1 /*
2 * Copyright (c) 1999, 2019, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/systemDictionary.hpp"
29 #include "classfile/vmSymbols.hpp"
30 #include "compiler/compileBroker.hpp"
31 #include "compiler/compileLog.hpp"
32 #include "gc/shared/barrierSet.hpp"
33 #include "jfr/support/jfrIntrinsics.hpp"
34 #include "memory/resourceArea.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/callGenerator.hpp"
40 #include "opto/castnode.hpp"
41 #include "opto/cfgnode.hpp"
42 #include "opto/convertnode.hpp"
43 #include "opto/countbitsnode.hpp"
44 #include "opto/intrinsicnode.hpp"
45 #include "opto/idealKit.hpp"
46 #include "opto/mathexactnode.hpp"
47 #include "opto/movenode.hpp"
48 #include "opto/mulnode.hpp"
49 #include "opto/narrowptrnode.hpp"
50 #include "opto/opaquenode.hpp"
51 #include "opto/parse.hpp"
52 #include "opto/runtime.hpp"
53 #include "opto/rootnode.hpp"
54 #include "opto/subnode.hpp"
55 #include "prims/nativeLookup.hpp"
56 #include "prims/unsafe.hpp"
57 #include "runtime/objectMonitor.hpp"
58 #include "runtime/sharedRuntime.hpp"
59 #include "utilities/macros.hpp"
60
61
62 class LibraryIntrinsic : public InlineCallGenerator {
63 // Extend the set of intrinsics known to the runtime:
64 public:
65 private:
66 bool _is_virtual;
67 bool _does_virtual_dispatch;
68 int8_t _predicates_count; // Intrinsic is predicated by several conditions
69 int8_t _last_predicate; // Last generated predicate
70 vmIntrinsics::ID _intrinsic_id;
71
72 public:
73 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
74 : InlineCallGenerator(m),
75 _is_virtual(is_virtual),
76 _does_virtual_dispatch(does_virtual_dispatch),
77 _predicates_count((int8_t)predicates_count),
78 _last_predicate((int8_t)-1),
79 _intrinsic_id(id)
80 {
81 }
82 virtual bool is_intrinsic() const { return true; }
83 virtual bool is_virtual() const { return _is_virtual; }
84 virtual bool is_predicated() const { return _predicates_count > 0; }
85 virtual int predicates_count() const { return _predicates_count; }
86 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; }
87 virtual JVMState* generate(JVMState* jvms);
88 virtual Node* generate_predicate(JVMState* jvms, int predicate);
89 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
90 };
91
92
93 // Local helper class for LibraryIntrinsic:
94 class LibraryCallKit : public GraphKit {
95 private:
96 LibraryIntrinsic* _intrinsic; // the library intrinsic being called
97 Node* _result; // the result node, if any
98 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted
99
100 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type);
101
102 public:
103 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
104 : GraphKit(jvms),
105 _intrinsic(intrinsic),
106 _result(NULL)
107 {
108 // Check if this is a root compile. In that case we don't have a caller.
109 if (!jvms->has_method()) {
110 _reexecute_sp = sp();
111 } else {
112 // Find out how many arguments the interpreter needs when deoptimizing
113 // and save the stack pointer value so it can used by uncommon_trap.
114 // We find the argument count by looking at the declared signature.
115 bool ignored_will_link;
116 ciSignature* declared_signature = NULL;
117 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
118 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
119 _reexecute_sp = sp() + nargs; // "push" arguments back on stack
120 }
121 }
122
123 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
124
125 ciMethod* caller() const { return jvms()->method(); }
126 int bci() const { return jvms()->bci(); }
127 LibraryIntrinsic* intrinsic() const { return _intrinsic; }
128 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); }
129 ciMethod* callee() const { return _intrinsic->method(); }
130
131 bool try_to_inline(int predicate);
132 Node* try_to_predicate(int predicate);
133
134 void push_result() {
135 // Push the result onto the stack.
136 if (!stopped() && result() != NULL) {
137 BasicType bt = result()->bottom_type()->basic_type();
138 push_node(bt, result());
139 }
140 }
141
142 private:
143 void fatal_unexpected_iid(vmIntrinsics::ID iid) {
144 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid));
145 }
146
147 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
148 void set_result(RegionNode* region, PhiNode* value);
149 Node* result() { return _result; }
150
151 virtual int reexecute_sp() { return _reexecute_sp; }
152
153 // Helper functions to inline natives
154 Node* generate_guard(Node* test, RegionNode* region, float true_prob);
155 Node* generate_slow_guard(Node* test, RegionNode* region);
156 Node* generate_fair_guard(Node* test, RegionNode* region);
157 Node* generate_negative_guard(Node* index, RegionNode* region,
158 // resulting CastII of index:
159 Node* *pos_index = NULL);
160 Node* generate_limit_guard(Node* offset, Node* subseq_length,
161 Node* array_length,
162 RegionNode* region);
163 void generate_string_range_check(Node* array, Node* offset,
164 Node* length, bool char_count);
165 Node* generate_current_thread(Node* &tls_output);
166 Node* load_mirror_from_klass(Node* klass);
167 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
168 RegionNode* region, int null_path,
169 int offset);
170 Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
171 RegionNode* region, int null_path) {
172 int offset = java_lang_Class::klass_offset_in_bytes();
173 return load_klass_from_mirror_common(mirror, never_see_null,
174 region, null_path,
175 offset);
176 }
177 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
178 RegionNode* region, int null_path) {
179 int offset = java_lang_Class::array_klass_offset_in_bytes();
180 return load_klass_from_mirror_common(mirror, never_see_null,
181 region, null_path,
182 offset);
183 }
184 Node* generate_access_flags_guard(Node* kls,
185 int modifier_mask, int modifier_bits,
186 RegionNode* region);
187 Node* generate_interface_guard(Node* kls, RegionNode* region);
188 Node* generate_array_guard(Node* kls, RegionNode* region) {
189 return generate_array_guard_common(kls, region, false, false);
190 }
191 Node* generate_non_array_guard(Node* kls, RegionNode* region) {
192 return generate_array_guard_common(kls, region, false, true);
193 }
194 Node* generate_objArray_guard(Node* kls, RegionNode* region) {
195 return generate_array_guard_common(kls, region, true, false);
196 }
197 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
198 return generate_array_guard_common(kls, region, true, true);
199 }
200 Node* generate_array_guard_common(Node* kls, RegionNode* region,
201 bool obj_array, bool not_array);
202 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
203 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
204 bool is_virtual = false, bool is_static = false);
205 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
206 return generate_method_call(method_id, false, true);
207 }
208 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
209 return generate_method_call(method_id, true, false);
210 }
211 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
212 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
213
214 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
215 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
216 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
217 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
218 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
219 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
220 bool inline_string_indexOfChar();
221 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
222 bool inline_string_toBytesU();
223 bool inline_string_getCharsU();
224 bool inline_string_copy(bool compress);
225 bool inline_string_char_access(bool is_store);
226 Node* round_double_node(Node* n);
227 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
228 bool inline_math_native(vmIntrinsics::ID id);
229 bool inline_math(vmIntrinsics::ID id);
230 bool inline_double_math(vmIntrinsics::ID id);
231 template <typename OverflowOp>
232 bool inline_math_overflow(Node* arg1, Node* arg2);
233 void inline_math_mathExact(Node* math, Node* test);
234 bool inline_math_addExactI(bool is_increment);
235 bool inline_math_addExactL(bool is_increment);
236 bool inline_math_multiplyExactI();
237 bool inline_math_multiplyExactL();
238 bool inline_math_multiplyHigh();
239 bool inline_math_negateExactI();
240 bool inline_math_negateExactL();
241 bool inline_math_subtractExactI(bool is_decrement);
242 bool inline_math_subtractExactL(bool is_decrement);
243 bool inline_min_max(vmIntrinsics::ID id);
244 bool inline_notify(vmIntrinsics::ID id);
245 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
246 // This returns Type::AnyPtr, RawPtr, or OopPtr.
247 int classify_unsafe_addr(Node* &base, Node* &offset, BasicType type);
248 Node* make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type = T_ILLEGAL, bool can_cast = false);
249
250 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
251 DecoratorSet mo_decorator_for_access_kind(AccessKind kind);
252 bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
253 static bool klass_needs_init_guard(Node* kls);
254 bool inline_unsafe_allocate();
255 bool inline_unsafe_newArray(bool uninitialized);
256 bool inline_unsafe_writeback0();
257 bool inline_unsafe_writebackSync0(bool is_pre);
258 bool inline_unsafe_copyMemory();
259 bool inline_native_currentThread();
260
261 bool inline_native_time_funcs(address method, const char* funcName);
262 #ifdef JFR_HAVE_INTRINSICS
263 bool inline_native_classID();
264 bool inline_native_getEventWriter();
265 #endif
266 bool inline_native_Class_query(vmIntrinsics::ID id);
267 bool inline_native_subtype_check();
268 bool inline_native_getLength();
269 bool inline_array_copyOf(bool is_copyOfRange);
270 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
271 bool inline_preconditions_checkIndex();
272 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array);
273 bool inline_native_clone(bool is_virtual);
274 bool inline_native_Reflection_getCallerClass();
275 // Helper function for inlining native object hash method
276 bool inline_native_hashcode(bool is_virtual, bool is_static);
277 bool inline_native_getClass();
278
279 // Helper functions for inlining arraycopy
280 bool inline_arraycopy();
281 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
282 RegionNode* slow_region);
283 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
284 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp,
285 uint new_idx);
286
287 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
288 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
289 bool inline_unsafe_fence(vmIntrinsics::ID id);
290 bool inline_onspinwait();
291 bool inline_fp_conversions(vmIntrinsics::ID id);
292 bool inline_number_methods(vmIntrinsics::ID id);
293 bool inline_reference_get();
294 bool inline_Class_cast();
295 bool inline_aescrypt_Block(vmIntrinsics::ID id);
296 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
297 bool inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id);
298 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
299 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
300 Node* inline_electronicCodeBook_AESCrypt_predicate(bool decrypting);
301 Node* inline_counterMode_AESCrypt_predicate();
302 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
303 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
304 bool inline_ghash_processBlocks();
305 bool inline_base64_encodeBlock();
306 bool inline_sha_implCompress(vmIntrinsics::ID id);
307 bool inline_digestBase_implCompressMB(int predicate);
308 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
309 bool long_state, address stubAddr, const char *stubName,
310 Node* src_start, Node* ofs, Node* limit);
311 Node* get_state_from_sha_object(Node *sha_object);
312 Node* get_state_from_sha5_object(Node *sha_object);
313 Node* inline_digestBase_implCompressMB_predicate(int predicate);
314 bool inline_encodeISOArray();
315 bool inline_updateCRC32();
316 bool inline_updateBytesCRC32();
317 bool inline_updateByteBufferCRC32();
318 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
319 bool inline_updateBytesCRC32C();
320 bool inline_updateDirectByteBufferCRC32C();
321 bool inline_updateBytesAdler32();
322 bool inline_updateByteBufferAdler32();
323 bool inline_multiplyToLen();
324 bool inline_hasNegatives();
325 bool inline_squareToLen();
326 bool inline_mulAdd();
327 bool inline_montgomeryMultiply();
328 bool inline_montgomerySquare();
329 bool inline_vectorizedMismatch();
330 bool inline_fma(vmIntrinsics::ID id);
331 bool inline_character_compare(vmIntrinsics::ID id);
332 bool inline_fp_min_max(vmIntrinsics::ID id);
333
334 bool inline_profileBoolean();
335 bool inline_isCompileConstant();
336 void clear_upper_avx() {
337 #ifdef X86
338 if (UseAVX >= 2) {
339 C->set_clear_upper_avx(true);
340 }
341 #endif
342 }
343 };
344
345 //---------------------------make_vm_intrinsic----------------------------
346 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
347 vmIntrinsics::ID id = m->intrinsic_id();
348 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
349
350 if (!m->is_loaded()) {
351 // Do not attempt to inline unloaded methods.
352 return NULL;
353 }
354
355 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
356 bool is_available = false;
357
358 {
359 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
360 // the compiler must transition to '_thread_in_vm' state because both
361 // methods access VM-internal data.
362 VM_ENTRY_MARK;
363 methodHandle mh(THREAD, m->get_Method());
364 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
365 !C->directive()->is_intrinsic_disabled(mh) &&
366 !vmIntrinsics::is_disabled_by_flags(mh);
367
368 }
369
370 if (is_available) {
371 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
372 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
373 return new LibraryIntrinsic(m, is_virtual,
374 vmIntrinsics::predicates_needed(id),
375 vmIntrinsics::does_virtual_dispatch(id),
376 (vmIntrinsics::ID) id);
377 } else {
378 return NULL;
379 }
380 }
381
382 //----------------------register_library_intrinsics-----------------------
383 // Initialize this file's data structures, for each Compile instance.
384 void Compile::register_library_intrinsics() {
385 // Nothing to do here.
386 }
387
388 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
389 LibraryCallKit kit(jvms, this);
390 Compile* C = kit.C;
391 int nodes = C->unique();
392 #ifndef PRODUCT
393 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
394 char buf[1000];
395 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
396 tty->print_cr("Intrinsic %s", str);
397 }
398 #endif
399 ciMethod* callee = kit.callee();
400 const int bci = kit.bci();
401
402 // Try to inline the intrinsic.
403 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
404 kit.try_to_inline(_last_predicate)) {
405 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
406 : "(intrinsic)";
407 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
408 if (C->print_intrinsics() || C->print_inlining()) {
409 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
410 }
411 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
412 if (C->log()) {
413 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
414 vmIntrinsics::name_at(intrinsic_id()),
415 (is_virtual() ? " virtual='1'" : ""),
416 C->unique() - nodes);
417 }
418 // Push the result from the inlined method onto the stack.
419 kit.push_result();
420 C->print_inlining_update(this);
421 return kit.transfer_exceptions_into_jvms();
422 }
423
424 // The intrinsic bailed out
425 if (jvms->has_method()) {
426 // Not a root compile.
427 const char* msg;
428 if (callee->intrinsic_candidate()) {
429 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
430 } else {
431 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
432 : "failed to inline (intrinsic), method not annotated";
433 }
434 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
435 if (C->print_intrinsics() || C->print_inlining()) {
436 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
437 }
438 } else {
439 // Root compile
440 ResourceMark rm;
441 stringStream msg_stream;
442 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
443 vmIntrinsics::name_at(intrinsic_id()),
444 is_virtual() ? " (virtual)" : "", bci);
445 const char *msg = msg_stream.as_string();
446 log_debug(jit, inlining)("%s", msg);
447 if (C->print_intrinsics() || C->print_inlining()) {
448 tty->print("%s", msg);
449 }
450 }
451 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
452 C->print_inlining_update(this);
453 return NULL;
454 }
455
456 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
457 LibraryCallKit kit(jvms, this);
458 Compile* C = kit.C;
459 int nodes = C->unique();
460 _last_predicate = predicate;
461 #ifndef PRODUCT
462 assert(is_predicated() && predicate < predicates_count(), "sanity");
463 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
464 char buf[1000];
465 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
466 tty->print_cr("Predicate for intrinsic %s", str);
467 }
468 #endif
469 ciMethod* callee = kit.callee();
470 const int bci = kit.bci();
471
472 Node* slow_ctl = kit.try_to_predicate(predicate);
473 if (!kit.failing()) {
474 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
475 : "(intrinsic, predicate)";
476 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
477 if (C->print_intrinsics() || C->print_inlining()) {
478 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
479 }
480 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
481 if (C->log()) {
482 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
483 vmIntrinsics::name_at(intrinsic_id()),
484 (is_virtual() ? " virtual='1'" : ""),
485 C->unique() - nodes);
486 }
487 return slow_ctl; // Could be NULL if the check folds.
488 }
489
490 // The intrinsic bailed out
491 if (jvms->has_method()) {
492 // Not a root compile.
493 const char* msg = "failed to generate predicate for intrinsic";
494 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
495 if (C->print_intrinsics() || C->print_inlining()) {
496 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
497 }
498 } else {
499 // Root compile
500 ResourceMark rm;
501 stringStream msg_stream;
502 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
503 vmIntrinsics::name_at(intrinsic_id()),
504 is_virtual() ? " (virtual)" : "", bci);
505 const char *msg = msg_stream.as_string();
506 log_debug(jit, inlining)("%s", msg);
507 if (C->print_intrinsics() || C->print_inlining()) {
508 C->print_inlining_stream()->print("%s", msg);
509 }
510 }
511 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
512 return NULL;
513 }
514
515 bool LibraryCallKit::try_to_inline(int predicate) {
516 // Handle symbolic names for otherwise undistinguished boolean switches:
517 const bool is_store = true;
518 const bool is_compress = true;
519 const bool is_static = true;
520 const bool is_volatile = true;
521
522 if (!jvms()->has_method()) {
523 // Root JVMState has a null method.
524 assert(map()->memory()->Opcode() == Op_Parm, "");
525 // Insert the memory aliasing node
526 set_all_memory(reset_memory());
527 }
528 assert(merged_memory(), "");
529
530
531 switch (intrinsic_id()) {
532 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
533 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
534 case vmIntrinsics::_getClass: return inline_native_getClass();
535
536 case vmIntrinsics::_ceil:
537 case vmIntrinsics::_floor:
538 case vmIntrinsics::_rint:
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
754 #ifdef JFR_HAVE_INTRINSICS
755 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime");
756 case vmIntrinsics::_getClassId: return inline_native_classID();
757 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
758 #endif
759 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
760 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
761 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
762 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
763 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
764 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
765 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
766 case vmIntrinsics::_getLength: return inline_native_getLength();
767 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
768 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
769 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
770 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
771 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
772 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
773
774 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
775 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
776
777 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
778
779 case vmIntrinsics::_isInstance:
780 case vmIntrinsics::_getModifiers:
781 case vmIntrinsics::_isInterface:
782 case vmIntrinsics::_isArray:
783 case vmIntrinsics::_isPrimitive:
784 case vmIntrinsics::_getSuperclass:
785 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
786
787 case vmIntrinsics::_floatToRawIntBits:
788 case vmIntrinsics::_floatToIntBits:
789 case vmIntrinsics::_intBitsToFloat:
790 case vmIntrinsics::_doubleToRawLongBits:
791 case vmIntrinsics::_doubleToLongBits:
792 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
793
794 case vmIntrinsics::_numberOfLeadingZeros_i:
795 case vmIntrinsics::_numberOfLeadingZeros_l:
796 case vmIntrinsics::_numberOfTrailingZeros_i:
797 case vmIntrinsics::_numberOfTrailingZeros_l:
798 case vmIntrinsics::_bitCount_i:
799 case vmIntrinsics::_bitCount_l:
800 case vmIntrinsics::_reverseBytes_i:
801 case vmIntrinsics::_reverseBytes_l:
802 case vmIntrinsics::_reverseBytes_s:
803 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
804
805 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
806
807 case vmIntrinsics::_Reference_get: return inline_reference_get();
808
809 case vmIntrinsics::_Class_cast: return inline_Class_cast();
810
811 case vmIntrinsics::_aescrypt_encryptBlock:
812 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
813
814 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
815 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
816 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
817
818 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
819 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
820 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
821
822 case vmIntrinsics::_counterMode_AESCrypt:
823 return inline_counterMode_AESCrypt(intrinsic_id());
824
825 case vmIntrinsics::_sha_implCompress:
826 case vmIntrinsics::_sha2_implCompress:
827 case vmIntrinsics::_sha5_implCompress:
828 return inline_sha_implCompress(intrinsic_id());
829
830 case vmIntrinsics::_digestBase_implCompressMB:
831 return inline_digestBase_implCompressMB(predicate);
832
833 case vmIntrinsics::_multiplyToLen:
834 return inline_multiplyToLen();
835
836 case vmIntrinsics::_squareToLen:
837 return inline_squareToLen();
838
839 case vmIntrinsics::_mulAdd:
840 return inline_mulAdd();
841
842 case vmIntrinsics::_montgomeryMultiply:
843 return inline_montgomeryMultiply();
844 case vmIntrinsics::_montgomerySquare:
845 return inline_montgomerySquare();
846
847 case vmIntrinsics::_vectorizedMismatch:
848 return inline_vectorizedMismatch();
849
850 case vmIntrinsics::_ghash_processBlocks:
851 return inline_ghash_processBlocks();
852 case vmIntrinsics::_base64_encodeBlock:
853 return inline_base64_encodeBlock();
854
855 case vmIntrinsics::_encodeISOArray:
856 case vmIntrinsics::_encodeByteISOArray:
857 return inline_encodeISOArray();
858
859 case vmIntrinsics::_updateCRC32:
860 return inline_updateCRC32();
861 case vmIntrinsics::_updateBytesCRC32:
862 return inline_updateBytesCRC32();
863 case vmIntrinsics::_updateByteBufferCRC32:
864 return inline_updateByteBufferCRC32();
865
866 case vmIntrinsics::_updateBytesCRC32C:
867 return inline_updateBytesCRC32C();
868 case vmIntrinsics::_updateDirectByteBufferCRC32C:
869 return inline_updateDirectByteBufferCRC32C();
870
871 case vmIntrinsics::_updateBytesAdler32:
872 return inline_updateBytesAdler32();
873 case vmIntrinsics::_updateByteBufferAdler32:
874 return inline_updateByteBufferAdler32();
875
876 case vmIntrinsics::_profileBoolean:
877 return inline_profileBoolean();
878 case vmIntrinsics::_isCompileConstant:
879 return inline_isCompileConstant();
880
881 case vmIntrinsics::_hasNegatives:
882 return inline_hasNegatives();
883
884 case vmIntrinsics::_fmaD:
885 case vmIntrinsics::_fmaF:
886 return inline_fma(intrinsic_id());
887
888 case vmIntrinsics::_isDigit:
889 case vmIntrinsics::_isLowerCase:
890 case vmIntrinsics::_isUpperCase:
891 case vmIntrinsics::_isWhitespace:
892 return inline_character_compare(intrinsic_id());
893
894 case vmIntrinsics::_maxF:
895 case vmIntrinsics::_minF:
896 case vmIntrinsics::_maxD:
897 case vmIntrinsics::_minD:
898 return inline_fp_min_max(intrinsic_id());
899
900 default:
901 // If you get here, it may be that someone has added a new intrinsic
902 // to the list in vmSymbols.hpp without implementing it here.
903 #ifndef PRODUCT
904 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
905 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
906 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
907 }
908 #endif
909 return false;
910 }
911 }
912
913 Node* LibraryCallKit::try_to_predicate(int predicate) {
914 if (!jvms()->has_method()) {
915 // Root JVMState has a null method.
916 assert(map()->memory()->Opcode() == Op_Parm, "");
917 // Insert the memory aliasing node
918 set_all_memory(reset_memory());
919 }
920 assert(merged_memory(), "");
921
922 switch (intrinsic_id()) {
923 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
924 return inline_cipherBlockChaining_AESCrypt_predicate(false);
925 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
926 return inline_cipherBlockChaining_AESCrypt_predicate(true);
927 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
928 return inline_electronicCodeBook_AESCrypt_predicate(false);
929 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
930 return inline_electronicCodeBook_AESCrypt_predicate(true);
931 case vmIntrinsics::_counterMode_AESCrypt:
932 return inline_counterMode_AESCrypt_predicate();
933 case vmIntrinsics::_digestBase_implCompressMB:
934 return inline_digestBase_implCompressMB_predicate(predicate);
935
936 default:
937 // If you get here, it may be that someone has added a new intrinsic
938 // to the list in vmSymbols.hpp without implementing it here.
939 #ifndef PRODUCT
940 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
941 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
942 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
943 }
944 #endif
945 Node* slow_ctl = control();
946 set_control(top()); // No fast path instrinsic
947 return slow_ctl;
948 }
949 }
950
951 //------------------------------set_result-------------------------------
952 // Helper function for finishing intrinsics.
953 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
954 record_for_igvn(region);
955 set_control(_gvn.transform(region));
956 set_result( _gvn.transform(value));
957 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
958 }
959
960 //------------------------------generate_guard---------------------------
961 // Helper function for generating guarded fast-slow graph structures.
962 // The given 'test', if true, guards a slow path. If the test fails
963 // then a fast path can be taken. (We generally hope it fails.)
964 // In all cases, GraphKit::control() is updated to the fast path.
965 // The returned value represents the control for the slow path.
966 // The return value is never 'top'; it is either a valid control
967 // or NULL if it is obvious that the slow path can never be taken.
968 // Also, if region and the slow control are not NULL, the slow edge
969 // is appended to the region.
970 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
971 if (stopped()) {
972 // Already short circuited.
973 return NULL;
974 }
975
976 // Build an if node and its projections.
977 // If test is true we take the slow path, which we assume is uncommon.
978 if (_gvn.type(test) == TypeInt::ZERO) {
979 // The slow branch is never taken. No need to build this guard.
980 return NULL;
981 }
982
983 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
984
985 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
986 if (if_slow == top()) {
987 // The slow branch is never taken. No need to build this guard.
988 return NULL;
989 }
990
991 if (region != NULL)
992 region->add_req(if_slow);
993
994 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
995 set_control(if_fast);
996
997 return if_slow;
998 }
999
1000 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
1001 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
1002 }
1003 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
1004 return generate_guard(test, region, PROB_FAIR);
1005 }
1006
1007 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
1008 Node* *pos_index) {
1009 if (stopped())
1010 return NULL; // already stopped
1011 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
1012 return NULL; // index is already adequately typed
1013 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
1014 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1015 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
1016 if (is_neg != NULL && pos_index != NULL) {
1017 // Emulate effect of Parse::adjust_map_after_if.
1018 Node* ccast = new CastIINode(index, TypeInt::POS);
1019 ccast->set_req(0, control());
1020 (*pos_index) = _gvn.transform(ccast);
1021 }
1022 return is_neg;
1023 }
1024
1025 // Make sure that 'position' is a valid limit index, in [0..length].
1026 // There are two equivalent plans for checking this:
1027 // A. (offset + copyLength) unsigned<= arrayLength
1028 // B. offset <= (arrayLength - copyLength)
1029 // We require that all of the values above, except for the sum and
1030 // difference, are already known to be non-negative.
1031 // Plan A is robust in the face of overflow, if offset and copyLength
1032 // are both hugely positive.
1033 //
1034 // Plan B is less direct and intuitive, but it does not overflow at
1035 // all, since the difference of two non-negatives is always
1036 // representable. Whenever Java methods must perform the equivalent
1037 // check they generally use Plan B instead of Plan A.
1038 // For the moment we use Plan A.
1039 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
1040 Node* subseq_length,
1041 Node* array_length,
1042 RegionNode* region) {
1043 if (stopped())
1044 return NULL; // already stopped
1045 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
1046 if (zero_offset && subseq_length->eqv_uncast(array_length))
1047 return NULL; // common case of whole-array copy
1048 Node* last = subseq_length;
1049 if (!zero_offset) // last += offset
1050 last = _gvn.transform(new AddINode(last, offset));
1051 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
1052 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1053 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
1054 return is_over;
1055 }
1056
1057 // Emit range checks for the given String.value byte array
1058 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
1059 if (stopped()) {
1060 return; // already stopped
1061 }
1062 RegionNode* bailout = new RegionNode(1);
1063 record_for_igvn(bailout);
1064 if (char_count) {
1065 // Convert char count to byte count
1066 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1067 }
1068
1069 // Offset and count must not be negative
1070 generate_negative_guard(offset, bailout);
1071 generate_negative_guard(count, bailout);
1072 // Offset + count must not exceed length of array
1073 generate_limit_guard(offset, count, load_array_length(array), bailout);
1074
1075 if (bailout->req() > 1) {
1076 PreserveJVMState pjvms(this);
1077 set_control(_gvn.transform(bailout));
1078 uncommon_trap(Deoptimization::Reason_intrinsic,
1079 Deoptimization::Action_maybe_recompile);
1080 }
1081 }
1082
1083 //--------------------------generate_current_thread--------------------
1084 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1085 ciKlass* thread_klass = env()->Thread_klass();
1086 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1087 Node* thread = _gvn.transform(new ThreadLocalNode());
1088 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1089 Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered);
1090 tls_output = thread;
1091 return threadObj;
1092 }
1093
1094
1095 //------------------------------make_string_method_node------------------------
1096 // Helper method for String intrinsic functions. This version is called with
1097 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1098 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1099 // containing the lengths of str1 and str2.
1100 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1101 Node* result = NULL;
1102 switch (opcode) {
1103 case Op_StrIndexOf:
1104 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1105 str1_start, cnt1, str2_start, cnt2, ae);
1106 break;
1107 case Op_StrComp:
1108 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1109 str1_start, cnt1, str2_start, cnt2, ae);
1110 break;
1111 case Op_StrEquals:
1112 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1113 // Use the constant length if there is one because optimized match rule may exist.
1114 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1115 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1116 break;
1117 default:
1118 ShouldNotReachHere();
1119 return NULL;
1120 }
1121
1122 // All these intrinsics have checks.
1123 C->set_has_split_ifs(true); // Has chance for split-if optimization
1124 clear_upper_avx();
1125
1126 return _gvn.transform(result);
1127 }
1128
1129 //------------------------------inline_string_compareTo------------------------
1130 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1131 Node* arg1 = argument(0);
1132 Node* arg2 = argument(1);
1133
1134 arg1 = must_be_not_null(arg1, true);
1135 arg2 = must_be_not_null(arg2, true);
1136
1137 arg1 = access_resolve(arg1, ACCESS_READ);
1138 arg2 = access_resolve(arg2, ACCESS_READ);
1139
1140 // Get start addr and length of first argument
1141 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1142 Node* arg1_cnt = load_array_length(arg1);
1143
1144 // Get start addr and length of second argument
1145 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1146 Node* arg2_cnt = load_array_length(arg2);
1147
1148 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1149 set_result(result);
1150 return true;
1151 }
1152
1153 //------------------------------inline_string_equals------------------------
1154 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1155 Node* arg1 = argument(0);
1156 Node* arg2 = argument(1);
1157
1158 // paths (plus control) merge
1159 RegionNode* region = new RegionNode(3);
1160 Node* phi = new PhiNode(region, TypeInt::BOOL);
1161
1162 if (!stopped()) {
1163
1164 arg1 = must_be_not_null(arg1, true);
1165 arg2 = must_be_not_null(arg2, true);
1166
1167 arg1 = access_resolve(arg1, ACCESS_READ);
1168 arg2 = access_resolve(arg2, ACCESS_READ);
1169
1170 // Get start addr and length of first argument
1171 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1172 Node* arg1_cnt = load_array_length(arg1);
1173
1174 // Get start addr and length of second argument
1175 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1176 Node* arg2_cnt = load_array_length(arg2);
1177
1178 // Check for arg1_cnt != arg2_cnt
1179 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1180 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1181 Node* if_ne = generate_slow_guard(bol, NULL);
1182 if (if_ne != NULL) {
1183 phi->init_req(2, intcon(0));
1184 region->init_req(2, if_ne);
1185 }
1186
1187 // Check for count == 0 is done by assembler code for StrEquals.
1188
1189 if (!stopped()) {
1190 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1191 phi->init_req(1, equals);
1192 region->init_req(1, control());
1193 }
1194 }
1195
1196 // post merge
1197 set_control(_gvn.transform(region));
1198 record_for_igvn(region);
1199
1200 set_result(_gvn.transform(phi));
1201 return true;
1202 }
1203
1204 //------------------------------inline_array_equals----------------------------
1205 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1206 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1207 Node* arg1 = argument(0);
1208 Node* arg2 = argument(1);
1209
1210 arg1 = access_resolve(arg1, ACCESS_READ);
1211 arg2 = access_resolve(arg2, ACCESS_READ);
1212
1213 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1214 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1215 clear_upper_avx();
1216
1217 return true;
1218 }
1219
1220 //------------------------------inline_hasNegatives------------------------------
1221 bool LibraryCallKit::inline_hasNegatives() {
1222 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1223 return false;
1224 }
1225
1226 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1227 // no receiver since it is static method
1228 Node* ba = argument(0);
1229 Node* offset = argument(1);
1230 Node* len = argument(2);
1231
1232 ba = must_be_not_null(ba, true);
1233
1234 // Range checks
1235 generate_string_range_check(ba, offset, len, false);
1236 if (stopped()) {
1237 return true;
1238 }
1239 ba = access_resolve(ba, ACCESS_READ);
1240 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1241 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1242 set_result(_gvn.transform(result));
1243 return true;
1244 }
1245
1246 bool LibraryCallKit::inline_preconditions_checkIndex() {
1247 Node* index = argument(0);
1248 Node* length = argument(1);
1249 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1250 return false;
1251 }
1252
1253 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1254 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1255
1256 {
1257 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1258 uncommon_trap(Deoptimization::Reason_intrinsic,
1259 Deoptimization::Action_make_not_entrant);
1260 }
1261
1262 if (stopped()) {
1263 return false;
1264 }
1265
1266 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1267 BoolTest::mask btest = BoolTest::lt;
1268 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1269 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1270 _gvn.set_type(rc, rc->Value(&_gvn));
1271 if (!rc_bool->is_Con()) {
1272 record_for_igvn(rc);
1273 }
1274 set_control(_gvn.transform(new IfTrueNode(rc)));
1275 {
1276 PreserveJVMState pjvms(this);
1277 set_control(_gvn.transform(new IfFalseNode(rc)));
1278 uncommon_trap(Deoptimization::Reason_range_check,
1279 Deoptimization::Action_make_not_entrant);
1280 }
1281
1282 if (stopped()) {
1283 return false;
1284 }
1285
1286 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1287 result->set_req(0, control());
1288 result = _gvn.transform(result);
1289 set_result(result);
1290 replace_in_map(index, result);
1291 clear_upper_avx();
1292 return true;
1293 }
1294
1295 //------------------------------inline_string_indexOf------------------------
1296 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1297 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1298 return false;
1299 }
1300 Node* src = argument(0);
1301 Node* tgt = argument(1);
1302
1303 // Make the merge point
1304 RegionNode* result_rgn = new RegionNode(4);
1305 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1306
1307 src = must_be_not_null(src, true);
1308 tgt = must_be_not_null(tgt, true);
1309
1310 src = access_resolve(src, ACCESS_READ);
1311 tgt = access_resolve(tgt, ACCESS_READ);
1312
1313 // Get start addr and length of source string
1314 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1315 Node* src_count = load_array_length(src);
1316
1317 // Get start addr and length of substring
1318 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1319 Node* tgt_count = load_array_length(tgt);
1320
1321 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1322 // Divide src size by 2 if String is UTF16 encoded
1323 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1324 }
1325 if (ae == StrIntrinsicNode::UU) {
1326 // Divide substring size by 2 if String is UTF16 encoded
1327 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1328 }
1329
1330 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1331 if (result != NULL) {
1332 result_phi->init_req(3, result);
1333 result_rgn->init_req(3, control());
1334 }
1335 set_control(_gvn.transform(result_rgn));
1336 record_for_igvn(result_rgn);
1337 set_result(_gvn.transform(result_phi));
1338
1339 return true;
1340 }
1341
1342 //-----------------------------inline_string_indexOf-----------------------
1343 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1344 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1345 return false;
1346 }
1347 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1348 return false;
1349 }
1350 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1351 Node* src = argument(0); // byte[]
1352 Node* src_count = argument(1); // char count
1353 Node* tgt = argument(2); // byte[]
1354 Node* tgt_count = argument(3); // char count
1355 Node* from_index = argument(4); // char index
1356
1357 src = must_be_not_null(src, true);
1358 tgt = must_be_not_null(tgt, true);
1359
1360 src = access_resolve(src, ACCESS_READ);
1361 tgt = access_resolve(tgt, ACCESS_READ);
1362
1363 // Multiply byte array index by 2 if String is UTF16 encoded
1364 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1365 src_count = _gvn.transform(new SubINode(src_count, from_index));
1366 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1367 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1368
1369 // Range checks
1370 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1371 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1372 if (stopped()) {
1373 return true;
1374 }
1375
1376 RegionNode* region = new RegionNode(5);
1377 Node* phi = new PhiNode(region, TypeInt::INT);
1378
1379 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1380 if (result != NULL) {
1381 // The result is index relative to from_index if substring was found, -1 otherwise.
1382 // Generate code which will fold into cmove.
1383 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1384 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1385
1386 Node* if_lt = generate_slow_guard(bol, NULL);
1387 if (if_lt != NULL) {
1388 // result == -1
1389 phi->init_req(3, result);
1390 region->init_req(3, if_lt);
1391 }
1392 if (!stopped()) {
1393 result = _gvn.transform(new AddINode(result, from_index));
1394 phi->init_req(4, result);
1395 region->init_req(4, control());
1396 }
1397 }
1398
1399 set_control(_gvn.transform(region));
1400 record_for_igvn(region);
1401 set_result(_gvn.transform(phi));
1402 clear_upper_avx();
1403
1404 return true;
1405 }
1406
1407 // Create StrIndexOfNode with fast path checks
1408 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1409 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1410 // Check for substr count > string count
1411 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1412 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1413 Node* if_gt = generate_slow_guard(bol, NULL);
1414 if (if_gt != NULL) {
1415 phi->init_req(1, intcon(-1));
1416 region->init_req(1, if_gt);
1417 }
1418 if (!stopped()) {
1419 // Check for substr count == 0
1420 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1421 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1422 Node* if_zero = generate_slow_guard(bol, NULL);
1423 if (if_zero != NULL) {
1424 phi->init_req(2, intcon(0));
1425 region->init_req(2, if_zero);
1426 }
1427 }
1428 if (!stopped()) {
1429 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1430 }
1431 return NULL;
1432 }
1433
1434 //-----------------------------inline_string_indexOfChar-----------------------
1435 bool LibraryCallKit::inline_string_indexOfChar() {
1436 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1437 return false;
1438 }
1439 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1440 return false;
1441 }
1442 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1443 Node* src = argument(0); // byte[]
1444 Node* tgt = argument(1); // tgt is int ch
1445 Node* from_index = argument(2);
1446 Node* max = argument(3);
1447
1448 src = must_be_not_null(src, true);
1449 src = access_resolve(src, ACCESS_READ);
1450
1451 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1452 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1453 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1454
1455 // Range checks
1456 generate_string_range_check(src, src_offset, src_count, true);
1457 if (stopped()) {
1458 return true;
1459 }
1460
1461 RegionNode* region = new RegionNode(3);
1462 Node* phi = new PhiNode(region, TypeInt::INT);
1463
1464 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1465 C->set_has_split_ifs(true); // Has chance for split-if optimization
1466 _gvn.transform(result);
1467
1468 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1469 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1470
1471 Node* if_lt = generate_slow_guard(bol, NULL);
1472 if (if_lt != NULL) {
1473 // result == -1
1474 phi->init_req(2, result);
1475 region->init_req(2, if_lt);
1476 }
1477 if (!stopped()) {
1478 result = _gvn.transform(new AddINode(result, from_index));
1479 phi->init_req(1, result);
1480 region->init_req(1, control());
1481 }
1482 set_control(_gvn.transform(region));
1483 record_for_igvn(region);
1484 set_result(_gvn.transform(phi));
1485
1486 return true;
1487 }
1488 //---------------------------inline_string_copy---------------------
1489 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1490 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1491 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1492 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1493 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1494 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1495 bool LibraryCallKit::inline_string_copy(bool compress) {
1496 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1497 return false;
1498 }
1499 int nargs = 5; // 2 oops, 3 ints
1500 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1501
1502 Node* src = argument(0);
1503 Node* src_offset = argument(1);
1504 Node* dst = argument(2);
1505 Node* dst_offset = argument(3);
1506 Node* length = argument(4);
1507
1508 // Check for allocation before we add nodes that would confuse
1509 // tightly_coupled_allocation()
1510 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1511
1512 // Figure out the size and type of the elements we will be copying.
1513 const Type* src_type = src->Value(&_gvn);
1514 const Type* dst_type = dst->Value(&_gvn);
1515 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1516 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1517 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1518 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1519 "Unsupported array types for inline_string_copy");
1520
1521 src = must_be_not_null(src, true);
1522 dst = must_be_not_null(dst, true);
1523
1524 // Convert char[] offsets to byte[] offsets
1525 bool convert_src = (compress && src_elem == T_BYTE);
1526 bool convert_dst = (!compress && dst_elem == T_BYTE);
1527 if (convert_src) {
1528 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1529 } else if (convert_dst) {
1530 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1531 }
1532
1533 // Range checks
1534 generate_string_range_check(src, src_offset, length, convert_src);
1535 generate_string_range_check(dst, dst_offset, length, convert_dst);
1536 if (stopped()) {
1537 return true;
1538 }
1539
1540 src = access_resolve(src, ACCESS_READ);
1541 dst = access_resolve(dst, ACCESS_WRITE);
1542
1543 Node* src_start = array_element_address(src, src_offset, src_elem);
1544 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1545 // 'src_start' points to src array + scaled offset
1546 // 'dst_start' points to dst array + scaled offset
1547 Node* count = NULL;
1548 if (compress) {
1549 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1550 } else {
1551 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1552 }
1553
1554 if (alloc != NULL) {
1555 if (alloc->maybe_set_complete(&_gvn)) {
1556 // "You break it, you buy it."
1557 InitializeNode* init = alloc->initialization();
1558 assert(init->is_complete(), "we just did this");
1559 init->set_complete_with_arraycopy();
1560 assert(dst->is_CheckCastPP(), "sanity");
1561 assert(dst->in(0)->in(0) == init, "dest pinned");
1562 }
1563 // Do not let stores that initialize this object be reordered with
1564 // a subsequent store that would make this object accessible by
1565 // other threads.
1566 // Record what AllocateNode this StoreStore protects so that
1567 // escape analysis can go from the MemBarStoreStoreNode to the
1568 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1569 // based on the escape status of the AllocateNode.
1570 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1571 }
1572 if (compress) {
1573 set_result(_gvn.transform(count));
1574 }
1575 clear_upper_avx();
1576
1577 return true;
1578 }
1579
1580 #ifdef _LP64
1581 #define XTOP ,top() /*additional argument*/
1582 #else //_LP64
1583 #define XTOP /*no additional argument*/
1584 #endif //_LP64
1585
1586 //------------------------inline_string_toBytesU--------------------------
1587 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1588 bool LibraryCallKit::inline_string_toBytesU() {
1589 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1590 return false;
1591 }
1592 // Get the arguments.
1593 Node* value = argument(0);
1594 Node* offset = argument(1);
1595 Node* length = argument(2);
1596
1597 Node* newcopy = NULL;
1598
1599 // Set the original stack and the reexecute bit for the interpreter to reexecute
1600 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1601 { PreserveReexecuteState preexecs(this);
1602 jvms()->set_should_reexecute(true);
1603
1604 // Check if a null path was taken unconditionally.
1605 value = null_check(value);
1606
1607 RegionNode* bailout = new RegionNode(1);
1608 record_for_igvn(bailout);
1609
1610 // Range checks
1611 generate_negative_guard(offset, bailout);
1612 generate_negative_guard(length, bailout);
1613 generate_limit_guard(offset, length, load_array_length(value), bailout);
1614 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1615 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1616
1617 if (bailout->req() > 1) {
1618 PreserveJVMState pjvms(this);
1619 set_control(_gvn.transform(bailout));
1620 uncommon_trap(Deoptimization::Reason_intrinsic,
1621 Deoptimization::Action_maybe_recompile);
1622 }
1623 if (stopped()) {
1624 return true;
1625 }
1626
1627 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1628 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1629 newcopy = new_array(klass_node, size, 0); // no arguments to push
1630 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1631
1632 // Calculate starting addresses.
1633 value = access_resolve(value, ACCESS_READ);
1634 Node* src_start = array_element_address(value, offset, T_CHAR);
1635 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1636
1637 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1638 const TypeInt* toffset = gvn().type(offset)->is_int();
1639 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1640
1641 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1642 const char* copyfunc_name = "arraycopy";
1643 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1644 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1645 OptoRuntime::fast_arraycopy_Type(),
1646 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1647 src_start, dst_start, ConvI2X(length) XTOP);
1648 // Do not let reads from the cloned object float above the arraycopy.
1649 if (alloc != NULL) {
1650 if (alloc->maybe_set_complete(&_gvn)) {
1651 // "You break it, you buy it."
1652 InitializeNode* init = alloc->initialization();
1653 assert(init->is_complete(), "we just did this");
1654 init->set_complete_with_arraycopy();
1655 assert(newcopy->is_CheckCastPP(), "sanity");
1656 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1657 }
1658 // Do not let stores that initialize this object be reordered with
1659 // a subsequent store that would make this object accessible by
1660 // other threads.
1661 // Record what AllocateNode this StoreStore protects so that
1662 // escape analysis can go from the MemBarStoreStoreNode to the
1663 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1664 // based on the escape status of the AllocateNode.
1665 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1666 } else {
1667 insert_mem_bar(Op_MemBarCPUOrder);
1668 }
1669 } // original reexecute is set back here
1670
1671 C->set_has_split_ifs(true); // Has chance for split-if optimization
1672 if (!stopped()) {
1673 set_result(newcopy);
1674 }
1675 clear_upper_avx();
1676
1677 return true;
1678 }
1679
1680 //------------------------inline_string_getCharsU--------------------------
1681 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1682 bool LibraryCallKit::inline_string_getCharsU() {
1683 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1684 return false;
1685 }
1686
1687 // Get the arguments.
1688 Node* src = argument(0);
1689 Node* src_begin = argument(1);
1690 Node* src_end = argument(2); // exclusive offset (i < src_end)
1691 Node* dst = argument(3);
1692 Node* dst_begin = argument(4);
1693
1694 // Check for allocation before we add nodes that would confuse
1695 // tightly_coupled_allocation()
1696 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1697
1698 // Check if a null path was taken unconditionally.
1699 src = null_check(src);
1700 dst = null_check(dst);
1701 if (stopped()) {
1702 return true;
1703 }
1704
1705 // Get length and convert char[] offset to byte[] offset
1706 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1707 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1708
1709 // Range checks
1710 generate_string_range_check(src, src_begin, length, true);
1711 generate_string_range_check(dst, dst_begin, length, false);
1712 if (stopped()) {
1713 return true;
1714 }
1715
1716 if (!stopped()) {
1717 src = access_resolve(src, ACCESS_READ);
1718 dst = access_resolve(dst, ACCESS_WRITE);
1719
1720 // Calculate starting addresses.
1721 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1722 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1723
1724 // Check if array addresses are aligned to HeapWordSize
1725 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1726 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1727 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1728 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1729
1730 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1731 const char* copyfunc_name = "arraycopy";
1732 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1733 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1734 OptoRuntime::fast_arraycopy_Type(),
1735 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1736 src_start, dst_start, ConvI2X(length) XTOP);
1737 // Do not let reads from the cloned object float above the arraycopy.
1738 if (alloc != NULL) {
1739 if (alloc->maybe_set_complete(&_gvn)) {
1740 // "You break it, you buy it."
1741 InitializeNode* init = alloc->initialization();
1742 assert(init->is_complete(), "we just did this");
1743 init->set_complete_with_arraycopy();
1744 assert(dst->is_CheckCastPP(), "sanity");
1745 assert(dst->in(0)->in(0) == init, "dest pinned");
1746 }
1747 // Do not let stores that initialize this object be reordered with
1748 // a subsequent store that would make this object accessible by
1749 // other threads.
1750 // Record what AllocateNode this StoreStore protects so that
1751 // escape analysis can go from the MemBarStoreStoreNode to the
1752 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1753 // based on the escape status of the AllocateNode.
1754 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1755 } else {
1756 insert_mem_bar(Op_MemBarCPUOrder);
1757 }
1758 }
1759
1760 C->set_has_split_ifs(true); // Has chance for split-if optimization
1761 return true;
1762 }
1763
1764 //----------------------inline_string_char_access----------------------------
1765 // Store/Load char to/from byte[] array.
1766 // static void StringUTF16.putChar(byte[] val, int index, int c)
1767 // static char StringUTF16.getChar(byte[] val, int index)
1768 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1769 Node* value = argument(0);
1770 Node* index = argument(1);
1771 Node* ch = is_store ? argument(2) : NULL;
1772
1773 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1774 // correctly requires matched array shapes.
1775 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1776 "sanity: byte[] and char[] bases agree");
1777 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1778 "sanity: byte[] and char[] scales agree");
1779
1780 // Bail when getChar over constants is requested: constant folding would
1781 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1782 // Java method would constant fold nicely instead.
1783 if (!is_store && value->is_Con() && index->is_Con()) {
1784 return false;
1785 }
1786
1787 value = must_be_not_null(value, true);
1788 value = access_resolve(value, is_store ? ACCESS_WRITE : ACCESS_READ);
1789
1790 Node* adr = array_element_address(value, index, T_CHAR);
1791 if (adr->is_top()) {
1792 return false;
1793 }
1794 if (is_store) {
1795 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1796 } else {
1797 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
1798 set_result(ch);
1799 }
1800 return true;
1801 }
1802
1803 //--------------------------round_double_node--------------------------------
1804 // Round a double node if necessary.
1805 Node* LibraryCallKit::round_double_node(Node* n) {
1806 if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1807 n = _gvn.transform(new RoundDoubleNode(0, n));
1808 return n;
1809 }
1810
1811 //------------------------------inline_math-----------------------------------
1812 // public static double Math.abs(double)
1813 // public static double Math.sqrt(double)
1814 // public static double Math.log(double)
1815 // public static double Math.log10(double)
1816 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1817 Node* arg = round_double_node(argument(0));
1818 Node* n = NULL;
1819 switch (id) {
1820 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1821 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1822 case vmIntrinsics::_ceil: n = new RoundDoubleModeNode(arg, makecon(TypeInt::make(2))); break;
1823 case vmIntrinsics::_floor: n = new RoundDoubleModeNode(arg, makecon(TypeInt::make(1))); break;
1824 case vmIntrinsics::_rint: n = new RoundDoubleModeNode(arg, makecon(TypeInt::make(0))); break;
1825 default: fatal_unexpected_iid(id); break;
1826 }
1827 set_result(_gvn.transform(n));
1828 return true;
1829 }
1830
1831 //------------------------------inline_math-----------------------------------
1832 // public static float Math.abs(float)
1833 // public static int Math.abs(int)
1834 // public static long Math.abs(long)
1835 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1836 Node* arg = argument(0);
1837 Node* n = NULL;
1838 switch (id) {
1839 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1840 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1841 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1842 default: fatal_unexpected_iid(id); break;
1843 }
1844 set_result(_gvn.transform(n));
1845 return true;
1846 }
1847
1848 //------------------------------runtime_math-----------------------------
1849 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1850 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1851 "must be (DD)D or (D)D type");
1852
1853 // Inputs
1854 Node* a = round_double_node(argument(0));
1855 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1856
1857 const TypePtr* no_memory_effects = NULL;
1858 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1859 no_memory_effects,
1860 a, top(), b, b ? top() : NULL);
1861 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1862 #ifdef ASSERT
1863 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1864 assert(value_top == top(), "second value must be top");
1865 #endif
1866
1867 set_result(value);
1868 return true;
1869 }
1870
1871 //------------------------------inline_math_native-----------------------------
1872 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1873 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1874 switch (id) {
1875 // These intrinsics are not properly supported on all hardware
1876 case vmIntrinsics::_dsin:
1877 return StubRoutines::dsin() != NULL ?
1878 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1879 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1880 case vmIntrinsics::_dcos:
1881 return StubRoutines::dcos() != NULL ?
1882 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1883 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1884 case vmIntrinsics::_dtan:
1885 return StubRoutines::dtan() != NULL ?
1886 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1887 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1888 case vmIntrinsics::_dlog:
1889 return StubRoutines::dlog() != NULL ?
1890 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1891 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1892 case vmIntrinsics::_dlog10:
1893 return StubRoutines::dlog10() != NULL ?
1894 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1895 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1896
1897 // These intrinsics are supported on all hardware
1898 case vmIntrinsics::_ceil:
1899 case vmIntrinsics::_floor:
1900 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1901 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1902 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1903 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1904 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1905 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1906
1907 case vmIntrinsics::_dexp:
1908 return StubRoutines::dexp() != NULL ?
1909 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1910 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1911 case vmIntrinsics::_dpow: {
1912 Node* exp = round_double_node(argument(2));
1913 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1914 if (d != NULL && d->getd() == 2.0) {
1915 // Special case: pow(x, 2.0) => x * x
1916 Node* base = round_double_node(argument(0));
1917 set_result(_gvn.transform(new MulDNode(base, base)));
1918 return true;
1919 }
1920 return StubRoutines::dpow() != NULL ?
1921 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1922 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1923 }
1924 #undef FN_PTR
1925
1926 // These intrinsics are not yet correctly implemented
1927 case vmIntrinsics::_datan2:
1928 return false;
1929
1930 default:
1931 fatal_unexpected_iid(id);
1932 return false;
1933 }
1934 }
1935
1936 static bool is_simple_name(Node* n) {
1937 return (n->req() == 1 // constant
1938 || (n->is_Type() && n->as_Type()->type()->singleton())
1939 || n->is_Proj() // parameter or return value
1940 || n->is_Phi() // local of some sort
1941 );
1942 }
1943
1944 //----------------------------inline_notify-----------------------------------*
1945 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1946 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1947 address func;
1948 if (id == vmIntrinsics::_notify) {
1949 func = OptoRuntime::monitor_notify_Java();
1950 } else {
1951 func = OptoRuntime::monitor_notifyAll_Java();
1952 }
1953 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1954 make_slow_call_ex(call, env()->Throwable_klass(), false);
1955 return true;
1956 }
1957
1958
1959 //----------------------------inline_min_max-----------------------------------
1960 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1961 set_result(generate_min_max(id, argument(0), argument(1)));
1962 return true;
1963 }
1964
1965 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1966 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1967 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1968 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1969 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1970
1971 {
1972 PreserveJVMState pjvms(this);
1973 PreserveReexecuteState preexecs(this);
1974 jvms()->set_should_reexecute(true);
1975
1976 set_control(slow_path);
1977 set_i_o(i_o());
1978
1979 uncommon_trap(Deoptimization::Reason_intrinsic,
1980 Deoptimization::Action_none);
1981 }
1982
1983 set_control(fast_path);
1984 set_result(math);
1985 }
1986
1987 template <typename OverflowOp>
1988 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1989 typedef typename OverflowOp::MathOp MathOp;
1990
1991 MathOp* mathOp = new MathOp(arg1, arg2);
1992 Node* operation = _gvn.transform( mathOp );
1993 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1994 inline_math_mathExact(operation, ofcheck);
1995 return true;
1996 }
1997
1998 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1999 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2000 }
2001
2002 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2003 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2004 }
2005
2006 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2007 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2008 }
2009
2010 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2011 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2012 }
2013
2014 bool LibraryCallKit::inline_math_negateExactI() {
2015 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2016 }
2017
2018 bool LibraryCallKit::inline_math_negateExactL() {
2019 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2020 }
2021
2022 bool LibraryCallKit::inline_math_multiplyExactI() {
2023 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2024 }
2025
2026 bool LibraryCallKit::inline_math_multiplyExactL() {
2027 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2028 }
2029
2030 bool LibraryCallKit::inline_math_multiplyHigh() {
2031 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2032 return true;
2033 }
2034
2035 Node*
2036 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2037 // These are the candidate return value:
2038 Node* xvalue = x0;
2039 Node* yvalue = y0;
2040
2041 if (xvalue == yvalue) {
2042 return xvalue;
2043 }
2044
2045 bool want_max = (id == vmIntrinsics::_max);
2046
2047 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
2048 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
2049 if (txvalue == NULL || tyvalue == NULL) return top();
2050 // This is not really necessary, but it is consistent with a
2051 // hypothetical MaxINode::Value method:
2052 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
2053
2054 // %%% This folding logic should (ideally) be in a different place.
2055 // Some should be inside IfNode, and there to be a more reliable
2056 // transformation of ?: style patterns into cmoves. We also want
2057 // more powerful optimizations around cmove and min/max.
2058
2059 // Try to find a dominating comparison of these guys.
2060 // It can simplify the index computation for Arrays.copyOf
2061 // and similar uses of System.arraycopy.
2062 // First, compute the normalized version of CmpI(x, y).
2063 int cmp_op = Op_CmpI;
2064 Node* xkey = xvalue;
2065 Node* ykey = yvalue;
2066 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
2067 if (ideal_cmpxy->is_Cmp()) {
2068 // E.g., if we have CmpI(length - offset, count),
2069 // it might idealize to CmpI(length, count + offset)
2070 cmp_op = ideal_cmpxy->Opcode();
2071 xkey = ideal_cmpxy->in(1);
2072 ykey = ideal_cmpxy->in(2);
2073 }
2074
2075 // Start by locating any relevant comparisons.
2076 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
2077 Node* cmpxy = NULL;
2078 Node* cmpyx = NULL;
2079 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
2080 Node* cmp = start_from->fast_out(k);
2081 if (cmp->outcnt() > 0 && // must have prior uses
2082 cmp->in(0) == NULL && // must be context-independent
2083 cmp->Opcode() == cmp_op) { // right kind of compare
2084 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
2085 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
2086 }
2087 }
2088
2089 const int NCMPS = 2;
2090 Node* cmps[NCMPS] = { cmpxy, cmpyx };
2091 int cmpn;
2092 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2093 if (cmps[cmpn] != NULL) break; // find a result
2094 }
2095 if (cmpn < NCMPS) {
2096 // Look for a dominating test that tells us the min and max.
2097 int depth = 0; // Limit search depth for speed
2098 Node* dom = control();
2099 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2100 if (++depth >= 100) break;
2101 Node* ifproj = dom;
2102 if (!ifproj->is_Proj()) continue;
2103 Node* iff = ifproj->in(0);
2104 if (!iff->is_If()) continue;
2105 Node* bol = iff->in(1);
2106 if (!bol->is_Bool()) continue;
2107 Node* cmp = bol->in(1);
2108 if (cmp == NULL) continue;
2109 for (cmpn = 0; cmpn < NCMPS; cmpn++)
2110 if (cmps[cmpn] == cmp) break;
2111 if (cmpn == NCMPS) continue;
2112 BoolTest::mask btest = bol->as_Bool()->_test._test;
2113 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
2114 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2115 // At this point, we know that 'x btest y' is true.
2116 switch (btest) {
2117 case BoolTest::eq:
2118 // They are proven equal, so we can collapse the min/max.
2119 // Either value is the answer. Choose the simpler.
2120 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2121 return yvalue;
2122 return xvalue;
2123 case BoolTest::lt: // x < y
2124 case BoolTest::le: // x <= y
2125 return (want_max ? yvalue : xvalue);
2126 case BoolTest::gt: // x > y
2127 case BoolTest::ge: // x >= y
2128 return (want_max ? xvalue : yvalue);
2129 default:
2130 break;
2131 }
2132 }
2133 }
2134
2135 // We failed to find a dominating test.
2136 // Let's pick a test that might GVN with prior tests.
2137 Node* best_bol = NULL;
2138 BoolTest::mask best_btest = BoolTest::illegal;
2139 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2140 Node* cmp = cmps[cmpn];
2141 if (cmp == NULL) continue;
2142 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2143 Node* bol = cmp->fast_out(j);
2144 if (!bol->is_Bool()) continue;
2145 BoolTest::mask btest = bol->as_Bool()->_test._test;
2146 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2147 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2148 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2149 best_bol = bol->as_Bool();
2150 best_btest = btest;
2151 }
2152 }
2153 }
2154
2155 Node* answer_if_true = NULL;
2156 Node* answer_if_false = NULL;
2157 switch (best_btest) {
2158 default:
2159 if (cmpxy == NULL)
2160 cmpxy = ideal_cmpxy;
2161 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2162 // and fall through:
2163 case BoolTest::lt: // x < y
2164 case BoolTest::le: // x <= y
2165 answer_if_true = (want_max ? yvalue : xvalue);
2166 answer_if_false = (want_max ? xvalue : yvalue);
2167 break;
2168 case BoolTest::gt: // x > y
2169 case BoolTest::ge: // x >= y
2170 answer_if_true = (want_max ? xvalue : yvalue);
2171 answer_if_false = (want_max ? yvalue : xvalue);
2172 break;
2173 }
2174
2175 jint hi, lo;
2176 if (want_max) {
2177 // We can sharpen the minimum.
2178 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2179 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2180 } else {
2181 // We can sharpen the maximum.
2182 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2183 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2184 }
2185
2186 // Use a flow-free graph structure, to avoid creating excess control edges
2187 // which could hinder other optimizations.
2188 // Since Math.min/max is often used with arraycopy, we want
2189 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2190 Node* cmov = CMoveNode::make(NULL, best_bol,
2191 answer_if_false, answer_if_true,
2192 TypeInt::make(lo, hi, widen));
2193
2194 return _gvn.transform(cmov);
2195
2196 /*
2197 // This is not as desirable as it may seem, since Min and Max
2198 // nodes do not have a full set of optimizations.
2199 // And they would interfere, anyway, with 'if' optimizations
2200 // and with CMoveI canonical forms.
2201 switch (id) {
2202 case vmIntrinsics::_min:
2203 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2204 case vmIntrinsics::_max:
2205 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2206 default:
2207 ShouldNotReachHere();
2208 }
2209 */
2210 }
2211
2212 inline int
2213 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2214 const TypePtr* base_type = TypePtr::NULL_PTR;
2215 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2216 if (base_type == NULL) {
2217 // Unknown type.
2218 return Type::AnyPtr;
2219 } else if (base_type == TypePtr::NULL_PTR) {
2220 // Since this is a NULL+long form, we have to switch to a rawptr.
2221 base = _gvn.transform(new CastX2PNode(offset));
2222 offset = MakeConX(0);
2223 return Type::RawPtr;
2224 } else if (base_type->base() == Type::RawPtr) {
2225 return Type::RawPtr;
2226 } else if (base_type->isa_oopptr()) {
2227 // Base is never null => always a heap address.
2228 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2229 return Type::OopPtr;
2230 }
2231 // Offset is small => always a heap address.
2232 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2233 if (offset_type != NULL &&
2234 base_type->offset() == 0 && // (should always be?)
2235 offset_type->_lo >= 0 &&
2236 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2237 return Type::OopPtr;
2238 } else if (type == T_OBJECT) {
2239 // off heap access to an oop doesn't make any sense. Has to be on
2240 // heap.
2241 return Type::OopPtr;
2242 }
2243 // Otherwise, it might either be oop+off or NULL+addr.
2244 return Type::AnyPtr;
2245 } else {
2246 // No information:
2247 return Type::AnyPtr;
2248 }
2249 }
2250
2251 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type, bool can_cast) {
2252 Node* uncasted_base = base;
2253 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2254 if (kind == Type::RawPtr) {
2255 return basic_plus_adr(top(), uncasted_base, offset);
2256 } else if (kind == Type::AnyPtr) {
2257 assert(base == uncasted_base, "unexpected base change");
2258 if (can_cast) {
2259 if (!_gvn.type(base)->speculative_maybe_null() &&
2260 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2261 // According to profiling, this access is always on
2262 // heap. Casting the base to not null and thus avoiding membars
2263 // around the access should allow better optimizations
2264 Node* null_ctl = top();
2265 base = null_check_oop(base, &null_ctl, true, true, true);
2266 assert(null_ctl->is_top(), "no null control here");
2267 return basic_plus_adr(base, offset);
2268 } else if (_gvn.type(base)->speculative_always_null() &&
2269 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2270 // According to profiling, this access is always off
2271 // heap.
2272 base = null_assert(base);
2273 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2274 offset = MakeConX(0);
2275 return basic_plus_adr(top(), raw_base, offset);
2276 }
2277 }
2278 // We don't know if it's an on heap or off heap access. Fall back
2279 // to raw memory access.
2280 base = access_resolve(base, decorators);
2281 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2282 return basic_plus_adr(top(), raw, offset);
2283 } else {
2284 assert(base == uncasted_base, "unexpected base change");
2285 // We know it's an on heap access so base can't be null
2286 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2287 base = must_be_not_null(base, true);
2288 }
2289 return basic_plus_adr(base, offset);
2290 }
2291 }
2292
2293 //--------------------------inline_number_methods-----------------------------
2294 // inline int Integer.numberOfLeadingZeros(int)
2295 // inline int Long.numberOfLeadingZeros(long)
2296 //
2297 // inline int Integer.numberOfTrailingZeros(int)
2298 // inline int Long.numberOfTrailingZeros(long)
2299 //
2300 // inline int Integer.bitCount(int)
2301 // inline int Long.bitCount(long)
2302 //
2303 // inline char Character.reverseBytes(char)
2304 // inline short Short.reverseBytes(short)
2305 // inline int Integer.reverseBytes(int)
2306 // inline long Long.reverseBytes(long)
2307 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2308 Node* arg = argument(0);
2309 Node* n = NULL;
2310 switch (id) {
2311 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2312 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2313 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2314 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2315 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2316 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2317 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2318 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2319 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2320 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2321 default: fatal_unexpected_iid(id); break;
2322 }
2323 set_result(_gvn.transform(n));
2324 return true;
2325 }
2326
2327 //----------------------------inline_unsafe_access----------------------------
2328
2329 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2330 // Attempt to infer a sharper value type from the offset and base type.
2331 ciKlass* sharpened_klass = NULL;
2332
2333 // See if it is an instance field, with an object type.
2334 if (alias_type->field() != NULL) {
2335 if (alias_type->field()->type()->is_klass()) {
2336 sharpened_klass = alias_type->field()->type()->as_klass();
2337 }
2338 }
2339
2340 // See if it is a narrow oop array.
2341 if (adr_type->isa_aryptr()) {
2342 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2343 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2344 if (elem_type != NULL) {
2345 sharpened_klass = elem_type->klass();
2346 }
2347 }
2348 }
2349
2350 // The sharpened class might be unloaded if there is no class loader
2351 // contraint in place.
2352 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2353 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2354
2355 #ifndef PRODUCT
2356 if (C->print_intrinsics() || C->print_inlining()) {
2357 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2358 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2359 }
2360 #endif
2361 // Sharpen the value type.
2362 return tjp;
2363 }
2364 return NULL;
2365 }
2366
2367 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2368 switch (kind) {
2369 case Relaxed:
2370 return MO_UNORDERED;
2371 case Opaque:
2372 return MO_RELAXED;
2373 case Acquire:
2374 return MO_ACQUIRE;
2375 case Release:
2376 return MO_RELEASE;
2377 case Volatile:
2378 return MO_SEQ_CST;
2379 default:
2380 ShouldNotReachHere();
2381 return 0;
2382 }
2383 }
2384
2385 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2386 if (callee()->is_static()) return false; // caller must have the capability!
2387 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2388 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2389 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2390 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2391
2392 if (is_reference_type(type)) {
2393 decorators |= ON_UNKNOWN_OOP_REF;
2394 }
2395
2396 if (unaligned) {
2397 decorators |= C2_UNALIGNED;
2398 }
2399
2400 #ifndef PRODUCT
2401 {
2402 ResourceMark rm;
2403 // Check the signatures.
2404 ciSignature* sig = callee()->signature();
2405 #ifdef ASSERT
2406 if (!is_store) {
2407 // Object getReference(Object base, int/long offset), etc.
2408 BasicType rtype = sig->return_type()->basic_type();
2409 assert(rtype == type, "getter must return the expected value");
2410 assert(sig->count() == 2, "oop getter has 2 arguments");
2411 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2412 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2413 } else {
2414 // void putReference(Object base, int/long offset, Object x), etc.
2415 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2416 assert(sig->count() == 3, "oop putter has 3 arguments");
2417 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2418 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2419 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2420 assert(vtype == type, "putter must accept the expected value");
2421 }
2422 #endif // ASSERT
2423 }
2424 #endif //PRODUCT
2425
2426 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2427
2428 Node* receiver = argument(0); // type: oop
2429
2430 // Build address expression.
2431 Node* adr;
2432 Node* heap_base_oop = top();
2433 Node* offset = top();
2434 Node* val;
2435
2436 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2437 Node* base = argument(1); // type: oop
2438 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2439 offset = argument(2); // type: long
2440 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2441 // to be plain byte offsets, which are also the same as those accepted
2442 // by oopDesc::field_addr.
2443 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2444 "fieldOffset must be byte-scaled");
2445 // 32-bit machines ignore the high half!
2446 offset = ConvL2X(offset);
2447 adr = make_unsafe_address(base, offset, is_store ? ACCESS_WRITE : ACCESS_READ, type, kind == Relaxed);
2448
2449 if (_gvn.type(base)->isa_ptr() != TypePtr::NULL_PTR) {
2450 heap_base_oop = base;
2451 } else if (type == T_OBJECT) {
2452 return false; // off-heap oop accesses are not supported
2453 }
2454
2455 // Can base be NULL? Otherwise, always on-heap access.
2456 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2457
2458 if (!can_access_non_heap) {
2459 decorators |= IN_HEAP;
2460 }
2461
2462 val = is_store ? argument(4) : NULL;
2463
2464 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2465 if (adr_type == TypePtr::NULL_PTR) {
2466 return false; // off-heap access with zero address
2467 }
2468
2469 // Try to categorize the address.
2470 Compile::AliasType* alias_type = C->alias_type(adr_type);
2471 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2472
2473 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2474 alias_type->adr_type() == TypeAryPtr::RANGE) {
2475 return false; // not supported
2476 }
2477
2478 bool mismatched = false;
2479 BasicType bt = alias_type->basic_type();
2480 if (bt != T_ILLEGAL) {
2481 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2482 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2483 // Alias type doesn't differentiate between byte[] and boolean[]).
2484 // Use address type to get the element type.
2485 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2486 }
2487 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2488 // accessing an array field with getReference is not a mismatch
2489 bt = T_OBJECT;
2490 }
2491 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2492 // Don't intrinsify mismatched object accesses
2493 return false;
2494 }
2495 mismatched = (bt != type);
2496 } else if (alias_type->adr_type()->isa_oopptr()) {
2497 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2498 }
2499
2500 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2501
2502 if (mismatched) {
2503 decorators |= C2_MISMATCHED;
2504 }
2505
2506 // First guess at the value type.
2507 const Type *value_type = Type::get_const_basic_type(type);
2508
2509 // Figure out the memory ordering.
2510 decorators |= mo_decorator_for_access_kind(kind);
2511
2512 if (!is_store && type == T_OBJECT) {
2513 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2514 if (tjp != NULL) {
2515 value_type = tjp;
2516 }
2517 }
2518
2519 receiver = null_check(receiver);
2520 if (stopped()) {
2521 return true;
2522 }
2523 // Heap pointers get a null-check from the interpreter,
2524 // as a courtesy. However, this is not guaranteed by Unsafe,
2525 // and it is not possible to fully distinguish unintended nulls
2526 // from intended ones in this API.
2527
2528 if (!is_store) {
2529 Node* p = NULL;
2530 // Try to constant fold a load from a constant field
2531 ciField* field = alias_type->field();
2532 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2533 // final or stable field
2534 p = make_constant_from_field(field, heap_base_oop);
2535 }
2536
2537 if (p == NULL) { // Could not constant fold the load
2538 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2539 // Normalize the value returned by getBoolean in the following cases
2540 if (type == T_BOOLEAN &&
2541 (mismatched ||
2542 heap_base_oop == top() || // - heap_base_oop is NULL or
2543 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2544 // and the unsafe access is made to large offset
2545 // (i.e., larger than the maximum offset necessary for any
2546 // field access)
2547 ) {
2548 IdealKit ideal = IdealKit(this);
2549 #define __ ideal.
2550 IdealVariable normalized_result(ideal);
2551 __ declarations_done();
2552 __ set(normalized_result, p);
2553 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2554 __ set(normalized_result, ideal.ConI(1));
2555 ideal.end_if();
2556 final_sync(ideal);
2557 p = __ value(normalized_result);
2558 #undef __
2559 }
2560 }
2561 if (type == T_ADDRESS) {
2562 p = gvn().transform(new CastP2XNode(NULL, p));
2563 p = ConvX2UL(p);
2564 }
2565 // The load node has the control of the preceding MemBarCPUOrder. All
2566 // following nodes will have the control of the MemBarCPUOrder inserted at
2567 // the end of this method. So, pushing the load onto the stack at a later
2568 // point is fine.
2569 set_result(p);
2570 } else {
2571 if (bt == T_ADDRESS) {
2572 // Repackage the long as a pointer.
2573 val = ConvL2X(val);
2574 val = gvn().transform(new CastX2PNode(val));
2575 }
2576 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2577 }
2578
2579 return true;
2580 }
2581
2582 //----------------------------inline_unsafe_load_store----------------------------
2583 // This method serves a couple of different customers (depending on LoadStoreKind):
2584 //
2585 // LS_cmp_swap:
2586 //
2587 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2588 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2589 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2590 //
2591 // LS_cmp_swap_weak:
2592 //
2593 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2594 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2595 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2596 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2597 //
2598 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2599 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2600 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2601 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2602 //
2603 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2604 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2605 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2606 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2607 //
2608 // LS_cmp_exchange:
2609 //
2610 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2611 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2612 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2613 //
2614 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2615 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2616 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2617 //
2618 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2619 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2620 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2621 //
2622 // LS_get_add:
2623 //
2624 // int getAndAddInt( Object o, long offset, int delta)
2625 // long getAndAddLong(Object o, long offset, long delta)
2626 //
2627 // LS_get_set:
2628 //
2629 // int getAndSet(Object o, long offset, int newValue)
2630 // long getAndSet(Object o, long offset, long newValue)
2631 // Object getAndSet(Object o, long offset, Object newValue)
2632 //
2633 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2634 // This basic scheme here is the same as inline_unsafe_access, but
2635 // differs in enough details that combining them would make the code
2636 // overly confusing. (This is a true fact! I originally combined
2637 // them, but even I was confused by it!) As much code/comments as
2638 // possible are retained from inline_unsafe_access though to make
2639 // the correspondences clearer. - dl
2640
2641 if (callee()->is_static()) return false; // caller must have the capability!
2642
2643 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2644 decorators |= mo_decorator_for_access_kind(access_kind);
2645
2646 #ifndef PRODUCT
2647 BasicType rtype;
2648 {
2649 ResourceMark rm;
2650 // Check the signatures.
2651 ciSignature* sig = callee()->signature();
2652 rtype = sig->return_type()->basic_type();
2653 switch(kind) {
2654 case LS_get_add:
2655 case LS_get_set: {
2656 // Check the signatures.
2657 #ifdef ASSERT
2658 assert(rtype == type, "get and set must return the expected type");
2659 assert(sig->count() == 3, "get and set has 3 arguments");
2660 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2661 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2662 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2663 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2664 #endif // ASSERT
2665 break;
2666 }
2667 case LS_cmp_swap:
2668 case LS_cmp_swap_weak: {
2669 // Check the signatures.
2670 #ifdef ASSERT
2671 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2672 assert(sig->count() == 4, "CAS has 4 arguments");
2673 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2674 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2675 #endif // ASSERT
2676 break;
2677 }
2678 case LS_cmp_exchange: {
2679 // Check the signatures.
2680 #ifdef ASSERT
2681 assert(rtype == type, "CAS must return the expected type");
2682 assert(sig->count() == 4, "CAS has 4 arguments");
2683 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2684 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2685 #endif // ASSERT
2686 break;
2687 }
2688 default:
2689 ShouldNotReachHere();
2690 }
2691 }
2692 #endif //PRODUCT
2693
2694 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2695
2696 // Get arguments:
2697 Node* receiver = NULL;
2698 Node* base = NULL;
2699 Node* offset = NULL;
2700 Node* oldval = NULL;
2701 Node* newval = NULL;
2702 switch(kind) {
2703 case LS_cmp_swap:
2704 case LS_cmp_swap_weak:
2705 case LS_cmp_exchange: {
2706 const bool two_slot_type = type2size[type] == 2;
2707 receiver = argument(0); // type: oop
2708 base = argument(1); // type: oop
2709 offset = argument(2); // type: long
2710 oldval = argument(4); // type: oop, int, or long
2711 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2712 break;
2713 }
2714 case LS_get_add:
2715 case LS_get_set: {
2716 receiver = argument(0); // type: oop
2717 base = argument(1); // type: oop
2718 offset = argument(2); // type: long
2719 oldval = NULL;
2720 newval = argument(4); // type: oop, int, or long
2721 break;
2722 }
2723 default:
2724 ShouldNotReachHere();
2725 }
2726
2727 // Build field offset expression.
2728 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2729 // to be plain byte offsets, which are also the same as those accepted
2730 // by oopDesc::field_addr.
2731 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2732 // 32-bit machines ignore the high half of long offsets
2733 offset = ConvL2X(offset);
2734 Node* adr = make_unsafe_address(base, offset, ACCESS_WRITE | ACCESS_READ, type, false);
2735 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2736
2737 Compile::AliasType* alias_type = C->alias_type(adr_type);
2738 BasicType bt = alias_type->basic_type();
2739 if (bt != T_ILLEGAL &&
2740 (is_reference_type(bt) != (type == T_OBJECT))) {
2741 // Don't intrinsify mismatched object accesses.
2742 return false;
2743 }
2744
2745 // For CAS, unlike inline_unsafe_access, there seems no point in
2746 // trying to refine types. Just use the coarse types here.
2747 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2748 const Type *value_type = Type::get_const_basic_type(type);
2749
2750 switch (kind) {
2751 case LS_get_set:
2752 case LS_cmp_exchange: {
2753 if (type == T_OBJECT) {
2754 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2755 if (tjp != NULL) {
2756 value_type = tjp;
2757 }
2758 }
2759 break;
2760 }
2761 case LS_cmp_swap:
2762 case LS_cmp_swap_weak:
2763 case LS_get_add:
2764 break;
2765 default:
2766 ShouldNotReachHere();
2767 }
2768
2769 // Null check receiver.
2770 receiver = null_check(receiver);
2771 if (stopped()) {
2772 return true;
2773 }
2774
2775 int alias_idx = C->get_alias_index(adr_type);
2776
2777 if (is_reference_type(type)) {
2778 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2779
2780 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2781 // could be delayed during Parse (for example, in adjust_map_after_if()).
2782 // Execute transformation here to avoid barrier generation in such case.
2783 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2784 newval = _gvn.makecon(TypePtr::NULL_PTR);
2785
2786 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2787 // Refine the value to a null constant, when it is known to be null
2788 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2789 }
2790 }
2791
2792 Node* result = NULL;
2793 switch (kind) {
2794 case LS_cmp_exchange: {
2795 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2796 oldval, newval, value_type, type, decorators);
2797 break;
2798 }
2799 case LS_cmp_swap_weak:
2800 decorators |= C2_WEAK_CMPXCHG;
2801 case LS_cmp_swap: {
2802 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2803 oldval, newval, value_type, type, decorators);
2804 break;
2805 }
2806 case LS_get_set: {
2807 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2808 newval, value_type, type, decorators);
2809 break;
2810 }
2811 case LS_get_add: {
2812 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2813 newval, value_type, type, decorators);
2814 break;
2815 }
2816 default:
2817 ShouldNotReachHere();
2818 }
2819
2820 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2821 set_result(result);
2822 return true;
2823 }
2824
2825 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2826 // Regardless of form, don't allow previous ld/st to move down,
2827 // then issue acquire, release, or volatile mem_bar.
2828 insert_mem_bar(Op_MemBarCPUOrder);
2829 switch(id) {
2830 case vmIntrinsics::_loadFence:
2831 insert_mem_bar(Op_LoadFence);
2832 return true;
2833 case vmIntrinsics::_storeFence:
2834 insert_mem_bar(Op_StoreFence);
2835 return true;
2836 case vmIntrinsics::_fullFence:
2837 insert_mem_bar(Op_MemBarVolatile);
2838 return true;
2839 default:
2840 fatal_unexpected_iid(id);
2841 return false;
2842 }
2843 }
2844
2845 bool LibraryCallKit::inline_onspinwait() {
2846 insert_mem_bar(Op_OnSpinWait);
2847 return true;
2848 }
2849
2850 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2851 if (!kls->is_Con()) {
2852 return true;
2853 }
2854 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
2855 if (klsptr == NULL) {
2856 return true;
2857 }
2858 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
2859 // don't need a guard for a klass that is already initialized
2860 return !ik->is_initialized();
2861 }
2862
2863 //----------------------------inline_unsafe_writeback0-------------------------
2864 // public native void Unsafe.writeback0(long address)
2865 bool LibraryCallKit::inline_unsafe_writeback0() {
2866 if (!Matcher::has_match_rule(Op_CacheWB)) {
2867 return false;
2868 }
2869 #ifndef PRODUCT
2870 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2871 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2872 ciSignature* sig = callee()->signature();
2873 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2874 #endif
2875 null_check_receiver(); // null-check, then ignore
2876 Node *addr = argument(1);
2877 addr = new CastX2PNode(addr);
2878 addr = _gvn.transform(addr);
2879 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2880 flush = _gvn.transform(flush);
2881 set_memory(flush, TypeRawPtr::BOTTOM);
2882 return true;
2883 }
2884
2885 //----------------------------inline_unsafe_writeback0-------------------------
2886 // public native void Unsafe.writeback0(long address)
2887 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2888 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2889 return false;
2890 }
2891 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2892 return false;
2893 }
2894 #ifndef PRODUCT
2895 assert(Matcher::has_match_rule(Op_CacheWB),
2896 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2897 : "found match rule for CacheWBPostSync but not CacheWB"));
2898
2899 #endif
2900 null_check_receiver(); // null-check, then ignore
2901 Node *sync;
2902 if (is_pre) {
2903 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2904 } else {
2905 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2906 }
2907 sync = _gvn.transform(sync);
2908 set_memory(sync, TypeRawPtr::BOTTOM);
2909 return true;
2910 }
2911
2912 //----------------------------inline_unsafe_allocate---------------------------
2913 // public native Object Unsafe.allocateInstance(Class<?> cls);
2914 bool LibraryCallKit::inline_unsafe_allocate() {
2915 if (callee()->is_static()) return false; // caller must have the capability!
2916
2917 null_check_receiver(); // null-check, then ignore
2918 Node* cls = null_check(argument(1));
2919 if (stopped()) return true;
2920
2921 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2922 kls = null_check(kls);
2923 if (stopped()) return true; // argument was like int.class
2924
2925 Node* test = NULL;
2926 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2927 // Note: The argument might still be an illegal value like
2928 // Serializable.class or Object[].class. The runtime will handle it.
2929 // But we must make an explicit check for initialization.
2930 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2931 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2932 // can generate code to load it as unsigned byte.
2933 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2934 Node* bits = intcon(InstanceKlass::fully_initialized);
2935 test = _gvn.transform(new SubINode(inst, bits));
2936 // The 'test' is non-zero if we need to take a slow path.
2937 }
2938
2939 Node* obj = new_instance(kls, test);
2940 set_result(obj);
2941 return true;
2942 }
2943
2944 //------------------------inline_native_time_funcs--------------
2945 // inline code for System.currentTimeMillis() and System.nanoTime()
2946 // these have the same type and signature
2947 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2948 const TypeFunc* tf = OptoRuntime::void_long_Type();
2949 const TypePtr* no_memory_effects = NULL;
2950 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2951 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2952 #ifdef ASSERT
2953 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2954 assert(value_top == top(), "second value must be top");
2955 #endif
2956 set_result(value);
2957 return true;
2958 }
2959
2960 #ifdef JFR_HAVE_INTRINSICS
2961
2962 /*
2963 * oop -> myklass
2964 * myklass->trace_id |= USED
2965 * return myklass->trace_id & ~0x3
2966 */
2967 bool LibraryCallKit::inline_native_classID() {
2968 Node* cls = null_check(argument(0), T_OBJECT);
2969 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2970 kls = null_check(kls, T_OBJECT);
2971
2972 ByteSize offset = KLASS_TRACE_ID_OFFSET;
2973 Node* insp = basic_plus_adr(kls, in_bytes(offset));
2974 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
2975
2976 Node* clsused = longcon(0x01l); // set the class bit
2977 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
2978 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
2979 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
2980
2981 #ifdef TRACE_ID_META_BITS
2982 Node* mbits = longcon(~TRACE_ID_META_BITS);
2983 tvalue = _gvn.transform(new AndLNode(tvalue, mbits));
2984 #endif
2985 #ifdef TRACE_ID_SHIFT
2986 Node* cbits = intcon(TRACE_ID_SHIFT);
2987 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits));
2988 #endif
2989
2990 set_result(tvalue);
2991 return true;
2992
2993 }
2994
2995 bool LibraryCallKit::inline_native_getEventWriter() {
2996 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
2997
2998 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
2999 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3000
3001 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3002
3003 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
3004 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
3005
3006 IfNode* iff_jobj_null =
3007 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
3008
3009 enum { _normal_path = 1,
3010 _null_path = 2,
3011 PATH_LIMIT };
3012
3013 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3014 PhiNode* result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3015
3016 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
3017 result_rgn->init_req(_null_path, jobj_is_null);
3018 result_val->init_req(_null_path, null());
3019
3020 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
3021 set_control(jobj_is_not_null);
3022 Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT,
3023 IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3024 result_rgn->init_req(_normal_path, control());
3025 result_val->init_req(_normal_path, res);
3026
3027 set_result(result_rgn, result_val);
3028
3029 return true;
3030 }
3031
3032 #endif // JFR_HAVE_INTRINSICS
3033
3034 //------------------------inline_native_currentThread------------------
3035 bool LibraryCallKit::inline_native_currentThread() {
3036 Node* junk = NULL;
3037 set_result(generate_current_thread(junk));
3038 return true;
3039 }
3040
3041 //---------------------------load_mirror_from_klass----------------------------
3042 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3043 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3044 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3045 Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3046 // mirror = ((OopHandle)mirror)->resolve();
3047 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3048 }
3049
3050 //-----------------------load_klass_from_mirror_common-------------------------
3051 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3052 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3053 // and branch to the given path on the region.
3054 // If never_see_null, take an uncommon trap on null, so we can optimistically
3055 // compile for the non-null case.
3056 // If the region is NULL, force never_see_null = true.
3057 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3058 bool never_see_null,
3059 RegionNode* region,
3060 int null_path,
3061 int offset) {
3062 if (region == NULL) never_see_null = true;
3063 Node* p = basic_plus_adr(mirror, offset);
3064 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3065 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3066 Node* null_ctl = top();
3067 kls = null_check_oop(kls, &null_ctl, never_see_null);
3068 if (region != NULL) {
3069 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3070 region->init_req(null_path, null_ctl);
3071 } else {
3072 assert(null_ctl == top(), "no loose ends");
3073 }
3074 return kls;
3075 }
3076
3077 //--------------------(inline_native_Class_query helpers)---------------------
3078 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3079 // Fall through if (mods & mask) == bits, take the guard otherwise.
3080 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3081 // Branch around if the given klass has the given modifier bit set.
3082 // Like generate_guard, adds a new path onto the region.
3083 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3084 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3085 Node* mask = intcon(modifier_mask);
3086 Node* bits = intcon(modifier_bits);
3087 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3088 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3089 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3090 return generate_fair_guard(bol, region);
3091 }
3092 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3093 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3094 }
3095
3096 //-------------------------inline_native_Class_query-------------------
3097 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3098 const Type* return_type = TypeInt::BOOL;
3099 Node* prim_return_value = top(); // what happens if it's a primitive class?
3100 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3101 bool expect_prim = false; // most of these guys expect to work on refs
3102
3103 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3104
3105 Node* mirror = argument(0);
3106 Node* obj = top();
3107
3108 switch (id) {
3109 case vmIntrinsics::_isInstance:
3110 // nothing is an instance of a primitive type
3111 prim_return_value = intcon(0);
3112 obj = argument(1);
3113 break;
3114 case vmIntrinsics::_getModifiers:
3115 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3116 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3117 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3118 break;
3119 case vmIntrinsics::_isInterface:
3120 prim_return_value = intcon(0);
3121 break;
3122 case vmIntrinsics::_isArray:
3123 prim_return_value = intcon(0);
3124 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3125 break;
3126 case vmIntrinsics::_isPrimitive:
3127 prim_return_value = intcon(1);
3128 expect_prim = true; // obviously
3129 break;
3130 case vmIntrinsics::_getSuperclass:
3131 prim_return_value = null();
3132 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3133 break;
3134 case vmIntrinsics::_getClassAccessFlags:
3135 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3136 return_type = TypeInt::INT; // not bool! 6297094
3137 break;
3138 default:
3139 fatal_unexpected_iid(id);
3140 break;
3141 }
3142
3143 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3144 if (mirror_con == NULL) return false; // cannot happen?
3145
3146 #ifndef PRODUCT
3147 if (C->print_intrinsics() || C->print_inlining()) {
3148 ciType* k = mirror_con->java_mirror_type();
3149 if (k) {
3150 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3151 k->print_name();
3152 tty->cr();
3153 }
3154 }
3155 #endif
3156
3157 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3158 RegionNode* region = new RegionNode(PATH_LIMIT);
3159 record_for_igvn(region);
3160 PhiNode* phi = new PhiNode(region, return_type);
3161
3162 // The mirror will never be null of Reflection.getClassAccessFlags, however
3163 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3164 // if it is. See bug 4774291.
3165
3166 // For Reflection.getClassAccessFlags(), the null check occurs in
3167 // the wrong place; see inline_unsafe_access(), above, for a similar
3168 // situation.
3169 mirror = null_check(mirror);
3170 // If mirror or obj is dead, only null-path is taken.
3171 if (stopped()) return true;
3172
3173 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3174
3175 // Now load the mirror's klass metaobject, and null-check it.
3176 // Side-effects region with the control path if the klass is null.
3177 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3178 // If kls is null, we have a primitive mirror.
3179 phi->init_req(_prim_path, prim_return_value);
3180 if (stopped()) { set_result(region, phi); return true; }
3181 bool safe_for_replace = (region->in(_prim_path) == top());
3182
3183 Node* p; // handy temp
3184 Node* null_ctl;
3185
3186 // Now that we have the non-null klass, we can perform the real query.
3187 // For constant classes, the query will constant-fold in LoadNode::Value.
3188 Node* query_value = top();
3189 switch (id) {
3190 case vmIntrinsics::_isInstance:
3191 // nothing is an instance of a primitive type
3192 query_value = gen_instanceof(obj, kls, safe_for_replace);
3193 break;
3194
3195 case vmIntrinsics::_getModifiers:
3196 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3197 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3198 break;
3199
3200 case vmIntrinsics::_isInterface:
3201 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3202 if (generate_interface_guard(kls, region) != NULL)
3203 // A guard was added. If the guard is taken, it was an interface.
3204 phi->add_req(intcon(1));
3205 // If we fall through, it's a plain class.
3206 query_value = intcon(0);
3207 break;
3208
3209 case vmIntrinsics::_isArray:
3210 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3211 if (generate_array_guard(kls, region) != NULL)
3212 // A guard was added. If the guard is taken, it was an array.
3213 phi->add_req(intcon(1));
3214 // If we fall through, it's a plain class.
3215 query_value = intcon(0);
3216 break;
3217
3218 case vmIntrinsics::_isPrimitive:
3219 query_value = intcon(0); // "normal" path produces false
3220 break;
3221
3222 case vmIntrinsics::_getSuperclass:
3223 // The rules here are somewhat unfortunate, but we can still do better
3224 // with random logic than with a JNI call.
3225 // Interfaces store null or Object as _super, but must report null.
3226 // Arrays store an intermediate super as _super, but must report Object.
3227 // Other types can report the actual _super.
3228 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3229 if (generate_interface_guard(kls, region) != NULL)
3230 // A guard was added. If the guard is taken, it was an interface.
3231 phi->add_req(null());
3232 if (generate_array_guard(kls, region) != NULL)
3233 // A guard was added. If the guard is taken, it was an array.
3234 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3235 // If we fall through, it's a plain class. Get its _super.
3236 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3237 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3238 null_ctl = top();
3239 kls = null_check_oop(kls, &null_ctl);
3240 if (null_ctl != top()) {
3241 // If the guard is taken, Object.superClass is null (both klass and mirror).
3242 region->add_req(null_ctl);
3243 phi ->add_req(null());
3244 }
3245 if (!stopped()) {
3246 query_value = load_mirror_from_klass(kls);
3247 }
3248 break;
3249
3250 case vmIntrinsics::_getClassAccessFlags:
3251 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3252 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3253 break;
3254
3255 default:
3256 fatal_unexpected_iid(id);
3257 break;
3258 }
3259
3260 // Fall-through is the normal case of a query to a real class.
3261 phi->init_req(1, query_value);
3262 region->init_req(1, control());
3263
3264 C->set_has_split_ifs(true); // Has chance for split-if optimization
3265 set_result(region, phi);
3266 return true;
3267 }
3268
3269 //-------------------------inline_Class_cast-------------------
3270 bool LibraryCallKit::inline_Class_cast() {
3271 Node* mirror = argument(0); // Class
3272 Node* obj = argument(1);
3273 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3274 if (mirror_con == NULL) {
3275 return false; // dead path (mirror->is_top()).
3276 }
3277 if (obj == NULL || obj->is_top()) {
3278 return false; // dead path
3279 }
3280 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3281
3282 // First, see if Class.cast() can be folded statically.
3283 // java_mirror_type() returns non-null for compile-time Class constants.
3284 ciType* tm = mirror_con->java_mirror_type();
3285 if (tm != NULL && tm->is_klass() &&
3286 tp != NULL && tp->klass() != NULL) {
3287 if (!tp->klass()->is_loaded()) {
3288 // Don't use intrinsic when class is not loaded.
3289 return false;
3290 } else {
3291 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3292 if (static_res == Compile::SSC_always_true) {
3293 // isInstance() is true - fold the code.
3294 set_result(obj);
3295 return true;
3296 } else if (static_res == Compile::SSC_always_false) {
3297 // Don't use intrinsic, have to throw ClassCastException.
3298 // If the reference is null, the non-intrinsic bytecode will
3299 // be optimized appropriately.
3300 return false;
3301 }
3302 }
3303 }
3304
3305 // Bailout intrinsic and do normal inlining if exception path is frequent.
3306 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3307 return false;
3308 }
3309
3310 // Generate dynamic checks.
3311 // Class.cast() is java implementation of _checkcast bytecode.
3312 // Do checkcast (Parse::do_checkcast()) optimizations here.
3313
3314 mirror = null_check(mirror);
3315 // If mirror is dead, only null-path is taken.
3316 if (stopped()) {
3317 return true;
3318 }
3319
3320 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3321 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3322 RegionNode* region = new RegionNode(PATH_LIMIT);
3323 record_for_igvn(region);
3324
3325 // Now load the mirror's klass metaobject, and null-check it.
3326 // If kls is null, we have a primitive mirror and
3327 // nothing is an instance of a primitive type.
3328 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3329
3330 Node* res = top();
3331 if (!stopped()) {
3332 Node* bad_type_ctrl = top();
3333 // Do checkcast optimizations.
3334 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3335 region->init_req(_bad_type_path, bad_type_ctrl);
3336 }
3337 if (region->in(_prim_path) != top() ||
3338 region->in(_bad_type_path) != top()) {
3339 // Let Interpreter throw ClassCastException.
3340 PreserveJVMState pjvms(this);
3341 set_control(_gvn.transform(region));
3342 uncommon_trap(Deoptimization::Reason_intrinsic,
3343 Deoptimization::Action_maybe_recompile);
3344 }
3345 if (!stopped()) {
3346 set_result(res);
3347 }
3348 return true;
3349 }
3350
3351
3352 //--------------------------inline_native_subtype_check------------------------
3353 // This intrinsic takes the JNI calls out of the heart of
3354 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3355 bool LibraryCallKit::inline_native_subtype_check() {
3356 // Pull both arguments off the stack.
3357 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3358 args[0] = argument(0);
3359 args[1] = argument(1);
3360 Node* klasses[2]; // corresponding Klasses: superk, subk
3361 klasses[0] = klasses[1] = top();
3362
3363 enum {
3364 // A full decision tree on {superc is prim, subc is prim}:
3365 _prim_0_path = 1, // {P,N} => false
3366 // {P,P} & superc!=subc => false
3367 _prim_same_path, // {P,P} & superc==subc => true
3368 _prim_1_path, // {N,P} => false
3369 _ref_subtype_path, // {N,N} & subtype check wins => true
3370 _both_ref_path, // {N,N} & subtype check loses => false
3371 PATH_LIMIT
3372 };
3373
3374 RegionNode* region = new RegionNode(PATH_LIMIT);
3375 Node* phi = new PhiNode(region, TypeInt::BOOL);
3376 record_for_igvn(region);
3377
3378 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3379 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3380 int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3381
3382 // First null-check both mirrors and load each mirror's klass metaobject.
3383 int which_arg;
3384 for (which_arg = 0; which_arg <= 1; which_arg++) {
3385 Node* arg = args[which_arg];
3386 arg = null_check(arg);
3387 if (stopped()) break;
3388 args[which_arg] = arg;
3389
3390 Node* p = basic_plus_adr(arg, class_klass_offset);
3391 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3392 klasses[which_arg] = _gvn.transform(kls);
3393 }
3394
3395 // Resolve oops to stable for CmpP below.
3396 args[0] = access_resolve(args[0], 0);
3397 args[1] = access_resolve(args[1], 0);
3398
3399 // Having loaded both klasses, test each for null.
3400 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3401 for (which_arg = 0; which_arg <= 1; which_arg++) {
3402 Node* kls = klasses[which_arg];
3403 Node* null_ctl = top();
3404 kls = null_check_oop(kls, &null_ctl, never_see_null);
3405 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3406 region->init_req(prim_path, null_ctl);
3407 if (stopped()) break;
3408 klasses[which_arg] = kls;
3409 }
3410
3411 if (!stopped()) {
3412 // now we have two reference types, in klasses[0..1]
3413 Node* subk = klasses[1]; // the argument to isAssignableFrom
3414 Node* superk = klasses[0]; // the receiver
3415 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3416 // now we have a successful reference subtype check
3417 region->set_req(_ref_subtype_path, control());
3418 }
3419
3420 // If both operands are primitive (both klasses null), then
3421 // we must return true when they are identical primitives.
3422 // It is convenient to test this after the first null klass check.
3423 set_control(region->in(_prim_0_path)); // go back to first null check
3424 if (!stopped()) {
3425 // Since superc is primitive, make a guard for the superc==subc case.
3426 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3427 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3428 generate_guard(bol_eq, region, PROB_FAIR);
3429 if (region->req() == PATH_LIMIT+1) {
3430 // A guard was added. If the added guard is taken, superc==subc.
3431 region->swap_edges(PATH_LIMIT, _prim_same_path);
3432 region->del_req(PATH_LIMIT);
3433 }
3434 region->set_req(_prim_0_path, control()); // Not equal after all.
3435 }
3436
3437 // these are the only paths that produce 'true':
3438 phi->set_req(_prim_same_path, intcon(1));
3439 phi->set_req(_ref_subtype_path, intcon(1));
3440
3441 // pull together the cases:
3442 assert(region->req() == PATH_LIMIT, "sane region");
3443 for (uint i = 1; i < region->req(); i++) {
3444 Node* ctl = region->in(i);
3445 if (ctl == NULL || ctl == top()) {
3446 region->set_req(i, top());
3447 phi ->set_req(i, top());
3448 } else if (phi->in(i) == NULL) {
3449 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3450 }
3451 }
3452
3453 set_control(_gvn.transform(region));
3454 set_result(_gvn.transform(phi));
3455 return true;
3456 }
3457
3458 //---------------------generate_array_guard_common------------------------
3459 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3460 bool obj_array, bool not_array) {
3461
3462 if (stopped()) {
3463 return NULL;
3464 }
3465
3466 // If obj_array/non_array==false/false:
3467 // Branch around if the given klass is in fact an array (either obj or prim).
3468 // If obj_array/non_array==false/true:
3469 // Branch around if the given klass is not an array klass of any kind.
3470 // If obj_array/non_array==true/true:
3471 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3472 // If obj_array/non_array==true/false:
3473 // Branch around if the kls is an oop array (Object[] or subtype)
3474 //
3475 // Like generate_guard, adds a new path onto the region.
3476 jint layout_con = 0;
3477 Node* layout_val = get_layout_helper(kls, layout_con);
3478 if (layout_val == NULL) {
3479 bool query = (obj_array
3480 ? Klass::layout_helper_is_objArray(layout_con)
3481 : Klass::layout_helper_is_array(layout_con));
3482 if (query == not_array) {
3483 return NULL; // never a branch
3484 } else { // always a branch
3485 Node* always_branch = control();
3486 if (region != NULL)
3487 region->add_req(always_branch);
3488 set_control(top());
3489 return always_branch;
3490 }
3491 }
3492 // Now test the correct condition.
3493 jint nval = (obj_array
3494 ? (jint)(Klass::_lh_array_tag_type_value
3495 << Klass::_lh_array_tag_shift)
3496 : Klass::_lh_neutral_value);
3497 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3498 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3499 // invert the test if we are looking for a non-array
3500 if (not_array) btest = BoolTest(btest).negate();
3501 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3502 return generate_fair_guard(bol, region);
3503 }
3504
3505
3506 //-----------------------inline_native_newArray--------------------------
3507 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3508 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3509 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3510 Node* mirror;
3511 Node* count_val;
3512 if (uninitialized) {
3513 mirror = argument(1);
3514 count_val = argument(2);
3515 } else {
3516 mirror = argument(0);
3517 count_val = argument(1);
3518 }
3519
3520 mirror = null_check(mirror);
3521 // If mirror or obj is dead, only null-path is taken.
3522 if (stopped()) return true;
3523
3524 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3525 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3526 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3527 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3528 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3529
3530 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3531 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3532 result_reg, _slow_path);
3533 Node* normal_ctl = control();
3534 Node* no_array_ctl = result_reg->in(_slow_path);
3535
3536 // Generate code for the slow case. We make a call to newArray().
3537 set_control(no_array_ctl);
3538 if (!stopped()) {
3539 // Either the input type is void.class, or else the
3540 // array klass has not yet been cached. Either the
3541 // ensuing call will throw an exception, or else it
3542 // will cache the array klass for next time.
3543 PreserveJVMState pjvms(this);
3544 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3545 Node* slow_result = set_results_for_java_call(slow_call);
3546 // this->control() comes from set_results_for_java_call
3547 result_reg->set_req(_slow_path, control());
3548 result_val->set_req(_slow_path, slow_result);
3549 result_io ->set_req(_slow_path, i_o());
3550 result_mem->set_req(_slow_path, reset_memory());
3551 }
3552
3553 set_control(normal_ctl);
3554 if (!stopped()) {
3555 // Normal case: The array type has been cached in the java.lang.Class.
3556 // The following call works fine even if the array type is polymorphic.
3557 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3558 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3559 result_reg->init_req(_normal_path, control());
3560 result_val->init_req(_normal_path, obj);
3561 result_io ->init_req(_normal_path, i_o());
3562 result_mem->init_req(_normal_path, reset_memory());
3563
3564 if (uninitialized) {
3565 // Mark the allocation so that zeroing is skipped
3566 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3567 alloc->maybe_set_complete(&_gvn);
3568 }
3569 }
3570
3571 // Return the combined state.
3572 set_i_o( _gvn.transform(result_io) );
3573 set_all_memory( _gvn.transform(result_mem));
3574
3575 C->set_has_split_ifs(true); // Has chance for split-if optimization
3576 set_result(result_reg, result_val);
3577 return true;
3578 }
3579
3580 //----------------------inline_native_getLength--------------------------
3581 // public static native int java.lang.reflect.Array.getLength(Object array);
3582 bool LibraryCallKit::inline_native_getLength() {
3583 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3584
3585 Node* array = null_check(argument(0));
3586 // If array is dead, only null-path is taken.
3587 if (stopped()) return true;
3588
3589 // Deoptimize if it is a non-array.
3590 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3591
3592 if (non_array != NULL) {
3593 PreserveJVMState pjvms(this);
3594 set_control(non_array);
3595 uncommon_trap(Deoptimization::Reason_intrinsic,
3596 Deoptimization::Action_maybe_recompile);
3597 }
3598
3599 // If control is dead, only non-array-path is taken.
3600 if (stopped()) return true;
3601
3602 // The works fine even if the array type is polymorphic.
3603 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3604 Node* result = load_array_length(array);
3605
3606 C->set_has_split_ifs(true); // Has chance for split-if optimization
3607 set_result(result);
3608 return true;
3609 }
3610
3611 //------------------------inline_array_copyOf----------------------------
3612 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3613 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3614 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3615 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3616
3617 // Get the arguments.
3618 Node* original = argument(0);
3619 Node* start = is_copyOfRange? argument(1): intcon(0);
3620 Node* end = is_copyOfRange? argument(2): argument(1);
3621 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3622
3623 Node* newcopy = NULL;
3624
3625 // Set the original stack and the reexecute bit for the interpreter to reexecute
3626 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3627 { PreserveReexecuteState preexecs(this);
3628 jvms()->set_should_reexecute(true);
3629
3630 array_type_mirror = null_check(array_type_mirror);
3631 original = null_check(original);
3632
3633 // Check if a null path was taken unconditionally.
3634 if (stopped()) return true;
3635
3636 Node* orig_length = load_array_length(original);
3637
3638 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3639 klass_node = null_check(klass_node);
3640
3641 RegionNode* bailout = new RegionNode(1);
3642 record_for_igvn(bailout);
3643
3644 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3645 // Bail out if that is so.
3646 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3647 if (not_objArray != NULL) {
3648 // Improve the klass node's type from the new optimistic assumption:
3649 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3650 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3651 Node* cast = new CastPPNode(klass_node, akls);
3652 cast->init_req(0, control());
3653 klass_node = _gvn.transform(cast);
3654 }
3655
3656 // Bail out if either start or end is negative.
3657 generate_negative_guard(start, bailout, &start);
3658 generate_negative_guard(end, bailout, &end);
3659
3660 Node* length = end;
3661 if (_gvn.type(start) != TypeInt::ZERO) {
3662 length = _gvn.transform(new SubINode(end, start));
3663 }
3664
3665 // Bail out if length is negative.
3666 // Without this the new_array would throw
3667 // NegativeArraySizeException but IllegalArgumentException is what
3668 // should be thrown
3669 generate_negative_guard(length, bailout, &length);
3670
3671 if (bailout->req() > 1) {
3672 PreserveJVMState pjvms(this);
3673 set_control(_gvn.transform(bailout));
3674 uncommon_trap(Deoptimization::Reason_intrinsic,
3675 Deoptimization::Action_maybe_recompile);
3676 }
3677
3678 if (!stopped()) {
3679 // How many elements will we copy from the original?
3680 // The answer is MinI(orig_length - start, length).
3681 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3682 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3683
3684 original = access_resolve(original, ACCESS_READ);
3685
3686 // Generate a direct call to the right arraycopy function(s).
3687 // We know the copy is disjoint but we might not know if the
3688 // oop stores need checking.
3689 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3690 // This will fail a store-check if x contains any non-nulls.
3691
3692 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3693 // loads/stores but it is legal only if we're sure the
3694 // Arrays.copyOf would succeed. So we need all input arguments
3695 // to the copyOf to be validated, including that the copy to the
3696 // new array won't trigger an ArrayStoreException. That subtype
3697 // check can be optimized if we know something on the type of
3698 // the input array from type speculation.
3699 if (_gvn.type(klass_node)->singleton()) {
3700 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3701 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3702
3703 int test = C->static_subtype_check(superk, subk);
3704 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3705 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3706 if (t_original->speculative_type() != NULL) {
3707 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3708 }
3709 }
3710 }
3711
3712 bool validated = false;
3713 // Reason_class_check rather than Reason_intrinsic because we
3714 // want to intrinsify even if this traps.
3715 if (!too_many_traps(Deoptimization::Reason_class_check)) {
3716 Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
3717 klass_node);
3718
3719 if (not_subtype_ctrl != top()) {
3720 PreserveJVMState pjvms(this);
3721 set_control(not_subtype_ctrl);
3722 uncommon_trap(Deoptimization::Reason_class_check,
3723 Deoptimization::Action_make_not_entrant);
3724 assert(stopped(), "Should be stopped");
3725 }
3726 validated = true;
3727 }
3728
3729 if (!stopped()) {
3730 newcopy = new_array(klass_node, length, 0); // no arguments to push
3731
3732 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
3733 load_object_klass(original), klass_node);
3734 if (!is_copyOfRange) {
3735 ac->set_copyof(validated);
3736 } else {
3737 ac->set_copyofrange(validated);
3738 }
3739 Node* n = _gvn.transform(ac);
3740 if (n == ac) {
3741 ac->connect_outputs(this);
3742 } else {
3743 assert(validated, "shouldn't transform if all arguments not validated");
3744 set_all_memory(n);
3745 }
3746 }
3747 }
3748 } // original reexecute is set back here
3749
3750 C->set_has_split_ifs(true); // Has chance for split-if optimization
3751 if (!stopped()) {
3752 set_result(newcopy);
3753 }
3754 return true;
3755 }
3756
3757
3758 //----------------------generate_virtual_guard---------------------------
3759 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
3760 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
3761 RegionNode* slow_region) {
3762 ciMethod* method = callee();
3763 int vtable_index = method->vtable_index();
3764 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3765 "bad index %d", vtable_index);
3766 // Get the Method* out of the appropriate vtable entry.
3767 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
3768 vtable_index*vtableEntry::size_in_bytes() +
3769 vtableEntry::method_offset_in_bytes();
3770 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
3771 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3772
3773 // Compare the target method with the expected method (e.g., Object.hashCode).
3774 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
3775
3776 Node* native_call = makecon(native_call_addr);
3777 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
3778 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
3779
3780 return generate_slow_guard(test_native, slow_region);
3781 }
3782
3783 //-----------------------generate_method_call----------------------------
3784 // Use generate_method_call to make a slow-call to the real
3785 // method if the fast path fails. An alternative would be to
3786 // use a stub like OptoRuntime::slow_arraycopy_Java.
3787 // This only works for expanding the current library call,
3788 // not another intrinsic. (E.g., don't use this for making an
3789 // arraycopy call inside of the copyOf intrinsic.)
3790 CallJavaNode*
3791 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
3792 // When compiling the intrinsic method itself, do not use this technique.
3793 guarantee(callee() != C->method(), "cannot make slow-call to self");
3794
3795 ciMethod* method = callee();
3796 // ensure the JVMS we have will be correct for this call
3797 guarantee(method_id == method->intrinsic_id(), "must match");
3798
3799 const TypeFunc* tf = TypeFunc::make(method);
3800 CallJavaNode* slow_call;
3801 if (is_static) {
3802 assert(!is_virtual, "");
3803 slow_call = new CallStaticJavaNode(C, tf,
3804 SharedRuntime::get_resolve_static_call_stub(),
3805 method, bci());
3806 } else if (is_virtual) {
3807 null_check_receiver();
3808 int vtable_index = Method::invalid_vtable_index;
3809 if (UseInlineCaches) {
3810 // Suppress the vtable call
3811 } else {
3812 // hashCode and clone are not a miranda methods,
3813 // so the vtable index is fixed.
3814 // No need to use the linkResolver to get it.
3815 vtable_index = method->vtable_index();
3816 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3817 "bad index %d", vtable_index);
3818 }
3819 slow_call = new CallDynamicJavaNode(tf,
3820 SharedRuntime::get_resolve_virtual_call_stub(),
3821 method, vtable_index, bci());
3822 } else { // neither virtual nor static: opt_virtual
3823 null_check_receiver();
3824 slow_call = new CallStaticJavaNode(C, tf,
3825 SharedRuntime::get_resolve_opt_virtual_call_stub(),
3826 method, bci());
3827 slow_call->set_optimized_virtual(true);
3828 }
3829 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
3830 // To be able to issue a direct call (optimized virtual or virtual)
3831 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
3832 // about the method being invoked should be attached to the call site to
3833 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
3834 slow_call->set_override_symbolic_info(true);
3835 }
3836 set_arguments_for_java_call(slow_call);
3837 set_edges_for_java_call(slow_call);
3838 return slow_call;
3839 }
3840
3841
3842 /**
3843 * Build special case code for calls to hashCode on an object. This call may
3844 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
3845 * slightly different code.
3846 */
3847 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
3848 assert(is_static == callee()->is_static(), "correct intrinsic selection");
3849 assert(!(is_virtual && is_static), "either virtual, special, or static");
3850
3851 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
3852
3853 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3854 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
3855 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3856 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3857 Node* obj = NULL;
3858 if (!is_static) {
3859 // Check for hashing null object
3860 obj = null_check_receiver();
3861 if (stopped()) return true; // unconditionally null
3862 result_reg->init_req(_null_path, top());
3863 result_val->init_req(_null_path, top());
3864 } else {
3865 // Do a null check, and return zero if null.
3866 // System.identityHashCode(null) == 0
3867 obj = argument(0);
3868 Node* null_ctl = top();
3869 obj = null_check_oop(obj, &null_ctl);
3870 result_reg->init_req(_null_path, null_ctl);
3871 result_val->init_req(_null_path, _gvn.intcon(0));
3872 }
3873
3874 // Unconditionally null? Then return right away.
3875 if (stopped()) {
3876 set_control( result_reg->in(_null_path));
3877 if (!stopped())
3878 set_result(result_val->in(_null_path));
3879 return true;
3880 }
3881
3882 // We only go to the fast case code if we pass a number of guards. The
3883 // paths which do not pass are accumulated in the slow_region.
3884 RegionNode* slow_region = new RegionNode(1);
3885 record_for_igvn(slow_region);
3886
3887 // If this is a virtual call, we generate a funny guard. We pull out
3888 // the vtable entry corresponding to hashCode() from the target object.
3889 // If the target method which we are calling happens to be the native
3890 // Object hashCode() method, we pass the guard. We do not need this
3891 // guard for non-virtual calls -- the caller is known to be the native
3892 // Object hashCode().
3893 if (is_virtual) {
3894 // After null check, get the object's klass.
3895 Node* obj_klass = load_object_klass(obj);
3896 generate_virtual_guard(obj_klass, slow_region);
3897 }
3898
3899 // Get the header out of the object, use LoadMarkNode when available
3900 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
3901 // The control of the load must be NULL. Otherwise, the load can move before
3902 // the null check after castPP removal.
3903 Node* no_ctrl = NULL;
3904 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
3905
3906 // Test the header to see if it is unlocked.
3907 Node *lock_mask = _gvn.MakeConX(markWord::biased_lock_mask_in_place);
3908 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
3909 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
3910 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
3911 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
3912
3913 generate_slow_guard(test_unlocked, slow_region);
3914
3915 // Get the hash value and check to see that it has been properly assigned.
3916 // We depend on hash_mask being at most 32 bits and avoid the use of
3917 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
3918 // vm: see markWord.hpp.
3919 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
3920 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
3921 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
3922 // This hack lets the hash bits live anywhere in the mark object now, as long
3923 // as the shift drops the relevant bits into the low 32 bits. Note that
3924 // Java spec says that HashCode is an int so there's no point in capturing
3925 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
3926 hshifted_header = ConvX2I(hshifted_header);
3927 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
3928
3929 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
3930 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
3931 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
3932
3933 generate_slow_guard(test_assigned, slow_region);
3934
3935 Node* init_mem = reset_memory();
3936 // fill in the rest of the null path:
3937 result_io ->init_req(_null_path, i_o());
3938 result_mem->init_req(_null_path, init_mem);
3939
3940 result_val->init_req(_fast_path, hash_val);
3941 result_reg->init_req(_fast_path, control());
3942 result_io ->init_req(_fast_path, i_o());
3943 result_mem->init_req(_fast_path, init_mem);
3944
3945 // Generate code for the slow case. We make a call to hashCode().
3946 set_control(_gvn.transform(slow_region));
3947 if (!stopped()) {
3948 // No need for PreserveJVMState, because we're using up the present state.
3949 set_all_memory(init_mem);
3950 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
3951 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
3952 Node* slow_result = set_results_for_java_call(slow_call);
3953 // this->control() comes from set_results_for_java_call
3954 result_reg->init_req(_slow_path, control());
3955 result_val->init_req(_slow_path, slow_result);
3956 result_io ->set_req(_slow_path, i_o());
3957 result_mem ->set_req(_slow_path, reset_memory());
3958 }
3959
3960 // Return the combined state.
3961 set_i_o( _gvn.transform(result_io) );
3962 set_all_memory( _gvn.transform(result_mem));
3963
3964 set_result(result_reg, result_val);
3965 return true;
3966 }
3967
3968 //---------------------------inline_native_getClass----------------------------
3969 // public final native Class<?> java.lang.Object.getClass();
3970 //
3971 // Build special case code for calls to getClass on an object.
3972 bool LibraryCallKit::inline_native_getClass() {
3973 Node* obj = null_check_receiver();
3974 if (stopped()) return true;
3975 set_result(load_mirror_from_klass(load_object_klass(obj)));
3976 return true;
3977 }
3978
3979 //-----------------inline_native_Reflection_getCallerClass---------------------
3980 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
3981 //
3982 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
3983 //
3984 // NOTE: This code must perform the same logic as JVM_GetCallerClass
3985 // in that it must skip particular security frames and checks for
3986 // caller sensitive methods.
3987 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
3988 #ifndef PRODUCT
3989 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3990 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
3991 }
3992 #endif
3993
3994 if (!jvms()->has_method()) {
3995 #ifndef PRODUCT
3996 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3997 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
3998 }
3999 #endif
4000 return false;
4001 }
4002
4003 // Walk back up the JVM state to find the caller at the required
4004 // depth.
4005 JVMState* caller_jvms = jvms();
4006
4007 // Cf. JVM_GetCallerClass
4008 // NOTE: Start the loop at depth 1 because the current JVM state does
4009 // not include the Reflection.getCallerClass() frame.
4010 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4011 ciMethod* m = caller_jvms->method();
4012 switch (n) {
4013 case 0:
4014 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4015 break;
4016 case 1:
4017 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4018 if (!m->caller_sensitive()) {
4019 #ifndef PRODUCT
4020 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4021 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4022 }
4023 #endif
4024 return false; // bail-out; let JVM_GetCallerClass do the work
4025 }
4026 break;
4027 default:
4028 if (!m->is_ignored_by_security_stack_walk()) {
4029 // We have reached the desired frame; return the holder class.
4030 // Acquire method holder as java.lang.Class and push as constant.
4031 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4032 ciInstance* caller_mirror = caller_klass->java_mirror();
4033 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4034
4035 #ifndef PRODUCT
4036 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4037 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());
4038 tty->print_cr(" JVM state at this point:");
4039 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4040 ciMethod* m = jvms()->of_depth(i)->method();
4041 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4042 }
4043 }
4044 #endif
4045 return true;
4046 }
4047 break;
4048 }
4049 }
4050
4051 #ifndef PRODUCT
4052 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4053 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4054 tty->print_cr(" JVM state at this point:");
4055 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4056 ciMethod* m = jvms()->of_depth(i)->method();
4057 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4058 }
4059 }
4060 #endif
4061
4062 return false; // bail-out; let JVM_GetCallerClass do the work
4063 }
4064
4065 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4066 Node* arg = argument(0);
4067 Node* result = NULL;
4068
4069 switch (id) {
4070 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4071 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4072 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4073 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4074
4075 case vmIntrinsics::_doubleToLongBits: {
4076 // two paths (plus control) merge in a wood
4077 RegionNode *r = new RegionNode(3);
4078 Node *phi = new PhiNode(r, TypeLong::LONG);
4079
4080 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4081 // Build the boolean node
4082 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4083
4084 // Branch either way.
4085 // NaN case is less traveled, which makes all the difference.
4086 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4087 Node *opt_isnan = _gvn.transform(ifisnan);
4088 assert( opt_isnan->is_If(), "Expect an IfNode");
4089 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4090 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4091
4092 set_control(iftrue);
4093
4094 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4095 Node *slow_result = longcon(nan_bits); // return NaN
4096 phi->init_req(1, _gvn.transform( slow_result ));
4097 r->init_req(1, iftrue);
4098
4099 // Else fall through
4100 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4101 set_control(iffalse);
4102
4103 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4104 r->init_req(2, iffalse);
4105
4106 // Post merge
4107 set_control(_gvn.transform(r));
4108 record_for_igvn(r);
4109
4110 C->set_has_split_ifs(true); // Has chance for split-if optimization
4111 result = phi;
4112 assert(result->bottom_type()->isa_long(), "must be");
4113 break;
4114 }
4115
4116 case vmIntrinsics::_floatToIntBits: {
4117 // two paths (plus control) merge in a wood
4118 RegionNode *r = new RegionNode(3);
4119 Node *phi = new PhiNode(r, TypeInt::INT);
4120
4121 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4122 // Build the boolean node
4123 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4124
4125 // Branch either way.
4126 // NaN case is less traveled, which makes all the difference.
4127 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4128 Node *opt_isnan = _gvn.transform(ifisnan);
4129 assert( opt_isnan->is_If(), "Expect an IfNode");
4130 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4131 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4132
4133 set_control(iftrue);
4134
4135 static const jint nan_bits = 0x7fc00000;
4136 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4137 phi->init_req(1, _gvn.transform( slow_result ));
4138 r->init_req(1, iftrue);
4139
4140 // Else fall through
4141 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4142 set_control(iffalse);
4143
4144 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4145 r->init_req(2, iffalse);
4146
4147 // Post merge
4148 set_control(_gvn.transform(r));
4149 record_for_igvn(r);
4150
4151 C->set_has_split_ifs(true); // Has chance for split-if optimization
4152 result = phi;
4153 assert(result->bottom_type()->isa_int(), "must be");
4154 break;
4155 }
4156
4157 default:
4158 fatal_unexpected_iid(id);
4159 break;
4160 }
4161 set_result(_gvn.transform(result));
4162 return true;
4163 }
4164
4165 //----------------------inline_unsafe_copyMemory-------------------------
4166 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4167 bool LibraryCallKit::inline_unsafe_copyMemory() {
4168 if (callee()->is_static()) return false; // caller must have the capability!
4169 null_check_receiver(); // null-check receiver
4170 if (stopped()) return true;
4171
4172 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4173
4174 Node* src_ptr = argument(1); // type: oop
4175 Node* src_off = ConvL2X(argument(2)); // type: long
4176 Node* dst_ptr = argument(4); // type: oop
4177 Node* dst_off = ConvL2X(argument(5)); // type: long
4178 Node* size = ConvL2X(argument(7)); // type: long
4179
4180 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4181 "fieldOffset must be byte-scaled");
4182
4183 src_ptr = access_resolve(src_ptr, ACCESS_READ);
4184 dst_ptr = access_resolve(dst_ptr, ACCESS_WRITE);
4185 Node* src = make_unsafe_address(src_ptr, src_off, ACCESS_READ);
4186 Node* dst = make_unsafe_address(dst_ptr, dst_off, ACCESS_WRITE);
4187
4188 // Conservatively insert a memory barrier on all memory slices.
4189 // Do not let writes of the copy source or destination float below the copy.
4190 insert_mem_bar(Op_MemBarCPUOrder);
4191
4192 Node* thread = _gvn.transform(new ThreadLocalNode());
4193 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4194 BasicType doing_unsafe_access_bt = T_BYTE;
4195 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4196
4197 // update volatile field
4198 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4199
4200 // Call it. Note that the length argument is not scaled.
4201 make_runtime_call(RC_LEAF|RC_NO_FP,
4202 OptoRuntime::fast_arraycopy_Type(),
4203 StubRoutines::unsafe_arraycopy(),
4204 "unsafe_arraycopy",
4205 TypeRawPtr::BOTTOM,
4206 src, dst, size XTOP);
4207
4208 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4209
4210 // Do not let reads of the copy destination float above the copy.
4211 insert_mem_bar(Op_MemBarCPUOrder);
4212
4213 return true;
4214 }
4215
4216 //------------------------clone_coping-----------------------------------
4217 // Helper function for inline_native_clone.
4218 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4219 assert(obj_size != NULL, "");
4220 Node* raw_obj = alloc_obj->in(1);
4221 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4222
4223 AllocateNode* alloc = NULL;
4224 if (ReduceBulkZeroing) {
4225 // We will be completely responsible for initializing this object -
4226 // mark Initialize node as complete.
4227 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4228 // The object was just allocated - there should be no any stores!
4229 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4230 // Mark as complete_with_arraycopy so that on AllocateNode
4231 // expansion, we know this AllocateNode is initialized by an array
4232 // copy and a StoreStore barrier exists after the array copy.
4233 alloc->initialization()->set_complete_with_arraycopy();
4234 }
4235
4236 // Copy the fastest available way.
4237 // TODO: generate fields copies for small objects instead.
4238 Node* size = _gvn.transform(obj_size);
4239
4240 access_clone(obj, alloc_obj, size, is_array);
4241
4242 // Do not let reads from the cloned object float above the arraycopy.
4243 if (alloc != NULL) {
4244 // Do not let stores that initialize this object be reordered with
4245 // a subsequent store that would make this object accessible by
4246 // other threads.
4247 // Record what AllocateNode this StoreStore protects so that
4248 // escape analysis can go from the MemBarStoreStoreNode to the
4249 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4250 // based on the escape status of the AllocateNode.
4251 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4252 } else {
4253 insert_mem_bar(Op_MemBarCPUOrder);
4254 }
4255 }
4256
4257 //------------------------inline_native_clone----------------------------
4258 // protected native Object java.lang.Object.clone();
4259 //
4260 // Here are the simple edge cases:
4261 // null receiver => normal trap
4262 // virtual and clone was overridden => slow path to out-of-line clone
4263 // not cloneable or finalizer => slow path to out-of-line Object.clone
4264 //
4265 // The general case has two steps, allocation and copying.
4266 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4267 //
4268 // Copying also has two cases, oop arrays and everything else.
4269 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4270 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4271 //
4272 // These steps fold up nicely if and when the cloned object's klass
4273 // can be sharply typed as an object array, a type array, or an instance.
4274 //
4275 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4276 PhiNode* result_val;
4277
4278 // Set the reexecute bit for the interpreter to reexecute
4279 // the bytecode that invokes Object.clone if deoptimization happens.
4280 { PreserveReexecuteState preexecs(this);
4281 jvms()->set_should_reexecute(true);
4282
4283 Node* obj = null_check_receiver();
4284 if (stopped()) return true;
4285
4286 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4287
4288 // If we are going to clone an instance, we need its exact type to
4289 // know the number and types of fields to convert the clone to
4290 // loads/stores. Maybe a speculative type can help us.
4291 if (!obj_type->klass_is_exact() &&
4292 obj_type->speculative_type() != NULL &&
4293 obj_type->speculative_type()->is_instance_klass()) {
4294 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4295 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4296 !spec_ik->has_injected_fields()) {
4297 ciKlass* k = obj_type->klass();
4298 if (!k->is_instance_klass() ||
4299 k->as_instance_klass()->is_interface() ||
4300 k->as_instance_klass()->has_subklass()) {
4301 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4302 }
4303 }
4304 }
4305
4306 Node* obj_klass = load_object_klass(obj);
4307 const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
4308 const TypeOopPtr* toop = ((tklass != NULL)
4309 ? tklass->as_instance_type()
4310 : TypeInstPtr::NOTNULL);
4311
4312 // Conservatively insert a memory barrier on all memory slices.
4313 // Do not let writes into the original float below the clone.
4314 insert_mem_bar(Op_MemBarCPUOrder);
4315
4316 // paths into result_reg:
4317 enum {
4318 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4319 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4320 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4321 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4322 PATH_LIMIT
4323 };
4324 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4325 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4326 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4327 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4328 record_for_igvn(result_reg);
4329
4330 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4331 if (array_ctl != NULL) {
4332 // It's an array.
4333 PreserveJVMState pjvms(this);
4334 set_control(array_ctl);
4335 Node* obj_length = load_array_length(obj);
4336 Node* obj_size = NULL;
4337 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size); // no arguments to push
4338
4339 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4340 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, BarrierSetC2::Parsing)) {
4341 // If it is an oop array, it requires very special treatment,
4342 // because gc barriers are required when accessing the array.
4343 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4344 if (is_obja != NULL) {
4345 PreserveJVMState pjvms2(this);
4346 set_control(is_obja);
4347 obj = access_resolve(obj, ACCESS_READ);
4348 // Generate a direct call to the right arraycopy function(s).
4349 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4350 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false);
4351 ac->set_cloneoop();
4352 Node* n = _gvn.transform(ac);
4353 assert(n == ac, "cannot disappear");
4354 ac->connect_outputs(this);
4355
4356 result_reg->init_req(_objArray_path, control());
4357 result_val->init_req(_objArray_path, alloc_obj);
4358 result_i_o ->set_req(_objArray_path, i_o());
4359 result_mem ->set_req(_objArray_path, reset_memory());
4360 }
4361 }
4362 // Otherwise, there are no barriers to worry about.
4363 // (We can dispense with card marks if we know the allocation
4364 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4365 // causes the non-eden paths to take compensating steps to
4366 // simulate a fresh allocation, so that no further
4367 // card marks are required in compiled code to initialize
4368 // the object.)
4369
4370 if (!stopped()) {
4371 copy_to_clone(obj, alloc_obj, obj_size, true);
4372
4373 // Present the results of the copy.
4374 result_reg->init_req(_array_path, control());
4375 result_val->init_req(_array_path, alloc_obj);
4376 result_i_o ->set_req(_array_path, i_o());
4377 result_mem ->set_req(_array_path, reset_memory());
4378 }
4379 }
4380
4381 // We only go to the instance fast case code if we pass a number of guards.
4382 // The paths which do not pass are accumulated in the slow_region.
4383 RegionNode* slow_region = new RegionNode(1);
4384 record_for_igvn(slow_region);
4385 if (!stopped()) {
4386 // It's an instance (we did array above). Make the slow-path tests.
4387 // If this is a virtual call, we generate a funny guard. We grab
4388 // the vtable entry corresponding to clone() from the target object.
4389 // If the target method which we are calling happens to be the
4390 // Object clone() method, we pass the guard. We do not need this
4391 // guard for non-virtual calls; the caller is known to be the native
4392 // Object clone().
4393 if (is_virtual) {
4394 generate_virtual_guard(obj_klass, slow_region);
4395 }
4396
4397 // The object must be easily cloneable and must not have a finalizer.
4398 // Both of these conditions may be checked in a single test.
4399 // We could optimize the test further, but we don't care.
4400 generate_access_flags_guard(obj_klass,
4401 // Test both conditions:
4402 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4403 // Must be cloneable but not finalizer:
4404 JVM_ACC_IS_CLONEABLE_FAST,
4405 slow_region);
4406 }
4407
4408 if (!stopped()) {
4409 // It's an instance, and it passed the slow-path tests.
4410 PreserveJVMState pjvms(this);
4411 Node* obj_size = NULL;
4412 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4413 // is reexecuted if deoptimization occurs and there could be problems when merging
4414 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4415 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4416
4417 copy_to_clone(obj, alloc_obj, obj_size, false);
4418
4419 // Present the results of the slow call.
4420 result_reg->init_req(_instance_path, control());
4421 result_val->init_req(_instance_path, alloc_obj);
4422 result_i_o ->set_req(_instance_path, i_o());
4423 result_mem ->set_req(_instance_path, reset_memory());
4424 }
4425
4426 // Generate code for the slow case. We make a call to clone().
4427 set_control(_gvn.transform(slow_region));
4428 if (!stopped()) {
4429 PreserveJVMState pjvms(this);
4430 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4431 // We need to deoptimize on exception (see comment above)
4432 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4433 // this->control() comes from set_results_for_java_call
4434 result_reg->init_req(_slow_path, control());
4435 result_val->init_req(_slow_path, slow_result);
4436 result_i_o ->set_req(_slow_path, i_o());
4437 result_mem ->set_req(_slow_path, reset_memory());
4438 }
4439
4440 // Return the combined state.
4441 set_control( _gvn.transform(result_reg));
4442 set_i_o( _gvn.transform(result_i_o));
4443 set_all_memory( _gvn.transform(result_mem));
4444 } // original reexecute is set back here
4445
4446 set_result(_gvn.transform(result_val));
4447 return true;
4448 }
4449
4450 // If we have a tightly coupled allocation, the arraycopy may take care
4451 // of the array initialization. If one of the guards we insert between
4452 // the allocation and the arraycopy causes a deoptimization, an
4453 // unitialized array will escape the compiled method. To prevent that
4454 // we set the JVM state for uncommon traps between the allocation and
4455 // the arraycopy to the state before the allocation so, in case of
4456 // deoptimization, we'll reexecute the allocation and the
4457 // initialization.
4458 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4459 if (alloc != NULL) {
4460 ciMethod* trap_method = alloc->jvms()->method();
4461 int trap_bci = alloc->jvms()->bci();
4462
4463 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4464 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4465 // Make sure there's no store between the allocation and the
4466 // arraycopy otherwise visible side effects could be rexecuted
4467 // in case of deoptimization and cause incorrect execution.
4468 bool no_interfering_store = true;
4469 Node* mem = alloc->in(TypeFunc::Memory);
4470 if (mem->is_MergeMem()) {
4471 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4472 Node* n = mms.memory();
4473 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4474 assert(n->is_Store(), "what else?");
4475 no_interfering_store = false;
4476 break;
4477 }
4478 }
4479 } else {
4480 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4481 Node* n = mms.memory();
4482 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4483 assert(n->is_Store(), "what else?");
4484 no_interfering_store = false;
4485 break;
4486 }
4487 }
4488 }
4489
4490 if (no_interfering_store) {
4491 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4492 uint size = alloc->req();
4493 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4494 old_jvms->set_map(sfpt);
4495 for (uint i = 0; i < size; i++) {
4496 sfpt->init_req(i, alloc->in(i));
4497 }
4498 // re-push array length for deoptimization
4499 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4500 old_jvms->set_sp(old_jvms->sp()+1);
4501 old_jvms->set_monoff(old_jvms->monoff()+1);
4502 old_jvms->set_scloff(old_jvms->scloff()+1);
4503 old_jvms->set_endoff(old_jvms->endoff()+1);
4504 old_jvms->set_should_reexecute(true);
4505
4506 sfpt->set_i_o(map()->i_o());
4507 sfpt->set_memory(map()->memory());
4508 sfpt->set_control(map()->control());
4509
4510 JVMState* saved_jvms = jvms();
4511 saved_reexecute_sp = _reexecute_sp;
4512
4513 set_jvms(sfpt->jvms());
4514 _reexecute_sp = jvms()->sp();
4515
4516 return saved_jvms;
4517 }
4518 }
4519 }
4520 return NULL;
4521 }
4522
4523 // In case of a deoptimization, we restart execution at the
4524 // allocation, allocating a new array. We would leave an uninitialized
4525 // array in the heap that GCs wouldn't expect. Move the allocation
4526 // after the traps so we don't allocate the array if we
4527 // deoptimize. This is possible because tightly_coupled_allocation()
4528 // guarantees there's no observer of the allocated array at this point
4529 // and the control flow is simple enough.
4530 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4531 int saved_reexecute_sp, uint new_idx) {
4532 if (saved_jvms != NULL && !stopped()) {
4533 assert(alloc != NULL, "only with a tightly coupled allocation");
4534 // restore JVM state to the state at the arraycopy
4535 saved_jvms->map()->set_control(map()->control());
4536 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4537 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4538 // If we've improved the types of some nodes (null check) while
4539 // emitting the guards, propagate them to the current state
4540 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4541 set_jvms(saved_jvms);
4542 _reexecute_sp = saved_reexecute_sp;
4543
4544 // Remove the allocation from above the guards
4545 CallProjections callprojs;
4546 alloc->extract_projections(&callprojs, true);
4547 InitializeNode* init = alloc->initialization();
4548 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4549 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4550 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4551 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4552
4553 // move the allocation here (after the guards)
4554 _gvn.hash_delete(alloc);
4555 alloc->set_req(TypeFunc::Control, control());
4556 alloc->set_req(TypeFunc::I_O, i_o());
4557 Node *mem = reset_memory();
4558 set_all_memory(mem);
4559 alloc->set_req(TypeFunc::Memory, mem);
4560 set_control(init->proj_out_or_null(TypeFunc::Control));
4561 set_i_o(callprojs.fallthrough_ioproj);
4562
4563 // Update memory as done in GraphKit::set_output_for_allocation()
4564 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4565 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4566 if (ary_type->isa_aryptr() && length_type != NULL) {
4567 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4568 }
4569 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4570 int elemidx = C->get_alias_index(telemref);
4571 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
4572 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
4573
4574 Node* allocx = _gvn.transform(alloc);
4575 assert(allocx == alloc, "where has the allocation gone?");
4576 assert(dest->is_CheckCastPP(), "not an allocation result?");
4577
4578 _gvn.hash_delete(dest);
4579 dest->set_req(0, control());
4580 Node* destx = _gvn.transform(dest);
4581 assert(destx == dest, "where has the allocation result gone?");
4582 }
4583 }
4584
4585
4586 //------------------------------inline_arraycopy-----------------------
4587 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4588 // Object dest, int destPos,
4589 // int length);
4590 bool LibraryCallKit::inline_arraycopy() {
4591 // Get the arguments.
4592 Node* src = argument(0); // type: oop
4593 Node* src_offset = argument(1); // type: int
4594 Node* dest = argument(2); // type: oop
4595 Node* dest_offset = argument(3); // type: int
4596 Node* length = argument(4); // type: int
4597
4598 uint new_idx = C->unique();
4599
4600 // Check for allocation before we add nodes that would confuse
4601 // tightly_coupled_allocation()
4602 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4603
4604 int saved_reexecute_sp = -1;
4605 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4606 // See arraycopy_restore_alloc_state() comment
4607 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4608 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4609 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
4610 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4611
4612 // The following tests must be performed
4613 // (1) src and dest are arrays.
4614 // (2) src and dest arrays must have elements of the same BasicType
4615 // (3) src and dest must not be null.
4616 // (4) src_offset must not be negative.
4617 // (5) dest_offset must not be negative.
4618 // (6) length must not be negative.
4619 // (7) src_offset + length must not exceed length of src.
4620 // (8) dest_offset + length must not exceed length of dest.
4621 // (9) each element of an oop array must be assignable
4622
4623 // (3) src and dest must not be null.
4624 // always do this here because we need the JVM state for uncommon traps
4625 Node* null_ctl = top();
4626 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4627 assert(null_ctl->is_top(), "no null control here");
4628 dest = null_check(dest, T_ARRAY);
4629
4630 if (!can_emit_guards) {
4631 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4632 // guards but the arraycopy node could still take advantage of a
4633 // tightly allocated allocation. tightly_coupled_allocation() is
4634 // called again to make sure it takes the null check above into
4635 // account: the null check is mandatory and if it caused an
4636 // uncommon trap to be emitted then the allocation can't be
4637 // considered tightly coupled in this context.
4638 alloc = tightly_coupled_allocation(dest, NULL);
4639 }
4640
4641 bool validated = false;
4642
4643 const Type* src_type = _gvn.type(src);
4644 const Type* dest_type = _gvn.type(dest);
4645 const TypeAryPtr* top_src = src_type->isa_aryptr();
4646 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4647
4648 // Do we have the type of src?
4649 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4650 // Do we have the type of dest?
4651 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4652 // Is the type for src from speculation?
4653 bool src_spec = false;
4654 // Is the type for dest from speculation?
4655 bool dest_spec = false;
4656
4657 if ((!has_src || !has_dest) && can_emit_guards) {
4658 // We don't have sufficient type information, let's see if
4659 // speculative types can help. We need to have types for both src
4660 // and dest so that it pays off.
4661
4662 // Do we already have or could we have type information for src
4663 bool could_have_src = has_src;
4664 // Do we already have or could we have type information for dest
4665 bool could_have_dest = has_dest;
4666
4667 ciKlass* src_k = NULL;
4668 if (!has_src) {
4669 src_k = src_type->speculative_type_not_null();
4670 if (src_k != NULL && src_k->is_array_klass()) {
4671 could_have_src = true;
4672 }
4673 }
4674
4675 ciKlass* dest_k = NULL;
4676 if (!has_dest) {
4677 dest_k = dest_type->speculative_type_not_null();
4678 if (dest_k != NULL && dest_k->is_array_klass()) {
4679 could_have_dest = true;
4680 }
4681 }
4682
4683 if (could_have_src && could_have_dest) {
4684 // This is going to pay off so emit the required guards
4685 if (!has_src) {
4686 src = maybe_cast_profiled_obj(src, src_k, true);
4687 src_type = _gvn.type(src);
4688 top_src = src_type->isa_aryptr();
4689 has_src = (top_src != NULL && top_src->klass() != NULL);
4690 src_spec = true;
4691 }
4692 if (!has_dest) {
4693 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4694 dest_type = _gvn.type(dest);
4695 top_dest = dest_type->isa_aryptr();
4696 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4697 dest_spec = true;
4698 }
4699 }
4700 }
4701
4702 if (has_src && has_dest && can_emit_guards) {
4703 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
4704 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
4705 if (is_reference_type(src_elem)) src_elem = T_OBJECT;
4706 if (is_reference_type(dest_elem)) dest_elem = T_OBJECT;
4707
4708 if (src_elem == dest_elem && src_elem == T_OBJECT) {
4709 // If both arrays are object arrays then having the exact types
4710 // for both will remove the need for a subtype check at runtime
4711 // before the call and may make it possible to pick a faster copy
4712 // routine (without a subtype check on every element)
4713 // Do we have the exact type of src?
4714 bool could_have_src = src_spec;
4715 // Do we have the exact type of dest?
4716 bool could_have_dest = dest_spec;
4717 ciKlass* src_k = top_src->klass();
4718 ciKlass* dest_k = top_dest->klass();
4719 if (!src_spec) {
4720 src_k = src_type->speculative_type_not_null();
4721 if (src_k != NULL && src_k->is_array_klass()) {
4722 could_have_src = true;
4723 }
4724 }
4725 if (!dest_spec) {
4726 dest_k = dest_type->speculative_type_not_null();
4727 if (dest_k != NULL && dest_k->is_array_klass()) {
4728 could_have_dest = true;
4729 }
4730 }
4731 if (could_have_src && could_have_dest) {
4732 // If we can have both exact types, emit the missing guards
4733 if (could_have_src && !src_spec) {
4734 src = maybe_cast_profiled_obj(src, src_k, true);
4735 }
4736 if (could_have_dest && !dest_spec) {
4737 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4738 }
4739 }
4740 }
4741 }
4742
4743 ciMethod* trap_method = method();
4744 int trap_bci = bci();
4745 if (saved_jvms != NULL) {
4746 trap_method = alloc->jvms()->method();
4747 trap_bci = alloc->jvms()->bci();
4748 }
4749
4750 bool negative_length_guard_generated = false;
4751
4752 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4753 can_emit_guards &&
4754 !src->is_top() && !dest->is_top()) {
4755 // validate arguments: enables transformation the ArrayCopyNode
4756 validated = true;
4757
4758 RegionNode* slow_region = new RegionNode(1);
4759 record_for_igvn(slow_region);
4760
4761 // (1) src and dest are arrays.
4762 generate_non_array_guard(load_object_klass(src), slow_region);
4763 generate_non_array_guard(load_object_klass(dest), slow_region);
4764
4765 // (2) src and dest arrays must have elements of the same BasicType
4766 // done at macro expansion or at Ideal transformation time
4767
4768 // (4) src_offset must not be negative.
4769 generate_negative_guard(src_offset, slow_region);
4770
4771 // (5) dest_offset must not be negative.
4772 generate_negative_guard(dest_offset, slow_region);
4773
4774 // (7) src_offset + length must not exceed length of src.
4775 generate_limit_guard(src_offset, length,
4776 load_array_length(src),
4777 slow_region);
4778
4779 // (8) dest_offset + length must not exceed length of dest.
4780 generate_limit_guard(dest_offset, length,
4781 load_array_length(dest),
4782 slow_region);
4783
4784 // (6) length must not be negative.
4785 // This is also checked in generate_arraycopy() during macro expansion, but
4786 // we also have to check it here for the case where the ArrayCopyNode will
4787 // be eliminated by Escape Analysis.
4788 if (EliminateAllocations) {
4789 generate_negative_guard(length, slow_region);
4790 negative_length_guard_generated = true;
4791 }
4792
4793 // (9) each element of an oop array must be assignable
4794 Node* src_klass = load_object_klass(src);
4795 Node* dest_klass = load_object_klass(dest);
4796 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
4797
4798 if (not_subtype_ctrl != top()) {
4799 PreserveJVMState pjvms(this);
4800 set_control(not_subtype_ctrl);
4801 uncommon_trap(Deoptimization::Reason_intrinsic,
4802 Deoptimization::Action_make_not_entrant);
4803 assert(stopped(), "Should be stopped");
4804 }
4805 {
4806 PreserveJVMState pjvms(this);
4807 set_control(_gvn.transform(slow_region));
4808 uncommon_trap(Deoptimization::Reason_intrinsic,
4809 Deoptimization::Action_make_not_entrant);
4810 assert(stopped(), "Should be stopped");
4811 }
4812
4813 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
4814 const Type *toop = TypeOopPtr::make_from_klass(dest_klass_t->klass());
4815 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
4816 }
4817
4818 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
4819
4820 if (stopped()) {
4821 return true;
4822 }
4823
4824 Node* new_src = access_resolve(src, ACCESS_READ);
4825 Node* new_dest = access_resolve(dest, ACCESS_WRITE);
4826
4827 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, new_src, src_offset, new_dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
4828 // Create LoadRange and LoadKlass nodes for use during macro expansion here
4829 // so the compiler has a chance to eliminate them: during macro expansion,
4830 // we have to set their control (CastPP nodes are eliminated).
4831 load_object_klass(src), load_object_klass(dest),
4832 load_array_length(src), load_array_length(dest));
4833
4834 ac->set_arraycopy(validated);
4835
4836 Node* n = _gvn.transform(ac);
4837 if (n == ac) {
4838 ac->connect_outputs(this);
4839 } else {
4840 assert(validated, "shouldn't transform if all arguments not validated");
4841 set_all_memory(n);
4842 }
4843 clear_upper_avx();
4844
4845
4846 return true;
4847 }
4848
4849
4850 // Helper function which determines if an arraycopy immediately follows
4851 // an allocation, with no intervening tests or other escapes for the object.
4852 AllocateArrayNode*
4853 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
4854 RegionNode* slow_region) {
4855 if (stopped()) return NULL; // no fast path
4856 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
4857
4858 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
4859 if (alloc == NULL) return NULL;
4860
4861 Node* rawmem = memory(Compile::AliasIdxRaw);
4862 // Is the allocation's memory state untouched?
4863 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
4864 // Bail out if there have been raw-memory effects since the allocation.
4865 // (Example: There might have been a call or safepoint.)
4866 return NULL;
4867 }
4868 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
4869 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
4870 return NULL;
4871 }
4872
4873 // There must be no unexpected observers of this allocation.
4874 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
4875 Node* obs = ptr->fast_out(i);
4876 if (obs != this->map()) {
4877 return NULL;
4878 }
4879 }
4880
4881 // This arraycopy must unconditionally follow the allocation of the ptr.
4882 Node* alloc_ctl = ptr->in(0);
4883 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
4884
4885 Node* ctl = control();
4886 while (ctl != alloc_ctl) {
4887 // There may be guards which feed into the slow_region.
4888 // Any other control flow means that we might not get a chance
4889 // to finish initializing the allocated object.
4890 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
4891 IfNode* iff = ctl->in(0)->as_If();
4892 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
4893 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
4894 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
4895 ctl = iff->in(0); // This test feeds the known slow_region.
4896 continue;
4897 }
4898 // One more try: Various low-level checks bottom out in
4899 // uncommon traps. If the debug-info of the trap omits
4900 // any reference to the allocation, as we've already
4901 // observed, then there can be no objection to the trap.
4902 bool found_trap = false;
4903 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
4904 Node* obs = not_ctl->fast_out(j);
4905 if (obs->in(0) == not_ctl && obs->is_Call() &&
4906 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
4907 found_trap = true; break;
4908 }
4909 }
4910 if (found_trap) {
4911 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
4912 continue;
4913 }
4914 }
4915 return NULL;
4916 }
4917
4918 // If we get this far, we have an allocation which immediately
4919 // precedes the arraycopy, and we can take over zeroing the new object.
4920 // The arraycopy will finish the initialization, and provide
4921 // a new control state to which we will anchor the destination pointer.
4922
4923 return alloc;
4924 }
4925
4926 //-------------inline_encodeISOArray-----------------------------------
4927 // encode char[] to byte[] in ISO_8859_1
4928 bool LibraryCallKit::inline_encodeISOArray() {
4929 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
4930 // no receiver since it is static method
4931 Node *src = argument(0);
4932 Node *src_offset = argument(1);
4933 Node *dst = argument(2);
4934 Node *dst_offset = argument(3);
4935 Node *length = argument(4);
4936
4937 src = must_be_not_null(src, true);
4938 dst = must_be_not_null(dst, true);
4939
4940 src = access_resolve(src, ACCESS_READ);
4941 dst = access_resolve(dst, ACCESS_WRITE);
4942
4943 const Type* src_type = src->Value(&_gvn);
4944 const Type* dst_type = dst->Value(&_gvn);
4945 const TypeAryPtr* top_src = src_type->isa_aryptr();
4946 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
4947 if (top_src == NULL || top_src->klass() == NULL ||
4948 top_dest == NULL || top_dest->klass() == NULL) {
4949 // failed array check
4950 return false;
4951 }
4952
4953 // Figure out the size and type of the elements we will be copying.
4954 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4955 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4956 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
4957 return false;
4958 }
4959
4960 Node* src_start = array_element_address(src, src_offset, T_CHAR);
4961 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
4962 // 'src_start' points to src array + scaled offset
4963 // 'dst_start' points to dst array + scaled offset
4964
4965 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
4966 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
4967 enc = _gvn.transform(enc);
4968 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
4969 set_memory(res_mem, mtype);
4970 set_result(enc);
4971 clear_upper_avx();
4972
4973 return true;
4974 }
4975
4976 //-------------inline_multiplyToLen-----------------------------------
4977 bool LibraryCallKit::inline_multiplyToLen() {
4978 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
4979
4980 address stubAddr = StubRoutines::multiplyToLen();
4981 if (stubAddr == NULL) {
4982 return false; // Intrinsic's stub is not implemented on this platform
4983 }
4984 const char* stubName = "multiplyToLen";
4985
4986 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
4987
4988 // no receiver because it is a static method
4989 Node* x = argument(0);
4990 Node* xlen = argument(1);
4991 Node* y = argument(2);
4992 Node* ylen = argument(3);
4993 Node* z = argument(4);
4994
4995 x = must_be_not_null(x, true);
4996 y = must_be_not_null(y, true);
4997
4998 x = access_resolve(x, ACCESS_READ);
4999 y = access_resolve(y, ACCESS_READ);
5000 z = access_resolve(z, ACCESS_WRITE);
5001
5002 const Type* x_type = x->Value(&_gvn);
5003 const Type* y_type = y->Value(&_gvn);
5004 const TypeAryPtr* top_x = x_type->isa_aryptr();
5005 const TypeAryPtr* top_y = y_type->isa_aryptr();
5006 if (top_x == NULL || top_x->klass() == NULL ||
5007 top_y == NULL || top_y->klass() == NULL) {
5008 // failed array check
5009 return false;
5010 }
5011
5012 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5013 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5014 if (x_elem != T_INT || y_elem != T_INT) {
5015 return false;
5016 }
5017
5018 // Set the original stack and the reexecute bit for the interpreter to reexecute
5019 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5020 // on the return from z array allocation in runtime.
5021 { PreserveReexecuteState preexecs(this);
5022 jvms()->set_should_reexecute(true);
5023
5024 Node* x_start = array_element_address(x, intcon(0), x_elem);
5025 Node* y_start = array_element_address(y, intcon(0), y_elem);
5026 // 'x_start' points to x array + scaled xlen
5027 // 'y_start' points to y array + scaled ylen
5028
5029 // Allocate the result array
5030 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5031 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5032 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5033
5034 IdealKit ideal(this);
5035
5036 #define __ ideal.
5037 Node* one = __ ConI(1);
5038 Node* zero = __ ConI(0);
5039 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5040 __ set(need_alloc, zero);
5041 __ set(z_alloc, z);
5042 __ if_then(z, BoolTest::eq, null()); {
5043 __ increment (need_alloc, one);
5044 } __ else_(); {
5045 // Update graphKit memory and control from IdealKit.
5046 sync_kit(ideal);
5047 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5048 cast->init_req(0, control());
5049 _gvn.set_type(cast, cast->bottom_type());
5050 C->record_for_igvn(cast);
5051
5052 Node* zlen_arg = load_array_length(cast);
5053 // Update IdealKit memory and control from graphKit.
5054 __ sync_kit(this);
5055 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5056 __ increment (need_alloc, one);
5057 } __ end_if();
5058 } __ end_if();
5059
5060 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5061 // Update graphKit memory and control from IdealKit.
5062 sync_kit(ideal);
5063 Node * narr = new_array(klass_node, zlen, 1);
5064 // Update IdealKit memory and control from graphKit.
5065 __ sync_kit(this);
5066 __ set(z_alloc, narr);
5067 } __ end_if();
5068
5069 sync_kit(ideal);
5070 z = __ value(z_alloc);
5071 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5072 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5073 // Final sync IdealKit and GraphKit.
5074 final_sync(ideal);
5075 #undef __
5076
5077 Node* z_start = array_element_address(z, intcon(0), T_INT);
5078
5079 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5080 OptoRuntime::multiplyToLen_Type(),
5081 stubAddr, stubName, TypePtr::BOTTOM,
5082 x_start, xlen, y_start, ylen, z_start, zlen);
5083 } // original reexecute is set back here
5084
5085 C->set_has_split_ifs(true); // Has chance for split-if optimization
5086 set_result(z);
5087 return true;
5088 }
5089
5090 //-------------inline_squareToLen------------------------------------
5091 bool LibraryCallKit::inline_squareToLen() {
5092 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5093
5094 address stubAddr = StubRoutines::squareToLen();
5095 if (stubAddr == NULL) {
5096 return false; // Intrinsic's stub is not implemented on this platform
5097 }
5098 const char* stubName = "squareToLen";
5099
5100 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5101
5102 Node* x = argument(0);
5103 Node* len = argument(1);
5104 Node* z = argument(2);
5105 Node* zlen = argument(3);
5106
5107 x = must_be_not_null(x, true);
5108 z = must_be_not_null(z, true);
5109
5110 x = access_resolve(x, ACCESS_READ);
5111 z = access_resolve(z, ACCESS_WRITE);
5112
5113 const Type* x_type = x->Value(&_gvn);
5114 const Type* z_type = z->Value(&_gvn);
5115 const TypeAryPtr* top_x = x_type->isa_aryptr();
5116 const TypeAryPtr* top_z = z_type->isa_aryptr();
5117 if (top_x == NULL || top_x->klass() == NULL ||
5118 top_z == NULL || top_z->klass() == NULL) {
5119 // failed array check
5120 return false;
5121 }
5122
5123 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5124 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5125 if (x_elem != T_INT || z_elem != T_INT) {
5126 return false;
5127 }
5128
5129
5130 Node* x_start = array_element_address(x, intcon(0), x_elem);
5131 Node* z_start = array_element_address(z, intcon(0), z_elem);
5132
5133 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5134 OptoRuntime::squareToLen_Type(),
5135 stubAddr, stubName, TypePtr::BOTTOM,
5136 x_start, len, z_start, zlen);
5137
5138 set_result(z);
5139 return true;
5140 }
5141
5142 //-------------inline_mulAdd------------------------------------------
5143 bool LibraryCallKit::inline_mulAdd() {
5144 assert(UseMulAddIntrinsic, "not implemented on this platform");
5145
5146 address stubAddr = StubRoutines::mulAdd();
5147 if (stubAddr == NULL) {
5148 return false; // Intrinsic's stub is not implemented on this platform
5149 }
5150 const char* stubName = "mulAdd";
5151
5152 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5153
5154 Node* out = argument(0);
5155 Node* in = argument(1);
5156 Node* offset = argument(2);
5157 Node* len = argument(3);
5158 Node* k = argument(4);
5159
5160 out = must_be_not_null(out, true);
5161
5162 in = access_resolve(in, ACCESS_READ);
5163 out = access_resolve(out, ACCESS_WRITE);
5164
5165 const Type* out_type = out->Value(&_gvn);
5166 const Type* in_type = in->Value(&_gvn);
5167 const TypeAryPtr* top_out = out_type->isa_aryptr();
5168 const TypeAryPtr* top_in = in_type->isa_aryptr();
5169 if (top_out == NULL || top_out->klass() == NULL ||
5170 top_in == NULL || top_in->klass() == NULL) {
5171 // failed array check
5172 return false;
5173 }
5174
5175 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5176 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5177 if (out_elem != T_INT || in_elem != T_INT) {
5178 return false;
5179 }
5180
5181 Node* outlen = load_array_length(out);
5182 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5183 Node* out_start = array_element_address(out, intcon(0), out_elem);
5184 Node* in_start = array_element_address(in, intcon(0), in_elem);
5185
5186 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5187 OptoRuntime::mulAdd_Type(),
5188 stubAddr, stubName, TypePtr::BOTTOM,
5189 out_start,in_start, new_offset, len, k);
5190 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5191 set_result(result);
5192 return true;
5193 }
5194
5195 //-------------inline_montgomeryMultiply-----------------------------------
5196 bool LibraryCallKit::inline_montgomeryMultiply() {
5197 address stubAddr = StubRoutines::montgomeryMultiply();
5198 if (stubAddr == NULL) {
5199 return false; // Intrinsic's stub is not implemented on this platform
5200 }
5201
5202 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5203 const char* stubName = "montgomery_multiply";
5204
5205 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5206
5207 Node* a = argument(0);
5208 Node* b = argument(1);
5209 Node* n = argument(2);
5210 Node* len = argument(3);
5211 Node* inv = argument(4);
5212 Node* m = argument(6);
5213
5214 a = access_resolve(a, ACCESS_READ);
5215 b = access_resolve(b, ACCESS_READ);
5216 n = access_resolve(n, ACCESS_READ);
5217 m = access_resolve(m, ACCESS_WRITE);
5218
5219 const Type* a_type = a->Value(&_gvn);
5220 const TypeAryPtr* top_a = a_type->isa_aryptr();
5221 const Type* b_type = b->Value(&_gvn);
5222 const TypeAryPtr* top_b = b_type->isa_aryptr();
5223 const Type* n_type = a->Value(&_gvn);
5224 const TypeAryPtr* top_n = n_type->isa_aryptr();
5225 const Type* m_type = a->Value(&_gvn);
5226 const TypeAryPtr* top_m = m_type->isa_aryptr();
5227 if (top_a == NULL || top_a->klass() == NULL ||
5228 top_b == NULL || top_b->klass() == NULL ||
5229 top_n == NULL || top_n->klass() == NULL ||
5230 top_m == NULL || top_m->klass() == NULL) {
5231 // failed array check
5232 return false;
5233 }
5234
5235 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5236 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5237 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5238 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5239 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5240 return false;
5241 }
5242
5243 // Make the call
5244 {
5245 Node* a_start = array_element_address(a, intcon(0), a_elem);
5246 Node* b_start = array_element_address(b, intcon(0), b_elem);
5247 Node* n_start = array_element_address(n, intcon(0), n_elem);
5248 Node* m_start = array_element_address(m, intcon(0), m_elem);
5249
5250 Node* call = make_runtime_call(RC_LEAF,
5251 OptoRuntime::montgomeryMultiply_Type(),
5252 stubAddr, stubName, TypePtr::BOTTOM,
5253 a_start, b_start, n_start, len, inv, top(),
5254 m_start);
5255 set_result(m);
5256 }
5257
5258 return true;
5259 }
5260
5261 bool LibraryCallKit::inline_montgomerySquare() {
5262 address stubAddr = StubRoutines::montgomerySquare();
5263 if (stubAddr == NULL) {
5264 return false; // Intrinsic's stub is not implemented on this platform
5265 }
5266
5267 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5268 const char* stubName = "montgomery_square";
5269
5270 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5271
5272 Node* a = argument(0);
5273 Node* n = argument(1);
5274 Node* len = argument(2);
5275 Node* inv = argument(3);
5276 Node* m = argument(5);
5277
5278 a = access_resolve(a, ACCESS_READ);
5279 n = access_resolve(n, ACCESS_READ);
5280 m = access_resolve(m, ACCESS_WRITE);
5281
5282 const Type* a_type = a->Value(&_gvn);
5283 const TypeAryPtr* top_a = a_type->isa_aryptr();
5284 const Type* n_type = a->Value(&_gvn);
5285 const TypeAryPtr* top_n = n_type->isa_aryptr();
5286 const Type* m_type = a->Value(&_gvn);
5287 const TypeAryPtr* top_m = m_type->isa_aryptr();
5288 if (top_a == NULL || top_a->klass() == NULL ||
5289 top_n == NULL || top_n->klass() == NULL ||
5290 top_m == NULL || top_m->klass() == NULL) {
5291 // failed array check
5292 return false;
5293 }
5294
5295 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5296 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5297 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5298 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5299 return false;
5300 }
5301
5302 // Make the call
5303 {
5304 Node* a_start = array_element_address(a, intcon(0), a_elem);
5305 Node* n_start = array_element_address(n, intcon(0), n_elem);
5306 Node* m_start = array_element_address(m, intcon(0), m_elem);
5307
5308 Node* call = make_runtime_call(RC_LEAF,
5309 OptoRuntime::montgomerySquare_Type(),
5310 stubAddr, stubName, TypePtr::BOTTOM,
5311 a_start, n_start, len, inv, top(),
5312 m_start);
5313 set_result(m);
5314 }
5315
5316 return true;
5317 }
5318
5319 //-------------inline_vectorizedMismatch------------------------------
5320 bool LibraryCallKit::inline_vectorizedMismatch() {
5321 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5322
5323 address stubAddr = StubRoutines::vectorizedMismatch();
5324 if (stubAddr == NULL) {
5325 return false; // Intrinsic's stub is not implemented on this platform
5326 }
5327 const char* stubName = "vectorizedMismatch";
5328 int size_l = callee()->signature()->size();
5329 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5330
5331 Node* obja = argument(0);
5332 Node* aoffset = argument(1);
5333 Node* objb = argument(3);
5334 Node* boffset = argument(4);
5335 Node* length = argument(6);
5336 Node* scale = argument(7);
5337
5338 const Type* a_type = obja->Value(&_gvn);
5339 const Type* b_type = objb->Value(&_gvn);
5340 const TypeAryPtr* top_a = a_type->isa_aryptr();
5341 const TypeAryPtr* top_b = b_type->isa_aryptr();
5342 if (top_a == NULL || top_a->klass() == NULL ||
5343 top_b == NULL || top_b->klass() == NULL) {
5344 // failed array check
5345 return false;
5346 }
5347
5348 Node* call;
5349 jvms()->set_should_reexecute(true);
5350
5351 obja = access_resolve(obja, ACCESS_READ);
5352 objb = access_resolve(objb, ACCESS_READ);
5353 Node* obja_adr = make_unsafe_address(obja, aoffset, ACCESS_READ);
5354 Node* objb_adr = make_unsafe_address(objb, boffset, ACCESS_READ);
5355
5356 call = make_runtime_call(RC_LEAF,
5357 OptoRuntime::vectorizedMismatch_Type(),
5358 stubAddr, stubName, TypePtr::BOTTOM,
5359 obja_adr, objb_adr, length, scale);
5360
5361 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5362 set_result(result);
5363 return true;
5364 }
5365
5366 /**
5367 * Calculate CRC32 for byte.
5368 * int java.util.zip.CRC32.update(int crc, int b)
5369 */
5370 bool LibraryCallKit::inline_updateCRC32() {
5371 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5372 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5373 // no receiver since it is static method
5374 Node* crc = argument(0); // type: int
5375 Node* b = argument(1); // type: int
5376
5377 /*
5378 * int c = ~ crc;
5379 * b = timesXtoThe32[(b ^ c) & 0xFF];
5380 * b = b ^ (c >>> 8);
5381 * crc = ~b;
5382 */
5383
5384 Node* M1 = intcon(-1);
5385 crc = _gvn.transform(new XorINode(crc, M1));
5386 Node* result = _gvn.transform(new XorINode(crc, b));
5387 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5388
5389 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5390 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5391 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5392 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5393
5394 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5395 result = _gvn.transform(new XorINode(crc, result));
5396 result = _gvn.transform(new XorINode(result, M1));
5397 set_result(result);
5398 return true;
5399 }
5400
5401 /**
5402 * Calculate CRC32 for byte[] array.
5403 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5404 */
5405 bool LibraryCallKit::inline_updateBytesCRC32() {
5406 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5407 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5408 // no receiver since it is static method
5409 Node* crc = argument(0); // type: int
5410 Node* src = argument(1); // type: oop
5411 Node* offset = argument(2); // type: int
5412 Node* length = argument(3); // type: int
5413
5414 const Type* src_type = src->Value(&_gvn);
5415 const TypeAryPtr* top_src = src_type->isa_aryptr();
5416 if (top_src == NULL || top_src->klass() == NULL) {
5417 // failed array check
5418 return false;
5419 }
5420
5421 // Figure out the size and type of the elements we will be copying.
5422 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5423 if (src_elem != T_BYTE) {
5424 return false;
5425 }
5426
5427 // 'src_start' points to src array + scaled offset
5428 src = must_be_not_null(src, true);
5429 src = access_resolve(src, ACCESS_READ);
5430 Node* src_start = array_element_address(src, offset, src_elem);
5431
5432 // We assume that range check is done by caller.
5433 // TODO: generate range check (offset+length < src.length) in debug VM.
5434
5435 // Call the stub.
5436 address stubAddr = StubRoutines::updateBytesCRC32();
5437 const char *stubName = "updateBytesCRC32";
5438
5439 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5440 stubAddr, stubName, TypePtr::BOTTOM,
5441 crc, src_start, length);
5442 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5443 set_result(result);
5444 return true;
5445 }
5446
5447 /**
5448 * Calculate CRC32 for ByteBuffer.
5449 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5450 */
5451 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5452 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5453 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5454 // no receiver since it is static method
5455 Node* crc = argument(0); // type: int
5456 Node* src = argument(1); // type: long
5457 Node* offset = argument(3); // type: int
5458 Node* length = argument(4); // type: int
5459
5460 src = ConvL2X(src); // adjust Java long to machine word
5461 Node* base = _gvn.transform(new CastX2PNode(src));
5462 offset = ConvI2X(offset);
5463
5464 // 'src_start' points to src array + scaled offset
5465 Node* src_start = basic_plus_adr(top(), base, offset);
5466
5467 // Call the stub.
5468 address stubAddr = StubRoutines::updateBytesCRC32();
5469 const char *stubName = "updateBytesCRC32";
5470
5471 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5472 stubAddr, stubName, TypePtr::BOTTOM,
5473 crc, src_start, length);
5474 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5475 set_result(result);
5476 return true;
5477 }
5478
5479 //------------------------------get_table_from_crc32c_class-----------------------
5480 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5481 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5482 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5483
5484 return table;
5485 }
5486
5487 //------------------------------inline_updateBytesCRC32C-----------------------
5488 //
5489 // Calculate CRC32C for byte[] array.
5490 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5491 //
5492 bool LibraryCallKit::inline_updateBytesCRC32C() {
5493 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5494 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5495 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5496 // no receiver since it is a static method
5497 Node* crc = argument(0); // type: int
5498 Node* src = argument(1); // type: oop
5499 Node* offset = argument(2); // type: int
5500 Node* end = argument(3); // type: int
5501
5502 Node* length = _gvn.transform(new SubINode(end, offset));
5503
5504 const Type* src_type = src->Value(&_gvn);
5505 const TypeAryPtr* top_src = src_type->isa_aryptr();
5506 if (top_src == NULL || top_src->klass() == NULL) {
5507 // failed array check
5508 return false;
5509 }
5510
5511 // Figure out the size and type of the elements we will be copying.
5512 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5513 if (src_elem != T_BYTE) {
5514 return false;
5515 }
5516
5517 // 'src_start' points to src array + scaled offset
5518 src = must_be_not_null(src, true);
5519 src = access_resolve(src, ACCESS_READ);
5520 Node* src_start = array_element_address(src, offset, src_elem);
5521
5522 // static final int[] byteTable in class CRC32C
5523 Node* table = get_table_from_crc32c_class(callee()->holder());
5524 table = must_be_not_null(table, true);
5525 table = access_resolve(table, ACCESS_READ);
5526 Node* table_start = array_element_address(table, intcon(0), T_INT);
5527
5528 // We assume that range check is done by caller.
5529 // TODO: generate range check (offset+length < src.length) in debug VM.
5530
5531 // Call the stub.
5532 address stubAddr = StubRoutines::updateBytesCRC32C();
5533 const char *stubName = "updateBytesCRC32C";
5534
5535 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5536 stubAddr, stubName, TypePtr::BOTTOM,
5537 crc, src_start, length, table_start);
5538 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5539 set_result(result);
5540 return true;
5541 }
5542
5543 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5544 //
5545 // Calculate CRC32C for DirectByteBuffer.
5546 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5547 //
5548 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5549 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5550 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5551 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5552 // no receiver since it is a static method
5553 Node* crc = argument(0); // type: int
5554 Node* src = argument(1); // type: long
5555 Node* offset = argument(3); // type: int
5556 Node* end = argument(4); // type: int
5557
5558 Node* length = _gvn.transform(new SubINode(end, offset));
5559
5560 src = ConvL2X(src); // adjust Java long to machine word
5561 Node* base = _gvn.transform(new CastX2PNode(src));
5562 offset = ConvI2X(offset);
5563
5564 // 'src_start' points to src array + scaled offset
5565 Node* src_start = basic_plus_adr(top(), base, offset);
5566
5567 // static final int[] byteTable in class CRC32C
5568 Node* table = get_table_from_crc32c_class(callee()->holder());
5569 table = must_be_not_null(table, true);
5570 table = access_resolve(table, ACCESS_READ);
5571 Node* table_start = array_element_address(table, intcon(0), T_INT);
5572
5573 // Call the stub.
5574 address stubAddr = StubRoutines::updateBytesCRC32C();
5575 const char *stubName = "updateBytesCRC32C";
5576
5577 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5578 stubAddr, stubName, TypePtr::BOTTOM,
5579 crc, src_start, length, table_start);
5580 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5581 set_result(result);
5582 return true;
5583 }
5584
5585 //------------------------------inline_updateBytesAdler32----------------------
5586 //
5587 // Calculate Adler32 checksum for byte[] array.
5588 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5589 //
5590 bool LibraryCallKit::inline_updateBytesAdler32() {
5591 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5592 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5593 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5594 // no receiver since it is static method
5595 Node* crc = argument(0); // type: int
5596 Node* src = argument(1); // type: oop
5597 Node* offset = argument(2); // type: int
5598 Node* length = argument(3); // type: int
5599
5600 const Type* src_type = src->Value(&_gvn);
5601 const TypeAryPtr* top_src = src_type->isa_aryptr();
5602 if (top_src == NULL || top_src->klass() == NULL) {
5603 // failed array check
5604 return false;
5605 }
5606
5607 // Figure out the size and type of the elements we will be copying.
5608 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5609 if (src_elem != T_BYTE) {
5610 return false;
5611 }
5612
5613 // 'src_start' points to src array + scaled offset
5614 src = access_resolve(src, ACCESS_READ);
5615 Node* src_start = array_element_address(src, offset, src_elem);
5616
5617 // We assume that range check is done by caller.
5618 // TODO: generate range check (offset+length < src.length) in debug VM.
5619
5620 // Call the stub.
5621 address stubAddr = StubRoutines::updateBytesAdler32();
5622 const char *stubName = "updateBytesAdler32";
5623
5624 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5625 stubAddr, stubName, TypePtr::BOTTOM,
5626 crc, src_start, length);
5627 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5628 set_result(result);
5629 return true;
5630 }
5631
5632 //------------------------------inline_updateByteBufferAdler32---------------
5633 //
5634 // Calculate Adler32 checksum for DirectByteBuffer.
5635 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5636 //
5637 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5638 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5639 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5640 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5641 // no receiver since it is static method
5642 Node* crc = argument(0); // type: int
5643 Node* src = argument(1); // type: long
5644 Node* offset = argument(3); // type: int
5645 Node* length = argument(4); // type: int
5646
5647 src = ConvL2X(src); // adjust Java long to machine word
5648 Node* base = _gvn.transform(new CastX2PNode(src));
5649 offset = ConvI2X(offset);
5650
5651 // 'src_start' points to src array + scaled offset
5652 Node* src_start = basic_plus_adr(top(), base, offset);
5653
5654 // Call the stub.
5655 address stubAddr = StubRoutines::updateBytesAdler32();
5656 const char *stubName = "updateBytesAdler32";
5657
5658 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5659 stubAddr, stubName, TypePtr::BOTTOM,
5660 crc, src_start, length);
5661
5662 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5663 set_result(result);
5664 return true;
5665 }
5666
5667 //----------------------------inline_reference_get----------------------------
5668 // public T java.lang.ref.Reference.get();
5669 bool LibraryCallKit::inline_reference_get() {
5670 const int referent_offset = java_lang_ref_Reference::referent_offset;
5671 guarantee(referent_offset > 0, "should have already been set");
5672
5673 // Get the argument:
5674 Node* reference_obj = null_check_receiver();
5675 if (stopped()) return true;
5676
5677 const TypeInstPtr* tinst = _gvn.type(reference_obj)->isa_instptr();
5678 assert(tinst != NULL, "obj is null");
5679 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5680 ciInstanceKlass* referenceKlass = tinst->klass()->as_instance_klass();
5681 ciField* field = referenceKlass->get_field_by_name(ciSymbol::make("referent"),
5682 ciSymbol::make("Ljava/lang/Object;"),
5683 false);
5684 assert (field != NULL, "undefined field");
5685
5686 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5687 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5688
5689 ciInstanceKlass* klass = env()->Object_klass();
5690 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5691
5692 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
5693 Node* result = access_load_at(reference_obj, adr, adr_type, object_type, T_OBJECT, decorators);
5694 // Add memory barrier to prevent commoning reads from this field
5695 // across safepoint since GC can change its value.
5696 insert_mem_bar(Op_MemBarCPUOrder);
5697
5698 set_result(result);
5699 return true;
5700 }
5701
5702
5703 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5704 bool is_exact=true, bool is_static=false,
5705 ciInstanceKlass * fromKls=NULL) {
5706 if (fromKls == NULL) {
5707 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5708 assert(tinst != NULL, "obj is null");
5709 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5710 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5711 fromKls = tinst->klass()->as_instance_klass();
5712 } else {
5713 assert(is_static, "only for static field access");
5714 }
5715 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5716 ciSymbol::make(fieldTypeString),
5717 is_static);
5718
5719 assert (field != NULL, "undefined field");
5720 if (field == NULL) return (Node *) NULL;
5721
5722 if (is_static) {
5723 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5724 fromObj = makecon(tip);
5725 }
5726
5727 // Next code copied from Parse::do_get_xxx():
5728
5729 // Compute address and memory type.
5730 int offset = field->offset_in_bytes();
5731 bool is_vol = field->is_volatile();
5732 ciType* field_klass = field->type();
5733 assert(field_klass->is_loaded(), "should be loaded");
5734 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5735 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5736 BasicType bt = field->layout_type();
5737
5738 // Build the resultant type of the load
5739 const Type *type;
5740 if (bt == T_OBJECT) {
5741 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5742 } else {
5743 type = Type::get_const_basic_type(bt);
5744 }
5745
5746 DecoratorSet decorators = IN_HEAP;
5747
5748 if (is_vol) {
5749 decorators |= MO_SEQ_CST;
5750 }
5751
5752 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
5753 }
5754
5755 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5756 bool is_exact = true, bool is_static = false,
5757 ciInstanceKlass * fromKls = NULL) {
5758 if (fromKls == NULL) {
5759 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5760 assert(tinst != NULL, "obj is null");
5761 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5762 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5763 fromKls = tinst->klass()->as_instance_klass();
5764 }
5765 else {
5766 assert(is_static, "only for static field access");
5767 }
5768 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5769 ciSymbol::make(fieldTypeString),
5770 is_static);
5771
5772 assert(field != NULL, "undefined field");
5773 assert(!field->is_volatile(), "not defined for volatile fields");
5774
5775 if (is_static) {
5776 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5777 fromObj = makecon(tip);
5778 }
5779
5780 // Next code copied from Parse::do_get_xxx():
5781
5782 // Compute address and memory type.
5783 int offset = field->offset_in_bytes();
5784 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5785
5786 return adr;
5787 }
5788
5789 //------------------------------inline_aescrypt_Block-----------------------
5790 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
5791 address stubAddr = NULL;
5792 const char *stubName;
5793 assert(UseAES, "need AES instruction support");
5794
5795 switch(id) {
5796 case vmIntrinsics::_aescrypt_encryptBlock:
5797 stubAddr = StubRoutines::aescrypt_encryptBlock();
5798 stubName = "aescrypt_encryptBlock";
5799 break;
5800 case vmIntrinsics::_aescrypt_decryptBlock:
5801 stubAddr = StubRoutines::aescrypt_decryptBlock();
5802 stubName = "aescrypt_decryptBlock";
5803 break;
5804 default:
5805 break;
5806 }
5807 if (stubAddr == NULL) return false;
5808
5809 Node* aescrypt_object = argument(0);
5810 Node* src = argument(1);
5811 Node* src_offset = argument(2);
5812 Node* dest = argument(3);
5813 Node* dest_offset = argument(4);
5814
5815 src = must_be_not_null(src, true);
5816 dest = must_be_not_null(dest, true);
5817
5818 src = access_resolve(src, ACCESS_READ);
5819 dest = access_resolve(dest, ACCESS_WRITE);
5820
5821 // (1) src and dest are arrays.
5822 const Type* src_type = src->Value(&_gvn);
5823 const Type* dest_type = dest->Value(&_gvn);
5824 const TypeAryPtr* top_src = src_type->isa_aryptr();
5825 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5826 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5827
5828 // for the quick and dirty code we will skip all the checks.
5829 // we are just trying to get the call to be generated.
5830 Node* src_start = src;
5831 Node* dest_start = dest;
5832 if (src_offset != NULL || dest_offset != NULL) {
5833 assert(src_offset != NULL && dest_offset != NULL, "");
5834 src_start = array_element_address(src, src_offset, T_BYTE);
5835 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5836 }
5837
5838 // now need to get the start of its expanded key array
5839 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5840 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5841 if (k_start == NULL) return false;
5842
5843 if (Matcher::pass_original_key_for_aes()) {
5844 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5845 // compatibility issues between Java key expansion and SPARC crypto instructions
5846 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5847 if (original_k_start == NULL) return false;
5848
5849 // Call the stub.
5850 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5851 stubAddr, stubName, TypePtr::BOTTOM,
5852 src_start, dest_start, k_start, original_k_start);
5853 } else {
5854 // Call the stub.
5855 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5856 stubAddr, stubName, TypePtr::BOTTOM,
5857 src_start, dest_start, k_start);
5858 }
5859
5860 return true;
5861 }
5862
5863 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
5864 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
5865 address stubAddr = NULL;
5866 const char *stubName = NULL;
5867
5868 assert(UseAES, "need AES instruction support");
5869
5870 switch(id) {
5871 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
5872 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
5873 stubName = "cipherBlockChaining_encryptAESCrypt";
5874 break;
5875 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
5876 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
5877 stubName = "cipherBlockChaining_decryptAESCrypt";
5878 break;
5879 default:
5880 break;
5881 }
5882 if (stubAddr == NULL) return false;
5883
5884 Node* cipherBlockChaining_object = argument(0);
5885 Node* src = argument(1);
5886 Node* src_offset = argument(2);
5887 Node* len = argument(3);
5888 Node* dest = argument(4);
5889 Node* dest_offset = argument(5);
5890
5891 src = must_be_not_null(src, false);
5892 dest = must_be_not_null(dest, false);
5893
5894 src = access_resolve(src, ACCESS_READ);
5895 dest = access_resolve(dest, ACCESS_WRITE);
5896
5897 // (1) src and dest are arrays.
5898 const Type* src_type = src->Value(&_gvn);
5899 const Type* dest_type = dest->Value(&_gvn);
5900 const TypeAryPtr* top_src = src_type->isa_aryptr();
5901 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5902 assert (top_src != NULL && top_src->klass() != NULL
5903 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5904
5905 // checks are the responsibility of the caller
5906 Node* src_start = src;
5907 Node* dest_start = dest;
5908 if (src_offset != NULL || dest_offset != NULL) {
5909 assert(src_offset != NULL && dest_offset != NULL, "");
5910 src_start = array_element_address(src, src_offset, T_BYTE);
5911 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5912 }
5913
5914 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
5915 // (because of the predicated logic executed earlier).
5916 // so we cast it here safely.
5917 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5918
5919 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
5920 if (embeddedCipherObj == NULL) return false;
5921
5922 // cast it to what we know it will be at runtime
5923 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
5924 assert(tinst != NULL, "CBC obj is null");
5925 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
5926 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5927 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
5928
5929 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
5930 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
5931 const TypeOopPtr* xtype = aklass->as_instance_type();
5932 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
5933 aescrypt_object = _gvn.transform(aescrypt_object);
5934
5935 // we need to get the start of the aescrypt_object's expanded key array
5936 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5937 if (k_start == NULL) return false;
5938
5939 // similarly, get the start address of the r vector
5940 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
5941 if (objRvec == NULL) return false;
5942 objRvec = access_resolve(objRvec, ACCESS_WRITE);
5943 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
5944
5945 Node* cbcCrypt;
5946 if (Matcher::pass_original_key_for_aes()) {
5947 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5948 // compatibility issues between Java key expansion and SPARC crypto instructions
5949 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5950 if (original_k_start == NULL) return false;
5951
5952 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
5953 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5954 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5955 stubAddr, stubName, TypePtr::BOTTOM,
5956 src_start, dest_start, k_start, r_start, len, original_k_start);
5957 } else {
5958 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
5959 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5960 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5961 stubAddr, stubName, TypePtr::BOTTOM,
5962 src_start, dest_start, k_start, r_start, len);
5963 }
5964
5965 // return cipher length (int)
5966 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
5967 set_result(retvalue);
5968 return true;
5969 }
5970
5971 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
5972 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
5973 address stubAddr = NULL;
5974 const char *stubName = NULL;
5975
5976 assert(UseAES, "need AES instruction support");
5977
5978 switch (id) {
5979 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
5980 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
5981 stubName = "electronicCodeBook_encryptAESCrypt";
5982 break;
5983 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
5984 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
5985 stubName = "electronicCodeBook_decryptAESCrypt";
5986 break;
5987 default:
5988 break;
5989 }
5990
5991 if (stubAddr == NULL) return false;
5992
5993 Node* electronicCodeBook_object = argument(0);
5994 Node* src = argument(1);
5995 Node* src_offset = argument(2);
5996 Node* len = argument(3);
5997 Node* dest = argument(4);
5998 Node* dest_offset = argument(5);
5999
6000 // (1) src and dest are arrays.
6001 const Type* src_type = src->Value(&_gvn);
6002 const Type* dest_type = dest->Value(&_gvn);
6003 const TypeAryPtr* top_src = src_type->isa_aryptr();
6004 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6005 assert(top_src != NULL && top_src->klass() != NULL
6006 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6007
6008 // checks are the responsibility of the caller
6009 Node* src_start = src;
6010 Node* dest_start = dest;
6011 if (src_offset != NULL || dest_offset != NULL) {
6012 assert(src_offset != NULL && dest_offset != NULL, "");
6013 src_start = array_element_address(src, src_offset, T_BYTE);
6014 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6015 }
6016
6017 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6018 // (because of the predicated logic executed earlier).
6019 // so we cast it here safely.
6020 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6021
6022 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6023 if (embeddedCipherObj == NULL) return false;
6024
6025 // cast it to what we know it will be at runtime
6026 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6027 assert(tinst != NULL, "ECB obj is null");
6028 assert(tinst->klass()->is_loaded(), "ECB obj is not loaded");
6029 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6030 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6031
6032 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6033 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6034 const TypeOopPtr* xtype = aklass->as_instance_type();
6035 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6036 aescrypt_object = _gvn.transform(aescrypt_object);
6037
6038 // we need to get the start of the aescrypt_object's expanded key array
6039 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6040 if (k_start == NULL) return false;
6041
6042 Node* ecbCrypt;
6043 if (Matcher::pass_original_key_for_aes()) {
6044 // no SPARC version for AES/ECB intrinsics now.
6045 return false;
6046 }
6047 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6048 ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6049 OptoRuntime::electronicCodeBook_aescrypt_Type(),
6050 stubAddr, stubName, TypePtr::BOTTOM,
6051 src_start, dest_start, k_start, len);
6052
6053 // return cipher length (int)
6054 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6055 set_result(retvalue);
6056 return true;
6057 }
6058
6059 //------------------------------inline_counterMode_AESCrypt-----------------------
6060 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6061 assert(UseAES, "need AES instruction support");
6062 if (!UseAESCTRIntrinsics) return false;
6063
6064 address stubAddr = NULL;
6065 const char *stubName = NULL;
6066 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6067 stubAddr = StubRoutines::counterMode_AESCrypt();
6068 stubName = "counterMode_AESCrypt";
6069 }
6070 if (stubAddr == NULL) return false;
6071
6072 Node* counterMode_object = argument(0);
6073 Node* src = argument(1);
6074 Node* src_offset = argument(2);
6075 Node* len = argument(3);
6076 Node* dest = argument(4);
6077 Node* dest_offset = argument(5);
6078
6079 src = access_resolve(src, ACCESS_READ);
6080 dest = access_resolve(dest, ACCESS_WRITE);
6081 counterMode_object = access_resolve(counterMode_object, ACCESS_WRITE);
6082
6083 // (1) src and dest are arrays.
6084 const Type* src_type = src->Value(&_gvn);
6085 const Type* dest_type = dest->Value(&_gvn);
6086 const TypeAryPtr* top_src = src_type->isa_aryptr();
6087 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6088 assert(top_src != NULL && top_src->klass() != NULL &&
6089 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6090
6091 // checks are the responsibility of the caller
6092 Node* src_start = src;
6093 Node* dest_start = dest;
6094 if (src_offset != NULL || dest_offset != NULL) {
6095 assert(src_offset != NULL && dest_offset != NULL, "");
6096 src_start = array_element_address(src, src_offset, T_BYTE);
6097 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6098 }
6099
6100 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6101 // (because of the predicated logic executed earlier).
6102 // so we cast it here safely.
6103 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6104 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6105 if (embeddedCipherObj == NULL) return false;
6106 // cast it to what we know it will be at runtime
6107 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6108 assert(tinst != NULL, "CTR obj is null");
6109 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6110 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6111 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6112 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6113 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6114 const TypeOopPtr* xtype = aklass->as_instance_type();
6115 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6116 aescrypt_object = _gvn.transform(aescrypt_object);
6117 // we need to get the start of the aescrypt_object's expanded key array
6118 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6119 if (k_start == NULL) return false;
6120 // similarly, get the start address of the r vector
6121 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6122 if (obj_counter == NULL) return false;
6123 obj_counter = access_resolve(obj_counter, ACCESS_WRITE);
6124 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6125
6126 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6127 if (saved_encCounter == NULL) return false;
6128 saved_encCounter = access_resolve(saved_encCounter, ACCESS_WRITE);
6129 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6130 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6131
6132 Node* ctrCrypt;
6133 if (Matcher::pass_original_key_for_aes()) {
6134 // no SPARC version for AES/CTR intrinsics now.
6135 return false;
6136 }
6137 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6138 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6139 OptoRuntime::counterMode_aescrypt_Type(),
6140 stubAddr, stubName, TypePtr::BOTTOM,
6141 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6142
6143 // return cipher length (int)
6144 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6145 set_result(retvalue);
6146 return true;
6147 }
6148
6149 //------------------------------get_key_start_from_aescrypt_object-----------------------
6150 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6151 #if defined(PPC64) || defined(S390)
6152 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6153 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6154 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6155 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6156 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6157 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6158 if (objSessionK == NULL) {
6159 return (Node *) NULL;
6160 }
6161 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6162 #else
6163 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6164 #endif // PPC64
6165 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6166 if (objAESCryptKey == NULL) return (Node *) NULL;
6167
6168 // now have the array, need to get the start address of the K array
6169 objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ);
6170 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6171 return k_start;
6172 }
6173
6174 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6175 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6176 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6177 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6178 if (objAESCryptKey == NULL) return (Node *) NULL;
6179
6180 // now have the array, need to get the start address of the lastKey array
6181 objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ);
6182 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6183 return original_k_start;
6184 }
6185
6186 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6187 // Return node representing slow path of predicate check.
6188 // the pseudo code we want to emulate with this predicate is:
6189 // for encryption:
6190 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6191 // for decryption:
6192 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6193 // note cipher==plain is more conservative than the original java code but that's OK
6194 //
6195 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6196 // The receiver was checked for NULL already.
6197 Node* objCBC = argument(0);
6198
6199 Node* src = argument(1);
6200 Node* dest = argument(4);
6201
6202 // Load embeddedCipher field of CipherBlockChaining object.
6203 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6204
6205 // get AESCrypt klass for instanceOf check
6206 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6207 // will have same classloader as CipherBlockChaining object
6208 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6209 assert(tinst != NULL, "CBCobj is null");
6210 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6211
6212 // we want to do an instanceof comparison against the AESCrypt class
6213 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6214 if (!klass_AESCrypt->is_loaded()) {
6215 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6216 Node* ctrl = control();
6217 set_control(top()); // no regular fast path
6218 return ctrl;
6219 }
6220
6221 src = must_be_not_null(src, true);
6222 dest = must_be_not_null(dest, true);
6223
6224 // Resolve oops to stable for CmpP below.
6225 src = access_resolve(src, 0);
6226 dest = access_resolve(dest, 0);
6227
6228 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6229
6230 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6231 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6232 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6233
6234 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6235
6236 // for encryption, we are done
6237 if (!decrypting)
6238 return instof_false; // even if it is NULL
6239
6240 // for decryption, we need to add a further check to avoid
6241 // taking the intrinsic path when cipher and plain are the same
6242 // see the original java code for why.
6243 RegionNode* region = new RegionNode(3);
6244 region->init_req(1, instof_false);
6245
6246 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6247 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6248 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6249 region->init_req(2, src_dest_conjoint);
6250
6251 record_for_igvn(region);
6252 return _gvn.transform(region);
6253 }
6254
6255 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
6256 // Return node representing slow path of predicate check.
6257 // the pseudo code we want to emulate with this predicate is:
6258 // for encryption:
6259 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6260 // for decryption:
6261 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6262 // note cipher==plain is more conservative than the original java code but that's OK
6263 //
6264 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
6265 // The receiver was checked for NULL already.
6266 Node* objECB = argument(0);
6267
6268 // Load embeddedCipher field of ElectronicCodeBook object.
6269 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6270
6271 // get AESCrypt klass for instanceOf check
6272 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6273 // will have same classloader as ElectronicCodeBook object
6274 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
6275 assert(tinst != NULL, "ECBobj is null");
6276 assert(tinst->klass()->is_loaded(), "ECBobj is not loaded");
6277
6278 // we want to do an instanceof comparison against the AESCrypt class
6279 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6280 if (!klass_AESCrypt->is_loaded()) {
6281 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6282 Node* ctrl = control();
6283 set_control(top()); // no regular fast path
6284 return ctrl;
6285 }
6286 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6287
6288 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6289 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6290 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6291
6292 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6293
6294 // for encryption, we are done
6295 if (!decrypting)
6296 return instof_false; // even if it is NULL
6297
6298 // for decryption, we need to add a further check to avoid
6299 // taking the intrinsic path when cipher and plain are the same
6300 // see the original java code for why.
6301 RegionNode* region = new RegionNode(3);
6302 region->init_req(1, instof_false);
6303 Node* src = argument(1);
6304 Node* dest = argument(4);
6305 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6306 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6307 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6308 region->init_req(2, src_dest_conjoint);
6309
6310 record_for_igvn(region);
6311 return _gvn.transform(region);
6312 }
6313
6314 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6315 // Return node representing slow path of predicate check.
6316 // the pseudo code we want to emulate with this predicate is:
6317 // for encryption:
6318 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6319 // for decryption:
6320 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6321 // note cipher==plain is more conservative than the original java code but that's OK
6322 //
6323
6324 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6325 // The receiver was checked for NULL already.
6326 Node* objCTR = argument(0);
6327
6328 // Load embeddedCipher field of CipherBlockChaining object.
6329 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6330
6331 // get AESCrypt klass for instanceOf check
6332 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6333 // will have same classloader as CipherBlockChaining object
6334 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6335 assert(tinst != NULL, "CTRobj is null");
6336 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6337
6338 // we want to do an instanceof comparison against the AESCrypt class
6339 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6340 if (!klass_AESCrypt->is_loaded()) {
6341 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6342 Node* ctrl = control();
6343 set_control(top()); // no regular fast path
6344 return ctrl;
6345 }
6346
6347 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6348 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6349 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6350 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6351 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6352
6353 return instof_false; // even if it is NULL
6354 }
6355
6356 //------------------------------inline_ghash_processBlocks
6357 bool LibraryCallKit::inline_ghash_processBlocks() {
6358 address stubAddr;
6359 const char *stubName;
6360 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6361
6362 stubAddr = StubRoutines::ghash_processBlocks();
6363 stubName = "ghash_processBlocks";
6364
6365 Node* data = argument(0);
6366 Node* offset = argument(1);
6367 Node* len = argument(2);
6368 Node* state = argument(3);
6369 Node* subkeyH = argument(4);
6370
6371 state = must_be_not_null(state, true);
6372 subkeyH = must_be_not_null(subkeyH, true);
6373 data = must_be_not_null(data, true);
6374
6375 state = access_resolve(state, ACCESS_WRITE);
6376 subkeyH = access_resolve(subkeyH, ACCESS_READ);
6377 data = access_resolve(data, ACCESS_READ);
6378
6379 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6380 assert(state_start, "state is NULL");
6381 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6382 assert(subkeyH_start, "subkeyH is NULL");
6383 Node* data_start = array_element_address(data, offset, T_BYTE);
6384 assert(data_start, "data is NULL");
6385
6386 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6387 OptoRuntime::ghash_processBlocks_Type(),
6388 stubAddr, stubName, TypePtr::BOTTOM,
6389 state_start, subkeyH_start, data_start, len);
6390 return true;
6391 }
6392
6393 bool LibraryCallKit::inline_base64_encodeBlock() {
6394 address stubAddr;
6395 const char *stubName;
6396 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6397 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6398 stubAddr = StubRoutines::base64_encodeBlock();
6399 stubName = "encodeBlock";
6400
6401 if (!stubAddr) return false;
6402 Node* base64obj = argument(0);
6403 Node* src = argument(1);
6404 Node* offset = argument(2);
6405 Node* len = argument(3);
6406 Node* dest = argument(4);
6407 Node* dp = argument(5);
6408 Node* isURL = argument(6);
6409
6410 src = must_be_not_null(src, true);
6411 src = access_resolve(src, ACCESS_READ);
6412 dest = must_be_not_null(dest, true);
6413 dest = access_resolve(dest, ACCESS_WRITE);
6414
6415 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6416 assert(src_start, "source array is NULL");
6417 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6418 assert(dest_start, "destination array is NULL");
6419
6420 Node* base64 = make_runtime_call(RC_LEAF,
6421 OptoRuntime::base64_encodeBlock_Type(),
6422 stubAddr, stubName, TypePtr::BOTTOM,
6423 src_start, offset, len, dest_start, dp, isURL);
6424 return true;
6425 }
6426
6427 //------------------------------inline_sha_implCompress-----------------------
6428 //
6429 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6430 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6431 //
6432 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6433 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6434 //
6435 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6436 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6437 //
6438 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6439 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6440
6441 Node* sha_obj = argument(0);
6442 Node* src = argument(1); // type oop
6443 Node* ofs = argument(2); // type int
6444
6445 const Type* src_type = src->Value(&_gvn);
6446 const TypeAryPtr* top_src = src_type->isa_aryptr();
6447 if (top_src == NULL || top_src->klass() == NULL) {
6448 // failed array check
6449 return false;
6450 }
6451 // Figure out the size and type of the elements we will be copying.
6452 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6453 if (src_elem != T_BYTE) {
6454 return false;
6455 }
6456 // 'src_start' points to src array + offset
6457 src = must_be_not_null(src, true);
6458 src = access_resolve(src, ACCESS_READ);
6459 Node* src_start = array_element_address(src, ofs, src_elem);
6460 Node* state = NULL;
6461 address stubAddr;
6462 const char *stubName;
6463
6464 switch(id) {
6465 case vmIntrinsics::_sha_implCompress:
6466 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6467 state = get_state_from_sha_object(sha_obj);
6468 stubAddr = StubRoutines::sha1_implCompress();
6469 stubName = "sha1_implCompress";
6470 break;
6471 case vmIntrinsics::_sha2_implCompress:
6472 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6473 state = get_state_from_sha_object(sha_obj);
6474 stubAddr = StubRoutines::sha256_implCompress();
6475 stubName = "sha256_implCompress";
6476 break;
6477 case vmIntrinsics::_sha5_implCompress:
6478 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6479 state = get_state_from_sha5_object(sha_obj);
6480 stubAddr = StubRoutines::sha512_implCompress();
6481 stubName = "sha512_implCompress";
6482 break;
6483 default:
6484 fatal_unexpected_iid(id);
6485 return false;
6486 }
6487 if (state == NULL) return false;
6488
6489 assert(stubAddr != NULL, "Stub is generated");
6490 if (stubAddr == NULL) return false;
6491
6492 // Call the stub.
6493 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6494 stubAddr, stubName, TypePtr::BOTTOM,
6495 src_start, state);
6496
6497 return true;
6498 }
6499
6500 //------------------------------inline_digestBase_implCompressMB-----------------------
6501 //
6502 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6503 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6504 //
6505 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6506 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6507 "need SHA1/SHA256/SHA512 instruction support");
6508 assert((uint)predicate < 3, "sanity");
6509 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6510
6511 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6512 Node* src = argument(1); // byte[] array
6513 Node* ofs = argument(2); // type int
6514 Node* limit = argument(3); // type int
6515
6516 const Type* src_type = src->Value(&_gvn);
6517 const TypeAryPtr* top_src = src_type->isa_aryptr();
6518 if (top_src == NULL || top_src->klass() == NULL) {
6519 // failed array check
6520 return false;
6521 }
6522 // Figure out the size and type of the elements we will be copying.
6523 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6524 if (src_elem != T_BYTE) {
6525 return false;
6526 }
6527 // 'src_start' points to src array + offset
6528 src = must_be_not_null(src, false);
6529 src = access_resolve(src, ACCESS_READ);
6530 Node* src_start = array_element_address(src, ofs, src_elem);
6531
6532 const char* klass_SHA_name = NULL;
6533 const char* stub_name = NULL;
6534 address stub_addr = NULL;
6535 bool long_state = false;
6536
6537 switch (predicate) {
6538 case 0:
6539 if (UseSHA1Intrinsics) {
6540 klass_SHA_name = "sun/security/provider/SHA";
6541 stub_name = "sha1_implCompressMB";
6542 stub_addr = StubRoutines::sha1_implCompressMB();
6543 }
6544 break;
6545 case 1:
6546 if (UseSHA256Intrinsics) {
6547 klass_SHA_name = "sun/security/provider/SHA2";
6548 stub_name = "sha256_implCompressMB";
6549 stub_addr = StubRoutines::sha256_implCompressMB();
6550 }
6551 break;
6552 case 2:
6553 if (UseSHA512Intrinsics) {
6554 klass_SHA_name = "sun/security/provider/SHA5";
6555 stub_name = "sha512_implCompressMB";
6556 stub_addr = StubRoutines::sha512_implCompressMB();
6557 long_state = true;
6558 }
6559 break;
6560 default:
6561 fatal("unknown SHA intrinsic predicate: %d", predicate);
6562 }
6563 if (klass_SHA_name != NULL) {
6564 assert(stub_addr != NULL, "Stub is generated");
6565 if (stub_addr == NULL) return false;
6566
6567 // get DigestBase klass to lookup for SHA klass
6568 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6569 assert(tinst != NULL, "digestBase_obj is not instance???");
6570 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6571
6572 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6573 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6574 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6575 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6576 }
6577 return false;
6578 }
6579 //------------------------------inline_sha_implCompressMB-----------------------
6580 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6581 bool long_state, address stubAddr, const char *stubName,
6582 Node* src_start, Node* ofs, Node* limit) {
6583 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6584 const TypeOopPtr* xtype = aklass->as_instance_type();
6585 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6586 sha_obj = _gvn.transform(sha_obj);
6587
6588 Node* state;
6589 if (long_state) {
6590 state = get_state_from_sha5_object(sha_obj);
6591 } else {
6592 state = get_state_from_sha_object(sha_obj);
6593 }
6594 if (state == NULL) return false;
6595
6596 // Call the stub.
6597 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6598 OptoRuntime::digestBase_implCompressMB_Type(),
6599 stubAddr, stubName, TypePtr::BOTTOM,
6600 src_start, state, ofs, limit);
6601 // return ofs (int)
6602 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6603 set_result(result);
6604
6605 return true;
6606 }
6607
6608 //------------------------------get_state_from_sha_object-----------------------
6609 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6610 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6611 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6612 if (sha_state == NULL) return (Node *) NULL;
6613
6614 // now have the array, need to get the start address of the state array
6615 sha_state = access_resolve(sha_state, ACCESS_WRITE);
6616 Node* state = array_element_address(sha_state, intcon(0), T_INT);
6617 return state;
6618 }
6619
6620 //------------------------------get_state_from_sha5_object-----------------------
6621 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6622 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6623 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6624 if (sha_state == NULL) return (Node *) NULL;
6625
6626 // now have the array, need to get the start address of the state array
6627 sha_state = access_resolve(sha_state, ACCESS_WRITE);
6628 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6629 return state;
6630 }
6631
6632 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6633 // Return node representing slow path of predicate check.
6634 // the pseudo code we want to emulate with this predicate is:
6635 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6636 //
6637 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6638 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6639 "need SHA1/SHA256/SHA512 instruction support");
6640 assert((uint)predicate < 3, "sanity");
6641
6642 // The receiver was checked for NULL already.
6643 Node* digestBaseObj = argument(0);
6644
6645 // get DigestBase klass for instanceOf check
6646 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6647 assert(tinst != NULL, "digestBaseObj is null");
6648 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6649
6650 const char* klass_SHA_name = NULL;
6651 switch (predicate) {
6652 case 0:
6653 if (UseSHA1Intrinsics) {
6654 // we want to do an instanceof comparison against the SHA class
6655 klass_SHA_name = "sun/security/provider/SHA";
6656 }
6657 break;
6658 case 1:
6659 if (UseSHA256Intrinsics) {
6660 // we want to do an instanceof comparison against the SHA2 class
6661 klass_SHA_name = "sun/security/provider/SHA2";
6662 }
6663 break;
6664 case 2:
6665 if (UseSHA512Intrinsics) {
6666 // we want to do an instanceof comparison against the SHA5 class
6667 klass_SHA_name = "sun/security/provider/SHA5";
6668 }
6669 break;
6670 default:
6671 fatal("unknown SHA intrinsic predicate: %d", predicate);
6672 }
6673
6674 ciKlass* klass_SHA = NULL;
6675 if (klass_SHA_name != NULL) {
6676 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6677 }
6678 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6679 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6680 Node* ctrl = control();
6681 set_control(top()); // no intrinsic path
6682 return ctrl;
6683 }
6684 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6685
6686 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6687 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6688 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6689 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6690
6691 return instof_false; // even if it is NULL
6692 }
6693
6694 //-------------inline_fma-----------------------------------
6695 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
6696 Node *a = NULL;
6697 Node *b = NULL;
6698 Node *c = NULL;
6699 Node* result = NULL;
6700 switch (id) {
6701 case vmIntrinsics::_fmaD:
6702 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
6703 // no receiver since it is static method
6704 a = round_double_node(argument(0));
6705 b = round_double_node(argument(2));
6706 c = round_double_node(argument(4));
6707 result = _gvn.transform(new FmaDNode(control(), a, b, c));
6708 break;
6709 case vmIntrinsics::_fmaF:
6710 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
6711 a = argument(0);
6712 b = argument(1);
6713 c = argument(2);
6714 result = _gvn.transform(new FmaFNode(control(), a, b, c));
6715 break;
6716 default:
6717 fatal_unexpected_iid(id); break;
6718 }
6719 set_result(result);
6720 return true;
6721 }
6722
6723 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
6724 // argument(0) is receiver
6725 Node* codePoint = argument(1);
6726 Node* n = NULL;
6727
6728 switch (id) {
6729 case vmIntrinsics::_isDigit :
6730 n = new DigitNode(control(), codePoint);
6731 break;
6732 case vmIntrinsics::_isLowerCase :
6733 n = new LowerCaseNode(control(), codePoint);
6734 break;
6735 case vmIntrinsics::_isUpperCase :
6736 n = new UpperCaseNode(control(), codePoint);
6737 break;
6738 case vmIntrinsics::_isWhitespace :
6739 n = new WhitespaceNode(control(), codePoint);
6740 break;
6741 default:
6742 fatal_unexpected_iid(id);
6743 }
6744
6745 set_result(_gvn.transform(n));
6746 return true;
6747 }
6748
6749 //------------------------------inline_fp_min_max------------------------------
6750 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
6751 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
6752
6753 // The intrinsic should be used only when the API branches aren't predictable,
6754 // the last one performing the most important comparison. The following heuristic
6755 // uses the branch statistics to eventually bail out if necessary.
6756
6757 ciMethodData *md = callee()->method_data();
6758
6759 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
6760 ciCallProfile cp = caller()->call_profile_at_bci(bci());
6761
6762 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
6763 // Bail out if the call-site didn't contribute enough to the statistics.
6764 return false;
6765 }
6766
6767 uint taken = 0, not_taken = 0;
6768
6769 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
6770 if (p->is_BranchData()) {
6771 taken = ((ciBranchData*)p)->taken();
6772 not_taken = ((ciBranchData*)p)->not_taken();
6773 }
6774 }
6775
6776 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
6777 balance = balance < 0 ? -balance : balance;
6778 if ( balance > 0.2 ) {
6779 // Bail out if the most important branch is predictable enough.
6780 return false;
6781 }
6782 }
6783 */
6784
6785 Node *a = NULL;
6786 Node *b = NULL;
6787 Node *n = NULL;
6788 switch (id) {
6789 case vmIntrinsics::_maxF:
6790 case vmIntrinsics::_minF:
6791 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
6792 a = argument(0);
6793 b = argument(1);
6794 break;
6795 case vmIntrinsics::_maxD:
6796 case vmIntrinsics::_minD:
6797 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
6798 a = round_double_node(argument(0));
6799 b = round_double_node(argument(2));
6800 break;
6801 default:
6802 fatal_unexpected_iid(id);
6803 break;
6804 }
6805 switch (id) {
6806 case vmIntrinsics::_maxF: n = new MaxFNode(a, b); break;
6807 case vmIntrinsics::_minF: n = new MinFNode(a, b); break;
6808 case vmIntrinsics::_maxD: n = new MaxDNode(a, b); break;
6809 case vmIntrinsics::_minD: n = new MinDNode(a, b); break;
6810 default: fatal_unexpected_iid(id); break;
6811 }
6812 set_result(_gvn.transform(n));
6813 return true;
6814 }
6815
6816 bool LibraryCallKit::inline_profileBoolean() {
6817 Node* counts = argument(1);
6818 const TypeAryPtr* ary = NULL;
6819 ciArray* aobj = NULL;
6820 if (counts->is_Con()
6821 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6822 && (aobj = ary->const_oop()->as_array()) != NULL
6823 && (aobj->length() == 2)) {
6824 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6825 jint false_cnt = aobj->element_value(0).as_int();
6826 jint true_cnt = aobj->element_value(1).as_int();
6827
6828 if (C->log() != NULL) {
6829 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6830 false_cnt, true_cnt);
6831 }
6832
6833 if (false_cnt + true_cnt == 0) {
6834 // According to profile, never executed.
6835 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6836 Deoptimization::Action_reinterpret);
6837 return true;
6838 }
6839
6840 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6841 // is a number of each value occurrences.
6842 Node* result = argument(0);
6843 if (false_cnt == 0 || true_cnt == 0) {
6844 // According to profile, one value has been never seen.
6845 int expected_val = (false_cnt == 0) ? 1 : 0;
6846
6847 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6848 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6849
6850 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6851 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6852 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6853
6854 { // Slow path: uncommon trap for never seen value and then reexecute
6855 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6856 // the value has been seen at least once.
6857 PreserveJVMState pjvms(this);
6858 PreserveReexecuteState preexecs(this);
6859 jvms()->set_should_reexecute(true);
6860
6861 set_control(slow_path);
6862 set_i_o(i_o());
6863
6864 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6865 Deoptimization::Action_reinterpret);
6866 }
6867 // The guard for never seen value enables sharpening of the result and
6868 // returning a constant. It allows to eliminate branches on the same value
6869 // later on.
6870 set_control(fast_path);
6871 result = intcon(expected_val);
6872 }
6873 // Stop profiling.
6874 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6875 // By replacing method body with profile data (represented as ProfileBooleanNode
6876 // on IR level) we effectively disable profiling.
6877 // It enables full speed execution once optimized code is generated.
6878 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6879 C->record_for_igvn(profile);
6880 set_result(profile);
6881 return true;
6882 } else {
6883 // Continue profiling.
6884 // Profile data isn't available at the moment. So, execute method's bytecode version.
6885 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6886 // is compiled and counters aren't available since corresponding MethodHandle
6887 // isn't a compile-time constant.
6888 return false;
6889 }
6890 }
6891
6892 bool LibraryCallKit::inline_isCompileConstant() {
6893 Node* n = argument(0);
6894 set_result(n->is_Con() ? intcon(1) : intcon(0));
6895 return true;
6896 }