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 "opto/valuetypenode.hpp"
56 #include "prims/nativeLookup.hpp"
57 #include "prims/unsafe.hpp"
58 #include "runtime/objectMonitor.hpp"
59 #include "runtime/sharedRuntime.hpp"
60 #include "utilities/macros.hpp"
61
62
63 class LibraryIntrinsic : public InlineCallGenerator {
64 // Extend the set of intrinsics known to the runtime:
65 public:
66 private:
67 bool _is_virtual;
68 bool _does_virtual_dispatch;
69 int8_t _predicates_count; // Intrinsic is predicated by several conditions
70 int8_t _last_predicate; // Last generated predicate
71 vmIntrinsics::ID _intrinsic_id;
72
73 public:
74 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
75 : InlineCallGenerator(m),
76 _is_virtual(is_virtual),
77 _does_virtual_dispatch(does_virtual_dispatch),
78 _predicates_count((int8_t)predicates_count),
79 _last_predicate((int8_t)-1),
80 _intrinsic_id(id)
81 {
82 }
83 virtual bool is_intrinsic() const { return true; }
84 virtual bool is_virtual() const { return _is_virtual; }
85 virtual bool is_predicated() const { return _predicates_count > 0; }
86 virtual int predicates_count() const { return _predicates_count; }
87 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; }
88 virtual JVMState* generate(JVMState* jvms);
89 virtual Node* generate_predicate(JVMState* jvms, int predicate);
90 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
91 };
92
93
94 // Local helper class for LibraryIntrinsic:
95 class LibraryCallKit : public GraphKit {
96 private:
97 LibraryIntrinsic* _intrinsic; // the library intrinsic being called
98 Node* _result; // the result node, if any
99 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted
100
101 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type);
102
103 public:
104 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
105 : GraphKit(jvms),
106 _intrinsic(intrinsic),
107 _result(NULL)
108 {
109 // Check if this is a root compile. In that case we don't have a caller.
110 if (!jvms->has_method()) {
111 _reexecute_sp = sp();
112 } else {
113 // Find out how many arguments the interpreter needs when deoptimizing
114 // and save the stack pointer value so it can used by uncommon_trap.
115 // We find the argument count by looking at the declared signature.
116 bool ignored_will_link;
117 ciSignature* declared_signature = NULL;
118 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
119 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
120 _reexecute_sp = sp() + nargs; // "push" arguments back on stack
121 }
122 }
123
124 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
125
126 ciMethod* caller() const { return jvms()->method(); }
127 int bci() const { return jvms()->bci(); }
128 LibraryIntrinsic* intrinsic() const { return _intrinsic; }
129 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); }
130 ciMethod* callee() const { return _intrinsic->method(); }
131
132 bool try_to_inline(int predicate);
133 Node* try_to_predicate(int predicate);
134
135 void push_result() {
136 // Push the result onto the stack.
137 if (!stopped() && result() != NULL) {
138 BasicType bt = result()->bottom_type()->basic_type();
139 push_node(bt, result());
140 }
141 }
142
143 private:
144 void fatal_unexpected_iid(vmIntrinsics::ID iid) {
145 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid));
146 }
147
148 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
149 void set_result(RegionNode* region, PhiNode* value);
150 Node* result() { return _result; }
151
152 virtual int reexecute_sp() { return _reexecute_sp; }
153
154 // Helper functions to inline natives
155 Node* generate_guard(Node* test, RegionNode* region, float true_prob);
156 Node* generate_slow_guard(Node* test, RegionNode* region);
157 Node* generate_fair_guard(Node* test, RegionNode* region);
158 Node* generate_negative_guard(Node* index, RegionNode* region,
159 // resulting CastII of index:
160 Node* *pos_index = NULL);
161 Node* generate_limit_guard(Node* offset, Node* subseq_length,
162 Node* array_length,
163 RegionNode* region);
164 void generate_string_range_check(Node* array, Node* offset,
165 Node* length, bool char_count);
166 Node* generate_current_thread(Node* &tls_output);
167 Node* load_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
189 enum ArrayKind {
190 AnyArray,
191 NonArray,
192 ObjectArray,
193 NonObjectArray,
194 TypeArray,
195 ValueArray
196 };
197
198 Node* generate_array_guard(Node* kls, RegionNode* region) {
199 return generate_array_guard_common(kls, region, AnyArray);
200 }
201 Node* generate_non_array_guard(Node* kls, RegionNode* region) {
202 return generate_array_guard_common(kls, region, NonArray);
203 }
204 Node* generate_objArray_guard(Node* kls, RegionNode* region) {
205 return generate_array_guard_common(kls, region, ObjectArray);
206 }
207 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
208 return generate_array_guard_common(kls, region, NonObjectArray);
209 }
210 Node* generate_typeArray_guard(Node* kls, RegionNode* region) {
211 return generate_array_guard_common(kls, region, TypeArray);
212 }
213 Node* generate_valueArray_guard(Node* kls, RegionNode* region) {
214 assert(ValueArrayFlatten, "can never be flattened");
215 return generate_array_guard_common(kls, region, ValueArray);
216 }
217 Node* generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind);
218 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
219 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
220 bool is_virtual = false, bool is_static = false);
221 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
222 return generate_method_call(method_id, false, true);
223 }
224 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
225 return generate_method_call(method_id, true, false);
226 }
227 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
228 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
229
230 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
231 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
232 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
233 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
234 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
235 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
236 bool inline_string_indexOfChar();
237 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
238 bool inline_string_toBytesU();
239 bool inline_string_getCharsU();
240 bool inline_string_copy(bool compress);
241 bool inline_string_char_access(bool is_store);
242 Node* round_double_node(Node* n);
243 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
244 bool inline_math_native(vmIntrinsics::ID id);
245 bool inline_math(vmIntrinsics::ID id);
246 bool inline_double_math(vmIntrinsics::ID id);
247 template <typename OverflowOp>
248 bool inline_math_overflow(Node* arg1, Node* arg2);
249 void inline_math_mathExact(Node* math, Node* test);
250 bool inline_math_addExactI(bool is_increment);
251 bool inline_math_addExactL(bool is_increment);
252 bool inline_math_multiplyExactI();
253 bool inline_math_multiplyExactL();
254 bool inline_math_multiplyHigh();
255 bool inline_math_negateExactI();
256 bool inline_math_negateExactL();
257 bool inline_math_subtractExactI(bool is_decrement);
258 bool inline_math_subtractExactL(bool is_decrement);
259 bool inline_min_max(vmIntrinsics::ID id);
260 bool inline_notify(vmIntrinsics::ID id);
261 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
262 // This returns Type::AnyPtr, RawPtr, or OopPtr.
263 int classify_unsafe_addr(Node* &base, Node* &offset, BasicType type);
264 Node* make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type = T_ILLEGAL, bool can_cast = false);
265
266 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
267 DecoratorSet mo_decorator_for_access_kind(AccessKind kind);
268 bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
269 static bool klass_needs_init_guard(Node* kls);
270 bool inline_unsafe_allocate();
271 bool inline_unsafe_newArray(bool uninitialized);
272 bool inline_unsafe_writeback0();
273 bool inline_unsafe_writebackSync0(bool is_pre);
274 bool inline_unsafe_copyMemory();
275 bool inline_unsafe_make_private_buffer();
276 bool inline_unsafe_finish_private_buffer();
277 bool inline_native_currentThread();
278
279 bool inline_native_time_funcs(address method, const char* funcName);
280 #ifdef JFR_HAVE_INTRINSICS
281 bool inline_native_classID();
282 bool inline_native_getEventWriter();
283 #endif
284 bool inline_native_isInterrupted();
285 bool inline_native_Class_query(vmIntrinsics::ID id);
286 bool inline_value_Class_conversion(vmIntrinsics::ID id);
287 bool inline_native_subtype_check();
288 bool inline_native_getLength();
289 bool inline_array_copyOf(bool is_copyOfRange);
290 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
291 bool inline_preconditions_checkIndex();
292 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array);
293 bool inline_native_clone(bool is_virtual);
294 bool inline_native_Reflection_getCallerClass();
295 // Helper function for inlining native object hash method
296 bool inline_native_hashcode(bool is_virtual, bool is_static);
297 bool inline_native_getClass();
298
299 // Helper functions for inlining arraycopy
300 bool inline_arraycopy();
301 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
302 RegionNode* slow_region);
303 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
304 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp,
305 uint new_idx);
306
307 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
308 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
309 bool inline_unsafe_fence(vmIntrinsics::ID id);
310 bool inline_onspinwait();
311 bool inline_fp_conversions(vmIntrinsics::ID id);
312 bool inline_number_methods(vmIntrinsics::ID id);
313 bool inline_reference_get();
314 bool inline_Class_cast();
315 bool inline_aescrypt_Block(vmIntrinsics::ID id);
316 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
317 bool inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id);
318 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
319 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
320 Node* inline_electronicCodeBook_AESCrypt_predicate(bool decrypting);
321 Node* inline_counterMode_AESCrypt_predicate();
322 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
323 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
324 bool inline_ghash_processBlocks();
325 bool inline_base64_encodeBlock();
326 bool inline_sha_implCompress(vmIntrinsics::ID id);
327 bool inline_digestBase_implCompressMB(int predicate);
328 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
329 bool long_state, address stubAddr, const char *stubName,
330 Node* src_start, Node* ofs, Node* limit);
331 Node* get_state_from_sha_object(Node *sha_object);
332 Node* get_state_from_sha5_object(Node *sha_object);
333 Node* inline_digestBase_implCompressMB_predicate(int predicate);
334 bool inline_encodeISOArray();
335 bool inline_updateCRC32();
336 bool inline_updateBytesCRC32();
337 bool inline_updateByteBufferCRC32();
338 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
339 bool inline_updateBytesCRC32C();
340 bool inline_updateDirectByteBufferCRC32C();
341 bool inline_updateBytesAdler32();
342 bool inline_updateByteBufferAdler32();
343 bool inline_multiplyToLen();
344 bool inline_hasNegatives();
345 bool inline_squareToLen();
346 bool inline_mulAdd();
347 bool inline_montgomeryMultiply();
348 bool inline_montgomerySquare();
349 bool inline_vectorizedMismatch();
350 bool inline_fma(vmIntrinsics::ID id);
351 bool inline_character_compare(vmIntrinsics::ID id);
352 bool inline_fp_min_max(vmIntrinsics::ID id);
353
354 bool inline_profileBoolean();
355 bool inline_isCompileConstant();
356 void clear_upper_avx() {
357 #ifdef X86
358 if (UseAVX >= 2) {
359 C->set_clear_upper_avx(true);
360 }
361 #endif
362 }
363 };
364
365 //---------------------------make_vm_intrinsic----------------------------
366 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
367 vmIntrinsics::ID id = m->intrinsic_id();
368 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
369
370 if (!m->is_loaded()) {
371 // Do not attempt to inline unloaded methods.
372 return NULL;
373 }
374
375 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
376 bool is_available = false;
377
378 {
379 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
380 // the compiler must transition to '_thread_in_vm' state because both
381 // methods access VM-internal data.
382 VM_ENTRY_MARK;
383 methodHandle mh(THREAD, m->get_Method());
384 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
385 !C->directive()->is_intrinsic_disabled(mh) &&
386 !vmIntrinsics::is_disabled_by_flags(mh);
387
388 }
389
390 if (is_available) {
391 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
392 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
393 return new LibraryIntrinsic(m, is_virtual,
394 vmIntrinsics::predicates_needed(id),
395 vmIntrinsics::does_virtual_dispatch(id),
396 (vmIntrinsics::ID) id);
397 } else {
398 return NULL;
399 }
400 }
401
402 //----------------------register_library_intrinsics-----------------------
403 // Initialize this file's data structures, for each Compile instance.
404 void Compile::register_library_intrinsics() {
405 // Nothing to do here.
406 }
407
408 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
409 LibraryCallKit kit(jvms, this);
410 Compile* C = kit.C;
411 int nodes = C->unique();
412 #ifndef PRODUCT
413 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
414 char buf[1000];
415 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
416 tty->print_cr("Intrinsic %s", str);
417 }
418 #endif
419 ciMethod* callee = kit.callee();
420 const int bci = kit.bci();
421
422 // Try to inline the intrinsic.
423 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
424 kit.try_to_inline(_last_predicate)) {
425 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
426 : "(intrinsic)";
427 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
428 if (C->print_intrinsics() || C->print_inlining()) {
429 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
430 }
431 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
432 if (C->log()) {
433 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
434 vmIntrinsics::name_at(intrinsic_id()),
435 (is_virtual() ? " virtual='1'" : ""),
436 C->unique() - nodes);
437 }
438 // Push the result from the inlined method onto the stack.
439 kit.push_result();
440 C->print_inlining_update(this);
441 return kit.transfer_exceptions_into_jvms();
442 }
443
444 // The intrinsic bailed out
445 if (jvms->has_method()) {
446 // Not a root compile.
447 const char* msg;
448 if (callee->intrinsic_candidate()) {
449 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
450 } else {
451 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
452 : "failed to inline (intrinsic), method not annotated";
453 }
454 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
455 if (C->print_intrinsics() || C->print_inlining()) {
456 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
457 }
458 } else {
459 // Root compile
460 ResourceMark rm;
461 stringStream msg_stream;
462 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
463 vmIntrinsics::name_at(intrinsic_id()),
464 is_virtual() ? " (virtual)" : "", bci);
465 const char *msg = msg_stream.as_string();
466 log_debug(jit, inlining)("%s", msg);
467 if (C->print_intrinsics() || C->print_inlining()) {
468 tty->print("%s", msg);
469 }
470 }
471 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
472 C->print_inlining_update(this);
473 return NULL;
474 }
475
476 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
477 LibraryCallKit kit(jvms, this);
478 Compile* C = kit.C;
479 int nodes = C->unique();
480 _last_predicate = predicate;
481 #ifndef PRODUCT
482 assert(is_predicated() && predicate < predicates_count(), "sanity");
483 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
484 char buf[1000];
485 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
486 tty->print_cr("Predicate for intrinsic %s", str);
487 }
488 #endif
489 ciMethod* callee = kit.callee();
490 const int bci = kit.bci();
491
492 Node* slow_ctl = kit.try_to_predicate(predicate);
493 if (!kit.failing()) {
494 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
495 : "(intrinsic, predicate)";
496 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
497 if (C->print_intrinsics() || C->print_inlining()) {
498 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
499 }
500 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
501 if (C->log()) {
502 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
503 vmIntrinsics::name_at(intrinsic_id()),
504 (is_virtual() ? " virtual='1'" : ""),
505 C->unique() - nodes);
506 }
507 return slow_ctl; // Could be NULL if the check folds.
508 }
509
510 // The intrinsic bailed out
511 if (jvms->has_method()) {
512 // Not a root compile.
513 const char* msg = "failed to generate predicate for intrinsic";
514 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
515 if (C->print_intrinsics() || C->print_inlining()) {
516 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
517 }
518 } else {
519 // Root compile
520 ResourceMark rm;
521 stringStream msg_stream;
522 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
523 vmIntrinsics::name_at(intrinsic_id()),
524 is_virtual() ? " (virtual)" : "", bci);
525 const char *msg = msg_stream.as_string();
526 log_debug(jit, inlining)("%s", msg);
527 if (C->print_intrinsics() || C->print_inlining()) {
528 C->print_inlining_stream()->print("%s", msg);
529 }
530 }
531 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
532 return NULL;
533 }
534
535 bool LibraryCallKit::try_to_inline(int predicate) {
536 // Handle symbolic names for otherwise undistinguished boolean switches:
537 const bool is_store = true;
538 const bool is_compress = true;
539 const bool is_static = true;
540 const bool is_volatile = true;
541
542 if (!jvms()->has_method()) {
543 // Root JVMState has a null method.
544 assert(map()->memory()->Opcode() == Op_Parm, "");
545 // Insert the memory aliasing node
546 set_all_memory(reset_memory());
547 }
548 assert(merged_memory(), "");
549
550
551 switch (intrinsic_id()) {
552 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
553 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
554 case vmIntrinsics::_getClass: return inline_native_getClass();
555
556 case vmIntrinsics::_ceil:
557 case vmIntrinsics::_floor:
558 case vmIntrinsics::_rint:
559 case vmIntrinsics::_dsin:
560 case vmIntrinsics::_dcos:
561 case vmIntrinsics::_dtan:
562 case vmIntrinsics::_dabs:
563 case vmIntrinsics::_fabs:
564 case vmIntrinsics::_iabs:
565 case vmIntrinsics::_labs:
566 case vmIntrinsics::_datan2:
567 case vmIntrinsics::_dsqrt:
568 case vmIntrinsics::_dexp:
569 case vmIntrinsics::_dlog:
570 case vmIntrinsics::_dlog10:
571 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id());
572
573 case vmIntrinsics::_min:
574 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
575
576 case vmIntrinsics::_notify:
577 case vmIntrinsics::_notifyAll:
578 return inline_notify(intrinsic_id());
579
580 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
581 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
582 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
583 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
584 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
585 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
586 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
587 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
588 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
589 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
590 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
591 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
592 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
593
594 case vmIntrinsics::_arraycopy: return inline_arraycopy();
595
596 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
597 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
598 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
599 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
600
601 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
602 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
603 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
604 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
605 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
606 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
607 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
608
609 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
610 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
611
612 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
613 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
614 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
615 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
616
617 case vmIntrinsics::_compressStringC:
618 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
619 case vmIntrinsics::_inflateStringC:
620 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
621
622 case vmIntrinsics::_makePrivateBuffer: return inline_unsafe_make_private_buffer();
623 case vmIntrinsics::_finishPrivateBuffer: return inline_unsafe_finish_private_buffer();
624 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
625 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
626 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
627 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
628 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
629 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
630 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
631 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
632 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
633 case vmIntrinsics::_getValue: return inline_unsafe_access(!is_store, T_VALUETYPE,Relaxed, false);
634
635 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
636 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
637 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
638 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
639 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
640 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
641 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
642 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
643 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
644 case vmIntrinsics::_putValue: return inline_unsafe_access( is_store, T_VALUETYPE,Relaxed, false);
645
646 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
647 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
648 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
649 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
650 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
651 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
652 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
653 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
654 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
655
656 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
657 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
658 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
659 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
660 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
661 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
662 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
663 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
664 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
665
666 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
667 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
668 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
669 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
670
671 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
672 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
673 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
674 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
675
676 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
677 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
678 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
679 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
680 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
681 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
682 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
683 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
684 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
685
686 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
687 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
688 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
689 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
690 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
691 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
692 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
693 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
694 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
695
696 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
697 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
698 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
699 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
700 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
701 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
702 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
703 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
704 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
705
706 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
707 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
708 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
709 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
710 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
711 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
712 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
713 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
714 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
715
716 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
717 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
718 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
719 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
720 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
721
722 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
723 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
724 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
725 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
726 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
727 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
728 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
729 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
730 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
731 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
732 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
733 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
734 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
735 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
736 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
737 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
738 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
739 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
740 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
741 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
742
743 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
744 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
745 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
746 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
747 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
748 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
749 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
750 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
751 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
752 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
753 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
754 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
755 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
756 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
757 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
758
759 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
760 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
761 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
762 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
763
764 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
765 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
766 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
767 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
768 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
769
770 case vmIntrinsics::_loadFence:
771 case vmIntrinsics::_storeFence:
772 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
773
774 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
775
776 case vmIntrinsics::_currentThread: return inline_native_currentThread();
777 case vmIntrinsics::_isInterrupted: return inline_native_isInterrupted();
778
779 #ifdef JFR_HAVE_INTRINSICS
780 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime");
781 case vmIntrinsics::_getClassId: return inline_native_classID();
782 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
783 #endif
784 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
785 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
786 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
787 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
788 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
789 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
790 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
791 case vmIntrinsics::_getLength: return inline_native_getLength();
792 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
793 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
794 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
795 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
796 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
797 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
798
799 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
800 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
801
802 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
803
804 case vmIntrinsics::_isInstance:
805 case vmIntrinsics::_getModifiers:
806 case vmIntrinsics::_isInterface:
807 case vmIntrinsics::_isArray:
808 case vmIntrinsics::_isPrimitive:
809 case vmIntrinsics::_getSuperclass:
810 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
811
812 case vmIntrinsics::_asPrimaryType:
813 case vmIntrinsics::_asIndirectType: return inline_value_Class_conversion(intrinsic_id());
814
815 case vmIntrinsics::_floatToRawIntBits:
816 case vmIntrinsics::_floatToIntBits:
817 case vmIntrinsics::_intBitsToFloat:
818 case vmIntrinsics::_doubleToRawLongBits:
819 case vmIntrinsics::_doubleToLongBits:
820 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
821
822 case vmIntrinsics::_numberOfLeadingZeros_i:
823 case vmIntrinsics::_numberOfLeadingZeros_l:
824 case vmIntrinsics::_numberOfTrailingZeros_i:
825 case vmIntrinsics::_numberOfTrailingZeros_l:
826 case vmIntrinsics::_bitCount_i:
827 case vmIntrinsics::_bitCount_l:
828 case vmIntrinsics::_reverseBytes_i:
829 case vmIntrinsics::_reverseBytes_l:
830 case vmIntrinsics::_reverseBytes_s:
831 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
832
833 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
834
835 case vmIntrinsics::_Reference_get: return inline_reference_get();
836
837 case vmIntrinsics::_Class_cast: return inline_Class_cast();
838
839 case vmIntrinsics::_aescrypt_encryptBlock:
840 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
841
842 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
843 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
844 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
845
846 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
847 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
848 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
849
850 case vmIntrinsics::_counterMode_AESCrypt:
851 return inline_counterMode_AESCrypt(intrinsic_id());
852
853 case vmIntrinsics::_sha_implCompress:
854 case vmIntrinsics::_sha2_implCompress:
855 case vmIntrinsics::_sha5_implCompress:
856 return inline_sha_implCompress(intrinsic_id());
857
858 case vmIntrinsics::_digestBase_implCompressMB:
859 return inline_digestBase_implCompressMB(predicate);
860
861 case vmIntrinsics::_multiplyToLen:
862 return inline_multiplyToLen();
863
864 case vmIntrinsics::_squareToLen:
865 return inline_squareToLen();
866
867 case vmIntrinsics::_mulAdd:
868 return inline_mulAdd();
869
870 case vmIntrinsics::_montgomeryMultiply:
871 return inline_montgomeryMultiply();
872 case vmIntrinsics::_montgomerySquare:
873 return inline_montgomerySquare();
874
875 case vmIntrinsics::_vectorizedMismatch:
876 return inline_vectorizedMismatch();
877
878 case vmIntrinsics::_ghash_processBlocks:
879 return inline_ghash_processBlocks();
880 case vmIntrinsics::_base64_encodeBlock:
881 return inline_base64_encodeBlock();
882
883 case vmIntrinsics::_encodeISOArray:
884 case vmIntrinsics::_encodeByteISOArray:
885 return inline_encodeISOArray();
886
887 case vmIntrinsics::_updateCRC32:
888 return inline_updateCRC32();
889 case vmIntrinsics::_updateBytesCRC32:
890 return inline_updateBytesCRC32();
891 case vmIntrinsics::_updateByteBufferCRC32:
892 return inline_updateByteBufferCRC32();
893
894 case vmIntrinsics::_updateBytesCRC32C:
895 return inline_updateBytesCRC32C();
896 case vmIntrinsics::_updateDirectByteBufferCRC32C:
897 return inline_updateDirectByteBufferCRC32C();
898
899 case vmIntrinsics::_updateBytesAdler32:
900 return inline_updateBytesAdler32();
901 case vmIntrinsics::_updateByteBufferAdler32:
902 return inline_updateByteBufferAdler32();
903
904 case vmIntrinsics::_profileBoolean:
905 return inline_profileBoolean();
906 case vmIntrinsics::_isCompileConstant:
907 return inline_isCompileConstant();
908
909 case vmIntrinsics::_hasNegatives:
910 return inline_hasNegatives();
911
912 case vmIntrinsics::_fmaD:
913 case vmIntrinsics::_fmaF:
914 return inline_fma(intrinsic_id());
915
916 case vmIntrinsics::_isDigit:
917 case vmIntrinsics::_isLowerCase:
918 case vmIntrinsics::_isUpperCase:
919 case vmIntrinsics::_isWhitespace:
920 return inline_character_compare(intrinsic_id());
921
922 case vmIntrinsics::_maxF:
923 case vmIntrinsics::_minF:
924 case vmIntrinsics::_maxD:
925 case vmIntrinsics::_minD:
926 return inline_fp_min_max(intrinsic_id());
927
928 default:
929 // If you get here, it may be that someone has added a new intrinsic
930 // to the list in vmSymbols.hpp without implementing it here.
931 #ifndef PRODUCT
932 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
933 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
934 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
935 }
936 #endif
937 return false;
938 }
939 }
940
941 Node* LibraryCallKit::try_to_predicate(int predicate) {
942 if (!jvms()->has_method()) {
943 // Root JVMState has a null method.
944 assert(map()->memory()->Opcode() == Op_Parm, "");
945 // Insert the memory aliasing node
946 set_all_memory(reset_memory());
947 }
948 assert(merged_memory(), "");
949
950 switch (intrinsic_id()) {
951 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
952 return inline_cipherBlockChaining_AESCrypt_predicate(false);
953 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
954 return inline_cipherBlockChaining_AESCrypt_predicate(true);
955 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
956 return inline_electronicCodeBook_AESCrypt_predicate(false);
957 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
958 return inline_electronicCodeBook_AESCrypt_predicate(true);
959 case vmIntrinsics::_counterMode_AESCrypt:
960 return inline_counterMode_AESCrypt_predicate();
961 case vmIntrinsics::_digestBase_implCompressMB:
962 return inline_digestBase_implCompressMB_predicate(predicate);
963
964 default:
965 // If you get here, it may be that someone has added a new intrinsic
966 // to the list in vmSymbols.hpp without implementing it here.
967 #ifndef PRODUCT
968 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
969 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
970 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
971 }
972 #endif
973 Node* slow_ctl = control();
974 set_control(top()); // No fast path instrinsic
975 return slow_ctl;
976 }
977 }
978
979 //------------------------------set_result-------------------------------
980 // Helper function for finishing intrinsics.
981 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
982 record_for_igvn(region);
983 set_control(_gvn.transform(region));
984 set_result( _gvn.transform(value));
985 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
986 }
987
988 //------------------------------generate_guard---------------------------
989 // Helper function for generating guarded fast-slow graph structures.
990 // The given 'test', if true, guards a slow path. If the test fails
991 // then a fast path can be taken. (We generally hope it fails.)
992 // In all cases, GraphKit::control() is updated to the fast path.
993 // The returned value represents the control for the slow path.
994 // The return value is never 'top'; it is either a valid control
995 // or NULL if it is obvious that the slow path can never be taken.
996 // Also, if region and the slow control are not NULL, the slow edge
997 // is appended to the region.
998 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
999 if (stopped()) {
1000 // Already short circuited.
1001 return NULL;
1002 }
1003
1004 // Build an if node and its projections.
1005 // If test is true we take the slow path, which we assume is uncommon.
1006 if (_gvn.type(test) == TypeInt::ZERO) {
1007 // The slow branch is never taken. No need to build this guard.
1008 return NULL;
1009 }
1010
1011 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
1012
1013 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
1014 if (if_slow == top()) {
1015 // The slow branch is never taken. No need to build this guard.
1016 return NULL;
1017 }
1018
1019 if (region != NULL)
1020 region->add_req(if_slow);
1021
1022 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
1023 set_control(if_fast);
1024
1025 return if_slow;
1026 }
1027
1028 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
1029 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
1030 }
1031 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
1032 return generate_guard(test, region, PROB_FAIR);
1033 }
1034
1035 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
1036 Node* *pos_index) {
1037 if (stopped())
1038 return NULL; // already stopped
1039 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
1040 return NULL; // index is already adequately typed
1041 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
1042 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1043 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
1044 if (is_neg != NULL && pos_index != NULL) {
1045 // Emulate effect of Parse::adjust_map_after_if.
1046 Node* ccast = new CastIINode(index, TypeInt::POS);
1047 ccast->set_req(0, control());
1048 (*pos_index) = _gvn.transform(ccast);
1049 }
1050 return is_neg;
1051 }
1052
1053 // Make sure that 'position' is a valid limit index, in [0..length].
1054 // There are two equivalent plans for checking this:
1055 // A. (offset + copyLength) unsigned<= arrayLength
1056 // B. offset <= (arrayLength - copyLength)
1057 // We require that all of the values above, except for the sum and
1058 // difference, are already known to be non-negative.
1059 // Plan A is robust in the face of overflow, if offset and copyLength
1060 // are both hugely positive.
1061 //
1062 // Plan B is less direct and intuitive, but it does not overflow at
1063 // all, since the difference of two non-negatives is always
1064 // representable. Whenever Java methods must perform the equivalent
1065 // check they generally use Plan B instead of Plan A.
1066 // For the moment we use Plan A.
1067 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
1068 Node* subseq_length,
1069 Node* array_length,
1070 RegionNode* region) {
1071 if (stopped())
1072 return NULL; // already stopped
1073 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
1074 if (zero_offset && subseq_length->eqv_uncast(array_length))
1075 return NULL; // common case of whole-array copy
1076 Node* last = subseq_length;
1077 if (!zero_offset) // last += offset
1078 last = _gvn.transform(new AddINode(last, offset));
1079 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
1080 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1081 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
1082 return is_over;
1083 }
1084
1085 // Emit range checks for the given String.value byte array
1086 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
1087 if (stopped()) {
1088 return; // already stopped
1089 }
1090 RegionNode* bailout = new RegionNode(1);
1091 record_for_igvn(bailout);
1092 if (char_count) {
1093 // Convert char count to byte count
1094 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1095 }
1096
1097 // Offset and count must not be negative
1098 generate_negative_guard(offset, bailout);
1099 generate_negative_guard(count, bailout);
1100 // Offset + count must not exceed length of array
1101 generate_limit_guard(offset, count, load_array_length(array), bailout);
1102
1103 if (bailout->req() > 1) {
1104 PreserveJVMState pjvms(this);
1105 set_control(_gvn.transform(bailout));
1106 uncommon_trap(Deoptimization::Reason_intrinsic,
1107 Deoptimization::Action_maybe_recompile);
1108 }
1109 }
1110
1111 //--------------------------generate_current_thread--------------------
1112 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1113 ciKlass* thread_klass = env()->Thread_klass();
1114 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1115 Node* thread = _gvn.transform(new ThreadLocalNode());
1116 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1117 Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered);
1118 tls_output = thread;
1119 return threadObj;
1120 }
1121
1122
1123 //------------------------------make_string_method_node------------------------
1124 // Helper method for String intrinsic functions. This version is called with
1125 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1126 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1127 // containing the lengths of str1 and str2.
1128 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1129 Node* result = NULL;
1130 switch (opcode) {
1131 case Op_StrIndexOf:
1132 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1133 str1_start, cnt1, str2_start, cnt2, ae);
1134 break;
1135 case Op_StrComp:
1136 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1137 str1_start, cnt1, str2_start, cnt2, ae);
1138 break;
1139 case Op_StrEquals:
1140 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1141 // Use the constant length if there is one because optimized match rule may exist.
1142 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1143 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1144 break;
1145 default:
1146 ShouldNotReachHere();
1147 return NULL;
1148 }
1149
1150 // All these intrinsics have checks.
1151 C->set_has_split_ifs(true); // Has chance for split-if optimization
1152 clear_upper_avx();
1153
1154 return _gvn.transform(result);
1155 }
1156
1157 //------------------------------inline_string_compareTo------------------------
1158 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1159 Node* arg1 = argument(0);
1160 Node* arg2 = argument(1);
1161
1162 arg1 = must_be_not_null(arg1, true);
1163 arg2 = must_be_not_null(arg2, true);
1164
1165 arg1 = access_resolve(arg1, ACCESS_READ);
1166 arg2 = access_resolve(arg2, ACCESS_READ);
1167
1168 // Get start addr and length of first argument
1169 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1170 Node* arg1_cnt = load_array_length(arg1);
1171
1172 // Get start addr and length of second argument
1173 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1174 Node* arg2_cnt = load_array_length(arg2);
1175
1176 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1177 set_result(result);
1178 return true;
1179 }
1180
1181 //------------------------------inline_string_equals------------------------
1182 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1183 Node* arg1 = argument(0);
1184 Node* arg2 = argument(1);
1185
1186 // paths (plus control) merge
1187 RegionNode* region = new RegionNode(3);
1188 Node* phi = new PhiNode(region, TypeInt::BOOL);
1189
1190 if (!stopped()) {
1191
1192 arg1 = must_be_not_null(arg1, true);
1193 arg2 = must_be_not_null(arg2, true);
1194
1195 arg1 = access_resolve(arg1, ACCESS_READ);
1196 arg2 = access_resolve(arg2, ACCESS_READ);
1197
1198 // Get start addr and length of first argument
1199 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1200 Node* arg1_cnt = load_array_length(arg1);
1201
1202 // Get start addr and length of second argument
1203 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1204 Node* arg2_cnt = load_array_length(arg2);
1205
1206 // Check for arg1_cnt != arg2_cnt
1207 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1208 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1209 Node* if_ne = generate_slow_guard(bol, NULL);
1210 if (if_ne != NULL) {
1211 phi->init_req(2, intcon(0));
1212 region->init_req(2, if_ne);
1213 }
1214
1215 // Check for count == 0 is done by assembler code for StrEquals.
1216
1217 if (!stopped()) {
1218 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1219 phi->init_req(1, equals);
1220 region->init_req(1, control());
1221 }
1222 }
1223
1224 // post merge
1225 set_control(_gvn.transform(region));
1226 record_for_igvn(region);
1227
1228 set_result(_gvn.transform(phi));
1229 return true;
1230 }
1231
1232 //------------------------------inline_array_equals----------------------------
1233 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1234 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1235 Node* arg1 = argument(0);
1236 Node* arg2 = argument(1);
1237
1238 arg1 = access_resolve(arg1, ACCESS_READ);
1239 arg2 = access_resolve(arg2, ACCESS_READ);
1240
1241 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1242 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1243 clear_upper_avx();
1244
1245 return true;
1246 }
1247
1248 //------------------------------inline_hasNegatives------------------------------
1249 bool LibraryCallKit::inline_hasNegatives() {
1250 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1251 return false;
1252 }
1253
1254 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1255 // no receiver since it is static method
1256 Node* ba = argument(0);
1257 Node* offset = argument(1);
1258 Node* len = argument(2);
1259
1260 ba = must_be_not_null(ba, true);
1261
1262 // Range checks
1263 generate_string_range_check(ba, offset, len, false);
1264 if (stopped()) {
1265 return true;
1266 }
1267 ba = access_resolve(ba, ACCESS_READ);
1268 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1269 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1270 set_result(_gvn.transform(result));
1271 return true;
1272 }
1273
1274 bool LibraryCallKit::inline_preconditions_checkIndex() {
1275 Node* index = argument(0);
1276 Node* length = argument(1);
1277 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1278 return false;
1279 }
1280
1281 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1282 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1283
1284 {
1285 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1286 uncommon_trap(Deoptimization::Reason_intrinsic,
1287 Deoptimization::Action_make_not_entrant);
1288 }
1289
1290 if (stopped()) {
1291 return false;
1292 }
1293
1294 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1295 BoolTest::mask btest = BoolTest::lt;
1296 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1297 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1298 _gvn.set_type(rc, rc->Value(&_gvn));
1299 if (!rc_bool->is_Con()) {
1300 record_for_igvn(rc);
1301 }
1302 set_control(_gvn.transform(new IfTrueNode(rc)));
1303 {
1304 PreserveJVMState pjvms(this);
1305 set_control(_gvn.transform(new IfFalseNode(rc)));
1306 uncommon_trap(Deoptimization::Reason_range_check,
1307 Deoptimization::Action_make_not_entrant);
1308 }
1309
1310 if (stopped()) {
1311 return false;
1312 }
1313
1314 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1315 result->set_req(0, control());
1316 result = _gvn.transform(result);
1317 set_result(result);
1318 replace_in_map(index, result);
1319 clear_upper_avx();
1320 return true;
1321 }
1322
1323 //------------------------------inline_string_indexOf------------------------
1324 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1325 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1326 return false;
1327 }
1328 Node* src = argument(0);
1329 Node* tgt = argument(1);
1330
1331 // Make the merge point
1332 RegionNode* result_rgn = new RegionNode(4);
1333 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1334
1335 src = must_be_not_null(src, true);
1336 tgt = must_be_not_null(tgt, true);
1337
1338 src = access_resolve(src, ACCESS_READ);
1339 tgt = access_resolve(tgt, ACCESS_READ);
1340
1341 // Get start addr and length of source string
1342 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1343 Node* src_count = load_array_length(src);
1344
1345 // Get start addr and length of substring
1346 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1347 Node* tgt_count = load_array_length(tgt);
1348
1349 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1350 // Divide src size by 2 if String is UTF16 encoded
1351 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1352 }
1353 if (ae == StrIntrinsicNode::UU) {
1354 // Divide substring size by 2 if String is UTF16 encoded
1355 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1356 }
1357
1358 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1359 if (result != NULL) {
1360 result_phi->init_req(3, result);
1361 result_rgn->init_req(3, control());
1362 }
1363 set_control(_gvn.transform(result_rgn));
1364 record_for_igvn(result_rgn);
1365 set_result(_gvn.transform(result_phi));
1366
1367 return true;
1368 }
1369
1370 //-----------------------------inline_string_indexOf-----------------------
1371 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1372 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1373 return false;
1374 }
1375 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1376 return false;
1377 }
1378 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1379 Node* src = argument(0); // byte[]
1380 Node* src_count = argument(1); // char count
1381 Node* tgt = argument(2); // byte[]
1382 Node* tgt_count = argument(3); // char count
1383 Node* from_index = argument(4); // char index
1384
1385 src = must_be_not_null(src, true);
1386 tgt = must_be_not_null(tgt, true);
1387
1388 src = access_resolve(src, ACCESS_READ);
1389 tgt = access_resolve(tgt, ACCESS_READ);
1390
1391 // Multiply byte array index by 2 if String is UTF16 encoded
1392 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1393 src_count = _gvn.transform(new SubINode(src_count, from_index));
1394 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1395 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1396
1397 // Range checks
1398 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1399 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1400 if (stopped()) {
1401 return true;
1402 }
1403
1404 RegionNode* region = new RegionNode(5);
1405 Node* phi = new PhiNode(region, TypeInt::INT);
1406
1407 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1408 if (result != NULL) {
1409 // The result is index relative to from_index if substring was found, -1 otherwise.
1410 // Generate code which will fold into cmove.
1411 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1412 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1413
1414 Node* if_lt = generate_slow_guard(bol, NULL);
1415 if (if_lt != NULL) {
1416 // result == -1
1417 phi->init_req(3, result);
1418 region->init_req(3, if_lt);
1419 }
1420 if (!stopped()) {
1421 result = _gvn.transform(new AddINode(result, from_index));
1422 phi->init_req(4, result);
1423 region->init_req(4, control());
1424 }
1425 }
1426
1427 set_control(_gvn.transform(region));
1428 record_for_igvn(region);
1429 set_result(_gvn.transform(phi));
1430 clear_upper_avx();
1431
1432 return true;
1433 }
1434
1435 // Create StrIndexOfNode with fast path checks
1436 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1437 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1438 // Check for substr count > string count
1439 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1440 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1441 Node* if_gt = generate_slow_guard(bol, NULL);
1442 if (if_gt != NULL) {
1443 phi->init_req(1, intcon(-1));
1444 region->init_req(1, if_gt);
1445 }
1446 if (!stopped()) {
1447 // Check for substr count == 0
1448 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1449 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1450 Node* if_zero = generate_slow_guard(bol, NULL);
1451 if (if_zero != NULL) {
1452 phi->init_req(2, intcon(0));
1453 region->init_req(2, if_zero);
1454 }
1455 }
1456 if (!stopped()) {
1457 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1458 }
1459 return NULL;
1460 }
1461
1462 //-----------------------------inline_string_indexOfChar-----------------------
1463 bool LibraryCallKit::inline_string_indexOfChar() {
1464 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1465 return false;
1466 }
1467 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1468 return false;
1469 }
1470 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1471 Node* src = argument(0); // byte[]
1472 Node* tgt = argument(1); // tgt is int ch
1473 Node* from_index = argument(2);
1474 Node* max = argument(3);
1475
1476 src = must_be_not_null(src, true);
1477 src = access_resolve(src, ACCESS_READ);
1478
1479 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1480 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1481 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1482
1483 // Range checks
1484 generate_string_range_check(src, src_offset, src_count, true);
1485 if (stopped()) {
1486 return true;
1487 }
1488
1489 RegionNode* region = new RegionNode(3);
1490 Node* phi = new PhiNode(region, TypeInt::INT);
1491
1492 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1493 C->set_has_split_ifs(true); // Has chance for split-if optimization
1494 _gvn.transform(result);
1495
1496 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1497 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1498
1499 Node* if_lt = generate_slow_guard(bol, NULL);
1500 if (if_lt != NULL) {
1501 // result == -1
1502 phi->init_req(2, result);
1503 region->init_req(2, if_lt);
1504 }
1505 if (!stopped()) {
1506 result = _gvn.transform(new AddINode(result, from_index));
1507 phi->init_req(1, result);
1508 region->init_req(1, control());
1509 }
1510 set_control(_gvn.transform(region));
1511 record_for_igvn(region);
1512 set_result(_gvn.transform(phi));
1513
1514 return true;
1515 }
1516 //---------------------------inline_string_copy---------------------
1517 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1518 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1519 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1520 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1521 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1522 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1523 bool LibraryCallKit::inline_string_copy(bool compress) {
1524 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1525 return false;
1526 }
1527 int nargs = 5; // 2 oops, 3 ints
1528 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1529
1530 Node* src = argument(0);
1531 Node* src_offset = argument(1);
1532 Node* dst = argument(2);
1533 Node* dst_offset = argument(3);
1534 Node* length = argument(4);
1535
1536 // Check for allocation before we add nodes that would confuse
1537 // tightly_coupled_allocation()
1538 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1539
1540 // Figure out the size and type of the elements we will be copying.
1541 const Type* src_type = src->Value(&_gvn);
1542 const Type* dst_type = dst->Value(&_gvn);
1543 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1544 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1545 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1546 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1547 "Unsupported array types for inline_string_copy");
1548
1549 src = must_be_not_null(src, true);
1550 dst = must_be_not_null(dst, true);
1551
1552 // Convert char[] offsets to byte[] offsets
1553 bool convert_src = (compress && src_elem == T_BYTE);
1554 bool convert_dst = (!compress && dst_elem == T_BYTE);
1555 if (convert_src) {
1556 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1557 } else if (convert_dst) {
1558 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1559 }
1560
1561 // Range checks
1562 generate_string_range_check(src, src_offset, length, convert_src);
1563 generate_string_range_check(dst, dst_offset, length, convert_dst);
1564 if (stopped()) {
1565 return true;
1566 }
1567
1568 src = access_resolve(src, ACCESS_READ);
1569 dst = access_resolve(dst, ACCESS_WRITE);
1570
1571 Node* src_start = array_element_address(src, src_offset, src_elem);
1572 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1573 // 'src_start' points to src array + scaled offset
1574 // 'dst_start' points to dst array + scaled offset
1575 Node* count = NULL;
1576 if (compress) {
1577 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1578 } else {
1579 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1580 }
1581
1582 if (alloc != NULL) {
1583 if (alloc->maybe_set_complete(&_gvn)) {
1584 // "You break it, you buy it."
1585 InitializeNode* init = alloc->initialization();
1586 assert(init->is_complete(), "we just did this");
1587 init->set_complete_with_arraycopy();
1588 assert(dst->is_CheckCastPP(), "sanity");
1589 assert(dst->in(0)->in(0) == init, "dest pinned");
1590 }
1591 // Do not let stores that initialize this object be reordered with
1592 // a subsequent store that would make this object accessible by
1593 // other threads.
1594 // Record what AllocateNode this StoreStore protects so that
1595 // escape analysis can go from the MemBarStoreStoreNode to the
1596 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1597 // based on the escape status of the AllocateNode.
1598 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1599 }
1600 if (compress) {
1601 set_result(_gvn.transform(count));
1602 }
1603 clear_upper_avx();
1604
1605 return true;
1606 }
1607
1608 #ifdef _LP64
1609 #define XTOP ,top() /*additional argument*/
1610 #else //_LP64
1611 #define XTOP /*no additional argument*/
1612 #endif //_LP64
1613
1614 //------------------------inline_string_toBytesU--------------------------
1615 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1616 bool LibraryCallKit::inline_string_toBytesU() {
1617 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1618 return false;
1619 }
1620 // Get the arguments.
1621 Node* value = argument(0);
1622 Node* offset = argument(1);
1623 Node* length = argument(2);
1624
1625 Node* newcopy = NULL;
1626
1627 // Set the original stack and the reexecute bit for the interpreter to reexecute
1628 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1629 { PreserveReexecuteState preexecs(this);
1630 jvms()->set_should_reexecute(true);
1631
1632 // Check if a null path was taken unconditionally.
1633 value = null_check(value);
1634
1635 RegionNode* bailout = new RegionNode(1);
1636 record_for_igvn(bailout);
1637
1638 // Range checks
1639 generate_negative_guard(offset, bailout);
1640 generate_negative_guard(length, bailout);
1641 generate_limit_guard(offset, length, load_array_length(value), bailout);
1642 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1643 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1644
1645 if (bailout->req() > 1) {
1646 PreserveJVMState pjvms(this);
1647 set_control(_gvn.transform(bailout));
1648 uncommon_trap(Deoptimization::Reason_intrinsic,
1649 Deoptimization::Action_maybe_recompile);
1650 }
1651 if (stopped()) {
1652 return true;
1653 }
1654
1655 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1656 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1657 newcopy = new_array(klass_node, size, 0); // no arguments to push
1658 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1659
1660 // Calculate starting addresses.
1661 value = access_resolve(value, ACCESS_READ);
1662 Node* src_start = array_element_address(value, offset, T_CHAR);
1663 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1664
1665 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1666 const TypeInt* toffset = gvn().type(offset)->is_int();
1667 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1668
1669 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1670 const char* copyfunc_name = "arraycopy";
1671 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1672 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1673 OptoRuntime::fast_arraycopy_Type(),
1674 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1675 src_start, dst_start, ConvI2X(length) XTOP);
1676 // Do not let reads from the cloned object float above the arraycopy.
1677 if (alloc != NULL) {
1678 if (alloc->maybe_set_complete(&_gvn)) {
1679 // "You break it, you buy it."
1680 InitializeNode* init = alloc->initialization();
1681 assert(init->is_complete(), "we just did this");
1682 init->set_complete_with_arraycopy();
1683 assert(newcopy->is_CheckCastPP(), "sanity");
1684 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1685 }
1686 // Do not let stores that initialize this object be reordered with
1687 // a subsequent store that would make this object accessible by
1688 // other threads.
1689 // Record what AllocateNode this StoreStore protects so that
1690 // escape analysis can go from the MemBarStoreStoreNode to the
1691 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1692 // based on the escape status of the AllocateNode.
1693 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1694 } else {
1695 insert_mem_bar(Op_MemBarCPUOrder);
1696 }
1697 } // original reexecute is set back here
1698
1699 C->set_has_split_ifs(true); // Has chance for split-if optimization
1700 if (!stopped()) {
1701 set_result(newcopy);
1702 }
1703 clear_upper_avx();
1704
1705 return true;
1706 }
1707
1708 //------------------------inline_string_getCharsU--------------------------
1709 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1710 bool LibraryCallKit::inline_string_getCharsU() {
1711 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1712 return false;
1713 }
1714
1715 // Get the arguments.
1716 Node* src = argument(0);
1717 Node* src_begin = argument(1);
1718 Node* src_end = argument(2); // exclusive offset (i < src_end)
1719 Node* dst = argument(3);
1720 Node* dst_begin = argument(4);
1721
1722 // Check for allocation before we add nodes that would confuse
1723 // tightly_coupled_allocation()
1724 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1725
1726 // Check if a null path was taken unconditionally.
1727 src = null_check(src);
1728 dst = null_check(dst);
1729 if (stopped()) {
1730 return true;
1731 }
1732
1733 // Get length and convert char[] offset to byte[] offset
1734 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1735 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1736
1737 // Range checks
1738 generate_string_range_check(src, src_begin, length, true);
1739 generate_string_range_check(dst, dst_begin, length, false);
1740 if (stopped()) {
1741 return true;
1742 }
1743
1744 if (!stopped()) {
1745 src = access_resolve(src, ACCESS_READ);
1746 dst = access_resolve(dst, ACCESS_WRITE);
1747
1748 // Calculate starting addresses.
1749 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1750 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1751
1752 // Check if array addresses are aligned to HeapWordSize
1753 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1754 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1755 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1756 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1757
1758 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1759 const char* copyfunc_name = "arraycopy";
1760 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1761 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1762 OptoRuntime::fast_arraycopy_Type(),
1763 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1764 src_start, dst_start, ConvI2X(length) XTOP);
1765 // Do not let reads from the cloned object float above the arraycopy.
1766 if (alloc != NULL) {
1767 if (alloc->maybe_set_complete(&_gvn)) {
1768 // "You break it, you buy it."
1769 InitializeNode* init = alloc->initialization();
1770 assert(init->is_complete(), "we just did this");
1771 init->set_complete_with_arraycopy();
1772 assert(dst->is_CheckCastPP(), "sanity");
1773 assert(dst->in(0)->in(0) == init, "dest pinned");
1774 }
1775 // Do not let stores that initialize this object be reordered with
1776 // a subsequent store that would make this object accessible by
1777 // other threads.
1778 // Record what AllocateNode this StoreStore protects so that
1779 // escape analysis can go from the MemBarStoreStoreNode to the
1780 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1781 // based on the escape status of the AllocateNode.
1782 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1783 } else {
1784 insert_mem_bar(Op_MemBarCPUOrder);
1785 }
1786 }
1787
1788 C->set_has_split_ifs(true); // Has chance for split-if optimization
1789 return true;
1790 }
1791
1792 //----------------------inline_string_char_access----------------------------
1793 // Store/Load char to/from byte[] array.
1794 // static void StringUTF16.putChar(byte[] val, int index, int c)
1795 // static char StringUTF16.getChar(byte[] val, int index)
1796 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1797 Node* value = argument(0);
1798 Node* index = argument(1);
1799 Node* ch = is_store ? argument(2) : NULL;
1800
1801 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1802 // correctly requires matched array shapes.
1803 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1804 "sanity: byte[] and char[] bases agree");
1805 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1806 "sanity: byte[] and char[] scales agree");
1807
1808 // Bail when getChar over constants is requested: constant folding would
1809 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1810 // Java method would constant fold nicely instead.
1811 if (!is_store && value->is_Con() && index->is_Con()) {
1812 return false;
1813 }
1814
1815 value = must_be_not_null(value, true);
1816 value = access_resolve(value, is_store ? ACCESS_WRITE : ACCESS_READ);
1817
1818 Node* adr = array_element_address(value, index, T_CHAR);
1819 if (adr->is_top()) {
1820 return false;
1821 }
1822 if (is_store) {
1823 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1824 } else {
1825 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
1826 set_result(ch);
1827 }
1828 return true;
1829 }
1830
1831 //--------------------------round_double_node--------------------------------
1832 // Round a double node if necessary.
1833 Node* LibraryCallKit::round_double_node(Node* n) {
1834 if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1835 n = _gvn.transform(new RoundDoubleNode(0, n));
1836 return n;
1837 }
1838
1839 //------------------------------inline_math-----------------------------------
1840 // public static double Math.abs(double)
1841 // public static double Math.sqrt(double)
1842 // public static double Math.log(double)
1843 // public static double Math.log10(double)
1844 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1845 Node* arg = round_double_node(argument(0));
1846 Node* n = NULL;
1847 switch (id) {
1848 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1849 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1850 case vmIntrinsics::_ceil: n = new RoundDoubleModeNode(arg, makecon(TypeInt::make(2))); break;
1851 case vmIntrinsics::_floor: n = new RoundDoubleModeNode(arg, makecon(TypeInt::make(1))); break;
1852 case vmIntrinsics::_rint: n = new RoundDoubleModeNode(arg, makecon(TypeInt::make(0))); break;
1853 default: fatal_unexpected_iid(id); break;
1854 }
1855 set_result(_gvn.transform(n));
1856 return true;
1857 }
1858
1859 //------------------------------inline_math-----------------------------------
1860 // public static float Math.abs(float)
1861 // public static int Math.abs(int)
1862 // public static long Math.abs(long)
1863 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1864 Node* arg = argument(0);
1865 Node* n = NULL;
1866 switch (id) {
1867 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1868 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1869 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1870 default: fatal_unexpected_iid(id); break;
1871 }
1872 set_result(_gvn.transform(n));
1873 return true;
1874 }
1875
1876 //------------------------------runtime_math-----------------------------
1877 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1878 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1879 "must be (DD)D or (D)D type");
1880
1881 // Inputs
1882 Node* a = round_double_node(argument(0));
1883 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1884
1885 const TypePtr* no_memory_effects = NULL;
1886 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1887 no_memory_effects,
1888 a, top(), b, b ? top() : NULL);
1889 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1890 #ifdef ASSERT
1891 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1892 assert(value_top == top(), "second value must be top");
1893 #endif
1894
1895 set_result(value);
1896 return true;
1897 }
1898
1899 //------------------------------inline_math_native-----------------------------
1900 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1901 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1902 switch (id) {
1903 // These intrinsics are not properly supported on all hardware
1904 case vmIntrinsics::_dsin:
1905 return StubRoutines::dsin() != NULL ?
1906 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1907 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1908 case vmIntrinsics::_dcos:
1909 return StubRoutines::dcos() != NULL ?
1910 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1911 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1912 case vmIntrinsics::_dtan:
1913 return StubRoutines::dtan() != NULL ?
1914 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1915 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1916 case vmIntrinsics::_dlog:
1917 return StubRoutines::dlog() != NULL ?
1918 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1919 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1920 case vmIntrinsics::_dlog10:
1921 return StubRoutines::dlog10() != NULL ?
1922 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1923 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1924
1925 // These intrinsics are supported on all hardware
1926 case vmIntrinsics::_ceil:
1927 case vmIntrinsics::_floor:
1928 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1929 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1930 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1931 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1932 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1933 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1934
1935 case vmIntrinsics::_dexp:
1936 return StubRoutines::dexp() != NULL ?
1937 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1938 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1939 case vmIntrinsics::_dpow: {
1940 Node* exp = round_double_node(argument(2));
1941 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1942 if (d != NULL && d->getd() == 2.0) {
1943 // Special case: pow(x, 2.0) => x * x
1944 Node* base = round_double_node(argument(0));
1945 set_result(_gvn.transform(new MulDNode(base, base)));
1946 return true;
1947 }
1948 return StubRoutines::dpow() != NULL ?
1949 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1950 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1951 }
1952 #undef FN_PTR
1953
1954 // These intrinsics are not yet correctly implemented
1955 case vmIntrinsics::_datan2:
1956 return false;
1957
1958 default:
1959 fatal_unexpected_iid(id);
1960 return false;
1961 }
1962 }
1963
1964 static bool is_simple_name(Node* n) {
1965 return (n->req() == 1 // constant
1966 || (n->is_Type() && n->as_Type()->type()->singleton())
1967 || n->is_Proj() // parameter or return value
1968 || n->is_Phi() // local of some sort
1969 );
1970 }
1971
1972 //----------------------------inline_notify-----------------------------------*
1973 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1974 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1975 address func;
1976 if (id == vmIntrinsics::_notify) {
1977 func = OptoRuntime::monitor_notify_Java();
1978 } else {
1979 func = OptoRuntime::monitor_notifyAll_Java();
1980 }
1981 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1982 make_slow_call_ex(call, env()->Throwable_klass(), false);
1983 return true;
1984 }
1985
1986
1987 //----------------------------inline_min_max-----------------------------------
1988 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1989 set_result(generate_min_max(id, argument(0), argument(1)));
1990 return true;
1991 }
1992
1993 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1994 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1995 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1996 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1997 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1998
1999 {
2000 PreserveJVMState pjvms(this);
2001 PreserveReexecuteState preexecs(this);
2002 jvms()->set_should_reexecute(true);
2003
2004 set_control(slow_path);
2005 set_i_o(i_o());
2006
2007 uncommon_trap(Deoptimization::Reason_intrinsic,
2008 Deoptimization::Action_none);
2009 }
2010
2011 set_control(fast_path);
2012 set_result(math);
2013 }
2014
2015 template <typename OverflowOp>
2016 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2017 typedef typename OverflowOp::MathOp MathOp;
2018
2019 MathOp* mathOp = new MathOp(arg1, arg2);
2020 Node* operation = _gvn.transform( mathOp );
2021 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2022 inline_math_mathExact(operation, ofcheck);
2023 return true;
2024 }
2025
2026 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2027 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2028 }
2029
2030 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2031 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2032 }
2033
2034 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2035 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2036 }
2037
2038 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2039 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2040 }
2041
2042 bool LibraryCallKit::inline_math_negateExactI() {
2043 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2044 }
2045
2046 bool LibraryCallKit::inline_math_negateExactL() {
2047 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2048 }
2049
2050 bool LibraryCallKit::inline_math_multiplyExactI() {
2051 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2052 }
2053
2054 bool LibraryCallKit::inline_math_multiplyExactL() {
2055 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2056 }
2057
2058 bool LibraryCallKit::inline_math_multiplyHigh() {
2059 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2060 return true;
2061 }
2062
2063 Node*
2064 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2065 // These are the candidate return value:
2066 Node* xvalue = x0;
2067 Node* yvalue = y0;
2068
2069 if (xvalue == yvalue) {
2070 return xvalue;
2071 }
2072
2073 bool want_max = (id == vmIntrinsics::_max);
2074
2075 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
2076 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
2077 if (txvalue == NULL || tyvalue == NULL) return top();
2078 // This is not really necessary, but it is consistent with a
2079 // hypothetical MaxINode::Value method:
2080 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
2081
2082 // %%% This folding logic should (ideally) be in a different place.
2083 // Some should be inside IfNode, and there to be a more reliable
2084 // transformation of ?: style patterns into cmoves. We also want
2085 // more powerful optimizations around cmove and min/max.
2086
2087 // Try to find a dominating comparison of these guys.
2088 // It can simplify the index computation for Arrays.copyOf
2089 // and similar uses of System.arraycopy.
2090 // First, compute the normalized version of CmpI(x, y).
2091 int cmp_op = Op_CmpI;
2092 Node* xkey = xvalue;
2093 Node* ykey = yvalue;
2094 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
2095 if (ideal_cmpxy->is_Cmp()) {
2096 // E.g., if we have CmpI(length - offset, count),
2097 // it might idealize to CmpI(length, count + offset)
2098 cmp_op = ideal_cmpxy->Opcode();
2099 xkey = ideal_cmpxy->in(1);
2100 ykey = ideal_cmpxy->in(2);
2101 }
2102
2103 // Start by locating any relevant comparisons.
2104 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
2105 Node* cmpxy = NULL;
2106 Node* cmpyx = NULL;
2107 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
2108 Node* cmp = start_from->fast_out(k);
2109 if (cmp->outcnt() > 0 && // must have prior uses
2110 cmp->in(0) == NULL && // must be context-independent
2111 cmp->Opcode() == cmp_op) { // right kind of compare
2112 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
2113 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
2114 }
2115 }
2116
2117 const int NCMPS = 2;
2118 Node* cmps[NCMPS] = { cmpxy, cmpyx };
2119 int cmpn;
2120 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2121 if (cmps[cmpn] != NULL) break; // find a result
2122 }
2123 if (cmpn < NCMPS) {
2124 // Look for a dominating test that tells us the min and max.
2125 int depth = 0; // Limit search depth for speed
2126 Node* dom = control();
2127 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2128 if (++depth >= 100) break;
2129 Node* ifproj = dom;
2130 if (!ifproj->is_Proj()) continue;
2131 Node* iff = ifproj->in(0);
2132 if (!iff->is_If()) continue;
2133 Node* bol = iff->in(1);
2134 if (!bol->is_Bool()) continue;
2135 Node* cmp = bol->in(1);
2136 if (cmp == NULL) continue;
2137 for (cmpn = 0; cmpn < NCMPS; cmpn++)
2138 if (cmps[cmpn] == cmp) break;
2139 if (cmpn == NCMPS) continue;
2140 BoolTest::mask btest = bol->as_Bool()->_test._test;
2141 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
2142 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2143 // At this point, we know that 'x btest y' is true.
2144 switch (btest) {
2145 case BoolTest::eq:
2146 // They are proven equal, so we can collapse the min/max.
2147 // Either value is the answer. Choose the simpler.
2148 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2149 return yvalue;
2150 return xvalue;
2151 case BoolTest::lt: // x < y
2152 case BoolTest::le: // x <= y
2153 return (want_max ? yvalue : xvalue);
2154 case BoolTest::gt: // x > y
2155 case BoolTest::ge: // x >= y
2156 return (want_max ? xvalue : yvalue);
2157 default:
2158 break;
2159 }
2160 }
2161 }
2162
2163 // We failed to find a dominating test.
2164 // Let's pick a test that might GVN with prior tests.
2165 Node* best_bol = NULL;
2166 BoolTest::mask best_btest = BoolTest::illegal;
2167 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2168 Node* cmp = cmps[cmpn];
2169 if (cmp == NULL) continue;
2170 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2171 Node* bol = cmp->fast_out(j);
2172 if (!bol->is_Bool()) continue;
2173 BoolTest::mask btest = bol->as_Bool()->_test._test;
2174 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2175 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2176 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2177 best_bol = bol->as_Bool();
2178 best_btest = btest;
2179 }
2180 }
2181 }
2182
2183 Node* answer_if_true = NULL;
2184 Node* answer_if_false = NULL;
2185 switch (best_btest) {
2186 default:
2187 if (cmpxy == NULL)
2188 cmpxy = ideal_cmpxy;
2189 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2190 // and fall through:
2191 case BoolTest::lt: // x < y
2192 case BoolTest::le: // x <= y
2193 answer_if_true = (want_max ? yvalue : xvalue);
2194 answer_if_false = (want_max ? xvalue : yvalue);
2195 break;
2196 case BoolTest::gt: // x > y
2197 case BoolTest::ge: // x >= y
2198 answer_if_true = (want_max ? xvalue : yvalue);
2199 answer_if_false = (want_max ? yvalue : xvalue);
2200 break;
2201 }
2202
2203 jint hi, lo;
2204 if (want_max) {
2205 // We can sharpen the minimum.
2206 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2207 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2208 } else {
2209 // We can sharpen the maximum.
2210 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2211 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2212 }
2213
2214 // Use a flow-free graph structure, to avoid creating excess control edges
2215 // which could hinder other optimizations.
2216 // Since Math.min/max is often used with arraycopy, we want
2217 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2218 Node* cmov = CMoveNode::make(NULL, best_bol,
2219 answer_if_false, answer_if_true,
2220 TypeInt::make(lo, hi, widen));
2221
2222 return _gvn.transform(cmov);
2223
2224 /*
2225 // This is not as desirable as it may seem, since Min and Max
2226 // nodes do not have a full set of optimizations.
2227 // And they would interfere, anyway, with 'if' optimizations
2228 // and with CMoveI canonical forms.
2229 switch (id) {
2230 case vmIntrinsics::_min:
2231 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2232 case vmIntrinsics::_max:
2233 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2234 default:
2235 ShouldNotReachHere();
2236 }
2237 */
2238 }
2239
2240 inline int
2241 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2242 const TypePtr* base_type = TypePtr::NULL_PTR;
2243 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2244 if (base_type == NULL) {
2245 // Unknown type.
2246 return Type::AnyPtr;
2247 } else if (base_type == TypePtr::NULL_PTR) {
2248 // Since this is a NULL+long form, we have to switch to a rawptr.
2249 base = _gvn.transform(new CastX2PNode(offset));
2250 offset = MakeConX(0);
2251 return Type::RawPtr;
2252 } else if (base_type->base() == Type::RawPtr) {
2253 return Type::RawPtr;
2254 } else if (base_type->isa_oopptr()) {
2255 // Base is never null => always a heap address.
2256 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2257 return Type::OopPtr;
2258 }
2259 // Offset is small => always a heap address.
2260 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2261 if (offset_type != NULL &&
2262 base_type->offset() == 0 && // (should always be?)
2263 offset_type->_lo >= 0 &&
2264 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2265 return Type::OopPtr;
2266 } else if (type == T_OBJECT) {
2267 // off heap access to an oop doesn't make any sense. Has to be on
2268 // heap.
2269 return Type::OopPtr;
2270 }
2271 // Otherwise, it might either be oop+off or NULL+addr.
2272 return Type::AnyPtr;
2273 } else {
2274 // No information:
2275 return Type::AnyPtr;
2276 }
2277 }
2278
2279 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type, bool can_cast) {
2280 Node* uncasted_base = base;
2281 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2282 if (kind == Type::RawPtr) {
2283 return basic_plus_adr(top(), uncasted_base, offset);
2284 } else if (kind == Type::AnyPtr) {
2285 assert(base == uncasted_base, "unexpected base change");
2286 if (can_cast) {
2287 if (!_gvn.type(base)->speculative_maybe_null() &&
2288 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2289 // According to profiling, this access is always on
2290 // heap. Casting the base to not null and thus avoiding membars
2291 // around the access should allow better optimizations
2292 Node* null_ctl = top();
2293 base = null_check_oop(base, &null_ctl, true, true, true);
2294 assert(null_ctl->is_top(), "no null control here");
2295 return basic_plus_adr(base, offset);
2296 } else if (_gvn.type(base)->speculative_always_null() &&
2297 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2298 // According to profiling, this access is always off
2299 // heap.
2300 base = null_assert(base);
2301 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2302 offset = MakeConX(0);
2303 return basic_plus_adr(top(), raw_base, offset);
2304 }
2305 }
2306 // We don't know if it's an on heap or off heap access. Fall back
2307 // to raw memory access.
2308 base = access_resolve(base, decorators);
2309 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2310 return basic_plus_adr(top(), raw, offset);
2311 } else {
2312 assert(base == uncasted_base, "unexpected base change");
2313 // We know it's an on heap access so base can't be null
2314 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2315 base = must_be_not_null(base, true);
2316 }
2317 return basic_plus_adr(base, offset);
2318 }
2319 }
2320
2321 //--------------------------inline_number_methods-----------------------------
2322 // inline int Integer.numberOfLeadingZeros(int)
2323 // inline int Long.numberOfLeadingZeros(long)
2324 //
2325 // inline int Integer.numberOfTrailingZeros(int)
2326 // inline int Long.numberOfTrailingZeros(long)
2327 //
2328 // inline int Integer.bitCount(int)
2329 // inline int Long.bitCount(long)
2330 //
2331 // inline char Character.reverseBytes(char)
2332 // inline short Short.reverseBytes(short)
2333 // inline int Integer.reverseBytes(int)
2334 // inline long Long.reverseBytes(long)
2335 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2336 Node* arg = argument(0);
2337 Node* n = NULL;
2338 switch (id) {
2339 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2340 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2341 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2342 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2343 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2344 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2345 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2346 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2347 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2348 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2349 default: fatal_unexpected_iid(id); break;
2350 }
2351 set_result(_gvn.transform(n));
2352 return true;
2353 }
2354
2355 //----------------------------inline_unsafe_access----------------------------
2356
2357 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2358 // Attempt to infer a sharper value type from the offset and base type.
2359 ciKlass* sharpened_klass = NULL;
2360
2361 // See if it is an instance field, with an object type.
2362 if (alias_type->field() != NULL) {
2363 if (alias_type->field()->type()->is_klass()) {
2364 sharpened_klass = alias_type->field()->type()->as_klass();
2365 }
2366 }
2367
2368 // See if it is a narrow oop array.
2369 if (adr_type->isa_aryptr()) {
2370 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2371 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2372 if (elem_type != NULL) {
2373 sharpened_klass = elem_type->klass();
2374 }
2375 }
2376 }
2377
2378 // The sharpened class might be unloaded if there is no class loader
2379 // contraint in place.
2380 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2381 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2382
2383 #ifndef PRODUCT
2384 if (C->print_intrinsics() || C->print_inlining()) {
2385 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2386 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2387 }
2388 #endif
2389 // Sharpen the value type.
2390 return tjp;
2391 }
2392 return NULL;
2393 }
2394
2395 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2396 switch (kind) {
2397 case Relaxed:
2398 return MO_UNORDERED;
2399 case Opaque:
2400 return MO_RELAXED;
2401 case Acquire:
2402 return MO_ACQUIRE;
2403 case Release:
2404 return MO_RELEASE;
2405 case Volatile:
2406 return MO_SEQ_CST;
2407 default:
2408 ShouldNotReachHere();
2409 return 0;
2410 }
2411 }
2412
2413 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2414 if (callee()->is_static()) return false; // caller must have the capability!
2415 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2416 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2417 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2418 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2419
2420 if (is_reference_type(type)) {
2421 decorators |= ON_UNKNOWN_OOP_REF;
2422 }
2423
2424 if (unaligned) {
2425 decorators |= C2_UNALIGNED;
2426 }
2427
2428 #ifndef PRODUCT
2429 {
2430 ResourceMark rm;
2431 // Check the signatures.
2432 ciSignature* sig = callee()->signature();
2433 #ifdef ASSERT
2434 if (!is_store) {
2435 // Object getReference(Object base, int/long offset), etc.
2436 BasicType rtype = sig->return_type()->basic_type();
2437 assert(rtype == type || (rtype == T_OBJECT && type == T_VALUETYPE), "getter must return the expected value");
2438 assert(sig->count() == 2 || (type == T_VALUETYPE && sig->count() == 3), "oop getter has 2 or 3 arguments");
2439 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2440 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2441 } else {
2442 // void putReference(Object base, int/long offset, Object x), etc.
2443 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2444 assert(sig->count() == 3 || (type == T_VALUETYPE && sig->count() == 4), "oop putter has 3 arguments");
2445 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2446 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2447 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2448 assert(vtype == type || (type == T_VALUETYPE && vtype == T_OBJECT), "putter must accept the expected value");
2449 }
2450 #endif // ASSERT
2451 }
2452 #endif //PRODUCT
2453
2454 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2455
2456 Node* receiver = argument(0); // type: oop
2457
2458 // Build address expression.
2459 Node* adr;
2460 Node* heap_base_oop = top();
2461 Node* offset = top();
2462 Node* val;
2463
2464 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2465 Node* base = argument(1); // type: oop
2466 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2467 offset = argument(2); // type: long
2468 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2469 // to be plain byte offsets, which are also the same as those accepted
2470 // by oopDesc::field_addr.
2471 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2472 "fieldOffset must be byte-scaled");
2473
2474 ciValueKlass* value_klass = NULL;
2475 if (type == T_VALUETYPE) {
2476 Node* cls = null_check(argument(4));
2477 if (stopped()) {
2478 return true;
2479 }
2480 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2481 const TypeKlassPtr* kls_t = _gvn.type(kls)->isa_klassptr();
2482 if (!kls_t->klass_is_exact()) {
2483 return false;
2484 }
2485 ciKlass* klass = kls_t->klass();
2486 if (!klass->is_valuetype()) {
2487 return false;
2488 }
2489 value_klass = klass->as_value_klass();
2490 }
2491
2492 receiver = null_check(receiver);
2493 if (stopped()) {
2494 return true;
2495 }
2496
2497 if (base->is_ValueType()) {
2498 ValueTypeNode* vt = base->as_ValueType();
2499
2500 if (is_store) {
2501 if (!vt->is_allocated(&_gvn) || !_gvn.type(vt)->is_valuetype()->larval()) {
2502 return false;
2503 }
2504 base = vt->get_oop();
2505 } else {
2506 if (offset->is_Con()) {
2507 long off = find_long_con(offset, 0);
2508 ciValueKlass* vk = vt->type()->value_klass();
2509 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2510 return false;
2511 }
2512
2513 ciField* f = vk->get_non_flattened_field_by_offset((int)off);
2514
2515 if (f != NULL) {
2516 BasicType bt = f->layout_type();
2517 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2518 bt = T_OBJECT;
2519 }
2520 if (bt == type) {
2521 if (bt != T_VALUETYPE || f->type() == value_klass) {
2522 set_result(vt->field_value_by_offset((int)off, false));
2523 return true;
2524 }
2525 }
2526 }
2527 }
2528 vt = vt->allocate(this)->as_ValueType();
2529 base = vt->get_oop();
2530 }
2531 }
2532
2533 // 32-bit machines ignore the high half!
2534 offset = ConvL2X(offset);
2535 adr = make_unsafe_address(base, offset, is_store ? ACCESS_WRITE : ACCESS_READ, type, kind == Relaxed);
2536
2537 if (_gvn.type(base)->isa_ptr() != TypePtr::NULL_PTR) {
2538 heap_base_oop = base;
2539 } else if (type == T_OBJECT || (value_klass != NULL && value_klass->has_object_fields())) {
2540 return false; // off-heap oop accesses are not supported
2541 }
2542
2543 // Can base be NULL? Otherwise, always on-heap access.
2544 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2545
2546 if (!can_access_non_heap) {
2547 decorators |= IN_HEAP;
2548 }
2549
2550 val = is_store ? argument(4 + (type == T_VALUETYPE ? 1 : 0)) : NULL;
2551
2552 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2553 if (adr_type == TypePtr::NULL_PTR) {
2554 return false; // off-heap access with zero address
2555 }
2556
2557 // Try to categorize the address.
2558 Compile::AliasType* alias_type = C->alias_type(adr_type);
2559 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2560
2561 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2562 alias_type->adr_type() == TypeAryPtr::RANGE) {
2563 return false; // not supported
2564 }
2565
2566 bool mismatched = false;
2567 BasicType bt = T_ILLEGAL;
2568 ciField* field = NULL;
2569 if (adr_type->isa_instptr()) {
2570 const TypeInstPtr* instptr = adr_type->is_instptr();
2571 ciInstanceKlass* k = instptr->klass()->as_instance_klass();
2572 int off = instptr->offset();
2573 if (instptr->const_oop() != NULL &&
2574 instptr->klass() == ciEnv::current()->Class_klass() &&
2575 instptr->offset() >= (instptr->klass()->as_instance_klass()->size_helper() * wordSize)) {
2576 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2577 field = k->get_field_by_offset(off, true);
2578 } else {
2579 field = k->get_non_flattened_field_by_offset(off);
2580 }
2581 if (field != NULL) {
2582 bt = field->layout_type();
2583 }
2584 assert(bt == alias_type->basic_type() || bt == T_VALUETYPE, "should match");
2585 if (field != NULL && bt == T_VALUETYPE && !field->is_flattened()) {
2586 bt = T_OBJECT;
2587 }
2588 } else {
2589 bt = alias_type->basic_type();
2590 }
2591
2592 if (bt != T_ILLEGAL) {
2593 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2594 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2595 // Alias type doesn't differentiate between byte[] and boolean[]).
2596 // Use address type to get the element type.
2597 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2598 }
2599 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2600 // accessing an array field with getReference is not a mismatch
2601 bt = T_OBJECT;
2602 }
2603 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2604 // Don't intrinsify mismatched object accesses
2605 return false;
2606 }
2607 mismatched = (bt != type);
2608 } else if (alias_type->adr_type()->isa_oopptr()) {
2609 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2610 }
2611
2612 if (type == T_VALUETYPE) {
2613 if (adr_type->isa_instptr()) {
2614 if (field == NULL || field->type() != value_klass) {
2615 mismatched = true;
2616 }
2617 } else if (adr_type->isa_aryptr()) {
2618 const Type* elem = adr_type->is_aryptr()->elem();
2619 if (!elem->isa_valuetype()) {
2620 mismatched = true;
2621 } else if (elem->value_klass() != value_klass) {
2622 mismatched = true;
2623 }
2624 }
2625 if (is_store) {
2626 const Type* val_t = _gvn.type(val);
2627 if (!val_t->isa_valuetype() || val_t->value_klass() != value_klass) {
2628 return false;
2629 }
2630 }
2631 }
2632
2633 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2634
2635 if (mismatched) {
2636 decorators |= C2_MISMATCHED;
2637 }
2638
2639 // First guess at the value type.
2640 const Type *value_type = Type::get_const_basic_type(type);
2641
2642 // Figure out the memory ordering.
2643 decorators |= mo_decorator_for_access_kind(kind);
2644
2645 if (!is_store) {
2646 if (type == T_OBJECT) {
2647 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2648 if (tjp != NULL) {
2649 value_type = tjp;
2650 }
2651 } else if (type == T_VALUETYPE) {
2652 value_type = NULL;
2653 }
2654 }
2655
2656 // Heap pointers get a null-check from the interpreter,
2657 // as a courtesy. However, this is not guaranteed by Unsafe,
2658 // and it is not possible to fully distinguish unintended nulls
2659 // from intended ones in this API.
2660
2661 if (!is_store) {
2662 Node* p = NULL;
2663 // Try to constant fold a load from a constant field
2664
2665 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2666 // final or stable field
2667 p = make_constant_from_field(field, heap_base_oop);
2668 }
2669
2670 if (p == NULL) { // Could not constant fold the load
2671 if (type == T_VALUETYPE) {
2672 if (adr_type->isa_instptr() && !mismatched) {
2673 ciInstanceKlass* holder = adr_type->is_instptr()->klass()->as_instance_klass();
2674 int offset = adr_type->is_instptr()->offset();
2675 p = ValueTypeNode::make_from_flattened(this, value_klass, base, base, holder, offset, decorators);
2676 } else {
2677 p = ValueTypeNode::make_from_flattened(this, value_klass, base, adr, NULL, 0, decorators);
2678 }
2679 } else {
2680 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2681 }
2682 // Normalize the value returned by getBoolean in the following cases
2683 if (type == T_BOOLEAN &&
2684 (mismatched ||
2685 heap_base_oop == top() || // - heap_base_oop is NULL or
2686 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2687 // and the unsafe access is made to large offset
2688 // (i.e., larger than the maximum offset necessary for any
2689 // field access)
2690 ) {
2691 IdealKit ideal = IdealKit(this);
2692 #define __ ideal.
2693 IdealVariable normalized_result(ideal);
2694 __ declarations_done();
2695 __ set(normalized_result, p);
2696 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2697 __ set(normalized_result, ideal.ConI(1));
2698 ideal.end_if();
2699 final_sync(ideal);
2700 p = __ value(normalized_result);
2701 #undef __
2702 }
2703 }
2704 if (type == T_ADDRESS) {
2705 p = gvn().transform(new CastP2XNode(NULL, p));
2706 p = ConvX2UL(p);
2707 }
2708 if (field != NULL && field->is_flattenable() && !field->is_flattened()) {
2709 // Load a non-flattened but flattenable value type from memory
2710 if (value_type->value_klass()->is_scalarizable()) {
2711 p = ValueTypeNode::make_from_oop(this, p, value_type->value_klass());
2712 } else {
2713 p = null2default(p, value_type->value_klass());
2714 }
2715 }
2716 // The load node has the control of the preceding MemBarCPUOrder. All
2717 // following nodes will have the control of the MemBarCPUOrder inserted at
2718 // the end of this method. So, pushing the load onto the stack at a later
2719 // point is fine.
2720 set_result(p);
2721 } else {
2722 if (bt == T_ADDRESS) {
2723 // Repackage the long as a pointer.
2724 val = ConvL2X(val);
2725 val = gvn().transform(new CastX2PNode(val));
2726 }
2727 if (type == T_VALUETYPE) {
2728 if (adr_type->isa_instptr() && !mismatched) {
2729 ciInstanceKlass* holder = adr_type->is_instptr()->klass()->as_instance_klass();
2730 int offset = adr_type->is_instptr()->offset();
2731 val->as_ValueType()->store_flattened(this, base, base, holder, offset, decorators);
2732 } else {
2733 val->as_ValueType()->store_flattened(this, base, adr, NULL, 0, decorators);
2734 }
2735 } else {
2736 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2737 }
2738 }
2739
2740 if (argument(1)->is_ValueType() && is_store) {
2741 Node* value = ValueTypeNode::make_from_oop(this, base, _gvn.type(base)->value_klass());
2742 value = value->as_ValueType()->make_larval(this, false);
2743 replace_in_map(argument(1), value);
2744 }
2745
2746 return true;
2747 }
2748
2749 bool LibraryCallKit::inline_unsafe_make_private_buffer() {
2750 Node* receiver = argument(0);
2751 Node* value = argument(1);
2752
2753 receiver = null_check(receiver);
2754 if (stopped()) {
2755 return true;
2756 }
2757
2758 if (!value->is_ValueType()) {
2759 return false;
2760 }
2761
2762 set_result(value->as_ValueType()->make_larval(this, true));
2763
2764 return true;
2765 }
2766
2767 bool LibraryCallKit::inline_unsafe_finish_private_buffer() {
2768 Node* receiver = argument(0);
2769 Node* buffer = argument(1);
2770
2771 receiver = null_check(receiver);
2772 if (stopped()) {
2773 return true;
2774 }
2775
2776 if (!buffer->is_ValueType()) {
2777 return false;
2778 }
2779
2780 ValueTypeNode* vt = buffer->as_ValueType();
2781 if (!vt->is_allocated(&_gvn) || !_gvn.type(vt)->is_valuetype()->larval()) {
2782 return false;
2783 }
2784
2785 set_result(vt->finish_larval(this));
2786
2787 return true;
2788 }
2789
2790 //----------------------------inline_unsafe_load_store----------------------------
2791 // This method serves a couple of different customers (depending on LoadStoreKind):
2792 //
2793 // LS_cmp_swap:
2794 //
2795 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2796 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2797 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2798 //
2799 // LS_cmp_swap_weak:
2800 //
2801 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2802 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2803 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2804 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2805 //
2806 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2807 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2808 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2809 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2810 //
2811 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2812 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2813 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2814 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2815 //
2816 // LS_cmp_exchange:
2817 //
2818 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2819 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2820 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2821 //
2822 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2823 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2824 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2825 //
2826 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2827 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2828 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2829 //
2830 // LS_get_add:
2831 //
2832 // int getAndAddInt( Object o, long offset, int delta)
2833 // long getAndAddLong(Object o, long offset, long delta)
2834 //
2835 // LS_get_set:
2836 //
2837 // int getAndSet(Object o, long offset, int newValue)
2838 // long getAndSet(Object o, long offset, long newValue)
2839 // Object getAndSet(Object o, long offset, Object newValue)
2840 //
2841 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2842 // This basic scheme here is the same as inline_unsafe_access, but
2843 // differs in enough details that combining them would make the code
2844 // overly confusing. (This is a true fact! I originally combined
2845 // them, but even I was confused by it!) As much code/comments as
2846 // possible are retained from inline_unsafe_access though to make
2847 // the correspondences clearer. - dl
2848
2849 if (callee()->is_static()) return false; // caller must have the capability!
2850
2851 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2852 decorators |= mo_decorator_for_access_kind(access_kind);
2853
2854 #ifndef PRODUCT
2855 BasicType rtype;
2856 {
2857 ResourceMark rm;
2858 // Check the signatures.
2859 ciSignature* sig = callee()->signature();
2860 rtype = sig->return_type()->basic_type();
2861 switch(kind) {
2862 case LS_get_add:
2863 case LS_get_set: {
2864 // Check the signatures.
2865 #ifdef ASSERT
2866 assert(rtype == type, "get and set must return the expected type");
2867 assert(sig->count() == 3, "get and set has 3 arguments");
2868 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2869 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2870 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2871 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2872 #endif // ASSERT
2873 break;
2874 }
2875 case LS_cmp_swap:
2876 case LS_cmp_swap_weak: {
2877 // Check the signatures.
2878 #ifdef ASSERT
2879 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2880 assert(sig->count() == 4, "CAS has 4 arguments");
2881 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2882 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2883 #endif // ASSERT
2884 break;
2885 }
2886 case LS_cmp_exchange: {
2887 // Check the signatures.
2888 #ifdef ASSERT
2889 assert(rtype == type, "CAS must return the expected type");
2890 assert(sig->count() == 4, "CAS has 4 arguments");
2891 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2892 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2893 #endif // ASSERT
2894 break;
2895 }
2896 default:
2897 ShouldNotReachHere();
2898 }
2899 }
2900 #endif //PRODUCT
2901
2902 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2903
2904 // Get arguments:
2905 Node* receiver = NULL;
2906 Node* base = NULL;
2907 Node* offset = NULL;
2908 Node* oldval = NULL;
2909 Node* newval = NULL;
2910 switch(kind) {
2911 case LS_cmp_swap:
2912 case LS_cmp_swap_weak:
2913 case LS_cmp_exchange: {
2914 const bool two_slot_type = type2size[type] == 2;
2915 receiver = argument(0); // type: oop
2916 base = argument(1); // type: oop
2917 offset = argument(2); // type: long
2918 oldval = argument(4); // type: oop, int, or long
2919 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2920 break;
2921 }
2922 case LS_get_add:
2923 case LS_get_set: {
2924 receiver = argument(0); // type: oop
2925 base = argument(1); // type: oop
2926 offset = argument(2); // type: long
2927 oldval = NULL;
2928 newval = argument(4); // type: oop, int, or long
2929 break;
2930 }
2931 default:
2932 ShouldNotReachHere();
2933 }
2934
2935 // Build field offset expression.
2936 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2937 // to be plain byte offsets, which are also the same as those accepted
2938 // by oopDesc::field_addr.
2939 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2940 // 32-bit machines ignore the high half of long offsets
2941 offset = ConvL2X(offset);
2942 Node* adr = make_unsafe_address(base, offset, ACCESS_WRITE | ACCESS_READ, type, false);
2943 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2944
2945 Compile::AliasType* alias_type = C->alias_type(adr_type);
2946 BasicType bt = alias_type->basic_type();
2947 if (bt != T_ILLEGAL &&
2948 (is_reference_type(bt) != (type == T_OBJECT))) {
2949 // Don't intrinsify mismatched object accesses.
2950 return false;
2951 }
2952
2953 // For CAS, unlike inline_unsafe_access, there seems no point in
2954 // trying to refine types. Just use the coarse types here.
2955 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2956 const Type *value_type = Type::get_const_basic_type(type);
2957
2958 switch (kind) {
2959 case LS_get_set:
2960 case LS_cmp_exchange: {
2961 if (type == T_OBJECT) {
2962 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2963 if (tjp != NULL) {
2964 value_type = tjp;
2965 }
2966 }
2967 break;
2968 }
2969 case LS_cmp_swap:
2970 case LS_cmp_swap_weak:
2971 case LS_get_add:
2972 break;
2973 default:
2974 ShouldNotReachHere();
2975 }
2976
2977 // Null check receiver.
2978 receiver = null_check(receiver);
2979 if (stopped()) {
2980 return true;
2981 }
2982
2983 int alias_idx = C->get_alias_index(adr_type);
2984
2985 if (is_reference_type(type)) {
2986 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2987
2988 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2989 // could be delayed during Parse (for example, in adjust_map_after_if()).
2990 // Execute transformation here to avoid barrier generation in such case.
2991 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2992 newval = _gvn.makecon(TypePtr::NULL_PTR);
2993
2994 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2995 // Refine the value to a null constant, when it is known to be null
2996 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2997 }
2998 }
2999
3000 Node* result = NULL;
3001 switch (kind) {
3002 case LS_cmp_exchange: {
3003 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
3004 oldval, newval, value_type, type, decorators);
3005 break;
3006 }
3007 case LS_cmp_swap_weak:
3008 decorators |= C2_WEAK_CMPXCHG;
3009 case LS_cmp_swap: {
3010 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
3011 oldval, newval, value_type, type, decorators);
3012 break;
3013 }
3014 case LS_get_set: {
3015 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
3016 newval, value_type, type, decorators);
3017 break;
3018 }
3019 case LS_get_add: {
3020 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
3021 newval, value_type, type, decorators);
3022 break;
3023 }
3024 default:
3025 ShouldNotReachHere();
3026 }
3027
3028 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3029 set_result(result);
3030 return true;
3031 }
3032
3033 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3034 // Regardless of form, don't allow previous ld/st to move down,
3035 // then issue acquire, release, or volatile mem_bar.
3036 insert_mem_bar(Op_MemBarCPUOrder);
3037 switch(id) {
3038 case vmIntrinsics::_loadFence:
3039 insert_mem_bar(Op_LoadFence);
3040 return true;
3041 case vmIntrinsics::_storeFence:
3042 insert_mem_bar(Op_StoreFence);
3043 return true;
3044 case vmIntrinsics::_fullFence:
3045 insert_mem_bar(Op_MemBarVolatile);
3046 return true;
3047 default:
3048 fatal_unexpected_iid(id);
3049 return false;
3050 }
3051 }
3052
3053 bool LibraryCallKit::inline_onspinwait() {
3054 insert_mem_bar(Op_OnSpinWait);
3055 return true;
3056 }
3057
3058 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3059 if (!kls->is_Con()) {
3060 return true;
3061 }
3062 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
3063 if (klsptr == NULL) {
3064 return true;
3065 }
3066 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
3067 // don't need a guard for a klass that is already initialized
3068 return !ik->is_initialized();
3069 }
3070
3071 //----------------------------inline_unsafe_writeback0-------------------------
3072 // public native void Unsafe.writeback0(long address)
3073 bool LibraryCallKit::inline_unsafe_writeback0() {
3074 if (!Matcher::has_match_rule(Op_CacheWB)) {
3075 return false;
3076 }
3077 #ifndef PRODUCT
3078 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
3079 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
3080 ciSignature* sig = callee()->signature();
3081 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
3082 #endif
3083 null_check_receiver(); // null-check, then ignore
3084 Node *addr = argument(1);
3085 addr = new CastX2PNode(addr);
3086 addr = _gvn.transform(addr);
3087 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
3088 flush = _gvn.transform(flush);
3089 set_memory(flush, TypeRawPtr::BOTTOM);
3090 return true;
3091 }
3092
3093 //----------------------------inline_unsafe_writeback0-------------------------
3094 // public native void Unsafe.writeback0(long address)
3095 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
3096 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
3097 return false;
3098 }
3099 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
3100 return false;
3101 }
3102 #ifndef PRODUCT
3103 assert(Matcher::has_match_rule(Op_CacheWB),
3104 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
3105 : "found match rule for CacheWBPostSync but not CacheWB"));
3106
3107 #endif
3108 null_check_receiver(); // null-check, then ignore
3109 Node *sync;
3110 if (is_pre) {
3111 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3112 } else {
3113 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3114 }
3115 sync = _gvn.transform(sync);
3116 set_memory(sync, TypeRawPtr::BOTTOM);
3117 return true;
3118 }
3119
3120 //----------------------------inline_unsafe_allocate---------------------------
3121 // public native Object Unsafe.allocateInstance(Class<?> cls);
3122 bool LibraryCallKit::inline_unsafe_allocate() {
3123 if (callee()->is_static()) return false; // caller must have the capability!
3124
3125 null_check_receiver(); // null-check, then ignore
3126 Node* cls = null_check(argument(1));
3127 if (stopped()) return true;
3128
3129 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3130 kls = null_check(kls);
3131 if (stopped()) return true; // argument was like int.class
3132
3133 Node* test = NULL;
3134 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3135 // Note: The argument might still be an illegal value like
3136 // Serializable.class or Object[].class. The runtime will handle it.
3137 // But we must make an explicit check for initialization.
3138 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3139 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3140 // can generate code to load it as unsigned byte.
3141 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3142 Node* bits = intcon(InstanceKlass::fully_initialized);
3143 test = _gvn.transform(new SubINode(inst, bits));
3144 // The 'test' is non-zero if we need to take a slow path.
3145 }
3146
3147 Node* obj = new_instance(kls, test);
3148 set_result(obj);
3149 return true;
3150 }
3151
3152 //------------------------inline_native_time_funcs--------------
3153 // inline code for System.currentTimeMillis() and System.nanoTime()
3154 // these have the same type and signature
3155 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3156 const TypeFunc* tf = OptoRuntime::void_long_Type();
3157 const TypePtr* no_memory_effects = NULL;
3158 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3159 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3160 #ifdef ASSERT
3161 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3162 assert(value_top == top(), "second value must be top");
3163 #endif
3164 set_result(value);
3165 return true;
3166 }
3167
3168 #ifdef JFR_HAVE_INTRINSICS
3169
3170 /*
3171 * oop -> myklass
3172 * myklass->trace_id |= USED
3173 * return myklass->trace_id & ~0x3
3174 */
3175 bool LibraryCallKit::inline_native_classID() {
3176 Node* cls = null_check(argument(0), T_OBJECT);
3177 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3178 kls = null_check(kls, T_OBJECT);
3179
3180 ByteSize offset = KLASS_TRACE_ID_OFFSET;
3181 Node* insp = basic_plus_adr(kls, in_bytes(offset));
3182 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
3183
3184 Node* clsused = longcon(0x01l); // set the class bit
3185 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
3186 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
3187 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
3188
3189 #ifdef TRACE_ID_META_BITS
3190 Node* mbits = longcon(~TRACE_ID_META_BITS);
3191 tvalue = _gvn.transform(new AndLNode(tvalue, mbits));
3192 #endif
3193 #ifdef TRACE_ID_SHIFT
3194 Node* cbits = intcon(TRACE_ID_SHIFT);
3195 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits));
3196 #endif
3197
3198 set_result(tvalue);
3199 return true;
3200
3201 }
3202
3203 bool LibraryCallKit::inline_native_getEventWriter() {
3204 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3205
3206 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
3207 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3208
3209 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3210
3211 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
3212 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
3213
3214 IfNode* iff_jobj_null =
3215 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
3216
3217 enum { _normal_path = 1,
3218 _null_path = 2,
3219 PATH_LIMIT };
3220
3221 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3222 PhiNode* result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3223
3224 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
3225 result_rgn->init_req(_null_path, jobj_is_null);
3226 result_val->init_req(_null_path, null());
3227
3228 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
3229 set_control(jobj_is_not_null);
3230 Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT,
3231 IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3232 result_rgn->init_req(_normal_path, control());
3233 result_val->init_req(_normal_path, res);
3234
3235 set_result(result_rgn, result_val);
3236
3237 return true;
3238 }
3239
3240 #endif // JFR_HAVE_INTRINSICS
3241
3242 //------------------------inline_native_currentThread------------------
3243 bool LibraryCallKit::inline_native_currentThread() {
3244 Node* junk = NULL;
3245 set_result(generate_current_thread(junk));
3246 return true;
3247 }
3248
3249 //------------------------inline_native_isInterrupted------------------
3250 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
3251 bool LibraryCallKit::inline_native_isInterrupted() {
3252 // Add a fast path to t.isInterrupted(clear_int):
3253 // (t == Thread.current() &&
3254 // (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
3255 // ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
3256 // So, in the common case that the interrupt bit is false,
3257 // we avoid making a call into the VM. Even if the interrupt bit
3258 // is true, if the clear_int argument is false, we avoid the VM call.
3259 // However, if the receiver is not currentThread, we must call the VM,
3260 // because there must be some locking done around the operation.
3261
3262 // We only go to the fast case code if we pass two guards.
3263 // Paths which do not pass are accumulated in the slow_region.
3264
3265 enum {
3266 no_int_result_path = 1, // t == Thread.current() && !TLS._osthread._interrupted
3267 no_clear_result_path = 2, // t == Thread.current() && TLS._osthread._interrupted && !clear_int
3268 slow_result_path = 3, // slow path: t.isInterrupted(clear_int)
3269 PATH_LIMIT
3270 };
3271
3272 // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
3273 // out of the function.
3274 insert_mem_bar(Op_MemBarCPUOrder);
3275
3276 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3277 PhiNode* result_val = new PhiNode(result_rgn, TypeInt::BOOL);
3278
3279 RegionNode* slow_region = new RegionNode(1);
3280 record_for_igvn(slow_region);
3281
3282 // (a) Receiving thread must be the current thread.
3283 Node* rec_thr = argument(0);
3284 Node* tls_ptr = NULL;
3285 Node* cur_thr = generate_current_thread(tls_ptr);
3286
3287 // Resolve oops to stable for CmpP below.
3288 cur_thr = access_resolve(cur_thr, 0);
3289 rec_thr = access_resolve(rec_thr, 0);
3290
3291 Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
3292 Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
3293
3294 generate_slow_guard(bol_thr, slow_region);
3295
3296 // (b) Interrupt bit on TLS must be false.
3297 Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
3298 Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3299 p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
3300
3301 // Set the control input on the field _interrupted read to prevent it floating up.
3302 Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
3303 Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
3304 Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
3305
3306 IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
3307
3308 // First fast path: if (!TLS._interrupted) return false;
3309 Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
3310 result_rgn->init_req(no_int_result_path, false_bit);
3311 result_val->init_req(no_int_result_path, intcon(0));
3312
3313 // drop through to next case
3314 set_control( _gvn.transform(new IfTrueNode(iff_bit)));
3315
3316 #ifndef _WINDOWS
3317 // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
3318 Node* clr_arg = argument(1);
3319 Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
3320 Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
3321 IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
3322
3323 // Second fast path: ... else if (!clear_int) return true;
3324 Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
3325 result_rgn->init_req(no_clear_result_path, false_arg);
3326 result_val->init_req(no_clear_result_path, intcon(1));
3327
3328 // drop through to next case
3329 set_control( _gvn.transform(new IfTrueNode(iff_arg)));
3330 #else
3331 // To return true on Windows you must read the _interrupted field
3332 // and check the event state i.e. take the slow path.
3333 #endif // _WINDOWS
3334
3335 // (d) Otherwise, go to the slow path.
3336 slow_region->add_req(control());
3337 set_control( _gvn.transform(slow_region));
3338
3339 if (stopped()) {
3340 // There is no slow path.
3341 result_rgn->init_req(slow_result_path, top());
3342 result_val->init_req(slow_result_path, top());
3343 } else {
3344 // non-virtual because it is a private non-static
3345 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
3346
3347 Node* slow_val = set_results_for_java_call(slow_call);
3348 // this->control() comes from set_results_for_java_call
3349
3350 Node* fast_io = slow_call->in(TypeFunc::I_O);
3351 Node* fast_mem = slow_call->in(TypeFunc::Memory);
3352
3353 // These two phis are pre-filled with copies of of the fast IO and Memory
3354 PhiNode* result_mem = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
3355 PhiNode* result_io = PhiNode::make(result_rgn, fast_io, Type::ABIO);
3356
3357 result_rgn->init_req(slow_result_path, control());
3358 result_io ->init_req(slow_result_path, i_o());
3359 result_mem->init_req(slow_result_path, reset_memory());
3360 result_val->init_req(slow_result_path, slow_val);
3361
3362 set_all_memory(_gvn.transform(result_mem));
3363 set_i_o( _gvn.transform(result_io));
3364 }
3365
3366 C->set_has_split_ifs(true); // Has chance for split-if optimization
3367 set_result(result_rgn, result_val);
3368 return true;
3369 }
3370
3371 //-----------------------load_klass_from_mirror_common-------------------------
3372 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3373 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3374 // and branch to the given path on the region.
3375 // If never_see_null, take an uncommon trap on null, so we can optimistically
3376 // compile for the non-null case.
3377 // If the region is NULL, force never_see_null = true.
3378 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3379 bool never_see_null,
3380 RegionNode* region,
3381 int null_path,
3382 int offset) {
3383 if (region == NULL) never_see_null = true;
3384 Node* p = basic_plus_adr(mirror, offset);
3385 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3386 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3387 Node* null_ctl = top();
3388 kls = null_check_oop(kls, &null_ctl, never_see_null);
3389 if (region != NULL) {
3390 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3391 region->init_req(null_path, null_ctl);
3392 } else {
3393 assert(null_ctl == top(), "no loose ends");
3394 }
3395 return kls;
3396 }
3397
3398 //--------------------(inline_native_Class_query helpers)---------------------
3399 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3400 // Fall through if (mods & mask) == bits, take the guard otherwise.
3401 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3402 // Branch around if the given klass has the given modifier bit set.
3403 // Like generate_guard, adds a new path onto the region.
3404 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3405 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3406 Node* mask = intcon(modifier_mask);
3407 Node* bits = intcon(modifier_bits);
3408 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3409 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3410 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3411 return generate_fair_guard(bol, region);
3412 }
3413 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3414 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3415 }
3416
3417 //-------------------------inline_native_Class_query-------------------
3418 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3419 const Type* return_type = TypeInt::BOOL;
3420 Node* prim_return_value = top(); // what happens if it's a primitive class?
3421 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3422 bool expect_prim = false; // most of these guys expect to work on refs
3423
3424 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3425
3426 Node* mirror = argument(0);
3427 Node* obj = top();
3428
3429 switch (id) {
3430 case vmIntrinsics::_isInstance:
3431 // nothing is an instance of a primitive type
3432 prim_return_value = intcon(0);
3433 obj = argument(1);
3434 break;
3435 case vmIntrinsics::_getModifiers:
3436 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3437 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3438 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3439 break;
3440 case vmIntrinsics::_isInterface:
3441 prim_return_value = intcon(0);
3442 break;
3443 case vmIntrinsics::_isArray:
3444 prim_return_value = intcon(0);
3445 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3446 break;
3447 case vmIntrinsics::_isPrimitive:
3448 prim_return_value = intcon(1);
3449 expect_prim = true; // obviously
3450 break;
3451 case vmIntrinsics::_getSuperclass:
3452 prim_return_value = null();
3453 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3454 break;
3455 case vmIntrinsics::_getClassAccessFlags:
3456 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3457 return_type = TypeInt::INT; // not bool! 6297094
3458 break;
3459 default:
3460 fatal_unexpected_iid(id);
3461 break;
3462 }
3463
3464 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3465 if (mirror_con == NULL) return false; // cannot happen?
3466
3467 #ifndef PRODUCT
3468 if (C->print_intrinsics() || C->print_inlining()) {
3469 ciType* k = mirror_con->java_mirror_type();
3470 if (k) {
3471 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3472 k->print_name();
3473 tty->cr();
3474 }
3475 }
3476 #endif
3477
3478 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3479 RegionNode* region = new RegionNode(PATH_LIMIT);
3480 record_for_igvn(region);
3481 PhiNode* phi = new PhiNode(region, return_type);
3482
3483 // The mirror will never be null of Reflection.getClassAccessFlags, however
3484 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3485 // if it is. See bug 4774291.
3486
3487 // For Reflection.getClassAccessFlags(), the null check occurs in
3488 // the wrong place; see inline_unsafe_access(), above, for a similar
3489 // situation.
3490 mirror = null_check(mirror);
3491 // If mirror or obj is dead, only null-path is taken.
3492 if (stopped()) return true;
3493
3494 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3495
3496 // Now load the mirror's klass metaobject, and null-check it.
3497 // Side-effects region with the control path if the klass is null.
3498 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3499 // If kls is null, we have a primitive mirror.
3500 phi->init_req(_prim_path, prim_return_value);
3501 if (stopped()) { set_result(region, phi); return true; }
3502 bool safe_for_replace = (region->in(_prim_path) == top());
3503
3504 Node* p; // handy temp
3505 Node* null_ctl;
3506
3507 // Now that we have the non-null klass, we can perform the real query.
3508 // For constant classes, the query will constant-fold in LoadNode::Value.
3509 Node* query_value = top();
3510 switch (id) {
3511 case vmIntrinsics::_isInstance:
3512 // nothing is an instance of a primitive type
3513 query_value = gen_instanceof(obj, kls, safe_for_replace);
3514 break;
3515
3516 case vmIntrinsics::_getModifiers:
3517 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3518 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3519 break;
3520
3521 case vmIntrinsics::_isInterface:
3522 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3523 if (generate_interface_guard(kls, region) != NULL)
3524 // A guard was added. If the guard is taken, it was an interface.
3525 phi->add_req(intcon(1));
3526 // If we fall through, it's a plain class.
3527 query_value = intcon(0);
3528 break;
3529
3530 case vmIntrinsics::_isArray:
3531 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3532 if (generate_array_guard(kls, region) != NULL)
3533 // A guard was added. If the guard is taken, it was an array.
3534 phi->add_req(intcon(1));
3535 // If we fall through, it's a plain class.
3536 query_value = intcon(0);
3537 break;
3538
3539 case vmIntrinsics::_isPrimitive:
3540 query_value = intcon(0); // "normal" path produces false
3541 break;
3542
3543 case vmIntrinsics::_getSuperclass:
3544 // The rules here are somewhat unfortunate, but we can still do better
3545 // with random logic than with a JNI call.
3546 // Interfaces store null or Object as _super, but must report null.
3547 // Arrays store an intermediate super as _super, but must report Object.
3548 // Other types can report the actual _super.
3549 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3550 if (generate_interface_guard(kls, region) != NULL)
3551 // A guard was added. If the guard is taken, it was an interface.
3552 phi->add_req(null());
3553 if (generate_array_guard(kls, region) != NULL)
3554 // A guard was added. If the guard is taken, it was an array.
3555 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3556 // If we fall through, it's a plain class. Get its _super.
3557 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3558 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3559 null_ctl = top();
3560 kls = null_check_oop(kls, &null_ctl);
3561 if (null_ctl != top()) {
3562 // If the guard is taken, Object.superClass is null (both klass and mirror).
3563 region->add_req(null_ctl);
3564 phi ->add_req(null());
3565 }
3566 if (!stopped()) {
3567 query_value = load_mirror_from_klass(kls);
3568 }
3569 break;
3570
3571 case vmIntrinsics::_getClassAccessFlags:
3572 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3573 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3574 break;
3575
3576 default:
3577 fatal_unexpected_iid(id);
3578 break;
3579 }
3580
3581 // Fall-through is the normal case of a query to a real class.
3582 phi->init_req(1, query_value);
3583 region->init_req(1, control());
3584
3585 C->set_has_split_ifs(true); // Has chance for split-if optimization
3586 set_result(region, phi);
3587 return true;
3588 }
3589
3590 //-------------------------inline_value_Class_conversion-------------------
3591 // public Class<T> java.lang.Class.asPrimaryType();
3592 // public Class<T> java.lang.Class.asIndirectType()
3593 bool LibraryCallKit::inline_value_Class_conversion(vmIntrinsics::ID id) {
3594 Node* mirror = argument(0); // Receiver Class
3595 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3596 if (mirror_con == NULL) {
3597 return false;
3598 }
3599
3600 bool is_indirect_type = true;
3601 ciType* tm = mirror_con->java_mirror_type(&is_indirect_type);
3602 if (tm != NULL) {
3603 Node* result = mirror;
3604 if (tm->is_valuetype()) {
3605 if (id == vmIntrinsics::_asPrimaryType && is_indirect_type) {
3606 result = _gvn.makecon(TypeInstPtr::make(tm->as_value_klass()->inline_mirror_instance()));
3607 } else if (id == vmIntrinsics::_asIndirectType && !is_indirect_type) {
3608 result = _gvn.makecon(TypeInstPtr::make(tm->as_value_klass()->indirect_mirror_instance()));
3609 }
3610 }
3611 set_result(result);
3612 return true;
3613 }
3614 return false;
3615 }
3616
3617 //-------------------------inline_Class_cast-------------------
3618 bool LibraryCallKit::inline_Class_cast() {
3619 Node* mirror = argument(0); // Class
3620 Node* obj = argument(1);
3621 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3622 if (mirror_con == NULL) {
3623 return false; // dead path (mirror->is_top()).
3624 }
3625 if (obj == NULL || obj->is_top()) {
3626 return false; // dead path
3627 }
3628
3629 ciKlass* obj_klass = NULL;
3630 if (obj->is_ValueType()) {
3631 obj_klass = _gvn.type(obj)->value_klass();
3632 } else {
3633 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3634 if (tp != NULL) {
3635 obj_klass = tp->klass();
3636 }
3637 }
3638
3639 // First, see if Class.cast() can be folded statically.
3640 // java_mirror_type() returns non-null for compile-time Class constants.
3641 bool is_indirect_type = true;
3642 ciType* tm = mirror_con->java_mirror_type(&is_indirect_type);
3643 if (!obj->is_ValueType() && !is_indirect_type) {
3644 obj = null_check(obj);
3645 if (stopped()) {
3646 return true;
3647 }
3648 }
3649 if (tm != NULL && tm->is_klass() && obj_klass != NULL) {
3650 if (!obj_klass->is_loaded()) {
3651 // Don't use intrinsic when class is not loaded.
3652 return false;
3653 } else {
3654 int static_res = C->static_subtype_check(tm->as_klass(), obj_klass);
3655 if (static_res == Compile::SSC_always_true) {
3656 // isInstance() is true - fold the code.
3657 set_result(obj);
3658 return true;
3659 } else if (static_res == Compile::SSC_always_false) {
3660 // Don't use intrinsic, have to throw ClassCastException.
3661 // If the reference is null, the non-intrinsic bytecode will
3662 // be optimized appropriately.
3663 return false;
3664 }
3665 }
3666 }
3667
3668 // Bailout intrinsic and do normal inlining if exception path is frequent.
3669 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3670 return false;
3671 }
3672
3673 // Generate dynamic checks.
3674 // Class.cast() is java implementation of _checkcast bytecode.
3675 // Do checkcast (Parse::do_checkcast()) optimizations here.
3676
3677 mirror = null_check(mirror);
3678 // If mirror is dead, only null-path is taken.
3679 if (stopped()) {
3680 return true;
3681 }
3682
3683 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3684 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
3685 RegionNode* region = new RegionNode(PATH_LIMIT);
3686 record_for_igvn(region);
3687
3688 // Now load the mirror's klass metaobject, and null-check it.
3689 // If kls is null, we have a primitive mirror and
3690 // nothing is an instance of a primitive type.
3691 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3692
3693 Node* res = top();
3694 if (!stopped()) {
3695 if (EnableValhalla && !obj->is_ValueType() && is_indirect_type) {
3696 // Check if (mirror == inline_mirror && obj == null)
3697 Node* is_val_mirror = generate_fair_guard(is_value_mirror(mirror), NULL);
3698 if (is_val_mirror != NULL) {
3699 RegionNode* r = new RegionNode(3);
3700 record_for_igvn(r);
3701 r->init_req(1, control());
3702
3703 // Casting to .val, check for null
3704 set_control(is_val_mirror);
3705 Node *null_ctr = top();
3706 null_check_oop(obj, &null_ctr);
3707 region->init_req(_npe_path, null_ctr);
3708 r->init_req(2, control());
3709
3710 set_control(_gvn.transform(r));
3711 }
3712 }
3713
3714 Node* bad_type_ctrl = top();
3715 // Do checkcast optimizations.
3716 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3717 region->init_req(_bad_type_path, bad_type_ctrl);
3718 }
3719 if (region->in(_prim_path) != top() ||
3720 region->in(_bad_type_path) != top() ||
3721 region->in(_npe_path) != top()) {
3722 // Let Interpreter throw ClassCastException.
3723 PreserveJVMState pjvms(this);
3724 set_control(_gvn.transform(region));
3725 uncommon_trap(Deoptimization::Reason_intrinsic,
3726 Deoptimization::Action_maybe_recompile);
3727 }
3728 if (!stopped()) {
3729 set_result(res);
3730 }
3731 return true;
3732 }
3733
3734
3735 //--------------------------inline_native_subtype_check------------------------
3736 // This intrinsic takes the JNI calls out of the heart of
3737 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3738 bool LibraryCallKit::inline_native_subtype_check() {
3739 // Pull both arguments off the stack.
3740 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3741 args[0] = argument(0);
3742 args[1] = argument(1);
3743 Node* klasses[2]; // corresponding Klasses: superk, subk
3744 klasses[0] = klasses[1] = top();
3745
3746 enum {
3747 // A full decision tree on {superc is prim, subc is prim}:
3748 _prim_0_path = 1, // {P,N} => false
3749 // {P,P} & superc!=subc => false
3750 _prim_same_path, // {P,P} & superc==subc => true
3751 _prim_1_path, // {N,P} => false
3752 _ref_subtype_path, // {N,N} & subtype check wins => true
3753 _both_ref_path, // {N,N} & subtype check loses => false
3754 PATH_LIMIT
3755 };
3756
3757 RegionNode* region = new RegionNode(PATH_LIMIT);
3758 RegionNode* prim_region = new RegionNode(2);
3759 Node* phi = new PhiNode(region, TypeInt::BOOL);
3760 record_for_igvn(region);
3761 record_for_igvn(prim_region);
3762
3763 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3764 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3765 int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3766
3767 // First null-check both mirrors and load each mirror's klass metaobject.
3768 int which_arg;
3769 for (which_arg = 0; which_arg <= 1; which_arg++) {
3770 Node* arg = args[which_arg];
3771 arg = null_check(arg);
3772 if (stopped()) break;
3773 args[which_arg] = arg;
3774
3775 Node* p = basic_plus_adr(arg, class_klass_offset);
3776 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3777 klasses[which_arg] = _gvn.transform(kls);
3778 }
3779
3780 // Resolve oops to stable for CmpP below.
3781 args[0] = access_resolve(args[0], 0);
3782 args[1] = access_resolve(args[1], 0);
3783
3784 // Having loaded both klasses, test each for null.
3785 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3786 for (which_arg = 0; which_arg <= 1; which_arg++) {
3787 Node* kls = klasses[which_arg];
3788 Node* null_ctl = top();
3789 kls = null_check_oop(kls, &null_ctl, never_see_null);
3790 if (which_arg == 0) {
3791 prim_region->init_req(1, null_ctl);
3792 } else {
3793 region->init_req(_prim_1_path, null_ctl);
3794 }
3795 if (stopped()) break;
3796 klasses[which_arg] = kls;
3797 }
3798
3799 if (!stopped()) {
3800 // now we have two reference types, in klasses[0..1]
3801 Node* subk = klasses[1]; // the argument to isAssignableFrom
3802 Node* superk = klasses[0]; // the receiver
3803 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3804 // If superc is a value mirror, we also need to check if superc == subc because
3805 // V? is not a subtype of V but due to subk == superk the subtype check will pass.
3806 generate_fair_guard(is_value_mirror(args[0]), prim_region);
3807 // now we have a successful reference subtype check
3808 region->set_req(_ref_subtype_path, control());
3809 }
3810
3811 // If both operands are primitive (both klasses null), then
3812 // we must return true when they are identical primitives.
3813 // It is convenient to test this after the first null klass check.
3814 // This path is also used if superc is a value mirror.
3815 set_control(_gvn.transform(prim_region));
3816 if (!stopped()) {
3817 // Since superc is primitive, make a guard for the superc==subc case.
3818 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3819 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3820 generate_fair_guard(bol_eq, region);
3821 if (region->req() == PATH_LIMIT+1) {
3822 // A guard was added. If the added guard is taken, superc==subc.
3823 region->swap_edges(PATH_LIMIT, _prim_same_path);
3824 region->del_req(PATH_LIMIT);
3825 }
3826 region->set_req(_prim_0_path, control()); // Not equal after all.
3827 }
3828
3829 // these are the only paths that produce 'true':
3830 phi->set_req(_prim_same_path, intcon(1));
3831 phi->set_req(_ref_subtype_path, intcon(1));
3832
3833 // pull together the cases:
3834 assert(region->req() == PATH_LIMIT, "sane region");
3835 for (uint i = 1; i < region->req(); i++) {
3836 Node* ctl = region->in(i);
3837 if (ctl == NULL || ctl == top()) {
3838 region->set_req(i, top());
3839 phi ->set_req(i, top());
3840 } else if (phi->in(i) == NULL) {
3841 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3842 }
3843 }
3844
3845 set_control(_gvn.transform(region));
3846 set_result(_gvn.transform(phi));
3847 return true;
3848 }
3849
3850 //---------------------generate_array_guard_common------------------------
3851 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind) {
3852
3853 if (stopped()) {
3854 return NULL;
3855 }
3856
3857 // Like generate_guard, adds a new path onto the region.
3858 jint layout_con = 0;
3859 Node* layout_val = get_layout_helper(kls, layout_con);
3860 if (layout_val == NULL) {
3861 bool query = 0;
3862 switch(kind) {
3863 case ObjectArray: query = Klass::layout_helper_is_objArray(layout_con); break;
3864 case NonObjectArray: query = !Klass::layout_helper_is_objArray(layout_con); break;
3865 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
3866 case ValueArray: query = Klass::layout_helper_is_valueArray(layout_con); break;
3867 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
3868 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
3869 default:
3870 ShouldNotReachHere();
3871 }
3872 if (!query) {
3873 return NULL; // never a branch
3874 } else { // always a branch
3875 Node* always_branch = control();
3876 if (region != NULL)
3877 region->add_req(always_branch);
3878 set_control(top());
3879 return always_branch;
3880 }
3881 }
3882 unsigned int value = 0;
3883 BoolTest::mask btest = BoolTest::illegal;
3884 switch(kind) {
3885 case ObjectArray:
3886 case NonObjectArray: {
3887 value = Klass::_lh_array_tag_obj_value;
3888 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
3889 btest = kind == ObjectArray ? BoolTest::eq : BoolTest::ne;
3890 break;
3891 }
3892 case TypeArray: {
3893 value = Klass::_lh_array_tag_type_value;
3894 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
3895 btest = BoolTest::eq;
3896 break;
3897 }
3898 case ValueArray: {
3899 value = Klass::_lh_array_tag_vt_value;
3900 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
3901 btest = BoolTest::eq;
3902 break;
3903 }
3904 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
3905 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
3906 default:
3907 ShouldNotReachHere();
3908 }
3909 // Now test the correct condition.
3910 jint nval = (jint)value;
3911 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3912 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3913 return generate_fair_guard(bol, region);
3914 }
3915
3916
3917 //-----------------------inline_native_newArray--------------------------
3918 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
3919 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3920 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3921 Node* mirror;
3922 Node* count_val;
3923 if (uninitialized) {
3924 mirror = argument(1);
3925 count_val = argument(2);
3926 } else {
3927 mirror = argument(0);
3928 count_val = argument(1);
3929 }
3930
3931 mirror = null_check(mirror);
3932 // If mirror or obj is dead, only null-path is taken.
3933 if (stopped()) return true;
3934
3935 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3936 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3937 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3938 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3939 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3940
3941 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3942 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3943 result_reg, _slow_path);
3944 Node* normal_ctl = control();
3945 Node* no_array_ctl = result_reg->in(_slow_path);
3946
3947 // Generate code for the slow case. We make a call to newArray().
3948 set_control(no_array_ctl);
3949 if (!stopped()) {
3950 // Either the input type is void.class, or else the
3951 // array klass has not yet been cached. Either the
3952 // ensuing call will throw an exception, or else it
3953 // will cache the array klass for next time.
3954 PreserveJVMState pjvms(this);
3955 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3956 Node* slow_result = set_results_for_java_call(slow_call);
3957 // this->control() comes from set_results_for_java_call
3958 result_reg->set_req(_slow_path, control());
3959 result_val->set_req(_slow_path, slow_result);
3960 result_io ->set_req(_slow_path, i_o());
3961 result_mem->set_req(_slow_path, reset_memory());
3962 }
3963
3964 set_control(normal_ctl);
3965 if (!stopped()) {
3966 // Normal case: The array type has been cached in the java.lang.Class.
3967 // The following call works fine even if the array type is polymorphic.
3968 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3969 Node* obj = new_array(klass_node, count_val, 0, NULL, false, mirror); // no arguments to push
3970 result_reg->init_req(_normal_path, control());
3971 result_val->init_req(_normal_path, obj);
3972 result_io ->init_req(_normal_path, i_o());
3973 result_mem->init_req(_normal_path, reset_memory());
3974
3975 if (uninitialized) {
3976 // Mark the allocation so that zeroing is skipped
3977 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3978 alloc->maybe_set_complete(&_gvn);
3979 }
3980 }
3981
3982 // Return the combined state.
3983 set_i_o( _gvn.transform(result_io) );
3984 set_all_memory( _gvn.transform(result_mem));
3985
3986 C->set_has_split_ifs(true); // Has chance for split-if optimization
3987 set_result(result_reg, result_val);
3988 return true;
3989 }
3990
3991 //----------------------inline_native_getLength--------------------------
3992 // public static native int java.lang.reflect.Array.getLength(Object array);
3993 bool LibraryCallKit::inline_native_getLength() {
3994 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3995
3996 Node* array = null_check(argument(0));
3997 // If array is dead, only null-path is taken.
3998 if (stopped()) return true;
3999
4000 // Deoptimize if it is a non-array.
4001 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
4002
4003 if (non_array != NULL) {
4004 PreserveJVMState pjvms(this);
4005 set_control(non_array);
4006 uncommon_trap(Deoptimization::Reason_intrinsic,
4007 Deoptimization::Action_maybe_recompile);
4008 }
4009
4010 // If control is dead, only non-array-path is taken.
4011 if (stopped()) return true;
4012
4013 // The works fine even if the array type is polymorphic.
4014 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4015 Node* result = load_array_length(array);
4016
4017 C->set_has_split_ifs(true); // Has chance for split-if optimization
4018 set_result(result);
4019 return true;
4020 }
4021
4022 //------------------------inline_array_copyOf----------------------------
4023 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4024 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4025 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4026 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4027
4028 // Get the arguments.
4029 Node* original = argument(0);
4030 Node* start = is_copyOfRange? argument(1): intcon(0);
4031 Node* end = is_copyOfRange? argument(2): argument(1);
4032 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4033
4034 const TypeAryPtr* original_t = _gvn.type(original)->isa_aryptr();
4035 const TypeInstPtr* mirror_t = _gvn.type(array_type_mirror)->isa_instptr();
4036 if (EnableValhalla && ValueArrayFlatten &&
4037 (original_t == NULL || mirror_t == NULL ||
4038 (mirror_t->java_mirror_type() == NULL &&
4039 (original_t->elem()->isa_valuetype() ||
4040 (original_t->elem()->make_oopptr() != NULL &&
4041 original_t->elem()->make_oopptr()->can_be_value_type()))))) {
4042 // We need to know statically if the copy is to a flattened array
4043 // or not but can't tell.
4044 return false;
4045 }
4046
4047 Node* newcopy = NULL;
4048
4049 // Set the original stack and the reexecute bit for the interpreter to reexecute
4050 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4051 { PreserveReexecuteState preexecs(this);
4052 jvms()->set_should_reexecute(true);
4053
4054 array_type_mirror = null_check(array_type_mirror);
4055 original = null_check(original);
4056
4057 // Check if a null path was taken unconditionally.
4058 if (stopped()) return true;
4059
4060 Node* orig_length = load_array_length(original);
4061
4062 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
4063 klass_node = null_check(klass_node);
4064
4065 RegionNode* bailout = new RegionNode(1);
4066 record_for_igvn(bailout);
4067
4068 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4069 // Bail out if that is so.
4070 // Value type array may have object field that would require a
4071 // write barrier. Conservatively, go to slow path.
4072 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4073 Node* not_objArray = !bs->array_copy_requires_gc_barriers(false, T_OBJECT, false, BarrierSetC2::Parsing) ?
4074 generate_typeArray_guard(klass_node, bailout) : generate_non_objArray_guard(klass_node, bailout);
4075 if (not_objArray != NULL) {
4076 // Improve the klass node's type from the new optimistic assumption:
4077 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4078 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0), false);
4079 Node* cast = new CastPPNode(klass_node, akls);
4080 cast->init_req(0, control());
4081 klass_node = _gvn.transform(cast);
4082 }
4083
4084 Node* original_kls = load_object_klass(original);
4085 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4086 // loads/stores but it is legal only if we're sure the
4087 // Arrays.copyOf would succeed. So we need all input arguments
4088 // to the copyOf to be validated, including that the copy to the
4089 // new array won't trigger an ArrayStoreException. That subtype
4090 // check can be optimized if we know something on the type of
4091 // the input array from type speculation.
4092 if (_gvn.type(klass_node)->singleton() && !stopped()) {
4093 ciKlass* subk = _gvn.type(original_kls)->is_klassptr()->klass();
4094 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
4095
4096 int test = C->static_subtype_check(superk, subk);
4097 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4098 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4099 if (t_original->speculative_type() != NULL) {
4100 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4101 original_kls = load_object_klass(original);
4102 }
4103 }
4104 }
4105
4106 if (ValueArrayFlatten) {
4107 // Either both or neither new array klass and original array
4108 // klass must be flattened
4109 Node* is_flat = generate_valueArray_guard(klass_node, NULL);
4110 if (!original_t->is_not_flat()) {
4111 generate_valueArray_guard(original_kls, bailout);
4112 }
4113 if (is_flat != NULL) {
4114 RegionNode* r = new RegionNode(2);
4115 record_for_igvn(r);
4116 r->init_req(1, control());
4117 set_control(is_flat);
4118 if (!original_t->is_not_flat()) {
4119 generate_valueArray_guard(original_kls, r);
4120 }
4121 bailout->add_req(control());
4122 set_control(_gvn.transform(r));
4123 }
4124 }
4125
4126 // Bail out if either start or end is negative.
4127 generate_negative_guard(start, bailout, &start);
4128 generate_negative_guard(end, bailout, &end);
4129
4130 Node* length = end;
4131 if (_gvn.type(start) != TypeInt::ZERO) {
4132 length = _gvn.transform(new SubINode(end, start));
4133 }
4134
4135 // Bail out if length is negative.
4136 // Without this the new_array would throw
4137 // NegativeArraySizeException but IllegalArgumentException is what
4138 // should be thrown
4139 generate_negative_guard(length, bailout, &length);
4140
4141 if (bailout->req() > 1) {
4142 PreserveJVMState pjvms(this);
4143 set_control(_gvn.transform(bailout));
4144 uncommon_trap(Deoptimization::Reason_intrinsic,
4145 Deoptimization::Action_maybe_recompile);
4146 }
4147
4148 if (!stopped()) {
4149 // How many elements will we copy from the original?
4150 // The answer is MinI(orig_length - start, length).
4151 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4152 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4153
4154 original = access_resolve(original, ACCESS_READ);
4155
4156 // Generate a direct call to the right arraycopy function(s).
4157 // We know the copy is disjoint but we might not know if the
4158 // oop stores need checking.
4159 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4160 // This will fail a store-check if x contains any non-nulls.
4161
4162 bool validated = false;
4163 // Reason_class_check rather than Reason_intrinsic because we
4164 // want to intrinsify even if this traps.
4165 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4166 Node* not_subtype_ctrl = gen_subtype_check(original_kls,
4167 klass_node);
4168
4169 if (not_subtype_ctrl != top()) {
4170 PreserveJVMState pjvms(this);
4171 set_control(not_subtype_ctrl);
4172 uncommon_trap(Deoptimization::Reason_class_check,
4173 Deoptimization::Action_make_not_entrant);
4174 assert(stopped(), "Should be stopped");
4175 }
4176 validated = true;
4177 }
4178
4179 if (!stopped()) {
4180 // Load element mirror
4181 Node* p = basic_plus_adr(array_type_mirror, java_lang_Class::component_mirror_offset_in_bytes());
4182 Node* elem_mirror = access_load_at(array_type_mirror, p, _gvn.type(p)->is_ptr(), TypeInstPtr::MIRROR, T_OBJECT, IN_HEAP);
4183
4184 newcopy = new_array(klass_node, length, 0, NULL, false, elem_mirror);
4185
4186 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
4187 original_kls, klass_node);
4188 if (!is_copyOfRange) {
4189 ac->set_copyof(validated);
4190 } else {
4191 ac->set_copyofrange(validated);
4192 }
4193 Node* n = _gvn.transform(ac);
4194 if (n == ac) {
4195 ac->connect_outputs(this);
4196 } else {
4197 assert(validated, "shouldn't transform if all arguments not validated");
4198 set_all_memory(n);
4199 }
4200 }
4201 }
4202 } // original reexecute is set back here
4203
4204 C->set_has_split_ifs(true); // Has chance for split-if optimization
4205 if (!stopped()) {
4206 set_result(newcopy);
4207 }
4208 return true;
4209 }
4210
4211
4212 //----------------------generate_virtual_guard---------------------------
4213 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4214 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4215 RegionNode* slow_region) {
4216 ciMethod* method = callee();
4217 int vtable_index = method->vtable_index();
4218 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4219 "bad index %d", vtable_index);
4220 // Get the Method* out of the appropriate vtable entry.
4221 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4222 vtable_index*vtableEntry::size_in_bytes() +
4223 vtableEntry::method_offset_in_bytes();
4224 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4225 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4226
4227 // Compare the target method with the expected method (e.g., Object.hashCode).
4228 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4229
4230 Node* native_call = makecon(native_call_addr);
4231 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4232 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4233
4234 return generate_slow_guard(test_native, slow_region);
4235 }
4236
4237 //-----------------------generate_method_call----------------------------
4238 // Use generate_method_call to make a slow-call to the real
4239 // method if the fast path fails. An alternative would be to
4240 // use a stub like OptoRuntime::slow_arraycopy_Java.
4241 // This only works for expanding the current library call,
4242 // not another intrinsic. (E.g., don't use this for making an
4243 // arraycopy call inside of the copyOf intrinsic.)
4244 CallJavaNode*
4245 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4246 // When compiling the intrinsic method itself, do not use this technique.
4247 guarantee(callee() != C->method(), "cannot make slow-call to self");
4248
4249 ciMethod* method = callee();
4250 // ensure the JVMS we have will be correct for this call
4251 guarantee(method_id == method->intrinsic_id(), "must match");
4252
4253 const TypeFunc* tf = TypeFunc::make(method);
4254 CallJavaNode* slow_call;
4255 if (is_static) {
4256 assert(!is_virtual, "");
4257 slow_call = new CallStaticJavaNode(C, tf,
4258 SharedRuntime::get_resolve_static_call_stub(),
4259 method, bci());
4260 } else if (is_virtual) {
4261 null_check_receiver();
4262 int vtable_index = Method::invalid_vtable_index;
4263 if (UseInlineCaches) {
4264 // Suppress the vtable call
4265 } else {
4266 // hashCode and clone are not a miranda methods,
4267 // so the vtable index is fixed.
4268 // No need to use the linkResolver to get it.
4269 vtable_index = method->vtable_index();
4270 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4271 "bad index %d", vtable_index);
4272 }
4273 slow_call = new CallDynamicJavaNode(tf,
4274 SharedRuntime::get_resolve_virtual_call_stub(),
4275 method, vtable_index, bci());
4276 } else { // neither virtual nor static: opt_virtual
4277 null_check_receiver();
4278 slow_call = new CallStaticJavaNode(C, tf,
4279 SharedRuntime::get_resolve_opt_virtual_call_stub(),
4280 method, bci());
4281 slow_call->set_optimized_virtual(true);
4282 }
4283 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4284 // To be able to issue a direct call (optimized virtual or virtual)
4285 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4286 // about the method being invoked should be attached to the call site to
4287 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4288 slow_call->set_override_symbolic_info(true);
4289 }
4290 set_arguments_for_java_call(slow_call);
4291 set_edges_for_java_call(slow_call);
4292 return slow_call;
4293 }
4294
4295
4296 /**
4297 * Build special case code for calls to hashCode on an object. This call may
4298 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4299 * slightly different code.
4300 */
4301 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4302 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4303 assert(!(is_virtual && is_static), "either virtual, special, or static");
4304
4305 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4306
4307 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4308 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4309 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4310 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4311 Node* obj = argument(0);
4312
4313 if (obj->is_ValueType() || gvn().type(obj)->is_valuetypeptr()) {
4314 return false;
4315 }
4316
4317 if (!is_static) {
4318 // Check for hashing null object
4319 obj = null_check_receiver();
4320 if (stopped()) return true; // unconditionally null
4321 result_reg->init_req(_null_path, top());
4322 result_val->init_req(_null_path, top());
4323 } else {
4324 // Do a null check, and return zero if null.
4325 // System.identityHashCode(null) == 0
4326 Node* null_ctl = top();
4327 obj = null_check_oop(obj, &null_ctl);
4328 result_reg->init_req(_null_path, null_ctl);
4329 result_val->init_req(_null_path, _gvn.intcon(0));
4330 }
4331
4332 // Unconditionally null? Then return right away.
4333 if (stopped()) {
4334 set_control( result_reg->in(_null_path));
4335 if (!stopped())
4336 set_result(result_val->in(_null_path));
4337 return true;
4338 }
4339
4340 // We only go to the fast case code if we pass a number of guards. The
4341 // paths which do not pass are accumulated in the slow_region.
4342 RegionNode* slow_region = new RegionNode(1);
4343 record_for_igvn(slow_region);
4344
4345 // If this is a virtual call, we generate a funny guard. We pull out
4346 // the vtable entry corresponding to hashCode() from the target object.
4347 // If the target method which we are calling happens to be the native
4348 // Object hashCode() method, we pass the guard. We do not need this
4349 // guard for non-virtual calls -- the caller is known to be the native
4350 // Object hashCode().
4351 if (is_virtual) {
4352 // After null check, get the object's klass.
4353 Node* obj_klass = load_object_klass(obj);
4354 generate_virtual_guard(obj_klass, slow_region);
4355 }
4356
4357 // Get the header out of the object, use LoadMarkNode when available
4358 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4359 // The control of the load must be NULL. Otherwise, the load can move before
4360 // the null check after castPP removal.
4361 Node* no_ctrl = NULL;
4362 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4363
4364 // Test the header to see if it is unlocked.
4365 // This also serves as guard against value types (they have the always_locked_pattern set).
4366 Node *lock_mask = _gvn.MakeConX(markWord::biased_lock_mask_in_place);
4367 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4368 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4369 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4370 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4371
4372 generate_slow_guard(test_unlocked, slow_region);
4373
4374 // Get the hash value and check to see that it has been properly assigned.
4375 // We depend on hash_mask being at most 32 bits and avoid the use of
4376 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4377 // vm: see markWord.hpp.
4378 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4379 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4380 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4381 // This hack lets the hash bits live anywhere in the mark object now, as long
4382 // as the shift drops the relevant bits into the low 32 bits. Note that
4383 // Java spec says that HashCode is an int so there's no point in capturing
4384 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4385 hshifted_header = ConvX2I(hshifted_header);
4386 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4387
4388 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4389 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4390 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4391
4392 generate_slow_guard(test_assigned, slow_region);
4393
4394 Node* init_mem = reset_memory();
4395 // fill in the rest of the null path:
4396 result_io ->init_req(_null_path, i_o());
4397 result_mem->init_req(_null_path, init_mem);
4398
4399 result_val->init_req(_fast_path, hash_val);
4400 result_reg->init_req(_fast_path, control());
4401 result_io ->init_req(_fast_path, i_o());
4402 result_mem->init_req(_fast_path, init_mem);
4403
4404 // Generate code for the slow case. We make a call to hashCode().
4405 set_control(_gvn.transform(slow_region));
4406 if (!stopped()) {
4407 // No need for PreserveJVMState, because we're using up the present state.
4408 set_all_memory(init_mem);
4409 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4410 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4411 Node* slow_result = set_results_for_java_call(slow_call);
4412 // this->control() comes from set_results_for_java_call
4413 result_reg->init_req(_slow_path, control());
4414 result_val->init_req(_slow_path, slow_result);
4415 result_io ->set_req(_slow_path, i_o());
4416 result_mem ->set_req(_slow_path, reset_memory());
4417 }
4418
4419 // Return the combined state.
4420 set_i_o( _gvn.transform(result_io) );
4421 set_all_memory( _gvn.transform(result_mem));
4422
4423 set_result(result_reg, result_val);
4424 return true;
4425 }
4426
4427 //---------------------------inline_native_getClass----------------------------
4428 // public final native Class<?> java.lang.Object.getClass();
4429 //
4430 // Build special case code for calls to getClass on an object.
4431 bool LibraryCallKit::inline_native_getClass() {
4432 Node* obj = argument(0);
4433 if (obj->is_ValueType()) {
4434 ciKlass* vk = _gvn.type(obj)->value_klass();
4435 set_result(makecon(TypeInstPtr::make(vk->java_mirror())));
4436 return true;
4437 }
4438 obj = null_check_receiver();
4439 if (stopped()) return true;
4440 set_result(load_mirror_from_klass(load_object_klass(obj)));
4441 return true;
4442 }
4443
4444 //-----------------inline_native_Reflection_getCallerClass---------------------
4445 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4446 //
4447 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4448 //
4449 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4450 // in that it must skip particular security frames and checks for
4451 // caller sensitive methods.
4452 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4453 #ifndef PRODUCT
4454 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4455 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4456 }
4457 #endif
4458
4459 if (!jvms()->has_method()) {
4460 #ifndef PRODUCT
4461 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4462 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4463 }
4464 #endif
4465 return false;
4466 }
4467
4468 // Walk back up the JVM state to find the caller at the required
4469 // depth.
4470 JVMState* caller_jvms = jvms();
4471
4472 // Cf. JVM_GetCallerClass
4473 // NOTE: Start the loop at depth 1 because the current JVM state does
4474 // not include the Reflection.getCallerClass() frame.
4475 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4476 ciMethod* m = caller_jvms->method();
4477 switch (n) {
4478 case 0:
4479 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4480 break;
4481 case 1:
4482 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4483 if (!m->caller_sensitive()) {
4484 #ifndef PRODUCT
4485 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4486 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4487 }
4488 #endif
4489 return false; // bail-out; let JVM_GetCallerClass do the work
4490 }
4491 break;
4492 default:
4493 if (!m->is_ignored_by_security_stack_walk()) {
4494 // We have reached the desired frame; return the holder class.
4495 // Acquire method holder as java.lang.Class and push as constant.
4496 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4497 ciInstance* caller_mirror = caller_klass->java_mirror();
4498 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4499
4500 #ifndef PRODUCT
4501 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4502 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());
4503 tty->print_cr(" JVM state at this point:");
4504 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4505 ciMethod* m = jvms()->of_depth(i)->method();
4506 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4507 }
4508 }
4509 #endif
4510 return true;
4511 }
4512 break;
4513 }
4514 }
4515
4516 #ifndef PRODUCT
4517 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4518 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4519 tty->print_cr(" JVM state at this point:");
4520 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4521 ciMethod* m = jvms()->of_depth(i)->method();
4522 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4523 }
4524 }
4525 #endif
4526
4527 return false; // bail-out; let JVM_GetCallerClass do the work
4528 }
4529
4530 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4531 Node* arg = argument(0);
4532 Node* result = NULL;
4533
4534 switch (id) {
4535 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4536 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4537 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4538 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4539
4540 case vmIntrinsics::_doubleToLongBits: {
4541 // two paths (plus control) merge in a wood
4542 RegionNode *r = new RegionNode(3);
4543 Node *phi = new PhiNode(r, TypeLong::LONG);
4544
4545 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4546 // Build the boolean node
4547 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4548
4549 // Branch either way.
4550 // NaN case is less traveled, which makes all the difference.
4551 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4552 Node *opt_isnan = _gvn.transform(ifisnan);
4553 assert( opt_isnan->is_If(), "Expect an IfNode");
4554 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4555 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4556
4557 set_control(iftrue);
4558
4559 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4560 Node *slow_result = longcon(nan_bits); // return NaN
4561 phi->init_req(1, _gvn.transform( slow_result ));
4562 r->init_req(1, iftrue);
4563
4564 // Else fall through
4565 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4566 set_control(iffalse);
4567
4568 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4569 r->init_req(2, iffalse);
4570
4571 // Post merge
4572 set_control(_gvn.transform(r));
4573 record_for_igvn(r);
4574
4575 C->set_has_split_ifs(true); // Has chance for split-if optimization
4576 result = phi;
4577 assert(result->bottom_type()->isa_long(), "must be");
4578 break;
4579 }
4580
4581 case vmIntrinsics::_floatToIntBits: {
4582 // two paths (plus control) merge in a wood
4583 RegionNode *r = new RegionNode(3);
4584 Node *phi = new PhiNode(r, TypeInt::INT);
4585
4586 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4587 // Build the boolean node
4588 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4589
4590 // Branch either way.
4591 // NaN case is less traveled, which makes all the difference.
4592 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4593 Node *opt_isnan = _gvn.transform(ifisnan);
4594 assert( opt_isnan->is_If(), "Expect an IfNode");
4595 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4596 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4597
4598 set_control(iftrue);
4599
4600 static const jint nan_bits = 0x7fc00000;
4601 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4602 phi->init_req(1, _gvn.transform( slow_result ));
4603 r->init_req(1, iftrue);
4604
4605 // Else fall through
4606 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4607 set_control(iffalse);
4608
4609 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4610 r->init_req(2, iffalse);
4611
4612 // Post merge
4613 set_control(_gvn.transform(r));
4614 record_for_igvn(r);
4615
4616 C->set_has_split_ifs(true); // Has chance for split-if optimization
4617 result = phi;
4618 assert(result->bottom_type()->isa_int(), "must be");
4619 break;
4620 }
4621
4622 default:
4623 fatal_unexpected_iid(id);
4624 break;
4625 }
4626 set_result(_gvn.transform(result));
4627 return true;
4628 }
4629
4630 //----------------------inline_unsafe_copyMemory-------------------------
4631 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4632 bool LibraryCallKit::inline_unsafe_copyMemory() {
4633 if (callee()->is_static()) return false; // caller must have the capability!
4634 null_check_receiver(); // null-check receiver
4635 if (stopped()) return true;
4636
4637 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4638
4639 Node* src_ptr = argument(1); // type: oop
4640 Node* src_off = ConvL2X(argument(2)); // type: long
4641 Node* dst_ptr = argument(4); // type: oop
4642 Node* dst_off = ConvL2X(argument(5)); // type: long
4643 Node* size = ConvL2X(argument(7)); // type: long
4644
4645 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4646 "fieldOffset must be byte-scaled");
4647
4648 src_ptr = access_resolve(src_ptr, ACCESS_READ);
4649 dst_ptr = access_resolve(dst_ptr, ACCESS_WRITE);
4650 Node* src = make_unsafe_address(src_ptr, src_off, ACCESS_READ);
4651 Node* dst = make_unsafe_address(dst_ptr, dst_off, ACCESS_WRITE);
4652
4653 // Conservatively insert a memory barrier on all memory slices.
4654 // Do not let writes of the copy source or destination float below the copy.
4655 insert_mem_bar(Op_MemBarCPUOrder);
4656
4657 Node* thread = _gvn.transform(new ThreadLocalNode());
4658 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4659 BasicType doing_unsafe_access_bt = T_BYTE;
4660 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4661
4662 // update volatile field
4663 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4664
4665 // Call it. Note that the length argument is not scaled.
4666 make_runtime_call(RC_LEAF|RC_NO_FP,
4667 OptoRuntime::fast_arraycopy_Type(),
4668 StubRoutines::unsafe_arraycopy(),
4669 "unsafe_arraycopy",
4670 TypeRawPtr::BOTTOM,
4671 src, dst, size XTOP);
4672
4673 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4674
4675 // Do not let reads of the copy destination float above the copy.
4676 insert_mem_bar(Op_MemBarCPUOrder);
4677
4678 return true;
4679 }
4680
4681 //------------------------clone_coping-----------------------------------
4682 // Helper function for inline_native_clone.
4683 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4684 assert(obj_size != NULL, "");
4685 Node* raw_obj = alloc_obj->in(1);
4686 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4687
4688 AllocateNode* alloc = NULL;
4689 if (ReduceBulkZeroing) {
4690 // We will be completely responsible for initializing this object -
4691 // mark Initialize node as complete.
4692 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4693 // The object was just allocated - there should be no any stores!
4694 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4695 // Mark as complete_with_arraycopy so that on AllocateNode
4696 // expansion, we know this AllocateNode is initialized by an array
4697 // copy and a StoreStore barrier exists after the array copy.
4698 alloc->initialization()->set_complete_with_arraycopy();
4699 }
4700
4701 // Copy the fastest available way.
4702 // TODO: generate fields copies for small objects instead.
4703 Node* size = _gvn.transform(obj_size);
4704
4705 // Exclude the header but include array length to copy by 8 bytes words.
4706 // Can't use base_offset_in_bytes(bt) since basic type is unknown.
4707 int base_off = is_array ? arrayOopDesc::length_offset_in_bytes() :
4708 instanceOopDesc::base_offset_in_bytes();
4709 // base_off:
4710 // 8 - 32-bit VM
4711 // 12 - 64-bit VM, compressed klass
4712 // 16 - 64-bit VM, normal klass
4713 if (base_off % BytesPerLong != 0) {
4714 assert(UseCompressedClassPointers, "");
4715 if (is_array) {
4716 // Exclude length to copy by 8 bytes words.
4717 base_off += sizeof(int);
4718 } else {
4719 // Include klass to copy by 8 bytes words.
4720 base_off = instanceOopDesc::klass_offset_in_bytes();
4721 }
4722 assert(base_off % BytesPerLong == 0, "expect 8 bytes alignment");
4723 }
4724 Node* src_base = basic_plus_adr(obj, base_off);
4725 Node* dst_base = basic_plus_adr(alloc_obj, base_off);
4726
4727 // Compute the length also, if needed:
4728 Node* countx = size;
4729 countx = _gvn.transform(new SubXNode(countx, MakeConX(base_off)));
4730 countx = _gvn.transform(new URShiftXNode(countx, intcon(LogBytesPerLong)));
4731
4732 access_clone(src_base, dst_base, countx, is_array);
4733
4734 // Do not let reads from the cloned object float above the arraycopy.
4735 if (alloc != NULL) {
4736 // Do not let stores that initialize this object be reordered with
4737 // a subsequent store that would make this object accessible by
4738 // other threads.
4739 // Record what AllocateNode this StoreStore protects so that
4740 // escape analysis can go from the MemBarStoreStoreNode to the
4741 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4742 // based on the escape status of the AllocateNode.
4743 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4744 } else {
4745 insert_mem_bar(Op_MemBarCPUOrder);
4746 }
4747 }
4748
4749 //------------------------inline_native_clone----------------------------
4750 // protected native Object java.lang.Object.clone();
4751 //
4752 // Here are the simple edge cases:
4753 // null receiver => normal trap
4754 // virtual and clone was overridden => slow path to out-of-line clone
4755 // not cloneable or finalizer => slow path to out-of-line Object.clone
4756 //
4757 // The general case has two steps, allocation and copying.
4758 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4759 //
4760 // Copying also has two cases, oop arrays and everything else.
4761 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4762 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4763 //
4764 // These steps fold up nicely if and when the cloned object's klass
4765 // can be sharply typed as an object array, a type array, or an instance.
4766 //
4767 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4768 PhiNode* result_val;
4769
4770 // Set the reexecute bit for the interpreter to reexecute
4771 // the bytecode that invokes Object.clone if deoptimization happens.
4772 { PreserveReexecuteState preexecs(this);
4773 jvms()->set_should_reexecute(true);
4774
4775 Node* obj = argument(0);
4776 if (obj->is_ValueType()) {
4777 return false;
4778 }
4779
4780 obj = null_check_receiver();
4781 if (stopped()) return true;
4782
4783 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4784
4785 // If we are going to clone an instance, we need its exact type to
4786 // know the number and types of fields to convert the clone to
4787 // loads/stores. Maybe a speculative type can help us.
4788 if (!obj_type->klass_is_exact() &&
4789 obj_type->speculative_type() != NULL &&
4790 obj_type->speculative_type()->is_instance_klass() &&
4791 !obj_type->speculative_type()->is_valuetype()) {
4792 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4793 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4794 !spec_ik->has_injected_fields()) {
4795 ciKlass* k = obj_type->klass();
4796 if (!k->is_instance_klass() ||
4797 k->as_instance_klass()->is_interface() ||
4798 k->as_instance_klass()->has_subklass()) {
4799 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4800 }
4801 }
4802 }
4803
4804 Node* obj_klass = load_object_klass(obj);
4805
4806 // Conservatively insert a memory barrier on all memory slices.
4807 // Do not let writes into the original float below the clone.
4808 insert_mem_bar(Op_MemBarCPUOrder);
4809
4810 // paths into result_reg:
4811 enum {
4812 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4813 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4814 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4815 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4816 PATH_LIMIT
4817 };
4818 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4819 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4820 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4821 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4822 record_for_igvn(result_reg);
4823
4824 // We only go to the fast case code if we pass a number of guards.
4825 // The paths which do not pass are accumulated in the slow_region.
4826 RegionNode* slow_region = new RegionNode(1);
4827 record_for_igvn(slow_region);
4828
4829 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4830 if (array_ctl != NULL) {
4831 // It's an array.
4832 PreserveJVMState pjvms(this);
4833 set_control(array_ctl);
4834
4835 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4836 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, BarrierSetC2::Parsing) &&
4837 (!obj_type->isa_aryptr() || !obj_type->is_aryptr()->is_not_flat())) {
4838 // Flattened value type array may have object field that would require a
4839 // write barrier. Conservatively, go to slow path.
4840 generate_valueArray_guard(obj_klass, slow_region);
4841 }
4842
4843 if (!stopped()) {
4844 Node* obj_length = load_array_length(obj);
4845 Node* obj_size = NULL;
4846 // Load element mirror
4847 Node* array_type_mirror = load_mirror_from_klass(obj_klass);
4848 Node* p = basic_plus_adr(array_type_mirror, java_lang_Class::component_mirror_offset_in_bytes());
4849 Node* elem_mirror = access_load_at(array_type_mirror, p, _gvn.type(p)->is_ptr(), TypeInstPtr::MIRROR, T_OBJECT, IN_HEAP);
4850
4851 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, false, elem_mirror);
4852
4853 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4854 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, BarrierSetC2::Parsing)) {
4855 // If it is an oop array, it requires very special treatment,
4856 // because gc barriers are required when accessing the array.
4857 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4858 if (is_obja != NULL) {
4859 PreserveJVMState pjvms2(this);
4860 set_control(is_obja);
4861 // Generate a direct call to the right arraycopy function(s).
4862 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4863 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false);
4864 ac->set_cloneoop();
4865 Node* n = _gvn.transform(ac);
4866 assert(n == ac, "cannot disappear");
4867 ac->connect_outputs(this);
4868
4869 result_reg->init_req(_objArray_path, control());
4870 result_val->init_req(_objArray_path, alloc_obj);
4871 result_i_o ->set_req(_objArray_path, i_o());
4872 result_mem ->set_req(_objArray_path, reset_memory());
4873 }
4874 }
4875
4876 // Otherwise, there are no barriers to worry about.
4877 // (We can dispense with card marks if we know the allocation
4878 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4879 // causes the non-eden paths to take compensating steps to
4880 // simulate a fresh allocation, so that no further
4881 // card marks are required in compiled code to initialize
4882 // the object.)
4883
4884 if (!stopped()) {
4885 copy_to_clone(obj, alloc_obj, obj_size, true);
4886
4887 // Present the results of the copy.
4888 result_reg->init_req(_array_path, control());
4889 result_val->init_req(_array_path, alloc_obj);
4890 result_i_o ->set_req(_array_path, i_o());
4891 result_mem ->set_req(_array_path, reset_memory());
4892 }
4893 }
4894 }
4895
4896 if (!stopped()) {
4897 // It's an instance (we did array above). Make the slow-path tests.
4898 // If this is a virtual call, we generate a funny guard. We grab
4899 // the vtable entry corresponding to clone() from the target object.
4900 // If the target method which we are calling happens to be the
4901 // Object clone() method, we pass the guard. We do not need this
4902 // guard for non-virtual calls; the caller is known to be the native
4903 // Object clone().
4904 if (is_virtual) {
4905 generate_virtual_guard(obj_klass, slow_region);
4906 }
4907
4908 // The object must be easily cloneable and must not have a finalizer.
4909 // Both of these conditions may be checked in a single test.
4910 // We could optimize the test further, but we don't care.
4911 generate_access_flags_guard(obj_klass,
4912 // Test both conditions:
4913 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4914 // Must be cloneable but not finalizer:
4915 JVM_ACC_IS_CLONEABLE_FAST,
4916 slow_region);
4917 }
4918
4919 if (!stopped()) {
4920 // It's an instance, and it passed the slow-path tests.
4921 PreserveJVMState pjvms(this);
4922 Node* obj_size = NULL;
4923 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4924 // is reexecuted if deoptimization occurs and there could be problems when merging
4925 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4926 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4927
4928 copy_to_clone(obj, alloc_obj, obj_size, false);
4929
4930 // Present the results of the slow call.
4931 result_reg->init_req(_instance_path, control());
4932 result_val->init_req(_instance_path, alloc_obj);
4933 result_i_o ->set_req(_instance_path, i_o());
4934 result_mem ->set_req(_instance_path, reset_memory());
4935 }
4936
4937 // Generate code for the slow case. We make a call to clone().
4938 set_control(_gvn.transform(slow_region));
4939 if (!stopped()) {
4940 PreserveJVMState pjvms(this);
4941 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4942 // We need to deoptimize on exception (see comment above)
4943 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4944 // this->control() comes from set_results_for_java_call
4945 result_reg->init_req(_slow_path, control());
4946 result_val->init_req(_slow_path, slow_result);
4947 result_i_o ->set_req(_slow_path, i_o());
4948 result_mem ->set_req(_slow_path, reset_memory());
4949 }
4950
4951 // Return the combined state.
4952 set_control( _gvn.transform(result_reg));
4953 set_i_o( _gvn.transform(result_i_o));
4954 set_all_memory( _gvn.transform(result_mem));
4955 } // original reexecute is set back here
4956
4957 set_result(_gvn.transform(result_val));
4958 return true;
4959 }
4960
4961 // If we have a tightly coupled allocation, the arraycopy may take care
4962 // of the array initialization. If one of the guards we insert between
4963 // the allocation and the arraycopy causes a deoptimization, an
4964 // unitialized array will escape the compiled method. To prevent that
4965 // we set the JVM state for uncommon traps between the allocation and
4966 // the arraycopy to the state before the allocation so, in case of
4967 // deoptimization, we'll reexecute the allocation and the
4968 // initialization.
4969 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4970 if (alloc != NULL) {
4971 ciMethod* trap_method = alloc->jvms()->method();
4972 int trap_bci = alloc->jvms()->bci();
4973
4974 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4975 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4976 // Make sure there's no store between the allocation and the
4977 // arraycopy otherwise visible side effects could be rexecuted
4978 // in case of deoptimization and cause incorrect execution.
4979 bool no_interfering_store = true;
4980 Node* mem = alloc->in(TypeFunc::Memory);
4981 if (mem->is_MergeMem()) {
4982 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4983 Node* n = mms.memory();
4984 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4985 assert(n->is_Store(), "what else?");
4986 no_interfering_store = false;
4987 break;
4988 }
4989 }
4990 } else {
4991 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4992 Node* n = mms.memory();
4993 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4994 assert(n->is_Store(), "what else?");
4995 no_interfering_store = false;
4996 break;
4997 }
4998 }
4999 }
5000
5001 if (no_interfering_store) {
5002 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5003 uint size = alloc->req();
5004 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5005 old_jvms->set_map(sfpt);
5006 for (uint i = 0; i < size; i++) {
5007 sfpt->init_req(i, alloc->in(i));
5008 }
5009 // re-push array length for deoptimization
5010 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5011 old_jvms->set_sp(old_jvms->sp()+1);
5012 old_jvms->set_monoff(old_jvms->monoff()+1);
5013 old_jvms->set_scloff(old_jvms->scloff()+1);
5014 old_jvms->set_endoff(old_jvms->endoff()+1);
5015 old_jvms->set_should_reexecute(true);
5016
5017 sfpt->set_i_o(map()->i_o());
5018 sfpt->set_memory(map()->memory());
5019 sfpt->set_control(map()->control());
5020
5021 JVMState* saved_jvms = jvms();
5022 saved_reexecute_sp = _reexecute_sp;
5023
5024 set_jvms(sfpt->jvms());
5025 _reexecute_sp = jvms()->sp();
5026
5027 return saved_jvms;
5028 }
5029 }
5030 }
5031 return NULL;
5032 }
5033
5034 // In case of a deoptimization, we restart execution at the
5035 // allocation, allocating a new array. We would leave an uninitialized
5036 // array in the heap that GCs wouldn't expect. Move the allocation
5037 // after the traps so we don't allocate the array if we
5038 // deoptimize. This is possible because tightly_coupled_allocation()
5039 // guarantees there's no observer of the allocated array at this point
5040 // and the control flow is simple enough.
5041 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
5042 int saved_reexecute_sp, uint new_idx) {
5043 if (saved_jvms != NULL && !stopped()) {
5044 assert(alloc != NULL, "only with a tightly coupled allocation");
5045 // restore JVM state to the state at the arraycopy
5046 saved_jvms->map()->set_control(map()->control());
5047 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
5048 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
5049 // If we've improved the types of some nodes (null check) while
5050 // emitting the guards, propagate them to the current state
5051 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
5052 set_jvms(saved_jvms);
5053 _reexecute_sp = saved_reexecute_sp;
5054
5055 // Remove the allocation from above the guards
5056 CallProjections* callprojs = alloc->extract_projections(true);
5057 InitializeNode* init = alloc->initialization();
5058 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5059 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5060 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5061 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5062
5063 // move the allocation here (after the guards)
5064 _gvn.hash_delete(alloc);
5065 alloc->set_req(TypeFunc::Control, control());
5066 alloc->set_req(TypeFunc::I_O, i_o());
5067 Node *mem = reset_memory();
5068 set_all_memory(mem);
5069 alloc->set_req(TypeFunc::Memory, mem);
5070 set_control(init->proj_out_or_null(TypeFunc::Control));
5071 set_i_o(callprojs->fallthrough_ioproj);
5072
5073 // Update memory as done in GraphKit::set_output_for_allocation()
5074 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5075 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5076 if (ary_type->isa_aryptr() && length_type != NULL) {
5077 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5078 }
5079 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5080 int elemidx = C->get_alias_index(telemref);
5081 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5082 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5083
5084 Node* allocx = _gvn.transform(alloc);
5085 assert(allocx == alloc, "where has the allocation gone?");
5086 assert(dest->is_CheckCastPP(), "not an allocation result?");
5087
5088 _gvn.hash_delete(dest);
5089 dest->set_req(0, control());
5090 Node* destx = _gvn.transform(dest);
5091 assert(destx == dest, "where has the allocation result gone?");
5092 }
5093 }
5094
5095
5096 //------------------------------inline_arraycopy-----------------------
5097 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5098 // Object dest, int destPos,
5099 // int length);
5100 bool LibraryCallKit::inline_arraycopy() {
5101 // Get the arguments.
5102 Node* src = argument(0); // type: oop
5103 Node* src_offset = argument(1); // type: int
5104 Node* dest = argument(2); // type: oop
5105 Node* dest_offset = argument(3); // type: int
5106 Node* length = argument(4); // type: int
5107
5108 uint new_idx = C->unique();
5109
5110 // Check for allocation before we add nodes that would confuse
5111 // tightly_coupled_allocation()
5112 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
5113
5114 int saved_reexecute_sp = -1;
5115 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5116 // See arraycopy_restore_alloc_state() comment
5117 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5118 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
5119 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
5120 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
5121
5122 // The following tests must be performed
5123 // (1) src and dest are arrays.
5124 // (2) src and dest arrays must have elements of the same BasicType
5125 // (3) src and dest must not be null.
5126 // (4) src_offset must not be negative.
5127 // (5) dest_offset must not be negative.
5128 // (6) length must not be negative.
5129 // (7) src_offset + length must not exceed length of src.
5130 // (8) dest_offset + length must not exceed length of dest.
5131 // (9) each element of an oop array must be assignable
5132
5133 // (3) src and dest must not be null.
5134 // always do this here because we need the JVM state for uncommon traps
5135 Node* null_ctl = top();
5136 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5137 assert(null_ctl->is_top(), "no null control here");
5138 dest = null_check(dest, T_ARRAY);
5139
5140 if (!can_emit_guards) {
5141 // if saved_jvms == NULL and alloc != NULL, we don't emit any
5142 // guards but the arraycopy node could still take advantage of a
5143 // tightly allocated allocation. tightly_coupled_allocation() is
5144 // called again to make sure it takes the null check above into
5145 // account: the null check is mandatory and if it caused an
5146 // uncommon trap to be emitted then the allocation can't be
5147 // considered tightly coupled in this context.
5148 alloc = tightly_coupled_allocation(dest, NULL);
5149 }
5150
5151 bool validated = false;
5152
5153 const Type* src_type = _gvn.type(src);
5154 const Type* dest_type = _gvn.type(dest);
5155 const TypeAryPtr* top_src = src_type->isa_aryptr();
5156 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5157
5158 // Do we have the type of src?
5159 bool has_src = (top_src != NULL && top_src->klass() != NULL);
5160 // Do we have the type of dest?
5161 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
5162 // Is the type for src from speculation?
5163 bool src_spec = false;
5164 // Is the type for dest from speculation?
5165 bool dest_spec = false;
5166
5167 if ((!has_src || !has_dest) && can_emit_guards) {
5168 // We don't have sufficient type information, let's see if
5169 // speculative types can help. We need to have types for both src
5170 // and dest so that it pays off.
5171
5172 // Do we already have or could we have type information for src
5173 bool could_have_src = has_src;
5174 // Do we already have or could we have type information for dest
5175 bool could_have_dest = has_dest;
5176
5177 ciKlass* src_k = NULL;
5178 if (!has_src) {
5179 src_k = src_type->speculative_type_not_null();
5180 if (src_k != NULL && src_k->is_array_klass()) {
5181 could_have_src = true;
5182 }
5183 }
5184
5185 ciKlass* dest_k = NULL;
5186 if (!has_dest) {
5187 dest_k = dest_type->speculative_type_not_null();
5188 if (dest_k != NULL && dest_k->is_array_klass()) {
5189 could_have_dest = true;
5190 }
5191 }
5192
5193 if (could_have_src && could_have_dest) {
5194 // This is going to pay off so emit the required guards
5195 if (!has_src) {
5196 src = maybe_cast_profiled_obj(src, src_k, true);
5197 src_type = _gvn.type(src);
5198 top_src = src_type->isa_aryptr();
5199 has_src = (top_src != NULL && top_src->klass() != NULL);
5200 src_spec = true;
5201 }
5202 if (!has_dest) {
5203 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5204 dest_type = _gvn.type(dest);
5205 top_dest = dest_type->isa_aryptr();
5206 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
5207 dest_spec = true;
5208 }
5209 }
5210 }
5211
5212 if (has_src && has_dest && can_emit_guards) {
5213 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
5214 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
5215 if (src_elem == T_ARRAY) src_elem = T_OBJECT;
5216 if (dest_elem == T_ARRAY) dest_elem = T_OBJECT;
5217
5218 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5219 // If both arrays are object arrays then having the exact types
5220 // for both will remove the need for a subtype check at runtime
5221 // before the call and may make it possible to pick a faster copy
5222 // routine (without a subtype check on every element)
5223 // Do we have the exact type of src?
5224 bool could_have_src = src_spec;
5225 // Do we have the exact type of dest?
5226 bool could_have_dest = dest_spec;
5227 ciKlass* src_k = top_src->klass();
5228 ciKlass* dest_k = top_dest->klass();
5229 if (!src_spec) {
5230 src_k = src_type->speculative_type_not_null();
5231 if (src_k != NULL && src_k->is_array_klass()) {
5232 could_have_src = true;
5233 }
5234 }
5235 if (!dest_spec) {
5236 dest_k = dest_type->speculative_type_not_null();
5237 if (dest_k != NULL && dest_k->is_array_klass()) {
5238 could_have_dest = true;
5239 }
5240 }
5241 if (could_have_src && could_have_dest) {
5242 // If we can have both exact types, emit the missing guards
5243 if (could_have_src && !src_spec) {
5244 src = maybe_cast_profiled_obj(src, src_k, true);
5245 }
5246 if (could_have_dest && !dest_spec) {
5247 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5248 }
5249 }
5250 }
5251 }
5252
5253 ciMethod* trap_method = method();
5254 int trap_bci = bci();
5255 if (saved_jvms != NULL) {
5256 trap_method = alloc->jvms()->method();
5257 trap_bci = alloc->jvms()->bci();
5258 }
5259
5260 bool negative_length_guard_generated = false;
5261
5262 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5263 can_emit_guards &&
5264 !src->is_top() && !dest->is_top()) {
5265 // validate arguments: enables transformation the ArrayCopyNode
5266 validated = true;
5267
5268 RegionNode* slow_region = new RegionNode(1);
5269 record_for_igvn(slow_region);
5270
5271 // (1) src and dest are arrays.
5272 generate_non_array_guard(load_object_klass(src), slow_region);
5273 generate_non_array_guard(load_object_klass(dest), slow_region);
5274
5275 // (2) src and dest arrays must have elements of the same BasicType
5276 // done at macro expansion or at Ideal transformation time
5277
5278 // (4) src_offset must not be negative.
5279 generate_negative_guard(src_offset, slow_region);
5280
5281 // (5) dest_offset must not be negative.
5282 generate_negative_guard(dest_offset, slow_region);
5283
5284 // (7) src_offset + length must not exceed length of src.
5285 generate_limit_guard(src_offset, length,
5286 load_array_length(src),
5287 slow_region);
5288
5289 // (8) dest_offset + length must not exceed length of dest.
5290 generate_limit_guard(dest_offset, length,
5291 load_array_length(dest),
5292 slow_region);
5293
5294 // (6) length must not be negative.
5295 // This is also checked in generate_arraycopy() during macro expansion, but
5296 // we also have to check it here for the case where the ArrayCopyNode will
5297 // be eliminated by Escape Analysis.
5298 if (EliminateAllocations) {
5299 generate_negative_guard(length, slow_region);
5300 negative_length_guard_generated = true;
5301 }
5302
5303 // (9) each element of an oop array must be assignable
5304 Node* src_klass = load_object_klass(src);
5305 Node* dest_klass = load_object_klass(dest);
5306 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
5307
5308 if (not_subtype_ctrl != top()) {
5309 PreserveJVMState pjvms(this);
5310 set_control(not_subtype_ctrl);
5311 uncommon_trap(Deoptimization::Reason_intrinsic,
5312 Deoptimization::Action_make_not_entrant);
5313 assert(stopped(), "Should be stopped");
5314 }
5315
5316 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5317 const Type* toop = TypeOopPtr::make_from_klass(dest_klass_t->klass());
5318 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5319 src_type = _gvn.type(src);
5320 top_src = src_type->isa_aryptr();
5321
5322 if (top_dest != NULL && !top_dest->elem()->isa_valuetype() && !top_dest->is_not_flat()) {
5323 generate_valueArray_guard(dest_klass, slow_region);
5324 }
5325
5326 if (top_src != NULL && !top_src->elem()->isa_valuetype() && !top_src->is_not_flat()) {
5327 generate_valueArray_guard(src_klass, slow_region);
5328 }
5329
5330 {
5331 PreserveJVMState pjvms(this);
5332 set_control(_gvn.transform(slow_region));
5333 uncommon_trap(Deoptimization::Reason_intrinsic,
5334 Deoptimization::Action_make_not_entrant);
5335 assert(stopped(), "Should be stopped");
5336 }
5337 }
5338
5339 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
5340
5341 if (stopped()) {
5342 return true;
5343 }
5344
5345 Node* new_src = access_resolve(src, ACCESS_READ);
5346 Node* new_dest = access_resolve(dest, ACCESS_WRITE);
5347
5348 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, new_src, src_offset, new_dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
5349 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5350 // so the compiler has a chance to eliminate them: during macro expansion,
5351 // we have to set their control (CastPP nodes are eliminated).
5352 load_object_klass(src), load_object_klass(dest),
5353 load_array_length(src), load_array_length(dest));
5354
5355 ac->set_arraycopy(validated);
5356
5357 Node* n = _gvn.transform(ac);
5358 if (n == ac) {
5359 ac->connect_outputs(this);
5360 } else {
5361 assert(validated, "shouldn't transform if all arguments not validated");
5362 set_all_memory(n);
5363 }
5364 clear_upper_avx();
5365
5366
5367 return true;
5368 }
5369
5370
5371 // Helper function which determines if an arraycopy immediately follows
5372 // an allocation, with no intervening tests or other escapes for the object.
5373 AllocateArrayNode*
5374 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
5375 RegionNode* slow_region) {
5376 if (stopped()) return NULL; // no fast path
5377 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
5378
5379 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5380 if (alloc == NULL) return NULL;
5381
5382 Node* rawmem = memory(Compile::AliasIdxRaw);
5383 // Is the allocation's memory state untouched?
5384 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5385 // Bail out if there have been raw-memory effects since the allocation.
5386 // (Example: There might have been a call or safepoint.)
5387 return NULL;
5388 }
5389 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5390 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5391 return NULL;
5392 }
5393
5394 // There must be no unexpected observers of this allocation.
5395 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5396 Node* obs = ptr->fast_out(i);
5397 if (obs != this->map()) {
5398 return NULL;
5399 }
5400 }
5401
5402 // This arraycopy must unconditionally follow the allocation of the ptr.
5403 Node* alloc_ctl = ptr->in(0);
5404 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
5405
5406 Node* ctl = control();
5407 while (ctl != alloc_ctl) {
5408 // There may be guards which feed into the slow_region.
5409 // Any other control flow means that we might not get a chance
5410 // to finish initializing the allocated object.
5411 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5412 IfNode* iff = ctl->in(0)->as_If();
5413 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
5414 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5415 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
5416 ctl = iff->in(0); // This test feeds the known slow_region.
5417 continue;
5418 }
5419 // One more try: Various low-level checks bottom out in
5420 // uncommon traps. If the debug-info of the trap omits
5421 // any reference to the allocation, as we've already
5422 // observed, then there can be no objection to the trap.
5423 bool found_trap = false;
5424 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5425 Node* obs = not_ctl->fast_out(j);
5426 if (obs->in(0) == not_ctl && obs->is_Call() &&
5427 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5428 found_trap = true; break;
5429 }
5430 }
5431 if (found_trap) {
5432 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
5433 continue;
5434 }
5435 }
5436 return NULL;
5437 }
5438
5439 // If we get this far, we have an allocation which immediately
5440 // precedes the arraycopy, and we can take over zeroing the new object.
5441 // The arraycopy will finish the initialization, and provide
5442 // a new control state to which we will anchor the destination pointer.
5443
5444 return alloc;
5445 }
5446
5447 //-------------inline_encodeISOArray-----------------------------------
5448 // encode char[] to byte[] in ISO_8859_1
5449 bool LibraryCallKit::inline_encodeISOArray() {
5450 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5451 // no receiver since it is static method
5452 Node *src = argument(0);
5453 Node *src_offset = argument(1);
5454 Node *dst = argument(2);
5455 Node *dst_offset = argument(3);
5456 Node *length = argument(4);
5457
5458 src = must_be_not_null(src, true);
5459 dst = must_be_not_null(dst, true);
5460
5461 src = access_resolve(src, ACCESS_READ);
5462 dst = access_resolve(dst, ACCESS_WRITE);
5463
5464 const Type* src_type = src->Value(&_gvn);
5465 const Type* dst_type = dst->Value(&_gvn);
5466 const TypeAryPtr* top_src = src_type->isa_aryptr();
5467 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5468 if (top_src == NULL || top_src->klass() == NULL ||
5469 top_dest == NULL || top_dest->klass() == NULL) {
5470 // failed array check
5471 return false;
5472 }
5473
5474 // Figure out the size and type of the elements we will be copying.
5475 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5476 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5477 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5478 return false;
5479 }
5480
5481 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5482 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5483 // 'src_start' points to src array + scaled offset
5484 // 'dst_start' points to dst array + scaled offset
5485
5486 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5487 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
5488 enc = _gvn.transform(enc);
5489 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5490 set_memory(res_mem, mtype);
5491 set_result(enc);
5492 clear_upper_avx();
5493
5494 return true;
5495 }
5496
5497 //-------------inline_multiplyToLen-----------------------------------
5498 bool LibraryCallKit::inline_multiplyToLen() {
5499 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5500
5501 address stubAddr = StubRoutines::multiplyToLen();
5502 if (stubAddr == NULL) {
5503 return false; // Intrinsic's stub is not implemented on this platform
5504 }
5505 const char* stubName = "multiplyToLen";
5506
5507 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5508
5509 // no receiver because it is a static method
5510 Node* x = argument(0);
5511 Node* xlen = argument(1);
5512 Node* y = argument(2);
5513 Node* ylen = argument(3);
5514 Node* z = argument(4);
5515
5516 x = must_be_not_null(x, true);
5517 y = must_be_not_null(y, true);
5518
5519 x = access_resolve(x, ACCESS_READ);
5520 y = access_resolve(y, ACCESS_READ);
5521 z = access_resolve(z, ACCESS_WRITE);
5522
5523 const Type* x_type = x->Value(&_gvn);
5524 const Type* y_type = y->Value(&_gvn);
5525 const TypeAryPtr* top_x = x_type->isa_aryptr();
5526 const TypeAryPtr* top_y = y_type->isa_aryptr();
5527 if (top_x == NULL || top_x->klass() == NULL ||
5528 top_y == NULL || top_y->klass() == NULL) {
5529 // failed array check
5530 return false;
5531 }
5532
5533 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5534 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5535 if (x_elem != T_INT || y_elem != T_INT) {
5536 return false;
5537 }
5538
5539 // Set the original stack and the reexecute bit for the interpreter to reexecute
5540 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5541 // on the return from z array allocation in runtime.
5542 { PreserveReexecuteState preexecs(this);
5543 jvms()->set_should_reexecute(true);
5544
5545 Node* x_start = array_element_address(x, intcon(0), x_elem);
5546 Node* y_start = array_element_address(y, intcon(0), y_elem);
5547 // 'x_start' points to x array + scaled xlen
5548 // 'y_start' points to y array + scaled ylen
5549
5550 // Allocate the result array
5551 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5552 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5553 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5554
5555 IdealKit ideal(this);
5556
5557 #define __ ideal.
5558 Node* one = __ ConI(1);
5559 Node* zero = __ ConI(0);
5560 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5561 __ set(need_alloc, zero);
5562 __ set(z_alloc, z);
5563 __ if_then(z, BoolTest::eq, null()); {
5564 __ increment (need_alloc, one);
5565 } __ else_(); {
5566 // Update graphKit memory and control from IdealKit.
5567 sync_kit(ideal);
5568 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5569 cast->init_req(0, control());
5570 _gvn.set_type(cast, cast->bottom_type());
5571 C->record_for_igvn(cast);
5572
5573 Node* zlen_arg = load_array_length(cast);
5574 // Update IdealKit memory and control from graphKit.
5575 __ sync_kit(this);
5576 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5577 __ increment (need_alloc, one);
5578 } __ end_if();
5579 } __ end_if();
5580
5581 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5582 // Update graphKit memory and control from IdealKit.
5583 sync_kit(ideal);
5584 Node * narr = new_array(klass_node, zlen, 1);
5585 // Update IdealKit memory and control from graphKit.
5586 __ sync_kit(this);
5587 __ set(z_alloc, narr);
5588 } __ end_if();
5589
5590 sync_kit(ideal);
5591 z = __ value(z_alloc);
5592 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5593 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5594 // Final sync IdealKit and GraphKit.
5595 final_sync(ideal);
5596 #undef __
5597
5598 Node* z_start = array_element_address(z, intcon(0), T_INT);
5599
5600 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5601 OptoRuntime::multiplyToLen_Type(),
5602 stubAddr, stubName, TypePtr::BOTTOM,
5603 x_start, xlen, y_start, ylen, z_start, zlen);
5604 } // original reexecute is set back here
5605
5606 C->set_has_split_ifs(true); // Has chance for split-if optimization
5607 set_result(z);
5608 return true;
5609 }
5610
5611 //-------------inline_squareToLen------------------------------------
5612 bool LibraryCallKit::inline_squareToLen() {
5613 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5614
5615 address stubAddr = StubRoutines::squareToLen();
5616 if (stubAddr == NULL) {
5617 return false; // Intrinsic's stub is not implemented on this platform
5618 }
5619 const char* stubName = "squareToLen";
5620
5621 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5622
5623 Node* x = argument(0);
5624 Node* len = argument(1);
5625 Node* z = argument(2);
5626 Node* zlen = argument(3);
5627
5628 x = must_be_not_null(x, true);
5629 z = must_be_not_null(z, true);
5630
5631 x = access_resolve(x, ACCESS_READ);
5632 z = access_resolve(z, ACCESS_WRITE);
5633
5634 const Type* x_type = x->Value(&_gvn);
5635 const Type* z_type = z->Value(&_gvn);
5636 const TypeAryPtr* top_x = x_type->isa_aryptr();
5637 const TypeAryPtr* top_z = z_type->isa_aryptr();
5638 if (top_x == NULL || top_x->klass() == NULL ||
5639 top_z == NULL || top_z->klass() == NULL) {
5640 // failed array check
5641 return false;
5642 }
5643
5644 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5645 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5646 if (x_elem != T_INT || z_elem != T_INT) {
5647 return false;
5648 }
5649
5650
5651 Node* x_start = array_element_address(x, intcon(0), x_elem);
5652 Node* z_start = array_element_address(z, intcon(0), z_elem);
5653
5654 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5655 OptoRuntime::squareToLen_Type(),
5656 stubAddr, stubName, TypePtr::BOTTOM,
5657 x_start, len, z_start, zlen);
5658
5659 set_result(z);
5660 return true;
5661 }
5662
5663 //-------------inline_mulAdd------------------------------------------
5664 bool LibraryCallKit::inline_mulAdd() {
5665 assert(UseMulAddIntrinsic, "not implemented on this platform");
5666
5667 address stubAddr = StubRoutines::mulAdd();
5668 if (stubAddr == NULL) {
5669 return false; // Intrinsic's stub is not implemented on this platform
5670 }
5671 const char* stubName = "mulAdd";
5672
5673 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5674
5675 Node* out = argument(0);
5676 Node* in = argument(1);
5677 Node* offset = argument(2);
5678 Node* len = argument(3);
5679 Node* k = argument(4);
5680
5681 out = must_be_not_null(out, true);
5682
5683 in = access_resolve(in, ACCESS_READ);
5684 out = access_resolve(out, ACCESS_WRITE);
5685
5686 const Type* out_type = out->Value(&_gvn);
5687 const Type* in_type = in->Value(&_gvn);
5688 const TypeAryPtr* top_out = out_type->isa_aryptr();
5689 const TypeAryPtr* top_in = in_type->isa_aryptr();
5690 if (top_out == NULL || top_out->klass() == NULL ||
5691 top_in == NULL || top_in->klass() == NULL) {
5692 // failed array check
5693 return false;
5694 }
5695
5696 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5697 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5698 if (out_elem != T_INT || in_elem != T_INT) {
5699 return false;
5700 }
5701
5702 Node* outlen = load_array_length(out);
5703 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5704 Node* out_start = array_element_address(out, intcon(0), out_elem);
5705 Node* in_start = array_element_address(in, intcon(0), in_elem);
5706
5707 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5708 OptoRuntime::mulAdd_Type(),
5709 stubAddr, stubName, TypePtr::BOTTOM,
5710 out_start,in_start, new_offset, len, k);
5711 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5712 set_result(result);
5713 return true;
5714 }
5715
5716 //-------------inline_montgomeryMultiply-----------------------------------
5717 bool LibraryCallKit::inline_montgomeryMultiply() {
5718 address stubAddr = StubRoutines::montgomeryMultiply();
5719 if (stubAddr == NULL) {
5720 return false; // Intrinsic's stub is not implemented on this platform
5721 }
5722
5723 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5724 const char* stubName = "montgomery_multiply";
5725
5726 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5727
5728 Node* a = argument(0);
5729 Node* b = argument(1);
5730 Node* n = argument(2);
5731 Node* len = argument(3);
5732 Node* inv = argument(4);
5733 Node* m = argument(6);
5734
5735 a = access_resolve(a, ACCESS_READ);
5736 b = access_resolve(b, ACCESS_READ);
5737 n = access_resolve(n, ACCESS_READ);
5738 m = access_resolve(m, ACCESS_WRITE);
5739
5740 const Type* a_type = a->Value(&_gvn);
5741 const TypeAryPtr* top_a = a_type->isa_aryptr();
5742 const Type* b_type = b->Value(&_gvn);
5743 const TypeAryPtr* top_b = b_type->isa_aryptr();
5744 const Type* n_type = a->Value(&_gvn);
5745 const TypeAryPtr* top_n = n_type->isa_aryptr();
5746 const Type* m_type = a->Value(&_gvn);
5747 const TypeAryPtr* top_m = m_type->isa_aryptr();
5748 if (top_a == NULL || top_a->klass() == NULL ||
5749 top_b == NULL || top_b->klass() == NULL ||
5750 top_n == NULL || top_n->klass() == NULL ||
5751 top_m == NULL || top_m->klass() == NULL) {
5752 // failed array check
5753 return false;
5754 }
5755
5756 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5757 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5758 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5759 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5760 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5761 return false;
5762 }
5763
5764 // Make the call
5765 {
5766 Node* a_start = array_element_address(a, intcon(0), a_elem);
5767 Node* b_start = array_element_address(b, intcon(0), b_elem);
5768 Node* n_start = array_element_address(n, intcon(0), n_elem);
5769 Node* m_start = array_element_address(m, intcon(0), m_elem);
5770
5771 Node* call = make_runtime_call(RC_LEAF,
5772 OptoRuntime::montgomeryMultiply_Type(),
5773 stubAddr, stubName, TypePtr::BOTTOM,
5774 a_start, b_start, n_start, len, inv, top(),
5775 m_start);
5776 set_result(m);
5777 }
5778
5779 return true;
5780 }
5781
5782 bool LibraryCallKit::inline_montgomerySquare() {
5783 address stubAddr = StubRoutines::montgomerySquare();
5784 if (stubAddr == NULL) {
5785 return false; // Intrinsic's stub is not implemented on this platform
5786 }
5787
5788 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5789 const char* stubName = "montgomery_square";
5790
5791 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5792
5793 Node* a = argument(0);
5794 Node* n = argument(1);
5795 Node* len = argument(2);
5796 Node* inv = argument(3);
5797 Node* m = argument(5);
5798
5799 a = access_resolve(a, ACCESS_READ);
5800 n = access_resolve(n, ACCESS_READ);
5801 m = access_resolve(m, ACCESS_WRITE);
5802
5803 const Type* a_type = a->Value(&_gvn);
5804 const TypeAryPtr* top_a = a_type->isa_aryptr();
5805 const Type* n_type = a->Value(&_gvn);
5806 const TypeAryPtr* top_n = n_type->isa_aryptr();
5807 const Type* m_type = a->Value(&_gvn);
5808 const TypeAryPtr* top_m = m_type->isa_aryptr();
5809 if (top_a == NULL || top_a->klass() == NULL ||
5810 top_n == NULL || top_n->klass() == NULL ||
5811 top_m == NULL || top_m->klass() == NULL) {
5812 // failed array check
5813 return false;
5814 }
5815
5816 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5817 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5818 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5819 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5820 return false;
5821 }
5822
5823 // Make the call
5824 {
5825 Node* a_start = array_element_address(a, intcon(0), a_elem);
5826 Node* n_start = array_element_address(n, intcon(0), n_elem);
5827 Node* m_start = array_element_address(m, intcon(0), m_elem);
5828
5829 Node* call = make_runtime_call(RC_LEAF,
5830 OptoRuntime::montgomerySquare_Type(),
5831 stubAddr, stubName, TypePtr::BOTTOM,
5832 a_start, n_start, len, inv, top(),
5833 m_start);
5834 set_result(m);
5835 }
5836
5837 return true;
5838 }
5839
5840 //-------------inline_vectorizedMismatch------------------------------
5841 bool LibraryCallKit::inline_vectorizedMismatch() {
5842 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5843
5844 address stubAddr = StubRoutines::vectorizedMismatch();
5845 if (stubAddr == NULL) {
5846 return false; // Intrinsic's stub is not implemented on this platform
5847 }
5848 const char* stubName = "vectorizedMismatch";
5849 int size_l = callee()->signature()->size();
5850 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5851
5852 Node* obja = argument(0);
5853 Node* aoffset = argument(1);
5854 Node* objb = argument(3);
5855 Node* boffset = argument(4);
5856 Node* length = argument(6);
5857 Node* scale = argument(7);
5858
5859 const Type* a_type = obja->Value(&_gvn);
5860 const Type* b_type = objb->Value(&_gvn);
5861 const TypeAryPtr* top_a = a_type->isa_aryptr();
5862 const TypeAryPtr* top_b = b_type->isa_aryptr();
5863 if (top_a == NULL || top_a->klass() == NULL ||
5864 top_b == NULL || top_b->klass() == NULL) {
5865 // failed array check
5866 return false;
5867 }
5868
5869 Node* call;
5870 jvms()->set_should_reexecute(true);
5871
5872 obja = access_resolve(obja, ACCESS_READ);
5873 objb = access_resolve(objb, ACCESS_READ);
5874 Node* obja_adr = make_unsafe_address(obja, aoffset, ACCESS_READ);
5875 Node* objb_adr = make_unsafe_address(objb, boffset, ACCESS_READ);
5876
5877 call = make_runtime_call(RC_LEAF,
5878 OptoRuntime::vectorizedMismatch_Type(),
5879 stubAddr, stubName, TypePtr::BOTTOM,
5880 obja_adr, objb_adr, length, scale);
5881
5882 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5883 set_result(result);
5884 return true;
5885 }
5886
5887 /**
5888 * Calculate CRC32 for byte.
5889 * int java.util.zip.CRC32.update(int crc, int b)
5890 */
5891 bool LibraryCallKit::inline_updateCRC32() {
5892 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5893 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5894 // no receiver since it is static method
5895 Node* crc = argument(0); // type: int
5896 Node* b = argument(1); // type: int
5897
5898 /*
5899 * int c = ~ crc;
5900 * b = timesXtoThe32[(b ^ c) & 0xFF];
5901 * b = b ^ (c >>> 8);
5902 * crc = ~b;
5903 */
5904
5905 Node* M1 = intcon(-1);
5906 crc = _gvn.transform(new XorINode(crc, M1));
5907 Node* result = _gvn.transform(new XorINode(crc, b));
5908 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5909
5910 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5911 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5912 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5913 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5914
5915 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5916 result = _gvn.transform(new XorINode(crc, result));
5917 result = _gvn.transform(new XorINode(result, M1));
5918 set_result(result);
5919 return true;
5920 }
5921
5922 /**
5923 * Calculate CRC32 for byte[] array.
5924 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5925 */
5926 bool LibraryCallKit::inline_updateBytesCRC32() {
5927 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5928 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5929 // no receiver since it is static method
5930 Node* crc = argument(0); // type: int
5931 Node* src = argument(1); // type: oop
5932 Node* offset = argument(2); // type: int
5933 Node* length = argument(3); // type: int
5934
5935 const Type* src_type = src->Value(&_gvn);
5936 const TypeAryPtr* top_src = src_type->isa_aryptr();
5937 if (top_src == NULL || top_src->klass() == NULL) {
5938 // failed array check
5939 return false;
5940 }
5941
5942 // Figure out the size and type of the elements we will be copying.
5943 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5944 if (src_elem != T_BYTE) {
5945 return false;
5946 }
5947
5948 // 'src_start' points to src array + scaled offset
5949 src = must_be_not_null(src, true);
5950 src = access_resolve(src, ACCESS_READ);
5951 Node* src_start = array_element_address(src, offset, src_elem);
5952
5953 // We assume that range check is done by caller.
5954 // TODO: generate range check (offset+length < src.length) in debug VM.
5955
5956 // Call the stub.
5957 address stubAddr = StubRoutines::updateBytesCRC32();
5958 const char *stubName = "updateBytesCRC32";
5959
5960 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5961 stubAddr, stubName, TypePtr::BOTTOM,
5962 crc, src_start, length);
5963 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5964 set_result(result);
5965 return true;
5966 }
5967
5968 /**
5969 * Calculate CRC32 for ByteBuffer.
5970 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5971 */
5972 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5973 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5974 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5975 // no receiver since it is static method
5976 Node* crc = argument(0); // type: int
5977 Node* src = argument(1); // type: long
5978 Node* offset = argument(3); // type: int
5979 Node* length = argument(4); // type: int
5980
5981 src = ConvL2X(src); // adjust Java long to machine word
5982 Node* base = _gvn.transform(new CastX2PNode(src));
5983 offset = ConvI2X(offset);
5984
5985 // 'src_start' points to src array + scaled offset
5986 Node* src_start = basic_plus_adr(top(), base, offset);
5987
5988 // Call the stub.
5989 address stubAddr = StubRoutines::updateBytesCRC32();
5990 const char *stubName = "updateBytesCRC32";
5991
5992 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5993 stubAddr, stubName, TypePtr::BOTTOM,
5994 crc, src_start, length);
5995 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5996 set_result(result);
5997 return true;
5998 }
5999
6000 //------------------------------get_table_from_crc32c_class-----------------------
6001 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6002 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
6003 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
6004
6005 return table;
6006 }
6007
6008 //------------------------------inline_updateBytesCRC32C-----------------------
6009 //
6010 // Calculate CRC32C for byte[] array.
6011 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6012 //
6013 bool LibraryCallKit::inline_updateBytesCRC32C() {
6014 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6015 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6016 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6017 // no receiver since it is a static method
6018 Node* crc = argument(0); // type: int
6019 Node* src = argument(1); // type: oop
6020 Node* offset = argument(2); // type: int
6021 Node* end = argument(3); // type: int
6022
6023 Node* length = _gvn.transform(new SubINode(end, offset));
6024
6025 const Type* src_type = src->Value(&_gvn);
6026 const TypeAryPtr* top_src = src_type->isa_aryptr();
6027 if (top_src == NULL || top_src->klass() == NULL) {
6028 // failed array check
6029 return false;
6030 }
6031
6032 // Figure out the size and type of the elements we will be copying.
6033 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6034 if (src_elem != T_BYTE) {
6035 return false;
6036 }
6037
6038 // 'src_start' points to src array + scaled offset
6039 src = must_be_not_null(src, true);
6040 src = access_resolve(src, ACCESS_READ);
6041 Node* src_start = array_element_address(src, offset, src_elem);
6042
6043 // static final int[] byteTable in class CRC32C
6044 Node* table = get_table_from_crc32c_class(callee()->holder());
6045 table = must_be_not_null(table, true);
6046 table = access_resolve(table, ACCESS_READ);
6047 Node* table_start = array_element_address(table, intcon(0), T_INT);
6048
6049 // We assume that range check is done by caller.
6050 // TODO: generate range check (offset+length < src.length) in debug VM.
6051
6052 // Call the stub.
6053 address stubAddr = StubRoutines::updateBytesCRC32C();
6054 const char *stubName = "updateBytesCRC32C";
6055
6056 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6057 stubAddr, stubName, TypePtr::BOTTOM,
6058 crc, src_start, length, table_start);
6059 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6060 set_result(result);
6061 return true;
6062 }
6063
6064 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6065 //
6066 // Calculate CRC32C for DirectByteBuffer.
6067 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6068 //
6069 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6070 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6071 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6072 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6073 // no receiver since it is a static method
6074 Node* crc = argument(0); // type: int
6075 Node* src = argument(1); // type: long
6076 Node* offset = argument(3); // type: int
6077 Node* end = argument(4); // type: int
6078
6079 Node* length = _gvn.transform(new SubINode(end, offset));
6080
6081 src = ConvL2X(src); // adjust Java long to machine word
6082 Node* base = _gvn.transform(new CastX2PNode(src));
6083 offset = ConvI2X(offset);
6084
6085 // 'src_start' points to src array + scaled offset
6086 Node* src_start = basic_plus_adr(top(), base, offset);
6087
6088 // static final int[] byteTable in class CRC32C
6089 Node* table = get_table_from_crc32c_class(callee()->holder());
6090 table = must_be_not_null(table, true);
6091 table = access_resolve(table, ACCESS_READ);
6092 Node* table_start = array_element_address(table, intcon(0), T_INT);
6093
6094 // Call the stub.
6095 address stubAddr = StubRoutines::updateBytesCRC32C();
6096 const char *stubName = "updateBytesCRC32C";
6097
6098 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6099 stubAddr, stubName, TypePtr::BOTTOM,
6100 crc, src_start, length, table_start);
6101 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6102 set_result(result);
6103 return true;
6104 }
6105
6106 //------------------------------inline_updateBytesAdler32----------------------
6107 //
6108 // Calculate Adler32 checksum for byte[] array.
6109 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6110 //
6111 bool LibraryCallKit::inline_updateBytesAdler32() {
6112 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
6113 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6114 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6115 // no receiver since it is static method
6116 Node* crc = argument(0); // type: int
6117 Node* src = argument(1); // type: oop
6118 Node* offset = argument(2); // type: int
6119 Node* length = argument(3); // type: int
6120
6121 const Type* src_type = src->Value(&_gvn);
6122 const TypeAryPtr* top_src = src_type->isa_aryptr();
6123 if (top_src == NULL || top_src->klass() == NULL) {
6124 // failed array check
6125 return false;
6126 }
6127
6128 // Figure out the size and type of the elements we will be copying.
6129 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6130 if (src_elem != T_BYTE) {
6131 return false;
6132 }
6133
6134 // 'src_start' points to src array + scaled offset
6135 src = access_resolve(src, ACCESS_READ);
6136 Node* src_start = array_element_address(src, offset, src_elem);
6137
6138 // We assume that range check is done by caller.
6139 // TODO: generate range check (offset+length < src.length) in debug VM.
6140
6141 // Call the stub.
6142 address stubAddr = StubRoutines::updateBytesAdler32();
6143 const char *stubName = "updateBytesAdler32";
6144
6145 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6146 stubAddr, stubName, TypePtr::BOTTOM,
6147 crc, src_start, length);
6148 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6149 set_result(result);
6150 return true;
6151 }
6152
6153 //------------------------------inline_updateByteBufferAdler32---------------
6154 //
6155 // Calculate Adler32 checksum for DirectByteBuffer.
6156 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6157 //
6158 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6159 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
6160 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6161 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6162 // no receiver since it is static method
6163 Node* crc = argument(0); // type: int
6164 Node* src = argument(1); // type: long
6165 Node* offset = argument(3); // type: int
6166 Node* length = argument(4); // type: int
6167
6168 src = ConvL2X(src); // adjust Java long to machine word
6169 Node* base = _gvn.transform(new CastX2PNode(src));
6170 offset = ConvI2X(offset);
6171
6172 // 'src_start' points to src array + scaled offset
6173 Node* src_start = basic_plus_adr(top(), base, offset);
6174
6175 // Call the stub.
6176 address stubAddr = StubRoutines::updateBytesAdler32();
6177 const char *stubName = "updateBytesAdler32";
6178
6179 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6180 stubAddr, stubName, TypePtr::BOTTOM,
6181 crc, src_start, length);
6182
6183 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6184 set_result(result);
6185 return true;
6186 }
6187
6188 //----------------------------inline_reference_get----------------------------
6189 // public T java.lang.ref.Reference.get();
6190 bool LibraryCallKit::inline_reference_get() {
6191 const int referent_offset = java_lang_ref_Reference::referent_offset;
6192 guarantee(referent_offset > 0, "should have already been set");
6193
6194 // Get the argument:
6195 Node* reference_obj = null_check_receiver();
6196 if (stopped()) return true;
6197
6198 const TypeInstPtr* tinst = _gvn.type(reference_obj)->isa_instptr();
6199 assert(tinst != NULL, "obj is null");
6200 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6201 ciInstanceKlass* referenceKlass = tinst->klass()->as_instance_klass();
6202 ciField* field = referenceKlass->get_field_by_name(ciSymbol::make("referent"),
6203 ciSymbol::make("Ljava/lang/Object;"),
6204 false);
6205 assert (field != NULL, "undefined field");
6206
6207 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
6208 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6209
6210 ciInstanceKlass* klass = env()->Object_klass();
6211 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
6212
6213 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6214 Node* result = access_load_at(reference_obj, adr, adr_type, object_type, T_OBJECT, decorators);
6215 // Add memory barrier to prevent commoning reads from this field
6216 // across safepoint since GC can change its value.
6217 insert_mem_bar(Op_MemBarCPUOrder);
6218
6219 set_result(result);
6220 return true;
6221 }
6222
6223
6224 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6225 bool is_exact=true, bool is_static=false,
6226 ciInstanceKlass * fromKls=NULL) {
6227 if (fromKls == NULL) {
6228 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6229 assert(tinst != NULL, "obj is null");
6230 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6231 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6232 fromKls = tinst->klass()->as_instance_klass();
6233 } else {
6234 assert(is_static, "only for static field access");
6235 }
6236 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6237 ciSymbol::make(fieldTypeString),
6238 is_static);
6239
6240 assert (field != NULL, "undefined field");
6241 if (field == NULL) return (Node *) NULL;
6242
6243 if (is_static) {
6244 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6245 fromObj = makecon(tip);
6246 }
6247
6248 // Next code copied from Parse::do_get_xxx():
6249
6250 // Compute address and memory type.
6251 int offset = field->offset_in_bytes();
6252 bool is_vol = field->is_volatile();
6253 ciType* field_klass = field->type();
6254 assert(field_klass->is_loaded(), "should be loaded");
6255 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6256 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6257 BasicType bt = field->layout_type();
6258
6259 // Build the resultant type of the load
6260 const Type *type;
6261 if (bt == T_OBJECT) {
6262 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6263 } else {
6264 type = Type::get_const_basic_type(bt);
6265 }
6266
6267 DecoratorSet decorators = IN_HEAP;
6268
6269 if (is_vol) {
6270 decorators |= MO_SEQ_CST;
6271 }
6272
6273 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
6274 }
6275
6276 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6277 bool is_exact = true, bool is_static = false,
6278 ciInstanceKlass * fromKls = NULL) {
6279 if (fromKls == NULL) {
6280 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6281 assert(tinst != NULL, "obj is null");
6282 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6283 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6284 fromKls = tinst->klass()->as_instance_klass();
6285 }
6286 else {
6287 assert(is_static, "only for static field access");
6288 }
6289 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6290 ciSymbol::make(fieldTypeString),
6291 is_static);
6292
6293 assert(field != NULL, "undefined field");
6294 assert(!field->is_volatile(), "not defined for volatile fields");
6295
6296 if (is_static) {
6297 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6298 fromObj = makecon(tip);
6299 }
6300
6301 // Next code copied from Parse::do_get_xxx():
6302
6303 // Compute address and memory type.
6304 int offset = field->offset_in_bytes();
6305 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6306
6307 return adr;
6308 }
6309
6310 //------------------------------inline_aescrypt_Block-----------------------
6311 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6312 address stubAddr = NULL;
6313 const char *stubName;
6314 assert(UseAES, "need AES instruction support");
6315
6316 switch(id) {
6317 case vmIntrinsics::_aescrypt_encryptBlock:
6318 stubAddr = StubRoutines::aescrypt_encryptBlock();
6319 stubName = "aescrypt_encryptBlock";
6320 break;
6321 case vmIntrinsics::_aescrypt_decryptBlock:
6322 stubAddr = StubRoutines::aescrypt_decryptBlock();
6323 stubName = "aescrypt_decryptBlock";
6324 break;
6325 default:
6326 break;
6327 }
6328 if (stubAddr == NULL) return false;
6329
6330 Node* aescrypt_object = argument(0);
6331 Node* src = argument(1);
6332 Node* src_offset = argument(2);
6333 Node* dest = argument(3);
6334 Node* dest_offset = argument(4);
6335
6336 src = must_be_not_null(src, true);
6337 dest = must_be_not_null(dest, true);
6338
6339 src = access_resolve(src, ACCESS_READ);
6340 dest = access_resolve(dest, ACCESS_WRITE);
6341
6342 // (1) src and dest are arrays.
6343 const Type* src_type = src->Value(&_gvn);
6344 const Type* dest_type = dest->Value(&_gvn);
6345 const TypeAryPtr* top_src = src_type->isa_aryptr();
6346 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6347 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6348
6349 // for the quick and dirty code we will skip all the checks.
6350 // we are just trying to get the call to be generated.
6351 Node* src_start = src;
6352 Node* dest_start = dest;
6353 if (src_offset != NULL || dest_offset != NULL) {
6354 assert(src_offset != NULL && dest_offset != NULL, "");
6355 src_start = array_element_address(src, src_offset, T_BYTE);
6356 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6357 }
6358
6359 // now need to get the start of its expanded key array
6360 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6361 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6362 if (k_start == NULL) return false;
6363
6364 if (Matcher::pass_original_key_for_aes()) {
6365 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6366 // compatibility issues between Java key expansion and SPARC crypto instructions
6367 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6368 if (original_k_start == NULL) return false;
6369
6370 // Call the stub.
6371 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6372 stubAddr, stubName, TypePtr::BOTTOM,
6373 src_start, dest_start, k_start, original_k_start);
6374 } else {
6375 // Call the stub.
6376 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6377 stubAddr, stubName, TypePtr::BOTTOM,
6378 src_start, dest_start, k_start);
6379 }
6380
6381 return true;
6382 }
6383
6384 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6385 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6386 address stubAddr = NULL;
6387 const char *stubName = NULL;
6388
6389 assert(UseAES, "need AES instruction support");
6390
6391 switch(id) {
6392 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6393 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6394 stubName = "cipherBlockChaining_encryptAESCrypt";
6395 break;
6396 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6397 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6398 stubName = "cipherBlockChaining_decryptAESCrypt";
6399 break;
6400 default:
6401 break;
6402 }
6403 if (stubAddr == NULL) return false;
6404
6405 Node* cipherBlockChaining_object = argument(0);
6406 Node* src = argument(1);
6407 Node* src_offset = argument(2);
6408 Node* len = argument(3);
6409 Node* dest = argument(4);
6410 Node* dest_offset = argument(5);
6411
6412 src = must_be_not_null(src, false);
6413 dest = must_be_not_null(dest, false);
6414
6415 src = access_resolve(src, ACCESS_READ);
6416 dest = access_resolve(dest, ACCESS_WRITE);
6417
6418 // (1) src and dest are arrays.
6419 const Type* src_type = src->Value(&_gvn);
6420 const Type* dest_type = dest->Value(&_gvn);
6421 const TypeAryPtr* top_src = src_type->isa_aryptr();
6422 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6423 assert (top_src != NULL && top_src->klass() != NULL
6424 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6425
6426 // checks are the responsibility of the caller
6427 Node* src_start = src;
6428 Node* dest_start = dest;
6429 if (src_offset != NULL || dest_offset != NULL) {
6430 assert(src_offset != NULL && dest_offset != NULL, "");
6431 src_start = array_element_address(src, src_offset, T_BYTE);
6432 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6433 }
6434
6435 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6436 // (because of the predicated logic executed earlier).
6437 // so we cast it here safely.
6438 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6439
6440 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6441 if (embeddedCipherObj == NULL) return false;
6442
6443 // cast it to what we know it will be at runtime
6444 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6445 assert(tinst != NULL, "CBC obj is null");
6446 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
6447 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6448 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6449
6450 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6451 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6452 const TypeOopPtr* xtype = aklass->as_instance_type();
6453 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6454 aescrypt_object = _gvn.transform(aescrypt_object);
6455
6456 // we need to get the start of the aescrypt_object's expanded key array
6457 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6458 if (k_start == NULL) return false;
6459
6460 // similarly, get the start address of the r vector
6461 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
6462 if (objRvec == NULL) return false;
6463 objRvec = access_resolve(objRvec, ACCESS_WRITE);
6464 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6465
6466 Node* cbcCrypt;
6467 if (Matcher::pass_original_key_for_aes()) {
6468 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6469 // compatibility issues between Java key expansion and SPARC crypto instructions
6470 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6471 if (original_k_start == NULL) return false;
6472
6473 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
6474 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6475 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6476 stubAddr, stubName, TypePtr::BOTTOM,
6477 src_start, dest_start, k_start, r_start, len, original_k_start);
6478 } else {
6479 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6480 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6481 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6482 stubAddr, stubName, TypePtr::BOTTOM,
6483 src_start, dest_start, k_start, r_start, len);
6484 }
6485
6486 // return cipher length (int)
6487 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6488 set_result(retvalue);
6489 return true;
6490 }
6491
6492 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
6493 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
6494 address stubAddr = NULL;
6495 const char *stubName = NULL;
6496
6497 assert(UseAES, "need AES instruction support");
6498
6499 switch (id) {
6500 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
6501 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
6502 stubName = "electronicCodeBook_encryptAESCrypt";
6503 break;
6504 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
6505 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
6506 stubName = "electronicCodeBook_decryptAESCrypt";
6507 break;
6508 default:
6509 break;
6510 }
6511
6512 if (stubAddr == NULL) return false;
6513
6514 Node* electronicCodeBook_object = argument(0);
6515 Node* src = argument(1);
6516 Node* src_offset = argument(2);
6517 Node* len = argument(3);
6518 Node* dest = argument(4);
6519 Node* dest_offset = argument(5);
6520
6521 // (1) src and dest are arrays.
6522 const Type* src_type = src->Value(&_gvn);
6523 const Type* dest_type = dest->Value(&_gvn);
6524 const TypeAryPtr* top_src = src_type->isa_aryptr();
6525 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6526 assert(top_src != NULL && top_src->klass() != NULL
6527 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6528
6529 // checks are the responsibility of the caller
6530 Node* src_start = src;
6531 Node* dest_start = dest;
6532 if (src_offset != NULL || dest_offset != NULL) {
6533 assert(src_offset != NULL && dest_offset != NULL, "");
6534 src_start = array_element_address(src, src_offset, T_BYTE);
6535 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6536 }
6537
6538 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6539 // (because of the predicated logic executed earlier).
6540 // so we cast it here safely.
6541 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6542
6543 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6544 if (embeddedCipherObj == NULL) return false;
6545
6546 // cast it to what we know it will be at runtime
6547 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6548 assert(tinst != NULL, "ECB obj is null");
6549 assert(tinst->klass()->is_loaded(), "ECB obj is not loaded");
6550 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6551 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6552
6553 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6554 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6555 const TypeOopPtr* xtype = aklass->as_instance_type();
6556 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6557 aescrypt_object = _gvn.transform(aescrypt_object);
6558
6559 // we need to get the start of the aescrypt_object's expanded key array
6560 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6561 if (k_start == NULL) return false;
6562
6563 Node* ecbCrypt;
6564 if (Matcher::pass_original_key_for_aes()) {
6565 // no SPARC version for AES/ECB intrinsics now.
6566 return false;
6567 }
6568 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6569 ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6570 OptoRuntime::electronicCodeBook_aescrypt_Type(),
6571 stubAddr, stubName, TypePtr::BOTTOM,
6572 src_start, dest_start, k_start, len);
6573
6574 // return cipher length (int)
6575 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6576 set_result(retvalue);
6577 return true;
6578 }
6579
6580 //------------------------------inline_counterMode_AESCrypt-----------------------
6581 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6582 assert(UseAES, "need AES instruction support");
6583 if (!UseAESCTRIntrinsics) return false;
6584
6585 address stubAddr = NULL;
6586 const char *stubName = NULL;
6587 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6588 stubAddr = StubRoutines::counterMode_AESCrypt();
6589 stubName = "counterMode_AESCrypt";
6590 }
6591 if (stubAddr == NULL) return false;
6592
6593 Node* counterMode_object = argument(0);
6594 Node* src = argument(1);
6595 Node* src_offset = argument(2);
6596 Node* len = argument(3);
6597 Node* dest = argument(4);
6598 Node* dest_offset = argument(5);
6599
6600 src = access_resolve(src, ACCESS_READ);
6601 dest = access_resolve(dest, ACCESS_WRITE);
6602 counterMode_object = access_resolve(counterMode_object, ACCESS_WRITE);
6603
6604 // (1) src and dest are arrays.
6605 const Type* src_type = src->Value(&_gvn);
6606 const Type* dest_type = dest->Value(&_gvn);
6607 const TypeAryPtr* top_src = src_type->isa_aryptr();
6608 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6609 assert(top_src != NULL && top_src->klass() != NULL &&
6610 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6611
6612 // checks are the responsibility of the caller
6613 Node* src_start = src;
6614 Node* dest_start = dest;
6615 if (src_offset != NULL || dest_offset != NULL) {
6616 assert(src_offset != NULL && dest_offset != NULL, "");
6617 src_start = array_element_address(src, src_offset, T_BYTE);
6618 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6619 }
6620
6621 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6622 // (because of the predicated logic executed earlier).
6623 // so we cast it here safely.
6624 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6625 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6626 if (embeddedCipherObj == NULL) return false;
6627 // cast it to what we know it will be at runtime
6628 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6629 assert(tinst != NULL, "CTR obj is null");
6630 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6631 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6632 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6633 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6634 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6635 const TypeOopPtr* xtype = aklass->as_instance_type();
6636 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6637 aescrypt_object = _gvn.transform(aescrypt_object);
6638 // we need to get the start of the aescrypt_object's expanded key array
6639 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6640 if (k_start == NULL) return false;
6641 // similarly, get the start address of the r vector
6642 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6643 if (obj_counter == NULL) return false;
6644 obj_counter = access_resolve(obj_counter, ACCESS_WRITE);
6645 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6646
6647 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6648 if (saved_encCounter == NULL) return false;
6649 saved_encCounter = access_resolve(saved_encCounter, ACCESS_WRITE);
6650 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6651 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6652
6653 Node* ctrCrypt;
6654 if (Matcher::pass_original_key_for_aes()) {
6655 // no SPARC version for AES/CTR intrinsics now.
6656 return false;
6657 }
6658 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6659 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6660 OptoRuntime::counterMode_aescrypt_Type(),
6661 stubAddr, stubName, TypePtr::BOTTOM,
6662 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6663
6664 // return cipher length (int)
6665 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6666 set_result(retvalue);
6667 return true;
6668 }
6669
6670 //------------------------------get_key_start_from_aescrypt_object-----------------------
6671 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6672 #if defined(PPC64) || defined(S390)
6673 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6674 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6675 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6676 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6677 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6678 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6679 if (objSessionK == NULL) {
6680 return (Node *) NULL;
6681 }
6682 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6683 #else
6684 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6685 #endif // PPC64
6686 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6687 if (objAESCryptKey == NULL) return (Node *) NULL;
6688
6689 // now have the array, need to get the start address of the K array
6690 objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ);
6691 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6692 return k_start;
6693 }
6694
6695 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6696 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6697 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6698 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6699 if (objAESCryptKey == NULL) return (Node *) NULL;
6700
6701 // now have the array, need to get the start address of the lastKey array
6702 objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ);
6703 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6704 return original_k_start;
6705 }
6706
6707 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6708 // Return node representing slow path of predicate check.
6709 // the pseudo code we want to emulate with this predicate is:
6710 // for encryption:
6711 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6712 // for decryption:
6713 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6714 // note cipher==plain is more conservative than the original java code but that's OK
6715 //
6716 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6717 // The receiver was checked for NULL already.
6718 Node* objCBC = argument(0);
6719
6720 Node* src = argument(1);
6721 Node* dest = argument(4);
6722
6723 // Load embeddedCipher field of CipherBlockChaining object.
6724 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6725
6726 // get AESCrypt klass for instanceOf check
6727 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6728 // will have same classloader as CipherBlockChaining object
6729 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6730 assert(tinst != NULL, "CBCobj is null");
6731 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6732
6733 // we want to do an instanceof comparison against the AESCrypt class
6734 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6735 if (!klass_AESCrypt->is_loaded()) {
6736 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6737 Node* ctrl = control();
6738 set_control(top()); // no regular fast path
6739 return ctrl;
6740 }
6741
6742 src = must_be_not_null(src, true);
6743 dest = must_be_not_null(dest, true);
6744
6745 // Resolve oops to stable for CmpP below.
6746 src = access_resolve(src, 0);
6747 dest = access_resolve(dest, 0);
6748
6749 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6750
6751 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6752 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6753 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6754
6755 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6756
6757 // for encryption, we are done
6758 if (!decrypting)
6759 return instof_false; // even if it is NULL
6760
6761 // for decryption, we need to add a further check to avoid
6762 // taking the intrinsic path when cipher and plain are the same
6763 // see the original java code for why.
6764 RegionNode* region = new RegionNode(3);
6765 region->init_req(1, instof_false);
6766
6767 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6768 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6769 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6770 region->init_req(2, src_dest_conjoint);
6771
6772 record_for_igvn(region);
6773 return _gvn.transform(region);
6774 }
6775
6776 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
6777 // Return node representing slow path of predicate check.
6778 // the pseudo code we want to emulate with this predicate is:
6779 // for encryption:
6780 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6781 // for decryption:
6782 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6783 // note cipher==plain is more conservative than the original java code but that's OK
6784 //
6785 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
6786 // The receiver was checked for NULL already.
6787 Node* objECB = argument(0);
6788
6789 // Load embeddedCipher field of ElectronicCodeBook object.
6790 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6791
6792 // get AESCrypt klass for instanceOf check
6793 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6794 // will have same classloader as ElectronicCodeBook object
6795 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
6796 assert(tinst != NULL, "ECBobj is null");
6797 assert(tinst->klass()->is_loaded(), "ECBobj is not loaded");
6798
6799 // we want to do an instanceof comparison against the AESCrypt class
6800 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6801 if (!klass_AESCrypt->is_loaded()) {
6802 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6803 Node* ctrl = control();
6804 set_control(top()); // no regular fast path
6805 return ctrl;
6806 }
6807 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6808
6809 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6810 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6811 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6812
6813 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6814
6815 // for encryption, we are done
6816 if (!decrypting)
6817 return instof_false; // even if it is NULL
6818
6819 // for decryption, we need to add a further check to avoid
6820 // taking the intrinsic path when cipher and plain are the same
6821 // see the original java code for why.
6822 RegionNode* region = new RegionNode(3);
6823 region->init_req(1, instof_false);
6824 Node* src = argument(1);
6825 Node* dest = argument(4);
6826 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6827 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6828 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6829 region->init_req(2, src_dest_conjoint);
6830
6831 record_for_igvn(region);
6832 return _gvn.transform(region);
6833 }
6834
6835 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6836 // Return node representing slow path of predicate check.
6837 // the pseudo code we want to emulate with this predicate is:
6838 // for encryption:
6839 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6840 // for decryption:
6841 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6842 // note cipher==plain is more conservative than the original java code but that's OK
6843 //
6844
6845 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6846 // The receiver was checked for NULL already.
6847 Node* objCTR = argument(0);
6848
6849 // Load embeddedCipher field of CipherBlockChaining object.
6850 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6851
6852 // get AESCrypt klass for instanceOf check
6853 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6854 // will have same classloader as CipherBlockChaining object
6855 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6856 assert(tinst != NULL, "CTRobj is null");
6857 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6858
6859 // we want to do an instanceof comparison against the AESCrypt class
6860 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6861 if (!klass_AESCrypt->is_loaded()) {
6862 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6863 Node* ctrl = control();
6864 set_control(top()); // no regular fast path
6865 return ctrl;
6866 }
6867
6868 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6869 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6870 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6871 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6872 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6873
6874 return instof_false; // even if it is NULL
6875 }
6876
6877 //------------------------------inline_ghash_processBlocks
6878 bool LibraryCallKit::inline_ghash_processBlocks() {
6879 address stubAddr;
6880 const char *stubName;
6881 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6882
6883 stubAddr = StubRoutines::ghash_processBlocks();
6884 stubName = "ghash_processBlocks";
6885
6886 Node* data = argument(0);
6887 Node* offset = argument(1);
6888 Node* len = argument(2);
6889 Node* state = argument(3);
6890 Node* subkeyH = argument(4);
6891
6892 state = must_be_not_null(state, true);
6893 subkeyH = must_be_not_null(subkeyH, true);
6894 data = must_be_not_null(data, true);
6895
6896 state = access_resolve(state, ACCESS_WRITE);
6897 subkeyH = access_resolve(subkeyH, ACCESS_READ);
6898 data = access_resolve(data, ACCESS_READ);
6899
6900 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6901 assert(state_start, "state is NULL");
6902 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6903 assert(subkeyH_start, "subkeyH is NULL");
6904 Node* data_start = array_element_address(data, offset, T_BYTE);
6905 assert(data_start, "data is NULL");
6906
6907 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6908 OptoRuntime::ghash_processBlocks_Type(),
6909 stubAddr, stubName, TypePtr::BOTTOM,
6910 state_start, subkeyH_start, data_start, len);
6911 return true;
6912 }
6913
6914 bool LibraryCallKit::inline_base64_encodeBlock() {
6915 address stubAddr;
6916 const char *stubName;
6917 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6918 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6919 stubAddr = StubRoutines::base64_encodeBlock();
6920 stubName = "encodeBlock";
6921
6922 if (!stubAddr) return false;
6923 Node* base64obj = argument(0);
6924 Node* src = argument(1);
6925 Node* offset = argument(2);
6926 Node* len = argument(3);
6927 Node* dest = argument(4);
6928 Node* dp = argument(5);
6929 Node* isURL = argument(6);
6930
6931 src = must_be_not_null(src, true);
6932 src = access_resolve(src, ACCESS_READ);
6933 dest = must_be_not_null(dest, true);
6934 dest = access_resolve(dest, ACCESS_WRITE);
6935
6936 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6937 assert(src_start, "source array is NULL");
6938 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6939 assert(dest_start, "destination array is NULL");
6940
6941 Node* base64 = make_runtime_call(RC_LEAF,
6942 OptoRuntime::base64_encodeBlock_Type(),
6943 stubAddr, stubName, TypePtr::BOTTOM,
6944 src_start, offset, len, dest_start, dp, isURL);
6945 return true;
6946 }
6947
6948 //------------------------------inline_sha_implCompress-----------------------
6949 //
6950 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6951 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6952 //
6953 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6954 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6955 //
6956 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6957 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6958 //
6959 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6960 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6961
6962 Node* sha_obj = argument(0);
6963 Node* src = argument(1); // type oop
6964 Node* ofs = argument(2); // type int
6965
6966 const Type* src_type = src->Value(&_gvn);
6967 const TypeAryPtr* top_src = src_type->isa_aryptr();
6968 if (top_src == NULL || top_src->klass() == NULL) {
6969 // failed array check
6970 return false;
6971 }
6972 // Figure out the size and type of the elements we will be copying.
6973 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6974 if (src_elem != T_BYTE) {
6975 return false;
6976 }
6977 // 'src_start' points to src array + offset
6978 src = must_be_not_null(src, true);
6979 src = access_resolve(src, ACCESS_READ);
6980 Node* src_start = array_element_address(src, ofs, src_elem);
6981 Node* state = NULL;
6982 address stubAddr;
6983 const char *stubName;
6984
6985 switch(id) {
6986 case vmIntrinsics::_sha_implCompress:
6987 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6988 state = get_state_from_sha_object(sha_obj);
6989 stubAddr = StubRoutines::sha1_implCompress();
6990 stubName = "sha1_implCompress";
6991 break;
6992 case vmIntrinsics::_sha2_implCompress:
6993 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6994 state = get_state_from_sha_object(sha_obj);
6995 stubAddr = StubRoutines::sha256_implCompress();
6996 stubName = "sha256_implCompress";
6997 break;
6998 case vmIntrinsics::_sha5_implCompress:
6999 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
7000 state = get_state_from_sha5_object(sha_obj);
7001 stubAddr = StubRoutines::sha512_implCompress();
7002 stubName = "sha512_implCompress";
7003 break;
7004 default:
7005 fatal_unexpected_iid(id);
7006 return false;
7007 }
7008 if (state == NULL) return false;
7009
7010 assert(stubAddr != NULL, "Stub is generated");
7011 if (stubAddr == NULL) return false;
7012
7013 // Call the stub.
7014 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
7015 stubAddr, stubName, TypePtr::BOTTOM,
7016 src_start, state);
7017
7018 return true;
7019 }
7020
7021 //------------------------------inline_digestBase_implCompressMB-----------------------
7022 //
7023 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
7024 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
7025 //
7026 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
7027 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
7028 "need SHA1/SHA256/SHA512 instruction support");
7029 assert((uint)predicate < 3, "sanity");
7030 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7031
7032 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
7033 Node* src = argument(1); // byte[] array
7034 Node* ofs = argument(2); // type int
7035 Node* limit = argument(3); // type int
7036
7037 const Type* src_type = src->Value(&_gvn);
7038 const TypeAryPtr* top_src = src_type->isa_aryptr();
7039 if (top_src == NULL || top_src->klass() == NULL) {
7040 // failed array check
7041 return false;
7042 }
7043 // Figure out the size and type of the elements we will be copying.
7044 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
7045 if (src_elem != T_BYTE) {
7046 return false;
7047 }
7048 // 'src_start' points to src array + offset
7049 src = must_be_not_null(src, false);
7050 src = access_resolve(src, ACCESS_READ);
7051 Node* src_start = array_element_address(src, ofs, src_elem);
7052
7053 const char* klass_SHA_name = NULL;
7054 const char* stub_name = NULL;
7055 address stub_addr = NULL;
7056 bool long_state = false;
7057
7058 switch (predicate) {
7059 case 0:
7060 if (UseSHA1Intrinsics) {
7061 klass_SHA_name = "sun/security/provider/SHA";
7062 stub_name = "sha1_implCompressMB";
7063 stub_addr = StubRoutines::sha1_implCompressMB();
7064 }
7065 break;
7066 case 1:
7067 if (UseSHA256Intrinsics) {
7068 klass_SHA_name = "sun/security/provider/SHA2";
7069 stub_name = "sha256_implCompressMB";
7070 stub_addr = StubRoutines::sha256_implCompressMB();
7071 }
7072 break;
7073 case 2:
7074 if (UseSHA512Intrinsics) {
7075 klass_SHA_name = "sun/security/provider/SHA5";
7076 stub_name = "sha512_implCompressMB";
7077 stub_addr = StubRoutines::sha512_implCompressMB();
7078 long_state = true;
7079 }
7080 break;
7081 default:
7082 fatal("unknown SHA intrinsic predicate: %d", predicate);
7083 }
7084 if (klass_SHA_name != NULL) {
7085 assert(stub_addr != NULL, "Stub is generated");
7086 if (stub_addr == NULL) return false;
7087
7088 // get DigestBase klass to lookup for SHA klass
7089 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7090 assert(tinst != NULL, "digestBase_obj is not instance???");
7091 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
7092
7093 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
7094 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
7095 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
7096 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
7097 }
7098 return false;
7099 }
7100 //------------------------------inline_sha_implCompressMB-----------------------
7101 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
7102 bool long_state, address stubAddr, const char *stubName,
7103 Node* src_start, Node* ofs, Node* limit) {
7104 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
7105 const TypeOopPtr* xtype = aklass->as_instance_type();
7106 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
7107 sha_obj = _gvn.transform(sha_obj);
7108
7109 Node* state;
7110 if (long_state) {
7111 state = get_state_from_sha5_object(sha_obj);
7112 } else {
7113 state = get_state_from_sha_object(sha_obj);
7114 }
7115 if (state == NULL) return false;
7116
7117 // Call the stub.
7118 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7119 OptoRuntime::digestBase_implCompressMB_Type(),
7120 stubAddr, stubName, TypePtr::BOTTOM,
7121 src_start, state, ofs, limit);
7122 // return ofs (int)
7123 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7124 set_result(result);
7125
7126 return true;
7127 }
7128
7129 //------------------------------get_state_from_sha_object-----------------------
7130 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
7131 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
7132 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
7133 if (sha_state == NULL) return (Node *) NULL;
7134
7135 // now have the array, need to get the start address of the state array
7136 sha_state = access_resolve(sha_state, ACCESS_WRITE);
7137 Node* state = array_element_address(sha_state, intcon(0), T_INT);
7138 return state;
7139 }
7140
7141 //------------------------------get_state_from_sha5_object-----------------------
7142 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
7143 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
7144 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
7145 if (sha_state == NULL) return (Node *) NULL;
7146
7147 // now have the array, need to get the start address of the state array
7148 sha_state = access_resolve(sha_state, ACCESS_WRITE);
7149 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
7150 return state;
7151 }
7152
7153 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
7154 // Return node representing slow path of predicate check.
7155 // the pseudo code we want to emulate with this predicate is:
7156 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
7157 //
7158 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
7159 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
7160 "need SHA1/SHA256/SHA512 instruction support");
7161 assert((uint)predicate < 3, "sanity");
7162
7163 // The receiver was checked for NULL already.
7164 Node* digestBaseObj = argument(0);
7165
7166 // get DigestBase klass for instanceOf check
7167 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
7168 assert(tinst != NULL, "digestBaseObj is null");
7169 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
7170
7171 const char* klass_SHA_name = NULL;
7172 switch (predicate) {
7173 case 0:
7174 if (UseSHA1Intrinsics) {
7175 // we want to do an instanceof comparison against the SHA class
7176 klass_SHA_name = "sun/security/provider/SHA";
7177 }
7178 break;
7179 case 1:
7180 if (UseSHA256Intrinsics) {
7181 // we want to do an instanceof comparison against the SHA2 class
7182 klass_SHA_name = "sun/security/provider/SHA2";
7183 }
7184 break;
7185 case 2:
7186 if (UseSHA512Intrinsics) {
7187 // we want to do an instanceof comparison against the SHA5 class
7188 klass_SHA_name = "sun/security/provider/SHA5";
7189 }
7190 break;
7191 default:
7192 fatal("unknown SHA intrinsic predicate: %d", predicate);
7193 }
7194
7195 ciKlass* klass_SHA = NULL;
7196 if (klass_SHA_name != NULL) {
7197 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
7198 }
7199 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
7200 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
7201 Node* ctrl = control();
7202 set_control(top()); // no intrinsic path
7203 return ctrl;
7204 }
7205 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
7206
7207 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
7208 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
7209 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7210 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
7211
7212 return instof_false; // even if it is NULL
7213 }
7214
7215 //-------------inline_fma-----------------------------------
7216 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
7217 Node *a = NULL;
7218 Node *b = NULL;
7219 Node *c = NULL;
7220 Node* result = NULL;
7221 switch (id) {
7222 case vmIntrinsics::_fmaD:
7223 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
7224 // no receiver since it is static method
7225 a = round_double_node(argument(0));
7226 b = round_double_node(argument(2));
7227 c = round_double_node(argument(4));
7228 result = _gvn.transform(new FmaDNode(control(), a, b, c));
7229 break;
7230 case vmIntrinsics::_fmaF:
7231 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
7232 a = argument(0);
7233 b = argument(1);
7234 c = argument(2);
7235 result = _gvn.transform(new FmaFNode(control(), a, b, c));
7236 break;
7237 default:
7238 fatal_unexpected_iid(id); break;
7239 }
7240 set_result(result);
7241 return true;
7242 }
7243
7244 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
7245 // argument(0) is receiver
7246 Node* codePoint = argument(1);
7247 Node* n = NULL;
7248
7249 switch (id) {
7250 case vmIntrinsics::_isDigit :
7251 n = new DigitNode(control(), codePoint);
7252 break;
7253 case vmIntrinsics::_isLowerCase :
7254 n = new LowerCaseNode(control(), codePoint);
7255 break;
7256 case vmIntrinsics::_isUpperCase :
7257 n = new UpperCaseNode(control(), codePoint);
7258 break;
7259 case vmIntrinsics::_isWhitespace :
7260 n = new WhitespaceNode(control(), codePoint);
7261 break;
7262 default:
7263 fatal_unexpected_iid(id);
7264 }
7265
7266 set_result(_gvn.transform(n));
7267 return true;
7268 }
7269
7270 //------------------------------inline_fp_min_max------------------------------
7271 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
7272 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
7273
7274 // The intrinsic should be used only when the API branches aren't predictable,
7275 // the last one performing the most important comparison. The following heuristic
7276 // uses the branch statistics to eventually bail out if necessary.
7277
7278 ciMethodData *md = callee()->method_data();
7279
7280 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
7281 ciCallProfile cp = caller()->call_profile_at_bci(bci());
7282
7283 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
7284 // Bail out if the call-site didn't contribute enough to the statistics.
7285 return false;
7286 }
7287
7288 uint taken = 0, not_taken = 0;
7289
7290 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
7291 if (p->is_BranchData()) {
7292 taken = ((ciBranchData*)p)->taken();
7293 not_taken = ((ciBranchData*)p)->not_taken();
7294 }
7295 }
7296
7297 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
7298 balance = balance < 0 ? -balance : balance;
7299 if ( balance > 0.2 ) {
7300 // Bail out if the most important branch is predictable enough.
7301 return false;
7302 }
7303 }
7304 */
7305
7306 Node *a = NULL;
7307 Node *b = NULL;
7308 Node *n = NULL;
7309 switch (id) {
7310 case vmIntrinsics::_maxF:
7311 case vmIntrinsics::_minF:
7312 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
7313 a = argument(0);
7314 b = argument(1);
7315 break;
7316 case vmIntrinsics::_maxD:
7317 case vmIntrinsics::_minD:
7318 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
7319 a = round_double_node(argument(0));
7320 b = round_double_node(argument(2));
7321 break;
7322 default:
7323 fatal_unexpected_iid(id);
7324 break;
7325 }
7326 switch (id) {
7327 case vmIntrinsics::_maxF: n = new MaxFNode(a, b); break;
7328 case vmIntrinsics::_minF: n = new MinFNode(a, b); break;
7329 case vmIntrinsics::_maxD: n = new MaxDNode(a, b); break;
7330 case vmIntrinsics::_minD: n = new MinDNode(a, b); break;
7331 default: fatal_unexpected_iid(id); break;
7332 }
7333 set_result(_gvn.transform(n));
7334 return true;
7335 }
7336
7337 bool LibraryCallKit::inline_profileBoolean() {
7338 Node* counts = argument(1);
7339 const TypeAryPtr* ary = NULL;
7340 ciArray* aobj = NULL;
7341 if (counts->is_Con()
7342 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
7343 && (aobj = ary->const_oop()->as_array()) != NULL
7344 && (aobj->length() == 2)) {
7345 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
7346 jint false_cnt = aobj->element_value(0).as_int();
7347 jint true_cnt = aobj->element_value(1).as_int();
7348
7349 if (C->log() != NULL) {
7350 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
7351 false_cnt, true_cnt);
7352 }
7353
7354 if (false_cnt + true_cnt == 0) {
7355 // According to profile, never executed.
7356 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7357 Deoptimization::Action_reinterpret);
7358 return true;
7359 }
7360
7361 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
7362 // is a number of each value occurrences.
7363 Node* result = argument(0);
7364 if (false_cnt == 0 || true_cnt == 0) {
7365 // According to profile, one value has been never seen.
7366 int expected_val = (false_cnt == 0) ? 1 : 0;
7367
7368 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
7369 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7370
7371 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
7372 Node* fast_path = _gvn.transform(new IfTrueNode(check));
7373 Node* slow_path = _gvn.transform(new IfFalseNode(check));
7374
7375 { // Slow path: uncommon trap for never seen value and then reexecute
7376 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
7377 // the value has been seen at least once.
7378 PreserveJVMState pjvms(this);
7379 PreserveReexecuteState preexecs(this);
7380 jvms()->set_should_reexecute(true);
7381
7382 set_control(slow_path);
7383 set_i_o(i_o());
7384
7385 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7386 Deoptimization::Action_reinterpret);
7387 }
7388 // The guard for never seen value enables sharpening of the result and
7389 // returning a constant. It allows to eliminate branches on the same value
7390 // later on.
7391 set_control(fast_path);
7392 result = intcon(expected_val);
7393 }
7394 // Stop profiling.
7395 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
7396 // By replacing method body with profile data (represented as ProfileBooleanNode
7397 // on IR level) we effectively disable profiling.
7398 // It enables full speed execution once optimized code is generated.
7399 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
7400 C->record_for_igvn(profile);
7401 set_result(profile);
7402 return true;
7403 } else {
7404 // Continue profiling.
7405 // Profile data isn't available at the moment. So, execute method's bytecode version.
7406 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
7407 // is compiled and counters aren't available since corresponding MethodHandle
7408 // isn't a compile-time constant.
7409 return false;
7410 }
7411 }
7412
7413 bool LibraryCallKit::inline_isCompileConstant() {
7414 Node* n = argument(0);
7415 set_result(n->is_Con() ? intcon(1) : intcon(0));
7416 return true;
7417 }