1 /*
2 * Copyright (c) 1999, 2020, 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/klass.inline.hpp"
36 #include "oops/objArrayKlass.hpp"
37 #include "opto/addnode.hpp"
38 #include "opto/arraycopynode.hpp"
39 #include "opto/c2compiler.hpp"
40 #include "opto/callGenerator.hpp"
41 #include "opto/castnode.hpp"
42 #include "opto/cfgnode.hpp"
43 #include "opto/convertnode.hpp"
44 #include "opto/countbitsnode.hpp"
45 #include "opto/intrinsicnode.hpp"
46 #include "opto/idealKit.hpp"
47 #include "opto/mathexactnode.hpp"
48 #include "opto/movenode.hpp"
49 #include "opto/mulnode.hpp"
50 #include "opto/narrowptrnode.hpp"
51 #include "opto/opaquenode.hpp"
52 #include "opto/parse.hpp"
53 #include "opto/runtime.hpp"
54 #include "opto/rootnode.hpp"
55 #include "opto/subnode.hpp"
56 #include "prims/nativeLookup.hpp"
57 #include "prims/unsafe.hpp"
58 #include "runtime/objectMonitor.hpp"
59 #include "runtime/sharedRuntime.hpp"
60 #include "utilities/macros.hpp"
61 #include "utilities/powerOfTwo.hpp"
62
63 class LibraryIntrinsic : public InlineCallGenerator {
64 // Extend the set of intrinsics known to the runtime:
65 public:
66 private:
67 bool _is_virtual;
68 bool _does_virtual_dispatch;
69 int8_t _predicates_count; // Intrinsic is predicated by several conditions
70 int8_t _last_predicate; // Last generated predicate
71 vmIntrinsics::ID _intrinsic_id;
72
73 public:
74 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
75 : InlineCallGenerator(m),
76 _is_virtual(is_virtual),
77 _does_virtual_dispatch(does_virtual_dispatch),
78 _predicates_count((int8_t)predicates_count),
79 _last_predicate((int8_t)-1),
80 _intrinsic_id(id)
81 {
82 }
83 virtual bool is_intrinsic() const { return true; }
84 virtual bool is_virtual() const { return _is_virtual; }
85 virtual bool is_predicated() const { return _predicates_count > 0; }
86 virtual int predicates_count() const { return _predicates_count; }
87 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; }
88 virtual JVMState* generate(JVMState* jvms);
89 virtual Node* generate_predicate(JVMState* jvms, int predicate);
90 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
91 };
92
93
94 // Local helper class for LibraryIntrinsic:
95 class LibraryCallKit : public GraphKit {
96 private:
97 LibraryIntrinsic* _intrinsic; // the library intrinsic being called
98 Node* _result; // the result node, if any
99 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted
100
101 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type);
102
103 public:
104 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
105 : GraphKit(jvms),
106 _intrinsic(intrinsic),
107 _result(NULL)
108 {
109 // Check if this is a root compile. In that case we don't have a caller.
110 if (!jvms->has_method()) {
111 _reexecute_sp = sp();
112 } else {
113 // Find out how many arguments the interpreter needs when deoptimizing
114 // and save the stack pointer value so it can used by uncommon_trap.
115 // We find the argument count by looking at the declared signature.
116 bool ignored_will_link;
117 ciSignature* declared_signature = NULL;
118 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
119 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
120 _reexecute_sp = sp() + nargs; // "push" arguments back on stack
121 }
122 }
123
124 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
125
126 ciMethod* caller() const { return jvms()->method(); }
127 int bci() const { return jvms()->bci(); }
128 LibraryIntrinsic* intrinsic() const { return _intrinsic; }
129 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); }
130 ciMethod* callee() const { return _intrinsic->method(); }
131
132 bool try_to_inline(int predicate);
133 Node* try_to_predicate(int predicate);
134
135 void push_result() {
136 // Push the result onto the stack.
137 if (!stopped() && result() != NULL) {
138 BasicType bt = result()->bottom_type()->basic_type();
139 push_node(bt, result());
140 }
141 }
142
143 private:
144 void fatal_unexpected_iid(vmIntrinsics::ID iid) {
145 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid));
146 }
147
148 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
149 void set_result(RegionNode* region, PhiNode* value);
150 Node* result() { return _result; }
151
152 virtual int reexecute_sp() { return _reexecute_sp; }
153
154 // Helper functions to inline natives
155 Node* generate_guard(Node* test, RegionNode* region, float true_prob);
156 Node* generate_slow_guard(Node* test, RegionNode* region);
157 Node* generate_fair_guard(Node* test, RegionNode* region);
158 Node* generate_negative_guard(Node* index, RegionNode* region,
159 // resulting CastII of index:
160 Node* *pos_index = NULL);
161 Node* generate_limit_guard(Node* offset, Node* subseq_length,
162 Node* array_length,
163 RegionNode* region);
164 void generate_string_range_check(Node* array, Node* offset,
165 Node* length, bool char_count);
166 Node* generate_current_thread(Node* &tls_output);
167 Node* load_mirror_from_klass(Node* klass);
168 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
169 RegionNode* region, int null_path,
170 int offset);
171 Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
172 RegionNode* region, int null_path) {
173 int offset = java_lang_Class::klass_offset();
174 return load_klass_from_mirror_common(mirror, never_see_null,
175 region, null_path,
176 offset);
177 }
178 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
179 RegionNode* region, int null_path) {
180 int offset = java_lang_Class::array_klass_offset();
181 return load_klass_from_mirror_common(mirror, never_see_null,
182 region, null_path,
183 offset);
184 }
185 Node* generate_access_flags_guard(Node* kls,
186 int modifier_mask, int modifier_bits,
187 RegionNode* region);
188 Node* generate_interface_guard(Node* kls, RegionNode* region);
189 Node* generate_hidden_class_guard(Node* kls, RegionNode* region);
190 Node* generate_array_guard(Node* kls, RegionNode* region) {
191 return generate_array_guard_common(kls, region, false, false);
192 }
193 Node* generate_non_array_guard(Node* kls, RegionNode* region) {
194 return generate_array_guard_common(kls, region, false, true);
195 }
196 Node* generate_objArray_guard(Node* kls, RegionNode* region) {
197 return generate_array_guard_common(kls, region, true, false);
198 }
199 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
200 return generate_array_guard_common(kls, region, true, true);
201 }
202 Node* generate_array_guard_common(Node* kls, RegionNode* region,
203 bool obj_array, bool not_array);
204 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
205 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
206 bool is_virtual = false, bool is_static = false);
207 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
208 return generate_method_call(method_id, false, true);
209 }
210 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
211 return generate_method_call(method_id, true, false);
212 }
213 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
214 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
215
216 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
217 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
218 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
219 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
220 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
221 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
222 bool inline_string_indexOfChar();
223 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
224 bool inline_string_toBytesU();
225 bool inline_string_getCharsU();
226 bool inline_string_copy(bool compress);
227 bool inline_string_char_access(bool is_store);
228 Node* round_double_node(Node* n);
229 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
230 bool inline_math_native(vmIntrinsics::ID id);
231 bool inline_math(vmIntrinsics::ID id);
232 bool inline_double_math(vmIntrinsics::ID id);
233 template <typename OverflowOp>
234 bool inline_math_overflow(Node* arg1, Node* arg2);
235 void inline_math_mathExact(Node* math, Node* test);
236 bool inline_math_addExactI(bool is_increment);
237 bool inline_math_addExactL(bool is_increment);
238 bool inline_math_multiplyExactI();
239 bool inline_math_multiplyExactL();
240 bool inline_math_multiplyHigh();
241 bool inline_math_negateExactI();
242 bool inline_math_negateExactL();
243 bool inline_math_subtractExactI(bool is_decrement);
244 bool inline_math_subtractExactL(bool is_decrement);
245 bool inline_min_max(vmIntrinsics::ID id);
246 bool inline_notify(vmIntrinsics::ID id);
247 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
248 // This returns Type::AnyPtr, RawPtr, or OopPtr.
249 int classify_unsafe_addr(Node* &base, Node* &offset, BasicType type);
250 Node* make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type = T_ILLEGAL, bool can_cast = false);
251
252 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
253 DecoratorSet mo_decorator_for_access_kind(AccessKind kind);
254 bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
255 static bool klass_needs_init_guard(Node* kls);
256 bool inline_unsafe_allocate();
257 bool inline_unsafe_newArray(bool uninitialized);
258 bool inline_unsafe_writeback0();
259 bool inline_unsafe_writebackSync0(bool is_pre);
260 bool inline_unsafe_copyMemory();
261 bool inline_native_currentThread();
262
263 bool inline_native_time_funcs(address method, const char* funcName);
264 #ifdef JFR_HAVE_INTRINSICS
265 bool inline_native_classID();
266 bool inline_native_getEventWriter();
267 #endif
268 bool inline_native_Class_query(vmIntrinsics::ID id);
269 bool inline_native_subtype_check();
270 bool inline_native_getLength();
271 bool inline_array_copyOf(bool is_copyOfRange);
272 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
273 bool inline_preconditions_checkIndex();
274 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array);
275 bool inline_native_clone(bool is_virtual);
276 bool inline_native_Reflection_getCallerClass();
277 // Helper function for inlining native object hash method
278 bool inline_native_hashcode(bool is_virtual, bool is_static);
279 bool inline_native_getClass();
280
281 // Helper functions for inlining arraycopy
282 bool inline_arraycopy();
283 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
284 RegionNode* slow_region);
285 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
286 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp,
287 uint new_idx);
288
289 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
290 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
291 bool inline_unsafe_fence(vmIntrinsics::ID id);
292 bool inline_onspinwait();
293 bool inline_fp_conversions(vmIntrinsics::ID id);
294 bool inline_number_methods(vmIntrinsics::ID id);
295 bool inline_reference_get();
296 bool inline_Class_cast();
297 bool inline_aescrypt_Block(vmIntrinsics::ID id);
298 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
299 bool inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id);
300 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
301 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
302 Node* inline_electronicCodeBook_AESCrypt_predicate(bool decrypting);
303 Node* inline_counterMode_AESCrypt_predicate();
304 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
305 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
306 bool inline_ghash_processBlocks();
307 bool inline_base64_encodeBlock();
308 bool inline_digestBase_implCompress(vmIntrinsics::ID id);
309 bool inline_digestBase_implCompressMB(int predicate);
310 bool inline_digestBase_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass,
311 const char* state_type, address stubAddr, const char *stubName,
312 Node* src_start, Node* ofs, Node* limit);
313 Node* get_state_from_digest_object(Node *digestBase_object, const char* state_type);
314 Node* get_digest_length_from_digest_object(Node *digestBase_object);
315 Node* inline_digestBase_implCompressMB_predicate(int predicate);
316 bool inline_encodeISOArray();
317 bool inline_updateCRC32();
318 bool inline_updateBytesCRC32();
319 bool inline_updateByteBufferCRC32();
320 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
321 bool inline_updateBytesCRC32C();
322 bool inline_updateDirectByteBufferCRC32C();
323 bool inline_updateBytesAdler32();
324 bool inline_updateByteBufferAdler32();
325 bool inline_multiplyToLen();
326 bool inline_hasNegatives();
327 bool inline_squareToLen();
328 bool inline_mulAdd();
329 bool inline_montgomeryMultiply();
330 bool inline_montgomerySquare();
331 bool inline_bigIntegerShift(bool isRightShift);
332 bool inline_vectorizedMismatch();
333 bool inline_fma(vmIntrinsics::ID id);
334 bool inline_character_compare(vmIntrinsics::ID id);
335 bool inline_fp_min_max(vmIntrinsics::ID id);
336
337 bool inline_profileBoolean();
338 bool inline_isCompileConstant();
339 void clear_upper_avx() {
340 #ifdef X86
341 if (UseAVX >= 2) {
342 C->set_clear_upper_avx(true);
343 }
344 #endif
345 }
346 };
347
348 //---------------------------make_vm_intrinsic----------------------------
349 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
350 vmIntrinsics::ID id = m->intrinsic_id();
351 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
352
353 if (!m->is_loaded()) {
354 // Do not attempt to inline unloaded methods.
355 return NULL;
356 }
357
358 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
359 bool is_available = false;
360
361 {
362 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
363 // the compiler must transition to '_thread_in_vm' state because both
364 // methods access VM-internal data.
365 VM_ENTRY_MARK;
366 methodHandle mh(THREAD, m->get_Method());
367 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
368 !C->directive()->is_intrinsic_disabled(mh) &&
369 !vmIntrinsics::is_disabled_by_flags(mh);
370
371 }
372
373 if (is_available) {
374 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
375 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
376 return new LibraryIntrinsic(m, is_virtual,
377 vmIntrinsics::predicates_needed(id),
378 vmIntrinsics::does_virtual_dispatch(id),
379 (vmIntrinsics::ID) id);
380 } else {
381 return NULL;
382 }
383 }
384
385 //----------------------register_library_intrinsics-----------------------
386 // Initialize this file's data structures, for each Compile instance.
387 void Compile::register_library_intrinsics() {
388 // Nothing to do here.
389 }
390
391 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
392 LibraryCallKit kit(jvms, this);
393 Compile* C = kit.C;
394 int nodes = C->unique();
395 #ifndef PRODUCT
396 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
397 char buf[1000];
398 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
399 tty->print_cr("Intrinsic %s", str);
400 }
401 #endif
402 ciMethod* callee = kit.callee();
403 const int bci = kit.bci();
404
405 // Try to inline the intrinsic.
406 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
407 kit.try_to_inline(_last_predicate)) {
408 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
409 : "(intrinsic)";
410 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
411 if (C->print_intrinsics() || C->print_inlining()) {
412 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
413 }
414 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
415 if (C->log()) {
416 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
417 vmIntrinsics::name_at(intrinsic_id()),
418 (is_virtual() ? " virtual='1'" : ""),
419 C->unique() - nodes);
420 }
421 // Push the result from the inlined method onto the stack.
422 kit.push_result();
423 C->print_inlining_update(this);
424 return kit.transfer_exceptions_into_jvms();
425 }
426
427 // The intrinsic bailed out
428 if (jvms->has_method()) {
429 // Not a root compile.
430 const char* msg;
431 if (callee->intrinsic_candidate()) {
432 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
433 } else {
434 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
435 : "failed to inline (intrinsic), method not annotated";
436 }
437 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
438 if (C->print_intrinsics() || C->print_inlining()) {
439 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
440 }
441 } else {
442 // Root compile
443 ResourceMark rm;
444 stringStream msg_stream;
445 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
446 vmIntrinsics::name_at(intrinsic_id()),
447 is_virtual() ? " (virtual)" : "", bci);
448 const char *msg = msg_stream.as_string();
449 log_debug(jit, inlining)("%s", msg);
450 if (C->print_intrinsics() || C->print_inlining()) {
451 tty->print("%s", msg);
452 }
453 }
454 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
455 C->print_inlining_update(this);
456 return NULL;
457 }
458
459 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
460 LibraryCallKit kit(jvms, this);
461 Compile* C = kit.C;
462 int nodes = C->unique();
463 _last_predicate = predicate;
464 #ifndef PRODUCT
465 assert(is_predicated() && predicate < predicates_count(), "sanity");
466 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
467 char buf[1000];
468 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
469 tty->print_cr("Predicate for intrinsic %s", str);
470 }
471 #endif
472 ciMethod* callee = kit.callee();
473 const int bci = kit.bci();
474
475 Node* slow_ctl = kit.try_to_predicate(predicate);
476 if (!kit.failing()) {
477 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
478 : "(intrinsic, predicate)";
479 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
480 if (C->print_intrinsics() || C->print_inlining()) {
481 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
482 }
483 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
484 if (C->log()) {
485 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
486 vmIntrinsics::name_at(intrinsic_id()),
487 (is_virtual() ? " virtual='1'" : ""),
488 C->unique() - nodes);
489 }
490 return slow_ctl; // Could be NULL if the check folds.
491 }
492
493 // The intrinsic bailed out
494 if (jvms->has_method()) {
495 // Not a root compile.
496 const char* msg = "failed to generate predicate for intrinsic";
497 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
498 if (C->print_intrinsics() || C->print_inlining()) {
499 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
500 }
501 } else {
502 // Root compile
503 ResourceMark rm;
504 stringStream msg_stream;
505 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
506 vmIntrinsics::name_at(intrinsic_id()),
507 is_virtual() ? " (virtual)" : "", bci);
508 const char *msg = msg_stream.as_string();
509 log_debug(jit, inlining)("%s", msg);
510 if (C->print_intrinsics() || C->print_inlining()) {
511 C->print_inlining_stream()->print("%s", msg);
512 }
513 }
514 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
515 return NULL;
516 }
517
518 bool LibraryCallKit::try_to_inline(int predicate) {
519 // Handle symbolic names for otherwise undistinguished boolean switches:
520 const bool is_store = true;
521 const bool is_compress = true;
522 const bool is_static = true;
523 const bool is_volatile = true;
524
525 if (!jvms()->has_method()) {
526 // Root JVMState has a null method.
527 assert(map()->memory()->Opcode() == Op_Parm, "");
528 // Insert the memory aliasing node
529 set_all_memory(reset_memory());
530 }
531 assert(merged_memory(), "");
532
533
534 switch (intrinsic_id()) {
535 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
536 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
537 case vmIntrinsics::_getClass: return inline_native_getClass();
538
539 case vmIntrinsics::_ceil:
540 case vmIntrinsics::_floor:
541 case vmIntrinsics::_rint:
542 case vmIntrinsics::_dsin:
543 case vmIntrinsics::_dcos:
544 case vmIntrinsics::_dtan:
545 case vmIntrinsics::_dabs:
546 case vmIntrinsics::_fabs:
547 case vmIntrinsics::_iabs:
548 case vmIntrinsics::_labs:
549 case vmIntrinsics::_datan2:
550 case vmIntrinsics::_dsqrt:
551 case vmIntrinsics::_dexp:
552 case vmIntrinsics::_dlog:
553 case vmIntrinsics::_dlog10:
554 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id());
555
556 case vmIntrinsics::_min:
557 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
558
559 case vmIntrinsics::_notify:
560 case vmIntrinsics::_notifyAll:
561 return inline_notify(intrinsic_id());
562
563 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
564 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
565 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
566 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
567 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
568 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
569 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
570 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
571 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
572 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
573 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
574 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
575 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
576
577 case vmIntrinsics::_arraycopy: return inline_arraycopy();
578
579 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
580 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
581 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
582 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
583
584 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
585 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
586 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
587 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
588 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
589 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
590 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
591
592 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
593 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
594
595 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
596 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
597 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
598 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
599
600 case vmIntrinsics::_compressStringC:
601 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
602 case vmIntrinsics::_inflateStringC:
603 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
604
605 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
606 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
607 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
608 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
609 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
610 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
611 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
612 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
613 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
614
615 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
616 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
617 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
618 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
619 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
620 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
621 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
622 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
623 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
624
625 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
626 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
627 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
628 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
629 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
630 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
631 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
632 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
633 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
634
635 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
636 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
637 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
638 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
639 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
640 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
641 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
642 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
643 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
644
645 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
646 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
647 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
648 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
649
650 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
651 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
652 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
653 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
654
655 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
656 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
657 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
658 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
659 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
660 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
661 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
662 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
663 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
664
665 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
666 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
667 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
668 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
669 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
670 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
671 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
672 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
673 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
674
675 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
676 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
677 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
678 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
679 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
680 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
681 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
682 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
683 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
684
685 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
686 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
687 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
688 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
689 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
690 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
691 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
692 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
693 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
694
695 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
696 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
697 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
698 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
699 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
700
701 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
702 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
703 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
704 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
705 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
706 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
707 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
708 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
709 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
710 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
711 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
712 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
713 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
714 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
715 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
716 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
717 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
718 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
719 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
720 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
721
722 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
723 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
724 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
725 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
726 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
727 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
728 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
729 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
730 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
731 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
732 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
733 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
734 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
735 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
736 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
737
738 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
739 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
740 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
741 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
742
743 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
744 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
745 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
746 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
747 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
748
749 case vmIntrinsics::_loadFence:
750 case vmIntrinsics::_storeFence:
751 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
752
753 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
754
755 case vmIntrinsics::_currentThread: return inline_native_currentThread();
756
757 #ifdef JFR_HAVE_INTRINSICS
758 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime");
759 case vmIntrinsics::_getClassId: return inline_native_classID();
760 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
761 #endif
762 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
763 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
764 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
765 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
766 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
767 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
768 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
769 case vmIntrinsics::_getLength: return inline_native_getLength();
770 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
771 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
772 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
773 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
774 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
775 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
776
777 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
778 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
779
780 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
781
782 case vmIntrinsics::_isInstance:
783 case vmIntrinsics::_getModifiers:
784 case vmIntrinsics::_isInterface:
785 case vmIntrinsics::_isArray:
786 case vmIntrinsics::_isPrimitive:
787 case vmIntrinsics::_isHidden:
788 case vmIntrinsics::_getSuperclass:
789 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
790
791 case vmIntrinsics::_floatToRawIntBits:
792 case vmIntrinsics::_floatToIntBits:
793 case vmIntrinsics::_intBitsToFloat:
794 case vmIntrinsics::_doubleToRawLongBits:
795 case vmIntrinsics::_doubleToLongBits:
796 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
797
798 case vmIntrinsics::_numberOfLeadingZeros_i:
799 case vmIntrinsics::_numberOfLeadingZeros_l:
800 case vmIntrinsics::_numberOfTrailingZeros_i:
801 case vmIntrinsics::_numberOfTrailingZeros_l:
802 case vmIntrinsics::_bitCount_i:
803 case vmIntrinsics::_bitCount_l:
804 case vmIntrinsics::_reverseBytes_i:
805 case vmIntrinsics::_reverseBytes_l:
806 case vmIntrinsics::_reverseBytes_s:
807 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
808
809 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
810
811 case vmIntrinsics::_Reference_get: return inline_reference_get();
812
813 case vmIntrinsics::_Class_cast: return inline_Class_cast();
814
815 case vmIntrinsics::_aescrypt_encryptBlock:
816 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
817
818 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
819 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
820 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
821
822 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
823 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
824 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
825
826 case vmIntrinsics::_counterMode_AESCrypt:
827 return inline_counterMode_AESCrypt(intrinsic_id());
828
829 case vmIntrinsics::_md5_implCompress:
830 case vmIntrinsics::_sha_implCompress:
831 case vmIntrinsics::_sha2_implCompress:
832 case vmIntrinsics::_sha5_implCompress:
833 case vmIntrinsics::_sha3_implCompress:
834 return inline_digestBase_implCompress(intrinsic_id());
835
836 case vmIntrinsics::_digestBase_implCompressMB:
837 return inline_digestBase_implCompressMB(predicate);
838
839 case vmIntrinsics::_multiplyToLen:
840 return inline_multiplyToLen();
841
842 case vmIntrinsics::_squareToLen:
843 return inline_squareToLen();
844
845 case vmIntrinsics::_mulAdd:
846 return inline_mulAdd();
847
848 case vmIntrinsics::_montgomeryMultiply:
849 return inline_montgomeryMultiply();
850 case vmIntrinsics::_montgomerySquare:
851 return inline_montgomerySquare();
852
853 case vmIntrinsics::_bigIntegerRightShiftWorker:
854 return inline_bigIntegerShift(true);
855 case vmIntrinsics::_bigIntegerLeftShiftWorker:
856 return inline_bigIntegerShift(false);
857
858 case vmIntrinsics::_vectorizedMismatch:
859 return inline_vectorizedMismatch();
860
861 case vmIntrinsics::_ghash_processBlocks:
862 return inline_ghash_processBlocks();
863 case vmIntrinsics::_base64_encodeBlock:
864 return inline_base64_encodeBlock();
865
866 case vmIntrinsics::_encodeISOArray:
867 case vmIntrinsics::_encodeByteISOArray:
868 return inline_encodeISOArray();
869
870 case vmIntrinsics::_updateCRC32:
871 return inline_updateCRC32();
872 case vmIntrinsics::_updateBytesCRC32:
873 return inline_updateBytesCRC32();
874 case vmIntrinsics::_updateByteBufferCRC32:
875 return inline_updateByteBufferCRC32();
876
877 case vmIntrinsics::_updateBytesCRC32C:
878 return inline_updateBytesCRC32C();
879 case vmIntrinsics::_updateDirectByteBufferCRC32C:
880 return inline_updateDirectByteBufferCRC32C();
881
882 case vmIntrinsics::_updateBytesAdler32:
883 return inline_updateBytesAdler32();
884 case vmIntrinsics::_updateByteBufferAdler32:
885 return inline_updateByteBufferAdler32();
886
887 case vmIntrinsics::_profileBoolean:
888 return inline_profileBoolean();
889 case vmIntrinsics::_isCompileConstant:
890 return inline_isCompileConstant();
891
892 case vmIntrinsics::_hasNegatives:
893 return inline_hasNegatives();
894
895 case vmIntrinsics::_fmaD:
896 case vmIntrinsics::_fmaF:
897 return inline_fma(intrinsic_id());
898
899 case vmIntrinsics::_isDigit:
900 case vmIntrinsics::_isLowerCase:
901 case vmIntrinsics::_isUpperCase:
902 case vmIntrinsics::_isWhitespace:
903 return inline_character_compare(intrinsic_id());
904
905 case vmIntrinsics::_maxF:
906 case vmIntrinsics::_minF:
907 case vmIntrinsics::_maxD:
908 case vmIntrinsics::_minD:
909 return inline_fp_min_max(intrinsic_id());
910
911 default:
912 // If you get here, it may be that someone has added a new intrinsic
913 // to the list in vmSymbols.hpp without implementing it here.
914 #ifndef PRODUCT
915 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
916 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
917 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
918 }
919 #endif
920 return false;
921 }
922 }
923
924 Node* LibraryCallKit::try_to_predicate(int predicate) {
925 if (!jvms()->has_method()) {
926 // Root JVMState has a null method.
927 assert(map()->memory()->Opcode() == Op_Parm, "");
928 // Insert the memory aliasing node
929 set_all_memory(reset_memory());
930 }
931 assert(merged_memory(), "");
932
933 switch (intrinsic_id()) {
934 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
935 return inline_cipherBlockChaining_AESCrypt_predicate(false);
936 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
937 return inline_cipherBlockChaining_AESCrypt_predicate(true);
938 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
939 return inline_electronicCodeBook_AESCrypt_predicate(false);
940 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
941 return inline_electronicCodeBook_AESCrypt_predicate(true);
942 case vmIntrinsics::_counterMode_AESCrypt:
943 return inline_counterMode_AESCrypt_predicate();
944 case vmIntrinsics::_digestBase_implCompressMB:
945 return inline_digestBase_implCompressMB_predicate(predicate);
946
947 default:
948 // If you get here, it may be that someone has added a new intrinsic
949 // to the list in vmSymbols.hpp without implementing it here.
950 #ifndef PRODUCT
951 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
952 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
953 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
954 }
955 #endif
956 Node* slow_ctl = control();
957 set_control(top()); // No fast path instrinsic
958 return slow_ctl;
959 }
960 }
961
962 //------------------------------set_result-------------------------------
963 // Helper function for finishing intrinsics.
964 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
965 record_for_igvn(region);
966 set_control(_gvn.transform(region));
967 set_result( _gvn.transform(value));
968 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
969 }
970
971 //------------------------------generate_guard---------------------------
972 // Helper function for generating guarded fast-slow graph structures.
973 // The given 'test', if true, guards a slow path. If the test fails
974 // then a fast path can be taken. (We generally hope it fails.)
975 // In all cases, GraphKit::control() is updated to the fast path.
976 // The returned value represents the control for the slow path.
977 // The return value is never 'top'; it is either a valid control
978 // or NULL if it is obvious that the slow path can never be taken.
979 // Also, if region and the slow control are not NULL, the slow edge
980 // is appended to the region.
981 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
982 if (stopped()) {
983 // Already short circuited.
984 return NULL;
985 }
986
987 // Build an if node and its projections.
988 // If test is true we take the slow path, which we assume is uncommon.
989 if (_gvn.type(test) == TypeInt::ZERO) {
990 // The slow branch is never taken. No need to build this guard.
991 return NULL;
992 }
993
994 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
995
996 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
997 if (if_slow == top()) {
998 // The slow branch is never taken. No need to build this guard.
999 return NULL;
1000 }
1001
1002 if (region != NULL)
1003 region->add_req(if_slow);
1004
1005 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
1006 set_control(if_fast);
1007
1008 return if_slow;
1009 }
1010
1011 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
1012 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
1013 }
1014 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
1015 return generate_guard(test, region, PROB_FAIR);
1016 }
1017
1018 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
1019 Node* *pos_index) {
1020 if (stopped())
1021 return NULL; // already stopped
1022 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
1023 return NULL; // index is already adequately typed
1024 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
1025 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1026 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
1027 if (is_neg != NULL && pos_index != NULL) {
1028 // Emulate effect of Parse::adjust_map_after_if.
1029 Node* ccast = new CastIINode(index, TypeInt::POS);
1030 ccast->set_req(0, control());
1031 (*pos_index) = _gvn.transform(ccast);
1032 }
1033 return is_neg;
1034 }
1035
1036 // Make sure that 'position' is a valid limit index, in [0..length].
1037 // There are two equivalent plans for checking this:
1038 // A. (offset + copyLength) unsigned<= arrayLength
1039 // B. offset <= (arrayLength - copyLength)
1040 // We require that all of the values above, except for the sum and
1041 // difference, are already known to be non-negative.
1042 // Plan A is robust in the face of overflow, if offset and copyLength
1043 // are both hugely positive.
1044 //
1045 // Plan B is less direct and intuitive, but it does not overflow at
1046 // all, since the difference of two non-negatives is always
1047 // representable. Whenever Java methods must perform the equivalent
1048 // check they generally use Plan B instead of Plan A.
1049 // For the moment we use Plan A.
1050 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
1051 Node* subseq_length,
1052 Node* array_length,
1053 RegionNode* region) {
1054 if (stopped())
1055 return NULL; // already stopped
1056 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
1057 if (zero_offset && subseq_length->eqv_uncast(array_length))
1058 return NULL; // common case of whole-array copy
1059 Node* last = subseq_length;
1060 if (!zero_offset) // last += offset
1061 last = _gvn.transform(new AddINode(last, offset));
1062 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
1063 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1064 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
1065 return is_over;
1066 }
1067
1068 // Emit range checks for the given String.value byte array
1069 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
1070 if (stopped()) {
1071 return; // already stopped
1072 }
1073 RegionNode* bailout = new RegionNode(1);
1074 record_for_igvn(bailout);
1075 if (char_count) {
1076 // Convert char count to byte count
1077 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1078 }
1079
1080 // Offset and count must not be negative
1081 generate_negative_guard(offset, bailout);
1082 generate_negative_guard(count, bailout);
1083 // Offset + count must not exceed length of array
1084 generate_limit_guard(offset, count, load_array_length(array), bailout);
1085
1086 if (bailout->req() > 1) {
1087 PreserveJVMState pjvms(this);
1088 set_control(_gvn.transform(bailout));
1089 uncommon_trap(Deoptimization::Reason_intrinsic,
1090 Deoptimization::Action_maybe_recompile);
1091 }
1092 }
1093
1094 //--------------------------generate_current_thread--------------------
1095 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1096 ciKlass* thread_klass = env()->Thread_klass();
1097 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1098 Node* thread = _gvn.transform(new ThreadLocalNode());
1099 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1100 tls_output = thread;
1101 Node* thread_obj_handle = LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(), TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
1102 thread_obj_handle = _gvn.transform(thread_obj_handle);
1103 return access_load(thread_obj_handle, thread_type, T_OBJECT, IN_NATIVE | C2_IMMUTABLE_MEMORY);
1104 }
1105
1106
1107 //------------------------------make_string_method_node------------------------
1108 // Helper method for String intrinsic functions. This version is called with
1109 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1110 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1111 // containing the lengths of str1 and str2.
1112 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1113 Node* result = NULL;
1114 switch (opcode) {
1115 case Op_StrIndexOf:
1116 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1117 str1_start, cnt1, str2_start, cnt2, ae);
1118 break;
1119 case Op_StrComp:
1120 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1121 str1_start, cnt1, str2_start, cnt2, ae);
1122 break;
1123 case Op_StrEquals:
1124 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1125 // Use the constant length if there is one because optimized match rule may exist.
1126 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1127 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1128 break;
1129 default:
1130 ShouldNotReachHere();
1131 return NULL;
1132 }
1133
1134 // All these intrinsics have checks.
1135 C->set_has_split_ifs(true); // Has chance for split-if optimization
1136 clear_upper_avx();
1137
1138 return _gvn.transform(result);
1139 }
1140
1141 //------------------------------inline_string_compareTo------------------------
1142 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1143 Node* arg1 = argument(0);
1144 Node* arg2 = argument(1);
1145
1146 arg1 = must_be_not_null(arg1, true);
1147 arg2 = must_be_not_null(arg2, true);
1148
1149 // Get start addr and length of first argument
1150 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1151 Node* arg1_cnt = load_array_length(arg1);
1152
1153 // Get start addr and length of second argument
1154 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1155 Node* arg2_cnt = load_array_length(arg2);
1156
1157 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1158 set_result(result);
1159 return true;
1160 }
1161
1162 //------------------------------inline_string_equals------------------------
1163 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1164 Node* arg1 = argument(0);
1165 Node* arg2 = argument(1);
1166
1167 // paths (plus control) merge
1168 RegionNode* region = new RegionNode(3);
1169 Node* phi = new PhiNode(region, TypeInt::BOOL);
1170
1171 if (!stopped()) {
1172
1173 arg1 = must_be_not_null(arg1, true);
1174 arg2 = must_be_not_null(arg2, true);
1175
1176 // Get start addr and length of first argument
1177 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1178 Node* arg1_cnt = load_array_length(arg1);
1179
1180 // Get start addr and length of second argument
1181 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1182 Node* arg2_cnt = load_array_length(arg2);
1183
1184 // Check for arg1_cnt != arg2_cnt
1185 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1186 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1187 Node* if_ne = generate_slow_guard(bol, NULL);
1188 if (if_ne != NULL) {
1189 phi->init_req(2, intcon(0));
1190 region->init_req(2, if_ne);
1191 }
1192
1193 // Check for count == 0 is done by assembler code for StrEquals.
1194
1195 if (!stopped()) {
1196 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1197 phi->init_req(1, equals);
1198 region->init_req(1, control());
1199 }
1200 }
1201
1202 // post merge
1203 set_control(_gvn.transform(region));
1204 record_for_igvn(region);
1205
1206 set_result(_gvn.transform(phi));
1207 return true;
1208 }
1209
1210 //------------------------------inline_array_equals----------------------------
1211 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1212 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1213 Node* arg1 = argument(0);
1214 Node* arg2 = argument(1);
1215
1216 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1217 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1218 clear_upper_avx();
1219
1220 return true;
1221 }
1222
1223 //------------------------------inline_hasNegatives------------------------------
1224 bool LibraryCallKit::inline_hasNegatives() {
1225 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1226 return false;
1227 }
1228
1229 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1230 // no receiver since it is static method
1231 Node* ba = argument(0);
1232 Node* offset = argument(1);
1233 Node* len = argument(2);
1234
1235 ba = must_be_not_null(ba, true);
1236
1237 // Range checks
1238 generate_string_range_check(ba, offset, len, false);
1239 if (stopped()) {
1240 return true;
1241 }
1242 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1243 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1244 set_result(_gvn.transform(result));
1245 return true;
1246 }
1247
1248 bool LibraryCallKit::inline_preconditions_checkIndex() {
1249 Node* index = argument(0);
1250 Node* length = argument(1);
1251 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1252 return false;
1253 }
1254
1255 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1256 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1257
1258 {
1259 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1260 uncommon_trap(Deoptimization::Reason_intrinsic,
1261 Deoptimization::Action_make_not_entrant);
1262 }
1263
1264 if (stopped()) {
1265 return false;
1266 }
1267
1268 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1269 BoolTest::mask btest = BoolTest::lt;
1270 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1271 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1272 _gvn.set_type(rc, rc->Value(&_gvn));
1273 if (!rc_bool->is_Con()) {
1274 record_for_igvn(rc);
1275 }
1276 set_control(_gvn.transform(new IfTrueNode(rc)));
1277 {
1278 PreserveJVMState pjvms(this);
1279 set_control(_gvn.transform(new IfFalseNode(rc)));
1280 uncommon_trap(Deoptimization::Reason_range_check,
1281 Deoptimization::Action_make_not_entrant);
1282 }
1283
1284 if (stopped()) {
1285 return false;
1286 }
1287
1288 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1289 result->set_req(0, control());
1290 result = _gvn.transform(result);
1291 set_result(result);
1292 replace_in_map(index, result);
1293 clear_upper_avx();
1294 return true;
1295 }
1296
1297 //------------------------------inline_string_indexOf------------------------
1298 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1299 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1300 return false;
1301 }
1302 Node* src = argument(0);
1303 Node* tgt = argument(1);
1304
1305 // Make the merge point
1306 RegionNode* result_rgn = new RegionNode(4);
1307 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1308
1309 src = must_be_not_null(src, true);
1310 tgt = must_be_not_null(tgt, true);
1311
1312 // Get start addr and length of source string
1313 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1314 Node* src_count = load_array_length(src);
1315
1316 // Get start addr and length of substring
1317 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1318 Node* tgt_count = load_array_length(tgt);
1319
1320 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1321 // Divide src size by 2 if String is UTF16 encoded
1322 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1323 }
1324 if (ae == StrIntrinsicNode::UU) {
1325 // Divide substring size by 2 if String is UTF16 encoded
1326 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1327 }
1328
1329 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1330 if (result != NULL) {
1331 result_phi->init_req(3, result);
1332 result_rgn->init_req(3, control());
1333 }
1334 set_control(_gvn.transform(result_rgn));
1335 record_for_igvn(result_rgn);
1336 set_result(_gvn.transform(result_phi));
1337
1338 return true;
1339 }
1340
1341 //-----------------------------inline_string_indexOf-----------------------
1342 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1343 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1344 return false;
1345 }
1346 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1347 return false;
1348 }
1349 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1350 Node* src = argument(0); // byte[]
1351 Node* src_count = argument(1); // char count
1352 Node* tgt = argument(2); // byte[]
1353 Node* tgt_count = argument(3); // char count
1354 Node* from_index = argument(4); // char index
1355
1356 src = must_be_not_null(src, true);
1357 tgt = must_be_not_null(tgt, true);
1358
1359 // Multiply byte array index by 2 if String is UTF16 encoded
1360 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1361 src_count = _gvn.transform(new SubINode(src_count, from_index));
1362 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1363 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1364
1365 // Range checks
1366 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1367 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1368 if (stopped()) {
1369 return true;
1370 }
1371
1372 RegionNode* region = new RegionNode(5);
1373 Node* phi = new PhiNode(region, TypeInt::INT);
1374
1375 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1376 if (result != NULL) {
1377 // The result is index relative to from_index if substring was found, -1 otherwise.
1378 // Generate code which will fold into cmove.
1379 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1380 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1381
1382 Node* if_lt = generate_slow_guard(bol, NULL);
1383 if (if_lt != NULL) {
1384 // result == -1
1385 phi->init_req(3, result);
1386 region->init_req(3, if_lt);
1387 }
1388 if (!stopped()) {
1389 result = _gvn.transform(new AddINode(result, from_index));
1390 phi->init_req(4, result);
1391 region->init_req(4, control());
1392 }
1393 }
1394
1395 set_control(_gvn.transform(region));
1396 record_for_igvn(region);
1397 set_result(_gvn.transform(phi));
1398 clear_upper_avx();
1399
1400 return true;
1401 }
1402
1403 // Create StrIndexOfNode with fast path checks
1404 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1405 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1406 // Check for substr count > string count
1407 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1408 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1409 Node* if_gt = generate_slow_guard(bol, NULL);
1410 if (if_gt != NULL) {
1411 phi->init_req(1, intcon(-1));
1412 region->init_req(1, if_gt);
1413 }
1414 if (!stopped()) {
1415 // Check for substr count == 0
1416 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1417 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1418 Node* if_zero = generate_slow_guard(bol, NULL);
1419 if (if_zero != NULL) {
1420 phi->init_req(2, intcon(0));
1421 region->init_req(2, if_zero);
1422 }
1423 }
1424 if (!stopped()) {
1425 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1426 }
1427 return NULL;
1428 }
1429
1430 //-----------------------------inline_string_indexOfChar-----------------------
1431 bool LibraryCallKit::inline_string_indexOfChar() {
1432 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1433 return false;
1434 }
1435 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1436 return false;
1437 }
1438 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1439 Node* src = argument(0); // byte[]
1440 Node* tgt = argument(1); // tgt is int ch
1441 Node* from_index = argument(2);
1442 Node* max = argument(3);
1443
1444 src = must_be_not_null(src, true);
1445
1446 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1447 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1448 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1449
1450 // Range checks
1451 generate_string_range_check(src, src_offset, src_count, true);
1452 if (stopped()) {
1453 return true;
1454 }
1455
1456 RegionNode* region = new RegionNode(3);
1457 Node* phi = new PhiNode(region, TypeInt::INT);
1458
1459 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1460 C->set_has_split_ifs(true); // Has chance for split-if optimization
1461 _gvn.transform(result);
1462
1463 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1464 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1465
1466 Node* if_lt = generate_slow_guard(bol, NULL);
1467 if (if_lt != NULL) {
1468 // result == -1
1469 phi->init_req(2, result);
1470 region->init_req(2, if_lt);
1471 }
1472 if (!stopped()) {
1473 result = _gvn.transform(new AddINode(result, from_index));
1474 phi->init_req(1, result);
1475 region->init_req(1, control());
1476 }
1477 set_control(_gvn.transform(region));
1478 record_for_igvn(region);
1479 set_result(_gvn.transform(phi));
1480
1481 return true;
1482 }
1483 //---------------------------inline_string_copy---------------------
1484 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1485 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1486 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1487 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1488 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1489 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1490 bool LibraryCallKit::inline_string_copy(bool compress) {
1491 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1492 return false;
1493 }
1494 int nargs = 5; // 2 oops, 3 ints
1495 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1496
1497 Node* src = argument(0);
1498 Node* src_offset = argument(1);
1499 Node* dst = argument(2);
1500 Node* dst_offset = argument(3);
1501 Node* length = argument(4);
1502
1503 // Check for allocation before we add nodes that would confuse
1504 // tightly_coupled_allocation()
1505 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1506
1507 // Figure out the size and type of the elements we will be copying.
1508 const Type* src_type = src->Value(&_gvn);
1509 const Type* dst_type = dst->Value(&_gvn);
1510 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1511 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1512 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1513 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1514 "Unsupported array types for inline_string_copy");
1515
1516 src = must_be_not_null(src, true);
1517 dst = must_be_not_null(dst, true);
1518
1519 // Convert char[] offsets to byte[] offsets
1520 bool convert_src = (compress && src_elem == T_BYTE);
1521 bool convert_dst = (!compress && dst_elem == T_BYTE);
1522 if (convert_src) {
1523 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1524 } else if (convert_dst) {
1525 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1526 }
1527
1528 // Range checks
1529 generate_string_range_check(src, src_offset, length, convert_src);
1530 generate_string_range_check(dst, dst_offset, length, convert_dst);
1531 if (stopped()) {
1532 return true;
1533 }
1534
1535 Node* src_start = array_element_address(src, src_offset, src_elem);
1536 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1537 // 'src_start' points to src array + scaled offset
1538 // 'dst_start' points to dst array + scaled offset
1539 Node* count = NULL;
1540 if (compress) {
1541 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1542 } else {
1543 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1544 }
1545
1546 if (alloc != NULL) {
1547 if (alloc->maybe_set_complete(&_gvn)) {
1548 // "You break it, you buy it."
1549 InitializeNode* init = alloc->initialization();
1550 assert(init->is_complete(), "we just did this");
1551 init->set_complete_with_arraycopy();
1552 assert(dst->is_CheckCastPP(), "sanity");
1553 assert(dst->in(0)->in(0) == init, "dest pinned");
1554 }
1555 // Do not let stores that initialize this object be reordered with
1556 // a subsequent store that would make this object accessible by
1557 // other threads.
1558 // Record what AllocateNode this StoreStore protects so that
1559 // escape analysis can go from the MemBarStoreStoreNode to the
1560 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1561 // based on the escape status of the AllocateNode.
1562 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1563 }
1564 if (compress) {
1565 set_result(_gvn.transform(count));
1566 }
1567 clear_upper_avx();
1568
1569 return true;
1570 }
1571
1572 #ifdef _LP64
1573 #define XTOP ,top() /*additional argument*/
1574 #else //_LP64
1575 #define XTOP /*no additional argument*/
1576 #endif //_LP64
1577
1578 //------------------------inline_string_toBytesU--------------------------
1579 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1580 bool LibraryCallKit::inline_string_toBytesU() {
1581 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1582 return false;
1583 }
1584 // Get the arguments.
1585 Node* value = argument(0);
1586 Node* offset = argument(1);
1587 Node* length = argument(2);
1588
1589 Node* newcopy = NULL;
1590
1591 // Set the original stack and the reexecute bit for the interpreter to reexecute
1592 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1593 { PreserveReexecuteState preexecs(this);
1594 jvms()->set_should_reexecute(true);
1595
1596 // Check if a null path was taken unconditionally.
1597 value = null_check(value);
1598
1599 RegionNode* bailout = new RegionNode(1);
1600 record_for_igvn(bailout);
1601
1602 // Range checks
1603 generate_negative_guard(offset, bailout);
1604 generate_negative_guard(length, bailout);
1605 generate_limit_guard(offset, length, load_array_length(value), bailout);
1606 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1607 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1608
1609 if (bailout->req() > 1) {
1610 PreserveJVMState pjvms(this);
1611 set_control(_gvn.transform(bailout));
1612 uncommon_trap(Deoptimization::Reason_intrinsic,
1613 Deoptimization::Action_maybe_recompile);
1614 }
1615 if (stopped()) {
1616 return true;
1617 }
1618
1619 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1620 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1621 newcopy = new_array(klass_node, size, 0); // no arguments to push
1622 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1623
1624 // Calculate starting addresses.
1625 Node* src_start = array_element_address(value, offset, T_CHAR);
1626 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1627
1628 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1629 const TypeInt* toffset = gvn().type(offset)->is_int();
1630 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1631
1632 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1633 const char* copyfunc_name = "arraycopy";
1634 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1635 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1636 OptoRuntime::fast_arraycopy_Type(),
1637 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1638 src_start, dst_start, ConvI2X(length) XTOP);
1639 // Do not let reads from the cloned object float above the arraycopy.
1640 if (alloc != NULL) {
1641 if (alloc->maybe_set_complete(&_gvn)) {
1642 // "You break it, you buy it."
1643 InitializeNode* init = alloc->initialization();
1644 assert(init->is_complete(), "we just did this");
1645 init->set_complete_with_arraycopy();
1646 assert(newcopy->is_CheckCastPP(), "sanity");
1647 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1648 }
1649 // Do not let stores that initialize this object be reordered with
1650 // a subsequent store that would make this object accessible by
1651 // other threads.
1652 // Record what AllocateNode this StoreStore protects so that
1653 // escape analysis can go from the MemBarStoreStoreNode to the
1654 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1655 // based on the escape status of the AllocateNode.
1656 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1657 } else {
1658 insert_mem_bar(Op_MemBarCPUOrder);
1659 }
1660 } // original reexecute is set back here
1661
1662 C->set_has_split_ifs(true); // Has chance for split-if optimization
1663 if (!stopped()) {
1664 set_result(newcopy);
1665 }
1666 clear_upper_avx();
1667
1668 return true;
1669 }
1670
1671 //------------------------inline_string_getCharsU--------------------------
1672 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1673 bool LibraryCallKit::inline_string_getCharsU() {
1674 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1675 return false;
1676 }
1677
1678 // Get the arguments.
1679 Node* src = argument(0);
1680 Node* src_begin = argument(1);
1681 Node* src_end = argument(2); // exclusive offset (i < src_end)
1682 Node* dst = argument(3);
1683 Node* dst_begin = argument(4);
1684
1685 // Check for allocation before we add nodes that would confuse
1686 // tightly_coupled_allocation()
1687 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1688
1689 // Check if a null path was taken unconditionally.
1690 src = null_check(src);
1691 dst = null_check(dst);
1692 if (stopped()) {
1693 return true;
1694 }
1695
1696 // Get length and convert char[] offset to byte[] offset
1697 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1698 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1699
1700 // Range checks
1701 generate_string_range_check(src, src_begin, length, true);
1702 generate_string_range_check(dst, dst_begin, length, false);
1703 if (stopped()) {
1704 return true;
1705 }
1706
1707 if (!stopped()) {
1708 // Calculate starting addresses.
1709 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1710 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1711
1712 // Check if array addresses are aligned to HeapWordSize
1713 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1714 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1715 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1716 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1717
1718 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1719 const char* copyfunc_name = "arraycopy";
1720 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1721 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1722 OptoRuntime::fast_arraycopy_Type(),
1723 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1724 src_start, dst_start, ConvI2X(length) XTOP);
1725 // Do not let reads from the cloned object float above the arraycopy.
1726 if (alloc != NULL) {
1727 if (alloc->maybe_set_complete(&_gvn)) {
1728 // "You break it, you buy it."
1729 InitializeNode* init = alloc->initialization();
1730 assert(init->is_complete(), "we just did this");
1731 init->set_complete_with_arraycopy();
1732 assert(dst->is_CheckCastPP(), "sanity");
1733 assert(dst->in(0)->in(0) == init, "dest pinned");
1734 }
1735 // Do not let stores that initialize this object be reordered with
1736 // a subsequent store that would make this object accessible by
1737 // other threads.
1738 // Record what AllocateNode this StoreStore protects so that
1739 // escape analysis can go from the MemBarStoreStoreNode to the
1740 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1741 // based on the escape status of the AllocateNode.
1742 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1743 } else {
1744 insert_mem_bar(Op_MemBarCPUOrder);
1745 }
1746 }
1747
1748 C->set_has_split_ifs(true); // Has chance for split-if optimization
1749 return true;
1750 }
1751
1752 //----------------------inline_string_char_access----------------------------
1753 // Store/Load char to/from byte[] array.
1754 // static void StringUTF16.putChar(byte[] val, int index, int c)
1755 // static char StringUTF16.getChar(byte[] val, int index)
1756 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1757 Node* value = argument(0);
1758 Node* index = argument(1);
1759 Node* ch = is_store ? argument(2) : NULL;
1760
1761 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1762 // correctly requires matched array shapes.
1763 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1764 "sanity: byte[] and char[] bases agree");
1765 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1766 "sanity: byte[] and char[] scales agree");
1767
1768 // Bail when getChar over constants is requested: constant folding would
1769 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1770 // Java method would constant fold nicely instead.
1771 if (!is_store && value->is_Con() && index->is_Con()) {
1772 return false;
1773 }
1774
1775 value = must_be_not_null(value, true);
1776
1777 Node* adr = array_element_address(value, index, T_CHAR);
1778 if (adr->is_top()) {
1779 return false;
1780 }
1781 if (is_store) {
1782 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1783 } else {
1784 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
1785 set_result(ch);
1786 }
1787 return true;
1788 }
1789
1790 //--------------------------round_double_node--------------------------------
1791 // Round a double node if necessary.
1792 Node* LibraryCallKit::round_double_node(Node* n) {
1793 if (Matcher::strict_fp_requires_explicit_rounding) {
1794 #ifdef IA32
1795 if (UseSSE < 2) {
1796 n = _gvn.transform(new RoundDoubleNode(NULL, n));
1797 }
1798 #else
1799 Unimplemented();
1800 #endif // IA32
1801 }
1802 return n;
1803 }
1804
1805 //------------------------------inline_math-----------------------------------
1806 // public static double Math.abs(double)
1807 // public static double Math.sqrt(double)
1808 // public static double Math.log(double)
1809 // public static double Math.log10(double)
1810 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1811 Node* arg = round_double_node(argument(0));
1812 Node* n = NULL;
1813 switch (id) {
1814 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1815 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1816 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1817 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1818 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1819 default: fatal_unexpected_iid(id); break;
1820 }
1821 set_result(_gvn.transform(n));
1822 return true;
1823 }
1824
1825 //------------------------------inline_math-----------------------------------
1826 // public static float Math.abs(float)
1827 // public static int Math.abs(int)
1828 // public static long Math.abs(long)
1829 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1830 Node* arg = argument(0);
1831 Node* n = NULL;
1832 switch (id) {
1833 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1834 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1835 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1836 default: fatal_unexpected_iid(id); break;
1837 }
1838 set_result(_gvn.transform(n));
1839 return true;
1840 }
1841
1842 //------------------------------runtime_math-----------------------------
1843 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1844 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1845 "must be (DD)D or (D)D type");
1846
1847 // Inputs
1848 Node* a = round_double_node(argument(0));
1849 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1850
1851 const TypePtr* no_memory_effects = NULL;
1852 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1853 no_memory_effects,
1854 a, top(), b, b ? top() : NULL);
1855 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1856 #ifdef ASSERT
1857 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1858 assert(value_top == top(), "second value must be top");
1859 #endif
1860
1861 set_result(value);
1862 return true;
1863 }
1864
1865 //------------------------------inline_math_native-----------------------------
1866 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1867 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1868 switch (id) {
1869 // These intrinsics are not properly supported on all hardware
1870 case vmIntrinsics::_dsin:
1871 return StubRoutines::dsin() != NULL ?
1872 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1873 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1874 case vmIntrinsics::_dcos:
1875 return StubRoutines::dcos() != NULL ?
1876 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1877 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1878 case vmIntrinsics::_dtan:
1879 return StubRoutines::dtan() != NULL ?
1880 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1881 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1882 case vmIntrinsics::_dlog:
1883 return StubRoutines::dlog() != NULL ?
1884 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1885 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1886 case vmIntrinsics::_dlog10:
1887 return StubRoutines::dlog10() != NULL ?
1888 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1889 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1890
1891 // These intrinsics are supported on all hardware
1892 case vmIntrinsics::_ceil:
1893 case vmIntrinsics::_floor:
1894 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1895 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1896 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1897 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1898 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1899 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1900
1901 case vmIntrinsics::_dexp:
1902 return StubRoutines::dexp() != NULL ?
1903 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1904 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1905 case vmIntrinsics::_dpow: {
1906 Node* exp = round_double_node(argument(2));
1907 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1908 if (d != NULL && d->getd() == 2.0) {
1909 // Special case: pow(x, 2.0) => x * x
1910 Node* base = round_double_node(argument(0));
1911 set_result(_gvn.transform(new MulDNode(base, base)));
1912 return true;
1913 }
1914 return StubRoutines::dpow() != NULL ?
1915 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1916 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1917 }
1918 #undef FN_PTR
1919
1920 // These intrinsics are not yet correctly implemented
1921 case vmIntrinsics::_datan2:
1922 return false;
1923
1924 default:
1925 fatal_unexpected_iid(id);
1926 return false;
1927 }
1928 }
1929
1930 static bool is_simple_name(Node* n) {
1931 return (n->req() == 1 // constant
1932 || (n->is_Type() && n->as_Type()->type()->singleton())
1933 || n->is_Proj() // parameter or return value
1934 || n->is_Phi() // local of some sort
1935 );
1936 }
1937
1938 //----------------------------inline_notify-----------------------------------*
1939 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1940 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1941 address func;
1942 if (id == vmIntrinsics::_notify) {
1943 func = OptoRuntime::monitor_notify_Java();
1944 } else {
1945 func = OptoRuntime::monitor_notifyAll_Java();
1946 }
1947 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1948 make_slow_call_ex(call, env()->Throwable_klass(), false);
1949 return true;
1950 }
1951
1952
1953 //----------------------------inline_min_max-----------------------------------
1954 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1955 set_result(generate_min_max(id, argument(0), argument(1)));
1956 return true;
1957 }
1958
1959 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1960 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1961 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1962 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1963 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1964
1965 {
1966 PreserveJVMState pjvms(this);
1967 PreserveReexecuteState preexecs(this);
1968 jvms()->set_should_reexecute(true);
1969
1970 set_control(slow_path);
1971 set_i_o(i_o());
1972
1973 uncommon_trap(Deoptimization::Reason_intrinsic,
1974 Deoptimization::Action_none);
1975 }
1976
1977 set_control(fast_path);
1978 set_result(math);
1979 }
1980
1981 template <typename OverflowOp>
1982 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1983 typedef typename OverflowOp::MathOp MathOp;
1984
1985 MathOp* mathOp = new MathOp(arg1, arg2);
1986 Node* operation = _gvn.transform( mathOp );
1987 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1988 inline_math_mathExact(operation, ofcheck);
1989 return true;
1990 }
1991
1992 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1993 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1994 }
1995
1996 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1997 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1998 }
1999
2000 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2001 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2002 }
2003
2004 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2005 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2006 }
2007
2008 bool LibraryCallKit::inline_math_negateExactI() {
2009 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2010 }
2011
2012 bool LibraryCallKit::inline_math_negateExactL() {
2013 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2014 }
2015
2016 bool LibraryCallKit::inline_math_multiplyExactI() {
2017 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2018 }
2019
2020 bool LibraryCallKit::inline_math_multiplyExactL() {
2021 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2022 }
2023
2024 bool LibraryCallKit::inline_math_multiplyHigh() {
2025 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2026 return true;
2027 }
2028
2029 Node*
2030 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2031 // These are the candidate return value:
2032 Node* xvalue = x0;
2033 Node* yvalue = y0;
2034
2035 if (xvalue == yvalue) {
2036 return xvalue;
2037 }
2038
2039 bool want_max = (id == vmIntrinsics::_max);
2040
2041 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
2042 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
2043 if (txvalue == NULL || tyvalue == NULL) return top();
2044 // This is not really necessary, but it is consistent with a
2045 // hypothetical MaxINode::Value method:
2046 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
2047
2048 // %%% This folding logic should (ideally) be in a different place.
2049 // Some should be inside IfNode, and there to be a more reliable
2050 // transformation of ?: style patterns into cmoves. We also want
2051 // more powerful optimizations around cmove and min/max.
2052
2053 // Try to find a dominating comparison of these guys.
2054 // It can simplify the index computation for Arrays.copyOf
2055 // and similar uses of System.arraycopy.
2056 // First, compute the normalized version of CmpI(x, y).
2057 int cmp_op = Op_CmpI;
2058 Node* xkey = xvalue;
2059 Node* ykey = yvalue;
2060 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
2061 if (ideal_cmpxy->is_Cmp()) {
2062 // E.g., if we have CmpI(length - offset, count),
2063 // it might idealize to CmpI(length, count + offset)
2064 cmp_op = ideal_cmpxy->Opcode();
2065 xkey = ideal_cmpxy->in(1);
2066 ykey = ideal_cmpxy->in(2);
2067 }
2068
2069 // Start by locating any relevant comparisons.
2070 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
2071 Node* cmpxy = NULL;
2072 Node* cmpyx = NULL;
2073 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
2074 Node* cmp = start_from->fast_out(k);
2075 if (cmp->outcnt() > 0 && // must have prior uses
2076 cmp->in(0) == NULL && // must be context-independent
2077 cmp->Opcode() == cmp_op) { // right kind of compare
2078 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
2079 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
2080 }
2081 }
2082
2083 const int NCMPS = 2;
2084 Node* cmps[NCMPS] = { cmpxy, cmpyx };
2085 int cmpn;
2086 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2087 if (cmps[cmpn] != NULL) break; // find a result
2088 }
2089 if (cmpn < NCMPS) {
2090 // Look for a dominating test that tells us the min and max.
2091 int depth = 0; // Limit search depth for speed
2092 Node* dom = control();
2093 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2094 if (++depth >= 100) break;
2095 Node* ifproj = dom;
2096 if (!ifproj->is_Proj()) continue;
2097 Node* iff = ifproj->in(0);
2098 if (!iff->is_If()) continue;
2099 Node* bol = iff->in(1);
2100 if (!bol->is_Bool()) continue;
2101 Node* cmp = bol->in(1);
2102 if (cmp == NULL) continue;
2103 for (cmpn = 0; cmpn < NCMPS; cmpn++)
2104 if (cmps[cmpn] == cmp) break;
2105 if (cmpn == NCMPS) continue;
2106 BoolTest::mask btest = bol->as_Bool()->_test._test;
2107 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
2108 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2109 // At this point, we know that 'x btest y' is true.
2110 switch (btest) {
2111 case BoolTest::eq:
2112 // They are proven equal, so we can collapse the min/max.
2113 // Either value is the answer. Choose the simpler.
2114 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2115 return yvalue;
2116 return xvalue;
2117 case BoolTest::lt: // x < y
2118 case BoolTest::le: // x <= y
2119 return (want_max ? yvalue : xvalue);
2120 case BoolTest::gt: // x > y
2121 case BoolTest::ge: // x >= y
2122 return (want_max ? xvalue : yvalue);
2123 default:
2124 break;
2125 }
2126 }
2127 }
2128
2129 // We failed to find a dominating test.
2130 // Let's pick a test that might GVN with prior tests.
2131 Node* best_bol = NULL;
2132 BoolTest::mask best_btest = BoolTest::illegal;
2133 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2134 Node* cmp = cmps[cmpn];
2135 if (cmp == NULL) continue;
2136 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2137 Node* bol = cmp->fast_out(j);
2138 if (!bol->is_Bool()) continue;
2139 BoolTest::mask btest = bol->as_Bool()->_test._test;
2140 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2141 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2142 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2143 best_bol = bol->as_Bool();
2144 best_btest = btest;
2145 }
2146 }
2147 }
2148
2149 Node* answer_if_true = NULL;
2150 Node* answer_if_false = NULL;
2151 switch (best_btest) {
2152 default:
2153 if (cmpxy == NULL)
2154 cmpxy = ideal_cmpxy;
2155 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2156 // and fall through:
2157 case BoolTest::lt: // x < y
2158 case BoolTest::le: // x <= y
2159 answer_if_true = (want_max ? yvalue : xvalue);
2160 answer_if_false = (want_max ? xvalue : yvalue);
2161 break;
2162 case BoolTest::gt: // x > y
2163 case BoolTest::ge: // x >= y
2164 answer_if_true = (want_max ? xvalue : yvalue);
2165 answer_if_false = (want_max ? yvalue : xvalue);
2166 break;
2167 }
2168
2169 jint hi, lo;
2170 if (want_max) {
2171 // We can sharpen the minimum.
2172 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2173 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2174 } else {
2175 // We can sharpen the maximum.
2176 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2177 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2178 }
2179
2180 // Use a flow-free graph structure, to avoid creating excess control edges
2181 // which could hinder other optimizations.
2182 // Since Math.min/max is often used with arraycopy, we want
2183 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2184 Node* cmov = CMoveNode::make(NULL, best_bol,
2185 answer_if_false, answer_if_true,
2186 TypeInt::make(lo, hi, widen));
2187
2188 return _gvn.transform(cmov);
2189
2190 /*
2191 // This is not as desirable as it may seem, since Min and Max
2192 // nodes do not have a full set of optimizations.
2193 // And they would interfere, anyway, with 'if' optimizations
2194 // and with CMoveI canonical forms.
2195 switch (id) {
2196 case vmIntrinsics::_min:
2197 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2198 case vmIntrinsics::_max:
2199 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2200 default:
2201 ShouldNotReachHere();
2202 }
2203 */
2204 }
2205
2206 inline int
2207 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2208 const TypePtr* base_type = TypePtr::NULL_PTR;
2209 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2210 if (base_type == NULL) {
2211 // Unknown type.
2212 return Type::AnyPtr;
2213 } else if (base_type == TypePtr::NULL_PTR) {
2214 // Since this is a NULL+long form, we have to switch to a rawptr.
2215 base = _gvn.transform(new CastX2PNode(offset));
2216 offset = MakeConX(0);
2217 return Type::RawPtr;
2218 } else if (base_type->base() == Type::RawPtr) {
2219 return Type::RawPtr;
2220 } else if (base_type->isa_oopptr()) {
2221 // Base is never null => always a heap address.
2222 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2223 return Type::OopPtr;
2224 }
2225 // Offset is small => always a heap address.
2226 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2227 if (offset_type != NULL &&
2228 base_type->offset() == 0 && // (should always be?)
2229 offset_type->_lo >= 0 &&
2230 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2231 return Type::OopPtr;
2232 } else if (type == T_OBJECT) {
2233 // off heap access to an oop doesn't make any sense. Has to be on
2234 // heap.
2235 return Type::OopPtr;
2236 }
2237 // Otherwise, it might either be oop+off or NULL+addr.
2238 return Type::AnyPtr;
2239 } else {
2240 // No information:
2241 return Type::AnyPtr;
2242 }
2243 }
2244
2245 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type, bool can_cast) {
2246 Node* uncasted_base = base;
2247 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2248 if (kind == Type::RawPtr) {
2249 return basic_plus_adr(top(), uncasted_base, offset);
2250 } else if (kind == Type::AnyPtr) {
2251 assert(base == uncasted_base, "unexpected base change");
2252 if (can_cast) {
2253 if (!_gvn.type(base)->speculative_maybe_null() &&
2254 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2255 // According to profiling, this access is always on
2256 // heap. Casting the base to not null and thus avoiding membars
2257 // around the access should allow better optimizations
2258 Node* null_ctl = top();
2259 base = null_check_oop(base, &null_ctl, true, true, true);
2260 assert(null_ctl->is_top(), "no null control here");
2261 return basic_plus_adr(base, offset);
2262 } else if (_gvn.type(base)->speculative_always_null() &&
2263 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2264 // According to profiling, this access is always off
2265 // heap.
2266 base = null_assert(base);
2267 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2268 offset = MakeConX(0);
2269 return basic_plus_adr(top(), raw_base, offset);
2270 }
2271 }
2272 // We don't know if it's an on heap or off heap access. Fall back
2273 // to raw memory access.
2274 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2275 return basic_plus_adr(top(), raw, offset);
2276 } else {
2277 assert(base == uncasted_base, "unexpected base change");
2278 // We know it's an on heap access so base can't be null
2279 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2280 base = must_be_not_null(base, true);
2281 }
2282 return basic_plus_adr(base, offset);
2283 }
2284 }
2285
2286 //--------------------------inline_number_methods-----------------------------
2287 // inline int Integer.numberOfLeadingZeros(int)
2288 // inline int Long.numberOfLeadingZeros(long)
2289 //
2290 // inline int Integer.numberOfTrailingZeros(int)
2291 // inline int Long.numberOfTrailingZeros(long)
2292 //
2293 // inline int Integer.bitCount(int)
2294 // inline int Long.bitCount(long)
2295 //
2296 // inline char Character.reverseBytes(char)
2297 // inline short Short.reverseBytes(short)
2298 // inline int Integer.reverseBytes(int)
2299 // inline long Long.reverseBytes(long)
2300 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2301 Node* arg = argument(0);
2302 Node* n = NULL;
2303 switch (id) {
2304 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2305 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2306 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2307 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2308 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2309 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2310 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2311 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2312 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2313 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2314 default: fatal_unexpected_iid(id); break;
2315 }
2316 set_result(_gvn.transform(n));
2317 return true;
2318 }
2319
2320 //----------------------------inline_unsafe_access----------------------------
2321
2322 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2323 // Attempt to infer a sharper value type from the offset and base type.
2324 ciKlass* sharpened_klass = NULL;
2325
2326 // See if it is an instance field, with an object type.
2327 if (alias_type->field() != NULL) {
2328 if (alias_type->field()->type()->is_klass()) {
2329 sharpened_klass = alias_type->field()->type()->as_klass();
2330 }
2331 }
2332
2333 // See if it is a narrow oop array.
2334 if (adr_type->isa_aryptr()) {
2335 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2336 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2337 if (elem_type != NULL) {
2338 sharpened_klass = elem_type->klass();
2339 }
2340 }
2341 }
2342
2343 // The sharpened class might be unloaded if there is no class loader
2344 // contraint in place.
2345 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2346 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2347
2348 #ifndef PRODUCT
2349 if (C->print_intrinsics() || C->print_inlining()) {
2350 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2351 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2352 }
2353 #endif
2354 // Sharpen the value type.
2355 return tjp;
2356 }
2357 return NULL;
2358 }
2359
2360 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2361 switch (kind) {
2362 case Relaxed:
2363 return MO_UNORDERED;
2364 case Opaque:
2365 return MO_RELAXED;
2366 case Acquire:
2367 return MO_ACQUIRE;
2368 case Release:
2369 return MO_RELEASE;
2370 case Volatile:
2371 return MO_SEQ_CST;
2372 default:
2373 ShouldNotReachHere();
2374 return 0;
2375 }
2376 }
2377
2378 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2379 if (callee()->is_static()) return false; // caller must have the capability!
2380 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2381 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2382 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2383 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2384
2385 if (is_reference_type(type)) {
2386 decorators |= ON_UNKNOWN_OOP_REF;
2387 }
2388
2389 if (unaligned) {
2390 decorators |= C2_UNALIGNED;
2391 }
2392
2393 #ifndef PRODUCT
2394 {
2395 ResourceMark rm;
2396 // Check the signatures.
2397 ciSignature* sig = callee()->signature();
2398 #ifdef ASSERT
2399 if (!is_store) {
2400 // Object getReference(Object base, int/long offset), etc.
2401 BasicType rtype = sig->return_type()->basic_type();
2402 assert(rtype == type, "getter must return the expected value");
2403 assert(sig->count() == 2, "oop getter has 2 arguments");
2404 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2405 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2406 } else {
2407 // void putReference(Object base, int/long offset, Object x), etc.
2408 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2409 assert(sig->count() == 3, "oop putter has 3 arguments");
2410 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2411 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2412 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2413 assert(vtype == type, "putter must accept the expected value");
2414 }
2415 #endif // ASSERT
2416 }
2417 #endif //PRODUCT
2418
2419 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2420
2421 Node* receiver = argument(0); // type: oop
2422
2423 // Build address expression.
2424 Node* adr;
2425 Node* heap_base_oop = top();
2426 Node* offset = top();
2427 Node* val;
2428
2429 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2430 Node* base = argument(1); // type: oop
2431 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2432 offset = argument(2); // type: long
2433 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2434 // to be plain byte offsets, which are also the same as those accepted
2435 // by oopDesc::field_addr.
2436 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2437 "fieldOffset must be byte-scaled");
2438 // 32-bit machines ignore the high half!
2439 offset = ConvL2X(offset);
2440 adr = make_unsafe_address(base, offset, is_store ? ACCESS_WRITE : ACCESS_READ, type, kind == Relaxed);
2441
2442 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2443 if (type != T_OBJECT) {
2444 decorators |= IN_NATIVE; // off-heap primitive access
2445 } else {
2446 return false; // off-heap oop accesses are not supported
2447 }
2448 } else {
2449 heap_base_oop = base; // on-heap or mixed access
2450 }
2451
2452 // Can base be NULL? Otherwise, always on-heap access.
2453 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2454
2455 if (!can_access_non_heap) {
2456 decorators |= IN_HEAP;
2457 }
2458
2459 val = is_store ? argument(4) : NULL;
2460
2461 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2462 if (adr_type == TypePtr::NULL_PTR) {
2463 return false; // off-heap access with zero address
2464 }
2465
2466 // Try to categorize the address.
2467 Compile::AliasType* alias_type = C->alias_type(adr_type);
2468 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2469
2470 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2471 alias_type->adr_type() == TypeAryPtr::RANGE) {
2472 return false; // not supported
2473 }
2474
2475 bool mismatched = false;
2476 BasicType bt = alias_type->basic_type();
2477 if (bt != T_ILLEGAL) {
2478 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2479 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2480 // Alias type doesn't differentiate between byte[] and boolean[]).
2481 // Use address type to get the element type.
2482 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2483 }
2484 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2485 // accessing an array field with getReference is not a mismatch
2486 bt = T_OBJECT;
2487 }
2488 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2489 // Don't intrinsify mismatched object accesses
2490 return false;
2491 }
2492 mismatched = (bt != type);
2493 } else if (alias_type->adr_type()->isa_oopptr()) {
2494 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2495 }
2496
2497 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2498
2499 if (mismatched) {
2500 decorators |= C2_MISMATCHED;
2501 }
2502
2503 // First guess at the value type.
2504 const Type *value_type = Type::get_const_basic_type(type);
2505
2506 // Figure out the memory ordering.
2507 decorators |= mo_decorator_for_access_kind(kind);
2508
2509 if (!is_store && type == T_OBJECT) {
2510 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2511 if (tjp != NULL) {
2512 value_type = tjp;
2513 }
2514 }
2515
2516 receiver = null_check(receiver);
2517 if (stopped()) {
2518 return true;
2519 }
2520 // Heap pointers get a null-check from the interpreter,
2521 // as a courtesy. However, this is not guaranteed by Unsafe,
2522 // and it is not possible to fully distinguish unintended nulls
2523 // from intended ones in this API.
2524
2525 if (!is_store) {
2526 Node* p = NULL;
2527 // Try to constant fold a load from a constant field
2528 ciField* field = alias_type->field();
2529 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2530 // final or stable field
2531 p = make_constant_from_field(field, heap_base_oop);
2532 }
2533
2534 if (p == NULL) { // Could not constant fold the load
2535 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2536 // Normalize the value returned by getBoolean in the following cases
2537 if (type == T_BOOLEAN &&
2538 (mismatched ||
2539 heap_base_oop == top() || // - heap_base_oop is NULL or
2540 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2541 // and the unsafe access is made to large offset
2542 // (i.e., larger than the maximum offset necessary for any
2543 // field access)
2544 ) {
2545 IdealKit ideal = IdealKit(this);
2546 #define __ ideal.
2547 IdealVariable normalized_result(ideal);
2548 __ declarations_done();
2549 __ set(normalized_result, p);
2550 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2551 __ set(normalized_result, ideal.ConI(1));
2552 ideal.end_if();
2553 final_sync(ideal);
2554 p = __ value(normalized_result);
2555 #undef __
2556 }
2557 }
2558 if (type == T_ADDRESS) {
2559 p = gvn().transform(new CastP2XNode(NULL, p));
2560 p = ConvX2UL(p);
2561 }
2562 // The load node has the control of the preceding MemBarCPUOrder. All
2563 // following nodes will have the control of the MemBarCPUOrder inserted at
2564 // the end of this method. So, pushing the load onto the stack at a later
2565 // point is fine.
2566 set_result(p);
2567 } else {
2568 if (bt == T_ADDRESS) {
2569 // Repackage the long as a pointer.
2570 val = ConvL2X(val);
2571 val = gvn().transform(new CastX2PNode(val));
2572 }
2573 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2574 }
2575
2576 return true;
2577 }
2578
2579 //----------------------------inline_unsafe_load_store----------------------------
2580 // This method serves a couple of different customers (depending on LoadStoreKind):
2581 //
2582 // LS_cmp_swap:
2583 //
2584 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2585 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2586 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2587 //
2588 // LS_cmp_swap_weak:
2589 //
2590 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2591 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2592 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2593 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2594 //
2595 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2596 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2597 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2598 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2599 //
2600 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2601 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2602 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2603 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2604 //
2605 // LS_cmp_exchange:
2606 //
2607 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2608 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2609 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2610 //
2611 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2612 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2613 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2614 //
2615 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2616 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2617 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2618 //
2619 // LS_get_add:
2620 //
2621 // int getAndAddInt( Object o, long offset, int delta)
2622 // long getAndAddLong(Object o, long offset, long delta)
2623 //
2624 // LS_get_set:
2625 //
2626 // int getAndSet(Object o, long offset, int newValue)
2627 // long getAndSet(Object o, long offset, long newValue)
2628 // Object getAndSet(Object o, long offset, Object newValue)
2629 //
2630 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2631 // This basic scheme here is the same as inline_unsafe_access, but
2632 // differs in enough details that combining them would make the code
2633 // overly confusing. (This is a true fact! I originally combined
2634 // them, but even I was confused by it!) As much code/comments as
2635 // possible are retained from inline_unsafe_access though to make
2636 // the correspondences clearer. - dl
2637
2638 if (callee()->is_static()) return false; // caller must have the capability!
2639
2640 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2641 decorators |= mo_decorator_for_access_kind(access_kind);
2642
2643 #ifndef PRODUCT
2644 BasicType rtype;
2645 {
2646 ResourceMark rm;
2647 // Check the signatures.
2648 ciSignature* sig = callee()->signature();
2649 rtype = sig->return_type()->basic_type();
2650 switch(kind) {
2651 case LS_get_add:
2652 case LS_get_set: {
2653 // Check the signatures.
2654 #ifdef ASSERT
2655 assert(rtype == type, "get and set must return the expected type");
2656 assert(sig->count() == 3, "get and set has 3 arguments");
2657 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2658 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2659 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2660 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2661 #endif // ASSERT
2662 break;
2663 }
2664 case LS_cmp_swap:
2665 case LS_cmp_swap_weak: {
2666 // Check the signatures.
2667 #ifdef ASSERT
2668 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2669 assert(sig->count() == 4, "CAS has 4 arguments");
2670 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2671 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2672 #endif // ASSERT
2673 break;
2674 }
2675 case LS_cmp_exchange: {
2676 // Check the signatures.
2677 #ifdef ASSERT
2678 assert(rtype == type, "CAS must return the expected type");
2679 assert(sig->count() == 4, "CAS has 4 arguments");
2680 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2681 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2682 #endif // ASSERT
2683 break;
2684 }
2685 default:
2686 ShouldNotReachHere();
2687 }
2688 }
2689 #endif //PRODUCT
2690
2691 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2692
2693 // Get arguments:
2694 Node* receiver = NULL;
2695 Node* base = NULL;
2696 Node* offset = NULL;
2697 Node* oldval = NULL;
2698 Node* newval = NULL;
2699 switch(kind) {
2700 case LS_cmp_swap:
2701 case LS_cmp_swap_weak:
2702 case LS_cmp_exchange: {
2703 const bool two_slot_type = type2size[type] == 2;
2704 receiver = argument(0); // type: oop
2705 base = argument(1); // type: oop
2706 offset = argument(2); // type: long
2707 oldval = argument(4); // type: oop, int, or long
2708 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2709 break;
2710 }
2711 case LS_get_add:
2712 case LS_get_set: {
2713 receiver = argument(0); // type: oop
2714 base = argument(1); // type: oop
2715 offset = argument(2); // type: long
2716 oldval = NULL;
2717 newval = argument(4); // type: oop, int, or long
2718 break;
2719 }
2720 default:
2721 ShouldNotReachHere();
2722 }
2723
2724 // Build field offset expression.
2725 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2726 // to be plain byte offsets, which are also the same as those accepted
2727 // by oopDesc::field_addr.
2728 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2729 // 32-bit machines ignore the high half of long offsets
2730 offset = ConvL2X(offset);
2731 Node* adr = make_unsafe_address(base, offset, ACCESS_WRITE | ACCESS_READ, type, false);
2732 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2733
2734 Compile::AliasType* alias_type = C->alias_type(adr_type);
2735 BasicType bt = alias_type->basic_type();
2736 if (bt != T_ILLEGAL &&
2737 (is_reference_type(bt) != (type == T_OBJECT))) {
2738 // Don't intrinsify mismatched object accesses.
2739 return false;
2740 }
2741
2742 // For CAS, unlike inline_unsafe_access, there seems no point in
2743 // trying to refine types. Just use the coarse types here.
2744 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2745 const Type *value_type = Type::get_const_basic_type(type);
2746
2747 switch (kind) {
2748 case LS_get_set:
2749 case LS_cmp_exchange: {
2750 if (type == T_OBJECT) {
2751 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2752 if (tjp != NULL) {
2753 value_type = tjp;
2754 }
2755 }
2756 break;
2757 }
2758 case LS_cmp_swap:
2759 case LS_cmp_swap_weak:
2760 case LS_get_add:
2761 break;
2762 default:
2763 ShouldNotReachHere();
2764 }
2765
2766 // Null check receiver.
2767 receiver = null_check(receiver);
2768 if (stopped()) {
2769 return true;
2770 }
2771
2772 int alias_idx = C->get_alias_index(adr_type);
2773
2774 if (is_reference_type(type)) {
2775 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2776
2777 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2778 // could be delayed during Parse (for example, in adjust_map_after_if()).
2779 // Execute transformation here to avoid barrier generation in such case.
2780 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2781 newval = _gvn.makecon(TypePtr::NULL_PTR);
2782
2783 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2784 // Refine the value to a null constant, when it is known to be null
2785 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2786 }
2787 }
2788
2789 Node* result = NULL;
2790 switch (kind) {
2791 case LS_cmp_exchange: {
2792 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2793 oldval, newval, value_type, type, decorators);
2794 break;
2795 }
2796 case LS_cmp_swap_weak:
2797 decorators |= C2_WEAK_CMPXCHG;
2798 case LS_cmp_swap: {
2799 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2800 oldval, newval, value_type, type, decorators);
2801 break;
2802 }
2803 case LS_get_set: {
2804 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2805 newval, value_type, type, decorators);
2806 break;
2807 }
2808 case LS_get_add: {
2809 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2810 newval, value_type, type, decorators);
2811 break;
2812 }
2813 default:
2814 ShouldNotReachHere();
2815 }
2816
2817 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2818 set_result(result);
2819 return true;
2820 }
2821
2822 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2823 // Regardless of form, don't allow previous ld/st to move down,
2824 // then issue acquire, release, or volatile mem_bar.
2825 insert_mem_bar(Op_MemBarCPUOrder);
2826 switch(id) {
2827 case vmIntrinsics::_loadFence:
2828 insert_mem_bar(Op_LoadFence);
2829 return true;
2830 case vmIntrinsics::_storeFence:
2831 insert_mem_bar(Op_StoreFence);
2832 return true;
2833 case vmIntrinsics::_fullFence:
2834 insert_mem_bar(Op_MemBarVolatile);
2835 return true;
2836 default:
2837 fatal_unexpected_iid(id);
2838 return false;
2839 }
2840 }
2841
2842 bool LibraryCallKit::inline_onspinwait() {
2843 insert_mem_bar(Op_OnSpinWait);
2844 return true;
2845 }
2846
2847 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2848 if (!kls->is_Con()) {
2849 return true;
2850 }
2851 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
2852 if (klsptr == NULL) {
2853 return true;
2854 }
2855 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
2856 // don't need a guard for a klass that is already initialized
2857 return !ik->is_initialized();
2858 }
2859
2860 //----------------------------inline_unsafe_writeback0-------------------------
2861 // public native void Unsafe.writeback0(long address)
2862 bool LibraryCallKit::inline_unsafe_writeback0() {
2863 if (!Matcher::has_match_rule(Op_CacheWB)) {
2864 return false;
2865 }
2866 #ifndef PRODUCT
2867 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2868 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2869 ciSignature* sig = callee()->signature();
2870 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2871 #endif
2872 null_check_receiver(); // null-check, then ignore
2873 Node *addr = argument(1);
2874 addr = new CastX2PNode(addr);
2875 addr = _gvn.transform(addr);
2876 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2877 flush = _gvn.transform(flush);
2878 set_memory(flush, TypeRawPtr::BOTTOM);
2879 return true;
2880 }
2881
2882 //----------------------------inline_unsafe_writeback0-------------------------
2883 // public native void Unsafe.writeback0(long address)
2884 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2885 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2886 return false;
2887 }
2888 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2889 return false;
2890 }
2891 #ifndef PRODUCT
2892 assert(Matcher::has_match_rule(Op_CacheWB),
2893 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2894 : "found match rule for CacheWBPostSync but not CacheWB"));
2895
2896 #endif
2897 null_check_receiver(); // null-check, then ignore
2898 Node *sync;
2899 if (is_pre) {
2900 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2901 } else {
2902 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2903 }
2904 sync = _gvn.transform(sync);
2905 set_memory(sync, TypeRawPtr::BOTTOM);
2906 return true;
2907 }
2908
2909 //----------------------------inline_unsafe_allocate---------------------------
2910 // public native Object Unsafe.allocateInstance(Class<?> cls);
2911 bool LibraryCallKit::inline_unsafe_allocate() {
2912 if (callee()->is_static()) return false; // caller must have the capability!
2913
2914 null_check_receiver(); // null-check, then ignore
2915 Node* cls = null_check(argument(1));
2916 if (stopped()) return true;
2917
2918 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2919 kls = null_check(kls);
2920 if (stopped()) return true; // argument was like int.class
2921
2922 Node* test = NULL;
2923 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2924 // Note: The argument might still be an illegal value like
2925 // Serializable.class or Object[].class. The runtime will handle it.
2926 // But we must make an explicit check for initialization.
2927 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2928 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2929 // can generate code to load it as unsigned byte.
2930 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2931 Node* bits = intcon(InstanceKlass::fully_initialized);
2932 test = _gvn.transform(new SubINode(inst, bits));
2933 // The 'test' is non-zero if we need to take a slow path.
2934 }
2935
2936 Node* obj = new_instance(kls, test);
2937 set_result(obj);
2938 return true;
2939 }
2940
2941 //------------------------inline_native_time_funcs--------------
2942 // inline code for System.currentTimeMillis() and System.nanoTime()
2943 // these have the same type and signature
2944 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2945 const TypeFunc* tf = OptoRuntime::void_long_Type();
2946 const TypePtr* no_memory_effects = NULL;
2947 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2948 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2949 #ifdef ASSERT
2950 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2951 assert(value_top == top(), "second value must be top");
2952 #endif
2953 set_result(value);
2954 return true;
2955 }
2956
2957 #ifdef JFR_HAVE_INTRINSICS
2958
2959 /*
2960 * oop -> myklass
2961 * myklass->trace_id |= USED
2962 * return myklass->trace_id & ~0x3
2963 */
2964 bool LibraryCallKit::inline_native_classID() {
2965 Node* cls = null_check(argument(0), T_OBJECT);
2966 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2967 kls = null_check(kls, T_OBJECT);
2968
2969 ByteSize offset = KLASS_TRACE_ID_OFFSET;
2970 Node* insp = basic_plus_adr(kls, in_bytes(offset));
2971 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
2972
2973 Node* clsused = longcon(0x01l); // set the class bit
2974 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
2975 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
2976 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
2977
2978 #ifdef TRACE_ID_META_BITS
2979 Node* mbits = longcon(~TRACE_ID_META_BITS);
2980 tvalue = _gvn.transform(new AndLNode(tvalue, mbits));
2981 #endif
2982 #ifdef TRACE_ID_SHIFT
2983 Node* cbits = intcon(TRACE_ID_SHIFT);
2984 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits));
2985 #endif
2986
2987 set_result(tvalue);
2988 return true;
2989
2990 }
2991
2992 bool LibraryCallKit::inline_native_getEventWriter() {
2993 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
2994
2995 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
2996 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
2997
2998 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
2999
3000 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
3001 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
3002
3003 IfNode* iff_jobj_null =
3004 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
3005
3006 enum { _normal_path = 1,
3007 _null_path = 2,
3008 PATH_LIMIT };
3009
3010 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3011 PhiNode* result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3012
3013 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
3014 result_rgn->init_req(_null_path, jobj_is_null);
3015 result_val->init_req(_null_path, null());
3016
3017 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
3018 set_control(jobj_is_not_null);
3019 Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT,
3020 IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3021 result_rgn->init_req(_normal_path, control());
3022 result_val->init_req(_normal_path, res);
3023
3024 set_result(result_rgn, result_val);
3025
3026 return true;
3027 }
3028
3029 #endif // JFR_HAVE_INTRINSICS
3030
3031 //------------------------inline_native_currentThread------------------
3032 bool LibraryCallKit::inline_native_currentThread() {
3033 Node* junk = NULL;
3034 set_result(generate_current_thread(junk));
3035 return true;
3036 }
3037
3038 //---------------------------load_mirror_from_klass----------------------------
3039 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3040 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3041 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3042 Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3043 // mirror = ((OopHandle)mirror)->resolve();
3044 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3045 }
3046
3047 //-----------------------load_klass_from_mirror_common-------------------------
3048 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3049 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3050 // and branch to the given path on the region.
3051 // If never_see_null, take an uncommon trap on null, so we can optimistically
3052 // compile for the non-null case.
3053 // If the region is NULL, force never_see_null = true.
3054 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3055 bool never_see_null,
3056 RegionNode* region,
3057 int null_path,
3058 int offset) {
3059 if (region == NULL) never_see_null = true;
3060 Node* p = basic_plus_adr(mirror, offset);
3061 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3062 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3063 Node* null_ctl = top();
3064 kls = null_check_oop(kls, &null_ctl, never_see_null);
3065 if (region != NULL) {
3066 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3067 region->init_req(null_path, null_ctl);
3068 } else {
3069 assert(null_ctl == top(), "no loose ends");
3070 }
3071 return kls;
3072 }
3073
3074 //--------------------(inline_native_Class_query helpers)---------------------
3075 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3076 // Fall through if (mods & mask) == bits, take the guard otherwise.
3077 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3078 // Branch around if the given klass has the given modifier bit set.
3079 // Like generate_guard, adds a new path onto the region.
3080 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3081 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3082 Node* mask = intcon(modifier_mask);
3083 Node* bits = intcon(modifier_bits);
3084 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3085 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3086 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3087 return generate_fair_guard(bol, region);
3088 }
3089 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3090 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3091 }
3092 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3093 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 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::_isHidden:
3131 prim_return_value = intcon(0);
3132 break;
3133 case vmIntrinsics::_getSuperclass:
3134 prim_return_value = null();
3135 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3136 break;
3137 case vmIntrinsics::_getClassAccessFlags:
3138 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3139 return_type = TypeInt::INT; // not bool! 6297094
3140 break;
3141 default:
3142 fatal_unexpected_iid(id);
3143 break;
3144 }
3145
3146 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3147 if (mirror_con == NULL) return false; // cannot happen?
3148
3149 #ifndef PRODUCT
3150 if (C->print_intrinsics() || C->print_inlining()) {
3151 ciType* k = mirror_con->java_mirror_type();
3152 if (k) {
3153 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3154 k->print_name();
3155 tty->cr();
3156 }
3157 }
3158 #endif
3159
3160 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3161 RegionNode* region = new RegionNode(PATH_LIMIT);
3162 record_for_igvn(region);
3163 PhiNode* phi = new PhiNode(region, return_type);
3164
3165 // The mirror will never be null of Reflection.getClassAccessFlags, however
3166 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3167 // if it is. See bug 4774291.
3168
3169 // For Reflection.getClassAccessFlags(), the null check occurs in
3170 // the wrong place; see inline_unsafe_access(), above, for a similar
3171 // situation.
3172 mirror = null_check(mirror);
3173 // If mirror or obj is dead, only null-path is taken.
3174 if (stopped()) return true;
3175
3176 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3177
3178 // Now load the mirror's klass metaobject, and null-check it.
3179 // Side-effects region with the control path if the klass is null.
3180 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3181 // If kls is null, we have a primitive mirror.
3182 phi->init_req(_prim_path, prim_return_value);
3183 if (stopped()) { set_result(region, phi); return true; }
3184 bool safe_for_replace = (region->in(_prim_path) == top());
3185
3186 Node* p; // handy temp
3187 Node* null_ctl;
3188
3189 // Now that we have the non-null klass, we can perform the real query.
3190 // For constant classes, the query will constant-fold in LoadNode::Value.
3191 Node* query_value = top();
3192 switch (id) {
3193 case vmIntrinsics::_isInstance:
3194 // nothing is an instance of a primitive type
3195 query_value = gen_instanceof(obj, kls, safe_for_replace);
3196 break;
3197
3198 case vmIntrinsics::_getModifiers:
3199 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3200 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3201 break;
3202
3203 case vmIntrinsics::_isInterface:
3204 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3205 if (generate_interface_guard(kls, region) != NULL)
3206 // A guard was added. If the guard is taken, it was an interface.
3207 phi->add_req(intcon(1));
3208 // If we fall through, it's a plain class.
3209 query_value = intcon(0);
3210 break;
3211
3212 case vmIntrinsics::_isArray:
3213 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3214 if (generate_array_guard(kls, region) != NULL)
3215 // A guard was added. If the guard is taken, it was an array.
3216 phi->add_req(intcon(1));
3217 // If we fall through, it's a plain class.
3218 query_value = intcon(0);
3219 break;
3220
3221 case vmIntrinsics::_isPrimitive:
3222 query_value = intcon(0); // "normal" path produces false
3223 break;
3224
3225 case vmIntrinsics::_isHidden:
3226 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
3227 if (generate_hidden_class_guard(kls, region) != NULL)
3228 // A guard was added. If the guard is taken, it was an hidden class.
3229 phi->add_req(intcon(1));
3230 // If we fall through, it's a plain class.
3231 query_value = intcon(0);
3232 break;
3233
3234
3235 case vmIntrinsics::_getSuperclass:
3236 // The rules here are somewhat unfortunate, but we can still do better
3237 // with random logic than with a JNI call.
3238 // Interfaces store null or Object as _super, but must report null.
3239 // Arrays store an intermediate super as _super, but must report Object.
3240 // Other types can report the actual _super.
3241 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3242 if (generate_interface_guard(kls, region) != NULL)
3243 // A guard was added. If the guard is taken, it was an interface.
3244 phi->add_req(null());
3245 if (generate_array_guard(kls, region) != NULL)
3246 // A guard was added. If the guard is taken, it was an array.
3247 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3248 // If we fall through, it's a plain class. Get its _super.
3249 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3250 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3251 null_ctl = top();
3252 kls = null_check_oop(kls, &null_ctl);
3253 if (null_ctl != top()) {
3254 // If the guard is taken, Object.superClass is null (both klass and mirror).
3255 region->add_req(null_ctl);
3256 phi ->add_req(null());
3257 }
3258 if (!stopped()) {
3259 query_value = load_mirror_from_klass(kls);
3260 }
3261 break;
3262
3263 case vmIntrinsics::_getClassAccessFlags:
3264 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3265 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3266 break;
3267
3268 default:
3269 fatal_unexpected_iid(id);
3270 break;
3271 }
3272
3273 // Fall-through is the normal case of a query to a real class.
3274 phi->init_req(1, query_value);
3275 region->init_req(1, control());
3276
3277 C->set_has_split_ifs(true); // Has chance for split-if optimization
3278 set_result(region, phi);
3279 return true;
3280 }
3281
3282 //-------------------------inline_Class_cast-------------------
3283 bool LibraryCallKit::inline_Class_cast() {
3284 Node* mirror = argument(0); // Class
3285 Node* obj = argument(1);
3286 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3287 if (mirror_con == NULL) {
3288 return false; // dead path (mirror->is_top()).
3289 }
3290 if (obj == NULL || obj->is_top()) {
3291 return false; // dead path
3292 }
3293 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3294
3295 // First, see if Class.cast() can be folded statically.
3296 // java_mirror_type() returns non-null for compile-time Class constants.
3297 ciType* tm = mirror_con->java_mirror_type();
3298 if (tm != NULL && tm->is_klass() &&
3299 tp != NULL && tp->klass() != NULL) {
3300 if (!tp->klass()->is_loaded()) {
3301 // Don't use intrinsic when class is not loaded.
3302 return false;
3303 } else {
3304 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3305 if (static_res == Compile::SSC_always_true) {
3306 // isInstance() is true - fold the code.
3307 set_result(obj);
3308 return true;
3309 } else if (static_res == Compile::SSC_always_false) {
3310 // Don't use intrinsic, have to throw ClassCastException.
3311 // If the reference is null, the non-intrinsic bytecode will
3312 // be optimized appropriately.
3313 return false;
3314 }
3315 }
3316 }
3317
3318 // Bailout intrinsic and do normal inlining if exception path is frequent.
3319 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3320 return false;
3321 }
3322
3323 // Generate dynamic checks.
3324 // Class.cast() is java implementation of _checkcast bytecode.
3325 // Do checkcast (Parse::do_checkcast()) optimizations here.
3326
3327 mirror = null_check(mirror);
3328 // If mirror is dead, only null-path is taken.
3329 if (stopped()) {
3330 return true;
3331 }
3332
3333 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3334 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3335 RegionNode* region = new RegionNode(PATH_LIMIT);
3336 record_for_igvn(region);
3337
3338 // Now load the mirror's klass metaobject, and null-check it.
3339 // If kls is null, we have a primitive mirror and
3340 // nothing is an instance of a primitive type.
3341 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3342
3343 Node* res = top();
3344 if (!stopped()) {
3345 Node* bad_type_ctrl = top();
3346 // Do checkcast optimizations.
3347 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3348 region->init_req(_bad_type_path, bad_type_ctrl);
3349 }
3350 if (region->in(_prim_path) != top() ||
3351 region->in(_bad_type_path) != top()) {
3352 // Let Interpreter throw ClassCastException.
3353 PreserveJVMState pjvms(this);
3354 set_control(_gvn.transform(region));
3355 uncommon_trap(Deoptimization::Reason_intrinsic,
3356 Deoptimization::Action_maybe_recompile);
3357 }
3358 if (!stopped()) {
3359 set_result(res);
3360 }
3361 return true;
3362 }
3363
3364
3365 //--------------------------inline_native_subtype_check------------------------
3366 // This intrinsic takes the JNI calls out of the heart of
3367 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3368 bool LibraryCallKit::inline_native_subtype_check() {
3369 // Pull both arguments off the stack.
3370 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3371 args[0] = argument(0);
3372 args[1] = argument(1);
3373 Node* klasses[2]; // corresponding Klasses: superk, subk
3374 klasses[0] = klasses[1] = top();
3375
3376 enum {
3377 // A full decision tree on {superc is prim, subc is prim}:
3378 _prim_0_path = 1, // {P,N} => false
3379 // {P,P} & superc!=subc => false
3380 _prim_same_path, // {P,P} & superc==subc => true
3381 _prim_1_path, // {N,P} => false
3382 _ref_subtype_path, // {N,N} & subtype check wins => true
3383 _both_ref_path, // {N,N} & subtype check loses => false
3384 PATH_LIMIT
3385 };
3386
3387 RegionNode* region = new RegionNode(PATH_LIMIT);
3388 Node* phi = new PhiNode(region, TypeInt::BOOL);
3389 record_for_igvn(region);
3390
3391 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3392 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3393 int class_klass_offset = java_lang_Class::klass_offset();
3394
3395 // First null-check both mirrors and load each mirror's klass metaobject.
3396 int which_arg;
3397 for (which_arg = 0; which_arg <= 1; which_arg++) {
3398 Node* arg = args[which_arg];
3399 arg = null_check(arg);
3400 if (stopped()) break;
3401 args[which_arg] = arg;
3402
3403 Node* p = basic_plus_adr(arg, class_klass_offset);
3404 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3405 klasses[which_arg] = _gvn.transform(kls);
3406 }
3407
3408 // Having loaded both klasses, test each for null.
3409 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3410 for (which_arg = 0; which_arg <= 1; which_arg++) {
3411 Node* kls = klasses[which_arg];
3412 Node* null_ctl = top();
3413 kls = null_check_oop(kls, &null_ctl, never_see_null);
3414 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3415 region->init_req(prim_path, null_ctl);
3416 if (stopped()) break;
3417 klasses[which_arg] = kls;
3418 }
3419
3420 if (!stopped()) {
3421 // now we have two reference types, in klasses[0..1]
3422 Node* subk = klasses[1]; // the argument to isAssignableFrom
3423 Node* superk = klasses[0]; // the receiver
3424 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3425 // now we have a successful reference subtype check
3426 region->set_req(_ref_subtype_path, control());
3427 }
3428
3429 // If both operands are primitive (both klasses null), then
3430 // we must return true when they are identical primitives.
3431 // It is convenient to test this after the first null klass check.
3432 set_control(region->in(_prim_0_path)); // go back to first null check
3433 if (!stopped()) {
3434 // Since superc is primitive, make a guard for the superc==subc case.
3435 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3436 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3437 generate_guard(bol_eq, region, PROB_FAIR);
3438 if (region->req() == PATH_LIMIT+1) {
3439 // A guard was added. If the added guard is taken, superc==subc.
3440 region->swap_edges(PATH_LIMIT, _prim_same_path);
3441 region->del_req(PATH_LIMIT);
3442 }
3443 region->set_req(_prim_0_path, control()); // Not equal after all.
3444 }
3445
3446 // these are the only paths that produce 'true':
3447 phi->set_req(_prim_same_path, intcon(1));
3448 phi->set_req(_ref_subtype_path, intcon(1));
3449
3450 // pull together the cases:
3451 assert(region->req() == PATH_LIMIT, "sane region");
3452 for (uint i = 1; i < region->req(); i++) {
3453 Node* ctl = region->in(i);
3454 if (ctl == NULL || ctl == top()) {
3455 region->set_req(i, top());
3456 phi ->set_req(i, top());
3457 } else if (phi->in(i) == NULL) {
3458 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3459 }
3460 }
3461
3462 set_control(_gvn.transform(region));
3463 set_result(_gvn.transform(phi));
3464 return true;
3465 }
3466
3467 //---------------------generate_array_guard_common------------------------
3468 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3469 bool obj_array, bool not_array) {
3470
3471 if (stopped()) {
3472 return NULL;
3473 }
3474
3475 // If obj_array/non_array==false/false:
3476 // Branch around if the given klass is in fact an array (either obj or prim).
3477 // If obj_array/non_array==false/true:
3478 // Branch around if the given klass is not an array klass of any kind.
3479 // If obj_array/non_array==true/true:
3480 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3481 // If obj_array/non_array==true/false:
3482 // Branch around if the kls is an oop array (Object[] or subtype)
3483 //
3484 // Like generate_guard, adds a new path onto the region.
3485 jint layout_con = 0;
3486 Node* layout_val = get_layout_helper(kls, layout_con);
3487 if (layout_val == NULL) {
3488 bool query = (obj_array
3489 ? Klass::layout_helper_is_objArray(layout_con)
3490 : Klass::layout_helper_is_array(layout_con));
3491 if (query == not_array) {
3492 return NULL; // never a branch
3493 } else { // always a branch
3494 Node* always_branch = control();
3495 if (region != NULL)
3496 region->add_req(always_branch);
3497 set_control(top());
3498 return always_branch;
3499 }
3500 }
3501 // Now test the correct condition.
3502 jint nval = (obj_array
3503 ? (jint)(Klass::_lh_array_tag_type_value
3504 << Klass::_lh_array_tag_shift)
3505 : Klass::_lh_neutral_value);
3506 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3507 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3508 // invert the test if we are looking for a non-array
3509 if (not_array) btest = BoolTest(btest).negate();
3510 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3511 return generate_fair_guard(bol, region);
3512 }
3513
3514
3515 //-----------------------inline_native_newArray--------------------------
3516 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3517 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3518 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3519 Node* mirror;
3520 Node* count_val;
3521 if (uninitialized) {
3522 mirror = argument(1);
3523 count_val = argument(2);
3524 } else {
3525 mirror = argument(0);
3526 count_val = argument(1);
3527 }
3528
3529 mirror = null_check(mirror);
3530 // If mirror or obj is dead, only null-path is taken.
3531 if (stopped()) return true;
3532
3533 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3534 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3535 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3536 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3537 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3538
3539 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3540 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3541 result_reg, _slow_path);
3542 Node* normal_ctl = control();
3543 Node* no_array_ctl = result_reg->in(_slow_path);
3544
3545 // Generate code for the slow case. We make a call to newArray().
3546 set_control(no_array_ctl);
3547 if (!stopped()) {
3548 // Either the input type is void.class, or else the
3549 // array klass has not yet been cached. Either the
3550 // ensuing call will throw an exception, or else it
3551 // will cache the array klass for next time.
3552 PreserveJVMState pjvms(this);
3553 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3554 Node* slow_result = set_results_for_java_call(slow_call);
3555 // this->control() comes from set_results_for_java_call
3556 result_reg->set_req(_slow_path, control());
3557 result_val->set_req(_slow_path, slow_result);
3558 result_io ->set_req(_slow_path, i_o());
3559 result_mem->set_req(_slow_path, reset_memory());
3560 }
3561
3562 set_control(normal_ctl);
3563 if (!stopped()) {
3564 // Normal case: The array type has been cached in the java.lang.Class.
3565 // The following call works fine even if the array type is polymorphic.
3566 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3567 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3568 result_reg->init_req(_normal_path, control());
3569 result_val->init_req(_normal_path, obj);
3570 result_io ->init_req(_normal_path, i_o());
3571 result_mem->init_req(_normal_path, reset_memory());
3572
3573 if (uninitialized) {
3574 // Mark the allocation so that zeroing is skipped
3575 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3576 alloc->maybe_set_complete(&_gvn);
3577 }
3578 }
3579
3580 // Return the combined state.
3581 set_i_o( _gvn.transform(result_io) );
3582 set_all_memory( _gvn.transform(result_mem));
3583
3584 C->set_has_split_ifs(true); // Has chance for split-if optimization
3585 set_result(result_reg, result_val);
3586 return true;
3587 }
3588
3589 //----------------------inline_native_getLength--------------------------
3590 // public static native int java.lang.reflect.Array.getLength(Object array);
3591 bool LibraryCallKit::inline_native_getLength() {
3592 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3593
3594 Node* array = null_check(argument(0));
3595 // If array is dead, only null-path is taken.
3596 if (stopped()) return true;
3597
3598 // Deoptimize if it is a non-array.
3599 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3600
3601 if (non_array != NULL) {
3602 PreserveJVMState pjvms(this);
3603 set_control(non_array);
3604 uncommon_trap(Deoptimization::Reason_intrinsic,
3605 Deoptimization::Action_maybe_recompile);
3606 }
3607
3608 // If control is dead, only non-array-path is taken.
3609 if (stopped()) return true;
3610
3611 // The works fine even if the array type is polymorphic.
3612 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3613 Node* result = load_array_length(array);
3614
3615 C->set_has_split_ifs(true); // Has chance for split-if optimization
3616 set_result(result);
3617 return true;
3618 }
3619
3620 //------------------------inline_array_copyOf----------------------------
3621 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3622 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3623 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3624 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3625
3626 // Get the arguments.
3627 Node* original = argument(0);
3628 Node* start = is_copyOfRange? argument(1): intcon(0);
3629 Node* end = is_copyOfRange? argument(2): argument(1);
3630 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3631
3632 Node* newcopy = NULL;
3633
3634 // Set the original stack and the reexecute bit for the interpreter to reexecute
3635 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3636 { PreserveReexecuteState preexecs(this);
3637 jvms()->set_should_reexecute(true);
3638
3639 array_type_mirror = null_check(array_type_mirror);
3640 original = null_check(original);
3641
3642 // Check if a null path was taken unconditionally.
3643 if (stopped()) return true;
3644
3645 Node* orig_length = load_array_length(original);
3646
3647 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3648 klass_node = null_check(klass_node);
3649
3650 RegionNode* bailout = new RegionNode(1);
3651 record_for_igvn(bailout);
3652
3653 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3654 // Bail out if that is so.
3655 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3656 if (not_objArray != NULL) {
3657 // Improve the klass node's type from the new optimistic assumption:
3658 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3659 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3660 Node* cast = new CastPPNode(klass_node, akls);
3661 cast->init_req(0, control());
3662 klass_node = _gvn.transform(cast);
3663 }
3664
3665 // Bail out if either start or end is negative.
3666 generate_negative_guard(start, bailout, &start);
3667 generate_negative_guard(end, bailout, &end);
3668
3669 Node* length = end;
3670 if (_gvn.type(start) != TypeInt::ZERO) {
3671 length = _gvn.transform(new SubINode(end, start));
3672 }
3673
3674 // Bail out if length is negative.
3675 // Without this the new_array would throw
3676 // NegativeArraySizeException but IllegalArgumentException is what
3677 // should be thrown
3678 generate_negative_guard(length, bailout, &length);
3679
3680 if (bailout->req() > 1) {
3681 PreserveJVMState pjvms(this);
3682 set_control(_gvn.transform(bailout));
3683 uncommon_trap(Deoptimization::Reason_intrinsic,
3684 Deoptimization::Action_maybe_recompile);
3685 }
3686
3687 if (!stopped()) {
3688 // How many elements will we copy from the original?
3689 // The answer is MinI(orig_length - start, length).
3690 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3691 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3692
3693 // Generate a direct call to the right arraycopy function(s).
3694 // We know the copy is disjoint but we might not know if the
3695 // oop stores need checking.
3696 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3697 // This will fail a store-check if x contains any non-nulls.
3698
3699 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3700 // loads/stores but it is legal only if we're sure the
3701 // Arrays.copyOf would succeed. So we need all input arguments
3702 // to the copyOf to be validated, including that the copy to the
3703 // new array won't trigger an ArrayStoreException. That subtype
3704 // check can be optimized if we know something on the type of
3705 // the input array from type speculation.
3706 if (_gvn.type(klass_node)->singleton()) {
3707 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3708 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3709
3710 int test = C->static_subtype_check(superk, subk);
3711 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3712 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3713 if (t_original->speculative_type() != NULL) {
3714 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3715 }
3716 }
3717 }
3718
3719 bool validated = false;
3720 // Reason_class_check rather than Reason_intrinsic because we
3721 // want to intrinsify even if this traps.
3722 if (!too_many_traps(Deoptimization::Reason_class_check)) {
3723 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
3724
3725 if (not_subtype_ctrl != top()) {
3726 PreserveJVMState pjvms(this);
3727 set_control(not_subtype_ctrl);
3728 uncommon_trap(Deoptimization::Reason_class_check,
3729 Deoptimization::Action_make_not_entrant);
3730 assert(stopped(), "Should be stopped");
3731 }
3732 validated = true;
3733 }
3734
3735 if (!stopped()) {
3736 newcopy = new_array(klass_node, length, 0); // no arguments to push
3737
3738 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
3739 load_object_klass(original), klass_node);
3740 if (!is_copyOfRange) {
3741 ac->set_copyof(validated);
3742 } else {
3743 ac->set_copyofrange(validated);
3744 }
3745 Node* n = _gvn.transform(ac);
3746 if (n == ac) {
3747 ac->connect_outputs(this);
3748 } else {
3749 assert(validated, "shouldn't transform if all arguments not validated");
3750 set_all_memory(n);
3751 }
3752 }
3753 }
3754 } // original reexecute is set back here
3755
3756 C->set_has_split_ifs(true); // Has chance for split-if optimization
3757 if (!stopped()) {
3758 set_result(newcopy);
3759 }
3760 return true;
3761 }
3762
3763
3764 //----------------------generate_virtual_guard---------------------------
3765 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
3766 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
3767 RegionNode* slow_region) {
3768 ciMethod* method = callee();
3769 int vtable_index = method->vtable_index();
3770 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3771 "bad index %d", vtable_index);
3772 // Get the Method* out of the appropriate vtable entry.
3773 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
3774 vtable_index*vtableEntry::size_in_bytes() +
3775 vtableEntry::method_offset_in_bytes();
3776 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
3777 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3778
3779 // Compare the target method with the expected method (e.g., Object.hashCode).
3780 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
3781
3782 Node* native_call = makecon(native_call_addr);
3783 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
3784 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
3785
3786 return generate_slow_guard(test_native, slow_region);
3787 }
3788
3789 //-----------------------generate_method_call----------------------------
3790 // Use generate_method_call to make a slow-call to the real
3791 // method if the fast path fails. An alternative would be to
3792 // use a stub like OptoRuntime::slow_arraycopy_Java.
3793 // This only works for expanding the current library call,
3794 // not another intrinsic. (E.g., don't use this for making an
3795 // arraycopy call inside of the copyOf intrinsic.)
3796 CallJavaNode*
3797 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
3798 // When compiling the intrinsic method itself, do not use this technique.
3799 guarantee(callee() != C->method(), "cannot make slow-call to self");
3800
3801 ciMethod* method = callee();
3802 // ensure the JVMS we have will be correct for this call
3803 guarantee(method_id == method->intrinsic_id(), "must match");
3804
3805 const TypeFunc* tf = TypeFunc::make(method);
3806 CallJavaNode* slow_call;
3807 if (is_static) {
3808 assert(!is_virtual, "");
3809 slow_call = new CallStaticJavaNode(C, tf,
3810 SharedRuntime::get_resolve_static_call_stub(),
3811 method, bci());
3812 } else if (is_virtual) {
3813 null_check_receiver();
3814 int vtable_index = Method::invalid_vtable_index;
3815 if (UseInlineCaches) {
3816 // Suppress the vtable call
3817 } else {
3818 // hashCode and clone are not a miranda methods,
3819 // so the vtable index is fixed.
3820 // No need to use the linkResolver to get it.
3821 vtable_index = method->vtable_index();
3822 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3823 "bad index %d", vtable_index);
3824 }
3825 slow_call = new CallDynamicJavaNode(tf,
3826 SharedRuntime::get_resolve_virtual_call_stub(),
3827 method, vtable_index, bci());
3828 } else { // neither virtual nor static: opt_virtual
3829 null_check_receiver();
3830 slow_call = new CallStaticJavaNode(C, tf,
3831 SharedRuntime::get_resolve_opt_virtual_call_stub(),
3832 method, bci());
3833 slow_call->set_optimized_virtual(true);
3834 }
3835 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
3836 // To be able to issue a direct call (optimized virtual or virtual)
3837 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
3838 // about the method being invoked should be attached to the call site to
3839 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
3840 slow_call->set_override_symbolic_info(true);
3841 }
3842 set_arguments_for_java_call(slow_call);
3843 set_edges_for_java_call(slow_call);
3844 return slow_call;
3845 }
3846
3847
3848 /**
3849 * Build special case code for calls to hashCode on an object. This call may
3850 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
3851 * slightly different code.
3852 */
3853 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
3854 assert(is_static == callee()->is_static(), "correct intrinsic selection");
3855 assert(!(is_virtual && is_static), "either virtual, special, or static");
3856
3857 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
3858
3859 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3860 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
3861 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3862 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3863 Node* obj = NULL;
3864 if (!is_static) {
3865 // Check for hashing null object
3866 obj = null_check_receiver();
3867 if (stopped()) return true; // unconditionally null
3868 result_reg->init_req(_null_path, top());
3869 result_val->init_req(_null_path, top());
3870 } else {
3871 // Do a null check, and return zero if null.
3872 // System.identityHashCode(null) == 0
3873 obj = argument(0);
3874 Node* null_ctl = top();
3875 obj = null_check_oop(obj, &null_ctl);
3876 result_reg->init_req(_null_path, null_ctl);
3877 result_val->init_req(_null_path, _gvn.intcon(0));
3878 }
3879
3880 // Unconditionally null? Then return right away.
3881 if (stopped()) {
3882 set_control( result_reg->in(_null_path));
3883 if (!stopped())
3884 set_result(result_val->in(_null_path));
3885 return true;
3886 }
3887
3888 // We only go to the fast case code if we pass a number of guards. The
3889 // paths which do not pass are accumulated in the slow_region.
3890 RegionNode* slow_region = new RegionNode(1);
3891 record_for_igvn(slow_region);
3892
3893 // If this is a virtual call, we generate a funny guard. We pull out
3894 // the vtable entry corresponding to hashCode() from the target object.
3895 // If the target method which we are calling happens to be the native
3896 // Object hashCode() method, we pass the guard. We do not need this
3897 // guard for non-virtual calls -- the caller is known to be the native
3898 // Object hashCode().
3899 if (is_virtual) {
3900 // After null check, get the object's klass.
3901 Node* obj_klass = load_object_klass(obj);
3902 generate_virtual_guard(obj_klass, slow_region);
3903 }
3904
3905 // Get the header out of the object, use LoadMarkNode when available
3906 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
3907 // The control of the load must be NULL. Otherwise, the load can move before
3908 // the null check after castPP removal.
3909 Node* no_ctrl = NULL;
3910 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
3911
3912 // Test the header to see if it is unlocked.
3913 Node *lock_mask = _gvn.MakeConX(markWord::biased_lock_mask_in_place);
3914 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
3915 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
3916 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
3917 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
3918
3919 generate_slow_guard(test_unlocked, slow_region);
3920
3921 // Get the hash value and check to see that it has been properly assigned.
3922 // We depend on hash_mask being at most 32 bits and avoid the use of
3923 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
3924 // vm: see markWord.hpp.
3925 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
3926 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
3927 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
3928 // This hack lets the hash bits live anywhere in the mark object now, as long
3929 // as the shift drops the relevant bits into the low 32 bits. Note that
3930 // Java spec says that HashCode is an int so there's no point in capturing
3931 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
3932 hshifted_header = ConvX2I(hshifted_header);
3933 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
3934
3935 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
3936 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
3937 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
3938
3939 generate_slow_guard(test_assigned, slow_region);
3940
3941 Node* init_mem = reset_memory();
3942 // fill in the rest of the null path:
3943 result_io ->init_req(_null_path, i_o());
3944 result_mem->init_req(_null_path, init_mem);
3945
3946 result_val->init_req(_fast_path, hash_val);
3947 result_reg->init_req(_fast_path, control());
3948 result_io ->init_req(_fast_path, i_o());
3949 result_mem->init_req(_fast_path, init_mem);
3950
3951 // Generate code for the slow case. We make a call to hashCode().
3952 set_control(_gvn.transform(slow_region));
3953 if (!stopped()) {
3954 // No need for PreserveJVMState, because we're using up the present state.
3955 set_all_memory(init_mem);
3956 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
3957 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
3958 Node* slow_result = set_results_for_java_call(slow_call);
3959 // this->control() comes from set_results_for_java_call
3960 result_reg->init_req(_slow_path, control());
3961 result_val->init_req(_slow_path, slow_result);
3962 result_io ->set_req(_slow_path, i_o());
3963 result_mem ->set_req(_slow_path, reset_memory());
3964 }
3965
3966 // Return the combined state.
3967 set_i_o( _gvn.transform(result_io) );
3968 set_all_memory( _gvn.transform(result_mem));
3969
3970 set_result(result_reg, result_val);
3971 return true;
3972 }
3973
3974 //---------------------------inline_native_getClass----------------------------
3975 // public final native Class<?> java.lang.Object.getClass();
3976 //
3977 // Build special case code for calls to getClass on an object.
3978 bool LibraryCallKit::inline_native_getClass() {
3979 Node* obj = null_check_receiver();
3980 if (stopped()) return true;
3981 set_result(load_mirror_from_klass(load_object_klass(obj)));
3982 return true;
3983 }
3984
3985 //-----------------inline_native_Reflection_getCallerClass---------------------
3986 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
3987 //
3988 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
3989 //
3990 // NOTE: This code must perform the same logic as JVM_GetCallerClass
3991 // in that it must skip particular security frames and checks for
3992 // caller sensitive methods.
3993 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
3994 #ifndef PRODUCT
3995 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3996 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
3997 }
3998 #endif
3999
4000 if (!jvms()->has_method()) {
4001 #ifndef PRODUCT
4002 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4003 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4004 }
4005 #endif
4006 return false;
4007 }
4008
4009 // Walk back up the JVM state to find the caller at the required
4010 // depth.
4011 JVMState* caller_jvms = jvms();
4012
4013 // Cf. JVM_GetCallerClass
4014 // NOTE: Start the loop at depth 1 because the current JVM state does
4015 // not include the Reflection.getCallerClass() frame.
4016 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4017 ciMethod* m = caller_jvms->method();
4018 switch (n) {
4019 case 0:
4020 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4021 break;
4022 case 1:
4023 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4024 if (!m->caller_sensitive()) {
4025 #ifndef PRODUCT
4026 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4027 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4028 }
4029 #endif
4030 return false; // bail-out; let JVM_GetCallerClass do the work
4031 }
4032 break;
4033 default:
4034 if (!m->is_ignored_by_security_stack_walk()) {
4035 // We have reached the desired frame; return the holder class.
4036 // Acquire method holder as java.lang.Class and push as constant.
4037 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4038 ciInstance* caller_mirror = caller_klass->java_mirror();
4039 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4040
4041 #ifndef PRODUCT
4042 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4043 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());
4044 tty->print_cr(" JVM state at this point:");
4045 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4046 ciMethod* m = jvms()->of_depth(i)->method();
4047 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4048 }
4049 }
4050 #endif
4051 return true;
4052 }
4053 break;
4054 }
4055 }
4056
4057 #ifndef PRODUCT
4058 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4059 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4060 tty->print_cr(" JVM state at this point:");
4061 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4062 ciMethod* m = jvms()->of_depth(i)->method();
4063 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4064 }
4065 }
4066 #endif
4067
4068 return false; // bail-out; let JVM_GetCallerClass do the work
4069 }
4070
4071 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4072 Node* arg = argument(0);
4073 Node* result = NULL;
4074
4075 switch (id) {
4076 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4077 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4078 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4079 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4080
4081 case vmIntrinsics::_doubleToLongBits: {
4082 // two paths (plus control) merge in a wood
4083 RegionNode *r = new RegionNode(3);
4084 Node *phi = new PhiNode(r, TypeLong::LONG);
4085
4086 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4087 // Build the boolean node
4088 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4089
4090 // Branch either way.
4091 // NaN case is less traveled, which makes all the difference.
4092 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4093 Node *opt_isnan = _gvn.transform(ifisnan);
4094 assert( opt_isnan->is_If(), "Expect an IfNode");
4095 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4096 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4097
4098 set_control(iftrue);
4099
4100 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4101 Node *slow_result = longcon(nan_bits); // return NaN
4102 phi->init_req(1, _gvn.transform( slow_result ));
4103 r->init_req(1, iftrue);
4104
4105 // Else fall through
4106 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4107 set_control(iffalse);
4108
4109 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4110 r->init_req(2, iffalse);
4111
4112 // Post merge
4113 set_control(_gvn.transform(r));
4114 record_for_igvn(r);
4115
4116 C->set_has_split_ifs(true); // Has chance for split-if optimization
4117 result = phi;
4118 assert(result->bottom_type()->isa_long(), "must be");
4119 break;
4120 }
4121
4122 case vmIntrinsics::_floatToIntBits: {
4123 // two paths (plus control) merge in a wood
4124 RegionNode *r = new RegionNode(3);
4125 Node *phi = new PhiNode(r, TypeInt::INT);
4126
4127 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4128 // Build the boolean node
4129 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4130
4131 // Branch either way.
4132 // NaN case is less traveled, which makes all the difference.
4133 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4134 Node *opt_isnan = _gvn.transform(ifisnan);
4135 assert( opt_isnan->is_If(), "Expect an IfNode");
4136 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4137 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4138
4139 set_control(iftrue);
4140
4141 static const jint nan_bits = 0x7fc00000;
4142 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4143 phi->init_req(1, _gvn.transform( slow_result ));
4144 r->init_req(1, iftrue);
4145
4146 // Else fall through
4147 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4148 set_control(iffalse);
4149
4150 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4151 r->init_req(2, iffalse);
4152
4153 // Post merge
4154 set_control(_gvn.transform(r));
4155 record_for_igvn(r);
4156
4157 C->set_has_split_ifs(true); // Has chance for split-if optimization
4158 result = phi;
4159 assert(result->bottom_type()->isa_int(), "must be");
4160 break;
4161 }
4162
4163 default:
4164 fatal_unexpected_iid(id);
4165 break;
4166 }
4167 set_result(_gvn.transform(result));
4168 return true;
4169 }
4170
4171 //----------------------inline_unsafe_copyMemory-------------------------
4172 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4173 bool LibraryCallKit::inline_unsafe_copyMemory() {
4174 if (callee()->is_static()) return false; // caller must have the capability!
4175 null_check_receiver(); // null-check receiver
4176 if (stopped()) return true;
4177
4178 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4179
4180 Node* src_ptr = argument(1); // type: oop
4181 Node* src_off = ConvL2X(argument(2)); // type: long
4182 Node* dst_ptr = argument(4); // type: oop
4183 Node* dst_off = ConvL2X(argument(5)); // type: long
4184 Node* size = ConvL2X(argument(7)); // type: long
4185
4186 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4187 "fieldOffset must be byte-scaled");
4188
4189 Node* src = make_unsafe_address(src_ptr, src_off, ACCESS_READ);
4190 Node* dst = make_unsafe_address(dst_ptr, dst_off, ACCESS_WRITE);
4191
4192 // Conservatively insert a memory barrier on all memory slices.
4193 // Do not let writes of the copy source or destination float below the copy.
4194 insert_mem_bar(Op_MemBarCPUOrder);
4195
4196 Node* thread = _gvn.transform(new ThreadLocalNode());
4197 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4198 BasicType doing_unsafe_access_bt = T_BYTE;
4199 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4200
4201 // update volatile field
4202 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4203
4204 // Call it. Note that the length argument is not scaled.
4205 make_runtime_call(RC_LEAF|RC_NO_FP,
4206 OptoRuntime::fast_arraycopy_Type(),
4207 StubRoutines::unsafe_arraycopy(),
4208 "unsafe_arraycopy",
4209 TypeRawPtr::BOTTOM,
4210 src, dst, size XTOP);
4211
4212 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4213
4214 // Do not let reads of the copy destination float above the copy.
4215 insert_mem_bar(Op_MemBarCPUOrder);
4216
4217 return true;
4218 }
4219
4220 //------------------------clone_coping-----------------------------------
4221 // Helper function for inline_native_clone.
4222 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4223 assert(obj_size != NULL, "");
4224 Node* raw_obj = alloc_obj->in(1);
4225 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4226
4227 AllocateNode* alloc = NULL;
4228 if (ReduceBulkZeroing) {
4229 // We will be completely responsible for initializing this object -
4230 // mark Initialize node as complete.
4231 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4232 // The object was just allocated - there should be no any stores!
4233 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4234 // Mark as complete_with_arraycopy so that on AllocateNode
4235 // expansion, we know this AllocateNode is initialized by an array
4236 // copy and a StoreStore barrier exists after the array copy.
4237 alloc->initialization()->set_complete_with_arraycopy();
4238 }
4239
4240 Node* size = _gvn.transform(obj_size);
4241 access_clone(obj, alloc_obj, size, is_array);
4242
4243 // Do not let reads from the cloned object float above the arraycopy.
4244 if (alloc != NULL) {
4245 // Do not let stores that initialize this object be reordered with
4246 // a subsequent store that would make this object accessible by
4247 // other threads.
4248 // Record what AllocateNode this StoreStore protects so that
4249 // escape analysis can go from the MemBarStoreStoreNode to the
4250 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4251 // based on the escape status of the AllocateNode.
4252 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4253 } else {
4254 insert_mem_bar(Op_MemBarCPUOrder);
4255 }
4256 }
4257
4258 //------------------------inline_native_clone----------------------------
4259 // protected native Object java.lang.Object.clone();
4260 //
4261 // Here are the simple edge cases:
4262 // null receiver => normal trap
4263 // virtual and clone was overridden => slow path to out-of-line clone
4264 // not cloneable or finalizer => slow path to out-of-line Object.clone
4265 //
4266 // The general case has two steps, allocation and copying.
4267 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4268 //
4269 // Copying also has two cases, oop arrays and everything else.
4270 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4271 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4272 //
4273 // These steps fold up nicely if and when the cloned object's klass
4274 // can be sharply typed as an object array, a type array, or an instance.
4275 //
4276 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4277 PhiNode* result_val;
4278
4279 // Set the reexecute bit for the interpreter to reexecute
4280 // the bytecode that invokes Object.clone if deoptimization happens.
4281 { PreserveReexecuteState preexecs(this);
4282 jvms()->set_should_reexecute(true);
4283
4284 Node* obj = null_check_receiver();
4285 if (stopped()) return true;
4286
4287 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4288
4289 // If we are going to clone an instance, we need its exact type to
4290 // know the number and types of fields to convert the clone to
4291 // loads/stores. Maybe a speculative type can help us.
4292 if (!obj_type->klass_is_exact() &&
4293 obj_type->speculative_type() != NULL &&
4294 obj_type->speculative_type()->is_instance_klass()) {
4295 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4296 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4297 !spec_ik->has_injected_fields()) {
4298 ciKlass* k = obj_type->klass();
4299 if (!k->is_instance_klass() ||
4300 k->as_instance_klass()->is_interface() ||
4301 k->as_instance_klass()->has_subklass()) {
4302 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4303 }
4304 }
4305 }
4306
4307 // Conservatively insert a memory barrier on all memory slices.
4308 // Do not let writes into the original float below the clone.
4309 insert_mem_bar(Op_MemBarCPUOrder);
4310
4311 // paths into result_reg:
4312 enum {
4313 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4314 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4315 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4316 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4317 PATH_LIMIT
4318 };
4319 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4320 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4321 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4322 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4323 record_for_igvn(result_reg);
4324
4325 Node* obj_klass = load_object_klass(obj);
4326 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4327 if (array_ctl != NULL) {
4328 // It's an array.
4329 PreserveJVMState pjvms(this);
4330 set_control(array_ctl);
4331 Node* obj_length = load_array_length(obj);
4332 Node* obj_size = NULL;
4333 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, /*deoptimize_on_exception=*/true);
4334
4335 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4336 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, BarrierSetC2::Parsing)) {
4337 // If it is an oop array, it requires very special treatment,
4338 // because gc barriers are required when accessing the array.
4339 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4340 if (is_obja != NULL) {
4341 PreserveJVMState pjvms2(this);
4342 set_control(is_obja);
4343 // Generate a direct call to the right arraycopy function(s).
4344 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4345 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false);
4346 ac->set_clone_oop_array();
4347 Node* n = _gvn.transform(ac);
4348 assert(n == ac, "cannot disappear");
4349 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
4350
4351 result_reg->init_req(_objArray_path, control());
4352 result_val->init_req(_objArray_path, alloc_obj);
4353 result_i_o ->set_req(_objArray_path, i_o());
4354 result_mem ->set_req(_objArray_path, reset_memory());
4355 }
4356 }
4357 // Otherwise, there are no barriers to worry about.
4358 // (We can dispense with card marks if we know the allocation
4359 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4360 // causes the non-eden paths to take compensating steps to
4361 // simulate a fresh allocation, so that no further
4362 // card marks are required in compiled code to initialize
4363 // the object.)
4364
4365 if (!stopped()) {
4366 copy_to_clone(obj, alloc_obj, obj_size, true);
4367
4368 // Present the results of the copy.
4369 result_reg->init_req(_array_path, control());
4370 result_val->init_req(_array_path, alloc_obj);
4371 result_i_o ->set_req(_array_path, i_o());
4372 result_mem ->set_req(_array_path, reset_memory());
4373 }
4374 }
4375
4376 // We only go to the instance fast case code if we pass a number of guards.
4377 // The paths which do not pass are accumulated in the slow_region.
4378 RegionNode* slow_region = new RegionNode(1);
4379 record_for_igvn(slow_region);
4380 if (!stopped()) {
4381 // It's an instance (we did array above). Make the slow-path tests.
4382 // If this is a virtual call, we generate a funny guard. We grab
4383 // the vtable entry corresponding to clone() from the target object.
4384 // If the target method which we are calling happens to be the
4385 // Object clone() method, we pass the guard. We do not need this
4386 // guard for non-virtual calls; the caller is known to be the native
4387 // Object clone().
4388 if (is_virtual) {
4389 generate_virtual_guard(obj_klass, slow_region);
4390 }
4391
4392 // The object must be easily cloneable and must not have a finalizer.
4393 // Both of these conditions may be checked in a single test.
4394 // We could optimize the test further, but we don't care.
4395 generate_access_flags_guard(obj_klass,
4396 // Test both conditions:
4397 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4398 // Must be cloneable but not finalizer:
4399 JVM_ACC_IS_CLONEABLE_FAST,
4400 slow_region);
4401 }
4402
4403 if (!stopped()) {
4404 // It's an instance, and it passed the slow-path tests.
4405 PreserveJVMState pjvms(this);
4406 Node* obj_size = NULL;
4407 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4408 // is reexecuted if deoptimization occurs and there could be problems when merging
4409 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4410 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4411
4412 copy_to_clone(obj, alloc_obj, obj_size, false);
4413
4414 // Present the results of the slow call.
4415 result_reg->init_req(_instance_path, control());
4416 result_val->init_req(_instance_path, alloc_obj);
4417 result_i_o ->set_req(_instance_path, i_o());
4418 result_mem ->set_req(_instance_path, reset_memory());
4419 }
4420
4421 // Generate code for the slow case. We make a call to clone().
4422 set_control(_gvn.transform(slow_region));
4423 if (!stopped()) {
4424 PreserveJVMState pjvms(this);
4425 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4426 // We need to deoptimize on exception (see comment above)
4427 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4428 // this->control() comes from set_results_for_java_call
4429 result_reg->init_req(_slow_path, control());
4430 result_val->init_req(_slow_path, slow_result);
4431 result_i_o ->set_req(_slow_path, i_o());
4432 result_mem ->set_req(_slow_path, reset_memory());
4433 }
4434
4435 // Return the combined state.
4436 set_control( _gvn.transform(result_reg));
4437 set_i_o( _gvn.transform(result_i_o));
4438 set_all_memory( _gvn.transform(result_mem));
4439 } // original reexecute is set back here
4440
4441 set_result(_gvn.transform(result_val));
4442 return true;
4443 }
4444
4445 // If we have a tightly coupled allocation, the arraycopy may take care
4446 // of the array initialization. If one of the guards we insert between
4447 // the allocation and the arraycopy causes a deoptimization, an
4448 // unitialized array will escape the compiled method. To prevent that
4449 // we set the JVM state for uncommon traps between the allocation and
4450 // the arraycopy to the state before the allocation so, in case of
4451 // deoptimization, we'll reexecute the allocation and the
4452 // initialization.
4453 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4454 if (alloc != NULL) {
4455 ciMethod* trap_method = alloc->jvms()->method();
4456 int trap_bci = alloc->jvms()->bci();
4457
4458 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4459 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4460 // Make sure there's no store between the allocation and the
4461 // arraycopy otherwise visible side effects could be rexecuted
4462 // in case of deoptimization and cause incorrect execution.
4463 bool no_interfering_store = true;
4464 Node* mem = alloc->in(TypeFunc::Memory);
4465 if (mem->is_MergeMem()) {
4466 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4467 Node* n = mms.memory();
4468 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4469 assert(n->is_Store(), "what else?");
4470 no_interfering_store = false;
4471 break;
4472 }
4473 }
4474 } else {
4475 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4476 Node* n = mms.memory();
4477 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4478 assert(n->is_Store(), "what else?");
4479 no_interfering_store = false;
4480 break;
4481 }
4482 }
4483 }
4484
4485 if (no_interfering_store) {
4486 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4487 uint size = alloc->req();
4488 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4489 old_jvms->set_map(sfpt);
4490 for (uint i = 0; i < size; i++) {
4491 sfpt->init_req(i, alloc->in(i));
4492 }
4493 // re-push array length for deoptimization
4494 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4495 old_jvms->set_sp(old_jvms->sp()+1);
4496 old_jvms->set_monoff(old_jvms->monoff()+1);
4497 old_jvms->set_scloff(old_jvms->scloff()+1);
4498 old_jvms->set_endoff(old_jvms->endoff()+1);
4499 old_jvms->set_should_reexecute(true);
4500
4501 sfpt->set_i_o(map()->i_o());
4502 sfpt->set_memory(map()->memory());
4503 sfpt->set_control(map()->control());
4504
4505 JVMState* saved_jvms = jvms();
4506 saved_reexecute_sp = _reexecute_sp;
4507
4508 set_jvms(sfpt->jvms());
4509 _reexecute_sp = jvms()->sp();
4510
4511 return saved_jvms;
4512 }
4513 }
4514 }
4515 return NULL;
4516 }
4517
4518 // In case of a deoptimization, we restart execution at the
4519 // allocation, allocating a new array. We would leave an uninitialized
4520 // array in the heap that GCs wouldn't expect. Move the allocation
4521 // after the traps so we don't allocate the array if we
4522 // deoptimize. This is possible because tightly_coupled_allocation()
4523 // guarantees there's no observer of the allocated array at this point
4524 // and the control flow is simple enough.
4525 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4526 int saved_reexecute_sp, uint new_idx) {
4527 if (saved_jvms != NULL && !stopped()) {
4528 assert(alloc != NULL, "only with a tightly coupled allocation");
4529 // restore JVM state to the state at the arraycopy
4530 saved_jvms->map()->set_control(map()->control());
4531 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4532 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4533 // If we've improved the types of some nodes (null check) while
4534 // emitting the guards, propagate them to the current state
4535 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4536 set_jvms(saved_jvms);
4537 _reexecute_sp = saved_reexecute_sp;
4538
4539 // Remove the allocation from above the guards
4540 CallProjections callprojs;
4541 alloc->extract_projections(&callprojs, true);
4542 InitializeNode* init = alloc->initialization();
4543 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4544 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4545 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4546 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4547
4548 // move the allocation here (after the guards)
4549 _gvn.hash_delete(alloc);
4550 alloc->set_req(TypeFunc::Control, control());
4551 alloc->set_req(TypeFunc::I_O, i_o());
4552 Node *mem = reset_memory();
4553 set_all_memory(mem);
4554 alloc->set_req(TypeFunc::Memory, mem);
4555 set_control(init->proj_out_or_null(TypeFunc::Control));
4556 set_i_o(callprojs.fallthrough_ioproj);
4557
4558 // Update memory as done in GraphKit::set_output_for_allocation()
4559 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4560 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4561 if (ary_type->isa_aryptr() && length_type != NULL) {
4562 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4563 }
4564 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4565 int elemidx = C->get_alias_index(telemref);
4566 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
4567 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
4568
4569 Node* allocx = _gvn.transform(alloc);
4570 assert(allocx == alloc, "where has the allocation gone?");
4571 assert(dest->is_CheckCastPP(), "not an allocation result?");
4572
4573 _gvn.hash_delete(dest);
4574 dest->set_req(0, control());
4575 Node* destx = _gvn.transform(dest);
4576 assert(destx == dest, "where has the allocation result gone?");
4577 }
4578 }
4579
4580
4581 //------------------------------inline_arraycopy-----------------------
4582 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4583 // Object dest, int destPos,
4584 // int length);
4585 bool LibraryCallKit::inline_arraycopy() {
4586 // Get the arguments.
4587 Node* src = argument(0); // type: oop
4588 Node* src_offset = argument(1); // type: int
4589 Node* dest = argument(2); // type: oop
4590 Node* dest_offset = argument(3); // type: int
4591 Node* length = argument(4); // type: int
4592
4593 uint new_idx = C->unique();
4594
4595 // Check for allocation before we add nodes that would confuse
4596 // tightly_coupled_allocation()
4597 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4598
4599 int saved_reexecute_sp = -1;
4600 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4601 // See arraycopy_restore_alloc_state() comment
4602 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4603 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4604 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
4605 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4606
4607 // The following tests must be performed
4608 // (1) src and dest are arrays.
4609 // (2) src and dest arrays must have elements of the same BasicType
4610 // (3) src and dest must not be null.
4611 // (4) src_offset must not be negative.
4612 // (5) dest_offset must not be negative.
4613 // (6) length must not be negative.
4614 // (7) src_offset + length must not exceed length of src.
4615 // (8) dest_offset + length must not exceed length of dest.
4616 // (9) each element of an oop array must be assignable
4617
4618 // (3) src and dest must not be null.
4619 // always do this here because we need the JVM state for uncommon traps
4620 Node* null_ctl = top();
4621 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4622 assert(null_ctl->is_top(), "no null control here");
4623 dest = null_check(dest, T_ARRAY);
4624
4625 if (!can_emit_guards) {
4626 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4627 // guards but the arraycopy node could still take advantage of a
4628 // tightly allocated allocation. tightly_coupled_allocation() is
4629 // called again to make sure it takes the null check above into
4630 // account: the null check is mandatory and if it caused an
4631 // uncommon trap to be emitted then the allocation can't be
4632 // considered tightly coupled in this context.
4633 alloc = tightly_coupled_allocation(dest, NULL);
4634 }
4635
4636 bool validated = false;
4637
4638 const Type* src_type = _gvn.type(src);
4639 const Type* dest_type = _gvn.type(dest);
4640 const TypeAryPtr* top_src = src_type->isa_aryptr();
4641 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4642
4643 // Do we have the type of src?
4644 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4645 // Do we have the type of dest?
4646 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4647 // Is the type for src from speculation?
4648 bool src_spec = false;
4649 // Is the type for dest from speculation?
4650 bool dest_spec = false;
4651
4652 if ((!has_src || !has_dest) && can_emit_guards) {
4653 // We don't have sufficient type information, let's see if
4654 // speculative types can help. We need to have types for both src
4655 // and dest so that it pays off.
4656
4657 // Do we already have or could we have type information for src
4658 bool could_have_src = has_src;
4659 // Do we already have or could we have type information for dest
4660 bool could_have_dest = has_dest;
4661
4662 ciKlass* src_k = NULL;
4663 if (!has_src) {
4664 src_k = src_type->speculative_type_not_null();
4665 if (src_k != NULL && src_k->is_array_klass()) {
4666 could_have_src = true;
4667 }
4668 }
4669
4670 ciKlass* dest_k = NULL;
4671 if (!has_dest) {
4672 dest_k = dest_type->speculative_type_not_null();
4673 if (dest_k != NULL && dest_k->is_array_klass()) {
4674 could_have_dest = true;
4675 }
4676 }
4677
4678 if (could_have_src && could_have_dest) {
4679 // This is going to pay off so emit the required guards
4680 if (!has_src) {
4681 src = maybe_cast_profiled_obj(src, src_k, true);
4682 src_type = _gvn.type(src);
4683 top_src = src_type->isa_aryptr();
4684 has_src = (top_src != NULL && top_src->klass() != NULL);
4685 src_spec = true;
4686 }
4687 if (!has_dest) {
4688 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4689 dest_type = _gvn.type(dest);
4690 top_dest = dest_type->isa_aryptr();
4691 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4692 dest_spec = true;
4693 }
4694 }
4695 }
4696
4697 if (has_src && has_dest && can_emit_guards) {
4698 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
4699 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
4700 if (is_reference_type(src_elem)) src_elem = T_OBJECT;
4701 if (is_reference_type(dest_elem)) dest_elem = T_OBJECT;
4702
4703 if (src_elem == dest_elem && src_elem == T_OBJECT) {
4704 // If both arrays are object arrays then having the exact types
4705 // for both will remove the need for a subtype check at runtime
4706 // before the call and may make it possible to pick a faster copy
4707 // routine (without a subtype check on every element)
4708 // Do we have the exact type of src?
4709 bool could_have_src = src_spec;
4710 // Do we have the exact type of dest?
4711 bool could_have_dest = dest_spec;
4712 ciKlass* src_k = top_src->klass();
4713 ciKlass* dest_k = top_dest->klass();
4714 if (!src_spec) {
4715 src_k = src_type->speculative_type_not_null();
4716 if (src_k != NULL && src_k->is_array_klass()) {
4717 could_have_src = true;
4718 }
4719 }
4720 if (!dest_spec) {
4721 dest_k = dest_type->speculative_type_not_null();
4722 if (dest_k != NULL && dest_k->is_array_klass()) {
4723 could_have_dest = true;
4724 }
4725 }
4726 if (could_have_src && could_have_dest) {
4727 // If we can have both exact types, emit the missing guards
4728 if (could_have_src && !src_spec) {
4729 src = maybe_cast_profiled_obj(src, src_k, true);
4730 }
4731 if (could_have_dest && !dest_spec) {
4732 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4733 }
4734 }
4735 }
4736 }
4737
4738 ciMethod* trap_method = method();
4739 int trap_bci = bci();
4740 if (saved_jvms != NULL) {
4741 trap_method = alloc->jvms()->method();
4742 trap_bci = alloc->jvms()->bci();
4743 }
4744
4745 bool negative_length_guard_generated = false;
4746
4747 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4748 can_emit_guards &&
4749 !src->is_top() && !dest->is_top()) {
4750 // validate arguments: enables transformation the ArrayCopyNode
4751 validated = true;
4752
4753 RegionNode* slow_region = new RegionNode(1);
4754 record_for_igvn(slow_region);
4755
4756 // (1) src and dest are arrays.
4757 generate_non_array_guard(load_object_klass(src), slow_region);
4758 generate_non_array_guard(load_object_klass(dest), slow_region);
4759
4760 // (2) src and dest arrays must have elements of the same BasicType
4761 // done at macro expansion or at Ideal transformation time
4762
4763 // (4) src_offset must not be negative.
4764 generate_negative_guard(src_offset, slow_region);
4765
4766 // (5) dest_offset must not be negative.
4767 generate_negative_guard(dest_offset, slow_region);
4768
4769 // (7) src_offset + length must not exceed length of src.
4770 generate_limit_guard(src_offset, length,
4771 load_array_length(src),
4772 slow_region);
4773
4774 // (8) dest_offset + length must not exceed length of dest.
4775 generate_limit_guard(dest_offset, length,
4776 load_array_length(dest),
4777 slow_region);
4778
4779 // (6) length must not be negative.
4780 // This is also checked in generate_arraycopy() during macro expansion, but
4781 // we also have to check it here for the case where the ArrayCopyNode will
4782 // be eliminated by Escape Analysis.
4783 if (EliminateAllocations) {
4784 generate_negative_guard(length, slow_region);
4785 negative_length_guard_generated = true;
4786 }
4787
4788 // (9) each element of an oop array must be assignable
4789 Node* dest_klass = load_object_klass(dest);
4790 if (src != dest) {
4791 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
4792
4793 if (not_subtype_ctrl != top()) {
4794 PreserveJVMState pjvms(this);
4795 set_control(not_subtype_ctrl);
4796 uncommon_trap(Deoptimization::Reason_intrinsic,
4797 Deoptimization::Action_make_not_entrant);
4798 assert(stopped(), "Should be stopped");
4799 }
4800 }
4801 {
4802 PreserveJVMState pjvms(this);
4803 set_control(_gvn.transform(slow_region));
4804 uncommon_trap(Deoptimization::Reason_intrinsic,
4805 Deoptimization::Action_make_not_entrant);
4806 assert(stopped(), "Should be stopped");
4807 }
4808
4809 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
4810 const Type *toop = TypeOopPtr::make_from_klass(dest_klass_t->klass());
4811 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
4812 }
4813
4814 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
4815
4816 if (stopped()) {
4817 return true;
4818 }
4819
4820 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
4821 // Create LoadRange and LoadKlass nodes for use during macro expansion here
4822 // so the compiler has a chance to eliminate them: during macro expansion,
4823 // we have to set their control (CastPP nodes are eliminated).
4824 load_object_klass(src), load_object_klass(dest),
4825 load_array_length(src), load_array_length(dest));
4826
4827 ac->set_arraycopy(validated);
4828
4829 Node* n = _gvn.transform(ac);
4830 if (n == ac) {
4831 ac->connect_outputs(this);
4832 } else {
4833 assert(validated, "shouldn't transform if all arguments not validated");
4834 set_all_memory(n);
4835 }
4836 clear_upper_avx();
4837
4838
4839 return true;
4840 }
4841
4842
4843 // Helper function which determines if an arraycopy immediately follows
4844 // an allocation, with no intervening tests or other escapes for the object.
4845 AllocateArrayNode*
4846 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
4847 RegionNode* slow_region) {
4848 if (stopped()) return NULL; // no fast path
4849 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
4850
4851 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
4852 if (alloc == NULL) return NULL;
4853
4854 Node* rawmem = memory(Compile::AliasIdxRaw);
4855 // Is the allocation's memory state untouched?
4856 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
4857 // Bail out if there have been raw-memory effects since the allocation.
4858 // (Example: There might have been a call or safepoint.)
4859 return NULL;
4860 }
4861 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
4862 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
4863 return NULL;
4864 }
4865
4866 // There must be no unexpected observers of this allocation.
4867 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
4868 Node* obs = ptr->fast_out(i);
4869 if (obs != this->map()) {
4870 return NULL;
4871 }
4872 }
4873
4874 // This arraycopy must unconditionally follow the allocation of the ptr.
4875 Node* alloc_ctl = ptr->in(0);
4876 Node* ctl = control();
4877 while (ctl != alloc_ctl) {
4878 // There may be guards which feed into the slow_region.
4879 // Any other control flow means that we might not get a chance
4880 // to finish initializing the allocated object.
4881 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
4882 IfNode* iff = ctl->in(0)->as_If();
4883 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
4884 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
4885 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
4886 ctl = iff->in(0); // This test feeds the known slow_region.
4887 continue;
4888 }
4889 // One more try: Various low-level checks bottom out in
4890 // uncommon traps. If the debug-info of the trap omits
4891 // any reference to the allocation, as we've already
4892 // observed, then there can be no objection to the trap.
4893 bool found_trap = false;
4894 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
4895 Node* obs = not_ctl->fast_out(j);
4896 if (obs->in(0) == not_ctl && obs->is_Call() &&
4897 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
4898 found_trap = true; break;
4899 }
4900 }
4901 if (found_trap) {
4902 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
4903 continue;
4904 }
4905 }
4906 return NULL;
4907 }
4908
4909 // If we get this far, we have an allocation which immediately
4910 // precedes the arraycopy, and we can take over zeroing the new object.
4911 // The arraycopy will finish the initialization, and provide
4912 // a new control state to which we will anchor the destination pointer.
4913
4914 return alloc;
4915 }
4916
4917 //-------------inline_encodeISOArray-----------------------------------
4918 // encode char[] to byte[] in ISO_8859_1
4919 bool LibraryCallKit::inline_encodeISOArray() {
4920 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
4921 // no receiver since it is static method
4922 Node *src = argument(0);
4923 Node *src_offset = argument(1);
4924 Node *dst = argument(2);
4925 Node *dst_offset = argument(3);
4926 Node *length = argument(4);
4927
4928 src = must_be_not_null(src, true);
4929 dst = must_be_not_null(dst, true);
4930
4931 const Type* src_type = src->Value(&_gvn);
4932 const Type* dst_type = dst->Value(&_gvn);
4933 const TypeAryPtr* top_src = src_type->isa_aryptr();
4934 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
4935 if (top_src == NULL || top_src->klass() == NULL ||
4936 top_dest == NULL || top_dest->klass() == NULL) {
4937 // failed array check
4938 return false;
4939 }
4940
4941 // Figure out the size and type of the elements we will be copying.
4942 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4943 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4944 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
4945 return false;
4946 }
4947
4948 Node* src_start = array_element_address(src, src_offset, T_CHAR);
4949 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
4950 // 'src_start' points to src array + scaled offset
4951 // 'dst_start' points to dst array + scaled offset
4952
4953 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
4954 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
4955 enc = _gvn.transform(enc);
4956 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
4957 set_memory(res_mem, mtype);
4958 set_result(enc);
4959 clear_upper_avx();
4960
4961 return true;
4962 }
4963
4964 //-------------inline_multiplyToLen-----------------------------------
4965 bool LibraryCallKit::inline_multiplyToLen() {
4966 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
4967
4968 address stubAddr = StubRoutines::multiplyToLen();
4969 if (stubAddr == NULL) {
4970 return false; // Intrinsic's stub is not implemented on this platform
4971 }
4972 const char* stubName = "multiplyToLen";
4973
4974 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
4975
4976 // no receiver because it is a static method
4977 Node* x = argument(0);
4978 Node* xlen = argument(1);
4979 Node* y = argument(2);
4980 Node* ylen = argument(3);
4981 Node* z = argument(4);
4982
4983 x = must_be_not_null(x, true);
4984 y = must_be_not_null(y, true);
4985
4986 const Type* x_type = x->Value(&_gvn);
4987 const Type* y_type = y->Value(&_gvn);
4988 const TypeAryPtr* top_x = x_type->isa_aryptr();
4989 const TypeAryPtr* top_y = y_type->isa_aryptr();
4990 if (top_x == NULL || top_x->klass() == NULL ||
4991 top_y == NULL || top_y->klass() == NULL) {
4992 // failed array check
4993 return false;
4994 }
4995
4996 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4997 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4998 if (x_elem != T_INT || y_elem != T_INT) {
4999 return false;
5000 }
5001
5002 // Set the original stack and the reexecute bit for the interpreter to reexecute
5003 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5004 // on the return from z array allocation in runtime.
5005 { PreserveReexecuteState preexecs(this);
5006 jvms()->set_should_reexecute(true);
5007
5008 Node* x_start = array_element_address(x, intcon(0), x_elem);
5009 Node* y_start = array_element_address(y, intcon(0), y_elem);
5010 // 'x_start' points to x array + scaled xlen
5011 // 'y_start' points to y array + scaled ylen
5012
5013 // Allocate the result array
5014 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5015 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5016 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5017
5018 IdealKit ideal(this);
5019
5020 #define __ ideal.
5021 Node* one = __ ConI(1);
5022 Node* zero = __ ConI(0);
5023 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5024 __ set(need_alloc, zero);
5025 __ set(z_alloc, z);
5026 __ if_then(z, BoolTest::eq, null()); {
5027 __ increment (need_alloc, one);
5028 } __ else_(); {
5029 // Update graphKit memory and control from IdealKit.
5030 sync_kit(ideal);
5031 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5032 cast->init_req(0, control());
5033 _gvn.set_type(cast, cast->bottom_type());
5034 C->record_for_igvn(cast);
5035
5036 Node* zlen_arg = load_array_length(cast);
5037 // Update IdealKit memory and control from graphKit.
5038 __ sync_kit(this);
5039 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5040 __ increment (need_alloc, one);
5041 } __ end_if();
5042 } __ end_if();
5043
5044 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5045 // Update graphKit memory and control from IdealKit.
5046 sync_kit(ideal);
5047 Node * narr = new_array(klass_node, zlen, 1);
5048 // Update IdealKit memory and control from graphKit.
5049 __ sync_kit(this);
5050 __ set(z_alloc, narr);
5051 } __ end_if();
5052
5053 sync_kit(ideal);
5054 z = __ value(z_alloc);
5055 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5056 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5057 // Final sync IdealKit and GraphKit.
5058 final_sync(ideal);
5059 #undef __
5060
5061 Node* z_start = array_element_address(z, intcon(0), T_INT);
5062
5063 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5064 OptoRuntime::multiplyToLen_Type(),
5065 stubAddr, stubName, TypePtr::BOTTOM,
5066 x_start, xlen, y_start, ylen, z_start, zlen);
5067 } // original reexecute is set back here
5068
5069 C->set_has_split_ifs(true); // Has chance for split-if optimization
5070 set_result(z);
5071 return true;
5072 }
5073
5074 //-------------inline_squareToLen------------------------------------
5075 bool LibraryCallKit::inline_squareToLen() {
5076 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5077
5078 address stubAddr = StubRoutines::squareToLen();
5079 if (stubAddr == NULL) {
5080 return false; // Intrinsic's stub is not implemented on this platform
5081 }
5082 const char* stubName = "squareToLen";
5083
5084 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5085
5086 Node* x = argument(0);
5087 Node* len = argument(1);
5088 Node* z = argument(2);
5089 Node* zlen = argument(3);
5090
5091 x = must_be_not_null(x, true);
5092 z = must_be_not_null(z, true);
5093
5094 const Type* x_type = x->Value(&_gvn);
5095 const Type* z_type = z->Value(&_gvn);
5096 const TypeAryPtr* top_x = x_type->isa_aryptr();
5097 const TypeAryPtr* top_z = z_type->isa_aryptr();
5098 if (top_x == NULL || top_x->klass() == NULL ||
5099 top_z == NULL || top_z->klass() == NULL) {
5100 // failed array check
5101 return false;
5102 }
5103
5104 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5105 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5106 if (x_elem != T_INT || z_elem != T_INT) {
5107 return false;
5108 }
5109
5110
5111 Node* x_start = array_element_address(x, intcon(0), x_elem);
5112 Node* z_start = array_element_address(z, intcon(0), z_elem);
5113
5114 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5115 OptoRuntime::squareToLen_Type(),
5116 stubAddr, stubName, TypePtr::BOTTOM,
5117 x_start, len, z_start, zlen);
5118
5119 set_result(z);
5120 return true;
5121 }
5122
5123 //-------------inline_mulAdd------------------------------------------
5124 bool LibraryCallKit::inline_mulAdd() {
5125 assert(UseMulAddIntrinsic, "not implemented on this platform");
5126
5127 address stubAddr = StubRoutines::mulAdd();
5128 if (stubAddr == NULL) {
5129 return false; // Intrinsic's stub is not implemented on this platform
5130 }
5131 const char* stubName = "mulAdd";
5132
5133 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5134
5135 Node* out = argument(0);
5136 Node* in = argument(1);
5137 Node* offset = argument(2);
5138 Node* len = argument(3);
5139 Node* k = argument(4);
5140
5141 out = must_be_not_null(out, true);
5142
5143 const Type* out_type = out->Value(&_gvn);
5144 const Type* in_type = in->Value(&_gvn);
5145 const TypeAryPtr* top_out = out_type->isa_aryptr();
5146 const TypeAryPtr* top_in = in_type->isa_aryptr();
5147 if (top_out == NULL || top_out->klass() == NULL ||
5148 top_in == NULL || top_in->klass() == NULL) {
5149 // failed array check
5150 return false;
5151 }
5152
5153 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5154 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5155 if (out_elem != T_INT || in_elem != T_INT) {
5156 return false;
5157 }
5158
5159 Node* outlen = load_array_length(out);
5160 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5161 Node* out_start = array_element_address(out, intcon(0), out_elem);
5162 Node* in_start = array_element_address(in, intcon(0), in_elem);
5163
5164 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5165 OptoRuntime::mulAdd_Type(),
5166 stubAddr, stubName, TypePtr::BOTTOM,
5167 out_start,in_start, new_offset, len, k);
5168 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5169 set_result(result);
5170 return true;
5171 }
5172
5173 //-------------inline_montgomeryMultiply-----------------------------------
5174 bool LibraryCallKit::inline_montgomeryMultiply() {
5175 address stubAddr = StubRoutines::montgomeryMultiply();
5176 if (stubAddr == NULL) {
5177 return false; // Intrinsic's stub is not implemented on this platform
5178 }
5179
5180 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5181 const char* stubName = "montgomery_multiply";
5182
5183 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5184
5185 Node* a = argument(0);
5186 Node* b = argument(1);
5187 Node* n = argument(2);
5188 Node* len = argument(3);
5189 Node* inv = argument(4);
5190 Node* m = argument(6);
5191
5192 const Type* a_type = a->Value(&_gvn);
5193 const TypeAryPtr* top_a = a_type->isa_aryptr();
5194 const Type* b_type = b->Value(&_gvn);
5195 const TypeAryPtr* top_b = b_type->isa_aryptr();
5196 const Type* n_type = a->Value(&_gvn);
5197 const TypeAryPtr* top_n = n_type->isa_aryptr();
5198 const Type* m_type = a->Value(&_gvn);
5199 const TypeAryPtr* top_m = m_type->isa_aryptr();
5200 if (top_a == NULL || top_a->klass() == NULL ||
5201 top_b == NULL || top_b->klass() == NULL ||
5202 top_n == NULL || top_n->klass() == NULL ||
5203 top_m == NULL || top_m->klass() == NULL) {
5204 // failed array check
5205 return false;
5206 }
5207
5208 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5209 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5210 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5211 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5212 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5213 return false;
5214 }
5215
5216 // Make the call
5217 {
5218 Node* a_start = array_element_address(a, intcon(0), a_elem);
5219 Node* b_start = array_element_address(b, intcon(0), b_elem);
5220 Node* n_start = array_element_address(n, intcon(0), n_elem);
5221 Node* m_start = array_element_address(m, intcon(0), m_elem);
5222
5223 Node* call = make_runtime_call(RC_LEAF,
5224 OptoRuntime::montgomeryMultiply_Type(),
5225 stubAddr, stubName, TypePtr::BOTTOM,
5226 a_start, b_start, n_start, len, inv, top(),
5227 m_start);
5228 set_result(m);
5229 }
5230
5231 return true;
5232 }
5233
5234 bool LibraryCallKit::inline_montgomerySquare() {
5235 address stubAddr = StubRoutines::montgomerySquare();
5236 if (stubAddr == NULL) {
5237 return false; // Intrinsic's stub is not implemented on this platform
5238 }
5239
5240 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5241 const char* stubName = "montgomery_square";
5242
5243 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5244
5245 Node* a = argument(0);
5246 Node* n = argument(1);
5247 Node* len = argument(2);
5248 Node* inv = argument(3);
5249 Node* m = argument(5);
5250
5251 const Type* a_type = a->Value(&_gvn);
5252 const TypeAryPtr* top_a = a_type->isa_aryptr();
5253 const Type* n_type = a->Value(&_gvn);
5254 const TypeAryPtr* top_n = n_type->isa_aryptr();
5255 const Type* m_type = a->Value(&_gvn);
5256 const TypeAryPtr* top_m = m_type->isa_aryptr();
5257 if (top_a == NULL || top_a->klass() == NULL ||
5258 top_n == NULL || top_n->klass() == NULL ||
5259 top_m == NULL || top_m->klass() == NULL) {
5260 // failed array check
5261 return false;
5262 }
5263
5264 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5265 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5266 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5267 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5268 return false;
5269 }
5270
5271 // Make the call
5272 {
5273 Node* a_start = array_element_address(a, intcon(0), a_elem);
5274 Node* n_start = array_element_address(n, intcon(0), n_elem);
5275 Node* m_start = array_element_address(m, intcon(0), m_elem);
5276
5277 Node* call = make_runtime_call(RC_LEAF,
5278 OptoRuntime::montgomerySquare_Type(),
5279 stubAddr, stubName, TypePtr::BOTTOM,
5280 a_start, n_start, len, inv, top(),
5281 m_start);
5282 set_result(m);
5283 }
5284
5285 return true;
5286 }
5287
5288 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
5289 address stubAddr = NULL;
5290 const char* stubName = NULL;
5291
5292 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
5293 if (stubAddr == NULL) {
5294 return false; // Intrinsic's stub is not implemented on this platform
5295 }
5296
5297 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
5298
5299 assert(callee()->signature()->size() == 5, "expected 5 arguments");
5300
5301 Node* newArr = argument(0);
5302 Node* oldArr = argument(1);
5303 Node* newIdx = argument(2);
5304 Node* shiftCount = argument(3);
5305 Node* numIter = argument(4);
5306
5307 const Type* newArr_type = newArr->Value(&_gvn);
5308 const TypeAryPtr* top_newArr = newArr_type->isa_aryptr();
5309 const Type* oldArr_type = oldArr->Value(&_gvn);
5310 const TypeAryPtr* top_oldArr = oldArr_type->isa_aryptr();
5311 if (top_newArr == NULL || top_newArr->klass() == NULL || top_oldArr == NULL
5312 || top_oldArr->klass() == NULL) {
5313 return false;
5314 }
5315
5316 BasicType newArr_elem = newArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5317 BasicType oldArr_elem = oldArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5318 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
5319 return false;
5320 }
5321
5322 // Make the call
5323 {
5324 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
5325 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
5326
5327 Node* call = make_runtime_call(RC_LEAF,
5328 OptoRuntime::bigIntegerShift_Type(),
5329 stubAddr,
5330 stubName,
5331 TypePtr::BOTTOM,
5332 newArr_start,
5333 oldArr_start,
5334 newIdx,
5335 shiftCount,
5336 numIter);
5337 }
5338
5339 return true;
5340 }
5341
5342 //-------------inline_vectorizedMismatch------------------------------
5343 bool LibraryCallKit::inline_vectorizedMismatch() {
5344 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5345
5346 address stubAddr = StubRoutines::vectorizedMismatch();
5347 if (stubAddr == NULL) {
5348 return false; // Intrinsic's stub is not implemented on this platform
5349 }
5350 const char* stubName = "vectorizedMismatch";
5351 int size_l = callee()->signature()->size();
5352 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5353
5354 Node* obja = argument(0);
5355 Node* aoffset = argument(1);
5356 Node* objb = argument(3);
5357 Node* boffset = argument(4);
5358 Node* length = argument(6);
5359 Node* scale = argument(7);
5360
5361 const Type* a_type = obja->Value(&_gvn);
5362 const Type* b_type = objb->Value(&_gvn);
5363 const TypeAryPtr* top_a = a_type->isa_aryptr();
5364 const TypeAryPtr* top_b = b_type->isa_aryptr();
5365 if (top_a == NULL || top_a->klass() == NULL ||
5366 top_b == NULL || top_b->klass() == NULL) {
5367 // failed array check
5368 return false;
5369 }
5370
5371 Node* call;
5372 jvms()->set_should_reexecute(true);
5373
5374 Node* obja_adr = make_unsafe_address(obja, aoffset, ACCESS_READ);
5375 Node* objb_adr = make_unsafe_address(objb, boffset, ACCESS_READ);
5376
5377 call = make_runtime_call(RC_LEAF,
5378 OptoRuntime::vectorizedMismatch_Type(),
5379 stubAddr, stubName, TypePtr::BOTTOM,
5380 obja_adr, objb_adr, length, scale);
5381
5382 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5383 set_result(result);
5384 return true;
5385 }
5386
5387 /**
5388 * Calculate CRC32 for byte.
5389 * int java.util.zip.CRC32.update(int crc, int b)
5390 */
5391 bool LibraryCallKit::inline_updateCRC32() {
5392 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5393 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5394 // no receiver since it is static method
5395 Node* crc = argument(0); // type: int
5396 Node* b = argument(1); // type: int
5397
5398 /*
5399 * int c = ~ crc;
5400 * b = timesXtoThe32[(b ^ c) & 0xFF];
5401 * b = b ^ (c >>> 8);
5402 * crc = ~b;
5403 */
5404
5405 Node* M1 = intcon(-1);
5406 crc = _gvn.transform(new XorINode(crc, M1));
5407 Node* result = _gvn.transform(new XorINode(crc, b));
5408 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5409
5410 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5411 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5412 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5413 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5414
5415 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5416 result = _gvn.transform(new XorINode(crc, result));
5417 result = _gvn.transform(new XorINode(result, M1));
5418 set_result(result);
5419 return true;
5420 }
5421
5422 /**
5423 * Calculate CRC32 for byte[] array.
5424 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5425 */
5426 bool LibraryCallKit::inline_updateBytesCRC32() {
5427 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5428 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5429 // no receiver since it is static method
5430 Node* crc = argument(0); // type: int
5431 Node* src = argument(1); // type: oop
5432 Node* offset = argument(2); // type: int
5433 Node* length = argument(3); // type: int
5434
5435 const Type* src_type = src->Value(&_gvn);
5436 const TypeAryPtr* top_src = src_type->isa_aryptr();
5437 if (top_src == NULL || top_src->klass() == NULL) {
5438 // failed array check
5439 return false;
5440 }
5441
5442 // Figure out the size and type of the elements we will be copying.
5443 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5444 if (src_elem != T_BYTE) {
5445 return false;
5446 }
5447
5448 // 'src_start' points to src array + scaled offset
5449 src = must_be_not_null(src, true);
5450 Node* src_start = array_element_address(src, offset, src_elem);
5451
5452 // We assume that range check is done by caller.
5453 // TODO: generate range check (offset+length < src.length) in debug VM.
5454
5455 // Call the stub.
5456 address stubAddr = StubRoutines::updateBytesCRC32();
5457 const char *stubName = "updateBytesCRC32";
5458
5459 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5460 stubAddr, stubName, TypePtr::BOTTOM,
5461 crc, src_start, length);
5462 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5463 set_result(result);
5464 return true;
5465 }
5466
5467 /**
5468 * Calculate CRC32 for ByteBuffer.
5469 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5470 */
5471 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5472 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5473 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5474 // no receiver since it is static method
5475 Node* crc = argument(0); // type: int
5476 Node* src = argument(1); // type: long
5477 Node* offset = argument(3); // type: int
5478 Node* length = argument(4); // type: int
5479
5480 src = ConvL2X(src); // adjust Java long to machine word
5481 Node* base = _gvn.transform(new CastX2PNode(src));
5482 offset = ConvI2X(offset);
5483
5484 // 'src_start' points to src array + scaled offset
5485 Node* src_start = basic_plus_adr(top(), base, offset);
5486
5487 // Call the stub.
5488 address stubAddr = StubRoutines::updateBytesCRC32();
5489 const char *stubName = "updateBytesCRC32";
5490
5491 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5492 stubAddr, stubName, TypePtr::BOTTOM,
5493 crc, src_start, length);
5494 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5495 set_result(result);
5496 return true;
5497 }
5498
5499 //------------------------------get_table_from_crc32c_class-----------------------
5500 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5501 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5502 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5503
5504 return table;
5505 }
5506
5507 //------------------------------inline_updateBytesCRC32C-----------------------
5508 //
5509 // Calculate CRC32C for byte[] array.
5510 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5511 //
5512 bool LibraryCallKit::inline_updateBytesCRC32C() {
5513 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5514 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5515 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5516 // no receiver since it is a static method
5517 Node* crc = argument(0); // type: int
5518 Node* src = argument(1); // type: oop
5519 Node* offset = argument(2); // type: int
5520 Node* end = argument(3); // type: int
5521
5522 Node* length = _gvn.transform(new SubINode(end, offset));
5523
5524 const Type* src_type = src->Value(&_gvn);
5525 const TypeAryPtr* top_src = src_type->isa_aryptr();
5526 if (top_src == NULL || top_src->klass() == NULL) {
5527 // failed array check
5528 return false;
5529 }
5530
5531 // Figure out the size and type of the elements we will be copying.
5532 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5533 if (src_elem != T_BYTE) {
5534 return false;
5535 }
5536
5537 // 'src_start' points to src array + scaled offset
5538 src = must_be_not_null(src, true);
5539 Node* src_start = array_element_address(src, offset, src_elem);
5540
5541 // static final int[] byteTable in class CRC32C
5542 Node* table = get_table_from_crc32c_class(callee()->holder());
5543 table = must_be_not_null(table, true);
5544 Node* table_start = array_element_address(table, intcon(0), T_INT);
5545
5546 // We assume that range check is done by caller.
5547 // TODO: generate range check (offset+length < src.length) in debug VM.
5548
5549 // Call the stub.
5550 address stubAddr = StubRoutines::updateBytesCRC32C();
5551 const char *stubName = "updateBytesCRC32C";
5552
5553 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5554 stubAddr, stubName, TypePtr::BOTTOM,
5555 crc, src_start, length, table_start);
5556 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5557 set_result(result);
5558 return true;
5559 }
5560
5561 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5562 //
5563 // Calculate CRC32C for DirectByteBuffer.
5564 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5565 //
5566 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5567 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5568 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5569 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5570 // no receiver since it is a static method
5571 Node* crc = argument(0); // type: int
5572 Node* src = argument(1); // type: long
5573 Node* offset = argument(3); // type: int
5574 Node* end = argument(4); // type: int
5575
5576 Node* length = _gvn.transform(new SubINode(end, offset));
5577
5578 src = ConvL2X(src); // adjust Java long to machine word
5579 Node* base = _gvn.transform(new CastX2PNode(src));
5580 offset = ConvI2X(offset);
5581
5582 // 'src_start' points to src array + scaled offset
5583 Node* src_start = basic_plus_adr(top(), base, offset);
5584
5585 // static final int[] byteTable in class CRC32C
5586 Node* table = get_table_from_crc32c_class(callee()->holder());
5587 table = must_be_not_null(table, true);
5588 Node* table_start = array_element_address(table, intcon(0), T_INT);
5589
5590 // Call the stub.
5591 address stubAddr = StubRoutines::updateBytesCRC32C();
5592 const char *stubName = "updateBytesCRC32C";
5593
5594 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5595 stubAddr, stubName, TypePtr::BOTTOM,
5596 crc, src_start, length, table_start);
5597 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5598 set_result(result);
5599 return true;
5600 }
5601
5602 //------------------------------inline_updateBytesAdler32----------------------
5603 //
5604 // Calculate Adler32 checksum for byte[] array.
5605 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5606 //
5607 bool LibraryCallKit::inline_updateBytesAdler32() {
5608 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5609 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5610 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5611 // no receiver since it is static method
5612 Node* crc = argument(0); // type: int
5613 Node* src = argument(1); // type: oop
5614 Node* offset = argument(2); // type: int
5615 Node* length = argument(3); // type: int
5616
5617 const Type* src_type = src->Value(&_gvn);
5618 const TypeAryPtr* top_src = src_type->isa_aryptr();
5619 if (top_src == NULL || top_src->klass() == NULL) {
5620 // failed array check
5621 return false;
5622 }
5623
5624 // Figure out the size and type of the elements we will be copying.
5625 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5626 if (src_elem != T_BYTE) {
5627 return false;
5628 }
5629
5630 // 'src_start' points to src array + scaled offset
5631 Node* src_start = array_element_address(src, offset, src_elem);
5632
5633 // We assume that range check is done by caller.
5634 // TODO: generate range check (offset+length < src.length) in debug VM.
5635
5636 // Call the stub.
5637 address stubAddr = StubRoutines::updateBytesAdler32();
5638 const char *stubName = "updateBytesAdler32";
5639
5640 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5641 stubAddr, stubName, TypePtr::BOTTOM,
5642 crc, src_start, length);
5643 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5644 set_result(result);
5645 return true;
5646 }
5647
5648 //------------------------------inline_updateByteBufferAdler32---------------
5649 //
5650 // Calculate Adler32 checksum for DirectByteBuffer.
5651 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5652 //
5653 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5654 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5655 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5656 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5657 // no receiver since it is static method
5658 Node* crc = argument(0); // type: int
5659 Node* src = argument(1); // type: long
5660 Node* offset = argument(3); // type: int
5661 Node* length = argument(4); // type: int
5662
5663 src = ConvL2X(src); // adjust Java long to machine word
5664 Node* base = _gvn.transform(new CastX2PNode(src));
5665 offset = ConvI2X(offset);
5666
5667 // 'src_start' points to src array + scaled offset
5668 Node* src_start = basic_plus_adr(top(), base, offset);
5669
5670 // Call the stub.
5671 address stubAddr = StubRoutines::updateBytesAdler32();
5672 const char *stubName = "updateBytesAdler32";
5673
5674 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5675 stubAddr, stubName, TypePtr::BOTTOM,
5676 crc, src_start, length);
5677
5678 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5679 set_result(result);
5680 return true;
5681 }
5682
5683 //----------------------------inline_reference_get----------------------------
5684 // public T java.lang.ref.Reference.get();
5685 bool LibraryCallKit::inline_reference_get() {
5686 const int referent_offset = java_lang_ref_Reference::referent_offset();
5687
5688 // Get the argument:
5689 Node* reference_obj = null_check_receiver();
5690 if (stopped()) return true;
5691
5692 const TypeInstPtr* tinst = _gvn.type(reference_obj)->isa_instptr();
5693 assert(tinst != NULL, "obj is null");
5694 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5695 ciInstanceKlass* referenceKlass = tinst->klass()->as_instance_klass();
5696 ciField* field = referenceKlass->get_field_by_name(ciSymbol::make("referent"),
5697 ciSymbol::make("Ljava/lang/Object;"),
5698 false);
5699 assert (field != NULL, "undefined field");
5700
5701 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5702 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5703
5704 ciInstanceKlass* klass = env()->Object_klass();
5705 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5706
5707 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
5708 Node* result = access_load_at(reference_obj, adr, adr_type, object_type, T_OBJECT, decorators);
5709 // Add memory barrier to prevent commoning reads from this field
5710 // across safepoint since GC can change its value.
5711 insert_mem_bar(Op_MemBarCPUOrder);
5712
5713 set_result(result);
5714 return true;
5715 }
5716
5717
5718 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5719 bool is_exact=true, bool is_static=false,
5720 ciInstanceKlass * fromKls=NULL) {
5721 if (fromKls == NULL) {
5722 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5723 assert(tinst != NULL, "obj is null");
5724 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5725 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5726 fromKls = tinst->klass()->as_instance_klass();
5727 } else {
5728 assert(is_static, "only for static field access");
5729 }
5730 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5731 ciSymbol::make(fieldTypeString),
5732 is_static);
5733
5734 assert (field != NULL, "undefined field");
5735 if (field == NULL) return (Node *) NULL;
5736
5737 if (is_static) {
5738 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5739 fromObj = makecon(tip);
5740 }
5741
5742 // Next code copied from Parse::do_get_xxx():
5743
5744 // Compute address and memory type.
5745 int offset = field->offset_in_bytes();
5746 bool is_vol = field->is_volatile();
5747 ciType* field_klass = field->type();
5748 assert(field_klass->is_loaded(), "should be loaded");
5749 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5750 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5751 BasicType bt = field->layout_type();
5752
5753 // Build the resultant type of the load
5754 const Type *type;
5755 if (bt == T_OBJECT) {
5756 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5757 } else {
5758 type = Type::get_const_basic_type(bt);
5759 }
5760
5761 DecoratorSet decorators = IN_HEAP;
5762
5763 if (is_vol) {
5764 decorators |= MO_SEQ_CST;
5765 }
5766
5767 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
5768 }
5769
5770 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5771 bool is_exact = true, bool is_static = false,
5772 ciInstanceKlass * fromKls = NULL) {
5773 if (fromKls == NULL) {
5774 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5775 assert(tinst != NULL, "obj is null");
5776 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5777 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5778 fromKls = tinst->klass()->as_instance_klass();
5779 }
5780 else {
5781 assert(is_static, "only for static field access");
5782 }
5783 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5784 ciSymbol::make(fieldTypeString),
5785 is_static);
5786
5787 assert(field != NULL, "undefined field");
5788 assert(!field->is_volatile(), "not defined for volatile fields");
5789
5790 if (is_static) {
5791 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5792 fromObj = makecon(tip);
5793 }
5794
5795 // Next code copied from Parse::do_get_xxx():
5796
5797 // Compute address and memory type.
5798 int offset = field->offset_in_bytes();
5799 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5800
5801 return adr;
5802 }
5803
5804 //------------------------------inline_aescrypt_Block-----------------------
5805 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
5806 address stubAddr = NULL;
5807 const char *stubName;
5808 assert(UseAES, "need AES instruction support");
5809
5810 switch(id) {
5811 case vmIntrinsics::_aescrypt_encryptBlock:
5812 stubAddr = StubRoutines::aescrypt_encryptBlock();
5813 stubName = "aescrypt_encryptBlock";
5814 break;
5815 case vmIntrinsics::_aescrypt_decryptBlock:
5816 stubAddr = StubRoutines::aescrypt_decryptBlock();
5817 stubName = "aescrypt_decryptBlock";
5818 break;
5819 default:
5820 break;
5821 }
5822 if (stubAddr == NULL) return false;
5823
5824 Node* aescrypt_object = argument(0);
5825 Node* src = argument(1);
5826 Node* src_offset = argument(2);
5827 Node* dest = argument(3);
5828 Node* dest_offset = argument(4);
5829
5830 src = must_be_not_null(src, true);
5831 dest = must_be_not_null(dest, true);
5832
5833 // (1) src and dest are arrays.
5834 const Type* src_type = src->Value(&_gvn);
5835 const Type* dest_type = dest->Value(&_gvn);
5836 const TypeAryPtr* top_src = src_type->isa_aryptr();
5837 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5838 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5839
5840 // for the quick and dirty code we will skip all the checks.
5841 // we are just trying to get the call to be generated.
5842 Node* src_start = src;
5843 Node* dest_start = dest;
5844 if (src_offset != NULL || dest_offset != NULL) {
5845 assert(src_offset != NULL && dest_offset != NULL, "");
5846 src_start = array_element_address(src, src_offset, T_BYTE);
5847 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5848 }
5849
5850 // now need to get the start of its expanded key array
5851 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5852 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5853 if (k_start == NULL) return false;
5854
5855 if (Matcher::pass_original_key_for_aes()) {
5856 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5857 // compatibility issues between Java key expansion and SPARC crypto instructions
5858 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5859 if (original_k_start == NULL) return false;
5860
5861 // Call the stub.
5862 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5863 stubAddr, stubName, TypePtr::BOTTOM,
5864 src_start, dest_start, k_start, original_k_start);
5865 } else {
5866 // Call the stub.
5867 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5868 stubAddr, stubName, TypePtr::BOTTOM,
5869 src_start, dest_start, k_start);
5870 }
5871
5872 return true;
5873 }
5874
5875 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
5876 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
5877 address stubAddr = NULL;
5878 const char *stubName = NULL;
5879
5880 assert(UseAES, "need AES instruction support");
5881
5882 switch(id) {
5883 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
5884 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
5885 stubName = "cipherBlockChaining_encryptAESCrypt";
5886 break;
5887 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
5888 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
5889 stubName = "cipherBlockChaining_decryptAESCrypt";
5890 break;
5891 default:
5892 break;
5893 }
5894 if (stubAddr == NULL) return false;
5895
5896 Node* cipherBlockChaining_object = argument(0);
5897 Node* src = argument(1);
5898 Node* src_offset = argument(2);
5899 Node* len = argument(3);
5900 Node* dest = argument(4);
5901 Node* dest_offset = argument(5);
5902
5903 src = must_be_not_null(src, false);
5904 dest = must_be_not_null(dest, false);
5905
5906 // (1) src and dest are arrays.
5907 const Type* src_type = src->Value(&_gvn);
5908 const Type* dest_type = dest->Value(&_gvn);
5909 const TypeAryPtr* top_src = src_type->isa_aryptr();
5910 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5911 assert (top_src != NULL && top_src->klass() != NULL
5912 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5913
5914 // checks are the responsibility of the caller
5915 Node* src_start = src;
5916 Node* dest_start = dest;
5917 if (src_offset != NULL || dest_offset != NULL) {
5918 assert(src_offset != NULL && dest_offset != NULL, "");
5919 src_start = array_element_address(src, src_offset, T_BYTE);
5920 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5921 }
5922
5923 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
5924 // (because of the predicated logic executed earlier).
5925 // so we cast it here safely.
5926 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5927
5928 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
5929 if (embeddedCipherObj == NULL) return false;
5930
5931 // cast it to what we know it will be at runtime
5932 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
5933 assert(tinst != NULL, "CBC obj is null");
5934 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
5935 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5936 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
5937
5938 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
5939 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
5940 const TypeOopPtr* xtype = aklass->as_instance_type();
5941 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
5942 aescrypt_object = _gvn.transform(aescrypt_object);
5943
5944 // we need to get the start of the aescrypt_object's expanded key array
5945 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5946 if (k_start == NULL) return false;
5947
5948 // similarly, get the start address of the r vector
5949 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
5950 if (objRvec == NULL) return false;
5951 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
5952
5953 Node* cbcCrypt;
5954 if (Matcher::pass_original_key_for_aes()) {
5955 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5956 // compatibility issues between Java key expansion and SPARC crypto instructions
5957 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5958 if (original_k_start == NULL) return false;
5959
5960 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
5961 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5962 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5963 stubAddr, stubName, TypePtr::BOTTOM,
5964 src_start, dest_start, k_start, r_start, len, original_k_start);
5965 } else {
5966 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
5967 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5968 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5969 stubAddr, stubName, TypePtr::BOTTOM,
5970 src_start, dest_start, k_start, r_start, len);
5971 }
5972
5973 // return cipher length (int)
5974 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
5975 set_result(retvalue);
5976 return true;
5977 }
5978
5979 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
5980 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
5981 address stubAddr = NULL;
5982 const char *stubName = NULL;
5983
5984 assert(UseAES, "need AES instruction support");
5985
5986 switch (id) {
5987 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
5988 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
5989 stubName = "electronicCodeBook_encryptAESCrypt";
5990 break;
5991 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
5992 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
5993 stubName = "electronicCodeBook_decryptAESCrypt";
5994 break;
5995 default:
5996 break;
5997 }
5998
5999 if (stubAddr == NULL) return false;
6000
6001 Node* electronicCodeBook_object = argument(0);
6002 Node* src = argument(1);
6003 Node* src_offset = argument(2);
6004 Node* len = argument(3);
6005 Node* dest = argument(4);
6006 Node* dest_offset = argument(5);
6007
6008 // (1) src and dest are arrays.
6009 const Type* src_type = src->Value(&_gvn);
6010 const Type* dest_type = dest->Value(&_gvn);
6011 const TypeAryPtr* top_src = src_type->isa_aryptr();
6012 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6013 assert(top_src != NULL && top_src->klass() != NULL
6014 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6015
6016 // checks are the responsibility of the caller
6017 Node* src_start = src;
6018 Node* dest_start = dest;
6019 if (src_offset != NULL || dest_offset != NULL) {
6020 assert(src_offset != NULL && dest_offset != NULL, "");
6021 src_start = array_element_address(src, src_offset, T_BYTE);
6022 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6023 }
6024
6025 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6026 // (because of the predicated logic executed earlier).
6027 // so we cast it here safely.
6028 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6029
6030 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6031 if (embeddedCipherObj == NULL) return false;
6032
6033 // cast it to what we know it will be at runtime
6034 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6035 assert(tinst != NULL, "ECB obj is null");
6036 assert(tinst->klass()->is_loaded(), "ECB obj is not loaded");
6037 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6038 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6039
6040 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6041 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6042 const TypeOopPtr* xtype = aklass->as_instance_type();
6043 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6044 aescrypt_object = _gvn.transform(aescrypt_object);
6045
6046 // we need to get the start of the aescrypt_object's expanded key array
6047 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6048 if (k_start == NULL) return false;
6049
6050 Node* ecbCrypt;
6051 if (Matcher::pass_original_key_for_aes()) {
6052 // no SPARC version for AES/ECB intrinsics now.
6053 return false;
6054 }
6055 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6056 ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6057 OptoRuntime::electronicCodeBook_aescrypt_Type(),
6058 stubAddr, stubName, TypePtr::BOTTOM,
6059 src_start, dest_start, k_start, len);
6060
6061 // return cipher length (int)
6062 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6063 set_result(retvalue);
6064 return true;
6065 }
6066
6067 //------------------------------inline_counterMode_AESCrypt-----------------------
6068 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6069 assert(UseAES, "need AES instruction support");
6070 if (!UseAESCTRIntrinsics) return false;
6071
6072 address stubAddr = NULL;
6073 const char *stubName = NULL;
6074 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6075 stubAddr = StubRoutines::counterMode_AESCrypt();
6076 stubName = "counterMode_AESCrypt";
6077 }
6078 if (stubAddr == NULL) return false;
6079
6080 Node* counterMode_object = argument(0);
6081 Node* src = argument(1);
6082 Node* src_offset = argument(2);
6083 Node* len = argument(3);
6084 Node* dest = argument(4);
6085 Node* dest_offset = argument(5);
6086
6087 // (1) src and dest are arrays.
6088 const Type* src_type = src->Value(&_gvn);
6089 const Type* dest_type = dest->Value(&_gvn);
6090 const TypeAryPtr* top_src = src_type->isa_aryptr();
6091 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6092 assert(top_src != NULL && top_src->klass() != NULL &&
6093 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6094
6095 // checks are the responsibility of the caller
6096 Node* src_start = src;
6097 Node* dest_start = dest;
6098 if (src_offset != NULL || dest_offset != NULL) {
6099 assert(src_offset != NULL && dest_offset != NULL, "");
6100 src_start = array_element_address(src, src_offset, T_BYTE);
6101 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6102 }
6103
6104 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6105 // (because of the predicated logic executed earlier).
6106 // so we cast it here safely.
6107 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6108 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6109 if (embeddedCipherObj == NULL) return false;
6110 // cast it to what we know it will be at runtime
6111 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6112 assert(tinst != NULL, "CTR obj is null");
6113 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6114 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6115 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6116 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6117 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6118 const TypeOopPtr* xtype = aklass->as_instance_type();
6119 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6120 aescrypt_object = _gvn.transform(aescrypt_object);
6121 // we need to get the start of the aescrypt_object's expanded key array
6122 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6123 if (k_start == NULL) return false;
6124 // similarly, get the start address of the r vector
6125 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6126 if (obj_counter == NULL) return false;
6127 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6128
6129 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6130 if (saved_encCounter == NULL) return false;
6131 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6132 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6133
6134 Node* ctrCrypt;
6135 if (Matcher::pass_original_key_for_aes()) {
6136 // no SPARC version for AES/CTR intrinsics now.
6137 return false;
6138 }
6139 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6140 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6141 OptoRuntime::counterMode_aescrypt_Type(),
6142 stubAddr, stubName, TypePtr::BOTTOM,
6143 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6144
6145 // return cipher length (int)
6146 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6147 set_result(retvalue);
6148 return true;
6149 }
6150
6151 //------------------------------get_key_start_from_aescrypt_object-----------------------
6152 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6153 #if defined(PPC64) || defined(S390)
6154 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6155 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6156 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6157 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6158 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6159 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6160 if (objSessionK == NULL) {
6161 return (Node *) NULL;
6162 }
6163 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6164 #else
6165 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6166 #endif // PPC64
6167 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6168 if (objAESCryptKey == NULL) return (Node *) NULL;
6169
6170 // now have the array, need to get the start address of the K array
6171 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6172 return k_start;
6173 }
6174
6175 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6176 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6177 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6178 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6179 if (objAESCryptKey == NULL) return (Node *) NULL;
6180
6181 // now have the array, need to get the start address of the lastKey array
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 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6226
6227 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6228 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6229 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6230
6231 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6232
6233 // for encryption, we are done
6234 if (!decrypting)
6235 return instof_false; // even if it is NULL
6236
6237 // for decryption, we need to add a further check to avoid
6238 // taking the intrinsic path when cipher and plain are the same
6239 // see the original java code for why.
6240 RegionNode* region = new RegionNode(3);
6241 region->init_req(1, instof_false);
6242
6243 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6244 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6245 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6246 region->init_req(2, src_dest_conjoint);
6247
6248 record_for_igvn(region);
6249 return _gvn.transform(region);
6250 }
6251
6252 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
6253 // Return node representing slow path of predicate check.
6254 // the pseudo code we want to emulate with this predicate is:
6255 // for encryption:
6256 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6257 // for decryption:
6258 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6259 // note cipher==plain is more conservative than the original java code but that's OK
6260 //
6261 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
6262 // The receiver was checked for NULL already.
6263 Node* objECB = argument(0);
6264
6265 // Load embeddedCipher field of ElectronicCodeBook object.
6266 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6267
6268 // get AESCrypt klass for instanceOf check
6269 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6270 // will have same classloader as ElectronicCodeBook object
6271 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
6272 assert(tinst != NULL, "ECBobj is null");
6273 assert(tinst->klass()->is_loaded(), "ECBobj is not loaded");
6274
6275 // we want to do an instanceof comparison against the AESCrypt class
6276 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6277 if (!klass_AESCrypt->is_loaded()) {
6278 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6279 Node* ctrl = control();
6280 set_control(top()); // no regular fast path
6281 return ctrl;
6282 }
6283 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6284
6285 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6286 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6287 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6288
6289 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6290
6291 // for encryption, we are done
6292 if (!decrypting)
6293 return instof_false; // even if it is NULL
6294
6295 // for decryption, we need to add a further check to avoid
6296 // taking the intrinsic path when cipher and plain are the same
6297 // see the original java code for why.
6298 RegionNode* region = new RegionNode(3);
6299 region->init_req(1, instof_false);
6300 Node* src = argument(1);
6301 Node* dest = argument(4);
6302 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6303 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6304 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6305 region->init_req(2, src_dest_conjoint);
6306
6307 record_for_igvn(region);
6308 return _gvn.transform(region);
6309 }
6310
6311 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6312 // Return node representing slow path of predicate check.
6313 // the pseudo code we want to emulate with this predicate is:
6314 // for encryption:
6315 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6316 // for decryption:
6317 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6318 // note cipher==plain is more conservative than the original java code but that's OK
6319 //
6320
6321 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6322 // The receiver was checked for NULL already.
6323 Node* objCTR = argument(0);
6324
6325 // Load embeddedCipher field of CipherBlockChaining object.
6326 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6327
6328 // get AESCrypt klass for instanceOf check
6329 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6330 // will have same classloader as CipherBlockChaining object
6331 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6332 assert(tinst != NULL, "CTRobj is null");
6333 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6334
6335 // we want to do an instanceof comparison against the AESCrypt class
6336 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6337 if (!klass_AESCrypt->is_loaded()) {
6338 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6339 Node* ctrl = control();
6340 set_control(top()); // no regular fast path
6341 return ctrl;
6342 }
6343
6344 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6345 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6346 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6347 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6348 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6349
6350 return instof_false; // even if it is NULL
6351 }
6352
6353 //------------------------------inline_ghash_processBlocks
6354 bool LibraryCallKit::inline_ghash_processBlocks() {
6355 address stubAddr;
6356 const char *stubName;
6357 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6358
6359 stubAddr = StubRoutines::ghash_processBlocks();
6360 stubName = "ghash_processBlocks";
6361
6362 Node* data = argument(0);
6363 Node* offset = argument(1);
6364 Node* len = argument(2);
6365 Node* state = argument(3);
6366 Node* subkeyH = argument(4);
6367
6368 state = must_be_not_null(state, true);
6369 subkeyH = must_be_not_null(subkeyH, true);
6370 data = must_be_not_null(data, true);
6371
6372 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6373 assert(state_start, "state is NULL");
6374 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6375 assert(subkeyH_start, "subkeyH is NULL");
6376 Node* data_start = array_element_address(data, offset, T_BYTE);
6377 assert(data_start, "data is NULL");
6378
6379 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6380 OptoRuntime::ghash_processBlocks_Type(),
6381 stubAddr, stubName, TypePtr::BOTTOM,
6382 state_start, subkeyH_start, data_start, len);
6383 return true;
6384 }
6385
6386 bool LibraryCallKit::inline_base64_encodeBlock() {
6387 address stubAddr;
6388 const char *stubName;
6389 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6390 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6391 stubAddr = StubRoutines::base64_encodeBlock();
6392 stubName = "encodeBlock";
6393
6394 if (!stubAddr) return false;
6395 Node* base64obj = argument(0);
6396 Node* src = argument(1);
6397 Node* offset = argument(2);
6398 Node* len = argument(3);
6399 Node* dest = argument(4);
6400 Node* dp = argument(5);
6401 Node* isURL = argument(6);
6402
6403 src = must_be_not_null(src, true);
6404 dest = must_be_not_null(dest, true);
6405
6406 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6407 assert(src_start, "source array is NULL");
6408 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6409 assert(dest_start, "destination array is NULL");
6410
6411 Node* base64 = make_runtime_call(RC_LEAF,
6412 OptoRuntime::base64_encodeBlock_Type(),
6413 stubAddr, stubName, TypePtr::BOTTOM,
6414 src_start, offset, len, dest_start, dp, isURL);
6415 return true;
6416 }
6417
6418 //------------------------------inline_digestBase_implCompress-----------------------
6419 //
6420 // Calculate MD5 for single-block byte[] array.
6421 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
6422 //
6423 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6424 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6425 //
6426 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6427 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6428 //
6429 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6430 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6431 //
6432 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
6433 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
6434 //
6435 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
6436 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6437
6438 Node* digestBase_obj = argument(0);
6439 Node* src = argument(1); // type oop
6440 Node* ofs = argument(2); // type int
6441
6442 const Type* src_type = src->Value(&_gvn);
6443 const TypeAryPtr* top_src = src_type->isa_aryptr();
6444 if (top_src == NULL || top_src->klass() == NULL) {
6445 // failed array check
6446 return false;
6447 }
6448 // Figure out the size and type of the elements we will be copying.
6449 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6450 if (src_elem != T_BYTE) {
6451 return false;
6452 }
6453 // 'src_start' points to src array + offset
6454 src = must_be_not_null(src, true);
6455 Node* src_start = array_element_address(src, ofs, src_elem);
6456 Node* state = NULL;
6457 Node* digest_length = NULL;
6458 address stubAddr;
6459 const char *stubName;
6460
6461 switch(id) {
6462 case vmIntrinsics::_md5_implCompress:
6463 assert(UseMD5Intrinsics, "need MD5 instruction support");
6464 state = get_state_from_digest_object(digestBase_obj, "[I");
6465 stubAddr = StubRoutines::md5_implCompress();
6466 stubName = "md5_implCompress";
6467 break;
6468 case vmIntrinsics::_sha_implCompress:
6469 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6470 state = get_state_from_digest_object(digestBase_obj, "[I");
6471 stubAddr = StubRoutines::sha1_implCompress();
6472 stubName = "sha1_implCompress";
6473 break;
6474 case vmIntrinsics::_sha2_implCompress:
6475 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6476 state = get_state_from_digest_object(digestBase_obj, "[I");
6477 stubAddr = StubRoutines::sha256_implCompress();
6478 stubName = "sha256_implCompress";
6479 break;
6480 case vmIntrinsics::_sha5_implCompress:
6481 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6482 state = get_state_from_digest_object(digestBase_obj, "[J");
6483 stubAddr = StubRoutines::sha512_implCompress();
6484 stubName = "sha512_implCompress";
6485 break;
6486 case vmIntrinsics::_sha3_implCompress:
6487 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
6488 state = get_state_from_digest_object(digestBase_obj, "[B");
6489 stubAddr = StubRoutines::sha3_implCompress();
6490 stubName = "sha3_implCompress";
6491 digest_length = get_digest_length_from_digest_object(digestBase_obj);
6492 if (digest_length == NULL) return false;
6493 break;
6494 default:
6495 fatal_unexpected_iid(id);
6496 return false;
6497 }
6498 if (state == NULL) return false;
6499
6500 assert(stubAddr != NULL, "Stub is generated");
6501 if (stubAddr == NULL) return false;
6502
6503 // Call the stub.
6504 Node* call;
6505 if (digest_length == NULL) {
6506 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
6507 stubAddr, stubName, TypePtr::BOTTOM,
6508 src_start, state);
6509 } else {
6510 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
6511 stubAddr, stubName, TypePtr::BOTTOM,
6512 src_start, state, digest_length);
6513 }
6514
6515 return true;
6516 }
6517
6518 //------------------------------inline_digestBase_implCompressMB-----------------------
6519 //
6520 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
6521 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6522 //
6523 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6524 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
6525 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
6526 assert((uint)predicate < 5, "sanity");
6527 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6528
6529 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6530 Node* src = argument(1); // byte[] array
6531 Node* ofs = argument(2); // type int
6532 Node* limit = argument(3); // type int
6533
6534 const Type* src_type = src->Value(&_gvn);
6535 const TypeAryPtr* top_src = src_type->isa_aryptr();
6536 if (top_src == NULL || top_src->klass() == NULL) {
6537 // failed array check
6538 return false;
6539 }
6540 // Figure out the size and type of the elements we will be copying.
6541 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6542 if (src_elem != T_BYTE) {
6543 return false;
6544 }
6545 // 'src_start' points to src array + offset
6546 src = must_be_not_null(src, false);
6547 Node* src_start = array_element_address(src, ofs, src_elem);
6548
6549 const char* klass_digestBase_name = NULL;
6550 const char* stub_name = NULL;
6551 address stub_addr = NULL;
6552 const char* state_type = "[I";
6553
6554 switch (predicate) {
6555 case 0:
6556 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
6557 klass_digestBase_name = "sun/security/provider/MD5";
6558 stub_name = "md5_implCompressMB";
6559 stub_addr = StubRoutines::md5_implCompressMB();
6560 }
6561 break;
6562 case 1:
6563 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
6564 klass_digestBase_name = "sun/security/provider/SHA";
6565 stub_name = "sha1_implCompressMB";
6566 stub_addr = StubRoutines::sha1_implCompressMB();
6567 }
6568 break;
6569 case 2:
6570 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
6571 klass_digestBase_name = "sun/security/provider/SHA2";
6572 stub_name = "sha256_implCompressMB";
6573 stub_addr = StubRoutines::sha256_implCompressMB();
6574 }
6575 break;
6576 case 3:
6577 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
6578 klass_digestBase_name = "sun/security/provider/SHA5";
6579 stub_name = "sha512_implCompressMB";
6580 stub_addr = StubRoutines::sha512_implCompressMB();
6581 state_type = "[J";
6582 }
6583 break;
6584 case 4:
6585 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
6586 klass_digestBase_name = "sun/security/provider/SHA3";
6587 stub_name = "sha3_implCompressMB";
6588 stub_addr = StubRoutines::sha3_implCompressMB();
6589 state_type = "[B";
6590 }
6591 break;
6592 default:
6593 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
6594 }
6595 if (klass_digestBase_name != NULL) {
6596 assert(stub_addr != NULL, "Stub is generated");
6597 if (stub_addr == NULL) return false;
6598
6599 // get DigestBase klass to lookup for SHA klass
6600 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6601 assert(tinst != NULL, "digestBase_obj is not instance???");
6602 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6603
6604 ciKlass* klass_digestBase = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
6605 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
6606 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
6607 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, state_type, stub_addr, stub_name, src_start, ofs, limit);
6608 }
6609 return false;
6610 }
6611
6612 //------------------------------inline_digestBase_implCompressMB-----------------------
6613 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
6614 const char* state_type, address stubAddr, const char *stubName,
6615 Node* src_start, Node* ofs, Node* limit) {
6616 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
6617 const TypeOopPtr* xtype = aklass->as_instance_type();
6618 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6619 digest_obj = _gvn.transform(digest_obj);
6620
6621 Node* state = get_state_from_digest_object(digest_obj, state_type);
6622 if (state == NULL) return false;
6623
6624 Node* digest_length = NULL;
6625 if (strcmp("sha3_implCompressMB", stubName) == 0) {
6626 digest_length = get_digest_length_from_digest_object(digest_obj);
6627 if (digest_length == NULL) return false;
6628 }
6629
6630 // Call the stub.
6631 Node* call;
6632 if (digest_length == NULL) {
6633 call = make_runtime_call(RC_LEAF|RC_NO_FP,
6634 OptoRuntime::digestBase_implCompressMB_Type(false),
6635 stubAddr, stubName, TypePtr::BOTTOM,
6636 src_start, state, ofs, limit);
6637 } else {
6638 call = make_runtime_call(RC_LEAF|RC_NO_FP,
6639 OptoRuntime::digestBase_implCompressMB_Type(true),
6640 stubAddr, stubName, TypePtr::BOTTOM,
6641 src_start, state, digest_length, ofs, limit);
6642 }
6643
6644 // return ofs (int)
6645 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6646 set_result(result);
6647
6648 return true;
6649 }
6650
6651 //------------------------------get_state_from_digest_object-----------------------
6652 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, const char *state_type) {
6653 Node* digest_state = load_field_from_object(digest_object, "state", state_type, /*is_exact*/ false);
6654 assert (digest_state != NULL, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
6655 if (digest_state == NULL) return (Node *) NULL;
6656
6657 // now have the array, need to get the start address of the state array
6658 Node* state = array_element_address(digest_state, intcon(0), T_INT);
6659 return state;
6660 }
6661
6662 //------------------------------get_digest_length_from_sha3_object----------------------------------
6663 Node * LibraryCallKit::get_digest_length_from_digest_object(Node *digest_object) {
6664 Node* digest_length = load_field_from_object(digest_object, "digestLength", "I", /*is_exact*/ false);
6665 assert (digest_length != NULL, "sanity");
6666 return digest_length;
6667 }
6668
6669 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6670 // Return node representing slow path of predicate check.
6671 // the pseudo code we want to emulate with this predicate is:
6672 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
6673 //
6674 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6675 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
6676 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
6677 assert((uint)predicate < 5, "sanity");
6678
6679 // The receiver was checked for NULL already.
6680 Node* digestBaseObj = argument(0);
6681
6682 // get DigestBase klass for instanceOf check
6683 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6684 assert(tinst != NULL, "digestBaseObj is null");
6685 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6686
6687 const char* klass_name = NULL;
6688 switch (predicate) {
6689 case 0:
6690 if (UseMD5Intrinsics) {
6691 // we want to do an instanceof comparison against the MD5 class
6692 klass_name = "sun/security/provider/MD5";
6693 }
6694 break;
6695 case 1:
6696 if (UseSHA1Intrinsics) {
6697 // we want to do an instanceof comparison against the SHA class
6698 klass_name = "sun/security/provider/SHA";
6699 }
6700 break;
6701 case 2:
6702 if (UseSHA256Intrinsics) {
6703 // we want to do an instanceof comparison against the SHA2 class
6704 klass_name = "sun/security/provider/SHA2";
6705 }
6706 break;
6707 case 3:
6708 if (UseSHA512Intrinsics) {
6709 // we want to do an instanceof comparison against the SHA5 class
6710 klass_name = "sun/security/provider/SHA5";
6711 }
6712 break;
6713 case 4:
6714 if (UseSHA3Intrinsics) {
6715 // we want to do an instanceof comparison against the SHA3 class
6716 klass_name = "sun/security/provider/SHA3";
6717 }
6718 break;
6719 default:
6720 fatal("unknown SHA intrinsic predicate: %d", predicate);
6721 }
6722
6723 ciKlass* klass = NULL;
6724 if (klass_name != NULL) {
6725 klass = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_name));
6726 }
6727 if ((klass == NULL) || !klass->is_loaded()) {
6728 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6729 Node* ctrl = control();
6730 set_control(top()); // no intrinsic path
6731 return ctrl;
6732 }
6733 ciInstanceKlass* instklass = klass->as_instance_klass();
6734
6735 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
6736 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6737 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6738 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6739
6740 return instof_false; // even if it is NULL
6741 }
6742
6743 //-------------inline_fma-----------------------------------
6744 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
6745 Node *a = NULL;
6746 Node *b = NULL;
6747 Node *c = NULL;
6748 Node* result = NULL;
6749 switch (id) {
6750 case vmIntrinsics::_fmaD:
6751 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
6752 // no receiver since it is static method
6753 a = round_double_node(argument(0));
6754 b = round_double_node(argument(2));
6755 c = round_double_node(argument(4));
6756 result = _gvn.transform(new FmaDNode(control(), a, b, c));
6757 break;
6758 case vmIntrinsics::_fmaF:
6759 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
6760 a = argument(0);
6761 b = argument(1);
6762 c = argument(2);
6763 result = _gvn.transform(new FmaFNode(control(), a, b, c));
6764 break;
6765 default:
6766 fatal_unexpected_iid(id); break;
6767 }
6768 set_result(result);
6769 return true;
6770 }
6771
6772 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
6773 // argument(0) is receiver
6774 Node* codePoint = argument(1);
6775 Node* n = NULL;
6776
6777 switch (id) {
6778 case vmIntrinsics::_isDigit :
6779 n = new DigitNode(control(), codePoint);
6780 break;
6781 case vmIntrinsics::_isLowerCase :
6782 n = new LowerCaseNode(control(), codePoint);
6783 break;
6784 case vmIntrinsics::_isUpperCase :
6785 n = new UpperCaseNode(control(), codePoint);
6786 break;
6787 case vmIntrinsics::_isWhitespace :
6788 n = new WhitespaceNode(control(), codePoint);
6789 break;
6790 default:
6791 fatal_unexpected_iid(id);
6792 }
6793
6794 set_result(_gvn.transform(n));
6795 return true;
6796 }
6797
6798 //------------------------------inline_fp_min_max------------------------------
6799 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
6800 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
6801
6802 // The intrinsic should be used only when the API branches aren't predictable,
6803 // the last one performing the most important comparison. The following heuristic
6804 // uses the branch statistics to eventually bail out if necessary.
6805
6806 ciMethodData *md = callee()->method_data();
6807
6808 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
6809 ciCallProfile cp = caller()->call_profile_at_bci(bci());
6810
6811 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
6812 // Bail out if the call-site didn't contribute enough to the statistics.
6813 return false;
6814 }
6815
6816 uint taken = 0, not_taken = 0;
6817
6818 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
6819 if (p->is_BranchData()) {
6820 taken = ((ciBranchData*)p)->taken();
6821 not_taken = ((ciBranchData*)p)->not_taken();
6822 }
6823 }
6824
6825 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
6826 balance = balance < 0 ? -balance : balance;
6827 if ( balance > 0.2 ) {
6828 // Bail out if the most important branch is predictable enough.
6829 return false;
6830 }
6831 }
6832 */
6833
6834 Node *a = NULL;
6835 Node *b = NULL;
6836 Node *n = NULL;
6837 switch (id) {
6838 case vmIntrinsics::_maxF:
6839 case vmIntrinsics::_minF:
6840 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
6841 a = argument(0);
6842 b = argument(1);
6843 break;
6844 case vmIntrinsics::_maxD:
6845 case vmIntrinsics::_minD:
6846 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
6847 a = round_double_node(argument(0));
6848 b = round_double_node(argument(2));
6849 break;
6850 default:
6851 fatal_unexpected_iid(id);
6852 break;
6853 }
6854 switch (id) {
6855 case vmIntrinsics::_maxF: n = new MaxFNode(a, b); break;
6856 case vmIntrinsics::_minF: n = new MinFNode(a, b); break;
6857 case vmIntrinsics::_maxD: n = new MaxDNode(a, b); break;
6858 case vmIntrinsics::_minD: n = new MinDNode(a, b); break;
6859 default: fatal_unexpected_iid(id); break;
6860 }
6861 set_result(_gvn.transform(n));
6862 return true;
6863 }
6864
6865 bool LibraryCallKit::inline_profileBoolean() {
6866 Node* counts = argument(1);
6867 const TypeAryPtr* ary = NULL;
6868 ciArray* aobj = NULL;
6869 if (counts->is_Con()
6870 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6871 && (aobj = ary->const_oop()->as_array()) != NULL
6872 && (aobj->length() == 2)) {
6873 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6874 jint false_cnt = aobj->element_value(0).as_int();
6875 jint true_cnt = aobj->element_value(1).as_int();
6876
6877 if (C->log() != NULL) {
6878 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6879 false_cnt, true_cnt);
6880 }
6881
6882 if (false_cnt + true_cnt == 0) {
6883 // According to profile, never executed.
6884 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6885 Deoptimization::Action_reinterpret);
6886 return true;
6887 }
6888
6889 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6890 // is a number of each value occurrences.
6891 Node* result = argument(0);
6892 if (false_cnt == 0 || true_cnt == 0) {
6893 // According to profile, one value has been never seen.
6894 int expected_val = (false_cnt == 0) ? 1 : 0;
6895
6896 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6897 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6898
6899 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6900 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6901 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6902
6903 { // Slow path: uncommon trap for never seen value and then reexecute
6904 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6905 // the value has been seen at least once.
6906 PreserveJVMState pjvms(this);
6907 PreserveReexecuteState preexecs(this);
6908 jvms()->set_should_reexecute(true);
6909
6910 set_control(slow_path);
6911 set_i_o(i_o());
6912
6913 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6914 Deoptimization::Action_reinterpret);
6915 }
6916 // The guard for never seen value enables sharpening of the result and
6917 // returning a constant. It allows to eliminate branches on the same value
6918 // later on.
6919 set_control(fast_path);
6920 result = intcon(expected_val);
6921 }
6922 // Stop profiling.
6923 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6924 // By replacing method body with profile data (represented as ProfileBooleanNode
6925 // on IR level) we effectively disable profiling.
6926 // It enables full speed execution once optimized code is generated.
6927 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6928 C->record_for_igvn(profile);
6929 set_result(profile);
6930 return true;
6931 } else {
6932 // Continue profiling.
6933 // Profile data isn't available at the moment. So, execute method's bytecode version.
6934 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6935 // is compiled and counters aren't available since corresponding MethodHandle
6936 // isn't a compile-time constant.
6937 return false;
6938 }
6939 }
6940
6941 bool LibraryCallKit::inline_isCompileConstant() {
6942 Node* n = argument(0);
6943 set_result(n->is_Con() ? intcon(1) : intcon(0));
6944 return true;
6945 }