1 /*
2 * Copyright (c) 1999, 2019, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/systemDictionary.hpp"
29 #include "classfile/vmSymbols.hpp"
30 #include "compiler/compileBroker.hpp"
31 #include "compiler/compileLog.hpp"
32 #include "gc/shared/barrierSet.hpp"
33 #include "jfr/support/jfrIntrinsics.hpp"
34 #include "memory/resourceArea.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/callGenerator.hpp"
40 #include "opto/castnode.hpp"
41 #include "opto/cfgnode.hpp"
42 #include "opto/convertnode.hpp"
43 #include "opto/countbitsnode.hpp"
44 #include "opto/intrinsicnode.hpp"
45 #include "opto/idealKit.hpp"
46 #include "opto/mathexactnode.hpp"
47 #include "opto/movenode.hpp"
48 #include "opto/mulnode.hpp"
49 #include "opto/narrowptrnode.hpp"
50 #include "opto/opaquenode.hpp"
51 #include "opto/parse.hpp"
52 #include "opto/runtime.hpp"
53 #include "opto/rootnode.hpp"
54 #include "opto/subnode.hpp"
55 #include "prims/nativeLookup.hpp"
56 #include "prims/unsafe.hpp"
57 #include "runtime/objectMonitor.hpp"
58 #include "runtime/sharedRuntime.hpp"
59 #include "utilities/macros.hpp"
60
61
62 class LibraryIntrinsic : public InlineCallGenerator {
63 // Extend the set of intrinsics known to the runtime:
64 public:
65 private:
66 bool _is_virtual;
67 bool _does_virtual_dispatch;
68 int8_t _predicates_count; // Intrinsic is predicated by several conditions
69 int8_t _last_predicate; // Last generated predicate
70 vmIntrinsics::ID _intrinsic_id;
71
72 public:
73 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
74 : InlineCallGenerator(m),
75 _is_virtual(is_virtual),
76 _does_virtual_dispatch(does_virtual_dispatch),
77 _predicates_count((int8_t)predicates_count),
78 _last_predicate((int8_t)-1),
79 _intrinsic_id(id)
80 {
81 }
82 virtual bool is_intrinsic() const { return true; }
83 virtual bool is_virtual() const { return _is_virtual; }
84 virtual bool is_predicated() const { return _predicates_count > 0; }
85 virtual int predicates_count() const { return _predicates_count; }
86 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; }
87 virtual JVMState* generate(JVMState* jvms);
88 virtual Node* generate_predicate(JVMState* jvms, int predicate);
89 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
90 };
91
92
93 // Local helper class for LibraryIntrinsic:
94 class LibraryCallKit : public GraphKit {
95 private:
96 LibraryIntrinsic* _intrinsic; // the library intrinsic being called
97 Node* _result; // the result node, if any
98 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted
99
100 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type);
101
102 public:
103 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
104 : GraphKit(jvms),
105 _intrinsic(intrinsic),
106 _result(NULL)
107 {
108 // Check if this is a root compile. In that case we don't have a caller.
109 if (!jvms->has_method()) {
110 _reexecute_sp = sp();
111 } else {
112 // Find out how many arguments the interpreter needs when deoptimizing
113 // and save the stack pointer value so it can used by uncommon_trap.
114 // We find the argument count by looking at the declared signature.
115 bool ignored_will_link;
116 ciSignature* declared_signature = NULL;
117 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
118 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
119 _reexecute_sp = sp() + nargs; // "push" arguments back on stack
120 }
121 }
122
123 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
124
125 ciMethod* caller() const { return jvms()->method(); }
126 int bci() const { return jvms()->bci(); }
127 LibraryIntrinsic* intrinsic() const { return _intrinsic; }
128 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); }
129 ciMethod* callee() const { return _intrinsic->method(); }
130
131 bool try_to_inline(int predicate);
132 Node* try_to_predicate(int predicate);
133
134 void push_result() {
135 // Push the result onto the stack.
136 if (!stopped() && result() != NULL) {
137 BasicType bt = result()->bottom_type()->basic_type();
138 push_node(bt, result());
139 }
140 }
141
142 private:
143 void fatal_unexpected_iid(vmIntrinsics::ID iid) {
144 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid));
145 }
146
147 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
148 void set_result(RegionNode* region, PhiNode* value);
149 Node* result() { return _result; }
150
151 virtual int reexecute_sp() { return _reexecute_sp; }
152
153 // Helper functions to inline natives
154 Node* generate_guard(Node* test, RegionNode* region, float true_prob);
155 Node* generate_slow_guard(Node* test, RegionNode* region);
156 Node* generate_fair_guard(Node* test, RegionNode* region);
157 Node* generate_negative_guard(Node* index, RegionNode* region,
158 // resulting CastII of index:
159 Node* *pos_index = NULL);
160 Node* generate_limit_guard(Node* offset, Node* subseq_length,
161 Node* array_length,
162 RegionNode* region);
163 void generate_string_range_check(Node* array, Node* offset,
164 Node* length, bool char_count);
165 Node* generate_current_thread(Node* &tls_output);
166 Node* load_mirror_from_klass(Node* klass);
167 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
168 RegionNode* region, int null_path,
169 int offset);
170 Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
171 RegionNode* region, int null_path) {
172 int offset = java_lang_Class::klass_offset_in_bytes();
173 return load_klass_from_mirror_common(mirror, never_see_null,
174 region, null_path,
175 offset);
176 }
177 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
178 RegionNode* region, int null_path) {
179 int offset = java_lang_Class::array_klass_offset_in_bytes();
180 return load_klass_from_mirror_common(mirror, never_see_null,
181 region, null_path,
182 offset);
183 }
184 Node* generate_access_flags_guard(Node* kls,
185 int modifier_mask, int modifier_bits,
186 RegionNode* region);
187 Node* generate_interface_guard(Node* kls, RegionNode* region);
188 Node* generate_array_guard(Node* kls, RegionNode* region) {
189 return generate_array_guard_common(kls, region, false, false);
190 }
191 Node* generate_non_array_guard(Node* kls, RegionNode* region) {
192 return generate_array_guard_common(kls, region, false, true);
193 }
194 Node* generate_objArray_guard(Node* kls, RegionNode* region) {
195 return generate_array_guard_common(kls, region, true, false);
196 }
197 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
198 return generate_array_guard_common(kls, region, true, true);
199 }
200 Node* generate_array_guard_common(Node* kls, RegionNode* region,
201 bool obj_array, bool not_array);
202 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
203 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
204 bool is_virtual = false, bool is_static = false);
205 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
206 return generate_method_call(method_id, false, true);
207 }
208 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
209 return generate_method_call(method_id, true, false);
210 }
211 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
212 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
213
214 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
215 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
216 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
217 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
218 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
219 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
220 bool inline_string_indexOfChar();
221 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
222 bool inline_string_toBytesU();
223 bool inline_string_getCharsU();
224 bool inline_string_copy(bool compress);
225 bool inline_string_char_access(bool is_store);
226 Node* round_double_node(Node* n);
227 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
228 bool inline_math_native(vmIntrinsics::ID id);
229 bool inline_math(vmIntrinsics::ID id);
230 bool inline_double_math(vmIntrinsics::ID id);
231 template <typename OverflowOp>
232 bool inline_math_overflow(Node* arg1, Node* arg2);
233 void inline_math_mathExact(Node* math, Node* test);
234 bool inline_math_addExactI(bool is_increment);
235 bool inline_math_addExactL(bool is_increment);
236 bool inline_math_multiplyExactI();
237 bool inline_math_multiplyExactL();
238 bool inline_math_multiplyHigh();
239 bool inline_math_negateExactI();
240 bool inline_math_negateExactL();
241 bool inline_math_subtractExactI(bool is_decrement);
242 bool inline_math_subtractExactL(bool is_decrement);
243 bool inline_min_max(vmIntrinsics::ID id);
244 bool inline_notify(vmIntrinsics::ID id);
245 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
246 // This returns Type::AnyPtr, RawPtr, or OopPtr.
247 int classify_unsafe_addr(Node* &base, Node* &offset, BasicType type);
248 Node* make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type = T_ILLEGAL, bool can_cast = false);
249
250 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
251 DecoratorSet mo_decorator_for_access_kind(AccessKind kind);
252 bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
253 static bool klass_needs_init_guard(Node* kls);
254 bool inline_unsafe_allocate();
255 bool inline_unsafe_newArray(bool uninitialized);
256 bool inline_unsafe_copyMemory();
257 bool inline_native_currentThread();
258
259 bool inline_native_time_funcs(address method, const char* funcName);
260 #ifdef JFR_HAVE_INTRINSICS
261 bool inline_native_classID();
262 bool inline_native_getEventWriter();
263 #endif
264 bool inline_native_isInterrupted();
265 bool inline_native_Class_query(vmIntrinsics::ID id);
266 bool inline_native_subtype_check();
267 bool inline_native_getLength();
268 bool inline_array_copyOf(bool is_copyOfRange);
269 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
270 bool inline_preconditions_checkIndex();
271 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array);
272 bool inline_native_clone(bool is_virtual);
273 bool inline_native_Reflection_getCallerClass();
274 // Helper function for inlining native object hash method
275 bool inline_native_hashcode(bool is_virtual, bool is_static);
276 bool inline_native_getClass();
277
278 // Helper functions for inlining arraycopy
279 bool inline_arraycopy();
280 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
281 RegionNode* slow_region);
282 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
283 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp,
284 uint new_idx);
285
286 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
287 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
288 bool inline_unsafe_fence(vmIntrinsics::ID id);
289 bool inline_onspinwait();
290 bool inline_fp_conversions(vmIntrinsics::ID id);
291 bool inline_number_methods(vmIntrinsics::ID id);
292 bool inline_reference_get();
293 bool inline_Class_cast();
294 bool inline_aescrypt_Block(vmIntrinsics::ID id);
295 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
296 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
297 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
298 Node* inline_counterMode_AESCrypt_predicate();
299 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
300 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
301 bool inline_ghash_processBlocks();
302 bool inline_base64_encodeBlock();
303 bool inline_sha_implCompress(vmIntrinsics::ID id);
304 bool inline_digestBase_implCompressMB(int predicate);
305 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
306 bool long_state, address stubAddr, const char *stubName,
307 Node* src_start, Node* ofs, Node* limit);
308 Node* get_state_from_sha_object(Node *sha_object);
309 Node* get_state_from_sha5_object(Node *sha_object);
310 Node* inline_digestBase_implCompressMB_predicate(int predicate);
311 bool inline_encodeISOArray();
312 bool inline_updateCRC32();
313 bool inline_updateBytesCRC32();
314 bool inline_updateByteBufferCRC32();
315 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
316 bool inline_updateBytesCRC32C();
317 bool inline_updateDirectByteBufferCRC32C();
318 bool inline_updateBytesAdler32();
319 bool inline_updateByteBufferAdler32();
320 bool inline_multiplyToLen();
321 bool inline_hasNegatives();
322 bool inline_squareToLen();
323 bool inline_mulAdd();
324 bool inline_montgomeryMultiply();
325 bool inline_montgomerySquare();
326 bool inline_vectorizedMismatch();
327 bool inline_fma(vmIntrinsics::ID id);
328 bool inline_character_compare(vmIntrinsics::ID id);
329 bool inline_fp_min_max(vmIntrinsics::ID id);
330
331 bool inline_profileBoolean();
332 bool inline_isCompileConstant();
333 bool inline_continuation(vmIntrinsics::ID id);
334 bool inline_get_frame_pointer();
335 bool inline_continuation_do_yield();
336 bool inline_continuation_jump();
337 bool inline_continuation_runLevel();
338 bool inline_continuation_do_continue();
339 void clear_upper_avx() {
340 #ifdef X86
341 if (UseAVX >= 2) {
342 C->set_clear_upper_avx(true);
343 }
344 #endif
345 }
346 };
347
348 //---------------------------make_vm_intrinsic----------------------------
349 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
350 vmIntrinsics::ID id = m->intrinsic_id();
351 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
352
353 if (!m->is_loaded()) {
354 // Do not attempt to inline unloaded methods.
355 return NULL;
356 }
357
358 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
359 bool is_available = false;
360
361 {
362 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
363 // the compiler must transition to '_thread_in_vm' state because both
364 // methods access VM-internal data.
365 VM_ENTRY_MARK;
366 methodHandle mh(THREAD, m->get_Method());
367 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
368 !C->directive()->is_intrinsic_disabled(mh) &&
369 !vmIntrinsics::is_disabled_by_flags(mh);
370
371 }
372
373 if (is_available) {
374 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
375 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
376 return new LibraryIntrinsic(m, is_virtual,
377 vmIntrinsics::predicates_needed(id),
378 vmIntrinsics::does_virtual_dispatch(id),
379 (vmIntrinsics::ID) id);
380 } else {
381 return NULL;
382 }
383 }
384
385 //----------------------register_library_intrinsics-----------------------
386 // Initialize this file's data structures, for each Compile instance.
387 void Compile::register_library_intrinsics() {
388 // Nothing to do here.
389 }
390
391 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
392 LibraryCallKit kit(jvms, this);
393 Compile* C = kit.C;
394 int nodes = C->unique();
395 #ifndef PRODUCT
396 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
397 char buf[1000];
398 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
399 tty->print_cr("Intrinsic %s", str);
400 }
401 #endif
402 ciMethod* callee = kit.callee();
403 const int bci = kit.bci();
404
405 // Try to inline the intrinsic.
406 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
407 kit.try_to_inline(_last_predicate)) {
408 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
409 : "(intrinsic)";
410 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
411 if (C->print_intrinsics() || C->print_inlining()) {
412 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
413 }
414 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
415 if (C->log()) {
416 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
417 vmIntrinsics::name_at(intrinsic_id()),
418 (is_virtual() ? " virtual='1'" : ""),
419 C->unique() - nodes);
420 }
421 // Push the result from the inlined method onto the stack.
422 kit.push_result();
423 C->print_inlining_update(this);
424 return kit.transfer_exceptions_into_jvms();
425 }
426
427 // The intrinsic bailed out
428 if (jvms->has_method()) {
429 // Not a root compile.
430 const char* msg;
431 if (callee->intrinsic_candidate()) {
432 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
433 } else {
434 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
435 : "failed to inline (intrinsic), method not annotated";
436 }
437 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
438 if (C->print_intrinsics() || C->print_inlining()) {
439 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
440 }
441 } else {
442 // Root compile
443 ResourceMark rm;
444 stringStream msg_stream;
445 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
446 vmIntrinsics::name_at(intrinsic_id()),
447 is_virtual() ? " (virtual)" : "", bci);
448 const char *msg = msg_stream.as_string();
449 log_debug(jit, inlining)("%s", msg);
450 if (C->print_intrinsics() || C->print_inlining()) {
451 tty->print("%s", msg);
452 }
453 }
454 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
455 C->print_inlining_update(this);
456 return NULL;
457 }
458
459 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
460 LibraryCallKit kit(jvms, this);
461 Compile* C = kit.C;
462 int nodes = C->unique();
463 _last_predicate = predicate;
464 #ifndef PRODUCT
465 assert(is_predicated() && predicate < predicates_count(), "sanity");
466 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
467 char buf[1000];
468 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
469 tty->print_cr("Predicate for intrinsic %s", str);
470 }
471 #endif
472 ciMethod* callee = kit.callee();
473 const int bci = kit.bci();
474
475 Node* slow_ctl = kit.try_to_predicate(predicate);
476 if (!kit.failing()) {
477 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
478 : "(intrinsic, predicate)";
479 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
480 if (C->print_intrinsics() || C->print_inlining()) {
481 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
482 }
483 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
484 if (C->log()) {
485 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
486 vmIntrinsics::name_at(intrinsic_id()),
487 (is_virtual() ? " virtual='1'" : ""),
488 C->unique() - nodes);
489 }
490 return slow_ctl; // Could be NULL if the check folds.
491 }
492
493 // The intrinsic bailed out
494 if (jvms->has_method()) {
495 // Not a root compile.
496 const char* msg = "failed to generate predicate for intrinsic";
497 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
498 if (C->print_intrinsics() || C->print_inlining()) {
499 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
500 }
501 } else {
502 // Root compile
503 ResourceMark rm;
504 stringStream msg_stream;
505 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
506 vmIntrinsics::name_at(intrinsic_id()),
507 is_virtual() ? " (virtual)" : "", bci);
508 const char *msg = msg_stream.as_string();
509 log_debug(jit, inlining)("%s", msg);
510 if (C->print_intrinsics() || C->print_inlining()) {
511 C->print_inlining_stream()->print("%s", msg);
512 }
513 }
514 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
515 return NULL;
516 }
517
518 bool LibraryCallKit::try_to_inline(int predicate) {
519 // Handle symbolic names for otherwise undistinguished boolean switches:
520 const bool is_store = true;
521 const bool is_compress = true;
522 const bool is_static = true;
523 const bool is_volatile = true;
524
525 if (!jvms()->has_method()) {
526 // Root JVMState has a null method.
527 assert(map()->memory()->Opcode() == Op_Parm, "");
528 // Insert the memory aliasing node
529 set_all_memory(reset_memory());
530 }
531 assert(merged_memory(), "");
532
533
534 switch (intrinsic_id()) {
535 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
536 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
537 case vmIntrinsics::_getClass: return inline_native_getClass();
538
539 case vmIntrinsics::_dsin:
540 case vmIntrinsics::_dcos:
541 case vmIntrinsics::_dtan:
542 case vmIntrinsics::_dabs:
543 case vmIntrinsics::_fabs:
544 case vmIntrinsics::_iabs:
545 case vmIntrinsics::_labs:
546 case vmIntrinsics::_datan2:
547 case vmIntrinsics::_dsqrt:
548 case vmIntrinsics::_dexp:
549 case vmIntrinsics::_dlog:
550 case vmIntrinsics::_dlog10:
551 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id());
552
553 case vmIntrinsics::_min:
554 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
555
556 case vmIntrinsics::_notify:
557 case vmIntrinsics::_notifyAll:
558 return inline_notify(intrinsic_id());
559
560 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
561 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
562 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
563 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
564 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
565 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
566 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
567 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
568 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
569 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
570 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
571 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
572 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
573
574 case vmIntrinsics::_arraycopy: return inline_arraycopy();
575
576 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
577 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
578 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
579 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
580
581 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
582 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
583 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
584 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
585 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
586 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
587 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
588
589 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
590 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
591
592 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
593 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
594 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
595 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
596
597 case vmIntrinsics::_compressStringC:
598 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
599 case vmIntrinsics::_inflateStringC:
600 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
601
602 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
603 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
604 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
605 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
606 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
607 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
608 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
609 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
610 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
611
612 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
613 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
614 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
615 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
616 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
617 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
618 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
619 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
620 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
621
622 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
623 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
624 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
625 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
626 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
627 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
628 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
629 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
630 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
631
632 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
633 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
634 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
635 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
636 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
637 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
638 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
639 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
640 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
641
642 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
643 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
644 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
645 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
646
647 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
648 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
649 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
650 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
651
652 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
653 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
654 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
655 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
656 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
657 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
658 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
659 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
660 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
661
662 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
663 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
664 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
665 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
666 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
667 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
668 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
669 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
670 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
671
672 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
673 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
674 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
675 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
676 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
677 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
678 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
679 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
680 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
681
682 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
683 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
684 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
685 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
686 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
687 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
688 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
689 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
690 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
691
692 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
693 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
694 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
695 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
696 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
697
698 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
699 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
700 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
701 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
702 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
703 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
704 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
705 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
706 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
707 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
708 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
709 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
710 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
711 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
712 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
713 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
714 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
715 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
716 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
717 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
718
719 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
720 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
721 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
722 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
723 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
724 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
725 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
726 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
727 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
728 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
729 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
730 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
731 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
732 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
733 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
734
735 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
736 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
737 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
738 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
739
740 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
741 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
742 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
743 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
744 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
745
746 case vmIntrinsics::_loadFence:
747 case vmIntrinsics::_storeFence:
748 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
749
750 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
751
752 case vmIntrinsics::_currentThread: return inline_native_currentThread();
753 case vmIntrinsics::_isInterrupted: return inline_native_isInterrupted();
754
755 #ifdef JFR_HAVE_INTRINSICS
756 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime");
757 case vmIntrinsics::_getClassId: return inline_native_classID();
758 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
759 #endif
760 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
761 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
762 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
763 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
764 case vmIntrinsics::_getLength: return inline_native_getLength();
765 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
766 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
767 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
768 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
769 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
770 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
771
772 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
773 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
774
775 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
776
777 case vmIntrinsics::_isInstance:
778 case vmIntrinsics::_getModifiers:
779 case vmIntrinsics::_isInterface:
780 case vmIntrinsics::_isArray:
781 case vmIntrinsics::_isPrimitive:
782 case vmIntrinsics::_getSuperclass:
783 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
784
785 case vmIntrinsics::_floatToRawIntBits:
786 case vmIntrinsics::_floatToIntBits:
787 case vmIntrinsics::_intBitsToFloat:
788 case vmIntrinsics::_doubleToRawLongBits:
789 case vmIntrinsics::_doubleToLongBits:
790 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
791
792 case vmIntrinsics::_numberOfLeadingZeros_i:
793 case vmIntrinsics::_numberOfLeadingZeros_l:
794 case vmIntrinsics::_numberOfTrailingZeros_i:
795 case vmIntrinsics::_numberOfTrailingZeros_l:
796 case vmIntrinsics::_bitCount_i:
797 case vmIntrinsics::_bitCount_l:
798 case vmIntrinsics::_reverseBytes_i:
799 case vmIntrinsics::_reverseBytes_l:
800 case vmIntrinsics::_reverseBytes_s:
801 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
802
803 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
804
805 case vmIntrinsics::_Reference_get: return inline_reference_get();
806
807 case vmIntrinsics::_Class_cast: return inline_Class_cast();
808
809 case vmIntrinsics::_aescrypt_encryptBlock:
810 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
811
812 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
813 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
814 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
815
816 case vmIntrinsics::_counterMode_AESCrypt:
817 return inline_counterMode_AESCrypt(intrinsic_id());
818
819 case vmIntrinsics::_sha_implCompress:
820 case vmIntrinsics::_sha2_implCompress:
821 case vmIntrinsics::_sha5_implCompress:
822 return inline_sha_implCompress(intrinsic_id());
823
824 case vmIntrinsics::_digestBase_implCompressMB:
825 return inline_digestBase_implCompressMB(predicate);
826
827 case vmIntrinsics::_multiplyToLen:
828 return inline_multiplyToLen();
829
830 case vmIntrinsics::_squareToLen:
831 return inline_squareToLen();
832
833 case vmIntrinsics::_mulAdd:
834 return inline_mulAdd();
835
836 case vmIntrinsics::_montgomeryMultiply:
837 return inline_montgomeryMultiply();
838 case vmIntrinsics::_montgomerySquare:
839 return inline_montgomerySquare();
840
841 case vmIntrinsics::_vectorizedMismatch:
842 return inline_vectorizedMismatch();
843
844 case vmIntrinsics::_ghash_processBlocks:
845 return inline_ghash_processBlocks();
846 case vmIntrinsics::_base64_encodeBlock:
847 return inline_base64_encodeBlock();
848
849 case vmIntrinsics::_encodeISOArray:
850 case vmIntrinsics::_encodeByteISOArray:
851 return inline_encodeISOArray();
852
853 case vmIntrinsics::_updateCRC32:
854 return inline_updateCRC32();
855 case vmIntrinsics::_updateBytesCRC32:
856 return inline_updateBytesCRC32();
857 case vmIntrinsics::_updateByteBufferCRC32:
858 return inline_updateByteBufferCRC32();
859
860 case vmIntrinsics::_updateBytesCRC32C:
861 return inline_updateBytesCRC32C();
862 case vmIntrinsics::_updateDirectByteBufferCRC32C:
863 return inline_updateDirectByteBufferCRC32C();
864
865 case vmIntrinsics::_updateBytesAdler32:
866 return inline_updateBytesAdler32();
867 case vmIntrinsics::_updateByteBufferAdler32:
868 return inline_updateByteBufferAdler32();
869
870 case vmIntrinsics::_profileBoolean:
871 return inline_profileBoolean();
872 case vmIntrinsics::_isCompileConstant:
873 return inline_isCompileConstant();
874
875 case vmIntrinsics::_hasNegatives:
876 return inline_hasNegatives();
877
878 case vmIntrinsics::_fmaD:
879 case vmIntrinsics::_fmaF:
880 return inline_fma(intrinsic_id());
881
882 case vmIntrinsics::_Continuation_getFP:
883 return inline_get_frame_pointer();
884
885 case vmIntrinsics::_Continuation_getSP:
886 case vmIntrinsics::_Continuation_getPC:
887 return inline_continuation(intrinsic_id());
888 case vmIntrinsics::_Continuation_doContinue:
889 return inline_continuation_do_continue();
890 case vmIntrinsics::_Continuation_doYield:
891 return inline_continuation_do_yield();
892 case vmIntrinsics::_Continuation_jump:
893 return inline_continuation_jump();
894 case vmIntrinsics::_Continuation_runLevel:
895 return inline_continuation_runLevel();
896
897 case vmIntrinsics::_isDigit:
898 case vmIntrinsics::_isLowerCase:
899 case vmIntrinsics::_isUpperCase:
900 case vmIntrinsics::_isWhitespace:
901 return inline_character_compare(intrinsic_id());
902
903 case vmIntrinsics::_maxF:
904 case vmIntrinsics::_minF:
905 case vmIntrinsics::_maxD:
906 case vmIntrinsics::_minD:
907 return inline_fp_min_max(intrinsic_id());
908
909 default:
910 // If you get here, it may be that someone has added a new intrinsic
911 // to the list in vmSymbols.hpp without implementing it here.
912 #ifndef PRODUCT
913 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
914 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
915 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
916 }
917 #endif
918 return false;
919 }
920 }
921
922 Node* LibraryCallKit::try_to_predicate(int predicate) {
923 if (!jvms()->has_method()) {
924 // Root JVMState has a null method.
925 assert(map()->memory()->Opcode() == Op_Parm, "");
926 // Insert the memory aliasing node
927 set_all_memory(reset_memory());
928 }
929 assert(merged_memory(), "");
930
931 switch (intrinsic_id()) {
932 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
933 return inline_cipherBlockChaining_AESCrypt_predicate(false);
934 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
935 return inline_cipherBlockChaining_AESCrypt_predicate(true);
936 case vmIntrinsics::_counterMode_AESCrypt:
937 return inline_counterMode_AESCrypt_predicate();
938 case vmIntrinsics::_digestBase_implCompressMB:
939 return inline_digestBase_implCompressMB_predicate(predicate);
940
941 default:
942 // If you get here, it may be that someone has added a new intrinsic
943 // to the list in vmSymbols.hpp without implementing it here.
944 #ifndef PRODUCT
945 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
946 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
947 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
948 }
949 #endif
950 Node* slow_ctl = control();
951 set_control(top()); // No fast path instrinsic
952 return slow_ctl;
953 }
954 }
955
956 //------------------------------set_result-------------------------------
957 // Helper function for finishing intrinsics.
958 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
959 record_for_igvn(region);
960 set_control(_gvn.transform(region));
961 set_result( _gvn.transform(value));
962 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
963 }
964
965 //------------------------------generate_guard---------------------------
966 // Helper function for generating guarded fast-slow graph structures.
967 // The given 'test', if true, guards a slow path. If the test fails
968 // then a fast path can be taken. (We generally hope it fails.)
969 // In all cases, GraphKit::control() is updated to the fast path.
970 // The returned value represents the control for the slow path.
971 // The return value is never 'top'; it is either a valid control
972 // or NULL if it is obvious that the slow path can never be taken.
973 // Also, if region and the slow control are not NULL, the slow edge
974 // is appended to the region.
975 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
976 if (stopped()) {
977 // Already short circuited.
978 return NULL;
979 }
980
981 // Build an if node and its projections.
982 // If test is true we take the slow path, which we assume is uncommon.
983 if (_gvn.type(test) == TypeInt::ZERO) {
984 // The slow branch is never taken. No need to build this guard.
985 return NULL;
986 }
987
988 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
989
990 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
991 if (if_slow == top()) {
992 // The slow branch is never taken. No need to build this guard.
993 return NULL;
994 }
995
996 if (region != NULL)
997 region->add_req(if_slow);
998
999 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
1000 set_control(if_fast);
1001
1002 return if_slow;
1003 }
1004
1005 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
1006 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
1007 }
1008 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
1009 return generate_guard(test, region, PROB_FAIR);
1010 }
1011
1012 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
1013 Node* *pos_index) {
1014 if (stopped())
1015 return NULL; // already stopped
1016 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
1017 return NULL; // index is already adequately typed
1018 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
1019 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1020 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
1021 if (is_neg != NULL && pos_index != NULL) {
1022 // Emulate effect of Parse::adjust_map_after_if.
1023 Node* ccast = new CastIINode(index, TypeInt::POS);
1024 ccast->set_req(0, control());
1025 (*pos_index) = _gvn.transform(ccast);
1026 }
1027 return is_neg;
1028 }
1029
1030 // Make sure that 'position' is a valid limit index, in [0..length].
1031 // There are two equivalent plans for checking this:
1032 // A. (offset + copyLength) unsigned<= arrayLength
1033 // B. offset <= (arrayLength - copyLength)
1034 // We require that all of the values above, except for the sum and
1035 // difference, are already known to be non-negative.
1036 // Plan A is robust in the face of overflow, if offset and copyLength
1037 // are both hugely positive.
1038 //
1039 // Plan B is less direct and intuitive, but it does not overflow at
1040 // all, since the difference of two non-negatives is always
1041 // representable. Whenever Java methods must perform the equivalent
1042 // check they generally use Plan B instead of Plan A.
1043 // For the moment we use Plan A.
1044 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
1045 Node* subseq_length,
1046 Node* array_length,
1047 RegionNode* region) {
1048 if (stopped())
1049 return NULL; // already stopped
1050 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
1051 if (zero_offset && subseq_length->eqv_uncast(array_length))
1052 return NULL; // common case of whole-array copy
1053 Node* last = subseq_length;
1054 if (!zero_offset) // last += offset
1055 last = _gvn.transform(new AddINode(last, offset));
1056 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
1057 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1058 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
1059 return is_over;
1060 }
1061
1062 // Emit range checks for the given String.value byte array
1063 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
1064 if (stopped()) {
1065 return; // already stopped
1066 }
1067 RegionNode* bailout = new RegionNode(1);
1068 record_for_igvn(bailout);
1069 if (char_count) {
1070 // Convert char count to byte count
1071 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1072 }
1073
1074 // Offset and count must not be negative
1075 generate_negative_guard(offset, bailout);
1076 generate_negative_guard(count, bailout);
1077 // Offset + count must not exceed length of array
1078 generate_limit_guard(offset, count, load_array_length(array), bailout);
1079
1080 if (bailout->req() > 1) {
1081 PreserveJVMState pjvms(this);
1082 set_control(_gvn.transform(bailout));
1083 uncommon_trap(Deoptimization::Reason_intrinsic,
1084 Deoptimization::Action_maybe_recompile);
1085 }
1086 }
1087
1088 //--------------------------generate_current_thread--------------------
1089 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1090 ciKlass* thread_klass = env()->Thread_klass();
1091 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1092 Node* thread = _gvn.transform(new ThreadLocalNode());
1093 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1094 Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered);
1095 tls_output = thread;
1096 return threadObj;
1097 }
1098
1099
1100 //------------------------------make_string_method_node------------------------
1101 // Helper method for String intrinsic functions. This version is called with
1102 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1103 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1104 // containing the lengths of str1 and str2.
1105 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1106 Node* result = NULL;
1107 switch (opcode) {
1108 case Op_StrIndexOf:
1109 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1110 str1_start, cnt1, str2_start, cnt2, ae);
1111 break;
1112 case Op_StrComp:
1113 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1114 str1_start, cnt1, str2_start, cnt2, ae);
1115 break;
1116 case Op_StrEquals:
1117 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1118 // Use the constant length if there is one because optimized match rule may exist.
1119 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1120 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1121 break;
1122 default:
1123 ShouldNotReachHere();
1124 return NULL;
1125 }
1126
1127 // All these intrinsics have checks.
1128 C->set_has_split_ifs(true); // Has chance for split-if optimization
1129 clear_upper_avx();
1130
1131 return _gvn.transform(result);
1132 }
1133
1134 //------------------------------inline_string_compareTo------------------------
1135 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1136 Node* arg1 = argument(0);
1137 Node* arg2 = argument(1);
1138
1139 arg1 = must_be_not_null(arg1, true);
1140 arg2 = must_be_not_null(arg2, true);
1141
1142 arg1 = access_resolve(arg1, ACCESS_READ);
1143 arg2 = access_resolve(arg2, ACCESS_READ);
1144
1145 // Get start addr and length of first argument
1146 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1147 Node* arg1_cnt = load_array_length(arg1);
1148
1149 // Get start addr and length of second argument
1150 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1151 Node* arg2_cnt = load_array_length(arg2);
1152
1153 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1154 set_result(result);
1155 return true;
1156 }
1157
1158 //------------------------------inline_string_equals------------------------
1159 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1160 Node* arg1 = argument(0);
1161 Node* arg2 = argument(1);
1162
1163 // paths (plus control) merge
1164 RegionNode* region = new RegionNode(3);
1165 Node* phi = new PhiNode(region, TypeInt::BOOL);
1166
1167 if (!stopped()) {
1168
1169 arg1 = must_be_not_null(arg1, true);
1170 arg2 = must_be_not_null(arg2, true);
1171
1172 arg1 = access_resolve(arg1, ACCESS_READ);
1173 arg2 = access_resolve(arg2, ACCESS_READ);
1174
1175 // Get start addr and length of first argument
1176 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1177 Node* arg1_cnt = load_array_length(arg1);
1178
1179 // Get start addr and length of second argument
1180 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1181 Node* arg2_cnt = load_array_length(arg2);
1182
1183 // Check for arg1_cnt != arg2_cnt
1184 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1185 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1186 Node* if_ne = generate_slow_guard(bol, NULL);
1187 if (if_ne != NULL) {
1188 phi->init_req(2, intcon(0));
1189 region->init_req(2, if_ne);
1190 }
1191
1192 // Check for count == 0 is done by assembler code for StrEquals.
1193
1194 if (!stopped()) {
1195 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1196 phi->init_req(1, equals);
1197 region->init_req(1, control());
1198 }
1199 }
1200
1201 // post merge
1202 set_control(_gvn.transform(region));
1203 record_for_igvn(region);
1204
1205 set_result(_gvn.transform(phi));
1206 return true;
1207 }
1208
1209 //------------------------------inline_array_equals----------------------------
1210 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1211 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1212 Node* arg1 = argument(0);
1213 Node* arg2 = argument(1);
1214
1215 arg1 = access_resolve(arg1, ACCESS_READ);
1216 arg2 = access_resolve(arg2, ACCESS_READ);
1217
1218 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1219 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1220 clear_upper_avx();
1221
1222 return true;
1223 }
1224
1225 //------------------------------inline_hasNegatives------------------------------
1226 bool LibraryCallKit::inline_hasNegatives() {
1227 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1228 return false;
1229 }
1230
1231 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1232 // no receiver since it is static method
1233 Node* ba = argument(0);
1234 Node* offset = argument(1);
1235 Node* len = argument(2);
1236
1237 ba = must_be_not_null(ba, true);
1238
1239 // Range checks
1240 generate_string_range_check(ba, offset, len, false);
1241 if (stopped()) {
1242 return true;
1243 }
1244 ba = access_resolve(ba, ACCESS_READ);
1245 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1246 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1247 set_result(_gvn.transform(result));
1248 return true;
1249 }
1250
1251 bool LibraryCallKit::inline_preconditions_checkIndex() {
1252 Node* index = argument(0);
1253 Node* length = argument(1);
1254 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1255 return false;
1256 }
1257
1258 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1259 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1260
1261 {
1262 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1263 uncommon_trap(Deoptimization::Reason_intrinsic,
1264 Deoptimization::Action_make_not_entrant);
1265 }
1266
1267 if (stopped()) {
1268 return false;
1269 }
1270
1271 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1272 BoolTest::mask btest = BoolTest::lt;
1273 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1274 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1275 _gvn.set_type(rc, rc->Value(&_gvn));
1276 if (!rc_bool->is_Con()) {
1277 record_for_igvn(rc);
1278 }
1279 set_control(_gvn.transform(new IfTrueNode(rc)));
1280 {
1281 PreserveJVMState pjvms(this);
1282 set_control(_gvn.transform(new IfFalseNode(rc)));
1283 uncommon_trap(Deoptimization::Reason_range_check,
1284 Deoptimization::Action_make_not_entrant);
1285 }
1286
1287 if (stopped()) {
1288 return false;
1289 }
1290
1291 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1292 result->set_req(0, control());
1293 result = _gvn.transform(result);
1294 set_result(result);
1295 replace_in_map(index, result);
1296 clear_upper_avx();
1297 return true;
1298 }
1299
1300 //------------------------------inline_string_indexOf------------------------
1301 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1302 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1303 return false;
1304 }
1305 Node* src = argument(0);
1306 Node* tgt = argument(1);
1307
1308 // Make the merge point
1309 RegionNode* result_rgn = new RegionNode(4);
1310 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1311
1312 src = must_be_not_null(src, true);
1313 tgt = must_be_not_null(tgt, true);
1314
1315 src = access_resolve(src, ACCESS_READ);
1316 tgt = access_resolve(tgt, ACCESS_READ);
1317
1318 // Get start addr and length of source string
1319 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1320 Node* src_count = load_array_length(src);
1321
1322 // Get start addr and length of substring
1323 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1324 Node* tgt_count = load_array_length(tgt);
1325
1326 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1327 // Divide src size by 2 if String is UTF16 encoded
1328 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1329 }
1330 if (ae == StrIntrinsicNode::UU) {
1331 // Divide substring size by 2 if String is UTF16 encoded
1332 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1333 }
1334
1335 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1336 if (result != NULL) {
1337 result_phi->init_req(3, result);
1338 result_rgn->init_req(3, control());
1339 }
1340 set_control(_gvn.transform(result_rgn));
1341 record_for_igvn(result_rgn);
1342 set_result(_gvn.transform(result_phi));
1343
1344 return true;
1345 }
1346
1347 //-----------------------------inline_string_indexOf-----------------------
1348 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1349 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1350 return false;
1351 }
1352 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1353 return false;
1354 }
1355 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1356 Node* src = argument(0); // byte[]
1357 Node* src_count = argument(1); // char count
1358 Node* tgt = argument(2); // byte[]
1359 Node* tgt_count = argument(3); // char count
1360 Node* from_index = argument(4); // char index
1361
1362 src = must_be_not_null(src, true);
1363 tgt = must_be_not_null(tgt, true);
1364
1365 src = access_resolve(src, ACCESS_READ);
1366 tgt = access_resolve(tgt, ACCESS_READ);
1367
1368 // Multiply byte array index by 2 if String is UTF16 encoded
1369 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1370 src_count = _gvn.transform(new SubINode(src_count, from_index));
1371 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1372 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1373
1374 // Range checks
1375 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1376 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1377 if (stopped()) {
1378 return true;
1379 }
1380
1381 RegionNode* region = new RegionNode(5);
1382 Node* phi = new PhiNode(region, TypeInt::INT);
1383
1384 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1385 if (result != NULL) {
1386 // The result is index relative to from_index if substring was found, -1 otherwise.
1387 // Generate code which will fold into cmove.
1388 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1389 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1390
1391 Node* if_lt = generate_slow_guard(bol, NULL);
1392 if (if_lt != NULL) {
1393 // result == -1
1394 phi->init_req(3, result);
1395 region->init_req(3, if_lt);
1396 }
1397 if (!stopped()) {
1398 result = _gvn.transform(new AddINode(result, from_index));
1399 phi->init_req(4, result);
1400 region->init_req(4, control());
1401 }
1402 }
1403
1404 set_control(_gvn.transform(region));
1405 record_for_igvn(region);
1406 set_result(_gvn.transform(phi));
1407 clear_upper_avx();
1408
1409 return true;
1410 }
1411
1412 // Create StrIndexOfNode with fast path checks
1413 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1414 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1415 // Check for substr count > string count
1416 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1417 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1418 Node* if_gt = generate_slow_guard(bol, NULL);
1419 if (if_gt != NULL) {
1420 phi->init_req(1, intcon(-1));
1421 region->init_req(1, if_gt);
1422 }
1423 if (!stopped()) {
1424 // Check for substr count == 0
1425 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1426 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1427 Node* if_zero = generate_slow_guard(bol, NULL);
1428 if (if_zero != NULL) {
1429 phi->init_req(2, intcon(0));
1430 region->init_req(2, if_zero);
1431 }
1432 }
1433 if (!stopped()) {
1434 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1435 }
1436 return NULL;
1437 }
1438
1439 //-----------------------------inline_string_indexOfChar-----------------------
1440 bool LibraryCallKit::inline_string_indexOfChar() {
1441 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1442 return false;
1443 }
1444 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1445 return false;
1446 }
1447 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1448 Node* src = argument(0); // byte[]
1449 Node* tgt = argument(1); // tgt is int ch
1450 Node* from_index = argument(2);
1451 Node* max = argument(3);
1452
1453 src = must_be_not_null(src, true);
1454 src = access_resolve(src, ACCESS_READ);
1455
1456 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1457 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1458 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1459
1460 // Range checks
1461 generate_string_range_check(src, src_offset, src_count, true);
1462 if (stopped()) {
1463 return true;
1464 }
1465
1466 RegionNode* region = new RegionNode(3);
1467 Node* phi = new PhiNode(region, TypeInt::INT);
1468
1469 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1470 C->set_has_split_ifs(true); // Has chance for split-if optimization
1471 _gvn.transform(result);
1472
1473 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1474 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1475
1476 Node* if_lt = generate_slow_guard(bol, NULL);
1477 if (if_lt != NULL) {
1478 // result == -1
1479 phi->init_req(2, result);
1480 region->init_req(2, if_lt);
1481 }
1482 if (!stopped()) {
1483 result = _gvn.transform(new AddINode(result, from_index));
1484 phi->init_req(1, result);
1485 region->init_req(1, control());
1486 }
1487 set_control(_gvn.transform(region));
1488 record_for_igvn(region);
1489 set_result(_gvn.transform(phi));
1490
1491 return true;
1492 }
1493 //---------------------------inline_string_copy---------------------
1494 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1495 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1496 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1497 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1498 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1499 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1500 bool LibraryCallKit::inline_string_copy(bool compress) {
1501 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1502 return false;
1503 }
1504 int nargs = 5; // 2 oops, 3 ints
1505 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1506
1507 Node* src = argument(0);
1508 Node* src_offset = argument(1);
1509 Node* dst = argument(2);
1510 Node* dst_offset = argument(3);
1511 Node* length = argument(4);
1512
1513 // Check for allocation before we add nodes that would confuse
1514 // tightly_coupled_allocation()
1515 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1516
1517 // Figure out the size and type of the elements we will be copying.
1518 const Type* src_type = src->Value(&_gvn);
1519 const Type* dst_type = dst->Value(&_gvn);
1520 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1521 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1522 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1523 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1524 "Unsupported array types for inline_string_copy");
1525
1526 src = must_be_not_null(src, true);
1527 dst = must_be_not_null(dst, true);
1528
1529 // Convert char[] offsets to byte[] offsets
1530 bool convert_src = (compress && src_elem == T_BYTE);
1531 bool convert_dst = (!compress && dst_elem == T_BYTE);
1532 if (convert_src) {
1533 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1534 } else if (convert_dst) {
1535 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1536 }
1537
1538 // Range checks
1539 generate_string_range_check(src, src_offset, length, convert_src);
1540 generate_string_range_check(dst, dst_offset, length, convert_dst);
1541 if (stopped()) {
1542 return true;
1543 }
1544
1545 src = access_resolve(src, ACCESS_READ);
1546 dst = access_resolve(dst, ACCESS_WRITE);
1547
1548 Node* src_start = array_element_address(src, src_offset, src_elem);
1549 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1550 // 'src_start' points to src array + scaled offset
1551 // 'dst_start' points to dst array + scaled offset
1552 Node* count = NULL;
1553 if (compress) {
1554 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1555 } else {
1556 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1557 }
1558
1559 if (alloc != NULL) {
1560 if (alloc->maybe_set_complete(&_gvn)) {
1561 // "You break it, you buy it."
1562 InitializeNode* init = alloc->initialization();
1563 assert(init->is_complete(), "we just did this");
1564 init->set_complete_with_arraycopy();
1565 assert(dst->is_CheckCastPP(), "sanity");
1566 assert(dst->in(0)->in(0) == init, "dest pinned");
1567 }
1568 // Do not let stores that initialize this object be reordered with
1569 // a subsequent store that would make this object accessible by
1570 // other threads.
1571 // Record what AllocateNode this StoreStore protects so that
1572 // escape analysis can go from the MemBarStoreStoreNode to the
1573 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1574 // based on the escape status of the AllocateNode.
1575 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1576 }
1577 if (compress) {
1578 set_result(_gvn.transform(count));
1579 }
1580 clear_upper_avx();
1581
1582 return true;
1583 }
1584
1585 #ifdef _LP64
1586 #define XTOP ,top() /*additional argument*/
1587 #else //_LP64
1588 #define XTOP /*no additional argument*/
1589 #endif //_LP64
1590
1591 //------------------------inline_string_toBytesU--------------------------
1592 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1593 bool LibraryCallKit::inline_string_toBytesU() {
1594 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1595 return false;
1596 }
1597 // Get the arguments.
1598 Node* value = argument(0);
1599 Node* offset = argument(1);
1600 Node* length = argument(2);
1601
1602 Node* newcopy = NULL;
1603
1604 // Set the original stack and the reexecute bit for the interpreter to reexecute
1605 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1606 { PreserveReexecuteState preexecs(this);
1607 jvms()->set_should_reexecute(true);
1608
1609 // Check if a null path was taken unconditionally.
1610 value = null_check(value);
1611
1612 RegionNode* bailout = new RegionNode(1);
1613 record_for_igvn(bailout);
1614
1615 // Range checks
1616 generate_negative_guard(offset, bailout);
1617 generate_negative_guard(length, bailout);
1618 generate_limit_guard(offset, length, load_array_length(value), bailout);
1619 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1620 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1621
1622 if (bailout->req() > 1) {
1623 PreserveJVMState pjvms(this);
1624 set_control(_gvn.transform(bailout));
1625 uncommon_trap(Deoptimization::Reason_intrinsic,
1626 Deoptimization::Action_maybe_recompile);
1627 }
1628 if (stopped()) {
1629 return true;
1630 }
1631
1632 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1633 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1634 newcopy = new_array(klass_node, size, 0); // no arguments to push
1635 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1636
1637 // Calculate starting addresses.
1638 value = access_resolve(value, ACCESS_READ);
1639 Node* src_start = array_element_address(value, offset, T_CHAR);
1640 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1641
1642 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1643 const TypeInt* toffset = gvn().type(offset)->is_int();
1644 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1645
1646 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1647 const char* copyfunc_name = "arraycopy";
1648 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1649 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1650 OptoRuntime::fast_arraycopy_Type(),
1651 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1652 src_start, dst_start, ConvI2X(length) XTOP);
1653 // Do not let reads from the cloned object float above the arraycopy.
1654 if (alloc != NULL) {
1655 if (alloc->maybe_set_complete(&_gvn)) {
1656 // "You break it, you buy it."
1657 InitializeNode* init = alloc->initialization();
1658 assert(init->is_complete(), "we just did this");
1659 init->set_complete_with_arraycopy();
1660 assert(newcopy->is_CheckCastPP(), "sanity");
1661 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1662 }
1663 // Do not let stores that initialize this object be reordered with
1664 // a subsequent store that would make this object accessible by
1665 // other threads.
1666 // Record what AllocateNode this StoreStore protects so that
1667 // escape analysis can go from the MemBarStoreStoreNode to the
1668 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1669 // based on the escape status of the AllocateNode.
1670 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1671 } else {
1672 insert_mem_bar(Op_MemBarCPUOrder);
1673 }
1674 } // original reexecute is set back here
1675
1676 C->set_has_split_ifs(true); // Has chance for split-if optimization
1677 if (!stopped()) {
1678 set_result(newcopy);
1679 }
1680 clear_upper_avx();
1681
1682 return true;
1683 }
1684
1685 //------------------------inline_string_getCharsU--------------------------
1686 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1687 bool LibraryCallKit::inline_string_getCharsU() {
1688 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1689 return false;
1690 }
1691
1692 // Get the arguments.
1693 Node* src = argument(0);
1694 Node* src_begin = argument(1);
1695 Node* src_end = argument(2); // exclusive offset (i < src_end)
1696 Node* dst = argument(3);
1697 Node* dst_begin = argument(4);
1698
1699 // Check for allocation before we add nodes that would confuse
1700 // tightly_coupled_allocation()
1701 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1702
1703 // Check if a null path was taken unconditionally.
1704 src = null_check(src);
1705 dst = null_check(dst);
1706 if (stopped()) {
1707 return true;
1708 }
1709
1710 // Get length and convert char[] offset to byte[] offset
1711 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1712 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1713
1714 // Range checks
1715 generate_string_range_check(src, src_begin, length, true);
1716 generate_string_range_check(dst, dst_begin, length, false);
1717 if (stopped()) {
1718 return true;
1719 }
1720
1721 if (!stopped()) {
1722 src = access_resolve(src, ACCESS_READ);
1723 dst = access_resolve(dst, ACCESS_WRITE);
1724
1725 // Calculate starting addresses.
1726 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1727 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1728
1729 // Check if array addresses are aligned to HeapWordSize
1730 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1731 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1732 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1733 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1734
1735 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1736 const char* copyfunc_name = "arraycopy";
1737 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1738 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1739 OptoRuntime::fast_arraycopy_Type(),
1740 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1741 src_start, dst_start, ConvI2X(length) XTOP);
1742 // Do not let reads from the cloned object float above the arraycopy.
1743 if (alloc != NULL) {
1744 if (alloc->maybe_set_complete(&_gvn)) {
1745 // "You break it, you buy it."
1746 InitializeNode* init = alloc->initialization();
1747 assert(init->is_complete(), "we just did this");
1748 init->set_complete_with_arraycopy();
1749 assert(dst->is_CheckCastPP(), "sanity");
1750 assert(dst->in(0)->in(0) == init, "dest pinned");
1751 }
1752 // Do not let stores that initialize this object be reordered with
1753 // a subsequent store that would make this object accessible by
1754 // other threads.
1755 // Record what AllocateNode this StoreStore protects so that
1756 // escape analysis can go from the MemBarStoreStoreNode to the
1757 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1758 // based on the escape status of the AllocateNode.
1759 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1760 } else {
1761 insert_mem_bar(Op_MemBarCPUOrder);
1762 }
1763 }
1764
1765 C->set_has_split_ifs(true); // Has chance for split-if optimization
1766 return true;
1767 }
1768
1769 //----------------------inline_string_char_access----------------------------
1770 // Store/Load char to/from byte[] array.
1771 // static void StringUTF16.putChar(byte[] val, int index, int c)
1772 // static char StringUTF16.getChar(byte[] val, int index)
1773 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1774 Node* value = argument(0);
1775 Node* index = argument(1);
1776 Node* ch = is_store ? argument(2) : NULL;
1777
1778 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1779 // correctly requires matched array shapes.
1780 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1781 "sanity: byte[] and char[] bases agree");
1782 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1783 "sanity: byte[] and char[] scales agree");
1784
1785 // Bail when getChar over constants is requested: constant folding would
1786 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1787 // Java method would constant fold nicely instead.
1788 if (!is_store && value->is_Con() && index->is_Con()) {
1789 return false;
1790 }
1791
1792 value = must_be_not_null(value, true);
1793 value = access_resolve(value, is_store ? ACCESS_WRITE : ACCESS_READ);
1794
1795 Node* adr = array_element_address(value, index, T_CHAR);
1796 if (adr->is_top()) {
1797 return false;
1798 }
1799 if (is_store) {
1800 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1801 } else {
1802 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
1803 set_result(ch);
1804 }
1805 return true;
1806 }
1807
1808 //--------------------------round_double_node--------------------------------
1809 // Round a double node if necessary.
1810 Node* LibraryCallKit::round_double_node(Node* n) {
1811 if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1812 n = _gvn.transform(new RoundDoubleNode(0, n));
1813 return n;
1814 }
1815
1816 //------------------------------inline_math-----------------------------------
1817 // public static double Math.abs(double)
1818 // public static double Math.sqrt(double)
1819 // public static double Math.log(double)
1820 // public static double Math.log10(double)
1821 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1822 Node* arg = round_double_node(argument(0));
1823 Node* n = NULL;
1824 switch (id) {
1825 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1826 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1827 default: fatal_unexpected_iid(id); break;
1828 }
1829 set_result(_gvn.transform(n));
1830 return true;
1831 }
1832
1833 //------------------------------inline_math-----------------------------------
1834 // public static float Math.abs(float)
1835 // public static int Math.abs(int)
1836 // public static long Math.abs(long)
1837 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1838 Node* arg = argument(0);
1839 Node* n = NULL;
1840 switch (id) {
1841 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1842 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1843 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1844 default: fatal_unexpected_iid(id); break;
1845 }
1846 set_result(_gvn.transform(n));
1847 return true;
1848 }
1849
1850 //------------------------------runtime_math-----------------------------
1851 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1852 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1853 "must be (DD)D or (D)D type");
1854
1855 // Inputs
1856 Node* a = round_double_node(argument(0));
1857 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1858
1859 const TypePtr* no_memory_effects = NULL;
1860 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1861 no_memory_effects,
1862 a, top(), b, b ? top() : NULL);
1863 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1864 #ifdef ASSERT
1865 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1866 assert(value_top == top(), "second value must be top");
1867 #endif
1868
1869 set_result(value);
1870 return true;
1871 }
1872
1873 //------------------------------inline_math_native-----------------------------
1874 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1875 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1876 switch (id) {
1877 // These intrinsics are not properly supported on all hardware
1878 case vmIntrinsics::_dsin:
1879 return StubRoutines::dsin() != NULL ?
1880 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1881 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1882 case vmIntrinsics::_dcos:
1883 return StubRoutines::dcos() != NULL ?
1884 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1885 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1886 case vmIntrinsics::_dtan:
1887 return StubRoutines::dtan() != NULL ?
1888 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1889 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1890 case vmIntrinsics::_dlog:
1891 return StubRoutines::dlog() != NULL ?
1892 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1893 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1894 case vmIntrinsics::_dlog10:
1895 return StubRoutines::dlog10() != NULL ?
1896 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1897 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1898
1899 // These intrinsics are supported on all hardware
1900 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1901 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1902 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1903 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1904 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1905
1906 case vmIntrinsics::_dexp:
1907 return StubRoutines::dexp() != NULL ?
1908 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1909 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1910 case vmIntrinsics::_dpow: {
1911 Node* exp = round_double_node(argument(2));
1912 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1913 if (d != NULL && d->getd() == 2.0) {
1914 // Special case: pow(x, 2.0) => x * x
1915 Node* base = round_double_node(argument(0));
1916 set_result(_gvn.transform(new MulDNode(base, base)));
1917 return true;
1918 }
1919 return StubRoutines::dpow() != NULL ?
1920 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1921 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1922 }
1923 #undef FN_PTR
1924
1925 // These intrinsics are not yet correctly implemented
1926 case vmIntrinsics::_datan2:
1927 return false;
1928
1929 default:
1930 fatal_unexpected_iid(id);
1931 return false;
1932 }
1933 }
1934
1935 static bool is_simple_name(Node* n) {
1936 return (n->req() == 1 // constant
1937 || (n->is_Type() && n->as_Type()->type()->singleton())
1938 || n->is_Proj() // parameter or return value
1939 || n->is_Phi() // local of some sort
1940 );
1941 }
1942
1943 //----------------------------inline_notify-----------------------------------*
1944 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1945 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1946 address func;
1947 if (id == vmIntrinsics::_notify) {
1948 func = OptoRuntime::monitor_notify_Java();
1949 } else {
1950 func = OptoRuntime::monitor_notifyAll_Java();
1951 }
1952 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1953 make_slow_call_ex(call, env()->Throwable_klass(), false);
1954 return true;
1955 }
1956
1957
1958 //----------------------------inline_min_max-----------------------------------
1959 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1960 set_result(generate_min_max(id, argument(0), argument(1)));
1961 return true;
1962 }
1963
1964 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1965 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1966 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1967 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1968 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1969
1970 {
1971 PreserveJVMState pjvms(this);
1972 PreserveReexecuteState preexecs(this);
1973 jvms()->set_should_reexecute(true);
1974
1975 set_control(slow_path);
1976 set_i_o(i_o());
1977
1978 uncommon_trap(Deoptimization::Reason_intrinsic,
1979 Deoptimization::Action_none);
1980 }
1981
1982 set_control(fast_path);
1983 set_result(math);
1984 }
1985
1986 template <typename OverflowOp>
1987 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1988 typedef typename OverflowOp::MathOp MathOp;
1989
1990 MathOp* mathOp = new MathOp(arg1, arg2);
1991 Node* operation = _gvn.transform( mathOp );
1992 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1993 inline_math_mathExact(operation, ofcheck);
1994 return true;
1995 }
1996
1997 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1998 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1999 }
2000
2001 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2002 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2003 }
2004
2005 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2006 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2007 }
2008
2009 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2010 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2011 }
2012
2013 bool LibraryCallKit::inline_math_negateExactI() {
2014 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2015 }
2016
2017 bool LibraryCallKit::inline_math_negateExactL() {
2018 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2019 }
2020
2021 bool LibraryCallKit::inline_math_multiplyExactI() {
2022 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2023 }
2024
2025 bool LibraryCallKit::inline_math_multiplyExactL() {
2026 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2027 }
2028
2029 bool LibraryCallKit::inline_math_multiplyHigh() {
2030 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2031 return true;
2032 }
2033
2034 Node*
2035 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2036 // These are the candidate return value:
2037 Node* xvalue = x0;
2038 Node* yvalue = y0;
2039
2040 if (xvalue == yvalue) {
2041 return xvalue;
2042 }
2043
2044 bool want_max = (id == vmIntrinsics::_max);
2045
2046 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
2047 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
2048 if (txvalue == NULL || tyvalue == NULL) return top();
2049 // This is not really necessary, but it is consistent with a
2050 // hypothetical MaxINode::Value method:
2051 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
2052
2053 // %%% This folding logic should (ideally) be in a different place.
2054 // Some should be inside IfNode, and there to be a more reliable
2055 // transformation of ?: style patterns into cmoves. We also want
2056 // more powerful optimizations around cmove and min/max.
2057
2058 // Try to find a dominating comparison of these guys.
2059 // It can simplify the index computation for Arrays.copyOf
2060 // and similar uses of System.arraycopy.
2061 // First, compute the normalized version of CmpI(x, y).
2062 int cmp_op = Op_CmpI;
2063 Node* xkey = xvalue;
2064 Node* ykey = yvalue;
2065 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
2066 if (ideal_cmpxy->is_Cmp()) {
2067 // E.g., if we have CmpI(length - offset, count),
2068 // it might idealize to CmpI(length, count + offset)
2069 cmp_op = ideal_cmpxy->Opcode();
2070 xkey = ideal_cmpxy->in(1);
2071 ykey = ideal_cmpxy->in(2);
2072 }
2073
2074 // Start by locating any relevant comparisons.
2075 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
2076 Node* cmpxy = NULL;
2077 Node* cmpyx = NULL;
2078 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
2079 Node* cmp = start_from->fast_out(k);
2080 if (cmp->outcnt() > 0 && // must have prior uses
2081 cmp->in(0) == NULL && // must be context-independent
2082 cmp->Opcode() == cmp_op) { // right kind of compare
2083 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
2084 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
2085 }
2086 }
2087
2088 const int NCMPS = 2;
2089 Node* cmps[NCMPS] = { cmpxy, cmpyx };
2090 int cmpn;
2091 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2092 if (cmps[cmpn] != NULL) break; // find a result
2093 }
2094 if (cmpn < NCMPS) {
2095 // Look for a dominating test that tells us the min and max.
2096 int depth = 0; // Limit search depth for speed
2097 Node* dom = control();
2098 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2099 if (++depth >= 100) break;
2100 Node* ifproj = dom;
2101 if (!ifproj->is_Proj()) continue;
2102 Node* iff = ifproj->in(0);
2103 if (!iff->is_If()) continue;
2104 Node* bol = iff->in(1);
2105 if (!bol->is_Bool()) continue;
2106 Node* cmp = bol->in(1);
2107 if (cmp == NULL) continue;
2108 for (cmpn = 0; cmpn < NCMPS; cmpn++)
2109 if (cmps[cmpn] == cmp) break;
2110 if (cmpn == NCMPS) continue;
2111 BoolTest::mask btest = bol->as_Bool()->_test._test;
2112 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
2113 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2114 // At this point, we know that 'x btest y' is true.
2115 switch (btest) {
2116 case BoolTest::eq:
2117 // They are proven equal, so we can collapse the min/max.
2118 // Either value is the answer. Choose the simpler.
2119 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2120 return yvalue;
2121 return xvalue;
2122 case BoolTest::lt: // x < y
2123 case BoolTest::le: // x <= y
2124 return (want_max ? yvalue : xvalue);
2125 case BoolTest::gt: // x > y
2126 case BoolTest::ge: // x >= y
2127 return (want_max ? xvalue : yvalue);
2128 default:
2129 break;
2130 }
2131 }
2132 }
2133
2134 // We failed to find a dominating test.
2135 // Let's pick a test that might GVN with prior tests.
2136 Node* best_bol = NULL;
2137 BoolTest::mask best_btest = BoolTest::illegal;
2138 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2139 Node* cmp = cmps[cmpn];
2140 if (cmp == NULL) continue;
2141 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2142 Node* bol = cmp->fast_out(j);
2143 if (!bol->is_Bool()) continue;
2144 BoolTest::mask btest = bol->as_Bool()->_test._test;
2145 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2146 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2147 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2148 best_bol = bol->as_Bool();
2149 best_btest = btest;
2150 }
2151 }
2152 }
2153
2154 Node* answer_if_true = NULL;
2155 Node* answer_if_false = NULL;
2156 switch (best_btest) {
2157 default:
2158 if (cmpxy == NULL)
2159 cmpxy = ideal_cmpxy;
2160 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2161 // and fall through:
2162 case BoolTest::lt: // x < y
2163 case BoolTest::le: // x <= y
2164 answer_if_true = (want_max ? yvalue : xvalue);
2165 answer_if_false = (want_max ? xvalue : yvalue);
2166 break;
2167 case BoolTest::gt: // x > y
2168 case BoolTest::ge: // x >= y
2169 answer_if_true = (want_max ? xvalue : yvalue);
2170 answer_if_false = (want_max ? yvalue : xvalue);
2171 break;
2172 }
2173
2174 jint hi, lo;
2175 if (want_max) {
2176 // We can sharpen the minimum.
2177 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2178 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2179 } else {
2180 // We can sharpen the maximum.
2181 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2182 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2183 }
2184
2185 // Use a flow-free graph structure, to avoid creating excess control edges
2186 // which could hinder other optimizations.
2187 // Since Math.min/max is often used with arraycopy, we want
2188 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2189 Node* cmov = CMoveNode::make(NULL, best_bol,
2190 answer_if_false, answer_if_true,
2191 TypeInt::make(lo, hi, widen));
2192
2193 return _gvn.transform(cmov);
2194
2195 /*
2196 // This is not as desirable as it may seem, since Min and Max
2197 // nodes do not have a full set of optimizations.
2198 // And they would interfere, anyway, with 'if' optimizations
2199 // and with CMoveI canonical forms.
2200 switch (id) {
2201 case vmIntrinsics::_min:
2202 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2203 case vmIntrinsics::_max:
2204 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2205 default:
2206 ShouldNotReachHere();
2207 }
2208 */
2209 }
2210
2211 inline int
2212 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2213 const TypePtr* base_type = TypePtr::NULL_PTR;
2214 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2215 if (base_type == NULL) {
2216 // Unknown type.
2217 return Type::AnyPtr;
2218 } else if (base_type == TypePtr::NULL_PTR) {
2219 // Since this is a NULL+long form, we have to switch to a rawptr.
2220 base = _gvn.transform(new CastX2PNode(offset));
2221 offset = MakeConX(0);
2222 return Type::RawPtr;
2223 } else if (base_type->base() == Type::RawPtr) {
2224 return Type::RawPtr;
2225 } else if (base_type->isa_oopptr()) {
2226 // Base is never null => always a heap address.
2227 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2228 return Type::OopPtr;
2229 }
2230 // Offset is small => always a heap address.
2231 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2232 if (offset_type != NULL &&
2233 base_type->offset() == 0 && // (should always be?)
2234 offset_type->_lo >= 0 &&
2235 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2236 return Type::OopPtr;
2237 } else if (type == T_OBJECT) {
2238 // off heap access to an oop doesn't make any sense. Has to be on
2239 // heap.
2240 return Type::OopPtr;
2241 }
2242 // Otherwise, it might either be oop+off or NULL+addr.
2243 return Type::AnyPtr;
2244 } else {
2245 // No information:
2246 return Type::AnyPtr;
2247 }
2248 }
2249
2250 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type, bool can_cast) {
2251 Node* uncasted_base = base;
2252 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2253 if (kind == Type::RawPtr) {
2254 return basic_plus_adr(top(), uncasted_base, offset);
2255 } else if (kind == Type::AnyPtr) {
2256 assert(base == uncasted_base, "unexpected base change");
2257 if (can_cast) {
2258 if (!_gvn.type(base)->speculative_maybe_null() &&
2259 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2260 // According to profiling, this access is always on
2261 // heap. Casting the base to not null and thus avoiding membars
2262 // around the access should allow better optimizations
2263 Node* null_ctl = top();
2264 base = null_check_oop(base, &null_ctl, true, true, true);
2265 assert(null_ctl->is_top(), "no null control here");
2266 return basic_plus_adr(base, offset);
2267 } else if (_gvn.type(base)->speculative_always_null() &&
2268 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2269 // According to profiling, this access is always off
2270 // heap.
2271 base = null_assert(base);
2272 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2273 offset = MakeConX(0);
2274 return basic_plus_adr(top(), raw_base, offset);
2275 }
2276 }
2277 // We don't know if it's an on heap or off heap access. Fall back
2278 // to raw memory access.
2279 base = access_resolve(base, decorators);
2280 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2281 return basic_plus_adr(top(), raw, offset);
2282 } else {
2283 assert(base == uncasted_base, "unexpected base change");
2284 // We know it's an on heap access so base can't be null
2285 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2286 base = must_be_not_null(base, true);
2287 }
2288 return basic_plus_adr(base, offset);
2289 }
2290 }
2291
2292 //--------------------------inline_number_methods-----------------------------
2293 // inline int Integer.numberOfLeadingZeros(int)
2294 // inline int Long.numberOfLeadingZeros(long)
2295 //
2296 // inline int Integer.numberOfTrailingZeros(int)
2297 // inline int Long.numberOfTrailingZeros(long)
2298 //
2299 // inline int Integer.bitCount(int)
2300 // inline int Long.bitCount(long)
2301 //
2302 // inline char Character.reverseBytes(char)
2303 // inline short Short.reverseBytes(short)
2304 // inline int Integer.reverseBytes(int)
2305 // inline long Long.reverseBytes(long)
2306 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2307 Node* arg = argument(0);
2308 Node* n = NULL;
2309 switch (id) {
2310 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2311 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2312 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2313 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2314 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2315 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2316 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2317 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2318 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2319 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2320 default: fatal_unexpected_iid(id); break;
2321 }
2322 set_result(_gvn.transform(n));
2323 return true;
2324 }
2325
2326 //----------------------------inline_unsafe_access----------------------------
2327
2328 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2329 // Attempt to infer a sharper value type from the offset and base type.
2330 ciKlass* sharpened_klass = NULL;
2331
2332 // See if it is an instance field, with an object type.
2333 if (alias_type->field() != NULL) {
2334 if (alias_type->field()->type()->is_klass()) {
2335 sharpened_klass = alias_type->field()->type()->as_klass();
2336 }
2337 }
2338
2339 // See if it is a narrow oop array.
2340 if (adr_type->isa_aryptr()) {
2341 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2342 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2343 if (elem_type != NULL) {
2344 sharpened_klass = elem_type->klass();
2345 }
2346 }
2347 }
2348
2349 // The sharpened class might be unloaded if there is no class loader
2350 // contraint in place.
2351 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2352 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2353
2354 #ifndef PRODUCT
2355 if (C->print_intrinsics() || C->print_inlining()) {
2356 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2357 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2358 }
2359 #endif
2360 // Sharpen the value type.
2361 return tjp;
2362 }
2363 return NULL;
2364 }
2365
2366 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2367 switch (kind) {
2368 case Relaxed:
2369 return MO_UNORDERED;
2370 case Opaque:
2371 return MO_RELAXED;
2372 case Acquire:
2373 return MO_ACQUIRE;
2374 case Release:
2375 return MO_RELEASE;
2376 case Volatile:
2377 return MO_SEQ_CST;
2378 default:
2379 ShouldNotReachHere();
2380 return 0;
2381 }
2382 }
2383
2384 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2385 if (callee()->is_static()) return false; // caller must have the capability!
2386 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2387 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2388 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2389 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2390
2391 if (type == T_OBJECT || type == T_ARRAY) {
2392 decorators |= ON_UNKNOWN_OOP_REF;
2393 }
2394
2395 if (unaligned) {
2396 decorators |= C2_UNALIGNED;
2397 }
2398
2399 #ifndef PRODUCT
2400 {
2401 ResourceMark rm;
2402 // Check the signatures.
2403 ciSignature* sig = callee()->signature();
2404 #ifdef ASSERT
2405 if (!is_store) {
2406 // Object getReference(Object base, int/long offset), etc.
2407 BasicType rtype = sig->return_type()->basic_type();
2408 assert(rtype == type, "getter must return the expected value");
2409 assert(sig->count() == 2, "oop getter has 2 arguments");
2410 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2411 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2412 } else {
2413 // void putReference(Object base, int/long offset, Object x), etc.
2414 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2415 assert(sig->count() == 3, "oop putter has 3 arguments");
2416 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2417 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2418 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2419 assert(vtype == type, "putter must accept the expected value");
2420 }
2421 #endif // ASSERT
2422 }
2423 #endif //PRODUCT
2424
2425 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2426
2427 Node* receiver = argument(0); // type: oop
2428
2429 // Build address expression.
2430 Node* adr;
2431 Node* heap_base_oop = top();
2432 Node* offset = top();
2433 Node* val;
2434
2435 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2436 Node* base = argument(1); // type: oop
2437 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2438 offset = argument(2); // type: long
2439 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2440 // to be plain byte offsets, which are also the same as those accepted
2441 // by oopDesc::field_addr.
2442 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2443 "fieldOffset must be byte-scaled");
2444 // 32-bit machines ignore the high half!
2445 offset = ConvL2X(offset);
2446 adr = make_unsafe_address(base, offset, is_store ? ACCESS_WRITE : ACCESS_READ, type, kind == Relaxed);
2447
2448 if (_gvn.type(base)->isa_ptr() != TypePtr::NULL_PTR) {
2449 heap_base_oop = base;
2450 } else if (type == T_OBJECT) {
2451 return false; // off-heap oop accesses are not supported
2452 }
2453
2454 // Can base be NULL? Otherwise, always on-heap access.
2455 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2456
2457 if (!can_access_non_heap) {
2458 decorators |= IN_HEAP;
2459 }
2460
2461 val = is_store ? argument(4) : NULL;
2462
2463 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2464 if (adr_type == TypePtr::NULL_PTR) {
2465 return false; // off-heap access with zero address
2466 }
2467
2468 // Try to categorize the address.
2469 Compile::AliasType* alias_type = C->alias_type(adr_type);
2470 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2471
2472 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2473 alias_type->adr_type() == TypeAryPtr::RANGE) {
2474 return false; // not supported
2475 }
2476
2477 bool mismatched = false;
2478 BasicType bt = alias_type->basic_type();
2479 if (bt != T_ILLEGAL) {
2480 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2481 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2482 // Alias type doesn't differentiate between byte[] and boolean[]).
2483 // Use address type to get the element type.
2484 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2485 }
2486 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2487 // accessing an array field with getReference is not a mismatch
2488 bt = T_OBJECT;
2489 }
2490 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2491 // Don't intrinsify mismatched object accesses
2492 return false;
2493 }
2494 mismatched = (bt != type);
2495 } else if (alias_type->adr_type()->isa_oopptr()) {
2496 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2497 }
2498
2499 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2500
2501 if (mismatched) {
2502 decorators |= C2_MISMATCHED;
2503 }
2504
2505 // First guess at the value type.
2506 const Type *value_type = Type::get_const_basic_type(type);
2507
2508 // Figure out the memory ordering.
2509 decorators |= mo_decorator_for_access_kind(kind);
2510
2511 if (!is_store && type == T_OBJECT) {
2512 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2513 if (tjp != NULL) {
2514 value_type = tjp;
2515 }
2516 }
2517
2518 receiver = null_check(receiver);
2519 if (stopped()) {
2520 return true;
2521 }
2522 // Heap pointers get a null-check from the interpreter,
2523 // as a courtesy. However, this is not guaranteed by Unsafe,
2524 // and it is not possible to fully distinguish unintended nulls
2525 // from intended ones in this API.
2526
2527 if (!is_store) {
2528 Node* p = NULL;
2529 // Try to constant fold a load from a constant field
2530 ciField* field = alias_type->field();
2531 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2532 // final or stable field
2533 p = make_constant_from_field(field, heap_base_oop);
2534 }
2535
2536 if (p == NULL) { // Could not constant fold the load
2537 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2538 // Normalize the value returned by getBoolean in the following cases
2539 if (type == T_BOOLEAN &&
2540 (mismatched ||
2541 heap_base_oop == top() || // - heap_base_oop is NULL or
2542 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2543 // and the unsafe access is made to large offset
2544 // (i.e., larger than the maximum offset necessary for any
2545 // field access)
2546 ) {
2547 IdealKit ideal = IdealKit(this);
2548 #define __ ideal.
2549 IdealVariable normalized_result(ideal);
2550 __ declarations_done();
2551 __ set(normalized_result, p);
2552 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2553 __ set(normalized_result, ideal.ConI(1));
2554 ideal.end_if();
2555 final_sync(ideal);
2556 p = __ value(normalized_result);
2557 #undef __
2558 }
2559 }
2560 if (type == T_ADDRESS) {
2561 p = gvn().transform(new CastP2XNode(NULL, p));
2562 p = ConvX2UL(p);
2563 }
2564 // The load node has the control of the preceding MemBarCPUOrder. All
2565 // following nodes will have the control of the MemBarCPUOrder inserted at
2566 // the end of this method. So, pushing the load onto the stack at a later
2567 // point is fine.
2568 set_result(p);
2569 } else {
2570 if (bt == T_ADDRESS) {
2571 // Repackage the long as a pointer.
2572 val = ConvL2X(val);
2573 val = gvn().transform(new CastX2PNode(val));
2574 }
2575 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2576 }
2577
2578 return true;
2579 }
2580
2581 //----------------------------inline_unsafe_load_store----------------------------
2582 // This method serves a couple of different customers (depending on LoadStoreKind):
2583 //
2584 // LS_cmp_swap:
2585 //
2586 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2587 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2588 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2589 //
2590 // LS_cmp_swap_weak:
2591 //
2592 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2593 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2594 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2595 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2596 //
2597 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2598 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2599 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2600 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2601 //
2602 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2603 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2604 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2605 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2606 //
2607 // LS_cmp_exchange:
2608 //
2609 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2610 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2611 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2612 //
2613 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2614 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2615 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2616 //
2617 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2618 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2619 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2620 //
2621 // LS_get_add:
2622 //
2623 // int getAndAddInt( Object o, long offset, int delta)
2624 // long getAndAddLong(Object o, long offset, long delta)
2625 //
2626 // LS_get_set:
2627 //
2628 // int getAndSet(Object o, long offset, int newValue)
2629 // long getAndSet(Object o, long offset, long newValue)
2630 // Object getAndSet(Object o, long offset, Object newValue)
2631 //
2632 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2633 // This basic scheme here is the same as inline_unsafe_access, but
2634 // differs in enough details that combining them would make the code
2635 // overly confusing. (This is a true fact! I originally combined
2636 // them, but even I was confused by it!) As much code/comments as
2637 // possible are retained from inline_unsafe_access though to make
2638 // the correspondences clearer. - dl
2639
2640 if (callee()->is_static()) return false; // caller must have the capability!
2641
2642 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2643 decorators |= mo_decorator_for_access_kind(access_kind);
2644
2645 #ifndef PRODUCT
2646 BasicType rtype;
2647 {
2648 ResourceMark rm;
2649 // Check the signatures.
2650 ciSignature* sig = callee()->signature();
2651 rtype = sig->return_type()->basic_type();
2652 switch(kind) {
2653 case LS_get_add:
2654 case LS_get_set: {
2655 // Check the signatures.
2656 #ifdef ASSERT
2657 assert(rtype == type, "get and set must return the expected type");
2658 assert(sig->count() == 3, "get and set has 3 arguments");
2659 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2660 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2661 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2662 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2663 #endif // ASSERT
2664 break;
2665 }
2666 case LS_cmp_swap:
2667 case LS_cmp_swap_weak: {
2668 // Check the signatures.
2669 #ifdef ASSERT
2670 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2671 assert(sig->count() == 4, "CAS has 4 arguments");
2672 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2673 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2674 #endif // ASSERT
2675 break;
2676 }
2677 case LS_cmp_exchange: {
2678 // Check the signatures.
2679 #ifdef ASSERT
2680 assert(rtype == type, "CAS must return the expected type");
2681 assert(sig->count() == 4, "CAS has 4 arguments");
2682 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2683 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2684 #endif // ASSERT
2685 break;
2686 }
2687 default:
2688 ShouldNotReachHere();
2689 }
2690 }
2691 #endif //PRODUCT
2692
2693 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2694
2695 // Get arguments:
2696 Node* receiver = NULL;
2697 Node* base = NULL;
2698 Node* offset = NULL;
2699 Node* oldval = NULL;
2700 Node* newval = NULL;
2701 switch(kind) {
2702 case LS_cmp_swap:
2703 case LS_cmp_swap_weak:
2704 case LS_cmp_exchange: {
2705 const bool two_slot_type = type2size[type] == 2;
2706 receiver = argument(0); // type: oop
2707 base = argument(1); // type: oop
2708 offset = argument(2); // type: long
2709 oldval = argument(4); // type: oop, int, or long
2710 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2711 break;
2712 }
2713 case LS_get_add:
2714 case LS_get_set: {
2715 receiver = argument(0); // type: oop
2716 base = argument(1); // type: oop
2717 offset = argument(2); // type: long
2718 oldval = NULL;
2719 newval = argument(4); // type: oop, int, or long
2720 break;
2721 }
2722 default:
2723 ShouldNotReachHere();
2724 }
2725
2726 // Build field offset expression.
2727 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2728 // to be plain byte offsets, which are also the same as those accepted
2729 // by oopDesc::field_addr.
2730 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2731 // 32-bit machines ignore the high half of long offsets
2732 offset = ConvL2X(offset);
2733 Node* adr = make_unsafe_address(base, offset, ACCESS_WRITE | ACCESS_READ, type, false);
2734 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2735
2736 Compile::AliasType* alias_type = C->alias_type(adr_type);
2737 BasicType bt = alias_type->basic_type();
2738 if (bt != T_ILLEGAL &&
2739 ((bt == T_OBJECT || bt == T_ARRAY) != (type == T_OBJECT))) {
2740 // Don't intrinsify mismatched object accesses.
2741 return false;
2742 }
2743
2744 // For CAS, unlike inline_unsafe_access, there seems no point in
2745 // trying to refine types. Just use the coarse types here.
2746 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2747 const Type *value_type = Type::get_const_basic_type(type);
2748
2749 switch (kind) {
2750 case LS_get_set:
2751 case LS_cmp_exchange: {
2752 if (type == T_OBJECT) {
2753 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2754 if (tjp != NULL) {
2755 value_type = tjp;
2756 }
2757 }
2758 break;
2759 }
2760 case LS_cmp_swap:
2761 case LS_cmp_swap_weak:
2762 case LS_get_add:
2763 break;
2764 default:
2765 ShouldNotReachHere();
2766 }
2767
2768 // Null check receiver.
2769 receiver = null_check(receiver);
2770 if (stopped()) {
2771 return true;
2772 }
2773
2774 int alias_idx = C->get_alias_index(adr_type);
2775
2776 if (type == T_OBJECT || type == T_ARRAY) {
2777 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2778
2779 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2780 // could be delayed during Parse (for example, in adjust_map_after_if()).
2781 // Execute transformation here to avoid barrier generation in such case.
2782 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2783 newval = _gvn.makecon(TypePtr::NULL_PTR);
2784
2785 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2786 // Refine the value to a null constant, when it is known to be null
2787 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2788 }
2789 }
2790
2791 Node* result = NULL;
2792 switch (kind) {
2793 case LS_cmp_exchange: {
2794 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2795 oldval, newval, value_type, type, decorators);
2796 break;
2797 }
2798 case LS_cmp_swap_weak:
2799 decorators |= C2_WEAK_CMPXCHG;
2800 case LS_cmp_swap: {
2801 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2802 oldval, newval, value_type, type, decorators);
2803 break;
2804 }
2805 case LS_get_set: {
2806 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2807 newval, value_type, type, decorators);
2808 break;
2809 }
2810 case LS_get_add: {
2811 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2812 newval, value_type, type, decorators);
2813 break;
2814 }
2815 default:
2816 ShouldNotReachHere();
2817 }
2818
2819 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2820 set_result(result);
2821 return true;
2822 }
2823
2824 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2825 // Regardless of form, don't allow previous ld/st to move down,
2826 // then issue acquire, release, or volatile mem_bar.
2827 insert_mem_bar(Op_MemBarCPUOrder);
2828 switch(id) {
2829 case vmIntrinsics::_loadFence:
2830 insert_mem_bar(Op_LoadFence);
2831 return true;
2832 case vmIntrinsics::_storeFence:
2833 insert_mem_bar(Op_StoreFence);
2834 return true;
2835 case vmIntrinsics::_fullFence:
2836 insert_mem_bar(Op_MemBarVolatile);
2837 return true;
2838 default:
2839 fatal_unexpected_iid(id);
2840 return false;
2841 }
2842 }
2843
2844 bool LibraryCallKit::inline_onspinwait() {
2845 insert_mem_bar(Op_OnSpinWait);
2846 return true;
2847 }
2848
2849 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2850 if (!kls->is_Con()) {
2851 return true;
2852 }
2853 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
2854 if (klsptr == NULL) {
2855 return true;
2856 }
2857 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
2858 // don't need a guard for a klass that is already initialized
2859 return !ik->is_initialized();
2860 }
2861
2862 //----------------------------inline_unsafe_allocate---------------------------
2863 // public native Object Unsafe.allocateInstance(Class<?> cls);
2864 bool LibraryCallKit::inline_unsafe_allocate() {
2865 if (callee()->is_static()) return false; // caller must have the capability!
2866
2867 null_check_receiver(); // null-check, then ignore
2868 Node* cls = null_check(argument(1));
2869 if (stopped()) return true;
2870
2871 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2872 kls = null_check(kls);
2873 if (stopped()) return true; // argument was like int.class
2874
2875 Node* test = NULL;
2876 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2877 // Note: The argument might still be an illegal value like
2878 // Serializable.class or Object[].class. The runtime will handle it.
2879 // But we must make an explicit check for initialization.
2880 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2881 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2882 // can generate code to load it as unsigned byte.
2883 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2884 Node* bits = intcon(InstanceKlass::fully_initialized);
2885 test = _gvn.transform(new SubINode(inst, bits));
2886 // The 'test' is non-zero if we need to take a slow path.
2887 }
2888
2889 Node* obj = new_instance(kls, test);
2890 set_result(obj);
2891 return true;
2892 }
2893
2894 //------------------------inline_native_time_funcs--------------
2895 // inline code for System.currentTimeMillis() and System.nanoTime()
2896 // these have the same type and signature
2897 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2898 const TypeFunc* tf = OptoRuntime::void_long_Type();
2899 const TypePtr* no_memory_effects = NULL;
2900 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2901 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2902 #ifdef ASSERT
2903 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2904 assert(value_top == top(), "second value must be top");
2905 #endif
2906 set_result(value);
2907 return true;
2908 }
2909
2910 #ifdef JFR_HAVE_INTRINSICS
2911
2912 /*
2913 * oop -> myklass
2914 * myklass->trace_id |= USED
2915 * return myklass->trace_id & ~0x3
2916 */
2917 bool LibraryCallKit::inline_native_classID() {
2918 Node* cls = null_check(argument(0), T_OBJECT);
2919 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2920 kls = null_check(kls, T_OBJECT);
2921
2922 ByteSize offset = KLASS_TRACE_ID_OFFSET;
2923 Node* insp = basic_plus_adr(kls, in_bytes(offset));
2924 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
2925
2926 Node* clsused = longcon(0x01l); // set the class bit
2927 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
2928 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
2929 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
2930
2931 #ifdef TRACE_ID_META_BITS
2932 Node* mbits = longcon(~TRACE_ID_META_BITS);
2933 tvalue = _gvn.transform(new AndLNode(tvalue, mbits));
2934 #endif
2935 #ifdef TRACE_ID_SHIFT
2936 Node* cbits = intcon(TRACE_ID_SHIFT);
2937 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits));
2938 #endif
2939
2940 set_result(tvalue);
2941 return true;
2942
2943 }
2944
2945 bool LibraryCallKit::inline_native_getEventWriter() {
2946 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
2947
2948 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
2949 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
2950
2951 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
2952
2953 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
2954 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
2955
2956 IfNode* iff_jobj_null =
2957 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
2958
2959 enum { _normal_path = 1,
2960 _null_path = 2,
2961 PATH_LIMIT };
2962
2963 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
2964 PhiNode* result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
2965
2966 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
2967 result_rgn->init_req(_null_path, jobj_is_null);
2968 result_val->init_req(_null_path, null());
2969
2970 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
2971 set_control(jobj_is_not_null);
2972 Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT,
2973 IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
2974 result_rgn->init_req(_normal_path, control());
2975 result_val->init_req(_normal_path, res);
2976
2977 set_result(result_rgn, result_val);
2978
2979 return true;
2980 }
2981
2982 #endif // JFR_HAVE_INTRINSICS
2983
2984 //------------------------inline_native_currentThread------------------
2985 bool LibraryCallKit::inline_native_currentThread() {
2986 Node* junk = NULL;
2987 set_result(generate_current_thread(junk));
2988 return true;
2989 }
2990
2991 //------------------------inline_native_isInterrupted------------------
2992 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
2993 bool LibraryCallKit::inline_native_isInterrupted() {
2994 // Add a fast path to t.isInterrupted(clear_int):
2995 // (t == Thread.current() &&
2996 // (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
2997 // ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
2998 // So, in the common case that the interrupt bit is false,
2999 // we avoid making a call into the VM. Even if the interrupt bit
3000 // is true, if the clear_int argument is false, we avoid the VM call.
3001 // However, if the receiver is not currentThread, we must call the VM,
3002 // because there must be some locking done around the operation.
3003
3004 // We only go to the fast case code if we pass two guards.
3005 // Paths which do not pass are accumulated in the slow_region.
3006
3007 enum {
3008 no_int_result_path = 1, // t == Thread.current() && !TLS._osthread._interrupted
3009 no_clear_result_path = 2, // t == Thread.current() && TLS._osthread._interrupted && !clear_int
3010 slow_result_path = 3, // slow path: t.isInterrupted(clear_int)
3011 PATH_LIMIT
3012 };
3013
3014 // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
3015 // out of the function.
3016 insert_mem_bar(Op_MemBarCPUOrder);
3017
3018 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3019 PhiNode* result_val = new PhiNode(result_rgn, TypeInt::BOOL);
3020
3021 RegionNode* slow_region = new RegionNode(1);
3022 record_for_igvn(slow_region);
3023
3024 // (a) Receiving thread must be the current thread.
3025 Node* rec_thr = argument(0);
3026 Node* tls_ptr = NULL;
3027 Node* cur_thr = generate_current_thread(tls_ptr);
3028
3029 // Resolve oops to stable for CmpP below.
3030 cur_thr = access_resolve(cur_thr, 0);
3031 rec_thr = access_resolve(rec_thr, 0);
3032
3033 Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
3034 Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
3035
3036 generate_slow_guard(bol_thr, slow_region);
3037
3038 // (b) Interrupt bit on TLS must be false.
3039 Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
3040 Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3041 p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
3042
3043 // Set the control input on the field _interrupted read to prevent it floating up.
3044 Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
3045 Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
3046 Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
3047
3048 IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
3049
3050 // First fast path: if (!TLS._interrupted) return false;
3051 Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
3052 result_rgn->init_req(no_int_result_path, false_bit);
3053 result_val->init_req(no_int_result_path, intcon(0));
3054
3055 // drop through to next case
3056 set_control( _gvn.transform(new IfTrueNode(iff_bit)));
3057
3058 #ifndef _WINDOWS
3059 // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
3060 Node* clr_arg = argument(1);
3061 Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
3062 Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
3063 IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
3064
3065 // Second fast path: ... else if (!clear_int) return true;
3066 Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
3067 result_rgn->init_req(no_clear_result_path, false_arg);
3068 result_val->init_req(no_clear_result_path, intcon(1));
3069
3070 // drop through to next case
3071 set_control( _gvn.transform(new IfTrueNode(iff_arg)));
3072 #else
3073 // To return true on Windows you must read the _interrupted field
3074 // and check the event state i.e. take the slow path.
3075 #endif // _WINDOWS
3076
3077 // (d) Otherwise, go to the slow path.
3078 slow_region->add_req(control());
3079 set_control( _gvn.transform(slow_region));
3080
3081 if (stopped()) {
3082 // There is no slow path.
3083 result_rgn->init_req(slow_result_path, top());
3084 result_val->init_req(slow_result_path, top());
3085 } else {
3086 // non-virtual because it is a private non-static
3087 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
3088
3089 Node* slow_val = set_results_for_java_call(slow_call);
3090 // this->control() comes from set_results_for_java_call
3091
3092 Node* fast_io = slow_call->in(TypeFunc::I_O);
3093 Node* fast_mem = slow_call->in(TypeFunc::Memory);
3094
3095 // These two phis are pre-filled with copies of of the fast IO and Memory
3096 PhiNode* result_mem = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
3097 PhiNode* result_io = PhiNode::make(result_rgn, fast_io, Type::ABIO);
3098
3099 result_rgn->init_req(slow_result_path, control());
3100 result_io ->init_req(slow_result_path, i_o());
3101 result_mem->init_req(slow_result_path, reset_memory());
3102 result_val->init_req(slow_result_path, slow_val);
3103
3104 set_all_memory(_gvn.transform(result_mem));
3105 set_i_o( _gvn.transform(result_io));
3106 }
3107
3108 C->set_has_split_ifs(true); // Has chance for split-if optimization
3109 set_result(result_rgn, result_val);
3110 return true;
3111 }
3112
3113 //---------------------------load_mirror_from_klass----------------------------
3114 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3115 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3116 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3117 Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3118 // mirror = ((OopHandle)mirror)->resolve();
3119 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3120 }
3121
3122 //-----------------------load_klass_from_mirror_common-------------------------
3123 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3124 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3125 // and branch to the given path on the region.
3126 // If never_see_null, take an uncommon trap on null, so we can optimistically
3127 // compile for the non-null case.
3128 // If the region is NULL, force never_see_null = true.
3129 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3130 bool never_see_null,
3131 RegionNode* region,
3132 int null_path,
3133 int offset) {
3134 if (region == NULL) never_see_null = true;
3135 Node* p = basic_plus_adr(mirror, offset);
3136 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3137 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3138 Node* null_ctl = top();
3139 kls = null_check_oop(kls, &null_ctl, never_see_null);
3140 if (region != NULL) {
3141 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3142 region->init_req(null_path, null_ctl);
3143 } else {
3144 assert(null_ctl == top(), "no loose ends");
3145 }
3146 return kls;
3147 }
3148
3149 //--------------------(inline_native_Class_query helpers)---------------------
3150 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3151 // Fall through if (mods & mask) == bits, take the guard otherwise.
3152 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3153 // Branch around if the given klass has the given modifier bit set.
3154 // Like generate_guard, adds a new path onto the region.
3155 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3156 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3157 Node* mask = intcon(modifier_mask);
3158 Node* bits = intcon(modifier_bits);
3159 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3160 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3161 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3162 return generate_fair_guard(bol, region);
3163 }
3164 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3165 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3166 }
3167
3168 //-------------------------inline_native_Class_query-------------------
3169 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3170 const Type* return_type = TypeInt::BOOL;
3171 Node* prim_return_value = top(); // what happens if it's a primitive class?
3172 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3173 bool expect_prim = false; // most of these guys expect to work on refs
3174
3175 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3176
3177 Node* mirror = argument(0);
3178 Node* obj = top();
3179
3180 switch (id) {
3181 case vmIntrinsics::_isInstance:
3182 // nothing is an instance of a primitive type
3183 prim_return_value = intcon(0);
3184 obj = argument(1);
3185 break;
3186 case vmIntrinsics::_getModifiers:
3187 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3188 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3189 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3190 break;
3191 case vmIntrinsics::_isInterface:
3192 prim_return_value = intcon(0);
3193 break;
3194 case vmIntrinsics::_isArray:
3195 prim_return_value = intcon(0);
3196 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3197 break;
3198 case vmIntrinsics::_isPrimitive:
3199 prim_return_value = intcon(1);
3200 expect_prim = true; // obviously
3201 break;
3202 case vmIntrinsics::_getSuperclass:
3203 prim_return_value = null();
3204 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3205 break;
3206 case vmIntrinsics::_getClassAccessFlags:
3207 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3208 return_type = TypeInt::INT; // not bool! 6297094
3209 break;
3210 default:
3211 fatal_unexpected_iid(id);
3212 break;
3213 }
3214
3215 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3216 if (mirror_con == NULL) return false; // cannot happen?
3217
3218 #ifndef PRODUCT
3219 if (C->print_intrinsics() || C->print_inlining()) {
3220 ciType* k = mirror_con->java_mirror_type();
3221 if (k) {
3222 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3223 k->print_name();
3224 tty->cr();
3225 }
3226 }
3227 #endif
3228
3229 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3230 RegionNode* region = new RegionNode(PATH_LIMIT);
3231 record_for_igvn(region);
3232 PhiNode* phi = new PhiNode(region, return_type);
3233
3234 // The mirror will never be null of Reflection.getClassAccessFlags, however
3235 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3236 // if it is. See bug 4774291.
3237
3238 // For Reflection.getClassAccessFlags(), the null check occurs in
3239 // the wrong place; see inline_unsafe_access(), above, for a similar
3240 // situation.
3241 mirror = null_check(mirror);
3242 // If mirror or obj is dead, only null-path is taken.
3243 if (stopped()) return true;
3244
3245 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3246
3247 // Now load the mirror's klass metaobject, and null-check it.
3248 // Side-effects region with the control path if the klass is null.
3249 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3250 // If kls is null, we have a primitive mirror.
3251 phi->init_req(_prim_path, prim_return_value);
3252 if (stopped()) { set_result(region, phi); return true; }
3253 bool safe_for_replace = (region->in(_prim_path) == top());
3254
3255 Node* p; // handy temp
3256 Node* null_ctl;
3257
3258 // Now that we have the non-null klass, we can perform the real query.
3259 // For constant classes, the query will constant-fold in LoadNode::Value.
3260 Node* query_value = top();
3261 switch (id) {
3262 case vmIntrinsics::_isInstance:
3263 // nothing is an instance of a primitive type
3264 query_value = gen_instanceof(obj, kls, safe_for_replace);
3265 break;
3266
3267 case vmIntrinsics::_getModifiers:
3268 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3269 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3270 break;
3271
3272 case vmIntrinsics::_isInterface:
3273 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3274 if (generate_interface_guard(kls, region) != NULL)
3275 // A guard was added. If the guard is taken, it was an interface.
3276 phi->add_req(intcon(1));
3277 // If we fall through, it's a plain class.
3278 query_value = intcon(0);
3279 break;
3280
3281 case vmIntrinsics::_isArray:
3282 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3283 if (generate_array_guard(kls, region) != NULL)
3284 // A guard was added. If the guard is taken, it was an array.
3285 phi->add_req(intcon(1));
3286 // If we fall through, it's a plain class.
3287 query_value = intcon(0);
3288 break;
3289
3290 case vmIntrinsics::_isPrimitive:
3291 query_value = intcon(0); // "normal" path produces false
3292 break;
3293
3294 case vmIntrinsics::_getSuperclass:
3295 // The rules here are somewhat unfortunate, but we can still do better
3296 // with random logic than with a JNI call.
3297 // Interfaces store null or Object as _super, but must report null.
3298 // Arrays store an intermediate super as _super, but must report Object.
3299 // Other types can report the actual _super.
3300 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3301 if (generate_interface_guard(kls, region) != NULL)
3302 // A guard was added. If the guard is taken, it was an interface.
3303 phi->add_req(null());
3304 if (generate_array_guard(kls, region) != NULL)
3305 // A guard was added. If the guard is taken, it was an array.
3306 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3307 // If we fall through, it's a plain class. Get its _super.
3308 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3309 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3310 null_ctl = top();
3311 kls = null_check_oop(kls, &null_ctl);
3312 if (null_ctl != top()) {
3313 // If the guard is taken, Object.superClass is null (both klass and mirror).
3314 region->add_req(null_ctl);
3315 phi ->add_req(null());
3316 }
3317 if (!stopped()) {
3318 query_value = load_mirror_from_klass(kls);
3319 }
3320 break;
3321
3322 case vmIntrinsics::_getClassAccessFlags:
3323 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3324 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3325 break;
3326
3327 default:
3328 fatal_unexpected_iid(id);
3329 break;
3330 }
3331
3332 // Fall-through is the normal case of a query to a real class.
3333 phi->init_req(1, query_value);
3334 region->init_req(1, control());
3335
3336 C->set_has_split_ifs(true); // Has chance for split-if optimization
3337 set_result(region, phi);
3338 return true;
3339 }
3340
3341 //-------------------------inline_Class_cast-------------------
3342 bool LibraryCallKit::inline_Class_cast() {
3343 Node* mirror = argument(0); // Class
3344 Node* obj = argument(1);
3345 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3346 if (mirror_con == NULL) {
3347 return false; // dead path (mirror->is_top()).
3348 }
3349 if (obj == NULL || obj->is_top()) {
3350 return false; // dead path
3351 }
3352 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3353
3354 // First, see if Class.cast() can be folded statically.
3355 // java_mirror_type() returns non-null for compile-time Class constants.
3356 ciType* tm = mirror_con->java_mirror_type();
3357 if (tm != NULL && tm->is_klass() &&
3358 tp != NULL && tp->klass() != NULL) {
3359 if (!tp->klass()->is_loaded()) {
3360 // Don't use intrinsic when class is not loaded.
3361 return false;
3362 } else {
3363 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3364 if (static_res == Compile::SSC_always_true) {
3365 // isInstance() is true - fold the code.
3366 set_result(obj);
3367 return true;
3368 } else if (static_res == Compile::SSC_always_false) {
3369 // Don't use intrinsic, have to throw ClassCastException.
3370 // If the reference is null, the non-intrinsic bytecode will
3371 // be optimized appropriately.
3372 return false;
3373 }
3374 }
3375 }
3376
3377 // Bailout intrinsic and do normal inlining if exception path is frequent.
3378 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3379 return false;
3380 }
3381
3382 // Generate dynamic checks.
3383 // Class.cast() is java implementation of _checkcast bytecode.
3384 // Do checkcast (Parse::do_checkcast()) optimizations here.
3385
3386 mirror = null_check(mirror);
3387 // If mirror is dead, only null-path is taken.
3388 if (stopped()) {
3389 return true;
3390 }
3391
3392 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3393 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3394 RegionNode* region = new RegionNode(PATH_LIMIT);
3395 record_for_igvn(region);
3396
3397 // Now load the mirror's klass metaobject, and null-check it.
3398 // If kls is null, we have a primitive mirror and
3399 // nothing is an instance of a primitive type.
3400 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3401
3402 Node* res = top();
3403 if (!stopped()) {
3404 Node* bad_type_ctrl = top();
3405 // Do checkcast optimizations.
3406 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3407 region->init_req(_bad_type_path, bad_type_ctrl);
3408 }
3409 if (region->in(_prim_path) != top() ||
3410 region->in(_bad_type_path) != top()) {
3411 // Let Interpreter throw ClassCastException.
3412 PreserveJVMState pjvms(this);
3413 set_control(_gvn.transform(region));
3414 uncommon_trap(Deoptimization::Reason_intrinsic,
3415 Deoptimization::Action_maybe_recompile);
3416 }
3417 if (!stopped()) {
3418 set_result(res);
3419 }
3420 return true;
3421 }
3422
3423
3424 //--------------------------inline_native_subtype_check------------------------
3425 // This intrinsic takes the JNI calls out of the heart of
3426 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3427 bool LibraryCallKit::inline_native_subtype_check() {
3428 // Pull both arguments off the stack.
3429 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3430 args[0] = argument(0);
3431 args[1] = argument(1);
3432 Node* klasses[2]; // corresponding Klasses: superk, subk
3433 klasses[0] = klasses[1] = top();
3434
3435 enum {
3436 // A full decision tree on {superc is prim, subc is prim}:
3437 _prim_0_path = 1, // {P,N} => false
3438 // {P,P} & superc!=subc => false
3439 _prim_same_path, // {P,P} & superc==subc => true
3440 _prim_1_path, // {N,P} => false
3441 _ref_subtype_path, // {N,N} & subtype check wins => true
3442 _both_ref_path, // {N,N} & subtype check loses => false
3443 PATH_LIMIT
3444 };
3445
3446 RegionNode* region = new RegionNode(PATH_LIMIT);
3447 Node* phi = new PhiNode(region, TypeInt::BOOL);
3448 record_for_igvn(region);
3449
3450 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3451 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3452 int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3453
3454 // First null-check both mirrors and load each mirror's klass metaobject.
3455 int which_arg;
3456 for (which_arg = 0; which_arg <= 1; which_arg++) {
3457 Node* arg = args[which_arg];
3458 arg = null_check(arg);
3459 if (stopped()) break;
3460 args[which_arg] = arg;
3461
3462 Node* p = basic_plus_adr(arg, class_klass_offset);
3463 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3464 klasses[which_arg] = _gvn.transform(kls);
3465 }
3466
3467 // Resolve oops to stable for CmpP below.
3468 args[0] = access_resolve(args[0], 0);
3469 args[1] = access_resolve(args[1], 0);
3470
3471 // Having loaded both klasses, test each for null.
3472 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3473 for (which_arg = 0; which_arg <= 1; which_arg++) {
3474 Node* kls = klasses[which_arg];
3475 Node* null_ctl = top();
3476 kls = null_check_oop(kls, &null_ctl, never_see_null);
3477 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3478 region->init_req(prim_path, null_ctl);
3479 if (stopped()) break;
3480 klasses[which_arg] = kls;
3481 }
3482
3483 if (!stopped()) {
3484 // now we have two reference types, in klasses[0..1]
3485 Node* subk = klasses[1]; // the argument to isAssignableFrom
3486 Node* superk = klasses[0]; // the receiver
3487 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3488 // now we have a successful reference subtype check
3489 region->set_req(_ref_subtype_path, control());
3490 }
3491
3492 // If both operands are primitive (both klasses null), then
3493 // we must return true when they are identical primitives.
3494 // It is convenient to test this after the first null klass check.
3495 set_control(region->in(_prim_0_path)); // go back to first null check
3496 if (!stopped()) {
3497 // Since superc is primitive, make a guard for the superc==subc case.
3498 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3499 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3500 generate_guard(bol_eq, region, PROB_FAIR);
3501 if (region->req() == PATH_LIMIT+1) {
3502 // A guard was added. If the added guard is taken, superc==subc.
3503 region->swap_edges(PATH_LIMIT, _prim_same_path);
3504 region->del_req(PATH_LIMIT);
3505 }
3506 region->set_req(_prim_0_path, control()); // Not equal after all.
3507 }
3508
3509 // these are the only paths that produce 'true':
3510 phi->set_req(_prim_same_path, intcon(1));
3511 phi->set_req(_ref_subtype_path, intcon(1));
3512
3513 // pull together the cases:
3514 assert(region->req() == PATH_LIMIT, "sane region");
3515 for (uint i = 1; i < region->req(); i++) {
3516 Node* ctl = region->in(i);
3517 if (ctl == NULL || ctl == top()) {
3518 region->set_req(i, top());
3519 phi ->set_req(i, top());
3520 } else if (phi->in(i) == NULL) {
3521 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3522 }
3523 }
3524
3525 set_control(_gvn.transform(region));
3526 set_result(_gvn.transform(phi));
3527 return true;
3528 }
3529
3530 //---------------------generate_array_guard_common------------------------
3531 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3532 bool obj_array, bool not_array) {
3533
3534 if (stopped()) {
3535 return NULL;
3536 }
3537
3538 // If obj_array/non_array==false/false:
3539 // Branch around if the given klass is in fact an array (either obj or prim).
3540 // If obj_array/non_array==false/true:
3541 // Branch around if the given klass is not an array klass of any kind.
3542 // If obj_array/non_array==true/true:
3543 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3544 // If obj_array/non_array==true/false:
3545 // Branch around if the kls is an oop array (Object[] or subtype)
3546 //
3547 // Like generate_guard, adds a new path onto the region.
3548 jint layout_con = 0;
3549 Node* layout_val = get_layout_helper(kls, layout_con);
3550 if (layout_val == NULL) {
3551 bool query = (obj_array
3552 ? Klass::layout_helper_is_objArray(layout_con)
3553 : Klass::layout_helper_is_array(layout_con));
3554 if (query == not_array) {
3555 return NULL; // never a branch
3556 } else { // always a branch
3557 Node* always_branch = control();
3558 if (region != NULL)
3559 region->add_req(always_branch);
3560 set_control(top());
3561 return always_branch;
3562 }
3563 }
3564 // Now test the correct condition.
3565 jint nval = (obj_array
3566 ? (jint)(Klass::_lh_array_tag_type_value
3567 << Klass::_lh_array_tag_shift)
3568 : Klass::_lh_neutral_value);
3569 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3570 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3571 // invert the test if we are looking for a non-array
3572 if (not_array) btest = BoolTest(btest).negate();
3573 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3574 return generate_fair_guard(bol, region);
3575 }
3576
3577
3578 //-----------------------inline_native_newArray--------------------------
3579 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3580 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3581 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3582 Node* mirror;
3583 Node* count_val;
3584 if (uninitialized) {
3585 mirror = argument(1);
3586 count_val = argument(2);
3587 } else {
3588 mirror = argument(0);
3589 count_val = argument(1);
3590 }
3591
3592 mirror = null_check(mirror);
3593 // If mirror or obj is dead, only null-path is taken.
3594 if (stopped()) return true;
3595
3596 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3597 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3598 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3599 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3600 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3601
3602 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3603 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3604 result_reg, _slow_path);
3605 Node* normal_ctl = control();
3606 Node* no_array_ctl = result_reg->in(_slow_path);
3607
3608 // Generate code for the slow case. We make a call to newArray().
3609 set_control(no_array_ctl);
3610 if (!stopped()) {
3611 // Either the input type is void.class, or else the
3612 // array klass has not yet been cached. Either the
3613 // ensuing call will throw an exception, or else it
3614 // will cache the array klass for next time.
3615 PreserveJVMState pjvms(this);
3616 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3617 Node* slow_result = set_results_for_java_call(slow_call);
3618 // this->control() comes from set_results_for_java_call
3619 result_reg->set_req(_slow_path, control());
3620 result_val->set_req(_slow_path, slow_result);
3621 result_io ->set_req(_slow_path, i_o());
3622 result_mem->set_req(_slow_path, reset_memory());
3623 }
3624
3625 set_control(normal_ctl);
3626 if (!stopped()) {
3627 // Normal case: The array type has been cached in the java.lang.Class.
3628 // The following call works fine even if the array type is polymorphic.
3629 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3630 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3631 result_reg->init_req(_normal_path, control());
3632 result_val->init_req(_normal_path, obj);
3633 result_io ->init_req(_normal_path, i_o());
3634 result_mem->init_req(_normal_path, reset_memory());
3635
3636 if (uninitialized) {
3637 // Mark the allocation so that zeroing is skipped
3638 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3639 alloc->maybe_set_complete(&_gvn);
3640 }
3641 }
3642
3643 // Return the combined state.
3644 set_i_o( _gvn.transform(result_io) );
3645 set_all_memory( _gvn.transform(result_mem));
3646
3647 C->set_has_split_ifs(true); // Has chance for split-if optimization
3648 set_result(result_reg, result_val);
3649 return true;
3650 }
3651
3652 //----------------------inline_native_getLength--------------------------
3653 // public static native int java.lang.reflect.Array.getLength(Object array);
3654 bool LibraryCallKit::inline_native_getLength() {
3655 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3656
3657 Node* array = null_check(argument(0));
3658 // If array is dead, only null-path is taken.
3659 if (stopped()) return true;
3660
3661 // Deoptimize if it is a non-array.
3662 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3663
3664 if (non_array != NULL) {
3665 PreserveJVMState pjvms(this);
3666 set_control(non_array);
3667 uncommon_trap(Deoptimization::Reason_intrinsic,
3668 Deoptimization::Action_maybe_recompile);
3669 }
3670
3671 // If control is dead, only non-array-path is taken.
3672 if (stopped()) return true;
3673
3674 // The works fine even if the array type is polymorphic.
3675 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3676 Node* result = load_array_length(array);
3677
3678 C->set_has_split_ifs(true); // Has chance for split-if optimization
3679 set_result(result);
3680 return true;
3681 }
3682
3683 //------------------------inline_array_copyOf----------------------------
3684 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3685 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3686 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3687 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3688
3689 // Get the arguments.
3690 Node* original = argument(0);
3691 Node* start = is_copyOfRange? argument(1): intcon(0);
3692 Node* end = is_copyOfRange? argument(2): argument(1);
3693 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3694
3695 Node* newcopy = NULL;
3696
3697 // Set the original stack and the reexecute bit for the interpreter to reexecute
3698 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3699 { PreserveReexecuteState preexecs(this);
3700 jvms()->set_should_reexecute(true);
3701
3702 array_type_mirror = null_check(array_type_mirror);
3703 original = null_check(original);
3704
3705 // Check if a null path was taken unconditionally.
3706 if (stopped()) return true;
3707
3708 Node* orig_length = load_array_length(original);
3709
3710 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3711 klass_node = null_check(klass_node);
3712
3713 RegionNode* bailout = new RegionNode(1);
3714 record_for_igvn(bailout);
3715
3716 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3717 // Bail out if that is so.
3718 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3719 if (not_objArray != NULL) {
3720 // Improve the klass node's type from the new optimistic assumption:
3721 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3722 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3723 Node* cast = new CastPPNode(klass_node, akls);
3724 cast->init_req(0, control());
3725 klass_node = _gvn.transform(cast);
3726 }
3727
3728 // Bail out if either start or end is negative.
3729 generate_negative_guard(start, bailout, &start);
3730 generate_negative_guard(end, bailout, &end);
3731
3732 Node* length = end;
3733 if (_gvn.type(start) != TypeInt::ZERO) {
3734 length = _gvn.transform(new SubINode(end, start));
3735 }
3736
3737 // Bail out if length is negative.
3738 // Without this the new_array would throw
3739 // NegativeArraySizeException but IllegalArgumentException is what
3740 // should be thrown
3741 generate_negative_guard(length, bailout, &length);
3742
3743 if (bailout->req() > 1) {
3744 PreserveJVMState pjvms(this);
3745 set_control(_gvn.transform(bailout));
3746 uncommon_trap(Deoptimization::Reason_intrinsic,
3747 Deoptimization::Action_maybe_recompile);
3748 }
3749
3750 if (!stopped()) {
3751 // How many elements will we copy from the original?
3752 // The answer is MinI(orig_length - start, length).
3753 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3754 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3755
3756 original = access_resolve(original, ACCESS_READ);
3757
3758 // Generate a direct call to the right arraycopy function(s).
3759 // We know the copy is disjoint but we might not know if the
3760 // oop stores need checking.
3761 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3762 // This will fail a store-check if x contains any non-nulls.
3763
3764 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3765 // loads/stores but it is legal only if we're sure the
3766 // Arrays.copyOf would succeed. So we need all input arguments
3767 // to the copyOf to be validated, including that the copy to the
3768 // new array won't trigger an ArrayStoreException. That subtype
3769 // check can be optimized if we know something on the type of
3770 // the input array from type speculation.
3771 if (_gvn.type(klass_node)->singleton()) {
3772 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3773 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3774
3775 int test = C->static_subtype_check(superk, subk);
3776 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3777 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3778 if (t_original->speculative_type() != NULL) {
3779 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3780 }
3781 }
3782 }
3783
3784 bool validated = false;
3785 // Reason_class_check rather than Reason_intrinsic because we
3786 // want to intrinsify even if this traps.
3787 if (!too_many_traps(Deoptimization::Reason_class_check)) {
3788 Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
3789 klass_node);
3790
3791 if (not_subtype_ctrl != top()) {
3792 PreserveJVMState pjvms(this);
3793 set_control(not_subtype_ctrl);
3794 uncommon_trap(Deoptimization::Reason_class_check,
3795 Deoptimization::Action_make_not_entrant);
3796 assert(stopped(), "Should be stopped");
3797 }
3798 validated = true;
3799 }
3800
3801 if (!stopped()) {
3802 newcopy = new_array(klass_node, length, 0); // no arguments to push
3803
3804 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
3805 load_object_klass(original), klass_node);
3806 if (!is_copyOfRange) {
3807 ac->set_copyof(validated);
3808 } else {
3809 ac->set_copyofrange(validated);
3810 }
3811 Node* n = _gvn.transform(ac);
3812 if (n == ac) {
3813 ac->connect_outputs(this);
3814 } else {
3815 assert(validated, "shouldn't transform if all arguments not validated");
3816 set_all_memory(n);
3817 }
3818 }
3819 }
3820 } // original reexecute is set back here
3821
3822 C->set_has_split_ifs(true); // Has chance for split-if optimization
3823 if (!stopped()) {
3824 set_result(newcopy);
3825 }
3826 return true;
3827 }
3828
3829
3830 //----------------------generate_virtual_guard---------------------------
3831 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
3832 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
3833 RegionNode* slow_region) {
3834 ciMethod* method = callee();
3835 int vtable_index = method->vtable_index();
3836 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3837 "bad index %d", vtable_index);
3838 // Get the Method* out of the appropriate vtable entry.
3839 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
3840 vtable_index*vtableEntry::size_in_bytes() +
3841 vtableEntry::method_offset_in_bytes();
3842 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
3843 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3844
3845 // Compare the target method with the expected method (e.g., Object.hashCode).
3846 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
3847
3848 Node* native_call = makecon(native_call_addr);
3849 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
3850 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
3851
3852 return generate_slow_guard(test_native, slow_region);
3853 }
3854
3855 //-----------------------generate_method_call----------------------------
3856 // Use generate_method_call to make a slow-call to the real
3857 // method if the fast path fails. An alternative would be to
3858 // use a stub like OptoRuntime::slow_arraycopy_Java.
3859 // This only works for expanding the current library call,
3860 // not another intrinsic. (E.g., don't use this for making an
3861 // arraycopy call inside of the copyOf intrinsic.)
3862 CallJavaNode*
3863 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
3864 // When compiling the intrinsic method itself, do not use this technique.
3865 guarantee(callee() != C->method(), "cannot make slow-call to self");
3866
3867 ciMethod* method = callee();
3868 // ensure the JVMS we have will be correct for this call
3869 guarantee(method_id == method->intrinsic_id(), "must match");
3870
3871 const TypeFunc* tf = TypeFunc::make(method);
3872 CallJavaNode* slow_call;
3873 if (is_static) {
3874 assert(!is_virtual, "");
3875 slow_call = new CallStaticJavaNode(C, tf,
3876 SharedRuntime::get_resolve_static_call_stub(),
3877 method, bci());
3878 } else if (is_virtual) {
3879 null_check_receiver();
3880 int vtable_index = Method::invalid_vtable_index;
3881 if (UseInlineCaches) {
3882 // Suppress the vtable call
3883 } else {
3884 // hashCode and clone are not a miranda methods,
3885 // so the vtable index is fixed.
3886 // No need to use the linkResolver to get it.
3887 vtable_index = method->vtable_index();
3888 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3889 "bad index %d", vtable_index);
3890 }
3891 slow_call = new CallDynamicJavaNode(tf,
3892 SharedRuntime::get_resolve_virtual_call_stub(),
3893 method, vtable_index, bci());
3894 } else { // neither virtual nor static: opt_virtual
3895 null_check_receiver();
3896 slow_call = new CallStaticJavaNode(C, tf,
3897 SharedRuntime::get_resolve_opt_virtual_call_stub(),
3898 method, bci());
3899 slow_call->set_optimized_virtual(true);
3900 }
3901 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
3902 // To be able to issue a direct call (optimized virtual or virtual)
3903 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
3904 // about the method being invoked should be attached to the call site to
3905 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
3906 slow_call->set_override_symbolic_info(true);
3907 }
3908 set_arguments_for_java_call(slow_call);
3909 set_edges_for_java_call(slow_call);
3910 return slow_call;
3911 }
3912
3913
3914 /**
3915 * Build special case code for calls to hashCode on an object. This call may
3916 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
3917 * slightly different code.
3918 */
3919 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
3920 assert(is_static == callee()->is_static(), "correct intrinsic selection");
3921 assert(!(is_virtual && is_static), "either virtual, special, or static");
3922
3923 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
3924
3925 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3926 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
3927 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3928 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3929 Node* obj = NULL;
3930 if (!is_static) {
3931 // Check for hashing null object
3932 obj = null_check_receiver();
3933 if (stopped()) return true; // unconditionally null
3934 result_reg->init_req(_null_path, top());
3935 result_val->init_req(_null_path, top());
3936 } else {
3937 // Do a null check, and return zero if null.
3938 // System.identityHashCode(null) == 0
3939 obj = argument(0);
3940 Node* null_ctl = top();
3941 obj = null_check_oop(obj, &null_ctl);
3942 result_reg->init_req(_null_path, null_ctl);
3943 result_val->init_req(_null_path, _gvn.intcon(0));
3944 }
3945
3946 // Unconditionally null? Then return right away.
3947 if (stopped()) {
3948 set_control( result_reg->in(_null_path));
3949 if (!stopped())
3950 set_result(result_val->in(_null_path));
3951 return true;
3952 }
3953
3954 // We only go to the fast case code if we pass a number of guards. The
3955 // paths which do not pass are accumulated in the slow_region.
3956 RegionNode* slow_region = new RegionNode(1);
3957 record_for_igvn(slow_region);
3958
3959 // If this is a virtual call, we generate a funny guard. We pull out
3960 // the vtable entry corresponding to hashCode() from the target object.
3961 // If the target method which we are calling happens to be the native
3962 // Object hashCode() method, we pass the guard. We do not need this
3963 // guard for non-virtual calls -- the caller is known to be the native
3964 // Object hashCode().
3965 if (is_virtual) {
3966 // After null check, get the object's klass.
3967 Node* obj_klass = load_object_klass(obj);
3968 generate_virtual_guard(obj_klass, slow_region);
3969 }
3970
3971 // Get the header out of the object, use LoadMarkNode when available
3972 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
3973 // The control of the load must be NULL. Otherwise, the load can move before
3974 // the null check after castPP removal.
3975 Node* no_ctrl = NULL;
3976 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
3977
3978 // Test the header to see if it is unlocked.
3979 Node *lock_mask = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
3980 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
3981 Node *unlocked_val = _gvn.MakeConX(markOopDesc::unlocked_value);
3982 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
3983 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
3984
3985 generate_slow_guard(test_unlocked, slow_region);
3986
3987 // Get the hash value and check to see that it has been properly assigned.
3988 // We depend on hash_mask being at most 32 bits and avoid the use of
3989 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
3990 // vm: see markOop.hpp.
3991 Node *hash_mask = _gvn.intcon(markOopDesc::hash_mask);
3992 Node *hash_shift = _gvn.intcon(markOopDesc::hash_shift);
3993 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
3994 // This hack lets the hash bits live anywhere in the mark object now, as long
3995 // as the shift drops the relevant bits into the low 32 bits. Note that
3996 // Java spec says that HashCode is an int so there's no point in capturing
3997 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
3998 hshifted_header = ConvX2I(hshifted_header);
3999 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4000
4001 Node *no_hash_val = _gvn.intcon(markOopDesc::no_hash);
4002 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4003 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4004
4005 generate_slow_guard(test_assigned, slow_region);
4006
4007 Node* init_mem = reset_memory();
4008 // fill in the rest of the null path:
4009 result_io ->init_req(_null_path, i_o());
4010 result_mem->init_req(_null_path, init_mem);
4011
4012 result_val->init_req(_fast_path, hash_val);
4013 result_reg->init_req(_fast_path, control());
4014 result_io ->init_req(_fast_path, i_o());
4015 result_mem->init_req(_fast_path, init_mem);
4016
4017 // Generate code for the slow case. We make a call to hashCode().
4018 set_control(_gvn.transform(slow_region));
4019 if (!stopped()) {
4020 // No need for PreserveJVMState, because we're using up the present state.
4021 set_all_memory(init_mem);
4022 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4023 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4024 Node* slow_result = set_results_for_java_call(slow_call);
4025 // this->control() comes from set_results_for_java_call
4026 result_reg->init_req(_slow_path, control());
4027 result_val->init_req(_slow_path, slow_result);
4028 result_io ->set_req(_slow_path, i_o());
4029 result_mem ->set_req(_slow_path, reset_memory());
4030 }
4031
4032 // Return the combined state.
4033 set_i_o( _gvn.transform(result_io) );
4034 set_all_memory( _gvn.transform(result_mem));
4035
4036 set_result(result_reg, result_val);
4037 return true;
4038 }
4039
4040 //---------------------------inline_native_getClass----------------------------
4041 // public final native Class<?> java.lang.Object.getClass();
4042 //
4043 // Build special case code for calls to getClass on an object.
4044 bool LibraryCallKit::inline_native_getClass() {
4045 Node* obj = null_check_receiver();
4046 if (stopped()) return true;
4047 set_result(load_mirror_from_klass(load_object_klass(obj)));
4048 return true;
4049 }
4050
4051 //-----------------inline_native_Reflection_getCallerClass---------------------
4052 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4053 //
4054 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4055 //
4056 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4057 // in that it must skip particular security frames and checks for
4058 // caller sensitive methods.
4059 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4060 #ifndef PRODUCT
4061 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4062 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4063 }
4064 #endif
4065
4066 if (!jvms()->has_method()) {
4067 #ifndef PRODUCT
4068 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4069 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4070 }
4071 #endif
4072 return false;
4073 }
4074
4075 // Walk back up the JVM state to find the caller at the required
4076 // depth.
4077 JVMState* caller_jvms = jvms();
4078
4079 // Cf. JVM_GetCallerClass
4080 // NOTE: Start the loop at depth 1 because the current JVM state does
4081 // not include the Reflection.getCallerClass() frame.
4082 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4083 ciMethod* m = caller_jvms->method();
4084 switch (n) {
4085 case 0:
4086 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4087 break;
4088 case 1:
4089 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4090 if (!m->caller_sensitive()) {
4091 #ifndef PRODUCT
4092 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4093 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4094 }
4095 #endif
4096 return false; // bail-out; let JVM_GetCallerClass do the work
4097 }
4098 break;
4099 default:
4100 if (!m->is_ignored_by_security_stack_walk()) {
4101 // We have reached the desired frame; return the holder class.
4102 // Acquire method holder as java.lang.Class and push as constant.
4103 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4104 ciInstance* caller_mirror = caller_klass->java_mirror();
4105 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4106
4107 #ifndef PRODUCT
4108 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4109 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());
4110 tty->print_cr(" JVM state at this point:");
4111 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4112 ciMethod* m = jvms()->of_depth(i)->method();
4113 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4114 }
4115 }
4116 #endif
4117 return true;
4118 }
4119 break;
4120 }
4121 }
4122
4123 #ifndef PRODUCT
4124 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4125 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4126 tty->print_cr(" JVM state at this point:");
4127 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4128 ciMethod* m = jvms()->of_depth(i)->method();
4129 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4130 }
4131 }
4132 #endif
4133
4134 return false; // bail-out; let JVM_GetCallerClass do the work
4135 }
4136
4137 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4138 Node* arg = argument(0);
4139 Node* result = NULL;
4140
4141 switch (id) {
4142 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4143 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4144 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4145 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4146
4147 case vmIntrinsics::_doubleToLongBits: {
4148 // two paths (plus control) merge in a wood
4149 RegionNode *r = new RegionNode(3);
4150 Node *phi = new PhiNode(r, TypeLong::LONG);
4151
4152 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4153 // Build the boolean node
4154 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4155
4156 // Branch either way.
4157 // NaN case is less traveled, which makes all the difference.
4158 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4159 Node *opt_isnan = _gvn.transform(ifisnan);
4160 assert( opt_isnan->is_If(), "Expect an IfNode");
4161 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4162 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4163
4164 set_control(iftrue);
4165
4166 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4167 Node *slow_result = longcon(nan_bits); // return NaN
4168 phi->init_req(1, _gvn.transform( slow_result ));
4169 r->init_req(1, iftrue);
4170
4171 // Else fall through
4172 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4173 set_control(iffalse);
4174
4175 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4176 r->init_req(2, iffalse);
4177
4178 // Post merge
4179 set_control(_gvn.transform(r));
4180 record_for_igvn(r);
4181
4182 C->set_has_split_ifs(true); // Has chance for split-if optimization
4183 result = phi;
4184 assert(result->bottom_type()->isa_long(), "must be");
4185 break;
4186 }
4187
4188 case vmIntrinsics::_floatToIntBits: {
4189 // two paths (plus control) merge in a wood
4190 RegionNode *r = new RegionNode(3);
4191 Node *phi = new PhiNode(r, TypeInt::INT);
4192
4193 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4194 // Build the boolean node
4195 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4196
4197 // Branch either way.
4198 // NaN case is less traveled, which makes all the difference.
4199 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4200 Node *opt_isnan = _gvn.transform(ifisnan);
4201 assert( opt_isnan->is_If(), "Expect an IfNode");
4202 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4203 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4204
4205 set_control(iftrue);
4206
4207 static const jint nan_bits = 0x7fc00000;
4208 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4209 phi->init_req(1, _gvn.transform( slow_result ));
4210 r->init_req(1, iftrue);
4211
4212 // Else fall through
4213 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4214 set_control(iffalse);
4215
4216 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4217 r->init_req(2, iffalse);
4218
4219 // Post merge
4220 set_control(_gvn.transform(r));
4221 record_for_igvn(r);
4222
4223 C->set_has_split_ifs(true); // Has chance for split-if optimization
4224 result = phi;
4225 assert(result->bottom_type()->isa_int(), "must be");
4226 break;
4227 }
4228
4229 default:
4230 fatal_unexpected_iid(id);
4231 break;
4232 }
4233 set_result(_gvn.transform(result));
4234 return true;
4235 }
4236
4237 //----------------------inline_unsafe_copyMemory-------------------------
4238 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4239 bool LibraryCallKit::inline_unsafe_copyMemory() {
4240 if (callee()->is_static()) return false; // caller must have the capability!
4241 null_check_receiver(); // null-check receiver
4242 if (stopped()) return true;
4243
4244 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4245
4246 Node* src_ptr = argument(1); // type: oop
4247 Node* src_off = ConvL2X(argument(2)); // type: long
4248 Node* dst_ptr = argument(4); // type: oop
4249 Node* dst_off = ConvL2X(argument(5)); // type: long
4250 Node* size = ConvL2X(argument(7)); // type: long
4251
4252 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4253 "fieldOffset must be byte-scaled");
4254
4255 src_ptr = access_resolve(src_ptr, ACCESS_READ);
4256 dst_ptr = access_resolve(dst_ptr, ACCESS_WRITE);
4257 Node* src = make_unsafe_address(src_ptr, src_off, ACCESS_READ);
4258 Node* dst = make_unsafe_address(dst_ptr, dst_off, ACCESS_WRITE);
4259
4260 // Conservatively insert a memory barrier on all memory slices.
4261 // Do not let writes of the copy source or destination float below the copy.
4262 insert_mem_bar(Op_MemBarCPUOrder);
4263
4264 Node* thread = _gvn.transform(new ThreadLocalNode());
4265 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4266 BasicType doing_unsafe_access_bt = T_BYTE;
4267 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4268
4269 // update volatile field
4270 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4271
4272 // Call it. Note that the length argument is not scaled.
4273 make_runtime_call(RC_LEAF|RC_NO_FP,
4274 OptoRuntime::fast_arraycopy_Type(),
4275 StubRoutines::unsafe_arraycopy(),
4276 "unsafe_arraycopy",
4277 TypeRawPtr::BOTTOM,
4278 src, dst, size XTOP);
4279
4280 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4281
4282 // Do not let reads of the copy destination float above the copy.
4283 insert_mem_bar(Op_MemBarCPUOrder);
4284
4285 return true;
4286 }
4287
4288 //------------------------clone_coping-----------------------------------
4289 // Helper function for inline_native_clone.
4290 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4291 assert(obj_size != NULL, "");
4292 Node* raw_obj = alloc_obj->in(1);
4293 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4294
4295 AllocateNode* alloc = NULL;
4296 if (ReduceBulkZeroing) {
4297 // We will be completely responsible for initializing this object -
4298 // mark Initialize node as complete.
4299 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4300 // The object was just allocated - there should be no any stores!
4301 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4302 // Mark as complete_with_arraycopy so that on AllocateNode
4303 // expansion, we know this AllocateNode is initialized by an array
4304 // copy and a StoreStore barrier exists after the array copy.
4305 alloc->initialization()->set_complete_with_arraycopy();
4306 }
4307
4308 // Copy the fastest available way.
4309 // TODO: generate fields copies for small objects instead.
4310 Node* size = _gvn.transform(obj_size);
4311
4312 access_clone(obj, alloc_obj, size, is_array);
4313
4314 // Do not let reads from the cloned object float above the arraycopy.
4315 if (alloc != NULL) {
4316 // Do not let stores that initialize this object be reordered with
4317 // a subsequent store that would make this object accessible by
4318 // other threads.
4319 // Record what AllocateNode this StoreStore protects so that
4320 // escape analysis can go from the MemBarStoreStoreNode to the
4321 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4322 // based on the escape status of the AllocateNode.
4323 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4324 } else {
4325 insert_mem_bar(Op_MemBarCPUOrder);
4326 }
4327 }
4328
4329 //------------------------inline_native_clone----------------------------
4330 // protected native Object java.lang.Object.clone();
4331 //
4332 // Here are the simple edge cases:
4333 // null receiver => normal trap
4334 // virtual and clone was overridden => slow path to out-of-line clone
4335 // not cloneable or finalizer => slow path to out-of-line Object.clone
4336 //
4337 // The general case has two steps, allocation and copying.
4338 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4339 //
4340 // Copying also has two cases, oop arrays and everything else.
4341 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4342 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4343 //
4344 // These steps fold up nicely if and when the cloned object's klass
4345 // can be sharply typed as an object array, a type array, or an instance.
4346 //
4347 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4348 PhiNode* result_val;
4349
4350 // Set the reexecute bit for the interpreter to reexecute
4351 // the bytecode that invokes Object.clone if deoptimization happens.
4352 { PreserveReexecuteState preexecs(this);
4353 jvms()->set_should_reexecute(true);
4354
4355 Node* obj = null_check_receiver();
4356 if (stopped()) return true;
4357
4358 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4359
4360 // If we are going to clone an instance, we need its exact type to
4361 // know the number and types of fields to convert the clone to
4362 // loads/stores. Maybe a speculative type can help us.
4363 if (!obj_type->klass_is_exact() &&
4364 obj_type->speculative_type() != NULL &&
4365 obj_type->speculative_type()->is_instance_klass()) {
4366 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4367 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4368 !spec_ik->has_injected_fields()) {
4369 ciKlass* k = obj_type->klass();
4370 if (!k->is_instance_klass() ||
4371 k->as_instance_klass()->is_interface() ||
4372 k->as_instance_klass()->has_subklass()) {
4373 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4374 }
4375 }
4376 }
4377
4378 Node* obj_klass = load_object_klass(obj);
4379 const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
4380 const TypeOopPtr* toop = ((tklass != NULL)
4381 ? tklass->as_instance_type()
4382 : TypeInstPtr::NOTNULL);
4383
4384 // Conservatively insert a memory barrier on all memory slices.
4385 // Do not let writes into the original float below the clone.
4386 insert_mem_bar(Op_MemBarCPUOrder);
4387
4388 // paths into result_reg:
4389 enum {
4390 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4391 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4392 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4393 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4394 PATH_LIMIT
4395 };
4396 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4397 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4398 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4399 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4400 record_for_igvn(result_reg);
4401
4402 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4403 if (array_ctl != NULL) {
4404 // It's an array.
4405 PreserveJVMState pjvms(this);
4406 set_control(array_ctl);
4407 Node* obj_length = load_array_length(obj);
4408 Node* obj_size = NULL;
4409 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size); // no arguments to push
4410
4411 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4412 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, BarrierSetC2::Parsing)) {
4413 // If it is an oop array, it requires very special treatment,
4414 // because gc barriers are required when accessing the array.
4415 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4416 if (is_obja != NULL) {
4417 PreserveJVMState pjvms2(this);
4418 set_control(is_obja);
4419 obj = access_resolve(obj, ACCESS_READ);
4420 // Generate a direct call to the right arraycopy function(s).
4421 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4422 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false);
4423 ac->set_cloneoop();
4424 Node* n = _gvn.transform(ac);
4425 assert(n == ac, "cannot disappear");
4426 ac->connect_outputs(this);
4427
4428 result_reg->init_req(_objArray_path, control());
4429 result_val->init_req(_objArray_path, alloc_obj);
4430 result_i_o ->set_req(_objArray_path, i_o());
4431 result_mem ->set_req(_objArray_path, reset_memory());
4432 }
4433 }
4434 // Otherwise, there are no barriers to worry about.
4435 // (We can dispense with card marks if we know the allocation
4436 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4437 // causes the non-eden paths to take compensating steps to
4438 // simulate a fresh allocation, so that no further
4439 // card marks are required in compiled code to initialize
4440 // the object.)
4441
4442 if (!stopped()) {
4443 copy_to_clone(obj, alloc_obj, obj_size, true);
4444
4445 // Present the results of the copy.
4446 result_reg->init_req(_array_path, control());
4447 result_val->init_req(_array_path, alloc_obj);
4448 result_i_o ->set_req(_array_path, i_o());
4449 result_mem ->set_req(_array_path, reset_memory());
4450 }
4451 }
4452
4453 // We only go to the instance fast case code if we pass a number of guards.
4454 // The paths which do not pass are accumulated in the slow_region.
4455 RegionNode* slow_region = new RegionNode(1);
4456 record_for_igvn(slow_region);
4457 if (!stopped()) {
4458 // It's an instance (we did array above). Make the slow-path tests.
4459 // If this is a virtual call, we generate a funny guard. We grab
4460 // the vtable entry corresponding to clone() from the target object.
4461 // If the target method which we are calling happens to be the
4462 // Object clone() method, we pass the guard. We do not need this
4463 // guard for non-virtual calls; the caller is known to be the native
4464 // Object clone().
4465 if (is_virtual) {
4466 generate_virtual_guard(obj_klass, slow_region);
4467 }
4468
4469 // The object must be easily cloneable and must not have a finalizer.
4470 // Both of these conditions may be checked in a single test.
4471 // We could optimize the test further, but we don't care.
4472 generate_access_flags_guard(obj_klass,
4473 // Test both conditions:
4474 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4475 // Must be cloneable but not finalizer:
4476 JVM_ACC_IS_CLONEABLE_FAST,
4477 slow_region);
4478 }
4479
4480 if (!stopped()) {
4481 // It's an instance, and it passed the slow-path tests.
4482 PreserveJVMState pjvms(this);
4483 Node* obj_size = NULL;
4484 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4485 // is reexecuted if deoptimization occurs and there could be problems when merging
4486 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4487 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4488
4489 copy_to_clone(obj, alloc_obj, obj_size, false);
4490
4491 // Present the results of the slow call.
4492 result_reg->init_req(_instance_path, control());
4493 result_val->init_req(_instance_path, alloc_obj);
4494 result_i_o ->set_req(_instance_path, i_o());
4495 result_mem ->set_req(_instance_path, reset_memory());
4496 }
4497
4498 // Generate code for the slow case. We make a call to clone().
4499 set_control(_gvn.transform(slow_region));
4500 if (!stopped()) {
4501 PreserveJVMState pjvms(this);
4502 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4503 // We need to deoptimize on exception (see comment above)
4504 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4505 // this->control() comes from set_results_for_java_call
4506 result_reg->init_req(_slow_path, control());
4507 result_val->init_req(_slow_path, slow_result);
4508 result_i_o ->set_req(_slow_path, i_o());
4509 result_mem ->set_req(_slow_path, reset_memory());
4510 }
4511
4512 // Return the combined state.
4513 set_control( _gvn.transform(result_reg));
4514 set_i_o( _gvn.transform(result_i_o));
4515 set_all_memory( _gvn.transform(result_mem));
4516 } // original reexecute is set back here
4517
4518 set_result(_gvn.transform(result_val));
4519 return true;
4520 }
4521
4522 // If we have a tightly coupled allocation, the arraycopy may take care
4523 // of the array initialization. If one of the guards we insert between
4524 // the allocation and the arraycopy causes a deoptimization, an
4525 // unitialized array will escape the compiled method. To prevent that
4526 // we set the JVM state for uncommon traps between the allocation and
4527 // the arraycopy to the state before the allocation so, in case of
4528 // deoptimization, we'll reexecute the allocation and the
4529 // initialization.
4530 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4531 if (alloc != NULL) {
4532 ciMethod* trap_method = alloc->jvms()->method();
4533 int trap_bci = alloc->jvms()->bci();
4534
4535 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4536 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4537 // Make sure there's no store between the allocation and the
4538 // arraycopy otherwise visible side effects could be rexecuted
4539 // in case of deoptimization and cause incorrect execution.
4540 bool no_interfering_store = true;
4541 Node* mem = alloc->in(TypeFunc::Memory);
4542 if (mem->is_MergeMem()) {
4543 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4544 Node* n = mms.memory();
4545 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4546 assert(n->is_Store(), "what else?");
4547 no_interfering_store = false;
4548 break;
4549 }
4550 }
4551 } else {
4552 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4553 Node* n = mms.memory();
4554 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4555 assert(n->is_Store(), "what else?");
4556 no_interfering_store = false;
4557 break;
4558 }
4559 }
4560 }
4561
4562 if (no_interfering_store) {
4563 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4564 uint size = alloc->req();
4565 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4566 old_jvms->set_map(sfpt);
4567 for (uint i = 0; i < size; i++) {
4568 sfpt->init_req(i, alloc->in(i));
4569 }
4570 // re-push array length for deoptimization
4571 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4572 old_jvms->set_sp(old_jvms->sp()+1);
4573 old_jvms->set_monoff(old_jvms->monoff()+1);
4574 old_jvms->set_scloff(old_jvms->scloff()+1);
4575 old_jvms->set_endoff(old_jvms->endoff()+1);
4576 old_jvms->set_should_reexecute(true);
4577
4578 sfpt->set_i_o(map()->i_o());
4579 sfpt->set_memory(map()->memory());
4580 sfpt->set_control(map()->control());
4581
4582 JVMState* saved_jvms = jvms();
4583 saved_reexecute_sp = _reexecute_sp;
4584
4585 set_jvms(sfpt->jvms());
4586 _reexecute_sp = jvms()->sp();
4587
4588 return saved_jvms;
4589 }
4590 }
4591 }
4592 return NULL;
4593 }
4594
4595 // In case of a deoptimization, we restart execution at the
4596 // allocation, allocating a new array. We would leave an uninitialized
4597 // array in the heap that GCs wouldn't expect. Move the allocation
4598 // after the traps so we don't allocate the array if we
4599 // deoptimize. This is possible because tightly_coupled_allocation()
4600 // guarantees there's no observer of the allocated array at this point
4601 // and the control flow is simple enough.
4602 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4603 int saved_reexecute_sp, uint new_idx) {
4604 if (saved_jvms != NULL && !stopped()) {
4605 assert(alloc != NULL, "only with a tightly coupled allocation");
4606 // restore JVM state to the state at the arraycopy
4607 saved_jvms->map()->set_control(map()->control());
4608 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4609 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4610 // If we've improved the types of some nodes (null check) while
4611 // emitting the guards, propagate them to the current state
4612 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4613 set_jvms(saved_jvms);
4614 _reexecute_sp = saved_reexecute_sp;
4615
4616 // Remove the allocation from above the guards
4617 CallProjections callprojs;
4618 alloc->extract_projections(&callprojs, true);
4619 InitializeNode* init = alloc->initialization();
4620 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4621 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4622 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4623 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4624
4625 // move the allocation here (after the guards)
4626 _gvn.hash_delete(alloc);
4627 alloc->set_req(TypeFunc::Control, control());
4628 alloc->set_req(TypeFunc::I_O, i_o());
4629 Node *mem = reset_memory();
4630 set_all_memory(mem);
4631 alloc->set_req(TypeFunc::Memory, mem);
4632 set_control(init->proj_out_or_null(TypeFunc::Control));
4633 set_i_o(callprojs.fallthrough_ioproj);
4634
4635 // Update memory as done in GraphKit::set_output_for_allocation()
4636 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4637 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4638 if (ary_type->isa_aryptr() && length_type != NULL) {
4639 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4640 }
4641 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4642 int elemidx = C->get_alias_index(telemref);
4643 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
4644 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
4645
4646 Node* allocx = _gvn.transform(alloc);
4647 assert(allocx == alloc, "where has the allocation gone?");
4648 assert(dest->is_CheckCastPP(), "not an allocation result?");
4649
4650 _gvn.hash_delete(dest);
4651 dest->set_req(0, control());
4652 Node* destx = _gvn.transform(dest);
4653 assert(destx == dest, "where has the allocation result gone?");
4654 }
4655 }
4656
4657
4658 //------------------------------inline_arraycopy-----------------------
4659 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4660 // Object dest, int destPos,
4661 // int length);
4662 bool LibraryCallKit::inline_arraycopy() {
4663 // Get the arguments.
4664 Node* src = argument(0); // type: oop
4665 Node* src_offset = argument(1); // type: int
4666 Node* dest = argument(2); // type: oop
4667 Node* dest_offset = argument(3); // type: int
4668 Node* length = argument(4); // type: int
4669
4670 uint new_idx = C->unique();
4671
4672 // Check for allocation before we add nodes that would confuse
4673 // tightly_coupled_allocation()
4674 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4675
4676 int saved_reexecute_sp = -1;
4677 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4678 // See arraycopy_restore_alloc_state() comment
4679 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4680 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4681 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
4682 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4683
4684 // The following tests must be performed
4685 // (1) src and dest are arrays.
4686 // (2) src and dest arrays must have elements of the same BasicType
4687 // (3) src and dest must not be null.
4688 // (4) src_offset must not be negative.
4689 // (5) dest_offset must not be negative.
4690 // (6) length must not be negative.
4691 // (7) src_offset + length must not exceed length of src.
4692 // (8) dest_offset + length must not exceed length of dest.
4693 // (9) each element of an oop array must be assignable
4694
4695 // (3) src and dest must not be null.
4696 // always do this here because we need the JVM state for uncommon traps
4697 Node* null_ctl = top();
4698 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4699 assert(null_ctl->is_top(), "no null control here");
4700 dest = null_check(dest, T_ARRAY);
4701
4702 if (!can_emit_guards) {
4703 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4704 // guards but the arraycopy node could still take advantage of a
4705 // tightly allocated allocation. tightly_coupled_allocation() is
4706 // called again to make sure it takes the null check above into
4707 // account: the null check is mandatory and if it caused an
4708 // uncommon trap to be emitted then the allocation can't be
4709 // considered tightly coupled in this context.
4710 alloc = tightly_coupled_allocation(dest, NULL);
4711 }
4712
4713 bool validated = false;
4714
4715 const Type* src_type = _gvn.type(src);
4716 const Type* dest_type = _gvn.type(dest);
4717 const TypeAryPtr* top_src = src_type->isa_aryptr();
4718 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4719
4720 // Do we have the type of src?
4721 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4722 // Do we have the type of dest?
4723 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4724 // Is the type for src from speculation?
4725 bool src_spec = false;
4726 // Is the type for dest from speculation?
4727 bool dest_spec = false;
4728
4729 if ((!has_src || !has_dest) && can_emit_guards) {
4730 // We don't have sufficient type information, let's see if
4731 // speculative types can help. We need to have types for both src
4732 // and dest so that it pays off.
4733
4734 // Do we already have or could we have type information for src
4735 bool could_have_src = has_src;
4736 // Do we already have or could we have type information for dest
4737 bool could_have_dest = has_dest;
4738
4739 ciKlass* src_k = NULL;
4740 if (!has_src) {
4741 src_k = src_type->speculative_type_not_null();
4742 if (src_k != NULL && src_k->is_array_klass()) {
4743 could_have_src = true;
4744 }
4745 }
4746
4747 ciKlass* dest_k = NULL;
4748 if (!has_dest) {
4749 dest_k = dest_type->speculative_type_not_null();
4750 if (dest_k != NULL && dest_k->is_array_klass()) {
4751 could_have_dest = true;
4752 }
4753 }
4754
4755 if (could_have_src && could_have_dest) {
4756 // This is going to pay off so emit the required guards
4757 if (!has_src) {
4758 src = maybe_cast_profiled_obj(src, src_k, true);
4759 src_type = _gvn.type(src);
4760 top_src = src_type->isa_aryptr();
4761 has_src = (top_src != NULL && top_src->klass() != NULL);
4762 src_spec = true;
4763 }
4764 if (!has_dest) {
4765 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4766 dest_type = _gvn.type(dest);
4767 top_dest = dest_type->isa_aryptr();
4768 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4769 dest_spec = true;
4770 }
4771 }
4772 }
4773
4774 if (has_src && has_dest && can_emit_guards) {
4775 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
4776 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
4777 if (src_elem == T_ARRAY) src_elem = T_OBJECT;
4778 if (dest_elem == T_ARRAY) dest_elem = T_OBJECT;
4779
4780 if (src_elem == dest_elem && src_elem == T_OBJECT) {
4781 // If both arrays are object arrays then having the exact types
4782 // for both will remove the need for a subtype check at runtime
4783 // before the call and may make it possible to pick a faster copy
4784 // routine (without a subtype check on every element)
4785 // Do we have the exact type of src?
4786 bool could_have_src = src_spec;
4787 // Do we have the exact type of dest?
4788 bool could_have_dest = dest_spec;
4789 ciKlass* src_k = top_src->klass();
4790 ciKlass* dest_k = top_dest->klass();
4791 if (!src_spec) {
4792 src_k = src_type->speculative_type_not_null();
4793 if (src_k != NULL && src_k->is_array_klass()) {
4794 could_have_src = true;
4795 }
4796 }
4797 if (!dest_spec) {
4798 dest_k = dest_type->speculative_type_not_null();
4799 if (dest_k != NULL && dest_k->is_array_klass()) {
4800 could_have_dest = true;
4801 }
4802 }
4803 if (could_have_src && could_have_dest) {
4804 // If we can have both exact types, emit the missing guards
4805 if (could_have_src && !src_spec) {
4806 src = maybe_cast_profiled_obj(src, src_k, true);
4807 }
4808 if (could_have_dest && !dest_spec) {
4809 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4810 }
4811 }
4812 }
4813 }
4814
4815 ciMethod* trap_method = method();
4816 int trap_bci = bci();
4817 if (saved_jvms != NULL) {
4818 trap_method = alloc->jvms()->method();
4819 trap_bci = alloc->jvms()->bci();
4820 }
4821
4822 bool negative_length_guard_generated = false;
4823
4824 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4825 can_emit_guards &&
4826 !src->is_top() && !dest->is_top()) {
4827 // validate arguments: enables transformation the ArrayCopyNode
4828 validated = true;
4829
4830 RegionNode* slow_region = new RegionNode(1);
4831 record_for_igvn(slow_region);
4832
4833 // (1) src and dest are arrays.
4834 generate_non_array_guard(load_object_klass(src), slow_region);
4835 generate_non_array_guard(load_object_klass(dest), slow_region);
4836
4837 // (2) src and dest arrays must have elements of the same BasicType
4838 // done at macro expansion or at Ideal transformation time
4839
4840 // (4) src_offset must not be negative.
4841 generate_negative_guard(src_offset, slow_region);
4842
4843 // (5) dest_offset must not be negative.
4844 generate_negative_guard(dest_offset, slow_region);
4845
4846 // (7) src_offset + length must not exceed length of src.
4847 generate_limit_guard(src_offset, length,
4848 load_array_length(src),
4849 slow_region);
4850
4851 // (8) dest_offset + length must not exceed length of dest.
4852 generate_limit_guard(dest_offset, length,
4853 load_array_length(dest),
4854 slow_region);
4855
4856 // (6) length must not be negative.
4857 // This is also checked in generate_arraycopy() during macro expansion, but
4858 // we also have to check it here for the case where the ArrayCopyNode will
4859 // be eliminated by Escape Analysis.
4860 if (EliminateAllocations) {
4861 generate_negative_guard(length, slow_region);
4862 negative_length_guard_generated = true;
4863 }
4864
4865 // (9) each element of an oop array must be assignable
4866 Node* src_klass = load_object_klass(src);
4867 Node* dest_klass = load_object_klass(dest);
4868 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
4869
4870 if (not_subtype_ctrl != top()) {
4871 PreserveJVMState pjvms(this);
4872 set_control(not_subtype_ctrl);
4873 uncommon_trap(Deoptimization::Reason_intrinsic,
4874 Deoptimization::Action_make_not_entrant);
4875 assert(stopped(), "Should be stopped");
4876 }
4877 {
4878 PreserveJVMState pjvms(this);
4879 set_control(_gvn.transform(slow_region));
4880 uncommon_trap(Deoptimization::Reason_intrinsic,
4881 Deoptimization::Action_make_not_entrant);
4882 assert(stopped(), "Should be stopped");
4883 }
4884
4885 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
4886 const Type *toop = TypeOopPtr::make_from_klass(dest_klass_t->klass());
4887 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
4888 }
4889
4890 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
4891
4892 if (stopped()) {
4893 return true;
4894 }
4895
4896 Node* new_src = access_resolve(src, ACCESS_READ);
4897 Node* new_dest = access_resolve(dest, ACCESS_WRITE);
4898
4899 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, new_src, src_offset, new_dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
4900 // Create LoadRange and LoadKlass nodes for use during macro expansion here
4901 // so the compiler has a chance to eliminate them: during macro expansion,
4902 // we have to set their control (CastPP nodes are eliminated).
4903 load_object_klass(src), load_object_klass(dest),
4904 load_array_length(src), load_array_length(dest));
4905
4906 ac->set_arraycopy(validated);
4907
4908 Node* n = _gvn.transform(ac);
4909 if (n == ac) {
4910 ac->connect_outputs(this);
4911 } else {
4912 assert(validated, "shouldn't transform if all arguments not validated");
4913 set_all_memory(n);
4914 }
4915 clear_upper_avx();
4916
4917
4918 return true;
4919 }
4920
4921
4922 // Helper function which determines if an arraycopy immediately follows
4923 // an allocation, with no intervening tests or other escapes for the object.
4924 AllocateArrayNode*
4925 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
4926 RegionNode* slow_region) {
4927 if (stopped()) return NULL; // no fast path
4928 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
4929
4930 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
4931 if (alloc == NULL) return NULL;
4932
4933 Node* rawmem = memory(Compile::AliasIdxRaw);
4934 // Is the allocation's memory state untouched?
4935 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
4936 // Bail out if there have been raw-memory effects since the allocation.
4937 // (Example: There might have been a call or safepoint.)
4938 return NULL;
4939 }
4940 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
4941 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
4942 return NULL;
4943 }
4944
4945 // There must be no unexpected observers of this allocation.
4946 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
4947 Node* obs = ptr->fast_out(i);
4948 if (obs != this->map()) {
4949 return NULL;
4950 }
4951 }
4952
4953 // This arraycopy must unconditionally follow the allocation of the ptr.
4954 Node* alloc_ctl = ptr->in(0);
4955 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
4956
4957 Node* ctl = control();
4958 while (ctl != alloc_ctl) {
4959 // There may be guards which feed into the slow_region.
4960 // Any other control flow means that we might not get a chance
4961 // to finish initializing the allocated object.
4962 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
4963 IfNode* iff = ctl->in(0)->as_If();
4964 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
4965 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
4966 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
4967 ctl = iff->in(0); // This test feeds the known slow_region.
4968 continue;
4969 }
4970 // One more try: Various low-level checks bottom out in
4971 // uncommon traps. If the debug-info of the trap omits
4972 // any reference to the allocation, as we've already
4973 // observed, then there can be no objection to the trap.
4974 bool found_trap = false;
4975 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
4976 Node* obs = not_ctl->fast_out(j);
4977 if (obs->in(0) == not_ctl && obs->is_Call() &&
4978 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
4979 found_trap = true; break;
4980 }
4981 }
4982 if (found_trap) {
4983 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
4984 continue;
4985 }
4986 }
4987 return NULL;
4988 }
4989
4990 // If we get this far, we have an allocation which immediately
4991 // precedes the arraycopy, and we can take over zeroing the new object.
4992 // The arraycopy will finish the initialization, and provide
4993 // a new control state to which we will anchor the destination pointer.
4994
4995 return alloc;
4996 }
4997
4998 //-------------inline_encodeISOArray-----------------------------------
4999 // encode char[] to byte[] in ISO_8859_1
5000 bool LibraryCallKit::inline_encodeISOArray() {
5001 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5002 // no receiver since it is static method
5003 Node *src = argument(0);
5004 Node *src_offset = argument(1);
5005 Node *dst = argument(2);
5006 Node *dst_offset = argument(3);
5007 Node *length = argument(4);
5008
5009 src = must_be_not_null(src, true);
5010 dst = must_be_not_null(dst, true);
5011
5012 src = access_resolve(src, ACCESS_READ);
5013 dst = access_resolve(dst, ACCESS_WRITE);
5014
5015 const Type* src_type = src->Value(&_gvn);
5016 const Type* dst_type = dst->Value(&_gvn);
5017 const TypeAryPtr* top_src = src_type->isa_aryptr();
5018 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5019 if (top_src == NULL || top_src->klass() == NULL ||
5020 top_dest == NULL || top_dest->klass() == NULL) {
5021 // failed array check
5022 return false;
5023 }
5024
5025 // Figure out the size and type of the elements we will be copying.
5026 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5027 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5028 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5029 return false;
5030 }
5031
5032 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5033 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5034 // 'src_start' points to src array + scaled offset
5035 // 'dst_start' points to dst array + scaled offset
5036
5037 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5038 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
5039 enc = _gvn.transform(enc);
5040 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5041 set_memory(res_mem, mtype);
5042 set_result(enc);
5043 clear_upper_avx();
5044
5045 return true;
5046 }
5047
5048 //-------------inline_multiplyToLen-----------------------------------
5049 bool LibraryCallKit::inline_multiplyToLen() {
5050 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5051
5052 address stubAddr = StubRoutines::multiplyToLen();
5053 if (stubAddr == NULL) {
5054 return false; // Intrinsic's stub is not implemented on this platform
5055 }
5056 const char* stubName = "multiplyToLen";
5057
5058 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5059
5060 // no receiver because it is a static method
5061 Node* x = argument(0);
5062 Node* xlen = argument(1);
5063 Node* y = argument(2);
5064 Node* ylen = argument(3);
5065 Node* z = argument(4);
5066
5067 x = must_be_not_null(x, true);
5068 y = must_be_not_null(y, true);
5069
5070 x = access_resolve(x, ACCESS_READ);
5071 y = access_resolve(y, ACCESS_READ);
5072 z = access_resolve(z, ACCESS_WRITE);
5073
5074 const Type* x_type = x->Value(&_gvn);
5075 const Type* y_type = y->Value(&_gvn);
5076 const TypeAryPtr* top_x = x_type->isa_aryptr();
5077 const TypeAryPtr* top_y = y_type->isa_aryptr();
5078 if (top_x == NULL || top_x->klass() == NULL ||
5079 top_y == NULL || top_y->klass() == NULL) {
5080 // failed array check
5081 return false;
5082 }
5083
5084 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5085 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5086 if (x_elem != T_INT || y_elem != T_INT) {
5087 return false;
5088 }
5089
5090 // Set the original stack and the reexecute bit for the interpreter to reexecute
5091 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5092 // on the return from z array allocation in runtime.
5093 { PreserveReexecuteState preexecs(this);
5094 jvms()->set_should_reexecute(true);
5095
5096 Node* x_start = array_element_address(x, intcon(0), x_elem);
5097 Node* y_start = array_element_address(y, intcon(0), y_elem);
5098 // 'x_start' points to x array + scaled xlen
5099 // 'y_start' points to y array + scaled ylen
5100
5101 // Allocate the result array
5102 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5103 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5104 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5105
5106 IdealKit ideal(this);
5107
5108 #define __ ideal.
5109 Node* one = __ ConI(1);
5110 Node* zero = __ ConI(0);
5111 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5112 __ set(need_alloc, zero);
5113 __ set(z_alloc, z);
5114 __ if_then(z, BoolTest::eq, null()); {
5115 __ increment (need_alloc, one);
5116 } __ else_(); {
5117 // Update graphKit memory and control from IdealKit.
5118 sync_kit(ideal);
5119 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5120 cast->init_req(0, control());
5121 _gvn.set_type(cast, cast->bottom_type());
5122 C->record_for_igvn(cast);
5123
5124 Node* zlen_arg = load_array_length(cast);
5125 // Update IdealKit memory and control from graphKit.
5126 __ sync_kit(this);
5127 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5128 __ increment (need_alloc, one);
5129 } __ end_if();
5130 } __ end_if();
5131
5132 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5133 // Update graphKit memory and control from IdealKit.
5134 sync_kit(ideal);
5135 Node * narr = new_array(klass_node, zlen, 1);
5136 // Update IdealKit memory and control from graphKit.
5137 __ sync_kit(this);
5138 __ set(z_alloc, narr);
5139 } __ end_if();
5140
5141 sync_kit(ideal);
5142 z = __ value(z_alloc);
5143 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5144 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5145 // Final sync IdealKit and GraphKit.
5146 final_sync(ideal);
5147 #undef __
5148
5149 Node* z_start = array_element_address(z, intcon(0), T_INT);
5150
5151 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5152 OptoRuntime::multiplyToLen_Type(),
5153 stubAddr, stubName, TypePtr::BOTTOM,
5154 x_start, xlen, y_start, ylen, z_start, zlen);
5155 } // original reexecute is set back here
5156
5157 C->set_has_split_ifs(true); // Has chance for split-if optimization
5158 set_result(z);
5159 return true;
5160 }
5161
5162 //-------------inline_squareToLen------------------------------------
5163 bool LibraryCallKit::inline_squareToLen() {
5164 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5165
5166 address stubAddr = StubRoutines::squareToLen();
5167 if (stubAddr == NULL) {
5168 return false; // Intrinsic's stub is not implemented on this platform
5169 }
5170 const char* stubName = "squareToLen";
5171
5172 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5173
5174 Node* x = argument(0);
5175 Node* len = argument(1);
5176 Node* z = argument(2);
5177 Node* zlen = argument(3);
5178
5179 x = must_be_not_null(x, true);
5180 z = must_be_not_null(z, true);
5181
5182 x = access_resolve(x, ACCESS_READ);
5183 z = access_resolve(z, ACCESS_WRITE);
5184
5185 const Type* x_type = x->Value(&_gvn);
5186 const Type* z_type = z->Value(&_gvn);
5187 const TypeAryPtr* top_x = x_type->isa_aryptr();
5188 const TypeAryPtr* top_z = z_type->isa_aryptr();
5189 if (top_x == NULL || top_x->klass() == NULL ||
5190 top_z == NULL || top_z->klass() == NULL) {
5191 // failed array check
5192 return false;
5193 }
5194
5195 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5196 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5197 if (x_elem != T_INT || z_elem != T_INT) {
5198 return false;
5199 }
5200
5201
5202 Node* x_start = array_element_address(x, intcon(0), x_elem);
5203 Node* z_start = array_element_address(z, intcon(0), z_elem);
5204
5205 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5206 OptoRuntime::squareToLen_Type(),
5207 stubAddr, stubName, TypePtr::BOTTOM,
5208 x_start, len, z_start, zlen);
5209
5210 set_result(z);
5211 return true;
5212 }
5213
5214 //-------------inline_mulAdd------------------------------------------
5215 bool LibraryCallKit::inline_mulAdd() {
5216 assert(UseMulAddIntrinsic, "not implemented on this platform");
5217
5218 address stubAddr = StubRoutines::mulAdd();
5219 if (stubAddr == NULL) {
5220 return false; // Intrinsic's stub is not implemented on this platform
5221 }
5222 const char* stubName = "mulAdd";
5223
5224 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5225
5226 Node* out = argument(0);
5227 Node* in = argument(1);
5228 Node* offset = argument(2);
5229 Node* len = argument(3);
5230 Node* k = argument(4);
5231
5232 out = must_be_not_null(out, true);
5233
5234 in = access_resolve(in, ACCESS_READ);
5235 out = access_resolve(out, ACCESS_WRITE);
5236
5237 const Type* out_type = out->Value(&_gvn);
5238 const Type* in_type = in->Value(&_gvn);
5239 const TypeAryPtr* top_out = out_type->isa_aryptr();
5240 const TypeAryPtr* top_in = in_type->isa_aryptr();
5241 if (top_out == NULL || top_out->klass() == NULL ||
5242 top_in == NULL || top_in->klass() == NULL) {
5243 // failed array check
5244 return false;
5245 }
5246
5247 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5248 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5249 if (out_elem != T_INT || in_elem != T_INT) {
5250 return false;
5251 }
5252
5253 Node* outlen = load_array_length(out);
5254 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5255 Node* out_start = array_element_address(out, intcon(0), out_elem);
5256 Node* in_start = array_element_address(in, intcon(0), in_elem);
5257
5258 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5259 OptoRuntime::mulAdd_Type(),
5260 stubAddr, stubName, TypePtr::BOTTOM,
5261 out_start,in_start, new_offset, len, k);
5262 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5263 set_result(result);
5264 return true;
5265 }
5266
5267 //-------------inline_montgomeryMultiply-----------------------------------
5268 bool LibraryCallKit::inline_montgomeryMultiply() {
5269 address stubAddr = StubRoutines::montgomeryMultiply();
5270 if (stubAddr == NULL) {
5271 return false; // Intrinsic's stub is not implemented on this platform
5272 }
5273
5274 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5275 const char* stubName = "montgomery_multiply";
5276
5277 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5278
5279 Node* a = argument(0);
5280 Node* b = argument(1);
5281 Node* n = argument(2);
5282 Node* len = argument(3);
5283 Node* inv = argument(4);
5284 Node* m = argument(6);
5285
5286 a = access_resolve(a, ACCESS_READ);
5287 b = access_resolve(b, ACCESS_READ);
5288 n = access_resolve(n, ACCESS_READ);
5289 m = access_resolve(m, ACCESS_WRITE);
5290
5291 const Type* a_type = a->Value(&_gvn);
5292 const TypeAryPtr* top_a = a_type->isa_aryptr();
5293 const Type* b_type = b->Value(&_gvn);
5294 const TypeAryPtr* top_b = b_type->isa_aryptr();
5295 const Type* n_type = a->Value(&_gvn);
5296 const TypeAryPtr* top_n = n_type->isa_aryptr();
5297 const Type* m_type = a->Value(&_gvn);
5298 const TypeAryPtr* top_m = m_type->isa_aryptr();
5299 if (top_a == NULL || top_a->klass() == NULL ||
5300 top_b == NULL || top_b->klass() == NULL ||
5301 top_n == NULL || top_n->klass() == NULL ||
5302 top_m == NULL || top_m->klass() == NULL) {
5303 // failed array check
5304 return false;
5305 }
5306
5307 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5308 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5309 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5310 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5311 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5312 return false;
5313 }
5314
5315 // Make the call
5316 {
5317 Node* a_start = array_element_address(a, intcon(0), a_elem);
5318 Node* b_start = array_element_address(b, intcon(0), b_elem);
5319 Node* n_start = array_element_address(n, intcon(0), n_elem);
5320 Node* m_start = array_element_address(m, intcon(0), m_elem);
5321
5322 Node* call = make_runtime_call(RC_LEAF,
5323 OptoRuntime::montgomeryMultiply_Type(),
5324 stubAddr, stubName, TypePtr::BOTTOM,
5325 a_start, b_start, n_start, len, inv, top(),
5326 m_start);
5327 set_result(m);
5328 }
5329
5330 return true;
5331 }
5332
5333 bool LibraryCallKit::inline_montgomerySquare() {
5334 address stubAddr = StubRoutines::montgomerySquare();
5335 if (stubAddr == NULL) {
5336 return false; // Intrinsic's stub is not implemented on this platform
5337 }
5338
5339 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5340 const char* stubName = "montgomery_square";
5341
5342 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5343
5344 Node* a = argument(0);
5345 Node* n = argument(1);
5346 Node* len = argument(2);
5347 Node* inv = argument(3);
5348 Node* m = argument(5);
5349
5350 a = access_resolve(a, ACCESS_READ);
5351 n = access_resolve(n, ACCESS_READ);
5352 m = access_resolve(m, ACCESS_WRITE);
5353
5354 const Type* a_type = a->Value(&_gvn);
5355 const TypeAryPtr* top_a = a_type->isa_aryptr();
5356 const Type* n_type = a->Value(&_gvn);
5357 const TypeAryPtr* top_n = n_type->isa_aryptr();
5358 const Type* m_type = a->Value(&_gvn);
5359 const TypeAryPtr* top_m = m_type->isa_aryptr();
5360 if (top_a == NULL || top_a->klass() == NULL ||
5361 top_n == NULL || top_n->klass() == NULL ||
5362 top_m == NULL || top_m->klass() == NULL) {
5363 // failed array check
5364 return false;
5365 }
5366
5367 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5368 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5369 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5370 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5371 return false;
5372 }
5373
5374 // Make the call
5375 {
5376 Node* a_start = array_element_address(a, intcon(0), a_elem);
5377 Node* n_start = array_element_address(n, intcon(0), n_elem);
5378 Node* m_start = array_element_address(m, intcon(0), m_elem);
5379
5380 Node* call = make_runtime_call(RC_LEAF,
5381 OptoRuntime::montgomerySquare_Type(),
5382 stubAddr, stubName, TypePtr::BOTTOM,
5383 a_start, n_start, len, inv, top(),
5384 m_start);
5385 set_result(m);
5386 }
5387
5388 return true;
5389 }
5390
5391 //-------------inline_vectorizedMismatch------------------------------
5392 bool LibraryCallKit::inline_vectorizedMismatch() {
5393 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5394
5395 address stubAddr = StubRoutines::vectorizedMismatch();
5396 if (stubAddr == NULL) {
5397 return false; // Intrinsic's stub is not implemented on this platform
5398 }
5399 const char* stubName = "vectorizedMismatch";
5400 int size_l = callee()->signature()->size();
5401 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5402
5403 Node* obja = argument(0);
5404 Node* aoffset = argument(1);
5405 Node* objb = argument(3);
5406 Node* boffset = argument(4);
5407 Node* length = argument(6);
5408 Node* scale = argument(7);
5409
5410 const Type* a_type = obja->Value(&_gvn);
5411 const Type* b_type = objb->Value(&_gvn);
5412 const TypeAryPtr* top_a = a_type->isa_aryptr();
5413 const TypeAryPtr* top_b = b_type->isa_aryptr();
5414 if (top_a == NULL || top_a->klass() == NULL ||
5415 top_b == NULL || top_b->klass() == NULL) {
5416 // failed array check
5417 return false;
5418 }
5419
5420 Node* call;
5421 jvms()->set_should_reexecute(true);
5422
5423 obja = access_resolve(obja, ACCESS_READ);
5424 objb = access_resolve(objb, ACCESS_READ);
5425 Node* obja_adr = make_unsafe_address(obja, aoffset, ACCESS_READ);
5426 Node* objb_adr = make_unsafe_address(objb, boffset, ACCESS_READ);
5427
5428 call = make_runtime_call(RC_LEAF,
5429 OptoRuntime::vectorizedMismatch_Type(),
5430 stubAddr, stubName, TypePtr::BOTTOM,
5431 obja_adr, objb_adr, length, scale);
5432
5433 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5434 set_result(result);
5435 return true;
5436 }
5437
5438 /**
5439 * Calculate CRC32 for byte.
5440 * int java.util.zip.CRC32.update(int crc, int b)
5441 */
5442 bool LibraryCallKit::inline_updateCRC32() {
5443 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5444 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5445 // no receiver since it is static method
5446 Node* crc = argument(0); // type: int
5447 Node* b = argument(1); // type: int
5448
5449 /*
5450 * int c = ~ crc;
5451 * b = timesXtoThe32[(b ^ c) & 0xFF];
5452 * b = b ^ (c >>> 8);
5453 * crc = ~b;
5454 */
5455
5456 Node* M1 = intcon(-1);
5457 crc = _gvn.transform(new XorINode(crc, M1));
5458 Node* result = _gvn.transform(new XorINode(crc, b));
5459 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5460
5461 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5462 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5463 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5464 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5465
5466 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5467 result = _gvn.transform(new XorINode(crc, result));
5468 result = _gvn.transform(new XorINode(result, M1));
5469 set_result(result);
5470 return true;
5471 }
5472
5473 /**
5474 * Calculate CRC32 for byte[] array.
5475 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5476 */
5477 bool LibraryCallKit::inline_updateBytesCRC32() {
5478 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5479 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5480 // no receiver since it is static method
5481 Node* crc = argument(0); // type: int
5482 Node* src = argument(1); // type: oop
5483 Node* offset = argument(2); // type: int
5484 Node* length = argument(3); // type: int
5485
5486 const Type* src_type = src->Value(&_gvn);
5487 const TypeAryPtr* top_src = src_type->isa_aryptr();
5488 if (top_src == NULL || top_src->klass() == NULL) {
5489 // failed array check
5490 return false;
5491 }
5492
5493 // Figure out the size and type of the elements we will be copying.
5494 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5495 if (src_elem != T_BYTE) {
5496 return false;
5497 }
5498
5499 // 'src_start' points to src array + scaled offset
5500 src = must_be_not_null(src, true);
5501 src = access_resolve(src, ACCESS_READ);
5502 Node* src_start = array_element_address(src, offset, src_elem);
5503
5504 // We assume that range check is done by caller.
5505 // TODO: generate range check (offset+length < src.length) in debug VM.
5506
5507 // Call the stub.
5508 address stubAddr = StubRoutines::updateBytesCRC32();
5509 const char *stubName = "updateBytesCRC32";
5510
5511 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5512 stubAddr, stubName, TypePtr::BOTTOM,
5513 crc, src_start, length);
5514 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5515 set_result(result);
5516 return true;
5517 }
5518
5519 /**
5520 * Calculate CRC32 for ByteBuffer.
5521 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5522 */
5523 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5524 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5525 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5526 // no receiver since it is static method
5527 Node* crc = argument(0); // type: int
5528 Node* src = argument(1); // type: long
5529 Node* offset = argument(3); // type: int
5530 Node* length = argument(4); // type: int
5531
5532 src = ConvL2X(src); // adjust Java long to machine word
5533 Node* base = _gvn.transform(new CastX2PNode(src));
5534 offset = ConvI2X(offset);
5535
5536 // 'src_start' points to src array + scaled offset
5537 Node* src_start = basic_plus_adr(top(), base, offset);
5538
5539 // Call the stub.
5540 address stubAddr = StubRoutines::updateBytesCRC32();
5541 const char *stubName = "updateBytesCRC32";
5542
5543 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5544 stubAddr, stubName, TypePtr::BOTTOM,
5545 crc, src_start, length);
5546 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5547 set_result(result);
5548 return true;
5549 }
5550
5551 //------------------------------get_table_from_crc32c_class-----------------------
5552 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5553 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5554 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5555
5556 return table;
5557 }
5558
5559 //------------------------------inline_updateBytesCRC32C-----------------------
5560 //
5561 // Calculate CRC32C for byte[] array.
5562 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5563 //
5564 bool LibraryCallKit::inline_updateBytesCRC32C() {
5565 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5566 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5567 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5568 // no receiver since it is a static method
5569 Node* crc = argument(0); // type: int
5570 Node* src = argument(1); // type: oop
5571 Node* offset = argument(2); // type: int
5572 Node* end = argument(3); // type: int
5573
5574 Node* length = _gvn.transform(new SubINode(end, offset));
5575
5576 const Type* src_type = src->Value(&_gvn);
5577 const TypeAryPtr* top_src = src_type->isa_aryptr();
5578 if (top_src == NULL || top_src->klass() == NULL) {
5579 // failed array check
5580 return false;
5581 }
5582
5583 // Figure out the size and type of the elements we will be copying.
5584 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5585 if (src_elem != T_BYTE) {
5586 return false;
5587 }
5588
5589 // 'src_start' points to src array + scaled offset
5590 src = must_be_not_null(src, true);
5591 src = access_resolve(src, ACCESS_READ);
5592 Node* src_start = array_element_address(src, offset, src_elem);
5593
5594 // static final int[] byteTable in class CRC32C
5595 Node* table = get_table_from_crc32c_class(callee()->holder());
5596 table = must_be_not_null(table, true);
5597 table = access_resolve(table, ACCESS_READ);
5598 Node* table_start = array_element_address(table, intcon(0), T_INT);
5599
5600 // We assume that range check is done by caller.
5601 // TODO: generate range check (offset+length < src.length) in debug VM.
5602
5603 // Call the stub.
5604 address stubAddr = StubRoutines::updateBytesCRC32C();
5605 const char *stubName = "updateBytesCRC32C";
5606
5607 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5608 stubAddr, stubName, TypePtr::BOTTOM,
5609 crc, src_start, length, table_start);
5610 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5611 set_result(result);
5612 return true;
5613 }
5614
5615 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5616 //
5617 // Calculate CRC32C for DirectByteBuffer.
5618 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5619 //
5620 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5621 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5622 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5623 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5624 // no receiver since it is a static method
5625 Node* crc = argument(0); // type: int
5626 Node* src = argument(1); // type: long
5627 Node* offset = argument(3); // type: int
5628 Node* end = argument(4); // type: int
5629
5630 Node* length = _gvn.transform(new SubINode(end, offset));
5631
5632 src = ConvL2X(src); // adjust Java long to machine word
5633 Node* base = _gvn.transform(new CastX2PNode(src));
5634 offset = ConvI2X(offset);
5635
5636 // 'src_start' points to src array + scaled offset
5637 Node* src_start = basic_plus_adr(top(), base, offset);
5638
5639 // static final int[] byteTable in class CRC32C
5640 Node* table = get_table_from_crc32c_class(callee()->holder());
5641 table = must_be_not_null(table, true);
5642 table = access_resolve(table, ACCESS_READ);
5643 Node* table_start = array_element_address(table, intcon(0), T_INT);
5644
5645 // Call the stub.
5646 address stubAddr = StubRoutines::updateBytesCRC32C();
5647 const char *stubName = "updateBytesCRC32C";
5648
5649 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5650 stubAddr, stubName, TypePtr::BOTTOM,
5651 crc, src_start, length, table_start);
5652 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5653 set_result(result);
5654 return true;
5655 }
5656
5657 //------------------------------inline_updateBytesAdler32----------------------
5658 //
5659 // Calculate Adler32 checksum for byte[] array.
5660 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5661 //
5662 bool LibraryCallKit::inline_updateBytesAdler32() {
5663 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5664 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5665 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5666 // no receiver since it is static method
5667 Node* crc = argument(0); // type: int
5668 Node* src = argument(1); // type: oop
5669 Node* offset = argument(2); // type: int
5670 Node* length = argument(3); // type: int
5671
5672 const Type* src_type = src->Value(&_gvn);
5673 const TypeAryPtr* top_src = src_type->isa_aryptr();
5674 if (top_src == NULL || top_src->klass() == NULL) {
5675 // failed array check
5676 return false;
5677 }
5678
5679 // Figure out the size and type of the elements we will be copying.
5680 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5681 if (src_elem != T_BYTE) {
5682 return false;
5683 }
5684
5685 // 'src_start' points to src array + scaled offset
5686 src = access_resolve(src, ACCESS_READ);
5687 Node* src_start = array_element_address(src, offset, src_elem);
5688
5689 // We assume that range check is done by caller.
5690 // TODO: generate range check (offset+length < src.length) in debug VM.
5691
5692 // Call the stub.
5693 address stubAddr = StubRoutines::updateBytesAdler32();
5694 const char *stubName = "updateBytesAdler32";
5695
5696 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5697 stubAddr, stubName, TypePtr::BOTTOM,
5698 crc, src_start, length);
5699 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5700 set_result(result);
5701 return true;
5702 }
5703
5704 //------------------------------inline_updateByteBufferAdler32---------------
5705 //
5706 // Calculate Adler32 checksum for DirectByteBuffer.
5707 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5708 //
5709 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5710 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5711 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5712 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5713 // no receiver since it is static method
5714 Node* crc = argument(0); // type: int
5715 Node* src = argument(1); // type: long
5716 Node* offset = argument(3); // type: int
5717 Node* length = argument(4); // type: int
5718
5719 src = ConvL2X(src); // adjust Java long to machine word
5720 Node* base = _gvn.transform(new CastX2PNode(src));
5721 offset = ConvI2X(offset);
5722
5723 // 'src_start' points to src array + scaled offset
5724 Node* src_start = basic_plus_adr(top(), base, offset);
5725
5726 // Call the stub.
5727 address stubAddr = StubRoutines::updateBytesAdler32();
5728 const char *stubName = "updateBytesAdler32";
5729
5730 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5731 stubAddr, stubName, TypePtr::BOTTOM,
5732 crc, src_start, length);
5733
5734 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5735 set_result(result);
5736 return true;
5737 }
5738
5739 //----------------------------inline_reference_get----------------------------
5740 // public T java.lang.ref.Reference.get();
5741 bool LibraryCallKit::inline_reference_get() {
5742 const int referent_offset = java_lang_ref_Reference::referent_offset;
5743 guarantee(referent_offset > 0, "should have already been set");
5744
5745 // Get the argument:
5746 Node* reference_obj = null_check_receiver();
5747 if (stopped()) return true;
5748
5749 const TypeInstPtr* tinst = _gvn.type(reference_obj)->isa_instptr();
5750 assert(tinst != NULL, "obj is null");
5751 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5752 ciInstanceKlass* referenceKlass = tinst->klass()->as_instance_klass();
5753 ciField* field = referenceKlass->get_field_by_name(ciSymbol::make("referent"),
5754 ciSymbol::make("Ljava/lang/Object;"),
5755 false);
5756 assert (field != NULL, "undefined field");
5757
5758 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5759 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5760
5761 ciInstanceKlass* klass = env()->Object_klass();
5762 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5763
5764 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
5765 Node* result = access_load_at(reference_obj, adr, adr_type, object_type, T_OBJECT, decorators);
5766 // Add memory barrier to prevent commoning reads from this field
5767 // across safepoint since GC can change its value.
5768 insert_mem_bar(Op_MemBarCPUOrder);
5769
5770 set_result(result);
5771 return true;
5772 }
5773
5774
5775 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5776 bool is_exact=true, bool is_static=false,
5777 ciInstanceKlass * fromKls=NULL) {
5778 if (fromKls == NULL) {
5779 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5780 assert(tinst != NULL, "obj is null");
5781 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5782 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5783 fromKls = tinst->klass()->as_instance_klass();
5784 } else {
5785 assert(is_static, "only for static field access");
5786 }
5787 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5788 ciSymbol::make(fieldTypeString),
5789 is_static);
5790
5791 assert (field != NULL, "undefined field");
5792 if (field == NULL) return (Node *) NULL;
5793
5794 if (is_static) {
5795 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5796 fromObj = makecon(tip);
5797 }
5798
5799 // Next code copied from Parse::do_get_xxx():
5800
5801 // Compute address and memory type.
5802 int offset = field->offset_in_bytes();
5803 bool is_vol = field->is_volatile();
5804 ciType* field_klass = field->type();
5805 assert(field_klass->is_loaded(), "should be loaded");
5806 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5807 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5808 BasicType bt = field->layout_type();
5809
5810 // Build the resultant type of the load
5811 const Type *type;
5812 if (bt == T_OBJECT) {
5813 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5814 } else {
5815 type = Type::get_const_basic_type(bt);
5816 }
5817
5818 DecoratorSet decorators = IN_HEAP;
5819
5820 if (is_vol) {
5821 decorators |= MO_SEQ_CST;
5822 }
5823
5824 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
5825 }
5826
5827 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5828 bool is_exact = true, bool is_static = false,
5829 ciInstanceKlass * fromKls = NULL) {
5830 if (fromKls == NULL) {
5831 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5832 assert(tinst != NULL, "obj is null");
5833 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5834 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5835 fromKls = tinst->klass()->as_instance_klass();
5836 }
5837 else {
5838 assert(is_static, "only for static field access");
5839 }
5840 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5841 ciSymbol::make(fieldTypeString),
5842 is_static);
5843
5844 assert(field != NULL, "undefined field");
5845 assert(!field->is_volatile(), "not defined for volatile fields");
5846
5847 if (is_static) {
5848 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5849 fromObj = makecon(tip);
5850 }
5851
5852 // Next code copied from Parse::do_get_xxx():
5853
5854 // Compute address and memory type.
5855 int offset = field->offset_in_bytes();
5856 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5857
5858 return adr;
5859 }
5860
5861 //------------------------------inline_aescrypt_Block-----------------------
5862 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
5863 address stubAddr = NULL;
5864 const char *stubName;
5865 assert(UseAES, "need AES instruction support");
5866
5867 switch(id) {
5868 case vmIntrinsics::_aescrypt_encryptBlock:
5869 stubAddr = StubRoutines::aescrypt_encryptBlock();
5870 stubName = "aescrypt_encryptBlock";
5871 break;
5872 case vmIntrinsics::_aescrypt_decryptBlock:
5873 stubAddr = StubRoutines::aescrypt_decryptBlock();
5874 stubName = "aescrypt_decryptBlock";
5875 break;
5876 default:
5877 break;
5878 }
5879 if (stubAddr == NULL) return false;
5880
5881 Node* aescrypt_object = argument(0);
5882 Node* src = argument(1);
5883 Node* src_offset = argument(2);
5884 Node* dest = argument(3);
5885 Node* dest_offset = argument(4);
5886
5887 src = must_be_not_null(src, true);
5888 dest = must_be_not_null(dest, true);
5889
5890 src = access_resolve(src, ACCESS_READ);
5891 dest = access_resolve(dest, ACCESS_WRITE);
5892
5893 // (1) src and dest are arrays.
5894 const Type* src_type = src->Value(&_gvn);
5895 const Type* dest_type = dest->Value(&_gvn);
5896 const TypeAryPtr* top_src = src_type->isa_aryptr();
5897 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5898 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5899
5900 // for the quick and dirty code we will skip all the checks.
5901 // we are just trying to get the call to be generated.
5902 Node* src_start = src;
5903 Node* dest_start = dest;
5904 if (src_offset != NULL || dest_offset != NULL) {
5905 assert(src_offset != NULL && dest_offset != NULL, "");
5906 src_start = array_element_address(src, src_offset, T_BYTE);
5907 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5908 }
5909
5910 // now need to get the start of its expanded key array
5911 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5912 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5913 if (k_start == NULL) return false;
5914
5915 if (Matcher::pass_original_key_for_aes()) {
5916 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5917 // compatibility issues between Java key expansion and SPARC crypto instructions
5918 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5919 if (original_k_start == NULL) return false;
5920
5921 // Call the stub.
5922 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5923 stubAddr, stubName, TypePtr::BOTTOM,
5924 src_start, dest_start, k_start, original_k_start);
5925 } else {
5926 // Call the stub.
5927 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5928 stubAddr, stubName, TypePtr::BOTTOM,
5929 src_start, dest_start, k_start);
5930 }
5931
5932 return true;
5933 }
5934
5935 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
5936 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
5937 address stubAddr = NULL;
5938 const char *stubName = NULL;
5939
5940 assert(UseAES, "need AES instruction support");
5941
5942 switch(id) {
5943 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
5944 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
5945 stubName = "cipherBlockChaining_encryptAESCrypt";
5946 break;
5947 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
5948 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
5949 stubName = "cipherBlockChaining_decryptAESCrypt";
5950 break;
5951 default:
5952 break;
5953 }
5954 if (stubAddr == NULL) return false;
5955
5956 Node* cipherBlockChaining_object = argument(0);
5957 Node* src = argument(1);
5958 Node* src_offset = argument(2);
5959 Node* len = argument(3);
5960 Node* dest = argument(4);
5961 Node* dest_offset = argument(5);
5962
5963 src = must_be_not_null(src, false);
5964 dest = must_be_not_null(dest, false);
5965
5966 src = access_resolve(src, ACCESS_READ);
5967 dest = access_resolve(dest, ACCESS_WRITE);
5968
5969 // (1) src and dest are arrays.
5970 const Type* src_type = src->Value(&_gvn);
5971 const Type* dest_type = dest->Value(&_gvn);
5972 const TypeAryPtr* top_src = src_type->isa_aryptr();
5973 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5974 assert (top_src != NULL && top_src->klass() != NULL
5975 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5976
5977 // checks are the responsibility of the caller
5978 Node* src_start = src;
5979 Node* dest_start = dest;
5980 if (src_offset != NULL || dest_offset != NULL) {
5981 assert(src_offset != NULL && dest_offset != NULL, "");
5982 src_start = array_element_address(src, src_offset, T_BYTE);
5983 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5984 }
5985
5986 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
5987 // (because of the predicated logic executed earlier).
5988 // so we cast it here safely.
5989 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5990
5991 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
5992 if (embeddedCipherObj == NULL) return false;
5993
5994 // cast it to what we know it will be at runtime
5995 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
5996 assert(tinst != NULL, "CBC obj is null");
5997 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
5998 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5999 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6000
6001 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6002 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6003 const TypeOopPtr* xtype = aklass->as_instance_type();
6004 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6005 aescrypt_object = _gvn.transform(aescrypt_object);
6006
6007 // we need to get the start of the aescrypt_object's expanded key array
6008 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6009 if (k_start == NULL) return false;
6010
6011 // similarly, get the start address of the r vector
6012 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
6013 if (objRvec == NULL) return false;
6014 objRvec = access_resolve(objRvec, ACCESS_WRITE);
6015 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6016
6017 Node* cbcCrypt;
6018 if (Matcher::pass_original_key_for_aes()) {
6019 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6020 // compatibility issues between Java key expansion and SPARC crypto instructions
6021 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6022 if (original_k_start == NULL) return false;
6023
6024 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
6025 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6026 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6027 stubAddr, stubName, TypePtr::BOTTOM,
6028 src_start, dest_start, k_start, r_start, len, original_k_start);
6029 } else {
6030 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6031 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6032 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6033 stubAddr, stubName, TypePtr::BOTTOM,
6034 src_start, dest_start, k_start, r_start, len);
6035 }
6036
6037 // return cipher length (int)
6038 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6039 set_result(retvalue);
6040 return true;
6041 }
6042
6043 //------------------------------inline_counterMode_AESCrypt-----------------------
6044 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6045 assert(UseAES, "need AES instruction support");
6046 if (!UseAESCTRIntrinsics) return false;
6047
6048 address stubAddr = NULL;
6049 const char *stubName = NULL;
6050 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6051 stubAddr = StubRoutines::counterMode_AESCrypt();
6052 stubName = "counterMode_AESCrypt";
6053 }
6054 if (stubAddr == NULL) return false;
6055
6056 Node* counterMode_object = argument(0);
6057 Node* src = argument(1);
6058 Node* src_offset = argument(2);
6059 Node* len = argument(3);
6060 Node* dest = argument(4);
6061 Node* dest_offset = argument(5);
6062
6063 src = access_resolve(src, ACCESS_READ);
6064 dest = access_resolve(dest, ACCESS_WRITE);
6065 counterMode_object = access_resolve(counterMode_object, ACCESS_WRITE);
6066
6067 // (1) src and dest are arrays.
6068 const Type* src_type = src->Value(&_gvn);
6069 const Type* dest_type = dest->Value(&_gvn);
6070 const TypeAryPtr* top_src = src_type->isa_aryptr();
6071 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6072 assert(top_src != NULL && top_src->klass() != NULL &&
6073 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6074
6075 // checks are the responsibility of the caller
6076 Node* src_start = src;
6077 Node* dest_start = dest;
6078 if (src_offset != NULL || dest_offset != NULL) {
6079 assert(src_offset != NULL && dest_offset != NULL, "");
6080 src_start = array_element_address(src, src_offset, T_BYTE);
6081 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6082 }
6083
6084 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6085 // (because of the predicated logic executed earlier).
6086 // so we cast it here safely.
6087 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6088 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6089 if (embeddedCipherObj == NULL) return false;
6090 // cast it to what we know it will be at runtime
6091 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6092 assert(tinst != NULL, "CTR obj is null");
6093 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6094 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6095 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6096 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6097 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6098 const TypeOopPtr* xtype = aklass->as_instance_type();
6099 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6100 aescrypt_object = _gvn.transform(aescrypt_object);
6101 // we need to get the start of the aescrypt_object's expanded key array
6102 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6103 if (k_start == NULL) return false;
6104 // similarly, get the start address of the r vector
6105 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6106 if (obj_counter == NULL) return false;
6107 obj_counter = access_resolve(obj_counter, ACCESS_WRITE);
6108 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6109
6110 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6111 if (saved_encCounter == NULL) return false;
6112 saved_encCounter = access_resolve(saved_encCounter, ACCESS_WRITE);
6113 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6114 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6115
6116 Node* ctrCrypt;
6117 if (Matcher::pass_original_key_for_aes()) {
6118 // no SPARC version for AES/CTR intrinsics now.
6119 return false;
6120 }
6121 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6122 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6123 OptoRuntime::counterMode_aescrypt_Type(),
6124 stubAddr, stubName, TypePtr::BOTTOM,
6125 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6126
6127 // return cipher length (int)
6128 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6129 set_result(retvalue);
6130 return true;
6131 }
6132
6133 //------------------------------get_key_start_from_aescrypt_object-----------------------
6134 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6135 #if defined(PPC64) || defined(S390)
6136 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6137 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6138 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6139 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6140 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6141 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6142 if (objSessionK == NULL) {
6143 return (Node *) NULL;
6144 }
6145 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6146 #else
6147 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6148 #endif // PPC64
6149 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6150 if (objAESCryptKey == NULL) return (Node *) NULL;
6151
6152 // now have the array, need to get the start address of the K array
6153 objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ);
6154 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6155 return k_start;
6156 }
6157
6158 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6159 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6160 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6161 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6162 if (objAESCryptKey == NULL) return (Node *) NULL;
6163
6164 // now have the array, need to get the start address of the lastKey array
6165 objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ);
6166 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6167 return original_k_start;
6168 }
6169
6170 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6171 // Return node representing slow path of predicate check.
6172 // the pseudo code we want to emulate with this predicate is:
6173 // for encryption:
6174 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6175 // for decryption:
6176 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6177 // note cipher==plain is more conservative than the original java code but that's OK
6178 //
6179 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6180 // The receiver was checked for NULL already.
6181 Node* objCBC = argument(0);
6182
6183 Node* src = argument(1);
6184 Node* dest = argument(4);
6185
6186 // Load embeddedCipher field of CipherBlockChaining object.
6187 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6188
6189 // get AESCrypt klass for instanceOf check
6190 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6191 // will have same classloader as CipherBlockChaining object
6192 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6193 assert(tinst != NULL, "CBCobj is null");
6194 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6195
6196 // we want to do an instanceof comparison against the AESCrypt class
6197 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6198 if (!klass_AESCrypt->is_loaded()) {
6199 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6200 Node* ctrl = control();
6201 set_control(top()); // no regular fast path
6202 return ctrl;
6203 }
6204
6205 src = must_be_not_null(src, true);
6206 dest = must_be_not_null(dest, true);
6207
6208 // Resolve oops to stable for CmpP below.
6209 src = access_resolve(src, 0);
6210 dest = access_resolve(dest, 0);
6211
6212 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6213
6214 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6215 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6216 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6217
6218 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6219
6220 // for encryption, we are done
6221 if (!decrypting)
6222 return instof_false; // even if it is NULL
6223
6224 // for decryption, we need to add a further check to avoid
6225 // taking the intrinsic path when cipher and plain are the same
6226 // see the original java code for why.
6227 RegionNode* region = new RegionNode(3);
6228 region->init_req(1, instof_false);
6229
6230 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6231 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6232 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6233 region->init_req(2, src_dest_conjoint);
6234
6235 record_for_igvn(region);
6236 return _gvn.transform(region);
6237 }
6238
6239 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6240 // Return node representing slow path of predicate check.
6241 // the pseudo code we want to emulate with this predicate is:
6242 // for encryption:
6243 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6244 // for decryption:
6245 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6246 // note cipher==plain is more conservative than the original java code but that's OK
6247 //
6248
6249 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6250 // The receiver was checked for NULL already.
6251 Node* objCTR = argument(0);
6252
6253 // Load embeddedCipher field of CipherBlockChaining object.
6254 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6255
6256 // get AESCrypt klass for instanceOf check
6257 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6258 // will have same classloader as CipherBlockChaining object
6259 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6260 assert(tinst != NULL, "CTRobj is null");
6261 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6262
6263 // we want to do an instanceof comparison against the AESCrypt class
6264 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6265 if (!klass_AESCrypt->is_loaded()) {
6266 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6267 Node* ctrl = control();
6268 set_control(top()); // no regular fast path
6269 return ctrl;
6270 }
6271
6272 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6273 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6274 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6275 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6276 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6277
6278 return instof_false; // even if it is NULL
6279 }
6280
6281 //------------------------------inline_ghash_processBlocks
6282 bool LibraryCallKit::inline_ghash_processBlocks() {
6283 address stubAddr;
6284 const char *stubName;
6285 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6286
6287 stubAddr = StubRoutines::ghash_processBlocks();
6288 stubName = "ghash_processBlocks";
6289
6290 Node* data = argument(0);
6291 Node* offset = argument(1);
6292 Node* len = argument(2);
6293 Node* state = argument(3);
6294 Node* subkeyH = argument(4);
6295
6296 state = must_be_not_null(state, true);
6297 subkeyH = must_be_not_null(subkeyH, true);
6298 data = must_be_not_null(data, true);
6299
6300 state = access_resolve(state, ACCESS_WRITE);
6301 subkeyH = access_resolve(subkeyH, ACCESS_READ);
6302 data = access_resolve(data, ACCESS_READ);
6303
6304 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6305 assert(state_start, "state is NULL");
6306 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6307 assert(subkeyH_start, "subkeyH is NULL");
6308 Node* data_start = array_element_address(data, offset, T_BYTE);
6309 assert(data_start, "data is NULL");
6310
6311 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6312 OptoRuntime::ghash_processBlocks_Type(),
6313 stubAddr, stubName, TypePtr::BOTTOM,
6314 state_start, subkeyH_start, data_start, len);
6315 return true;
6316 }
6317
6318 bool LibraryCallKit::inline_base64_encodeBlock() {
6319 address stubAddr;
6320 const char *stubName;
6321 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6322 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6323 stubAddr = StubRoutines::base64_encodeBlock();
6324 stubName = "encodeBlock";
6325
6326 if (!stubAddr) return false;
6327 Node* base64obj = argument(0);
6328 Node* src = argument(1);
6329 Node* offset = argument(2);
6330 Node* len = argument(3);
6331 Node* dest = argument(4);
6332 Node* dp = argument(5);
6333 Node* isURL = argument(6);
6334
6335 src = must_be_not_null(src, true);
6336 src = access_resolve(src, ACCESS_READ);
6337 dest = must_be_not_null(dest, true);
6338 dest = access_resolve(dest, ACCESS_WRITE);
6339
6340 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6341 assert(src_start, "source array is NULL");
6342 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6343 assert(dest_start, "destination array is NULL");
6344
6345 Node* base64 = make_runtime_call(RC_LEAF,
6346 OptoRuntime::base64_encodeBlock_Type(),
6347 stubAddr, stubName, TypePtr::BOTTOM,
6348 src_start, offset, len, dest_start, dp, isURL);
6349 return true;
6350 }
6351
6352 //------------------------------inline_sha_implCompress-----------------------
6353 //
6354 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6355 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6356 //
6357 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6358 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6359 //
6360 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6361 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6362 //
6363 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6364 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6365
6366 Node* sha_obj = argument(0);
6367 Node* src = argument(1); // type oop
6368 Node* ofs = argument(2); // type int
6369
6370 const Type* src_type = src->Value(&_gvn);
6371 const TypeAryPtr* top_src = src_type->isa_aryptr();
6372 if (top_src == NULL || top_src->klass() == NULL) {
6373 // failed array check
6374 return false;
6375 }
6376 // Figure out the size and type of the elements we will be copying.
6377 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6378 if (src_elem != T_BYTE) {
6379 return false;
6380 }
6381 // 'src_start' points to src array + offset
6382 src = must_be_not_null(src, true);
6383 src = access_resolve(src, ACCESS_READ);
6384 Node* src_start = array_element_address(src, ofs, src_elem);
6385 Node* state = NULL;
6386 address stubAddr;
6387 const char *stubName;
6388
6389 switch(id) {
6390 case vmIntrinsics::_sha_implCompress:
6391 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6392 state = get_state_from_sha_object(sha_obj);
6393 stubAddr = StubRoutines::sha1_implCompress();
6394 stubName = "sha1_implCompress";
6395 break;
6396 case vmIntrinsics::_sha2_implCompress:
6397 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6398 state = get_state_from_sha_object(sha_obj);
6399 stubAddr = StubRoutines::sha256_implCompress();
6400 stubName = "sha256_implCompress";
6401 break;
6402 case vmIntrinsics::_sha5_implCompress:
6403 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6404 state = get_state_from_sha5_object(sha_obj);
6405 stubAddr = StubRoutines::sha512_implCompress();
6406 stubName = "sha512_implCompress";
6407 break;
6408 default:
6409 fatal_unexpected_iid(id);
6410 return false;
6411 }
6412 if (state == NULL) return false;
6413
6414 assert(stubAddr != NULL, "Stub is generated");
6415 if (stubAddr == NULL) return false;
6416
6417 // Call the stub.
6418 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6419 stubAddr, stubName, TypePtr::BOTTOM,
6420 src_start, state);
6421
6422 return true;
6423 }
6424
6425 //------------------------------inline_digestBase_implCompressMB-----------------------
6426 //
6427 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6428 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6429 //
6430 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6431 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6432 "need SHA1/SHA256/SHA512 instruction support");
6433 assert((uint)predicate < 3, "sanity");
6434 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6435
6436 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6437 Node* src = argument(1); // byte[] array
6438 Node* ofs = argument(2); // type int
6439 Node* limit = argument(3); // type int
6440
6441 const Type* src_type = src->Value(&_gvn);
6442 const TypeAryPtr* top_src = src_type->isa_aryptr();
6443 if (top_src == NULL || top_src->klass() == NULL) {
6444 // failed array check
6445 return false;
6446 }
6447 // Figure out the size and type of the elements we will be copying.
6448 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6449 if (src_elem != T_BYTE) {
6450 return false;
6451 }
6452 // 'src_start' points to src array + offset
6453 src = must_be_not_null(src, false);
6454 src = access_resolve(src, ACCESS_READ);
6455 Node* src_start = array_element_address(src, ofs, src_elem);
6456
6457 const char* klass_SHA_name = NULL;
6458 const char* stub_name = NULL;
6459 address stub_addr = NULL;
6460 bool long_state = false;
6461
6462 switch (predicate) {
6463 case 0:
6464 if (UseSHA1Intrinsics) {
6465 klass_SHA_name = "sun/security/provider/SHA";
6466 stub_name = "sha1_implCompressMB";
6467 stub_addr = StubRoutines::sha1_implCompressMB();
6468 }
6469 break;
6470 case 1:
6471 if (UseSHA256Intrinsics) {
6472 klass_SHA_name = "sun/security/provider/SHA2";
6473 stub_name = "sha256_implCompressMB";
6474 stub_addr = StubRoutines::sha256_implCompressMB();
6475 }
6476 break;
6477 case 2:
6478 if (UseSHA512Intrinsics) {
6479 klass_SHA_name = "sun/security/provider/SHA5";
6480 stub_name = "sha512_implCompressMB";
6481 stub_addr = StubRoutines::sha512_implCompressMB();
6482 long_state = true;
6483 }
6484 break;
6485 default:
6486 fatal("unknown SHA intrinsic predicate: %d", predicate);
6487 }
6488 if (klass_SHA_name != NULL) {
6489 assert(stub_addr != NULL, "Stub is generated");
6490 if (stub_addr == NULL) return false;
6491
6492 // get DigestBase klass to lookup for SHA klass
6493 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6494 assert(tinst != NULL, "digestBase_obj is not instance???");
6495 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6496
6497 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6498 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6499 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6500 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6501 }
6502 return false;
6503 }
6504 //------------------------------inline_sha_implCompressMB-----------------------
6505 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6506 bool long_state, address stubAddr, const char *stubName,
6507 Node* src_start, Node* ofs, Node* limit) {
6508 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6509 const TypeOopPtr* xtype = aklass->as_instance_type();
6510 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6511 sha_obj = _gvn.transform(sha_obj);
6512
6513 Node* state;
6514 if (long_state) {
6515 state = get_state_from_sha5_object(sha_obj);
6516 } else {
6517 state = get_state_from_sha_object(sha_obj);
6518 }
6519 if (state == NULL) return false;
6520
6521 // Call the stub.
6522 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6523 OptoRuntime::digestBase_implCompressMB_Type(),
6524 stubAddr, stubName, TypePtr::BOTTOM,
6525 src_start, state, ofs, limit);
6526 // return ofs (int)
6527 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6528 set_result(result);
6529
6530 return true;
6531 }
6532
6533 //------------------------------get_state_from_sha_object-----------------------
6534 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6535 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6536 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6537 if (sha_state == NULL) return (Node *) NULL;
6538
6539 // now have the array, need to get the start address of the state array
6540 sha_state = access_resolve(sha_state, ACCESS_WRITE);
6541 Node* state = array_element_address(sha_state, intcon(0), T_INT);
6542 return state;
6543 }
6544
6545 //------------------------------get_state_from_sha5_object-----------------------
6546 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6547 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6548 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6549 if (sha_state == NULL) return (Node *) NULL;
6550
6551 // now have the array, need to get the start address of the state array
6552 sha_state = access_resolve(sha_state, ACCESS_WRITE);
6553 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6554 return state;
6555 }
6556
6557 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6558 // Return node representing slow path of predicate check.
6559 // the pseudo code we want to emulate with this predicate is:
6560 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6561 //
6562 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6563 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6564 "need SHA1/SHA256/SHA512 instruction support");
6565 assert((uint)predicate < 3, "sanity");
6566
6567 // The receiver was checked for NULL already.
6568 Node* digestBaseObj = argument(0);
6569
6570 // get DigestBase klass for instanceOf check
6571 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6572 assert(tinst != NULL, "digestBaseObj is null");
6573 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6574
6575 const char* klass_SHA_name = NULL;
6576 switch (predicate) {
6577 case 0:
6578 if (UseSHA1Intrinsics) {
6579 // we want to do an instanceof comparison against the SHA class
6580 klass_SHA_name = "sun/security/provider/SHA";
6581 }
6582 break;
6583 case 1:
6584 if (UseSHA256Intrinsics) {
6585 // we want to do an instanceof comparison against the SHA2 class
6586 klass_SHA_name = "sun/security/provider/SHA2";
6587 }
6588 break;
6589 case 2:
6590 if (UseSHA512Intrinsics) {
6591 // we want to do an instanceof comparison against the SHA5 class
6592 klass_SHA_name = "sun/security/provider/SHA5";
6593 }
6594 break;
6595 default:
6596 fatal("unknown SHA intrinsic predicate: %d", predicate);
6597 }
6598
6599 ciKlass* klass_SHA = NULL;
6600 if (klass_SHA_name != NULL) {
6601 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6602 }
6603 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6604 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6605 Node* ctrl = control();
6606 set_control(top()); // no intrinsic path
6607 return ctrl;
6608 }
6609 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6610
6611 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6612 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6613 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6614 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6615
6616 return instof_false; // even if it is NULL
6617 }
6618
6619 // long Continuations::getFP() ()J
6620 bool LibraryCallKit::inline_get_frame_pointer() {
6621 Node *frame = _gvn.transform(new GetFPNode(control()));
6622 set_result(frame);
6623 return true;
6624 }
6625
6626 bool LibraryCallKit::inline_continuation(vmIntrinsics::ID id) {
6627 address call_addr = NULL;
6628 const char *name = NULL;
6629
6630 switch (id) {
6631 case vmIntrinsics::_Continuation_getSP: call_addr = StubRoutines::cont_getSP(); name = "getSP"; break;
6632 case vmIntrinsics::_Continuation_getPC: call_addr = StubRoutines::cont_getPC(); name = "getPC"; break;
6633 case vmIntrinsics::_Continuation_getFP: call_addr = OptoRuntime::continuation_getFP_Java(); name = "getFP"; break;
6634 default: fatal("error"); return false;
6635 }
6636
6637 const TypeFunc* tf = OptoRuntime::void_long_Type();
6638 const TypePtr* no_memory_effects = NULL;
6639 Node* call = make_runtime_call(RC_LEAF, tf, call_addr, name, no_memory_effects);
6640 Node* value = _gvn.transform(new ProjNode(call, TypeFunc::Parms+0));
6641 #ifdef ASSERT
6642 Node* value_top = _gvn.transform(new ProjNode(call, TypeFunc::Parms+1));
6643 assert(value_top == top(), "second value must be top");
6644 #endif
6645 set_result(value);
6646 return true;
6647 }
6648
6649 bool LibraryCallKit::inline_continuation_runLevel() {
6650 set_result(intcon(2));
6651 return true;
6652 }
6653
6654 bool LibraryCallKit::inline_continuation_do_continue() {
6655 address call_addr = StubRoutines::cont_thaw();
6656 const TypeFunc* tf = OptoRuntime::void_void_Type();
6657 Node* call = make_runtime_call(RC_NO_LEAF, tf, call_addr, "doContinue", TypeRawPtr::BOTTOM);
6658 assert (call->is_CallJava(), "");
6659 Node* result = set_results_for_java_call((CallJavaNode*)call, false, /* deoptimize */ false);
6660 return true;
6661 }
6662
6663 bool LibraryCallKit::inline_continuation_do_yield() {
6664 address call_addr = StubRoutines::cont_doYield();
6665 // Node* arg0 = argument(0); // type int - scopes
6666 Node* arg0 = intcon(0); // type int - from interpreter
6667 const TypeFunc* tf = OptoRuntime::continuation_doYield_Type();
6668 Node* call = make_runtime_call(RC_NO_LEAF, tf, call_addr, "doYield", TypeRawPtr::BOTTOM, arg0);
6669 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6670 set_result(result);
6671 return true;
6672 }
6673
6674 bool LibraryCallKit::inline_continuation_jump() {
6675 address call_addr = StubRoutines::cont_jump();
6676 Node* sp = argument(0); // type long
6677 Node* fp = argument(2); // type long
6678 Node* pc = argument(4); // type long
6679 const TypeFunc* tf = OptoRuntime::continuation_jump_Type();
6680 Node* call = make_runtime_call(RC_LEAF, tf, call_addr, "jump", TypeRawPtr::BOTTOM, sp, fp, pc);
6681 return true;
6682 }
6683
6684 //-------------inline_fma-----------------------------------
6685 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
6686 Node *a = NULL;
6687 Node *b = NULL;
6688 Node *c = NULL;
6689 Node* result = NULL;
6690 switch (id) {
6691 case vmIntrinsics::_fmaD:
6692 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
6693 // no receiver since it is static method
6694 a = round_double_node(argument(0));
6695 b = round_double_node(argument(2));
6696 c = round_double_node(argument(4));
6697 result = _gvn.transform(new FmaDNode(control(), a, b, c));
6698 break;
6699 case vmIntrinsics::_fmaF:
6700 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
6701 a = argument(0);
6702 b = argument(1);
6703 c = argument(2);
6704 result = _gvn.transform(new FmaFNode(control(), a, b, c));
6705 break;
6706 default:
6707 fatal_unexpected_iid(id); break;
6708 }
6709 set_result(result);
6710 return true;
6711 }
6712
6713 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
6714 // argument(0) is receiver
6715 Node* codePoint = argument(1);
6716 Node* n = NULL;
6717
6718 switch (id) {
6719 case vmIntrinsics::_isDigit :
6720 n = new DigitNode(control(), codePoint);
6721 break;
6722 case vmIntrinsics::_isLowerCase :
6723 n = new LowerCaseNode(control(), codePoint);
6724 break;
6725 case vmIntrinsics::_isUpperCase :
6726 n = new UpperCaseNode(control(), codePoint);
6727 break;
6728 case vmIntrinsics::_isWhitespace :
6729 n = new WhitespaceNode(control(), codePoint);
6730 break;
6731 default:
6732 fatal_unexpected_iid(id);
6733 }
6734
6735 set_result(_gvn.transform(n));
6736 return true;
6737 }
6738
6739 //------------------------------inline_fp_min_max------------------------------
6740 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
6741 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
6742
6743 // The intrinsic should be used only when the API branches aren't predictable,
6744 // the last one performing the most important comparison. The following heuristic
6745 // uses the branch statistics to eventually bail out if necessary.
6746
6747 ciMethodData *md = callee()->method_data();
6748
6749 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
6750 ciCallProfile cp = caller()->call_profile_at_bci(bci());
6751
6752 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
6753 // Bail out if the call-site didn't contribute enough to the statistics.
6754 return false;
6755 }
6756
6757 uint taken = 0, not_taken = 0;
6758
6759 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
6760 if (p->is_BranchData()) {
6761 taken = ((ciBranchData*)p)->taken();
6762 not_taken = ((ciBranchData*)p)->not_taken();
6763 }
6764 }
6765
6766 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
6767 balance = balance < 0 ? -balance : balance;
6768 if ( balance > 0.2 ) {
6769 // Bail out if the most important branch is predictable enough.
6770 return false;
6771 }
6772 }
6773 */
6774
6775 Node *a = NULL;
6776 Node *b = NULL;
6777 Node *n = NULL;
6778 switch (id) {
6779 case vmIntrinsics::_maxF:
6780 case vmIntrinsics::_minF:
6781 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
6782 a = argument(0);
6783 b = argument(1);
6784 break;
6785 case vmIntrinsics::_maxD:
6786 case vmIntrinsics::_minD:
6787 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
6788 a = round_double_node(argument(0));
6789 b = round_double_node(argument(2));
6790 break;
6791 default:
6792 fatal_unexpected_iid(id);
6793 break;
6794 }
6795 switch (id) {
6796 case vmIntrinsics::_maxF: n = new MaxFNode(a, b); break;
6797 case vmIntrinsics::_minF: n = new MinFNode(a, b); break;
6798 case vmIntrinsics::_maxD: n = new MaxDNode(a, b); break;
6799 case vmIntrinsics::_minD: n = new MinDNode(a, b); break;
6800 default: fatal_unexpected_iid(id); break;
6801 }
6802 set_result(_gvn.transform(n));
6803 return true;
6804 }
6805
6806 bool LibraryCallKit::inline_profileBoolean() {
6807 Node* counts = argument(1);
6808 const TypeAryPtr* ary = NULL;
6809 ciArray* aobj = NULL;
6810 if (counts->is_Con()
6811 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6812 && (aobj = ary->const_oop()->as_array()) != NULL
6813 && (aobj->length() == 2)) {
6814 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6815 jint false_cnt = aobj->element_value(0).as_int();
6816 jint true_cnt = aobj->element_value(1).as_int();
6817
6818 if (C->log() != NULL) {
6819 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6820 false_cnt, true_cnt);
6821 }
6822
6823 if (false_cnt + true_cnt == 0) {
6824 // According to profile, never executed.
6825 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6826 Deoptimization::Action_reinterpret);
6827 return true;
6828 }
6829
6830 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6831 // is a number of each value occurrences.
6832 Node* result = argument(0);
6833 if (false_cnt == 0 || true_cnt == 0) {
6834 // According to profile, one value has been never seen.
6835 int expected_val = (false_cnt == 0) ? 1 : 0;
6836
6837 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6838 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6839
6840 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6841 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6842 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6843
6844 { // Slow path: uncommon trap for never seen value and then reexecute
6845 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6846 // the value has been seen at least once.
6847 PreserveJVMState pjvms(this);
6848 PreserveReexecuteState preexecs(this);
6849 jvms()->set_should_reexecute(true);
6850
6851 set_control(slow_path);
6852 set_i_o(i_o());
6853
6854 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6855 Deoptimization::Action_reinterpret);
6856 }
6857 // The guard for never seen value enables sharpening of the result and
6858 // returning a constant. It allows to eliminate branches on the same value
6859 // later on.
6860 set_control(fast_path);
6861 result = intcon(expected_val);
6862 }
6863 // Stop profiling.
6864 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6865 // By replacing method body with profile data (represented as ProfileBooleanNode
6866 // on IR level) we effectively disable profiling.
6867 // It enables full speed execution once optimized code is generated.
6868 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6869 C->record_for_igvn(profile);
6870 set_result(profile);
6871 return true;
6872 } else {
6873 // Continue profiling.
6874 // Profile data isn't available at the moment. So, execute method's bytecode version.
6875 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6876 // is compiled and counters aren't available since corresponding MethodHandle
6877 // isn't a compile-time constant.
6878 return false;
6879 }
6880 }
6881
6882 bool LibraryCallKit::inline_isCompileConstant() {
6883 Node* n = argument(0);
6884 set_result(n->is_Con() ? intcon(1) : intcon(0));
6885 return true;
6886 }