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: return inline_math_native(intrinsic_id());
555
556 case vmIntrinsics::_min:
557 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
558
559 case vmIntrinsics::_notify:
560 case vmIntrinsics::_notifyAll:
561 return inline_notify(intrinsic_id());
562
563 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
564 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
565 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
566 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
567 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
568 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
569 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
570 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
571 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
572 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
573 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
574 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
575 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
576
577 case vmIntrinsics::_arraycopy: return inline_arraycopy();
578
579 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
580 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
581 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
582 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
583
584 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
585 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
586 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
587 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
588 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
589 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
590 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
591
592 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
593 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
594
595 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
596 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
597 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
598 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
599
600 case vmIntrinsics::_compressStringC:
601 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
602 case vmIntrinsics::_inflateStringC:
603 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
604
605 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
606 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
607 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
608 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
609 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
610 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
611 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
612 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
613 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
614
615 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
616 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
617 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
618 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
619 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
620 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
621 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
622 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
623 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
624
625 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
626 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
627 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
628 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
629 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
630 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
631 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
632 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
633 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
634
635 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
636 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
637 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
638 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
639 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
640 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
641 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
642 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
643 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
644
645 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
646 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
647 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
648 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
649
650 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
651 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
652 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
653 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
654
655 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
656 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
657 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
658 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
659 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
660 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
661 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
662 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
663 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
664
665 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
666 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
667 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
668 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
669 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
670 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
671 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
672 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
673 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
674
675 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
676 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
677 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
678 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
679 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
680 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
681 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
682 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
683 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
684
685 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
686 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
687 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
688 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
689 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
690 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
691 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
692 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
693 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
694
695 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
696 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
697 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
698 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
699 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
700
701 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
702 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
703 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
704 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
705 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
706 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
707 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
708 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
709 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
710 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
711 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
712 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
713 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
714 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
715 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
716 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
717 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
718 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
719 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
720 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
721
722 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
723 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
724 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
725 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
726 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
727 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
728 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
729 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
730 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
731 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
732 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
733 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
734 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
735 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
736 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
737
738 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
739 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
740 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
741 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
742
743 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
744 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
745 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
746 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
747 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
748
749 case vmIntrinsics::_loadFence:
750 case vmIntrinsics::_storeFence:
751 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
752
753 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
754
755 case vmIntrinsics::_currentThread: return inline_native_currentThread();
756
757 #ifdef JFR_HAVE_INTRINSICS
758 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime");
759 case vmIntrinsics::_getClassId: return inline_native_classID();
760 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
761 #endif
762 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
763 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
764 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
765 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
766 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
767 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
768 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
769 case vmIntrinsics::_getLength: return inline_native_getLength();
770 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
771 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
772 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
773 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
774 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
775 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
776
777 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
778 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
779
780 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
781
782 case vmIntrinsics::_isInstance:
783 case vmIntrinsics::_getModifiers:
784 case vmIntrinsics::_isInterface:
785 case vmIntrinsics::_isArray:
786 case vmIntrinsics::_isPrimitive:
787 case vmIntrinsics::_isHidden:
788 case vmIntrinsics::_getSuperclass:
789 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
790
791 case vmIntrinsics::_floatToRawIntBits:
792 case vmIntrinsics::_floatToIntBits:
793 case vmIntrinsics::_intBitsToFloat:
794 case vmIntrinsics::_doubleToRawLongBits:
795 case vmIntrinsics::_doubleToLongBits:
796 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
797
798 case vmIntrinsics::_numberOfLeadingZeros_i:
799 case vmIntrinsics::_numberOfLeadingZeros_l:
800 case vmIntrinsics::_numberOfTrailingZeros_i:
801 case vmIntrinsics::_numberOfTrailingZeros_l:
802 case vmIntrinsics::_bitCount_i:
803 case vmIntrinsics::_bitCount_l:
804 case vmIntrinsics::_reverseBytes_i:
805 case vmIntrinsics::_reverseBytes_l:
806 case vmIntrinsics::_reverseBytes_s:
807 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
808
809 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
810
811 case vmIntrinsics::_Reference_get: return inline_reference_get();
812
813 case vmIntrinsics::_Class_cast: return inline_Class_cast();
814
815 case vmIntrinsics::_aescrypt_encryptBlock:
816 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
817
818 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
819 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
820 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
821
822 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
823 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
824 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
825
826 case vmIntrinsics::_counterMode_AESCrypt:
827 return inline_counterMode_AESCrypt(intrinsic_id());
828
829 case vmIntrinsics::_md5_implCompress:
830 case vmIntrinsics::_sha_implCompress:
831 case vmIntrinsics::_sha2_implCompress:
832 case vmIntrinsics::_sha5_implCompress:
833 return inline_digestBase_implCompress(intrinsic_id());
834
835 case vmIntrinsics::_digestBase_implCompressMB:
836 return inline_digestBase_implCompressMB(predicate);
837
838 case vmIntrinsics::_multiplyToLen:
839 return inline_multiplyToLen();
840
841 case vmIntrinsics::_squareToLen:
842 return inline_squareToLen();
843
844 case vmIntrinsics::_mulAdd:
845 return inline_mulAdd();
846
847 case vmIntrinsics::_montgomeryMultiply:
848 return inline_montgomeryMultiply();
849 case vmIntrinsics::_montgomerySquare:
850 return inline_montgomerySquare();
851
852 case vmIntrinsics::_bigIntegerRightShiftWorker:
853 return inline_bigIntegerShift(true);
854 case vmIntrinsics::_bigIntegerLeftShiftWorker:
855 return inline_bigIntegerShift(false);
856
857 case vmIntrinsics::_vectorizedMismatch:
858 return inline_vectorizedMismatch();
859
860 case vmIntrinsics::_ghash_processBlocks:
861 return inline_ghash_processBlocks();
862 case vmIntrinsics::_base64_encodeBlock:
863 return inline_base64_encodeBlock();
864
865 case vmIntrinsics::_encodeISOArray:
866 case vmIntrinsics::_encodeByteISOArray:
867 return inline_encodeISOArray();
868
869 case vmIntrinsics::_updateCRC32:
870 return inline_updateCRC32();
871 case vmIntrinsics::_updateBytesCRC32:
872 return inline_updateBytesCRC32();
873 case vmIntrinsics::_updateByteBufferCRC32:
874 return inline_updateByteBufferCRC32();
875
876 case vmIntrinsics::_updateBytesCRC32C:
877 return inline_updateBytesCRC32C();
878 case vmIntrinsics::_updateDirectByteBufferCRC32C:
879 return inline_updateDirectByteBufferCRC32C();
880
881 case vmIntrinsics::_updateBytesAdler32:
882 return inline_updateBytesAdler32();
883 case vmIntrinsics::_updateByteBufferAdler32:
884 return inline_updateByteBufferAdler32();
885
886 case vmIntrinsics::_profileBoolean:
887 return inline_profileBoolean();
888 case vmIntrinsics::_isCompileConstant:
889 return inline_isCompileConstant();
890
891 case vmIntrinsics::_hasNegatives:
892 return inline_hasNegatives();
893
894 case vmIntrinsics::_fmaD:
895 case vmIntrinsics::_fmaF:
896 return inline_fma(intrinsic_id());
897
898 case vmIntrinsics::_isDigit:
899 case vmIntrinsics::_isLowerCase:
900 case vmIntrinsics::_isUpperCase:
901 case vmIntrinsics::_isWhitespace:
902 return inline_character_compare(intrinsic_id());
903
904 case vmIntrinsics::_maxF:
905 case vmIntrinsics::_minF:
906 case vmIntrinsics::_maxD:
907 case vmIntrinsics::_minD:
908 return inline_fp_min_max(intrinsic_id());
909
910 default:
911 // If you get here, it may be that someone has added a new intrinsic
912 // to the list in vmSymbols.hpp without implementing it here.
913 #ifndef PRODUCT
914 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
915 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
916 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
917 }
918 #endif
919 return false;
920 }
921 }
922
923 Node* LibraryCallKit::try_to_predicate(int predicate) {
924 if (!jvms()->has_method()) {
925 // Root JVMState has a null method.
926 assert(map()->memory()->Opcode() == Op_Parm, "");
927 // Insert the memory aliasing node
928 set_all_memory(reset_memory());
929 }
930 assert(merged_memory(), "");
931
932 switch (intrinsic_id()) {
933 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
934 return inline_cipherBlockChaining_AESCrypt_predicate(false);
935 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
936 return inline_cipherBlockChaining_AESCrypt_predicate(true);
937 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
938 return inline_electronicCodeBook_AESCrypt_predicate(false);
939 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
940 return inline_electronicCodeBook_AESCrypt_predicate(true);
941 case vmIntrinsics::_counterMode_AESCrypt:
942 return inline_counterMode_AESCrypt_predicate();
943 case vmIntrinsics::_digestBase_implCompressMB:
944 return inline_digestBase_implCompressMB_predicate(predicate);
945
946 default:
947 // If you get here, it may be that someone has added a new intrinsic
948 // to the list in vmSymbols.hpp without implementing it here.
949 #ifndef PRODUCT
950 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
951 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
952 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
953 }
954 #endif
955 Node* slow_ctl = control();
956 set_control(top()); // No fast path instrinsic
957 return slow_ctl;
958 }
959 }
960
961 //------------------------------set_result-------------------------------
962 // Helper function for finishing intrinsics.
963 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
964 record_for_igvn(region);
965 set_control(_gvn.transform(region));
966 set_result( _gvn.transform(value));
967 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
968 }
969
970 //------------------------------generate_guard---------------------------
971 // Helper function for generating guarded fast-slow graph structures.
972 // The given 'test', if true, guards a slow path. If the test fails
973 // then a fast path can be taken. (We generally hope it fails.)
974 // In all cases, GraphKit::control() is updated to the fast path.
975 // The returned value represents the control for the slow path.
976 // The return value is never 'top'; it is either a valid control
977 // or NULL if it is obvious that the slow path can never be taken.
978 // Also, if region and the slow control are not NULL, the slow edge
979 // is appended to the region.
980 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
981 if (stopped()) {
982 // Already short circuited.
983 return NULL;
984 }
985
986 // Build an if node and its projections.
987 // If test is true we take the slow path, which we assume is uncommon.
988 if (_gvn.type(test) == TypeInt::ZERO) {
989 // The slow branch is never taken. No need to build this guard.
990 return NULL;
991 }
992
993 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
994
995 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
996 if (if_slow == top()) {
997 // The slow branch is never taken. No need to build this guard.
998 return NULL;
999 }
1000
1001 if (region != NULL)
1002 region->add_req(if_slow);
1003
1004 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
1005 set_control(if_fast);
1006
1007 return if_slow;
1008 }
1009
1010 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
1011 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
1012 }
1013 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
1014 return generate_guard(test, region, PROB_FAIR);
1015 }
1016
1017 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
1018 Node* *pos_index) {
1019 if (stopped())
1020 return NULL; // already stopped
1021 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
1022 return NULL; // index is already adequately typed
1023 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
1024 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1025 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
1026 if (is_neg != NULL && pos_index != NULL) {
1027 // Emulate effect of Parse::adjust_map_after_if.
1028 Node* ccast = new CastIINode(index, TypeInt::POS);
1029 ccast->set_req(0, control());
1030 (*pos_index) = _gvn.transform(ccast);
1031 }
1032 return is_neg;
1033 }
1034
1035 // Make sure that 'position' is a valid limit index, in [0..length].
1036 // There are two equivalent plans for checking this:
1037 // A. (offset + copyLength) unsigned<= arrayLength
1038 // B. offset <= (arrayLength - copyLength)
1039 // We require that all of the values above, except for the sum and
1040 // difference, are already known to be non-negative.
1041 // Plan A is robust in the face of overflow, if offset and copyLength
1042 // are both hugely positive.
1043 //
1044 // Plan B is less direct and intuitive, but it does not overflow at
1045 // all, since the difference of two non-negatives is always
1046 // representable. Whenever Java methods must perform the equivalent
1047 // check they generally use Plan B instead of Plan A.
1048 // For the moment we use Plan A.
1049 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
1050 Node* subseq_length,
1051 Node* array_length,
1052 RegionNode* region) {
1053 if (stopped())
1054 return NULL; // already stopped
1055 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
1056 if (zero_offset && subseq_length->eqv_uncast(array_length))
1057 return NULL; // common case of whole-array copy
1058 Node* last = subseq_length;
1059 if (!zero_offset) // last += offset
1060 last = _gvn.transform(new AddINode(last, offset));
1061 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
1062 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1063 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
1064 return is_over;
1065 }
1066
1067 // Emit range checks for the given String.value byte array
1068 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
1069 if (stopped()) {
1070 return; // already stopped
1071 }
1072 RegionNode* bailout = new RegionNode(1);
1073 record_for_igvn(bailout);
1074 if (char_count) {
1075 // Convert char count to byte count
1076 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1077 }
1078
1079 // Offset and count must not be negative
1080 generate_negative_guard(offset, bailout);
1081 generate_negative_guard(count, bailout);
1082 // Offset + count must not exceed length of array
1083 generate_limit_guard(offset, count, load_array_length(array), bailout);
1084
1085 if (bailout->req() > 1) {
1086 PreserveJVMState pjvms(this);
1087 set_control(_gvn.transform(bailout));
1088 uncommon_trap(Deoptimization::Reason_intrinsic,
1089 Deoptimization::Action_maybe_recompile);
1090 }
1091 }
1092
1093 //--------------------------generate_current_thread--------------------
1094 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1095 ciKlass* thread_klass = env()->Thread_klass();
1096 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1097 Node* thread = _gvn.transform(new ThreadLocalNode());
1098 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1099 tls_output = thread;
1100 Node* thread_obj_handle = LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(), TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
1101 thread_obj_handle = _gvn.transform(thread_obj_handle);
1102 return access_load(thread_obj_handle, thread_type, T_OBJECT, IN_NATIVE | C2_IMMUTABLE_MEMORY);
1103 }
1104
1105
1106 //------------------------------make_string_method_node------------------------
1107 // Helper method for String intrinsic functions. This version is called with
1108 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1109 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1110 // containing the lengths of str1 and str2.
1111 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1112 Node* result = NULL;
1113 switch (opcode) {
1114 case Op_StrIndexOf:
1115 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1116 str1_start, cnt1, str2_start, cnt2, ae);
1117 break;
1118 case Op_StrComp:
1119 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1120 str1_start, cnt1, str2_start, cnt2, ae);
1121 break;
1122 case Op_StrEquals:
1123 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1124 // Use the constant length if there is one because optimized match rule may exist.
1125 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1126 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1127 break;
1128 default:
1129 ShouldNotReachHere();
1130 return NULL;
1131 }
1132
1133 // All these intrinsics have checks.
1134 C->set_has_split_ifs(true); // Has chance for split-if optimization
1135 clear_upper_avx();
1136
1137 return _gvn.transform(result);
1138 }
1139
1140 //------------------------------inline_string_compareTo------------------------
1141 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1142 Node* arg1 = argument(0);
1143 Node* arg2 = argument(1);
1144
1145 arg1 = must_be_not_null(arg1, true);
1146 arg2 = must_be_not_null(arg2, true);
1147
1148 // Get start addr and length of first argument
1149 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1150 Node* arg1_cnt = load_array_length(arg1);
1151
1152 // Get start addr and length of second argument
1153 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1154 Node* arg2_cnt = load_array_length(arg2);
1155
1156 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1157 set_result(result);
1158 return true;
1159 }
1160
1161 //------------------------------inline_string_equals------------------------
1162 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1163 Node* arg1 = argument(0);
1164 Node* arg2 = argument(1);
1165
1166 // paths (plus control) merge
1167 RegionNode* region = new RegionNode(3);
1168 Node* phi = new PhiNode(region, TypeInt::BOOL);
1169
1170 if (!stopped()) {
1171
1172 arg1 = must_be_not_null(arg1, true);
1173 arg2 = must_be_not_null(arg2, true);
1174
1175 // Get start addr and length of first argument
1176 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1177 Node* arg1_cnt = load_array_length(arg1);
1178
1179 // Get start addr and length of second argument
1180 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1181 Node* arg2_cnt = load_array_length(arg2);
1182
1183 // Check for arg1_cnt != arg2_cnt
1184 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1185 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1186 Node* if_ne = generate_slow_guard(bol, NULL);
1187 if (if_ne != NULL) {
1188 phi->init_req(2, intcon(0));
1189 region->init_req(2, if_ne);
1190 }
1191
1192 // Check for count == 0 is done by assembler code for StrEquals.
1193
1194 if (!stopped()) {
1195 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1196 phi->init_req(1, equals);
1197 region->init_req(1, control());
1198 }
1199 }
1200
1201 // post merge
1202 set_control(_gvn.transform(region));
1203 record_for_igvn(region);
1204
1205 set_result(_gvn.transform(phi));
1206 return true;
1207 }
1208
1209 //------------------------------inline_array_equals----------------------------
1210 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1211 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1212 Node* arg1 = argument(0);
1213 Node* arg2 = argument(1);
1214
1215 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1216 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1217 clear_upper_avx();
1218
1219 return true;
1220 }
1221
1222 //------------------------------inline_hasNegatives------------------------------
1223 bool LibraryCallKit::inline_hasNegatives() {
1224 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1225 return false;
1226 }
1227
1228 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1229 // no receiver since it is static method
1230 Node* ba = argument(0);
1231 Node* offset = argument(1);
1232 Node* len = argument(2);
1233
1234 ba = must_be_not_null(ba, true);
1235
1236 // Range checks
1237 generate_string_range_check(ba, offset, len, false);
1238 if (stopped()) {
1239 return true;
1240 }
1241 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1242 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1243 set_result(_gvn.transform(result));
1244 return true;
1245 }
1246
1247 bool LibraryCallKit::inline_preconditions_checkIndex() {
1248 Node* index = argument(0);
1249 Node* length = argument(1);
1250 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1251 return false;
1252 }
1253
1254 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1255 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1256
1257 {
1258 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1259 uncommon_trap(Deoptimization::Reason_intrinsic,
1260 Deoptimization::Action_make_not_entrant);
1261 }
1262
1263 if (stopped()) {
1264 return false;
1265 }
1266
1267 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1268 BoolTest::mask btest = BoolTest::lt;
1269 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1270 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1271 _gvn.set_type(rc, rc->Value(&_gvn));
1272 if (!rc_bool->is_Con()) {
1273 record_for_igvn(rc);
1274 }
1275 set_control(_gvn.transform(new IfTrueNode(rc)));
1276 {
1277 PreserveJVMState pjvms(this);
1278 set_control(_gvn.transform(new IfFalseNode(rc)));
1279 uncommon_trap(Deoptimization::Reason_range_check,
1280 Deoptimization::Action_make_not_entrant);
1281 }
1282
1283 if (stopped()) {
1284 return false;
1285 }
1286
1287 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1288 result->set_req(0, control());
1289 result = _gvn.transform(result);
1290 set_result(result);
1291 replace_in_map(index, result);
1292 clear_upper_avx();
1293 return true;
1294 }
1295
1296 //------------------------------inline_string_indexOf------------------------
1297 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1298 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1299 return false;
1300 }
1301 Node* src = argument(0);
1302 Node* tgt = argument(1);
1303
1304 // Make the merge point
1305 RegionNode* result_rgn = new RegionNode(4);
1306 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1307
1308 src = must_be_not_null(src, true);
1309 tgt = must_be_not_null(tgt, true);
1310
1311 // Get start addr and length of source string
1312 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1313 Node* src_count = load_array_length(src);
1314
1315 // Get start addr and length of substring
1316 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1317 Node* tgt_count = load_array_length(tgt);
1318
1319 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1320 // Divide src size by 2 if String is UTF16 encoded
1321 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1322 }
1323 if (ae == StrIntrinsicNode::UU) {
1324 // Divide substring size by 2 if String is UTF16 encoded
1325 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1326 }
1327
1328 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1329 if (result != NULL) {
1330 result_phi->init_req(3, result);
1331 result_rgn->init_req(3, control());
1332 }
1333 set_control(_gvn.transform(result_rgn));
1334 record_for_igvn(result_rgn);
1335 set_result(_gvn.transform(result_phi));
1336
1337 return true;
1338 }
1339
1340 //-----------------------------inline_string_indexOf-----------------------
1341 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1342 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1343 return false;
1344 }
1345 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1346 return false;
1347 }
1348 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1349 Node* src = argument(0); // byte[]
1350 Node* src_count = argument(1); // char count
1351 Node* tgt = argument(2); // byte[]
1352 Node* tgt_count = argument(3); // char count
1353 Node* from_index = argument(4); // char index
1354
1355 src = must_be_not_null(src, true);
1356 tgt = must_be_not_null(tgt, true);
1357
1358 // Multiply byte array index by 2 if String is UTF16 encoded
1359 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1360 src_count = _gvn.transform(new SubINode(src_count, from_index));
1361 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1362 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1363
1364 // Range checks
1365 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1366 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1367 if (stopped()) {
1368 return true;
1369 }
1370
1371 RegionNode* region = new RegionNode(5);
1372 Node* phi = new PhiNode(region, TypeInt::INT);
1373
1374 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1375 if (result != NULL) {
1376 // The result is index relative to from_index if substring was found, -1 otherwise.
1377 // Generate code which will fold into cmove.
1378 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1379 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1380
1381 Node* if_lt = generate_slow_guard(bol, NULL);
1382 if (if_lt != NULL) {
1383 // result == -1
1384 phi->init_req(3, result);
1385 region->init_req(3, if_lt);
1386 }
1387 if (!stopped()) {
1388 result = _gvn.transform(new AddINode(result, from_index));
1389 phi->init_req(4, result);
1390 region->init_req(4, control());
1391 }
1392 }
1393
1394 set_control(_gvn.transform(region));
1395 record_for_igvn(region);
1396 set_result(_gvn.transform(phi));
1397 clear_upper_avx();
1398
1399 return true;
1400 }
1401
1402 // Create StrIndexOfNode with fast path checks
1403 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1404 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1405 // Check for substr count > string count
1406 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1407 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1408 Node* if_gt = generate_slow_guard(bol, NULL);
1409 if (if_gt != NULL) {
1410 phi->init_req(1, intcon(-1));
1411 region->init_req(1, if_gt);
1412 }
1413 if (!stopped()) {
1414 // Check for substr count == 0
1415 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1416 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1417 Node* if_zero = generate_slow_guard(bol, NULL);
1418 if (if_zero != NULL) {
1419 phi->init_req(2, intcon(0));
1420 region->init_req(2, if_zero);
1421 }
1422 }
1423 if (!stopped()) {
1424 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1425 }
1426 return NULL;
1427 }
1428
1429 //-----------------------------inline_string_indexOfChar-----------------------
1430 bool LibraryCallKit::inline_string_indexOfChar() {
1431 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1432 return false;
1433 }
1434 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1435 return false;
1436 }
1437 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1438 Node* src = argument(0); // byte[]
1439 Node* tgt = argument(1); // tgt is int ch
1440 Node* from_index = argument(2);
1441 Node* max = argument(3);
1442
1443 src = must_be_not_null(src, true);
1444
1445 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1446 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1447 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1448
1449 // Range checks
1450 generate_string_range_check(src, src_offset, src_count, true);
1451 if (stopped()) {
1452 return true;
1453 }
1454
1455 RegionNode* region = new RegionNode(3);
1456 Node* phi = new PhiNode(region, TypeInt::INT);
1457
1458 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1459 C->set_has_split_ifs(true); // Has chance for split-if optimization
1460 _gvn.transform(result);
1461
1462 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1463 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1464
1465 Node* if_lt = generate_slow_guard(bol, NULL);
1466 if (if_lt != NULL) {
1467 // result == -1
1468 phi->init_req(2, result);
1469 region->init_req(2, if_lt);
1470 }
1471 if (!stopped()) {
1472 result = _gvn.transform(new AddINode(result, from_index));
1473 phi->init_req(1, result);
1474 region->init_req(1, control());
1475 }
1476 set_control(_gvn.transform(region));
1477 record_for_igvn(region);
1478 set_result(_gvn.transform(phi));
1479
1480 return true;
1481 }
1482 //---------------------------inline_string_copy---------------------
1483 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1484 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1485 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1486 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1487 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1488 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1489 bool LibraryCallKit::inline_string_copy(bool compress) {
1490 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1491 return false;
1492 }
1493 int nargs = 5; // 2 oops, 3 ints
1494 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1495
1496 Node* src = argument(0);
1497 Node* src_offset = argument(1);
1498 Node* dst = argument(2);
1499 Node* dst_offset = argument(3);
1500 Node* length = argument(4);
1501
1502 // Check for allocation before we add nodes that would confuse
1503 // tightly_coupled_allocation()
1504 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1505
1506 // Figure out the size and type of the elements we will be copying.
1507 const Type* src_type = src->Value(&_gvn);
1508 const Type* dst_type = dst->Value(&_gvn);
1509 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1510 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1511 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1512 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1513 "Unsupported array types for inline_string_copy");
1514
1515 src = must_be_not_null(src, true);
1516 dst = must_be_not_null(dst, true);
1517
1518 // Convert char[] offsets to byte[] offsets
1519 bool convert_src = (compress && src_elem == T_BYTE);
1520 bool convert_dst = (!compress && dst_elem == T_BYTE);
1521 if (convert_src) {
1522 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1523 } else if (convert_dst) {
1524 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1525 }
1526
1527 // Range checks
1528 generate_string_range_check(src, src_offset, length, convert_src);
1529 generate_string_range_check(dst, dst_offset, length, convert_dst);
1530 if (stopped()) {
1531 return true;
1532 }
1533
1534 Node* src_start = array_element_address(src, src_offset, src_elem);
1535 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1536 // 'src_start' points to src array + scaled offset
1537 // 'dst_start' points to dst array + scaled offset
1538 Node* count = NULL;
1539 if (compress) {
1540 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1541 } else {
1542 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1543 }
1544
1545 if (alloc != NULL) {
1546 if (alloc->maybe_set_complete(&_gvn)) {
1547 // "You break it, you buy it."
1548 InitializeNode* init = alloc->initialization();
1549 assert(init->is_complete(), "we just did this");
1550 init->set_complete_with_arraycopy();
1551 assert(dst->is_CheckCastPP(), "sanity");
1552 assert(dst->in(0)->in(0) == init, "dest pinned");
1553 }
1554 // Do not let stores that initialize this object be reordered with
1555 // a subsequent store that would make this object accessible by
1556 // other threads.
1557 // Record what AllocateNode this StoreStore protects so that
1558 // escape analysis can go from the MemBarStoreStoreNode to the
1559 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1560 // based on the escape status of the AllocateNode.
1561 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1562 }
1563 if (compress) {
1564 set_result(_gvn.transform(count));
1565 }
1566 clear_upper_avx();
1567
1568 return true;
1569 }
1570
1571 #ifdef _LP64
1572 #define XTOP ,top() /*additional argument*/
1573 #else //_LP64
1574 #define XTOP /*no additional argument*/
1575 #endif //_LP64
1576
1577 //------------------------inline_string_toBytesU--------------------------
1578 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1579 bool LibraryCallKit::inline_string_toBytesU() {
1580 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1581 return false;
1582 }
1583 // Get the arguments.
1584 Node* value = argument(0);
1585 Node* offset = argument(1);
1586 Node* length = argument(2);
1587
1588 Node* newcopy = NULL;
1589
1590 // Set the original stack and the reexecute bit for the interpreter to reexecute
1591 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1592 { PreserveReexecuteState preexecs(this);
1593 jvms()->set_should_reexecute(true);
1594
1595 // Check if a null path was taken unconditionally.
1596 value = null_check(value);
1597
1598 RegionNode* bailout = new RegionNode(1);
1599 record_for_igvn(bailout);
1600
1601 // Range checks
1602 generate_negative_guard(offset, bailout);
1603 generate_negative_guard(length, bailout);
1604 generate_limit_guard(offset, length, load_array_length(value), bailout);
1605 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1606 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1607
1608 if (bailout->req() > 1) {
1609 PreserveJVMState pjvms(this);
1610 set_control(_gvn.transform(bailout));
1611 uncommon_trap(Deoptimization::Reason_intrinsic,
1612 Deoptimization::Action_maybe_recompile);
1613 }
1614 if (stopped()) {
1615 return true;
1616 }
1617
1618 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1619 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1620 newcopy = new_array(klass_node, size, 0); // no arguments to push
1621 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1622
1623 // Calculate starting addresses.
1624 Node* src_start = array_element_address(value, offset, T_CHAR);
1625 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1626
1627 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1628 const TypeInt* toffset = gvn().type(offset)->is_int();
1629 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1630
1631 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1632 const char* copyfunc_name = "arraycopy";
1633 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1634 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1635 OptoRuntime::fast_arraycopy_Type(),
1636 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1637 src_start, dst_start, ConvI2X(length) XTOP);
1638 // Do not let reads from the cloned object float above the arraycopy.
1639 if (alloc != NULL) {
1640 if (alloc->maybe_set_complete(&_gvn)) {
1641 // "You break it, you buy it."
1642 InitializeNode* init = alloc->initialization();
1643 assert(init->is_complete(), "we just did this");
1644 init->set_complete_with_arraycopy();
1645 assert(newcopy->is_CheckCastPP(), "sanity");
1646 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1647 }
1648 // Do not let stores that initialize this object be reordered with
1649 // a subsequent store that would make this object accessible by
1650 // other threads.
1651 // Record what AllocateNode this StoreStore protects so that
1652 // escape analysis can go from the MemBarStoreStoreNode to the
1653 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1654 // based on the escape status of the AllocateNode.
1655 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1656 } else {
1657 insert_mem_bar(Op_MemBarCPUOrder);
1658 }
1659 } // original reexecute is set back here
1660
1661 C->set_has_split_ifs(true); // Has chance for split-if optimization
1662 if (!stopped()) {
1663 set_result(newcopy);
1664 }
1665 clear_upper_avx();
1666
1667 return true;
1668 }
1669
1670 //------------------------inline_string_getCharsU--------------------------
1671 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1672 bool LibraryCallKit::inline_string_getCharsU() {
1673 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1674 return false;
1675 }
1676
1677 // Get the arguments.
1678 Node* src = argument(0);
1679 Node* src_begin = argument(1);
1680 Node* src_end = argument(2); // exclusive offset (i < src_end)
1681 Node* dst = argument(3);
1682 Node* dst_begin = argument(4);
1683
1684 // Check for allocation before we add nodes that would confuse
1685 // tightly_coupled_allocation()
1686 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1687
1688 // Check if a null path was taken unconditionally.
1689 src = null_check(src);
1690 dst = null_check(dst);
1691 if (stopped()) {
1692 return true;
1693 }
1694
1695 // Get length and convert char[] offset to byte[] offset
1696 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1697 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1698
1699 // Range checks
1700 generate_string_range_check(src, src_begin, length, true);
1701 generate_string_range_check(dst, dst_begin, length, false);
1702 if (stopped()) {
1703 return true;
1704 }
1705
1706 if (!stopped()) {
1707 // Calculate starting addresses.
1708 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1709 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1710
1711 // Check if array addresses are aligned to HeapWordSize
1712 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1713 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1714 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1715 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1716
1717 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1718 const char* copyfunc_name = "arraycopy";
1719 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1720 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1721 OptoRuntime::fast_arraycopy_Type(),
1722 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1723 src_start, dst_start, ConvI2X(length) XTOP);
1724 // Do not let reads from the cloned object float above the arraycopy.
1725 if (alloc != NULL) {
1726 if (alloc->maybe_set_complete(&_gvn)) {
1727 // "You break it, you buy it."
1728 InitializeNode* init = alloc->initialization();
1729 assert(init->is_complete(), "we just did this");
1730 init->set_complete_with_arraycopy();
1731 assert(dst->is_CheckCastPP(), "sanity");
1732 assert(dst->in(0)->in(0) == init, "dest pinned");
1733 }
1734 // Do not let stores that initialize this object be reordered with
1735 // a subsequent store that would make this object accessible by
1736 // other threads.
1737 // Record what AllocateNode this StoreStore protects so that
1738 // escape analysis can go from the MemBarStoreStoreNode to the
1739 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1740 // based on the escape status of the AllocateNode.
1741 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1742 } else {
1743 insert_mem_bar(Op_MemBarCPUOrder);
1744 }
1745 }
1746
1747 C->set_has_split_ifs(true); // Has chance for split-if optimization
1748 return true;
1749 }
1750
1751 //----------------------inline_string_char_access----------------------------
1752 // Store/Load char to/from byte[] array.
1753 // static void StringUTF16.putChar(byte[] val, int index, int c)
1754 // static char StringUTF16.getChar(byte[] val, int index)
1755 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1756 Node* value = argument(0);
1757 Node* index = argument(1);
1758 Node* ch = is_store ? argument(2) : NULL;
1759
1760 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1761 // correctly requires matched array shapes.
1762 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1763 "sanity: byte[] and char[] bases agree");
1764 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1765 "sanity: byte[] and char[] scales agree");
1766
1767 // Bail when getChar over constants is requested: constant folding would
1768 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1769 // Java method would constant fold nicely instead.
1770 if (!is_store && value->is_Con() && index->is_Con()) {
1771 return false;
1772 }
1773
1774 value = must_be_not_null(value, true);
1775
1776 Node* adr = array_element_address(value, index, T_CHAR);
1777 if (adr->is_top()) {
1778 return false;
1779 }
1780 if (is_store) {
1781 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1782 } else {
1783 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
1784 set_result(ch);
1785 }
1786 return true;
1787 }
1788
1789 //--------------------------round_double_node--------------------------------
1790 // Round a double node if necessary.
1791 Node* LibraryCallKit::round_double_node(Node* n) {
1792 if (Matcher::strict_fp_requires_explicit_rounding) {
1793 #ifdef IA32
1794 if (UseSSE < 2) {
1795 n = _gvn.transform(new RoundDoubleNode(NULL, n));
1796 }
1797 #else
1798 Unimplemented();
1799 #endif // IA32
1800 }
1801 return n;
1802 }
1803
1804 //------------------------------inline_math-----------------------------------
1805 // public static double Math.abs(double)
1806 // public static double Math.sqrt(double)
1807 // public static double Math.log(double)
1808 // public static double Math.log10(double)
1809 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1810 Node* arg = round_double_node(argument(0));
1811 Node* n = NULL;
1812 switch (id) {
1813 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1814 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1815 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1816 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1817 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1818 default: fatal_unexpected_iid(id); break;
1819 }
1820 set_result(_gvn.transform(n));
1821 return true;
1822 }
1823
1824 //------------------------------inline_math-----------------------------------
1825 // public static float Math.abs(float)
1826 // public static int Math.abs(int)
1827 // public static long Math.abs(long)
1828 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1829 Node* arg = argument(0);
1830 Node* n = NULL;
1831 switch (id) {
1832 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1833 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1834 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1835 default: fatal_unexpected_iid(id); break;
1836 }
1837 set_result(_gvn.transform(n));
1838 return true;
1839 }
1840
1841 //------------------------------runtime_math-----------------------------
1842 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1843 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1844 "must be (DD)D or (D)D type");
1845
1846 // Inputs
1847 Node* a = round_double_node(argument(0));
1848 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1849
1850 const TypePtr* no_memory_effects = NULL;
1851 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1852 no_memory_effects,
1853 a, top(), b, b ? top() : NULL);
1854 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1855 #ifdef ASSERT
1856 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1857 assert(value_top == top(), "second value must be top");
1858 #endif
1859
1860 set_result(value);
1861 return true;
1862 }
1863
1864 //------------------------------inline_math_native-----------------------------
1865 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1866 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1867 switch (id) {
1868 // These intrinsics are not properly supported on all hardware
1869 case vmIntrinsics::_dsin:
1870 return StubRoutines::dsin() != NULL ?
1871 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1872 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1873 case vmIntrinsics::_dcos:
1874 return StubRoutines::dcos() != NULL ?
1875 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1876 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1877 case vmIntrinsics::_dtan:
1878 return StubRoutines::dtan() != NULL ?
1879 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1880 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1881 case vmIntrinsics::_dlog:
1882 return StubRoutines::dlog() != NULL ?
1883 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1884 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1885 case vmIntrinsics::_dlog10:
1886 return StubRoutines::dlog10() != NULL ?
1887 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1888 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1889
1890 // These intrinsics are supported on all hardware
1891 case vmIntrinsics::_ceil:
1892 case vmIntrinsics::_floor:
1893 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1894 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1895 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1896 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1897 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1898 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1899
1900 case vmIntrinsics::_dexp:
1901 return StubRoutines::dexp() != NULL ?
1902 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1903 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1904 case vmIntrinsics::_dpow: {
1905 Node* exp = round_double_node(argument(2));
1906 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1907 if (d != NULL && d->getd() == 2.0) {
1908 // Special case: pow(x, 2.0) => x * x
1909 Node* base = round_double_node(argument(0));
1910 set_result(_gvn.transform(new MulDNode(base, base)));
1911 return true;
1912 }
1913 return StubRoutines::dpow() != NULL ?
1914 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1915 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1916 }
1917 #undef FN_PTR
1918
1919 // These intrinsics are not yet correctly implemented
1920 case vmIntrinsics::_datan2:
1921 return false;
1922
1923 default:
1924 fatal_unexpected_iid(id);
1925 return false;
1926 }
1927 }
1928
1929 static bool is_simple_name(Node* n) {
1930 return (n->req() == 1 // constant
1931 || (n->is_Type() && n->as_Type()->type()->singleton())
1932 || n->is_Proj() // parameter or return value
1933 || n->is_Phi() // local of some sort
1934 );
1935 }
1936
1937 //----------------------------inline_notify-----------------------------------*
1938 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1939 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1940 address func;
1941 if (id == vmIntrinsics::_notify) {
1942 func = OptoRuntime::monitor_notify_Java();
1943 } else {
1944 func = OptoRuntime::monitor_notifyAll_Java();
1945 }
1946 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1947 make_slow_call_ex(call, env()->Throwable_klass(), false);
1948 return true;
1949 }
1950
1951
1952 //----------------------------inline_min_max-----------------------------------
1953 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1954 set_result(generate_min_max(id, argument(0), argument(1)));
1955 return true;
1956 }
1957
1958 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1959 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1960 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1961 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1962 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1963
1964 {
1965 PreserveJVMState pjvms(this);
1966 PreserveReexecuteState preexecs(this);
1967 jvms()->set_should_reexecute(true);
1968
1969 set_control(slow_path);
1970 set_i_o(i_o());
1971
1972 uncommon_trap(Deoptimization::Reason_intrinsic,
1973 Deoptimization::Action_none);
1974 }
1975
1976 set_control(fast_path);
1977 set_result(math);
1978 }
1979
1980 template <typename OverflowOp>
1981 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1982 typedef typename OverflowOp::MathOp MathOp;
1983
1984 MathOp* mathOp = new MathOp(arg1, arg2);
1985 Node* operation = _gvn.transform( mathOp );
1986 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1987 inline_math_mathExact(operation, ofcheck);
1988 return true;
1989 }
1990
1991 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1992 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1993 }
1994
1995 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1996 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1997 }
1998
1999 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2000 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2001 }
2002
2003 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2004 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2005 }
2006
2007 bool LibraryCallKit::inline_math_negateExactI() {
2008 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2009 }
2010
2011 bool LibraryCallKit::inline_math_negateExactL() {
2012 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2013 }
2014
2015 bool LibraryCallKit::inline_math_multiplyExactI() {
2016 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2017 }
2018
2019 bool LibraryCallKit::inline_math_multiplyExactL() {
2020 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2021 }
2022
2023 bool LibraryCallKit::inline_math_multiplyHigh() {
2024 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2025 return true;
2026 }
2027
2028 Node*
2029 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2030 // These are the candidate return value:
2031 Node* xvalue = x0;
2032 Node* yvalue = y0;
2033
2034 if (xvalue == yvalue) {
2035 return xvalue;
2036 }
2037
2038 bool want_max = (id == vmIntrinsics::_max);
2039
2040 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
2041 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
2042 if (txvalue == NULL || tyvalue == NULL) return top();
2043 // This is not really necessary, but it is consistent with a
2044 // hypothetical MaxINode::Value method:
2045 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
2046
2047 // %%% This folding logic should (ideally) be in a different place.
2048 // Some should be inside IfNode, and there to be a more reliable
2049 // transformation of ?: style patterns into cmoves. We also want
2050 // more powerful optimizations around cmove and min/max.
2051
2052 // Try to find a dominating comparison of these guys.
2053 // It can simplify the index computation for Arrays.copyOf
2054 // and similar uses of System.arraycopy.
2055 // First, compute the normalized version of CmpI(x, y).
2056 int cmp_op = Op_CmpI;
2057 Node* xkey = xvalue;
2058 Node* ykey = yvalue;
2059 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
2060 if (ideal_cmpxy->is_Cmp()) {
2061 // E.g., if we have CmpI(length - offset, count),
2062 // it might idealize to CmpI(length, count + offset)
2063 cmp_op = ideal_cmpxy->Opcode();
2064 xkey = ideal_cmpxy->in(1);
2065 ykey = ideal_cmpxy->in(2);
2066 }
2067
2068 // Start by locating any relevant comparisons.
2069 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
2070 Node* cmpxy = NULL;
2071 Node* cmpyx = NULL;
2072 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
2073 Node* cmp = start_from->fast_out(k);
2074 if (cmp->outcnt() > 0 && // must have prior uses
2075 cmp->in(0) == NULL && // must be context-independent
2076 cmp->Opcode() == cmp_op) { // right kind of compare
2077 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
2078 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
2079 }
2080 }
2081
2082 const int NCMPS = 2;
2083 Node* cmps[NCMPS] = { cmpxy, cmpyx };
2084 int cmpn;
2085 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2086 if (cmps[cmpn] != NULL) break; // find a result
2087 }
2088 if (cmpn < NCMPS) {
2089 // Look for a dominating test that tells us the min and max.
2090 int depth = 0; // Limit search depth for speed
2091 Node* dom = control();
2092 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2093 if (++depth >= 100) break;
2094 Node* ifproj = dom;
2095 if (!ifproj->is_Proj()) continue;
2096 Node* iff = ifproj->in(0);
2097 if (!iff->is_If()) continue;
2098 Node* bol = iff->in(1);
2099 if (!bol->is_Bool()) continue;
2100 Node* cmp = bol->in(1);
2101 if (cmp == NULL) continue;
2102 for (cmpn = 0; cmpn < NCMPS; cmpn++)
2103 if (cmps[cmpn] == cmp) break;
2104 if (cmpn == NCMPS) continue;
2105 BoolTest::mask btest = bol->as_Bool()->_test._test;
2106 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
2107 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2108 // At this point, we know that 'x btest y' is true.
2109 switch (btest) {
2110 case BoolTest::eq:
2111 // They are proven equal, so we can collapse the min/max.
2112 // Either value is the answer. Choose the simpler.
2113 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2114 return yvalue;
2115 return xvalue;
2116 case BoolTest::lt: // x < y
2117 case BoolTest::le: // x <= y
2118 return (want_max ? yvalue : xvalue);
2119 case BoolTest::gt: // x > y
2120 case BoolTest::ge: // x >= y
2121 return (want_max ? xvalue : yvalue);
2122 default:
2123 break;
2124 }
2125 }
2126 }
2127
2128 // We failed to find a dominating test.
2129 // Let's pick a test that might GVN with prior tests.
2130 Node* best_bol = NULL;
2131 BoolTest::mask best_btest = BoolTest::illegal;
2132 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2133 Node* cmp = cmps[cmpn];
2134 if (cmp == NULL) continue;
2135 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2136 Node* bol = cmp->fast_out(j);
2137 if (!bol->is_Bool()) continue;
2138 BoolTest::mask btest = bol->as_Bool()->_test._test;
2139 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2140 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2141 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2142 best_bol = bol->as_Bool();
2143 best_btest = btest;
2144 }
2145 }
2146 }
2147
2148 Node* answer_if_true = NULL;
2149 Node* answer_if_false = NULL;
2150 switch (best_btest) {
2151 default:
2152 if (cmpxy == NULL)
2153 cmpxy = ideal_cmpxy;
2154 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2155 // and fall through:
2156 case BoolTest::lt: // x < y
2157 case BoolTest::le: // x <= y
2158 answer_if_true = (want_max ? yvalue : xvalue);
2159 answer_if_false = (want_max ? xvalue : yvalue);
2160 break;
2161 case BoolTest::gt: // x > y
2162 case BoolTest::ge: // x >= y
2163 answer_if_true = (want_max ? xvalue : yvalue);
2164 answer_if_false = (want_max ? yvalue : xvalue);
2165 break;
2166 }
2167
2168 jint hi, lo;
2169 if (want_max) {
2170 // We can sharpen the minimum.
2171 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2172 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2173 } else {
2174 // We can sharpen the maximum.
2175 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2176 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2177 }
2178
2179 // Use a flow-free graph structure, to avoid creating excess control edges
2180 // which could hinder other optimizations.
2181 // Since Math.min/max is often used with arraycopy, we want
2182 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2183 Node* cmov = CMoveNode::make(NULL, best_bol,
2184 answer_if_false, answer_if_true,
2185 TypeInt::make(lo, hi, widen));
2186
2187 return _gvn.transform(cmov);
2188
2189 /*
2190 // This is not as desirable as it may seem, since Min and Max
2191 // nodes do not have a full set of optimizations.
2192 // And they would interfere, anyway, with 'if' optimizations
2193 // and with CMoveI canonical forms.
2194 switch (id) {
2195 case vmIntrinsics::_min:
2196 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2197 case vmIntrinsics::_max:
2198 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2199 default:
2200 ShouldNotReachHere();
2201 }
2202 */
2203 }
2204
2205 inline int
2206 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2207 const TypePtr* base_type = TypePtr::NULL_PTR;
2208 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2209 if (base_type == NULL) {
2210 // Unknown type.
2211 return Type::AnyPtr;
2212 } else if (base_type == TypePtr::NULL_PTR) {
2213 // Since this is a NULL+long form, we have to switch to a rawptr.
2214 base = _gvn.transform(new CastX2PNode(offset));
2215 offset = MakeConX(0);
2216 return Type::RawPtr;
2217 } else if (base_type->base() == Type::RawPtr) {
2218 return Type::RawPtr;
2219 } else if (base_type->isa_oopptr()) {
2220 // Base is never null => always a heap address.
2221 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2222 return Type::OopPtr;
2223 }
2224 // Offset is small => always a heap address.
2225 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2226 if (offset_type != NULL &&
2227 base_type->offset() == 0 && // (should always be?)
2228 offset_type->_lo >= 0 &&
2229 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2230 return Type::OopPtr;
2231 } else if (type == T_OBJECT) {
2232 // off heap access to an oop doesn't make any sense. Has to be on
2233 // heap.
2234 return Type::OopPtr;
2235 }
2236 // Otherwise, it might either be oop+off or NULL+addr.
2237 return Type::AnyPtr;
2238 } else {
2239 // No information:
2240 return Type::AnyPtr;
2241 }
2242 }
2243
2244 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type, bool can_cast) {
2245 Node* uncasted_base = base;
2246 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2247 if (kind == Type::RawPtr) {
2248 return basic_plus_adr(top(), uncasted_base, offset);
2249 } else if (kind == Type::AnyPtr) {
2250 assert(base == uncasted_base, "unexpected base change");
2251 if (can_cast) {
2252 if (!_gvn.type(base)->speculative_maybe_null() &&
2253 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2254 // According to profiling, this access is always on
2255 // heap. Casting the base to not null and thus avoiding membars
2256 // around the access should allow better optimizations
2257 Node* null_ctl = top();
2258 base = null_check_oop(base, &null_ctl, true, true, true);
2259 assert(null_ctl->is_top(), "no null control here");
2260 return basic_plus_adr(base, offset);
2261 } else if (_gvn.type(base)->speculative_always_null() &&
2262 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2263 // According to profiling, this access is always off
2264 // heap.
2265 base = null_assert(base);
2266 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2267 offset = MakeConX(0);
2268 return basic_plus_adr(top(), raw_base, offset);
2269 }
2270 }
2271 // We don't know if it's an on heap or off heap access. Fall back
2272 // to raw memory access.
2273 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2274 return basic_plus_adr(top(), raw, offset);
2275 } else {
2276 assert(base == uncasted_base, "unexpected base change");
2277 // We know it's an on heap access so base can't be null
2278 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2279 base = must_be_not_null(base, true);
2280 }
2281 return basic_plus_adr(base, offset);
2282 }
2283 }
2284
2285 //--------------------------inline_number_methods-----------------------------
2286 // inline int Integer.numberOfLeadingZeros(int)
2287 // inline int Long.numberOfLeadingZeros(long)
2288 //
2289 // inline int Integer.numberOfTrailingZeros(int)
2290 // inline int Long.numberOfTrailingZeros(long)
2291 //
2292 // inline int Integer.bitCount(int)
2293 // inline int Long.bitCount(long)
2294 //
2295 // inline char Character.reverseBytes(char)
2296 // inline short Short.reverseBytes(short)
2297 // inline int Integer.reverseBytes(int)
2298 // inline long Long.reverseBytes(long)
2299 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2300 Node* arg = argument(0);
2301 Node* n = NULL;
2302 switch (id) {
2303 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2304 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2305 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2306 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2307 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2308 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2309 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2310 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2311 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2312 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2313 default: fatal_unexpected_iid(id); break;
2314 }
2315 set_result(_gvn.transform(n));
2316 return true;
2317 }
2318
2319 //----------------------------inline_unsafe_access----------------------------
2320
2321 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2322 // Attempt to infer a sharper value type from the offset and base type.
2323 ciKlass* sharpened_klass = NULL;
2324
2325 // See if it is an instance field, with an object type.
2326 if (alias_type->field() != NULL) {
2327 if (alias_type->field()->type()->is_klass()) {
2328 sharpened_klass = alias_type->field()->type()->as_klass();
2329 }
2330 }
2331
2332 // See if it is a narrow oop array.
2333 if (adr_type->isa_aryptr()) {
2334 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2335 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2336 if (elem_type != NULL) {
2337 sharpened_klass = elem_type->klass();
2338 }
2339 }
2340 }
2341
2342 // The sharpened class might be unloaded if there is no class loader
2343 // contraint in place.
2344 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2345 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2346
2347 #ifndef PRODUCT
2348 if (C->print_intrinsics() || C->print_inlining()) {
2349 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2350 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2351 }
2352 #endif
2353 // Sharpen the value type.
2354 return tjp;
2355 }
2356 return NULL;
2357 }
2358
2359 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2360 switch (kind) {
2361 case Relaxed:
2362 return MO_UNORDERED;
2363 case Opaque:
2364 return MO_RELAXED;
2365 case Acquire:
2366 return MO_ACQUIRE;
2367 case Release:
2368 return MO_RELEASE;
2369 case Volatile:
2370 return MO_SEQ_CST;
2371 default:
2372 ShouldNotReachHere();
2373 return 0;
2374 }
2375 }
2376
2377 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2378 if (callee()->is_static()) return false; // caller must have the capability!
2379 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2380 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2381 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2382 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2383
2384 if (is_reference_type(type)) {
2385 decorators |= ON_UNKNOWN_OOP_REF;
2386 }
2387
2388 if (unaligned) {
2389 decorators |= C2_UNALIGNED;
2390 }
2391
2392 #ifndef PRODUCT
2393 {
2394 ResourceMark rm;
2395 // Check the signatures.
2396 ciSignature* sig = callee()->signature();
2397 #ifdef ASSERT
2398 if (!is_store) {
2399 // Object getReference(Object base, int/long offset), etc.
2400 BasicType rtype = sig->return_type()->basic_type();
2401 assert(rtype == type, "getter must return the expected value");
2402 assert(sig->count() == 2, "oop getter has 2 arguments");
2403 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2404 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2405 } else {
2406 // void putReference(Object base, int/long offset, Object x), etc.
2407 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2408 assert(sig->count() == 3, "oop putter has 3 arguments");
2409 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2410 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2411 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2412 assert(vtype == type, "putter must accept the expected value");
2413 }
2414 #endif // ASSERT
2415 }
2416 #endif //PRODUCT
2417
2418 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2419
2420 Node* receiver = argument(0); // type: oop
2421
2422 // Build address expression.
2423 Node* adr;
2424 Node* heap_base_oop = top();
2425 Node* offset = top();
2426 Node* val;
2427
2428 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2429 Node* base = argument(1); // type: oop
2430 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2431 offset = argument(2); // type: long
2432 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2433 // to be plain byte offsets, which are also the same as those accepted
2434 // by oopDesc::field_addr.
2435 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2436 "fieldOffset must be byte-scaled");
2437 // 32-bit machines ignore the high half!
2438 offset = ConvL2X(offset);
2439 adr = make_unsafe_address(base, offset, is_store ? ACCESS_WRITE : ACCESS_READ, type, kind == Relaxed);
2440
2441 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2442 if (type != T_OBJECT) {
2443 decorators |= IN_NATIVE; // off-heap primitive access
2444 } else {
2445 return false; // off-heap oop accesses are not supported
2446 }
2447 } else {
2448 heap_base_oop = base; // on-heap or mixed access
2449 }
2450
2451 // Can base be NULL? Otherwise, always on-heap access.
2452 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2453
2454 if (!can_access_non_heap) {
2455 decorators |= IN_HEAP;
2456 }
2457
2458 val = is_store ? argument(4) : NULL;
2459
2460 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2461 if (adr_type == TypePtr::NULL_PTR) {
2462 return false; // off-heap access with zero address
2463 }
2464
2465 // Try to categorize the address.
2466 Compile::AliasType* alias_type = C->alias_type(adr_type);
2467 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2468
2469 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2470 alias_type->adr_type() == TypeAryPtr::RANGE) {
2471 return false; // not supported
2472 }
2473
2474 bool mismatched = false;
2475 BasicType bt = alias_type->basic_type();
2476 if (bt != T_ILLEGAL) {
2477 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2478 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2479 // Alias type doesn't differentiate between byte[] and boolean[]).
2480 // Use address type to get the element type.
2481 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2482 }
2483 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2484 // accessing an array field with getReference is not a mismatch
2485 bt = T_OBJECT;
2486 }
2487 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2488 // Don't intrinsify mismatched object accesses
2489 return false;
2490 }
2491 mismatched = (bt != type);
2492 } else if (alias_type->adr_type()->isa_oopptr()) {
2493 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2494 }
2495
2496 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2497
2498 if (mismatched) {
2499 decorators |= C2_MISMATCHED;
2500 }
2501
2502 // First guess at the value type.
2503 const Type *value_type = Type::get_const_basic_type(type);
2504
2505 // Figure out the memory ordering.
2506 decorators |= mo_decorator_for_access_kind(kind);
2507
2508 if (!is_store && type == T_OBJECT) {
2509 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2510 if (tjp != NULL) {
2511 value_type = tjp;
2512 }
2513 }
2514
2515 receiver = null_check(receiver);
2516 if (stopped()) {
2517 return true;
2518 }
2519 // Heap pointers get a null-check from the interpreter,
2520 // as a courtesy. However, this is not guaranteed by Unsafe,
2521 // and it is not possible to fully distinguish unintended nulls
2522 // from intended ones in this API.
2523
2524 if (!is_store) {
2525 Node* p = NULL;
2526 // Try to constant fold a load from a constant field
2527 ciField* field = alias_type->field();
2528 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2529 // final or stable field
2530 p = make_constant_from_field(field, heap_base_oop);
2531 }
2532
2533 if (p == NULL) { // Could not constant fold the load
2534 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2535 // Normalize the value returned by getBoolean in the following cases
2536 if (type == T_BOOLEAN &&
2537 (mismatched ||
2538 heap_base_oop == top() || // - heap_base_oop is NULL or
2539 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2540 // and the unsafe access is made to large offset
2541 // (i.e., larger than the maximum offset necessary for any
2542 // field access)
2543 ) {
2544 IdealKit ideal = IdealKit(this);
2545 #define __ ideal.
2546 IdealVariable normalized_result(ideal);
2547 __ declarations_done();
2548 __ set(normalized_result, p);
2549 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2550 __ set(normalized_result, ideal.ConI(1));
2551 ideal.end_if();
2552 final_sync(ideal);
2553 p = __ value(normalized_result);
2554 #undef __
2555 }
2556 }
2557 if (type == T_ADDRESS) {
2558 p = gvn().transform(new CastP2XNode(NULL, p));
2559 p = ConvX2UL(p);
2560 }
2561 // The load node has the control of the preceding MemBarCPUOrder. All
2562 // following nodes will have the control of the MemBarCPUOrder inserted at
2563 // the end of this method. So, pushing the load onto the stack at a later
2564 // point is fine.
2565 set_result(p);
2566 } else {
2567 if (bt == T_ADDRESS) {
2568 // Repackage the long as a pointer.
2569 val = ConvL2X(val);
2570 val = gvn().transform(new CastX2PNode(val));
2571 }
2572 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2573 }
2574
2575 return true;
2576 }
2577
2578 //----------------------------inline_unsafe_load_store----------------------------
2579 // This method serves a couple of different customers (depending on LoadStoreKind):
2580 //
2581 // LS_cmp_swap:
2582 //
2583 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2584 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2585 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2586 //
2587 // LS_cmp_swap_weak:
2588 //
2589 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2590 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2591 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2592 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2593 //
2594 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2595 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2596 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2597 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2598 //
2599 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2600 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2601 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2602 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2603 //
2604 // LS_cmp_exchange:
2605 //
2606 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2607 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2608 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2609 //
2610 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2611 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2612 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2613 //
2614 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2615 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2616 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2617 //
2618 // LS_get_add:
2619 //
2620 // int getAndAddInt( Object o, long offset, int delta)
2621 // long getAndAddLong(Object o, long offset, long delta)
2622 //
2623 // LS_get_set:
2624 //
2625 // int getAndSet(Object o, long offset, int newValue)
2626 // long getAndSet(Object o, long offset, long newValue)
2627 // Object getAndSet(Object o, long offset, Object newValue)
2628 //
2629 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2630 // This basic scheme here is the same as inline_unsafe_access, but
2631 // differs in enough details that combining them would make the code
2632 // overly confusing. (This is a true fact! I originally combined
2633 // them, but even I was confused by it!) As much code/comments as
2634 // possible are retained from inline_unsafe_access though to make
2635 // the correspondences clearer. - dl
2636
2637 if (callee()->is_static()) return false; // caller must have the capability!
2638
2639 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2640 decorators |= mo_decorator_for_access_kind(access_kind);
2641
2642 #ifndef PRODUCT
2643 BasicType rtype;
2644 {
2645 ResourceMark rm;
2646 // Check the signatures.
2647 ciSignature* sig = callee()->signature();
2648 rtype = sig->return_type()->basic_type();
2649 switch(kind) {
2650 case LS_get_add:
2651 case LS_get_set: {
2652 // Check the signatures.
2653 #ifdef ASSERT
2654 assert(rtype == type, "get and set must return the expected type");
2655 assert(sig->count() == 3, "get and set has 3 arguments");
2656 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2657 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2658 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2659 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2660 #endif // ASSERT
2661 break;
2662 }
2663 case LS_cmp_swap:
2664 case LS_cmp_swap_weak: {
2665 // Check the signatures.
2666 #ifdef ASSERT
2667 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2668 assert(sig->count() == 4, "CAS has 4 arguments");
2669 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2670 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2671 #endif // ASSERT
2672 break;
2673 }
2674 case LS_cmp_exchange: {
2675 // Check the signatures.
2676 #ifdef ASSERT
2677 assert(rtype == type, "CAS must return the expected type");
2678 assert(sig->count() == 4, "CAS has 4 arguments");
2679 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2680 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2681 #endif // ASSERT
2682 break;
2683 }
2684 default:
2685 ShouldNotReachHere();
2686 }
2687 }
2688 #endif //PRODUCT
2689
2690 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2691
2692 // Get arguments:
2693 Node* receiver = NULL;
2694 Node* base = NULL;
2695 Node* offset = NULL;
2696 Node* oldval = NULL;
2697 Node* newval = NULL;
2698 switch(kind) {
2699 case LS_cmp_swap:
2700 case LS_cmp_swap_weak:
2701 case LS_cmp_exchange: {
2702 const bool two_slot_type = type2size[type] == 2;
2703 receiver = argument(0); // type: oop
2704 base = argument(1); // type: oop
2705 offset = argument(2); // type: long
2706 oldval = argument(4); // type: oop, int, or long
2707 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2708 break;
2709 }
2710 case LS_get_add:
2711 case LS_get_set: {
2712 receiver = argument(0); // type: oop
2713 base = argument(1); // type: oop
2714 offset = argument(2); // type: long
2715 oldval = NULL;
2716 newval = argument(4); // type: oop, int, or long
2717 break;
2718 }
2719 default:
2720 ShouldNotReachHere();
2721 }
2722
2723 // Build field offset expression.
2724 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2725 // to be plain byte offsets, which are also the same as those accepted
2726 // by oopDesc::field_addr.
2727 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2728 // 32-bit machines ignore the high half of long offsets
2729 offset = ConvL2X(offset);
2730 Node* adr = make_unsafe_address(base, offset, ACCESS_WRITE | ACCESS_READ, type, false);
2731 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2732
2733 Compile::AliasType* alias_type = C->alias_type(adr_type);
2734 BasicType bt = alias_type->basic_type();
2735 if (bt != T_ILLEGAL &&
2736 (is_reference_type(bt) != (type == T_OBJECT))) {
2737 // Don't intrinsify mismatched object accesses.
2738 return false;
2739 }
2740
2741 // For CAS, unlike inline_unsafe_access, there seems no point in
2742 // trying to refine types. Just use the coarse types here.
2743 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2744 const Type *value_type = Type::get_const_basic_type(type);
2745
2746 switch (kind) {
2747 case LS_get_set:
2748 case LS_cmp_exchange: {
2749 if (type == T_OBJECT) {
2750 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2751 if (tjp != NULL) {
2752 value_type = tjp;
2753 }
2754 }
2755 break;
2756 }
2757 case LS_cmp_swap:
2758 case LS_cmp_swap_weak:
2759 case LS_get_add:
2760 break;
2761 default:
2762 ShouldNotReachHere();
2763 }
2764
2765 // Null check receiver.
2766 receiver = null_check(receiver);
2767 if (stopped()) {
2768 return true;
2769 }
2770
2771 int alias_idx = C->get_alias_index(adr_type);
2772
2773 if (is_reference_type(type)) {
2774 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2775
2776 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2777 // could be delayed during Parse (for example, in adjust_map_after_if()).
2778 // Execute transformation here to avoid barrier generation in such case.
2779 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2780 newval = _gvn.makecon(TypePtr::NULL_PTR);
2781
2782 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2783 // Refine the value to a null constant, when it is known to be null
2784 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2785 }
2786 }
2787
2788 Node* result = NULL;
2789 switch (kind) {
2790 case LS_cmp_exchange: {
2791 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2792 oldval, newval, value_type, type, decorators);
2793 break;
2794 }
2795 case LS_cmp_swap_weak:
2796 decorators |= C2_WEAK_CMPXCHG;
2797 case LS_cmp_swap: {
2798 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2799 oldval, newval, value_type, type, decorators);
2800 break;
2801 }
2802 case LS_get_set: {
2803 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2804 newval, value_type, type, decorators);
2805 break;
2806 }
2807 case LS_get_add: {
2808 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2809 newval, value_type, type, decorators);
2810 break;
2811 }
2812 default:
2813 ShouldNotReachHere();
2814 }
2815
2816 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2817 set_result(result);
2818 return true;
2819 }
2820
2821 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2822 // Regardless of form, don't allow previous ld/st to move down,
2823 // then issue acquire, release, or volatile mem_bar.
2824 insert_mem_bar(Op_MemBarCPUOrder);
2825 switch(id) {
2826 case vmIntrinsics::_loadFence:
2827 insert_mem_bar(Op_LoadFence);
2828 return true;
2829 case vmIntrinsics::_storeFence:
2830 insert_mem_bar(Op_StoreFence);
2831 return true;
2832 case vmIntrinsics::_fullFence:
2833 insert_mem_bar(Op_MemBarVolatile);
2834 return true;
2835 default:
2836 fatal_unexpected_iid(id);
2837 return false;
2838 }
2839 }
2840
2841 bool LibraryCallKit::inline_onspinwait() {
2842 insert_mem_bar(Op_OnSpinWait);
2843 return true;
2844 }
2845
2846 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2847 if (!kls->is_Con()) {
2848 return true;
2849 }
2850 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
2851 if (klsptr == NULL) {
2852 return true;
2853 }
2854 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
2855 // don't need a guard for a klass that is already initialized
2856 return !ik->is_initialized();
2857 }
2858
2859 //----------------------------inline_unsafe_writeback0-------------------------
2860 // public native void Unsafe.writeback0(long address)
2861 bool LibraryCallKit::inline_unsafe_writeback0() {
2862 if (!Matcher::has_match_rule(Op_CacheWB)) {
2863 return false;
2864 }
2865 #ifndef PRODUCT
2866 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2867 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2868 ciSignature* sig = callee()->signature();
2869 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2870 #endif
2871 null_check_receiver(); // null-check, then ignore
2872 Node *addr = argument(1);
2873 addr = new CastX2PNode(addr);
2874 addr = _gvn.transform(addr);
2875 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2876 flush = _gvn.transform(flush);
2877 set_memory(flush, TypeRawPtr::BOTTOM);
2878 return true;
2879 }
2880
2881 //----------------------------inline_unsafe_writeback0-------------------------
2882 // public native void Unsafe.writeback0(long address)
2883 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2884 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2885 return false;
2886 }
2887 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2888 return false;
2889 }
2890 #ifndef PRODUCT
2891 assert(Matcher::has_match_rule(Op_CacheWB),
2892 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2893 : "found match rule for CacheWBPostSync but not CacheWB"));
2894
2895 #endif
2896 null_check_receiver(); // null-check, then ignore
2897 Node *sync;
2898 if (is_pre) {
2899 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2900 } else {
2901 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2902 }
2903 sync = _gvn.transform(sync);
2904 set_memory(sync, TypeRawPtr::BOTTOM);
2905 return true;
2906 }
2907
2908 //----------------------------inline_unsafe_allocate---------------------------
2909 // public native Object Unsafe.allocateInstance(Class<?> cls);
2910 bool LibraryCallKit::inline_unsafe_allocate() {
2911 if (callee()->is_static()) return false; // caller must have the capability!
2912
2913 null_check_receiver(); // null-check, then ignore
2914 Node* cls = null_check(argument(1));
2915 if (stopped()) return true;
2916
2917 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2918 kls = null_check(kls);
2919 if (stopped()) return true; // argument was like int.class
2920
2921 Node* test = NULL;
2922 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2923 // Note: The argument might still be an illegal value like
2924 // Serializable.class or Object[].class. The runtime will handle it.
2925 // But we must make an explicit check for initialization.
2926 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2927 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2928 // can generate code to load it as unsigned byte.
2929 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2930 Node* bits = intcon(InstanceKlass::fully_initialized);
2931 test = _gvn.transform(new SubINode(inst, bits));
2932 // The 'test' is non-zero if we need to take a slow path.
2933 }
2934
2935 Node* obj = new_instance(kls, test);
2936 set_result(obj);
2937 return true;
2938 }
2939
2940 //------------------------inline_native_time_funcs--------------
2941 // inline code for System.currentTimeMillis() and System.nanoTime()
2942 // these have the same type and signature
2943 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2944 const TypeFunc* tf = OptoRuntime::void_long_Type();
2945 const TypePtr* no_memory_effects = NULL;
2946 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2947 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2948 #ifdef ASSERT
2949 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2950 assert(value_top == top(), "second value must be top");
2951 #endif
2952 set_result(value);
2953 return true;
2954 }
2955
2956 #ifdef JFR_HAVE_INTRINSICS
2957
2958 /*
2959 * oop -> myklass
2960 * myklass->trace_id |= USED
2961 * return myklass->trace_id & ~0x3
2962 */
2963 bool LibraryCallKit::inline_native_classID() {
2964 Node* cls = null_check(argument(0), T_OBJECT);
2965 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2966 kls = null_check(kls, T_OBJECT);
2967
2968 ByteSize offset = KLASS_TRACE_ID_OFFSET;
2969 Node* insp = basic_plus_adr(kls, in_bytes(offset));
2970 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
2971
2972 Node* clsused = longcon(0x01l); // set the class bit
2973 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
2974 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
2975 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
2976
2977 #ifdef TRACE_ID_META_BITS
2978 Node* mbits = longcon(~TRACE_ID_META_BITS);
2979 tvalue = _gvn.transform(new AndLNode(tvalue, mbits));
2980 #endif
2981 #ifdef TRACE_ID_SHIFT
2982 Node* cbits = intcon(TRACE_ID_SHIFT);
2983 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits));
2984 #endif
2985
2986 set_result(tvalue);
2987 return true;
2988
2989 }
2990
2991 bool LibraryCallKit::inline_native_getEventWriter() {
2992 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
2993
2994 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
2995 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
2996
2997 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
2998
2999 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
3000 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
3001
3002 IfNode* iff_jobj_null =
3003 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
3004
3005 enum { _normal_path = 1,
3006 _null_path = 2,
3007 PATH_LIMIT };
3008
3009 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3010 PhiNode* result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3011
3012 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
3013 result_rgn->init_req(_null_path, jobj_is_null);
3014 result_val->init_req(_null_path, null());
3015
3016 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
3017 set_control(jobj_is_not_null);
3018 Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT,
3019 IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3020 result_rgn->init_req(_normal_path, control());
3021 result_val->init_req(_normal_path, res);
3022
3023 set_result(result_rgn, result_val);
3024
3025 return true;
3026 }
3027
3028 #endif // JFR_HAVE_INTRINSICS
3029
3030 //------------------------inline_native_currentThread------------------
3031 bool LibraryCallKit::inline_native_currentThread() {
3032 Node* junk = NULL;
3033 set_result(generate_current_thread(junk));
3034 return true;
3035 }
3036
3037 //---------------------------load_mirror_from_klass----------------------------
3038 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3039 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3040 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3041 Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3042 // mirror = ((OopHandle)mirror)->resolve();
3043 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3044 }
3045
3046 //-----------------------load_klass_from_mirror_common-------------------------
3047 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3048 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3049 // and branch to the given path on the region.
3050 // If never_see_null, take an uncommon trap on null, so we can optimistically
3051 // compile for the non-null case.
3052 // If the region is NULL, force never_see_null = true.
3053 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3054 bool never_see_null,
3055 RegionNode* region,
3056 int null_path,
3057 int offset) {
3058 if (region == NULL) never_see_null = true;
3059 Node* p = basic_plus_adr(mirror, offset);
3060 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3061 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3062 Node* null_ctl = top();
3063 kls = null_check_oop(kls, &null_ctl, never_see_null);
3064 if (region != NULL) {
3065 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3066 region->init_req(null_path, null_ctl);
3067 } else {
3068 assert(null_ctl == top(), "no loose ends");
3069 }
3070 return kls;
3071 }
3072
3073 //--------------------(inline_native_Class_query helpers)---------------------
3074 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3075 // Fall through if (mods & mask) == bits, take the guard otherwise.
3076 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3077 // Branch around if the given klass has the given modifier bit set.
3078 // Like generate_guard, adds a new path onto the region.
3079 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3080 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3081 Node* mask = intcon(modifier_mask);
3082 Node* bits = intcon(modifier_bits);
3083 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3084 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3085 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3086 return generate_fair_guard(bol, region);
3087 }
3088 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3089 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3090 }
3091 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3092 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
3093 }
3094
3095 //-------------------------inline_native_Class_query-------------------
3096 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3097 const Type* return_type = TypeInt::BOOL;
3098 Node* prim_return_value = top(); // what happens if it's a primitive class?
3099 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3100 bool expect_prim = false; // most of these guys expect to work on refs
3101
3102 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3103
3104 Node* mirror = argument(0);
3105 Node* obj = top();
3106
3107 switch (id) {
3108 case vmIntrinsics::_isInstance:
3109 // nothing is an instance of a primitive type
3110 prim_return_value = intcon(0);
3111 obj = argument(1);
3112 break;
3113 case vmIntrinsics::_getModifiers:
3114 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3115 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3116 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3117 break;
3118 case vmIntrinsics::_isInterface:
3119 prim_return_value = intcon(0);
3120 break;
3121 case vmIntrinsics::_isArray:
3122 prim_return_value = intcon(0);
3123 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3124 break;
3125 case vmIntrinsics::_isPrimitive:
3126 prim_return_value = intcon(1);
3127 expect_prim = true; // obviously
3128 break;
3129 case vmIntrinsics::_isHidden:
3130 prim_return_value = intcon(0);
3131 break;
3132 case vmIntrinsics::_getSuperclass:
3133 prim_return_value = null();
3134 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3135 break;
3136 case vmIntrinsics::_getClassAccessFlags:
3137 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3138 return_type = TypeInt::INT; // not bool! 6297094
3139 break;
3140 default:
3141 fatal_unexpected_iid(id);
3142 break;
3143 }
3144
3145 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3146 if (mirror_con == NULL) return false; // cannot happen?
3147
3148 #ifndef PRODUCT
3149 if (C->print_intrinsics() || C->print_inlining()) {
3150 ciType* k = mirror_con->java_mirror_type();
3151 if (k) {
3152 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3153 k->print_name();
3154 tty->cr();
3155 }
3156 }
3157 #endif
3158
3159 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3160 RegionNode* region = new RegionNode(PATH_LIMIT);
3161 record_for_igvn(region);
3162 PhiNode* phi = new PhiNode(region, return_type);
3163
3164 // The mirror will never be null of Reflection.getClassAccessFlags, however
3165 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3166 // if it is. See bug 4774291.
3167
3168 // For Reflection.getClassAccessFlags(), the null check occurs in
3169 // the wrong place; see inline_unsafe_access(), above, for a similar
3170 // situation.
3171 mirror = null_check(mirror);
3172 // If mirror or obj is dead, only null-path is taken.
3173 if (stopped()) return true;
3174
3175 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3176
3177 // Now load the mirror's klass metaobject, and null-check it.
3178 // Side-effects region with the control path if the klass is null.
3179 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3180 // If kls is null, we have a primitive mirror.
3181 phi->init_req(_prim_path, prim_return_value);
3182 if (stopped()) { set_result(region, phi); return true; }
3183 bool safe_for_replace = (region->in(_prim_path) == top());
3184
3185 Node* p; // handy temp
3186 Node* null_ctl;
3187
3188 // Now that we have the non-null klass, we can perform the real query.
3189 // For constant classes, the query will constant-fold in LoadNode::Value.
3190 Node* query_value = top();
3191 switch (id) {
3192 case vmIntrinsics::_isInstance:
3193 // nothing is an instance of a primitive type
3194 query_value = gen_instanceof(obj, kls, safe_for_replace);
3195 break;
3196
3197 case vmIntrinsics::_getModifiers:
3198 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3199 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3200 break;
3201
3202 case vmIntrinsics::_isInterface:
3203 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3204 if (generate_interface_guard(kls, region) != NULL)
3205 // A guard was added. If the guard is taken, it was an interface.
3206 phi->add_req(intcon(1));
3207 // If we fall through, it's a plain class.
3208 query_value = intcon(0);
3209 break;
3210
3211 case vmIntrinsics::_isArray:
3212 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3213 if (generate_array_guard(kls, region) != NULL)
3214 // A guard was added. If the guard is taken, it was an array.
3215 phi->add_req(intcon(1));
3216 // If we fall through, it's a plain class.
3217 query_value = intcon(0);
3218 break;
3219
3220 case vmIntrinsics::_isPrimitive:
3221 query_value = intcon(0); // "normal" path produces false
3222 break;
3223
3224 case vmIntrinsics::_isHidden:
3225 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
3226 if (generate_hidden_class_guard(kls, region) != NULL)
3227 // A guard was added. If the guard is taken, it was an hidden class.
3228 phi->add_req(intcon(1));
3229 // If we fall through, it's a plain class.
3230 query_value = intcon(0);
3231 break;
3232
3233
3234 case vmIntrinsics::_getSuperclass:
3235 // The rules here are somewhat unfortunate, but we can still do better
3236 // with random logic than with a JNI call.
3237 // Interfaces store null or Object as _super, but must report null.
3238 // Arrays store an intermediate super as _super, but must report Object.
3239 // Other types can report the actual _super.
3240 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3241 if (generate_interface_guard(kls, region) != NULL)
3242 // A guard was added. If the guard is taken, it was an interface.
3243 phi->add_req(null());
3244 if (generate_array_guard(kls, region) != NULL)
3245 // A guard was added. If the guard is taken, it was an array.
3246 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3247 // If we fall through, it's a plain class. Get its _super.
3248 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3249 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3250 null_ctl = top();
3251 kls = null_check_oop(kls, &null_ctl);
3252 if (null_ctl != top()) {
3253 // If the guard is taken, Object.superClass is null (both klass and mirror).
3254 region->add_req(null_ctl);
3255 phi ->add_req(null());
3256 }
3257 if (!stopped()) {
3258 query_value = load_mirror_from_klass(kls);
3259 }
3260 break;
3261
3262 case vmIntrinsics::_getClassAccessFlags:
3263 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3264 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3265 break;
3266
3267 default:
3268 fatal_unexpected_iid(id);
3269 break;
3270 }
3271
3272 // Fall-through is the normal case of a query to a real class.
3273 phi->init_req(1, query_value);
3274 region->init_req(1, control());
3275
3276 C->set_has_split_ifs(true); // Has chance for split-if optimization
3277 set_result(region, phi);
3278 return true;
3279 }
3280
3281 //-------------------------inline_Class_cast-------------------
3282 bool LibraryCallKit::inline_Class_cast() {
3283 Node* mirror = argument(0); // Class
3284 Node* obj = argument(1);
3285 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3286 if (mirror_con == NULL) {
3287 return false; // dead path (mirror->is_top()).
3288 }
3289 if (obj == NULL || obj->is_top()) {
3290 return false; // dead path
3291 }
3292 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3293
3294 // First, see if Class.cast() can be folded statically.
3295 // java_mirror_type() returns non-null for compile-time Class constants.
3296 ciType* tm = mirror_con->java_mirror_type();
3297 if (tm != NULL && tm->is_klass() &&
3298 tp != NULL && tp->klass() != NULL) {
3299 if (!tp->klass()->is_loaded()) {
3300 // Don't use intrinsic when class is not loaded.
3301 return false;
3302 } else {
3303 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3304 if (static_res == Compile::SSC_always_true) {
3305 // isInstance() is true - fold the code.
3306 set_result(obj);
3307 return true;
3308 } else if (static_res == Compile::SSC_always_false) {
3309 // Don't use intrinsic, have to throw ClassCastException.
3310 // If the reference is null, the non-intrinsic bytecode will
3311 // be optimized appropriately.
3312 return false;
3313 }
3314 }
3315 }
3316
3317 // Bailout intrinsic and do normal inlining if exception path is frequent.
3318 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3319 return false;
3320 }
3321
3322 // Generate dynamic checks.
3323 // Class.cast() is java implementation of _checkcast bytecode.
3324 // Do checkcast (Parse::do_checkcast()) optimizations here.
3325
3326 mirror = null_check(mirror);
3327 // If mirror is dead, only null-path is taken.
3328 if (stopped()) {
3329 return true;
3330 }
3331
3332 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3333 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3334 RegionNode* region = new RegionNode(PATH_LIMIT);
3335 record_for_igvn(region);
3336
3337 // Now load the mirror's klass metaobject, and null-check it.
3338 // If kls is null, we have a primitive mirror and
3339 // nothing is an instance of a primitive type.
3340 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3341
3342 Node* res = top();
3343 if (!stopped()) {
3344 Node* bad_type_ctrl = top();
3345 // Do checkcast optimizations.
3346 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3347 region->init_req(_bad_type_path, bad_type_ctrl);
3348 }
3349 if (region->in(_prim_path) != top() ||
3350 region->in(_bad_type_path) != top()) {
3351 // Let Interpreter throw ClassCastException.
3352 PreserveJVMState pjvms(this);
3353 set_control(_gvn.transform(region));
3354 uncommon_trap(Deoptimization::Reason_intrinsic,
3355 Deoptimization::Action_maybe_recompile);
3356 }
3357 if (!stopped()) {
3358 set_result(res);
3359 }
3360 return true;
3361 }
3362
3363
3364 //--------------------------inline_native_subtype_check------------------------
3365 // This intrinsic takes the JNI calls out of the heart of
3366 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3367 bool LibraryCallKit::inline_native_subtype_check() {
3368 // Pull both arguments off the stack.
3369 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3370 args[0] = argument(0);
3371 args[1] = argument(1);
3372 Node* klasses[2]; // corresponding Klasses: superk, subk
3373 klasses[0] = klasses[1] = top();
3374
3375 enum {
3376 // A full decision tree on {superc is prim, subc is prim}:
3377 _prim_0_path = 1, // {P,N} => false
3378 // {P,P} & superc!=subc => false
3379 _prim_same_path, // {P,P} & superc==subc => true
3380 _prim_1_path, // {N,P} => false
3381 _ref_subtype_path, // {N,N} & subtype check wins => true
3382 _both_ref_path, // {N,N} & subtype check loses => false
3383 PATH_LIMIT
3384 };
3385
3386 RegionNode* region = new RegionNode(PATH_LIMIT);
3387 Node* phi = new PhiNode(region, TypeInt::BOOL);
3388 record_for_igvn(region);
3389
3390 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3391 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3392 int class_klass_offset = java_lang_Class::klass_offset();
3393
3394 // First null-check both mirrors and load each mirror's klass metaobject.
3395 int which_arg;
3396 for (which_arg = 0; which_arg <= 1; which_arg++) {
3397 Node* arg = args[which_arg];
3398 arg = null_check(arg);
3399 if (stopped()) break;
3400 args[which_arg] = arg;
3401
3402 Node* p = basic_plus_adr(arg, class_klass_offset);
3403 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3404 klasses[which_arg] = _gvn.transform(kls);
3405 }
3406
3407 // Having loaded both klasses, test each for null.
3408 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3409 for (which_arg = 0; which_arg <= 1; which_arg++) {
3410 Node* kls = klasses[which_arg];
3411 Node* null_ctl = top();
3412 kls = null_check_oop(kls, &null_ctl, never_see_null);
3413 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3414 region->init_req(prim_path, null_ctl);
3415 if (stopped()) break;
3416 klasses[which_arg] = kls;
3417 }
3418
3419 if (!stopped()) {
3420 // now we have two reference types, in klasses[0..1]
3421 Node* subk = klasses[1]; // the argument to isAssignableFrom
3422 Node* superk = klasses[0]; // the receiver
3423 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3424 // now we have a successful reference subtype check
3425 region->set_req(_ref_subtype_path, control());
3426 }
3427
3428 // If both operands are primitive (both klasses null), then
3429 // we must return true when they are identical primitives.
3430 // It is convenient to test this after the first null klass check.
3431 set_control(region->in(_prim_0_path)); // go back to first null check
3432 if (!stopped()) {
3433 // Since superc is primitive, make a guard for the superc==subc case.
3434 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3435 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3436 generate_guard(bol_eq, region, PROB_FAIR);
3437 if (region->req() == PATH_LIMIT+1) {
3438 // A guard was added. If the added guard is taken, superc==subc.
3439 region->swap_edges(PATH_LIMIT, _prim_same_path);
3440 region->del_req(PATH_LIMIT);
3441 }
3442 region->set_req(_prim_0_path, control()); // Not equal after all.
3443 }
3444
3445 // these are the only paths that produce 'true':
3446 phi->set_req(_prim_same_path, intcon(1));
3447 phi->set_req(_ref_subtype_path, intcon(1));
3448
3449 // pull together the cases:
3450 assert(region->req() == PATH_LIMIT, "sane region");
3451 for (uint i = 1; i < region->req(); i++) {
3452 Node* ctl = region->in(i);
3453 if (ctl == NULL || ctl == top()) {
3454 region->set_req(i, top());
3455 phi ->set_req(i, top());
3456 } else if (phi->in(i) == NULL) {
3457 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3458 }
3459 }
3460
3461 set_control(_gvn.transform(region));
3462 set_result(_gvn.transform(phi));
3463 return true;
3464 }
3465
3466 //---------------------generate_array_guard_common------------------------
3467 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3468 bool obj_array, bool not_array) {
3469
3470 if (stopped()) {
3471 return NULL;
3472 }
3473
3474 // If obj_array/non_array==false/false:
3475 // Branch around if the given klass is in fact an array (either obj or prim).
3476 // If obj_array/non_array==false/true:
3477 // Branch around if the given klass is not an array klass of any kind.
3478 // If obj_array/non_array==true/true:
3479 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3480 // If obj_array/non_array==true/false:
3481 // Branch around if the kls is an oop array (Object[] or subtype)
3482 //
3483 // Like generate_guard, adds a new path onto the region.
3484 jint layout_con = 0;
3485 Node* layout_val = get_layout_helper(kls, layout_con);
3486 if (layout_val == NULL) {
3487 bool query = (obj_array
3488 ? Klass::layout_helper_is_objArray(layout_con)
3489 : Klass::layout_helper_is_array(layout_con));
3490 if (query == not_array) {
3491 return NULL; // never a branch
3492 } else { // always a branch
3493 Node* always_branch = control();
3494 if (region != NULL)
3495 region->add_req(always_branch);
3496 set_control(top());
3497 return always_branch;
3498 }
3499 }
3500 // Now test the correct condition.
3501 jint nval = (obj_array
3502 ? (jint)(Klass::_lh_array_tag_type_value
3503 << Klass::_lh_array_tag_shift)
3504 : Klass::_lh_neutral_value);
3505 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3506 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3507 // invert the test if we are looking for a non-array
3508 if (not_array) btest = BoolTest(btest).negate();
3509 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3510 return generate_fair_guard(bol, region);
3511 }
3512
3513
3514 //-----------------------inline_native_newArray--------------------------
3515 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3516 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3517 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3518 Node* mirror;
3519 Node* count_val;
3520 if (uninitialized) {
3521 mirror = argument(1);
3522 count_val = argument(2);
3523 } else {
3524 mirror = argument(0);
3525 count_val = argument(1);
3526 }
3527
3528 mirror = null_check(mirror);
3529 // If mirror or obj is dead, only null-path is taken.
3530 if (stopped()) return true;
3531
3532 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3533 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3534 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3535 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3536 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3537
3538 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3539 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3540 result_reg, _slow_path);
3541 Node* normal_ctl = control();
3542 Node* no_array_ctl = result_reg->in(_slow_path);
3543
3544 // Generate code for the slow case. We make a call to newArray().
3545 set_control(no_array_ctl);
3546 if (!stopped()) {
3547 // Either the input type is void.class, or else the
3548 // array klass has not yet been cached. Either the
3549 // ensuing call will throw an exception, or else it
3550 // will cache the array klass for next time.
3551 PreserveJVMState pjvms(this);
3552 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3553 Node* slow_result = set_results_for_java_call(slow_call);
3554 // this->control() comes from set_results_for_java_call
3555 result_reg->set_req(_slow_path, control());
3556 result_val->set_req(_slow_path, slow_result);
3557 result_io ->set_req(_slow_path, i_o());
3558 result_mem->set_req(_slow_path, reset_memory());
3559 }
3560
3561 set_control(normal_ctl);
3562 if (!stopped()) {
3563 // Normal case: The array type has been cached in the java.lang.Class.
3564 // The following call works fine even if the array type is polymorphic.
3565 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3566 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3567 result_reg->init_req(_normal_path, control());
3568 result_val->init_req(_normal_path, obj);
3569 result_io ->init_req(_normal_path, i_o());
3570 result_mem->init_req(_normal_path, reset_memory());
3571
3572 if (uninitialized) {
3573 // Mark the allocation so that zeroing is skipped
3574 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3575 alloc->maybe_set_complete(&_gvn);
3576 }
3577 }
3578
3579 // Return the combined state.
3580 set_i_o( _gvn.transform(result_io) );
3581 set_all_memory( _gvn.transform(result_mem));
3582
3583 C->set_has_split_ifs(true); // Has chance for split-if optimization
3584 set_result(result_reg, result_val);
3585 return true;
3586 }
3587
3588 //----------------------inline_native_getLength--------------------------
3589 // public static native int java.lang.reflect.Array.getLength(Object array);
3590 bool LibraryCallKit::inline_native_getLength() {
3591 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3592
3593 Node* array = null_check(argument(0));
3594 // If array is dead, only null-path is taken.
3595 if (stopped()) return true;
3596
3597 // Deoptimize if it is a non-array.
3598 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3599
3600 if (non_array != NULL) {
3601 PreserveJVMState pjvms(this);
3602 set_control(non_array);
3603 uncommon_trap(Deoptimization::Reason_intrinsic,
3604 Deoptimization::Action_maybe_recompile);
3605 }
3606
3607 // If control is dead, only non-array-path is taken.
3608 if (stopped()) return true;
3609
3610 // The works fine even if the array type is polymorphic.
3611 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3612 Node* result = load_array_length(array);
3613
3614 C->set_has_split_ifs(true); // Has chance for split-if optimization
3615 set_result(result);
3616 return true;
3617 }
3618
3619 //------------------------inline_array_copyOf----------------------------
3620 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3621 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3622 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3623 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3624
3625 // Get the arguments.
3626 Node* original = argument(0);
3627 Node* start = is_copyOfRange? argument(1): intcon(0);
3628 Node* end = is_copyOfRange? argument(2): argument(1);
3629 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3630
3631 Node* newcopy = NULL;
3632
3633 // Set the original stack and the reexecute bit for the interpreter to reexecute
3634 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3635 { PreserveReexecuteState preexecs(this);
3636 jvms()->set_should_reexecute(true);
3637
3638 array_type_mirror = null_check(array_type_mirror);
3639 original = null_check(original);
3640
3641 // Check if a null path was taken unconditionally.
3642 if (stopped()) return true;
3643
3644 Node* orig_length = load_array_length(original);
3645
3646 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3647 klass_node = null_check(klass_node);
3648
3649 RegionNode* bailout = new RegionNode(1);
3650 record_for_igvn(bailout);
3651
3652 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3653 // Bail out if that is so.
3654 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3655 if (not_objArray != NULL) {
3656 // Improve the klass node's type from the new optimistic assumption:
3657 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3658 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3659 Node* cast = new CastPPNode(klass_node, akls);
3660 cast->init_req(0, control());
3661 klass_node = _gvn.transform(cast);
3662 }
3663
3664 // Bail out if either start or end is negative.
3665 generate_negative_guard(start, bailout, &start);
3666 generate_negative_guard(end, bailout, &end);
3667
3668 Node* length = end;
3669 if (_gvn.type(start) != TypeInt::ZERO) {
3670 length = _gvn.transform(new SubINode(end, start));
3671 }
3672
3673 // Bail out if length is negative.
3674 // Without this the new_array would throw
3675 // NegativeArraySizeException but IllegalArgumentException is what
3676 // should be thrown
3677 generate_negative_guard(length, bailout, &length);
3678
3679 if (bailout->req() > 1) {
3680 PreserveJVMState pjvms(this);
3681 set_control(_gvn.transform(bailout));
3682 uncommon_trap(Deoptimization::Reason_intrinsic,
3683 Deoptimization::Action_maybe_recompile);
3684 }
3685
3686 if (!stopped()) {
3687 // How many elements will we copy from the original?
3688 // The answer is MinI(orig_length - start, length).
3689 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3690 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3691
3692 // Generate a direct call to the right arraycopy function(s).
3693 // We know the copy is disjoint but we might not know if the
3694 // oop stores need checking.
3695 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3696 // This will fail a store-check if x contains any non-nulls.
3697
3698 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3699 // loads/stores but it is legal only if we're sure the
3700 // Arrays.copyOf would succeed. So we need all input arguments
3701 // to the copyOf to be validated, including that the copy to the
3702 // new array won't trigger an ArrayStoreException. That subtype
3703 // check can be optimized if we know something on the type of
3704 // the input array from type speculation.
3705 if (_gvn.type(klass_node)->singleton()) {
3706 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3707 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3708
3709 int test = C->static_subtype_check(superk, subk);
3710 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3711 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3712 if (t_original->speculative_type() != NULL) {
3713 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3714 }
3715 }
3716 }
3717
3718 bool validated = false;
3719 // Reason_class_check rather than Reason_intrinsic because we
3720 // want to intrinsify even if this traps.
3721 if (!too_many_traps(Deoptimization::Reason_class_check)) {
3722 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
3723
3724 if (not_subtype_ctrl != top()) {
3725 PreserveJVMState pjvms(this);
3726 set_control(not_subtype_ctrl);
3727 uncommon_trap(Deoptimization::Reason_class_check,
3728 Deoptimization::Action_make_not_entrant);
3729 assert(stopped(), "Should be stopped");
3730 }
3731 validated = true;
3732 }
3733
3734 if (!stopped()) {
3735 newcopy = new_array(klass_node, length, 0); // no arguments to push
3736
3737 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
3738 load_object_klass(original), klass_node);
3739 if (!is_copyOfRange) {
3740 ac->set_copyof(validated);
3741 } else {
3742 ac->set_copyofrange(validated);
3743 }
3744 Node* n = _gvn.transform(ac);
3745 if (n == ac) {
3746 ac->connect_outputs(this);
3747 } else {
3748 assert(validated, "shouldn't transform if all arguments not validated");
3749 set_all_memory(n);
3750 }
3751 }
3752 }
3753 } // original reexecute is set back here
3754
3755 C->set_has_split_ifs(true); // Has chance for split-if optimization
3756 if (!stopped()) {
3757 set_result(newcopy);
3758 }
3759 return true;
3760 }
3761
3762
3763 //----------------------generate_virtual_guard---------------------------
3764 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
3765 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
3766 RegionNode* slow_region) {
3767 ciMethod* method = callee();
3768 int vtable_index = method->vtable_index();
3769 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3770 "bad index %d", vtable_index);
3771 // Get the Method* out of the appropriate vtable entry.
3772 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
3773 vtable_index*vtableEntry::size_in_bytes() +
3774 vtableEntry::method_offset_in_bytes();
3775 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
3776 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3777
3778 // Compare the target method with the expected method (e.g., Object.hashCode).
3779 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
3780
3781 Node* native_call = makecon(native_call_addr);
3782 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
3783 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
3784
3785 return generate_slow_guard(test_native, slow_region);
3786 }
3787
3788 //-----------------------generate_method_call----------------------------
3789 // Use generate_method_call to make a slow-call to the real
3790 // method if the fast path fails. An alternative would be to
3791 // use a stub like OptoRuntime::slow_arraycopy_Java.
3792 // This only works for expanding the current library call,
3793 // not another intrinsic. (E.g., don't use this for making an
3794 // arraycopy call inside of the copyOf intrinsic.)
3795 CallJavaNode*
3796 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
3797 // When compiling the intrinsic method itself, do not use this technique.
3798 guarantee(callee() != C->method(), "cannot make slow-call to self");
3799
3800 ciMethod* method = callee();
3801 // ensure the JVMS we have will be correct for this call
3802 guarantee(method_id == method->intrinsic_id(), "must match");
3803
3804 const TypeFunc* tf = TypeFunc::make(method);
3805 CallJavaNode* slow_call;
3806 if (is_static) {
3807 assert(!is_virtual, "");
3808 slow_call = new CallStaticJavaNode(C, tf,
3809 SharedRuntime::get_resolve_static_call_stub(),
3810 method, bci());
3811 } else if (is_virtual) {
3812 null_check_receiver();
3813 int vtable_index = Method::invalid_vtable_index;
3814 if (UseInlineCaches) {
3815 // Suppress the vtable call
3816 } else {
3817 // hashCode and clone are not a miranda methods,
3818 // so the vtable index is fixed.
3819 // No need to use the linkResolver to get it.
3820 vtable_index = method->vtable_index();
3821 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3822 "bad index %d", vtable_index);
3823 }
3824 slow_call = new CallDynamicJavaNode(tf,
3825 SharedRuntime::get_resolve_virtual_call_stub(),
3826 method, vtable_index, bci());
3827 } else { // neither virtual nor static: opt_virtual
3828 null_check_receiver();
3829 slow_call = new CallStaticJavaNode(C, tf,
3830 SharedRuntime::get_resolve_opt_virtual_call_stub(),
3831 method, bci());
3832 slow_call->set_optimized_virtual(true);
3833 }
3834 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
3835 // To be able to issue a direct call (optimized virtual or virtual)
3836 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
3837 // about the method being invoked should be attached to the call site to
3838 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
3839 slow_call->set_override_symbolic_info(true);
3840 }
3841 set_arguments_for_java_call(slow_call);
3842 set_edges_for_java_call(slow_call);
3843 return slow_call;
3844 }
3845
3846
3847 /**
3848 * Build special case code for calls to hashCode on an object. This call may
3849 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
3850 * slightly different code.
3851 */
3852 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
3853 assert(is_static == callee()->is_static(), "correct intrinsic selection");
3854 assert(!(is_virtual && is_static), "either virtual, special, or static");
3855
3856 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
3857
3858 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3859 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
3860 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3861 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3862 Node* obj = NULL;
3863 if (!is_static) {
3864 // Check for hashing null object
3865 obj = null_check_receiver();
3866 if (stopped()) return true; // unconditionally null
3867 result_reg->init_req(_null_path, top());
3868 result_val->init_req(_null_path, top());
3869 } else {
3870 // Do a null check, and return zero if null.
3871 // System.identityHashCode(null) == 0
3872 obj = argument(0);
3873 Node* null_ctl = top();
3874 obj = null_check_oop(obj, &null_ctl);
3875 result_reg->init_req(_null_path, null_ctl);
3876 result_val->init_req(_null_path, _gvn.intcon(0));
3877 }
3878
3879 // Unconditionally null? Then return right away.
3880 if (stopped()) {
3881 set_control( result_reg->in(_null_path));
3882 if (!stopped())
3883 set_result(result_val->in(_null_path));
3884 return true;
3885 }
3886
3887 // We only go to the fast case code if we pass a number of guards. The
3888 // paths which do not pass are accumulated in the slow_region.
3889 RegionNode* slow_region = new RegionNode(1);
3890 record_for_igvn(slow_region);
3891
3892 // If this is a virtual call, we generate a funny guard. We pull out
3893 // the vtable entry corresponding to hashCode() from the target object.
3894 // If the target method which we are calling happens to be the native
3895 // Object hashCode() method, we pass the guard. We do not need this
3896 // guard for non-virtual calls -- the caller is known to be the native
3897 // Object hashCode().
3898 if (is_virtual) {
3899 // After null check, get the object's klass.
3900 Node* obj_klass = load_object_klass(obj);
3901 generate_virtual_guard(obj_klass, slow_region);
3902 }
3903
3904 // Get the header out of the object, use LoadMarkNode when available
3905 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
3906 // The control of the load must be NULL. Otherwise, the load can move before
3907 // the null check after castPP removal.
3908 Node* no_ctrl = NULL;
3909 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
3910
3911 // Test the header to see if it is unlocked.
3912 Node *lock_mask = _gvn.MakeConX(markWord::biased_lock_mask_in_place);
3913 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
3914 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
3915 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
3916 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
3917
3918 generate_slow_guard(test_unlocked, slow_region);
3919
3920 // Get the hash value and check to see that it has been properly assigned.
3921 // We depend on hash_mask being at most 32 bits and avoid the use of
3922 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
3923 // vm: see markWord.hpp.
3924 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
3925 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
3926 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
3927 // This hack lets the hash bits live anywhere in the mark object now, as long
3928 // as the shift drops the relevant bits into the low 32 bits. Note that
3929 // Java spec says that HashCode is an int so there's no point in capturing
3930 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
3931 hshifted_header = ConvX2I(hshifted_header);
3932 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
3933
3934 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
3935 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
3936 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
3937
3938 generate_slow_guard(test_assigned, slow_region);
3939
3940 Node* init_mem = reset_memory();
3941 // fill in the rest of the null path:
3942 result_io ->init_req(_null_path, i_o());
3943 result_mem->init_req(_null_path, init_mem);
3944
3945 result_val->init_req(_fast_path, hash_val);
3946 result_reg->init_req(_fast_path, control());
3947 result_io ->init_req(_fast_path, i_o());
3948 result_mem->init_req(_fast_path, init_mem);
3949
3950 // Generate code for the slow case. We make a call to hashCode().
3951 set_control(_gvn.transform(slow_region));
3952 if (!stopped()) {
3953 // No need for PreserveJVMState, because we're using up the present state.
3954 set_all_memory(init_mem);
3955 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
3956 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
3957 Node* slow_result = set_results_for_java_call(slow_call);
3958 // this->control() comes from set_results_for_java_call
3959 result_reg->init_req(_slow_path, control());
3960 result_val->init_req(_slow_path, slow_result);
3961 result_io ->set_req(_slow_path, i_o());
3962 result_mem ->set_req(_slow_path, reset_memory());
3963 }
3964
3965 // Return the combined state.
3966 set_i_o( _gvn.transform(result_io) );
3967 set_all_memory( _gvn.transform(result_mem));
3968
3969 set_result(result_reg, result_val);
3970 return true;
3971 }
3972
3973 //---------------------------inline_native_getClass----------------------------
3974 // public final native Class<?> java.lang.Object.getClass();
3975 //
3976 // Build special case code for calls to getClass on an object.
3977 bool LibraryCallKit::inline_native_getClass() {
3978 Node* obj = null_check_receiver();
3979 if (stopped()) return true;
3980 set_result(load_mirror_from_klass(load_object_klass(obj)));
3981 return true;
3982 }
3983
3984 //-----------------inline_native_Reflection_getCallerClass---------------------
3985 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
3986 //
3987 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
3988 //
3989 // NOTE: This code must perform the same logic as JVM_GetCallerClass
3990 // in that it must skip particular security frames and checks for
3991 // caller sensitive methods.
3992 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
3993 #ifndef PRODUCT
3994 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3995 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
3996 }
3997 #endif
3998
3999 if (!jvms()->has_method()) {
4000 #ifndef PRODUCT
4001 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4002 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4003 }
4004 #endif
4005 return false;
4006 }
4007
4008 // Walk back up the JVM state to find the caller at the required
4009 // depth.
4010 JVMState* caller_jvms = jvms();
4011
4012 // Cf. JVM_GetCallerClass
4013 // NOTE: Start the loop at depth 1 because the current JVM state does
4014 // not include the Reflection.getCallerClass() frame.
4015 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4016 ciMethod* m = caller_jvms->method();
4017 switch (n) {
4018 case 0:
4019 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4020 break;
4021 case 1:
4022 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4023 if (!m->caller_sensitive()) {
4024 #ifndef PRODUCT
4025 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4026 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4027 }
4028 #endif
4029 return false; // bail-out; let JVM_GetCallerClass do the work
4030 }
4031 break;
4032 default:
4033 if (!m->is_ignored_by_security_stack_walk()) {
4034 // We have reached the desired frame; return the holder class.
4035 // Acquire method holder as java.lang.Class and push as constant.
4036 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4037 ciInstance* caller_mirror = caller_klass->java_mirror();
4038 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4039
4040 #ifndef PRODUCT
4041 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4042 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());
4043 tty->print_cr(" JVM state at this point:");
4044 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4045 ciMethod* m = jvms()->of_depth(i)->method();
4046 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4047 }
4048 }
4049 #endif
4050 return true;
4051 }
4052 break;
4053 }
4054 }
4055
4056 #ifndef PRODUCT
4057 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4058 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4059 tty->print_cr(" JVM state at this point:");
4060 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4061 ciMethod* m = jvms()->of_depth(i)->method();
4062 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4063 }
4064 }
4065 #endif
4066
4067 return false; // bail-out; let JVM_GetCallerClass do the work
4068 }
4069
4070 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4071 Node* arg = argument(0);
4072 Node* result = NULL;
4073
4074 switch (id) {
4075 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4076 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4077 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4078 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4079
4080 case vmIntrinsics::_doubleToLongBits: {
4081 // two paths (plus control) merge in a wood
4082 RegionNode *r = new RegionNode(3);
4083 Node *phi = new PhiNode(r, TypeLong::LONG);
4084
4085 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4086 // Build the boolean node
4087 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4088
4089 // Branch either way.
4090 // NaN case is less traveled, which makes all the difference.
4091 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4092 Node *opt_isnan = _gvn.transform(ifisnan);
4093 assert( opt_isnan->is_If(), "Expect an IfNode");
4094 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4095 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4096
4097 set_control(iftrue);
4098
4099 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4100 Node *slow_result = longcon(nan_bits); // return NaN
4101 phi->init_req(1, _gvn.transform( slow_result ));
4102 r->init_req(1, iftrue);
4103
4104 // Else fall through
4105 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4106 set_control(iffalse);
4107
4108 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4109 r->init_req(2, iffalse);
4110
4111 // Post merge
4112 set_control(_gvn.transform(r));
4113 record_for_igvn(r);
4114
4115 C->set_has_split_ifs(true); // Has chance for split-if optimization
4116 result = phi;
4117 assert(result->bottom_type()->isa_long(), "must be");
4118 break;
4119 }
4120
4121 case vmIntrinsics::_floatToIntBits: {
4122 // two paths (plus control) merge in a wood
4123 RegionNode *r = new RegionNode(3);
4124 Node *phi = new PhiNode(r, TypeInt::INT);
4125
4126 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4127 // Build the boolean node
4128 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4129
4130 // Branch either way.
4131 // NaN case is less traveled, which makes all the difference.
4132 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4133 Node *opt_isnan = _gvn.transform(ifisnan);
4134 assert( opt_isnan->is_If(), "Expect an IfNode");
4135 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4136 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4137
4138 set_control(iftrue);
4139
4140 static const jint nan_bits = 0x7fc00000;
4141 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4142 phi->init_req(1, _gvn.transform( slow_result ));
4143 r->init_req(1, iftrue);
4144
4145 // Else fall through
4146 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4147 set_control(iffalse);
4148
4149 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4150 r->init_req(2, iffalse);
4151
4152 // Post merge
4153 set_control(_gvn.transform(r));
4154 record_for_igvn(r);
4155
4156 C->set_has_split_ifs(true); // Has chance for split-if optimization
4157 result = phi;
4158 assert(result->bottom_type()->isa_int(), "must be");
4159 break;
4160 }
4161
4162 default:
4163 fatal_unexpected_iid(id);
4164 break;
4165 }
4166 set_result(_gvn.transform(result));
4167 return true;
4168 }
4169
4170 //----------------------inline_unsafe_copyMemory-------------------------
4171 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4172 bool LibraryCallKit::inline_unsafe_copyMemory() {
4173 if (callee()->is_static()) return false; // caller must have the capability!
4174 null_check_receiver(); // null-check receiver
4175 if (stopped()) return true;
4176
4177 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4178
4179 Node* src_ptr = argument(1); // type: oop
4180 Node* src_off = ConvL2X(argument(2)); // type: long
4181 Node* dst_ptr = argument(4); // type: oop
4182 Node* dst_off = ConvL2X(argument(5)); // type: long
4183 Node* size = ConvL2X(argument(7)); // type: long
4184
4185 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4186 "fieldOffset must be byte-scaled");
4187
4188 Node* src = make_unsafe_address(src_ptr, src_off, ACCESS_READ);
4189 Node* dst = make_unsafe_address(dst_ptr, dst_off, ACCESS_WRITE);
4190
4191 // Conservatively insert a memory barrier on all memory slices.
4192 // Do not let writes of the copy source or destination float below the copy.
4193 insert_mem_bar(Op_MemBarCPUOrder);
4194
4195 Node* thread = _gvn.transform(new ThreadLocalNode());
4196 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4197 BasicType doing_unsafe_access_bt = T_BYTE;
4198 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4199
4200 // update volatile field
4201 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4202
4203 // Call it. Note that the length argument is not scaled.
4204 make_runtime_call(RC_LEAF|RC_NO_FP,
4205 OptoRuntime::fast_arraycopy_Type(),
4206 StubRoutines::unsafe_arraycopy(),
4207 "unsafe_arraycopy",
4208 TypeRawPtr::BOTTOM,
4209 src, dst, size XTOP);
4210
4211 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4212
4213 // Do not let reads of the copy destination float above the copy.
4214 insert_mem_bar(Op_MemBarCPUOrder);
4215
4216 return true;
4217 }
4218
4219 //------------------------clone_coping-----------------------------------
4220 // Helper function for inline_native_clone.
4221 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4222 assert(obj_size != NULL, "");
4223 Node* raw_obj = alloc_obj->in(1);
4224 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4225
4226 AllocateNode* alloc = NULL;
4227 if (ReduceBulkZeroing) {
4228 // We will be completely responsible for initializing this object -
4229 // mark Initialize node as complete.
4230 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4231 // The object was just allocated - there should be no any stores!
4232 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4233 // Mark as complete_with_arraycopy so that on AllocateNode
4234 // expansion, we know this AllocateNode is initialized by an array
4235 // copy and a StoreStore barrier exists after the array copy.
4236 alloc->initialization()->set_complete_with_arraycopy();
4237 }
4238
4239 Node* size = _gvn.transform(obj_size);
4240 access_clone(obj, alloc_obj, size, is_array);
4241
4242 // Do not let reads from the cloned object float above the arraycopy.
4243 if (alloc != NULL) {
4244 // Do not let stores that initialize this object be reordered with
4245 // a subsequent store that would make this object accessible by
4246 // other threads.
4247 // Record what AllocateNode this StoreStore protects so that
4248 // escape analysis can go from the MemBarStoreStoreNode to the
4249 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4250 // based on the escape status of the AllocateNode.
4251 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4252 } else {
4253 insert_mem_bar(Op_MemBarCPUOrder);
4254 }
4255 }
4256
4257 //------------------------inline_native_clone----------------------------
4258 // protected native Object java.lang.Object.clone();
4259 //
4260 // Here are the simple edge cases:
4261 // null receiver => normal trap
4262 // virtual and clone was overridden => slow path to out-of-line clone
4263 // not cloneable or finalizer => slow path to out-of-line Object.clone
4264 //
4265 // The general case has two steps, allocation and copying.
4266 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4267 //
4268 // Copying also has two cases, oop arrays and everything else.
4269 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4270 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4271 //
4272 // These steps fold up nicely if and when the cloned object's klass
4273 // can be sharply typed as an object array, a type array, or an instance.
4274 //
4275 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4276 PhiNode* result_val;
4277
4278 // Set the reexecute bit for the interpreter to reexecute
4279 // the bytecode that invokes Object.clone if deoptimization happens.
4280 { PreserveReexecuteState preexecs(this);
4281 jvms()->set_should_reexecute(true);
4282
4283 Node* obj = null_check_receiver();
4284 if (stopped()) return true;
4285
4286 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4287
4288 // If we are going to clone an instance, we need its exact type to
4289 // know the number and types of fields to convert the clone to
4290 // loads/stores. Maybe a speculative type can help us.
4291 if (!obj_type->klass_is_exact() &&
4292 obj_type->speculative_type() != NULL &&
4293 obj_type->speculative_type()->is_instance_klass()) {
4294 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4295 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4296 !spec_ik->has_injected_fields()) {
4297 ciKlass* k = obj_type->klass();
4298 if (!k->is_instance_klass() ||
4299 k->as_instance_klass()->is_interface() ||
4300 k->as_instance_klass()->has_subklass()) {
4301 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4302 }
4303 }
4304 }
4305
4306 // Conservatively insert a memory barrier on all memory slices.
4307 // Do not let writes into the original float below the clone.
4308 insert_mem_bar(Op_MemBarCPUOrder);
4309
4310 // paths into result_reg:
4311 enum {
4312 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4313 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4314 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4315 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4316 PATH_LIMIT
4317 };
4318 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4319 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4320 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4321 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4322 record_for_igvn(result_reg);
4323
4324 Node* obj_klass = load_object_klass(obj);
4325 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4326 if (array_ctl != NULL) {
4327 // It's an array.
4328 PreserveJVMState pjvms(this);
4329 set_control(array_ctl);
4330 Node* obj_length = load_array_length(obj);
4331 Node* obj_size = NULL;
4332 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, /*deoptimize_on_exception=*/true);
4333
4334 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4335 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, BarrierSetC2::Parsing)) {
4336 // If it is an oop array, it requires very special treatment,
4337 // because gc barriers are required when accessing the array.
4338 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4339 if (is_obja != NULL) {
4340 PreserveJVMState pjvms2(this);
4341 set_control(is_obja);
4342 // Generate a direct call to the right arraycopy function(s).
4343 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4344 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false);
4345 ac->set_clone_oop_array();
4346 Node* n = _gvn.transform(ac);
4347 assert(n == ac, "cannot disappear");
4348 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
4349
4350 result_reg->init_req(_objArray_path, control());
4351 result_val->init_req(_objArray_path, alloc_obj);
4352 result_i_o ->set_req(_objArray_path, i_o());
4353 result_mem ->set_req(_objArray_path, reset_memory());
4354 }
4355 }
4356 // Otherwise, there are no barriers to worry about.
4357 // (We can dispense with card marks if we know the allocation
4358 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4359 // causes the non-eden paths to take compensating steps to
4360 // simulate a fresh allocation, so that no further
4361 // card marks are required in compiled code to initialize
4362 // the object.)
4363
4364 if (!stopped()) {
4365 copy_to_clone(obj, alloc_obj, obj_size, true);
4366
4367 // Present the results of the copy.
4368 result_reg->init_req(_array_path, control());
4369 result_val->init_req(_array_path, alloc_obj);
4370 result_i_o ->set_req(_array_path, i_o());
4371 result_mem ->set_req(_array_path, reset_memory());
4372 }
4373 }
4374
4375 // We only go to the instance fast case code if we pass a number of guards.
4376 // The paths which do not pass are accumulated in the slow_region.
4377 RegionNode* slow_region = new RegionNode(1);
4378 record_for_igvn(slow_region);
4379 if (!stopped()) {
4380 // It's an instance (we did array above). Make the slow-path tests.
4381 // If this is a virtual call, we generate a funny guard. We grab
4382 // the vtable entry corresponding to clone() from the target object.
4383 // If the target method which we are calling happens to be the
4384 // Object clone() method, we pass the guard. We do not need this
4385 // guard for non-virtual calls; the caller is known to be the native
4386 // Object clone().
4387 if (is_virtual) {
4388 generate_virtual_guard(obj_klass, slow_region);
4389 }
4390
4391 // The object must be easily cloneable and must not have a finalizer.
4392 // Both of these conditions may be checked in a single test.
4393 // We could optimize the test further, but we don't care.
4394 generate_access_flags_guard(obj_klass,
4395 // Test both conditions:
4396 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4397 // Must be cloneable but not finalizer:
4398 JVM_ACC_IS_CLONEABLE_FAST,
4399 slow_region);
4400 }
4401
4402 if (!stopped()) {
4403 // It's an instance, and it passed the slow-path tests.
4404 PreserveJVMState pjvms(this);
4405 Node* obj_size = NULL;
4406 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4407 // is reexecuted if deoptimization occurs and there could be problems when merging
4408 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4409 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4410
4411 copy_to_clone(obj, alloc_obj, obj_size, false);
4412
4413 // Present the results of the slow call.
4414 result_reg->init_req(_instance_path, control());
4415 result_val->init_req(_instance_path, alloc_obj);
4416 result_i_o ->set_req(_instance_path, i_o());
4417 result_mem ->set_req(_instance_path, reset_memory());
4418 }
4419
4420 // Generate code for the slow case. We make a call to clone().
4421 set_control(_gvn.transform(slow_region));
4422 if (!stopped()) {
4423 PreserveJVMState pjvms(this);
4424 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4425 // We need to deoptimize on exception (see comment above)
4426 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4427 // this->control() comes from set_results_for_java_call
4428 result_reg->init_req(_slow_path, control());
4429 result_val->init_req(_slow_path, slow_result);
4430 result_i_o ->set_req(_slow_path, i_o());
4431 result_mem ->set_req(_slow_path, reset_memory());
4432 }
4433
4434 // Return the combined state.
4435 set_control( _gvn.transform(result_reg));
4436 set_i_o( _gvn.transform(result_i_o));
4437 set_all_memory( _gvn.transform(result_mem));
4438 } // original reexecute is set back here
4439
4440 set_result(_gvn.transform(result_val));
4441 return true;
4442 }
4443
4444 // If we have a tightly coupled allocation, the arraycopy may take care
4445 // of the array initialization. If one of the guards we insert between
4446 // the allocation and the arraycopy causes a deoptimization, an
4447 // unitialized array will escape the compiled method. To prevent that
4448 // we set the JVM state for uncommon traps between the allocation and
4449 // the arraycopy to the state before the allocation so, in case of
4450 // deoptimization, we'll reexecute the allocation and the
4451 // initialization.
4452 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4453 if (alloc != NULL) {
4454 ciMethod* trap_method = alloc->jvms()->method();
4455 int trap_bci = alloc->jvms()->bci();
4456
4457 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4458 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4459 // Make sure there's no store between the allocation and the
4460 // arraycopy otherwise visible side effects could be rexecuted
4461 // in case of deoptimization and cause incorrect execution.
4462 bool no_interfering_store = true;
4463 Node* mem = alloc->in(TypeFunc::Memory);
4464 if (mem->is_MergeMem()) {
4465 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4466 Node* n = mms.memory();
4467 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4468 assert(n->is_Store(), "what else?");
4469 no_interfering_store = false;
4470 break;
4471 }
4472 }
4473 } else {
4474 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4475 Node* n = mms.memory();
4476 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4477 assert(n->is_Store(), "what else?");
4478 no_interfering_store = false;
4479 break;
4480 }
4481 }
4482 }
4483
4484 if (no_interfering_store) {
4485 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4486 uint size = alloc->req();
4487 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4488 old_jvms->set_map(sfpt);
4489 for (uint i = 0; i < size; i++) {
4490 sfpt->init_req(i, alloc->in(i));
4491 }
4492 // re-push array length for deoptimization
4493 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4494 old_jvms->set_sp(old_jvms->sp()+1);
4495 old_jvms->set_monoff(old_jvms->monoff()+1);
4496 old_jvms->set_scloff(old_jvms->scloff()+1);
4497 old_jvms->set_endoff(old_jvms->endoff()+1);
4498 old_jvms->set_should_reexecute(true);
4499
4500 sfpt->set_i_o(map()->i_o());
4501 sfpt->set_memory(map()->memory());
4502 sfpt->set_control(map()->control());
4503
4504 JVMState* saved_jvms = jvms();
4505 saved_reexecute_sp = _reexecute_sp;
4506
4507 set_jvms(sfpt->jvms());
4508 _reexecute_sp = jvms()->sp();
4509
4510 return saved_jvms;
4511 }
4512 }
4513 }
4514 return NULL;
4515 }
4516
4517 // In case of a deoptimization, we restart execution at the
4518 // allocation, allocating a new array. We would leave an uninitialized
4519 // array in the heap that GCs wouldn't expect. Move the allocation
4520 // after the traps so we don't allocate the array if we
4521 // deoptimize. This is possible because tightly_coupled_allocation()
4522 // guarantees there's no observer of the allocated array at this point
4523 // and the control flow is simple enough.
4524 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4525 int saved_reexecute_sp, uint new_idx) {
4526 if (saved_jvms != NULL && !stopped()) {
4527 assert(alloc != NULL, "only with a tightly coupled allocation");
4528 // restore JVM state to the state at the arraycopy
4529 saved_jvms->map()->set_control(map()->control());
4530 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4531 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4532 // If we've improved the types of some nodes (null check) while
4533 // emitting the guards, propagate them to the current state
4534 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4535 set_jvms(saved_jvms);
4536 _reexecute_sp = saved_reexecute_sp;
4537
4538 // Remove the allocation from above the guards
4539 CallProjections callprojs;
4540 alloc->extract_projections(&callprojs, true);
4541 InitializeNode* init = alloc->initialization();
4542 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4543 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4544 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4545 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4546
4547 // move the allocation here (after the guards)
4548 _gvn.hash_delete(alloc);
4549 alloc->set_req(TypeFunc::Control, control());
4550 alloc->set_req(TypeFunc::I_O, i_o());
4551 Node *mem = reset_memory();
4552 set_all_memory(mem);
4553 alloc->set_req(TypeFunc::Memory, mem);
4554 set_control(init->proj_out_or_null(TypeFunc::Control));
4555 set_i_o(callprojs.fallthrough_ioproj);
4556
4557 // Update memory as done in GraphKit::set_output_for_allocation()
4558 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4559 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4560 if (ary_type->isa_aryptr() && length_type != NULL) {
4561 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4562 }
4563 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4564 int elemidx = C->get_alias_index(telemref);
4565 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
4566 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
4567
4568 Node* allocx = _gvn.transform(alloc);
4569 assert(allocx == alloc, "where has the allocation gone?");
4570 assert(dest->is_CheckCastPP(), "not an allocation result?");
4571
4572 _gvn.hash_delete(dest);
4573 dest->set_req(0, control());
4574 Node* destx = _gvn.transform(dest);
4575 assert(destx == dest, "where has the allocation result gone?");
4576 }
4577 }
4578
4579
4580 //------------------------------inline_arraycopy-----------------------
4581 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4582 // Object dest, int destPos,
4583 // int length);
4584 bool LibraryCallKit::inline_arraycopy() {
4585 // Get the arguments.
4586 Node* src = argument(0); // type: oop
4587 Node* src_offset = argument(1); // type: int
4588 Node* dest = argument(2); // type: oop
4589 Node* dest_offset = argument(3); // type: int
4590 Node* length = argument(4); // type: int
4591
4592 uint new_idx = C->unique();
4593
4594 // Check for allocation before we add nodes that would confuse
4595 // tightly_coupled_allocation()
4596 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4597
4598 int saved_reexecute_sp = -1;
4599 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4600 // See arraycopy_restore_alloc_state() comment
4601 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4602 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4603 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
4604 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4605
4606 // The following tests must be performed
4607 // (1) src and dest are arrays.
4608 // (2) src and dest arrays must have elements of the same BasicType
4609 // (3) src and dest must not be null.
4610 // (4) src_offset must not be negative.
4611 // (5) dest_offset must not be negative.
4612 // (6) length must not be negative.
4613 // (7) src_offset + length must not exceed length of src.
4614 // (8) dest_offset + length must not exceed length of dest.
4615 // (9) each element of an oop array must be assignable
4616
4617 // (3) src and dest must not be null.
4618 // always do this here because we need the JVM state for uncommon traps
4619 Node* null_ctl = top();
4620 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4621 assert(null_ctl->is_top(), "no null control here");
4622 dest = null_check(dest, T_ARRAY);
4623
4624 if (!can_emit_guards) {
4625 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4626 // guards but the arraycopy node could still take advantage of a
4627 // tightly allocated allocation. tightly_coupled_allocation() is
4628 // called again to make sure it takes the null check above into
4629 // account: the null check is mandatory and if it caused an
4630 // uncommon trap to be emitted then the allocation can't be
4631 // considered tightly coupled in this context.
4632 alloc = tightly_coupled_allocation(dest, NULL);
4633 }
4634
4635 bool validated = false;
4636
4637 const Type* src_type = _gvn.type(src);
4638 const Type* dest_type = _gvn.type(dest);
4639 const TypeAryPtr* top_src = src_type->isa_aryptr();
4640 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4641
4642 // Do we have the type of src?
4643 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4644 // Do we have the type of dest?
4645 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4646 // Is the type for src from speculation?
4647 bool src_spec = false;
4648 // Is the type for dest from speculation?
4649 bool dest_spec = false;
4650
4651 if ((!has_src || !has_dest) && can_emit_guards) {
4652 // We don't have sufficient type information, let's see if
4653 // speculative types can help. We need to have types for both src
4654 // and dest so that it pays off.
4655
4656 // Do we already have or could we have type information for src
4657 bool could_have_src = has_src;
4658 // Do we already have or could we have type information for dest
4659 bool could_have_dest = has_dest;
4660
4661 ciKlass* src_k = NULL;
4662 if (!has_src) {
4663 src_k = src_type->speculative_type_not_null();
4664 if (src_k != NULL && src_k->is_array_klass()) {
4665 could_have_src = true;
4666 }
4667 }
4668
4669 ciKlass* dest_k = NULL;
4670 if (!has_dest) {
4671 dest_k = dest_type->speculative_type_not_null();
4672 if (dest_k != NULL && dest_k->is_array_klass()) {
4673 could_have_dest = true;
4674 }
4675 }
4676
4677 if (could_have_src && could_have_dest) {
4678 // This is going to pay off so emit the required guards
4679 if (!has_src) {
4680 src = maybe_cast_profiled_obj(src, src_k, true);
4681 src_type = _gvn.type(src);
4682 top_src = src_type->isa_aryptr();
4683 has_src = (top_src != NULL && top_src->klass() != NULL);
4684 src_spec = true;
4685 }
4686 if (!has_dest) {
4687 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4688 dest_type = _gvn.type(dest);
4689 top_dest = dest_type->isa_aryptr();
4690 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4691 dest_spec = true;
4692 }
4693 }
4694 }
4695
4696 if (has_src && has_dest && can_emit_guards) {
4697 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
4698 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
4699 if (is_reference_type(src_elem)) src_elem = T_OBJECT;
4700 if (is_reference_type(dest_elem)) dest_elem = T_OBJECT;
4701
4702 if (src_elem == dest_elem && src_elem == T_OBJECT) {
4703 // If both arrays are object arrays then having the exact types
4704 // for both will remove the need for a subtype check at runtime
4705 // before the call and may make it possible to pick a faster copy
4706 // routine (without a subtype check on every element)
4707 // Do we have the exact type of src?
4708 bool could_have_src = src_spec;
4709 // Do we have the exact type of dest?
4710 bool could_have_dest = dest_spec;
4711 ciKlass* src_k = top_src->klass();
4712 ciKlass* dest_k = top_dest->klass();
4713 if (!src_spec) {
4714 src_k = src_type->speculative_type_not_null();
4715 if (src_k != NULL && src_k->is_array_klass()) {
4716 could_have_src = true;
4717 }
4718 }
4719 if (!dest_spec) {
4720 dest_k = dest_type->speculative_type_not_null();
4721 if (dest_k != NULL && dest_k->is_array_klass()) {
4722 could_have_dest = true;
4723 }
4724 }
4725 if (could_have_src && could_have_dest) {
4726 // If we can have both exact types, emit the missing guards
4727 if (could_have_src && !src_spec) {
4728 src = maybe_cast_profiled_obj(src, src_k, true);
4729 }
4730 if (could_have_dest && !dest_spec) {
4731 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4732 }
4733 }
4734 }
4735 }
4736
4737 ciMethod* trap_method = method();
4738 int trap_bci = bci();
4739 if (saved_jvms != NULL) {
4740 trap_method = alloc->jvms()->method();
4741 trap_bci = alloc->jvms()->bci();
4742 }
4743
4744 bool negative_length_guard_generated = false;
4745
4746 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4747 can_emit_guards &&
4748 !src->is_top() && !dest->is_top()) {
4749 // validate arguments: enables transformation the ArrayCopyNode
4750 validated = true;
4751
4752 RegionNode* slow_region = new RegionNode(1);
4753 record_for_igvn(slow_region);
4754
4755 // (1) src and dest are arrays.
4756 generate_non_array_guard(load_object_klass(src), slow_region);
4757 generate_non_array_guard(load_object_klass(dest), slow_region);
4758
4759 // (2) src and dest arrays must have elements of the same BasicType
4760 // done at macro expansion or at Ideal transformation time
4761
4762 // (4) src_offset must not be negative.
4763 generate_negative_guard(src_offset, slow_region);
4764
4765 // (5) dest_offset must not be negative.
4766 generate_negative_guard(dest_offset, slow_region);
4767
4768 // (7) src_offset + length must not exceed length of src.
4769 generate_limit_guard(src_offset, length,
4770 load_array_length(src),
4771 slow_region);
4772
4773 // (8) dest_offset + length must not exceed length of dest.
4774 generate_limit_guard(dest_offset, length,
4775 load_array_length(dest),
4776 slow_region);
4777
4778 // (6) length must not be negative.
4779 // This is also checked in generate_arraycopy() during macro expansion, but
4780 // we also have to check it here for the case where the ArrayCopyNode will
4781 // be eliminated by Escape Analysis.
4782 if (EliminateAllocations) {
4783 generate_negative_guard(length, slow_region);
4784 negative_length_guard_generated = true;
4785 }
4786
4787 // (9) each element of an oop array must be assignable
4788 Node* dest_klass = load_object_klass(dest);
4789 if (src != dest) {
4790 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
4791
4792 if (not_subtype_ctrl != top()) {
4793 PreserveJVMState pjvms(this);
4794 set_control(not_subtype_ctrl);
4795 uncommon_trap(Deoptimization::Reason_intrinsic,
4796 Deoptimization::Action_make_not_entrant);
4797 assert(stopped(), "Should be stopped");
4798 }
4799 }
4800 {
4801 PreserveJVMState pjvms(this);
4802 set_control(_gvn.transform(slow_region));
4803 uncommon_trap(Deoptimization::Reason_intrinsic,
4804 Deoptimization::Action_make_not_entrant);
4805 assert(stopped(), "Should be stopped");
4806 }
4807
4808 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
4809 const Type *toop = TypeOopPtr::make_from_klass(dest_klass_t->klass());
4810 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
4811 }
4812
4813 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
4814
4815 if (stopped()) {
4816 return true;
4817 }
4818
4819 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
4820 // Create LoadRange and LoadKlass nodes for use during macro expansion here
4821 // so the compiler has a chance to eliminate them: during macro expansion,
4822 // we have to set their control (CastPP nodes are eliminated).
4823 load_object_klass(src), load_object_klass(dest),
4824 load_array_length(src), load_array_length(dest));
4825
4826 ac->set_arraycopy(validated);
4827
4828 Node* n = _gvn.transform(ac);
4829 if (n == ac) {
4830 ac->connect_outputs(this);
4831 } else {
4832 assert(validated, "shouldn't transform if all arguments not validated");
4833 set_all_memory(n);
4834 }
4835 clear_upper_avx();
4836
4837
4838 return true;
4839 }
4840
4841
4842 // Helper function which determines if an arraycopy immediately follows
4843 // an allocation, with no intervening tests or other escapes for the object.
4844 AllocateArrayNode*
4845 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
4846 RegionNode* slow_region) {
4847 if (stopped()) return NULL; // no fast path
4848 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
4849
4850 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
4851 if (alloc == NULL) return NULL;
4852
4853 Node* rawmem = memory(Compile::AliasIdxRaw);
4854 // Is the allocation's memory state untouched?
4855 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
4856 // Bail out if there have been raw-memory effects since the allocation.
4857 // (Example: There might have been a call or safepoint.)
4858 return NULL;
4859 }
4860 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
4861 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
4862 return NULL;
4863 }
4864
4865 // There must be no unexpected observers of this allocation.
4866 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
4867 Node* obs = ptr->fast_out(i);
4868 if (obs != this->map()) {
4869 return NULL;
4870 }
4871 }
4872
4873 // This arraycopy must unconditionally follow the allocation of the ptr.
4874 Node* alloc_ctl = ptr->in(0);
4875 Node* ctl = control();
4876 while (ctl != alloc_ctl) {
4877 // There may be guards which feed into the slow_region.
4878 // Any other control flow means that we might not get a chance
4879 // to finish initializing the allocated object.
4880 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
4881 IfNode* iff = ctl->in(0)->as_If();
4882 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
4883 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
4884 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
4885 ctl = iff->in(0); // This test feeds the known slow_region.
4886 continue;
4887 }
4888 // One more try: Various low-level checks bottom out in
4889 // uncommon traps. If the debug-info of the trap omits
4890 // any reference to the allocation, as we've already
4891 // observed, then there can be no objection to the trap.
4892 bool found_trap = false;
4893 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
4894 Node* obs = not_ctl->fast_out(j);
4895 if (obs->in(0) == not_ctl && obs->is_Call() &&
4896 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
4897 found_trap = true; break;
4898 }
4899 }
4900 if (found_trap) {
4901 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
4902 continue;
4903 }
4904 }
4905 return NULL;
4906 }
4907
4908 // If we get this far, we have an allocation which immediately
4909 // precedes the arraycopy, and we can take over zeroing the new object.
4910 // The arraycopy will finish the initialization, and provide
4911 // a new control state to which we will anchor the destination pointer.
4912
4913 return alloc;
4914 }
4915
4916 //-------------inline_encodeISOArray-----------------------------------
4917 // encode char[] to byte[] in ISO_8859_1
4918 bool LibraryCallKit::inline_encodeISOArray() {
4919 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
4920 // no receiver since it is static method
4921 Node *src = argument(0);
4922 Node *src_offset = argument(1);
4923 Node *dst = argument(2);
4924 Node *dst_offset = argument(3);
4925 Node *length = argument(4);
4926
4927 src = must_be_not_null(src, true);
4928 dst = must_be_not_null(dst, true);
4929
4930 const Type* src_type = src->Value(&_gvn);
4931 const Type* dst_type = dst->Value(&_gvn);
4932 const TypeAryPtr* top_src = src_type->isa_aryptr();
4933 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
4934 if (top_src == NULL || top_src->klass() == NULL ||
4935 top_dest == NULL || top_dest->klass() == NULL) {
4936 // failed array check
4937 return false;
4938 }
4939
4940 // Figure out the size and type of the elements we will be copying.
4941 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4942 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4943 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
4944 return false;
4945 }
4946
4947 Node* src_start = array_element_address(src, src_offset, T_CHAR);
4948 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
4949 // 'src_start' points to src array + scaled offset
4950 // 'dst_start' points to dst array + scaled offset
4951
4952 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
4953 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
4954 enc = _gvn.transform(enc);
4955 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
4956 set_memory(res_mem, mtype);
4957 set_result(enc);
4958 clear_upper_avx();
4959
4960 return true;
4961 }
4962
4963 //-------------inline_multiplyToLen-----------------------------------
4964 bool LibraryCallKit::inline_multiplyToLen() {
4965 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
4966
4967 address stubAddr = StubRoutines::multiplyToLen();
4968 if (stubAddr == NULL) {
4969 return false; // Intrinsic's stub is not implemented on this platform
4970 }
4971 const char* stubName = "multiplyToLen";
4972
4973 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
4974
4975 // no receiver because it is a static method
4976 Node* x = argument(0);
4977 Node* xlen = argument(1);
4978 Node* y = argument(2);
4979 Node* ylen = argument(3);
4980 Node* z = argument(4);
4981
4982 x = must_be_not_null(x, true);
4983 y = must_be_not_null(y, true);
4984
4985 const Type* x_type = x->Value(&_gvn);
4986 const Type* y_type = y->Value(&_gvn);
4987 const TypeAryPtr* top_x = x_type->isa_aryptr();
4988 const TypeAryPtr* top_y = y_type->isa_aryptr();
4989 if (top_x == NULL || top_x->klass() == NULL ||
4990 top_y == NULL || top_y->klass() == NULL) {
4991 // failed array check
4992 return false;
4993 }
4994
4995 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4996 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4997 if (x_elem != T_INT || y_elem != T_INT) {
4998 return false;
4999 }
5000
5001 // Set the original stack and the reexecute bit for the interpreter to reexecute
5002 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5003 // on the return from z array allocation in runtime.
5004 { PreserveReexecuteState preexecs(this);
5005 jvms()->set_should_reexecute(true);
5006
5007 Node* x_start = array_element_address(x, intcon(0), x_elem);
5008 Node* y_start = array_element_address(y, intcon(0), y_elem);
5009 // 'x_start' points to x array + scaled xlen
5010 // 'y_start' points to y array + scaled ylen
5011
5012 // Allocate the result array
5013 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5014 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5015 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5016
5017 IdealKit ideal(this);
5018
5019 #define __ ideal.
5020 Node* one = __ ConI(1);
5021 Node* zero = __ ConI(0);
5022 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5023 __ set(need_alloc, zero);
5024 __ set(z_alloc, z);
5025 __ if_then(z, BoolTest::eq, null()); {
5026 __ increment (need_alloc, one);
5027 } __ else_(); {
5028 // Update graphKit memory and control from IdealKit.
5029 sync_kit(ideal);
5030 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5031 cast->init_req(0, control());
5032 _gvn.set_type(cast, cast->bottom_type());
5033 C->record_for_igvn(cast);
5034
5035 Node* zlen_arg = load_array_length(cast);
5036 // Update IdealKit memory and control from graphKit.
5037 __ sync_kit(this);
5038 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5039 __ increment (need_alloc, one);
5040 } __ end_if();
5041 } __ end_if();
5042
5043 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5044 // Update graphKit memory and control from IdealKit.
5045 sync_kit(ideal);
5046 Node * narr = new_array(klass_node, zlen, 1);
5047 // Update IdealKit memory and control from graphKit.
5048 __ sync_kit(this);
5049 __ set(z_alloc, narr);
5050 } __ end_if();
5051
5052 sync_kit(ideal);
5053 z = __ value(z_alloc);
5054 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5055 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5056 // Final sync IdealKit and GraphKit.
5057 final_sync(ideal);
5058 #undef __
5059
5060 Node* z_start = array_element_address(z, intcon(0), T_INT);
5061
5062 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5063 OptoRuntime::multiplyToLen_Type(),
5064 stubAddr, stubName, TypePtr::BOTTOM,
5065 x_start, xlen, y_start, ylen, z_start, zlen);
5066 } // original reexecute is set back here
5067
5068 C->set_has_split_ifs(true); // Has chance for split-if optimization
5069 set_result(z);
5070 return true;
5071 }
5072
5073 //-------------inline_squareToLen------------------------------------
5074 bool LibraryCallKit::inline_squareToLen() {
5075 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5076
5077 address stubAddr = StubRoutines::squareToLen();
5078 if (stubAddr == NULL) {
5079 return false; // Intrinsic's stub is not implemented on this platform
5080 }
5081 const char* stubName = "squareToLen";
5082
5083 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5084
5085 Node* x = argument(0);
5086 Node* len = argument(1);
5087 Node* z = argument(2);
5088 Node* zlen = argument(3);
5089
5090 x = must_be_not_null(x, true);
5091 z = must_be_not_null(z, true);
5092
5093 const Type* x_type = x->Value(&_gvn);
5094 const Type* z_type = z->Value(&_gvn);
5095 const TypeAryPtr* top_x = x_type->isa_aryptr();
5096 const TypeAryPtr* top_z = z_type->isa_aryptr();
5097 if (top_x == NULL || top_x->klass() == NULL ||
5098 top_z == NULL || top_z->klass() == NULL) {
5099 // failed array check
5100 return false;
5101 }
5102
5103 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5104 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5105 if (x_elem != T_INT || z_elem != T_INT) {
5106 return false;
5107 }
5108
5109
5110 Node* x_start = array_element_address(x, intcon(0), x_elem);
5111 Node* z_start = array_element_address(z, intcon(0), z_elem);
5112
5113 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5114 OptoRuntime::squareToLen_Type(),
5115 stubAddr, stubName, TypePtr::BOTTOM,
5116 x_start, len, z_start, zlen);
5117
5118 set_result(z);
5119 return true;
5120 }
5121
5122 //-------------inline_mulAdd------------------------------------------
5123 bool LibraryCallKit::inline_mulAdd() {
5124 assert(UseMulAddIntrinsic, "not implemented on this platform");
5125
5126 address stubAddr = StubRoutines::mulAdd();
5127 if (stubAddr == NULL) {
5128 return false; // Intrinsic's stub is not implemented on this platform
5129 }
5130 const char* stubName = "mulAdd";
5131
5132 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5133
5134 Node* out = argument(0);
5135 Node* in = argument(1);
5136 Node* offset = argument(2);
5137 Node* len = argument(3);
5138 Node* k = argument(4);
5139
5140 out = must_be_not_null(out, true);
5141
5142 const Type* out_type = out->Value(&_gvn);
5143 const Type* in_type = in->Value(&_gvn);
5144 const TypeAryPtr* top_out = out_type->isa_aryptr();
5145 const TypeAryPtr* top_in = in_type->isa_aryptr();
5146 if (top_out == NULL || top_out->klass() == NULL ||
5147 top_in == NULL || top_in->klass() == NULL) {
5148 // failed array check
5149 return false;
5150 }
5151
5152 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5153 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5154 if (out_elem != T_INT || in_elem != T_INT) {
5155 return false;
5156 }
5157
5158 Node* outlen = load_array_length(out);
5159 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5160 Node* out_start = array_element_address(out, intcon(0), out_elem);
5161 Node* in_start = array_element_address(in, intcon(0), in_elem);
5162
5163 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5164 OptoRuntime::mulAdd_Type(),
5165 stubAddr, stubName, TypePtr::BOTTOM,
5166 out_start,in_start, new_offset, len, k);
5167 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5168 set_result(result);
5169 return true;
5170 }
5171
5172 //-------------inline_montgomeryMultiply-----------------------------------
5173 bool LibraryCallKit::inline_montgomeryMultiply() {
5174 address stubAddr = StubRoutines::montgomeryMultiply();
5175 if (stubAddr == NULL) {
5176 return false; // Intrinsic's stub is not implemented on this platform
5177 }
5178
5179 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5180 const char* stubName = "montgomery_multiply";
5181
5182 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5183
5184 Node* a = argument(0);
5185 Node* b = argument(1);
5186 Node* n = argument(2);
5187 Node* len = argument(3);
5188 Node* inv = argument(4);
5189 Node* m = argument(6);
5190
5191 const Type* a_type = a->Value(&_gvn);
5192 const TypeAryPtr* top_a = a_type->isa_aryptr();
5193 const Type* b_type = b->Value(&_gvn);
5194 const TypeAryPtr* top_b = b_type->isa_aryptr();
5195 const Type* n_type = a->Value(&_gvn);
5196 const TypeAryPtr* top_n = n_type->isa_aryptr();
5197 const Type* m_type = a->Value(&_gvn);
5198 const TypeAryPtr* top_m = m_type->isa_aryptr();
5199 if (top_a == NULL || top_a->klass() == NULL ||
5200 top_b == NULL || top_b->klass() == NULL ||
5201 top_n == NULL || top_n->klass() == NULL ||
5202 top_m == NULL || top_m->klass() == NULL) {
5203 // failed array check
5204 return false;
5205 }
5206
5207 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5208 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5209 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5210 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5211 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5212 return false;
5213 }
5214
5215 // Make the call
5216 {
5217 Node* a_start = array_element_address(a, intcon(0), a_elem);
5218 Node* b_start = array_element_address(b, intcon(0), b_elem);
5219 Node* n_start = array_element_address(n, intcon(0), n_elem);
5220 Node* m_start = array_element_address(m, intcon(0), m_elem);
5221
5222 Node* call = make_runtime_call(RC_LEAF,
5223 OptoRuntime::montgomeryMultiply_Type(),
5224 stubAddr, stubName, TypePtr::BOTTOM,
5225 a_start, b_start, n_start, len, inv, top(),
5226 m_start);
5227 set_result(m);
5228 }
5229
5230 return true;
5231 }
5232
5233 bool LibraryCallKit::inline_montgomerySquare() {
5234 address stubAddr = StubRoutines::montgomerySquare();
5235 if (stubAddr == NULL) {
5236 return false; // Intrinsic's stub is not implemented on this platform
5237 }
5238
5239 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5240 const char* stubName = "montgomery_square";
5241
5242 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5243
5244 Node* a = argument(0);
5245 Node* n = argument(1);
5246 Node* len = argument(2);
5247 Node* inv = argument(3);
5248 Node* m = argument(5);
5249
5250 const Type* a_type = a->Value(&_gvn);
5251 const TypeAryPtr* top_a = a_type->isa_aryptr();
5252 const Type* n_type = a->Value(&_gvn);
5253 const TypeAryPtr* top_n = n_type->isa_aryptr();
5254 const Type* m_type = a->Value(&_gvn);
5255 const TypeAryPtr* top_m = m_type->isa_aryptr();
5256 if (top_a == NULL || top_a->klass() == NULL ||
5257 top_n == NULL || top_n->klass() == NULL ||
5258 top_m == NULL || top_m->klass() == NULL) {
5259 // failed array check
5260 return false;
5261 }
5262
5263 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5264 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5265 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5266 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5267 return false;
5268 }
5269
5270 // Make the call
5271 {
5272 Node* a_start = array_element_address(a, intcon(0), a_elem);
5273 Node* n_start = array_element_address(n, intcon(0), n_elem);
5274 Node* m_start = array_element_address(m, intcon(0), m_elem);
5275
5276 Node* call = make_runtime_call(RC_LEAF,
5277 OptoRuntime::montgomerySquare_Type(),
5278 stubAddr, stubName, TypePtr::BOTTOM,
5279 a_start, n_start, len, inv, top(),
5280 m_start);
5281 set_result(m);
5282 }
5283
5284 return true;
5285 }
5286
5287 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
5288 address stubAddr = NULL;
5289 const char* stubName = NULL;
5290
5291 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
5292 if (stubAddr == NULL) {
5293 return false; // Intrinsic's stub is not implemented on this platform
5294 }
5295
5296 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
5297
5298 assert(callee()->signature()->size() == 5, "expected 5 arguments");
5299
5300 Node* newArr = argument(0);
5301 Node* oldArr = argument(1);
5302 Node* newIdx = argument(2);
5303 Node* shiftCount = argument(3);
5304 Node* numIter = argument(4);
5305
5306 const Type* newArr_type = newArr->Value(&_gvn);
5307 const TypeAryPtr* top_newArr = newArr_type->isa_aryptr();
5308 const Type* oldArr_type = oldArr->Value(&_gvn);
5309 const TypeAryPtr* top_oldArr = oldArr_type->isa_aryptr();
5310 if (top_newArr == NULL || top_newArr->klass() == NULL || top_oldArr == NULL
5311 || top_oldArr->klass() == NULL) {
5312 return false;
5313 }
5314
5315 BasicType newArr_elem = newArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5316 BasicType oldArr_elem = oldArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5317 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
5318 return false;
5319 }
5320
5321 // Make the call
5322 {
5323 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
5324 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
5325
5326 Node* call = make_runtime_call(RC_LEAF,
5327 OptoRuntime::bigIntegerShift_Type(),
5328 stubAddr,
5329 stubName,
5330 TypePtr::BOTTOM,
5331 newArr_start,
5332 oldArr_start,
5333 newIdx,
5334 shiftCount,
5335 numIter);
5336 }
5337
5338 return true;
5339 }
5340
5341 //-------------inline_vectorizedMismatch------------------------------
5342 bool LibraryCallKit::inline_vectorizedMismatch() {
5343 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5344
5345 address stubAddr = StubRoutines::vectorizedMismatch();
5346 if (stubAddr == NULL) {
5347 return false; // Intrinsic's stub is not implemented on this platform
5348 }
5349 const char* stubName = "vectorizedMismatch";
5350 int size_l = callee()->signature()->size();
5351 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5352
5353 Node* obja = argument(0);
5354 Node* aoffset = argument(1);
5355 Node* objb = argument(3);
5356 Node* boffset = argument(4);
5357 Node* length = argument(6);
5358 Node* scale = argument(7);
5359
5360 const Type* a_type = obja->Value(&_gvn);
5361 const Type* b_type = objb->Value(&_gvn);
5362 const TypeAryPtr* top_a = a_type->isa_aryptr();
5363 const TypeAryPtr* top_b = b_type->isa_aryptr();
5364 if (top_a == NULL || top_a->klass() == NULL ||
5365 top_b == NULL || top_b->klass() == NULL) {
5366 // failed array check
5367 return false;
5368 }
5369
5370 Node* call;
5371 jvms()->set_should_reexecute(true);
5372
5373 Node* obja_adr = make_unsafe_address(obja, aoffset, ACCESS_READ);
5374 Node* objb_adr = make_unsafe_address(objb, boffset, ACCESS_READ);
5375
5376 call = make_runtime_call(RC_LEAF,
5377 OptoRuntime::vectorizedMismatch_Type(),
5378 stubAddr, stubName, TypePtr::BOTTOM,
5379 obja_adr, objb_adr, length, scale);
5380
5381 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5382 set_result(result);
5383 return true;
5384 }
5385
5386 /**
5387 * Calculate CRC32 for byte.
5388 * int java.util.zip.CRC32.update(int crc, int b)
5389 */
5390 bool LibraryCallKit::inline_updateCRC32() {
5391 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5392 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5393 // no receiver since it is static method
5394 Node* crc = argument(0); // type: int
5395 Node* b = argument(1); // type: int
5396
5397 /*
5398 * int c = ~ crc;
5399 * b = timesXtoThe32[(b ^ c) & 0xFF];
5400 * b = b ^ (c >>> 8);
5401 * crc = ~b;
5402 */
5403
5404 Node* M1 = intcon(-1);
5405 crc = _gvn.transform(new XorINode(crc, M1));
5406 Node* result = _gvn.transform(new XorINode(crc, b));
5407 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5408
5409 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5410 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5411 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5412 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5413
5414 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5415 result = _gvn.transform(new XorINode(crc, result));
5416 result = _gvn.transform(new XorINode(result, M1));
5417 set_result(result);
5418 return true;
5419 }
5420
5421 /**
5422 * Calculate CRC32 for byte[] array.
5423 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5424 */
5425 bool LibraryCallKit::inline_updateBytesCRC32() {
5426 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5427 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5428 // no receiver since it is static method
5429 Node* crc = argument(0); // type: int
5430 Node* src = argument(1); // type: oop
5431 Node* offset = argument(2); // type: int
5432 Node* length = argument(3); // type: int
5433
5434 const Type* src_type = src->Value(&_gvn);
5435 const TypeAryPtr* top_src = src_type->isa_aryptr();
5436 if (top_src == NULL || top_src->klass() == NULL) {
5437 // failed array check
5438 return false;
5439 }
5440
5441 // Figure out the size and type of the elements we will be copying.
5442 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5443 if (src_elem != T_BYTE) {
5444 return false;
5445 }
5446
5447 // 'src_start' points to src array + scaled offset
5448 src = must_be_not_null(src, true);
5449 Node* src_start = array_element_address(src, offset, src_elem);
5450
5451 // We assume that range check is done by caller.
5452 // TODO: generate range check (offset+length < src.length) in debug VM.
5453
5454 // Call the stub.
5455 address stubAddr = StubRoutines::updateBytesCRC32();
5456 const char *stubName = "updateBytesCRC32";
5457
5458 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5459 stubAddr, stubName, TypePtr::BOTTOM,
5460 crc, src_start, length);
5461 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5462 set_result(result);
5463 return true;
5464 }
5465
5466 /**
5467 * Calculate CRC32 for ByteBuffer.
5468 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5469 */
5470 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5471 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5472 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5473 // no receiver since it is static method
5474 Node* crc = argument(0); // type: int
5475 Node* src = argument(1); // type: long
5476 Node* offset = argument(3); // type: int
5477 Node* length = argument(4); // type: int
5478
5479 src = ConvL2X(src); // adjust Java long to machine word
5480 Node* base = _gvn.transform(new CastX2PNode(src));
5481 offset = ConvI2X(offset);
5482
5483 // 'src_start' points to src array + scaled offset
5484 Node* src_start = basic_plus_adr(top(), base, offset);
5485
5486 // Call the stub.
5487 address stubAddr = StubRoutines::updateBytesCRC32();
5488 const char *stubName = "updateBytesCRC32";
5489
5490 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5491 stubAddr, stubName, TypePtr::BOTTOM,
5492 crc, src_start, length);
5493 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5494 set_result(result);
5495 return true;
5496 }
5497
5498 //------------------------------get_table_from_crc32c_class-----------------------
5499 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5500 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5501 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5502
5503 return table;
5504 }
5505
5506 //------------------------------inline_updateBytesCRC32C-----------------------
5507 //
5508 // Calculate CRC32C for byte[] array.
5509 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5510 //
5511 bool LibraryCallKit::inline_updateBytesCRC32C() {
5512 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5513 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5514 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5515 // no receiver since it is a static method
5516 Node* crc = argument(0); // type: int
5517 Node* src = argument(1); // type: oop
5518 Node* offset = argument(2); // type: int
5519 Node* end = argument(3); // type: int
5520
5521 Node* length = _gvn.transform(new SubINode(end, offset));
5522
5523 const Type* src_type = src->Value(&_gvn);
5524 const TypeAryPtr* top_src = src_type->isa_aryptr();
5525 if (top_src == NULL || top_src->klass() == NULL) {
5526 // failed array check
5527 return false;
5528 }
5529
5530 // Figure out the size and type of the elements we will be copying.
5531 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5532 if (src_elem != T_BYTE) {
5533 return false;
5534 }
5535
5536 // 'src_start' points to src array + scaled offset
5537 src = must_be_not_null(src, true);
5538 Node* src_start = array_element_address(src, offset, src_elem);
5539
5540 // static final int[] byteTable in class CRC32C
5541 Node* table = get_table_from_crc32c_class(callee()->holder());
5542 table = must_be_not_null(table, true);
5543 Node* table_start = array_element_address(table, intcon(0), T_INT);
5544
5545 // We assume that range check is done by caller.
5546 // TODO: generate range check (offset+length < src.length) in debug VM.
5547
5548 // Call the stub.
5549 address stubAddr = StubRoutines::updateBytesCRC32C();
5550 const char *stubName = "updateBytesCRC32C";
5551
5552 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5553 stubAddr, stubName, TypePtr::BOTTOM,
5554 crc, src_start, length, table_start);
5555 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5556 set_result(result);
5557 return true;
5558 }
5559
5560 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5561 //
5562 // Calculate CRC32C for DirectByteBuffer.
5563 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5564 //
5565 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5566 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5567 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5568 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5569 // no receiver since it is a static method
5570 Node* crc = argument(0); // type: int
5571 Node* src = argument(1); // type: long
5572 Node* offset = argument(3); // type: int
5573 Node* end = argument(4); // type: int
5574
5575 Node* length = _gvn.transform(new SubINode(end, offset));
5576
5577 src = ConvL2X(src); // adjust Java long to machine word
5578 Node* base = _gvn.transform(new CastX2PNode(src));
5579 offset = ConvI2X(offset);
5580
5581 // 'src_start' points to src array + scaled offset
5582 Node* src_start = basic_plus_adr(top(), base, offset);
5583
5584 // static final int[] byteTable in class CRC32C
5585 Node* table = get_table_from_crc32c_class(callee()->holder());
5586 table = must_be_not_null(table, true);
5587 Node* table_start = array_element_address(table, intcon(0), T_INT);
5588
5589 // Call the stub.
5590 address stubAddr = StubRoutines::updateBytesCRC32C();
5591 const char *stubName = "updateBytesCRC32C";
5592
5593 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5594 stubAddr, stubName, TypePtr::BOTTOM,
5595 crc, src_start, length, table_start);
5596 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5597 set_result(result);
5598 return true;
5599 }
5600
5601 //------------------------------inline_updateBytesAdler32----------------------
5602 //
5603 // Calculate Adler32 checksum for byte[] array.
5604 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5605 //
5606 bool LibraryCallKit::inline_updateBytesAdler32() {
5607 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5608 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5609 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5610 // no receiver since it is static method
5611 Node* crc = argument(0); // type: int
5612 Node* src = argument(1); // type: oop
5613 Node* offset = argument(2); // type: int
5614 Node* length = argument(3); // type: int
5615
5616 const Type* src_type = src->Value(&_gvn);
5617 const TypeAryPtr* top_src = src_type->isa_aryptr();
5618 if (top_src == NULL || top_src->klass() == NULL) {
5619 // failed array check
5620 return false;
5621 }
5622
5623 // Figure out the size and type of the elements we will be copying.
5624 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5625 if (src_elem != T_BYTE) {
5626 return false;
5627 }
5628
5629 // 'src_start' points to src array + scaled offset
5630 Node* src_start = array_element_address(src, offset, src_elem);
5631
5632 // We assume that range check is done by caller.
5633 // TODO: generate range check (offset+length < src.length) in debug VM.
5634
5635 // Call the stub.
5636 address stubAddr = StubRoutines::updateBytesAdler32();
5637 const char *stubName = "updateBytesAdler32";
5638
5639 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5640 stubAddr, stubName, TypePtr::BOTTOM,
5641 crc, src_start, length);
5642 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5643 set_result(result);
5644 return true;
5645 }
5646
5647 //------------------------------inline_updateByteBufferAdler32---------------
5648 //
5649 // Calculate Adler32 checksum for DirectByteBuffer.
5650 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5651 //
5652 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5653 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5654 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5655 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5656 // no receiver since it is static method
5657 Node* crc = argument(0); // type: int
5658 Node* src = argument(1); // type: long
5659 Node* offset = argument(3); // type: int
5660 Node* length = argument(4); // type: int
5661
5662 src = ConvL2X(src); // adjust Java long to machine word
5663 Node* base = _gvn.transform(new CastX2PNode(src));
5664 offset = ConvI2X(offset);
5665
5666 // 'src_start' points to src array + scaled offset
5667 Node* src_start = basic_plus_adr(top(), base, offset);
5668
5669 // Call the stub.
5670 address stubAddr = StubRoutines::updateBytesAdler32();
5671 const char *stubName = "updateBytesAdler32";
5672
5673 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5674 stubAddr, stubName, TypePtr::BOTTOM,
5675 crc, src_start, length);
5676
5677 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5678 set_result(result);
5679 return true;
5680 }
5681
5682 //----------------------------inline_reference_get----------------------------
5683 // public T java.lang.ref.Reference.get();
5684 bool LibraryCallKit::inline_reference_get() {
5685 const int referent_offset = java_lang_ref_Reference::referent_offset();
5686
5687 // Get the argument:
5688 Node* reference_obj = null_check_receiver();
5689 if (stopped()) return true;
5690
5691 const TypeInstPtr* tinst = _gvn.type(reference_obj)->isa_instptr();
5692 assert(tinst != NULL, "obj is null");
5693 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5694 ciInstanceKlass* referenceKlass = tinst->klass()->as_instance_klass();
5695 ciField* field = referenceKlass->get_field_by_name(ciSymbol::make("referent"),
5696 ciSymbol::make("Ljava/lang/Object;"),
5697 false);
5698 assert (field != NULL, "undefined field");
5699
5700 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5701 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5702
5703 ciInstanceKlass* klass = env()->Object_klass();
5704 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5705
5706 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
5707 Node* result = access_load_at(reference_obj, adr, adr_type, object_type, T_OBJECT, decorators);
5708 // Add memory barrier to prevent commoning reads from this field
5709 // across safepoint since GC can change its value.
5710 insert_mem_bar(Op_MemBarCPUOrder);
5711
5712 set_result(result);
5713 return true;
5714 }
5715
5716
5717 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5718 bool is_exact=true, bool is_static=false,
5719 ciInstanceKlass * fromKls=NULL) {
5720 if (fromKls == NULL) {
5721 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5722 assert(tinst != NULL, "obj is null");
5723 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5724 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5725 fromKls = tinst->klass()->as_instance_klass();
5726 } else {
5727 assert(is_static, "only for static field access");
5728 }
5729 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5730 ciSymbol::make(fieldTypeString),
5731 is_static);
5732
5733 assert (field != NULL, "undefined field");
5734 if (field == NULL) return (Node *) NULL;
5735
5736 if (is_static) {
5737 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5738 fromObj = makecon(tip);
5739 }
5740
5741 // Next code copied from Parse::do_get_xxx():
5742
5743 // Compute address and memory type.
5744 int offset = field->offset_in_bytes();
5745 bool is_vol = field->is_volatile();
5746 ciType* field_klass = field->type();
5747 assert(field_klass->is_loaded(), "should be loaded");
5748 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5749 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5750 BasicType bt = field->layout_type();
5751
5752 // Build the resultant type of the load
5753 const Type *type;
5754 if (bt == T_OBJECT) {
5755 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5756 } else {
5757 type = Type::get_const_basic_type(bt);
5758 }
5759
5760 DecoratorSet decorators = IN_HEAP;
5761
5762 if (is_vol) {
5763 decorators |= MO_SEQ_CST;
5764 }
5765
5766 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
5767 }
5768
5769 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5770 bool is_exact = true, bool is_static = false,
5771 ciInstanceKlass * fromKls = NULL) {
5772 if (fromKls == NULL) {
5773 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5774 assert(tinst != NULL, "obj is null");
5775 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5776 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5777 fromKls = tinst->klass()->as_instance_klass();
5778 }
5779 else {
5780 assert(is_static, "only for static field access");
5781 }
5782 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5783 ciSymbol::make(fieldTypeString),
5784 is_static);
5785
5786 assert(field != NULL, "undefined field");
5787 assert(!field->is_volatile(), "not defined for volatile fields");
5788
5789 if (is_static) {
5790 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5791 fromObj = makecon(tip);
5792 }
5793
5794 // Next code copied from Parse::do_get_xxx():
5795
5796 // Compute address and memory type.
5797 int offset = field->offset_in_bytes();
5798 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5799
5800 return adr;
5801 }
5802
5803 //------------------------------inline_aescrypt_Block-----------------------
5804 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
5805 address stubAddr = NULL;
5806 const char *stubName;
5807 assert(UseAES, "need AES instruction support");
5808
5809 switch(id) {
5810 case vmIntrinsics::_aescrypt_encryptBlock:
5811 stubAddr = StubRoutines::aescrypt_encryptBlock();
5812 stubName = "aescrypt_encryptBlock";
5813 break;
5814 case vmIntrinsics::_aescrypt_decryptBlock:
5815 stubAddr = StubRoutines::aescrypt_decryptBlock();
5816 stubName = "aescrypt_decryptBlock";
5817 break;
5818 default:
5819 break;
5820 }
5821 if (stubAddr == NULL) return false;
5822
5823 Node* aescrypt_object = argument(0);
5824 Node* src = argument(1);
5825 Node* src_offset = argument(2);
5826 Node* dest = argument(3);
5827 Node* dest_offset = argument(4);
5828
5829 src = must_be_not_null(src, true);
5830 dest = must_be_not_null(dest, true);
5831
5832 // (1) src and dest are arrays.
5833 const Type* src_type = src->Value(&_gvn);
5834 const Type* dest_type = dest->Value(&_gvn);
5835 const TypeAryPtr* top_src = src_type->isa_aryptr();
5836 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5837 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5838
5839 // for the quick and dirty code we will skip all the checks.
5840 // we are just trying to get the call to be generated.
5841 Node* src_start = src;
5842 Node* dest_start = dest;
5843 if (src_offset != NULL || dest_offset != NULL) {
5844 assert(src_offset != NULL && dest_offset != NULL, "");
5845 src_start = array_element_address(src, src_offset, T_BYTE);
5846 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5847 }
5848
5849 // now need to get the start of its expanded key array
5850 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5851 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5852 if (k_start == NULL) return false;
5853
5854 if (Matcher::pass_original_key_for_aes()) {
5855 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5856 // compatibility issues between Java key expansion and SPARC crypto instructions
5857 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5858 if (original_k_start == NULL) return false;
5859
5860 // Call the stub.
5861 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5862 stubAddr, stubName, TypePtr::BOTTOM,
5863 src_start, dest_start, k_start, original_k_start);
5864 } else {
5865 // Call the stub.
5866 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5867 stubAddr, stubName, TypePtr::BOTTOM,
5868 src_start, dest_start, k_start);
5869 }
5870
5871 return true;
5872 }
5873
5874 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
5875 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
5876 address stubAddr = NULL;
5877 const char *stubName = NULL;
5878
5879 assert(UseAES, "need AES instruction support");
5880
5881 switch(id) {
5882 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
5883 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
5884 stubName = "cipherBlockChaining_encryptAESCrypt";
5885 break;
5886 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
5887 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
5888 stubName = "cipherBlockChaining_decryptAESCrypt";
5889 break;
5890 default:
5891 break;
5892 }
5893 if (stubAddr == NULL) return false;
5894
5895 Node* cipherBlockChaining_object = argument(0);
5896 Node* src = argument(1);
5897 Node* src_offset = argument(2);
5898 Node* len = argument(3);
5899 Node* dest = argument(4);
5900 Node* dest_offset = argument(5);
5901
5902 src = must_be_not_null(src, false);
5903 dest = must_be_not_null(dest, false);
5904
5905 // (1) src and dest are arrays.
5906 const Type* src_type = src->Value(&_gvn);
5907 const Type* dest_type = dest->Value(&_gvn);
5908 const TypeAryPtr* top_src = src_type->isa_aryptr();
5909 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5910 assert (top_src != NULL && top_src->klass() != NULL
5911 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5912
5913 // checks are the responsibility of the caller
5914 Node* src_start = src;
5915 Node* dest_start = dest;
5916 if (src_offset != NULL || dest_offset != NULL) {
5917 assert(src_offset != NULL && dest_offset != NULL, "");
5918 src_start = array_element_address(src, src_offset, T_BYTE);
5919 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5920 }
5921
5922 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
5923 // (because of the predicated logic executed earlier).
5924 // so we cast it here safely.
5925 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5926
5927 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
5928 if (embeddedCipherObj == NULL) return false;
5929
5930 // cast it to what we know it will be at runtime
5931 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
5932 assert(tinst != NULL, "CBC obj is null");
5933 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
5934 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5935 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
5936
5937 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
5938 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
5939 const TypeOopPtr* xtype = aklass->as_instance_type();
5940 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
5941 aescrypt_object = _gvn.transform(aescrypt_object);
5942
5943 // we need to get the start of the aescrypt_object's expanded key array
5944 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5945 if (k_start == NULL) return false;
5946
5947 // similarly, get the start address of the r vector
5948 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
5949 if (objRvec == NULL) return false;
5950 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
5951
5952 Node* cbcCrypt;
5953 if (Matcher::pass_original_key_for_aes()) {
5954 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5955 // compatibility issues between Java key expansion and SPARC crypto instructions
5956 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5957 if (original_k_start == NULL) return false;
5958
5959 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
5960 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5961 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5962 stubAddr, stubName, TypePtr::BOTTOM,
5963 src_start, dest_start, k_start, r_start, len, original_k_start);
5964 } else {
5965 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
5966 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5967 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5968 stubAddr, stubName, TypePtr::BOTTOM,
5969 src_start, dest_start, k_start, r_start, len);
5970 }
5971
5972 // return cipher length (int)
5973 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
5974 set_result(retvalue);
5975 return true;
5976 }
5977
5978 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
5979 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
5980 address stubAddr = NULL;
5981 const char *stubName = NULL;
5982
5983 assert(UseAES, "need AES instruction support");
5984
5985 switch (id) {
5986 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
5987 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
5988 stubName = "electronicCodeBook_encryptAESCrypt";
5989 break;
5990 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
5991 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
5992 stubName = "electronicCodeBook_decryptAESCrypt";
5993 break;
5994 default:
5995 break;
5996 }
5997
5998 if (stubAddr == NULL) return false;
5999
6000 Node* electronicCodeBook_object = argument(0);
6001 Node* src = argument(1);
6002 Node* src_offset = argument(2);
6003 Node* len = argument(3);
6004 Node* dest = argument(4);
6005 Node* dest_offset = argument(5);
6006
6007 // (1) src and dest are arrays.
6008 const Type* src_type = src->Value(&_gvn);
6009 const Type* dest_type = dest->Value(&_gvn);
6010 const TypeAryPtr* top_src = src_type->isa_aryptr();
6011 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6012 assert(top_src != NULL && top_src->klass() != NULL
6013 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6014
6015 // checks are the responsibility of the caller
6016 Node* src_start = src;
6017 Node* dest_start = dest;
6018 if (src_offset != NULL || dest_offset != NULL) {
6019 assert(src_offset != NULL && dest_offset != NULL, "");
6020 src_start = array_element_address(src, src_offset, T_BYTE);
6021 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6022 }
6023
6024 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6025 // (because of the predicated logic executed earlier).
6026 // so we cast it here safely.
6027 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6028
6029 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6030 if (embeddedCipherObj == NULL) return false;
6031
6032 // cast it to what we know it will be at runtime
6033 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6034 assert(tinst != NULL, "ECB obj is null");
6035 assert(tinst->klass()->is_loaded(), "ECB obj is not loaded");
6036 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6037 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6038
6039 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6040 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6041 const TypeOopPtr* xtype = aklass->as_instance_type();
6042 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6043 aescrypt_object = _gvn.transform(aescrypt_object);
6044
6045 // we need to get the start of the aescrypt_object's expanded key array
6046 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6047 if (k_start == NULL) return false;
6048
6049 Node* ecbCrypt;
6050 if (Matcher::pass_original_key_for_aes()) {
6051 // no SPARC version for AES/ECB intrinsics now.
6052 return false;
6053 }
6054 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6055 ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6056 OptoRuntime::electronicCodeBook_aescrypt_Type(),
6057 stubAddr, stubName, TypePtr::BOTTOM,
6058 src_start, dest_start, k_start, len);
6059
6060 // return cipher length (int)
6061 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6062 set_result(retvalue);
6063 return true;
6064 }
6065
6066 //------------------------------inline_counterMode_AESCrypt-----------------------
6067 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6068 assert(UseAES, "need AES instruction support");
6069 if (!UseAESCTRIntrinsics) return false;
6070
6071 address stubAddr = NULL;
6072 const char *stubName = NULL;
6073 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6074 stubAddr = StubRoutines::counterMode_AESCrypt();
6075 stubName = "counterMode_AESCrypt";
6076 }
6077 if (stubAddr == NULL) return false;
6078
6079 Node* counterMode_object = argument(0);
6080 Node* src = argument(1);
6081 Node* src_offset = argument(2);
6082 Node* len = argument(3);
6083 Node* dest = argument(4);
6084 Node* dest_offset = argument(5);
6085
6086 // (1) src and dest are arrays.
6087 const Type* src_type = src->Value(&_gvn);
6088 const Type* dest_type = dest->Value(&_gvn);
6089 const TypeAryPtr* top_src = src_type->isa_aryptr();
6090 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6091 assert(top_src != NULL && top_src->klass() != NULL &&
6092 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6093
6094 // checks are the responsibility of the caller
6095 Node* src_start = src;
6096 Node* dest_start = dest;
6097 if (src_offset != NULL || dest_offset != NULL) {
6098 assert(src_offset != NULL && dest_offset != NULL, "");
6099 src_start = array_element_address(src, src_offset, T_BYTE);
6100 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6101 }
6102
6103 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6104 // (because of the predicated logic executed earlier).
6105 // so we cast it here safely.
6106 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6107 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6108 if (embeddedCipherObj == NULL) return false;
6109 // cast it to what we know it will be at runtime
6110 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6111 assert(tinst != NULL, "CTR obj is null");
6112 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6113 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6114 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6115 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6116 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6117 const TypeOopPtr* xtype = aklass->as_instance_type();
6118 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6119 aescrypt_object = _gvn.transform(aescrypt_object);
6120 // we need to get the start of the aescrypt_object's expanded key array
6121 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6122 if (k_start == NULL) return false;
6123 // similarly, get the start address of the r vector
6124 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6125 if (obj_counter == NULL) return false;
6126 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6127
6128 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6129 if (saved_encCounter == NULL) return false;
6130 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6131 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6132
6133 Node* ctrCrypt;
6134 if (Matcher::pass_original_key_for_aes()) {
6135 // no SPARC version for AES/CTR intrinsics now.
6136 return false;
6137 }
6138 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6139 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6140 OptoRuntime::counterMode_aescrypt_Type(),
6141 stubAddr, stubName, TypePtr::BOTTOM,
6142 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6143
6144 // return cipher length (int)
6145 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6146 set_result(retvalue);
6147 return true;
6148 }
6149
6150 //------------------------------get_key_start_from_aescrypt_object-----------------------
6151 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6152 #if defined(PPC64) || defined(S390)
6153 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6154 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6155 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6156 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6157 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6158 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6159 if (objSessionK == NULL) {
6160 return (Node *) NULL;
6161 }
6162 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6163 #else
6164 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6165 #endif // PPC64
6166 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6167 if (objAESCryptKey == NULL) return (Node *) NULL;
6168
6169 // now have the array, need to get the start address of the K array
6170 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6171 return k_start;
6172 }
6173
6174 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6175 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6176 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6177 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6178 if (objAESCryptKey == NULL) return (Node *) NULL;
6179
6180 // now have the array, need to get the start address of the lastKey array
6181 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6182 return original_k_start;
6183 }
6184
6185 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6186 // Return node representing slow path of predicate check.
6187 // the pseudo code we want to emulate with this predicate is:
6188 // for encryption:
6189 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6190 // for decryption:
6191 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6192 // note cipher==plain is more conservative than the original java code but that's OK
6193 //
6194 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6195 // The receiver was checked for NULL already.
6196 Node* objCBC = argument(0);
6197
6198 Node* src = argument(1);
6199 Node* dest = argument(4);
6200
6201 // Load embeddedCipher field of CipherBlockChaining object.
6202 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6203
6204 // get AESCrypt klass for instanceOf check
6205 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6206 // will have same classloader as CipherBlockChaining object
6207 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6208 assert(tinst != NULL, "CBCobj is null");
6209 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6210
6211 // we want to do an instanceof comparison against the AESCrypt class
6212 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6213 if (!klass_AESCrypt->is_loaded()) {
6214 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6215 Node* ctrl = control();
6216 set_control(top()); // no regular fast path
6217 return ctrl;
6218 }
6219
6220 src = must_be_not_null(src, true);
6221 dest = must_be_not_null(dest, true);
6222
6223 // Resolve oops to stable for CmpP below.
6224 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6225
6226 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6227 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6228 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6229
6230 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6231
6232 // for encryption, we are done
6233 if (!decrypting)
6234 return instof_false; // even if it is NULL
6235
6236 // for decryption, we need to add a further check to avoid
6237 // taking the intrinsic path when cipher and plain are the same
6238 // see the original java code for why.
6239 RegionNode* region = new RegionNode(3);
6240 region->init_req(1, instof_false);
6241
6242 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6243 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6244 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6245 region->init_req(2, src_dest_conjoint);
6246
6247 record_for_igvn(region);
6248 return _gvn.transform(region);
6249 }
6250
6251 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
6252 // Return node representing slow path of predicate check.
6253 // the pseudo code we want to emulate with this predicate is:
6254 // for encryption:
6255 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6256 // for decryption:
6257 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6258 // note cipher==plain is more conservative than the original java code but that's OK
6259 //
6260 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
6261 // The receiver was checked for NULL already.
6262 Node* objECB = argument(0);
6263
6264 // Load embeddedCipher field of ElectronicCodeBook object.
6265 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6266
6267 // get AESCrypt klass for instanceOf check
6268 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6269 // will have same classloader as ElectronicCodeBook object
6270 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
6271 assert(tinst != NULL, "ECBobj is null");
6272 assert(tinst->klass()->is_loaded(), "ECBobj is not loaded");
6273
6274 // we want to do an instanceof comparison against the AESCrypt class
6275 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6276 if (!klass_AESCrypt->is_loaded()) {
6277 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6278 Node* ctrl = control();
6279 set_control(top()); // no regular fast path
6280 return ctrl;
6281 }
6282 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6283
6284 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6285 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6286 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6287
6288 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6289
6290 // for encryption, we are done
6291 if (!decrypting)
6292 return instof_false; // even if it is NULL
6293
6294 // for decryption, we need to add a further check to avoid
6295 // taking the intrinsic path when cipher and plain are the same
6296 // see the original java code for why.
6297 RegionNode* region = new RegionNode(3);
6298 region->init_req(1, instof_false);
6299 Node* src = argument(1);
6300 Node* dest = argument(4);
6301 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6302 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6303 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6304 region->init_req(2, src_dest_conjoint);
6305
6306 record_for_igvn(region);
6307 return _gvn.transform(region);
6308 }
6309
6310 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6311 // Return node representing slow path of predicate check.
6312 // the pseudo code we want to emulate with this predicate is:
6313 // for encryption:
6314 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6315 // for decryption:
6316 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6317 // note cipher==plain is more conservative than the original java code but that's OK
6318 //
6319
6320 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6321 // The receiver was checked for NULL already.
6322 Node* objCTR = argument(0);
6323
6324 // Load embeddedCipher field of CipherBlockChaining object.
6325 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6326
6327 // get AESCrypt klass for instanceOf check
6328 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6329 // will have same classloader as CipherBlockChaining object
6330 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6331 assert(tinst != NULL, "CTRobj is null");
6332 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6333
6334 // we want to do an instanceof comparison against the AESCrypt class
6335 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6336 if (!klass_AESCrypt->is_loaded()) {
6337 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6338 Node* ctrl = control();
6339 set_control(top()); // no regular fast path
6340 return ctrl;
6341 }
6342
6343 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6344 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6345 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6346 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6347 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6348
6349 return instof_false; // even if it is NULL
6350 }
6351
6352 //------------------------------inline_ghash_processBlocks
6353 bool LibraryCallKit::inline_ghash_processBlocks() {
6354 address stubAddr;
6355 const char *stubName;
6356 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6357
6358 stubAddr = StubRoutines::ghash_processBlocks();
6359 stubName = "ghash_processBlocks";
6360
6361 Node* data = argument(0);
6362 Node* offset = argument(1);
6363 Node* len = argument(2);
6364 Node* state = argument(3);
6365 Node* subkeyH = argument(4);
6366
6367 state = must_be_not_null(state, true);
6368 subkeyH = must_be_not_null(subkeyH, true);
6369 data = must_be_not_null(data, true);
6370
6371 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6372 assert(state_start, "state is NULL");
6373 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6374 assert(subkeyH_start, "subkeyH is NULL");
6375 Node* data_start = array_element_address(data, offset, T_BYTE);
6376 assert(data_start, "data is NULL");
6377
6378 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6379 OptoRuntime::ghash_processBlocks_Type(),
6380 stubAddr, stubName, TypePtr::BOTTOM,
6381 state_start, subkeyH_start, data_start, len);
6382 return true;
6383 }
6384
6385 bool LibraryCallKit::inline_base64_encodeBlock() {
6386 address stubAddr;
6387 const char *stubName;
6388 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6389 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6390 stubAddr = StubRoutines::base64_encodeBlock();
6391 stubName = "encodeBlock";
6392
6393 if (!stubAddr) return false;
6394 Node* base64obj = argument(0);
6395 Node* src = argument(1);
6396 Node* offset = argument(2);
6397 Node* len = argument(3);
6398 Node* dest = argument(4);
6399 Node* dp = argument(5);
6400 Node* isURL = argument(6);
6401
6402 src = must_be_not_null(src, true);
6403 dest = must_be_not_null(dest, true);
6404
6405 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6406 assert(src_start, "source array is NULL");
6407 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6408 assert(dest_start, "destination array is NULL");
6409
6410 Node* base64 = make_runtime_call(RC_LEAF,
6411 OptoRuntime::base64_encodeBlock_Type(),
6412 stubAddr, stubName, TypePtr::BOTTOM,
6413 src_start, offset, len, dest_start, dp, isURL);
6414 return true;
6415 }
6416
6417 //------------------------------inline_digestBase_implCompress-----------------------
6418 //
6419 // Calculate MD5 for single-block byte[] array.
6420 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
6421 //
6422 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6423 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6424 //
6425 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6426 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6427 //
6428 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6429 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6430 //
6431 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
6432 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6433
6434 Node* digestBase_obj = argument(0);
6435 Node* src = argument(1); // type oop
6436 Node* ofs = argument(2); // type int
6437
6438 const Type* src_type = src->Value(&_gvn);
6439 const TypeAryPtr* top_src = src_type->isa_aryptr();
6440 if (top_src == NULL || top_src->klass() == NULL) {
6441 // failed array check
6442 return false;
6443 }
6444 // Figure out the size and type of the elements we will be copying.
6445 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6446 if (src_elem != T_BYTE) {
6447 return false;
6448 }
6449 // 'src_start' points to src array + offset
6450 src = must_be_not_null(src, true);
6451 Node* src_start = array_element_address(src, ofs, src_elem);
6452 Node* state = NULL;
6453 address stubAddr;
6454 const char *stubName;
6455
6456 switch(id) {
6457 case vmIntrinsics::_md5_implCompress:
6458 assert(UseMD5Intrinsics, "need MD5 instruction support");
6459 state = get_state_from_digest_object(digestBase_obj);
6460 stubAddr = StubRoutines::md5_implCompress();
6461 stubName = "md5_implCompress";
6462 break;
6463 case vmIntrinsics::_sha_implCompress:
6464 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6465 state = get_state_from_digest_object(digestBase_obj);
6466 stubAddr = StubRoutines::sha1_implCompress();
6467 stubName = "sha1_implCompress";
6468 break;
6469 case vmIntrinsics::_sha2_implCompress:
6470 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6471 state = get_state_from_digest_object(digestBase_obj);
6472 stubAddr = StubRoutines::sha256_implCompress();
6473 stubName = "sha256_implCompress";
6474 break;
6475 case vmIntrinsics::_sha5_implCompress:
6476 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6477 state = get_long_state_from_digest_object(digestBase_obj);
6478 stubAddr = StubRoutines::sha512_implCompress();
6479 stubName = "sha512_implCompress";
6480 break;
6481 default:
6482 fatal_unexpected_iid(id);
6483 return false;
6484 }
6485 if (state == NULL) return false;
6486
6487 assert(stubAddr != NULL, "Stub is generated");
6488 if (stubAddr == NULL) return false;
6489
6490 // Call the stub.
6491 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(),
6492 stubAddr, stubName, TypePtr::BOTTOM,
6493 src_start, state);
6494
6495 return true;
6496 }
6497
6498 //------------------------------inline_digestBase_implCompressMB-----------------------
6499 //
6500 // Calculate MD5/SHA/SHA2/SHA5 for multi-block byte[] array.
6501 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6502 //
6503 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6504 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6505 "need MD5/SHA1/SHA256/SHA512 instruction support");
6506 assert((uint)predicate < 4, "sanity");
6507 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6508
6509 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6510 Node* src = argument(1); // byte[] array
6511 Node* ofs = argument(2); // type int
6512 Node* limit = argument(3); // type int
6513
6514 const Type* src_type = src->Value(&_gvn);
6515 const TypeAryPtr* top_src = src_type->isa_aryptr();
6516 if (top_src == NULL || top_src->klass() == NULL) {
6517 // failed array check
6518 return false;
6519 }
6520 // Figure out the size and type of the elements we will be copying.
6521 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6522 if (src_elem != T_BYTE) {
6523 return false;
6524 }
6525 // 'src_start' points to src array + offset
6526 src = must_be_not_null(src, false);
6527 Node* src_start = array_element_address(src, ofs, src_elem);
6528
6529 const char* klass_digestBase_name = NULL;
6530 const char* stub_name = NULL;
6531 address stub_addr = NULL;
6532 bool long_state = false;
6533
6534 switch (predicate) {
6535 case 0:
6536 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
6537 klass_digestBase_name = "sun/security/provider/MD5";
6538 stub_name = "md5_implCompressMB";
6539 stub_addr = StubRoutines::md5_implCompressMB();
6540 }
6541 break;
6542 case 1:
6543 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
6544 klass_digestBase_name = "sun/security/provider/SHA";
6545 stub_name = "sha1_implCompressMB";
6546 stub_addr = StubRoutines::sha1_implCompressMB();
6547 }
6548 break;
6549 case 2:
6550 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
6551 klass_digestBase_name = "sun/security/provider/SHA2";
6552 stub_name = "sha256_implCompressMB";
6553 stub_addr = StubRoutines::sha256_implCompressMB();
6554 }
6555 break;
6556 case 3:
6557 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
6558 klass_digestBase_name = "sun/security/provider/SHA5";
6559 stub_name = "sha512_implCompressMB";
6560 stub_addr = StubRoutines::sha512_implCompressMB();
6561 long_state = true;
6562 }
6563 break;
6564 default:
6565 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
6566 }
6567 if (klass_digestBase_name != NULL) {
6568 assert(stub_addr != NULL, "Stub is generated");
6569 if (stub_addr == NULL) return false;
6570
6571 // get DigestBase klass to lookup for SHA klass
6572 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6573 assert(tinst != NULL, "digestBase_obj is not instance???");
6574 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6575
6576 ciKlass* klass_digestBase = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
6577 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
6578 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
6579 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, long_state, stub_addr, stub_name, src_start, ofs, limit);
6580 }
6581 return false;
6582 }
6583
6584 //------------------------------inline_digestBase_implCompressMB-----------------------
6585 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
6586 bool long_state, address stubAddr, const char *stubName,
6587 Node* src_start, Node* ofs, Node* limit) {
6588 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
6589 const TypeOopPtr* xtype = aklass->as_instance_type();
6590 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6591 digest_obj = _gvn.transform(digest_obj);
6592
6593 Node* state;
6594 if (long_state) {
6595 state = get_long_state_from_digest_object(digest_obj);
6596 } else {
6597 state = get_state_from_digest_object(digest_obj);
6598 }
6599 if (state == NULL) return false;
6600
6601 // Call the stub.
6602 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6603 OptoRuntime::digestBase_implCompressMB_Type(),
6604 stubAddr, stubName, TypePtr::BOTTOM,
6605 src_start, state, ofs, limit);
6606 // return ofs (int)
6607 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6608 set_result(result);
6609
6610 return true;
6611 }
6612
6613 //------------------------------get_state_from_digest_object-----------------------
6614 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object) {
6615 Node* digest_state = load_field_from_object(digest_object, "state", "[I", /*is_exact*/ false);
6616 assert (digest_state != NULL, "wrong version of sun.security.provider.MD5/SHA/SHA2");
6617 if (digest_state == NULL) return (Node *) NULL;
6618
6619 // now have the array, need to get the start address of the state array
6620 Node* state = array_element_address(digest_state, intcon(0), T_INT);
6621 return state;
6622 }
6623
6624 //------------------------------get_long_state_from_digest_object-----------------------
6625 Node * LibraryCallKit::get_long_state_from_digest_object(Node *digest_object) {
6626 Node* digest_state = load_field_from_object(digest_object, "state", "[J", /*is_exact*/ false);
6627 assert (digest_state != NULL, "wrong version of sun.security.provider.SHA5");
6628 if (digest_state == NULL) return (Node *) NULL;
6629
6630 // now have the array, need to get the start address of the state array
6631 Node* state = array_element_address(digest_state, intcon(0), T_LONG);
6632 return state;
6633 }
6634
6635 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6636 // Return node representing slow path of predicate check.
6637 // the pseudo code we want to emulate with this predicate is:
6638 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6639 //
6640 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6641 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6642 "need MD5/SHA1/SHA256/SHA512 instruction support");
6643 assert((uint)predicate < 4, "sanity");
6644
6645 // The receiver was checked for NULL already.
6646 Node* digestBaseObj = argument(0);
6647
6648 // get DigestBase klass for instanceOf check
6649 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6650 assert(tinst != NULL, "digestBaseObj is null");
6651 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6652
6653 const char* klass_name = NULL;
6654 switch (predicate) {
6655 case 0:
6656 if (UseMD5Intrinsics) {
6657 // we want to do an instanceof comparison against the MD5 class
6658 klass_name = "sun/security/provider/MD5";
6659 }
6660 break;
6661 case 1:
6662 if (UseSHA1Intrinsics) {
6663 // we want to do an instanceof comparison against the SHA class
6664 klass_name = "sun/security/provider/SHA";
6665 }
6666 break;
6667 case 2:
6668 if (UseSHA256Intrinsics) {
6669 // we want to do an instanceof comparison against the SHA2 class
6670 klass_name = "sun/security/provider/SHA2";
6671 }
6672 break;
6673 case 3:
6674 if (UseSHA512Intrinsics) {
6675 // we want to do an instanceof comparison against the SHA5 class
6676 klass_name = "sun/security/provider/SHA5";
6677 }
6678 break;
6679 default:
6680 fatal("unknown SHA intrinsic predicate: %d", predicate);
6681 }
6682
6683 ciKlass* klass = NULL;
6684 if (klass_name != NULL) {
6685 klass = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_name));
6686 }
6687 if ((klass == NULL) || !klass->is_loaded()) {
6688 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6689 Node* ctrl = control();
6690 set_control(top()); // no intrinsic path
6691 return ctrl;
6692 }
6693 ciInstanceKlass* instklass = klass->as_instance_klass();
6694
6695 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
6696 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6697 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6698 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6699
6700 return instof_false; // even if it is NULL
6701 }
6702
6703 //-------------inline_fma-----------------------------------
6704 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
6705 Node *a = NULL;
6706 Node *b = NULL;
6707 Node *c = NULL;
6708 Node* result = NULL;
6709 switch (id) {
6710 case vmIntrinsics::_fmaD:
6711 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
6712 // no receiver since it is static method
6713 a = round_double_node(argument(0));
6714 b = round_double_node(argument(2));
6715 c = round_double_node(argument(4));
6716 result = _gvn.transform(new FmaDNode(control(), a, b, c));
6717 break;
6718 case vmIntrinsics::_fmaF:
6719 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
6720 a = argument(0);
6721 b = argument(1);
6722 c = argument(2);
6723 result = _gvn.transform(new FmaFNode(control(), a, b, c));
6724 break;
6725 default:
6726 fatal_unexpected_iid(id); break;
6727 }
6728 set_result(result);
6729 return true;
6730 }
6731
6732 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
6733 // argument(0) is receiver
6734 Node* codePoint = argument(1);
6735 Node* n = NULL;
6736
6737 switch (id) {
6738 case vmIntrinsics::_isDigit :
6739 n = new DigitNode(control(), codePoint);
6740 break;
6741 case vmIntrinsics::_isLowerCase :
6742 n = new LowerCaseNode(control(), codePoint);
6743 break;
6744 case vmIntrinsics::_isUpperCase :
6745 n = new UpperCaseNode(control(), codePoint);
6746 break;
6747 case vmIntrinsics::_isWhitespace :
6748 n = new WhitespaceNode(control(), codePoint);
6749 break;
6750 default:
6751 fatal_unexpected_iid(id);
6752 }
6753
6754 set_result(_gvn.transform(n));
6755 return true;
6756 }
6757
6758 //------------------------------inline_fp_min_max------------------------------
6759 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
6760 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
6761
6762 // The intrinsic should be used only when the API branches aren't predictable,
6763 // the last one performing the most important comparison. The following heuristic
6764 // uses the branch statistics to eventually bail out if necessary.
6765
6766 ciMethodData *md = callee()->method_data();
6767
6768 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
6769 ciCallProfile cp = caller()->call_profile_at_bci(bci());
6770
6771 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
6772 // Bail out if the call-site didn't contribute enough to the statistics.
6773 return false;
6774 }
6775
6776 uint taken = 0, not_taken = 0;
6777
6778 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
6779 if (p->is_BranchData()) {
6780 taken = ((ciBranchData*)p)->taken();
6781 not_taken = ((ciBranchData*)p)->not_taken();
6782 }
6783 }
6784
6785 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
6786 balance = balance < 0 ? -balance : balance;
6787 if ( balance > 0.2 ) {
6788 // Bail out if the most important branch is predictable enough.
6789 return false;
6790 }
6791 }
6792 */
6793
6794 Node *a = NULL;
6795 Node *b = NULL;
6796 Node *n = NULL;
6797 switch (id) {
6798 case vmIntrinsics::_maxF:
6799 case vmIntrinsics::_minF:
6800 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
6801 a = argument(0);
6802 b = argument(1);
6803 break;
6804 case vmIntrinsics::_maxD:
6805 case vmIntrinsics::_minD:
6806 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
6807 a = round_double_node(argument(0));
6808 b = round_double_node(argument(2));
6809 break;
6810 default:
6811 fatal_unexpected_iid(id);
6812 break;
6813 }
6814 switch (id) {
6815 case vmIntrinsics::_maxF: n = new MaxFNode(a, b); break;
6816 case vmIntrinsics::_minF: n = new MinFNode(a, b); break;
6817 case vmIntrinsics::_maxD: n = new MaxDNode(a, b); break;
6818 case vmIntrinsics::_minD: n = new MinDNode(a, b); break;
6819 default: fatal_unexpected_iid(id); break;
6820 }
6821 set_result(_gvn.transform(n));
6822 return true;
6823 }
6824
6825 bool LibraryCallKit::inline_profileBoolean() {
6826 Node* counts = argument(1);
6827 const TypeAryPtr* ary = NULL;
6828 ciArray* aobj = NULL;
6829 if (counts->is_Con()
6830 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6831 && (aobj = ary->const_oop()->as_array()) != NULL
6832 && (aobj->length() == 2)) {
6833 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6834 jint false_cnt = aobj->element_value(0).as_int();
6835 jint true_cnt = aobj->element_value(1).as_int();
6836
6837 if (C->log() != NULL) {
6838 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6839 false_cnt, true_cnt);
6840 }
6841
6842 if (false_cnt + true_cnt == 0) {
6843 // According to profile, never executed.
6844 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6845 Deoptimization::Action_reinterpret);
6846 return true;
6847 }
6848
6849 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6850 // is a number of each value occurrences.
6851 Node* result = argument(0);
6852 if (false_cnt == 0 || true_cnt == 0) {
6853 // According to profile, one value has been never seen.
6854 int expected_val = (false_cnt == 0) ? 1 : 0;
6855
6856 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6857 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6858
6859 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6860 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6861 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6862
6863 { // Slow path: uncommon trap for never seen value and then reexecute
6864 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6865 // the value has been seen at least once.
6866 PreserveJVMState pjvms(this);
6867 PreserveReexecuteState preexecs(this);
6868 jvms()->set_should_reexecute(true);
6869
6870 set_control(slow_path);
6871 set_i_o(i_o());
6872
6873 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6874 Deoptimization::Action_reinterpret);
6875 }
6876 // The guard for never seen value enables sharpening of the result and
6877 // returning a constant. It allows to eliminate branches on the same value
6878 // later on.
6879 set_control(fast_path);
6880 result = intcon(expected_val);
6881 }
6882 // Stop profiling.
6883 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6884 // By replacing method body with profile data (represented as ProfileBooleanNode
6885 // on IR level) we effectively disable profiling.
6886 // It enables full speed execution once optimized code is generated.
6887 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6888 C->record_for_igvn(profile);
6889 set_result(profile);
6890 return true;
6891 } else {
6892 // Continue profiling.
6893 // Profile data isn't available at the moment. So, execute method's bytecode version.
6894 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6895 // is compiled and counters aren't available since corresponding MethodHandle
6896 // isn't a compile-time constant.
6897 return false;
6898 }
6899 }
6900
6901 bool LibraryCallKit::inline_isCompileConstant() {
6902 Node* n = argument(0);
6903 set_result(n->is_Con() ? intcon(1) : intcon(0));
6904 return true;
6905 }