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