------------------------------
1835 // public static float Math.abs(float)
1836 // public static int Math.abs(int)
1837 // public static long Math.abs(long)
1838 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1839 Node* arg = argument(0);
1840 Node* n = NULL;
1841 switch (id) {
1842 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1843 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1844 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1845 default: fatal_unexpected_iid(id); break;
1846 }
1847 set_result(_gvn.transform(n));
1848 return true;
1849 }
1850
1851 //------------------------------runtime_math-----------------------------
1852 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1853 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1854 "must be (DD)D or (D)D type");
1855
1856 // Inputs
1857 Node* a = round_double_node(argument(0));
1858 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1859
1860 const TypePtr* no_memory_effects = NULL;
1861 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1862 no_memory_effects,
1863 a, top(), b, b ? top() : NULL);
1864 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1865 #ifdef ASSERT
1866 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1867 assert(value_top == top(), "second value must be top");
1868 #endif
1869
1870 set_result(value);
1871 return true;
1872 }
1873
1874 //------------------------------inline_math_native-----------------------------
1875 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1876 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1877 switch (id) {
1878 // These intrinsics are not properly supported on all hardware
1879 case vmIntrinsics::_dsin:
1880 return StubRoutines::dsin() != NULL ?
1881 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1882 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1883 case vmIntrinsics::_dcos:
1884 return StubRoutines::dcos() != NULL ?
1885 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1886 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1887 case vmIntrinsics::_dtan:
1888 return StubRoutines::dtan() != NULL ?
1889 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1890 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1891 case vmIntrinsics::_dlog:
1892 return StubRoutines::dlog() != NULL ?
1893 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1894 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1895 case vmIntrinsics::_dlog10:
1896 return StubRoutines::dlog10() != NULL ?
1897 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1898 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1899
1900 // These intrinsics are supported on all hardware
1901 case vmIntrinsics::_ceil:
1902 case vmIntrinsics::_floor:
1903 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1904 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1905 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1906 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1907 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1908 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1909
1910 case vmIntrinsics::_dexp:
1911 return StubRoutines::dexp() != NULL ?
1912 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1913 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1914 case vmIntrinsics::_dpow: {
1915 Node* exp = round_double_node(argument(2));
1916 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1917 if (d != NULL && d->getd() == 2.0) {
1918 // Special case: pow(x, 2.0) => x * x
1919 Node* base = round_double_node(argument(0));
1920 set_result(_gvn.transform(new MulDNode(base, base)));
1921 return true;
1922 }
1923 return StubRoutines::dpow() != NULL ?
1924 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1925 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1926 }
1927 #undef FN_PTR
1928
1929 // These intrinsics are not yet correctly implemented
1930 case vmIntrinsics::_datan2:
1931 return false;
1932
1933 default:
1934 fatal_unexpected_iid(id);
1935 return false;
1936 }
1937 }
1938
1939 static bool is_simple_name(Node* n) {
1940 return (n->req() == 1 // constant
1941 || (n->is_Type() && n->as_Type()->type()->singleton())
1942 || n->is_Proj() // parameter or return value
1943 || n->is_Phi() // local of some sort
1944 );
1945 }
1946
1947 //----------------------------inline_notify-----------------------------------*
1948 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1949 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1950 address func;
1951 if (id == vmIntrinsics::_notify) {
1952 func = OptoRuntime::monitor_notify_Java();
1953 } else {
1954 func = OptoRuntime::monitor_notifyAll_Java();
1955 }
1956 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1957 make_slow_call_ex(call, env()->Throwable_klass(), false);
1958 return true;
1959 }
1960
1961
1962 //----------------------------inline_min_max-----------------------------------
1963 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1964 set_result(generate_min_max(id, argument(0), argument(1)));
1965 return true;
1966 }
1967
1968 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1969 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1970 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1971 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1972 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1973
1974 {
1975 PreserveJVMState pjvms(this);
1976 PreserveReexecuteState preexecs(this);
1977 jvms()->set_should_reexecute(true);
1978
1979 set_control(slow_path);
1980 set_i_o(i_o());
1981
1982 uncommon_trap(Deoptimization::Reason_intrinsic,
1983 Deoptimization::Action_none);
1984 }
1985
1986 set_control(fast_path);
1987 set_result(math);
1988 }
1989
1990 template <typename OverflowOp>
1991 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1992 typedef typename OverflowOp::MathOp MathOp;
1993
1994 MathOp* mathOp = new MathOp(arg1, arg2);
1995 Node* operation = _gvn.transform( mathOp );
1996 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1997 inline_math_mathExact(operation, ofcheck);
1998 return true;
1999 }
2000
2001 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2002 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2003 }
2004
2005 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2006 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2007 }
2008
2009 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2010 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2011 }
2012
2013 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2014 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2015 }
2016
2017 bool LibraryCallKit::inline_math_negateExactI() {
2018 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2019 }
2020
2021 bool LibraryCallKit::inline_math_negateExactL() {
2022 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2023 }
2024
2025 bool LibraryCallKit::inline_math_multiplyExactI() {
2026 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2027 }
2028
2029 bool LibraryCallKit::inline_math_multiplyExactL() {
2030 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2031 }
2032
2033 bool LibraryCallKit::inline_math_multiplyHigh() {
2034 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2035 return true;
2036 }
2037
2038 Node*
2039 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2040 // These are the candidate return value:
2041 Node* xvalue = x0;
2042 Node* yvalue = y0;
2043
2044 if (xvalue == yvalue) {
2045 return xvalue;
2046 }
2047
2048 bool want_max = (id == vmIntrinsics::_max);
2049
2050 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
2051 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
2052 if (txvalue == NULL || tyvalue == NULL) return top();
2053 // This is not really necessary, but it is consistent with a
2054 // hypothetical MaxINode::Value method:
2055 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
2056
2057 // %%% This folding logic should (ideally) be in a different place.
2058 // Some should be inside IfNode, and there to be a more reliable
2059 // transformation of ?: style patterns into cmoves. We also want
2060 // more powerful optimizations around cmove and min/max.
2061
2062 // Try to find a dominating comparison of these guys.
2063 // It can simplify the index computation for Arrays.copyOf
2064 // and similar uses of System.arraycopy.
2065 // First, compute the normalized version of CmpI(x, y).
2066 int cmp_op = Op_CmpI;
2067 Node* xkey = xvalue;
2068 Node* ykey = yvalue;
2069 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
2070 if (ideal_cmpxy->is_Cmp()) {
2071 // E.g., if we have CmpI(length - offset, count),
2072 // it might idealize to CmpI(length, count + offset)
2073 cmp_op = ideal_cmpxy->Opcode();
2074 xkey = ideal_cmpxy->in(1);
2075 ykey = ideal_cmpxy->in(2);
2076 }
2077
2078 // Start by locating any relevant comparisons.
2079 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
2080 Node* cmpxy = NULL;
2081 Node* cmpyx = NULL;
2082 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
2083 Node* cmp = start_from->fast_out(k);
2084 if (cmp->outcnt() > 0 && // must have prior uses
2085 cmp->in(0) == NULL && // must be context-independent
2086 cmp->Opcode() == cmp_op) { // right kind of compare
2087 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
2088 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
2089 }
2090 }
2091
2092 const int NCMPS = 2;
2093 Node* cmps[NCMPS] = { cmpxy, cmpyx };
2094 int cmpn;
2095 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2096 if (cmps[cmpn] != NULL) break; // find a result
2097 }
2098 if (cmpn < NCMPS) {
2099 // Look for a dominating test that tells us the min and max.
2100 int depth = 0; // Limit search depth for speed
2101 Node* dom = control();
2102 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2103 if (++depth >= 100) break;
2104 Node* ifproj = dom;
2105 if (!ifproj->is_Proj()) continue;
2106 Node* iff = ifproj->in(0);
2107 if (!iff->is_If()) continue;
2108 Node* bol = iff->in(1);
2109 if (!bol->is_Bool()) continue;
2110 Node* cmp = bol->in(1);
2111 if (cmp == NULL) continue;
2112 for (cmpn = 0; cmpn < NCMPS; cmpn++)
2113 if (cmps[cmpn] == cmp) break;
2114 if (cmpn == NCMPS) continue;
2115 BoolTest::mask btest = bol->as_Bool()->_test._test;
2116 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
2117 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2118 // At this point, we know that 'x btest y' is true.
2119 switch (btest) {
2120 case BoolTest::eq:
2121 // They are proven equal, so we can collapse the min/max.
2122 // Either value is the answer. Choose the simpler.
2123 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2124 return yvalue;
2125 return xvalue;
2126 case BoolTest::lt: // x < y
2127 case BoolTest::le: // x <= y
2128 return (want_max ? yvalue : xvalue);
2129 case BoolTest::gt: // x > y
2130 case BoolTest::ge: // x >= y
2131 return (want_max ? xvalue : yvalue);
2132 default:
2133 break;
2134 }
2135 }
2136 }
2137
2138 // We failed to find a dominating test.
2139 // Let's pick a test that might GVN with prior tests.
2140 Node* best_bol = NULL;
2141 BoolTest::mask best_btest = BoolTest::illegal;
2142 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2143 Node* cmp = cmps[cmpn];
2144 if (cmp == NULL) continue;
2145 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2146 Node* bol = cmp->fast_out(j);
2147 if (!bol->is_Bool()) continue;
2148 BoolTest::mask btest = bol->as_Bool()->_test._test;
2149 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2150 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2151 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2152 best_bol = bol->as_Bool();
2153 best_btest = btest;
2154 }
2155 }
2156 }
2157
2158 Node* answer_if_true = NULL;
2159 Node* answer_if_false = NULL;
2160 switch (best_btest) {
2161 default:
2162 if (cmpxy == NULL)
2163 cmpxy = ideal_cmpxy;
2164 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2165 // and fall through:
2166 case BoolTest::lt: // x < y
2167 case BoolTest::le: // x <= y
2168 answer_if_true = (want_max ? yvalue : xvalue);
2169 answer_if_false = (want_max ? xvalue : yvalue);
2170 break;
2171 case BoolTest::gt: // x > y
2172 case BoolTest::ge: // x >= y
2173 answer_if_true = (want_max ? xvalue : yvalue);
2174 answer_if_false = (want_max ? yvalue : xvalue);
2175 break;
2176 }
2177
2178 jint hi, lo;
2179 if (want_max) {
2180 // We can sharpen the minimum.
2181 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2182 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2183 } else {
2184 // We can sharpen the maximum.
2185 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2186 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2187 }
2188
2189 // Use a flow-free graph structure, to avoid creating excess control edges
2190 // which could hinder other optimizations.
2191 // Since Math.min/max is often used with arraycopy, we want
2192 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2193 Node* cmov = CMoveNode::make(NULL, best_bol,
2194 answer_if_false, answer_if_true,
2195 TypeInt::make(lo, hi, widen));
2196
2197 return _gvn.transform(cmov);
2198
2199 /*
2200 // This is not as desirable as it may seem, since Min and Max
2201 // nodes do not have a full set of optimizations.
2202 // And they would interfere, anyway, with 'if' optimizations
2203 // and with CMoveI canonical forms.
2204 switch (id) {
2205 case vmIntrinsics::_min:
2206 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2207 case vmIntrinsics::_max:
2208 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2209 default:
2210 ShouldNotReachHere();
2211 }
2212 */
2213 }
2214
2215 inline int
2216 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2217 const TypePtr* base_type = TypePtr::NULL_PTR;
2218 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2219 if (base_type == NULL) {
2220 // Unknown type.
2221 return Type::AnyPtr;
2222 } else if (base_type == TypePtr::NULL_PTR) {
2223 // Since this is a NULL+long form, we have to switch to a rawptr.
2224 base = _gvn.transform(new CastX2PNode(offset));
2225 offset = MakeConX(0);
2226 return Type::RawPtr;
2227 } else if (base_type->base() == Type::RawPtr) {
2228 return Type::RawPtr;
2229 } else if (base_type->isa_oopptr()) {
2230 // Base is never null => always a heap address.
2231 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2232 return Type::OopPtr;
2233 }
2234 // Offset is small => always a heap address.
2235 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2236 if (offset_type != NULL &&
2237 base_type->offset() == 0 && // (should always be?)
2238 offset_type->_lo >= 0 &&
2239 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2240 return Type::OopPtr;
2241 } else if (type == T_OBJECT) {
2242 // off heap access to an oop doesn't make any sense. Has to be on
2243 // heap.
2244 return Type::OopPtr;
2245 }
2246 // Otherwise, it might either be oop+off or NULL+addr.
2247 return Type::AnyPtr;
2248 } else {
2249 // No information:
2250 return Type::AnyPtr;
2251 }
2252 }
2253
2254 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type, bool can_cast) {
2255 Node* uncasted_base = base;
2256 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2257 if (kind == Type::RawPtr) {
2258 return basic_plus_adr(top(), uncasted_base, offset);
2259 } else if (kind == Type::AnyPtr) {
2260 assert(base == uncasted_base, "unexpected base change");
2261 if (can_cast) {
2262 if (!_gvn.type(base)->speculative_maybe_null() &&
2263 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2264 // According to profiling, this access is always on
2265 // heap. Casting the base to not null and thus avoiding membars
2266 // around the access should allow better optimizations
2267 Node* null_ctl = top();
2268 base = null_check_oop(base, &null_ctl, true, true, true);
2269 assert(null_ctl->is_top(), "no null control here");
2270 return basic_plus_adr(base, offset);
2271 } else if (_gvn.type(base)->speculative_always_null() &&
2272 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2273 // According to profiling, this access is always off
2274 // heap.
2275 base = null_assert(base);
2276 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2277 offset = MakeConX(0);
2278 return basic_plus_adr(top(), raw_base, offset);
2279 }
2280 }
2281 // We don't know if it's an on heap or off heap access. Fall back
2282 // to raw memory access.
2283 base = access_resolve(base, decorators);
2284 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2285 return basic_plus_adr(top(), raw, offset);
2286 } else {
2287 assert(base == uncasted_base, "unexpected base change");
2288 // We know it's an on heap access so base can't be null
2289 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2290 base = must_be_not_null(base, true);
2291 }
2292 return basic_plus_adr(base, offset);
2293 }
2294 }
2295
2296 //--------------------------inline_number_methods-----------------------------
2297 // inline int Integer.numberOfLeadingZeros(int)
2298 // inline int Long.numberOfLeadingZeros(long)
2299 //
2300 // inline int Integer.numberOfTrailingZeros(int)
2301 // inline int Long.numberOfTrailingZeros(long)
2302 //
2303 // inline int Integer.bitCount(int)
2304 // inline int Long.bitCount(long)
2305 //
2306 // inline char Character.reverseBytes(char)
2307 // inline short Short.reverseBytes(short)
2308 // inline int Integer.reverseBytes(int)
2309 // inline long Long.reverseBytes(long)
2310 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2311 Node* arg = argument(0);
2312 Node* n = NULL;
2313 switch (id) {
2314 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2315 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2316 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2317 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2318 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2319 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2320 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2321 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2322 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2323 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2324 default: fatal_unexpected_iid(id); break;
2325 }
2326 set_result(_gvn.transform(n));
2327 return true;
2328 }
2329
2330 //----------------------------inline_unsafe_access----------------------------
2331
2332 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2333 // Attempt to infer a sharper value type from the offset and base type.
2334 ciKlass* sharpened_klass = NULL;
2335
2336 // See if it is an instance field, with an object type.
2337 if (alias_type->field() != NULL) {
2338 if (alias_type->field()->type()->is_klass()) {
2339 sharpened_klass = alias_type->field()->type()->as_klass();
2340 }
2341 }
2342
2343 // See if it is a narrow oop array.
2344 if (adr_type->isa_aryptr()) {
2345 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2346 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2347 if (elem_type != NULL) {
2348 sharpened_klass = elem_type->klass();
2349 }
2350 }
2351 }
2352
2353 // The sharpened class might be unloaded if there is no class loader
2354 // contraint in place.
2355 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2356 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2357
2358 #ifndef PRODUCT
2359 if (C->print_intrinsics() || C->print_inlining()) {
2360 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2361 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2362 }
2363 #endif
2364 // Sharpen the value type.
2365 return tjp;
2366 }
2367 return NULL;
2368 }
2369
2370 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2371 switch (kind) {
2372 case Relaxed:
2373 return MO_UNORDERED;
2374 case Opaque:
2375 return MO_RELAXED;
2376 case Acquire:
2377 return MO_ACQUIRE;
2378 case Release:
2379 return MO_RELEASE;
2380 case Volatile:
2381 return MO_SEQ_CST;
2382 default:
2383 ShouldNotReachHere();
2384 return 0;
2385 }
2386 }
2387
2388 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2389 if (callee()->is_static()) return false; // caller must have the capability!
2390 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2391 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2392 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2393 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2394
2395 if (is_reference_type(type)) {
2396 decorators |= ON_UNKNOWN_OOP_REF;
2397 }
2398
2399 if (unaligned) {
2400 decorators |= C2_UNALIGNED;
2401 }
2402
2403 #ifndef PRODUCT
2404 {
2405 ResourceMark rm;
2406 // Check the signatures.
2407 ciSignature* sig = callee()->signature();
2408 #ifdef ASSERT
2409 if (!is_store) {
2410 // Object getReference(Object base, int/long offset), etc.
2411 BasicType rtype = sig->return_type()->basic_type();
2412 assert(rtype == type, "getter must return the expected value");
2413 assert(sig->count() == 2, "oop getter has 2 arguments");
2414 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2415 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2416 } else {
2417 // void putReference(Object base, int/long offset, Object x), etc.
2418 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2419 assert(sig->count() == 3, "oop putter has 3 arguments");
2420 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2421 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2422 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2423 assert(vtype == type, "putter must accept the expected value");
2424 }
2425 #endif // ASSERT
2426 }
2427 #endif //PRODUCT
2428
2429 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2430
2431 Node* receiver = argument(0); // type: oop
2432
2433 // Build address expression.
2434 Node* adr;
2435 Node* heap_base_oop = top();
2436 Node* offset = top();
2437 Node* val;
2438
2439 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2440 Node* base = argument(1); // type: oop
2441 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2442 offset = argument(2); // type: long
2443 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2444 // to be plain byte offsets, which are also the same as those accepted
2445 // by oopDesc::field_addr.
2446 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2447 "fieldOffset must be byte-scaled");
2448 // 32-bit machines ignore the high half!
2449 offset = ConvL2X(offset);
2450 adr = make_unsafe_address(base, offset, is_store ? ACCESS_WRITE : ACCESS_READ, type, kind == Relaxed);
2451
2452 if (_gvn.type(base)->isa_ptr() != TypePtr::NULL_PTR) {
2453 heap_base_oop = base;
2454 } else if (type == T_OBJECT) {
2455 return false; // off-heap oop accesses are not supported
2456 }
2457
2458 // Can base be NULL? Otherwise, always on-heap access.
2459 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2460
2461 if (!can_access_non_heap) {
2462 decorators |= IN_HEAP;
2463 }
2464
2465 val = is_store ? argument(4) : NULL;
2466
2467 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2468 if (adr_type == TypePtr::NULL_PTR) {
2469 return false; // off-heap access with zero address
2470 }
2471
2472 // Try to categorize the address.
2473 Compile::AliasType* alias_type = C->alias_type(adr_type);
2474 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2475
2476 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2477 alias_type->adr_type() == TypeAryPtr::RANGE) {
2478 return false; // not supported
2479 }
2480
2481 bool mismatched = false;
2482 BasicType bt = alias_type->basic_type();
2483 if (bt != T_ILLEGAL) {
2484 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2485 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2486 // Alias type doesn't differentiate between byte[] and boolean[]).
2487 // Use address type to get the element type.
2488 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2489 }
2490 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2491 // accessing an array field with getReference is not a mismatch
2492 bt = T_OBJECT;
2493 }
2494 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2495 // Don't intrinsify mismatched object accesses
2496 return false;
2497 }
2498 mismatched = (bt != type);
2499 } else if (alias_type->adr_type()->isa_oopptr()) {
2500 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2501 }
2502
2503 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2504
2505 if (mismatched) {
2506 decorators |= C2_MISMATCHED;
2507 }
2508
2509 // First guess at the value type.
2510 const Type *value_type = Type::get_const_basic_type(type);
2511
2512 // Figure out the memory ordering.
2513 decorators |= mo_decorator_for_access_kind(kind);
2514
2515 if (!is_store && type == T_OBJECT) {
2516 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2517 if (tjp != NULL) {
2518 value_type = tjp;
2519 }
2520 }
2521
2522 receiver = null_check(receiver);
2523 if (stopped()) {
2524 return true;
2525 }
2526 // Heap pointers get a null-check from the interpreter,
2527 // as a courtesy. However, this is not guaranteed by Unsafe,
2528 // and it is not possible to fully distinguish unintended nulls
2529 // from intended ones in this API.
2530
2531 if (!is_store) {
2532 Node* p = NULL;
2533 // Try to constant fold a load from a constant field
2534 ciField* field = alias_type->field();
2535 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2536 // final or stable field
2537 p = make_constant_from_field(field, heap_base_oop);
2538 }
2539
2540 if (p == NULL) { // Could not constant fold the load
2541 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2542 // Normalize the value returned by getBoolean in the following cases
2543 if (type == T_BOOLEAN &&
2544 (mismatched ||
2545 heap_base_oop == top() || // - heap_base_oop is NULL or
2546 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2547 // and the unsafe access is made to large offset
2548 // (i.e., larger than the maximum offset necessary for any
2549 // field access)
2550 ) {
2551 IdealKit ideal = IdealKit(this);
2552 #define __ ideal.
2553 IdealVariable normalized_result(ideal);
2554 __ declarations_done();
2555 __ set(normalized_result, p);
2556 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2557 __ set(normalized_result, ideal.ConI(1));
2558 ideal.end_if();
2559 final_sync(ideal);
2560 p = __ value(normalized_result);
2561 #undef __
2562 }
2563 }
2564 if (type == T_ADDRESS) {
2565 p = gvn().transform(new CastP2XNode(NULL, p));
2566 p = ConvX2UL(p);
2567 }
2568 // The load node has the control of the preceding MemBarCPUOrder. All
2569 // following nodes will have the control of the MemBarCPUOrder inserted at
2570 // the end of this method. So, pushing the load onto the stack at a later
2571 // point is fine.
2572 set_result(p);
2573 } else {
2574 if (bt == T_ADDRESS) {
2575 // Repackage the long as a pointer.
2576 val = ConvL2X(val);
2577 val = gvn().transform(new CastX2PNode(val));
2578 }
2579 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2580 }
2581
2582 return true;
2583 }
2584
2585 //----------------------------inline_unsafe_load_store----------------------------
2586 // This method serves a couple of different customers (depending on LoadStoreKind):
2587 //
2588 // LS_cmp_swap:
2589 //
2590 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2591 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2592 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2593 //
2594 // LS_cmp_swap_weak:
2595 //
2596 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2597 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2598 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2599 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2600 //
2601 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2602 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2603 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2604 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2605 //
2606 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2607 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2608 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2609 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2610 //
2611 // LS_cmp_exchange:
2612 //
2613 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2614 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2615 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2616 //
2617 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2618 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2619 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2620 //
2621 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2622 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2623 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2624 //
2625 // LS_get_add:
2626 //
2627 // int getAndAddInt( Object o, long offset, int delta)
2628 // long getAndAddLong(Object o, long offset, long delta)
2629 //
2630 // LS_get_set:
2631 //
2632 // int getAndSet(Object o, long offset, int newValue)
2633 // long getAndSet(Object o, long offset, long newValue)
2634 // Object getAndSet(Object o, long offset, Object newValue)
2635 //
2636 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2637 // This basic scheme here is the same as inline_unsafe_access, but
2638 // differs in enough details that combining them would make the code
2639 // overly confusing. (This is a true fact! I originally combined
2640 // them, but even I was confused by it!) As much code/comments as
2641 // possible are retained from inline_unsafe_access though to make
2642 // the correspondences clearer. - dl
2643
2644 if (callee()->is_static()) return false; // caller must have the capability!
2645
2646 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2647 decorators |= mo_decorator_for_access_kind(access_kind);
2648
2649 #ifndef PRODUCT
2650 BasicType rtype;
2651 {
2652 ResourceMark rm;
2653 // Check the signatures.
2654 ciSignature* sig = callee()->signature();
2655 rtype = sig->return_type()->basic_type();
2656 switch(kind) {
2657 case LS_get_add:
2658 case LS_get_set: {
2659 // Check the signatures.
2660 #ifdef ASSERT
2661 assert(rtype == type, "get and set must return the expected type");
2662 assert(sig->count() == 3, "get and set has 3 arguments");
2663 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2664 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2665 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2666 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2667 #endif // ASSERT
2668 break;
2669 }
2670 case LS_cmp_swap:
2671 case LS_cmp_swap_weak: {
2672 // Check the signatures.
2673 #ifdef ASSERT
2674 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2675 assert(sig->count() == 4, "CAS has 4 arguments");
2676 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2677 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2678 #endif // ASSERT
2679 break;
2680 }
2681 case LS_cmp_exchange: {
2682 // Check the signatures.
2683 #ifdef ASSERT
2684 assert(rtype == type, "CAS must return the expected type");
2685 assert(sig->count() == 4, "CAS has 4 arguments");
2686 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2687 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2688 #endif // ASSERT
2689 break;
2690 }
2691 default:
2692 ShouldNotReachHere();
2693 }
2694 }
2695 #endif //PRODUCT
2696
2697 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2698
2699 // Get arguments:
2700 Node* receiver = NULL;
2701 Node* base = NULL;
2702 Node* offset = NULL;
2703 Node* oldval = NULL;
2704 Node* newval = NULL;
2705 switch(kind) {
2706 case LS_cmp_swap:
2707 case LS_cmp_swap_weak:
2708 case LS_cmp_exchange: {
2709 const bool two_slot_type = type2size[type] == 2;
2710 receiver = argument(0); // type: oop
2711 base = argument(1); // type: oop
2712 offset = argument(2); // type: long
2713 oldval = argument(4); // type: oop, int, or long
2714 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2715 break;
2716 }
2717 case LS_get_add:
2718 case LS_get_set: {
2719 receiver = argument(0); // type: oop
2720 base = argument(1); // type: oop
2721 offset = argument(2); // type: long
2722 oldval = NULL;
2723 newval = argument(4); // type: oop, int, or long
2724 break;
2725 }
2726 default:
2727 ShouldNotReachHere();
2728 }
2729
2730 // Build field offset expression.
2731 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2732 // to be plain byte offsets, which are also the same as those accepted
2733 // by oopDesc::field_addr.
2734 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2735 // 32-bit machines ignore the high half of long offsets
2736 offset = ConvL2X(offset);
2737 Node* adr = make_unsafe_address(base, offset, ACCESS_WRITE | ACCESS_READ, type, false);
2738 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2739
2740 Compile::AliasType* alias_type = C->alias_type(adr_type);
2741 BasicType bt = alias_type->basic_type();
2742 if (bt != T_ILLEGAL &&
2743 (is_reference_type(bt) != (type == T_OBJECT))) {
2744 // Don't intrinsify mismatched object accesses.
2745 return false;
2746 }
2747
2748 // For CAS, unlike inline_unsafe_access, there seems no point in
2749 // trying to refine types. Just use the coarse types here.
2750 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2751 const Type *value_type = Type::get_const_basic_type(type);
2752
2753 switch (kind) {
2754 case LS_get_set:
2755 case LS_cmp_exchange: {
2756 if (type == T_OBJECT) {
2757 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2758 if (tjp != NULL) {
2759 value_type = tjp;
2760 }
2761 }
2762 break;
2763 }
2764 case LS_cmp_swap:
2765 case LS_cmp_swap_weak:
2766 case LS_get_add:
2767 break;
2768 default:
2769 ShouldNotReachHere();
2770 }
2771
2772 // Null check receiver.
2773 receiver = null_check(receiver);
2774 if (stopped()) {
2775 return true;
2776 }
2777
2778 int alias_idx = C->get_alias_index(adr_type);
2779
2780 if (is_reference_type(type)) {
2781 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2782
2783 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2784 // could be delayed during Parse (for example, in adjust_map_after_if()).
2785 // Execute transformation here to avoid barrier generation in such case.
2786 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2787 newval = _gvn.makecon(TypePtr::NULL_PTR);
2788
2789 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2790 // Refine the value to a null constant, when it is known to be null
2791 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2792 }
2793 }
2794
2795 Node* result = NULL;
2796 switch (kind) {
2797 case LS_cmp_exchange: {
2798 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2799 oldval, newval, value_type, type, decorators);
2800 break;
2801 }
2802 case LS_cmp_swap_weak:
2803 decorators |= C2_WEAK_CMPXCHG;
2804 case LS_cmp_swap: {
2805 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2806 oldval, newval, value_type, type, decorators);
2807 break;
2808 }
2809 case LS_get_set: {
2810 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2811 newval, value_type, type, decorators);
2812 break;
2813 }
2814 case LS_get_add: {
2815 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2816 newval, value_type, type, decorators);
2817 break;
2818 }
2819 default:
2820 ShouldNotReachHere();
2821 }
2822
2823 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2824 set_result(result);
2825 return true;
2826 }
2827
2828 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2829 // Regardless of form, don't allow previous ld/st to move down,
2830 // then issue acquire, release, or volatile mem_bar.
2831 insert_mem_bar(Op_MemBarCPUOrder);
2832 switch(id) {
2833 case vmIntrinsics::_loadFence:
2834 insert_mem_bar(Op_LoadFence);
2835 return true;
2836 case vmIntrinsics::_storeFence:
2837 insert_mem_bar(Op_StoreFence);
2838 return true;
2839 case vmIntrinsics::_fullFence:
2840 insert_mem_bar(Op_MemBarVolatile);
2841 return true;
2842 default:
2843 fatal_unexpected_iid(id);
2844 return false;
2845 }
2846 }
2847
2848 bool LibraryCallKit::inline_onspinwait() {
2849 insert_mem_bar(Op_OnSpinWait);
2850 return true;
2851 }
2852
2853 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2854 if (!kls->is_Con()) {
2855 return true;
2856 }
2857 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
2858 if (klsptr == NULL) {
2859 return true;
2860 }
2861 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
2862 // don't need a guard for a klass that is already initialized
2863 return !ik->is_initialized();
2864 }
2865
2866 //----------------------------inline_unsafe_writeback0-------------------------
2867 // public native void Unsafe.writeback0(long address)
2868 bool LibraryCallKit::inline_unsafe_writeback0() {
2869 if (!Matcher::has_match_rule(Op_CacheWB)) {
2870 return false;
2871 }
2872 #ifndef PRODUCT
2873 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2874 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2875 ciSignature* sig = callee()->signature();
2876 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2877 #endif
2878 null_check_receiver(); // null-check, then ignore
2879 Node *addr = argument(1);
2880 addr = new CastX2PNode(addr);
2881 addr = _gvn.transform(addr);
2882 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2883 flush = _gvn.transform(flush);
2884 set_memory(flush, TypeRawPtr::BOTTOM);
2885 return true;
2886 }
2887
2888 //----------------------------inline_unsafe_writeback0-------------------------
2889 // public native void Unsafe.writeback0(long address)
2890 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2891 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2892 return false;
2893 }
2894 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2895 return false;
2896 }
2897 #ifndef PRODUCT
2898 assert(Matcher::has_match_rule(Op_CacheWB),
2899 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2900 : "found match rule for CacheWBPostSync but not CacheWB"));
2901
2902 #endif
2903 null_check_receiver(); // null-check, then ignore
2904 Node *sync;
2905 if (is_pre) {
2906 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2907 } else {
2908 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2909 }
2910 sync = _gvn.transform(sync);
2911 set_memory(sync, TypeRawPtr::BOTTOM);
2912 return true;
2913 }
2914
2915 //----------------------------inline_unsafe_allocate---------------------------
2916 // public native Object Unsafe.allocateInstance(Class<?> cls);
2917 bool LibraryCallKit::inline_unsafe_allocate() {
2918 if (callee()->is_static()) return false; // caller must have the capability!
2919
2920 null_check_receiver(); // null-check, then ignore
2921 Node* cls = null_check(argument(1));
2922 if (stopped()) return true;
2923
2924 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2925 kls = null_check(kls);
2926 if (stopped()) return true; // argument was like int.class
2927
2928 Node* test = NULL;
2929 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2930 // Note: The argument might still be an illegal value like
2931 // Serializable.class or Object[].class. The runtime will handle it.
2932 // But we must make an explicit check for initialization.
2933 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2934 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2935 // can generate code to load it as unsigned byte.
2936 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2937 Node* bits = intcon(InstanceKlass::fully_initialized);
2938 test = _gvn.transform(new SubINode(inst, bits));
2939 // The 'test' is non-zero if we need to take a slow path.
2940 }
2941
2942 Node* obj = new_instance(kls, test);
2943 set_result(obj);
2944 return true;
2945 }
2946
2947 //------------------------inline_native_time_funcs--------------
2948 // inline code for System.currentTimeMillis() and System.nanoTime()
2949 // these have the same type and signature
2950 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2951 const TypeFunc* tf = OptoRuntime::void_long_Type();
2952 const TypePtr* no_memory_effects = NULL;
2953 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2954 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2955 #ifdef ASSERT
2956 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2957 assert(value_top == top(), "second value must be top");
2958 #endif
2959 set_result(value);
2960 return true;
2961 }
2962
2963 #ifdef JFR_HAVE_INTRINSICS
2964
2965 /*
2966 * oop -> myklass
2967 * myklass->trace_id |= USED
2968 * return myklass->trace_id & ~0x3
2969 */
2970 bool LibraryCallKit::inline_native_classID() {
2971 Node* cls = null_check(argument(0), T_OBJECT);
2972 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2973 kls = null_check(kls, T_OBJECT);
2974
2975 ByteSize offset = KLASS_TRACE_ID_OFFSET;
2976 Node* insp = basic_plus_adr(kls, in_bytes(offset));
2977 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
2978
2979 Node* clsused = longcon(0x01l); // set the class bit
2980 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
2981 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
2982 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
2983
2984 #ifdef TRACE_ID_META_BITS
2985 Node* mbits = longcon(~TRACE_ID_META_BITS);
2986 tvalue = _gvn.transform(new AndLNode(tvalue, mbits));
2987 #endif
2988 #ifdef TRACE_ID_SHIFT
2989 Node* cbits = intcon(TRACE_ID_SHIFT);
2990 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits));
2991 #endif
2992
2993 set_result(tvalue);
2994 return true;
2995
2996 }
2997
2998 bool LibraryCallKit::inline_native_getEventWriter() {
2999 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3000
3001 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
3002 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3003
3004 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3005
3006 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
3007 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
3008
3009 IfNode* iff_jobj_null =
3010 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
3011
3012 enum { _normal_path = 1,
3013 _null_path = 2,
3014 PATH_LIMIT };
3015
3016 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3017 PhiNode* result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3018
3019 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
3020 result_rgn->init_req(_null_path, jobj_is_null);
3021 result_val->init_req(_null_path, null());
3022
3023 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
3024 set_control(jobj_is_not_null);
3025 Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT,
3026 IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3027 result_rgn->init_req(_normal_path, control());
3028 result_val->init_req(_normal_path, res);
3029
3030 set_result(result_rgn, result_val);
3031
3032 return true;
3033 }
3034
3035 #endif // JFR_HAVE_INTRINSICS
3036
3037 //------------------------inline_native_currentThread------------------
3038 bool LibraryCallKit::inline_native_currentThread() {
3039 Node* junk = NULL;
3040 set_result(generate_current_thread(junk));
3041 return true;
3042 }
3043
3044 //------------------------inline_native_isInterrupted------------------
3045 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
3046 bool LibraryCallKit::inline_native_isInterrupted() {
3047 // Add a fast path to t.isInterrupted(clear_int):
3048 // (t == Thread.current() &&
3049 // (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
3050 // ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
3051 // So, in the common case that the interrupt bit is false,
3052 // we avoid making a call into the VM. Even if the interrupt bit
3053 // is true, if the clear_int argument is false, we avoid the VM call.
3054 // However, if the receiver is not currentThread, we must call the VM,
3055 // because there must be some locking done around the operation.
3056
3057 // We only go to the fast case code if we pass two guards.
3058 // Paths which do not pass are accumulated in the slow_region.
3059
3060 enum {
3061 no_int_result_path = 1, // t == Thread.current() && !TLS._osthread._interrupted
3062 no_clear_result_path = 2, // t == Thread.current() && TLS._osthread._interrupted && !clear_int
3063 slow_result_path = 3, // slow path: t.isInterrupted(clear_int)
3064 PATH_LIMIT
3065 };
3066
3067 // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
3068 // out of the function.
3069 insert_mem_bar(Op_MemBarCPUOrder);
3070
3071 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3072 PhiNode* result_val = new PhiNode(result_rgn, TypeInt::BOOL);
3073
3074 RegionNode* slow_region = new RegionNode(1);
3075 record_for_igvn(slow_region);
3076
3077 // (a) Receiving thread must be the current thread.
3078 Node* rec_thr = argument(0);
3079 Node* tls_ptr = NULL;
3080 Node* cur_thr = generate_current_thread(tls_ptr);
3081
3082 // Resolve oops to stable for CmpP below.
3083 cur_thr = access_resolve(cur_thr, 0);
3084 rec_thr = access_resolve(rec_thr, 0);
3085
3086 Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
3087 Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
3088
3089 generate_slow_guard(bol_thr, slow_region);
3090
3091 // (b) Interrupt bit on TLS must be false.
3092 Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
3093 Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3094 p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
3095
3096 // Set the control input on the field _interrupted read to prevent it floating up.
3097 Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
3098 Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
3099 Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
3100
3101 IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
3102
3103 // First fast path: if (!TLS._interrupted) return false;
3104 Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
3105 result_rgn->init_req(no_int_result_path, false_bit);
3106 result_val->init_req(no_int_result_path, intcon(0));
3107
3108 // drop through to next case
3109 set_control( _gvn.transform(new IfTrueNode(iff_bit)));
3110
3111 #ifndef _WINDOWS
3112 // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
3113 Node* clr_arg = argument(1);
3114 Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
3115 Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
3116 IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
3117
3118 // Second fast path: ... else if (!clear_int) return true;
3119 Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
3120 result_rgn->init_req(no_clear_result_path, false_arg);
3121 result_val->init_req(no_clear_result_path, intcon(1));
3122
3123 // drop through to next case
3124 set_control( _gvn.transform(new IfTrueNode(iff_arg)));
3125 #else
3126 // To return true on Windows you must read the _interrupted field
3127 // and check the event state i.e. take the slow path.
3128 #endif // _WINDOWS
3129
3130 // (d) Otherwise, go to the slow path.
3131 slow_region->add_req(control());
3132 set_control( _gvn.transform(slow_region));
3133
3134 if (stopped()) {
3135 // There is no slow path.
3136 result_rgn->init_req(slow_result_path, top());
3137 result_val->init_req(slow_result_path, top());
3138 } else {
3139 // non-virtual because it is a private non-static
3140 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
3141
3142 Node* slow_val = set_results_for_java_call(slow_call);
3143 // this->control() comes from set_results_for_java_call
3144
3145 Node* fast_io = slow_call->in(TypeFunc::I_O);
3146 Node* fast_mem = slow_call->in(TypeFunc::Memory);
3147
3148 // These two phis are pre-filled with copies of of the fast IO and Memory
3149 PhiNode* result_mem = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
3150 PhiNode* result_io = PhiNode::make(result_rgn, fast_io, Type::ABIO);
3151
3152 result_rgn->init_req(slow_result_path, control());
3153 result_io ->init_req(slow_result_path, i_o());
3154 result_mem->init_req(slow_result_path, reset_memory());
3155 result_val->init_req(slow_result_path, slow_val);
3156
3157 set_all_memory(_gvn.transform(result_mem));
3158 set_i_o( _gvn.transform(result_io));
3159 }
3160
3161 C->set_has_split_ifs(true); // Has chance for split-if optimization
3162 set_result(result_rgn, result_val);
3163 return true;
3164 }
3165
3166 //---------------------------load_mirror_from_klass----------------------------
3167 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3168 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3169 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3170 Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3171 // mirror = ((OopHandle)mirror)->resolve();
3172 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3173 }
3174
3175 //-----------------------load_klass_from_mirror_common-------------------------
3176 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3177 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3178 // and branch to the given path on the region.
3179 // If never_see_null, take an uncommon trap on null, so we can optimistically
3180 // compile for the non-null case.
3181 // If the region is NULL, force never_see_null = true.
3182 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3183 bool never_see_null,
3184 RegionNode* region,
3185 int null_path,
3186 int offset) {
3187 if (region == NULL) never_see_null = true;
3188 Node* p = basic_plus_adr(mirror, offset);
3189 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3190 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3191 Node* null_ctl = top();
3192 kls = null_check_oop(kls, &null_ctl, never_see_null);
3193 if (region != NULL) {
3194 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3195 region->init_req(null_path, null_ctl);
3196 } else {
3197 assert(null_ctl == top(), "no loose ends");
3198 }
3199 return kls;
3200 }
3201
3202 //--------------------(inline_native_Class_query helpers)---------------------
3203 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3204 // Fall through if (mods & mask) == bits, take the guard otherwise.
3205 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3206 // Branch around if the given klass has the given modifier bit set.
3207 // Like generate_guard, adds a new path onto the region.
3208 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3209 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3210 Node* mask = intcon(modifier_mask);
3211 Node* bits = intcon(modifier_bits);
3212 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3213 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3214 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3215 return generate_fair_guard(bol, region);
3216 }
3217 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3218 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3219 }
3220
3221 //-------------------------inline_native_Class_query-------------------
3222 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3223 const Type* return_type = TypeInt::BOOL;
3224 Node* prim_return_value = top(); // what happens if it's a primitive class?
3225 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3226 bool expect_prim = false; // most of these guys expect to work on refs
3227
3228 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3229
3230 Node* mirror = argument(0);
3231 Node* obj = top();
3232
3233 switch (id) {
3234 case vmIntrinsics::_isInstance:
3235 // nothing is an instance of a primitive type
3236 prim_return_value = intcon(0);
3237 obj = argument(1);
3238 break;
3239 case vmIntrinsics::_getModifiers:
3240 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3241 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3242 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3243 break;
3244 case vmIntrinsics::_isInterface:
3245 prim_return_value = intcon(0);
3246 break;
3247 case vmIntrinsics::_isArray:
3248 prim_return_value = intcon(0);
3249 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3250 break;
3251 case vmIntrinsics::_isPrimitive:
3252 prim_return_value = intcon(1);
3253 expect_prim = true; // obviously
3254 break;
3255 case vmIntrinsics::_getSuperclass:
3256 prim_return_value = null();
3257 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3258 break;
3259 case vmIntrinsics::_getClassAccessFlags:
3260 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3261 return_type = TypeInt::INT; // not bool! 6297094
3262 break;
3263 default:
3264 fatal_unexpected_iid(id);
3265 break;
3266 }
3267
3268 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3269 if (mirror_con == NULL) return false; // cannot happen?
3270
3271 #ifndef PRODUCT
3272 if (C->print_intrinsics() || C->print_inlining()) {
3273 ciType* k = mirror_con->java_mirror_type();
3274 if (k) {
3275 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3276 k->print_name();
3277 tty->cr();
3278 }
3279 }
3280 #endif
3281
3282 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3283 RegionNode* region = new RegionNode(PATH_LIMIT);
3284 record_for_igvn(region);
3285 PhiNode* phi = new PhiNode(region, return_type);
3286
3287 // The mirror will never be null of Reflection.getClassAccessFlags, however
3288 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3289 // if it is. See bug 4774291.
3290
3291 // For Reflection.getClassAccessFlags(), the null check occurs in
3292 // the wrong place; see inline_unsafe_access(), above, for a similar
3293 // situation.
3294 mirror = null_check(mirror);
3295 // If mirror or obj is dead, only null-path is taken.
3296 if (stopped()) return true;
3297
3298 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3299
3300 // Now load the mirror's klass metaobject, and null-check it.
3301 // Side-effects region with the control path if the klass is null.
3302 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3303 // If kls is null, we have a primitive mirror.
3304 phi->init_req(_prim_path, prim_return_value);
3305 if (stopped()) { set_result(region, phi); return true; }
3306 bool safe_for_replace = (region->in(_prim_path) == top());
3307
3308 Node* p; // handy temp
3309 Node* null_ctl;
3310
3311 // Now that we have the non-null klass, we can perform the real query.
3312 // For constant classes, the query will constant-fold in LoadNode::Value.
3313 Node* query_value = top();
3314 switch (id) {
3315 case vmIntrinsics::_isInstance:
3316 // nothing is an instance of a primitive type
3317 query_value = gen_instanceof(obj, kls, safe_for_replace);
3318 break;
3319
3320 case vmIntrinsics::_getModifiers:
3321 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3322 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3323 break;
3324
3325 case vmIntrinsics::_isInterface:
3326 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3327 if (generate_interface_guard(kls, region) != NULL)
3328 // A guard was added. If the guard is taken, it was an interface.
3329 phi->add_req(intcon(1));
3330 // If we fall through, it's a plain class.
3331 query_value = intcon(0);
3332 break;
3333
3334 case vmIntrinsics::_isArray:
3335 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3336 if (generate_array_guard(kls, region) != NULL)
3337 // A guard was added. If the guard is taken, it was an array.
3338 phi->add_req(intcon(1));
3339 // If we fall through, it's a plain class.
3340 query_value = intcon(0);
3341 break;
3342
3343 case vmIntrinsics::_isPrimitive:
3344 query_value = intcon(0); // "normal" path produces false
3345 break;
3346
3347 case vmIntrinsics::_getSuperclass:
3348 // The rules here are somewhat unfortunate, but we can still do better
3349 // with random logic than with a JNI call.
3350 // Interfaces store null or Object as _super, but must report null.
3351 // Arrays store an intermediate super as _super, but must report Object.
3352 // Other types can report the actual _super.
3353 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3354 if (generate_interface_guard(kls, region) != NULL)
3355 // A guard was added. If the guard is taken, it was an interface.
3356 phi->add_req(null());
3357 if (generate_array_guard(kls, region) != NULL)
3358 // A guard was added. If the guard is taken, it was an array.
3359 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3360 // If we fall through, it's a plain class. Get its _super.
3361 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3362 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3363 null_ctl = top();
3364 kls = null_check_oop(kls, &null_ctl);
3365 if (null_ctl != top()) {
3366 // If the guard is taken, Object.superClass is null (both klass and mirror).
3367 region->add_req(null_ctl);
3368 phi ->add_req(null());
3369 }
3370 if (!stopped()) {
3371 query_value = load_mirror_from_klass(kls);
3372 }
3373 break;
3374
3375 case vmIntrinsics::_getClassAccessFlags:
3376 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3377 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3378 break;
3379
3380 default:
3381 fatal_unexpected_iid(id);
3382 break;
3383 }
3384
3385 // Fall-through is the normal case of a query to a real class.
3386 phi->init_req(1, query_value);
3387 region->init_req(1, control());
3388
3389 C->set_has_split_ifs(true); // Has chance for split-if optimization
3390 set_result(region, phi);
3391 return true;
3392 }
3393
3394 //-------------------------inline_Class_cast-------------------
3395 bool LibraryCallKit::inline_Class_cast() {
3396 Node* mirror = argument(0); // Class
3397 Node* obj = argument(1);
3398 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3399 if (mirror_con == NULL) {
3400 return false; // dead path (mirror->is_top()).
3401 }
3402 if (obj == NULL || obj->is_top()) {
3403 return false; // dead path
3404 }
3405 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3406
3407 // First, see if Class.cast() can be folded statically.
3408 // java_mirror_type() returns non-null for compile-time Class constants.
3409 ciType* tm = mirror_con->java_mirror_type();
3410 if (tm != NULL && tm->is_klass() &&
3411 tp != NULL && tp->klass() != NULL) {
3412 if (!tp->klass()->is_loaded()) {
3413 // Don't use intrinsic when class is not loaded.
3414 return false;
3415 } else {
3416 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3417 if (static_res == Compile::SSC_always_true) {
3418 // isInstance() is true - fold the code.
3419 set_result(obj);
3420 return true;
3421 } else if (static_res == Compile::SSC_always_false) {
3422 // Don't use intrinsic, have to throw ClassCastException.
3423 // If the reference is null, the non-intrinsic bytecode will
3424 // be optimized appropriately.
3425 return false;
3426 }
3427 }
3428 }
3429
3430 // Bailout intrinsic and do normal inlining if exception path is frequent.
3431 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3432 return false;
3433 }
3434
3435 // Generate dynamic checks.
3436 // Class.cast() is java implementation of _checkcast bytecode.
3437 // Do checkcast (Parse::do_checkcast()) optimizations here.
3438
3439 mirror = null_check(mirror);
3440 // If mirror is dead, only null-path is taken.
3441 if (stopped()) {
3442 return true;
3443 }
3444
3445 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3446 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3447 RegionNode* region = new RegionNode(PATH_LIMIT);
3448 record_for_igvn(region);
3449
3450 // Now load the mirror's klass metaobject, and null-check it.
3451 // If kls is null, we have a primitive mirror and
3452 // nothing is an instance of a primitive type.
3453 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3454
3455 Node* res = top();
3456 if (!stopped()) {
3457 Node* bad_type_ctrl = top();
3458 // Do checkcast optimizations.
3459 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3460 region->init_req(_bad_type_path, bad_type_ctrl);
3461 }
3462 if (region->in(_prim_path) != top() ||
3463 region->in(_bad_type_path) != top()) {
3464 // Let Interpreter throw ClassCastException.
3465 PreserveJVMState pjvms(this);
3466 set_control(_gvn.transform(region));
3467 uncommon_trap(Deoptimization::Reason_intrinsic,
3468 Deoptimization::Action_maybe_recompile);
3469 }
3470 if (!stopped()) {
3471 set_result(res);
3472 }
3473 return true;
3474 }
3475
3476
3477 //--------------------------inline_native_subtype_check------------------------
3478 // This intrinsic takes the JNI calls out of the heart of
3479 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3480 bool LibraryCallKit::inline_native_subtype_check() {
3481 // Pull both arguments off the stack.
3482 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3483 args[0] = argument(0);
3484 args[1] = argument(1);
3485 Node* klasses[2]; // corresponding Klasses: superk, subk
3486 klasses[0] = klasses[1] = top();
3487
3488 enum {
3489 // A full decision tree on {superc is prim, subc is prim}:
3490 _prim_0_path = 1, // {P,N} => false
3491 // {P,P} & superc!=subc => false
3492 _prim_same_path, // {P,P} & superc==subc => true
3493 _prim_1_path, // {N,P} => false
3494 _ref_subtype_path, // {N,N} & subtype check wins => true
3495 _both_ref_path, // {N,N} & subtype check loses => false
3496 PATH_LIMIT
3497 };
3498
3499 RegionNode* region = new RegionNode(PATH_LIMIT);
3500 Node* phi = new PhiNode(region, TypeInt::BOOL);
3501 record_for_igvn(region);
3502
3503 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3504 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3505 int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3506
3507 // First null-check both mirrors and load each mirror's klass metaobject.
3508 int which_arg;
3509 for (which_arg = 0; which_arg <= 1; which_arg++) {
3510 Node* arg = args[which_arg];
3511 arg = null_check(arg);
3512 if (stopped()) break;
3513 args[which_arg] = arg;
3514
3515 Node* p = basic_plus_adr(arg, class_klass_offset);
3516 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3517 klasses[which_arg] = _gvn.transform(kls);
3518 }
3519
3520 // Resolve oops to stable for CmpP below.
3521 args[0] = access_resolve(args[0], 0);
3522 args[1] = access_resolve(args[1], 0);
3523
3524 // Having loaded both klasses, test each for null.
3525 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3526 for (which_arg = 0; which_arg <= 1; which_arg++) {
3527 Node* kls = klasses[which_arg];
3528 Node* null_ctl = top();
3529 kls = null_check_oop(kls, &null_ctl, never_see_null);
3530 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3531 region->init_req(prim_path, null_ctl);
3532 if (stopped()) break;
3533 klasses[which_arg] = kls;
3534 }
3535
3536 if (!stopped()) {
3537 // now we have two reference types, in klasses[0..1]
3538 Node* subk = klasses[1]; // the argument to isAssignableFrom
3539 Node* superk = klasses[0]; // the receiver
3540 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3541 // now we have a successful reference subtype check
3542 region->set_req(_ref_subtype_path, control());
3543 }
3544
3545 // If both operands are primitive (both klasses null), then
3546 // we must return true when they are identical primitives.
3547 // It is convenient to test this after the first null klass check.
3548 set_control(region->in(_prim_0_path)); // go back to first null check
3549 if (!stopped()) {
3550 // Since superc is primitive, make a guard for the superc==subc case.
3551 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3552 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3553 generate_guard(bol_eq, region, PROB_FAIR);
3554 if (region->req() == PATH_LIMIT+1) {
3555 // A guard was added. If the added guard is taken, superc==subc.
3556 region->swap_edges(PATH_LIMIT, _prim_same_path);
3557 region->del_req(PATH_LIMIT);
3558 }
3559 region->set_req(_prim_0_path, control()); // Not equal after all.
3560 }
3561
3562 // these are the only paths that produce 'true':
3563 phi->set_req(_prim_same_path, intcon(1));
3564 phi->set_req(_ref_subtype_path, intcon(1));
3565
3566 // pull together the cases:
3567 assert(region->req() == PATH_LIMIT, "sane region");
3568 for (uint i = 1; i < region->req(); i++) {
3569 Node* ctl = region->in(i);
3570 if (ctl == NULL || ctl == top()) {
3571 region->set_req(i, top());
3572 phi ->set_req(i, top());
3573 } else if (phi->in(i) == NULL) {
3574 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3575 }
3576 }
3577
3578 set_control(_gvn.transform(region));
3579 set_result(_gvn.transform(phi));
3580 return true;
3581 }
3582
3583 //---------------------generate_array_guard_common------------------------
3584 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3585 bool obj_array, bool not_array) {
3586
3587 if (stopped()) {
3588 return NULL;
3589 }
3590
3591 // If obj_array/non_array==false/false:
3592 // Branch around if the given klass is in fact an array (either obj or prim).
3593 // If obj_array/non_array==false/true:
3594 // Branch around if the given klass is not an array klass of any kind.
3595 // If obj_array/non_array==true/true:
3596 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3597 // If obj_array/non_array==true/false:
3598 // Branch around if the kls is an oop array (Object[] or subtype)
3599 //
3600 // Like generate_guard, adds a new path onto the region.
3601 jint layout_con = 0;
3602 Node* layout_val = get_layout_helper(kls, layout_con);
3603 if (layout_val == NULL) {
3604 bool query = (obj_array
3605 ? Klass::layout_helper_is_objArray(layout_con)
3606 : Klass::layout_helper_is_array(layout_con));
3607 if (query == not_array) {
3608 return NULL; // never a branch
3609 } else { // always a branch
3610 Node* always_branch = control();
3611 if (region != NULL)
3612 region->add_req(always_branch);
3613 set_control(top());
3614 return always_branch;
3615 }
3616 }
3617 // Now test the correct condition.
3618 jint nval = (obj_array
3619 ? (jint)(Klass::_lh_array_tag_type_value
3620 << Klass::_lh_array_tag_shift)
3621 : Klass::_lh_neutral_value);
3622 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3623 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3624 // invert the test if we are looking for a non-array
3625 if (not_array) btest = BoolTest(btest).negate();
3626 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3627 return generate_fair_guard(bol, region);
3628 }
3629
3630
3631 //-----------------------inline_native_newArray--------------------------
3632 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3633 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3634 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3635 Node* mirror;
3636 Node* count_val;
3637 if (uninitialized) {
3638 mirror = argument(1);
3639 count_val = argument(2);
3640 } else {
3641 mirror = argument(0);
3642 count_val = argument(1);
3643 }
3644
3645 mirror = null_check(mirror);
3646 // If mirror or obj is dead, only null-path is taken.
3647 if (stopped()) return true;
3648
3649 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3650 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3651 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3652 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3653 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3654
3655 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3656 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3657 result_reg, _slow_path);
3658 Node* normal_ctl = control();
3659 Node* no_array_ctl = result_reg->in(_slow_path);
3660
3661 // Generate code for the slow case. We make a call to newArray().
3662 set_control(no_array_ctl);
3663 if (!stopped()) {
3664 // Either the input type is void.class, or else the
3665 // array klass has not yet been cached. Either the
3666 // ensuing call will throw an exception, or else it
3667 // will cache the array klass for next time.
3668 PreserveJVMState pjvms(this);
3669 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3670 Node* slow_result = set_results_for_java_call(slow_call);
3671 // this->control() comes from set_results_for_java_call
3672 result_reg->set_req(_slow_path, control());
3673 result_val->set_req(_slow_path, slow_result);
3674 result_io ->set_req(_slow_path, i_o());
3675 result_mem->set_req(_slow_path, reset_memory());
3676 }
3677
3678 set_control(normal_ctl);
3679 if (!stopped()) {
3680 // Normal case: The array type has been cached in the java.lang.Class.
3681 // The following call works fine even if the array type is polymorphic.
3682 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3683 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3684 result_reg->init_req(_normal_path, control());
3685 result_val->init_req(_normal_path, obj);
3686 result_io ->init_req(_normal_path, i_o());
3687 result_mem->init_req(_normal_path, reset_memory());
3688
3689 if (uninitialized) {
3690 // Mark the allocation so that zeroing is skipped
3691 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3692 alloc->maybe_set_complete(&_gvn);
3693 }
3694 }
3695
3696 // Return the combined state.
3697 set_i_o( _gvn.transform(result_io) );
3698 set_all_memory( _gvn.transform(result_mem));
3699
3700 C->set_has_split_ifs(true); // Has chance for split-if optimization
3701 set_result(result_reg, result_val);
3702 return true;
3703 }
3704
3705 //----------------------inline_native_getLength--------------------------
3706 // public static native int java.lang.reflect.Array.getLength(Object array);
3707 bool LibraryCallKit::inline_native_getLength() {
3708 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3709
3710 Node* array = null_check(argument(0));
3711 // If array is dead, only null-path is taken.
3712 if (stopped()) return true;
3713
3714 // Deoptimize if it is a non-array.
3715 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3716
3717 if (non_array != NULL) {
3718 PreserveJVMState pjvms(this);
3719 set_control(non_array);
3720 uncommon_trap(Deoptimization::Reason_intrinsic,
3721 Deoptimization::Action_maybe_recompile);
3722 }
3723
3724 // If control is dead, only non-array-path is taken.
3725 if (stopped()) return true;
3726
3727 // The works fine even if the array type is polymorphic.
3728 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3729 Node* result = load_array_length(array);
3730
3731 C->set_has_split_ifs(true); // Has chance for split-if optimization
3732 set_result(result);
3733 return true;
3734 }
3735
3736 //------------------------inline_array_copyOf----------------------------
3737 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3738 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3739 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3740 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3741
3742 // Get the arguments.
3743 Node* original = argument(0);
3744 Node* start = is_copyOfRange? argument(1): intcon(0);
3745 Node* end = is_copyOfRange? argument(2): argument(1);
3746 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3747
3748 Node* newcopy = NULL;
3749
3750 // Set the original stack and the reexecute bit for the interpreter to reexecute
3751 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3752 { PreserveReexecuteState preexecs(this);
3753 jvms()->set_should_reexecute(true);
3754
3755 array_type_mirror = null_check(array_type_mirror);
3756 original = null_check(original);
3757
3758 // Check if a null path was taken unconditionally.
3759 if (stopped()) return true;
3760
3761 Node* orig_length = load_array_length(original);
3762
3763 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3764 klass_node = null_check(klass_node);
3765
3766 RegionNode* bailout = new RegionNode(1);
3767 record_for_igvn(bailout);
3768
3769 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3770 // Bail out if that is so.
3771 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3772 if (not_objArray != NULL) {
3773 // Improve the klass node's type from the new optimistic assumption:
3774 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3775 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3776 Node* cast = new CastPPNode(klass_node, akls);
3777 cast->init_req(0, control());
3778 klass_node = _gvn.transform(cast);
3779 }
3780
3781 // Bail out if either start or end is negative.
3782 generate_negative_guard(start, bailout, &start);
3783 generate_negative_guard(end, bailout, &end);
3784
3785 Node* length = end;
3786 if (_gvn.type(start) != TypeInt::ZERO) {
3787 length = _gvn.transform(new SubINode(end, start));
3788 }
3789
3790 // Bail out if length is negative.
3791 // Without this the new_array would throw
3792 // NegativeArraySizeException but IllegalArgumentException is what
3793 // should be thrown
3794 generate_negative_guard(length, bailout, &length);
3795
3796 if (bailout->req() > 1) {
3797 PreserveJVMState pjvms(this);
3798 set_control(_gvn.transform(bailout));
3799 uncommon_trap(Deoptimization::Reason_intrinsic,
3800 Deoptimization::Action_maybe_recompile);
3801 }
3802
3803 if (!stopped()) {
3804 // How many elements will we copy from the original?
3805 // The answer is MinI(orig_length - start, length).
3806 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3807 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3808
3809 original = access_resolve(original, ACCESS_READ);
3810
3811 // Generate a direct call to the right arraycopy function(s).
3812 // We know the copy is disjoint but we might not know if the
3813 // oop stores need checking.
3814 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3815 // This will fail a store-check if x contains any non-nulls.
3816
3817 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3818 // loads/stores but it is legal only if we're sure the
3819 // Arrays.copyOf would succeed. So we need all input arguments
3820 // to the copyOf to be validated, including that the copy to the
3821 // new array won't trigger an ArrayStoreException. That subtype
3822 // check can be optimized if we know something on the type of
3823 // the input array from type speculation.
3824 if (_gvn.type(klass_node)->singleton()) {
3825 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3826 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3827
3828 int test = C->static_subtype_check(superk, subk);
3829 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3830 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3831 if (t_original->speculative_type() != NULL) {
3832 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3833 }
3834 }
3835 }
3836
3837 bool validated = false;
3838 // Reason_class_check rather than Reason_intrinsic because we
3839 // want to intrinsify even if this traps.
3840 if (!too_many_traps(Deoptimization::Reason_class_check)) {
3841 Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
3842 klass_node);
3843
3844 if (not_subtype_ctrl != top()) {
3845 PreserveJVMState pjvms(this);
3846 set_control(not_subtype_ctrl);
3847 uncommon_trap(Deoptimization::Reason_class_check,
3848 Deoptimization::Action_make_not_entrant);
3849 assert(stopped(), "Should be stopped");
3850 }
3851 validated = true;
3852 }
3853
3854 if (!stopped()) {
3855 newcopy = new_array(klass_node, length, 0); // no arguments to push
3856
3857 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
3858 load_object_klass(original), klass_node);
3859 if (!is_copyOfRange) {
3860 ac->set_copyof(validated);
3861 } else {
3862 ac->set_copyofrange(validated);
3863 }
3864 Node* n = _gvn.transform(ac);
3865 if (n == ac) {
3866 ac->connect_outputs(this);
3867 } else {
3868 assert(validated, "shouldn't transform if all arguments not validated");
3869 set_all_memory(n);
3870 }
3871 }
3872 }
3873 } // original reexecute is set back here
3874
3875 C->set_has_split_ifs(true); // Has chance for split-if optimization
3876 if (!stopped()) {
3877 set_result(newcopy);
3878 }
3879 return true;
3880 }
3881
3882
3883 //----------------------generate_virtual_guard---------------------------
3884 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
3885 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
3886 RegionNode* slow_region) {
3887 ciMethod* method = callee();
3888 int vtable_index = method->vtable_index();
3889 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3890 "bad index %d", vtable_index);
3891 // Get the Method* out of the appropriate vtable entry.
3892 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
3893 vtable_index*vtableEntry::size_in_bytes() +
3894 vtableEntry::method_offset_in_bytes();
3895 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
3896 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3897
3898 // Compare the target method with the expected method (e.g., Object.hashCode).
3899 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
3900
3901 Node* native_call = makecon(native_call_addr);
3902 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
3903 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
3904
3905 return generate_slow_guard(test_native, slow_region);
3906 }
3907
3908 //-----------------------generate_method_call----------------------------
3909 // Use generate_method_call to make a slow-call to the real
3910 // method if the fast path fails. An alternative would be to
3911 // use a stub like OptoRuntime::slow_arraycopy_Java.
3912 // This only works for expanding the current library call,
3913 // not another intrinsic. (E.g., don't use this for making an
3914 // arraycopy call inside of the copyOf intrinsic.)
3915 CallJavaNode*
3916 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
3917 // When compiling the intrinsic method itself, do not use this technique.
3918 guarantee(callee() != C->method(), "cannot make slow-call to self");
3919
3920 ciMethod* method = callee();
3921 // ensure the JVMS we have will be correct for this call
3922 guarantee(method_id == method->intrinsic_id(), "must match");
3923
3924 const TypeFunc* tf = TypeFunc::make(method);
3925 CallJavaNode* slow_call;
3926 if (is_static) {
3927 assert(!is_virtual, "");
3928 slow_call = new CallStaticJavaNode(C, tf,
3929 SharedRuntime::get_resolve_static_call_stub(),
3930 method, bci());
3931 } else if (is_virtual) {
3932 null_check_receiver();
3933 int vtable_index = Method::invalid_vtable_index;
3934 if (UseInlineCaches) {
3935 // Suppress the vtable call
3936 } else {
3937 // hashCode and clone are not a miranda methods,
3938 // so the vtable index is fixed.
3939 // No need to use the linkResolver to get it.
3940 vtable_index = method->vtable_index();
3941 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3942 "bad index %d", vtable_index);
3943 }
3944 slow_call = new CallDynamicJavaNode(tf,
3945 SharedRuntime::get_resolve_virtual_call_stub(),
3946 method, vtable_index, bci());
3947 } else { // neither virtual nor static: opt_virtual
3948 null_check_receiver();
3949 slow_call = new CallStaticJavaNode(C, tf,
3950 SharedRuntime::get_resolve_opt_virtual_call_stub(),
3951 method, bci());
3952 slow_call->set_optimized_virtual(true);
3953 }
3954 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
3955 // To be able to issue a direct call (optimized virtual or virtual)
3956 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
3957 // about the method being invoked should be attached to the call site to
3958 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
3959 slow_call->set_override_symbolic_info(true);
3960 }
3961 set_arguments_for_java_call(slow_call);
3962 set_edges_for_java_call(slow_call);
3963 return slow_call;
3964 }
3965
3966
3967 /**
3968 * Build special case code for calls to hashCode on an object. This call may
3969 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
3970 * slightly different code.
3971 */
3972 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
3973 assert(is_static == callee()->is_static(), "correct intrinsic selection");
3974 assert(!(is_virtual && is_static), "either virtual, special, or static");
3975
3976 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
3977
3978 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3979 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
3980 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3981 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3982 Node* obj = NULL;
3983 if (!is_static) {
3984 // Check for hashing null object
3985 obj = null_check_receiver();
3986 if (stopped()) return true; // unconditionally null
3987 result_reg->init_req(_null_path, top());
3988 result_val->init_req(_null_path, top());
3989 } else {
3990 // Do a null check, and return zero if null.
3991 // System.identityHashCode(null) == 0
3992 obj = argument(0);
3993 Node* null_ctl = top();
3994 obj = null_check_oop(obj, &null_ctl);
3995 result_reg->init_req(_null_path, null_ctl);
3996 result_val->init_req(_null_path, _gvn.intcon(0));
3997 }
3998
3999 // Unconditionally null? Then return right away.
4000 if (stopped()) {
4001 set_control( result_reg->in(_null_path));
4002 if (!stopped())
4003 set_result(result_val->in(_null_path));
4004 return true;
4005 }
4006
4007 // We only go to the fast case code if we pass a number of guards. The
4008 // paths which do not pass are accumulated in the slow_region.
4009 RegionNode* slow_region = new RegionNode(1);
4010 record_for_igvn(slow_region);
4011
4012 // If this is a virtual call, we generate a funny guard. We pull out
4013 // the vtable entry corresponding to hashCode() from the target object.
4014 // If the target method which we are calling happens to be the native
4015 // Object hashCode() method, we pass the guard. We do not need this
4016 // guard for non-virtual calls -- the caller is known to be the native
4017 // Object hashCode().
4018 if (is_virtual) {
4019 // After null check, get the object's klass.
4020 Node* obj_klass = load_object_klass(obj);
4021 generate_virtual_guard(obj_klass, slow_region);
4022 }
4023
4024 // Get the header out of the object, use LoadMarkNode when available
4025 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4026 // The control of the load must be NULL. Otherwise, the load can move before
4027 // the null check after castPP removal.
4028 Node* no_ctrl = NULL;
4029 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4030
4031 // Test the header to see if it is unlocked.
4032 Node *lock_mask = _gvn.MakeConX(markWord::biased_lock_mask_in_place);
4033 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4034 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4035 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4036 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4037
4038 generate_slow_guard(test_unlocked, slow_region);
4039
4040 // Get the hash value and check to see that it has been properly assigned.
4041 // We depend on hash_mask being at most 32 bits and avoid the use of
4042 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4043 // vm: see markWord.hpp.
4044 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4045 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4046 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4047 // This hack lets the hash bits live anywhere in the mark object now, as long
4048 // as the shift drops the relevant bits into the low 32 bits. Note that
4049 // Java spec says that HashCode is an int so there's no point in capturing
4050 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4051 hshifted_header = ConvX2I(hshifted_header);
4052 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4053
4054 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4055 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4056 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4057
4058 generate_slow_guard(test_assigned, slow_region);
4059
4060 Node* init_mem = reset_memory();
4061 // fill in the rest of the null path:
4062 result_io ->init_req(_null_path, i_o());
4063 result_mem->init_req(_null_path, init_mem);
4064
4065 result_val->init_req(_fast_path, hash_val);
4066 result_reg->init_req(_fast_path, control());
4067 result_io ->init_req(_fast_path, i_o());
4068 result_mem->init_req(_fast_path, init_mem);
4069
4070 // Generate code for the slow case. We make a call to hashCode().
4071 set_control(_gvn.transform(slow_region));
4072 if (!stopped()) {
4073 // No need for PreserveJVMState, because we're using up the present state.
4074 set_all_memory(init_mem);
4075 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4076 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4077 Node* slow_result = set_results_for_java_call(slow_call);
4078 // this->control() comes from set_results_for_java_call
4079 result_reg->init_req(_slow_path, control());
4080 result_val->init_req(_slow_path, slow_result);
4081 result_io ->set_req(_slow_path, i_o());
4082 result_mem ->set_req(_slow_path, reset_memory());
4083 }
4084
4085 // Return the combined state.
4086 set_i_o( _gvn.transform(result_io) );
4087 set_all_memory( _gvn.transform(result_mem));
4088
4089 set_result(result_reg, result_val);
4090 return true;
4091 }
4092
4093 //---------------------------inline_native_getClass----------------------------
4094 // public final native Class<?> java.lang.Object.getClass();
4095 //
4096 // Build special case code for calls to getClass on an object.
4097 bool LibraryCallKit::inline_native_getClass() {
4098 Node* obj = null_check_receiver();
4099 if (stopped()) return true;
4100 set_result(load_mirror_from_klass(load_object_klass(obj)));
4101 return true;
4102 }
4103
4104 //-----------------inline_native_Reflection_getCallerClass---------------------
4105 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4106 //
4107 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4108 //
4109 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4110 // in that it must skip particular security frames and checks for
4111 // caller sensitive methods.
4112 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4113 #ifndef PRODUCT
4114 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4115 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4116 }
4117 #endif
4118
4119 if (!jvms()->has_method()) {
4120 #ifndef PRODUCT
4121 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4122 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4123 }
4124 #endif
4125 return false;
4126 }
4127
4128 // Walk back up the JVM state to find the caller at the required
4129 // depth.
4130 JVMState* caller_jvms = jvms();
4131
4132 // Cf. JVM_GetCallerClass
4133 // NOTE: Start the loop at depth 1 because the current JVM state does
4134 // not include the Reflection.getCallerClass() frame.
4135 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4136 ciMethod* m = caller_jvms->method();
4137 switch (n) {
4138 case 0:
4139 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4140 break;
4141 case 1:
4142 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4143 if (!m->caller_sensitive()) {
4144 #ifndef PRODUCT
4145 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4146 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4147 }
4148 #endif
4149 return false; // bail-out; let JVM_GetCallerClass do the work
4150 }
4151 break;
4152 default:
4153 if (!m->is_ignored_by_security_stack_walk()) {
4154 // We have reached the desired frame; return the holder class.
4155 // Acquire method holder as java.lang.Class and push as constant.
4156 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4157 ciInstance* caller_mirror = caller_klass->java_mirror();
4158 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4159
4160 #ifndef PRODUCT
4161 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4162 tty->print_cr(" Succeeded: caller = %d) %s.%s, JVMS depth = %d", n, caller_klass->name()->as_utf8(), caller_jvms->method()->name()->as_utf8(), jvms()->depth());
4163 tty->print_cr(" JVM state at this point:");
4164 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4165 ciMethod* m = jvms()->of_depth(i)->method();
4166 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4167 }
4168 }
4169 #endif
4170 return true;
4171 }
4172 break;
4173 }
4174 }
4175
4176 #ifndef PRODUCT
4177 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4178 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4179 tty->print_cr(" JVM state at this point:");
4180 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4181 ciMethod* m = jvms()->of_depth(i)->method();
4182 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4183 }
4184 }
4185 #endif
4186
4187 return false; // bail-out; let JVM_GetCallerClass do the work
4188 }
4189
4190 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4191 Node* arg = argument(0);
4192 Node* result = NULL;
4193
4194 switch (id) {
4195 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4196 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4197 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4198 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4199
4200 case vmIntrinsics::_doubleToLongBits: {
4201 // two paths (plus control) merge in a wood
4202 RegionNode *r = new RegionNode(3);
4203 Node *phi = new PhiNode(r, TypeLong::LONG);
4204
4205 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4206 // Build the boolean node
4207 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4208
4209 // Branch either way.
4210 // NaN case is less traveled, which makes all the difference.
4211 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4212 Node *opt_isnan = _gvn.transform(ifisnan);
4213 assert( opt_isnan->is_If(), "Expect an IfNode");
4214 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4215 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4216
4217 set_control(iftrue);
4218
4219 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4220 Node *slow_result = longcon(nan_bits); // return NaN
4221 phi->init_req(1, _gvn.transform( slow_result ));
4222 r->init_req(1, iftrue);
4223
4224 // Else fall through
4225 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4226 set_control(iffalse);
4227
4228 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4229 r->init_req(2, iffalse);
4230
4231 // Post merge
4232 set_control(_gvn.transform(r));
4233 record_for_igvn(r);
4234
4235 C->set_has_split_ifs(true); // Has chance for split-if optimization
4236 result = phi;
4237 assert(result->bottom_type()->isa_long(), "must be");
4238 break;
4239 }
4240
4241 case vmIntrinsics::_floatToIntBits: {
4242 // two paths (plus control) merge in a wood
4243 RegionNode *r = new RegionNode(3);
4244 Node *phi = new PhiNode(r, TypeInt::INT);
4245
4246 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4247 // Build the boolean node
4248 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4249
4250 // Branch either way.
4251 // NaN case is less traveled, which makes all the difference.
4252 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4253 Node *opt_isnan = _gvn.transform(ifisnan);
4254 assert( opt_isnan->is_If(), "Expect an IfNode");
4255 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4256 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4257
4258 set_control(iftrue);
4259
4260 static const jint nan_bits = 0x7fc00000;
4261 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4262 phi->init_req(1, _gvn.transform( slow_result ));
4263 r->init_req(1, iftrue);
4264
4265 // Else fall through
4266 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4267 set_control(iffalse);
4268
4269 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4270 r->init_req(2, iffalse);
4271
4272 // Post merge
4273 set_control(_gvn.transform(r));
4274 record_for_igvn(r);
4275
4276 C->set_has_split_ifs(true); // Has chance for split-if optimization
4277 result = phi;
4278 assert(result->bottom_type()->isa_int(), "must be");
4279 break;
4280 }
4281
4282 default:
4283 fatal_unexpected_iid(id);
4284 break;
4285 }
4286 set_result(_gvn.transform(result));
4287 return true;
4288 }
4289
4290 //----------------------inline_unsafe_copyMemory-------------------------
4291 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4292 bool LibraryCallKit::inline_unsafe_copyMemory() {
4293 if (callee()->is_static()) return false; // caller must have the capability!
4294 null_check_receiver(); // null-check receiver
4295 if (stopped()) return true;
4296
4297 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4298
4299 Node* src_ptr = argument(1); // type: oop
4300 Node* src_off = ConvL2X(argument(2)); // type: long
4301 Node* dst_ptr = argument(4); // type: oop
4302 Node* dst_off = ConvL2X(argument(5)); // type: long
4303 Node* size = ConvL2X(argument(7)); // type: long
4304
4305 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4306 "fieldOffset must be byte-scaled");
4307
4308 src_ptr = access_resolve(src_ptr, ACCESS_READ);
4309 dst_ptr = access_resolve(dst_ptr, ACCESS_WRITE);
4310 Node* src = make_unsafe_address(src_ptr, src_off, ACCESS_READ);
4311 Node* dst = make_unsafe_address(dst_ptr, dst_off, ACCESS_WRITE);
4312
4313 // Conservatively insert a memory barrier on all memory slices.
4314 // Do not let writes of the copy source or destination float below the copy.
4315 insert_mem_bar(Op_MemBarCPUOrder);
4316
4317 Node* thread = _gvn.transform(new ThreadLocalNode());
4318 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4319 BasicType doing_unsafe_access_bt = T_BYTE;
4320 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4321
4322 // update volatile field
4323 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4324
4325 // Call it. Note that the length argument is not scaled.
4326 make_runtime_call(RC_LEAF|RC_NO_FP,
4327 OptoRuntime::fast_arraycopy_Type(),
4328 StubRoutines::unsafe_arraycopy(),
4329 "unsafe_arraycopy",
4330 TypeRawPtr::BOTTOM,
4331 src, dst, size XTOP);
4332
4333 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4334
4335 // Do not let reads of the copy destination float above the copy.
4336 insert_mem_bar(Op_MemBarCPUOrder);
4337
4338 return true;
4339 }
4340
4341 //------------------------clone_coping-----------------------------------
4342 // Helper function for inline_native_clone.
4343 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4344 assert(obj_size != NULL, "");
4345 Node* raw_obj = alloc_obj->in(1);
4346 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4347
4348 AllocateNode* alloc = NULL;
4349 if (ReduceBulkZeroing) {
4350 // We will be completely responsible for initializing this object -
4351 // mark Initialize node as complete.
4352 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4353 // The object was just allocated - there should be no any stores!
4354 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4355 // Mark as complete_with_arraycopy so that on AllocateNode
4356 // expansion, we know this AllocateNode is initialized by an array
4357 // copy and a StoreStore barrier exists after the array copy.
4358 alloc->initialization()->set_complete_with_arraycopy();
4359 }
4360
4361 // Copy the fastest available way.
4362 // TODO: generate fields copies for small objects instead.
4363 Node* size = _gvn.transform(obj_size);
4364
4365 access_clone(obj, alloc_obj, size, is_array);
4366
4367 // Do not let reads from the cloned object float above the arraycopy.
4368 if (alloc != NULL) {
4369 // Do not let stores that initialize this object be reordered with
4370 // a subsequent store that would make this object accessible by
4371 // other threads.
4372 // Record what AllocateNode this StoreStore protects so that
4373 // escape analysis can go from the MemBarStoreStoreNode to the
4374 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4375 // based on the escape status of the AllocateNode.
4376 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4377 } else {
4378 insert_mem_bar(Op_MemBarCPUOrder);
4379 }
4380 }
4381
4382 //------------------------inline_native_clone----------------------------
4383 // protected native Object java.lang.Object.clone();
4384 //
4385 // Here are the simple edge cases:
4386 // null receiver => normal trap
4387 // virtual and clone was overridden => slow path to out-of-line clone
4388 // not cloneable or finalizer => slow path to out-of-line Object.clone
4389 //
4390 // The general case has two steps, allocation and copying.
4391 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4392 //
4393 // Copying also has two cases, oop arrays and everything else.
4394 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4395 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4396 //
4397 // These steps fold up nicely if and when the cloned object's klass
4398 // can be sharply typed as an object array, a type array, or an instance.
4399 //
4400 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4401 PhiNode* result_val;
4402
4403 // Set the reexecute bit for the interpreter to reexecute
4404 // the bytecode that invokes Object.clone if deoptimization happens.
4405 { PreserveReexecuteState preexecs(this);
4406 jvms()->set_should_reexecute(true);
4407
4408 Node* obj = null_check_receiver();
4409 if (stopped()) return true;
4410
4411 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4412
4413 // If we are going to clone an instance, we need its exact type to
4414 // know the number and types of fields to convert the clone to
4415 // loads/stores. Maybe a speculative type can help us.
4416 if (!obj_type->klass_is_exact() &&
4417 obj_type->speculative_type() != NULL &&
4418 obj_type->speculative_type()->is_instance_klass()) {
4419 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4420 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4421 !spec_ik->has_injected_fields()) {
4422 ciKlass* k = obj_type->klass();
4423 if (!k->is_instance_klass() ||
4424 k->as_instance_klass()->is_interface() ||
4425 k->as_instance_klass()->has_subklass()) {
4426 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4427 }
4428 }
4429 }
4430
4431 Node* obj_klass = load_object_klass(obj);
4432 const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
4433 const TypeOopPtr* toop = ((tklass != NULL)
4434 ? tklass->as_instance_type()
4435 : TypeInstPtr::NOTNULL);
4436
4437 // Conservatively insert a memory barrier on all memory slices.
4438 // Do not let writes into the original float below the clone.
4439 insert_mem_bar(Op_MemBarCPUOrder);
4440
4441 // paths into result_reg:
4442 enum {
4443 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4444 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4445 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4446 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4447 PATH_LIMIT
4448 };
4449 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4450 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4451 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4452 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4453 record_for_igvn(result_reg);
4454
4455 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4456 if (array_ctl != NULL) {
4457 // It's an array.
4458 PreserveJVMState pjvms(this);
4459 set_control(array_ctl);
4460 Node* obj_length = load_array_length(obj);
4461 Node* obj_size = NULL;
4462 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size); // no arguments to push
4463
4464 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4465 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, BarrierSetC2::Parsing)) {
4466 // If it is an oop array, it requires very special treatment,
4467 // because gc barriers are required when accessing the array.
4468 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4469 if (is_obja != NULL) {
4470 PreserveJVMState pjvms2(this);
4471 set_control(is_obja);
4472 obj = access_resolve(obj, ACCESS_READ);
4473 // Generate a direct call to the right arraycopy function(s).
4474 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4475 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false);
4476 ac->set_cloneoop();
4477 Node* n = _gvn.transform(ac);
4478 assert(n == ac, "cannot disappear");
4479 ac->connect_outputs(this);
4480
4481 result_reg->init_req(_objArray_path, control());
4482 result_val->init_req(_objArray_path, alloc_obj);
4483 result_i_o ->set_req(_objArray_path, i_o());
4484 result_mem ->set_req(_objArray_path, reset_memory());
4485 }
4486 }
4487 // Otherwise, there are no barriers to worry about.
4488 // (We can dispense with card marks if we know the allocation
4489 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4490 // causes the non-eden paths to take compensating steps to
4491 // simulate a fresh allocation, so that no further
4492 // card marks are required in compiled code to initialize
4493 // the object.)
4494
4495 if (!stopped()) {
4496 copy_to_clone(obj, alloc_obj, obj_size, true);
4497
4498 // Present the results of the copy.
4499 result_reg->init_req(_array_path, control());
4500 result_val->init_req(_array_path, alloc_obj);
4501 result_i_o ->set_req(_array_path, i_o());
4502 result_mem ->set_req(_array_path, reset_memory());
4503 }
4504 }
4505
4506 // We only go to the instance fast case code if we pass a number of guards.
4507 // The paths which do not pass are accumulated in the slow_region.
4508 RegionNode* slow_region = new RegionNode(1);
4509 record_for_igvn(slow_region);
4510 if (!stopped()) {
4511 // It's an instance (we did array above). Make the slow-path tests.
4512 // If this is a virtual call, we generate a funny guard. We grab
4513 // the vtable entry corresponding to clone() from the target object.
4514 // If the target method which we are calling happens to be the
4515 // Object clone() method, we pass the guard. We do not need this
4516 // guard for non-virtual calls; the caller is known to be the native
4517 // Object clone().
4518 if (is_virtual) {
4519 generate_virtual_guard(obj_klass, slow_region);
4520 }
4521
4522 // The object must be easily cloneable and must not have a finalizer.
4523 // Both of these conditions may be checked in a single test.
4524 // We could optimize the test further, but we don't care.
4525 generate_access_flags_guard(obj_klass,
4526 // Test both conditions:
4527 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4528 // Must be cloneable but not finalizer:
4529 JVM_ACC_IS_CLONEABLE_FAST,
4530 slow_region);
4531 }
4532
4533 if (!stopped()) {
4534 // It's an instance, and it passed the slow-path tests.
4535 PreserveJVMState pjvms(this);
4536 Node* obj_size = NULL;
4537 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4538 // is reexecuted if deoptimization occurs and there could be problems when merging
4539 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4540 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4541
4542 copy_to_clone(obj, alloc_obj, obj_size, false);
4543
4544 // Present the results of the slow call.
4545 result_reg->init_req(_instance_path, control());
4546 result_val->init_req(_instance_path, alloc_obj);
4547 result_i_o ->set_req(_instance_path, i_o());
4548 result_mem ->set_req(_instance_path, reset_memory());
4549 }
4550
4551 // Generate code for the slow case. We make a call to clone().
4552 set_control(_gvn.transform(slow_region));
4553 if (!stopped()) {
4554 PreserveJVMState pjvms(this);
4555 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4556 // We need to deoptimize on exception (see comment above)
4557 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4558 // this->control() comes from set_results_for_java_call
4559 result_reg->init_req(_slow_path, control());
4560 result_val->init_req(_slow_path, slow_result);
4561 result_i_o ->set_req(_slow_path, i_o());
4562 result_mem ->set_req(_slow_path, reset_memory());
4563 }
4564
4565 // Return the combined state.
4566 set_control( _gvn.transform(result_reg));
4567 set_i_o( _gvn.transform(result_i_o));
4568 set_all_memory( _gvn.transform(result_mem));
4569 } // original reexecute is set back here
4570
4571 set_result(_gvn.transform(result_val));
4572 return true;
4573 }
4574
4575 // If we have a tightly coupled allocation, the arraycopy may take care
4576 // of the array initialization. If one of the guards we insert between
4577 // the allocation and the arraycopy causes a deoptimization, an
4578 // unitialized array will escape the compiled method. To prevent that
4579 // we set the JVM state for uncommon traps between the allocation and
4580 // the arraycopy to the state before the allocation so, in case of
4581 // deoptimization, we'll reexecute the allocation and the
4582 // initialization.
4583 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4584 if (alloc != NULL) {
4585 ciMethod* trap_method = alloc->jvms()->method();
4586 int trap_bci = alloc->jvms()->bci();
4587
4588 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4589 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4590 // Make sure there's no store between the allocation and the
4591 // arraycopy otherwise visible side effects could be rexecuted
4592 // in case of deoptimization and cause incorrect execution.
4593 bool no_interfering_store = true;
4594 Node* mem = alloc->in(TypeFunc::Memory);
4595 if (mem->is_MergeMem()) {
4596 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4597 Node* n = mms.memory();
4598 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4599 assert(n->is_Store(), "what else?");
4600 no_interfering_store = false;
4601 break;
4602 }
4603 }
4604 } else {
4605 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4606 Node* n = mms.memory();
4607 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4608 assert(n->is_Store(), "what else?");
4609 no_interfering_store = false;
4610 break;
4611 }
4612 }
4613 }
4614
4615 if (no_interfering_store) {
4616 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4617 uint size = alloc->req();
4618 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4619 old_jvms->set_map(sfpt);
4620 for (uint i = 0; i < size; i++) {
4621 sfpt->init_req(i, alloc->in(i));
4622 }
4623 // re-push array length for deoptimization
4624 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4625 old_jvms->set_sp(old_jvms->sp()+1);
4626 old_jvms->set_monoff(old_jvms->monoff()+1);
4627 old_jvms->set_scloff(old_jvms->scloff()+1);
4628 old_jvms->set_endoff(old_jvms->endoff()+1);
4629 old_jvms->set_should_reexecute(true);
4630
4631 sfpt->set_i_o(map()->i_o());
4632 sfpt->set_memory(map()->memory());
4633 sfpt->set_control(map()->control());
4634
4635 JVMState* saved_jvms = jvms();
4636 saved_reexecute_sp = _reexecute_sp;
4637
4638 set_jvms(sfpt->jvms());
4639 _reexecute_sp = jvms()->sp();
4640
4641 return saved_jvms;
4642 }
4643 }
4644 }
4645 return NULL;
4646 }
4647
4648 // In case of a deoptimization, we restart execution at the
4649 // allocation, allocating a new array. We would leave an uninitialized
4650 // array in the heap that GCs wouldn't expect. Move the allocation
4651 // after the traps so we don't allocate the array if we
4652 // deoptimize. This is possible because tightly_coupled_allocation()
4653 // guarantees there's no observer of the allocated array at this point
4654 // and the control flow is simple enough.
4655 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4656 int saved_reexecute_sp, uint new_idx) {
4657 if (saved_jvms != NULL && !stopped()) {
4658 assert(alloc != NULL, "only with a tightly coupled allocation");
4659 // restore JVM state to the state at the arraycopy
4660 saved_jvms->map()->set_control(map()->control());
4661 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4662 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4663 // If we've improved the types of some nodes (null check) while
4664 // emitting the guards, propagate them to the current state
4665 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4666 set_jvms(saved_jvms);
4667 _reexecute_sp = saved_reexecute_sp;
4668
4669 // Remove the allocation from above the guards
4670 CallProjections callprojs;
4671 alloc->extract_projections(&callprojs, true);
4672 InitializeNode* init = alloc->initialization();
4673 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4674 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4675 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4676 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4677
4678 // move the allocation here (after the guards)
4679 _gvn.hash_delete(alloc);
4680 alloc->set_req(TypeFunc::Control, control());
4681 alloc->set_req(TypeFunc::I_O, i_o());
4682 Node *mem = reset_memory();
4683 set_all_memory(mem);
4684 alloc->set_req(TypeFunc::Memory, mem);
4685 set_control(init->proj_out_or_null(TypeFunc::Control));
4686 set_i_o(callprojs.fallthrough_ioproj);
4687
4688 // Update memory as done in GraphKit::set_output_for_allocation()
4689 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4690 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4691 if (ary_type->isa_aryptr() && length_type != NULL) {
4692 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4693 }
4694 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4695 int elemidx = C->get_alias_index(telemref);
4696 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
4697 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
4698
4699 Node* allocx = _gvn.transform(alloc);
4700 assert(allocx == alloc, "where has the allocation gone?");
4701 assert(dest->is_CheckCastPP(), "not an allocation result?");
4702
4703 _gvn.hash_delete(dest);
4704 dest->set_req(0, control());
4705 Node* destx = _gvn.transform(dest);
4706 assert(destx == dest, "where has the allocation result gone?");
4707 }
4708 }
4709
4710
4711 //------------------------------inline_arraycopy-----------------------
4712 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4713 // Object dest, int destPos,
4714 // int length);
4715 bool LibraryCallKit::inline_arraycopy() {
4716 // Get the arguments.
4717 Node* src = argument(0); // type: oop
4718 Node* src_offset = argument(1); // type: int
4719 Node* dest = argument(2); // type: oop
4720 Node* dest_offset = argument(3); // type: int
4721 Node* length = argument(4); // type: int
4722
4723 uint new_idx = C->unique();
4724
4725 // Check for allocation before we add nodes that would confuse
4726 // tightly_coupled_allocation()
4727 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4728
4729 int saved_reexecute_sp = -1;
4730 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4731 // See arraycopy_restore_alloc_state() comment
4732 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4733 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4734 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
4735 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4736
4737 // The following tests must be performed
4738 // (1) src and dest are arrays.
4739 // (2) src and dest arrays must have elements of the same BasicType
4740 // (3) src and dest must not be null.
4741 // (4) src_offset must not be negative.
4742 // (5) dest_offset must not be negative.
4743 // (6) length must not be negative.
4744 // (7) src_offset + length must not exceed length of src.
4745 // (8) dest_offset + length must not exceed length of dest.
4746 // (9) each element of an oop array must be assignable
4747
4748 // (3) src and dest must not be null.
4749 // always do this here because we need the JVM state for uncommon traps
4750 Node* null_ctl = top();
4751 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4752 assert(null_ctl->is_top(), "no null control here");
4753 dest = null_check(dest, T_ARRAY);
4754
4755 if (!can_emit_guards) {
4756 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4757 // guards but the arraycopy node could still take advantage of a
4758 // tightly allocated allocation. tightly_coupled_allocation() is
4759 // called again to make sure it takes the null check above into
4760 // account: the null check is mandatory and if it caused an
4761 // uncommon trap to be emitted then the allocation can't be
4762 // considered tightly coupled in this context.
4763 alloc = tightly_coupled_allocation(dest, NULL);
4764 }
4765
4766 bool validated = false;
4767
4768 const Type* src_type = _gvn.type(src);
4769 const Type* dest_type = _gvn.type(dest);
4770 const TypeAryPtr* top_src = src_type->isa_aryptr();
4771 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4772
4773 // Do we have the type of src?
4774 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4775 // Do we have the type of dest?
4776 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4777 // Is the type for src from speculation?
4778 bool src_spec = false;
4779 // Is the type for dest from speculation?
4780 bool dest_spec = false;
4781
4782 if ((!has_src || !has_dest) && can_emit_guards) {
4783 // We don't have sufficient type information, let's see if
4784 // speculative types can help. We need to have types for both src
4785 // and dest so that it pays off.
4786
4787 // Do we already have or could we have type information for src
4788 bool could_have_src = has_src;
4789 // Do we already have or could we have type information for dest
4790 bool could_have_dest = has_dest;
4791
4792 ciKlass* src_k = NULL;
4793 if (!has_src) {
4794 src_k = src_type->speculative_type_not_null();
4795 if (src_k != NULL && src_k->is_array_klass()) {
4796 could_have_src = true;
4797 }
4798 }
4799
4800 ciKlass* dest_k = NULL;
4801 if (!has_dest) {
4802 dest_k = dest_type->speculative_type_not_null();
4803 if (dest_k != NULL && dest_k->is_array_klass()) {
4804 could_have_dest = true;
4805 }
4806 }
4807
4808 if (could_have_src && could_have_dest) {
4809 // This is going to pay off so emit the required guards
4810 if (!has_src) {
4811 src = maybe_cast_profiled_obj(src, src_k, true);
4812 src_type = _gvn.type(src);
4813 top_src = src_type->isa_aryptr();
4814 has_src = (top_src != NULL && top_src->klass() != NULL);
4815 src_spec = true;
4816 }
4817 if (!has_dest) {
4818 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4819 dest_type = _gvn.type(dest);
4820 top_dest = dest_type->isa_aryptr();
4821 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4822 dest_spec = true;
4823 }
4824 }
4825 }
4826
4827 if (has_src && has_dest && can_emit_guards) {
4828 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
4829 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
4830 if (is_reference_type(src_elem)) src_elem = T_OBJECT;
4831 if (is_reference_type(dest_elem)) dest_elem = T_OBJECT;
4832
4833 if (src_elem == dest_elem && src_elem == T_OBJECT) {
4834 // If both arrays are object arrays then having the exact types
4835 // for both will remove the need for a subtype check at runtime
4836 // before the call and may make it possible to pick a faster copy
4837 // routine (without a subtype check on every element)
4838 // Do we have the exact type of src?
4839 bool could_have_src = src_spec;
4840 // Do we have the exact type of dest?
4841 bool could_have_dest = dest_spec;
4842 ciKlass* src_k = top_src->klass();
4843 ciKlass* dest_k = top_dest->klass();
4844 if (!src_spec) {
4845 src_k = src_type->speculative_type_not_null();
4846 if (src_k != NULL && src_k->is_array_klass()) {
4847 could_have_src = true;
4848 }
4849 }
4850 if (!dest_spec) {
4851 dest_k = dest_type->speculative_type_not_null();
4852 if (dest_k != NULL && dest_k->is_array_klass()) {
4853 could_have_dest = true;
4854 }
4855 }
4856 if (could_have_src && could_have_dest) {
4857 // If we can have both exact types, emit the missing guards
4858 if (could_have_src && !src_spec) {
4859 src = maybe_cast_profiled_obj(src, src_k, true);
4860 }
4861 if (could_have_dest && !dest_spec) {
4862 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4863 }
4864 }
4865 }
4866 }
4867
4868 ciMethod* trap_method = method();
4869 int trap_bci = bci();
4870 if (saved_jvms != NULL) {
4871 trap_method = alloc->jvms()->method();
4872 trap_bci = alloc->jvms()->bci();
4873 }
4874
4875 bool negative_length_guard_generated = false;
4876
4877 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4878 can_emit_guards &&
4879 !src->is_top() && !dest->is_top()) {
4880 // validate arguments: enables transformation the ArrayCopyNode
4881 validated = true;
4882
4883 RegionNode* slow_region = new RegionNode(1);
4884 record_for_igvn(slow_region);
4885
4886 // (1) src and dest are arrays.
4887 generate_non_array_guard(load_object_klass(src), slow_region);
4888 generate_non_array_guard(load_object_klass(dest), slow_region);
4889
4890 // (2) src and dest arrays must have elements of the same BasicType
4891 // done at macro expansion or at Ideal transformation time
4892
4893 // (4) src_offset must not be negative.
4894 generate_negative_guard(src_offset, slow_region);
4895
4896 // (5) dest_offset must not be negative.
4897 generate_negative_guard(dest_offset, slow_region);
4898
4899 // (7) src_offset + length must not exceed length of src.
4900 generate_limit_guard(src_offset, length,
4901 load_array_length(src),
4902 slow_region);
4903
4904 // (8) dest_offset + length must not exceed length of dest.
4905 generate_limit_guard(dest_offset, length,
4906 load_array_length(dest),
4907 slow_region);
4908
4909 // (6) length must not be negative.
4910 // This is also checked in generate_arraycopy() during macro expansion, but
4911 // we also have to check it here for the case where the ArrayCopyNode will
4912 // be eliminated by Escape Analysis.
4913 if (EliminateAllocations) {
4914 generate_negative_guard(length, slow_region);
4915 negative_length_guard_generated = true;
4916 }
4917
4918 // (9) each element of an oop array must be assignable
4919 Node* src_klass = load_object_klass(src);
4920 Node* dest_klass = load_object_klass(dest);
4921 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
4922
4923 if (not_subtype_ctrl != top()) {
4924 PreserveJVMState pjvms(this);
4925 set_control(not_subtype_ctrl);
4926 uncommon_trap(Deoptimization::Reason_intrinsic,
4927 Deoptimization::Action_make_not_entrant);
4928 assert(stopped(), "Should be stopped");
4929 }
4930 {
4931 PreserveJVMState pjvms(this);
4932 set_control(_gvn.transform(slow_region));
4933 uncommon_trap(Deoptimization::Reason_intrinsic,
4934 Deoptimization::Action_make_not_entrant);
4935 assert(stopped(), "Should be stopped");
4936 }
4937
4938 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
4939 const Type *toop = TypeOopPtr::make_from_klass(dest_klass_t->klass());
4940 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
4941 }
4942
4943 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
4944
4945 if (stopped()) {
4946 return true;
4947 }
4948
4949 Node* new_src = access_resolve(src, ACCESS_READ);
4950 Node* new_dest = access_resolve(dest, ACCESS_WRITE);
4951
4952 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, new_src, src_offset, new_dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
4953 // Create LoadRange and LoadKlass nodes for use during macro expansion here
4954 // so the compiler has a chance to eliminate them: during macro expansion,
4955 // we have to set their control (CastPP nodes are eliminated).
4956 load_object_klass(src), load_object_klass(dest),
4957 load_array_length(src), load_array_length(dest));
4958
4959 ac->set_arraycopy(validated);
4960
4961 Node* n = _gvn.transform(ac);
4962 if (n == ac) {
4963 ac->connect_outputs(this);
4964 } else {
4965 assert(validated, "shouldn't transform if all arguments not validated");
4966 set_all_memory(n);
4967 }
4968 clear_upper_avx();
4969
4970
4971 return true;
4972 }
4973
4974
4975 // Helper function which determines if an arraycopy immediately follows
4976 // an allocation, with no intervening tests or other escapes for the object.
4977 AllocateArrayNode*
4978 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
4979 RegionNode* slow_region) {
4980 if (stopped()) return NULL; // no fast path
4981 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
4982
4983 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
4984 if (alloc == NULL) return NULL;
4985
4986 Node* rawmem = memory(Compile::AliasIdxRaw);
4987 // Is the allocation's memory state untouched?
4988 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
4989 // Bail out if there have been raw-memory effects since the allocation.
4990 // (Example: There might have been a call or safepoint.)
4991 return NULL;
4992 }
4993 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
4994 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
4995 return NULL;
4996 }
4997
4998 // There must be no unexpected observers of this allocation.
4999 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5000 Node* obs = ptr->fast_out(i);
5001 if (obs != this->map()) {
5002 return NULL;
5003 }
5004 }
5005
5006 // This arraycopy must unconditionally follow the allocation of the ptr.
5007 Node* alloc_ctl = ptr->in(0);
5008 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
5009
5010 Node* ctl = control();
5011 while (ctl != alloc_ctl) {
5012 // There may be guards which feed into the slow_region.
5013 // Any other control flow means that we might not get a chance
5014 // to finish initializing the allocated object.
5015 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5016 IfNode* iff = ctl->in(0)->as_If();
5017 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
5018 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5019 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
5020 ctl = iff->in(0); // This test feeds the known slow_region.
5021 continue;
5022 }
5023 // One more try: Various low-level checks bottom out in
5024 // uncommon traps. If the debug-info of the trap omits
5025 // any reference to the allocation, as we've already
5026 // observed, then there can be no objection to the trap.
5027 bool found_trap = false;
5028 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5029 Node* obs = not_ctl->fast_out(j);
5030 if (obs->in(0) == not_ctl && obs->is_Call() &&
5031 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5032 found_trap = true; break;
5033 }
5034 }
5035 if (found_trap) {
5036 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
5037 continue;
5038 }
5039 }
5040 return NULL;
5041 }
5042
5043 // If we get this far, we have an allocation which immediately
5044 // precedes the arraycopy, and we can take over zeroing the new object.
5045 // The arraycopy will finish the initialization, and provide
5046 // a new control state to which we will anchor the destination pointer.
5047
5048 return alloc;
5049 }
5050
5051 //-------------inline_encodeISOArray-----------------------------------
5052 // encode char[] to byte[] in ISO_8859_1
5053 bool LibraryCallKit::inline_encodeISOArray() {
5054 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5055 // no receiver since it is static method
5056 Node *src = argument(0);
5057 Node *src_offset = argument(1);
5058 Node *dst = argument(2);
5059 Node *dst_offset = argument(3);
5060 Node *length = argument(4);
5061
5062 src = must_be_not_null(src, true);
5063 dst = must_be_not_null(dst, true);
5064
5065 src = access_resolve(src, ACCESS_READ);
5066 dst = access_resolve(dst, ACCESS_WRITE);
5067
5068 const Type* src_type = src->Value(&_gvn);
5069 const Type* dst_type = dst->Value(&_gvn);
5070 const TypeAryPtr* top_src = src_type->isa_aryptr();
5071 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5072 if (top_src == NULL || top_src->klass() == NULL ||
5073 top_dest == NULL || top_dest->klass() == NULL) {
5074 // failed array check
5075 return false;
5076 }
5077
5078 // Figure out the size and type of the elements we will be copying.
5079 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5080 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5081 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5082 return false;
5083 }
5084
5085 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5086 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5087 // 'src_start' points to src array + scaled offset
5088 // 'dst_start' points to dst array + scaled offset
5089
5090 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5091 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
5092 enc = _gvn.transform(enc);
5093 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5094 set_memory(res_mem, mtype);
5095 set_result(enc);
5096 clear_upper_avx();
5097
5098 return true;
5099 }
5100
5101 //-------------inline_multiplyToLen-----------------------------------
5102 bool LibraryCallKit::inline_multiplyToLen() {
5103 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5104
5105 address stubAddr = StubRoutines::multiplyToLen();
5106 if (stubAddr == NULL) {
5107 return false; // Intrinsic's stub is not implemented on this platform
5108 }
5109 const char* stubName = "multiplyToLen";
5110
5111 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5112
5113 // no receiver because it is a static method
5114 Node* x = argument(0);
5115 Node* xlen = argument(1);
5116 Node* y = argument(2);
5117 Node* ylen = argument(3);
5118 Node* z = argument(4);
5119
5120 x = must_be_not_null(x, true);
5121 y = must_be_not_null(y, true);
5122
5123 x = access_resolve(x, ACCESS_READ);
5124 y = access_resolve(y, ACCESS_READ);
5125 z = access_resolve(z, ACCESS_WRITE);
5126
5127 const Type* x_type = x->Value(&_gvn);
5128 const Type* y_type = y->Value(&_gvn);
5129 const TypeAryPtr* top_x = x_type->isa_aryptr();
5130 const TypeAryPtr* top_y = y_type->isa_aryptr();
5131 if (top_x == NULL || top_x->klass() == NULL ||
5132 top_y == NULL || top_y->klass() == NULL) {
5133 // failed array check
5134 return false;
5135 }
5136
5137 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5138 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5139 if (x_elem != T_INT || y_elem != T_INT) {
5140 return false;
5141 }
5142
5143 // Set the original stack and the reexecute bit for the interpreter to reexecute
5144 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5145 // on the return from z array allocation in runtime.
5146 { PreserveReexecuteState preexecs(this);
5147 jvms()->set_should_reexecute(true);
5148
5149 Node* x_start = array_element_address(x, intcon(0), x_elem);
5150 Node* y_start = array_element_address(y, intcon(0), y_elem);
5151 // 'x_start' points to x array + scaled xlen
5152 // 'y_start' points to y array + scaled ylen
5153
5154 // Allocate the result array
5155 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5156 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5157 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5158
5159 IdealKit ideal(this);
5160
5161 #define __ ideal.
5162 Node* one = __ ConI(1);
5163 Node* zero = __ ConI(0);
5164 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5165 __ set(need_alloc, zero);
5166 __ set(z_alloc, z);
5167 __ if_then(z, BoolTest::eq, null()); {
5168 __ increment (need_alloc, one);
5169 } __ else_(); {
5170 // Update graphKit memory and control from IdealKit.
5171 sync_kit(ideal);
5172 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5173 cast->init_req(0, control());
5174 _gvn.set_type(cast, cast->bottom_type());
5175 C->record_for_igvn(cast);
5176
5177 Node* zlen_arg = load_array_length(cast);
5178 // Update IdealKit memory and control from graphKit.
5179 __ sync_kit(this);
5180 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5181 __ increment (need_alloc, one);
5182 } __ end_if();
5183 } __ end_if();
5184
5185 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5186 // Update graphKit memory and control from IdealKit.
5187 sync_kit(ideal);
5188 Node * narr = new_array(klass_node, zlen, 1);
5189 // Update IdealKit memory and control from graphKit.
5190 __ sync_kit(this);
5191 __ set(z_alloc, narr);
5192 } __ end_if();
5193
5194 sync_kit(ideal);
5195 z = __ value(z_alloc);
5196 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5197 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5198 // Final sync IdealKit and GraphKit.
5199 final_sync(ideal);
5200 #undef __
5201
5202 Node* z_start = array_element_address(z, intcon(0), T_INT);
5203
5204 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5205 OptoRuntime::multiplyToLen_Type(),
5206 stubAddr, stubName, TypePtr::BOTTOM,
5207 x_start, xlen, y_start, ylen, z_start, zlen);
5208 } // original reexecute is set back here
5209
5210 C->set_has_split_ifs(true); // Has chance for split-if optimization
5211 set_result(z);
5212 return true;
5213 }
5214
5215 //-------------inline_squareToLen------------------------------------
5216 bool LibraryCallKit::inline_squareToLen() {
5217 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5218
5219 address stubAddr = StubRoutines::squareToLen();
5220 if (stubAddr == NULL) {
5221 return false; // Intrinsic's stub is not implemented on this platform
5222 }
5223 const char* stubName = "squareToLen";
5224
5225 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5226
5227 Node* x = argument(0);
5228 Node* len = argument(1);
5229 Node* z = argument(2);
5230 Node* zlen = argument(3);
5231
5232 x = must_be_not_null(x, true);
5233 z = must_be_not_null(z, true);
5234
5235 x = access_resolve(x, ACCESS_READ);
5236 z = access_resolve(z, ACCESS_WRITE);
5237
5238 const Type* x_type = x->Value(&_gvn);
5239 const Type* z_type = z->Value(&_gvn);
5240 const TypeAryPtr* top_x = x_type->isa_aryptr();
5241 const TypeAryPtr* top_z = z_type->isa_aryptr();
5242 if (top_x == NULL || top_x->klass() == NULL ||
5243 top_z == NULL || top_z->klass() == NULL) {
5244 // failed array check
5245 return false;
5246 }
5247
5248 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5249 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5250 if (x_elem != T_INT || z_elem != T_INT) {
5251 return false;
5252 }
5253
5254
5255 Node* x_start = array_element_address(x, intcon(0), x_elem);
5256 Node* z_start = array_element_address(z, intcon(0), z_elem);
5257
5258 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5259 OptoRuntime::squareToLen_Type(),
5260 stubAddr, stubName, TypePtr::BOTTOM,
5261 x_start, len, z_start, zlen);
5262
5263 set_result(z);
5264 return true;
5265 }
5266
5267 //-------------inline_mulAdd------------------------------------------
5268 bool LibraryCallKit::inline_mulAdd() {
5269 assert(UseMulAddIntrinsic, "not implemented on this platform");
5270
5271 address stubAddr = StubRoutines::mulAdd();
5272 if (stubAddr == NULL) {
5273 return false; // Intrinsic's stub is not implemented on this platform
5274 }
5275 const char* stubName = "mulAdd";
5276
5277 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5278
5279 Node* out = argument(0);
5280 Node* in = argument(1);
5281 Node* offset = argument(2);
5282 Node* len = argument(3);
5283 Node* k = argument(4);
5284
5285 out = must_be_not_null(out, true);
5286
5287 in = access_resolve(in, ACCESS_READ);
5288 out = access_resolve(out, ACCESS_WRITE);
5289
5290 const Type* out_type = out->Value(&_gvn);
5291 const Type* in_type = in->Value(&_gvn);
5292 const TypeAryPtr* top_out = out_type->isa_aryptr();
5293 const TypeAryPtr* top_in = in_type->isa_aryptr();
5294 if (top_out == NULL || top_out->klass() == NULL ||
5295 top_in == NULL || top_in->klass() == NULL) {
5296 // failed array check
5297 return false;
5298 }
5299
5300 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5301 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5302 if (out_elem != T_INT || in_elem != T_INT) {
5303 return false;
5304 }
5305
5306 Node* outlen = load_array_length(out);
5307 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5308 Node* out_start = array_element_address(out, intcon(0), out_elem);
5309 Node* in_start = array_element_address(in, intcon(0), in_elem);
5310
5311 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5312 OptoRuntime::mulAdd_Type(),
5313 stubAddr, stubName, TypePtr::BOTTOM,
5314 out_start,in_start, new_offset, len, k);
5315 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5316 set_result(result);
5317 return true;
5318 }
5319
5320 //-------------inline_montgomeryMultiply-----------------------------------
5321 bool LibraryCallKit::inline_montgomeryMultiply() {
5322 address stubAddr = StubRoutines::montgomeryMultiply();
5323 if (stubAddr == NULL) {
5324 return false; // Intrinsic's stub is not implemented on this platform
5325 }
5326
5327 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5328 const char* stubName = "montgomery_multiply";
5329
5330 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5331
5332 Node* a = argument(0);
5333 Node* b = argument(1);
5334 Node* n = argument(2);
5335 Node* len = argument(3);
5336 Node* inv = argument(4);
5337 Node* m = argument(6);
5338
5339 a = access_resolve(a, ACCESS_READ);
5340 b = access_resolve(b, ACCESS_READ);
5341 n = access_resolve(n, ACCESS_READ);
5342 m = access_resolve(m, ACCESS_WRITE);
5343
5344 const Type* a_type = a->Value(&_gvn);
5345 const TypeAryPtr* top_a = a_type->isa_aryptr();
5346 const Type* b_type = b->Value(&_gvn);
5347 const TypeAryPtr* top_b = b_type->isa_aryptr();
5348 const Type* n_type = a->Value(&_gvn);
5349 const TypeAryPtr* top_n = n_type->isa_aryptr();
5350 const Type* m_type = a->Value(&_gvn);
5351 const TypeAryPtr* top_m = m_type->isa_aryptr();
5352 if (top_a == NULL || top_a->klass() == NULL ||
5353 top_b == NULL || top_b->klass() == NULL ||
5354 top_n == NULL || top_n->klass() == NULL ||
5355 top_m == NULL || top_m->klass() == NULL) {
5356 // failed array check
5357 return false;
5358 }
5359
5360 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5361 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5362 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5363 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5364 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5365 return false;
5366 }
5367
5368 // Make the call
5369 {
5370 Node* a_start = array_element_address(a, intcon(0), a_elem);
5371 Node* b_start = array_element_address(b, intcon(0), b_elem);
5372 Node* n_start = array_element_address(n, intcon(0), n_elem);
5373 Node* m_start = array_element_address(m, intcon(0), m_elem);
5374
5375 Node* call = make_runtime_call(RC_LEAF,
5376 OptoRuntime::montgomeryMultiply_Type(),
5377 stubAddr, stubName, TypePtr::BOTTOM,
5378 a_start, b_start, n_start, len, inv, top(),
5379 m_start);
5380 set_result(m);
5381 }
5382
5383 return true;
5384 }
5385
5386 bool LibraryCallKit::inline_montgomerySquare() {
5387 address stubAddr = StubRoutines::montgomerySquare();
5388 if (stubAddr == NULL) {
5389 return false; // Intrinsic's stub is not implemented on this platform
5390 }
5391
5392 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5393 const char* stubName = "montgomery_square";
5394
5395 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5396
5397 Node* a = argument(0);
5398 Node* n = argument(1);
5399 Node* len = argument(2);
5400 Node* inv = argument(3);
5401 Node* m = argument(5);
5402
5403 a = access_resolve(a, ACCESS_READ);
5404 n = access_resolve(n, ACCESS_READ);
5405 m = access_resolve(m, ACCESS_WRITE);
5406
5407 const Type* a_type = a->Value(&_gvn);
5408 const TypeAryPtr* top_a = a_type->isa_aryptr();
5409 const Type* n_type = a->Value(&_gvn);
5410 const TypeAryPtr* top_n = n_type->isa_aryptr();
5411 const Type* m_type = a->Value(&_gvn);
5412 const TypeAryPtr* top_m = m_type->isa_aryptr();
5413 if (top_a == NULL || top_a->klass() == NULL ||
5414 top_n == NULL || top_n->klass() == NULL ||
5415 top_m == NULL || top_m->klass() == NULL) {
5416 // failed array check
5417 return false;
5418 }
5419
5420 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5421 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5422 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5423 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5424 return false;
5425 }
5426
5427 // Make the call
5428 {
5429 Node* a_start = array_element_address(a, intcon(0), a_elem);
5430 Node* n_start = array_element_address(n, intcon(0), n_elem);
5431 Node* m_start = array_element_address(m, intcon(0), m_elem);
5432
5433 Node* call = make_runtime_call(RC_LEAF,
5434 OptoRuntime::montgomerySquare_Type(),
5435 stubAddr, stubName, TypePtr::BOTTOM,
5436 a_start, n_start, len, inv, top(),
5437 m_start);
5438 set_result(m);
5439 }
5440
5441 return true;
5442 }
5443
5444 //-------------inline_vectorizedMismatch------------------------------
5445 bool LibraryCallKit::inline_vectorizedMismatch() {
5446 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5447
5448 address stubAddr = StubRoutines::vectorizedMismatch();
5449 if (stubAddr == NULL) {
5450 return false; // Intrinsic's stub is not implemented on this platform
5451 }
5452 const char* stubName = "vectorizedMismatch";
5453 int size_l = callee()->signature()->size();
5454 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5455
5456 Node* obja = argument(0);
5457 Node* aoffset = argument(1);
5458 Node* objb = argument(3);
5459 Node* boffset = argument(4);
5460 Node* length = argument(6);
5461 Node* scale = argument(7);
5462
5463 const Type* a_type = obja->Value(&_gvn);
5464 const Type* b_type = objb->Value(&_gvn);
5465 const TypeAryPtr* top_a = a_type->isa_aryptr();
5466 const TypeAryPtr* top_b = b_type->isa_aryptr();
5467 if (top_a == NULL || top_a->klass() == NULL ||
5468 top_b == NULL || top_b->klass() == NULL) {
5469 // failed array check
5470 return false;
5471 }
5472
5473 Node* call;
5474 jvms()->set_should_reexecute(true);
5475
5476 obja = access_resolve(obja, ACCESS_READ);
5477 objb = access_resolve(objb, ACCESS_READ);
5478 Node* obja_adr = make_unsafe_address(obja, aoffset, ACCESS_READ);
5479 Node* objb_adr = make_unsafe_address(objb, boffset, ACCESS_READ);
5480
5481 call = make_runtime_call(RC_LEAF,
5482 OptoRuntime::vectorizedMismatch_Type(),
5483 stubAddr, stubName, TypePtr::BOTTOM,
5484 obja_adr, objb_adr, length, scale);
5485
5486 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5487 set_result(result);
5488 return true;
5489 }
5490
5491 /**
5492 * Calculate CRC32 for byte.
5493 * int java.util.zip.CRC32.update(int crc, int b)
5494 */
5495 bool LibraryCallKit::inline_updateCRC32() {
5496 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5497 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5498 // no receiver since it is static method
5499 Node* crc = argument(0); // type: int
5500 Node* b = argument(1); // type: int
5501
5502 /*
5503 * int c = ~ crc;
5504 * b = timesXtoThe32[(b ^ c) & 0xFF];
5505 * b = b ^ (c >>> 8);
5506 * crc = ~b;
5507 */
5508
5509 Node* M1 = intcon(-1);
5510 crc = _gvn.transform(new XorINode(crc, M1));
5511 Node* result = _gvn.transform(new XorINode(crc, b));
5512 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5513
5514 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5515 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5516 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5517 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5518
5519 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5520 result = _gvn.transform(new XorINode(crc, result));
5521 result = _gvn.transform(new XorINode(result, M1));
5522 set_result(result);
5523 return true;
5524 }
5525
5526 /**
5527 * Calculate CRC32 for byte[] array.
5528 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5529 */
5530 bool LibraryCallKit::inline_updateBytesCRC32() {
5531 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5532 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5533 // no receiver since it is static method
5534 Node* crc = argument(0); // type: int
5535 Node* src = argument(1); // type: oop
5536 Node* offset = argument(2); // type: int
5537 Node* length = argument(3); // type: int
5538
5539 const Type* src_type = src->Value(&_gvn);
5540 const TypeAryPtr* top_src = src_type->isa_aryptr();
5541 if (top_src == NULL || top_src->klass() == NULL) {
5542 // failed array check
5543 return false;
5544 }
5545
5546 // Figure out the size and type of the elements we will be copying.
5547 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5548 if (src_elem != T_BYTE) {
5549 return false;
5550 }
5551
5552 // 'src_start' points to src array + scaled offset
5553 src = must_be_not_null(src, true);
5554 src = access_resolve(src, ACCESS_READ);
5555 Node* src_start = array_element_address(src, offset, src_elem);
5556
5557 // We assume that range check is done by caller.
5558 // TODO: generate range check (offset+length < src.length) in debug VM.
5559
5560 // Call the stub.
5561 address stubAddr = StubRoutines::updateBytesCRC32();
5562 const char *stubName = "updateBytesCRC32";
5563
5564 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5565 stubAddr, stubName, TypePtr::BOTTOM,
5566 crc, src_start, length);
5567 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5568 set_result(result);
5569 return true;
5570 }
5571
5572 /**
5573 * Calculate CRC32 for ByteBuffer.
5574 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5575 */
5576 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5577 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5578 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5579 // no receiver since it is static method
5580 Node* crc = argument(0); // type: int
5581 Node* src = argument(1); // type: long
5582 Node* offset = argument(3); // type: int
5583 Node* length = argument(4); // type: int
5584
5585 src = ConvL2X(src); // adjust Java long to machine word
5586 Node* base = _gvn.transform(new CastX2PNode(src));
5587 offset = ConvI2X(offset);
5588
5589 // 'src_start' points to src array + scaled offset
5590 Node* src_start = basic_plus_adr(top(), base, offset);
5591
5592 // Call the stub.
5593 address stubAddr = StubRoutines::updateBytesCRC32();
5594 const char *stubName = "updateBytesCRC32";
5595
5596 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5597 stubAddr, stubName, TypePtr::BOTTOM,
5598 crc, src_start, length);
5599 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5600 set_result(result);
5601 return true;
5602 }
5603
5604 //------------------------------get_table_from_crc32c_class-----------------------
5605 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5606 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5607 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5608
5609 return table;
5610 }
5611
5612 //------------------------------inline_updateBytesCRC32C-----------------------
5613 //
5614 // Calculate CRC32C for byte[] array.
5615 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5616 //
5617 bool LibraryCallKit::inline_updateBytesCRC32C() {
5618 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5619 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5620 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5621 // no receiver since it is a static method
5622 Node* crc = argument(0); // type: int
5623 Node* src = argument(1); // type: oop
5624 Node* offset = argument(2); // type: int
5625 Node* end = argument(3); // type: int
5626
5627 Node* length = _gvn.transform(new SubINode(end, offset));
5628
5629 const Type* src_type = src->Value(&_gvn);
5630 const TypeAryPtr* top_src = src_type->isa_aryptr();
5631 if (top_src == NULL || top_src->klass() == NULL) {
5632 // failed array check
5633 return false;
5634 }
5635
5636 // Figure out the size and type of the elements we will be copying.
5637 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5638 if (src_elem != T_BYTE) {
5639 return false;
5640 }
5641
5642 // 'src_start' points to src array + scaled offset
5643 src = must_be_not_null(src, true);
5644 src = access_resolve(src, ACCESS_READ);
5645 Node* src_start = array_element_address(src, offset, src_elem);
5646
5647 // static final int[] byteTable in class CRC32C
5648 Node* table = get_table_from_crc32c_class(callee()->holder());
5649 table = must_be_not_null(table, true);
5650 table = access_resolve(table, ACCESS_READ);
5651 Node* table_start = array_element_address(table, intcon(0), T_INT);
5652
5653 // We assume that range check is done by caller.
5654 // TODO: generate range check (offset+length < src.length) in debug VM.
5655
5656 // Call the stub.
5657 address stubAddr = StubRoutines::updateBytesCRC32C();
5658 const char *stubName = "updateBytesCRC32C";
5659
5660 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5661 stubAddr, stubName, TypePtr::BOTTOM,
5662 crc, src_start, length, table_start);
5663 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5664 set_result(result);
5665 return true;
5666 }
5667
5668 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5669 //
5670 // Calculate CRC32C for DirectByteBuffer.
5671 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5672 //
5673 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5674 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5675 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5676 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5677 // no receiver since it is a static method
5678 Node* crc = argument(0); // type: int
5679 Node* src = argument(1); // type: long
5680 Node* offset = argument(3); // type: int
5681 Node* end = argument(4); // type: int
5682
5683 Node* length = _gvn.transform(new SubINode(end, offset));
5684
5685 src = ConvL2X(src); // adjust Java long to machine word
5686 Node* base = _gvn.transform(new CastX2PNode(src));
5687 offset = ConvI2X(offset);
5688
5689 // 'src_start' points to src array + scaled offset
5690 Node* src_start = basic_plus_adr(top(), base, offset);
5691
5692 // static final int[] byteTable in class CRC32C
5693 Node* table = get_table_from_crc32c_class(callee()->holder());
5694 table = must_be_not_null(table, true);
5695 table = access_resolve(table, ACCESS_READ);
5696 Node* table_start = array_element_address(table, intcon(0), T_INT);
5697
5698 // Call the stub.
5699 address stubAddr = StubRoutines::updateBytesCRC32C();
5700 const char *stubName = "updateBytesCRC32C";
5701
5702 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5703 stubAddr, stubName, TypePtr::BOTTOM,
5704 crc, src_start, length, table_start);
5705 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5706 set_result(result);
5707 return true;
5708 }
5709
5710 //------------------------------inline_updateBytesAdler32----------------------
5711 //
5712 // Calculate Adler32 checksum for byte[] array.
5713 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5714 //
5715 bool LibraryCallKit::inline_updateBytesAdler32() {
5716 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5717 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5718 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5719 // no receiver since it is static method
5720 Node* crc = argument(0); // type: int
5721 Node* src = argument(1); // type: oop
5722 Node* offset = argument(2); // type: int
5723 Node* length = argument(3); // type: int
5724
5725 const Type* src_type = src->Value(&_gvn);
5726 const TypeAryPtr* top_src = src_type->isa_aryptr();
5727 if (top_src == NULL || top_src->klass() == NULL) {
5728 // failed array check
5729 return false;
5730 }
5731
5732 // Figure out the size and type of the elements we will be copying.
5733 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5734 if (src_elem != T_BYTE) {
5735 return false;
5736 }
5737
5738 // 'src_start' points to src array + scaled offset
5739 src = access_resolve(src, ACCESS_READ);
5740 Node* src_start = array_element_address(src, offset, src_elem);
5741
5742 // We assume that range check is done by caller.
5743 // TODO: generate range check (offset+length < src.length) in debug VM.
5744
5745 // Call the stub.
5746 address stubAddr = StubRoutines::updateBytesAdler32();
5747 const char *stubName = "updateBytesAdler32";
5748
5749 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5750 stubAddr, stubName, TypePtr::BOTTOM,
5751 crc, src_start, length);
5752 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5753 set_result(result);
5754 return true;
5755 }
5756
5757 //------------------------------inline_updateByteBufferAdler32---------------
5758 //
5759 // Calculate Adler32 checksum for DirectByteBuffer.
5760 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5761 //
5762 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5763 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5764 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5765 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5766 // no receiver since it is static method
5767 Node* crc = argument(0); // type: int
5768 Node* src = argument(1); // type: long
5769 Node* offset = argument(3); // type: int
5770 Node* length = argument(4); // type: int
5771
5772 src = ConvL2X(src); // adjust Java long to machine word
5773 Node* base = _gvn.transform(new CastX2PNode(src));
5774 offset = ConvI2X(offset);
5775
5776 // 'src_start' points to src array + scaled offset
5777 Node* src_start = basic_plus_adr(top(), base, offset);
5778
5779 // Call the stub.
5780 address stubAddr = StubRoutines::updateBytesAdler32();
5781 const char *stubName = "updateBytesAdler32";
5782
5783 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5784 stubAddr, stubName, TypePtr::BOTTOM,
5785 crc, src_start, length);
5786
5787 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5788 set_result(result);
5789 return true;
5790 }
5791
5792 //----------------------------inline_reference_get----------------------------
5793 // public T java.lang.ref.Reference.get();
5794 bool LibraryCallKit::inline_reference_get() {
5795 const int referent_offset = java_lang_ref_Reference::referent_offset;
5796 guarantee(referent_offset > 0, "should have already been set");
5797
5798 // Get the argument:
5799 Node* reference_obj = null_check_receiver();
5800 if (stopped()) return true;
5801
5802 const TypeInstPtr* tinst = _gvn.type(reference_obj)->isa_instptr();
5803 assert(tinst != NULL, "obj is null");
5804 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5805 ciInstanceKlass* referenceKlass = tinst->klass()->as_instance_klass();
5806 ciField* field = referenceKlass->get_field_by_name(ciSymbol::make("referent"),
5807 ciSymbol::make("Ljava/lang/Object;"),
5808 false);
5809 assert (field != NULL, "undefined field");
5810
5811 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5812 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5813
5814 ciInstanceKlass* klass = env()->Object_klass();
5815 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5816
5817 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
5818 Node* result = access_load_at(reference_obj, adr, adr_type, object_type, T_OBJECT, decorators);
5819 // Add memory barrier to prevent commoning reads from this field
5820 // across safepoint since GC can change its value.
5821 insert_mem_bar(Op_MemBarCPUOrder);
5822
5823 set_result(result);
5824 return true;
5825 }
5826
5827
5828 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5829 bool is_exact=true, bool is_static=false,
5830 ciInstanceKlass * fromKls=NULL) {
5831 if (fromKls == NULL) {
5832 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5833 assert(tinst != NULL, "obj is null");
5834 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5835 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5836 fromKls = tinst->klass()->as_instance_klass();
5837 } else {
5838 assert(is_static, "only for static field access");
5839 }
5840 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5841 ciSymbol::make(fieldTypeString),
5842 is_static);
5843
5844 assert (field != NULL, "undefined field");
5845 if (field == NULL) return (Node *) NULL;
5846
5847 if (is_static) {
5848 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5849 fromObj = makecon(tip);
5850 }
5851
5852 // Next code copied from Parse::do_get_xxx():
5853
5854 // Compute address and memory type.
5855 int offset = field->offset_in_bytes();
5856 bool is_vol = field->is_volatile();
5857 ciType* field_klass = field->type();
5858 assert(field_klass->is_loaded(), "should be loaded");
5859 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5860 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5861 BasicType bt = field->layout_type();
5862
5863 // Build the resultant type of the load
5864 const Type *type;
5865 if (bt == T_OBJECT) {
5866 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5867 } else {
5868 type = Type::get_const_basic_type(bt);
5869 }
5870
5871 DecoratorSet decorators = IN_HEAP;
5872
5873 if (is_vol) {
5874 decorators |= MO_SEQ_CST;
5875 }
5876
5877 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
5878 }
5879
5880 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5881 bool is_exact = true, bool is_static = false,
5882 ciInstanceKlass * fromKls = NULL) {
5883 if (fromKls == NULL) {
5884 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5885 assert(tinst != NULL, "obj is null");
5886 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5887 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5888 fromKls = tinst->klass()->as_instance_klass();
5889 }
5890 else {
5891 assert(is_static, "only for static field access");
5892 }
5893 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5894 ciSymbol::make(fieldTypeString),
5895 is_static);
5896
5897 assert(field != NULL, "undefined field");
5898 assert(!field->is_volatile(), "not defined for volatile fields");
5899
5900 if (is_static) {
5901 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5902 fromObj = makecon(tip);
5903 }
5904
5905 // Next code copied from Parse::do_get_xxx():
5906
5907 // Compute address and memory type.
5908 int offset = field->offset_in_bytes();
5909 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5910
5911 return adr;
5912 }
5913
5914 //------------------------------inline_aescrypt_Block-----------------------
5915 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
5916 address stubAddr = NULL;
5917 const char *stubName;
5918 assert(UseAES, "need AES instruction support");
5919
5920 switch(id) {
5921 case vmIntrinsics::_aescrypt_encryptBlock:
5922 stubAddr = StubRoutines::aescrypt_encryptBlock();
5923 stubName = "aescrypt_encryptBlock";
5924 break;
5925 case vmIntrinsics::_aescrypt_decryptBlock:
5926 stubAddr = StubRoutines::aescrypt_decryptBlock();
5927 stubName = "aescrypt_decryptBlock";
5928 break;
5929 default:
5930 break;
5931 }
5932 if (stubAddr == NULL) return false;
5933
5934 Node* aescrypt_object = argument(0);
5935 Node* src = argument(1);
5936 Node* src_offset = argument(2);
5937 Node* dest = argument(3);
5938 Node* dest_offset = argument(4);
5939
5940 src = must_be_not_null(src, true);
5941 dest = must_be_not_null(dest, true);
5942
5943 src = access_resolve(src, ACCESS_READ);
5944 dest = access_resolve(dest, ACCESS_WRITE);
5945
5946 // (1) src and dest are arrays.
5947 const Type* src_type = src->Value(&_gvn);
5948 const Type* dest_type = dest->Value(&_gvn);
5949 const TypeAryPtr* top_src = src_type->isa_aryptr();
5950 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5951 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5952
5953 // for the quick and dirty code we will skip all the checks.
5954 // we are just trying to get the call to be generated.
5955 Node* src_start = src;
5956 Node* dest_start = dest;
5957 if (src_offset != NULL || dest_offset != NULL) {
5958 assert(src_offset != NULL && dest_offset != NULL, "");
5959 src_start = array_element_address(src, src_offset, T_BYTE);
5960 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5961 }
5962
5963 // now need to get the start of its expanded key array
5964 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5965 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5966 if (k_start == NULL) return false;
5967
5968 if (Matcher::pass_original_key_for_aes()) {
5969 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5970 // compatibility issues between Java key expansion and SPARC crypto instructions
5971 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5972 if (original_k_start == NULL) return false;
5973
5974 // Call the stub.
5975 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5976 stubAddr, stubName, TypePtr::BOTTOM,
5977 src_start, dest_start, k_start, original_k_start);
5978 } else {
5979 // Call the stub.
5980 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5981 stubAddr, stubName, TypePtr::BOTTOM,
5982 src_start, dest_start, k_start);
5983 }
5984
5985 return true;
5986 }
5987
5988 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
5989 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
5990 address stubAddr = NULL;
5991 const char *stubName = NULL;
5992
5993 assert(UseAES, "need AES instruction support");
5994
5995 switch(id) {
5996 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
5997 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
5998 stubName = "cipherBlockChaining_encryptAESCrypt";
5999 break;
6000 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6001 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6002 stubName = "cipherBlockChaining_decryptAESCrypt";
6003 break;
6004 default:
6005 break;
6006 }
6007 if (stubAddr == NULL) return false;
6008
6009 Node* cipherBlockChaining_object = argument(0);
6010 Node* src = argument(1);
6011 Node* src_offset = argument(2);
6012 Node* len = argument(3);
6013 Node* dest = argument(4);
6014 Node* dest_offset = argument(5);
6015
6016 src = must_be_not_null(src, false);
6017 dest = must_be_not_null(dest, false);
6018
6019 src = access_resolve(src, ACCESS_READ);
6020 dest = access_resolve(dest, ACCESS_WRITE);
6021
6022 // (1) src and dest are arrays.
6023 const Type* src_type = src->Value(&_gvn);
6024 const Type* dest_type = dest->Value(&_gvn);
6025 const TypeAryPtr* top_src = src_type->isa_aryptr();
6026 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6027 assert (top_src != NULL && top_src->klass() != NULL
6028 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6029
6030 // checks are the responsibility of the caller
6031 Node* src_start = src;
6032 Node* dest_start = dest;
6033 if (src_offset != NULL || dest_offset != NULL) {
6034 assert(src_offset != NULL && dest_offset != NULL, "");
6035 src_start = array_element_address(src, src_offset, T_BYTE);
6036 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6037 }
6038
6039 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6040 // (because of the predicated logic executed earlier).
6041 // so we cast it here safely.
6042 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6043
6044 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6045 if (embeddedCipherObj == NULL) return false;
6046
6047 // cast it to what we know it will be at runtime
6048 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6049 assert(tinst != NULL, "CBC obj is null");
6050 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
6051 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6052 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6053
6054 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6055 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6056 const TypeOopPtr* xtype = aklass->as_instance_type();
6057 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6058 aescrypt_object = _gvn.transform(aescrypt_object);
6059
6060 // we need to get the start of the aescrypt_object's expanded key array
6061 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6062 if (k_start == NULL) return false;
6063
6064 // similarly, get the start address of the r vector
6065 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
6066 if (objRvec == NULL) return false;
6067 objRvec = access_resolve(objRvec, ACCESS_WRITE);
6068 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6069
6070 Node* cbcCrypt;
6071 if (Matcher::pass_original_key_for_aes()) {
6072 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6073 // compatibility issues between Java key expansion and SPARC crypto instructions
6074 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6075 if (original_k_start == NULL) return false;
6076
6077 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
6078 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6079 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6080 stubAddr, stubName, TypePtr::BOTTOM,
6081 src_start, dest_start, k_start, r_start, len, original_k_start);
6082 } else {
6083 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6084 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6085 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6086 stubAddr, stubName, TypePtr::BOTTOM,
6087 src_start, dest_start, k_start, r_start, len);
6088 }
6089
6090 // return cipher length (int)
6091 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6092 set_result(retvalue);
6093 return true;
6094 }
6095
6096 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
6097 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
6098 address stubAddr = NULL;
6099 const char *stubName = NULL;
6100
6101 assert(UseAES, "need AES instruction support");
6102
6103 switch (id) {
6104 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
6105 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
6106 stubName = "electronicCodeBook_encryptAESCrypt";
6107 break;
6108 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
6109 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
6110 stubName = "electronicCodeBook_decryptAESCrypt";
6111 break;
6112 default:
6113 break;
6114 }
6115
6116 if (stubAddr == NULL) return false;
6117
6118 Node* electronicCodeBook_object = argument(0);
6119 Node* src = argument(1);
6120 Node* src_offset = argument(2);
6121 Node* len = argument(3);
6122 Node* dest = argument(4);
6123 Node* dest_offset = argument(5);
6124
6125 // (1) src and dest are arrays.
6126 const Type* src_type = src->Value(&_gvn);
6127 const Type* dest_type = dest->Value(&_gvn);
6128 const TypeAryPtr* top_src = src_type->isa_aryptr();
6129 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6130 assert(top_src != NULL && top_src->klass() != NULL
6131 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6132
6133 // checks are the responsibility of the caller
6134 Node* src_start = src;
6135 Node* dest_start = dest;
6136 if (src_offset != NULL || dest_offset != NULL) {
6137 assert(src_offset != NULL && dest_offset != NULL, "");
6138 src_start = array_element_address(src, src_offset, T_BYTE);
6139 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6140 }
6141
6142 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6143 // (because of the predicated logic executed earlier).
6144 // so we cast it here safely.
6145 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6146
6147 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6148 if (embeddedCipherObj == NULL) return false;
6149
6150 // cast it to what we know it will be at runtime
6151 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6152 assert(tinst != NULL, "ECB obj is null");
6153 assert(tinst->klass()->is_loaded(), "ECB obj is not loaded");
6154 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6155 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6156
6157 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6158 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6159 const TypeOopPtr* xtype = aklass->as_instance_type();
6160 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6161 aescrypt_object = _gvn.transform(aescrypt_object);
6162
6163 // we need to get the start of the aescrypt_object's expanded key array
6164 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6165 if (k_start == NULL) return false;
6166
6167 Node* ecbCrypt;
6168 if (Matcher::pass_original_key_for_aes()) {
6169 // no SPARC version for AES/ECB intrinsics now.
6170 return false;
6171 }
6172 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6173 ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6174 OptoRuntime::electronicCodeBook_aescrypt_Type(),
6175 stubAddr, stubName, TypePtr::BOTTOM,
6176 src_start, dest_start, k_start, len);
6177
6178 // return cipher length (int)
6179 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6180 set_result(retvalue);
6181 return true;
6182 }
6183
6184 //------------------------------inline_counterMode_AESCrypt-----------------------
6185 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6186 assert(UseAES, "need AES instruction support");
6187 if (!UseAESCTRIntrinsics) return false;
6188
6189 address stubAddr = NULL;
6190 const char *stubName = NULL;
6191 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6192 stubAddr = StubRoutines::counterMode_AESCrypt();
6193 stubName = "counterMode_AESCrypt";
6194 }
6195 if (stubAddr == NULL) return false;
6196
6197 Node* counterMode_object = argument(0);
6198 Node* src = argument(1);
6199 Node* src_offset = argument(2);
6200 Node* len = argument(3);
6201 Node* dest = argument(4);
6202 Node* dest_offset = argument(5);
6203
6204 src = access_resolve(src, ACCESS_READ);
6205 dest = access_resolve(dest, ACCESS_WRITE);
6206 counterMode_object = access_resolve(counterMode_object, ACCESS_WRITE);
6207
6208 // (1) src and dest are arrays.
6209 const Type* src_type = src->Value(&_gvn);
6210 const Type* dest_type = dest->Value(&_gvn);
6211 const TypeAryPtr* top_src = src_type->isa_aryptr();
6212 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6213 assert(top_src != NULL && top_src->klass() != NULL &&
6214 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6215
6216 // checks are the responsibility of the caller
6217 Node* src_start = src;
6218 Node* dest_start = dest;
6219 if (src_offset != NULL || dest_offset != NULL) {
6220 assert(src_offset != NULL && dest_offset != NULL, "");
6221 src_start = array_element_address(src, src_offset, T_BYTE);
6222 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6223 }
6224
6225 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6226 // (because of the predicated logic executed earlier).
6227 // so we cast it here safely.
6228 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6229 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6230 if (embeddedCipherObj == NULL) return false;
6231 // cast it to what we know it will be at runtime
6232 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6233 assert(tinst != NULL, "CTR obj is null");
6234 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6235 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6236 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6237 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6238 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6239 const TypeOopPtr* xtype = aklass->as_instance_type();
6240 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6241 aescrypt_object = _gvn.transform(aescrypt_object);
6242 // we need to get the start of the aescrypt_object's expanded key array
6243 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6244 if (k_start == NULL) return false;
6245 // similarly, get the start address of the r vector
6246 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6247 if (obj_counter == NULL) return false;
6248 obj_counter = access_resolve(obj_counter, ACCESS_WRITE);
6249 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6250
6251 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6252 if (saved_encCounter == NULL) return false;
6253 saved_encCounter = access_resolve(saved_encCounter, ACCESS_WRITE);
6254 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6255 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6256
6257 Node* ctrCrypt;
6258 if (Matcher::pass_original_key_for_aes()) {
6259 // no SPARC version for AES/CTR intrinsics now.
6260 return false;
6261 }
6262 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6263 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6264 OptoRuntime::counterMode_aescrypt_Type(),
6265 stubAddr, stubName, TypePtr::BOTTOM,
6266 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6267
6268 // return cipher length (int)
6269 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6270 set_result(retvalue);
6271 return true;
6272 }
6273
6274 //------------------------------get_key_start_from_aescrypt_object-----------------------
6275 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6276 #if defined(PPC64) || defined(S390)
6277 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6278 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6279 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6280 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6281 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6282 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6283 if (objSessionK == NULL) {
6284 return (Node *) NULL;
6285 }
6286 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6287 #else
6288 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6289 #endif // PPC64
6290 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6291 if (objAESCryptKey == NULL) return (Node *) NULL;
6292
6293 // now have the array, need to get the start address of the K array
6294 objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ);
6295 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6296 return k_start;
6297 }
6298
6299 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6300 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6301 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6302 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6303 if (objAESCryptKey == NULL) return (Node *) NULL;
6304
6305 // now have the array, need to get the start address of the lastKey array
6306 objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ);
6307 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6308 return original_k_start;
6309 }
6310
6311 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6312 // Return node representing slow path of predicate check.
6313 // the pseudo code we want to emulate with this predicate is:
6314 // for encryption:
6315 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6316 // for decryption:
6317 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6318 // note cipher==plain is more conservative than the original java code but that's OK
6319 //
6320 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6321 // The receiver was checked for NULL already.
6322 Node* objCBC = argument(0);
6323
6324 Node* src = argument(1);
6325 Node* dest = argument(4);
6326
6327 // Load embeddedCipher field of CipherBlockChaining object.
6328 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6329
6330 // get AESCrypt klass for instanceOf check
6331 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6332 // will have same classloader as CipherBlockChaining object
6333 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6334 assert(tinst != NULL, "CBCobj is null");
6335 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6336
6337 // we want to do an instanceof comparison against the AESCrypt class
6338 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6339 if (!klass_AESCrypt->is_loaded()) {
6340 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6341 Node* ctrl = control();
6342 set_control(top()); // no regular fast path
6343 return ctrl;
6344 }
6345
6346 src = must_be_not_null(src, true);
6347 dest = must_be_not_null(dest, true);
6348
6349 // Resolve oops to stable for CmpP below.
6350 src = access_resolve(src, 0);
6351 dest = access_resolve(dest, 0);
6352
6353 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6354
6355 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6356 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6357 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6358
6359 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6360
6361 // for encryption, we are done
6362 if (!decrypting)
6363 return instof_false; // even if it is NULL
6364
6365 // for decryption, we need to add a further check to avoid
6366 // taking the intrinsic path when cipher and plain are the same
6367 // see the original java code for why.
6368 RegionNode* region = new RegionNode(3);
6369 region->init_req(1, instof_false);
6370
6371 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6372 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6373 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6374 region->init_req(2, src_dest_conjoint);
6375
6376 record_for_igvn(region);
6377 return _gvn.transform(region);
6378 }
6379
6380 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
6381 // Return node representing slow path of predicate check.
6382 // the pseudo code we want to emulate with this predicate is:
6383 // for encryption:
6384 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6385 // for decryption:
6386 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6387 // note cipher==plain is more conservative than the original java code but that's OK
6388 //
6389 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
6390 // The receiver was checked for NULL already.
6391 Node* objECB = argument(0);
6392
6393 // Load embeddedCipher field of ElectronicCodeBook object.
6394 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6395
6396 // get AESCrypt klass for instanceOf check
6397 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6398 // will have same classloader as ElectronicCodeBook object
6399 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
6400 assert(tinst != NULL, "ECBobj is null");
6401 assert(tinst->klass()->is_loaded(), "ECBobj is not loaded");
6402
6403 // we want to do an instanceof comparison against the AESCrypt class
6404 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6405 if (!klass_AESCrypt->is_loaded()) {
6406 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6407 Node* ctrl = control();
6408 set_control(top()); // no regular fast path
6409 return ctrl;
6410 }
6411 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6412
6413 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6414 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6415 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6416
6417 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6418
6419 // for encryption, we are done
6420 if (!decrypting)
6421 return instof_false; // even if it is NULL
6422
6423 // for decryption, we need to add a further check to avoid
6424 // taking the intrinsic path when cipher and plain are the same
6425 // see the original java code for why.
6426 RegionNode* region = new RegionNode(3);
6427 region->init_req(1, instof_false);
6428 Node* src = argument(1);
6429 Node* dest = argument(4);
6430 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6431 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6432 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6433 region->init_req(2, src_dest_conjoint);
6434
6435 record_for_igvn(region);
6436 return _gvn.transform(region);
6437 }
6438
6439 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6440 // Return node representing slow path of predicate check.
6441 // the pseudo code we want to emulate with this predicate is:
6442 // for encryption:
6443 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6444 // for decryption:
6445 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6446 // note cipher==plain is more conservative than the original java code but that's OK
6447 //
6448
6449 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6450 // The receiver was checked for NULL already.
6451 Node* objCTR = argument(0);
6452
6453 // Load embeddedCipher field of CipherBlockChaining object.
6454 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6455
6456 // get AESCrypt klass for instanceOf check
6457 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6458 // will have same classloader as CipherBlockChaining object
6459 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6460 assert(tinst != NULL, "CTRobj is null");
6461 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6462
6463 // we want to do an instanceof comparison against the AESCrypt class
6464 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6465 if (!klass_AESCrypt->is_loaded()) {
6466 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6467 Node* ctrl = control();
6468 set_control(top()); // no regular fast path
6469 return ctrl;
6470 }
6471
6472 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6473 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6474 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6475 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6476 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6477
6478 return instof_false; // even if it is NULL
6479 }
6480
6481 //------------------------------inline_ghash_processBlocks
6482 bool LibraryCallKit::inline_ghash_processBlocks() {
6483 address stubAddr;
6484 const char *stubName;
6485 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6486
6487 stubAddr = StubRoutines::ghash_processBlocks();
6488 stubName = "ghash_processBlocks";
6489
6490 Node* data = argument(0);
6491 Node* offset = argument(1);
6492 Node* len = argument(2);
6493 Node* state = argument(3);
6494 Node* subkeyH = argument(4);
6495
6496 state = must_be_not_null(state, true);
6497 subkeyH = must_be_not_null(subkeyH, true);
6498 data = must_be_not_null(data, true);
6499
6500 state = access_resolve(state, ACCESS_WRITE);
6501 subkeyH = access_resolve(subkeyH, ACCESS_READ);
6502 data = access_resolve(data, ACCESS_READ);
6503
6504 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6505 assert(state_start, "state is NULL");
6506 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6507 assert(subkeyH_start, "subkeyH is NULL");
6508 Node* data_start = array_element_address(data, offset, T_BYTE);
6509 assert(data_start, "data is NULL");
6510
6511 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6512 OptoRuntime::ghash_processBlocks_Type(),
6513 stubAddr, stubName, TypePtr::BOTTOM,
6514 state_start, subkeyH_start, data_start, len);
6515 return true;
6516 }
6517
6518 bool LibraryCallKit::inline_base64_encodeBlock() {
6519 address stubAddr;
6520 const char *stubName;
6521 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6522 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6523 stubAddr = StubRoutines::base64_encodeBlock();
6524 stubName = "encodeBlock";
6525
6526 if (!stubAddr) return false;
6527 Node* base64obj = argument(0);
6528 Node* src = argument(1);
6529 Node* offset = argument(2);
6530 Node* len = argument(3);
6531 Node* dest = argument(4);
6532 Node* dp = argument(5);
6533 Node* isURL = argument(6);
6534
6535 src = must_be_not_null(src, true);
6536 src = access_resolve(src, ACCESS_READ);
6537 dest = must_be_not_null(dest, true);
6538 dest = access_resolve(dest, ACCESS_WRITE);
6539
6540 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6541 assert(src_start, "source array is NULL");
6542 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6543 assert(dest_start, "destination array is NULL");
6544
6545 Node* base64 = make_runtime_call(RC_LEAF,
6546 OptoRuntime::base64_encodeBlock_Type(),
6547 stubAddr, stubName, TypePtr::BOTTOM,
6548 src_start, offset, len, dest_start, dp, isURL);
6549 return true;
6550 }
6551
6552 //------------------------------inline_sha_implCompress-----------------------
6553 //
6554 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6555 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6556 //
6557 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6558 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6559 //
6560 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6561 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6562 //
6563 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6564 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6565
6566 Node* sha_obj = argument(0);
6567 Node* src = argument(1); // type oop
6568 Node* ofs = argument(2); // type int
6569
6570 const Type* src_type = src->Value(&_gvn);
6571 const TypeAryPtr* top_src = src_type->isa_aryptr();
6572 if (top_src == NULL || top_src->klass() == NULL) {
6573 // failed array check
6574 return false;
6575 }
6576 // Figure out the size and type of the elements we will be copying.
6577 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6578 if (src_elem != T_BYTE) {
6579 return false;
6580 }
6581 // 'src_start' points to src array + offset
6582 src = must_be_not_null(src, true);
6583 src = access_resolve(src, ACCESS_READ);
6584 Node* src_start = array_element_address(src, ofs, src_elem);
6585 Node* state = NULL;
6586 address stubAddr;
6587 const char *stubName;
6588
6589 switch(id) {
6590 case vmIntrinsics::_sha_implCompress:
6591 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6592 state = get_state_from_sha_object(sha_obj);
6593 stubAddr = StubRoutines::sha1_implCompress();
6594 stubName = "sha1_implCompress";
6595 break;
6596 case vmIntrinsics::_sha2_implCompress:
6597 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6598 state = get_state_from_sha_object(sha_obj);
6599 stubAddr = StubRoutines::sha256_implCompress();
6600 stubName = "sha256_implCompress";
6601 break;
6602 case vmIntrinsics::_sha5_implCompress:
6603 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6604 state = get_state_from_sha5_object(sha_obj);
6605 stubAddr = StubRoutines::sha512_implCompress();
6606 stubName = "sha512_implCompress";
6607 break;
6608 default:
6609 fatal_unexpected_iid(id);
6610 return false;
6611 }
6612 if (state == NULL) return false;
6613
6614 assert(stubAddr != NULL, "Stub is generated");
6615 if (stubAddr == NULL) return false;
6616
6617 // Call the stub.
6618 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6619 stubAddr, stubName, TypePtr::BOTTOM,
6620 src_start, state);
6621
6622 return true;
6623 }
6624
6625 //------------------------------inline_digestBase_implCompressMB-----------------------
6626 //
6627 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6628 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6629 //
6630 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6631 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6632 "need SHA1/SHA256/SHA512 instruction support");
6633 assert((uint)predicate < 3, "sanity");
6634 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6635
6636 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6637 Node* src = argument(1); // byte[] array
6638 Node* ofs = argument(2); // type int
6639 Node* limit = argument(3); // type int
6640
6641 const Type* src_type = src->Value(&_gvn);
6642 const TypeAryPtr* top_src = src_type->isa_aryptr();
6643 if (top_src == NULL || top_src->klass() == NULL) {
6644 // failed array check
6645 return false;
6646 }
6647 // Figure out the size and type of the elements we will be copying.
6648 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6649 if (src_elem != T_BYTE) {
6650 return false;
6651 }
6652 // 'src_start' points to src array + offset
6653 src = must_be_not_null(src, false);
6654 src = access_resolve(src, ACCESS_READ);
6655 Node* src_start = array_element_address(src, ofs, src_elem);
6656
6657 const char* klass_SHA_name = NULL;
6658 const char* stub_name = NULL;
6659 address stub_addr = NULL;
6660 bool long_state = false;
6661
6662 switch (predicate) {
6663 case 0:
6664 if (UseSHA1Intrinsics) {
6665 klass_SHA_name = "sun/security/provider/SHA";
6666 stub_name = "sha1_implCompressMB";
6667 stub_addr = StubRoutines::sha1_implCompressMB();
6668 }
6669 break;
6670 case 1:
6671 if (UseSHA256Intrinsics) {
6672 klass_SHA_name = "sun/security/provider/SHA2";
6673 stub_name = "sha256_implCompressMB";
6674 stub_addr = StubRoutines::sha256_implCompressMB();
6675 }
6676 break;
6677 case 2:
6678 if (UseSHA512Intrinsics) {
6679 klass_SHA_name = "sun/security/provider/SHA5";
6680 stub_name = "sha512_implCompressMB";
6681 stub_addr = StubRoutines::sha512_implCompressMB();
6682 long_state = true;
6683 }
6684 break;
6685 default:
6686 fatal("unknown SHA intrinsic predicate: %d", predicate);
6687 }
6688 if (klass_SHA_name != NULL) {
6689 assert(stub_addr != NULL, "Stub is generated");
6690 if (stub_addr == NULL) return false;
6691
6692 // get DigestBase klass to lookup for SHA klass
6693 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6694 assert(tinst != NULL, "digestBase_obj is not instance???");
6695 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6696
6697 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6698 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6699 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6700 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6701 }
6702 return false;
6703 }
6704 //------------------------------inline_sha_implCompressMB-----------------------
6705 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6706 bool long_state, address stubAddr, const char *stubName,
6707 Node* src_start, Node* ofs, Node* limit) {
6708 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6709 const TypeOopPtr* xtype = aklass->as_instance_type();
6710 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6711 sha_obj = _gvn.transform(sha_obj);
6712
6713 Node* state;
6714 if (long_state) {
6715 state = get_state_from_sha5_object(sha_obj);
6716 } else {
6717 state = get_state_from_sha_object(sha_obj);
6718 }
6719 if (state == NULL) return false;
6720
6721 // Call the stub.
6722 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6723 OptoRuntime::digestBase_implCompressMB_Type(),
6724 stubAddr, stubName, TypePtr::BOTTOM,
6725 src_start, state, ofs, limit);
6726 // return ofs (int)
6727 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6728 set_result(result);
6729
6730 return true;
6731 }
6732
6733 //------------------------------get_state_from_sha_object-----------------------
6734 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6735 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6736 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6737 if (sha_state == NULL) return (Node *) NULL;
6738
6739 // now have the array, need to get the start address of the state array
6740 sha_state = access_resolve(sha_state, ACCESS_WRITE);
6741 Node* state = array_element_address(sha_state, intcon(0), T_INT);
6742 return state;
6743 }
6744
6745 //------------------------------get_state_from_sha5_object-----------------------
6746 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6747 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6748 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6749 if (sha_state == NULL) return (Node *) NULL;
6750
6751 // now have the array, need to get the start address of the state array
6752 sha_state = access_resolve(sha_state, ACCESS_WRITE);
6753 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6754 return state;
6755 }
6756
6757 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6758 // Return node representing slow path of predicate check.
6759 // the pseudo code we want to emulate with this predicate is:
6760 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6761 //
6762 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6763 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6764 "need SHA1/SHA256/SHA512 instruction support");
6765 assert((uint)predicate < 3, "sanity");
6766
6767 // The receiver was checked for NULL already.
6768 Node* digestBaseObj = argument(0);
6769
6770 // get DigestBase klass for instanceOf check
6771 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6772 assert(tinst != NULL, "digestBaseObj is null");
6773 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6774
6775 const char* klass_SHA_name = NULL;
6776 switch (predicate) {
6777 case 0:
6778 if (UseSHA1Intrinsics) {
6779 // we want to do an instanceof comparison against the SHA class
6780 klass_SHA_name = "sun/security/provider/SHA";
6781 }
6782 break;
6783 case 1:
6784 if (UseSHA256Intrinsics) {
6785 // we want to do an instanceof comparison against the SHA2 class
6786 klass_SHA_name = "sun/security/provider/SHA2";
6787 }
6788 break;
6789 case 2:
6790 if (UseSHA512Intrinsics) {
6791 // we want to do an instanceof comparison against the SHA5 class
6792 klass_SHA_name = "sun/security/provider/SHA5";
6793 }
6794 break;
6795 default:
6796 fatal("unknown SHA intrinsic predicate: %d", predicate);
6797 }
6798
6799 ciKlass* klass_SHA = NULL;
6800 if (klass_SHA_name != NULL) {
6801 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6802 }
6803 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6804 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6805 Node* ctrl = control();
6806 set_control(top()); // no intrinsic path
6807 return ctrl;
6808 }
6809 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6810
6811 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6812 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6813 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6814 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6815
6816 return instof_false; // even if it is NULL
6817 }
6818
6819 //-------------inline_fma-----------------------------------
6820 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
6821 Node *a = NULL;
6822 Node *b = NULL;
6823 Node *c = NULL;
6824 Node* result = NULL;
6825 switch (id) {
6826 case vmIntrinsics::_fmaD:
6827 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
6828 // no receiver since it is static method
6829 a = round_double_node(argument(0));
6830 b = round_double_node(argument(2));
6831 c = round_double_node(argument(4));
6832 result = _gvn.transform(new FmaDNode(control(), a, b, c));
6833 break;
6834 case vmIntrinsics::_fmaF:
6835 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
6836 a = argument(0);
6837 b = argument(1);
6838 c = argument(2);
6839 result = _gvn.transform(new FmaFNode(control(), a, b, c));
6840 break;
6841 default:
6842 fatal_unexpected_iid(id); break;
6843 }
6844 set_result(result);
6845 return true;
6846 }
6847
6848 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
6849 // argument(0) is receiver
6850 Node* codePoint = argument(1);
6851 Node* n = NULL;
6852
6853 switch (id) {
6854 case vmIntrinsics::_isDigit :
6855 n = new DigitNode(control(), codePoint);
6856 break;
6857 case vmIntrinsics::_isLowerCase :
6858 n = new LowerCaseNode(control(), codePoint);
6859 break;
6860 case vmIntrinsics::_isUpperCase :
6861 n = new UpperCaseNode(control(), codePoint);
6862 break;
6863 case vmIntrinsics::_isWhitespace :
6864 n = new WhitespaceNode(control(), codePoint);
6865 break;
6866 default:
6867 fatal_unexpected_iid(id);
6868 }
6869
6870 set_result(_gvn.transform(n));
6871 return true;
6872 }
6873
6874 //------------------------------inline_fp_min_max------------------------------
6875 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
6876 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
6877
6878 // The intrinsic should be used only when the API branches aren't predictable,
6879 // the last one performing the most important comparison. The following heuristic
6880 // uses the branch statistics to eventually bail out if necessary.
6881
6882 ciMethodData *md = callee()->method_data();
6883
6884 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
6885 ciCallProfile cp = caller()->call_profile_at_bci(bci());
6886
6887 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
6888 // Bail out if the call-site didn't contribute enough to the statistics.
6889 return false;
6890 }
6891
6892 uint taken = 0, not_taken = 0;
6893
6894 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
6895 if (p->is_BranchData()) {
6896 taken = ((ciBranchData*)p)->taken();
6897 not_taken = ((ciBranchData*)p)->not_taken();
6898 }
6899 }
6900
6901 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
6902 balance = balance < 0 ? -balance : balance;
6903 if ( balance > 0.2 ) {
6904 // Bail out if the most important branch is predictable enough.
6905 return false;
6906 }
6907 }
6908 */
6909
6910 Node *a = NULL;
6911 Node *b = NULL;
6912 Node *n = NULL;
6913 switch (id) {
6914 case vmIntrinsics::_maxF:
6915 case vmIntrinsics::_minF:
6916 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
6917 a = argument(0);
6918 b = argument(1);
6919 break;
6920 case vmIntrinsics::_maxD:
6921 case vmIntrinsics::_minD:
6922 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
6923 a = round_double_node(argument(0));
6924 b = round_double_node(argument(2));
6925 break;
6926 default:
6927 fatal_unexpected_iid(id);
6928 break;
6929 }
6930 switch (id) {
6931 case vmIntrinsics::_maxF: n = new MaxFNode(a, b); break;
6932 case vmIntrinsics::_minF: n = new MinFNode(a, b); break;
6933 case vmIntrinsics::_maxD: n = new MaxDNode(a, b); break;
6934 case vmIntrinsics::_minD: n = new MinDNode(a, b); break;
6935 default: fatal_unexpected_iid(id); break;
6936 }
6937 set_result(_gvn.transform(n));
6938 return true;
6939 }
6940
6941 bool LibraryCallKit::inline_profileBoolean() {
6942 Node* counts = argument(1);
6943 const TypeAryPtr* ary = NULL;
6944 ciArray* aobj = NULL;
6945 if (counts->is_Con()
6946 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6947 && (aobj = ary->const_oop()->as_array()) != NULL
6948 && (aobj->length() == 2)) {
6949 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6950 jint false_cnt = aobj->element_value(0).as_int();
6951 jint true_cnt = aobj->element_value(1).as_int();
6952
6953 if (C->log() != NULL) {
6954 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6955 false_cnt, true_cnt);
6956 }
6957
6958 if (false_cnt + true_cnt == 0) {
6959 // According to profile, never executed.
6960 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6961 Deoptimization::Action_reinterpret);
6962 return true;
6963 }
6964
6965 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6966 // is a number of each value occurrences.
6967 Node* result = argument(0);
6968 if (false_cnt == 0 || true_cnt == 0) {
6969 // According to profile, one value has been never seen.
6970 int expected_val = (false_cnt == 0) ? 1 : 0;
6971
6972 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6973 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6974
6975 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6976 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6977 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6978
6979 { // Slow path: uncommon trap for never seen value and then reexecute
6980 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6981 // the value has been seen at least once.
6982 PreserveJVMState pjvms(this);
6983 PreserveReexecuteState preexecs(this);
6984 jvms()->set_should_reexecute(true);
6985
6986 set_control(slow_path);
6987 set_i_o(i_o());
6988
6989 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6990 Deoptimization::Action_reinterpret);
6991 }
6992 // The guard for never seen value enables sharpening of the result and
6993 // returning a constant. It allows to eliminate branches on the same value
6994 // later on.
6995 set_control(fast_path);
6996 result = intcon(expected_val);
6997 }
6998 // Stop profiling.
6999 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
7000 // By replacing method body with profile data (represented as ProfileBooleanNode
7001 // on IR level) we effectively disable profiling.
7002 // It enables full speed execution once optimized code is generated.
7003 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
7004 C->record_for_igvn(profile);
7005 set_result(profile);
7006 return true;
7007 } else {
7008 // Continue profiling.
7009 // Profile data isn't available at the moment. So, execute method's bytecode version.
7010 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
7011 // is compiled and counters aren't available since corresponding MethodHandle
7012 // isn't a compile-time constant.
7013 return false;
7014 }
7015 }
7016
7017 bool LibraryCallKit::inline_isCompileConstant() {
7018 Node* n = argument(0);
7019 set_result(n->is_Con() ? intcon(1) : intcon(0));
7020 return true;
7021 }
--- EOF ---