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