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 bool long_state, address stubAddr, const char *stubName,
312 Node* src_start, Node* ofs, Node* limit);
313 Node* get_state_from_digest_object(Node *digestBase_object);
314 Node* get_long_state_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:
555 case vmIntrinsics::_dsignum:
556 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
557
558 case vmIntrinsics::_min:
559 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
560
561 case vmIntrinsics::_notify:
562 case vmIntrinsics::_notifyAll:
563 return inline_notify(intrinsic_id());
564
565 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
566 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
567 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
568 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
569 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
570 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
571 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
572 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
573 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
574 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
575 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
576 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
577 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
578
579 case vmIntrinsics::_arraycopy: return inline_arraycopy();
580
581 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
582 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
583 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
584 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
585
586 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
587 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
588 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
589 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
590 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
591 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
592 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
593
594 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
595 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
596
597 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
598 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
599 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
600 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
601
602 case vmIntrinsics::_compressStringC:
603 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
604 case vmIntrinsics::_inflateStringC:
605 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
606
607 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
608 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
609 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
610 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
611 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
612 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
613 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
614 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
615 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
616
617 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
618 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
619 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
620 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
621 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
622 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
623 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
624 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
625 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
626
627 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
628 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
629 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
630 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
631 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
632 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
633 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
634 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
635 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
636
637 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
638 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
639 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
640 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
641 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
642 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
643 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
644 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
645 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
646
647 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
648 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
649 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
650 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
651
652 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
653 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
654 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
655 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
656
657 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
658 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
659 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
660 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
661 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
662 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
663 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
664 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
665 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
666
667 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
668 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
669 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
670 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
671 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
672 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
673 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
674 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
675 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
676
677 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
678 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
679 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
680 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
681 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
682 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
683 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
684 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
685 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
686
687 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
688 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
689 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
690 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
691 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
692 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
693 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
694 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
695 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
696
697 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
698 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
699 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
700 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
701 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
702
703 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
704 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
705 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
706 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
707 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
708 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
709 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
710 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
711 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
712 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
713 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
714 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
715 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
716 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
717 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
718 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
719 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
720 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
721 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
722 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
723
724 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
725 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
726 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
727 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
728 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
729 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
730 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
731 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
732 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
733 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
734 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
735 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
736 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
737 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
738 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
739
740 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
741 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
742 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
743 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
744
745 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
746 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
747 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
748 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
749 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
750
751 case vmIntrinsics::_loadFence:
752 case vmIntrinsics::_storeFence:
753 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
754
755 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
756
757 case vmIntrinsics::_currentThread: return inline_native_currentThread();
758
759 #ifdef JFR_HAVE_INTRINSICS
760 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime");
761 case vmIntrinsics::_getClassId: return inline_native_classID();
762 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
763 #endif
764 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
765 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
766 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
767 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
768 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
769 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
770 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
771 case vmIntrinsics::_getLength: return inline_native_getLength();
772 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
773 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
774 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
775 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
776 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
777 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
778
779 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
780 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
781
782 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
783
784 case vmIntrinsics::_isInstance:
785 case vmIntrinsics::_getModifiers:
786 case vmIntrinsics::_isInterface:
787 case vmIntrinsics::_isArray:
788 case vmIntrinsics::_isPrimitive:
789 case vmIntrinsics::_isHidden:
790 case vmIntrinsics::_getSuperclass:
791 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
792
793 case vmIntrinsics::_floatToRawIntBits:
794 case vmIntrinsics::_floatToIntBits:
795 case vmIntrinsics::_intBitsToFloat:
796 case vmIntrinsics::_doubleToRawLongBits:
797 case vmIntrinsics::_doubleToLongBits:
798 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
799
800 case vmIntrinsics::_numberOfLeadingZeros_i:
801 case vmIntrinsics::_numberOfLeadingZeros_l:
802 case vmIntrinsics::_numberOfTrailingZeros_i:
803 case vmIntrinsics::_numberOfTrailingZeros_l:
804 case vmIntrinsics::_bitCount_i:
805 case vmIntrinsics::_bitCount_l:
806 case vmIntrinsics::_reverseBytes_i:
807 case vmIntrinsics::_reverseBytes_l:
808 case vmIntrinsics::_reverseBytes_s:
809 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
810
811 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
812
813 case vmIntrinsics::_Reference_get: return inline_reference_get();
814
815 case vmIntrinsics::_Class_cast: return inline_Class_cast();
816
817 case vmIntrinsics::_aescrypt_encryptBlock:
818 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
819
820 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
821 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
822 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
823
824 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
825 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
826 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
827
828 case vmIntrinsics::_counterMode_AESCrypt:
829 return inline_counterMode_AESCrypt(intrinsic_id());
830
831 case vmIntrinsics::_md5_implCompress:
832 case vmIntrinsics::_sha_implCompress:
833 case vmIntrinsics::_sha2_implCompress:
834 case vmIntrinsics::_sha5_implCompress:
835 return inline_digestBase_implCompress(intrinsic_id());
836
837 case vmIntrinsics::_digestBase_implCompressMB:
838 return inline_digestBase_implCompressMB(predicate);
839
840 case vmIntrinsics::_multiplyToLen:
841 return inline_multiplyToLen();
842
843 case vmIntrinsics::_squareToLen:
844 return inline_squareToLen();
845
846 case vmIntrinsics::_mulAdd:
847 return inline_mulAdd();
848
849 case vmIntrinsics::_montgomeryMultiply:
850 return inline_montgomeryMultiply();
851 case vmIntrinsics::_montgomerySquare:
852 return inline_montgomerySquare();
853
854 case vmIntrinsics::_bigIntegerRightShiftWorker:
855 return inline_bigIntegerShift(true);
856 case vmIntrinsics::_bigIntegerLeftShiftWorker:
857 return inline_bigIntegerShift(false);
858
859 case vmIntrinsics::_vectorizedMismatch:
860 return inline_vectorizedMismatch();
861
862 case vmIntrinsics::_ghash_processBlocks:
863 return inline_ghash_processBlocks();
864 case vmIntrinsics::_base64_encodeBlock:
865 return inline_base64_encodeBlock();
866
867 case vmIntrinsics::_encodeISOArray:
868 case vmIntrinsics::_encodeByteISOArray:
869 return inline_encodeISOArray();
870
871 case vmIntrinsics::_updateCRC32:
872 return inline_updateCRC32();
873 case vmIntrinsics::_updateBytesCRC32:
874 return inline_updateBytesCRC32();
875 case vmIntrinsics::_updateByteBufferCRC32:
876 return inline_updateByteBufferCRC32();
877
878 case vmIntrinsics::_updateBytesCRC32C:
879 return inline_updateBytesCRC32C();
880 case vmIntrinsics::_updateDirectByteBufferCRC32C:
881 return inline_updateDirectByteBufferCRC32C();
882
883 case vmIntrinsics::_updateBytesAdler32:
884 return inline_updateBytesAdler32();
885 case vmIntrinsics::_updateByteBufferAdler32:
886 return inline_updateByteBufferAdler32();
887
888 case vmIntrinsics::_profileBoolean:
889 return inline_profileBoolean();
890 case vmIntrinsics::_isCompileConstant:
891 return inline_isCompileConstant();
892
893 case vmIntrinsics::_hasNegatives:
894 return inline_hasNegatives();
895
896 case vmIntrinsics::_fmaD:
897 case vmIntrinsics::_fmaF:
898 return inline_fma(intrinsic_id());
899
900 case vmIntrinsics::_isDigit:
901 case vmIntrinsics::_isLowerCase:
902 case vmIntrinsics::_isUpperCase:
903 case vmIntrinsics::_isWhitespace:
904 return inline_character_compare(intrinsic_id());
905
906 case vmIntrinsics::_maxF:
907 case vmIntrinsics::_minF:
908 case vmIntrinsics::_maxD:
909 case vmIntrinsics::_minD:
910 return inline_fp_min_max(intrinsic_id());
911
912 default:
913 // If you get here, it may be that someone has added a new intrinsic
914 // to the list in vmSymbols.hpp without implementing it here.
915 #ifndef PRODUCT
916 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
917 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
918 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
919 }
920 #endif
921 return false;
922 }
923 }
924
925 Node* LibraryCallKit::try_to_predicate(int predicate) {
926 if (!jvms()->has_method()) {
927 // Root JVMState has a null method.
928 assert(map()->memory()->Opcode() == Op_Parm, "");
929 // Insert the memory aliasing node
930 set_all_memory(reset_memory());
931 }
932 assert(merged_memory(), "");
933
934 switch (intrinsic_id()) {
935 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
936 return inline_cipherBlockChaining_AESCrypt_predicate(false);
937 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
938 return inline_cipherBlockChaining_AESCrypt_predicate(true);
939 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
940 return inline_electronicCodeBook_AESCrypt_predicate(false);
941 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
942 return inline_electronicCodeBook_AESCrypt_predicate(true);
943 case vmIntrinsics::_counterMode_AESCrypt:
944 return inline_counterMode_AESCrypt_predicate();
945 case vmIntrinsics::_digestBase_implCompressMB:
946 return inline_digestBase_implCompressMB_predicate(predicate);
947
948 default:
949 // If you get here, it may be that someone has added a new intrinsic
950 // to the list in vmSymbols.hpp without implementing it here.
951 #ifndef PRODUCT
952 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
953 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
954 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
955 }
956 #endif
957 Node* slow_ctl = control();
958 set_control(top()); // No fast path instrinsic
959 return slow_ctl;
960 }
961 }
962
963 //------------------------------set_result-------------------------------
964 // Helper function for finishing intrinsics.
965 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
966 record_for_igvn(region);
967 set_control(_gvn.transform(region));
968 set_result( _gvn.transform(value));
969 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
970 }
971
972 //------------------------------generate_guard---------------------------
973 // Helper function for generating guarded fast-slow graph structures.
974 // The given 'test', if true, guards a slow path. If the test fails
975 // then a fast path can be taken. (We generally hope it fails.)
976 // In all cases, GraphKit::control() is updated to the fast path.
977 // The returned value represents the control for the slow path.
978 // The return value is never 'top'; it is either a valid control
979 // or NULL if it is obvious that the slow path can never be taken.
980 // Also, if region and the slow control are not NULL, the slow edge
981 // is appended to the region.
982 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
983 if (stopped()) {
984 // Already short circuited.
985 return NULL;
986 }
987
988 // Build an if node and its projections.
989 // If test is true we take the slow path, which we assume is uncommon.
990 if (_gvn.type(test) == TypeInt::ZERO) {
991 // The slow branch is never taken. No need to build this guard.
992 return NULL;
993 }
994
995 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
996
997 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
998 if (if_slow == top()) {
999 // The slow branch is never taken. No need to build this guard.
1000 return NULL;
1001 }
1002
1003 if (region != NULL)
1004 region->add_req(if_slow);
1005
1006 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
1007 set_control(if_fast);
1008
1009 return if_slow;
1010 }
1011
1012 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
1013 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
1014 }
1015 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
1016 return generate_guard(test, region, PROB_FAIR);
1017 }
1018
1019 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
1020 Node* *pos_index) {
1021 if (stopped())
1022 return NULL; // already stopped
1023 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
1024 return NULL; // index is already adequately typed
1025 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
1026 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1027 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
1028 if (is_neg != NULL && pos_index != NULL) {
1029 // Emulate effect of Parse::adjust_map_after_if.
1030 Node* ccast = new CastIINode(index, TypeInt::POS);
1031 ccast->set_req(0, control());
1032 (*pos_index) = _gvn.transform(ccast);
1033 }
1034 return is_neg;
1035 }
1036
1037 // Make sure that 'position' is a valid limit index, in [0..length].
1038 // There are two equivalent plans for checking this:
1039 // A. (offset + copyLength) unsigned<= arrayLength
1040 // B. offset <= (arrayLength - copyLength)
1041 // We require that all of the values above, except for the sum and
1042 // difference, are already known to be non-negative.
1043 // Plan A is robust in the face of overflow, if offset and copyLength
1044 // are both hugely positive.
1045 //
1046 // Plan B is less direct and intuitive, but it does not overflow at
1047 // all, since the difference of two non-negatives is always
1048 // representable. Whenever Java methods must perform the equivalent
1049 // check they generally use Plan B instead of Plan A.
1050 // For the moment we use Plan A.
1051 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
1052 Node* subseq_length,
1053 Node* array_length,
1054 RegionNode* region) {
1055 if (stopped())
1056 return NULL; // already stopped
1057 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
1058 if (zero_offset && subseq_length->eqv_uncast(array_length))
1059 return NULL; // common case of whole-array copy
1060 Node* last = subseq_length;
1061 if (!zero_offset) // last += offset
1062 last = _gvn.transform(new AddINode(last, offset));
1063 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
1064 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1065 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
1066 return is_over;
1067 }
1068
1069 // Emit range checks for the given String.value byte array
1070 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
1071 if (stopped()) {
1072 return; // already stopped
1073 }
1074 RegionNode* bailout = new RegionNode(1);
1075 record_for_igvn(bailout);
1076 if (char_count) {
1077 // Convert char count to byte count
1078 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1079 }
1080
1081 // Offset and count must not be negative
1082 generate_negative_guard(offset, bailout);
1083 generate_negative_guard(count, bailout);
1084 // Offset + count must not exceed length of array
1085 generate_limit_guard(offset, count, load_array_length(array), bailout);
1086
1087 if (bailout->req() > 1) {
1088 PreserveJVMState pjvms(this);
1089 set_control(_gvn.transform(bailout));
1090 uncommon_trap(Deoptimization::Reason_intrinsic,
1091 Deoptimization::Action_maybe_recompile);
1092 }
1093 }
1094
1095 //--------------------------generate_current_thread--------------------
1096 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1097 ciKlass* thread_klass = env()->Thread_klass();
1098 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1099 Node* thread = _gvn.transform(new ThreadLocalNode());
1100 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1101 tls_output = thread;
1102 Node* thread_obj_handle = LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(), TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
1103 thread_obj_handle = _gvn.transform(thread_obj_handle);
1104 return access_load(thread_obj_handle, thread_type, T_OBJECT, IN_NATIVE | C2_IMMUTABLE_MEMORY);
1105 }
1106
1107
1108 //------------------------------make_string_method_node------------------------
1109 // Helper method for String intrinsic functions. This version is called with
1110 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1111 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1112 // containing the lengths of str1 and str2.
1113 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1114 Node* result = NULL;
1115 switch (opcode) {
1116 case Op_StrIndexOf:
1117 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1118 str1_start, cnt1, str2_start, cnt2, ae);
1119 break;
1120 case Op_StrComp:
1121 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1122 str1_start, cnt1, str2_start, cnt2, ae);
1123 break;
1124 case Op_StrEquals:
1125 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1126 // Use the constant length if there is one because optimized match rule may exist.
1127 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1128 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1129 break;
1130 default:
1131 ShouldNotReachHere();
1132 return NULL;
1133 }
1134
1135 // All these intrinsics have checks.
1136 C->set_has_split_ifs(true); // Has chance for split-if optimization
1137 clear_upper_avx();
1138
1139 return _gvn.transform(result);
1140 }
1141
1142 //------------------------------inline_string_compareTo------------------------
1143 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1144 Node* arg1 = argument(0);
1145 Node* arg2 = argument(1);
1146
1147 arg1 = must_be_not_null(arg1, true);
1148 arg2 = must_be_not_null(arg2, true);
1149
1150 // Get start addr and length of first argument
1151 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1152 Node* arg1_cnt = load_array_length(arg1);
1153
1154 // Get start addr and length of second argument
1155 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1156 Node* arg2_cnt = load_array_length(arg2);
1157
1158 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1159 set_result(result);
1160 return true;
1161 }
1162
1163 //------------------------------inline_string_equals------------------------
1164 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1165 Node* arg1 = argument(0);
1166 Node* arg2 = argument(1);
1167
1168 // paths (plus control) merge
1169 RegionNode* region = new RegionNode(3);
1170 Node* phi = new PhiNode(region, TypeInt::BOOL);
1171
1172 if (!stopped()) {
1173
1174 arg1 = must_be_not_null(arg1, true);
1175 arg2 = must_be_not_null(arg2, true);
1176
1177 // Get start addr and length of first argument
1178 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1179 Node* arg1_cnt = load_array_length(arg1);
1180
1181 // Get start addr and length of second argument
1182 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1183 Node* arg2_cnt = load_array_length(arg2);
1184
1185 // Check for arg1_cnt != arg2_cnt
1186 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1187 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1188 Node* if_ne = generate_slow_guard(bol, NULL);
1189 if (if_ne != NULL) {
1190 phi->init_req(2, intcon(0));
1191 region->init_req(2, if_ne);
1192 }
1193
1194 // Check for count == 0 is done by assembler code for StrEquals.
1195
1196 if (!stopped()) {
1197 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1198 phi->init_req(1, equals);
1199 region->init_req(1, control());
1200 }
1201 }
1202
1203 // post merge
1204 set_control(_gvn.transform(region));
1205 record_for_igvn(region);
1206
1207 set_result(_gvn.transform(phi));
1208 return true;
1209 }
1210
1211 //------------------------------inline_array_equals----------------------------
1212 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1213 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1214 Node* arg1 = argument(0);
1215 Node* arg2 = argument(1);
1216
1217 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1218 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1219 clear_upper_avx();
1220
1221 return true;
1222 }
1223
1224 //------------------------------inline_hasNegatives------------------------------
1225 bool LibraryCallKit::inline_hasNegatives() {
1226 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1227 return false;
1228 }
1229
1230 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1231 // no receiver since it is static method
1232 Node* ba = argument(0);
1233 Node* offset = argument(1);
1234 Node* len = argument(2);
1235
1236 ba = must_be_not_null(ba, true);
1237
1238 // Range checks
1239 generate_string_range_check(ba, offset, len, false);
1240 if (stopped()) {
1241 return true;
1242 }
1243 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1244 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1245 set_result(_gvn.transform(result));
1246 return true;
1247 }
1248
1249 bool LibraryCallKit::inline_preconditions_checkIndex() {
1250 Node* index = argument(0);
1251 Node* length = argument(1);
1252 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1253 return false;
1254 }
1255
1256 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1257 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1258
1259 {
1260 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1261 uncommon_trap(Deoptimization::Reason_intrinsic,
1262 Deoptimization::Action_make_not_entrant);
1263 }
1264
1265 if (stopped()) {
1266 return false;
1267 }
1268
1269 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1270 BoolTest::mask btest = BoolTest::lt;
1271 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1272 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1273 _gvn.set_type(rc, rc->Value(&_gvn));
1274 if (!rc_bool->is_Con()) {
1275 record_for_igvn(rc);
1276 }
1277 set_control(_gvn.transform(new IfTrueNode(rc)));
1278 {
1279 PreserveJVMState pjvms(this);
1280 set_control(_gvn.transform(new IfFalseNode(rc)));
1281 uncommon_trap(Deoptimization::Reason_range_check,
1282 Deoptimization::Action_make_not_entrant);
1283 }
1284
1285 if (stopped()) {
1286 return false;
1287 }
1288
1289 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1290 result->set_req(0, control());
1291 result = _gvn.transform(result);
1292 set_result(result);
1293 replace_in_map(index, result);
1294 clear_upper_avx();
1295 return true;
1296 }
1297
1298 //------------------------------inline_string_indexOf------------------------
1299 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1300 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1301 return false;
1302 }
1303 Node* src = argument(0);
1304 Node* tgt = argument(1);
1305
1306 // Make the merge point
1307 RegionNode* result_rgn = new RegionNode(4);
1308 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1309
1310 src = must_be_not_null(src, true);
1311 tgt = must_be_not_null(tgt, true);
1312
1313 // Get start addr and length of source string
1314 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1315 Node* src_count = load_array_length(src);
1316
1317 // Get start addr and length of substring
1318 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1319 Node* tgt_count = load_array_length(tgt);
1320
1321 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1322 // Divide src size by 2 if String is UTF16 encoded
1323 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1324 }
1325 if (ae == StrIntrinsicNode::UU) {
1326 // Divide substring size by 2 if String is UTF16 encoded
1327 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1328 }
1329
1330 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1331 if (result != NULL) {
1332 result_phi->init_req(3, result);
1333 result_rgn->init_req(3, control());
1334 }
1335 set_control(_gvn.transform(result_rgn));
1336 record_for_igvn(result_rgn);
1337 set_result(_gvn.transform(result_phi));
1338
1339 return true;
1340 }
1341
1342 //-----------------------------inline_string_indexOf-----------------------
1343 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1344 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1345 return false;
1346 }
1347 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1348 return false;
1349 }
1350 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1351 Node* src = argument(0); // byte[]
1352 Node* src_count = argument(1); // char count
1353 Node* tgt = argument(2); // byte[]
1354 Node* tgt_count = argument(3); // char count
1355 Node* from_index = argument(4); // char index
1356
1357 src = must_be_not_null(src, true);
1358 tgt = must_be_not_null(tgt, true);
1359
1360 // Multiply byte array index by 2 if String is UTF16 encoded
1361 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1362 src_count = _gvn.transform(new SubINode(src_count, from_index));
1363 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1364 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1365
1366 // Range checks
1367 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1368 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1369 if (stopped()) {
1370 return true;
1371 }
1372
1373 RegionNode* region = new RegionNode(5);
1374 Node* phi = new PhiNode(region, TypeInt::INT);
1375
1376 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1377 if (result != NULL) {
1378 // The result is index relative to from_index if substring was found, -1 otherwise.
1379 // Generate code which will fold into cmove.
1380 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1381 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1382
1383 Node* if_lt = generate_slow_guard(bol, NULL);
1384 if (if_lt != NULL) {
1385 // result == -1
1386 phi->init_req(3, result);
1387 region->init_req(3, if_lt);
1388 }
1389 if (!stopped()) {
1390 result = _gvn.transform(new AddINode(result, from_index));
1391 phi->init_req(4, result);
1392 region->init_req(4, control());
1393 }
1394 }
1395
1396 set_control(_gvn.transform(region));
1397 record_for_igvn(region);
1398 set_result(_gvn.transform(phi));
1399 clear_upper_avx();
1400
1401 return true;
1402 }
1403
1404 // Create StrIndexOfNode with fast path checks
1405 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1406 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1407 // Check for substr count > string count
1408 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1409 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1410 Node* if_gt = generate_slow_guard(bol, NULL);
1411 if (if_gt != NULL) {
1412 phi->init_req(1, intcon(-1));
1413 region->init_req(1, if_gt);
1414 }
1415 if (!stopped()) {
1416 // Check for substr count == 0
1417 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1418 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1419 Node* if_zero = generate_slow_guard(bol, NULL);
1420 if (if_zero != NULL) {
1421 phi->init_req(2, intcon(0));
1422 region->init_req(2, if_zero);
1423 }
1424 }
1425 if (!stopped()) {
1426 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1427 }
1428 return NULL;
1429 }
1430
1431 //-----------------------------inline_string_indexOfChar-----------------------
1432 bool LibraryCallKit::inline_string_indexOfChar() {
1433 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1434 return false;
1435 }
1436 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1437 return false;
1438 }
1439 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1440 Node* src = argument(0); // byte[]
1441 Node* tgt = argument(1); // tgt is int ch
1442 Node* from_index = argument(2);
1443 Node* max = argument(3);
1444
1445 src = must_be_not_null(src, true);
1446
1447 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1448 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1449 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1450
1451 // Range checks
1452 generate_string_range_check(src, src_offset, src_count, true);
1453 if (stopped()) {
1454 return true;
1455 }
1456
1457 RegionNode* region = new RegionNode(3);
1458 Node* phi = new PhiNode(region, TypeInt::INT);
1459
1460 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1461 C->set_has_split_ifs(true); // Has chance for split-if optimization
1462 _gvn.transform(result);
1463
1464 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1465 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1466
1467 Node* if_lt = generate_slow_guard(bol, NULL);
1468 if (if_lt != NULL) {
1469 // result == -1
1470 phi->init_req(2, result);
1471 region->init_req(2, if_lt);
1472 }
1473 if (!stopped()) {
1474 result = _gvn.transform(new AddINode(result, from_index));
1475 phi->init_req(1, result);
1476 region->init_req(1, control());
1477 }
1478 set_control(_gvn.transform(region));
1479 record_for_igvn(region);
1480 set_result(_gvn.transform(phi));
1481
1482 return true;
1483 }
1484 //---------------------------inline_string_copy---------------------
1485 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1486 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1487 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1488 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1489 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1490 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1491 bool LibraryCallKit::inline_string_copy(bool compress) {
1492 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1493 return false;
1494 }
1495 int nargs = 5; // 2 oops, 3 ints
1496 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1497
1498 Node* src = argument(0);
1499 Node* src_offset = argument(1);
1500 Node* dst = argument(2);
1501 Node* dst_offset = argument(3);
1502 Node* length = argument(4);
1503
1504 // Check for allocation before we add nodes that would confuse
1505 // tightly_coupled_allocation()
1506 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1507
1508 // Figure out the size and type of the elements we will be copying.
1509 const Type* src_type = src->Value(&_gvn);
1510 const Type* dst_type = dst->Value(&_gvn);
1511 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1512 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1513 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1514 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1515 "Unsupported array types for inline_string_copy");
1516
1517 src = must_be_not_null(src, true);
1518 dst = must_be_not_null(dst, true);
1519
1520 // Convert char[] offsets to byte[] offsets
1521 bool convert_src = (compress && src_elem == T_BYTE);
1522 bool convert_dst = (!compress && dst_elem == T_BYTE);
1523 if (convert_src) {
1524 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1525 } else if (convert_dst) {
1526 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1527 }
1528
1529 // Range checks
1530 generate_string_range_check(src, src_offset, length, convert_src);
1531 generate_string_range_check(dst, dst_offset, length, convert_dst);
1532 if (stopped()) {
1533 return true;
1534 }
1535
1536 Node* src_start = array_element_address(src, src_offset, src_elem);
1537 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1538 // 'src_start' points to src array + scaled offset
1539 // 'dst_start' points to dst array + scaled offset
1540 Node* count = NULL;
1541 if (compress) {
1542 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1543 } else {
1544 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1545 }
1546
1547 if (alloc != NULL) {
1548 if (alloc->maybe_set_complete(&_gvn)) {
1549 // "You break it, you buy it."
1550 InitializeNode* init = alloc->initialization();
1551 assert(init->is_complete(), "we just did this");
1552 init->set_complete_with_arraycopy();
1553 assert(dst->is_CheckCastPP(), "sanity");
1554 assert(dst->in(0)->in(0) == init, "dest pinned");
1555 }
1556 // Do not let stores that initialize this object be reordered with
1557 // a subsequent store that would make this object accessible by
1558 // other threads.
1559 // Record what AllocateNode this StoreStore protects so that
1560 // escape analysis can go from the MemBarStoreStoreNode to the
1561 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1562 // based on the escape status of the AllocateNode.
1563 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1564 }
1565 if (compress) {
1566 set_result(_gvn.transform(count));
1567 }
1568 clear_upper_avx();
1569
1570 return true;
1571 }
1572
1573 #ifdef _LP64
1574 #define XTOP ,top() /*additional argument*/
1575 #else //_LP64
1576 #define XTOP /*no additional argument*/
1577 #endif //_LP64
1578
1579 //------------------------inline_string_toBytesU--------------------------
1580 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1581 bool LibraryCallKit::inline_string_toBytesU() {
1582 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1583 return false;
1584 }
1585 // Get the arguments.
1586 Node* value = argument(0);
1587 Node* offset = argument(1);
1588 Node* length = argument(2);
1589
1590 Node* newcopy = NULL;
1591
1592 // Set the original stack and the reexecute bit for the interpreter to reexecute
1593 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1594 { PreserveReexecuteState preexecs(this);
1595 jvms()->set_should_reexecute(true);
1596
1597 // Check if a null path was taken unconditionally.
1598 value = null_check(value);
1599
1600 RegionNode* bailout = new RegionNode(1);
1601 record_for_igvn(bailout);
1602
1603 // Range checks
1604 generate_negative_guard(offset, bailout);
1605 generate_negative_guard(length, bailout);
1606 generate_limit_guard(offset, length, load_array_length(value), bailout);
1607 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1608 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1609
1610 if (bailout->req() > 1) {
1611 PreserveJVMState pjvms(this);
1612 set_control(_gvn.transform(bailout));
1613 uncommon_trap(Deoptimization::Reason_intrinsic,
1614 Deoptimization::Action_maybe_recompile);
1615 }
1616 if (stopped()) {
1617 return true;
1618 }
1619
1620 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1621 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1622 newcopy = new_array(klass_node, size, 0); // no arguments to push
1623 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1624
1625 // Calculate starting addresses.
1626 Node* src_start = array_element_address(value, offset, T_CHAR);
1627 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1628
1629 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1630 const TypeInt* toffset = gvn().type(offset)->is_int();
1631 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1632
1633 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1634 const char* copyfunc_name = "arraycopy";
1635 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1636 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1637 OptoRuntime::fast_arraycopy_Type(),
1638 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1639 src_start, dst_start, ConvI2X(length) XTOP);
1640 // Do not let reads from the cloned object float above the arraycopy.
1641 if (alloc != NULL) {
1642 if (alloc->maybe_set_complete(&_gvn)) {
1643 // "You break it, you buy it."
1644 InitializeNode* init = alloc->initialization();
1645 assert(init->is_complete(), "we just did this");
1646 init->set_complete_with_arraycopy();
1647 assert(newcopy->is_CheckCastPP(), "sanity");
1648 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1649 }
1650 // Do not let stores that initialize this object be reordered with
1651 // a subsequent store that would make this object accessible by
1652 // other threads.
1653 // Record what AllocateNode this StoreStore protects so that
1654 // escape analysis can go from the MemBarStoreStoreNode to the
1655 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1656 // based on the escape status of the AllocateNode.
1657 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1658 } else {
1659 insert_mem_bar(Op_MemBarCPUOrder);
1660 }
1661 } // original reexecute is set back here
1662
1663 C->set_has_split_ifs(true); // Has chance for split-if optimization
1664 if (!stopped()) {
1665 set_result(newcopy);
1666 }
1667 clear_upper_avx();
1668
1669 return true;
1670 }
1671
1672 //------------------------inline_string_getCharsU--------------------------
1673 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1674 bool LibraryCallKit::inline_string_getCharsU() {
1675 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1676 return false;
1677 }
1678
1679 // Get the arguments.
1680 Node* src = argument(0);
1681 Node* src_begin = argument(1);
1682 Node* src_end = argument(2); // exclusive offset (i < src_end)
1683 Node* dst = argument(3);
1684 Node* dst_begin = argument(4);
1685
1686 // Check for allocation before we add nodes that would confuse
1687 // tightly_coupled_allocation()
1688 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1689
1690 // Check if a null path was taken unconditionally.
1691 src = null_check(src);
1692 dst = null_check(dst);
1693 if (stopped()) {
1694 return true;
1695 }
1696
1697 // Get length and convert char[] offset to byte[] offset
1698 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1699 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1700
1701 // Range checks
1702 generate_string_range_check(src, src_begin, length, true);
1703 generate_string_range_check(dst, dst_begin, length, false);
1704 if (stopped()) {
1705 return true;
1706 }
1707
1708 if (!stopped()) {
1709 // Calculate starting addresses.
1710 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1711 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1712
1713 // Check if array addresses are aligned to HeapWordSize
1714 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1715 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1716 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1717 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1718
1719 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1720 const char* copyfunc_name = "arraycopy";
1721 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1722 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1723 OptoRuntime::fast_arraycopy_Type(),
1724 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1725 src_start, dst_start, ConvI2X(length) XTOP);
1726 // Do not let reads from the cloned object float above the arraycopy.
1727 if (alloc != NULL) {
1728 if (alloc->maybe_set_complete(&_gvn)) {
1729 // "You break it, you buy it."
1730 InitializeNode* init = alloc->initialization();
1731 assert(init->is_complete(), "we just did this");
1732 init->set_complete_with_arraycopy();
1733 assert(dst->is_CheckCastPP(), "sanity");
1734 assert(dst->in(0)->in(0) == init, "dest pinned");
1735 }
1736 // Do not let stores that initialize this object be reordered with
1737 // a subsequent store that would make this object accessible by
1738 // other threads.
1739 // Record what AllocateNode this StoreStore protects so that
1740 // escape analysis can go from the MemBarStoreStoreNode to the
1741 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1742 // based on the escape status of the AllocateNode.
1743 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1744 } else {
1745 insert_mem_bar(Op_MemBarCPUOrder);
1746 }
1747 }
1748
1749 C->set_has_split_ifs(true); // Has chance for split-if optimization
1750 return true;
1751 }
1752
1753 //----------------------inline_string_char_access----------------------------
1754 // Store/Load char to/from byte[] array.
1755 // static void StringUTF16.putChar(byte[] val, int index, int c)
1756 // static char StringUTF16.getChar(byte[] val, int index)
1757 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1758 Node* value = argument(0);
1759 Node* index = argument(1);
1760 Node* ch = is_store ? argument(2) : NULL;
1761
1762 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1763 // correctly requires matched array shapes.
1764 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1765 "sanity: byte[] and char[] bases agree");
1766 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1767 "sanity: byte[] and char[] scales agree");
1768
1769 // Bail when getChar over constants is requested: constant folding would
1770 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1771 // Java method would constant fold nicely instead.
1772 if (!is_store && value->is_Con() && index->is_Con()) {
1773 return false;
1774 }
1775
1776 value = must_be_not_null(value, true);
1777
1778 Node* adr = array_element_address(value, index, T_CHAR);
1779 if (adr->is_top()) {
1780 return false;
1781 }
1782 if (is_store) {
1783 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1784 } else {
1785 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
1786 set_result(ch);
1787 }
1788 return true;
1789 }
1790
1791 //--------------------------round_double_node--------------------------------
1792 // Round a double node if necessary.
1793 Node* LibraryCallKit::round_double_node(Node* n) {
1794 if (Matcher::strict_fp_requires_explicit_rounding) {
1795 #ifdef IA32
1796 if (UseSSE < 2) {
1797 n = _gvn.transform(new RoundDoubleNode(NULL, n));
1798 }
1799 #else
1800 Unimplemented();
1801 #endif // IA32
1802 }
1803 return n;
1804 }
1805
1806 //------------------------------inline_math-----------------------------------
1807 // public static double Math.abs(double)
1808 // public static double Math.sqrt(double)
1809 // public static double Math.log(double)
1810 // public static double Math.log10(double)
1811 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1812 Node* arg = round_double_node(argument(0));
1813 Node* n = NULL;
1814 switch (id) {
1815 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1816 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1817 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1818 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1819 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1820 case vmIntrinsics::_dsignum: n = new SignumDNode(arg); break;
1821 default: fatal_unexpected_iid(id); break;
1822 }
1823 set_result(_gvn.transform(n));
1824 return true;
1825 }
1826
1827 //------------------------------inline_math-----------------------------------
1828 // public static float Math.abs(float)
1829 // public static int Math.abs(int)
1830 // public static long Math.abs(long)
1831 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1832 Node* arg = argument(0);
1833 Node* n = NULL;
1834 switch (id) {
1835 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1836 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1837 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1838 case vmIntrinsics::_fsignum: n = new SignumFNode(arg); break;
1839 default: fatal_unexpected_iid(id); break;
1840 }
1841 set_result(_gvn.transform(n));
1842 return true;
1843 }
1844
1845 //------------------------------runtime_math-----------------------------
1846 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1847 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1848 "must be (DD)D or (D)D type");
1849
1850 // Inputs
1851 Node* a = round_double_node(argument(0));
1852 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1853
1854 const TypePtr* no_memory_effects = NULL;
1855 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1856 no_memory_effects,
1857 a, top(), b, b ? top() : NULL);
1858 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1859 #ifdef ASSERT
1860 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1861 assert(value_top == top(), "second value must be top");
1862 #endif
1863
1864 set_result(value);
1865 return true;
1866 }
1867
1868 //------------------------------inline_math_native-----------------------------
1869 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1870 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1871 switch (id) {
1872 // These intrinsics are not properly supported on all hardware
1873 case vmIntrinsics::_dsin:
1874 return StubRoutines::dsin() != NULL ?
1875 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1876 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1877 case vmIntrinsics::_dcos:
1878 return StubRoutines::dcos() != NULL ?
1879 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1880 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1881 case vmIntrinsics::_dtan:
1882 return StubRoutines::dtan() != NULL ?
1883 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1884 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1885 case vmIntrinsics::_dlog:
1886 return StubRoutines::dlog() != NULL ?
1887 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1888 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1889 case vmIntrinsics::_dlog10:
1890 return StubRoutines::dlog10() != NULL ?
1891 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1892 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1893
1894 // These intrinsics are supported on all hardware
1895 case vmIntrinsics::_ceil:
1896 case vmIntrinsics::_floor:
1897 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1898 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1899 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1900 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1901 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1902 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1903
1904 case vmIntrinsics::_dexp:
1905 return StubRoutines::dexp() != NULL ?
1906 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1907 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1908 case vmIntrinsics::_dpow: {
1909 Node* exp = round_double_node(argument(2));
1910 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1911 if (d != NULL && d->getd() == 2.0) {
1912 // Special case: pow(x, 2.0) => x * x
1913 Node* base = round_double_node(argument(0));
1914 set_result(_gvn.transform(new MulDNode(base, base)));
1915 return true;
1916 }
1917 return StubRoutines::dpow() != NULL ?
1918 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1919 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1920 }
1921 #undef FN_PTR
1922
1923 case vmIntrinsics::_dsignum: return inline_double_math(id);
1924 case vmIntrinsics::_fsignum: return inline_math(id);
1925
1926 // These intrinsics are not yet correctly implemented
1927 case vmIntrinsics::_datan2:
1928 return false;
1929
1930 default:
1931 fatal_unexpected_iid(id);
1932 return false;
1933 }
1934 }
1935
1936 static bool is_simple_name(Node* n) {
1937 return (n->req() == 1 // constant
1938 || (n->is_Type() && n->as_Type()->type()->singleton())
1939 || n->is_Proj() // parameter or return value
1940 || n->is_Phi() // local of some sort
1941 );
1942 }
1943
1944 //----------------------------inline_notify-----------------------------------*
1945 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1946 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1947 address func;
1948 if (id == vmIntrinsics::_notify) {
1949 func = OptoRuntime::monitor_notify_Java();
1950 } else {
1951 func = OptoRuntime::monitor_notifyAll_Java();
1952 }
1953 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1954 make_slow_call_ex(call, env()->Throwable_klass(), false);
1955 return true;
1956 }
1957
1958
1959 //----------------------------inline_min_max-----------------------------------
1960 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1961 set_result(generate_min_max(id, argument(0), argument(1)));
1962 return true;
1963 }
1964
1965 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1966 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1967 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1968 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1969 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1970
1971 {
1972 PreserveJVMState pjvms(this);
1973 PreserveReexecuteState preexecs(this);
1974 jvms()->set_should_reexecute(true);
1975
1976 set_control(slow_path);
1977 set_i_o(i_o());
1978
1979 uncommon_trap(Deoptimization::Reason_intrinsic,
1980 Deoptimization::Action_none);
1981 }
1982
1983 set_control(fast_path);
1984 set_result(math);
1985 }
1986
1987 template <typename OverflowOp>
1988 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1989 typedef typename OverflowOp::MathOp MathOp;
1990
1991 MathOp* mathOp = new MathOp(arg1, arg2);
1992 Node* operation = _gvn.transform( mathOp );
1993 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1994 inline_math_mathExact(operation, ofcheck);
1995 return true;
1996 }
1997
1998 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1999 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2000 }
2001
2002 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2003 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2004 }
2005
2006 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2007 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2008 }
2009
2010 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2011 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2012 }
2013
2014 bool LibraryCallKit::inline_math_negateExactI() {
2015 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2016 }
2017
2018 bool LibraryCallKit::inline_math_negateExactL() {
2019 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2020 }
2021
2022 bool LibraryCallKit::inline_math_multiplyExactI() {
2023 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2024 }
2025
2026 bool LibraryCallKit::inline_math_multiplyExactL() {
2027 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2028 }
2029
2030 bool LibraryCallKit::inline_math_multiplyHigh() {
2031 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2032 return true;
2033 }
2034
2035 Node*
2036 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2037 // These are the candidate return value:
2038 Node* xvalue = x0;
2039 Node* yvalue = y0;
2040
2041 if (xvalue == yvalue) {
2042 return xvalue;
2043 }
2044
2045 bool want_max = (id == vmIntrinsics::_max);
2046
2047 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
2048 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
2049 if (txvalue == NULL || tyvalue == NULL) return top();
2050 // This is not really necessary, but it is consistent with a
2051 // hypothetical MaxINode::Value method:
2052 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
2053
2054 // %%% This folding logic should (ideally) be in a different place.
2055 // Some should be inside IfNode, and there to be a more reliable
2056 // transformation of ?: style patterns into cmoves. We also want
2057 // more powerful optimizations around cmove and min/max.
2058
2059 // Try to find a dominating comparison of these guys.
2060 // It can simplify the index computation for Arrays.copyOf
2061 // and similar uses of System.arraycopy.
2062 // First, compute the normalized version of CmpI(x, y).
2063 int cmp_op = Op_CmpI;
2064 Node* xkey = xvalue;
2065 Node* ykey = yvalue;
2066 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
2067 if (ideal_cmpxy->is_Cmp()) {
2068 // E.g., if we have CmpI(length - offset, count),
2069 // it might idealize to CmpI(length, count + offset)
2070 cmp_op = ideal_cmpxy->Opcode();
2071 xkey = ideal_cmpxy->in(1);
2072 ykey = ideal_cmpxy->in(2);
2073 }
2074
2075 // Start by locating any relevant comparisons.
2076 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
2077 Node* cmpxy = NULL;
2078 Node* cmpyx = NULL;
2079 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
2080 Node* cmp = start_from->fast_out(k);
2081 if (cmp->outcnt() > 0 && // must have prior uses
2082 cmp->in(0) == NULL && // must be context-independent
2083 cmp->Opcode() == cmp_op) { // right kind of compare
2084 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
2085 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
2086 }
2087 }
2088
2089 const int NCMPS = 2;
2090 Node* cmps[NCMPS] = { cmpxy, cmpyx };
2091 int cmpn;
2092 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2093 if (cmps[cmpn] != NULL) break; // find a result
2094 }
2095 if (cmpn < NCMPS) {
2096 // Look for a dominating test that tells us the min and max.
2097 int depth = 0; // Limit search depth for speed
2098 Node* dom = control();
2099 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2100 if (++depth >= 100) break;
2101 Node* ifproj = dom;
2102 if (!ifproj->is_Proj()) continue;
2103 Node* iff = ifproj->in(0);
2104 if (!iff->is_If()) continue;
2105 Node* bol = iff->in(1);
2106 if (!bol->is_Bool()) continue;
2107 Node* cmp = bol->in(1);
2108 if (cmp == NULL) continue;
2109 for (cmpn = 0; cmpn < NCMPS; cmpn++)
2110 if (cmps[cmpn] == cmp) break;
2111 if (cmpn == NCMPS) continue;
2112 BoolTest::mask btest = bol->as_Bool()->_test._test;
2113 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
2114 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2115 // At this point, we know that 'x btest y' is true.
2116 switch (btest) {
2117 case BoolTest::eq:
2118 // They are proven equal, so we can collapse the min/max.
2119 // Either value is the answer. Choose the simpler.
2120 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2121 return yvalue;
2122 return xvalue;
2123 case BoolTest::lt: // x < y
2124 case BoolTest::le: // x <= y
2125 return (want_max ? yvalue : xvalue);
2126 case BoolTest::gt: // x > y
2127 case BoolTest::ge: // x >= y
2128 return (want_max ? xvalue : yvalue);
2129 default:
2130 break;
2131 }
2132 }
2133 }
2134
2135 // We failed to find a dominating test.
2136 // Let's pick a test that might GVN with prior tests.
2137 Node* best_bol = NULL;
2138 BoolTest::mask best_btest = BoolTest::illegal;
2139 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2140 Node* cmp = cmps[cmpn];
2141 if (cmp == NULL) continue;
2142 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2143 Node* bol = cmp->fast_out(j);
2144 if (!bol->is_Bool()) continue;
2145 BoolTest::mask btest = bol->as_Bool()->_test._test;
2146 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2147 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2148 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2149 best_bol = bol->as_Bool();
2150 best_btest = btest;
2151 }
2152 }
2153 }
2154
2155 Node* answer_if_true = NULL;
2156 Node* answer_if_false = NULL;
2157 switch (best_btest) {
2158 default:
2159 if (cmpxy == NULL)
2160 cmpxy = ideal_cmpxy;
2161 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2162 // and fall through:
2163 case BoolTest::lt: // x < y
2164 case BoolTest::le: // x <= y
2165 answer_if_true = (want_max ? yvalue : xvalue);
2166 answer_if_false = (want_max ? xvalue : yvalue);
2167 break;
2168 case BoolTest::gt: // x > y
2169 case BoolTest::ge: // x >= y
2170 answer_if_true = (want_max ? xvalue : yvalue);
2171 answer_if_false = (want_max ? yvalue : xvalue);
2172 break;
2173 }
2174
2175 jint hi, lo;
2176 if (want_max) {
2177 // We can sharpen the minimum.
2178 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2179 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2180 } else {
2181 // We can sharpen the maximum.
2182 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2183 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2184 }
2185
2186 // Use a flow-free graph structure, to avoid creating excess control edges
2187 // which could hinder other optimizations.
2188 // Since Math.min/max is often used with arraycopy, we want
2189 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2190 Node* cmov = CMoveNode::make(NULL, best_bol,
2191 answer_if_false, answer_if_true,
2192 TypeInt::make(lo, hi, widen));
2193
2194 return _gvn.transform(cmov);
2195
2196 /*
2197 // This is not as desirable as it may seem, since Min and Max
2198 // nodes do not have a full set of optimizations.
2199 // And they would interfere, anyway, with 'if' optimizations
2200 // and with CMoveI canonical forms.
2201 switch (id) {
2202 case vmIntrinsics::_min:
2203 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2204 case vmIntrinsics::_max:
2205 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2206 default:
2207 ShouldNotReachHere();
2208 }
2209 */
2210 }
2211
2212 inline int
2213 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2214 const TypePtr* base_type = TypePtr::NULL_PTR;
2215 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2216 if (base_type == NULL) {
2217 // Unknown type.
2218 return Type::AnyPtr;
2219 } else if (base_type == TypePtr::NULL_PTR) {
2220 // Since this is a NULL+long form, we have to switch to a rawptr.
2221 base = _gvn.transform(new CastX2PNode(offset));
2222 offset = MakeConX(0);
2223 return Type::RawPtr;
2224 } else if (base_type->base() == Type::RawPtr) {
2225 return Type::RawPtr;
2226 } else if (base_type->isa_oopptr()) {
2227 // Base is never null => always a heap address.
2228 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2229 return Type::OopPtr;
2230 }
2231 // Offset is small => always a heap address.
2232 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2233 if (offset_type != NULL &&
2234 base_type->offset() == 0 && // (should always be?)
2235 offset_type->_lo >= 0 &&
2236 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2237 return Type::OopPtr;
2238 } else if (type == T_OBJECT) {
2239 // off heap access to an oop doesn't make any sense. Has to be on
2240 // heap.
2241 return Type::OopPtr;
2242 }
2243 // Otherwise, it might either be oop+off or NULL+addr.
2244 return Type::AnyPtr;
2245 } else {
2246 // No information:
2247 return Type::AnyPtr;
2248 }
2249 }
2250
2251 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type, bool can_cast) {
2252 Node* uncasted_base = base;
2253 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2254 if (kind == Type::RawPtr) {
2255 return basic_plus_adr(top(), uncasted_base, offset);
2256 } else if (kind == Type::AnyPtr) {
2257 assert(base == uncasted_base, "unexpected base change");
2258 if (can_cast) {
2259 if (!_gvn.type(base)->speculative_maybe_null() &&
2260 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2261 // According to profiling, this access is always on
2262 // heap. Casting the base to not null and thus avoiding membars
2263 // around the access should allow better optimizations
2264 Node* null_ctl = top();
2265 base = null_check_oop(base, &null_ctl, true, true, true);
2266 assert(null_ctl->is_top(), "no null control here");
2267 return basic_plus_adr(base, offset);
2268 } else if (_gvn.type(base)->speculative_always_null() &&
2269 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2270 // According to profiling, this access is always off
2271 // heap.
2272 base = null_assert(base);
2273 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2274 offset = MakeConX(0);
2275 return basic_plus_adr(top(), raw_base, offset);
2276 }
2277 }
2278 // We don't know if it's an on heap or off heap access. Fall back
2279 // to raw memory access.
2280 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2281 return basic_plus_adr(top(), raw, offset);
2282 } else {
2283 assert(base == uncasted_base, "unexpected base change");
2284 // We know it's an on heap access so base can't be null
2285 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2286 base = must_be_not_null(base, true);
2287 }
2288 return basic_plus_adr(base, offset);
2289 }
2290 }
2291
2292 //--------------------------inline_number_methods-----------------------------
2293 // inline int Integer.numberOfLeadingZeros(int)
2294 // inline int Long.numberOfLeadingZeros(long)
2295 //
2296 // inline int Integer.numberOfTrailingZeros(int)
2297 // inline int Long.numberOfTrailingZeros(long)
2298 //
2299 // inline int Integer.bitCount(int)
2300 // inline int Long.bitCount(long)
2301 //
2302 // inline char Character.reverseBytes(char)
2303 // inline short Short.reverseBytes(short)
2304 // inline int Integer.reverseBytes(int)
2305 // inline long Long.reverseBytes(long)
2306 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2307 Node* arg = argument(0);
2308 Node* n = NULL;
2309 switch (id) {
2310 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2311 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2312 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2313 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2314 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2315 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2316 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2317 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2318 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2319 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2320 default: fatal_unexpected_iid(id); break;
2321 }
2322 set_result(_gvn.transform(n));
2323 return true;
2324 }
2325
2326 //----------------------------inline_unsafe_access----------------------------
2327
2328 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2329 // Attempt to infer a sharper value type from the offset and base type.
2330 ciKlass* sharpened_klass = NULL;
2331
2332 // See if it is an instance field, with an object type.
2333 if (alias_type->field() != NULL) {
2334 if (alias_type->field()->type()->is_klass()) {
2335 sharpened_klass = alias_type->field()->type()->as_klass();
2336 }
2337 }
2338
2339 // See if it is a narrow oop array.
2340 if (adr_type->isa_aryptr()) {
2341 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2342 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2343 if (elem_type != NULL) {
2344 sharpened_klass = elem_type->klass();
2345 }
2346 }
2347 }
2348
2349 // The sharpened class might be unloaded if there is no class loader
2350 // contraint in place.
2351 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2352 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2353
2354 #ifndef PRODUCT
2355 if (C->print_intrinsics() || C->print_inlining()) {
2356 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2357 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2358 }
2359 #endif
2360 // Sharpen the value type.
2361 return tjp;
2362 }
2363 return NULL;
2364 }
2365
2366 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2367 switch (kind) {
2368 case Relaxed:
2369 return MO_UNORDERED;
2370 case Opaque:
2371 return MO_RELAXED;
2372 case Acquire:
2373 return MO_ACQUIRE;
2374 case Release:
2375 return MO_RELEASE;
2376 case Volatile:
2377 return MO_SEQ_CST;
2378 default:
2379 ShouldNotReachHere();
2380 return 0;
2381 }
2382 }
2383
2384 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2385 if (callee()->is_static()) return false; // caller must have the capability!
2386 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2387 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2388 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2389 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2390
2391 if (is_reference_type(type)) {
2392 decorators |= ON_UNKNOWN_OOP_REF;
2393 }
2394
2395 if (unaligned) {
2396 decorators |= C2_UNALIGNED;
2397 }
2398
2399 #ifndef PRODUCT
2400 {
2401 ResourceMark rm;
2402 // Check the signatures.
2403 ciSignature* sig = callee()->signature();
2404 #ifdef ASSERT
2405 if (!is_store) {
2406 // Object getReference(Object base, int/long offset), etc.
2407 BasicType rtype = sig->return_type()->basic_type();
2408 assert(rtype == type, "getter must return the expected value");
2409 assert(sig->count() == 2, "oop getter has 2 arguments");
2410 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2411 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2412 } else {
2413 // void putReference(Object base, int/long offset, Object x), etc.
2414 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2415 assert(sig->count() == 3, "oop putter has 3 arguments");
2416 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2417 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2418 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2419 assert(vtype == type, "putter must accept the expected value");
2420 }
2421 #endif // ASSERT
2422 }
2423 #endif //PRODUCT
2424
2425 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2426
2427 Node* receiver = argument(0); // type: oop
2428
2429 // Build address expression.
2430 Node* adr;
2431 Node* heap_base_oop = top();
2432 Node* offset = top();
2433 Node* val;
2434
2435 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2436 Node* base = argument(1); // type: oop
2437 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2438 offset = argument(2); // type: long
2439 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2440 // to be plain byte offsets, which are also the same as those accepted
2441 // by oopDesc::field_addr.
2442 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2443 "fieldOffset must be byte-scaled");
2444 // 32-bit machines ignore the high half!
2445 offset = ConvL2X(offset);
2446 adr = make_unsafe_address(base, offset, is_store ? ACCESS_WRITE : ACCESS_READ, type, kind == Relaxed);
2447
2448 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2449 if (type != T_OBJECT) {
2450 decorators |= IN_NATIVE; // off-heap primitive access
2451 } else {
2452 return false; // off-heap oop accesses are not supported
2453 }
2454 } else {
2455 heap_base_oop = base; // on-heap or mixed access
2456 }
2457
2458 // Can base be NULL? Otherwise, always on-heap access.
2459 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2460
2461 if (!can_access_non_heap) {
2462 decorators |= IN_HEAP;
2463 }
2464
2465 val = is_store ? argument(4) : NULL;
2466
2467 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2468 if (adr_type == TypePtr::NULL_PTR) {
2469 return false; // off-heap access with zero address
2470 }
2471
2472 // Try to categorize the address.
2473 Compile::AliasType* alias_type = C->alias_type(adr_type);
2474 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2475
2476 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2477 alias_type->adr_type() == TypeAryPtr::RANGE) {
2478 return false; // not supported
2479 }
2480
2481 bool mismatched = false;
2482 BasicType bt = alias_type->basic_type();
2483 if (bt != T_ILLEGAL) {
2484 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2485 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2486 // Alias type doesn't differentiate between byte[] and boolean[]).
2487 // Use address type to get the element type.
2488 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2489 }
2490 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2491 // accessing an array field with getReference is not a mismatch
2492 bt = T_OBJECT;
2493 }
2494 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2495 // Don't intrinsify mismatched object accesses
2496 return false;
2497 }
2498 mismatched = (bt != type);
2499 } else if (alias_type->adr_type()->isa_oopptr()) {
2500 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2501 }
2502
2503 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2504
2505 if (mismatched) {
2506 decorators |= C2_MISMATCHED;
2507 }
2508
2509 // First guess at the value type.
2510 const Type *value_type = Type::get_const_basic_type(type);
2511
2512 // Figure out the memory ordering.
2513 decorators |= mo_decorator_for_access_kind(kind);
2514
2515 if (!is_store && type == T_OBJECT) {
2516 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2517 if (tjp != NULL) {
2518 value_type = tjp;
2519 }
2520 }
2521
2522 receiver = null_check(receiver);
2523 if (stopped()) {
2524 return true;
2525 }
2526 // Heap pointers get a null-check from the interpreter,
2527 // as a courtesy. However, this is not guaranteed by Unsafe,
2528 // and it is not possible to fully distinguish unintended nulls
2529 // from intended ones in this API.
2530
2531 if (!is_store) {
2532 Node* p = NULL;
2533 // Try to constant fold a load from a constant field
2534 ciField* field = alias_type->field();
2535 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2536 // final or stable field
2537 p = make_constant_from_field(field, heap_base_oop);
2538 }
2539
2540 if (p == NULL) { // Could not constant fold the load
2541 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2542 // Normalize the value returned by getBoolean in the following cases
2543 if (type == T_BOOLEAN &&
2544 (mismatched ||
2545 heap_base_oop == top() || // - heap_base_oop is NULL or
2546 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2547 // and the unsafe access is made to large offset
2548 // (i.e., larger than the maximum offset necessary for any
2549 // field access)
2550 ) {
2551 IdealKit ideal = IdealKit(this);
2552 #define __ ideal.
2553 IdealVariable normalized_result(ideal);
2554 __ declarations_done();
2555 __ set(normalized_result, p);
2556 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2557 __ set(normalized_result, ideal.ConI(1));
2558 ideal.end_if();
2559 final_sync(ideal);
2560 p = __ value(normalized_result);
2561 #undef __
2562 }
2563 }
2564 if (type == T_ADDRESS) {
2565 p = gvn().transform(new CastP2XNode(NULL, p));
2566 p = ConvX2UL(p);
2567 }
2568 // The load node has the control of the preceding MemBarCPUOrder. All
2569 // following nodes will have the control of the MemBarCPUOrder inserted at
2570 // the end of this method. So, pushing the load onto the stack at a later
2571 // point is fine.
2572 set_result(p);
2573 } else {
2574 if (bt == T_ADDRESS) {
2575 // Repackage the long as a pointer.
2576 val = ConvL2X(val);
2577 val = gvn().transform(new CastX2PNode(val));
2578 }
2579 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2580 }
2581
2582 return true;
2583 }
2584
2585 //----------------------------inline_unsafe_load_store----------------------------
2586 // This method serves a couple of different customers (depending on LoadStoreKind):
2587 //
2588 // LS_cmp_swap:
2589 //
2590 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2591 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2592 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2593 //
2594 // LS_cmp_swap_weak:
2595 //
2596 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2597 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2598 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2599 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2600 //
2601 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2602 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2603 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2604 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2605 //
2606 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2607 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2608 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2609 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2610 //
2611 // LS_cmp_exchange:
2612 //
2613 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2614 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2615 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2616 //
2617 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2618 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2619 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2620 //
2621 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2622 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2623 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2624 //
2625 // LS_get_add:
2626 //
2627 // int getAndAddInt( Object o, long offset, int delta)
2628 // long getAndAddLong(Object o, long offset, long delta)
2629 //
2630 // LS_get_set:
2631 //
2632 // int getAndSet(Object o, long offset, int newValue)
2633 // long getAndSet(Object o, long offset, long newValue)
2634 // Object getAndSet(Object o, long offset, Object newValue)
2635 //
2636 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2637 // This basic scheme here is the same as inline_unsafe_access, but
2638 // differs in enough details that combining them would make the code
2639 // overly confusing. (This is a true fact! I originally combined
2640 // them, but even I was confused by it!) As much code/comments as
2641 // possible are retained from inline_unsafe_access though to make
2642 // the correspondences clearer. - dl
2643
2644 if (callee()->is_static()) return false; // caller must have the capability!
2645
2646 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2647 decorators |= mo_decorator_for_access_kind(access_kind);
2648
2649 #ifndef PRODUCT
2650 BasicType rtype;
2651 {
2652 ResourceMark rm;
2653 // Check the signatures.
2654 ciSignature* sig = callee()->signature();
2655 rtype = sig->return_type()->basic_type();
2656 switch(kind) {
2657 case LS_get_add:
2658 case LS_get_set: {
2659 // Check the signatures.
2660 #ifdef ASSERT
2661 assert(rtype == type, "get and set must return the expected type");
2662 assert(sig->count() == 3, "get and set has 3 arguments");
2663 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2664 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2665 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2666 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2667 #endif // ASSERT
2668 break;
2669 }
2670 case LS_cmp_swap:
2671 case LS_cmp_swap_weak: {
2672 // Check the signatures.
2673 #ifdef ASSERT
2674 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2675 assert(sig->count() == 4, "CAS has 4 arguments");
2676 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2677 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2678 #endif // ASSERT
2679 break;
2680 }
2681 case LS_cmp_exchange: {
2682 // Check the signatures.
2683 #ifdef ASSERT
2684 assert(rtype == type, "CAS must return the expected type");
2685 assert(sig->count() == 4, "CAS has 4 arguments");
2686 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2687 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2688 #endif // ASSERT
2689 break;
2690 }
2691 default:
2692 ShouldNotReachHere();
2693 }
2694 }
2695 #endif //PRODUCT
2696
2697 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2698
2699 // Get arguments:
2700 Node* receiver = NULL;
2701 Node* base = NULL;
2702 Node* offset = NULL;
2703 Node* oldval = NULL;
2704 Node* newval = NULL;
2705 switch(kind) {
2706 case LS_cmp_swap:
2707 case LS_cmp_swap_weak:
2708 case LS_cmp_exchange: {
2709 const bool two_slot_type = type2size[type] == 2;
2710 receiver = argument(0); // type: oop
2711 base = argument(1); // type: oop
2712 offset = argument(2); // type: long
2713 oldval = argument(4); // type: oop, int, or long
2714 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2715 break;
2716 }
2717 case LS_get_add:
2718 case LS_get_set: {
2719 receiver = argument(0); // type: oop
2720 base = argument(1); // type: oop
2721 offset = argument(2); // type: long
2722 oldval = NULL;
2723 newval = argument(4); // type: oop, int, or long
2724 break;
2725 }
2726 default:
2727 ShouldNotReachHere();
2728 }
2729
2730 // Build field offset expression.
2731 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2732 // to be plain byte offsets, which are also the same as those accepted
2733 // by oopDesc::field_addr.
2734 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2735 // 32-bit machines ignore the high half of long offsets
2736 offset = ConvL2X(offset);
2737 Node* adr = make_unsafe_address(base, offset, ACCESS_WRITE | ACCESS_READ, type, false);
2738 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2739
2740 Compile::AliasType* alias_type = C->alias_type(adr_type);
2741 BasicType bt = alias_type->basic_type();
2742 if (bt != T_ILLEGAL &&
2743 (is_reference_type(bt) != (type == T_OBJECT))) {
2744 // Don't intrinsify mismatched object accesses.
2745 return false;
2746 }
2747
2748 // For CAS, unlike inline_unsafe_access, there seems no point in
2749 // trying to refine types. Just use the coarse types here.
2750 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2751 const Type *value_type = Type::get_const_basic_type(type);
2752
2753 switch (kind) {
2754 case LS_get_set:
2755 case LS_cmp_exchange: {
2756 if (type == T_OBJECT) {
2757 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2758 if (tjp != NULL) {
2759 value_type = tjp;
2760 }
2761 }
2762 break;
2763 }
2764 case LS_cmp_swap:
2765 case LS_cmp_swap_weak:
2766 case LS_get_add:
2767 break;
2768 default:
2769 ShouldNotReachHere();
2770 }
2771
2772 // Null check receiver.
2773 receiver = null_check(receiver);
2774 if (stopped()) {
2775 return true;
2776 }
2777
2778 int alias_idx = C->get_alias_index(adr_type);
2779
2780 if (is_reference_type(type)) {
2781 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2782
2783 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2784 // could be delayed during Parse (for example, in adjust_map_after_if()).
2785 // Execute transformation here to avoid barrier generation in such case.
2786 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2787 newval = _gvn.makecon(TypePtr::NULL_PTR);
2788
2789 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2790 // Refine the value to a null constant, when it is known to be null
2791 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2792 }
2793 }
2794
2795 Node* result = NULL;
2796 switch (kind) {
2797 case LS_cmp_exchange: {
2798 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2799 oldval, newval, value_type, type, decorators);
2800 break;
2801 }
2802 case LS_cmp_swap_weak:
2803 decorators |= C2_WEAK_CMPXCHG;
2804 case LS_cmp_swap: {
2805 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2806 oldval, newval, value_type, type, decorators);
2807 break;
2808 }
2809 case LS_get_set: {
2810 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2811 newval, value_type, type, decorators);
2812 break;
2813 }
2814 case LS_get_add: {
2815 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2816 newval, value_type, type, decorators);
2817 break;
2818 }
2819 default:
2820 ShouldNotReachHere();
2821 }
2822
2823 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2824 set_result(result);
2825 return true;
2826 }
2827
2828 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2829 // Regardless of form, don't allow previous ld/st to move down,
2830 // then issue acquire, release, or volatile mem_bar.
2831 insert_mem_bar(Op_MemBarCPUOrder);
2832 switch(id) {
2833 case vmIntrinsics::_loadFence:
2834 insert_mem_bar(Op_LoadFence);
2835 return true;
2836 case vmIntrinsics::_storeFence:
2837 insert_mem_bar(Op_StoreFence);
2838 return true;
2839 case vmIntrinsics::_fullFence:
2840 insert_mem_bar(Op_MemBarVolatile);
2841 return true;
2842 default:
2843 fatal_unexpected_iid(id);
2844 return false;
2845 }
2846 }
2847
2848 bool LibraryCallKit::inline_onspinwait() {
2849 insert_mem_bar(Op_OnSpinWait);
2850 return true;
2851 }
2852
2853 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2854 if (!kls->is_Con()) {
2855 return true;
2856 }
2857 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
2858 if (klsptr == NULL) {
2859 return true;
2860 }
2861 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
2862 // don't need a guard for a klass that is already initialized
2863 return !ik->is_initialized();
2864 }
2865
2866 //----------------------------inline_unsafe_writeback0-------------------------
2867 // public native void Unsafe.writeback0(long address)
2868 bool LibraryCallKit::inline_unsafe_writeback0() {
2869 if (!Matcher::has_match_rule(Op_CacheWB)) {
2870 return false;
2871 }
2872 #ifndef PRODUCT
2873 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2874 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2875 ciSignature* sig = callee()->signature();
2876 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2877 #endif
2878 null_check_receiver(); // null-check, then ignore
2879 Node *addr = argument(1);
2880 addr = new CastX2PNode(addr);
2881 addr = _gvn.transform(addr);
2882 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2883 flush = _gvn.transform(flush);
2884 set_memory(flush, TypeRawPtr::BOTTOM);
2885 return true;
2886 }
2887
2888 //----------------------------inline_unsafe_writeback0-------------------------
2889 // public native void Unsafe.writeback0(long address)
2890 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2891 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2892 return false;
2893 }
2894 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2895 return false;
2896 }
2897 #ifndef PRODUCT
2898 assert(Matcher::has_match_rule(Op_CacheWB),
2899 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2900 : "found match rule for CacheWBPostSync but not CacheWB"));
2901
2902 #endif
2903 null_check_receiver(); // null-check, then ignore
2904 Node *sync;
2905 if (is_pre) {
2906 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2907 } else {
2908 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2909 }
2910 sync = _gvn.transform(sync);
2911 set_memory(sync, TypeRawPtr::BOTTOM);
2912 return true;
2913 }
2914
2915 //----------------------------inline_unsafe_allocate---------------------------
2916 // public native Object Unsafe.allocateInstance(Class<?> cls);
2917 bool LibraryCallKit::inline_unsafe_allocate() {
2918 if (callee()->is_static()) return false; // caller must have the capability!
2919
2920 null_check_receiver(); // null-check, then ignore
2921 Node* cls = null_check(argument(1));
2922 if (stopped()) return true;
2923
2924 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2925 kls = null_check(kls);
2926 if (stopped()) return true; // argument was like int.class
2927
2928 Node* test = NULL;
2929 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2930 // Note: The argument might still be an illegal value like
2931 // Serializable.class or Object[].class. The runtime will handle it.
2932 // But we must make an explicit check for initialization.
2933 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2934 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2935 // can generate code to load it as unsigned byte.
2936 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2937 Node* bits = intcon(InstanceKlass::fully_initialized);
2938 test = _gvn.transform(new SubINode(inst, bits));
2939 // The 'test' is non-zero if we need to take a slow path.
2940 }
2941
2942 Node* obj = new_instance(kls, test);
2943 set_result(obj);
2944 return true;
2945 }
2946
2947 //------------------------inline_native_time_funcs--------------
2948 // inline code for System.currentTimeMillis() and System.nanoTime()
2949 // these have the same type and signature
2950 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2951 const TypeFunc* tf = OptoRuntime::void_long_Type();
2952 const TypePtr* no_memory_effects = NULL;
2953 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2954 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2955 #ifdef ASSERT
2956 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2957 assert(value_top == top(), "second value must be top");
2958 #endif
2959 set_result(value);
2960 return true;
2961 }
2962
2963 #ifdef JFR_HAVE_INTRINSICS
2964
2965 /*
2966 * oop -> myklass
2967 * myklass->trace_id |= USED
2968 * return myklass->trace_id & ~0x3
2969 */
2970 bool LibraryCallKit::inline_native_classID() {
2971 Node* cls = null_check(argument(0), T_OBJECT);
2972 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2973 kls = null_check(kls, T_OBJECT);
2974
2975 ByteSize offset = KLASS_TRACE_ID_OFFSET;
2976 Node* insp = basic_plus_adr(kls, in_bytes(offset));
2977 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
2978
2979 Node* clsused = longcon(0x01l); // set the class bit
2980 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
2981 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
2982 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
2983
2984 #ifdef TRACE_ID_META_BITS
2985 Node* mbits = longcon(~TRACE_ID_META_BITS);
2986 tvalue = _gvn.transform(new AndLNode(tvalue, mbits));
2987 #endif
2988 #ifdef TRACE_ID_SHIFT
2989 Node* cbits = intcon(TRACE_ID_SHIFT);
2990 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits));
2991 #endif
2992
2993 set_result(tvalue);
2994 return true;
2995
2996 }
2997
2998 bool LibraryCallKit::inline_native_getEventWriter() {
2999 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3000
3001 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
3002 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3003
3004 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3005
3006 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
3007 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
3008
3009 IfNode* iff_jobj_null =
3010 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
3011
3012 enum { _normal_path = 1,
3013 _null_path = 2,
3014 PATH_LIMIT };
3015
3016 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3017 PhiNode* result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3018
3019 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
3020 result_rgn->init_req(_null_path, jobj_is_null);
3021 result_val->init_req(_null_path, null());
3022
3023 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
3024 set_control(jobj_is_not_null);
3025 Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT,
3026 IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3027 result_rgn->init_req(_normal_path, control());
3028 result_val->init_req(_normal_path, res);
3029
3030 set_result(result_rgn, result_val);
3031
3032 return true;
3033 }
3034
3035 #endif // JFR_HAVE_INTRINSICS
3036
3037 //------------------------inline_native_currentThread------------------
3038 bool LibraryCallKit::inline_native_currentThread() {
3039 Node* junk = NULL;
3040 set_result(generate_current_thread(junk));
3041 return true;
3042 }
3043
3044 //---------------------------load_mirror_from_klass----------------------------
3045 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3046 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3047 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3048 Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3049 // mirror = ((OopHandle)mirror)->resolve();
3050 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3051 }
3052
3053 //-----------------------load_klass_from_mirror_common-------------------------
3054 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3055 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3056 // and branch to the given path on the region.
3057 // If never_see_null, take an uncommon trap on null, so we can optimistically
3058 // compile for the non-null case.
3059 // If the region is NULL, force never_see_null = true.
3060 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3061 bool never_see_null,
3062 RegionNode* region,
3063 int null_path,
3064 int offset) {
3065 if (region == NULL) never_see_null = true;
3066 Node* p = basic_plus_adr(mirror, offset);
3067 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3068 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3069 Node* null_ctl = top();
3070 kls = null_check_oop(kls, &null_ctl, never_see_null);
3071 if (region != NULL) {
3072 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3073 region->init_req(null_path, null_ctl);
3074 } else {
3075 assert(null_ctl == top(), "no loose ends");
3076 }
3077 return kls;
3078 }
3079
3080 //--------------------(inline_native_Class_query helpers)---------------------
3081 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3082 // Fall through if (mods & mask) == bits, take the guard otherwise.
3083 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3084 // Branch around if the given klass has the given modifier bit set.
3085 // Like generate_guard, adds a new path onto the region.
3086 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3087 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3088 Node* mask = intcon(modifier_mask);
3089 Node* bits = intcon(modifier_bits);
3090 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3091 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3092 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3093 return generate_fair_guard(bol, region);
3094 }
3095 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3096 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3097 }
3098 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3099 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
3100 }
3101
3102 //-------------------------inline_native_Class_query-------------------
3103 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3104 const Type* return_type = TypeInt::BOOL;
3105 Node* prim_return_value = top(); // what happens if it's a primitive class?
3106 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3107 bool expect_prim = false; // most of these guys expect to work on refs
3108
3109 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3110
3111 Node* mirror = argument(0);
3112 Node* obj = top();
3113
3114 switch (id) {
3115 case vmIntrinsics::_isInstance:
3116 // nothing is an instance of a primitive type
3117 prim_return_value = intcon(0);
3118 obj = argument(1);
3119 break;
3120 case vmIntrinsics::_getModifiers:
3121 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3122 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3123 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3124 break;
3125 case vmIntrinsics::_isInterface:
3126 prim_return_value = intcon(0);
3127 break;
3128 case vmIntrinsics::_isArray:
3129 prim_return_value = intcon(0);
3130 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3131 break;
3132 case vmIntrinsics::_isPrimitive:
3133 prim_return_value = intcon(1);
3134 expect_prim = true; // obviously
3135 break;
3136 case vmIntrinsics::_isHidden:
3137 prim_return_value = intcon(0);
3138 break;
3139 case vmIntrinsics::_getSuperclass:
3140 prim_return_value = null();
3141 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3142 break;
3143 case vmIntrinsics::_getClassAccessFlags:
3144 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3145 return_type = TypeInt::INT; // not bool! 6297094
3146 break;
3147 default:
3148 fatal_unexpected_iid(id);
3149 break;
3150 }
3151
3152 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3153 if (mirror_con == NULL) return false; // cannot happen?
3154
3155 #ifndef PRODUCT
3156 if (C->print_intrinsics() || C->print_inlining()) {
3157 ciType* k = mirror_con->java_mirror_type();
3158 if (k) {
3159 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3160 k->print_name();
3161 tty->cr();
3162 }
3163 }
3164 #endif
3165
3166 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3167 RegionNode* region = new RegionNode(PATH_LIMIT);
3168 record_for_igvn(region);
3169 PhiNode* phi = new PhiNode(region, return_type);
3170
3171 // The mirror will never be null of Reflection.getClassAccessFlags, however
3172 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3173 // if it is. See bug 4774291.
3174
3175 // For Reflection.getClassAccessFlags(), the null check occurs in
3176 // the wrong place; see inline_unsafe_access(), above, for a similar
3177 // situation.
3178 mirror = null_check(mirror);
3179 // If mirror or obj is dead, only null-path is taken.
3180 if (stopped()) return true;
3181
3182 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3183
3184 // Now load the mirror's klass metaobject, and null-check it.
3185 // Side-effects region with the control path if the klass is null.
3186 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3187 // If kls is null, we have a primitive mirror.
3188 phi->init_req(_prim_path, prim_return_value);
3189 if (stopped()) { set_result(region, phi); return true; }
3190 bool safe_for_replace = (region->in(_prim_path) == top());
3191
3192 Node* p; // handy temp
3193 Node* null_ctl;
3194
3195 // Now that we have the non-null klass, we can perform the real query.
3196 // For constant classes, the query will constant-fold in LoadNode::Value.
3197 Node* query_value = top();
3198 switch (id) {
3199 case vmIntrinsics::_isInstance:
3200 // nothing is an instance of a primitive type
3201 query_value = gen_instanceof(obj, kls, safe_for_replace);
3202 break;
3203
3204 case vmIntrinsics::_getModifiers:
3205 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3206 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3207 break;
3208
3209 case vmIntrinsics::_isInterface:
3210 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3211 if (generate_interface_guard(kls, region) != NULL)
3212 // A guard was added. If the guard is taken, it was an interface.
3213 phi->add_req(intcon(1));
3214 // If we fall through, it's a plain class.
3215 query_value = intcon(0);
3216 break;
3217
3218 case vmIntrinsics::_isArray:
3219 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3220 if (generate_array_guard(kls, region) != NULL)
3221 // A guard was added. If the guard is taken, it was an array.
3222 phi->add_req(intcon(1));
3223 // If we fall through, it's a plain class.
3224 query_value = intcon(0);
3225 break;
3226
3227 case vmIntrinsics::_isPrimitive:
3228 query_value = intcon(0); // "normal" path produces false
3229 break;
3230
3231 case vmIntrinsics::_isHidden:
3232 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
3233 if (generate_hidden_class_guard(kls, region) != NULL)
3234 // A guard was added. If the guard is taken, it was an hidden class.
3235 phi->add_req(intcon(1));
3236 // If we fall through, it's a plain class.
3237 query_value = intcon(0);
3238 break;
3239
3240
3241 case vmIntrinsics::_getSuperclass:
3242 // The rules here are somewhat unfortunate, but we can still do better
3243 // with random logic than with a JNI call.
3244 // Interfaces store null or Object as _super, but must report null.
3245 // Arrays store an intermediate super as _super, but must report Object.
3246 // Other types can report the actual _super.
3247 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3248 if (generate_interface_guard(kls, region) != NULL)
3249 // A guard was added. If the guard is taken, it was an interface.
3250 phi->add_req(null());
3251 if (generate_array_guard(kls, region) != NULL)
3252 // A guard was added. If the guard is taken, it was an array.
3253 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3254 // If we fall through, it's a plain class. Get its _super.
3255 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3256 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3257 null_ctl = top();
3258 kls = null_check_oop(kls, &null_ctl);
3259 if (null_ctl != top()) {
3260 // If the guard is taken, Object.superClass is null (both klass and mirror).
3261 region->add_req(null_ctl);
3262 phi ->add_req(null());
3263 }
3264 if (!stopped()) {
3265 query_value = load_mirror_from_klass(kls);
3266 }
3267 break;
3268
3269 case vmIntrinsics::_getClassAccessFlags:
3270 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3271 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3272 break;
3273
3274 default:
3275 fatal_unexpected_iid(id);
3276 break;
3277 }
3278
3279 // Fall-through is the normal case of a query to a real class.
3280 phi->init_req(1, query_value);
3281 region->init_req(1, control());
3282
3283 C->set_has_split_ifs(true); // Has chance for split-if optimization
3284 set_result(region, phi);
3285 return true;
3286 }
3287
3288 //-------------------------inline_Class_cast-------------------
3289 bool LibraryCallKit::inline_Class_cast() {
3290 Node* mirror = argument(0); // Class
3291 Node* obj = argument(1);
3292 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3293 if (mirror_con == NULL) {
3294 return false; // dead path (mirror->is_top()).
3295 }
3296 if (obj == NULL || obj->is_top()) {
3297 return false; // dead path
3298 }
3299 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3300
3301 // First, see if Class.cast() can be folded statically.
3302 // java_mirror_type() returns non-null for compile-time Class constants.
3303 ciType* tm = mirror_con->java_mirror_type();
3304 if (tm != NULL && tm->is_klass() &&
3305 tp != NULL && tp->klass() != NULL) {
3306 if (!tp->klass()->is_loaded()) {
3307 // Don't use intrinsic when class is not loaded.
3308 return false;
3309 } else {
3310 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3311 if (static_res == Compile::SSC_always_true) {
3312 // isInstance() is true - fold the code.
3313 set_result(obj);
3314 return true;
3315 } else if (static_res == Compile::SSC_always_false) {
3316 // Don't use intrinsic, have to throw ClassCastException.
3317 // If the reference is null, the non-intrinsic bytecode will
3318 // be optimized appropriately.
3319 return false;
3320 }
3321 }
3322 }
3323
3324 // Bailout intrinsic and do normal inlining if exception path is frequent.
3325 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3326 return false;
3327 }
3328
3329 // Generate dynamic checks.
3330 // Class.cast() is java implementation of _checkcast bytecode.
3331 // Do checkcast (Parse::do_checkcast()) optimizations here.
3332
3333 mirror = null_check(mirror);
3334 // If mirror is dead, only null-path is taken.
3335 if (stopped()) {
3336 return true;
3337 }
3338
3339 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3340 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3341 RegionNode* region = new RegionNode(PATH_LIMIT);
3342 record_for_igvn(region);
3343
3344 // Now load the mirror's klass metaobject, and null-check it.
3345 // If kls is null, we have a primitive mirror and
3346 // nothing is an instance of a primitive type.
3347 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3348
3349 Node* res = top();
3350 if (!stopped()) {
3351 Node* bad_type_ctrl = top();
3352 // Do checkcast optimizations.
3353 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3354 region->init_req(_bad_type_path, bad_type_ctrl);
3355 }
3356 if (region->in(_prim_path) != top() ||
3357 region->in(_bad_type_path) != top()) {
3358 // Let Interpreter throw ClassCastException.
3359 PreserveJVMState pjvms(this);
3360 set_control(_gvn.transform(region));
3361 uncommon_trap(Deoptimization::Reason_intrinsic,
3362 Deoptimization::Action_maybe_recompile);
3363 }
3364 if (!stopped()) {
3365 set_result(res);
3366 }
3367 return true;
3368 }
3369
3370
3371 //--------------------------inline_native_subtype_check------------------------
3372 // This intrinsic takes the JNI calls out of the heart of
3373 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3374 bool LibraryCallKit::inline_native_subtype_check() {
3375 // Pull both arguments off the stack.
3376 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3377 args[0] = argument(0);
3378 args[1] = argument(1);
3379 Node* klasses[2]; // corresponding Klasses: superk, subk
3380 klasses[0] = klasses[1] = top();
3381
3382 enum {
3383 // A full decision tree on {superc is prim, subc is prim}:
3384 _prim_0_path = 1, // {P,N} => false
3385 // {P,P} & superc!=subc => false
3386 _prim_same_path, // {P,P} & superc==subc => true
3387 _prim_1_path, // {N,P} => false
3388 _ref_subtype_path, // {N,N} & subtype check wins => true
3389 _both_ref_path, // {N,N} & subtype check loses => false
3390 PATH_LIMIT
3391 };
3392
3393 RegionNode* region = new RegionNode(PATH_LIMIT);
3394 Node* phi = new PhiNode(region, TypeInt::BOOL);
3395 record_for_igvn(region);
3396
3397 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3398 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3399 int class_klass_offset = java_lang_Class::klass_offset();
3400
3401 // First null-check both mirrors and load each mirror's klass metaobject.
3402 int which_arg;
3403 for (which_arg = 0; which_arg <= 1; which_arg++) {
3404 Node* arg = args[which_arg];
3405 arg = null_check(arg);
3406 if (stopped()) break;
3407 args[which_arg] = arg;
3408
3409 Node* p = basic_plus_adr(arg, class_klass_offset);
3410 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3411 klasses[which_arg] = _gvn.transform(kls);
3412 }
3413
3414 // Having loaded both klasses, test each for null.
3415 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3416 for (which_arg = 0; which_arg <= 1; which_arg++) {
3417 Node* kls = klasses[which_arg];
3418 Node* null_ctl = top();
3419 kls = null_check_oop(kls, &null_ctl, never_see_null);
3420 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3421 region->init_req(prim_path, null_ctl);
3422 if (stopped()) break;
3423 klasses[which_arg] = kls;
3424 }
3425
3426 if (!stopped()) {
3427 // now we have two reference types, in klasses[0..1]
3428 Node* subk = klasses[1]; // the argument to isAssignableFrom
3429 Node* superk = klasses[0]; // the receiver
3430 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3431 // now we have a successful reference subtype check
3432 region->set_req(_ref_subtype_path, control());
3433 }
3434
3435 // If both operands are primitive (both klasses null), then
3436 // we must return true when they are identical primitives.
3437 // It is convenient to test this after the first null klass check.
3438 set_control(region->in(_prim_0_path)); // go back to first null check
3439 if (!stopped()) {
3440 // Since superc is primitive, make a guard for the superc==subc case.
3441 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3442 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3443 generate_guard(bol_eq, region, PROB_FAIR);
3444 if (region->req() == PATH_LIMIT+1) {
3445 // A guard was added. If the added guard is taken, superc==subc.
3446 region->swap_edges(PATH_LIMIT, _prim_same_path);
3447 region->del_req(PATH_LIMIT);
3448 }
3449 region->set_req(_prim_0_path, control()); // Not equal after all.
3450 }
3451
3452 // these are the only paths that produce 'true':
3453 phi->set_req(_prim_same_path, intcon(1));
3454 phi->set_req(_ref_subtype_path, intcon(1));
3455
3456 // pull together the cases:
3457 assert(region->req() == PATH_LIMIT, "sane region");
3458 for (uint i = 1; i < region->req(); i++) {
3459 Node* ctl = region->in(i);
3460 if (ctl == NULL || ctl == top()) {
3461 region->set_req(i, top());
3462 phi ->set_req(i, top());
3463 } else if (phi->in(i) == NULL) {
3464 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3465 }
3466 }
3467
3468 set_control(_gvn.transform(region));
3469 set_result(_gvn.transform(phi));
3470 return true;
3471 }
3472
3473 //---------------------generate_array_guard_common------------------------
3474 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3475 bool obj_array, bool not_array) {
3476
3477 if (stopped()) {
3478 return NULL;
3479 }
3480
3481 // If obj_array/non_array==false/false:
3482 // Branch around if the given klass is in fact an array (either obj or prim).
3483 // If obj_array/non_array==false/true:
3484 // Branch around if the given klass is not an array klass of any kind.
3485 // If obj_array/non_array==true/true:
3486 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3487 // If obj_array/non_array==true/false:
3488 // Branch around if the kls is an oop array (Object[] or subtype)
3489 //
3490 // Like generate_guard, adds a new path onto the region.
3491 jint layout_con = 0;
3492 Node* layout_val = get_layout_helper(kls, layout_con);
3493 if (layout_val == NULL) {
3494 bool query = (obj_array
3495 ? Klass::layout_helper_is_objArray(layout_con)
3496 : Klass::layout_helper_is_array(layout_con));
3497 if (query == not_array) {
3498 return NULL; // never a branch
3499 } else { // always a branch
3500 Node* always_branch = control();
3501 if (region != NULL)
3502 region->add_req(always_branch);
3503 set_control(top());
3504 return always_branch;
3505 }
3506 }
3507 // Now test the correct condition.
3508 jint nval = (obj_array
3509 ? (jint)(Klass::_lh_array_tag_type_value
3510 << Klass::_lh_array_tag_shift)
3511 : Klass::_lh_neutral_value);
3512 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3513 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3514 // invert the test if we are looking for a non-array
3515 if (not_array) btest = BoolTest(btest).negate();
3516 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3517 return generate_fair_guard(bol, region);
3518 }
3519
3520
3521 //-----------------------inline_native_newArray--------------------------
3522 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3523 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3524 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3525 Node* mirror;
3526 Node* count_val;
3527 if (uninitialized) {
3528 mirror = argument(1);
3529 count_val = argument(2);
3530 } else {
3531 mirror = argument(0);
3532 count_val = argument(1);
3533 }
3534
3535 mirror = null_check(mirror);
3536 // If mirror or obj is dead, only null-path is taken.
3537 if (stopped()) return true;
3538
3539 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3540 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3541 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3542 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3543 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3544
3545 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3546 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3547 result_reg, _slow_path);
3548 Node* normal_ctl = control();
3549 Node* no_array_ctl = result_reg->in(_slow_path);
3550
3551 // Generate code for the slow case. We make a call to newArray().
3552 set_control(no_array_ctl);
3553 if (!stopped()) {
3554 // Either the input type is void.class, or else the
3555 // array klass has not yet been cached. Either the
3556 // ensuing call will throw an exception, or else it
3557 // will cache the array klass for next time.
3558 PreserveJVMState pjvms(this);
3559 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3560 Node* slow_result = set_results_for_java_call(slow_call);
3561 // this->control() comes from set_results_for_java_call
3562 result_reg->set_req(_slow_path, control());
3563 result_val->set_req(_slow_path, slow_result);
3564 result_io ->set_req(_slow_path, i_o());
3565 result_mem->set_req(_slow_path, reset_memory());
3566 }
3567
3568 set_control(normal_ctl);
3569 if (!stopped()) {
3570 // Normal case: The array type has been cached in the java.lang.Class.
3571 // The following call works fine even if the array type is polymorphic.
3572 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3573 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3574 result_reg->init_req(_normal_path, control());
3575 result_val->init_req(_normal_path, obj);
3576 result_io ->init_req(_normal_path, i_o());
3577 result_mem->init_req(_normal_path, reset_memory());
3578
3579 if (uninitialized) {
3580 // Mark the allocation so that zeroing is skipped
3581 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3582 alloc->maybe_set_complete(&_gvn);
3583 }
3584 }
3585
3586 // Return the combined state.
3587 set_i_o( _gvn.transform(result_io) );
3588 set_all_memory( _gvn.transform(result_mem));
3589
3590 C->set_has_split_ifs(true); // Has chance for split-if optimization
3591 set_result(result_reg, result_val);
3592 return true;
3593 }
3594
3595 //----------------------inline_native_getLength--------------------------
3596 // public static native int java.lang.reflect.Array.getLength(Object array);
3597 bool LibraryCallKit::inline_native_getLength() {
3598 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3599
3600 Node* array = null_check(argument(0));
3601 // If array is dead, only null-path is taken.
3602 if (stopped()) return true;
3603
3604 // Deoptimize if it is a non-array.
3605 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3606
3607 if (non_array != NULL) {
3608 PreserveJVMState pjvms(this);
3609 set_control(non_array);
3610 uncommon_trap(Deoptimization::Reason_intrinsic,
3611 Deoptimization::Action_maybe_recompile);
3612 }
3613
3614 // If control is dead, only non-array-path is taken.
3615 if (stopped()) return true;
3616
3617 // The works fine even if the array type is polymorphic.
3618 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3619 Node* result = load_array_length(array);
3620
3621 C->set_has_split_ifs(true); // Has chance for split-if optimization
3622 set_result(result);
3623 return true;
3624 }
3625
3626 //------------------------inline_array_copyOf----------------------------
3627 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3628 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3629 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3630 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3631
3632 // Get the arguments.
3633 Node* original = argument(0);
3634 Node* start = is_copyOfRange? argument(1): intcon(0);
3635 Node* end = is_copyOfRange? argument(2): argument(1);
3636 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3637
3638 Node* newcopy = NULL;
3639
3640 // Set the original stack and the reexecute bit for the interpreter to reexecute
3641 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3642 { PreserveReexecuteState preexecs(this);
3643 jvms()->set_should_reexecute(true);
3644
3645 array_type_mirror = null_check(array_type_mirror);
3646 original = null_check(original);
3647
3648 // Check if a null path was taken unconditionally.
3649 if (stopped()) return true;
3650
3651 Node* orig_length = load_array_length(original);
3652
3653 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3654 klass_node = null_check(klass_node);
3655
3656 RegionNode* bailout = new RegionNode(1);
3657 record_for_igvn(bailout);
3658
3659 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3660 // Bail out if that is so.
3661 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3662 if (not_objArray != NULL) {
3663 // Improve the klass node's type from the new optimistic assumption:
3664 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3665 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3666 Node* cast = new CastPPNode(klass_node, akls);
3667 cast->init_req(0, control());
3668 klass_node = _gvn.transform(cast);
3669 }
3670
3671 // Bail out if either start or end is negative.
3672 generate_negative_guard(start, bailout, &start);
3673 generate_negative_guard(end, bailout, &end);
3674
3675 Node* length = end;
3676 if (_gvn.type(start) != TypeInt::ZERO) {
3677 length = _gvn.transform(new SubINode(end, start));
3678 }
3679
3680 // Bail out if length is negative.
3681 // Without this the new_array would throw
3682 // NegativeArraySizeException but IllegalArgumentException is what
3683 // should be thrown
3684 generate_negative_guard(length, bailout, &length);
3685
3686 if (bailout->req() > 1) {
3687 PreserveJVMState pjvms(this);
3688 set_control(_gvn.transform(bailout));
3689 uncommon_trap(Deoptimization::Reason_intrinsic,
3690 Deoptimization::Action_maybe_recompile);
3691 }
3692
3693 if (!stopped()) {
3694 // How many elements will we copy from the original?
3695 // The answer is MinI(orig_length - start, length).
3696 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3697 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3698
3699 // Generate a direct call to the right arraycopy function(s).
3700 // We know the copy is disjoint but we might not know if the
3701 // oop stores need checking.
3702 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3703 // This will fail a store-check if x contains any non-nulls.
3704
3705 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3706 // loads/stores but it is legal only if we're sure the
3707 // Arrays.copyOf would succeed. So we need all input arguments
3708 // to the copyOf to be validated, including that the copy to the
3709 // new array won't trigger an ArrayStoreException. That subtype
3710 // check can be optimized if we know something on the type of
3711 // the input array from type speculation.
3712 if (_gvn.type(klass_node)->singleton()) {
3713 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3714 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3715
3716 int test = C->static_subtype_check(superk, subk);
3717 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3718 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3719 if (t_original->speculative_type() != NULL) {
3720 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3721 }
3722 }
3723 }
3724
3725 bool validated = false;
3726 // Reason_class_check rather than Reason_intrinsic because we
3727 // want to intrinsify even if this traps.
3728 if (!too_many_traps(Deoptimization::Reason_class_check)) {
3729 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
3730
3731 if (not_subtype_ctrl != top()) {
3732 PreserveJVMState pjvms(this);
3733 set_control(not_subtype_ctrl);
3734 uncommon_trap(Deoptimization::Reason_class_check,
3735 Deoptimization::Action_make_not_entrant);
3736 assert(stopped(), "Should be stopped");
3737 }
3738 validated = true;
3739 }
3740
3741 if (!stopped()) {
3742 newcopy = new_array(klass_node, length, 0); // no arguments to push
3743
3744 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
3745 load_object_klass(original), klass_node);
3746 if (!is_copyOfRange) {
3747 ac->set_copyof(validated);
3748 } else {
3749 ac->set_copyofrange(validated);
3750 }
3751 Node* n = _gvn.transform(ac);
3752 if (n == ac) {
3753 ac->connect_outputs(this);
3754 } else {
3755 assert(validated, "shouldn't transform if all arguments not validated");
3756 set_all_memory(n);
3757 }
3758 }
3759 }
3760 } // original reexecute is set back here
3761
3762 C->set_has_split_ifs(true); // Has chance for split-if optimization
3763 if (!stopped()) {
3764 set_result(newcopy);
3765 }
3766 return true;
3767 }
3768
3769
3770 //----------------------generate_virtual_guard---------------------------
3771 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
3772 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
3773 RegionNode* slow_region) {
3774 ciMethod* method = callee();
3775 int vtable_index = method->vtable_index();
3776 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3777 "bad index %d", vtable_index);
3778 // Get the Method* out of the appropriate vtable entry.
3779 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
3780 vtable_index*vtableEntry::size_in_bytes() +
3781 vtableEntry::method_offset_in_bytes();
3782 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
3783 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3784
3785 // Compare the target method with the expected method (e.g., Object.hashCode).
3786 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
3787
3788 Node* native_call = makecon(native_call_addr);
3789 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
3790 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
3791
3792 return generate_slow_guard(test_native, slow_region);
3793 }
3794
3795 //-----------------------generate_method_call----------------------------
3796 // Use generate_method_call to make a slow-call to the real
3797 // method if the fast path fails. An alternative would be to
3798 // use a stub like OptoRuntime::slow_arraycopy_Java.
3799 // This only works for expanding the current library call,
3800 // not another intrinsic. (E.g., don't use this for making an
3801 // arraycopy call inside of the copyOf intrinsic.)
3802 CallJavaNode*
3803 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
3804 // When compiling the intrinsic method itself, do not use this technique.
3805 guarantee(callee() != C->method(), "cannot make slow-call to self");
3806
3807 ciMethod* method = callee();
3808 // ensure the JVMS we have will be correct for this call
3809 guarantee(method_id == method->intrinsic_id(), "must match");
3810
3811 const TypeFunc* tf = TypeFunc::make(method);
3812 CallJavaNode* slow_call;
3813 if (is_static) {
3814 assert(!is_virtual, "");
3815 slow_call = new CallStaticJavaNode(C, tf,
3816 SharedRuntime::get_resolve_static_call_stub(),
3817 method, bci());
3818 } else if (is_virtual) {
3819 null_check_receiver();
3820 int vtable_index = Method::invalid_vtable_index;
3821 if (UseInlineCaches) {
3822 // Suppress the vtable call
3823 } else {
3824 // hashCode and clone are not a miranda methods,
3825 // so the vtable index is fixed.
3826 // No need to use the linkResolver to get it.
3827 vtable_index = method->vtable_index();
3828 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3829 "bad index %d", vtable_index);
3830 }
3831 slow_call = new CallDynamicJavaNode(tf,
3832 SharedRuntime::get_resolve_virtual_call_stub(),
3833 method, vtable_index, bci());
3834 } else { // neither virtual nor static: opt_virtual
3835 null_check_receiver();
3836 slow_call = new CallStaticJavaNode(C, tf,
3837 SharedRuntime::get_resolve_opt_virtual_call_stub(),
3838 method, bci());
3839 slow_call->set_optimized_virtual(true);
3840 }
3841 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
3842 // To be able to issue a direct call (optimized virtual or virtual)
3843 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
3844 // about the method being invoked should be attached to the call site to
3845 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
3846 slow_call->set_override_symbolic_info(true);
3847 }
3848 set_arguments_for_java_call(slow_call);
3849 set_edges_for_java_call(slow_call);
3850 return slow_call;
3851 }
3852
3853
3854 /**
3855 * Build special case code for calls to hashCode on an object. This call may
3856 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
3857 * slightly different code.
3858 */
3859 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
3860 assert(is_static == callee()->is_static(), "correct intrinsic selection");
3861 assert(!(is_virtual && is_static), "either virtual, special, or static");
3862
3863 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
3864
3865 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3866 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
3867 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3868 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3869 Node* obj = NULL;
3870 if (!is_static) {
3871 // Check for hashing null object
3872 obj = null_check_receiver();
3873 if (stopped()) return true; // unconditionally null
3874 result_reg->init_req(_null_path, top());
3875 result_val->init_req(_null_path, top());
3876 } else {
3877 // Do a null check, and return zero if null.
3878 // System.identityHashCode(null) == 0
3879 obj = argument(0);
3880 Node* null_ctl = top();
3881 obj = null_check_oop(obj, &null_ctl);
3882 result_reg->init_req(_null_path, null_ctl);
3883 result_val->init_req(_null_path, _gvn.intcon(0));
3884 }
3885
3886 // Unconditionally null? Then return right away.
3887 if (stopped()) {
3888 set_control( result_reg->in(_null_path));
3889 if (!stopped())
3890 set_result(result_val->in(_null_path));
3891 return true;
3892 }
3893
3894 // We only go to the fast case code if we pass a number of guards. The
3895 // paths which do not pass are accumulated in the slow_region.
3896 RegionNode* slow_region = new RegionNode(1);
3897 record_for_igvn(slow_region);
3898
3899 // If this is a virtual call, we generate a funny guard. We pull out
3900 // the vtable entry corresponding to hashCode() from the target object.
3901 // If the target method which we are calling happens to be the native
3902 // Object hashCode() method, we pass the guard. We do not need this
3903 // guard for non-virtual calls -- the caller is known to be the native
3904 // Object hashCode().
3905 if (is_virtual) {
3906 // After null check, get the object's klass.
3907 Node* obj_klass = load_object_klass(obj);
3908 generate_virtual_guard(obj_klass, slow_region);
3909 }
3910
3911 // Get the header out of the object, use LoadMarkNode when available
3912 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
3913 // The control of the load must be NULL. Otherwise, the load can move before
3914 // the null check after castPP removal.
3915 Node* no_ctrl = NULL;
3916 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
3917
3918 // Test the header to see if it is unlocked.
3919 Node *lock_mask = _gvn.MakeConX(markWord::biased_lock_mask_in_place);
3920 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
3921 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
3922 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
3923 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
3924
3925 generate_slow_guard(test_unlocked, slow_region);
3926
3927 // Get the hash value and check to see that it has been properly assigned.
3928 // We depend on hash_mask being at most 32 bits and avoid the use of
3929 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
3930 // vm: see markWord.hpp.
3931 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
3932 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
3933 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
3934 // This hack lets the hash bits live anywhere in the mark object now, as long
3935 // as the shift drops the relevant bits into the low 32 bits. Note that
3936 // Java spec says that HashCode is an int so there's no point in capturing
3937 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
3938 hshifted_header = ConvX2I(hshifted_header);
3939 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
3940
3941 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
3942 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
3943 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
3944
3945 generate_slow_guard(test_assigned, slow_region);
3946
3947 Node* init_mem = reset_memory();
3948 // fill in the rest of the null path:
3949 result_io ->init_req(_null_path, i_o());
3950 result_mem->init_req(_null_path, init_mem);
3951
3952 result_val->init_req(_fast_path, hash_val);
3953 result_reg->init_req(_fast_path, control());
3954 result_io ->init_req(_fast_path, i_o());
3955 result_mem->init_req(_fast_path, init_mem);
3956
3957 // Generate code for the slow case. We make a call to hashCode().
3958 set_control(_gvn.transform(slow_region));
3959 if (!stopped()) {
3960 // No need for PreserveJVMState, because we're using up the present state.
3961 set_all_memory(init_mem);
3962 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
3963 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
3964 Node* slow_result = set_results_for_java_call(slow_call);
3965 // this->control() comes from set_results_for_java_call
3966 result_reg->init_req(_slow_path, control());
3967 result_val->init_req(_slow_path, slow_result);
3968 result_io ->set_req(_slow_path, i_o());
3969 result_mem ->set_req(_slow_path, reset_memory());
3970 }
3971
3972 // Return the combined state.
3973 set_i_o( _gvn.transform(result_io) );
3974 set_all_memory( _gvn.transform(result_mem));
3975
3976 set_result(result_reg, result_val);
3977 return true;
3978 }
3979
3980 //---------------------------inline_native_getClass----------------------------
3981 // public final native Class<?> java.lang.Object.getClass();
3982 //
3983 // Build special case code for calls to getClass on an object.
3984 bool LibraryCallKit::inline_native_getClass() {
3985 Node* obj = null_check_receiver();
3986 if (stopped()) return true;
3987 set_result(load_mirror_from_klass(load_object_klass(obj)));
3988 return true;
3989 }
3990
3991 //-----------------inline_native_Reflection_getCallerClass---------------------
3992 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
3993 //
3994 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
3995 //
3996 // NOTE: This code must perform the same logic as JVM_GetCallerClass
3997 // in that it must skip particular security frames and checks for
3998 // caller sensitive methods.
3999 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4000 #ifndef PRODUCT
4001 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4002 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4003 }
4004 #endif
4005
4006 if (!jvms()->has_method()) {
4007 #ifndef PRODUCT
4008 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4009 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4010 }
4011 #endif
4012 return false;
4013 }
4014
4015 // Walk back up the JVM state to find the caller at the required
4016 // depth.
4017 JVMState* caller_jvms = jvms();
4018
4019 // Cf. JVM_GetCallerClass
4020 // NOTE: Start the loop at depth 1 because the current JVM state does
4021 // not include the Reflection.getCallerClass() frame.
4022 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4023 ciMethod* m = caller_jvms->method();
4024 switch (n) {
4025 case 0:
4026 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4027 break;
4028 case 1:
4029 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4030 if (!m->caller_sensitive()) {
4031 #ifndef PRODUCT
4032 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4033 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4034 }
4035 #endif
4036 return false; // bail-out; let JVM_GetCallerClass do the work
4037 }
4038 break;
4039 default:
4040 if (!m->is_ignored_by_security_stack_walk()) {
4041 // We have reached the desired frame; return the holder class.
4042 // Acquire method holder as java.lang.Class and push as constant.
4043 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4044 ciInstance* caller_mirror = caller_klass->java_mirror();
4045 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4046
4047 #ifndef PRODUCT
4048 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4049 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());
4050 tty->print_cr(" JVM state at this point:");
4051 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4052 ciMethod* m = jvms()->of_depth(i)->method();
4053 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4054 }
4055 }
4056 #endif
4057 return true;
4058 }
4059 break;
4060 }
4061 }
4062
4063 #ifndef PRODUCT
4064 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4065 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4066 tty->print_cr(" JVM state at this point:");
4067 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4068 ciMethod* m = jvms()->of_depth(i)->method();
4069 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4070 }
4071 }
4072 #endif
4073
4074 return false; // bail-out; let JVM_GetCallerClass do the work
4075 }
4076
4077 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4078 Node* arg = argument(0);
4079 Node* result = NULL;
4080
4081 switch (id) {
4082 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4083 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4084 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4085 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4086
4087 case vmIntrinsics::_doubleToLongBits: {
4088 // two paths (plus control) merge in a wood
4089 RegionNode *r = new RegionNode(3);
4090 Node *phi = new PhiNode(r, TypeLong::LONG);
4091
4092 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4093 // Build the boolean node
4094 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4095
4096 // Branch either way.
4097 // NaN case is less traveled, which makes all the difference.
4098 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4099 Node *opt_isnan = _gvn.transform(ifisnan);
4100 assert( opt_isnan->is_If(), "Expect an IfNode");
4101 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4102 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4103
4104 set_control(iftrue);
4105
4106 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4107 Node *slow_result = longcon(nan_bits); // return NaN
4108 phi->init_req(1, _gvn.transform( slow_result ));
4109 r->init_req(1, iftrue);
4110
4111 // Else fall through
4112 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4113 set_control(iffalse);
4114
4115 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4116 r->init_req(2, iffalse);
4117
4118 // Post merge
4119 set_control(_gvn.transform(r));
4120 record_for_igvn(r);
4121
4122 C->set_has_split_ifs(true); // Has chance for split-if optimization
4123 result = phi;
4124 assert(result->bottom_type()->isa_long(), "must be");
4125 break;
4126 }
4127
4128 case vmIntrinsics::_floatToIntBits: {
4129 // two paths (plus control) merge in a wood
4130 RegionNode *r = new RegionNode(3);
4131 Node *phi = new PhiNode(r, TypeInt::INT);
4132
4133 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4134 // Build the boolean node
4135 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4136
4137 // Branch either way.
4138 // NaN case is less traveled, which makes all the difference.
4139 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4140 Node *opt_isnan = _gvn.transform(ifisnan);
4141 assert( opt_isnan->is_If(), "Expect an IfNode");
4142 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4143 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4144
4145 set_control(iftrue);
4146
4147 static const jint nan_bits = 0x7fc00000;
4148 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4149 phi->init_req(1, _gvn.transform( slow_result ));
4150 r->init_req(1, iftrue);
4151
4152 // Else fall through
4153 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4154 set_control(iffalse);
4155
4156 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4157 r->init_req(2, iffalse);
4158
4159 // Post merge
4160 set_control(_gvn.transform(r));
4161 record_for_igvn(r);
4162
4163 C->set_has_split_ifs(true); // Has chance for split-if optimization
4164 result = phi;
4165 assert(result->bottom_type()->isa_int(), "must be");
4166 break;
4167 }
4168
4169 default:
4170 fatal_unexpected_iid(id);
4171 break;
4172 }
4173 set_result(_gvn.transform(result));
4174 return true;
4175 }
4176
4177 //----------------------inline_unsafe_copyMemory-------------------------
4178 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4179 bool LibraryCallKit::inline_unsafe_copyMemory() {
4180 if (callee()->is_static()) return false; // caller must have the capability!
4181 null_check_receiver(); // null-check receiver
4182 if (stopped()) return true;
4183
4184 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4185
4186 Node* src_ptr = argument(1); // type: oop
4187 Node* src_off = ConvL2X(argument(2)); // type: long
4188 Node* dst_ptr = argument(4); // type: oop
4189 Node* dst_off = ConvL2X(argument(5)); // type: long
4190 Node* size = ConvL2X(argument(7)); // type: long
4191
4192 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4193 "fieldOffset must be byte-scaled");
4194
4195 Node* src = make_unsafe_address(src_ptr, src_off, ACCESS_READ);
4196 Node* dst = make_unsafe_address(dst_ptr, dst_off, ACCESS_WRITE);
4197
4198 // Conservatively insert a memory barrier on all memory slices.
4199 // Do not let writes of the copy source or destination float below the copy.
4200 insert_mem_bar(Op_MemBarCPUOrder);
4201
4202 Node* thread = _gvn.transform(new ThreadLocalNode());
4203 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4204 BasicType doing_unsafe_access_bt = T_BYTE;
4205 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4206
4207 // update volatile field
4208 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4209
4210 // Call it. Note that the length argument is not scaled.
4211 make_runtime_call(RC_LEAF|RC_NO_FP,
4212 OptoRuntime::fast_arraycopy_Type(),
4213 StubRoutines::unsafe_arraycopy(),
4214 "unsafe_arraycopy",
4215 TypeRawPtr::BOTTOM,
4216 src, dst, size XTOP);
4217
4218 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4219
4220 // Do not let reads of the copy destination float above the copy.
4221 insert_mem_bar(Op_MemBarCPUOrder);
4222
4223 return true;
4224 }
4225
4226 //------------------------clone_coping-----------------------------------
4227 // Helper function for inline_native_clone.
4228 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4229 assert(obj_size != NULL, "");
4230 Node* raw_obj = alloc_obj->in(1);
4231 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4232
4233 AllocateNode* alloc = NULL;
4234 if (ReduceBulkZeroing) {
4235 // We will be completely responsible for initializing this object -
4236 // mark Initialize node as complete.
4237 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4238 // The object was just allocated - there should be no any stores!
4239 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4240 // Mark as complete_with_arraycopy so that on AllocateNode
4241 // expansion, we know this AllocateNode is initialized by an array
4242 // copy and a StoreStore barrier exists after the array copy.
4243 alloc->initialization()->set_complete_with_arraycopy();
4244 }
4245
4246 Node* size = _gvn.transform(obj_size);
4247 access_clone(obj, alloc_obj, size, is_array);
4248
4249 // Do not let reads from the cloned object float above the arraycopy.
4250 if (alloc != NULL) {
4251 // Do not let stores that initialize this object be reordered with
4252 // a subsequent store that would make this object accessible by
4253 // other threads.
4254 // Record what AllocateNode this StoreStore protects so that
4255 // escape analysis can go from the MemBarStoreStoreNode to the
4256 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4257 // based on the escape status of the AllocateNode.
4258 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4259 } else {
4260 insert_mem_bar(Op_MemBarCPUOrder);
4261 }
4262 }
4263
4264 //------------------------inline_native_clone----------------------------
4265 // protected native Object java.lang.Object.clone();
4266 //
4267 // Here are the simple edge cases:
4268 // null receiver => normal trap
4269 // virtual and clone was overridden => slow path to out-of-line clone
4270 // not cloneable or finalizer => slow path to out-of-line Object.clone
4271 //
4272 // The general case has two steps, allocation and copying.
4273 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4274 //
4275 // Copying also has two cases, oop arrays and everything else.
4276 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4277 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4278 //
4279 // These steps fold up nicely if and when the cloned object's klass
4280 // can be sharply typed as an object array, a type array, or an instance.
4281 //
4282 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4283 PhiNode* result_val;
4284
4285 // Set the reexecute bit for the interpreter to reexecute
4286 // the bytecode that invokes Object.clone if deoptimization happens.
4287 { PreserveReexecuteState preexecs(this);
4288 jvms()->set_should_reexecute(true);
4289
4290 Node* obj = null_check_receiver();
4291 if (stopped()) return true;
4292
4293 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4294
4295 // If we are going to clone an instance, we need its exact type to
4296 // know the number and types of fields to convert the clone to
4297 // loads/stores. Maybe a speculative type can help us.
4298 if (!obj_type->klass_is_exact() &&
4299 obj_type->speculative_type() != NULL &&
4300 obj_type->speculative_type()->is_instance_klass()) {
4301 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4302 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4303 !spec_ik->has_injected_fields()) {
4304 ciKlass* k = obj_type->klass();
4305 if (!k->is_instance_klass() ||
4306 k->as_instance_klass()->is_interface() ||
4307 k->as_instance_klass()->has_subklass()) {
4308 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4309 }
4310 }
4311 }
4312
4313 // Conservatively insert a memory barrier on all memory slices.
4314 // Do not let writes into the original float below the clone.
4315 insert_mem_bar(Op_MemBarCPUOrder);
4316
4317 // paths into result_reg:
4318 enum {
4319 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4320 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4321 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4322 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4323 PATH_LIMIT
4324 };
4325 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4326 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4327 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4328 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4329 record_for_igvn(result_reg);
4330
4331 Node* obj_klass = load_object_klass(obj);
4332 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4333 if (array_ctl != NULL) {
4334 // It's an array.
4335 PreserveJVMState pjvms(this);
4336 set_control(array_ctl);
4337 Node* obj_length = load_array_length(obj);
4338 Node* obj_size = NULL;
4339 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, /*deoptimize_on_exception=*/true);
4340
4341 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4342 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, BarrierSetC2::Parsing)) {
4343 // If it is an oop array, it requires very special treatment,
4344 // because gc barriers are required when accessing the array.
4345 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4346 if (is_obja != NULL) {
4347 PreserveJVMState pjvms2(this);
4348 set_control(is_obja);
4349 // Generate a direct call to the right arraycopy function(s).
4350 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4351 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false);
4352 ac->set_clone_oop_array();
4353 Node* n = _gvn.transform(ac);
4354 assert(n == ac, "cannot disappear");
4355 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
4356
4357 result_reg->init_req(_objArray_path, control());
4358 result_val->init_req(_objArray_path, alloc_obj);
4359 result_i_o ->set_req(_objArray_path, i_o());
4360 result_mem ->set_req(_objArray_path, reset_memory());
4361 }
4362 }
4363 // Otherwise, there are no barriers to worry about.
4364 // (We can dispense with card marks if we know the allocation
4365 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4366 // causes the non-eden paths to take compensating steps to
4367 // simulate a fresh allocation, so that no further
4368 // card marks are required in compiled code to initialize
4369 // the object.)
4370
4371 if (!stopped()) {
4372 copy_to_clone(obj, alloc_obj, obj_size, true);
4373
4374 // Present the results of the copy.
4375 result_reg->init_req(_array_path, control());
4376 result_val->init_req(_array_path, alloc_obj);
4377 result_i_o ->set_req(_array_path, i_o());
4378 result_mem ->set_req(_array_path, reset_memory());
4379 }
4380 }
4381
4382 // We only go to the instance fast case code if we pass a number of guards.
4383 // The paths which do not pass are accumulated in the slow_region.
4384 RegionNode* slow_region = new RegionNode(1);
4385 record_for_igvn(slow_region);
4386 if (!stopped()) {
4387 // It's an instance (we did array above). Make the slow-path tests.
4388 // If this is a virtual call, we generate a funny guard. We grab
4389 // the vtable entry corresponding to clone() from the target object.
4390 // If the target method which we are calling happens to be the
4391 // Object clone() method, we pass the guard. We do not need this
4392 // guard for non-virtual calls; the caller is known to be the native
4393 // Object clone().
4394 if (is_virtual) {
4395 generate_virtual_guard(obj_klass, slow_region);
4396 }
4397
4398 // The object must be easily cloneable and must not have a finalizer.
4399 // Both of these conditions may be checked in a single test.
4400 // We could optimize the test further, but we don't care.
4401 generate_access_flags_guard(obj_klass,
4402 // Test both conditions:
4403 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4404 // Must be cloneable but not finalizer:
4405 JVM_ACC_IS_CLONEABLE_FAST,
4406 slow_region);
4407 }
4408
4409 if (!stopped()) {
4410 // It's an instance, and it passed the slow-path tests.
4411 PreserveJVMState pjvms(this);
4412 Node* obj_size = NULL;
4413 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4414 // is reexecuted if deoptimization occurs and there could be problems when merging
4415 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4416 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4417
4418 copy_to_clone(obj, alloc_obj, obj_size, false);
4419
4420 // Present the results of the slow call.
4421 result_reg->init_req(_instance_path, control());
4422 result_val->init_req(_instance_path, alloc_obj);
4423 result_i_o ->set_req(_instance_path, i_o());
4424 result_mem ->set_req(_instance_path, reset_memory());
4425 }
4426
4427 // Generate code for the slow case. We make a call to clone().
4428 set_control(_gvn.transform(slow_region));
4429 if (!stopped()) {
4430 PreserveJVMState pjvms(this);
4431 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4432 // We need to deoptimize on exception (see comment above)
4433 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4434 // this->control() comes from set_results_for_java_call
4435 result_reg->init_req(_slow_path, control());
4436 result_val->init_req(_slow_path, slow_result);
4437 result_i_o ->set_req(_slow_path, i_o());
4438 result_mem ->set_req(_slow_path, reset_memory());
4439 }
4440
4441 // Return the combined state.
4442 set_control( _gvn.transform(result_reg));
4443 set_i_o( _gvn.transform(result_i_o));
4444 set_all_memory( _gvn.transform(result_mem));
4445 } // original reexecute is set back here
4446
4447 set_result(_gvn.transform(result_val));
4448 return true;
4449 }
4450
4451 // If we have a tightly coupled allocation, the arraycopy may take care
4452 // of the array initialization. If one of the guards we insert between
4453 // the allocation and the arraycopy causes a deoptimization, an
4454 // unitialized array will escape the compiled method. To prevent that
4455 // we set the JVM state for uncommon traps between the allocation and
4456 // the arraycopy to the state before the allocation so, in case of
4457 // deoptimization, we'll reexecute the allocation and the
4458 // initialization.
4459 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4460 if (alloc != NULL) {
4461 ciMethod* trap_method = alloc->jvms()->method();
4462 int trap_bci = alloc->jvms()->bci();
4463
4464 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4465 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4466 // Make sure there's no store between the allocation and the
4467 // arraycopy otherwise visible side effects could be rexecuted
4468 // in case of deoptimization and cause incorrect execution.
4469 bool no_interfering_store = true;
4470 Node* mem = alloc->in(TypeFunc::Memory);
4471 if (mem->is_MergeMem()) {
4472 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4473 Node* n = mms.memory();
4474 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4475 assert(n->is_Store(), "what else?");
4476 no_interfering_store = false;
4477 break;
4478 }
4479 }
4480 } else {
4481 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4482 Node* n = mms.memory();
4483 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4484 assert(n->is_Store(), "what else?");
4485 no_interfering_store = false;
4486 break;
4487 }
4488 }
4489 }
4490
4491 if (no_interfering_store) {
4492 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4493 uint size = alloc->req();
4494 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4495 old_jvms->set_map(sfpt);
4496 for (uint i = 0; i < size; i++) {
4497 sfpt->init_req(i, alloc->in(i));
4498 }
4499 // re-push array length for deoptimization
4500 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4501 old_jvms->set_sp(old_jvms->sp()+1);
4502 old_jvms->set_monoff(old_jvms->monoff()+1);
4503 old_jvms->set_scloff(old_jvms->scloff()+1);
4504 old_jvms->set_endoff(old_jvms->endoff()+1);
4505 old_jvms->set_should_reexecute(true);
4506
4507 sfpt->set_i_o(map()->i_o());
4508 sfpt->set_memory(map()->memory());
4509 sfpt->set_control(map()->control());
4510
4511 JVMState* saved_jvms = jvms();
4512 saved_reexecute_sp = _reexecute_sp;
4513
4514 set_jvms(sfpt->jvms());
4515 _reexecute_sp = jvms()->sp();
4516
4517 return saved_jvms;
4518 }
4519 }
4520 }
4521 return NULL;
4522 }
4523
4524 // In case of a deoptimization, we restart execution at the
4525 // allocation, allocating a new array. We would leave an uninitialized
4526 // array in the heap that GCs wouldn't expect. Move the allocation
4527 // after the traps so we don't allocate the array if we
4528 // deoptimize. This is possible because tightly_coupled_allocation()
4529 // guarantees there's no observer of the allocated array at this point
4530 // and the control flow is simple enough.
4531 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4532 int saved_reexecute_sp, uint new_idx) {
4533 if (saved_jvms != NULL && !stopped()) {
4534 assert(alloc != NULL, "only with a tightly coupled allocation");
4535 // restore JVM state to the state at the arraycopy
4536 saved_jvms->map()->set_control(map()->control());
4537 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4538 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4539 // If we've improved the types of some nodes (null check) while
4540 // emitting the guards, propagate them to the current state
4541 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4542 set_jvms(saved_jvms);
4543 _reexecute_sp = saved_reexecute_sp;
4544
4545 // Remove the allocation from above the guards
4546 CallProjections callprojs;
4547 alloc->extract_projections(&callprojs, true);
4548 InitializeNode* init = alloc->initialization();
4549 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4550 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4551 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4552 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4553
4554 // move the allocation here (after the guards)
4555 _gvn.hash_delete(alloc);
4556 alloc->set_req(TypeFunc::Control, control());
4557 alloc->set_req(TypeFunc::I_O, i_o());
4558 Node *mem = reset_memory();
4559 set_all_memory(mem);
4560 alloc->set_req(TypeFunc::Memory, mem);
4561 set_control(init->proj_out_or_null(TypeFunc::Control));
4562 set_i_o(callprojs.fallthrough_ioproj);
4563
4564 // Update memory as done in GraphKit::set_output_for_allocation()
4565 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4566 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4567 if (ary_type->isa_aryptr() && length_type != NULL) {
4568 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4569 }
4570 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4571 int elemidx = C->get_alias_index(telemref);
4572 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
4573 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
4574
4575 Node* allocx = _gvn.transform(alloc);
4576 assert(allocx == alloc, "where has the allocation gone?");
4577 assert(dest->is_CheckCastPP(), "not an allocation result?");
4578
4579 _gvn.hash_delete(dest);
4580 dest->set_req(0, control());
4581 Node* destx = _gvn.transform(dest);
4582 assert(destx == dest, "where has the allocation result gone?");
4583 }
4584 }
4585
4586
4587 //------------------------------inline_arraycopy-----------------------
4588 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4589 // Object dest, int destPos,
4590 // int length);
4591 bool LibraryCallKit::inline_arraycopy() {
4592 // Get the arguments.
4593 Node* src = argument(0); // type: oop
4594 Node* src_offset = argument(1); // type: int
4595 Node* dest = argument(2); // type: oop
4596 Node* dest_offset = argument(3); // type: int
4597 Node* length = argument(4); // type: int
4598
4599 uint new_idx = C->unique();
4600
4601 // Check for allocation before we add nodes that would confuse
4602 // tightly_coupled_allocation()
4603 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4604
4605 int saved_reexecute_sp = -1;
4606 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4607 // See arraycopy_restore_alloc_state() comment
4608 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4609 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4610 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
4611 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4612
4613 // The following tests must be performed
4614 // (1) src and dest are arrays.
4615 // (2) src and dest arrays must have elements of the same BasicType
4616 // (3) src and dest must not be null.
4617 // (4) src_offset must not be negative.
4618 // (5) dest_offset must not be negative.
4619 // (6) length must not be negative.
4620 // (7) src_offset + length must not exceed length of src.
4621 // (8) dest_offset + length must not exceed length of dest.
4622 // (9) each element of an oop array must be assignable
4623
4624 // (3) src and dest must not be null.
4625 // always do this here because we need the JVM state for uncommon traps
4626 Node* null_ctl = top();
4627 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4628 assert(null_ctl->is_top(), "no null control here");
4629 dest = null_check(dest, T_ARRAY);
4630
4631 if (!can_emit_guards) {
4632 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4633 // guards but the arraycopy node could still take advantage of a
4634 // tightly allocated allocation. tightly_coupled_allocation() is
4635 // called again to make sure it takes the null check above into
4636 // account: the null check is mandatory and if it caused an
4637 // uncommon trap to be emitted then the allocation can't be
4638 // considered tightly coupled in this context.
4639 alloc = tightly_coupled_allocation(dest, NULL);
4640 }
4641
4642 bool validated = false;
4643
4644 const Type* src_type = _gvn.type(src);
4645 const Type* dest_type = _gvn.type(dest);
4646 const TypeAryPtr* top_src = src_type->isa_aryptr();
4647 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4648
4649 // Do we have the type of src?
4650 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4651 // Do we have the type of dest?
4652 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4653 // Is the type for src from speculation?
4654 bool src_spec = false;
4655 // Is the type for dest from speculation?
4656 bool dest_spec = false;
4657
4658 if ((!has_src || !has_dest) && can_emit_guards) {
4659 // We don't have sufficient type information, let's see if
4660 // speculative types can help. We need to have types for both src
4661 // and dest so that it pays off.
4662
4663 // Do we already have or could we have type information for src
4664 bool could_have_src = has_src;
4665 // Do we already have or could we have type information for dest
4666 bool could_have_dest = has_dest;
4667
4668 ciKlass* src_k = NULL;
4669 if (!has_src) {
4670 src_k = src_type->speculative_type_not_null();
4671 if (src_k != NULL && src_k->is_array_klass()) {
4672 could_have_src = true;
4673 }
4674 }
4675
4676 ciKlass* dest_k = NULL;
4677 if (!has_dest) {
4678 dest_k = dest_type->speculative_type_not_null();
4679 if (dest_k != NULL && dest_k->is_array_klass()) {
4680 could_have_dest = true;
4681 }
4682 }
4683
4684 if (could_have_src && could_have_dest) {
4685 // This is going to pay off so emit the required guards
4686 if (!has_src) {
4687 src = maybe_cast_profiled_obj(src, src_k, true);
4688 src_type = _gvn.type(src);
4689 top_src = src_type->isa_aryptr();
4690 has_src = (top_src != NULL && top_src->klass() != NULL);
4691 src_spec = true;
4692 }
4693 if (!has_dest) {
4694 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4695 dest_type = _gvn.type(dest);
4696 top_dest = dest_type->isa_aryptr();
4697 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4698 dest_spec = true;
4699 }
4700 }
4701 }
4702
4703 if (has_src && has_dest && can_emit_guards) {
4704 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
4705 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
4706 if (is_reference_type(src_elem)) src_elem = T_OBJECT;
4707 if (is_reference_type(dest_elem)) dest_elem = T_OBJECT;
4708
4709 if (src_elem == dest_elem && src_elem == T_OBJECT) {
4710 // If both arrays are object arrays then having the exact types
4711 // for both will remove the need for a subtype check at runtime
4712 // before the call and may make it possible to pick a faster copy
4713 // routine (without a subtype check on every element)
4714 // Do we have the exact type of src?
4715 bool could_have_src = src_spec;
4716 // Do we have the exact type of dest?
4717 bool could_have_dest = dest_spec;
4718 ciKlass* src_k = top_src->klass();
4719 ciKlass* dest_k = top_dest->klass();
4720 if (!src_spec) {
4721 src_k = src_type->speculative_type_not_null();
4722 if (src_k != NULL && src_k->is_array_klass()) {
4723 could_have_src = true;
4724 }
4725 }
4726 if (!dest_spec) {
4727 dest_k = dest_type->speculative_type_not_null();
4728 if (dest_k != NULL && dest_k->is_array_klass()) {
4729 could_have_dest = true;
4730 }
4731 }
4732 if (could_have_src && could_have_dest) {
4733 // If we can have both exact types, emit the missing guards
4734 if (could_have_src && !src_spec) {
4735 src = maybe_cast_profiled_obj(src, src_k, true);
4736 }
4737 if (could_have_dest && !dest_spec) {
4738 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4739 }
4740 }
4741 }
4742 }
4743
4744 ciMethod* trap_method = method();
4745 int trap_bci = bci();
4746 if (saved_jvms != NULL) {
4747 trap_method = alloc->jvms()->method();
4748 trap_bci = alloc->jvms()->bci();
4749 }
4750
4751 bool negative_length_guard_generated = false;
4752
4753 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4754 can_emit_guards &&
4755 !src->is_top() && !dest->is_top()) {
4756 // validate arguments: enables transformation the ArrayCopyNode
4757 validated = true;
4758
4759 RegionNode* slow_region = new RegionNode(1);
4760 record_for_igvn(slow_region);
4761
4762 // (1) src and dest are arrays.
4763 generate_non_array_guard(load_object_klass(src), slow_region);
4764 generate_non_array_guard(load_object_klass(dest), slow_region);
4765
4766 // (2) src and dest arrays must have elements of the same BasicType
4767 // done at macro expansion or at Ideal transformation time
4768
4769 // (4) src_offset must not be negative.
4770 generate_negative_guard(src_offset, slow_region);
4771
4772 // (5) dest_offset must not be negative.
4773 generate_negative_guard(dest_offset, slow_region);
4774
4775 // (7) src_offset + length must not exceed length of src.
4776 generate_limit_guard(src_offset, length,
4777 load_array_length(src),
4778 slow_region);
4779
4780 // (8) dest_offset + length must not exceed length of dest.
4781 generate_limit_guard(dest_offset, length,
4782 load_array_length(dest),
4783 slow_region);
4784
4785 // (6) length must not be negative.
4786 // This is also checked in generate_arraycopy() during macro expansion, but
4787 // we also have to check it here for the case where the ArrayCopyNode will
4788 // be eliminated by Escape Analysis.
4789 if (EliminateAllocations) {
4790 generate_negative_guard(length, slow_region);
4791 negative_length_guard_generated = true;
4792 }
4793
4794 // (9) each element of an oop array must be assignable
4795 Node* dest_klass = load_object_klass(dest);
4796 if (src != dest) {
4797 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
4798
4799 if (not_subtype_ctrl != top()) {
4800 PreserveJVMState pjvms(this);
4801 set_control(not_subtype_ctrl);
4802 uncommon_trap(Deoptimization::Reason_intrinsic,
4803 Deoptimization::Action_make_not_entrant);
4804 assert(stopped(), "Should be stopped");
4805 }
4806 }
4807 {
4808 PreserveJVMState pjvms(this);
4809 set_control(_gvn.transform(slow_region));
4810 uncommon_trap(Deoptimization::Reason_intrinsic,
4811 Deoptimization::Action_make_not_entrant);
4812 assert(stopped(), "Should be stopped");
4813 }
4814
4815 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
4816 const Type *toop = TypeOopPtr::make_from_klass(dest_klass_t->klass());
4817 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
4818 }
4819
4820 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
4821
4822 if (stopped()) {
4823 return true;
4824 }
4825
4826 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
4827 // Create LoadRange and LoadKlass nodes for use during macro expansion here
4828 // so the compiler has a chance to eliminate them: during macro expansion,
4829 // we have to set their control (CastPP nodes are eliminated).
4830 load_object_klass(src), load_object_klass(dest),
4831 load_array_length(src), load_array_length(dest));
4832
4833 ac->set_arraycopy(validated);
4834
4835 Node* n = _gvn.transform(ac);
4836 if (n == ac) {
4837 ac->connect_outputs(this);
4838 } else {
4839 assert(validated, "shouldn't transform if all arguments not validated");
4840 set_all_memory(n);
4841 }
4842 clear_upper_avx();
4843
4844
4845 return true;
4846 }
4847
4848
4849 // Helper function which determines if an arraycopy immediately follows
4850 // an allocation, with no intervening tests or other escapes for the object.
4851 AllocateArrayNode*
4852 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
4853 RegionNode* slow_region) {
4854 if (stopped()) return NULL; // no fast path
4855 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
4856
4857 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
4858 if (alloc == NULL) return NULL;
4859
4860 Node* rawmem = memory(Compile::AliasIdxRaw);
4861 // Is the allocation's memory state untouched?
4862 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
4863 // Bail out if there have been raw-memory effects since the allocation.
4864 // (Example: There might have been a call or safepoint.)
4865 return NULL;
4866 }
4867 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
4868 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
4869 return NULL;
4870 }
4871
4872 // There must be no unexpected observers of this allocation.
4873 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
4874 Node* obs = ptr->fast_out(i);
4875 if (obs != this->map()) {
4876 return NULL;
4877 }
4878 }
4879
4880 // This arraycopy must unconditionally follow the allocation of the ptr.
4881 Node* alloc_ctl = ptr->in(0);
4882 Node* ctl = control();
4883 while (ctl != alloc_ctl) {
4884 // There may be guards which feed into the slow_region.
4885 // Any other control flow means that we might not get a chance
4886 // to finish initializing the allocated object.
4887 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
4888 IfNode* iff = ctl->in(0)->as_If();
4889 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
4890 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
4891 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
4892 ctl = iff->in(0); // This test feeds the known slow_region.
4893 continue;
4894 }
4895 // One more try: Various low-level checks bottom out in
4896 // uncommon traps. If the debug-info of the trap omits
4897 // any reference to the allocation, as we've already
4898 // observed, then there can be no objection to the trap.
4899 bool found_trap = false;
4900 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
4901 Node* obs = not_ctl->fast_out(j);
4902 if (obs->in(0) == not_ctl && obs->is_Call() &&
4903 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
4904 found_trap = true; break;
4905 }
4906 }
4907 if (found_trap) {
4908 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
4909 continue;
4910 }
4911 }
4912 return NULL;
4913 }
4914
4915 // If we get this far, we have an allocation which immediately
4916 // precedes the arraycopy, and we can take over zeroing the new object.
4917 // The arraycopy will finish the initialization, and provide
4918 // a new control state to which we will anchor the destination pointer.
4919
4920 return alloc;
4921 }
4922
4923 //-------------inline_encodeISOArray-----------------------------------
4924 // encode char[] to byte[] in ISO_8859_1
4925 bool LibraryCallKit::inline_encodeISOArray() {
4926 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
4927 // no receiver since it is static method
4928 Node *src = argument(0);
4929 Node *src_offset = argument(1);
4930 Node *dst = argument(2);
4931 Node *dst_offset = argument(3);
4932 Node *length = argument(4);
4933
4934 src = must_be_not_null(src, true);
4935 dst = must_be_not_null(dst, true);
4936
4937 const Type* src_type = src->Value(&_gvn);
4938 const Type* dst_type = dst->Value(&_gvn);
4939 const TypeAryPtr* top_src = src_type->isa_aryptr();
4940 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
4941 if (top_src == NULL || top_src->klass() == NULL ||
4942 top_dest == NULL || top_dest->klass() == NULL) {
4943 // failed array check
4944 return false;
4945 }
4946
4947 // Figure out the size and type of the elements we will be copying.
4948 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4949 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4950 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
4951 return false;
4952 }
4953
4954 Node* src_start = array_element_address(src, src_offset, T_CHAR);
4955 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
4956 // 'src_start' points to src array + scaled offset
4957 // 'dst_start' points to dst array + scaled offset
4958
4959 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
4960 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
4961 enc = _gvn.transform(enc);
4962 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
4963 set_memory(res_mem, mtype);
4964 set_result(enc);
4965 clear_upper_avx();
4966
4967 return true;
4968 }
4969
4970 //-------------inline_multiplyToLen-----------------------------------
4971 bool LibraryCallKit::inline_multiplyToLen() {
4972 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
4973
4974 address stubAddr = StubRoutines::multiplyToLen();
4975 if (stubAddr == NULL) {
4976 return false; // Intrinsic's stub is not implemented on this platform
4977 }
4978 const char* stubName = "multiplyToLen";
4979
4980 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
4981
4982 // no receiver because it is a static method
4983 Node* x = argument(0);
4984 Node* xlen = argument(1);
4985 Node* y = argument(2);
4986 Node* ylen = argument(3);
4987 Node* z = argument(4);
4988
4989 x = must_be_not_null(x, true);
4990 y = must_be_not_null(y, true);
4991
4992 const Type* x_type = x->Value(&_gvn);
4993 const Type* y_type = y->Value(&_gvn);
4994 const TypeAryPtr* top_x = x_type->isa_aryptr();
4995 const TypeAryPtr* top_y = y_type->isa_aryptr();
4996 if (top_x == NULL || top_x->klass() == NULL ||
4997 top_y == NULL || top_y->klass() == NULL) {
4998 // failed array check
4999 return false;
5000 }
5001
5002 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5003 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5004 if (x_elem != T_INT || y_elem != T_INT) {
5005 return false;
5006 }
5007
5008 // Set the original stack and the reexecute bit for the interpreter to reexecute
5009 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5010 // on the return from z array allocation in runtime.
5011 { PreserveReexecuteState preexecs(this);
5012 jvms()->set_should_reexecute(true);
5013
5014 Node* x_start = array_element_address(x, intcon(0), x_elem);
5015 Node* y_start = array_element_address(y, intcon(0), y_elem);
5016 // 'x_start' points to x array + scaled xlen
5017 // 'y_start' points to y array + scaled ylen
5018
5019 // Allocate the result array
5020 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5021 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5022 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5023
5024 IdealKit ideal(this);
5025
5026 #define __ ideal.
5027 Node* one = __ ConI(1);
5028 Node* zero = __ ConI(0);
5029 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5030 __ set(need_alloc, zero);
5031 __ set(z_alloc, z);
5032 __ if_then(z, BoolTest::eq, null()); {
5033 __ increment (need_alloc, one);
5034 } __ else_(); {
5035 // Update graphKit memory and control from IdealKit.
5036 sync_kit(ideal);
5037 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5038 cast->init_req(0, control());
5039 _gvn.set_type(cast, cast->bottom_type());
5040 C->record_for_igvn(cast);
5041
5042 Node* zlen_arg = load_array_length(cast);
5043 // Update IdealKit memory and control from graphKit.
5044 __ sync_kit(this);
5045 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5046 __ increment (need_alloc, one);
5047 } __ end_if();
5048 } __ end_if();
5049
5050 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5051 // Update graphKit memory and control from IdealKit.
5052 sync_kit(ideal);
5053 Node * narr = new_array(klass_node, zlen, 1);
5054 // Update IdealKit memory and control from graphKit.
5055 __ sync_kit(this);
5056 __ set(z_alloc, narr);
5057 } __ end_if();
5058
5059 sync_kit(ideal);
5060 z = __ value(z_alloc);
5061 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5062 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5063 // Final sync IdealKit and GraphKit.
5064 final_sync(ideal);
5065 #undef __
5066
5067 Node* z_start = array_element_address(z, intcon(0), T_INT);
5068
5069 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5070 OptoRuntime::multiplyToLen_Type(),
5071 stubAddr, stubName, TypePtr::BOTTOM,
5072 x_start, xlen, y_start, ylen, z_start, zlen);
5073 } // original reexecute is set back here
5074
5075 C->set_has_split_ifs(true); // Has chance for split-if optimization
5076 set_result(z);
5077 return true;
5078 }
5079
5080 //-------------inline_squareToLen------------------------------------
5081 bool LibraryCallKit::inline_squareToLen() {
5082 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5083
5084 address stubAddr = StubRoutines::squareToLen();
5085 if (stubAddr == NULL) {
5086 return false; // Intrinsic's stub is not implemented on this platform
5087 }
5088 const char* stubName = "squareToLen";
5089
5090 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5091
5092 Node* x = argument(0);
5093 Node* len = argument(1);
5094 Node* z = argument(2);
5095 Node* zlen = argument(3);
5096
5097 x = must_be_not_null(x, true);
5098 z = must_be_not_null(z, true);
5099
5100 const Type* x_type = x->Value(&_gvn);
5101 const Type* z_type = z->Value(&_gvn);
5102 const TypeAryPtr* top_x = x_type->isa_aryptr();
5103 const TypeAryPtr* top_z = z_type->isa_aryptr();
5104 if (top_x == NULL || top_x->klass() == NULL ||
5105 top_z == NULL || top_z->klass() == NULL) {
5106 // failed array check
5107 return false;
5108 }
5109
5110 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5111 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5112 if (x_elem != T_INT || z_elem != T_INT) {
5113 return false;
5114 }
5115
5116
5117 Node* x_start = array_element_address(x, intcon(0), x_elem);
5118 Node* z_start = array_element_address(z, intcon(0), z_elem);
5119
5120 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5121 OptoRuntime::squareToLen_Type(),
5122 stubAddr, stubName, TypePtr::BOTTOM,
5123 x_start, len, z_start, zlen);
5124
5125 set_result(z);
5126 return true;
5127 }
5128
5129 //-------------inline_mulAdd------------------------------------------
5130 bool LibraryCallKit::inline_mulAdd() {
5131 assert(UseMulAddIntrinsic, "not implemented on this platform");
5132
5133 address stubAddr = StubRoutines::mulAdd();
5134 if (stubAddr == NULL) {
5135 return false; // Intrinsic's stub is not implemented on this platform
5136 }
5137 const char* stubName = "mulAdd";
5138
5139 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5140
5141 Node* out = argument(0);
5142 Node* in = argument(1);
5143 Node* offset = argument(2);
5144 Node* len = argument(3);
5145 Node* k = argument(4);
5146
5147 out = must_be_not_null(out, true);
5148
5149 const Type* out_type = out->Value(&_gvn);
5150 const Type* in_type = in->Value(&_gvn);
5151 const TypeAryPtr* top_out = out_type->isa_aryptr();
5152 const TypeAryPtr* top_in = in_type->isa_aryptr();
5153 if (top_out == NULL || top_out->klass() == NULL ||
5154 top_in == NULL || top_in->klass() == NULL) {
5155 // failed array check
5156 return false;
5157 }
5158
5159 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5160 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5161 if (out_elem != T_INT || in_elem != T_INT) {
5162 return false;
5163 }
5164
5165 Node* outlen = load_array_length(out);
5166 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5167 Node* out_start = array_element_address(out, intcon(0), out_elem);
5168 Node* in_start = array_element_address(in, intcon(0), in_elem);
5169
5170 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5171 OptoRuntime::mulAdd_Type(),
5172 stubAddr, stubName, TypePtr::BOTTOM,
5173 out_start,in_start, new_offset, len, k);
5174 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5175 set_result(result);
5176 return true;
5177 }
5178
5179 //-------------inline_montgomeryMultiply-----------------------------------
5180 bool LibraryCallKit::inline_montgomeryMultiply() {
5181 address stubAddr = StubRoutines::montgomeryMultiply();
5182 if (stubAddr == NULL) {
5183 return false; // Intrinsic's stub is not implemented on this platform
5184 }
5185
5186 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5187 const char* stubName = "montgomery_multiply";
5188
5189 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5190
5191 Node* a = argument(0);
5192 Node* b = argument(1);
5193 Node* n = argument(2);
5194 Node* len = argument(3);
5195 Node* inv = argument(4);
5196 Node* m = argument(6);
5197
5198 const Type* a_type = a->Value(&_gvn);
5199 const TypeAryPtr* top_a = a_type->isa_aryptr();
5200 const Type* b_type = b->Value(&_gvn);
5201 const TypeAryPtr* top_b = b_type->isa_aryptr();
5202 const Type* n_type = a->Value(&_gvn);
5203 const TypeAryPtr* top_n = n_type->isa_aryptr();
5204 const Type* m_type = a->Value(&_gvn);
5205 const TypeAryPtr* top_m = m_type->isa_aryptr();
5206 if (top_a == NULL || top_a->klass() == NULL ||
5207 top_b == NULL || top_b->klass() == NULL ||
5208 top_n == NULL || top_n->klass() == NULL ||
5209 top_m == NULL || top_m->klass() == NULL) {
5210 // failed array check
5211 return false;
5212 }
5213
5214 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5215 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5216 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5217 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5218 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5219 return false;
5220 }
5221
5222 // Make the call
5223 {
5224 Node* a_start = array_element_address(a, intcon(0), a_elem);
5225 Node* b_start = array_element_address(b, intcon(0), b_elem);
5226 Node* n_start = array_element_address(n, intcon(0), n_elem);
5227 Node* m_start = array_element_address(m, intcon(0), m_elem);
5228
5229 Node* call = make_runtime_call(RC_LEAF,
5230 OptoRuntime::montgomeryMultiply_Type(),
5231 stubAddr, stubName, TypePtr::BOTTOM,
5232 a_start, b_start, n_start, len, inv, top(),
5233 m_start);
5234 set_result(m);
5235 }
5236
5237 return true;
5238 }
5239
5240 bool LibraryCallKit::inline_montgomerySquare() {
5241 address stubAddr = StubRoutines::montgomerySquare();
5242 if (stubAddr == NULL) {
5243 return false; // Intrinsic's stub is not implemented on this platform
5244 }
5245
5246 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5247 const char* stubName = "montgomery_square";
5248
5249 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5250
5251 Node* a = argument(0);
5252 Node* n = argument(1);
5253 Node* len = argument(2);
5254 Node* inv = argument(3);
5255 Node* m = argument(5);
5256
5257 const Type* a_type = a->Value(&_gvn);
5258 const TypeAryPtr* top_a = a_type->isa_aryptr();
5259 const Type* n_type = a->Value(&_gvn);
5260 const TypeAryPtr* top_n = n_type->isa_aryptr();
5261 const Type* m_type = a->Value(&_gvn);
5262 const TypeAryPtr* top_m = m_type->isa_aryptr();
5263 if (top_a == NULL || top_a->klass() == NULL ||
5264 top_n == NULL || top_n->klass() == NULL ||
5265 top_m == NULL || top_m->klass() == NULL) {
5266 // failed array check
5267 return false;
5268 }
5269
5270 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5271 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5272 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5273 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5274 return false;
5275 }
5276
5277 // Make the call
5278 {
5279 Node* a_start = array_element_address(a, intcon(0), a_elem);
5280 Node* n_start = array_element_address(n, intcon(0), n_elem);
5281 Node* m_start = array_element_address(m, intcon(0), m_elem);
5282
5283 Node* call = make_runtime_call(RC_LEAF,
5284 OptoRuntime::montgomerySquare_Type(),
5285 stubAddr, stubName, TypePtr::BOTTOM,
5286 a_start, n_start, len, inv, top(),
5287 m_start);
5288 set_result(m);
5289 }
5290
5291 return true;
5292 }
5293
5294 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
5295 address stubAddr = NULL;
5296 const char* stubName = NULL;
5297
5298 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
5299 if (stubAddr == NULL) {
5300 return false; // Intrinsic's stub is not implemented on this platform
5301 }
5302
5303 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
5304
5305 assert(callee()->signature()->size() == 5, "expected 5 arguments");
5306
5307 Node* newArr = argument(0);
5308 Node* oldArr = argument(1);
5309 Node* newIdx = argument(2);
5310 Node* shiftCount = argument(3);
5311 Node* numIter = argument(4);
5312
5313 const Type* newArr_type = newArr->Value(&_gvn);
5314 const TypeAryPtr* top_newArr = newArr_type->isa_aryptr();
5315 const Type* oldArr_type = oldArr->Value(&_gvn);
5316 const TypeAryPtr* top_oldArr = oldArr_type->isa_aryptr();
5317 if (top_newArr == NULL || top_newArr->klass() == NULL || top_oldArr == NULL
5318 || top_oldArr->klass() == NULL) {
5319 return false;
5320 }
5321
5322 BasicType newArr_elem = newArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5323 BasicType oldArr_elem = oldArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5324 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
5325 return false;
5326 }
5327
5328 // Make the call
5329 {
5330 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
5331 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
5332
5333 Node* call = make_runtime_call(RC_LEAF,
5334 OptoRuntime::bigIntegerShift_Type(),
5335 stubAddr,
5336 stubName,
5337 TypePtr::BOTTOM,
5338 newArr_start,
5339 oldArr_start,
5340 newIdx,
5341 shiftCount,
5342 numIter);
5343 }
5344
5345 return true;
5346 }
5347
5348 //-------------inline_vectorizedMismatch------------------------------
5349 bool LibraryCallKit::inline_vectorizedMismatch() {
5350 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5351
5352 address stubAddr = StubRoutines::vectorizedMismatch();
5353 if (stubAddr == NULL) {
5354 return false; // Intrinsic's stub is not implemented on this platform
5355 }
5356 const char* stubName = "vectorizedMismatch";
5357 int size_l = callee()->signature()->size();
5358 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5359
5360 Node* obja = argument(0);
5361 Node* aoffset = argument(1);
5362 Node* objb = argument(3);
5363 Node* boffset = argument(4);
5364 Node* length = argument(6);
5365 Node* scale = argument(7);
5366
5367 const Type* a_type = obja->Value(&_gvn);
5368 const Type* b_type = objb->Value(&_gvn);
5369 const TypeAryPtr* top_a = a_type->isa_aryptr();
5370 const TypeAryPtr* top_b = b_type->isa_aryptr();
5371 if (top_a == NULL || top_a->klass() == NULL ||
5372 top_b == NULL || top_b->klass() == NULL) {
5373 // failed array check
5374 return false;
5375 }
5376
5377 Node* call;
5378 jvms()->set_should_reexecute(true);
5379
5380 Node* obja_adr = make_unsafe_address(obja, aoffset, ACCESS_READ);
5381 Node* objb_adr = make_unsafe_address(objb, boffset, ACCESS_READ);
5382
5383 call = make_runtime_call(RC_LEAF,
5384 OptoRuntime::vectorizedMismatch_Type(),
5385 stubAddr, stubName, TypePtr::BOTTOM,
5386 obja_adr, objb_adr, length, scale);
5387
5388 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5389 set_result(result);
5390 return true;
5391 }
5392
5393 /**
5394 * Calculate CRC32 for byte.
5395 * int java.util.zip.CRC32.update(int crc, int b)
5396 */
5397 bool LibraryCallKit::inline_updateCRC32() {
5398 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5399 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5400 // no receiver since it is static method
5401 Node* crc = argument(0); // type: int
5402 Node* b = argument(1); // type: int
5403
5404 /*
5405 * int c = ~ crc;
5406 * b = timesXtoThe32[(b ^ c) & 0xFF];
5407 * b = b ^ (c >>> 8);
5408 * crc = ~b;
5409 */
5410
5411 Node* M1 = intcon(-1);
5412 crc = _gvn.transform(new XorINode(crc, M1));
5413 Node* result = _gvn.transform(new XorINode(crc, b));
5414 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5415
5416 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5417 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5418 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5419 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5420
5421 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5422 result = _gvn.transform(new XorINode(crc, result));
5423 result = _gvn.transform(new XorINode(result, M1));
5424 set_result(result);
5425 return true;
5426 }
5427
5428 /**
5429 * Calculate CRC32 for byte[] array.
5430 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5431 */
5432 bool LibraryCallKit::inline_updateBytesCRC32() {
5433 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5434 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5435 // no receiver since it is static method
5436 Node* crc = argument(0); // type: int
5437 Node* src = argument(1); // type: oop
5438 Node* offset = argument(2); // type: int
5439 Node* length = argument(3); // type: int
5440
5441 const Type* src_type = src->Value(&_gvn);
5442 const TypeAryPtr* top_src = src_type->isa_aryptr();
5443 if (top_src == NULL || top_src->klass() == NULL) {
5444 // failed array check
5445 return false;
5446 }
5447
5448 // Figure out the size and type of the elements we will be copying.
5449 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5450 if (src_elem != T_BYTE) {
5451 return false;
5452 }
5453
5454 // 'src_start' points to src array + scaled offset
5455 src = must_be_not_null(src, true);
5456 Node* src_start = array_element_address(src, offset, src_elem);
5457
5458 // We assume that range check is done by caller.
5459 // TODO: generate range check (offset+length < src.length) in debug VM.
5460
5461 // Call the stub.
5462 address stubAddr = StubRoutines::updateBytesCRC32();
5463 const char *stubName = "updateBytesCRC32";
5464
5465 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5466 stubAddr, stubName, TypePtr::BOTTOM,
5467 crc, src_start, length);
5468 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5469 set_result(result);
5470 return true;
5471 }
5472
5473 /**
5474 * Calculate CRC32 for ByteBuffer.
5475 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5476 */
5477 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5478 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5479 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5480 // no receiver since it is static method
5481 Node* crc = argument(0); // type: int
5482 Node* src = argument(1); // type: long
5483 Node* offset = argument(3); // type: int
5484 Node* length = argument(4); // type: int
5485
5486 src = ConvL2X(src); // adjust Java long to machine word
5487 Node* base = _gvn.transform(new CastX2PNode(src));
5488 offset = ConvI2X(offset);
5489
5490 // 'src_start' points to src array + scaled offset
5491 Node* src_start = basic_plus_adr(top(), base, offset);
5492
5493 // Call the stub.
5494 address stubAddr = StubRoutines::updateBytesCRC32();
5495 const char *stubName = "updateBytesCRC32";
5496
5497 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5498 stubAddr, stubName, TypePtr::BOTTOM,
5499 crc, src_start, length);
5500 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5501 set_result(result);
5502 return true;
5503 }
5504
5505 //------------------------------get_table_from_crc32c_class-----------------------
5506 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5507 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5508 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5509
5510 return table;
5511 }
5512
5513 //------------------------------inline_updateBytesCRC32C-----------------------
5514 //
5515 // Calculate CRC32C for byte[] array.
5516 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5517 //
5518 bool LibraryCallKit::inline_updateBytesCRC32C() {
5519 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5520 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5521 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5522 // no receiver since it is a static method
5523 Node* crc = argument(0); // type: int
5524 Node* src = argument(1); // type: oop
5525 Node* offset = argument(2); // type: int
5526 Node* end = argument(3); // type: int
5527
5528 Node* length = _gvn.transform(new SubINode(end, offset));
5529
5530 const Type* src_type = src->Value(&_gvn);
5531 const TypeAryPtr* top_src = src_type->isa_aryptr();
5532 if (top_src == NULL || top_src->klass() == NULL) {
5533 // failed array check
5534 return false;
5535 }
5536
5537 // Figure out the size and type of the elements we will be copying.
5538 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5539 if (src_elem != T_BYTE) {
5540 return false;
5541 }
5542
5543 // 'src_start' points to src array + scaled offset
5544 src = must_be_not_null(src, true);
5545 Node* src_start = array_element_address(src, offset, src_elem);
5546
5547 // static final int[] byteTable in class CRC32C
5548 Node* table = get_table_from_crc32c_class(callee()->holder());
5549 table = must_be_not_null(table, true);
5550 Node* table_start = array_element_address(table, intcon(0), T_INT);
5551
5552 // We assume that range check is done by caller.
5553 // TODO: generate range check (offset+length < src.length) in debug VM.
5554
5555 // Call the stub.
5556 address stubAddr = StubRoutines::updateBytesCRC32C();
5557 const char *stubName = "updateBytesCRC32C";
5558
5559 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5560 stubAddr, stubName, TypePtr::BOTTOM,
5561 crc, src_start, length, table_start);
5562 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5563 set_result(result);
5564 return true;
5565 }
5566
5567 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5568 //
5569 // Calculate CRC32C for DirectByteBuffer.
5570 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5571 //
5572 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5573 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5574 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5575 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5576 // no receiver since it is a static method
5577 Node* crc = argument(0); // type: int
5578 Node* src = argument(1); // type: long
5579 Node* offset = argument(3); // type: int
5580 Node* end = argument(4); // type: int
5581
5582 Node* length = _gvn.transform(new SubINode(end, offset));
5583
5584 src = ConvL2X(src); // adjust Java long to machine word
5585 Node* base = _gvn.transform(new CastX2PNode(src));
5586 offset = ConvI2X(offset);
5587
5588 // 'src_start' points to src array + scaled offset
5589 Node* src_start = basic_plus_adr(top(), base, offset);
5590
5591 // static final int[] byteTable in class CRC32C
5592 Node* table = get_table_from_crc32c_class(callee()->holder());
5593 table = must_be_not_null(table, true);
5594 Node* table_start = array_element_address(table, intcon(0), T_INT);
5595
5596 // Call the stub.
5597 address stubAddr = StubRoutines::updateBytesCRC32C();
5598 const char *stubName = "updateBytesCRC32C";
5599
5600 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5601 stubAddr, stubName, TypePtr::BOTTOM,
5602 crc, src_start, length, table_start);
5603 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5604 set_result(result);
5605 return true;
5606 }
5607
5608 //------------------------------inline_updateBytesAdler32----------------------
5609 //
5610 // Calculate Adler32 checksum for byte[] array.
5611 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5612 //
5613 bool LibraryCallKit::inline_updateBytesAdler32() {
5614 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5615 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5616 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5617 // no receiver since it is static method
5618 Node* crc = argument(0); // type: int
5619 Node* src = argument(1); // type: oop
5620 Node* offset = argument(2); // type: int
5621 Node* length = argument(3); // type: int
5622
5623 const Type* src_type = src->Value(&_gvn);
5624 const TypeAryPtr* top_src = src_type->isa_aryptr();
5625 if (top_src == NULL || top_src->klass() == NULL) {
5626 // failed array check
5627 return false;
5628 }
5629
5630 // Figure out the size and type of the elements we will be copying.
5631 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5632 if (src_elem != T_BYTE) {
5633 return false;
5634 }
5635
5636 // 'src_start' points to src array + scaled offset
5637 Node* src_start = array_element_address(src, offset, src_elem);
5638
5639 // We assume that range check is done by caller.
5640 // TODO: generate range check (offset+length < src.length) in debug VM.
5641
5642 // Call the stub.
5643 address stubAddr = StubRoutines::updateBytesAdler32();
5644 const char *stubName = "updateBytesAdler32";
5645
5646 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5647 stubAddr, stubName, TypePtr::BOTTOM,
5648 crc, src_start, length);
5649 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5650 set_result(result);
5651 return true;
5652 }
5653
5654 //------------------------------inline_updateByteBufferAdler32---------------
5655 //
5656 // Calculate Adler32 checksum for DirectByteBuffer.
5657 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5658 //
5659 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5660 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5661 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5662 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5663 // no receiver since it is static method
5664 Node* crc = argument(0); // type: int
5665 Node* src = argument(1); // type: long
5666 Node* offset = argument(3); // type: int
5667 Node* length = argument(4); // type: int
5668
5669 src = ConvL2X(src); // adjust Java long to machine word
5670 Node* base = _gvn.transform(new CastX2PNode(src));
5671 offset = ConvI2X(offset);
5672
5673 // 'src_start' points to src array + scaled offset
5674 Node* src_start = basic_plus_adr(top(), base, offset);
5675
5676 // Call the stub.
5677 address stubAddr = StubRoutines::updateBytesAdler32();
5678 const char *stubName = "updateBytesAdler32";
5679
5680 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5681 stubAddr, stubName, TypePtr::BOTTOM,
5682 crc, src_start, length);
5683
5684 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5685 set_result(result);
5686 return true;
5687 }
5688
5689 //----------------------------inline_reference_get----------------------------
5690 // public T java.lang.ref.Reference.get();
5691 bool LibraryCallKit::inline_reference_get() {
5692 const int referent_offset = java_lang_ref_Reference::referent_offset();
5693
5694 // Get the argument:
5695 Node* reference_obj = null_check_receiver();
5696 if (stopped()) return true;
5697
5698 const TypeInstPtr* tinst = _gvn.type(reference_obj)->isa_instptr();
5699 assert(tinst != NULL, "obj is null");
5700 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5701 ciInstanceKlass* referenceKlass = tinst->klass()->as_instance_klass();
5702 ciField* field = referenceKlass->get_field_by_name(ciSymbol::make("referent"),
5703 ciSymbol::make("Ljava/lang/Object;"),
5704 false);
5705 assert (field != NULL, "undefined field");
5706
5707 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5708 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5709
5710 ciInstanceKlass* klass = env()->Object_klass();
5711 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5712
5713 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
5714 Node* result = access_load_at(reference_obj, adr, adr_type, object_type, T_OBJECT, decorators);
5715 // Add memory barrier to prevent commoning reads from this field
5716 // across safepoint since GC can change its value.
5717 insert_mem_bar(Op_MemBarCPUOrder);
5718
5719 set_result(result);
5720 return true;
5721 }
5722
5723
5724 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5725 bool is_exact=true, bool is_static=false,
5726 ciInstanceKlass * fromKls=NULL) {
5727 if (fromKls == NULL) {
5728 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5729 assert(tinst != NULL, "obj is null");
5730 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5731 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5732 fromKls = tinst->klass()->as_instance_klass();
5733 } else {
5734 assert(is_static, "only for static field access");
5735 }
5736 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5737 ciSymbol::make(fieldTypeString),
5738 is_static);
5739
5740 assert (field != NULL, "undefined field");
5741 if (field == NULL) return (Node *) NULL;
5742
5743 if (is_static) {
5744 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5745 fromObj = makecon(tip);
5746 }
5747
5748 // Next code copied from Parse::do_get_xxx():
5749
5750 // Compute address and memory type.
5751 int offset = field->offset_in_bytes();
5752 bool is_vol = field->is_volatile();
5753 ciType* field_klass = field->type();
5754 assert(field_klass->is_loaded(), "should be loaded");
5755 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5756 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5757 BasicType bt = field->layout_type();
5758
5759 // Build the resultant type of the load
5760 const Type *type;
5761 if (bt == T_OBJECT) {
5762 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5763 } else {
5764 type = Type::get_const_basic_type(bt);
5765 }
5766
5767 DecoratorSet decorators = IN_HEAP;
5768
5769 if (is_vol) {
5770 decorators |= MO_SEQ_CST;
5771 }
5772
5773 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
5774 }
5775
5776 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5777 bool is_exact = true, bool is_static = false,
5778 ciInstanceKlass * fromKls = NULL) {
5779 if (fromKls == NULL) {
5780 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5781 assert(tinst != NULL, "obj is null");
5782 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5783 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5784 fromKls = tinst->klass()->as_instance_klass();
5785 }
5786 else {
5787 assert(is_static, "only for static field access");
5788 }
5789 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5790 ciSymbol::make(fieldTypeString),
5791 is_static);
5792
5793 assert(field != NULL, "undefined field");
5794 assert(!field->is_volatile(), "not defined for volatile fields");
5795
5796 if (is_static) {
5797 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5798 fromObj = makecon(tip);
5799 }
5800
5801 // Next code copied from Parse::do_get_xxx():
5802
5803 // Compute address and memory type.
5804 int offset = field->offset_in_bytes();
5805 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5806
5807 return adr;
5808 }
5809
5810 //------------------------------inline_aescrypt_Block-----------------------
5811 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
5812 address stubAddr = NULL;
5813 const char *stubName;
5814 assert(UseAES, "need AES instruction support");
5815
5816 switch(id) {
5817 case vmIntrinsics::_aescrypt_encryptBlock:
5818 stubAddr = StubRoutines::aescrypt_encryptBlock();
5819 stubName = "aescrypt_encryptBlock";
5820 break;
5821 case vmIntrinsics::_aescrypt_decryptBlock:
5822 stubAddr = StubRoutines::aescrypt_decryptBlock();
5823 stubName = "aescrypt_decryptBlock";
5824 break;
5825 default:
5826 break;
5827 }
5828 if (stubAddr == NULL) return false;
5829
5830 Node* aescrypt_object = argument(0);
5831 Node* src = argument(1);
5832 Node* src_offset = argument(2);
5833 Node* dest = argument(3);
5834 Node* dest_offset = argument(4);
5835
5836 src = must_be_not_null(src, true);
5837 dest = must_be_not_null(dest, true);
5838
5839 // (1) src and dest are arrays.
5840 const Type* src_type = src->Value(&_gvn);
5841 const Type* dest_type = dest->Value(&_gvn);
5842 const TypeAryPtr* top_src = src_type->isa_aryptr();
5843 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5844 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5845
5846 // for the quick and dirty code we will skip all the checks.
5847 // we are just trying to get the call to be generated.
5848 Node* src_start = src;
5849 Node* dest_start = dest;
5850 if (src_offset != NULL || dest_offset != NULL) {
5851 assert(src_offset != NULL && dest_offset != NULL, "");
5852 src_start = array_element_address(src, src_offset, T_BYTE);
5853 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5854 }
5855
5856 // now need to get the start of its expanded key array
5857 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5858 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5859 if (k_start == NULL) return false;
5860
5861 if (Matcher::pass_original_key_for_aes()) {
5862 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5863 // compatibility issues between Java key expansion and SPARC crypto instructions
5864 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5865 if (original_k_start == NULL) return false;
5866
5867 // Call the stub.
5868 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5869 stubAddr, stubName, TypePtr::BOTTOM,
5870 src_start, dest_start, k_start, original_k_start);
5871 } else {
5872 // Call the stub.
5873 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5874 stubAddr, stubName, TypePtr::BOTTOM,
5875 src_start, dest_start, k_start);
5876 }
5877
5878 return true;
5879 }
5880
5881 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
5882 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
5883 address stubAddr = NULL;
5884 const char *stubName = NULL;
5885
5886 assert(UseAES, "need AES instruction support");
5887
5888 switch(id) {
5889 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
5890 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
5891 stubName = "cipherBlockChaining_encryptAESCrypt";
5892 break;
5893 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
5894 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
5895 stubName = "cipherBlockChaining_decryptAESCrypt";
5896 break;
5897 default:
5898 break;
5899 }
5900 if (stubAddr == NULL) return false;
5901
5902 Node* cipherBlockChaining_object = argument(0);
5903 Node* src = argument(1);
5904 Node* src_offset = argument(2);
5905 Node* len = argument(3);
5906 Node* dest = argument(4);
5907 Node* dest_offset = argument(5);
5908
5909 src = must_be_not_null(src, false);
5910 dest = must_be_not_null(dest, false);
5911
5912 // (1) src and dest are arrays.
5913 const Type* src_type = src->Value(&_gvn);
5914 const Type* dest_type = dest->Value(&_gvn);
5915 const TypeAryPtr* top_src = src_type->isa_aryptr();
5916 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5917 assert (top_src != NULL && top_src->klass() != NULL
5918 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5919
5920 // checks are the responsibility of the caller
5921 Node* src_start = src;
5922 Node* dest_start = dest;
5923 if (src_offset != NULL || dest_offset != NULL) {
5924 assert(src_offset != NULL && dest_offset != NULL, "");
5925 src_start = array_element_address(src, src_offset, T_BYTE);
5926 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5927 }
5928
5929 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
5930 // (because of the predicated logic executed earlier).
5931 // so we cast it here safely.
5932 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5933
5934 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
5935 if (embeddedCipherObj == NULL) return false;
5936
5937 // cast it to what we know it will be at runtime
5938 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
5939 assert(tinst != NULL, "CBC obj is null");
5940 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
5941 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5942 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
5943
5944 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
5945 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
5946 const TypeOopPtr* xtype = aklass->as_instance_type();
5947 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
5948 aescrypt_object = _gvn.transform(aescrypt_object);
5949
5950 // we need to get the start of the aescrypt_object's expanded key array
5951 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5952 if (k_start == NULL) return false;
5953
5954 // similarly, get the start address of the r vector
5955 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
5956 if (objRvec == NULL) return false;
5957 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
5958
5959 Node* cbcCrypt;
5960 if (Matcher::pass_original_key_for_aes()) {
5961 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5962 // compatibility issues between Java key expansion and SPARC crypto instructions
5963 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5964 if (original_k_start == NULL) return false;
5965
5966 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
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, original_k_start);
5971 } else {
5972 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
5973 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5974 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5975 stubAddr, stubName, TypePtr::BOTTOM,
5976 src_start, dest_start, k_start, r_start, len);
5977 }
5978
5979 // return cipher length (int)
5980 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
5981 set_result(retvalue);
5982 return true;
5983 }
5984
5985 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
5986 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
5987 address stubAddr = NULL;
5988 const char *stubName = NULL;
5989
5990 assert(UseAES, "need AES instruction support");
5991
5992 switch (id) {
5993 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
5994 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
5995 stubName = "electronicCodeBook_encryptAESCrypt";
5996 break;
5997 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
5998 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
5999 stubName = "electronicCodeBook_decryptAESCrypt";
6000 break;
6001 default:
6002 break;
6003 }
6004
6005 if (stubAddr == NULL) return false;
6006
6007 Node* electronicCodeBook_object = argument(0);
6008 Node* src = argument(1);
6009 Node* src_offset = argument(2);
6010 Node* len = argument(3);
6011 Node* dest = argument(4);
6012 Node* dest_offset = argument(5);
6013
6014 // (1) src and dest are arrays.
6015 const Type* src_type = src->Value(&_gvn);
6016 const Type* dest_type = dest->Value(&_gvn);
6017 const TypeAryPtr* top_src = src_type->isa_aryptr();
6018 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6019 assert(top_src != NULL && top_src->klass() != NULL
6020 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6021
6022 // checks are the responsibility of the caller
6023 Node* src_start = src;
6024 Node* dest_start = dest;
6025 if (src_offset != NULL || dest_offset != NULL) {
6026 assert(src_offset != NULL && dest_offset != NULL, "");
6027 src_start = array_element_address(src, src_offset, T_BYTE);
6028 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6029 }
6030
6031 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6032 // (because of the predicated logic executed earlier).
6033 // so we cast it here safely.
6034 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6035
6036 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6037 if (embeddedCipherObj == NULL) return false;
6038
6039 // cast it to what we know it will be at runtime
6040 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6041 assert(tinst != NULL, "ECB obj is null");
6042 assert(tinst->klass()->is_loaded(), "ECB obj is not loaded");
6043 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6044 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6045
6046 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6047 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6048 const TypeOopPtr* xtype = aklass->as_instance_type();
6049 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6050 aescrypt_object = _gvn.transform(aescrypt_object);
6051
6052 // we need to get the start of the aescrypt_object's expanded key array
6053 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6054 if (k_start == NULL) return false;
6055
6056 Node* ecbCrypt;
6057 if (Matcher::pass_original_key_for_aes()) {
6058 // no SPARC version for AES/ECB intrinsics now.
6059 return false;
6060 }
6061 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6062 ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6063 OptoRuntime::electronicCodeBook_aescrypt_Type(),
6064 stubAddr, stubName, TypePtr::BOTTOM,
6065 src_start, dest_start, k_start, len);
6066
6067 // return cipher length (int)
6068 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6069 set_result(retvalue);
6070 return true;
6071 }
6072
6073 //------------------------------inline_counterMode_AESCrypt-----------------------
6074 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6075 assert(UseAES, "need AES instruction support");
6076 if (!UseAESCTRIntrinsics) return false;
6077
6078 address stubAddr = NULL;
6079 const char *stubName = NULL;
6080 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6081 stubAddr = StubRoutines::counterMode_AESCrypt();
6082 stubName = "counterMode_AESCrypt";
6083 }
6084 if (stubAddr == NULL) return false;
6085
6086 Node* counterMode_object = argument(0);
6087 Node* src = argument(1);
6088 Node* src_offset = argument(2);
6089 Node* len = argument(3);
6090 Node* dest = argument(4);
6091 Node* dest_offset = argument(5);
6092
6093 // (1) src and dest are arrays.
6094 const Type* src_type = src->Value(&_gvn);
6095 const Type* dest_type = dest->Value(&_gvn);
6096 const TypeAryPtr* top_src = src_type->isa_aryptr();
6097 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6098 assert(top_src != NULL && top_src->klass() != NULL &&
6099 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6100
6101 // checks are the responsibility of the caller
6102 Node* src_start = src;
6103 Node* dest_start = dest;
6104 if (src_offset != NULL || dest_offset != NULL) {
6105 assert(src_offset != NULL && dest_offset != NULL, "");
6106 src_start = array_element_address(src, src_offset, T_BYTE);
6107 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6108 }
6109
6110 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6111 // (because of the predicated logic executed earlier).
6112 // so we cast it here safely.
6113 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6114 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6115 if (embeddedCipherObj == NULL) return false;
6116 // cast it to what we know it will be at runtime
6117 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6118 assert(tinst != NULL, "CTR obj is null");
6119 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6120 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6121 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6122 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6123 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6124 const TypeOopPtr* xtype = aklass->as_instance_type();
6125 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6126 aescrypt_object = _gvn.transform(aescrypt_object);
6127 // we need to get the start of the aescrypt_object's expanded key array
6128 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6129 if (k_start == NULL) return false;
6130 // similarly, get the start address of the r vector
6131 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6132 if (obj_counter == NULL) return false;
6133 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6134
6135 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6136 if (saved_encCounter == NULL) return false;
6137 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6138 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6139
6140 Node* ctrCrypt;
6141 if (Matcher::pass_original_key_for_aes()) {
6142 // no SPARC version for AES/CTR intrinsics now.
6143 return false;
6144 }
6145 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6146 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6147 OptoRuntime::counterMode_aescrypt_Type(),
6148 stubAddr, stubName, TypePtr::BOTTOM,
6149 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6150
6151 // return cipher length (int)
6152 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6153 set_result(retvalue);
6154 return true;
6155 }
6156
6157 //------------------------------get_key_start_from_aescrypt_object-----------------------
6158 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6159 #if defined(PPC64) || defined(S390)
6160 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6161 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6162 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6163 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6164 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6165 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6166 if (objSessionK == NULL) {
6167 return (Node *) NULL;
6168 }
6169 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6170 #else
6171 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6172 #endif // PPC64
6173 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6174 if (objAESCryptKey == NULL) return (Node *) NULL;
6175
6176 // now have the array, need to get the start address of the K array
6177 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6178 return k_start;
6179 }
6180
6181 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6182 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6183 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6184 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6185 if (objAESCryptKey == NULL) return (Node *) NULL;
6186
6187 // now have the array, need to get the start address of the lastKey array
6188 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6189 return original_k_start;
6190 }
6191
6192 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6193 // Return node representing slow path of predicate check.
6194 // the pseudo code we want to emulate with this predicate is:
6195 // for encryption:
6196 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6197 // for decryption:
6198 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6199 // note cipher==plain is more conservative than the original java code but that's OK
6200 //
6201 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6202 // The receiver was checked for NULL already.
6203 Node* objCBC = argument(0);
6204
6205 Node* src = argument(1);
6206 Node* dest = argument(4);
6207
6208 // Load embeddedCipher field of CipherBlockChaining object.
6209 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6210
6211 // get AESCrypt klass for instanceOf check
6212 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6213 // will have same classloader as CipherBlockChaining object
6214 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6215 assert(tinst != NULL, "CBCobj is null");
6216 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6217
6218 // we want to do an instanceof comparison against the AESCrypt class
6219 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6220 if (!klass_AESCrypt->is_loaded()) {
6221 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6222 Node* ctrl = control();
6223 set_control(top()); // no regular fast path
6224 return ctrl;
6225 }
6226
6227 src = must_be_not_null(src, true);
6228 dest = must_be_not_null(dest, true);
6229
6230 // Resolve oops to stable for CmpP below.
6231 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6232
6233 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6234 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6235 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6236
6237 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6238
6239 // for encryption, we are done
6240 if (!decrypting)
6241 return instof_false; // even if it is NULL
6242
6243 // for decryption, we need to add a further check to avoid
6244 // taking the intrinsic path when cipher and plain are the same
6245 // see the original java code for why.
6246 RegionNode* region = new RegionNode(3);
6247 region->init_req(1, instof_false);
6248
6249 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6250 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6251 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6252 region->init_req(2, src_dest_conjoint);
6253
6254 record_for_igvn(region);
6255 return _gvn.transform(region);
6256 }
6257
6258 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
6259 // Return node representing slow path of predicate check.
6260 // the pseudo code we want to emulate with this predicate is:
6261 // for encryption:
6262 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6263 // for decryption:
6264 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6265 // note cipher==plain is more conservative than the original java code but that's OK
6266 //
6267 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
6268 // The receiver was checked for NULL already.
6269 Node* objECB = argument(0);
6270
6271 // Load embeddedCipher field of ElectronicCodeBook object.
6272 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6273
6274 // get AESCrypt klass for instanceOf check
6275 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6276 // will have same classloader as ElectronicCodeBook object
6277 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
6278 assert(tinst != NULL, "ECBobj is null");
6279 assert(tinst->klass()->is_loaded(), "ECBobj is not loaded");
6280
6281 // we want to do an instanceof comparison against the AESCrypt class
6282 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6283 if (!klass_AESCrypt->is_loaded()) {
6284 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6285 Node* ctrl = control();
6286 set_control(top()); // no regular fast path
6287 return ctrl;
6288 }
6289 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6290
6291 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6292 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6293 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6294
6295 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6296
6297 // for encryption, we are done
6298 if (!decrypting)
6299 return instof_false; // even if it is NULL
6300
6301 // for decryption, we need to add a further check to avoid
6302 // taking the intrinsic path when cipher and plain are the same
6303 // see the original java code for why.
6304 RegionNode* region = new RegionNode(3);
6305 region->init_req(1, instof_false);
6306 Node* src = argument(1);
6307 Node* dest = argument(4);
6308 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6309 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6310 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6311 region->init_req(2, src_dest_conjoint);
6312
6313 record_for_igvn(region);
6314 return _gvn.transform(region);
6315 }
6316
6317 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6318 // Return node representing slow path of predicate check.
6319 // the pseudo code we want to emulate with this predicate is:
6320 // for encryption:
6321 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6322 // for decryption:
6323 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6324 // note cipher==plain is more conservative than the original java code but that's OK
6325 //
6326
6327 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6328 // The receiver was checked for NULL already.
6329 Node* objCTR = argument(0);
6330
6331 // Load embeddedCipher field of CipherBlockChaining object.
6332 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6333
6334 // get AESCrypt klass for instanceOf check
6335 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6336 // will have same classloader as CipherBlockChaining object
6337 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6338 assert(tinst != NULL, "CTRobj is null");
6339 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6340
6341 // we want to do an instanceof comparison against the AESCrypt class
6342 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6343 if (!klass_AESCrypt->is_loaded()) {
6344 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6345 Node* ctrl = control();
6346 set_control(top()); // no regular fast path
6347 return ctrl;
6348 }
6349
6350 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6351 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6352 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6353 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6354 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6355
6356 return instof_false; // even if it is NULL
6357 }
6358
6359 //------------------------------inline_ghash_processBlocks
6360 bool LibraryCallKit::inline_ghash_processBlocks() {
6361 address stubAddr;
6362 const char *stubName;
6363 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6364
6365 stubAddr = StubRoutines::ghash_processBlocks();
6366 stubName = "ghash_processBlocks";
6367
6368 Node* data = argument(0);
6369 Node* offset = argument(1);
6370 Node* len = argument(2);
6371 Node* state = argument(3);
6372 Node* subkeyH = argument(4);
6373
6374 state = must_be_not_null(state, true);
6375 subkeyH = must_be_not_null(subkeyH, true);
6376 data = must_be_not_null(data, true);
6377
6378 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6379 assert(state_start, "state is NULL");
6380 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6381 assert(subkeyH_start, "subkeyH is NULL");
6382 Node* data_start = array_element_address(data, offset, T_BYTE);
6383 assert(data_start, "data is NULL");
6384
6385 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6386 OptoRuntime::ghash_processBlocks_Type(),
6387 stubAddr, stubName, TypePtr::BOTTOM,
6388 state_start, subkeyH_start, data_start, len);
6389 return true;
6390 }
6391
6392 bool LibraryCallKit::inline_base64_encodeBlock() {
6393 address stubAddr;
6394 const char *stubName;
6395 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6396 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6397 stubAddr = StubRoutines::base64_encodeBlock();
6398 stubName = "encodeBlock";
6399
6400 if (!stubAddr) return false;
6401 Node* base64obj = argument(0);
6402 Node* src = argument(1);
6403 Node* offset = argument(2);
6404 Node* len = argument(3);
6405 Node* dest = argument(4);
6406 Node* dp = argument(5);
6407 Node* isURL = argument(6);
6408
6409 src = must_be_not_null(src, true);
6410 dest = must_be_not_null(dest, true);
6411
6412 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6413 assert(src_start, "source array is NULL");
6414 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6415 assert(dest_start, "destination array is NULL");
6416
6417 Node* base64 = make_runtime_call(RC_LEAF,
6418 OptoRuntime::base64_encodeBlock_Type(),
6419 stubAddr, stubName, TypePtr::BOTTOM,
6420 src_start, offset, len, dest_start, dp, isURL);
6421 return true;
6422 }
6423
6424 //------------------------------inline_digestBase_implCompress-----------------------
6425 //
6426 // Calculate MD5 for single-block byte[] array.
6427 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
6428 //
6429 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6430 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6431 //
6432 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6433 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6434 //
6435 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6436 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6437 //
6438 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
6439 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6440
6441 Node* digestBase_obj = argument(0);
6442 Node* src = argument(1); // type oop
6443 Node* ofs = argument(2); // type int
6444
6445 const Type* src_type = src->Value(&_gvn);
6446 const TypeAryPtr* top_src = src_type->isa_aryptr();
6447 if (top_src == NULL || top_src->klass() == NULL) {
6448 // failed array check
6449 return false;
6450 }
6451 // Figure out the size and type of the elements we will be copying.
6452 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6453 if (src_elem != T_BYTE) {
6454 return false;
6455 }
6456 // 'src_start' points to src array + offset
6457 src = must_be_not_null(src, true);
6458 Node* src_start = array_element_address(src, ofs, src_elem);
6459 Node* state = NULL;
6460 address stubAddr;
6461 const char *stubName;
6462
6463 switch(id) {
6464 case vmIntrinsics::_md5_implCompress:
6465 assert(UseMD5Intrinsics, "need MD5 instruction support");
6466 state = get_state_from_digest_object(digestBase_obj);
6467 stubAddr = StubRoutines::md5_implCompress();
6468 stubName = "md5_implCompress";
6469 break;
6470 case vmIntrinsics::_sha_implCompress:
6471 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6472 state = get_state_from_digest_object(digestBase_obj);
6473 stubAddr = StubRoutines::sha1_implCompress();
6474 stubName = "sha1_implCompress";
6475 break;
6476 case vmIntrinsics::_sha2_implCompress:
6477 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6478 state = get_state_from_digest_object(digestBase_obj);
6479 stubAddr = StubRoutines::sha256_implCompress();
6480 stubName = "sha256_implCompress";
6481 break;
6482 case vmIntrinsics::_sha5_implCompress:
6483 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6484 state = get_long_state_from_digest_object(digestBase_obj);
6485 stubAddr = StubRoutines::sha512_implCompress();
6486 stubName = "sha512_implCompress";
6487 break;
6488 default:
6489 fatal_unexpected_iid(id);
6490 return false;
6491 }
6492 if (state == NULL) return false;
6493
6494 assert(stubAddr != NULL, "Stub is generated");
6495 if (stubAddr == NULL) return false;
6496
6497 // Call the stub.
6498 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(),
6499 stubAddr, stubName, TypePtr::BOTTOM,
6500 src_start, state);
6501
6502 return true;
6503 }
6504
6505 //------------------------------inline_digestBase_implCompressMB-----------------------
6506 //
6507 // Calculate MD5/SHA/SHA2/SHA5 for multi-block byte[] array.
6508 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6509 //
6510 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6511 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6512 "need MD5/SHA1/SHA256/SHA512 instruction support");
6513 assert((uint)predicate < 4, "sanity");
6514 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6515
6516 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6517 Node* src = argument(1); // byte[] array
6518 Node* ofs = argument(2); // type int
6519 Node* limit = argument(3); // type int
6520
6521 const Type* src_type = src->Value(&_gvn);
6522 const TypeAryPtr* top_src = src_type->isa_aryptr();
6523 if (top_src == NULL || top_src->klass() == NULL) {
6524 // failed array check
6525 return false;
6526 }
6527 // Figure out the size and type of the elements we will be copying.
6528 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6529 if (src_elem != T_BYTE) {
6530 return false;
6531 }
6532 // 'src_start' points to src array + offset
6533 src = must_be_not_null(src, false);
6534 Node* src_start = array_element_address(src, ofs, src_elem);
6535
6536 const char* klass_digestBase_name = NULL;
6537 const char* stub_name = NULL;
6538 address stub_addr = NULL;
6539 bool long_state = false;
6540
6541 switch (predicate) {
6542 case 0:
6543 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
6544 klass_digestBase_name = "sun/security/provider/MD5";
6545 stub_name = "md5_implCompressMB";
6546 stub_addr = StubRoutines::md5_implCompressMB();
6547 }
6548 break;
6549 case 1:
6550 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
6551 klass_digestBase_name = "sun/security/provider/SHA";
6552 stub_name = "sha1_implCompressMB";
6553 stub_addr = StubRoutines::sha1_implCompressMB();
6554 }
6555 break;
6556 case 2:
6557 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
6558 klass_digestBase_name = "sun/security/provider/SHA2";
6559 stub_name = "sha256_implCompressMB";
6560 stub_addr = StubRoutines::sha256_implCompressMB();
6561 }
6562 break;
6563 case 3:
6564 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
6565 klass_digestBase_name = "sun/security/provider/SHA5";
6566 stub_name = "sha512_implCompressMB";
6567 stub_addr = StubRoutines::sha512_implCompressMB();
6568 long_state = true;
6569 }
6570 break;
6571 default:
6572 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
6573 }
6574 if (klass_digestBase_name != NULL) {
6575 assert(stub_addr != NULL, "Stub is generated");
6576 if (stub_addr == NULL) return false;
6577
6578 // get DigestBase klass to lookup for SHA klass
6579 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6580 assert(tinst != NULL, "digestBase_obj is not instance???");
6581 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6582
6583 ciKlass* klass_digestBase = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
6584 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
6585 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
6586 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, long_state, stub_addr, stub_name, src_start, ofs, limit);
6587 }
6588 return false;
6589 }
6590
6591 //------------------------------inline_digestBase_implCompressMB-----------------------
6592 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
6593 bool long_state, address stubAddr, const char *stubName,
6594 Node* src_start, Node* ofs, Node* limit) {
6595 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
6596 const TypeOopPtr* xtype = aklass->as_instance_type();
6597 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6598 digest_obj = _gvn.transform(digest_obj);
6599
6600 Node* state;
6601 if (long_state) {
6602 state = get_long_state_from_digest_object(digest_obj);
6603 } else {
6604 state = get_state_from_digest_object(digest_obj);
6605 }
6606 if (state == NULL) return false;
6607
6608 // Call the stub.
6609 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6610 OptoRuntime::digestBase_implCompressMB_Type(),
6611 stubAddr, stubName, TypePtr::BOTTOM,
6612 src_start, state, ofs, limit);
6613 // return ofs (int)
6614 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6615 set_result(result);
6616
6617 return true;
6618 }
6619
6620 //------------------------------get_state_from_digest_object-----------------------
6621 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object) {
6622 Node* digest_state = load_field_from_object(digest_object, "state", "[I", /*is_exact*/ false);
6623 assert (digest_state != NULL, "wrong version of sun.security.provider.MD5/SHA/SHA2");
6624 if (digest_state == NULL) return (Node *) NULL;
6625
6626 // now have the array, need to get the start address of the state array
6627 Node* state = array_element_address(digest_state, intcon(0), T_INT);
6628 return state;
6629 }
6630
6631 //------------------------------get_long_state_from_digest_object-----------------------
6632 Node * LibraryCallKit::get_long_state_from_digest_object(Node *digest_object) {
6633 Node* digest_state = load_field_from_object(digest_object, "state", "[J", /*is_exact*/ false);
6634 assert (digest_state != NULL, "wrong version of sun.security.provider.SHA5");
6635 if (digest_state == NULL) return (Node *) NULL;
6636
6637 // now have the array, need to get the start address of the state array
6638 Node* state = array_element_address(digest_state, intcon(0), T_LONG);
6639 return state;
6640 }
6641
6642 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6643 // Return node representing slow path of predicate check.
6644 // the pseudo code we want to emulate with this predicate is:
6645 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6646 //
6647 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6648 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6649 "need MD5/SHA1/SHA256/SHA512 instruction support");
6650 assert((uint)predicate < 4, "sanity");
6651
6652 // The receiver was checked for NULL already.
6653 Node* digestBaseObj = argument(0);
6654
6655 // get DigestBase klass for instanceOf check
6656 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6657 assert(tinst != NULL, "digestBaseObj is null");
6658 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6659
6660 const char* klass_name = NULL;
6661 switch (predicate) {
6662 case 0:
6663 if (UseMD5Intrinsics) {
6664 // we want to do an instanceof comparison against the MD5 class
6665 klass_name = "sun/security/provider/MD5";
6666 }
6667 break;
6668 case 1:
6669 if (UseSHA1Intrinsics) {
6670 // we want to do an instanceof comparison against the SHA class
6671 klass_name = "sun/security/provider/SHA";
6672 }
6673 break;
6674 case 2:
6675 if (UseSHA256Intrinsics) {
6676 // we want to do an instanceof comparison against the SHA2 class
6677 klass_name = "sun/security/provider/SHA2";
6678 }
6679 break;
6680 case 3:
6681 if (UseSHA512Intrinsics) {
6682 // we want to do an instanceof comparison against the SHA5 class
6683 klass_name = "sun/security/provider/SHA5";
6684 }
6685 break;
6686 default:
6687 fatal("unknown SHA intrinsic predicate: %d", predicate);
6688 }
6689
6690 ciKlass* klass = NULL;
6691 if (klass_name != NULL) {
6692 klass = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_name));
6693 }
6694 if ((klass == NULL) || !klass->is_loaded()) {
6695 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6696 Node* ctrl = control();
6697 set_control(top()); // no intrinsic path
6698 return ctrl;
6699 }
6700 ciInstanceKlass* instklass = klass->as_instance_klass();
6701
6702 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
6703 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6704 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6705 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6706
6707 return instof_false; // even if it is NULL
6708 }
6709
6710 //-------------inline_fma-----------------------------------
6711 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
6712 Node *a = NULL;
6713 Node *b = NULL;
6714 Node *c = NULL;
6715 Node* result = NULL;
6716 switch (id) {
6717 case vmIntrinsics::_fmaD:
6718 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
6719 // no receiver since it is static method
6720 a = round_double_node(argument(0));
6721 b = round_double_node(argument(2));
6722 c = round_double_node(argument(4));
6723 result = _gvn.transform(new FmaDNode(control(), a, b, c));
6724 break;
6725 case vmIntrinsics::_fmaF:
6726 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
6727 a = argument(0);
6728 b = argument(1);
6729 c = argument(2);
6730 result = _gvn.transform(new FmaFNode(control(), a, b, c));
6731 break;
6732 default:
6733 fatal_unexpected_iid(id); break;
6734 }
6735 set_result(result);
6736 return true;
6737 }
6738
6739 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
6740 // argument(0) is receiver
6741 Node* codePoint = argument(1);
6742 Node* n = NULL;
6743
6744 switch (id) {
6745 case vmIntrinsics::_isDigit :
6746 n = new DigitNode(control(), codePoint);
6747 break;
6748 case vmIntrinsics::_isLowerCase :
6749 n = new LowerCaseNode(control(), codePoint);
6750 break;
6751 case vmIntrinsics::_isUpperCase :
6752 n = new UpperCaseNode(control(), codePoint);
6753 break;
6754 case vmIntrinsics::_isWhitespace :
6755 n = new WhitespaceNode(control(), codePoint);
6756 break;
6757 default:
6758 fatal_unexpected_iid(id);
6759 }
6760
6761 set_result(_gvn.transform(n));
6762 return true;
6763 }
6764
6765 //------------------------------inline_fp_min_max------------------------------
6766 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
6767 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
6768
6769 // The intrinsic should be used only when the API branches aren't predictable,
6770 // the last one performing the most important comparison. The following heuristic
6771 // uses the branch statistics to eventually bail out if necessary.
6772
6773 ciMethodData *md = callee()->method_data();
6774
6775 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
6776 ciCallProfile cp = caller()->call_profile_at_bci(bci());
6777
6778 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
6779 // Bail out if the call-site didn't contribute enough to the statistics.
6780 return false;
6781 }
6782
6783 uint taken = 0, not_taken = 0;
6784
6785 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
6786 if (p->is_BranchData()) {
6787 taken = ((ciBranchData*)p)->taken();
6788 not_taken = ((ciBranchData*)p)->not_taken();
6789 }
6790 }
6791
6792 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
6793 balance = balance < 0 ? -balance : balance;
6794 if ( balance > 0.2 ) {
6795 // Bail out if the most important branch is predictable enough.
6796 return false;
6797 }
6798 }
6799 */
6800
6801 Node *a = NULL;
6802 Node *b = NULL;
6803 Node *n = NULL;
6804 switch (id) {
6805 case vmIntrinsics::_maxF:
6806 case vmIntrinsics::_minF:
6807 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
6808 a = argument(0);
6809 b = argument(1);
6810 break;
6811 case vmIntrinsics::_maxD:
6812 case vmIntrinsics::_minD:
6813 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
6814 a = round_double_node(argument(0));
6815 b = round_double_node(argument(2));
6816 break;
6817 default:
6818 fatal_unexpected_iid(id);
6819 break;
6820 }
6821 switch (id) {
6822 case vmIntrinsics::_maxF: n = new MaxFNode(a, b); break;
6823 case vmIntrinsics::_minF: n = new MinFNode(a, b); break;
6824 case vmIntrinsics::_maxD: n = new MaxDNode(a, b); break;
6825 case vmIntrinsics::_minD: n = new MinDNode(a, b); break;
6826 default: fatal_unexpected_iid(id); break;
6827 }
6828 set_result(_gvn.transform(n));
6829 return true;
6830 }
6831
6832 bool LibraryCallKit::inline_profileBoolean() {
6833 Node* counts = argument(1);
6834 const TypeAryPtr* ary = NULL;
6835 ciArray* aobj = NULL;
6836 if (counts->is_Con()
6837 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6838 && (aobj = ary->const_oop()->as_array()) != NULL
6839 && (aobj->length() == 2)) {
6840 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6841 jint false_cnt = aobj->element_value(0).as_int();
6842 jint true_cnt = aobj->element_value(1).as_int();
6843
6844 if (C->log() != NULL) {
6845 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6846 false_cnt, true_cnt);
6847 }
6848
6849 if (false_cnt + true_cnt == 0) {
6850 // According to profile, never executed.
6851 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6852 Deoptimization::Action_reinterpret);
6853 return true;
6854 }
6855
6856 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6857 // is a number of each value occurrences.
6858 Node* result = argument(0);
6859 if (false_cnt == 0 || true_cnt == 0) {
6860 // According to profile, one value has been never seen.
6861 int expected_val = (false_cnt == 0) ? 1 : 0;
6862
6863 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6864 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6865
6866 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6867 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6868 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6869
6870 { // Slow path: uncommon trap for never seen value and then reexecute
6871 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6872 // the value has been seen at least once.
6873 PreserveJVMState pjvms(this);
6874 PreserveReexecuteState preexecs(this);
6875 jvms()->set_should_reexecute(true);
6876
6877 set_control(slow_path);
6878 set_i_o(i_o());
6879
6880 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6881 Deoptimization::Action_reinterpret);
6882 }
6883 // The guard for never seen value enables sharpening of the result and
6884 // returning a constant. It allows to eliminate branches on the same value
6885 // later on.
6886 set_control(fast_path);
6887 result = intcon(expected_val);
6888 }
6889 // Stop profiling.
6890 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6891 // By replacing method body with profile data (represented as ProfileBooleanNode
6892 // on IR level) we effectively disable profiling.
6893 // It enables full speed execution once optimized code is generated.
6894 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6895 C->record_for_igvn(profile);
6896 set_result(profile);
6897 return true;
6898 } else {
6899 // Continue profiling.
6900 // Profile data isn't available at the moment. So, execute method's bytecode version.
6901 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6902 // is compiled and counters aren't available since corresponding MethodHandle
6903 // isn't a compile-time constant.
6904 return false;
6905 }
6906 }
6907
6908 bool LibraryCallKit::inline_isCompileConstant() {
6909 Node* n = argument(0);
6910 set_result(n->is_Con() ? intcon(1) : intcon(0));
6911 return true;
6912 }