1 /*
2 * Copyright (c) 1999, 2019, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/systemDictionary.hpp"
29 #include "classfile/vmSymbols.hpp"
30 #include "compiler/compileBroker.hpp"
31 #include "compiler/compileLog.hpp"
32 #include "gc/shared/barrierSet.hpp"
33 #include "jfr/support/jfrIntrinsics.hpp"
34 #include "memory/resourceArea.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/callGenerator.hpp"
40 #include "opto/castnode.hpp"
41 #include "opto/cfgnode.hpp"
42 #include "opto/convertnode.hpp"
43 #include "opto/countbitsnode.hpp"
44 #include "opto/intrinsicnode.hpp"
45 #include "opto/idealKit.hpp"
46 #include "opto/mathexactnode.hpp"
47 #include "opto/movenode.hpp"
48 #include "opto/mulnode.hpp"
49 #include "opto/narrowptrnode.hpp"
50 #include "opto/opaquenode.hpp"
51 #include "opto/parse.hpp"
52 #include "opto/runtime.hpp"
53 #include "opto/rootnode.hpp"
54 #include "opto/subnode.hpp"
55 #include "opto/valuetypenode.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
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_klass_from_mirror_common(Node* mirror, bool never_see_null,
168 RegionNode* region, int null_path,
169 int offset);
170 Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
171 RegionNode* region, int null_path) {
172 int offset = java_lang_Class::klass_offset_in_bytes();
173 return load_klass_from_mirror_common(mirror, never_see_null,
174 region, null_path,
175 offset);
176 }
177 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
178 RegionNode* region, int null_path) {
179 int offset = java_lang_Class::array_klass_offset_in_bytes();
180 return load_klass_from_mirror_common(mirror, never_see_null,
181 region, null_path,
182 offset);
183 }
184 Node* generate_access_flags_guard(Node* kls,
185 int modifier_mask, int modifier_bits,
186 RegionNode* region);
187 Node* generate_interface_guard(Node* kls, RegionNode* region);
188 Node* generate_value_guard(Node* kls, RegionNode* region);
189
190 enum ArrayKind {
191 AnyArray,
192 NonArray,
193 ObjectArray,
194 NonObjectArray,
195 TypeArray,
196 ValueArray
197 };
198
199 Node* generate_array_guard(Node* kls, RegionNode* region) {
200 return generate_array_guard_common(kls, region, AnyArray);
201 }
202 Node* generate_non_array_guard(Node* kls, RegionNode* region) {
203 return generate_array_guard_common(kls, region, NonArray);
204 }
205 Node* generate_objArray_guard(Node* kls, RegionNode* region) {
206 return generate_array_guard_common(kls, region, ObjectArray);
207 }
208 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
209 return generate_array_guard_common(kls, region, NonObjectArray);
210 }
211 Node* generate_typeArray_guard(Node* kls, RegionNode* region) {
212 return generate_array_guard_common(kls, region, TypeArray);
213 }
214 Node* generate_valueArray_guard(Node* kls, RegionNode* region) {
215 assert(ValueArrayFlatten, "can never be flattened");
216 return generate_array_guard_common(kls, region, ValueArray);
217 }
218 Node* generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind);
219 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
220 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
221 bool is_virtual = false, bool is_static = false);
222 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
223 return generate_method_call(method_id, false, true);
224 }
225 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
226 return generate_method_call(method_id, true, false);
227 }
228 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
229 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
230
231 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
232 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
233 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
234 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
235 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
236 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
237 bool inline_string_indexOfChar();
238 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
239 bool inline_string_toBytesU();
240 bool inline_string_getCharsU();
241 bool inline_string_copy(bool compress);
242 bool inline_string_char_access(bool is_store);
243 Node* round_double_node(Node* n);
244 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
245 bool inline_math_native(vmIntrinsics::ID id);
246 bool inline_math(vmIntrinsics::ID id);
247 bool inline_double_math(vmIntrinsics::ID id);
248 template <typename OverflowOp>
249 bool inline_math_overflow(Node* arg1, Node* arg2);
250 void inline_math_mathExact(Node* math, Node* test);
251 bool inline_math_addExactI(bool is_increment);
252 bool inline_math_addExactL(bool is_increment);
253 bool inline_math_multiplyExactI();
254 bool inline_math_multiplyExactL();
255 bool inline_math_multiplyHigh();
256 bool inline_math_negateExactI();
257 bool inline_math_negateExactL();
258 bool inline_math_subtractExactI(bool is_decrement);
259 bool inline_math_subtractExactL(bool is_decrement);
260 bool inline_min_max(vmIntrinsics::ID id);
261 bool inline_notify(vmIntrinsics::ID id);
262 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
263 // This returns Type::AnyPtr, RawPtr, or OopPtr.
264 int classify_unsafe_addr(Node* &base, Node* &offset, BasicType type);
265 Node* make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type = T_ILLEGAL, bool can_cast = false);
266
267 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
268 DecoratorSet mo_decorator_for_access_kind(AccessKind kind);
269 bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
270 static bool klass_needs_init_guard(Node* kls);
271 bool inline_unsafe_allocate();
272 bool inline_unsafe_newArray(bool uninitialized);
273 bool inline_unsafe_writeback0();
274 bool inline_unsafe_writebackSync0(bool is_pre);
275 bool inline_unsafe_copyMemory();
276 bool inline_unsafe_make_private_buffer();
277 bool inline_unsafe_finish_private_buffer();
278 bool inline_native_currentThread();
279
280 bool inline_native_time_funcs(address method, const char* funcName);
281 #ifdef JFR_HAVE_INTRINSICS
282 bool inline_native_classID();
283 bool inline_native_getEventWriter();
284 #endif
285 bool inline_native_isInterrupted();
286 bool inline_native_Class_query(vmIntrinsics::ID id);
287 bool inline_value_Class_conversion(vmIntrinsics::ID id);
288 bool inline_native_subtype_check();
289 bool inline_native_getLength();
290 bool inline_array_copyOf(bool is_copyOfRange);
291 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
292 bool inline_preconditions_checkIndex();
293 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array);
294 bool inline_native_clone(bool is_virtual);
295 bool inline_native_Reflection_getCallerClass();
296 // Helper function for inlining native object hash method
297 bool inline_native_hashcode(bool is_virtual, bool is_static);
298 bool inline_native_getClass();
299
300 // Helper functions for inlining arraycopy
301 bool inline_arraycopy();
302 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
303 RegionNode* slow_region);
304 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
305 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp,
306 uint new_idx);
307
308 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
309 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
310 bool inline_unsafe_fence(vmIntrinsics::ID id);
311 bool inline_onspinwait();
312 bool inline_fp_conversions(vmIntrinsics::ID id);
313 bool inline_number_methods(vmIntrinsics::ID id);
314 bool inline_reference_get();
315 bool inline_Class_cast();
316 bool inline_aescrypt_Block(vmIntrinsics::ID id);
317 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
318 bool inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id);
319 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
320 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
321 Node* inline_electronicCodeBook_AESCrypt_predicate(bool decrypting);
322 Node* inline_counterMode_AESCrypt_predicate();
323 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
324 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
325 bool inline_ghash_processBlocks();
326 bool inline_base64_encodeBlock();
327 bool inline_sha_implCompress(vmIntrinsics::ID id);
328 bool inline_digestBase_implCompressMB(int predicate);
329 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
330 bool long_state, address stubAddr, const char *stubName,
331 Node* src_start, Node* ofs, Node* limit);
332 Node* get_state_from_sha_object(Node *sha_object);
333 Node* get_state_from_sha5_object(Node *sha_object);
334 Node* inline_digestBase_implCompressMB_predicate(int predicate);
335 bool inline_encodeISOArray();
336 bool inline_updateCRC32();
337 bool inline_updateBytesCRC32();
338 bool inline_updateByteBufferCRC32();
339 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
340 bool inline_updateBytesCRC32C();
341 bool inline_updateDirectByteBufferCRC32C();
342 bool inline_updateBytesAdler32();
343 bool inline_updateByteBufferAdler32();
344 bool inline_multiplyToLen();
345 bool inline_hasNegatives();
346 bool inline_squareToLen();
347 bool inline_mulAdd();
348 bool inline_montgomeryMultiply();
349 bool inline_montgomerySquare();
350 bool inline_vectorizedMismatch();
351 bool inline_fma(vmIntrinsics::ID id);
352 bool inline_character_compare(vmIntrinsics::ID id);
353 bool inline_fp_min_max(vmIntrinsics::ID id);
354
355 bool inline_profileBoolean();
356 bool inline_isCompileConstant();
357 void clear_upper_avx() {
358 #ifdef X86
359 if (UseAVX >= 2) {
360 C->set_clear_upper_avx(true);
361 }
362 #endif
363 }
364 };
365
366 //---------------------------make_vm_intrinsic----------------------------
367 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
368 vmIntrinsics::ID id = m->intrinsic_id();
369 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
370
371 if (!m->is_loaded()) {
372 // Do not attempt to inline unloaded methods.
373 return NULL;
374 }
375
376 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
377 bool is_available = false;
378
379 {
380 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
381 // the compiler must transition to '_thread_in_vm' state because both
382 // methods access VM-internal data.
383 VM_ENTRY_MARK;
384 methodHandle mh(THREAD, m->get_Method());
385 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
386 !C->directive()->is_intrinsic_disabled(mh) &&
387 !vmIntrinsics::is_disabled_by_flags(mh);
388
389 }
390
391 if (is_available) {
392 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
393 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
394 return new LibraryIntrinsic(m, is_virtual,
395 vmIntrinsics::predicates_needed(id),
396 vmIntrinsics::does_virtual_dispatch(id),
397 (vmIntrinsics::ID) id);
398 } else {
399 return NULL;
400 }
401 }
402
403 //----------------------register_library_intrinsics-----------------------
404 // Initialize this file's data structures, for each Compile instance.
405 void Compile::register_library_intrinsics() {
406 // Nothing to do here.
407 }
408
409 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
410 LibraryCallKit kit(jvms, this);
411 Compile* C = kit.C;
412 int nodes = C->unique();
413 #ifndef PRODUCT
414 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
415 char buf[1000];
416 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
417 tty->print_cr("Intrinsic %s", str);
418 }
419 #endif
420 ciMethod* callee = kit.callee();
421 const int bci = kit.bci();
422
423 // Try to inline the intrinsic.
424 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
425 kit.try_to_inline(_last_predicate)) {
426 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
427 : "(intrinsic)";
428 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
429 if (C->print_intrinsics() || C->print_inlining()) {
430 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
431 }
432 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
433 if (C->log()) {
434 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
435 vmIntrinsics::name_at(intrinsic_id()),
436 (is_virtual() ? " virtual='1'" : ""),
437 C->unique() - nodes);
438 }
439 // Push the result from the inlined method onto the stack.
440 kit.push_result();
441 C->print_inlining_update(this);
442 return kit.transfer_exceptions_into_jvms();
443 }
444
445 // The intrinsic bailed out
446 if (jvms->has_method()) {
447 // Not a root compile.
448 const char* msg;
449 if (callee->intrinsic_candidate()) {
450 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
451 } else {
452 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
453 : "failed to inline (intrinsic), method not annotated";
454 }
455 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
456 if (C->print_intrinsics() || C->print_inlining()) {
457 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
458 }
459 } else {
460 // Root compile
461 ResourceMark rm;
462 stringStream msg_stream;
463 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
464 vmIntrinsics::name_at(intrinsic_id()),
465 is_virtual() ? " (virtual)" : "", bci);
466 const char *msg = msg_stream.as_string();
467 log_debug(jit, inlining)("%s", msg);
468 if (C->print_intrinsics() || C->print_inlining()) {
469 tty->print("%s", msg);
470 }
471 }
472 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
473 C->print_inlining_update(this);
474 return NULL;
475 }
476
477 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
478 LibraryCallKit kit(jvms, this);
479 Compile* C = kit.C;
480 int nodes = C->unique();
481 _last_predicate = predicate;
482 #ifndef PRODUCT
483 assert(is_predicated() && predicate < predicates_count(), "sanity");
484 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
485 char buf[1000];
486 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
487 tty->print_cr("Predicate for intrinsic %s", str);
488 }
489 #endif
490 ciMethod* callee = kit.callee();
491 const int bci = kit.bci();
492
493 Node* slow_ctl = kit.try_to_predicate(predicate);
494 if (!kit.failing()) {
495 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
496 : "(intrinsic, predicate)";
497 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
498 if (C->print_intrinsics() || C->print_inlining()) {
499 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
500 }
501 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
502 if (C->log()) {
503 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
504 vmIntrinsics::name_at(intrinsic_id()),
505 (is_virtual() ? " virtual='1'" : ""),
506 C->unique() - nodes);
507 }
508 return slow_ctl; // Could be NULL if the check folds.
509 }
510
511 // The intrinsic bailed out
512 if (jvms->has_method()) {
513 // Not a root compile.
514 const char* msg = "failed to generate predicate for intrinsic";
515 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
516 if (C->print_intrinsics() || C->print_inlining()) {
517 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
518 }
519 } else {
520 // Root compile
521 ResourceMark rm;
522 stringStream msg_stream;
523 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
524 vmIntrinsics::name_at(intrinsic_id()),
525 is_virtual() ? " (virtual)" : "", bci);
526 const char *msg = msg_stream.as_string();
527 log_debug(jit, inlining)("%s", msg);
528 if (C->print_intrinsics() || C->print_inlining()) {
529 C->print_inlining_stream()->print("%s", msg);
530 }
531 }
532 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
533 return NULL;
534 }
535
536 bool LibraryCallKit::try_to_inline(int predicate) {
537 // Handle symbolic names for otherwise undistinguished boolean switches:
538 const bool is_store = true;
539 const bool is_compress = true;
540 const bool is_static = true;
541 const bool is_volatile = true;
542
543 if (!jvms()->has_method()) {
544 // Root JVMState has a null method.
545 assert(map()->memory()->Opcode() == Op_Parm, "");
546 // Insert the memory aliasing node
547 set_all_memory(reset_memory());
548 }
549 assert(merged_memory(), "");
550
551
552 switch (intrinsic_id()) {
553 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
554 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
555 case vmIntrinsics::_getClass: return inline_native_getClass();
556
557 case vmIntrinsics::_ceil:
558 case vmIntrinsics::_floor:
559 case vmIntrinsics::_rint:
560 case vmIntrinsics::_dsin:
561 case vmIntrinsics::_dcos:
562 case vmIntrinsics::_dtan:
563 case vmIntrinsics::_dabs:
564 case vmIntrinsics::_fabs:
565 case vmIntrinsics::_iabs:
566 case vmIntrinsics::_labs:
567 case vmIntrinsics::_datan2:
568 case vmIntrinsics::_dsqrt:
569 case vmIntrinsics::_dexp:
570 case vmIntrinsics::_dlog:
571 case vmIntrinsics::_dlog10:
572 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id());
573
574 case vmIntrinsics::_min:
575 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
576
577 case vmIntrinsics::_notify:
578 case vmIntrinsics::_notifyAll:
579 return inline_notify(intrinsic_id());
580
581 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
582 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
583 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
584 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
585 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
586 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
587 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
588 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
589 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
590 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
591 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
592 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
593 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
594
595 case vmIntrinsics::_arraycopy: return inline_arraycopy();
596
597 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
598 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
599 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
600 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
601
602 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
603 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
604 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
605 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
606 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
607 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
608 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
609
610 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
611 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
612
613 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
614 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
615 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
616 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
617
618 case vmIntrinsics::_compressStringC:
619 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
620 case vmIntrinsics::_inflateStringC:
621 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
622
623 case vmIntrinsics::_makePrivateBuffer: return inline_unsafe_make_private_buffer();
624 case vmIntrinsics::_finishPrivateBuffer: return inline_unsafe_finish_private_buffer();
625 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
626 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
627 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
628 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
629 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
630 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
631 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
632 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
633 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
634 case vmIntrinsics::_getValue: return inline_unsafe_access(!is_store, T_VALUETYPE,Relaxed, false);
635
636 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
637 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
638 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
639 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
640 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
641 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
642 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
643 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
644 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
645 case vmIntrinsics::_putValue: return inline_unsafe_access( is_store, T_VALUETYPE,Relaxed, false);
646
647 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
648 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
649 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
650 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
651 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
652 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
653 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
654 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
655 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
656
657 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
658 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
659 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
660 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
661 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
662 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
663 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
664 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
665 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
666
667 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
668 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
669 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
670 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
671
672 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
673 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
674 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
675 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
676
677 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
678 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
679 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
680 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
681 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
682 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
683 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
684 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
685 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
686
687 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
688 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
689 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
690 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
691 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
692 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
693 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
694 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
695 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
696
697 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
698 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
699 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
700 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
701 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
702 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
703 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
704 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
705 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
706
707 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
708 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
709 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
710 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
711 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
712 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
713 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
714 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
715 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
716
717 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
718 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
719 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
720 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
721 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
722
723 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
724 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
725 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
726 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
727 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
728 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
729 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
730 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
731 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
732 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
733 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
734 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
735 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
736 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
737 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
738 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
739 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
740 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
741 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
742 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
743
744 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
745 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
746 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
747 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
748 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
749 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
750 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
751 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
752 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
753 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
754 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
755 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
756 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
757 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
758 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
759
760 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
761 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
762 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
763 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
764
765 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
766 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
767 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
768 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
769 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
770
771 case vmIntrinsics::_loadFence:
772 case vmIntrinsics::_storeFence:
773 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
774
775 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
776
777 case vmIntrinsics::_currentThread: return inline_native_currentThread();
778 case vmIntrinsics::_isInterrupted: return inline_native_isInterrupted();
779
780 #ifdef JFR_HAVE_INTRINSICS
781 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime");
782 case vmIntrinsics::_getClassId: return inline_native_classID();
783 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
784 #endif
785 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
786 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
787 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
788 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
789 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
790 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
791 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
792 case vmIntrinsics::_getLength: return inline_native_getLength();
793 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
794 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
795 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
796 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
797 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
798 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
799
800 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
801 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
802
803 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
804
805 case vmIntrinsics::_isInstance:
806 case vmIntrinsics::_getModifiers:
807 case vmIntrinsics::_isInterface:
808 case vmIntrinsics::_isArray:
809 case vmIntrinsics::_isPrimitive:
810 case vmIntrinsics::_getSuperclass:
811 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
812
813 case vmIntrinsics::_asPrimaryType:
814 case vmIntrinsics::_asIndirectType: return inline_value_Class_conversion(intrinsic_id());
815
816 case vmIntrinsics::_floatToRawIntBits:
817 case vmIntrinsics::_floatToIntBits:
818 case vmIntrinsics::_intBitsToFloat:
819 case vmIntrinsics::_doubleToRawLongBits:
820 case vmIntrinsics::_doubleToLongBits:
821 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
822
823 case vmIntrinsics::_numberOfLeadingZeros_i:
824 case vmIntrinsics::_numberOfLeadingZeros_l:
825 case vmIntrinsics::_numberOfTrailingZeros_i:
826 case vmIntrinsics::_numberOfTrailingZeros_l:
827 case vmIntrinsics::_bitCount_i:
828 case vmIntrinsics::_bitCount_l:
829 case vmIntrinsics::_reverseBytes_i:
830 case vmIntrinsics::_reverseBytes_l:
831 case vmIntrinsics::_reverseBytes_s:
832 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
833
834 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
835
836 case vmIntrinsics::_Reference_get: return inline_reference_get();
837
838 case vmIntrinsics::_Class_cast: return inline_Class_cast();
839
840 case vmIntrinsics::_aescrypt_encryptBlock:
841 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
842
843 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
844 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
845 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
846
847 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
848 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
849 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
850
851 case vmIntrinsics::_counterMode_AESCrypt:
852 return inline_counterMode_AESCrypt(intrinsic_id());
853
854 case vmIntrinsics::_sha_implCompress:
855 case vmIntrinsics::_sha2_implCompress:
856 case vmIntrinsics::_sha5_implCompress:
857 return inline_sha_implCompress(intrinsic_id());
858
859 case vmIntrinsics::_digestBase_implCompressMB:
860 return inline_digestBase_implCompressMB(predicate);
861
862 case vmIntrinsics::_multiplyToLen:
863 return inline_multiplyToLen();
864
865 case vmIntrinsics::_squareToLen:
866 return inline_squareToLen();
867
868 case vmIntrinsics::_mulAdd:
869 return inline_mulAdd();
870
871 case vmIntrinsics::_montgomeryMultiply:
872 return inline_montgomeryMultiply();
873 case vmIntrinsics::_montgomerySquare:
874 return inline_montgomerySquare();
875
876 case vmIntrinsics::_vectorizedMismatch:
877 return inline_vectorizedMismatch();
878
879 case vmIntrinsics::_ghash_processBlocks:
880 return inline_ghash_processBlocks();
881 case vmIntrinsics::_base64_encodeBlock:
882 return inline_base64_encodeBlock();
883
884 case vmIntrinsics::_encodeISOArray:
885 case vmIntrinsics::_encodeByteISOArray:
886 return inline_encodeISOArray();
887
888 case vmIntrinsics::_updateCRC32:
889 return inline_updateCRC32();
890 case vmIntrinsics::_updateBytesCRC32:
891 return inline_updateBytesCRC32();
892 case vmIntrinsics::_updateByteBufferCRC32:
893 return inline_updateByteBufferCRC32();
894
895 case vmIntrinsics::_updateBytesCRC32C:
896 return inline_updateBytesCRC32C();
897 case vmIntrinsics::_updateDirectByteBufferCRC32C:
898 return inline_updateDirectByteBufferCRC32C();
899
900 case vmIntrinsics::_updateBytesAdler32:
901 return inline_updateBytesAdler32();
902 case vmIntrinsics::_updateByteBufferAdler32:
903 return inline_updateByteBufferAdler32();
904
905 case vmIntrinsics::_profileBoolean:
906 return inline_profileBoolean();
907 case vmIntrinsics::_isCompileConstant:
908 return inline_isCompileConstant();
909
910 case vmIntrinsics::_hasNegatives:
911 return inline_hasNegatives();
912
913 case vmIntrinsics::_fmaD:
914 case vmIntrinsics::_fmaF:
915 return inline_fma(intrinsic_id());
916
917 case vmIntrinsics::_isDigit:
918 case vmIntrinsics::_isLowerCase:
919 case vmIntrinsics::_isUpperCase:
920 case vmIntrinsics::_isWhitespace:
921 return inline_character_compare(intrinsic_id());
922
923 case vmIntrinsics::_maxF:
924 case vmIntrinsics::_minF:
925 case vmIntrinsics::_maxD:
926 case vmIntrinsics::_minD:
927 return inline_fp_min_max(intrinsic_id());
928
929 default:
930 // If you get here, it may be that someone has added a new intrinsic
931 // to the list in vmSymbols.hpp without implementing it here.
932 #ifndef PRODUCT
933 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
934 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
935 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
936 }
937 #endif
938 return false;
939 }
940 }
941
942 Node* LibraryCallKit::try_to_predicate(int predicate) {
943 if (!jvms()->has_method()) {
944 // Root JVMState has a null method.
945 assert(map()->memory()->Opcode() == Op_Parm, "");
946 // Insert the memory aliasing node
947 set_all_memory(reset_memory());
948 }
949 assert(merged_memory(), "");
950
951 switch (intrinsic_id()) {
952 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
953 return inline_cipherBlockChaining_AESCrypt_predicate(false);
954 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
955 return inline_cipherBlockChaining_AESCrypt_predicate(true);
956 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
957 return inline_electronicCodeBook_AESCrypt_predicate(false);
958 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
959 return inline_electronicCodeBook_AESCrypt_predicate(true);
960 case vmIntrinsics::_counterMode_AESCrypt:
961 return inline_counterMode_AESCrypt_predicate();
962 case vmIntrinsics::_digestBase_implCompressMB:
963 return inline_digestBase_implCompressMB_predicate(predicate);
964
965 default:
966 // If you get here, it may be that someone has added a new intrinsic
967 // to the list in vmSymbols.hpp without implementing it here.
968 #ifndef PRODUCT
969 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
970 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
971 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
972 }
973 #endif
974 Node* slow_ctl = control();
975 set_control(top()); // No fast path instrinsic
976 return slow_ctl;
977 }
978 }
979
980 //------------------------------set_result-------------------------------
981 // Helper function for finishing intrinsics.
982 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
983 record_for_igvn(region);
984 set_control(_gvn.transform(region));
985 set_result( _gvn.transform(value));
986 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
987 }
988
989 //------------------------------generate_guard---------------------------
990 // Helper function for generating guarded fast-slow graph structures.
991 // The given 'test', if true, guards a slow path. If the test fails
992 // then a fast path can be taken. (We generally hope it fails.)
993 // In all cases, GraphKit::control() is updated to the fast path.
994 // The returned value represents the control for the slow path.
995 // The return value is never 'top'; it is either a valid control
996 // or NULL if it is obvious that the slow path can never be taken.
997 // Also, if region and the slow control are not NULL, the slow edge
998 // is appended to the region.
999 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
1000 if (stopped()) {
1001 // Already short circuited.
1002 return NULL;
1003 }
1004
1005 // Build an if node and its projections.
1006 // If test is true we take the slow path, which we assume is uncommon.
1007 if (_gvn.type(test) == TypeInt::ZERO) {
1008 // The slow branch is never taken. No need to build this guard.
1009 return NULL;
1010 }
1011
1012 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
1013
1014 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
1015 if (if_slow == top()) {
1016 // The slow branch is never taken. No need to build this guard.
1017 return NULL;
1018 }
1019
1020 if (region != NULL)
1021 region->add_req(if_slow);
1022
1023 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
1024 set_control(if_fast);
1025
1026 return if_slow;
1027 }
1028
1029 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
1030 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
1031 }
1032 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
1033 return generate_guard(test, region, PROB_FAIR);
1034 }
1035
1036 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
1037 Node* *pos_index) {
1038 if (stopped())
1039 return NULL; // already stopped
1040 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
1041 return NULL; // index is already adequately typed
1042 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
1043 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1044 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
1045 if (is_neg != NULL && pos_index != NULL) {
1046 // Emulate effect of Parse::adjust_map_after_if.
1047 Node* ccast = new CastIINode(index, TypeInt::POS);
1048 ccast->set_req(0, control());
1049 (*pos_index) = _gvn.transform(ccast);
1050 }
1051 return is_neg;
1052 }
1053
1054 // Make sure that 'position' is a valid limit index, in [0..length].
1055 // There are two equivalent plans for checking this:
1056 // A. (offset + copyLength) unsigned<= arrayLength
1057 // B. offset <= (arrayLength - copyLength)
1058 // We require that all of the values above, except for the sum and
1059 // difference, are already known to be non-negative.
1060 // Plan A is robust in the face of overflow, if offset and copyLength
1061 // are both hugely positive.
1062 //
1063 // Plan B is less direct and intuitive, but it does not overflow at
1064 // all, since the difference of two non-negatives is always
1065 // representable. Whenever Java methods must perform the equivalent
1066 // check they generally use Plan B instead of Plan A.
1067 // For the moment we use Plan A.
1068 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
1069 Node* subseq_length,
1070 Node* array_length,
1071 RegionNode* region) {
1072 if (stopped())
1073 return NULL; // already stopped
1074 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
1075 if (zero_offset && subseq_length->eqv_uncast(array_length))
1076 return NULL; // common case of whole-array copy
1077 Node* last = subseq_length;
1078 if (!zero_offset) // last += offset
1079 last = _gvn.transform(new AddINode(last, offset));
1080 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
1081 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1082 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
1083 return is_over;
1084 }
1085
1086 // Emit range checks for the given String.value byte array
1087 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
1088 if (stopped()) {
1089 return; // already stopped
1090 }
1091 RegionNode* bailout = new RegionNode(1);
1092 record_for_igvn(bailout);
1093 if (char_count) {
1094 // Convert char count to byte count
1095 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1096 }
1097
1098 // Offset and count must not be negative
1099 generate_negative_guard(offset, bailout);
1100 generate_negative_guard(count, bailout);
1101 // Offset + count must not exceed length of array
1102 generate_limit_guard(offset, count, load_array_length(array), bailout);
1103
1104 if (bailout->req() > 1) {
1105 PreserveJVMState pjvms(this);
1106 set_control(_gvn.transform(bailout));
1107 uncommon_trap(Deoptimization::Reason_intrinsic,
1108 Deoptimization::Action_maybe_recompile);
1109 }
1110 }
1111
1112 //--------------------------generate_current_thread--------------------
1113 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1114 ciKlass* thread_klass = env()->Thread_klass();
1115 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1116 Node* thread = _gvn.transform(new ThreadLocalNode());
1117 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1118 Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered);
1119 tls_output = thread;
1120 return threadObj;
1121 }
1122
1123
1124 //------------------------------make_string_method_node------------------------
1125 // Helper method for String intrinsic functions. This version is called with
1126 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1127 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1128 // containing the lengths of str1 and str2.
1129 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1130 Node* result = NULL;
1131 switch (opcode) {
1132 case Op_StrIndexOf:
1133 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1134 str1_start, cnt1, str2_start, cnt2, ae);
1135 break;
1136 case Op_StrComp:
1137 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1138 str1_start, cnt1, str2_start, cnt2, ae);
1139 break;
1140 case Op_StrEquals:
1141 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1142 // Use the constant length if there is one because optimized match rule may exist.
1143 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1144 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1145 break;
1146 default:
1147 ShouldNotReachHere();
1148 return NULL;
1149 }
1150
1151 // All these intrinsics have checks.
1152 C->set_has_split_ifs(true); // Has chance for split-if optimization
1153 clear_upper_avx();
1154
1155 return _gvn.transform(result);
1156 }
1157
1158 //------------------------------inline_string_compareTo------------------------
1159 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1160 Node* arg1 = argument(0);
1161 Node* arg2 = argument(1);
1162
1163 arg1 = must_be_not_null(arg1, true);
1164 arg2 = must_be_not_null(arg2, true);
1165
1166 arg1 = access_resolve(arg1, ACCESS_READ);
1167 arg2 = access_resolve(arg2, ACCESS_READ);
1168
1169 // Get start addr and length of first argument
1170 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1171 Node* arg1_cnt = load_array_length(arg1);
1172
1173 // Get start addr and length of second argument
1174 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1175 Node* arg2_cnt = load_array_length(arg2);
1176
1177 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1178 set_result(result);
1179 return true;
1180 }
1181
1182 //------------------------------inline_string_equals------------------------
1183 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1184 Node* arg1 = argument(0);
1185 Node* arg2 = argument(1);
1186
1187 // paths (plus control) merge
1188 RegionNode* region = new RegionNode(3);
1189 Node* phi = new PhiNode(region, TypeInt::BOOL);
1190
1191 if (!stopped()) {
1192
1193 arg1 = must_be_not_null(arg1, true);
1194 arg2 = must_be_not_null(arg2, true);
1195
1196 arg1 = access_resolve(arg1, ACCESS_READ);
1197 arg2 = access_resolve(arg2, ACCESS_READ);
1198
1199 // Get start addr and length of first argument
1200 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1201 Node* arg1_cnt = load_array_length(arg1);
1202
1203 // Get start addr and length of second argument
1204 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1205 Node* arg2_cnt = load_array_length(arg2);
1206
1207 // Check for arg1_cnt != arg2_cnt
1208 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1209 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1210 Node* if_ne = generate_slow_guard(bol, NULL);
1211 if (if_ne != NULL) {
1212 phi->init_req(2, intcon(0));
1213 region->init_req(2, if_ne);
1214 }
1215
1216 // Check for count == 0 is done by assembler code for StrEquals.
1217
1218 if (!stopped()) {
1219 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1220 phi->init_req(1, equals);
1221 region->init_req(1, control());
1222 }
1223 }
1224
1225 // post merge
1226 set_control(_gvn.transform(region));
1227 record_for_igvn(region);
1228
1229 set_result(_gvn.transform(phi));
1230 return true;
1231 }
1232
1233 //------------------------------inline_array_equals----------------------------
1234 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1235 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1236 Node* arg1 = argument(0);
1237 Node* arg2 = argument(1);
1238
1239 arg1 = access_resolve(arg1, ACCESS_READ);
1240 arg2 = access_resolve(arg2, ACCESS_READ);
1241
1242 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1243 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1244 clear_upper_avx();
1245
1246 return true;
1247 }
1248
1249 //------------------------------inline_hasNegatives------------------------------
1250 bool LibraryCallKit::inline_hasNegatives() {
1251 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1252 return false;
1253 }
1254
1255 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1256 // no receiver since it is static method
1257 Node* ba = argument(0);
1258 Node* offset = argument(1);
1259 Node* len = argument(2);
1260
1261 ba = must_be_not_null(ba, true);
1262
1263 // Range checks
1264 generate_string_range_check(ba, offset, len, false);
1265 if (stopped()) {
1266 return true;
1267 }
1268 ba = access_resolve(ba, ACCESS_READ);
1269 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1270 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1271 set_result(_gvn.transform(result));
1272 return true;
1273 }
1274
1275 bool LibraryCallKit::inline_preconditions_checkIndex() {
1276 Node* index = argument(0);
1277 Node* length = argument(1);
1278 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1279 return false;
1280 }
1281
1282 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1283 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1284
1285 {
1286 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1287 uncommon_trap(Deoptimization::Reason_intrinsic,
1288 Deoptimization::Action_make_not_entrant);
1289 }
1290
1291 if (stopped()) {
1292 return false;
1293 }
1294
1295 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1296 BoolTest::mask btest = BoolTest::lt;
1297 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1298 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1299 _gvn.set_type(rc, rc->Value(&_gvn));
1300 if (!rc_bool->is_Con()) {
1301 record_for_igvn(rc);
1302 }
1303 set_control(_gvn.transform(new IfTrueNode(rc)));
1304 {
1305 PreserveJVMState pjvms(this);
1306 set_control(_gvn.transform(new IfFalseNode(rc)));
1307 uncommon_trap(Deoptimization::Reason_range_check,
1308 Deoptimization::Action_make_not_entrant);
1309 }
1310
1311 if (stopped()) {
1312 return false;
1313 }
1314
1315 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1316 result->set_req(0, control());
1317 result = _gvn.transform(result);
1318 set_result(result);
1319 replace_in_map(index, result);
1320 clear_upper_avx();
1321 return true;
1322 }
1323
1324 //------------------------------inline_string_indexOf------------------------
1325 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1326 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1327 return false;
1328 }
1329 Node* src = argument(0);
1330 Node* tgt = argument(1);
1331
1332 // Make the merge point
1333 RegionNode* result_rgn = new RegionNode(4);
1334 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1335
1336 src = must_be_not_null(src, true);
1337 tgt = must_be_not_null(tgt, true);
1338
1339 src = access_resolve(src, ACCESS_READ);
1340 tgt = access_resolve(tgt, ACCESS_READ);
1341
1342 // Get start addr and length of source string
1343 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1344 Node* src_count = load_array_length(src);
1345
1346 // Get start addr and length of substring
1347 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1348 Node* tgt_count = load_array_length(tgt);
1349
1350 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1351 // Divide src size by 2 if String is UTF16 encoded
1352 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1353 }
1354 if (ae == StrIntrinsicNode::UU) {
1355 // Divide substring size by 2 if String is UTF16 encoded
1356 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1357 }
1358
1359 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1360 if (result != NULL) {
1361 result_phi->init_req(3, result);
1362 result_rgn->init_req(3, control());
1363 }
1364 set_control(_gvn.transform(result_rgn));
1365 record_for_igvn(result_rgn);
1366 set_result(_gvn.transform(result_phi));
1367
1368 return true;
1369 }
1370
1371 //-----------------------------inline_string_indexOf-----------------------
1372 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1373 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1374 return false;
1375 }
1376 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1377 return false;
1378 }
1379 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1380 Node* src = argument(0); // byte[]
1381 Node* src_count = argument(1); // char count
1382 Node* tgt = argument(2); // byte[]
1383 Node* tgt_count = argument(3); // char count
1384 Node* from_index = argument(4); // char index
1385
1386 src = must_be_not_null(src, true);
1387 tgt = must_be_not_null(tgt, true);
1388
1389 src = access_resolve(src, ACCESS_READ);
1390 tgt = access_resolve(tgt, ACCESS_READ);
1391
1392 // Multiply byte array index by 2 if String is UTF16 encoded
1393 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1394 src_count = _gvn.transform(new SubINode(src_count, from_index));
1395 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1396 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1397
1398 // Range checks
1399 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1400 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1401 if (stopped()) {
1402 return true;
1403 }
1404
1405 RegionNode* region = new RegionNode(5);
1406 Node* phi = new PhiNode(region, TypeInt::INT);
1407
1408 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1409 if (result != NULL) {
1410 // The result is index relative to from_index if substring was found, -1 otherwise.
1411 // Generate code which will fold into cmove.
1412 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1413 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1414
1415 Node* if_lt = generate_slow_guard(bol, NULL);
1416 if (if_lt != NULL) {
1417 // result == -1
1418 phi->init_req(3, result);
1419 region->init_req(3, if_lt);
1420 }
1421 if (!stopped()) {
1422 result = _gvn.transform(new AddINode(result, from_index));
1423 phi->init_req(4, result);
1424 region->init_req(4, control());
1425 }
1426 }
1427
1428 set_control(_gvn.transform(region));
1429 record_for_igvn(region);
1430 set_result(_gvn.transform(phi));
1431 clear_upper_avx();
1432
1433 return true;
1434 }
1435
1436 // Create StrIndexOfNode with fast path checks
1437 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1438 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1439 // Check for substr count > string count
1440 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1441 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1442 Node* if_gt = generate_slow_guard(bol, NULL);
1443 if (if_gt != NULL) {
1444 phi->init_req(1, intcon(-1));
1445 region->init_req(1, if_gt);
1446 }
1447 if (!stopped()) {
1448 // Check for substr count == 0
1449 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1450 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1451 Node* if_zero = generate_slow_guard(bol, NULL);
1452 if (if_zero != NULL) {
1453 phi->init_req(2, intcon(0));
1454 region->init_req(2, if_zero);
1455 }
1456 }
1457 if (!stopped()) {
1458 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1459 }
1460 return NULL;
1461 }
1462
1463 //-----------------------------inline_string_indexOfChar-----------------------
1464 bool LibraryCallKit::inline_string_indexOfChar() {
1465 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1466 return false;
1467 }
1468 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1469 return false;
1470 }
1471 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1472 Node* src = argument(0); // byte[]
1473 Node* tgt = argument(1); // tgt is int ch
1474 Node* from_index = argument(2);
1475 Node* max = argument(3);
1476
1477 src = must_be_not_null(src, true);
1478 src = access_resolve(src, ACCESS_READ);
1479
1480 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1481 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1482 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1483
1484 // Range checks
1485 generate_string_range_check(src, src_offset, src_count, true);
1486 if (stopped()) {
1487 return true;
1488 }
1489
1490 RegionNode* region = new RegionNode(3);
1491 Node* phi = new PhiNode(region, TypeInt::INT);
1492
1493 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1494 C->set_has_split_ifs(true); // Has chance for split-if optimization
1495 _gvn.transform(result);
1496
1497 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1498 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1499
1500 Node* if_lt = generate_slow_guard(bol, NULL);
1501 if (if_lt != NULL) {
1502 // result == -1
1503 phi->init_req(2, result);
1504 region->init_req(2, if_lt);
1505 }
1506 if (!stopped()) {
1507 result = _gvn.transform(new AddINode(result, from_index));
1508 phi->init_req(1, result);
1509 region->init_req(1, control());
1510 }
1511 set_control(_gvn.transform(region));
1512 record_for_igvn(region);
1513 set_result(_gvn.transform(phi));
1514
1515 return true;
1516 }
1517 //---------------------------inline_string_copy---------------------
1518 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1519 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1520 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1521 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1522 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1523 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1524 bool LibraryCallKit::inline_string_copy(bool compress) {
1525 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1526 return false;
1527 }
1528 int nargs = 5; // 2 oops, 3 ints
1529 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1530
1531 Node* src = argument(0);
1532 Node* src_offset = argument(1);
1533 Node* dst = argument(2);
1534 Node* dst_offset = argument(3);
1535 Node* length = argument(4);
1536
1537 // Check for allocation before we add nodes that would confuse
1538 // tightly_coupled_allocation()
1539 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1540
1541 // Figure out the size and type of the elements we will be copying.
1542 const Type* src_type = src->Value(&_gvn);
1543 const Type* dst_type = dst->Value(&_gvn);
1544 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1545 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1546 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1547 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1548 "Unsupported array types for inline_string_copy");
1549
1550 src = must_be_not_null(src, true);
1551 dst = must_be_not_null(dst, true);
1552
1553 // Convert char[] offsets to byte[] offsets
1554 bool convert_src = (compress && src_elem == T_BYTE);
1555 bool convert_dst = (!compress && dst_elem == T_BYTE);
1556 if (convert_src) {
1557 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1558 } else if (convert_dst) {
1559 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1560 }
1561
1562 // Range checks
1563 generate_string_range_check(src, src_offset, length, convert_src);
1564 generate_string_range_check(dst, dst_offset, length, convert_dst);
1565 if (stopped()) {
1566 return true;
1567 }
1568
1569 src = access_resolve(src, ACCESS_READ);
1570 dst = access_resolve(dst, ACCESS_WRITE);
1571
1572 Node* src_start = array_element_address(src, src_offset, src_elem);
1573 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1574 // 'src_start' points to src array + scaled offset
1575 // 'dst_start' points to dst array + scaled offset
1576 Node* count = NULL;
1577 if (compress) {
1578 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1579 } else {
1580 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1581 }
1582
1583 if (alloc != NULL) {
1584 if (alloc->maybe_set_complete(&_gvn)) {
1585 // "You break it, you buy it."
1586 InitializeNode* init = alloc->initialization();
1587 assert(init->is_complete(), "we just did this");
1588 init->set_complete_with_arraycopy();
1589 assert(dst->is_CheckCastPP(), "sanity");
1590 assert(dst->in(0)->in(0) == init, "dest pinned");
1591 }
1592 // Do not let stores that initialize this object be reordered with
1593 // a subsequent store that would make this object accessible by
1594 // other threads.
1595 // Record what AllocateNode this StoreStore protects so that
1596 // escape analysis can go from the MemBarStoreStoreNode to the
1597 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1598 // based on the escape status of the AllocateNode.
1599 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1600 }
1601 if (compress) {
1602 set_result(_gvn.transform(count));
1603 }
1604 clear_upper_avx();
1605
1606 return true;
1607 }
1608
1609 #ifdef _LP64
1610 #define XTOP ,top() /*additional argument*/
1611 #else //_LP64
1612 #define XTOP /*no additional argument*/
1613 #endif //_LP64
1614
1615 //------------------------inline_string_toBytesU--------------------------
1616 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1617 bool LibraryCallKit::inline_string_toBytesU() {
1618 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1619 return false;
1620 }
1621 // Get the arguments.
1622 Node* value = argument(0);
1623 Node* offset = argument(1);
1624 Node* length = argument(2);
1625
1626 Node* newcopy = NULL;
1627
1628 // Set the original stack and the reexecute bit for the interpreter to reexecute
1629 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1630 { PreserveReexecuteState preexecs(this);
1631 jvms()->set_should_reexecute(true);
1632
1633 // Check if a null path was taken unconditionally.
1634 value = null_check(value);
1635
1636 RegionNode* bailout = new RegionNode(1);
1637 record_for_igvn(bailout);
1638
1639 // Range checks
1640 generate_negative_guard(offset, bailout);
1641 generate_negative_guard(length, bailout);
1642 generate_limit_guard(offset, length, load_array_length(value), bailout);
1643 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1644 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1645
1646 if (bailout->req() > 1) {
1647 PreserveJVMState pjvms(this);
1648 set_control(_gvn.transform(bailout));
1649 uncommon_trap(Deoptimization::Reason_intrinsic,
1650 Deoptimization::Action_maybe_recompile);
1651 }
1652 if (stopped()) {
1653 return true;
1654 }
1655
1656 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1657 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1658 newcopy = new_array(klass_node, size, 0); // no arguments to push
1659 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1660
1661 // Calculate starting addresses.
1662 value = access_resolve(value, ACCESS_READ);
1663 Node* src_start = array_element_address(value, offset, T_CHAR);
1664 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1665
1666 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1667 const TypeInt* toffset = gvn().type(offset)->is_int();
1668 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1669
1670 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1671 const char* copyfunc_name = "arraycopy";
1672 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1673 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1674 OptoRuntime::fast_arraycopy_Type(),
1675 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1676 src_start, dst_start, ConvI2X(length) XTOP);
1677 // Do not let reads from the cloned object float above the arraycopy.
1678 if (alloc != NULL) {
1679 if (alloc->maybe_set_complete(&_gvn)) {
1680 // "You break it, you buy it."
1681 InitializeNode* init = alloc->initialization();
1682 assert(init->is_complete(), "we just did this");
1683 init->set_complete_with_arraycopy();
1684 assert(newcopy->is_CheckCastPP(), "sanity");
1685 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1686 }
1687 // Do not let stores that initialize this object be reordered with
1688 // a subsequent store that would make this object accessible by
1689 // other threads.
1690 // Record what AllocateNode this StoreStore protects so that
1691 // escape analysis can go from the MemBarStoreStoreNode to the
1692 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1693 // based on the escape status of the AllocateNode.
1694 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1695 } else {
1696 insert_mem_bar(Op_MemBarCPUOrder);
1697 }
1698 } // original reexecute is set back here
1699
1700 C->set_has_split_ifs(true); // Has chance for split-if optimization
1701 if (!stopped()) {
1702 set_result(newcopy);
1703 }
1704 clear_upper_avx();
1705
1706 return true;
1707 }
1708
1709 //------------------------inline_string_getCharsU--------------------------
1710 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1711 bool LibraryCallKit::inline_string_getCharsU() {
1712 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1713 return false;
1714 }
1715
1716 // Get the arguments.
1717 Node* src = argument(0);
1718 Node* src_begin = argument(1);
1719 Node* src_end = argument(2); // exclusive offset (i < src_end)
1720 Node* dst = argument(3);
1721 Node* dst_begin = argument(4);
1722
1723 // Check for allocation before we add nodes that would confuse
1724 // tightly_coupled_allocation()
1725 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1726
1727 // Check if a null path was taken unconditionally.
1728 src = null_check(src);
1729 dst = null_check(dst);
1730 if (stopped()) {
1731 return true;
1732 }
1733
1734 // Get length and convert char[] offset to byte[] offset
1735 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1736 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1737
1738 // Range checks
1739 generate_string_range_check(src, src_begin, length, true);
1740 generate_string_range_check(dst, dst_begin, length, false);
1741 if (stopped()) {
1742 return true;
1743 }
1744
1745 if (!stopped()) {
1746 src = access_resolve(src, ACCESS_READ);
1747 dst = access_resolve(dst, ACCESS_WRITE);
1748
1749 // Calculate starting addresses.
1750 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1751 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1752
1753 // Check if array addresses are aligned to HeapWordSize
1754 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1755 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1756 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1757 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1758
1759 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1760 const char* copyfunc_name = "arraycopy";
1761 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1762 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1763 OptoRuntime::fast_arraycopy_Type(),
1764 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1765 src_start, dst_start, ConvI2X(length) XTOP);
1766 // Do not let reads from the cloned object float above the arraycopy.
1767 if (alloc != NULL) {
1768 if (alloc->maybe_set_complete(&_gvn)) {
1769 // "You break it, you buy it."
1770 InitializeNode* init = alloc->initialization();
1771 assert(init->is_complete(), "we just did this");
1772 init->set_complete_with_arraycopy();
1773 assert(dst->is_CheckCastPP(), "sanity");
1774 assert(dst->in(0)->in(0) == init, "dest pinned");
1775 }
1776 // Do not let stores that initialize this object be reordered with
1777 // a subsequent store that would make this object accessible by
1778 // other threads.
1779 // Record what AllocateNode this StoreStore protects so that
1780 // escape analysis can go from the MemBarStoreStoreNode to the
1781 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1782 // based on the escape status of the AllocateNode.
1783 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1784 } else {
1785 insert_mem_bar(Op_MemBarCPUOrder);
1786 }
1787 }
1788
1789 C->set_has_split_ifs(true); // Has chance for split-if optimization
1790 return true;
1791 }
1792
1793 //----------------------inline_string_char_access----------------------------
1794 // Store/Load char to/from byte[] array.
1795 // static void StringUTF16.putChar(byte[] val, int index, int c)
1796 // static char StringUTF16.getChar(byte[] val, int index)
1797 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1798 Node* value = argument(0);
1799 Node* index = argument(1);
1800 Node* ch = is_store ? argument(2) : NULL;
1801
1802 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1803 // correctly requires matched array shapes.
1804 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1805 "sanity: byte[] and char[] bases agree");
1806 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1807 "sanity: byte[] and char[] scales agree");
1808
1809 // Bail when getChar over constants is requested: constant folding would
1810 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1811 // Java method would constant fold nicely instead.
1812 if (!is_store && value->is_Con() && index->is_Con()) {
1813 return false;
1814 }
1815
1816 value = must_be_not_null(value, true);
1817 value = access_resolve(value, is_store ? ACCESS_WRITE : ACCESS_READ);
1818
1819 Node* adr = array_element_address(value, index, T_CHAR);
1820 if (adr->is_top()) {
1821 return false;
1822 }
1823 if (is_store) {
1824 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1825 } else {
1826 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
1827 set_result(ch);
1828 }
1829 return true;
1830 }
1831
1832 //--------------------------round_double_node--------------------------------
1833 // Round a double node if necessary.
1834 Node* LibraryCallKit::round_double_node(Node* n) {
1835 if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1836 n = _gvn.transform(new RoundDoubleNode(0, n));
1837 return n;
1838 }
1839
1840 //------------------------------inline_math-----------------------------------
1841 // public static double Math.abs(double)
1842 // public static double Math.sqrt(double)
1843 // public static double Math.log(double)
1844 // public static double Math.log10(double)
1845 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1846 Node* arg = round_double_node(argument(0));
1847 Node* n = NULL;
1848 switch (id) {
1849 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1850 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1851 case vmIntrinsics::_ceil: n = new RoundDoubleModeNode(arg, makecon(TypeInt::make(2))); break;
1852 case vmIntrinsics::_floor: n = new RoundDoubleModeNode(arg, makecon(TypeInt::make(1))); break;
1853 case vmIntrinsics::_rint: n = new RoundDoubleModeNode(arg, makecon(TypeInt::make(0))); break;
1854 default: fatal_unexpected_iid(id); break;
1855 }
1856 set_result(_gvn.transform(n));
1857 return true;
1858 }
1859
1860 //------------------------------inline_math-----------------------------------
1861 // public static float Math.abs(float)
1862 // public static int Math.abs(int)
1863 // public static long Math.abs(long)
1864 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1865 Node* arg = argument(0);
1866 Node* n = NULL;
1867 switch (id) {
1868 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1869 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1870 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1871 default: fatal_unexpected_iid(id); break;
1872 }
1873 set_result(_gvn.transform(n));
1874 return true;
1875 }
1876
1877 //------------------------------runtime_math-----------------------------
1878 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1879 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1880 "must be (DD)D or (D)D type");
1881
1882 // Inputs
1883 Node* a = round_double_node(argument(0));
1884 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1885
1886 const TypePtr* no_memory_effects = NULL;
1887 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1888 no_memory_effects,
1889 a, top(), b, b ? top() : NULL);
1890 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1891 #ifdef ASSERT
1892 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1893 assert(value_top == top(), "second value must be top");
1894 #endif
1895
1896 set_result(value);
1897 return true;
1898 }
1899
1900 //------------------------------inline_math_native-----------------------------
1901 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1902 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1903 switch (id) {
1904 // These intrinsics are not properly supported on all hardware
1905 case vmIntrinsics::_dsin:
1906 return StubRoutines::dsin() != NULL ?
1907 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1908 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1909 case vmIntrinsics::_dcos:
1910 return StubRoutines::dcos() != NULL ?
1911 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1912 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1913 case vmIntrinsics::_dtan:
1914 return StubRoutines::dtan() != NULL ?
1915 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1916 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1917 case vmIntrinsics::_dlog:
1918 return StubRoutines::dlog() != NULL ?
1919 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1920 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1921 case vmIntrinsics::_dlog10:
1922 return StubRoutines::dlog10() != NULL ?
1923 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1924 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1925
1926 // These intrinsics are supported on all hardware
1927 case vmIntrinsics::_ceil:
1928 case vmIntrinsics::_floor:
1929 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1930 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1931 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1932 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1933 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1934 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1935
1936 case vmIntrinsics::_dexp:
1937 return StubRoutines::dexp() != NULL ?
1938 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1939 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1940 case vmIntrinsics::_dpow: {
1941 Node* exp = round_double_node(argument(2));
1942 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1943 if (d != NULL && d->getd() == 2.0) {
1944 // Special case: pow(x, 2.0) => x * x
1945 Node* base = round_double_node(argument(0));
1946 set_result(_gvn.transform(new MulDNode(base, base)));
1947 return true;
1948 }
1949 return StubRoutines::dpow() != NULL ?
1950 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1951 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1952 }
1953 #undef FN_PTR
1954
1955 // These intrinsics are not yet correctly implemented
1956 case vmIntrinsics::_datan2:
1957 return false;
1958
1959 default:
1960 fatal_unexpected_iid(id);
1961 return false;
1962 }
1963 }
1964
1965 static bool is_simple_name(Node* n) {
1966 return (n->req() == 1 // constant
1967 || (n->is_Type() && n->as_Type()->type()->singleton())
1968 || n->is_Proj() // parameter or return value
1969 || n->is_Phi() // local of some sort
1970 );
1971 }
1972
1973 //----------------------------inline_notify-----------------------------------*
1974 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1975 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1976 address func;
1977 if (id == vmIntrinsics::_notify) {
1978 func = OptoRuntime::monitor_notify_Java();
1979 } else {
1980 func = OptoRuntime::monitor_notifyAll_Java();
1981 }
1982 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1983 make_slow_call_ex(call, env()->Throwable_klass(), false);
1984 return true;
1985 }
1986
1987
1988 //----------------------------inline_min_max-----------------------------------
1989 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1990 set_result(generate_min_max(id, argument(0), argument(1)));
1991 return true;
1992 }
1993
1994 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1995 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1996 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1997 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1998 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1999
2000 {
2001 PreserveJVMState pjvms(this);
2002 PreserveReexecuteState preexecs(this);
2003 jvms()->set_should_reexecute(true);
2004
2005 set_control(slow_path);
2006 set_i_o(i_o());
2007
2008 uncommon_trap(Deoptimization::Reason_intrinsic,
2009 Deoptimization::Action_none);
2010 }
2011
2012 set_control(fast_path);
2013 set_result(math);
2014 }
2015
2016 template <typename OverflowOp>
2017 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2018 typedef typename OverflowOp::MathOp MathOp;
2019
2020 MathOp* mathOp = new MathOp(arg1, arg2);
2021 Node* operation = _gvn.transform( mathOp );
2022 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2023 inline_math_mathExact(operation, ofcheck);
2024 return true;
2025 }
2026
2027 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2028 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2029 }
2030
2031 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2032 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2033 }
2034
2035 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2036 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2037 }
2038
2039 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2040 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2041 }
2042
2043 bool LibraryCallKit::inline_math_negateExactI() {
2044 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2045 }
2046
2047 bool LibraryCallKit::inline_math_negateExactL() {
2048 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2049 }
2050
2051 bool LibraryCallKit::inline_math_multiplyExactI() {
2052 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2053 }
2054
2055 bool LibraryCallKit::inline_math_multiplyExactL() {
2056 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2057 }
2058
2059 bool LibraryCallKit::inline_math_multiplyHigh() {
2060 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2061 return true;
2062 }
2063
2064 Node*
2065 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2066 // These are the candidate return value:
2067 Node* xvalue = x0;
2068 Node* yvalue = y0;
2069
2070 if (xvalue == yvalue) {
2071 return xvalue;
2072 }
2073
2074 bool want_max = (id == vmIntrinsics::_max);
2075
2076 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
2077 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
2078 if (txvalue == NULL || tyvalue == NULL) return top();
2079 // This is not really necessary, but it is consistent with a
2080 // hypothetical MaxINode::Value method:
2081 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
2082
2083 // %%% This folding logic should (ideally) be in a different place.
2084 // Some should be inside IfNode, and there to be a more reliable
2085 // transformation of ?: style patterns into cmoves. We also want
2086 // more powerful optimizations around cmove and min/max.
2087
2088 // Try to find a dominating comparison of these guys.
2089 // It can simplify the index computation for Arrays.copyOf
2090 // and similar uses of System.arraycopy.
2091 // First, compute the normalized version of CmpI(x, y).
2092 int cmp_op = Op_CmpI;
2093 Node* xkey = xvalue;
2094 Node* ykey = yvalue;
2095 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
2096 if (ideal_cmpxy->is_Cmp()) {
2097 // E.g., if we have CmpI(length - offset, count),
2098 // it might idealize to CmpI(length, count + offset)
2099 cmp_op = ideal_cmpxy->Opcode();
2100 xkey = ideal_cmpxy->in(1);
2101 ykey = ideal_cmpxy->in(2);
2102 }
2103
2104 // Start by locating any relevant comparisons.
2105 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
2106 Node* cmpxy = NULL;
2107 Node* cmpyx = NULL;
2108 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
2109 Node* cmp = start_from->fast_out(k);
2110 if (cmp->outcnt() > 0 && // must have prior uses
2111 cmp->in(0) == NULL && // must be context-independent
2112 cmp->Opcode() == cmp_op) { // right kind of compare
2113 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
2114 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
2115 }
2116 }
2117
2118 const int NCMPS = 2;
2119 Node* cmps[NCMPS] = { cmpxy, cmpyx };
2120 int cmpn;
2121 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2122 if (cmps[cmpn] != NULL) break; // find a result
2123 }
2124 if (cmpn < NCMPS) {
2125 // Look for a dominating test that tells us the min and max.
2126 int depth = 0; // Limit search depth for speed
2127 Node* dom = control();
2128 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2129 if (++depth >= 100) break;
2130 Node* ifproj = dom;
2131 if (!ifproj->is_Proj()) continue;
2132 Node* iff = ifproj->in(0);
2133 if (!iff->is_If()) continue;
2134 Node* bol = iff->in(1);
2135 if (!bol->is_Bool()) continue;
2136 Node* cmp = bol->in(1);
2137 if (cmp == NULL) continue;
2138 for (cmpn = 0; cmpn < NCMPS; cmpn++)
2139 if (cmps[cmpn] == cmp) break;
2140 if (cmpn == NCMPS) continue;
2141 BoolTest::mask btest = bol->as_Bool()->_test._test;
2142 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
2143 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2144 // At this point, we know that 'x btest y' is true.
2145 switch (btest) {
2146 case BoolTest::eq:
2147 // They are proven equal, so we can collapse the min/max.
2148 // Either value is the answer. Choose the simpler.
2149 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2150 return yvalue;
2151 return xvalue;
2152 case BoolTest::lt: // x < y
2153 case BoolTest::le: // x <= y
2154 return (want_max ? yvalue : xvalue);
2155 case BoolTest::gt: // x > y
2156 case BoolTest::ge: // x >= y
2157 return (want_max ? xvalue : yvalue);
2158 default:
2159 break;
2160 }
2161 }
2162 }
2163
2164 // We failed to find a dominating test.
2165 // Let's pick a test that might GVN with prior tests.
2166 Node* best_bol = NULL;
2167 BoolTest::mask best_btest = BoolTest::illegal;
2168 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2169 Node* cmp = cmps[cmpn];
2170 if (cmp == NULL) continue;
2171 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2172 Node* bol = cmp->fast_out(j);
2173 if (!bol->is_Bool()) continue;
2174 BoolTest::mask btest = bol->as_Bool()->_test._test;
2175 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2176 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2177 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2178 best_bol = bol->as_Bool();
2179 best_btest = btest;
2180 }
2181 }
2182 }
2183
2184 Node* answer_if_true = NULL;
2185 Node* answer_if_false = NULL;
2186 switch (best_btest) {
2187 default:
2188 if (cmpxy == NULL)
2189 cmpxy = ideal_cmpxy;
2190 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2191 // and fall through:
2192 case BoolTest::lt: // x < y
2193 case BoolTest::le: // x <= y
2194 answer_if_true = (want_max ? yvalue : xvalue);
2195 answer_if_false = (want_max ? xvalue : yvalue);
2196 break;
2197 case BoolTest::gt: // x > y
2198 case BoolTest::ge: // x >= y
2199 answer_if_true = (want_max ? xvalue : yvalue);
2200 answer_if_false = (want_max ? yvalue : xvalue);
2201 break;
2202 }
2203
2204 jint hi, lo;
2205 if (want_max) {
2206 // We can sharpen the minimum.
2207 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2208 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2209 } else {
2210 // We can sharpen the maximum.
2211 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2212 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2213 }
2214
2215 // Use a flow-free graph structure, to avoid creating excess control edges
2216 // which could hinder other optimizations.
2217 // Since Math.min/max is often used with arraycopy, we want
2218 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2219 Node* cmov = CMoveNode::make(NULL, best_bol,
2220 answer_if_false, answer_if_true,
2221 TypeInt::make(lo, hi, widen));
2222
2223 return _gvn.transform(cmov);
2224
2225 /*
2226 // This is not as desirable as it may seem, since Min and Max
2227 // nodes do not have a full set of optimizations.
2228 // And they would interfere, anyway, with 'if' optimizations
2229 // and with CMoveI canonical forms.
2230 switch (id) {
2231 case vmIntrinsics::_min:
2232 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2233 case vmIntrinsics::_max:
2234 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2235 default:
2236 ShouldNotReachHere();
2237 }
2238 */
2239 }
2240
2241 inline int
2242 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2243 const TypePtr* base_type = TypePtr::NULL_PTR;
2244 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2245 if (base_type == NULL) {
2246 // Unknown type.
2247 return Type::AnyPtr;
2248 } else if (base_type == TypePtr::NULL_PTR) {
2249 // Since this is a NULL+long form, we have to switch to a rawptr.
2250 base = _gvn.transform(new CastX2PNode(offset));
2251 offset = MakeConX(0);
2252 return Type::RawPtr;
2253 } else if (base_type->base() == Type::RawPtr) {
2254 return Type::RawPtr;
2255 } else if (base_type->isa_oopptr()) {
2256 // Base is never null => always a heap address.
2257 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2258 return Type::OopPtr;
2259 }
2260 // Offset is small => always a heap address.
2261 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2262 if (offset_type != NULL &&
2263 base_type->offset() == 0 && // (should always be?)
2264 offset_type->_lo >= 0 &&
2265 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2266 return Type::OopPtr;
2267 } else if (type == T_OBJECT) {
2268 // off heap access to an oop doesn't make any sense. Has to be on
2269 // heap.
2270 return Type::OopPtr;
2271 }
2272 // Otherwise, it might either be oop+off or NULL+addr.
2273 return Type::AnyPtr;
2274 } else {
2275 // No information:
2276 return Type::AnyPtr;
2277 }
2278 }
2279
2280 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type, bool can_cast) {
2281 Node* uncasted_base = base;
2282 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2283 if (kind == Type::RawPtr) {
2284 return basic_plus_adr(top(), uncasted_base, offset);
2285 } else if (kind == Type::AnyPtr) {
2286 assert(base == uncasted_base, "unexpected base change");
2287 if (can_cast) {
2288 if (!_gvn.type(base)->speculative_maybe_null() &&
2289 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2290 // According to profiling, this access is always on
2291 // heap. Casting the base to not null and thus avoiding membars
2292 // around the access should allow better optimizations
2293 Node* null_ctl = top();
2294 base = null_check_oop(base, &null_ctl, true, true, true);
2295 assert(null_ctl->is_top(), "no null control here");
2296 return basic_plus_adr(base, offset);
2297 } else if (_gvn.type(base)->speculative_always_null() &&
2298 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2299 // According to profiling, this access is always off
2300 // heap.
2301 base = null_assert(base);
2302 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2303 offset = MakeConX(0);
2304 return basic_plus_adr(top(), raw_base, offset);
2305 }
2306 }
2307 // We don't know if it's an on heap or off heap access. Fall back
2308 // to raw memory access.
2309 base = access_resolve(base, decorators);
2310 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2311 return basic_plus_adr(top(), raw, offset);
2312 } else {
2313 assert(base == uncasted_base, "unexpected base change");
2314 // We know it's an on heap access so base can't be null
2315 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2316 base = must_be_not_null(base, true);
2317 }
2318 return basic_plus_adr(base, offset);
2319 }
2320 }
2321
2322 //--------------------------inline_number_methods-----------------------------
2323 // inline int Integer.numberOfLeadingZeros(int)
2324 // inline int Long.numberOfLeadingZeros(long)
2325 //
2326 // inline int Integer.numberOfTrailingZeros(int)
2327 // inline int Long.numberOfTrailingZeros(long)
2328 //
2329 // inline int Integer.bitCount(int)
2330 // inline int Long.bitCount(long)
2331 //
2332 // inline char Character.reverseBytes(char)
2333 // inline short Short.reverseBytes(short)
2334 // inline int Integer.reverseBytes(int)
2335 // inline long Long.reverseBytes(long)
2336 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2337 Node* arg = argument(0);
2338 Node* n = NULL;
2339 switch (id) {
2340 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2341 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2342 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2343 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2344 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2345 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2346 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2347 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2348 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2349 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2350 default: fatal_unexpected_iid(id); break;
2351 }
2352 set_result(_gvn.transform(n));
2353 return true;
2354 }
2355
2356 //----------------------------inline_unsafe_access----------------------------
2357
2358 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2359 // Attempt to infer a sharper value type from the offset and base type.
2360 ciKlass* sharpened_klass = NULL;
2361
2362 // See if it is an instance field, with an object type.
2363 if (alias_type->field() != NULL) {
2364 if (alias_type->field()->type()->is_klass()) {
2365 sharpened_klass = alias_type->field()->type()->as_klass();
2366 }
2367 }
2368
2369 // See if it is a narrow oop array.
2370 if (adr_type->isa_aryptr()) {
2371 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2372 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2373 if (elem_type != NULL) {
2374 sharpened_klass = elem_type->klass();
2375 }
2376 }
2377 }
2378
2379 // The sharpened class might be unloaded if there is no class loader
2380 // contraint in place.
2381 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2382 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2383
2384 #ifndef PRODUCT
2385 if (C->print_intrinsics() || C->print_inlining()) {
2386 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2387 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2388 }
2389 #endif
2390 // Sharpen the value type.
2391 return tjp;
2392 }
2393 return NULL;
2394 }
2395
2396 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2397 switch (kind) {
2398 case Relaxed:
2399 return MO_UNORDERED;
2400 case Opaque:
2401 return MO_RELAXED;
2402 case Acquire:
2403 return MO_ACQUIRE;
2404 case Release:
2405 return MO_RELEASE;
2406 case Volatile:
2407 return MO_SEQ_CST;
2408 default:
2409 ShouldNotReachHere();
2410 return 0;
2411 }
2412 }
2413
2414 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2415 if (callee()->is_static()) return false; // caller must have the capability!
2416 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2417 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2418 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2419 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2420
2421 if (is_reference_type(type)) {
2422 decorators |= ON_UNKNOWN_OOP_REF;
2423 }
2424
2425 if (unaligned) {
2426 decorators |= C2_UNALIGNED;
2427 }
2428
2429 #ifndef PRODUCT
2430 {
2431 ResourceMark rm;
2432 // Check the signatures.
2433 ciSignature* sig = callee()->signature();
2434 #ifdef ASSERT
2435 if (!is_store) {
2436 // Object getReference(Object base, int/long offset), etc.
2437 BasicType rtype = sig->return_type()->basic_type();
2438 assert(rtype == type || (rtype == T_OBJECT && type == T_VALUETYPE), "getter must return the expected value");
2439 assert(sig->count() == 2 || (type == T_VALUETYPE && sig->count() == 3), "oop getter has 2 or 3 arguments");
2440 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2441 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2442 } else {
2443 // void putReference(Object base, int/long offset, Object x), etc.
2444 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2445 assert(sig->count() == 3 || (type == T_VALUETYPE && sig->count() == 4), "oop putter has 3 arguments");
2446 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2447 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2448 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2449 assert(vtype == type || (type == T_VALUETYPE && vtype == T_OBJECT), "putter must accept the expected value");
2450 }
2451 #endif // ASSERT
2452 }
2453 #endif //PRODUCT
2454
2455 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2456
2457 Node* receiver = argument(0); // type: oop
2458
2459 // Build address expression.
2460 Node* adr;
2461 Node* heap_base_oop = top();
2462 Node* offset = top();
2463 Node* val;
2464
2465 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2466 Node* base = argument(1); // type: oop
2467 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2468 offset = argument(2); // type: long
2469 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2470 // to be plain byte offsets, which are also the same as those accepted
2471 // by oopDesc::field_addr.
2472 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2473 "fieldOffset must be byte-scaled");
2474
2475 ciValueKlass* value_klass = NULL;
2476 if (type == T_VALUETYPE) {
2477 Node* cls = null_check(argument(4));
2478 if (stopped()) {
2479 return true;
2480 }
2481 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2482 const TypeKlassPtr* kls_t = _gvn.type(kls)->isa_klassptr();
2483 if (!kls_t->klass_is_exact()) {
2484 return false;
2485 }
2486 ciKlass* klass = kls_t->klass();
2487 if (!klass->is_valuetype()) {
2488 return false;
2489 }
2490 value_klass = klass->as_value_klass();
2491 }
2492
2493 receiver = null_check(receiver);
2494 if (stopped()) {
2495 return true;
2496 }
2497
2498 if (base->is_ValueType()) {
2499 ValueTypeNode* vt = base->as_ValueType();
2500
2501 if (is_store) {
2502 if (!vt->is_allocated(&_gvn) || !_gvn.type(vt)->is_valuetype()->larval()) {
2503 return false;
2504 }
2505 base = vt->get_oop();
2506 } else {
2507 if (offset->is_Con()) {
2508 long off = find_long_con(offset, 0);
2509 ciValueKlass* vk = vt->type()->value_klass();
2510 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2511 return false;
2512 }
2513
2514 ciField* f = vk->get_non_flattened_field_by_offset((int)off);
2515
2516 if (f != NULL) {
2517 BasicType bt = f->layout_type();
2518 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2519 bt = T_OBJECT;
2520 }
2521 if (bt == type) {
2522 if (bt != T_VALUETYPE || f->type() == value_klass) {
2523 set_result(vt->field_value_by_offset((int)off, false));
2524 return true;
2525 }
2526 }
2527 }
2528 }
2529 vt = vt->allocate(this)->as_ValueType();
2530 base = vt->get_oop();
2531 }
2532 }
2533
2534 // 32-bit machines ignore the high half!
2535 offset = ConvL2X(offset);
2536 adr = make_unsafe_address(base, offset, is_store ? ACCESS_WRITE : ACCESS_READ, type, kind == Relaxed);
2537
2538 if (_gvn.type(base)->isa_ptr() != TypePtr::NULL_PTR) {
2539 heap_base_oop = base;
2540 } else if (type == T_OBJECT || (value_klass != NULL && value_klass->has_object_fields())) {
2541 return false; // off-heap oop accesses are not supported
2542 }
2543
2544 // Can base be NULL? Otherwise, always on-heap access.
2545 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2546
2547 if (!can_access_non_heap) {
2548 decorators |= IN_HEAP;
2549 }
2550
2551 val = is_store ? argument(4 + (type == T_VALUETYPE ? 1 : 0)) : NULL;
2552
2553 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2554 if (adr_type == TypePtr::NULL_PTR) {
2555 return false; // off-heap access with zero address
2556 }
2557
2558 // Try to categorize the address.
2559 Compile::AliasType* alias_type = C->alias_type(adr_type);
2560 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2561
2562 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2563 alias_type->adr_type() == TypeAryPtr::RANGE) {
2564 return false; // not supported
2565 }
2566
2567 bool mismatched = false;
2568 BasicType bt = T_ILLEGAL;
2569 ciField* field = NULL;
2570 if (adr_type->isa_instptr()) {
2571 const TypeInstPtr* instptr = adr_type->is_instptr();
2572 ciInstanceKlass* k = instptr->klass()->as_instance_klass();
2573 int off = instptr->offset();
2574 if (instptr->const_oop() != NULL &&
2575 instptr->klass() == ciEnv::current()->Class_klass() &&
2576 instptr->offset() >= (instptr->klass()->as_instance_klass()->size_helper() * wordSize)) {
2577 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2578 field = k->get_field_by_offset(off, true);
2579 } else {
2580 field = k->get_non_flattened_field_by_offset(off);
2581 }
2582 if (field != NULL) {
2583 bt = field->layout_type();
2584 }
2585 assert(bt == alias_type->basic_type() || bt == T_VALUETYPE, "should match");
2586 if (field != NULL && bt == T_VALUETYPE && !field->is_flattened()) {
2587 bt = T_OBJECT;
2588 }
2589 } else {
2590 bt = alias_type->basic_type();
2591 }
2592
2593 if (bt != T_ILLEGAL) {
2594 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2595 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2596 // Alias type doesn't differentiate between byte[] and boolean[]).
2597 // Use address type to get the element type.
2598 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2599 }
2600 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2601 // accessing an array field with getReference is not a mismatch
2602 bt = T_OBJECT;
2603 }
2604 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2605 // Don't intrinsify mismatched object accesses
2606 return false;
2607 }
2608 mismatched = (bt != type);
2609 } else if (alias_type->adr_type()->isa_oopptr()) {
2610 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2611 }
2612
2613 if (type == T_VALUETYPE) {
2614 if (adr_type->isa_instptr()) {
2615 if (field == NULL || field->type() != value_klass) {
2616 mismatched = true;
2617 }
2618 } else if (adr_type->isa_aryptr()) {
2619 const Type* elem = adr_type->is_aryptr()->elem();
2620 if (!elem->isa_valuetype()) {
2621 mismatched = true;
2622 } else if (elem->value_klass() != value_klass) {
2623 mismatched = true;
2624 }
2625 }
2626 if (is_store) {
2627 const Type* val_t = _gvn.type(val);
2628 if (!val_t->isa_valuetype() || val_t->value_klass() != value_klass) {
2629 return false;
2630 }
2631 }
2632 }
2633
2634 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2635
2636 if (mismatched) {
2637 decorators |= C2_MISMATCHED;
2638 }
2639
2640 // First guess at the value type.
2641 const Type *value_type = Type::get_const_basic_type(type);
2642
2643 // Figure out the memory ordering.
2644 decorators |= mo_decorator_for_access_kind(kind);
2645
2646 if (!is_store) {
2647 if (type == T_OBJECT) {
2648 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2649 if (tjp != NULL) {
2650 value_type = tjp;
2651 }
2652 } else if (type == T_VALUETYPE) {
2653 value_type = NULL;
2654 }
2655 }
2656
2657 // Heap pointers get a null-check from the interpreter,
2658 // as a courtesy. However, this is not guaranteed by Unsafe,
2659 // and it is not possible to fully distinguish unintended nulls
2660 // from intended ones in this API.
2661
2662 if (!is_store) {
2663 Node* p = NULL;
2664 // Try to constant fold a load from a constant field
2665
2666 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2667 // final or stable field
2668 p = make_constant_from_field(field, heap_base_oop);
2669 }
2670
2671 if (p == NULL) { // Could not constant fold the load
2672 if (type == T_VALUETYPE) {
2673 if (adr_type->isa_instptr() && !mismatched) {
2674 ciInstanceKlass* holder = adr_type->is_instptr()->klass()->as_instance_klass();
2675 int offset = adr_type->is_instptr()->offset();
2676 p = ValueTypeNode::make_from_flattened(this, value_klass, base, base, holder, offset, decorators);
2677 } else {
2678 p = ValueTypeNode::make_from_flattened(this, value_klass, base, adr, NULL, 0, decorators);
2679 }
2680 } else {
2681 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2682 }
2683 // Normalize the value returned by getBoolean in the following cases
2684 if (type == T_BOOLEAN &&
2685 (mismatched ||
2686 heap_base_oop == top() || // - heap_base_oop is NULL or
2687 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2688 // and the unsafe access is made to large offset
2689 // (i.e., larger than the maximum offset necessary for any
2690 // field access)
2691 ) {
2692 IdealKit ideal = IdealKit(this);
2693 #define __ ideal.
2694 IdealVariable normalized_result(ideal);
2695 __ declarations_done();
2696 __ set(normalized_result, p);
2697 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2698 __ set(normalized_result, ideal.ConI(1));
2699 ideal.end_if();
2700 final_sync(ideal);
2701 p = __ value(normalized_result);
2702 #undef __
2703 }
2704 }
2705 if (type == T_ADDRESS) {
2706 p = gvn().transform(new CastP2XNode(NULL, p));
2707 p = ConvX2UL(p);
2708 }
2709 if (field != NULL && field->is_flattenable() && !field->is_flattened()) {
2710 // Load a non-flattened but flattenable value type from memory
2711 if (value_type->value_klass()->is_scalarizable()) {
2712 p = ValueTypeNode::make_from_oop(this, p, value_type->value_klass());
2713 } else {
2714 p = null2default(p, value_type->value_klass());
2715 }
2716 }
2717 // The load node has the control of the preceding MemBarCPUOrder. All
2718 // following nodes will have the control of the MemBarCPUOrder inserted at
2719 // the end of this method. So, pushing the load onto the stack at a later
2720 // point is fine.
2721 set_result(p);
2722 } else {
2723 if (bt == T_ADDRESS) {
2724 // Repackage the long as a pointer.
2725 val = ConvL2X(val);
2726 val = gvn().transform(new CastX2PNode(val));
2727 }
2728 if (type == T_VALUETYPE) {
2729 if (adr_type->isa_instptr() && !mismatched) {
2730 ciInstanceKlass* holder = adr_type->is_instptr()->klass()->as_instance_klass();
2731 int offset = adr_type->is_instptr()->offset();
2732 val->as_ValueType()->store_flattened(this, base, base, holder, offset, decorators);
2733 } else {
2734 val->as_ValueType()->store_flattened(this, base, adr, NULL, 0, decorators);
2735 }
2736 } else {
2737 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2738 }
2739 }
2740
2741 if (argument(1)->is_ValueType() && is_store) {
2742 Node* value = ValueTypeNode::make_from_oop(this, base, _gvn.type(base)->value_klass());
2743 value = value->as_ValueType()->make_larval(this, false);
2744 replace_in_map(argument(1), value);
2745 }
2746
2747 return true;
2748 }
2749
2750 bool LibraryCallKit::inline_unsafe_make_private_buffer() {
2751 Node* receiver = argument(0);
2752 Node* value = argument(1);
2753
2754 receiver = null_check(receiver);
2755 if (stopped()) {
2756 return true;
2757 }
2758
2759 if (!value->is_ValueType()) {
2760 return false;
2761 }
2762
2763 set_result(value->as_ValueType()->make_larval(this, true));
2764
2765 return true;
2766 }
2767
2768 bool LibraryCallKit::inline_unsafe_finish_private_buffer() {
2769 Node* receiver = argument(0);
2770 Node* buffer = argument(1);
2771
2772 receiver = null_check(receiver);
2773 if (stopped()) {
2774 return true;
2775 }
2776
2777 if (!buffer->is_ValueType()) {
2778 return false;
2779 }
2780
2781 ValueTypeNode* vt = buffer->as_ValueType();
2782 if (!vt->is_allocated(&_gvn) || !_gvn.type(vt)->is_valuetype()->larval()) {
2783 return false;
2784 }
2785
2786 set_result(vt->finish_larval(this));
2787
2788 return true;
2789 }
2790
2791 //----------------------------inline_unsafe_load_store----------------------------
2792 // This method serves a couple of different customers (depending on LoadStoreKind):
2793 //
2794 // LS_cmp_swap:
2795 //
2796 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2797 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2798 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2799 //
2800 // LS_cmp_swap_weak:
2801 //
2802 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2803 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2804 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2805 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2806 //
2807 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2808 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2809 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2810 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2811 //
2812 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2813 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2814 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2815 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2816 //
2817 // LS_cmp_exchange:
2818 //
2819 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2820 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2821 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2822 //
2823 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2824 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2825 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2826 //
2827 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2828 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2829 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2830 //
2831 // LS_get_add:
2832 //
2833 // int getAndAddInt( Object o, long offset, int delta)
2834 // long getAndAddLong(Object o, long offset, long delta)
2835 //
2836 // LS_get_set:
2837 //
2838 // int getAndSet(Object o, long offset, int newValue)
2839 // long getAndSet(Object o, long offset, long newValue)
2840 // Object getAndSet(Object o, long offset, Object newValue)
2841 //
2842 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2843 // This basic scheme here is the same as inline_unsafe_access, but
2844 // differs in enough details that combining them would make the code
2845 // overly confusing. (This is a true fact! I originally combined
2846 // them, but even I was confused by it!) As much code/comments as
2847 // possible are retained from inline_unsafe_access though to make
2848 // the correspondences clearer. - dl
2849
2850 if (callee()->is_static()) return false; // caller must have the capability!
2851
2852 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2853 decorators |= mo_decorator_for_access_kind(access_kind);
2854
2855 #ifndef PRODUCT
2856 BasicType rtype;
2857 {
2858 ResourceMark rm;
2859 // Check the signatures.
2860 ciSignature* sig = callee()->signature();
2861 rtype = sig->return_type()->basic_type();
2862 switch(kind) {
2863 case LS_get_add:
2864 case LS_get_set: {
2865 // Check the signatures.
2866 #ifdef ASSERT
2867 assert(rtype == type, "get and set must return the expected type");
2868 assert(sig->count() == 3, "get and set has 3 arguments");
2869 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2870 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2871 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2872 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2873 #endif // ASSERT
2874 break;
2875 }
2876 case LS_cmp_swap:
2877 case LS_cmp_swap_weak: {
2878 // Check the signatures.
2879 #ifdef ASSERT
2880 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2881 assert(sig->count() == 4, "CAS has 4 arguments");
2882 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2883 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2884 #endif // ASSERT
2885 break;
2886 }
2887 case LS_cmp_exchange: {
2888 // Check the signatures.
2889 #ifdef ASSERT
2890 assert(rtype == type, "CAS must return the expected type");
2891 assert(sig->count() == 4, "CAS has 4 arguments");
2892 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2893 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2894 #endif // ASSERT
2895 break;
2896 }
2897 default:
2898 ShouldNotReachHere();
2899 }
2900 }
2901 #endif //PRODUCT
2902
2903 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2904
2905 // Get arguments:
2906 Node* receiver = NULL;
2907 Node* base = NULL;
2908 Node* offset = NULL;
2909 Node* oldval = NULL;
2910 Node* newval = NULL;
2911 switch(kind) {
2912 case LS_cmp_swap:
2913 case LS_cmp_swap_weak:
2914 case LS_cmp_exchange: {
2915 const bool two_slot_type = type2size[type] == 2;
2916 receiver = argument(0); // type: oop
2917 base = argument(1); // type: oop
2918 offset = argument(2); // type: long
2919 oldval = argument(4); // type: oop, int, or long
2920 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2921 break;
2922 }
2923 case LS_get_add:
2924 case LS_get_set: {
2925 receiver = argument(0); // type: oop
2926 base = argument(1); // type: oop
2927 offset = argument(2); // type: long
2928 oldval = NULL;
2929 newval = argument(4); // type: oop, int, or long
2930 break;
2931 }
2932 default:
2933 ShouldNotReachHere();
2934 }
2935
2936 // Build field offset expression.
2937 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2938 // to be plain byte offsets, which are also the same as those accepted
2939 // by oopDesc::field_addr.
2940 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2941 // 32-bit machines ignore the high half of long offsets
2942 offset = ConvL2X(offset);
2943 Node* adr = make_unsafe_address(base, offset, ACCESS_WRITE | ACCESS_READ, type, false);
2944 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2945
2946 Compile::AliasType* alias_type = C->alias_type(adr_type);
2947 BasicType bt = alias_type->basic_type();
2948 if (bt != T_ILLEGAL &&
2949 (is_reference_type(bt) != (type == T_OBJECT))) {
2950 // Don't intrinsify mismatched object accesses.
2951 return false;
2952 }
2953
2954 // For CAS, unlike inline_unsafe_access, there seems no point in
2955 // trying to refine types. Just use the coarse types here.
2956 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2957 const Type *value_type = Type::get_const_basic_type(type);
2958
2959 switch (kind) {
2960 case LS_get_set:
2961 case LS_cmp_exchange: {
2962 if (type == T_OBJECT) {
2963 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2964 if (tjp != NULL) {
2965 value_type = tjp;
2966 }
2967 }
2968 break;
2969 }
2970 case LS_cmp_swap:
2971 case LS_cmp_swap_weak:
2972 case LS_get_add:
2973 break;
2974 default:
2975 ShouldNotReachHere();
2976 }
2977
2978 // Null check receiver.
2979 receiver = null_check(receiver);
2980 if (stopped()) {
2981 return true;
2982 }
2983
2984 int alias_idx = C->get_alias_index(adr_type);
2985
2986 if (is_reference_type(type)) {
2987 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2988
2989 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2990 // could be delayed during Parse (for example, in adjust_map_after_if()).
2991 // Execute transformation here to avoid barrier generation in such case.
2992 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2993 newval = _gvn.makecon(TypePtr::NULL_PTR);
2994
2995 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2996 // Refine the value to a null constant, when it is known to be null
2997 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2998 }
2999 }
3000
3001 Node* result = NULL;
3002 switch (kind) {
3003 case LS_cmp_exchange: {
3004 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
3005 oldval, newval, value_type, type, decorators);
3006 break;
3007 }
3008 case LS_cmp_swap_weak:
3009 decorators |= C2_WEAK_CMPXCHG;
3010 case LS_cmp_swap: {
3011 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
3012 oldval, newval, value_type, type, decorators);
3013 break;
3014 }
3015 case LS_get_set: {
3016 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
3017 newval, value_type, type, decorators);
3018 break;
3019 }
3020 case LS_get_add: {
3021 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
3022 newval, value_type, type, decorators);
3023 break;
3024 }
3025 default:
3026 ShouldNotReachHere();
3027 }
3028
3029 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3030 set_result(result);
3031 return true;
3032 }
3033
3034 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3035 // Regardless of form, don't allow previous ld/st to move down,
3036 // then issue acquire, release, or volatile mem_bar.
3037 insert_mem_bar(Op_MemBarCPUOrder);
3038 switch(id) {
3039 case vmIntrinsics::_loadFence:
3040 insert_mem_bar(Op_LoadFence);
3041 return true;
3042 case vmIntrinsics::_storeFence:
3043 insert_mem_bar(Op_StoreFence);
3044 return true;
3045 case vmIntrinsics::_fullFence:
3046 insert_mem_bar(Op_MemBarVolatile);
3047 return true;
3048 default:
3049 fatal_unexpected_iid(id);
3050 return false;
3051 }
3052 }
3053
3054 bool LibraryCallKit::inline_onspinwait() {
3055 insert_mem_bar(Op_OnSpinWait);
3056 return true;
3057 }
3058
3059 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3060 if (!kls->is_Con()) {
3061 return true;
3062 }
3063 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
3064 if (klsptr == NULL) {
3065 return true;
3066 }
3067 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
3068 // don't need a guard for a klass that is already initialized
3069 return !ik->is_initialized();
3070 }
3071
3072 //----------------------------inline_unsafe_writeback0-------------------------
3073 // public native void Unsafe.writeback0(long address)
3074 bool LibraryCallKit::inline_unsafe_writeback0() {
3075 if (!Matcher::has_match_rule(Op_CacheWB)) {
3076 return false;
3077 }
3078 #ifndef PRODUCT
3079 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
3080 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
3081 ciSignature* sig = callee()->signature();
3082 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
3083 #endif
3084 null_check_receiver(); // null-check, then ignore
3085 Node *addr = argument(1);
3086 addr = new CastX2PNode(addr);
3087 addr = _gvn.transform(addr);
3088 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
3089 flush = _gvn.transform(flush);
3090 set_memory(flush, TypeRawPtr::BOTTOM);
3091 return true;
3092 }
3093
3094 //----------------------------inline_unsafe_writeback0-------------------------
3095 // public native void Unsafe.writeback0(long address)
3096 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
3097 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
3098 return false;
3099 }
3100 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
3101 return false;
3102 }
3103 #ifndef PRODUCT
3104 assert(Matcher::has_match_rule(Op_CacheWB),
3105 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
3106 : "found match rule for CacheWBPostSync but not CacheWB"));
3107
3108 #endif
3109 null_check_receiver(); // null-check, then ignore
3110 Node *sync;
3111 if (is_pre) {
3112 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3113 } else {
3114 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3115 }
3116 sync = _gvn.transform(sync);
3117 set_memory(sync, TypeRawPtr::BOTTOM);
3118 return true;
3119 }
3120
3121 //----------------------------inline_unsafe_allocate---------------------------
3122 // public native Object Unsafe.allocateInstance(Class<?> cls);
3123 bool LibraryCallKit::inline_unsafe_allocate() {
3124 if (callee()->is_static()) return false; // caller must have the capability!
3125
3126 null_check_receiver(); // null-check, then ignore
3127 Node* cls = null_check(argument(1));
3128 if (stopped()) return true;
3129
3130 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3131 kls = null_check(kls);
3132 if (stopped()) return true; // argument was like int.class
3133
3134 Node* test = NULL;
3135 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3136 // Note: The argument might still be an illegal value like
3137 // Serializable.class or Object[].class. The runtime will handle it.
3138 // But we must make an explicit check for initialization.
3139 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3140 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3141 // can generate code to load it as unsigned byte.
3142 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3143 Node* bits = intcon(InstanceKlass::fully_initialized);
3144 test = _gvn.transform(new SubINode(inst, bits));
3145 // The 'test' is non-zero if we need to take a slow path.
3146 }
3147
3148 Node* obj = new_instance(kls, test);
3149 set_result(obj);
3150 return true;
3151 }
3152
3153 //------------------------inline_native_time_funcs--------------
3154 // inline code for System.currentTimeMillis() and System.nanoTime()
3155 // these have the same type and signature
3156 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3157 const TypeFunc* tf = OptoRuntime::void_long_Type();
3158 const TypePtr* no_memory_effects = NULL;
3159 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3160 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3161 #ifdef ASSERT
3162 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3163 assert(value_top == top(), "second value must be top");
3164 #endif
3165 set_result(value);
3166 return true;
3167 }
3168
3169 #ifdef JFR_HAVE_INTRINSICS
3170
3171 /*
3172 * oop -> myklass
3173 * myklass->trace_id |= USED
3174 * return myklass->trace_id & ~0x3
3175 */
3176 bool LibraryCallKit::inline_native_classID() {
3177 Node* cls = null_check(argument(0), T_OBJECT);
3178 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3179 kls = null_check(kls, T_OBJECT);
3180
3181 ByteSize offset = KLASS_TRACE_ID_OFFSET;
3182 Node* insp = basic_plus_adr(kls, in_bytes(offset));
3183 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
3184
3185 Node* clsused = longcon(0x01l); // set the class bit
3186 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
3187 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
3188 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
3189
3190 #ifdef TRACE_ID_META_BITS
3191 Node* mbits = longcon(~TRACE_ID_META_BITS);
3192 tvalue = _gvn.transform(new AndLNode(tvalue, mbits));
3193 #endif
3194 #ifdef TRACE_ID_SHIFT
3195 Node* cbits = intcon(TRACE_ID_SHIFT);
3196 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits));
3197 #endif
3198
3199 set_result(tvalue);
3200 return true;
3201
3202 }
3203
3204 bool LibraryCallKit::inline_native_getEventWriter() {
3205 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3206
3207 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
3208 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3209
3210 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3211
3212 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
3213 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
3214
3215 IfNode* iff_jobj_null =
3216 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
3217
3218 enum { _normal_path = 1,
3219 _null_path = 2,
3220 PATH_LIMIT };
3221
3222 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3223 PhiNode* result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3224
3225 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
3226 result_rgn->init_req(_null_path, jobj_is_null);
3227 result_val->init_req(_null_path, null());
3228
3229 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
3230 set_control(jobj_is_not_null);
3231 Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT,
3232 IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3233 result_rgn->init_req(_normal_path, control());
3234 result_val->init_req(_normal_path, res);
3235
3236 set_result(result_rgn, result_val);
3237
3238 return true;
3239 }
3240
3241 #endif // JFR_HAVE_INTRINSICS
3242
3243 //------------------------inline_native_currentThread------------------
3244 bool LibraryCallKit::inline_native_currentThread() {
3245 Node* junk = NULL;
3246 set_result(generate_current_thread(junk));
3247 return true;
3248 }
3249
3250 //------------------------inline_native_isInterrupted------------------
3251 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
3252 bool LibraryCallKit::inline_native_isInterrupted() {
3253 // Add a fast path to t.isInterrupted(clear_int):
3254 // (t == Thread.current() &&
3255 // (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
3256 // ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
3257 // So, in the common case that the interrupt bit is false,
3258 // we avoid making a call into the VM. Even if the interrupt bit
3259 // is true, if the clear_int argument is false, we avoid the VM call.
3260 // However, if the receiver is not currentThread, we must call the VM,
3261 // because there must be some locking done around the operation.
3262
3263 // We only go to the fast case code if we pass two guards.
3264 // Paths which do not pass are accumulated in the slow_region.
3265
3266 enum {
3267 no_int_result_path = 1, // t == Thread.current() && !TLS._osthread._interrupted
3268 no_clear_result_path = 2, // t == Thread.current() && TLS._osthread._interrupted && !clear_int
3269 slow_result_path = 3, // slow path: t.isInterrupted(clear_int)
3270 PATH_LIMIT
3271 };
3272
3273 // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
3274 // out of the function.
3275 insert_mem_bar(Op_MemBarCPUOrder);
3276
3277 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3278 PhiNode* result_val = new PhiNode(result_rgn, TypeInt::BOOL);
3279
3280 RegionNode* slow_region = new RegionNode(1);
3281 record_for_igvn(slow_region);
3282
3283 // (a) Receiving thread must be the current thread.
3284 Node* rec_thr = argument(0);
3285 Node* tls_ptr = NULL;
3286 Node* cur_thr = generate_current_thread(tls_ptr);
3287
3288 // Resolve oops to stable for CmpP below.
3289 cur_thr = access_resolve(cur_thr, 0);
3290 rec_thr = access_resolve(rec_thr, 0);
3291
3292 Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
3293 Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
3294
3295 generate_slow_guard(bol_thr, slow_region);
3296
3297 // (b) Interrupt bit on TLS must be false.
3298 Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
3299 Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3300 p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
3301
3302 // Set the control input on the field _interrupted read to prevent it floating up.
3303 Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
3304 Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
3305 Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
3306
3307 IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
3308
3309 // First fast path: if (!TLS._interrupted) return false;
3310 Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
3311 result_rgn->init_req(no_int_result_path, false_bit);
3312 result_val->init_req(no_int_result_path, intcon(0));
3313
3314 // drop through to next case
3315 set_control( _gvn.transform(new IfTrueNode(iff_bit)));
3316
3317 #ifndef _WINDOWS
3318 // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
3319 Node* clr_arg = argument(1);
3320 Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
3321 Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
3322 IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
3323
3324 // Second fast path: ... else if (!clear_int) return true;
3325 Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
3326 result_rgn->init_req(no_clear_result_path, false_arg);
3327 result_val->init_req(no_clear_result_path, intcon(1));
3328
3329 // drop through to next case
3330 set_control( _gvn.transform(new IfTrueNode(iff_arg)));
3331 #else
3332 // To return true on Windows you must read the _interrupted field
3333 // and check the event state i.e. take the slow path.
3334 #endif // _WINDOWS
3335
3336 // (d) Otherwise, go to the slow path.
3337 slow_region->add_req(control());
3338 set_control( _gvn.transform(slow_region));
3339
3340 if (stopped()) {
3341 // There is no slow path.
3342 result_rgn->init_req(slow_result_path, top());
3343 result_val->init_req(slow_result_path, top());
3344 } else {
3345 // non-virtual because it is a private non-static
3346 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
3347
3348 Node* slow_val = set_results_for_java_call(slow_call);
3349 // this->control() comes from set_results_for_java_call
3350
3351 Node* fast_io = slow_call->in(TypeFunc::I_O);
3352 Node* fast_mem = slow_call->in(TypeFunc::Memory);
3353
3354 // These two phis are pre-filled with copies of of the fast IO and Memory
3355 PhiNode* result_mem = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
3356 PhiNode* result_io = PhiNode::make(result_rgn, fast_io, Type::ABIO);
3357
3358 result_rgn->init_req(slow_result_path, control());
3359 result_io ->init_req(slow_result_path, i_o());
3360 result_mem->init_req(slow_result_path, reset_memory());
3361 result_val->init_req(slow_result_path, slow_val);
3362
3363 set_all_memory(_gvn.transform(result_mem));
3364 set_i_o( _gvn.transform(result_io));
3365 }
3366
3367 C->set_has_split_ifs(true); // Has chance for split-if optimization
3368 set_result(result_rgn, result_val);
3369 return true;
3370 }
3371
3372 //-----------------------load_klass_from_mirror_common-------------------------
3373 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3374 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3375 // and branch to the given path on the region.
3376 // If never_see_null, take an uncommon trap on null, so we can optimistically
3377 // compile for the non-null case.
3378 // If the region is NULL, force never_see_null = true.
3379 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3380 bool never_see_null,
3381 RegionNode* region,
3382 int null_path,
3383 int offset) {
3384 if (region == NULL) never_see_null = true;
3385 Node* p = basic_plus_adr(mirror, offset);
3386 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3387 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3388 Node* null_ctl = top();
3389 kls = null_check_oop(kls, &null_ctl, never_see_null);
3390 if (region != NULL) {
3391 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3392 region->init_req(null_path, null_ctl);
3393 } else {
3394 assert(null_ctl == top(), "no loose ends");
3395 }
3396 return kls;
3397 }
3398
3399 //--------------------(inline_native_Class_query helpers)---------------------
3400 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3401 // Fall through if (mods & mask) == bits, take the guard otherwise.
3402 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3403 // Branch around if the given klass has the given modifier bit set.
3404 // Like generate_guard, adds a new path onto the region.
3405 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3406 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3407 Node* mask = intcon(modifier_mask);
3408 Node* bits = intcon(modifier_bits);
3409 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3410 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3411 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3412 return generate_fair_guard(bol, region);
3413 }
3414 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3415 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3416 }
3417
3418 Node* LibraryCallKit::generate_value_guard(Node* kls, RegionNode* region) {
3419 return generate_access_flags_guard(kls, JVM_ACC_VALUE, 0, region);
3420 }
3421
3422 //-------------------------inline_native_Class_query-------------------
3423 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3424 const Type* return_type = TypeInt::BOOL;
3425 Node* prim_return_value = top(); // what happens if it's a primitive class?
3426 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3427 bool expect_prim = false; // most of these guys expect to work on refs
3428
3429 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3430
3431 Node* mirror = argument(0);
3432 Node* obj = top();
3433
3434 switch (id) {
3435 case vmIntrinsics::_isInstance:
3436 // nothing is an instance of a primitive type
3437 prim_return_value = intcon(0);
3438 obj = argument(1);
3439 break;
3440 case vmIntrinsics::_getModifiers:
3441 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3442 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3443 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3444 break;
3445 case vmIntrinsics::_isInterface:
3446 prim_return_value = intcon(0);
3447 break;
3448 case vmIntrinsics::_isArray:
3449 prim_return_value = intcon(0);
3450 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3451 break;
3452 case vmIntrinsics::_isPrimitive:
3453 prim_return_value = intcon(1);
3454 expect_prim = true; // obviously
3455 break;
3456 case vmIntrinsics::_getSuperclass:
3457 prim_return_value = null();
3458 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3459 break;
3460 case vmIntrinsics::_getClassAccessFlags:
3461 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3462 return_type = TypeInt::INT; // not bool! 6297094
3463 break;
3464 default:
3465 fatal_unexpected_iid(id);
3466 break;
3467 }
3468
3469 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3470 if (mirror_con == NULL) return false; // cannot happen?
3471
3472 #ifndef PRODUCT
3473 if (C->print_intrinsics() || C->print_inlining()) {
3474 ciType* k = mirror_con->java_mirror_type();
3475 if (k) {
3476 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3477 k->print_name();
3478 tty->cr();
3479 }
3480 }
3481 #endif
3482
3483 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3484 RegionNode* region = new RegionNode(PATH_LIMIT);
3485 record_for_igvn(region);
3486 PhiNode* phi = new PhiNode(region, return_type);
3487
3488 // The mirror will never be null of Reflection.getClassAccessFlags, however
3489 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3490 // if it is. See bug 4774291.
3491
3492 // For Reflection.getClassAccessFlags(), the null check occurs in
3493 // the wrong place; see inline_unsafe_access(), above, for a similar
3494 // situation.
3495 mirror = null_check(mirror);
3496 // If mirror or obj is dead, only null-path is taken.
3497 if (stopped()) return true;
3498
3499 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3500
3501 // Now load the mirror's klass metaobject, and null-check it.
3502 // Side-effects region with the control path if the klass is null.
3503 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3504 // If kls is null, we have a primitive mirror.
3505 phi->init_req(_prim_path, prim_return_value);
3506 if (stopped()) { set_result(region, phi); return true; }
3507 bool safe_for_replace = (region->in(_prim_path) == top());
3508
3509 Node* p; // handy temp
3510 Node* null_ctl;
3511
3512 // Now that we have the non-null klass, we can perform the real query.
3513 // For constant classes, the query will constant-fold in LoadNode::Value.
3514 Node* query_value = top();
3515 switch (id) {
3516 case vmIntrinsics::_isInstance:
3517 // nothing is an instance of a primitive type
3518 query_value = gen_instanceof(obj, kls, safe_for_replace);
3519 break;
3520
3521 case vmIntrinsics::_getModifiers:
3522 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3523 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3524 break;
3525
3526 case vmIntrinsics::_isInterface:
3527 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3528 if (generate_interface_guard(kls, region) != NULL)
3529 // A guard was added. If the guard is taken, it was an interface.
3530 phi->add_req(intcon(1));
3531 // If we fall through, it's a plain class.
3532 query_value = intcon(0);
3533 break;
3534
3535 case vmIntrinsics::_isArray:
3536 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3537 if (generate_array_guard(kls, region) != NULL)
3538 // A guard was added. If the guard is taken, it was an array.
3539 phi->add_req(intcon(1));
3540 // If we fall through, it's a plain class.
3541 query_value = intcon(0);
3542 break;
3543
3544 case vmIntrinsics::_isPrimitive:
3545 query_value = intcon(0); // "normal" path produces false
3546 break;
3547
3548 case vmIntrinsics::_getSuperclass:
3549 // The rules here are somewhat unfortunate, but we can still do better
3550 // with random logic than with a JNI call.
3551 // Interfaces store null or Object as _super, but must report null.
3552 // Arrays store an intermediate super as _super, but must report Object.
3553 // Other types can report the actual _super.
3554 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3555 if (generate_interface_guard(kls, region) != NULL)
3556 // A guard was added. If the guard is taken, it was an interface.
3557 phi->add_req(null());
3558 if (generate_array_guard(kls, region) != NULL)
3559 // A guard was added. If the guard is taken, it was an array.
3560 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3561 // If we fall through, it's a plain class. Get its _super.
3562 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3563 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3564 null_ctl = top();
3565 kls = null_check_oop(kls, &null_ctl);
3566 if (null_ctl != top()) {
3567 // If the guard is taken, Object.superClass is null (both klass and mirror).
3568 region->add_req(null_ctl);
3569 phi ->add_req(null());
3570 }
3571 if (!stopped()) {
3572 query_value = load_mirror_from_klass(kls);
3573 }
3574 break;
3575
3576 case vmIntrinsics::_getClassAccessFlags:
3577 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3578 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3579 break;
3580
3581 default:
3582 fatal_unexpected_iid(id);
3583 break;
3584 }
3585
3586 // Fall-through is the normal case of a query to a real class.
3587 phi->init_req(1, query_value);
3588 region->init_req(1, control());
3589
3590 C->set_has_split_ifs(true); // Has chance for split-if optimization
3591 set_result(region, phi);
3592 return true;
3593 }
3594
3595 //-------------------------inline_value_Class_conversion-------------------
3596 // public Class<T> java.lang.Class.asPrimaryType();
3597 // public Class<T> java.lang.Class.asIndirectType()
3598 bool LibraryCallKit::inline_value_Class_conversion(vmIntrinsics::ID id) {
3599 Node* mirror = argument(0); // Receiver Class
3600 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3601 if (mirror_con == NULL) {
3602 return false;
3603 }
3604
3605 bool is_indirect_type = true;
3606 ciType* tm = mirror_con->java_mirror_type(&is_indirect_type);
3607 if (tm != NULL) {
3608 Node* result = mirror;
3609 if (tm->is_valuetype()) {
3610 if (id == vmIntrinsics::_asPrimaryType && is_indirect_type) {
3611 result = _gvn.makecon(TypeInstPtr::make(tm->as_value_klass()->inline_mirror_instance()));
3612 } else if (id == vmIntrinsics::_asIndirectType && !is_indirect_type) {
3613 result = _gvn.makecon(TypeInstPtr::make(tm->as_value_klass()->indirect_mirror_instance()));
3614 }
3615 }
3616 set_result(result);
3617 return true;
3618 }
3619 return false;
3620 }
3621
3622 //-------------------------inline_Class_cast-------------------
3623 bool LibraryCallKit::inline_Class_cast() {
3624 Node* mirror = argument(0); // Class
3625 Node* obj = argument(1);
3626 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3627 if (mirror_con == NULL) {
3628 return false; // dead path (mirror->is_top()).
3629 }
3630 if (obj == NULL || obj->is_top()) {
3631 return false; // dead path
3632 }
3633
3634 ciKlass* obj_klass = NULL;
3635 if (obj->is_ValueType()) {
3636 obj_klass = _gvn.type(obj)->value_klass();
3637 } else {
3638 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3639 if (tp != NULL) {
3640 obj_klass = tp->klass();
3641 }
3642 }
3643
3644 // First, see if Class.cast() can be folded statically.
3645 // java_mirror_type() returns non-null for compile-time Class constants.
3646 bool is_indirect_type = true;
3647 ciType* tm = mirror_con->java_mirror_type(&is_indirect_type);
3648 if (!obj->is_ValueType() && !is_indirect_type) {
3649 obj = null_check(obj);
3650 if (stopped()) {
3651 return true;
3652 }
3653 }
3654 if (tm != NULL && tm->is_klass() && obj_klass != NULL) {
3655 if (!obj_klass->is_loaded()) {
3656 // Don't use intrinsic when class is not loaded.
3657 return false;
3658 } else {
3659 int static_res = C->static_subtype_check(tm->as_klass(), obj_klass);
3660 if (static_res == Compile::SSC_always_true) {
3661 // isInstance() is true - fold the code.
3662 set_result(obj);
3663 return true;
3664 } else if (static_res == Compile::SSC_always_false) {
3665 // Don't use intrinsic, have to throw ClassCastException.
3666 // If the reference is null, the non-intrinsic bytecode will
3667 // be optimized appropriately.
3668 return false;
3669 }
3670 }
3671 }
3672
3673 // Bailout intrinsic and do normal inlining if exception path is frequent.
3674 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3675 return false;
3676 }
3677
3678 // Generate dynamic checks.
3679 // Class.cast() is java implementation of _checkcast bytecode.
3680 // Do checkcast (Parse::do_checkcast()) optimizations here.
3681
3682 mirror = null_check(mirror);
3683 // If mirror is dead, only null-path is taken.
3684 if (stopped()) {
3685 return true;
3686 }
3687
3688 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3689 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
3690 RegionNode* region = new RegionNode(PATH_LIMIT);
3691 record_for_igvn(region);
3692
3693 // Now load the mirror's klass metaobject, and null-check it.
3694 // If kls is null, we have a primitive mirror and
3695 // nothing is an instance of a primitive type.
3696 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3697
3698 Node* res = top();
3699 if (!stopped()) {
3700 if (EnableValhalla && !obj->is_ValueType() && is_indirect_type) {
3701 // Check if (mirror == inline_mirror && obj == null)
3702 Node* is_val_mirror = generate_fair_guard(is_value_mirror(mirror), NULL);
3703 if (is_val_mirror != NULL) {
3704 RegionNode* r = new RegionNode(3);
3705 record_for_igvn(r);
3706 r->init_req(1, control());
3707
3708 // Casting to .val, check for null
3709 set_control(is_val_mirror);
3710 Node *null_ctr = top();
3711 null_check_oop(obj, &null_ctr);
3712 region->init_req(_npe_path, null_ctr);
3713 r->init_req(2, control());
3714
3715 set_control(_gvn.transform(r));
3716 }
3717 }
3718
3719 Node* bad_type_ctrl = top();
3720 // Do checkcast optimizations.
3721 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3722 region->init_req(_bad_type_path, bad_type_ctrl);
3723 }
3724 if (region->in(_prim_path) != top() ||
3725 region->in(_bad_type_path) != top() ||
3726 region->in(_npe_path) != top()) {
3727 // Let Interpreter throw ClassCastException.
3728 PreserveJVMState pjvms(this);
3729 set_control(_gvn.transform(region));
3730 uncommon_trap(Deoptimization::Reason_intrinsic,
3731 Deoptimization::Action_maybe_recompile);
3732 }
3733 if (!stopped()) {
3734 set_result(res);
3735 }
3736 return true;
3737 }
3738
3739
3740 //--------------------------inline_native_subtype_check------------------------
3741 // This intrinsic takes the JNI calls out of the heart of
3742 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3743 bool LibraryCallKit::inline_native_subtype_check() {
3744 // Pull both arguments off the stack.
3745 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3746 args[0] = argument(0);
3747 args[1] = argument(1);
3748 Node* klasses[2]; // corresponding Klasses: superk, subk
3749 klasses[0] = klasses[1] = top();
3750
3751 enum {
3752 // A full decision tree on {superc is prim, subc is prim}:
3753 _prim_0_path = 1, // {P,N} => false
3754 // {P,P} & superc!=subc => false
3755 _prim_same_path, // {P,P} & superc==subc => true
3756 _prim_1_path, // {N,P} => false
3757 _ref_subtype_path, // {N,N} & subtype check wins => true
3758 _both_ref_path, // {N,N} & subtype check loses => false
3759 PATH_LIMIT
3760 };
3761
3762 RegionNode* region = new RegionNode(PATH_LIMIT);
3763 RegionNode* prim_region = new RegionNode(2);
3764 Node* phi = new PhiNode(region, TypeInt::BOOL);
3765 record_for_igvn(region);
3766 record_for_igvn(prim_region);
3767
3768 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3769 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3770 int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3771
3772 // First null-check both mirrors and load each mirror's klass metaobject.
3773 int which_arg;
3774 for (which_arg = 0; which_arg <= 1; which_arg++) {
3775 Node* arg = args[which_arg];
3776 arg = null_check(arg);
3777 if (stopped()) break;
3778 args[which_arg] = arg;
3779
3780 Node* p = basic_plus_adr(arg, class_klass_offset);
3781 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3782 klasses[which_arg] = _gvn.transform(kls);
3783 }
3784
3785 // Resolve oops to stable for CmpP below.
3786 args[0] = access_resolve(args[0], 0);
3787 args[1] = access_resolve(args[1], 0);
3788
3789 // Having loaded both klasses, test each for null.
3790 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3791 for (which_arg = 0; which_arg <= 1; which_arg++) {
3792 Node* kls = klasses[which_arg];
3793 Node* null_ctl = top();
3794 kls = null_check_oop(kls, &null_ctl, never_see_null);
3795 if (which_arg == 0) {
3796 prim_region->init_req(1, null_ctl);
3797 } else {
3798 region->init_req(_prim_1_path, null_ctl);
3799 }
3800 if (stopped()) break;
3801 klasses[which_arg] = kls;
3802 }
3803
3804 if (!stopped()) {
3805 // now we have two reference types, in klasses[0..1]
3806 Node* subk = klasses[1]; // the argument to isAssignableFrom
3807 Node* superk = klasses[0]; // the receiver
3808 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3809 // If superc is a value mirror, we also need to check if superc == subc because
3810 // V? is not a subtype of V but due to subk == superk the subtype check will pass.
3811 generate_fair_guard(is_value_mirror(args[0]), prim_region);
3812 // now we have a successful reference subtype check
3813 region->set_req(_ref_subtype_path, control());
3814 }
3815
3816 // If both operands are primitive (both klasses null), then
3817 // we must return true when they are identical primitives.
3818 // It is convenient to test this after the first null klass check.
3819 // This path is also used if superc is a value mirror.
3820 set_control(_gvn.transform(prim_region));
3821 if (!stopped()) {
3822 // Since superc is primitive, make a guard for the superc==subc case.
3823 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3824 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3825 generate_fair_guard(bol_eq, region);
3826 if (region->req() == PATH_LIMIT+1) {
3827 // A guard was added. If the added guard is taken, superc==subc.
3828 region->swap_edges(PATH_LIMIT, _prim_same_path);
3829 region->del_req(PATH_LIMIT);
3830 }
3831 region->set_req(_prim_0_path, control()); // Not equal after all.
3832 }
3833
3834 // these are the only paths that produce 'true':
3835 phi->set_req(_prim_same_path, intcon(1));
3836 phi->set_req(_ref_subtype_path, intcon(1));
3837
3838 // pull together the cases:
3839 assert(region->req() == PATH_LIMIT, "sane region");
3840 for (uint i = 1; i < region->req(); i++) {
3841 Node* ctl = region->in(i);
3842 if (ctl == NULL || ctl == top()) {
3843 region->set_req(i, top());
3844 phi ->set_req(i, top());
3845 } else if (phi->in(i) == NULL) {
3846 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3847 }
3848 }
3849
3850 set_control(_gvn.transform(region));
3851 set_result(_gvn.transform(phi));
3852 return true;
3853 }
3854
3855 //---------------------generate_array_guard_common------------------------
3856 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind) {
3857
3858 if (stopped()) {
3859 return NULL;
3860 }
3861
3862 // Like generate_guard, adds a new path onto the region.
3863 jint layout_con = 0;
3864 Node* layout_val = get_layout_helper(kls, layout_con);
3865 if (layout_val == NULL) {
3866 bool query = 0;
3867 switch(kind) {
3868 case ObjectArray: query = Klass::layout_helper_is_objArray(layout_con); break;
3869 case NonObjectArray: query = !Klass::layout_helper_is_objArray(layout_con); break;
3870 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
3871 case ValueArray: query = Klass::layout_helper_is_valueArray(layout_con); break;
3872 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
3873 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
3874 default:
3875 ShouldNotReachHere();
3876 }
3877 if (!query) {
3878 return NULL; // never a branch
3879 } else { // always a branch
3880 Node* always_branch = control();
3881 if (region != NULL)
3882 region->add_req(always_branch);
3883 set_control(top());
3884 return always_branch;
3885 }
3886 }
3887 unsigned int value = 0;
3888 BoolTest::mask btest = BoolTest::illegal;
3889 switch(kind) {
3890 case ObjectArray:
3891 case NonObjectArray: {
3892 value = Klass::_lh_array_tag_obj_value;
3893 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
3894 btest = kind == ObjectArray ? BoolTest::eq : BoolTest::ne;
3895 break;
3896 }
3897 case TypeArray: {
3898 value = Klass::_lh_array_tag_type_value;
3899 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
3900 btest = BoolTest::eq;
3901 break;
3902 }
3903 case ValueArray: {
3904 value = Klass::_lh_array_tag_vt_value;
3905 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
3906 btest = BoolTest::eq;
3907 break;
3908 }
3909 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
3910 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
3911 default:
3912 ShouldNotReachHere();
3913 }
3914 // Now test the correct condition.
3915 jint nval = (jint)value;
3916 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3917 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3918 return generate_fair_guard(bol, region);
3919 }
3920
3921
3922 //-----------------------inline_native_newArray--------------------------
3923 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
3924 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3925 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3926 Node* mirror;
3927 Node* count_val;
3928 if (uninitialized) {
3929 mirror = argument(1);
3930 count_val = argument(2);
3931 } else {
3932 mirror = argument(0);
3933 count_val = argument(1);
3934 }
3935
3936 mirror = null_check(mirror);
3937 // If mirror or obj is dead, only null-path is taken.
3938 if (stopped()) return true;
3939
3940 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3941 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3942 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3943 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3944 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3945
3946 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3947 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3948 result_reg, _slow_path);
3949 Node* normal_ctl = control();
3950 Node* no_array_ctl = result_reg->in(_slow_path);
3951
3952 // Generate code for the slow case. We make a call to newArray().
3953 set_control(no_array_ctl);
3954 if (!stopped()) {
3955 // Either the input type is void.class, or else the
3956 // array klass has not yet been cached. Either the
3957 // ensuing call will throw an exception, or else it
3958 // will cache the array klass for next time.
3959 PreserveJVMState pjvms(this);
3960 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3961 Node* slow_result = set_results_for_java_call(slow_call);
3962 // this->control() comes from set_results_for_java_call
3963 result_reg->set_req(_slow_path, control());
3964 result_val->set_req(_slow_path, slow_result);
3965 result_io ->set_req(_slow_path, i_o());
3966 result_mem->set_req(_slow_path, reset_memory());
3967 }
3968
3969 set_control(normal_ctl);
3970 if (!stopped()) {
3971 // Normal case: The array type has been cached in the java.lang.Class.
3972 // The following call works fine even if the array type is polymorphic.
3973 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3974 Node* obj = new_array(klass_node, count_val, 0, NULL, false, mirror); // no arguments to push
3975 result_reg->init_req(_normal_path, control());
3976 result_val->init_req(_normal_path, obj);
3977 result_io ->init_req(_normal_path, i_o());
3978 result_mem->init_req(_normal_path, reset_memory());
3979
3980 if (uninitialized) {
3981 // Mark the allocation so that zeroing is skipped
3982 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3983 alloc->maybe_set_complete(&_gvn);
3984 }
3985 }
3986
3987 // Return the combined state.
3988 set_i_o( _gvn.transform(result_io) );
3989 set_all_memory( _gvn.transform(result_mem));
3990
3991 C->set_has_split_ifs(true); // Has chance for split-if optimization
3992 set_result(result_reg, result_val);
3993 return true;
3994 }
3995
3996 //----------------------inline_native_getLength--------------------------
3997 // public static native int java.lang.reflect.Array.getLength(Object array);
3998 bool LibraryCallKit::inline_native_getLength() {
3999 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4000
4001 Node* array = null_check(argument(0));
4002 // If array is dead, only null-path is taken.
4003 if (stopped()) return true;
4004
4005 // Deoptimize if it is a non-array.
4006 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
4007
4008 if (non_array != NULL) {
4009 PreserveJVMState pjvms(this);
4010 set_control(non_array);
4011 uncommon_trap(Deoptimization::Reason_intrinsic,
4012 Deoptimization::Action_maybe_recompile);
4013 }
4014
4015 // If control is dead, only non-array-path is taken.
4016 if (stopped()) return true;
4017
4018 // The works fine even if the array type is polymorphic.
4019 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4020 Node* result = load_array_length(array);
4021
4022 C->set_has_split_ifs(true); // Has chance for split-if optimization
4023 set_result(result);
4024 return true;
4025 }
4026
4027 //------------------------inline_array_copyOf----------------------------
4028 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4029 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4030 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4031 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4032
4033 // Get the arguments.
4034 Node* original = argument(0);
4035 Node* start = is_copyOfRange? argument(1): intcon(0);
4036 Node* end = is_copyOfRange? argument(2): argument(1);
4037 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4038
4039 const TypeAryPtr* original_t = _gvn.type(original)->isa_aryptr();
4040 const TypeInstPtr* mirror_t = _gvn.type(array_type_mirror)->isa_instptr();
4041 if (EnableValhalla && ValueArrayFlatten &&
4042 (original_t == NULL || mirror_t == NULL ||
4043 (mirror_t->java_mirror_type() == NULL &&
4044 (original_t->elem()->isa_valuetype() ||
4045 (original_t->elem()->make_oopptr() != NULL &&
4046 original_t->elem()->make_oopptr()->can_be_value_type()))))) {
4047 // We need to know statically if the copy is to a flattened array
4048 // or not but can't tell.
4049 return false;
4050 }
4051
4052 Node* newcopy = NULL;
4053
4054 // Set the original stack and the reexecute bit for the interpreter to reexecute
4055 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4056 { PreserveReexecuteState preexecs(this);
4057 jvms()->set_should_reexecute(true);
4058
4059 array_type_mirror = null_check(array_type_mirror);
4060 original = null_check(original);
4061
4062 // Check if a null path was taken unconditionally.
4063 if (stopped()) return true;
4064
4065 Node* orig_length = load_array_length(original);
4066
4067 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
4068 klass_node = null_check(klass_node);
4069
4070 RegionNode* bailout = new RegionNode(1);
4071 record_for_igvn(bailout);
4072
4073 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4074 // Bail out if that is so.
4075 // Value type array may have object field that would require a
4076 // write barrier. Conservatively, go to slow path.
4077 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4078 Node* not_objArray = !bs->array_copy_requires_gc_barriers(false, T_OBJECT, false, BarrierSetC2::Parsing) ?
4079 generate_typeArray_guard(klass_node, bailout) : generate_non_objArray_guard(klass_node, bailout);
4080 if (not_objArray != NULL) {
4081 // Improve the klass node's type from the new optimistic assumption:
4082 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4083 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0), false);
4084 Node* cast = new CastPPNode(klass_node, akls);
4085 cast->init_req(0, control());
4086 klass_node = _gvn.transform(cast);
4087 }
4088
4089 Node* original_kls = load_object_klass(original);
4090 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4091 // loads/stores but it is legal only if we're sure the
4092 // Arrays.copyOf would succeed. So we need all input arguments
4093 // to the copyOf to be validated, including that the copy to the
4094 // new array won't trigger an ArrayStoreException. That subtype
4095 // check can be optimized if we know something on the type of
4096 // the input array from type speculation.
4097 if (_gvn.type(klass_node)->singleton() && !stopped()) {
4098 ciKlass* subk = _gvn.type(original_kls)->is_klassptr()->klass();
4099 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
4100
4101 int test = C->static_subtype_check(superk, subk);
4102 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4103 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4104 if (t_original->speculative_type() != NULL) {
4105 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4106 original_kls = load_object_klass(original);
4107 }
4108 }
4109 }
4110
4111 if (ValueArrayFlatten) {
4112 // Either both or neither new array klass and original array
4113 // klass must be flattened
4114 Node* is_flat = generate_valueArray_guard(klass_node, NULL);
4115 if (!original_t->is_not_flat()) {
4116 generate_valueArray_guard(original_kls, bailout);
4117 }
4118 if (is_flat != NULL) {
4119 RegionNode* r = new RegionNode(2);
4120 record_for_igvn(r);
4121 r->init_req(1, control());
4122 set_control(is_flat);
4123 if (!original_t->is_not_flat()) {
4124 generate_valueArray_guard(original_kls, r);
4125 }
4126 bailout->add_req(control());
4127 set_control(_gvn.transform(r));
4128 }
4129 }
4130
4131 // Bail out if either start or end is negative.
4132 generate_negative_guard(start, bailout, &start);
4133 generate_negative_guard(end, bailout, &end);
4134
4135 Node* length = end;
4136 if (_gvn.type(start) != TypeInt::ZERO) {
4137 length = _gvn.transform(new SubINode(end, start));
4138 }
4139
4140 // Bail out if length is negative.
4141 // Without this the new_array would throw
4142 // NegativeArraySizeException but IllegalArgumentException is what
4143 // should be thrown
4144 generate_negative_guard(length, bailout, &length);
4145
4146 if (bailout->req() > 1) {
4147 PreserveJVMState pjvms(this);
4148 set_control(_gvn.transform(bailout));
4149 uncommon_trap(Deoptimization::Reason_intrinsic,
4150 Deoptimization::Action_maybe_recompile);
4151 }
4152
4153 if (!stopped()) {
4154 // How many elements will we copy from the original?
4155 // The answer is MinI(orig_length - start, length).
4156 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4157 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4158
4159 original = access_resolve(original, ACCESS_READ);
4160
4161 // Generate a direct call to the right arraycopy function(s).
4162 // We know the copy is disjoint but we might not know if the
4163 // oop stores need checking.
4164 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4165 // This will fail a store-check if x contains any non-nulls.
4166
4167 bool validated = false;
4168 // Reason_class_check rather than Reason_intrinsic because we
4169 // want to intrinsify even if this traps.
4170 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4171 Node* not_subtype_ctrl = gen_subtype_check(original_kls,
4172 klass_node);
4173
4174 if (not_subtype_ctrl != top()) {
4175 PreserveJVMState pjvms(this);
4176 set_control(not_subtype_ctrl);
4177 uncommon_trap(Deoptimization::Reason_class_check,
4178 Deoptimization::Action_make_not_entrant);
4179 assert(stopped(), "Should be stopped");
4180 }
4181 validated = true;
4182 }
4183
4184 if (!stopped()) {
4185 // Load element mirror
4186 Node* p = basic_plus_adr(array_type_mirror, java_lang_Class::component_mirror_offset_in_bytes());
4187 Node* elem_mirror = access_load_at(array_type_mirror, p, _gvn.type(p)->is_ptr(), TypeInstPtr::MIRROR, T_OBJECT, IN_HEAP);
4188
4189 newcopy = new_array(klass_node, length, 0, NULL, false, elem_mirror);
4190
4191 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
4192 original_kls, klass_node);
4193 if (!is_copyOfRange) {
4194 ac->set_copyof(validated);
4195 } else {
4196 ac->set_copyofrange(validated);
4197 }
4198 Node* n = _gvn.transform(ac);
4199 if (n == ac) {
4200 ac->connect_outputs(this);
4201 } else {
4202 assert(validated, "shouldn't transform if all arguments not validated");
4203 set_all_memory(n);
4204 }
4205 }
4206 }
4207 } // original reexecute is set back here
4208
4209 C->set_has_split_ifs(true); // Has chance for split-if optimization
4210 if (!stopped()) {
4211 set_result(newcopy);
4212 }
4213 return true;
4214 }
4215
4216
4217 //----------------------generate_virtual_guard---------------------------
4218 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4219 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4220 RegionNode* slow_region) {
4221 ciMethod* method = callee();
4222 int vtable_index = method->vtable_index();
4223 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4224 "bad index %d", vtable_index);
4225 // Get the Method* out of the appropriate vtable entry.
4226 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4227 vtable_index*vtableEntry::size_in_bytes() +
4228 vtableEntry::method_offset_in_bytes();
4229 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4230 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4231
4232 // Compare the target method with the expected method (e.g., Object.hashCode).
4233 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4234
4235 Node* native_call = makecon(native_call_addr);
4236 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4237 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4238
4239 return generate_slow_guard(test_native, slow_region);
4240 }
4241
4242 //-----------------------generate_method_call----------------------------
4243 // Use generate_method_call to make a slow-call to the real
4244 // method if the fast path fails. An alternative would be to
4245 // use a stub like OptoRuntime::slow_arraycopy_Java.
4246 // This only works for expanding the current library call,
4247 // not another intrinsic. (E.g., don't use this for making an
4248 // arraycopy call inside of the copyOf intrinsic.)
4249 CallJavaNode*
4250 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4251 // When compiling the intrinsic method itself, do not use this technique.
4252 guarantee(callee() != C->method(), "cannot make slow-call to self");
4253
4254 ciMethod* method = callee();
4255 // ensure the JVMS we have will be correct for this call
4256 guarantee(method_id == method->intrinsic_id(), "must match");
4257
4258 const TypeFunc* tf = TypeFunc::make(method);
4259 CallJavaNode* slow_call;
4260 if (is_static) {
4261 assert(!is_virtual, "");
4262 slow_call = new CallStaticJavaNode(C, tf,
4263 SharedRuntime::get_resolve_static_call_stub(),
4264 method, bci());
4265 } else if (is_virtual) {
4266 null_check_receiver();
4267 int vtable_index = Method::invalid_vtable_index;
4268 if (UseInlineCaches) {
4269 // Suppress the vtable call
4270 } else {
4271 // hashCode and clone are not a miranda methods,
4272 // so the vtable index is fixed.
4273 // No need to use the linkResolver to get it.
4274 vtable_index = method->vtable_index();
4275 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4276 "bad index %d", vtable_index);
4277 }
4278 slow_call = new CallDynamicJavaNode(tf,
4279 SharedRuntime::get_resolve_virtual_call_stub(),
4280 method, vtable_index, bci());
4281 } else { // neither virtual nor static: opt_virtual
4282 null_check_receiver();
4283 slow_call = new CallStaticJavaNode(C, tf,
4284 SharedRuntime::get_resolve_opt_virtual_call_stub(),
4285 method, bci());
4286 slow_call->set_optimized_virtual(true);
4287 }
4288 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4289 // To be able to issue a direct call (optimized virtual or virtual)
4290 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4291 // about the method being invoked should be attached to the call site to
4292 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4293 slow_call->set_override_symbolic_info(true);
4294 }
4295 set_arguments_for_java_call(slow_call);
4296 set_edges_for_java_call(slow_call);
4297 return slow_call;
4298 }
4299
4300
4301 /**
4302 * Build special case code for calls to hashCode on an object. This call may
4303 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4304 * slightly different code.
4305 */
4306 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4307 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4308 assert(!(is_virtual && is_static), "either virtual, special, or static");
4309
4310 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4311
4312 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4313 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4314 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4315 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4316 Node* obj = argument(0);
4317
4318 if (obj->is_ValueType() || gvn().type(obj)->is_valuetypeptr()) {
4319 return false;
4320 }
4321
4322 if (!is_static) {
4323 // Check for hashing null object
4324 obj = null_check_receiver();
4325 if (stopped()) return true; // unconditionally null
4326 result_reg->init_req(_null_path, top());
4327 result_val->init_req(_null_path, top());
4328 } else {
4329 // Do a null check, and return zero if null.
4330 // System.identityHashCode(null) == 0
4331 Node* null_ctl = top();
4332 obj = null_check_oop(obj, &null_ctl);
4333 result_reg->init_req(_null_path, null_ctl);
4334 result_val->init_req(_null_path, _gvn.intcon(0));
4335 }
4336
4337 // Unconditionally null? Then return right away.
4338 if (stopped()) {
4339 set_control( result_reg->in(_null_path));
4340 if (!stopped())
4341 set_result(result_val->in(_null_path));
4342 return true;
4343 }
4344
4345 // We only go to the fast case code if we pass a number of guards. The
4346 // paths which do not pass are accumulated in the slow_region.
4347 RegionNode* slow_region = new RegionNode(1);
4348 record_for_igvn(slow_region);
4349
4350 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4351 assert(!obj_type->isa_valuetype() || !obj_type->is_valuetypeptr(), "no value type here");
4352 if (is_static && obj_type->can_be_value_type()) {
4353 Node* obj_klass = load_object_klass(obj);
4354 generate_value_guard(obj_klass, slow_region);
4355 }
4356
4357 // If this is a virtual call, we generate a funny guard. We pull out
4358 // the vtable entry corresponding to hashCode() from the target object.
4359 // If the target method which we are calling happens to be the native
4360 // Object hashCode() method, we pass the guard. We do not need this
4361 // guard for non-virtual calls -- the caller is known to be the native
4362 // Object hashCode().
4363 if (is_virtual) {
4364 // After null check, get the object's klass.
4365 Node* obj_klass = load_object_klass(obj);
4366 generate_virtual_guard(obj_klass, slow_region);
4367 }
4368
4369 // Get the header out of the object, use LoadMarkNode when available
4370 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4371 // The control of the load must be NULL. Otherwise, the load can move before
4372 // the null check after castPP removal.
4373 Node* no_ctrl = NULL;
4374 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4375
4376 // Test the header to see if it is unlocked.
4377 Node *lock_mask = _gvn.MakeConX(markWord::biased_lock_mask_in_place);
4378 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4379 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4380 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4381 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4382
4383 generate_slow_guard(test_unlocked, slow_region);
4384
4385 // Get the hash value and check to see that it has been properly assigned.
4386 // We depend on hash_mask being at most 32 bits and avoid the use of
4387 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4388 // vm: see markWord.hpp.
4389 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4390 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4391 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4392 // This hack lets the hash bits live anywhere in the mark object now, as long
4393 // as the shift drops the relevant bits into the low 32 bits. Note that
4394 // Java spec says that HashCode is an int so there's no point in capturing
4395 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4396 hshifted_header = ConvX2I(hshifted_header);
4397 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4398
4399 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4400 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4401 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4402
4403 generate_slow_guard(test_assigned, slow_region);
4404
4405 Node* init_mem = reset_memory();
4406 // fill in the rest of the null path:
4407 result_io ->init_req(_null_path, i_o());
4408 result_mem->init_req(_null_path, init_mem);
4409
4410 result_val->init_req(_fast_path, hash_val);
4411 result_reg->init_req(_fast_path, control());
4412 result_io ->init_req(_fast_path, i_o());
4413 result_mem->init_req(_fast_path, init_mem);
4414
4415 // Generate code for the slow case. We make a call to hashCode().
4416 set_control(_gvn.transform(slow_region));
4417 if (!stopped()) {
4418 // No need for PreserveJVMState, because we're using up the present state.
4419 set_all_memory(init_mem);
4420 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4421 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4422 Node* slow_result = set_results_for_java_call(slow_call);
4423 // this->control() comes from set_results_for_java_call
4424 result_reg->init_req(_slow_path, control());
4425 result_val->init_req(_slow_path, slow_result);
4426 result_io ->set_req(_slow_path, i_o());
4427 result_mem ->set_req(_slow_path, reset_memory());
4428 }
4429
4430 // Return the combined state.
4431 set_i_o( _gvn.transform(result_io) );
4432 set_all_memory( _gvn.transform(result_mem));
4433
4434 set_result(result_reg, result_val);
4435 return true;
4436 }
4437
4438 //---------------------------inline_native_getClass----------------------------
4439 // public final native Class<?> java.lang.Object.getClass();
4440 //
4441 // Build special case code for calls to getClass on an object.
4442 bool LibraryCallKit::inline_native_getClass() {
4443 Node* obj = argument(0);
4444 if (obj->is_ValueType()) {
4445 ciKlass* vk = _gvn.type(obj)->value_klass();
4446 set_result(makecon(TypeInstPtr::make(vk->java_mirror())));
4447 return true;
4448 }
4449 obj = null_check_receiver();
4450 if (stopped()) return true;
4451 set_result(load_mirror_from_klass(load_object_klass(obj)));
4452 return true;
4453 }
4454
4455 //-----------------inline_native_Reflection_getCallerClass---------------------
4456 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4457 //
4458 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4459 //
4460 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4461 // in that it must skip particular security frames and checks for
4462 // caller sensitive methods.
4463 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4464 #ifndef PRODUCT
4465 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4466 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4467 }
4468 #endif
4469
4470 if (!jvms()->has_method()) {
4471 #ifndef PRODUCT
4472 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4473 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4474 }
4475 #endif
4476 return false;
4477 }
4478
4479 // Walk back up the JVM state to find the caller at the required
4480 // depth.
4481 JVMState* caller_jvms = jvms();
4482
4483 // Cf. JVM_GetCallerClass
4484 // NOTE: Start the loop at depth 1 because the current JVM state does
4485 // not include the Reflection.getCallerClass() frame.
4486 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4487 ciMethod* m = caller_jvms->method();
4488 switch (n) {
4489 case 0:
4490 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4491 break;
4492 case 1:
4493 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4494 if (!m->caller_sensitive()) {
4495 #ifndef PRODUCT
4496 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4497 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4498 }
4499 #endif
4500 return false; // bail-out; let JVM_GetCallerClass do the work
4501 }
4502 break;
4503 default:
4504 if (!m->is_ignored_by_security_stack_walk()) {
4505 // We have reached the desired frame; return the holder class.
4506 // Acquire method holder as java.lang.Class and push as constant.
4507 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4508 ciInstance* caller_mirror = caller_klass->java_mirror();
4509 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4510
4511 #ifndef PRODUCT
4512 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4513 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());
4514 tty->print_cr(" JVM state at this point:");
4515 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4516 ciMethod* m = jvms()->of_depth(i)->method();
4517 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4518 }
4519 }
4520 #endif
4521 return true;
4522 }
4523 break;
4524 }
4525 }
4526
4527 #ifndef PRODUCT
4528 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4529 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4530 tty->print_cr(" JVM state at this point:");
4531 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4532 ciMethod* m = jvms()->of_depth(i)->method();
4533 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4534 }
4535 }
4536 #endif
4537
4538 return false; // bail-out; let JVM_GetCallerClass do the work
4539 }
4540
4541 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4542 Node* arg = argument(0);
4543 Node* result = NULL;
4544
4545 switch (id) {
4546 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4547 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4548 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4549 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4550
4551 case vmIntrinsics::_doubleToLongBits: {
4552 // two paths (plus control) merge in a wood
4553 RegionNode *r = new RegionNode(3);
4554 Node *phi = new PhiNode(r, TypeLong::LONG);
4555
4556 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4557 // Build the boolean node
4558 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4559
4560 // Branch either way.
4561 // NaN case is less traveled, which makes all the difference.
4562 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4563 Node *opt_isnan = _gvn.transform(ifisnan);
4564 assert( opt_isnan->is_If(), "Expect an IfNode");
4565 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4566 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4567
4568 set_control(iftrue);
4569
4570 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4571 Node *slow_result = longcon(nan_bits); // return NaN
4572 phi->init_req(1, _gvn.transform( slow_result ));
4573 r->init_req(1, iftrue);
4574
4575 // Else fall through
4576 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4577 set_control(iffalse);
4578
4579 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4580 r->init_req(2, iffalse);
4581
4582 // Post merge
4583 set_control(_gvn.transform(r));
4584 record_for_igvn(r);
4585
4586 C->set_has_split_ifs(true); // Has chance for split-if optimization
4587 result = phi;
4588 assert(result->bottom_type()->isa_long(), "must be");
4589 break;
4590 }
4591
4592 case vmIntrinsics::_floatToIntBits: {
4593 // two paths (plus control) merge in a wood
4594 RegionNode *r = new RegionNode(3);
4595 Node *phi = new PhiNode(r, TypeInt::INT);
4596
4597 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4598 // Build the boolean node
4599 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4600
4601 // Branch either way.
4602 // NaN case is less traveled, which makes all the difference.
4603 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4604 Node *opt_isnan = _gvn.transform(ifisnan);
4605 assert( opt_isnan->is_If(), "Expect an IfNode");
4606 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4607 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4608
4609 set_control(iftrue);
4610
4611 static const jint nan_bits = 0x7fc00000;
4612 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4613 phi->init_req(1, _gvn.transform( slow_result ));
4614 r->init_req(1, iftrue);
4615
4616 // Else fall through
4617 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4618 set_control(iffalse);
4619
4620 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4621 r->init_req(2, iffalse);
4622
4623 // Post merge
4624 set_control(_gvn.transform(r));
4625 record_for_igvn(r);
4626
4627 C->set_has_split_ifs(true); // Has chance for split-if optimization
4628 result = phi;
4629 assert(result->bottom_type()->isa_int(), "must be");
4630 break;
4631 }
4632
4633 default:
4634 fatal_unexpected_iid(id);
4635 break;
4636 }
4637 set_result(_gvn.transform(result));
4638 return true;
4639 }
4640
4641 //----------------------inline_unsafe_copyMemory-------------------------
4642 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4643 bool LibraryCallKit::inline_unsafe_copyMemory() {
4644 if (callee()->is_static()) return false; // caller must have the capability!
4645 null_check_receiver(); // null-check receiver
4646 if (stopped()) return true;
4647
4648 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4649
4650 Node* src_ptr = argument(1); // type: oop
4651 Node* src_off = ConvL2X(argument(2)); // type: long
4652 Node* dst_ptr = argument(4); // type: oop
4653 Node* dst_off = ConvL2X(argument(5)); // type: long
4654 Node* size = ConvL2X(argument(7)); // type: long
4655
4656 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4657 "fieldOffset must be byte-scaled");
4658
4659 src_ptr = access_resolve(src_ptr, ACCESS_READ);
4660 dst_ptr = access_resolve(dst_ptr, ACCESS_WRITE);
4661 Node* src = make_unsafe_address(src_ptr, src_off, ACCESS_READ);
4662 Node* dst = make_unsafe_address(dst_ptr, dst_off, ACCESS_WRITE);
4663
4664 // Conservatively insert a memory barrier on all memory slices.
4665 // Do not let writes of the copy source or destination float below the copy.
4666 insert_mem_bar(Op_MemBarCPUOrder);
4667
4668 Node* thread = _gvn.transform(new ThreadLocalNode());
4669 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4670 BasicType doing_unsafe_access_bt = T_BYTE;
4671 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4672
4673 // update volatile field
4674 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4675
4676 // Call it. Note that the length argument is not scaled.
4677 make_runtime_call(RC_LEAF|RC_NO_FP,
4678 OptoRuntime::fast_arraycopy_Type(),
4679 StubRoutines::unsafe_arraycopy(),
4680 "unsafe_arraycopy",
4681 TypeRawPtr::BOTTOM,
4682 src, dst, size XTOP);
4683
4684 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4685
4686 // Do not let reads of the copy destination float above the copy.
4687 insert_mem_bar(Op_MemBarCPUOrder);
4688
4689 return true;
4690 }
4691
4692 //------------------------clone_coping-----------------------------------
4693 // Helper function for inline_native_clone.
4694 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4695 assert(obj_size != NULL, "");
4696 Node* raw_obj = alloc_obj->in(1);
4697 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4698
4699 AllocateNode* alloc = NULL;
4700 if (ReduceBulkZeroing) {
4701 // We will be completely responsible for initializing this object -
4702 // mark Initialize node as complete.
4703 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4704 // The object was just allocated - there should be no any stores!
4705 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4706 // Mark as complete_with_arraycopy so that on AllocateNode
4707 // expansion, we know this AllocateNode is initialized by an array
4708 // copy and a StoreStore barrier exists after the array copy.
4709 alloc->initialization()->set_complete_with_arraycopy();
4710 }
4711
4712 // Copy the fastest available way.
4713 // TODO: generate fields copies for small objects instead.
4714 Node* size = _gvn.transform(obj_size);
4715
4716 // Exclude the header but include array length to copy by 8 bytes words.
4717 // Can't use base_offset_in_bytes(bt) since basic type is unknown.
4718 int base_off = is_array ? arrayOopDesc::length_offset_in_bytes() :
4719 instanceOopDesc::base_offset_in_bytes();
4720 // base_off:
4721 // 8 - 32-bit VM
4722 // 12 - 64-bit VM, compressed klass
4723 // 16 - 64-bit VM, normal klass
4724 if (base_off % BytesPerLong != 0) {
4725 assert(UseCompressedClassPointers, "");
4726 if (is_array) {
4727 // Exclude length to copy by 8 bytes words.
4728 base_off += sizeof(int);
4729 } else {
4730 // Include klass to copy by 8 bytes words.
4731 base_off = instanceOopDesc::klass_offset_in_bytes();
4732 }
4733 assert(base_off % BytesPerLong == 0, "expect 8 bytes alignment");
4734 }
4735 Node* src_base = basic_plus_adr(obj, base_off);
4736 Node* dst_base = basic_plus_adr(alloc_obj, base_off);
4737
4738 // Compute the length also, if needed:
4739 Node* countx = size;
4740 countx = _gvn.transform(new SubXNode(countx, MakeConX(base_off)));
4741 countx = _gvn.transform(new URShiftXNode(countx, intcon(LogBytesPerLong)));
4742
4743 access_clone(src_base, dst_base, countx, is_array);
4744
4745 // Do not let reads from the cloned object float above the arraycopy.
4746 if (alloc != NULL) {
4747 // Do not let stores that initialize this object be reordered with
4748 // a subsequent store that would make this object accessible by
4749 // other threads.
4750 // Record what AllocateNode this StoreStore protects so that
4751 // escape analysis can go from the MemBarStoreStoreNode to the
4752 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4753 // based on the escape status of the AllocateNode.
4754 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4755 } else {
4756 insert_mem_bar(Op_MemBarCPUOrder);
4757 }
4758 }
4759
4760 //------------------------inline_native_clone----------------------------
4761 // protected native Object java.lang.Object.clone();
4762 //
4763 // Here are the simple edge cases:
4764 // null receiver => normal trap
4765 // virtual and clone was overridden => slow path to out-of-line clone
4766 // not cloneable or finalizer => slow path to out-of-line Object.clone
4767 //
4768 // The general case has two steps, allocation and copying.
4769 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4770 //
4771 // Copying also has two cases, oop arrays and everything else.
4772 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4773 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4774 //
4775 // These steps fold up nicely if and when the cloned object's klass
4776 // can be sharply typed as an object array, a type array, or an instance.
4777 //
4778 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4779 PhiNode* result_val;
4780
4781 // Set the reexecute bit for the interpreter to reexecute
4782 // the bytecode that invokes Object.clone if deoptimization happens.
4783 { PreserveReexecuteState preexecs(this);
4784 jvms()->set_should_reexecute(true);
4785
4786 Node* obj = argument(0);
4787 if (obj->is_ValueType()) {
4788 return false;
4789 }
4790
4791 obj = null_check_receiver();
4792 if (stopped()) return true;
4793
4794 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4795
4796 // If we are going to clone an instance, we need its exact type to
4797 // know the number and types of fields to convert the clone to
4798 // loads/stores. Maybe a speculative type can help us.
4799 if (!obj_type->klass_is_exact() &&
4800 obj_type->speculative_type() != NULL &&
4801 obj_type->speculative_type()->is_instance_klass() &&
4802 !obj_type->speculative_type()->is_valuetype()) {
4803 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4804 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4805 !spec_ik->has_injected_fields()) {
4806 ciKlass* k = obj_type->klass();
4807 if (!k->is_instance_klass() ||
4808 k->as_instance_klass()->is_interface() ||
4809 k->as_instance_klass()->has_subklass()) {
4810 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4811 }
4812 }
4813 }
4814
4815 Node* obj_klass = load_object_klass(obj);
4816
4817 // Conservatively insert a memory barrier on all memory slices.
4818 // Do not let writes into the original float below the clone.
4819 insert_mem_bar(Op_MemBarCPUOrder);
4820
4821 // paths into result_reg:
4822 enum {
4823 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4824 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4825 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4826 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4827 PATH_LIMIT
4828 };
4829 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4830 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4831 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4832 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4833 record_for_igvn(result_reg);
4834
4835 // We only go to the fast case code if we pass a number of guards.
4836 // The paths which do not pass are accumulated in the slow_region.
4837 RegionNode* slow_region = new RegionNode(1);
4838 record_for_igvn(slow_region);
4839
4840 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4841 if (array_ctl != NULL) {
4842 // It's an array.
4843 PreserveJVMState pjvms(this);
4844 set_control(array_ctl);
4845
4846 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4847 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, BarrierSetC2::Parsing) &&
4848 (!obj_type->isa_aryptr() || !obj_type->is_aryptr()->is_not_flat())) {
4849 // Flattened value type array may have object field that would require a
4850 // write barrier. Conservatively, go to slow path.
4851 generate_valueArray_guard(obj_klass, slow_region);
4852 }
4853
4854 if (!stopped()) {
4855 Node* obj_length = load_array_length(obj);
4856 Node* obj_size = NULL;
4857 // Load element mirror
4858 Node* array_type_mirror = load_mirror_from_klass(obj_klass);
4859 Node* p = basic_plus_adr(array_type_mirror, java_lang_Class::component_mirror_offset_in_bytes());
4860 Node* elem_mirror = access_load_at(array_type_mirror, p, _gvn.type(p)->is_ptr(), TypeInstPtr::MIRROR, T_OBJECT, IN_HEAP);
4861
4862 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, false, elem_mirror);
4863
4864 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4865 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, BarrierSetC2::Parsing)) {
4866 // If it is an oop array, it requires very special treatment,
4867 // because gc barriers are required when accessing the array.
4868 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4869 if (is_obja != NULL) {
4870 PreserveJVMState pjvms2(this);
4871 set_control(is_obja);
4872 // Generate a direct call to the right arraycopy function(s).
4873 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4874 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false);
4875 ac->set_cloneoop();
4876 Node* n = _gvn.transform(ac);
4877 assert(n == ac, "cannot disappear");
4878 ac->connect_outputs(this);
4879
4880 result_reg->init_req(_objArray_path, control());
4881 result_val->init_req(_objArray_path, alloc_obj);
4882 result_i_o ->set_req(_objArray_path, i_o());
4883 result_mem ->set_req(_objArray_path, reset_memory());
4884 }
4885 }
4886
4887 // Otherwise, there are no barriers to worry about.
4888 // (We can dispense with card marks if we know the allocation
4889 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4890 // causes the non-eden paths to take compensating steps to
4891 // simulate a fresh allocation, so that no further
4892 // card marks are required in compiled code to initialize
4893 // the object.)
4894
4895 if (!stopped()) {
4896 copy_to_clone(obj, alloc_obj, obj_size, true);
4897
4898 // Present the results of the copy.
4899 result_reg->init_req(_array_path, control());
4900 result_val->init_req(_array_path, alloc_obj);
4901 result_i_o ->set_req(_array_path, i_o());
4902 result_mem ->set_req(_array_path, reset_memory());
4903 }
4904 }
4905 }
4906
4907 if (!stopped()) {
4908 // It's an instance (we did array above). Make the slow-path tests.
4909 // If this is a virtual call, we generate a funny guard. We grab
4910 // the vtable entry corresponding to clone() from the target object.
4911 // If the target method which we are calling happens to be the
4912 // Object clone() method, we pass the guard. We do not need this
4913 // guard for non-virtual calls; the caller is known to be the native
4914 // Object clone().
4915 if (is_virtual) {
4916 generate_virtual_guard(obj_klass, slow_region);
4917 }
4918
4919 // The object must be easily cloneable and must not have a finalizer.
4920 // Both of these conditions may be checked in a single test.
4921 // We could optimize the test further, but we don't care.
4922 generate_access_flags_guard(obj_klass,
4923 // Test both conditions:
4924 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4925 // Must be cloneable but not finalizer:
4926 JVM_ACC_IS_CLONEABLE_FAST,
4927 slow_region);
4928 }
4929
4930 if (!stopped()) {
4931 // It's an instance, and it passed the slow-path tests.
4932 PreserveJVMState pjvms(this);
4933 Node* obj_size = NULL;
4934 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4935 // is reexecuted if deoptimization occurs and there could be problems when merging
4936 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4937 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4938
4939 copy_to_clone(obj, alloc_obj, obj_size, false);
4940
4941 // Present the results of the slow call.
4942 result_reg->init_req(_instance_path, control());
4943 result_val->init_req(_instance_path, alloc_obj);
4944 result_i_o ->set_req(_instance_path, i_o());
4945 result_mem ->set_req(_instance_path, reset_memory());
4946 }
4947
4948 // Generate code for the slow case. We make a call to clone().
4949 set_control(_gvn.transform(slow_region));
4950 if (!stopped()) {
4951 PreserveJVMState pjvms(this);
4952 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4953 // We need to deoptimize on exception (see comment above)
4954 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4955 // this->control() comes from set_results_for_java_call
4956 result_reg->init_req(_slow_path, control());
4957 result_val->init_req(_slow_path, slow_result);
4958 result_i_o ->set_req(_slow_path, i_o());
4959 result_mem ->set_req(_slow_path, reset_memory());
4960 }
4961
4962 // Return the combined state.
4963 set_control( _gvn.transform(result_reg));
4964 set_i_o( _gvn.transform(result_i_o));
4965 set_all_memory( _gvn.transform(result_mem));
4966 } // original reexecute is set back here
4967
4968 set_result(_gvn.transform(result_val));
4969 return true;
4970 }
4971
4972 // If we have a tightly coupled allocation, the arraycopy may take care
4973 // of the array initialization. If one of the guards we insert between
4974 // the allocation and the arraycopy causes a deoptimization, an
4975 // unitialized array will escape the compiled method. To prevent that
4976 // we set the JVM state for uncommon traps between the allocation and
4977 // the arraycopy to the state before the allocation so, in case of
4978 // deoptimization, we'll reexecute the allocation and the
4979 // initialization.
4980 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4981 if (alloc != NULL) {
4982 ciMethod* trap_method = alloc->jvms()->method();
4983 int trap_bci = alloc->jvms()->bci();
4984
4985 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4986 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4987 // Make sure there's no store between the allocation and the
4988 // arraycopy otherwise visible side effects could be rexecuted
4989 // in case of deoptimization and cause incorrect execution.
4990 bool no_interfering_store = true;
4991 Node* mem = alloc->in(TypeFunc::Memory);
4992 if (mem->is_MergeMem()) {
4993 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4994 Node* n = mms.memory();
4995 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4996 assert(n->is_Store(), "what else?");
4997 no_interfering_store = false;
4998 break;
4999 }
5000 }
5001 } else {
5002 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5003 Node* n = mms.memory();
5004 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5005 assert(n->is_Store(), "what else?");
5006 no_interfering_store = false;
5007 break;
5008 }
5009 }
5010 }
5011
5012 if (no_interfering_store) {
5013 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5014 uint size = alloc->req();
5015 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5016 old_jvms->set_map(sfpt);
5017 for (uint i = 0; i < size; i++) {
5018 sfpt->init_req(i, alloc->in(i));
5019 }
5020 // re-push array length for deoptimization
5021 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5022 old_jvms->set_sp(old_jvms->sp()+1);
5023 old_jvms->set_monoff(old_jvms->monoff()+1);
5024 old_jvms->set_scloff(old_jvms->scloff()+1);
5025 old_jvms->set_endoff(old_jvms->endoff()+1);
5026 old_jvms->set_should_reexecute(true);
5027
5028 sfpt->set_i_o(map()->i_o());
5029 sfpt->set_memory(map()->memory());
5030 sfpt->set_control(map()->control());
5031
5032 JVMState* saved_jvms = jvms();
5033 saved_reexecute_sp = _reexecute_sp;
5034
5035 set_jvms(sfpt->jvms());
5036 _reexecute_sp = jvms()->sp();
5037
5038 return saved_jvms;
5039 }
5040 }
5041 }
5042 return NULL;
5043 }
5044
5045 // In case of a deoptimization, we restart execution at the
5046 // allocation, allocating a new array. We would leave an uninitialized
5047 // array in the heap that GCs wouldn't expect. Move the allocation
5048 // after the traps so we don't allocate the array if we
5049 // deoptimize. This is possible because tightly_coupled_allocation()
5050 // guarantees there's no observer of the allocated array at this point
5051 // and the control flow is simple enough.
5052 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
5053 int saved_reexecute_sp, uint new_idx) {
5054 if (saved_jvms != NULL && !stopped()) {
5055 assert(alloc != NULL, "only with a tightly coupled allocation");
5056 // restore JVM state to the state at the arraycopy
5057 saved_jvms->map()->set_control(map()->control());
5058 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
5059 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
5060 // If we've improved the types of some nodes (null check) while
5061 // emitting the guards, propagate them to the current state
5062 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
5063 set_jvms(saved_jvms);
5064 _reexecute_sp = saved_reexecute_sp;
5065
5066 // Remove the allocation from above the guards
5067 CallProjections* callprojs = alloc->extract_projections(true);
5068 InitializeNode* init = alloc->initialization();
5069 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5070 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5071 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5072 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5073
5074 // move the allocation here (after the guards)
5075 _gvn.hash_delete(alloc);
5076 alloc->set_req(TypeFunc::Control, control());
5077 alloc->set_req(TypeFunc::I_O, i_o());
5078 Node *mem = reset_memory();
5079 set_all_memory(mem);
5080 alloc->set_req(TypeFunc::Memory, mem);
5081 set_control(init->proj_out_or_null(TypeFunc::Control));
5082 set_i_o(callprojs->fallthrough_ioproj);
5083
5084 // Update memory as done in GraphKit::set_output_for_allocation()
5085 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5086 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5087 if (ary_type->isa_aryptr() && length_type != NULL) {
5088 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5089 }
5090 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5091 int elemidx = C->get_alias_index(telemref);
5092 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5093 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5094
5095 Node* allocx = _gvn.transform(alloc);
5096 assert(allocx == alloc, "where has the allocation gone?");
5097 assert(dest->is_CheckCastPP(), "not an allocation result?");
5098
5099 _gvn.hash_delete(dest);
5100 dest->set_req(0, control());
5101 Node* destx = _gvn.transform(dest);
5102 assert(destx == dest, "where has the allocation result gone?");
5103 }
5104 }
5105
5106
5107 //------------------------------inline_arraycopy-----------------------
5108 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5109 // Object dest, int destPos,
5110 // int length);
5111 bool LibraryCallKit::inline_arraycopy() {
5112 // Get the arguments.
5113 Node* src = argument(0); // type: oop
5114 Node* src_offset = argument(1); // type: int
5115 Node* dest = argument(2); // type: oop
5116 Node* dest_offset = argument(3); // type: int
5117 Node* length = argument(4); // type: int
5118
5119 uint new_idx = C->unique();
5120
5121 // Check for allocation before we add nodes that would confuse
5122 // tightly_coupled_allocation()
5123 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
5124
5125 int saved_reexecute_sp = -1;
5126 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5127 // See arraycopy_restore_alloc_state() comment
5128 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5129 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
5130 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
5131 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
5132
5133 // The following tests must be performed
5134 // (1) src and dest are arrays.
5135 // (2) src and dest arrays must have elements of the same BasicType
5136 // (3) src and dest must not be null.
5137 // (4) src_offset must not be negative.
5138 // (5) dest_offset must not be negative.
5139 // (6) length must not be negative.
5140 // (7) src_offset + length must not exceed length of src.
5141 // (8) dest_offset + length must not exceed length of dest.
5142 // (9) each element of an oop array must be assignable
5143
5144 // (3) src and dest must not be null.
5145 // always do this here because we need the JVM state for uncommon traps
5146 Node* null_ctl = top();
5147 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5148 assert(null_ctl->is_top(), "no null control here");
5149 dest = null_check(dest, T_ARRAY);
5150
5151 if (!can_emit_guards) {
5152 // if saved_jvms == NULL and alloc != NULL, we don't emit any
5153 // guards but the arraycopy node could still take advantage of a
5154 // tightly allocated allocation. tightly_coupled_allocation() is
5155 // called again to make sure it takes the null check above into
5156 // account: the null check is mandatory and if it caused an
5157 // uncommon trap to be emitted then the allocation can't be
5158 // considered tightly coupled in this context.
5159 alloc = tightly_coupled_allocation(dest, NULL);
5160 }
5161
5162 bool validated = false;
5163
5164 const Type* src_type = _gvn.type(src);
5165 const Type* dest_type = _gvn.type(dest);
5166 const TypeAryPtr* top_src = src_type->isa_aryptr();
5167 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5168
5169 // Do we have the type of src?
5170 bool has_src = (top_src != NULL && top_src->klass() != NULL);
5171 // Do we have the type of dest?
5172 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
5173 // Is the type for src from speculation?
5174 bool src_spec = false;
5175 // Is the type for dest from speculation?
5176 bool dest_spec = false;
5177
5178 if ((!has_src || !has_dest) && can_emit_guards) {
5179 // We don't have sufficient type information, let's see if
5180 // speculative types can help. We need to have types for both src
5181 // and dest so that it pays off.
5182
5183 // Do we already have or could we have type information for src
5184 bool could_have_src = has_src;
5185 // Do we already have or could we have type information for dest
5186 bool could_have_dest = has_dest;
5187
5188 ciKlass* src_k = NULL;
5189 if (!has_src) {
5190 src_k = src_type->speculative_type_not_null();
5191 if (src_k != NULL && src_k->is_array_klass()) {
5192 could_have_src = true;
5193 }
5194 }
5195
5196 ciKlass* dest_k = NULL;
5197 if (!has_dest) {
5198 dest_k = dest_type->speculative_type_not_null();
5199 if (dest_k != NULL && dest_k->is_array_klass()) {
5200 could_have_dest = true;
5201 }
5202 }
5203
5204 if (could_have_src && could_have_dest) {
5205 // This is going to pay off so emit the required guards
5206 if (!has_src) {
5207 src = maybe_cast_profiled_obj(src, src_k, true);
5208 src_type = _gvn.type(src);
5209 top_src = src_type->isa_aryptr();
5210 has_src = (top_src != NULL && top_src->klass() != NULL);
5211 src_spec = true;
5212 }
5213 if (!has_dest) {
5214 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5215 dest_type = _gvn.type(dest);
5216 top_dest = dest_type->isa_aryptr();
5217 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
5218 dest_spec = true;
5219 }
5220 }
5221 }
5222
5223 if (has_src && has_dest && can_emit_guards) {
5224 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
5225 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
5226 if (src_elem == T_ARRAY) src_elem = T_OBJECT;
5227 if (dest_elem == T_ARRAY) dest_elem = T_OBJECT;
5228
5229 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5230 // If both arrays are object arrays then having the exact types
5231 // for both will remove the need for a subtype check at runtime
5232 // before the call and may make it possible to pick a faster copy
5233 // routine (without a subtype check on every element)
5234 // Do we have the exact type of src?
5235 bool could_have_src = src_spec;
5236 // Do we have the exact type of dest?
5237 bool could_have_dest = dest_spec;
5238 ciKlass* src_k = top_src->klass();
5239 ciKlass* dest_k = top_dest->klass();
5240 if (!src_spec) {
5241 src_k = src_type->speculative_type_not_null();
5242 if (src_k != NULL && src_k->is_array_klass()) {
5243 could_have_src = true;
5244 }
5245 }
5246 if (!dest_spec) {
5247 dest_k = dest_type->speculative_type_not_null();
5248 if (dest_k != NULL && dest_k->is_array_klass()) {
5249 could_have_dest = true;
5250 }
5251 }
5252 if (could_have_src && could_have_dest) {
5253 // If we can have both exact types, emit the missing guards
5254 if (could_have_src && !src_spec) {
5255 src = maybe_cast_profiled_obj(src, src_k, true);
5256 }
5257 if (could_have_dest && !dest_spec) {
5258 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5259 }
5260 }
5261 }
5262 }
5263
5264 ciMethod* trap_method = method();
5265 int trap_bci = bci();
5266 if (saved_jvms != NULL) {
5267 trap_method = alloc->jvms()->method();
5268 trap_bci = alloc->jvms()->bci();
5269 }
5270
5271 bool negative_length_guard_generated = false;
5272
5273 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5274 can_emit_guards &&
5275 !src->is_top() && !dest->is_top()) {
5276 // validate arguments: enables transformation the ArrayCopyNode
5277 validated = true;
5278
5279 RegionNode* slow_region = new RegionNode(1);
5280 record_for_igvn(slow_region);
5281
5282 // (1) src and dest are arrays.
5283 generate_non_array_guard(load_object_klass(src), slow_region);
5284 generate_non_array_guard(load_object_klass(dest), slow_region);
5285
5286 // (2) src and dest arrays must have elements of the same BasicType
5287 // done at macro expansion or at Ideal transformation time
5288
5289 // (4) src_offset must not be negative.
5290 generate_negative_guard(src_offset, slow_region);
5291
5292 // (5) dest_offset must not be negative.
5293 generate_negative_guard(dest_offset, slow_region);
5294
5295 // (7) src_offset + length must not exceed length of src.
5296 generate_limit_guard(src_offset, length,
5297 load_array_length(src),
5298 slow_region);
5299
5300 // (8) dest_offset + length must not exceed length of dest.
5301 generate_limit_guard(dest_offset, length,
5302 load_array_length(dest),
5303 slow_region);
5304
5305 // (6) length must not be negative.
5306 // This is also checked in generate_arraycopy() during macro expansion, but
5307 // we also have to check it here for the case where the ArrayCopyNode will
5308 // be eliminated by Escape Analysis.
5309 if (EliminateAllocations) {
5310 generate_negative_guard(length, slow_region);
5311 negative_length_guard_generated = true;
5312 }
5313
5314 // (9) each element of an oop array must be assignable
5315 Node* src_klass = load_object_klass(src);
5316 Node* dest_klass = load_object_klass(dest);
5317 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
5318
5319 if (not_subtype_ctrl != top()) {
5320 PreserveJVMState pjvms(this);
5321 set_control(not_subtype_ctrl);
5322 uncommon_trap(Deoptimization::Reason_intrinsic,
5323 Deoptimization::Action_make_not_entrant);
5324 assert(stopped(), "Should be stopped");
5325 }
5326
5327 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5328 const Type* toop = TypeOopPtr::make_from_klass(dest_klass_t->klass());
5329 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5330 src_type = _gvn.type(src);
5331 top_src = src_type->isa_aryptr();
5332
5333 if (top_dest != NULL && !top_dest->elem()->isa_valuetype() && !top_dest->is_not_flat()) {
5334 generate_valueArray_guard(dest_klass, slow_region);
5335 }
5336
5337 if (top_src != NULL && !top_src->elem()->isa_valuetype() && !top_src->is_not_flat()) {
5338 generate_valueArray_guard(src_klass, slow_region);
5339 }
5340
5341 {
5342 PreserveJVMState pjvms(this);
5343 set_control(_gvn.transform(slow_region));
5344 uncommon_trap(Deoptimization::Reason_intrinsic,
5345 Deoptimization::Action_make_not_entrant);
5346 assert(stopped(), "Should be stopped");
5347 }
5348 }
5349
5350 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
5351
5352 if (stopped()) {
5353 return true;
5354 }
5355
5356 Node* new_src = access_resolve(src, ACCESS_READ);
5357 Node* new_dest = access_resolve(dest, ACCESS_WRITE);
5358
5359 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, new_src, src_offset, new_dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
5360 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5361 // so the compiler has a chance to eliminate them: during macro expansion,
5362 // we have to set their control (CastPP nodes are eliminated).
5363 load_object_klass(src), load_object_klass(dest),
5364 load_array_length(src), load_array_length(dest));
5365
5366 ac->set_arraycopy(validated);
5367
5368 Node* n = _gvn.transform(ac);
5369 if (n == ac) {
5370 ac->connect_outputs(this);
5371 } else {
5372 assert(validated, "shouldn't transform if all arguments not validated");
5373 set_all_memory(n);
5374 }
5375 clear_upper_avx();
5376
5377
5378 return true;
5379 }
5380
5381
5382 // Helper function which determines if an arraycopy immediately follows
5383 // an allocation, with no intervening tests or other escapes for the object.
5384 AllocateArrayNode*
5385 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
5386 RegionNode* slow_region) {
5387 if (stopped()) return NULL; // no fast path
5388 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
5389
5390 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5391 if (alloc == NULL) return NULL;
5392
5393 Node* rawmem = memory(Compile::AliasIdxRaw);
5394 // Is the allocation's memory state untouched?
5395 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5396 // Bail out if there have been raw-memory effects since the allocation.
5397 // (Example: There might have been a call or safepoint.)
5398 return NULL;
5399 }
5400 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5401 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5402 return NULL;
5403 }
5404
5405 // There must be no unexpected observers of this allocation.
5406 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5407 Node* obs = ptr->fast_out(i);
5408 if (obs != this->map()) {
5409 return NULL;
5410 }
5411 }
5412
5413 // This arraycopy must unconditionally follow the allocation of the ptr.
5414 Node* alloc_ctl = ptr->in(0);
5415 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
5416
5417 Node* ctl = control();
5418 while (ctl != alloc_ctl) {
5419 // There may be guards which feed into the slow_region.
5420 // Any other control flow means that we might not get a chance
5421 // to finish initializing the allocated object.
5422 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5423 IfNode* iff = ctl->in(0)->as_If();
5424 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
5425 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5426 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
5427 ctl = iff->in(0); // This test feeds the known slow_region.
5428 continue;
5429 }
5430 // One more try: Various low-level checks bottom out in
5431 // uncommon traps. If the debug-info of the trap omits
5432 // any reference to the allocation, as we've already
5433 // observed, then there can be no objection to the trap.
5434 bool found_trap = false;
5435 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5436 Node* obs = not_ctl->fast_out(j);
5437 if (obs->in(0) == not_ctl && obs->is_Call() &&
5438 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5439 found_trap = true; break;
5440 }
5441 }
5442 if (found_trap) {
5443 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
5444 continue;
5445 }
5446 }
5447 return NULL;
5448 }
5449
5450 // If we get this far, we have an allocation which immediately
5451 // precedes the arraycopy, and we can take over zeroing the new object.
5452 // The arraycopy will finish the initialization, and provide
5453 // a new control state to which we will anchor the destination pointer.
5454
5455 return alloc;
5456 }
5457
5458 //-------------inline_encodeISOArray-----------------------------------
5459 // encode char[] to byte[] in ISO_8859_1
5460 bool LibraryCallKit::inline_encodeISOArray() {
5461 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5462 // no receiver since it is static method
5463 Node *src = argument(0);
5464 Node *src_offset = argument(1);
5465 Node *dst = argument(2);
5466 Node *dst_offset = argument(3);
5467 Node *length = argument(4);
5468
5469 src = must_be_not_null(src, true);
5470 dst = must_be_not_null(dst, true);
5471
5472 src = access_resolve(src, ACCESS_READ);
5473 dst = access_resolve(dst, ACCESS_WRITE);
5474
5475 const Type* src_type = src->Value(&_gvn);
5476 const Type* dst_type = dst->Value(&_gvn);
5477 const TypeAryPtr* top_src = src_type->isa_aryptr();
5478 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5479 if (top_src == NULL || top_src->klass() == NULL ||
5480 top_dest == NULL || top_dest->klass() == NULL) {
5481 // failed array check
5482 return false;
5483 }
5484
5485 // Figure out the size and type of the elements we will be copying.
5486 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5487 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5488 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5489 return false;
5490 }
5491
5492 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5493 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5494 // 'src_start' points to src array + scaled offset
5495 // 'dst_start' points to dst array + scaled offset
5496
5497 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5498 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
5499 enc = _gvn.transform(enc);
5500 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5501 set_memory(res_mem, mtype);
5502 set_result(enc);
5503 clear_upper_avx();
5504
5505 return true;
5506 }
5507
5508 //-------------inline_multiplyToLen-----------------------------------
5509 bool LibraryCallKit::inline_multiplyToLen() {
5510 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5511
5512 address stubAddr = StubRoutines::multiplyToLen();
5513 if (stubAddr == NULL) {
5514 return false; // Intrinsic's stub is not implemented on this platform
5515 }
5516 const char* stubName = "multiplyToLen";
5517
5518 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5519
5520 // no receiver because it is a static method
5521 Node* x = argument(0);
5522 Node* xlen = argument(1);
5523 Node* y = argument(2);
5524 Node* ylen = argument(3);
5525 Node* z = argument(4);
5526
5527 x = must_be_not_null(x, true);
5528 y = must_be_not_null(y, true);
5529
5530 x = access_resolve(x, ACCESS_READ);
5531 y = access_resolve(y, ACCESS_READ);
5532 z = access_resolve(z, ACCESS_WRITE);
5533
5534 const Type* x_type = x->Value(&_gvn);
5535 const Type* y_type = y->Value(&_gvn);
5536 const TypeAryPtr* top_x = x_type->isa_aryptr();
5537 const TypeAryPtr* top_y = y_type->isa_aryptr();
5538 if (top_x == NULL || top_x->klass() == NULL ||
5539 top_y == NULL || top_y->klass() == NULL) {
5540 // failed array check
5541 return false;
5542 }
5543
5544 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5545 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5546 if (x_elem != T_INT || y_elem != T_INT) {
5547 return false;
5548 }
5549
5550 // Set the original stack and the reexecute bit for the interpreter to reexecute
5551 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5552 // on the return from z array allocation in runtime.
5553 { PreserveReexecuteState preexecs(this);
5554 jvms()->set_should_reexecute(true);
5555
5556 Node* x_start = array_element_address(x, intcon(0), x_elem);
5557 Node* y_start = array_element_address(y, intcon(0), y_elem);
5558 // 'x_start' points to x array + scaled xlen
5559 // 'y_start' points to y array + scaled ylen
5560
5561 // Allocate the result array
5562 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5563 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5564 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5565
5566 IdealKit ideal(this);
5567
5568 #define __ ideal.
5569 Node* one = __ ConI(1);
5570 Node* zero = __ ConI(0);
5571 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5572 __ set(need_alloc, zero);
5573 __ set(z_alloc, z);
5574 __ if_then(z, BoolTest::eq, null()); {
5575 __ increment (need_alloc, one);
5576 } __ else_(); {
5577 // Update graphKit memory and control from IdealKit.
5578 sync_kit(ideal);
5579 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5580 cast->init_req(0, control());
5581 _gvn.set_type(cast, cast->bottom_type());
5582 C->record_for_igvn(cast);
5583
5584 Node* zlen_arg = load_array_length(cast);
5585 // Update IdealKit memory and control from graphKit.
5586 __ sync_kit(this);
5587 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5588 __ increment (need_alloc, one);
5589 } __ end_if();
5590 } __ end_if();
5591
5592 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5593 // Update graphKit memory and control from IdealKit.
5594 sync_kit(ideal);
5595 Node * narr = new_array(klass_node, zlen, 1);
5596 // Update IdealKit memory and control from graphKit.
5597 __ sync_kit(this);
5598 __ set(z_alloc, narr);
5599 } __ end_if();
5600
5601 sync_kit(ideal);
5602 z = __ value(z_alloc);
5603 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5604 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5605 // Final sync IdealKit and GraphKit.
5606 final_sync(ideal);
5607 #undef __
5608
5609 Node* z_start = array_element_address(z, intcon(0), T_INT);
5610
5611 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5612 OptoRuntime::multiplyToLen_Type(),
5613 stubAddr, stubName, TypePtr::BOTTOM,
5614 x_start, xlen, y_start, ylen, z_start, zlen);
5615 } // original reexecute is set back here
5616
5617 C->set_has_split_ifs(true); // Has chance for split-if optimization
5618 set_result(z);
5619 return true;
5620 }
5621
5622 //-------------inline_squareToLen------------------------------------
5623 bool LibraryCallKit::inline_squareToLen() {
5624 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5625
5626 address stubAddr = StubRoutines::squareToLen();
5627 if (stubAddr == NULL) {
5628 return false; // Intrinsic's stub is not implemented on this platform
5629 }
5630 const char* stubName = "squareToLen";
5631
5632 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5633
5634 Node* x = argument(0);
5635 Node* len = argument(1);
5636 Node* z = argument(2);
5637 Node* zlen = argument(3);
5638
5639 x = must_be_not_null(x, true);
5640 z = must_be_not_null(z, true);
5641
5642 x = access_resolve(x, ACCESS_READ);
5643 z = access_resolve(z, ACCESS_WRITE);
5644
5645 const Type* x_type = x->Value(&_gvn);
5646 const Type* z_type = z->Value(&_gvn);
5647 const TypeAryPtr* top_x = x_type->isa_aryptr();
5648 const TypeAryPtr* top_z = z_type->isa_aryptr();
5649 if (top_x == NULL || top_x->klass() == NULL ||
5650 top_z == NULL || top_z->klass() == NULL) {
5651 // failed array check
5652 return false;
5653 }
5654
5655 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5656 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5657 if (x_elem != T_INT || z_elem != T_INT) {
5658 return false;
5659 }
5660
5661
5662 Node* x_start = array_element_address(x, intcon(0), x_elem);
5663 Node* z_start = array_element_address(z, intcon(0), z_elem);
5664
5665 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5666 OptoRuntime::squareToLen_Type(),
5667 stubAddr, stubName, TypePtr::BOTTOM,
5668 x_start, len, z_start, zlen);
5669
5670 set_result(z);
5671 return true;
5672 }
5673
5674 //-------------inline_mulAdd------------------------------------------
5675 bool LibraryCallKit::inline_mulAdd() {
5676 assert(UseMulAddIntrinsic, "not implemented on this platform");
5677
5678 address stubAddr = StubRoutines::mulAdd();
5679 if (stubAddr == NULL) {
5680 return false; // Intrinsic's stub is not implemented on this platform
5681 }
5682 const char* stubName = "mulAdd";
5683
5684 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5685
5686 Node* out = argument(0);
5687 Node* in = argument(1);
5688 Node* offset = argument(2);
5689 Node* len = argument(3);
5690 Node* k = argument(4);
5691
5692 out = must_be_not_null(out, true);
5693
5694 in = access_resolve(in, ACCESS_READ);
5695 out = access_resolve(out, ACCESS_WRITE);
5696
5697 const Type* out_type = out->Value(&_gvn);
5698 const Type* in_type = in->Value(&_gvn);
5699 const TypeAryPtr* top_out = out_type->isa_aryptr();
5700 const TypeAryPtr* top_in = in_type->isa_aryptr();
5701 if (top_out == NULL || top_out->klass() == NULL ||
5702 top_in == NULL || top_in->klass() == NULL) {
5703 // failed array check
5704 return false;
5705 }
5706
5707 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5708 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5709 if (out_elem != T_INT || in_elem != T_INT) {
5710 return false;
5711 }
5712
5713 Node* outlen = load_array_length(out);
5714 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5715 Node* out_start = array_element_address(out, intcon(0), out_elem);
5716 Node* in_start = array_element_address(in, intcon(0), in_elem);
5717
5718 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5719 OptoRuntime::mulAdd_Type(),
5720 stubAddr, stubName, TypePtr::BOTTOM,
5721 out_start,in_start, new_offset, len, k);
5722 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5723 set_result(result);
5724 return true;
5725 }
5726
5727 //-------------inline_montgomeryMultiply-----------------------------------
5728 bool LibraryCallKit::inline_montgomeryMultiply() {
5729 address stubAddr = StubRoutines::montgomeryMultiply();
5730 if (stubAddr == NULL) {
5731 return false; // Intrinsic's stub is not implemented on this platform
5732 }
5733
5734 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5735 const char* stubName = "montgomery_multiply";
5736
5737 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5738
5739 Node* a = argument(0);
5740 Node* b = argument(1);
5741 Node* n = argument(2);
5742 Node* len = argument(3);
5743 Node* inv = argument(4);
5744 Node* m = argument(6);
5745
5746 a = access_resolve(a, ACCESS_READ);
5747 b = access_resolve(b, ACCESS_READ);
5748 n = access_resolve(n, ACCESS_READ);
5749 m = access_resolve(m, ACCESS_WRITE);
5750
5751 const Type* a_type = a->Value(&_gvn);
5752 const TypeAryPtr* top_a = a_type->isa_aryptr();
5753 const Type* b_type = b->Value(&_gvn);
5754 const TypeAryPtr* top_b = b_type->isa_aryptr();
5755 const Type* n_type = a->Value(&_gvn);
5756 const TypeAryPtr* top_n = n_type->isa_aryptr();
5757 const Type* m_type = a->Value(&_gvn);
5758 const TypeAryPtr* top_m = m_type->isa_aryptr();
5759 if (top_a == NULL || top_a->klass() == NULL ||
5760 top_b == NULL || top_b->klass() == NULL ||
5761 top_n == NULL || top_n->klass() == NULL ||
5762 top_m == NULL || top_m->klass() == NULL) {
5763 // failed array check
5764 return false;
5765 }
5766
5767 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5768 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5769 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5770 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5771 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5772 return false;
5773 }
5774
5775 // Make the call
5776 {
5777 Node* a_start = array_element_address(a, intcon(0), a_elem);
5778 Node* b_start = array_element_address(b, intcon(0), b_elem);
5779 Node* n_start = array_element_address(n, intcon(0), n_elem);
5780 Node* m_start = array_element_address(m, intcon(0), m_elem);
5781
5782 Node* call = make_runtime_call(RC_LEAF,
5783 OptoRuntime::montgomeryMultiply_Type(),
5784 stubAddr, stubName, TypePtr::BOTTOM,
5785 a_start, b_start, n_start, len, inv, top(),
5786 m_start);
5787 set_result(m);
5788 }
5789
5790 return true;
5791 }
5792
5793 bool LibraryCallKit::inline_montgomerySquare() {
5794 address stubAddr = StubRoutines::montgomerySquare();
5795 if (stubAddr == NULL) {
5796 return false; // Intrinsic's stub is not implemented on this platform
5797 }
5798
5799 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5800 const char* stubName = "montgomery_square";
5801
5802 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5803
5804 Node* a = argument(0);
5805 Node* n = argument(1);
5806 Node* len = argument(2);
5807 Node* inv = argument(3);
5808 Node* m = argument(5);
5809
5810 a = access_resolve(a, ACCESS_READ);
5811 n = access_resolve(n, ACCESS_READ);
5812 m = access_resolve(m, ACCESS_WRITE);
5813
5814 const Type* a_type = a->Value(&_gvn);
5815 const TypeAryPtr* top_a = a_type->isa_aryptr();
5816 const Type* n_type = a->Value(&_gvn);
5817 const TypeAryPtr* top_n = n_type->isa_aryptr();
5818 const Type* m_type = a->Value(&_gvn);
5819 const TypeAryPtr* top_m = m_type->isa_aryptr();
5820 if (top_a == NULL || top_a->klass() == NULL ||
5821 top_n == NULL || top_n->klass() == NULL ||
5822 top_m == NULL || top_m->klass() == NULL) {
5823 // failed array check
5824 return false;
5825 }
5826
5827 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5828 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5829 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5830 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5831 return false;
5832 }
5833
5834 // Make the call
5835 {
5836 Node* a_start = array_element_address(a, intcon(0), a_elem);
5837 Node* n_start = array_element_address(n, intcon(0), n_elem);
5838 Node* m_start = array_element_address(m, intcon(0), m_elem);
5839
5840 Node* call = make_runtime_call(RC_LEAF,
5841 OptoRuntime::montgomerySquare_Type(),
5842 stubAddr, stubName, TypePtr::BOTTOM,
5843 a_start, n_start, len, inv, top(),
5844 m_start);
5845 set_result(m);
5846 }
5847
5848 return true;
5849 }
5850
5851 //-------------inline_vectorizedMismatch------------------------------
5852 bool LibraryCallKit::inline_vectorizedMismatch() {
5853 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5854
5855 address stubAddr = StubRoutines::vectorizedMismatch();
5856 if (stubAddr == NULL) {
5857 return false; // Intrinsic's stub is not implemented on this platform
5858 }
5859 const char* stubName = "vectorizedMismatch";
5860 int size_l = callee()->signature()->size();
5861 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5862
5863 Node* obja = argument(0);
5864 Node* aoffset = argument(1);
5865 Node* objb = argument(3);
5866 Node* boffset = argument(4);
5867 Node* length = argument(6);
5868 Node* scale = argument(7);
5869
5870 const Type* a_type = obja->Value(&_gvn);
5871 const Type* b_type = objb->Value(&_gvn);
5872 const TypeAryPtr* top_a = a_type->isa_aryptr();
5873 const TypeAryPtr* top_b = b_type->isa_aryptr();
5874 if (top_a == NULL || top_a->klass() == NULL ||
5875 top_b == NULL || top_b->klass() == NULL) {
5876 // failed array check
5877 return false;
5878 }
5879
5880 Node* call;
5881 jvms()->set_should_reexecute(true);
5882
5883 obja = access_resolve(obja, ACCESS_READ);
5884 objb = access_resolve(objb, ACCESS_READ);
5885 Node* obja_adr = make_unsafe_address(obja, aoffset, ACCESS_READ);
5886 Node* objb_adr = make_unsafe_address(objb, boffset, ACCESS_READ);
5887
5888 call = make_runtime_call(RC_LEAF,
5889 OptoRuntime::vectorizedMismatch_Type(),
5890 stubAddr, stubName, TypePtr::BOTTOM,
5891 obja_adr, objb_adr, length, scale);
5892
5893 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5894 set_result(result);
5895 return true;
5896 }
5897
5898 /**
5899 * Calculate CRC32 for byte.
5900 * int java.util.zip.CRC32.update(int crc, int b)
5901 */
5902 bool LibraryCallKit::inline_updateCRC32() {
5903 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5904 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5905 // no receiver since it is static method
5906 Node* crc = argument(0); // type: int
5907 Node* b = argument(1); // type: int
5908
5909 /*
5910 * int c = ~ crc;
5911 * b = timesXtoThe32[(b ^ c) & 0xFF];
5912 * b = b ^ (c >>> 8);
5913 * crc = ~b;
5914 */
5915
5916 Node* M1 = intcon(-1);
5917 crc = _gvn.transform(new XorINode(crc, M1));
5918 Node* result = _gvn.transform(new XorINode(crc, b));
5919 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5920
5921 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5922 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5923 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5924 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5925
5926 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5927 result = _gvn.transform(new XorINode(crc, result));
5928 result = _gvn.transform(new XorINode(result, M1));
5929 set_result(result);
5930 return true;
5931 }
5932
5933 /**
5934 * Calculate CRC32 for byte[] array.
5935 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5936 */
5937 bool LibraryCallKit::inline_updateBytesCRC32() {
5938 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5939 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5940 // no receiver since it is static method
5941 Node* crc = argument(0); // type: int
5942 Node* src = argument(1); // type: oop
5943 Node* offset = argument(2); // type: int
5944 Node* length = argument(3); // type: int
5945
5946 const Type* src_type = src->Value(&_gvn);
5947 const TypeAryPtr* top_src = src_type->isa_aryptr();
5948 if (top_src == NULL || top_src->klass() == NULL) {
5949 // failed array check
5950 return false;
5951 }
5952
5953 // Figure out the size and type of the elements we will be copying.
5954 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5955 if (src_elem != T_BYTE) {
5956 return false;
5957 }
5958
5959 // 'src_start' points to src array + scaled offset
5960 src = must_be_not_null(src, true);
5961 src = access_resolve(src, ACCESS_READ);
5962 Node* src_start = array_element_address(src, offset, src_elem);
5963
5964 // We assume that range check is done by caller.
5965 // TODO: generate range check (offset+length < src.length) in debug VM.
5966
5967 // Call the stub.
5968 address stubAddr = StubRoutines::updateBytesCRC32();
5969 const char *stubName = "updateBytesCRC32";
5970
5971 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5972 stubAddr, stubName, TypePtr::BOTTOM,
5973 crc, src_start, length);
5974 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5975 set_result(result);
5976 return true;
5977 }
5978
5979 /**
5980 * Calculate CRC32 for ByteBuffer.
5981 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5982 */
5983 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5984 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5985 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5986 // no receiver since it is static method
5987 Node* crc = argument(0); // type: int
5988 Node* src = argument(1); // type: long
5989 Node* offset = argument(3); // type: int
5990 Node* length = argument(4); // type: int
5991
5992 src = ConvL2X(src); // adjust Java long to machine word
5993 Node* base = _gvn.transform(new CastX2PNode(src));
5994 offset = ConvI2X(offset);
5995
5996 // 'src_start' points to src array + scaled offset
5997 Node* src_start = basic_plus_adr(top(), base, offset);
5998
5999 // Call the stub.
6000 address stubAddr = StubRoutines::updateBytesCRC32();
6001 const char *stubName = "updateBytesCRC32";
6002
6003 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6004 stubAddr, stubName, TypePtr::BOTTOM,
6005 crc, src_start, length);
6006 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6007 set_result(result);
6008 return true;
6009 }
6010
6011 //------------------------------get_table_from_crc32c_class-----------------------
6012 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6013 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
6014 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
6015
6016 return table;
6017 }
6018
6019 //------------------------------inline_updateBytesCRC32C-----------------------
6020 //
6021 // Calculate CRC32C for byte[] array.
6022 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6023 //
6024 bool LibraryCallKit::inline_updateBytesCRC32C() {
6025 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6026 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6027 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6028 // no receiver since it is a static method
6029 Node* crc = argument(0); // type: int
6030 Node* src = argument(1); // type: oop
6031 Node* offset = argument(2); // type: int
6032 Node* end = argument(3); // type: int
6033
6034 Node* length = _gvn.transform(new SubINode(end, offset));
6035
6036 const Type* src_type = src->Value(&_gvn);
6037 const TypeAryPtr* top_src = src_type->isa_aryptr();
6038 if (top_src == NULL || top_src->klass() == NULL) {
6039 // failed array check
6040 return false;
6041 }
6042
6043 // Figure out the size and type of the elements we will be copying.
6044 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6045 if (src_elem != T_BYTE) {
6046 return false;
6047 }
6048
6049 // 'src_start' points to src array + scaled offset
6050 src = must_be_not_null(src, true);
6051 src = access_resolve(src, ACCESS_READ);
6052 Node* src_start = array_element_address(src, offset, src_elem);
6053
6054 // static final int[] byteTable in class CRC32C
6055 Node* table = get_table_from_crc32c_class(callee()->holder());
6056 table = must_be_not_null(table, true);
6057 table = access_resolve(table, ACCESS_READ);
6058 Node* table_start = array_element_address(table, intcon(0), T_INT);
6059
6060 // We assume that range check is done by caller.
6061 // TODO: generate range check (offset+length < src.length) in debug VM.
6062
6063 // Call the stub.
6064 address stubAddr = StubRoutines::updateBytesCRC32C();
6065 const char *stubName = "updateBytesCRC32C";
6066
6067 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6068 stubAddr, stubName, TypePtr::BOTTOM,
6069 crc, src_start, length, table_start);
6070 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6071 set_result(result);
6072 return true;
6073 }
6074
6075 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6076 //
6077 // Calculate CRC32C for DirectByteBuffer.
6078 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6079 //
6080 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6081 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6082 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6083 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6084 // no receiver since it is a static method
6085 Node* crc = argument(0); // type: int
6086 Node* src = argument(1); // type: long
6087 Node* offset = argument(3); // type: int
6088 Node* end = argument(4); // type: int
6089
6090 Node* length = _gvn.transform(new SubINode(end, offset));
6091
6092 src = ConvL2X(src); // adjust Java long to machine word
6093 Node* base = _gvn.transform(new CastX2PNode(src));
6094 offset = ConvI2X(offset);
6095
6096 // 'src_start' points to src array + scaled offset
6097 Node* src_start = basic_plus_adr(top(), base, offset);
6098
6099 // static final int[] byteTable in class CRC32C
6100 Node* table = get_table_from_crc32c_class(callee()->holder());
6101 table = must_be_not_null(table, true);
6102 table = access_resolve(table, ACCESS_READ);
6103 Node* table_start = array_element_address(table, intcon(0), T_INT);
6104
6105 // Call the stub.
6106 address stubAddr = StubRoutines::updateBytesCRC32C();
6107 const char *stubName = "updateBytesCRC32C";
6108
6109 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6110 stubAddr, stubName, TypePtr::BOTTOM,
6111 crc, src_start, length, table_start);
6112 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6113 set_result(result);
6114 return true;
6115 }
6116
6117 //------------------------------inline_updateBytesAdler32----------------------
6118 //
6119 // Calculate Adler32 checksum for byte[] array.
6120 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6121 //
6122 bool LibraryCallKit::inline_updateBytesAdler32() {
6123 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
6124 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6125 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6126 // no receiver since it is static method
6127 Node* crc = argument(0); // type: int
6128 Node* src = argument(1); // type: oop
6129 Node* offset = argument(2); // type: int
6130 Node* length = argument(3); // type: int
6131
6132 const Type* src_type = src->Value(&_gvn);
6133 const TypeAryPtr* top_src = src_type->isa_aryptr();
6134 if (top_src == NULL || top_src->klass() == NULL) {
6135 // failed array check
6136 return false;
6137 }
6138
6139 // Figure out the size and type of the elements we will be copying.
6140 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6141 if (src_elem != T_BYTE) {
6142 return false;
6143 }
6144
6145 // 'src_start' points to src array + scaled offset
6146 src = access_resolve(src, ACCESS_READ);
6147 Node* src_start = array_element_address(src, offset, src_elem);
6148
6149 // We assume that range check is done by caller.
6150 // TODO: generate range check (offset+length < src.length) in debug VM.
6151
6152 // Call the stub.
6153 address stubAddr = StubRoutines::updateBytesAdler32();
6154 const char *stubName = "updateBytesAdler32";
6155
6156 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6157 stubAddr, stubName, TypePtr::BOTTOM,
6158 crc, src_start, length);
6159 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6160 set_result(result);
6161 return true;
6162 }
6163
6164 //------------------------------inline_updateByteBufferAdler32---------------
6165 //
6166 // Calculate Adler32 checksum for DirectByteBuffer.
6167 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6168 //
6169 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6170 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
6171 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6172 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6173 // no receiver since it is static method
6174 Node* crc = argument(0); // type: int
6175 Node* src = argument(1); // type: long
6176 Node* offset = argument(3); // type: int
6177 Node* length = argument(4); // type: int
6178
6179 src = ConvL2X(src); // adjust Java long to machine word
6180 Node* base = _gvn.transform(new CastX2PNode(src));
6181 offset = ConvI2X(offset);
6182
6183 // 'src_start' points to src array + scaled offset
6184 Node* src_start = basic_plus_adr(top(), base, offset);
6185
6186 // Call the stub.
6187 address stubAddr = StubRoutines::updateBytesAdler32();
6188 const char *stubName = "updateBytesAdler32";
6189
6190 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6191 stubAddr, stubName, TypePtr::BOTTOM,
6192 crc, src_start, length);
6193
6194 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6195 set_result(result);
6196 return true;
6197 }
6198
6199 //----------------------------inline_reference_get----------------------------
6200 // public T java.lang.ref.Reference.get();
6201 bool LibraryCallKit::inline_reference_get() {
6202 const int referent_offset = java_lang_ref_Reference::referent_offset;
6203 guarantee(referent_offset > 0, "should have already been set");
6204
6205 // Get the argument:
6206 Node* reference_obj = null_check_receiver();
6207 if (stopped()) return true;
6208
6209 const TypeInstPtr* tinst = _gvn.type(reference_obj)->isa_instptr();
6210 assert(tinst != NULL, "obj is null");
6211 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6212 ciInstanceKlass* referenceKlass = tinst->klass()->as_instance_klass();
6213 ciField* field = referenceKlass->get_field_by_name(ciSymbol::make("referent"),
6214 ciSymbol::make("Ljava/lang/Object;"),
6215 false);
6216 assert (field != NULL, "undefined field");
6217
6218 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
6219 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6220
6221 ciInstanceKlass* klass = env()->Object_klass();
6222 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
6223
6224 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6225 Node* result = access_load_at(reference_obj, adr, adr_type, object_type, T_OBJECT, decorators);
6226 // Add memory barrier to prevent commoning reads from this field
6227 // across safepoint since GC can change its value.
6228 insert_mem_bar(Op_MemBarCPUOrder);
6229
6230 set_result(result);
6231 return true;
6232 }
6233
6234
6235 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6236 bool is_exact=true, bool is_static=false,
6237 ciInstanceKlass * fromKls=NULL) {
6238 if (fromKls == NULL) {
6239 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6240 assert(tinst != NULL, "obj is null");
6241 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6242 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6243 fromKls = tinst->klass()->as_instance_klass();
6244 } else {
6245 assert(is_static, "only for static field access");
6246 }
6247 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6248 ciSymbol::make(fieldTypeString),
6249 is_static);
6250
6251 assert (field != NULL, "undefined field");
6252 if (field == NULL) return (Node *) NULL;
6253
6254 if (is_static) {
6255 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6256 fromObj = makecon(tip);
6257 }
6258
6259 // Next code copied from Parse::do_get_xxx():
6260
6261 // Compute address and memory type.
6262 int offset = field->offset_in_bytes();
6263 bool is_vol = field->is_volatile();
6264 ciType* field_klass = field->type();
6265 assert(field_klass->is_loaded(), "should be loaded");
6266 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6267 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6268 BasicType bt = field->layout_type();
6269
6270 // Build the resultant type of the load
6271 const Type *type;
6272 if (bt == T_OBJECT) {
6273 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6274 } else {
6275 type = Type::get_const_basic_type(bt);
6276 }
6277
6278 DecoratorSet decorators = IN_HEAP;
6279
6280 if (is_vol) {
6281 decorators |= MO_SEQ_CST;
6282 }
6283
6284 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
6285 }
6286
6287 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6288 bool is_exact = true, bool is_static = false,
6289 ciInstanceKlass * fromKls = NULL) {
6290 if (fromKls == NULL) {
6291 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6292 assert(tinst != NULL, "obj is null");
6293 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6294 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6295 fromKls = tinst->klass()->as_instance_klass();
6296 }
6297 else {
6298 assert(is_static, "only for static field access");
6299 }
6300 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6301 ciSymbol::make(fieldTypeString),
6302 is_static);
6303
6304 assert(field != NULL, "undefined field");
6305 assert(!field->is_volatile(), "not defined for volatile fields");
6306
6307 if (is_static) {
6308 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6309 fromObj = makecon(tip);
6310 }
6311
6312 // Next code copied from Parse::do_get_xxx():
6313
6314 // Compute address and memory type.
6315 int offset = field->offset_in_bytes();
6316 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6317
6318 return adr;
6319 }
6320
6321 //------------------------------inline_aescrypt_Block-----------------------
6322 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6323 address stubAddr = NULL;
6324 const char *stubName;
6325 assert(UseAES, "need AES instruction support");
6326
6327 switch(id) {
6328 case vmIntrinsics::_aescrypt_encryptBlock:
6329 stubAddr = StubRoutines::aescrypt_encryptBlock();
6330 stubName = "aescrypt_encryptBlock";
6331 break;
6332 case vmIntrinsics::_aescrypt_decryptBlock:
6333 stubAddr = StubRoutines::aescrypt_decryptBlock();
6334 stubName = "aescrypt_decryptBlock";
6335 break;
6336 default:
6337 break;
6338 }
6339 if (stubAddr == NULL) return false;
6340
6341 Node* aescrypt_object = argument(0);
6342 Node* src = argument(1);
6343 Node* src_offset = argument(2);
6344 Node* dest = argument(3);
6345 Node* dest_offset = argument(4);
6346
6347 src = must_be_not_null(src, true);
6348 dest = must_be_not_null(dest, true);
6349
6350 src = access_resolve(src, ACCESS_READ);
6351 dest = access_resolve(dest, ACCESS_WRITE);
6352
6353 // (1) src and dest are arrays.
6354 const Type* src_type = src->Value(&_gvn);
6355 const Type* dest_type = dest->Value(&_gvn);
6356 const TypeAryPtr* top_src = src_type->isa_aryptr();
6357 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6358 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6359
6360 // for the quick and dirty code we will skip all the checks.
6361 // we are just trying to get the call to be generated.
6362 Node* src_start = src;
6363 Node* dest_start = dest;
6364 if (src_offset != NULL || dest_offset != NULL) {
6365 assert(src_offset != NULL && dest_offset != NULL, "");
6366 src_start = array_element_address(src, src_offset, T_BYTE);
6367 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6368 }
6369
6370 // now need to get the start of its expanded key array
6371 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6372 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6373 if (k_start == NULL) return false;
6374
6375 if (Matcher::pass_original_key_for_aes()) {
6376 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6377 // compatibility issues between Java key expansion and SPARC crypto instructions
6378 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6379 if (original_k_start == NULL) return false;
6380
6381 // Call the stub.
6382 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6383 stubAddr, stubName, TypePtr::BOTTOM,
6384 src_start, dest_start, k_start, original_k_start);
6385 } else {
6386 // Call the stub.
6387 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6388 stubAddr, stubName, TypePtr::BOTTOM,
6389 src_start, dest_start, k_start);
6390 }
6391
6392 return true;
6393 }
6394
6395 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6396 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6397 address stubAddr = NULL;
6398 const char *stubName = NULL;
6399
6400 assert(UseAES, "need AES instruction support");
6401
6402 switch(id) {
6403 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6404 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6405 stubName = "cipherBlockChaining_encryptAESCrypt";
6406 break;
6407 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6408 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6409 stubName = "cipherBlockChaining_decryptAESCrypt";
6410 break;
6411 default:
6412 break;
6413 }
6414 if (stubAddr == NULL) return false;
6415
6416 Node* cipherBlockChaining_object = argument(0);
6417 Node* src = argument(1);
6418 Node* src_offset = argument(2);
6419 Node* len = argument(3);
6420 Node* dest = argument(4);
6421 Node* dest_offset = argument(5);
6422
6423 src = must_be_not_null(src, false);
6424 dest = must_be_not_null(dest, false);
6425
6426 src = access_resolve(src, ACCESS_READ);
6427 dest = access_resolve(dest, ACCESS_WRITE);
6428
6429 // (1) src and dest are arrays.
6430 const Type* src_type = src->Value(&_gvn);
6431 const Type* dest_type = dest->Value(&_gvn);
6432 const TypeAryPtr* top_src = src_type->isa_aryptr();
6433 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6434 assert (top_src != NULL && top_src->klass() != NULL
6435 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6436
6437 // checks are the responsibility of the caller
6438 Node* src_start = src;
6439 Node* dest_start = dest;
6440 if (src_offset != NULL || dest_offset != NULL) {
6441 assert(src_offset != NULL && dest_offset != NULL, "");
6442 src_start = array_element_address(src, src_offset, T_BYTE);
6443 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6444 }
6445
6446 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6447 // (because of the predicated logic executed earlier).
6448 // so we cast it here safely.
6449 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6450
6451 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6452 if (embeddedCipherObj == NULL) return false;
6453
6454 // cast it to what we know it will be at runtime
6455 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6456 assert(tinst != NULL, "CBC obj is null");
6457 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
6458 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6459 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6460
6461 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6462 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6463 const TypeOopPtr* xtype = aklass->as_instance_type();
6464 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6465 aescrypt_object = _gvn.transform(aescrypt_object);
6466
6467 // we need to get the start of the aescrypt_object's expanded key array
6468 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6469 if (k_start == NULL) return false;
6470
6471 // similarly, get the start address of the r vector
6472 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
6473 if (objRvec == NULL) return false;
6474 objRvec = access_resolve(objRvec, ACCESS_WRITE);
6475 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6476
6477 Node* cbcCrypt;
6478 if (Matcher::pass_original_key_for_aes()) {
6479 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6480 // compatibility issues between Java key expansion and SPARC crypto instructions
6481 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6482 if (original_k_start == NULL) return false;
6483
6484 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
6485 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6486 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6487 stubAddr, stubName, TypePtr::BOTTOM,
6488 src_start, dest_start, k_start, r_start, len, original_k_start);
6489 } else {
6490 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6491 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6492 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6493 stubAddr, stubName, TypePtr::BOTTOM,
6494 src_start, dest_start, k_start, r_start, len);
6495 }
6496
6497 // return cipher length (int)
6498 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6499 set_result(retvalue);
6500 return true;
6501 }
6502
6503 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
6504 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
6505 address stubAddr = NULL;
6506 const char *stubName = NULL;
6507
6508 assert(UseAES, "need AES instruction support");
6509
6510 switch (id) {
6511 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
6512 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
6513 stubName = "electronicCodeBook_encryptAESCrypt";
6514 break;
6515 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
6516 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
6517 stubName = "electronicCodeBook_decryptAESCrypt";
6518 break;
6519 default:
6520 break;
6521 }
6522
6523 if (stubAddr == NULL) return false;
6524
6525 Node* electronicCodeBook_object = argument(0);
6526 Node* src = argument(1);
6527 Node* src_offset = argument(2);
6528 Node* len = argument(3);
6529 Node* dest = argument(4);
6530 Node* dest_offset = argument(5);
6531
6532 // (1) src and dest are arrays.
6533 const Type* src_type = src->Value(&_gvn);
6534 const Type* dest_type = dest->Value(&_gvn);
6535 const TypeAryPtr* top_src = src_type->isa_aryptr();
6536 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6537 assert(top_src != NULL && top_src->klass() != NULL
6538 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6539
6540 // checks are the responsibility of the caller
6541 Node* src_start = src;
6542 Node* dest_start = dest;
6543 if (src_offset != NULL || dest_offset != NULL) {
6544 assert(src_offset != NULL && dest_offset != NULL, "");
6545 src_start = array_element_address(src, src_offset, T_BYTE);
6546 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6547 }
6548
6549 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6550 // (because of the predicated logic executed earlier).
6551 // so we cast it here safely.
6552 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6553
6554 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6555 if (embeddedCipherObj == NULL) return false;
6556
6557 // cast it to what we know it will be at runtime
6558 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6559 assert(tinst != NULL, "ECB obj is null");
6560 assert(tinst->klass()->is_loaded(), "ECB obj is not loaded");
6561 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6562 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6563
6564 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6565 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6566 const TypeOopPtr* xtype = aklass->as_instance_type();
6567 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6568 aescrypt_object = _gvn.transform(aescrypt_object);
6569
6570 // we need to get the start of the aescrypt_object's expanded key array
6571 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6572 if (k_start == NULL) return false;
6573
6574 Node* ecbCrypt;
6575 if (Matcher::pass_original_key_for_aes()) {
6576 // no SPARC version for AES/ECB intrinsics now.
6577 return false;
6578 }
6579 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6580 ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6581 OptoRuntime::electronicCodeBook_aescrypt_Type(),
6582 stubAddr, stubName, TypePtr::BOTTOM,
6583 src_start, dest_start, k_start, len);
6584
6585 // return cipher length (int)
6586 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6587 set_result(retvalue);
6588 return true;
6589 }
6590
6591 //------------------------------inline_counterMode_AESCrypt-----------------------
6592 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6593 assert(UseAES, "need AES instruction support");
6594 if (!UseAESCTRIntrinsics) return false;
6595
6596 address stubAddr = NULL;
6597 const char *stubName = NULL;
6598 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6599 stubAddr = StubRoutines::counterMode_AESCrypt();
6600 stubName = "counterMode_AESCrypt";
6601 }
6602 if (stubAddr == NULL) return false;
6603
6604 Node* counterMode_object = argument(0);
6605 Node* src = argument(1);
6606 Node* src_offset = argument(2);
6607 Node* len = argument(3);
6608 Node* dest = argument(4);
6609 Node* dest_offset = argument(5);
6610
6611 src = access_resolve(src, ACCESS_READ);
6612 dest = access_resolve(dest, ACCESS_WRITE);
6613 counterMode_object = access_resolve(counterMode_object, ACCESS_WRITE);
6614
6615 // (1) src and dest are arrays.
6616 const Type* src_type = src->Value(&_gvn);
6617 const Type* dest_type = dest->Value(&_gvn);
6618 const TypeAryPtr* top_src = src_type->isa_aryptr();
6619 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6620 assert(top_src != NULL && top_src->klass() != NULL &&
6621 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6622
6623 // checks are the responsibility of the caller
6624 Node* src_start = src;
6625 Node* dest_start = dest;
6626 if (src_offset != NULL || dest_offset != NULL) {
6627 assert(src_offset != NULL && dest_offset != NULL, "");
6628 src_start = array_element_address(src, src_offset, T_BYTE);
6629 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6630 }
6631
6632 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6633 // (because of the predicated logic executed earlier).
6634 // so we cast it here safely.
6635 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6636 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6637 if (embeddedCipherObj == NULL) return false;
6638 // cast it to what we know it will be at runtime
6639 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6640 assert(tinst != NULL, "CTR obj is null");
6641 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6642 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6643 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6644 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6645 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6646 const TypeOopPtr* xtype = aklass->as_instance_type();
6647 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6648 aescrypt_object = _gvn.transform(aescrypt_object);
6649 // we need to get the start of the aescrypt_object's expanded key array
6650 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6651 if (k_start == NULL) return false;
6652 // similarly, get the start address of the r vector
6653 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6654 if (obj_counter == NULL) return false;
6655 obj_counter = access_resolve(obj_counter, ACCESS_WRITE);
6656 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6657
6658 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6659 if (saved_encCounter == NULL) return false;
6660 saved_encCounter = access_resolve(saved_encCounter, ACCESS_WRITE);
6661 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6662 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6663
6664 Node* ctrCrypt;
6665 if (Matcher::pass_original_key_for_aes()) {
6666 // no SPARC version for AES/CTR intrinsics now.
6667 return false;
6668 }
6669 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6670 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6671 OptoRuntime::counterMode_aescrypt_Type(),
6672 stubAddr, stubName, TypePtr::BOTTOM,
6673 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6674
6675 // return cipher length (int)
6676 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6677 set_result(retvalue);
6678 return true;
6679 }
6680
6681 //------------------------------get_key_start_from_aescrypt_object-----------------------
6682 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6683 #if defined(PPC64) || defined(S390)
6684 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6685 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6686 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6687 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6688 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6689 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6690 if (objSessionK == NULL) {
6691 return (Node *) NULL;
6692 }
6693 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6694 #else
6695 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6696 #endif // PPC64
6697 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6698 if (objAESCryptKey == NULL) return (Node *) NULL;
6699
6700 // now have the array, need to get the start address of the K array
6701 objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ);
6702 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6703 return k_start;
6704 }
6705
6706 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6707 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6708 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6709 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6710 if (objAESCryptKey == NULL) return (Node *) NULL;
6711
6712 // now have the array, need to get the start address of the lastKey array
6713 objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ);
6714 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6715 return original_k_start;
6716 }
6717
6718 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6719 // Return node representing slow path of predicate check.
6720 // the pseudo code we want to emulate with this predicate is:
6721 // for encryption:
6722 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6723 // for decryption:
6724 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6725 // note cipher==plain is more conservative than the original java code but that's OK
6726 //
6727 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6728 // The receiver was checked for NULL already.
6729 Node* objCBC = argument(0);
6730
6731 Node* src = argument(1);
6732 Node* dest = argument(4);
6733
6734 // Load embeddedCipher field of CipherBlockChaining object.
6735 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6736
6737 // get AESCrypt klass for instanceOf check
6738 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6739 // will have same classloader as CipherBlockChaining object
6740 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6741 assert(tinst != NULL, "CBCobj is null");
6742 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6743
6744 // we want to do an instanceof comparison against the AESCrypt class
6745 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6746 if (!klass_AESCrypt->is_loaded()) {
6747 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6748 Node* ctrl = control();
6749 set_control(top()); // no regular fast path
6750 return ctrl;
6751 }
6752
6753 src = must_be_not_null(src, true);
6754 dest = must_be_not_null(dest, true);
6755
6756 // Resolve oops to stable for CmpP below.
6757 src = access_resolve(src, 0);
6758 dest = access_resolve(dest, 0);
6759
6760 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6761
6762 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6763 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6764 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6765
6766 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6767
6768 // for encryption, we are done
6769 if (!decrypting)
6770 return instof_false; // even if it is NULL
6771
6772 // for decryption, we need to add a further check to avoid
6773 // taking the intrinsic path when cipher and plain are the same
6774 // see the original java code for why.
6775 RegionNode* region = new RegionNode(3);
6776 region->init_req(1, instof_false);
6777
6778 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6779 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6780 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6781 region->init_req(2, src_dest_conjoint);
6782
6783 record_for_igvn(region);
6784 return _gvn.transform(region);
6785 }
6786
6787 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
6788 // Return node representing slow path of predicate check.
6789 // the pseudo code we want to emulate with this predicate is:
6790 // for encryption:
6791 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6792 // for decryption:
6793 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6794 // note cipher==plain is more conservative than the original java code but that's OK
6795 //
6796 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
6797 // The receiver was checked for NULL already.
6798 Node* objECB = argument(0);
6799
6800 // Load embeddedCipher field of ElectronicCodeBook object.
6801 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6802
6803 // get AESCrypt klass for instanceOf check
6804 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6805 // will have same classloader as ElectronicCodeBook object
6806 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
6807 assert(tinst != NULL, "ECBobj is null");
6808 assert(tinst->klass()->is_loaded(), "ECBobj is not loaded");
6809
6810 // we want to do an instanceof comparison against the AESCrypt class
6811 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6812 if (!klass_AESCrypt->is_loaded()) {
6813 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6814 Node* ctrl = control();
6815 set_control(top()); // no regular fast path
6816 return ctrl;
6817 }
6818 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6819
6820 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6821 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6822 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6823
6824 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6825
6826 // for encryption, we are done
6827 if (!decrypting)
6828 return instof_false; // even if it is NULL
6829
6830 // for decryption, we need to add a further check to avoid
6831 // taking the intrinsic path when cipher and plain are the same
6832 // see the original java code for why.
6833 RegionNode* region = new RegionNode(3);
6834 region->init_req(1, instof_false);
6835 Node* src = argument(1);
6836 Node* dest = argument(4);
6837 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6838 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6839 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6840 region->init_req(2, src_dest_conjoint);
6841
6842 record_for_igvn(region);
6843 return _gvn.transform(region);
6844 }
6845
6846 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6847 // Return node representing slow path of predicate check.
6848 // the pseudo code we want to emulate with this predicate is:
6849 // for encryption:
6850 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6851 // for decryption:
6852 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6853 // note cipher==plain is more conservative than the original java code but that's OK
6854 //
6855
6856 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6857 // The receiver was checked for NULL already.
6858 Node* objCTR = argument(0);
6859
6860 // Load embeddedCipher field of CipherBlockChaining object.
6861 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6862
6863 // get AESCrypt klass for instanceOf check
6864 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6865 // will have same classloader as CipherBlockChaining object
6866 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6867 assert(tinst != NULL, "CTRobj is null");
6868 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6869
6870 // we want to do an instanceof comparison against the AESCrypt class
6871 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6872 if (!klass_AESCrypt->is_loaded()) {
6873 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6874 Node* ctrl = control();
6875 set_control(top()); // no regular fast path
6876 return ctrl;
6877 }
6878
6879 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6880 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6881 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6882 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6883 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6884
6885 return instof_false; // even if it is NULL
6886 }
6887
6888 //------------------------------inline_ghash_processBlocks
6889 bool LibraryCallKit::inline_ghash_processBlocks() {
6890 address stubAddr;
6891 const char *stubName;
6892 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6893
6894 stubAddr = StubRoutines::ghash_processBlocks();
6895 stubName = "ghash_processBlocks";
6896
6897 Node* data = argument(0);
6898 Node* offset = argument(1);
6899 Node* len = argument(2);
6900 Node* state = argument(3);
6901 Node* subkeyH = argument(4);
6902
6903 state = must_be_not_null(state, true);
6904 subkeyH = must_be_not_null(subkeyH, true);
6905 data = must_be_not_null(data, true);
6906
6907 state = access_resolve(state, ACCESS_WRITE);
6908 subkeyH = access_resolve(subkeyH, ACCESS_READ);
6909 data = access_resolve(data, ACCESS_READ);
6910
6911 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6912 assert(state_start, "state is NULL");
6913 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6914 assert(subkeyH_start, "subkeyH is NULL");
6915 Node* data_start = array_element_address(data, offset, T_BYTE);
6916 assert(data_start, "data is NULL");
6917
6918 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6919 OptoRuntime::ghash_processBlocks_Type(),
6920 stubAddr, stubName, TypePtr::BOTTOM,
6921 state_start, subkeyH_start, data_start, len);
6922 return true;
6923 }
6924
6925 bool LibraryCallKit::inline_base64_encodeBlock() {
6926 address stubAddr;
6927 const char *stubName;
6928 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6929 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6930 stubAddr = StubRoutines::base64_encodeBlock();
6931 stubName = "encodeBlock";
6932
6933 if (!stubAddr) return false;
6934 Node* base64obj = argument(0);
6935 Node* src = argument(1);
6936 Node* offset = argument(2);
6937 Node* len = argument(3);
6938 Node* dest = argument(4);
6939 Node* dp = argument(5);
6940 Node* isURL = argument(6);
6941
6942 src = must_be_not_null(src, true);
6943 src = access_resolve(src, ACCESS_READ);
6944 dest = must_be_not_null(dest, true);
6945 dest = access_resolve(dest, ACCESS_WRITE);
6946
6947 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6948 assert(src_start, "source array is NULL");
6949 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6950 assert(dest_start, "destination array is NULL");
6951
6952 Node* base64 = make_runtime_call(RC_LEAF,
6953 OptoRuntime::base64_encodeBlock_Type(),
6954 stubAddr, stubName, TypePtr::BOTTOM,
6955 src_start, offset, len, dest_start, dp, isURL);
6956 return true;
6957 }
6958
6959 //------------------------------inline_sha_implCompress-----------------------
6960 //
6961 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6962 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6963 //
6964 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6965 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6966 //
6967 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6968 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6969 //
6970 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6971 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6972
6973 Node* sha_obj = argument(0);
6974 Node* src = argument(1); // type oop
6975 Node* ofs = argument(2); // type int
6976
6977 const Type* src_type = src->Value(&_gvn);
6978 const TypeAryPtr* top_src = src_type->isa_aryptr();
6979 if (top_src == NULL || top_src->klass() == NULL) {
6980 // failed array check
6981 return false;
6982 }
6983 // Figure out the size and type of the elements we will be copying.
6984 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6985 if (src_elem != T_BYTE) {
6986 return false;
6987 }
6988 // 'src_start' points to src array + offset
6989 src = must_be_not_null(src, true);
6990 src = access_resolve(src, ACCESS_READ);
6991 Node* src_start = array_element_address(src, ofs, src_elem);
6992 Node* state = NULL;
6993 address stubAddr;
6994 const char *stubName;
6995
6996 switch(id) {
6997 case vmIntrinsics::_sha_implCompress:
6998 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6999 state = get_state_from_sha_object(sha_obj);
7000 stubAddr = StubRoutines::sha1_implCompress();
7001 stubName = "sha1_implCompress";
7002 break;
7003 case vmIntrinsics::_sha2_implCompress:
7004 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
7005 state = get_state_from_sha_object(sha_obj);
7006 stubAddr = StubRoutines::sha256_implCompress();
7007 stubName = "sha256_implCompress";
7008 break;
7009 case vmIntrinsics::_sha5_implCompress:
7010 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
7011 state = get_state_from_sha5_object(sha_obj);
7012 stubAddr = StubRoutines::sha512_implCompress();
7013 stubName = "sha512_implCompress";
7014 break;
7015 default:
7016 fatal_unexpected_iid(id);
7017 return false;
7018 }
7019 if (state == NULL) return false;
7020
7021 assert(stubAddr != NULL, "Stub is generated");
7022 if (stubAddr == NULL) return false;
7023
7024 // Call the stub.
7025 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
7026 stubAddr, stubName, TypePtr::BOTTOM,
7027 src_start, state);
7028
7029 return true;
7030 }
7031
7032 //------------------------------inline_digestBase_implCompressMB-----------------------
7033 //
7034 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
7035 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
7036 //
7037 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
7038 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
7039 "need SHA1/SHA256/SHA512 instruction support");
7040 assert((uint)predicate < 3, "sanity");
7041 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7042
7043 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
7044 Node* src = argument(1); // byte[] array
7045 Node* ofs = argument(2); // type int
7046 Node* limit = argument(3); // type int
7047
7048 const Type* src_type = src->Value(&_gvn);
7049 const TypeAryPtr* top_src = src_type->isa_aryptr();
7050 if (top_src == NULL || top_src->klass() == NULL) {
7051 // failed array check
7052 return false;
7053 }
7054 // Figure out the size and type of the elements we will be copying.
7055 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
7056 if (src_elem != T_BYTE) {
7057 return false;
7058 }
7059 // 'src_start' points to src array + offset
7060 src = must_be_not_null(src, false);
7061 src = access_resolve(src, ACCESS_READ);
7062 Node* src_start = array_element_address(src, ofs, src_elem);
7063
7064 const char* klass_SHA_name = NULL;
7065 const char* stub_name = NULL;
7066 address stub_addr = NULL;
7067 bool long_state = false;
7068
7069 switch (predicate) {
7070 case 0:
7071 if (UseSHA1Intrinsics) {
7072 klass_SHA_name = "sun/security/provider/SHA";
7073 stub_name = "sha1_implCompressMB";
7074 stub_addr = StubRoutines::sha1_implCompressMB();
7075 }
7076 break;
7077 case 1:
7078 if (UseSHA256Intrinsics) {
7079 klass_SHA_name = "sun/security/provider/SHA2";
7080 stub_name = "sha256_implCompressMB";
7081 stub_addr = StubRoutines::sha256_implCompressMB();
7082 }
7083 break;
7084 case 2:
7085 if (UseSHA512Intrinsics) {
7086 klass_SHA_name = "sun/security/provider/SHA5";
7087 stub_name = "sha512_implCompressMB";
7088 stub_addr = StubRoutines::sha512_implCompressMB();
7089 long_state = true;
7090 }
7091 break;
7092 default:
7093 fatal("unknown SHA intrinsic predicate: %d", predicate);
7094 }
7095 if (klass_SHA_name != NULL) {
7096 assert(stub_addr != NULL, "Stub is generated");
7097 if (stub_addr == NULL) return false;
7098
7099 // get DigestBase klass to lookup for SHA klass
7100 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7101 assert(tinst != NULL, "digestBase_obj is not instance???");
7102 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
7103
7104 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
7105 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
7106 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
7107 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
7108 }
7109 return false;
7110 }
7111 //------------------------------inline_sha_implCompressMB-----------------------
7112 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
7113 bool long_state, address stubAddr, const char *stubName,
7114 Node* src_start, Node* ofs, Node* limit) {
7115 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
7116 const TypeOopPtr* xtype = aklass->as_instance_type();
7117 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
7118 sha_obj = _gvn.transform(sha_obj);
7119
7120 Node* state;
7121 if (long_state) {
7122 state = get_state_from_sha5_object(sha_obj);
7123 } else {
7124 state = get_state_from_sha_object(sha_obj);
7125 }
7126 if (state == NULL) return false;
7127
7128 // Call the stub.
7129 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7130 OptoRuntime::digestBase_implCompressMB_Type(),
7131 stubAddr, stubName, TypePtr::BOTTOM,
7132 src_start, state, ofs, limit);
7133 // return ofs (int)
7134 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7135 set_result(result);
7136
7137 return true;
7138 }
7139
7140 //------------------------------get_state_from_sha_object-----------------------
7141 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
7142 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
7143 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
7144 if (sha_state == NULL) return (Node *) NULL;
7145
7146 // now have the array, need to get the start address of the state array
7147 sha_state = access_resolve(sha_state, ACCESS_WRITE);
7148 Node* state = array_element_address(sha_state, intcon(0), T_INT);
7149 return state;
7150 }
7151
7152 //------------------------------get_state_from_sha5_object-----------------------
7153 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
7154 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
7155 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
7156 if (sha_state == NULL) return (Node *) NULL;
7157
7158 // now have the array, need to get the start address of the state array
7159 sha_state = access_resolve(sha_state, ACCESS_WRITE);
7160 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
7161 return state;
7162 }
7163
7164 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
7165 // Return node representing slow path of predicate check.
7166 // the pseudo code we want to emulate with this predicate is:
7167 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
7168 //
7169 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
7170 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
7171 "need SHA1/SHA256/SHA512 instruction support");
7172 assert((uint)predicate < 3, "sanity");
7173
7174 // The receiver was checked for NULL already.
7175 Node* digestBaseObj = argument(0);
7176
7177 // get DigestBase klass for instanceOf check
7178 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
7179 assert(tinst != NULL, "digestBaseObj is null");
7180 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
7181
7182 const char* klass_SHA_name = NULL;
7183 switch (predicate) {
7184 case 0:
7185 if (UseSHA1Intrinsics) {
7186 // we want to do an instanceof comparison against the SHA class
7187 klass_SHA_name = "sun/security/provider/SHA";
7188 }
7189 break;
7190 case 1:
7191 if (UseSHA256Intrinsics) {
7192 // we want to do an instanceof comparison against the SHA2 class
7193 klass_SHA_name = "sun/security/provider/SHA2";
7194 }
7195 break;
7196 case 2:
7197 if (UseSHA512Intrinsics) {
7198 // we want to do an instanceof comparison against the SHA5 class
7199 klass_SHA_name = "sun/security/provider/SHA5";
7200 }
7201 break;
7202 default:
7203 fatal("unknown SHA intrinsic predicate: %d", predicate);
7204 }
7205
7206 ciKlass* klass_SHA = NULL;
7207 if (klass_SHA_name != NULL) {
7208 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
7209 }
7210 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
7211 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
7212 Node* ctrl = control();
7213 set_control(top()); // no intrinsic path
7214 return ctrl;
7215 }
7216 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
7217
7218 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
7219 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
7220 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7221 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
7222
7223 return instof_false; // even if it is NULL
7224 }
7225
7226 //-------------inline_fma-----------------------------------
7227 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
7228 Node *a = NULL;
7229 Node *b = NULL;
7230 Node *c = NULL;
7231 Node* result = NULL;
7232 switch (id) {
7233 case vmIntrinsics::_fmaD:
7234 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
7235 // no receiver since it is static method
7236 a = round_double_node(argument(0));
7237 b = round_double_node(argument(2));
7238 c = round_double_node(argument(4));
7239 result = _gvn.transform(new FmaDNode(control(), a, b, c));
7240 break;
7241 case vmIntrinsics::_fmaF:
7242 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
7243 a = argument(0);
7244 b = argument(1);
7245 c = argument(2);
7246 result = _gvn.transform(new FmaFNode(control(), a, b, c));
7247 break;
7248 default:
7249 fatal_unexpected_iid(id); break;
7250 }
7251 set_result(result);
7252 return true;
7253 }
7254
7255 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
7256 // argument(0) is receiver
7257 Node* codePoint = argument(1);
7258 Node* n = NULL;
7259
7260 switch (id) {
7261 case vmIntrinsics::_isDigit :
7262 n = new DigitNode(control(), codePoint);
7263 break;
7264 case vmIntrinsics::_isLowerCase :
7265 n = new LowerCaseNode(control(), codePoint);
7266 break;
7267 case vmIntrinsics::_isUpperCase :
7268 n = new UpperCaseNode(control(), codePoint);
7269 break;
7270 case vmIntrinsics::_isWhitespace :
7271 n = new WhitespaceNode(control(), codePoint);
7272 break;
7273 default:
7274 fatal_unexpected_iid(id);
7275 }
7276
7277 set_result(_gvn.transform(n));
7278 return true;
7279 }
7280
7281 //------------------------------inline_fp_min_max------------------------------
7282 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
7283 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
7284
7285 // The intrinsic should be used only when the API branches aren't predictable,
7286 // the last one performing the most important comparison. The following heuristic
7287 // uses the branch statistics to eventually bail out if necessary.
7288
7289 ciMethodData *md = callee()->method_data();
7290
7291 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
7292 ciCallProfile cp = caller()->call_profile_at_bci(bci());
7293
7294 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
7295 // Bail out if the call-site didn't contribute enough to the statistics.
7296 return false;
7297 }
7298
7299 uint taken = 0, not_taken = 0;
7300
7301 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
7302 if (p->is_BranchData()) {
7303 taken = ((ciBranchData*)p)->taken();
7304 not_taken = ((ciBranchData*)p)->not_taken();
7305 }
7306 }
7307
7308 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
7309 balance = balance < 0 ? -balance : balance;
7310 if ( balance > 0.2 ) {
7311 // Bail out if the most important branch is predictable enough.
7312 return false;
7313 }
7314 }
7315 */
7316
7317 Node *a = NULL;
7318 Node *b = NULL;
7319 Node *n = NULL;
7320 switch (id) {
7321 case vmIntrinsics::_maxF:
7322 case vmIntrinsics::_minF:
7323 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
7324 a = argument(0);
7325 b = argument(1);
7326 break;
7327 case vmIntrinsics::_maxD:
7328 case vmIntrinsics::_minD:
7329 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
7330 a = round_double_node(argument(0));
7331 b = round_double_node(argument(2));
7332 break;
7333 default:
7334 fatal_unexpected_iid(id);
7335 break;
7336 }
7337 switch (id) {
7338 case vmIntrinsics::_maxF: n = new MaxFNode(a, b); break;
7339 case vmIntrinsics::_minF: n = new MinFNode(a, b); break;
7340 case vmIntrinsics::_maxD: n = new MaxDNode(a, b); break;
7341 case vmIntrinsics::_minD: n = new MinDNode(a, b); break;
7342 default: fatal_unexpected_iid(id); break;
7343 }
7344 set_result(_gvn.transform(n));
7345 return true;
7346 }
7347
7348 bool LibraryCallKit::inline_profileBoolean() {
7349 Node* counts = argument(1);
7350 const TypeAryPtr* ary = NULL;
7351 ciArray* aobj = NULL;
7352 if (counts->is_Con()
7353 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
7354 && (aobj = ary->const_oop()->as_array()) != NULL
7355 && (aobj->length() == 2)) {
7356 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
7357 jint false_cnt = aobj->element_value(0).as_int();
7358 jint true_cnt = aobj->element_value(1).as_int();
7359
7360 if (C->log() != NULL) {
7361 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
7362 false_cnt, true_cnt);
7363 }
7364
7365 if (false_cnt + true_cnt == 0) {
7366 // According to profile, never executed.
7367 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7368 Deoptimization::Action_reinterpret);
7369 return true;
7370 }
7371
7372 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
7373 // is a number of each value occurrences.
7374 Node* result = argument(0);
7375 if (false_cnt == 0 || true_cnt == 0) {
7376 // According to profile, one value has been never seen.
7377 int expected_val = (false_cnt == 0) ? 1 : 0;
7378
7379 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
7380 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7381
7382 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
7383 Node* fast_path = _gvn.transform(new IfTrueNode(check));
7384 Node* slow_path = _gvn.transform(new IfFalseNode(check));
7385
7386 { // Slow path: uncommon trap for never seen value and then reexecute
7387 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
7388 // the value has been seen at least once.
7389 PreserveJVMState pjvms(this);
7390 PreserveReexecuteState preexecs(this);
7391 jvms()->set_should_reexecute(true);
7392
7393 set_control(slow_path);
7394 set_i_o(i_o());
7395
7396 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7397 Deoptimization::Action_reinterpret);
7398 }
7399 // The guard for never seen value enables sharpening of the result and
7400 // returning a constant. It allows to eliminate branches on the same value
7401 // later on.
7402 set_control(fast_path);
7403 result = intcon(expected_val);
7404 }
7405 // Stop profiling.
7406 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
7407 // By replacing method body with profile data (represented as ProfileBooleanNode
7408 // on IR level) we effectively disable profiling.
7409 // It enables full speed execution once optimized code is generated.
7410 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
7411 C->record_for_igvn(profile);
7412 set_result(profile);
7413 return true;
7414 } else {
7415 // Continue profiling.
7416 // Profile data isn't available at the moment. So, execute method's bytecode version.
7417 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
7418 // is compiled and counters aren't available since corresponding MethodHandle
7419 // isn't a compile-time constant.
7420 return false;
7421 }
7422 }
7423
7424 bool LibraryCallKit::inline_isCompileConstant() {
7425 Node* n = argument(0);
7426 set_result(n->is_Con() ? intcon(1) : intcon(0));
7427 return true;
7428 }