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/klass.inline.hpp"
36 #include "oops/objArrayKlass.hpp"
37 #include "opto/addnode.hpp"
38 #include "opto/arraycopynode.hpp"
39 #include "opto/c2compiler.hpp"
40 #include "opto/callGenerator.hpp"
41 #include "opto/castnode.hpp"
42 #include "opto/cfgnode.hpp"
43 #include "opto/convertnode.hpp"
44 #include "opto/countbitsnode.hpp"
45 #include "opto/intrinsicnode.hpp"
46 #include "opto/idealKit.hpp"
47 #include "opto/mathexactnode.hpp"
48 #include "opto/movenode.hpp"
49 #include "opto/mulnode.hpp"
50 #include "opto/narrowptrnode.hpp"
51 #include "opto/opaquenode.hpp"
52 #include "opto/parse.hpp"
53 #include "opto/runtime.hpp"
54 #include "opto/rootnode.hpp"
55 #include "opto/subnode.hpp"
56 #include "prims/nativeLookup.hpp"
57 #include "prims/unsafe.hpp"
58 #include "runtime/objectMonitor.hpp"
59 #include "runtime/sharedRuntime.hpp"
60 #include "utilities/macros.hpp"
61
62
63 class LibraryIntrinsic : public InlineCallGenerator {
64 // Extend the set of intrinsics known to the runtime:
65 public:
66 private:
67 bool _is_virtual;
68 bool _does_virtual_dispatch;
69 int8_t _predicates_count; // Intrinsic is predicated by several conditions
70 int8_t _last_predicate; // Last generated predicate
71 vmIntrinsics::ID _intrinsic_id;
72
73 public:
74 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
75 : InlineCallGenerator(m),
76 _is_virtual(is_virtual),
77 _does_virtual_dispatch(does_virtual_dispatch),
78 _predicates_count((int8_t)predicates_count),
79 _last_predicate((int8_t)-1),
80 _intrinsic_id(id)
81 {
82 }
83 virtual bool is_intrinsic() const { return true; }
84 virtual bool is_virtual() const { return _is_virtual; }
85 virtual bool is_predicated() const { return _predicates_count > 0; }
86 virtual int predicates_count() const { return _predicates_count; }
87 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; }
88 virtual JVMState* generate(JVMState* jvms);
89 virtual Node* generate_predicate(JVMState* jvms, int predicate);
90 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
91 };
92
93
94 // Local helper class for LibraryIntrinsic:
95 class LibraryCallKit : public GraphKit {
96 private:
97 LibraryIntrinsic* _intrinsic; // the library intrinsic being called
98 Node* _result; // the result node, if any
99 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted
100
101 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type);
102
103 public:
104 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
105 : GraphKit(jvms),
106 _intrinsic(intrinsic),
107 _result(NULL)
108 {
109 // Check if this is a root compile. In that case we don't have a caller.
110 if (!jvms->has_method()) {
111 _reexecute_sp = sp();
112 } else {
113 // Find out how many arguments the interpreter needs when deoptimizing
114 // and save the stack pointer value so it can used by uncommon_trap.
115 // We find the argument count by looking at the declared signature.
116 bool ignored_will_link;
117 ciSignature* declared_signature = NULL;
118 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
119 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
120 _reexecute_sp = sp() + nargs; // "push" arguments back on stack
121 }
122 }
123
124 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
125
126 ciMethod* caller() const { return jvms()->method(); }
127 int bci() const { return jvms()->bci(); }
128 LibraryIntrinsic* intrinsic() const { return _intrinsic; }
129 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); }
130 ciMethod* callee() const { return _intrinsic->method(); }
131
132 bool try_to_inline(int predicate);
133 Node* try_to_predicate(int predicate);
134
135 void push_result() {
136 // Push the result onto the stack.
137 if (!stopped() && result() != NULL) {
138 BasicType bt = result()->bottom_type()->basic_type();
139 push_node(bt, result());
140 }
141 }
142
143 private:
144 void fatal_unexpected_iid(vmIntrinsics::ID iid) {
145 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid));
146 }
147
148 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
149 void set_result(RegionNode* region, PhiNode* value);
150 Node* result() { return _result; }
151
152 virtual int reexecute_sp() { return _reexecute_sp; }
153
154 // Helper functions to inline natives
155 Node* generate_guard(Node* test, RegionNode* region, float true_prob);
156 Node* generate_slow_guard(Node* test, RegionNode* region);
157 Node* generate_fair_guard(Node* test, RegionNode* region);
158 Node* generate_negative_guard(Node* index, RegionNode* region,
159 // resulting CastII of index:
160 Node* *pos_index = NULL);
161 Node* generate_limit_guard(Node* offset, Node* subseq_length,
162 Node* array_length,
163 RegionNode* region);
164 void generate_string_range_check(Node* array, Node* offset,
165 Node* length, bool char_count);
166 Node* generate_current_thread(Node* &tls_output);
167 Node* load_mirror_from_klass(Node* klass);
168 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
169 RegionNode* region, int null_path,
170 int offset);
171 Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
172 RegionNode* region, int null_path) {
173 int offset = java_lang_Class::klass_offset_in_bytes();
174 return load_klass_from_mirror_common(mirror, never_see_null,
175 region, null_path,
176 offset);
177 }
178 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
179 RegionNode* region, int null_path) {
180 int offset = java_lang_Class::array_klass_offset_in_bytes();
181 return load_klass_from_mirror_common(mirror, never_see_null,
182 region, null_path,
183 offset);
184 }
185 Node* generate_access_flags_guard(Node* kls,
186 int modifier_mask, int modifier_bits,
187 RegionNode* region);
188 Node* generate_interface_guard(Node* kls, RegionNode* region);
189 Node* generate_array_guard(Node* kls, RegionNode* region) {
190 return generate_array_guard_common(kls, region, false, false);
191 }
192 Node* generate_non_array_guard(Node* kls, RegionNode* region) {
193 return generate_array_guard_common(kls, region, false, true);
194 }
195 Node* generate_objArray_guard(Node* kls, RegionNode* region) {
196 return generate_array_guard_common(kls, region, true, false);
197 }
198 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
199 return generate_array_guard_common(kls, region, true, true);
200 }
201 Node* generate_array_guard_common(Node* kls, RegionNode* region,
202 bool obj_array, bool not_array);
203 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
204 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
205 bool is_virtual = false, bool is_static = false);
206 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
207 return generate_method_call(method_id, false, true);
208 }
209 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
210 return generate_method_call(method_id, true, false);
211 }
212 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
213 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
214
215 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
216 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
217 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
218 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
219 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
220 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
221 bool inline_string_indexOfChar();
222 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
223 bool inline_string_toBytesU();
224 bool inline_string_getCharsU();
225 bool inline_string_copy(bool compress);
226 bool inline_string_char_access(bool is_store);
227 Node* round_double_node(Node* n);
228 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
229 bool inline_math_native(vmIntrinsics::ID id);
230 bool inline_math(vmIntrinsics::ID id);
231 bool inline_double_math(vmIntrinsics::ID id);
232 template <typename OverflowOp>
233 bool inline_math_overflow(Node* arg1, Node* arg2);
234 void inline_math_mathExact(Node* math, Node* test);
235 bool inline_math_addExactI(bool is_increment);
236 bool inline_math_addExactL(bool is_increment);
237 bool inline_math_multiplyExactI();
238 bool inline_math_multiplyExactL();
239 bool inline_math_multiplyHigh();
240 bool inline_math_negateExactI();
241 bool inline_math_negateExactL();
242 bool inline_math_subtractExactI(bool is_decrement);
243 bool inline_math_subtractExactL(bool is_decrement);
244 bool inline_min_max(vmIntrinsics::ID id);
245 bool inline_notify(vmIntrinsics::ID id);
246 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
247 // This returns Type::AnyPtr, RawPtr, or OopPtr.
248 int classify_unsafe_addr(Node* &base, Node* &offset, BasicType type);
249 Node* make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type = T_ILLEGAL, bool can_cast = false);
250
251 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
252 DecoratorSet mo_decorator_for_access_kind(AccessKind kind);
253 bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
254 static bool klass_needs_init_guard(Node* kls);
255 bool inline_unsafe_allocate();
256 bool inline_unsafe_newArray(bool uninitialized);
257 bool inline_unsafe_writeback0();
258 bool inline_unsafe_writebackSync0(bool is_pre);
259 bool inline_unsafe_copyMemory();
260 bool inline_native_currentThread();
261
262 bool inline_native_time_funcs(address method, const char* funcName);
263 #ifdef JFR_HAVE_INTRINSICS
264 bool inline_native_classID();
265 bool inline_native_getEventWriter();
266 #endif
267 bool inline_native_Class_query(vmIntrinsics::ID id);
268 bool inline_native_subtype_check();
269 bool inline_native_getLength();
270 bool inline_array_copyOf(bool is_copyOfRange);
271 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
272 bool inline_preconditions_checkIndex();
273 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array);
274 bool inline_native_clone(bool is_virtual);
275 bool inline_native_Reflection_getCallerClass();
276 // Helper function for inlining native object hash method
277 bool inline_native_hashcode(bool is_virtual, bool is_static);
278 bool inline_native_getClass();
279
280 // Helper functions for inlining arraycopy
281 bool inline_arraycopy();
282 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
283 RegionNode* slow_region);
284 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
285 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp,
286 uint new_idx);
287
288 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
289 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
290 bool inline_unsafe_fence(vmIntrinsics::ID id);
291 bool inline_onspinwait();
292 bool inline_fp_conversions(vmIntrinsics::ID id);
293 bool inline_number_methods(vmIntrinsics::ID id);
294 bool inline_reference_get();
295 bool inline_Class_cast();
296 bool inline_aescrypt_Block(vmIntrinsics::ID id);
297 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
298 bool inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id);
299 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
300 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
301 Node* inline_electronicCodeBook_AESCrypt_predicate(bool decrypting);
302 Node* inline_counterMode_AESCrypt_predicate();
303 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
304 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
305 bool inline_ghash_processBlocks();
306 bool inline_base64_encodeBlock();
307 bool inline_sha_implCompress(vmIntrinsics::ID id);
308 bool inline_digestBase_implCompressMB(int predicate);
309 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
310 bool long_state, address stubAddr, const char *stubName,
311 Node* src_start, Node* ofs, Node* limit);
312 Node* get_state_from_sha_object(Node *sha_object);
313 Node* get_state_from_sha5_object(Node *sha_object);
314 Node* inline_digestBase_implCompressMB_predicate(int predicate);
315 bool inline_encodeISOArray();
316 bool inline_updateCRC32();
317 bool inline_updateBytesCRC32();
318 bool inline_updateByteBufferCRC32();
319 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
320 bool inline_updateBytesCRC32C();
321 bool inline_updateDirectByteBufferCRC32C();
322 bool inline_updateBytesAdler32();
323 bool inline_updateByteBufferAdler32();
324 bool inline_multiplyToLen();
325 bool inline_hasNegatives();
326 bool inline_squareToLen();
327 bool inline_mulAdd();
328 bool inline_montgomeryMultiply();
329 bool inline_montgomerySquare();
330 bool inline_vectorizedMismatch();
331 bool inline_fma(vmIntrinsics::ID id);
332 bool inline_character_compare(vmIntrinsics::ID id);
333 bool inline_fp_min_max(vmIntrinsics::ID id);
334
335 bool inline_profileBoolean();
336 bool inline_isCompileConstant();
337 void clear_upper_avx() {
338 #ifdef X86
339 if (UseAVX >= 2) {
340 C->set_clear_upper_avx(true);
341 }
342 #endif
343 }
344 };
345
346 //---------------------------make_vm_intrinsic----------------------------
347 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
348 vmIntrinsics::ID id = m->intrinsic_id();
349 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
350
351 if (!m->is_loaded()) {
352 // Do not attempt to inline unloaded methods.
353 return NULL;
354 }
355
356 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
357 bool is_available = false;
358
359 {
360 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
361 // the compiler must transition to '_thread_in_vm' state because both
362 // methods access VM-internal data.
363 VM_ENTRY_MARK;
364 methodHandle mh(THREAD, m->get_Method());
365 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
366 !C->directive()->is_intrinsic_disabled(mh) &&
367 !vmIntrinsics::is_disabled_by_flags(mh);
368
369 }
370
371 if (is_available) {
372 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
373 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
374 return new LibraryIntrinsic(m, is_virtual,
375 vmIntrinsics::predicates_needed(id),
376 vmIntrinsics::does_virtual_dispatch(id),
377 (vmIntrinsics::ID) id);
378 } else {
379 return NULL;
380 }
381 }
382
383 //----------------------register_library_intrinsics-----------------------
384 // Initialize this file's data structures, for each Compile instance.
385 void Compile::register_library_intrinsics() {
386 // Nothing to do here.
387 }
388
389 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
390 LibraryCallKit kit(jvms, this);
391 Compile* C = kit.C;
392 int nodes = C->unique();
393 #ifndef PRODUCT
394 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
395 char buf[1000];
396 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
397 tty->print_cr("Intrinsic %s", str);
398 }
399 #endif
400 ciMethod* callee = kit.callee();
401 const int bci = kit.bci();
402
403 // Try to inline the intrinsic.
404 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
405 kit.try_to_inline(_last_predicate)) {
406 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
407 : "(intrinsic)";
408 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
409 if (C->print_intrinsics() || C->print_inlining()) {
410 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
411 }
412 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
413 if (C->log()) {
414 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
415 vmIntrinsics::name_at(intrinsic_id()),
416 (is_virtual() ? " virtual='1'" : ""),
417 C->unique() - nodes);
418 }
419 // Push the result from the inlined method onto the stack.
420 kit.push_result();
421 C->print_inlining_update(this);
422 return kit.transfer_exceptions_into_jvms();
423 }
424
425 // The intrinsic bailed out
426 if (jvms->has_method()) {
427 // Not a root compile.
428 const char* msg;
429 if (callee->intrinsic_candidate()) {
430 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
431 } else {
432 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
433 : "failed to inline (intrinsic), method not annotated";
434 }
435 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
436 if (C->print_intrinsics() || C->print_inlining()) {
437 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
438 }
439 } else {
440 // Root compile
441 ResourceMark rm;
442 stringStream msg_stream;
443 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
444 vmIntrinsics::name_at(intrinsic_id()),
445 is_virtual() ? " (virtual)" : "", bci);
446 const char *msg = msg_stream.as_string();
447 log_debug(jit, inlining)("%s", msg);
448 if (C->print_intrinsics() || C->print_inlining()) {
449 tty->print("%s", msg);
450 }
451 }
452 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
453 C->print_inlining_update(this);
454 return NULL;
455 }
456
457 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
458 LibraryCallKit kit(jvms, this);
459 Compile* C = kit.C;
460 int nodes = C->unique();
461 _last_predicate = predicate;
462 #ifndef PRODUCT
463 assert(is_predicated() && predicate < predicates_count(), "sanity");
464 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
465 char buf[1000];
466 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
467 tty->print_cr("Predicate for intrinsic %s", str);
468 }
469 #endif
470 ciMethod* callee = kit.callee();
471 const int bci = kit.bci();
472
473 Node* slow_ctl = kit.try_to_predicate(predicate);
474 if (!kit.failing()) {
475 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
476 : "(intrinsic, predicate)";
477 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
478 if (C->print_intrinsics() || C->print_inlining()) {
479 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
480 }
481 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
482 if (C->log()) {
483 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
484 vmIntrinsics::name_at(intrinsic_id()),
485 (is_virtual() ? " virtual='1'" : ""),
486 C->unique() - nodes);
487 }
488 return slow_ctl; // Could be NULL if the check folds.
489 }
490
491 // The intrinsic bailed out
492 if (jvms->has_method()) {
493 // Not a root compile.
494 const char* msg = "failed to generate predicate for intrinsic";
495 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
496 if (C->print_intrinsics() || C->print_inlining()) {
497 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
498 }
499 } else {
500 // Root compile
501 ResourceMark rm;
502 stringStream msg_stream;
503 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
504 vmIntrinsics::name_at(intrinsic_id()),
505 is_virtual() ? " (virtual)" : "", bci);
506 const char *msg = msg_stream.as_string();
507 log_debug(jit, inlining)("%s", msg);
508 if (C->print_intrinsics() || C->print_inlining()) {
509 C->print_inlining_stream()->print("%s", msg);
510 }
511 }
512 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
513 return NULL;
514 }
515
516 bool LibraryCallKit::try_to_inline(int predicate) {
517 // Handle symbolic names for otherwise undistinguished boolean switches:
518 const bool is_store = true;
519 const bool is_compress = true;
520 const bool is_static = true;
521 const bool is_volatile = true;
522
523 if (!jvms()->has_method()) {
524 // Root JVMState has a null method.
525 assert(map()->memory()->Opcode() == Op_Parm, "");
526 // Insert the memory aliasing node
527 set_all_memory(reset_memory());
528 }
529 assert(merged_memory(), "");
530
531
532 switch (intrinsic_id()) {
533 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
534 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
535 case vmIntrinsics::_getClass: return inline_native_getClass();
536
537 case vmIntrinsics::_ceil:
538 case vmIntrinsics::_floor:
539 case vmIntrinsics::_rint:
540 case vmIntrinsics::_dsin:
541 case vmIntrinsics::_dcos:
542 case vmIntrinsics::_dtan:
543 case vmIntrinsics::_dabs:
544 case vmIntrinsics::_fabs:
545 case vmIntrinsics::_iabs:
546 case vmIntrinsics::_labs:
547 case vmIntrinsics::_datan2:
548 case vmIntrinsics::_dsqrt:
549 case vmIntrinsics::_dexp:
550 case vmIntrinsics::_dlog:
551 case vmIntrinsics::_dlog10:
552 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id());
553
554 case vmIntrinsics::_min:
555 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
556
557 case vmIntrinsics::_notify:
558 case vmIntrinsics::_notifyAll:
559 return inline_notify(intrinsic_id());
560
561 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
562 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
563 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
564 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
565 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
566 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
567 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
568 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
569 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
570 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
571 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
572 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
573 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
574
575 case vmIntrinsics::_arraycopy: return inline_arraycopy();
576
577 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
578 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
579 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
580 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
581
582 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
583 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
584 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
585 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
586 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
587 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
588 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
589
590 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
591 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
592
593 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
594 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
595 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
596 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
597
598 case vmIntrinsics::_compressStringC:
599 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
600 case vmIntrinsics::_inflateStringC:
601 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
602
603 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
604 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
605 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
606 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
607 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
608 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
609 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
610 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
611 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
612
613 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
614 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
615 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
616 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
617 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
618 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
619 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
620 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
621 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
622
623 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
624 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
625 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
626 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
627 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
628 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
629 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
630 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
631 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
632
633 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
634 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
635 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
636 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
637 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
638 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
639 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
640 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
641 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
642
643 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
644 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
645 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
646 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
647
648 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
649 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
650 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
651 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
652
653 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
654 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
655 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
656 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
657 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
658 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
659 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
660 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
661 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
662
663 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
664 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
665 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
666 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
667 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
668 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
669 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
670 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
671 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
672
673 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
674 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
675 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
676 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
677 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
678 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
679 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
680 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
681 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
682
683 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
684 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
685 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
686 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
687 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
688 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
689 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
690 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
691 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
692
693 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
694 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
695 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
696 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
697 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
698
699 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
700 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
701 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
702 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
703 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
704 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
705 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
706 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
707 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
708 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
709 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
710 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
711 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
712 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
713 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
714 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
715 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
716 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
717 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
718 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
719
720 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
721 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
722 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
723 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
724 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
725 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
726 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
727 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
728 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
729 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
730 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
731 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
732 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
733 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
734 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
735
736 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
737 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
738 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
739 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
740
741 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
742 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
743 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
744 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
745 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
746
747 case vmIntrinsics::_loadFence:
748 case vmIntrinsics::_storeFence:
749 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
750
751 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
752
753 case vmIntrinsics::_currentThread: return inline_native_currentThread();
754
755 #ifdef JFR_HAVE_INTRINSICS
756 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime");
757 case vmIntrinsics::_getClassId: return inline_native_classID();
758 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
759 #endif
760 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
761 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
762 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
763 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
764 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
765 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
766 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
767 case vmIntrinsics::_getLength: return inline_native_getLength();
768 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
769 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
770 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
771 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
772 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
773 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
774
775 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
776 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
777
778 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
779
780 case vmIntrinsics::_isInstance:
781 case vmIntrinsics::_getModifiers:
782 case vmIntrinsics::_isInterface:
783 case vmIntrinsics::_isArray:
784 case vmIntrinsics::_isPrimitive:
785 case vmIntrinsics::_getSuperclass:
786 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
787
788 case vmIntrinsics::_floatToRawIntBits:
789 case vmIntrinsics::_floatToIntBits:
790 case vmIntrinsics::_intBitsToFloat:
791 case vmIntrinsics::_doubleToRawLongBits:
792 case vmIntrinsics::_doubleToLongBits:
793 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
794
795 case vmIntrinsics::_numberOfLeadingZeros_i:
796 case vmIntrinsics::_numberOfLeadingZeros_l:
797 case vmIntrinsics::_numberOfTrailingZeros_i:
798 case vmIntrinsics::_numberOfTrailingZeros_l:
799 case vmIntrinsics::_bitCount_i:
800 case vmIntrinsics::_bitCount_l:
801 case vmIntrinsics::_reverseBytes_i:
802 case vmIntrinsics::_reverseBytes_l:
803 case vmIntrinsics::_reverseBytes_s:
804 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
805
806 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
807
808 case vmIntrinsics::_Reference_get: return inline_reference_get();
809
810 case vmIntrinsics::_Class_cast: return inline_Class_cast();
811
812 case vmIntrinsics::_aescrypt_encryptBlock:
813 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
814
815 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
816 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
817 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
818
819 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
820 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
821 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
822
823 case vmIntrinsics::_counterMode_AESCrypt:
824 return inline_counterMode_AESCrypt(intrinsic_id());
825
826 case vmIntrinsics::_sha_implCompress:
827 case vmIntrinsics::_sha2_implCompress:
828 case vmIntrinsics::_sha5_implCompress:
829 return inline_sha_implCompress(intrinsic_id());
830
831 case vmIntrinsics::_digestBase_implCompressMB:
832 return inline_digestBase_implCompressMB(predicate);
833
834 case vmIntrinsics::_multiplyToLen:
835 return inline_multiplyToLen();
836
837 case vmIntrinsics::_squareToLen:
838 return inline_squareToLen();
839
840 case vmIntrinsics::_mulAdd:
841 return inline_mulAdd();
842
843 case vmIntrinsics::_montgomeryMultiply:
844 return inline_montgomeryMultiply();
845 case vmIntrinsics::_montgomerySquare:
846 return inline_montgomerySquare();
847
848 case vmIntrinsics::_vectorizedMismatch:
849 return inline_vectorizedMismatch();
850
851 case vmIntrinsics::_ghash_processBlocks:
852 return inline_ghash_processBlocks();
853 case vmIntrinsics::_base64_encodeBlock:
854 return inline_base64_encodeBlock();
855
856 case vmIntrinsics::_encodeISOArray:
857 case vmIntrinsics::_encodeByteISOArray:
858 return inline_encodeISOArray();
859
860 case vmIntrinsics::_updateCRC32:
861 return inline_updateCRC32();
862 case vmIntrinsics::_updateBytesCRC32:
863 return inline_updateBytesCRC32();
864 case vmIntrinsics::_updateByteBufferCRC32:
865 return inline_updateByteBufferCRC32();
866
867 case vmIntrinsics::_updateBytesCRC32C:
868 return inline_updateBytesCRC32C();
869 case vmIntrinsics::_updateDirectByteBufferCRC32C:
870 return inline_updateDirectByteBufferCRC32C();
871
872 case vmIntrinsics::_updateBytesAdler32:
873 return inline_updateBytesAdler32();
874 case vmIntrinsics::_updateByteBufferAdler32:
875 return inline_updateByteBufferAdler32();
876
877 case vmIntrinsics::_profileBoolean:
878 return inline_profileBoolean();
879 case vmIntrinsics::_isCompileConstant:
880 return inline_isCompileConstant();
881
882 case vmIntrinsics::_hasNegatives:
883 return inline_hasNegatives();
884
885 case vmIntrinsics::_fmaD:
886 case vmIntrinsics::_fmaF:
887 return inline_fma(intrinsic_id());
888
889 case vmIntrinsics::_isDigit:
890 case vmIntrinsics::_isLowerCase:
891 case vmIntrinsics::_isUpperCase:
892 case vmIntrinsics::_isWhitespace:
893 return inline_character_compare(intrinsic_id());
894
895 case vmIntrinsics::_maxF:
896 case vmIntrinsics::_minF:
897 case vmIntrinsics::_maxD:
898 case vmIntrinsics::_minD:
899 return inline_fp_min_max(intrinsic_id());
900
901 default:
902 // If you get here, it may be that someone has added a new intrinsic
903 // to the list in vmSymbols.hpp without implementing it here.
904 #ifndef PRODUCT
905 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
906 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
907 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
908 }
909 #endif
910 return false;
911 }
912 }
913
914 Node* LibraryCallKit::try_to_predicate(int predicate) {
915 if (!jvms()->has_method()) {
916 // Root JVMState has a null method.
917 assert(map()->memory()->Opcode() == Op_Parm, "");
918 // Insert the memory aliasing node
919 set_all_memory(reset_memory());
920 }
921 assert(merged_memory(), "");
922
923 switch (intrinsic_id()) {
924 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
925 return inline_cipherBlockChaining_AESCrypt_predicate(false);
926 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
927 return inline_cipherBlockChaining_AESCrypt_predicate(true);
928 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
929 return inline_electronicCodeBook_AESCrypt_predicate(false);
930 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
931 return inline_electronicCodeBook_AESCrypt_predicate(true);
932 case vmIntrinsics::_counterMode_AESCrypt:
933 return inline_counterMode_AESCrypt_predicate();
934 case vmIntrinsics::_digestBase_implCompressMB:
935 return inline_digestBase_implCompressMB_predicate(predicate);
936
937 default:
938 // If you get here, it may be that someone has added a new intrinsic
939 // to the list in vmSymbols.hpp without implementing it here.
940 #ifndef PRODUCT
941 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
942 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
943 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
944 }
945 #endif
946 Node* slow_ctl = control();
947 set_control(top()); // No fast path instrinsic
948 return slow_ctl;
949 }
950 }
951
952 //------------------------------set_result-------------------------------
953 // Helper function for finishing intrinsics.
954 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
955 record_for_igvn(region);
956 set_control(_gvn.transform(region));
957 set_result( _gvn.transform(value));
958 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
959 }
960
961 //------------------------------generate_guard---------------------------
962 // Helper function for generating guarded fast-slow graph structures.
963 // The given 'test', if true, guards a slow path. If the test fails
964 // then a fast path can be taken. (We generally hope it fails.)
965 // In all cases, GraphKit::control() is updated to the fast path.
966 // The returned value represents the control for the slow path.
967 // The return value is never 'top'; it is either a valid control
968 // or NULL if it is obvious that the slow path can never be taken.
969 // Also, if region and the slow control are not NULL, the slow edge
970 // is appended to the region.
971 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
972 if (stopped()) {
973 // Already short circuited.
974 return NULL;
975 }
976
977 // Build an if node and its projections.
978 // If test is true we take the slow path, which we assume is uncommon.
979 if (_gvn.type(test) == TypeInt::ZERO) {
980 // The slow branch is never taken. No need to build this guard.
981 return NULL;
982 }
983
984 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
985
986 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
987 if (if_slow == top()) {
988 // The slow branch is never taken. No need to build this guard.
989 return NULL;
990 }
991
992 if (region != NULL)
993 region->add_req(if_slow);
994
995 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
996 set_control(if_fast);
997
998 return if_slow;
999 }
1000
1001 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
1002 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
1003 }
1004 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
1005 return generate_guard(test, region, PROB_FAIR);
1006 }
1007
1008 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
1009 Node* *pos_index) {
1010 if (stopped())
1011 return NULL; // already stopped
1012 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
1013 return NULL; // index is already adequately typed
1014 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
1015 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1016 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
1017 if (is_neg != NULL && pos_index != NULL) {
1018 // Emulate effect of Parse::adjust_map_after_if.
1019 Node* ccast = new CastIINode(index, TypeInt::POS);
1020 ccast->set_req(0, control());
1021 (*pos_index) = _gvn.transform(ccast);
1022 }
1023 return is_neg;
1024 }
1025
1026 // Make sure that 'position' is a valid limit index, in [0..length].
1027 // There are two equivalent plans for checking this:
1028 // A. (offset + copyLength) unsigned<= arrayLength
1029 // B. offset <= (arrayLength - copyLength)
1030 // We require that all of the values above, except for the sum and
1031 // difference, are already known to be non-negative.
1032 // Plan A is robust in the face of overflow, if offset and copyLength
1033 // are both hugely positive.
1034 //
1035 // Plan B is less direct and intuitive, but it does not overflow at
1036 // all, since the difference of two non-negatives is always
1037 // representable. Whenever Java methods must perform the equivalent
1038 // check they generally use Plan B instead of Plan A.
1039 // For the moment we use Plan A.
1040 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
1041 Node* subseq_length,
1042 Node* array_length,
1043 RegionNode* region) {
1044 if (stopped())
1045 return NULL; // already stopped
1046 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
1047 if (zero_offset && subseq_length->eqv_uncast(array_length))
1048 return NULL; // common case of whole-array copy
1049 Node* last = subseq_length;
1050 if (!zero_offset) // last += offset
1051 last = _gvn.transform(new AddINode(last, offset));
1052 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
1053 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1054 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
1055 return is_over;
1056 }
1057
1058 // Emit range checks for the given String.value byte array
1059 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
1060 if (stopped()) {
1061 return; // already stopped
1062 }
1063 RegionNode* bailout = new RegionNode(1);
1064 record_for_igvn(bailout);
1065 if (char_count) {
1066 // Convert char count to byte count
1067 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1068 }
1069
1070 // Offset and count must not be negative
1071 generate_negative_guard(offset, bailout);
1072 generate_negative_guard(count, bailout);
1073 // Offset + count must not exceed length of array
1074 generate_limit_guard(offset, count, load_array_length(array), bailout);
1075
1076 if (bailout->req() > 1) {
1077 PreserveJVMState pjvms(this);
1078 set_control(_gvn.transform(bailout));
1079 uncommon_trap(Deoptimization::Reason_intrinsic,
1080 Deoptimization::Action_maybe_recompile);
1081 }
1082 }
1083
1084 //--------------------------generate_current_thread--------------------
1085 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1086 ciKlass* thread_klass = env()->Thread_klass();
1087 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1088 Node* thread = _gvn.transform(new ThreadLocalNode());
1089 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1090 Node* threadObj = _gvn.transform(LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(), thread_type, T_OBJECT, MemNode::unordered));
1091 tls_output = thread;
1092 return threadObj;
1093 }
1094
1095
1096 //------------------------------make_string_method_node------------------------
1097 // Helper method for String intrinsic functions. This version is called with
1098 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1099 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1100 // containing the lengths of str1 and str2.
1101 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1102 Node* result = NULL;
1103 switch (opcode) {
1104 case Op_StrIndexOf:
1105 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1106 str1_start, cnt1, str2_start, cnt2, ae);
1107 break;
1108 case Op_StrComp:
1109 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1110 str1_start, cnt1, str2_start, cnt2, ae);
1111 break;
1112 case Op_StrEquals:
1113 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1114 // Use the constant length if there is one because optimized match rule may exist.
1115 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1116 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1117 break;
1118 default:
1119 ShouldNotReachHere();
1120 return NULL;
1121 }
1122
1123 // All these intrinsics have checks.
1124 C->set_has_split_ifs(true); // Has chance for split-if optimization
1125 clear_upper_avx();
1126
1127 return _gvn.transform(result);
1128 }
1129
1130 //------------------------------inline_string_compareTo------------------------
1131 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1132 Node* arg1 = argument(0);
1133 Node* arg2 = argument(1);
1134
1135 arg1 = must_be_not_null(arg1, true);
1136 arg2 = must_be_not_null(arg2, true);
1137
1138 arg1 = access_resolve(arg1, ACCESS_READ);
1139 arg2 = access_resolve(arg2, ACCESS_READ);
1140
1141 // Get start addr and length of first argument
1142 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1143 Node* arg1_cnt = load_array_length(arg1);
1144
1145 // Get start addr and length of second argument
1146 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1147 Node* arg2_cnt = load_array_length(arg2);
1148
1149 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1150 set_result(result);
1151 return true;
1152 }
1153
1154 //------------------------------inline_string_equals------------------------
1155 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1156 Node* arg1 = argument(0);
1157 Node* arg2 = argument(1);
1158
1159 // paths (plus control) merge
1160 RegionNode* region = new RegionNode(3);
1161 Node* phi = new PhiNode(region, TypeInt::BOOL);
1162
1163 if (!stopped()) {
1164
1165 arg1 = must_be_not_null(arg1, true);
1166 arg2 = must_be_not_null(arg2, true);
1167
1168 arg1 = access_resolve(arg1, ACCESS_READ);
1169 arg2 = access_resolve(arg2, ACCESS_READ);
1170
1171 // Get start addr and length of first argument
1172 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1173 Node* arg1_cnt = load_array_length(arg1);
1174
1175 // Get start addr and length of second argument
1176 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1177 Node* arg2_cnt = load_array_length(arg2);
1178
1179 // Check for arg1_cnt != arg2_cnt
1180 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1181 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1182 Node* if_ne = generate_slow_guard(bol, NULL);
1183 if (if_ne != NULL) {
1184 phi->init_req(2, intcon(0));
1185 region->init_req(2, if_ne);
1186 }
1187
1188 // Check for count == 0 is done by assembler code for StrEquals.
1189
1190 if (!stopped()) {
1191 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1192 phi->init_req(1, equals);
1193 region->init_req(1, control());
1194 }
1195 }
1196
1197 // post merge
1198 set_control(_gvn.transform(region));
1199 record_for_igvn(region);
1200
1201 set_result(_gvn.transform(phi));
1202 return true;
1203 }
1204
1205 //------------------------------inline_array_equals----------------------------
1206 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1207 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1208 Node* arg1 = argument(0);
1209 Node* arg2 = argument(1);
1210
1211 arg1 = access_resolve(arg1, ACCESS_READ);
1212 arg2 = access_resolve(arg2, ACCESS_READ);
1213
1214 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1215 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1216 clear_upper_avx();
1217
1218 return true;
1219 }
1220
1221 //------------------------------inline_hasNegatives------------------------------
1222 bool LibraryCallKit::inline_hasNegatives() {
1223 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1224 return false;
1225 }
1226
1227 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1228 // no receiver since it is static method
1229 Node* ba = argument(0);
1230 Node* offset = argument(1);
1231 Node* len = argument(2);
1232
1233 ba = must_be_not_null(ba, true);
1234
1235 // Range checks
1236 generate_string_range_check(ba, offset, len, false);
1237 if (stopped()) {
1238 return true;
1239 }
1240 ba = access_resolve(ba, ACCESS_READ);
1241 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1242 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1243 set_result(_gvn.transform(result));
1244 return true;
1245 }
1246
1247 bool LibraryCallKit::inline_preconditions_checkIndex() {
1248 Node* index = argument(0);
1249 Node* length = argument(1);
1250 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1251 return false;
1252 }
1253
1254 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1255 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1256
1257 {
1258 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1259 uncommon_trap(Deoptimization::Reason_intrinsic,
1260 Deoptimization::Action_make_not_entrant);
1261 }
1262
1263 if (stopped()) {
1264 return false;
1265 }
1266
1267 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1268 BoolTest::mask btest = BoolTest::lt;
1269 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1270 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1271 _gvn.set_type(rc, rc->Value(&_gvn));
1272 if (!rc_bool->is_Con()) {
1273 record_for_igvn(rc);
1274 }
1275 set_control(_gvn.transform(new IfTrueNode(rc)));
1276 {
1277 PreserveJVMState pjvms(this);
1278 set_control(_gvn.transform(new IfFalseNode(rc)));
1279 uncommon_trap(Deoptimization::Reason_range_check,
1280 Deoptimization::Action_make_not_entrant);
1281 }
1282
1283 if (stopped()) {
1284 return false;
1285 }
1286
1287 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1288 result->set_req(0, control());
1289 result = _gvn.transform(result);
1290 set_result(result);
1291 replace_in_map(index, result);
1292 clear_upper_avx();
1293 return true;
1294 }
1295
1296 //------------------------------inline_string_indexOf------------------------
1297 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1298 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1299 return false;
1300 }
1301 Node* src = argument(0);
1302 Node* tgt = argument(1);
1303
1304 // Make the merge point
1305 RegionNode* result_rgn = new RegionNode(4);
1306 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1307
1308 src = must_be_not_null(src, true);
1309 tgt = must_be_not_null(tgt, true);
1310
1311 src = access_resolve(src, ACCESS_READ);
1312 tgt = access_resolve(tgt, ACCESS_READ);
1313
1314 // Get start addr and length of source string
1315 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1316 Node* src_count = load_array_length(src);
1317
1318 // Get start addr and length of substring
1319 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1320 Node* tgt_count = load_array_length(tgt);
1321
1322 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1323 // Divide src size by 2 if String is UTF16 encoded
1324 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1325 }
1326 if (ae == StrIntrinsicNode::UU) {
1327 // Divide substring size by 2 if String is UTF16 encoded
1328 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1329 }
1330
1331 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1332 if (result != NULL) {
1333 result_phi->init_req(3, result);
1334 result_rgn->init_req(3, control());
1335 }
1336 set_control(_gvn.transform(result_rgn));
1337 record_for_igvn(result_rgn);
1338 set_result(_gvn.transform(result_phi));
1339
1340 return true;
1341 }
1342
1343 //-----------------------------inline_string_indexOf-----------------------
1344 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1345 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1346 return false;
1347 }
1348 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1349 return false;
1350 }
1351 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1352 Node* src = argument(0); // byte[]
1353 Node* src_count = argument(1); // char count
1354 Node* tgt = argument(2); // byte[]
1355 Node* tgt_count = argument(3); // char count
1356 Node* from_index = argument(4); // char index
1357
1358 src = must_be_not_null(src, true);
1359 tgt = must_be_not_null(tgt, true);
1360
1361 src = access_resolve(src, ACCESS_READ);
1362 tgt = access_resolve(tgt, ACCESS_READ);
1363
1364 // Multiply byte array index by 2 if String is UTF16 encoded
1365 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1366 src_count = _gvn.transform(new SubINode(src_count, from_index));
1367 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1368 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1369
1370 // Range checks
1371 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1372 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1373 if (stopped()) {
1374 return true;
1375 }
1376
1377 RegionNode* region = new RegionNode(5);
1378 Node* phi = new PhiNode(region, TypeInt::INT);
1379
1380 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1381 if (result != NULL) {
1382 // The result is index relative to from_index if substring was found, -1 otherwise.
1383 // Generate code which will fold into cmove.
1384 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1385 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1386
1387 Node* if_lt = generate_slow_guard(bol, NULL);
1388 if (if_lt != NULL) {
1389 // result == -1
1390 phi->init_req(3, result);
1391 region->init_req(3, if_lt);
1392 }
1393 if (!stopped()) {
1394 result = _gvn.transform(new AddINode(result, from_index));
1395 phi->init_req(4, result);
1396 region->init_req(4, control());
1397 }
1398 }
1399
1400 set_control(_gvn.transform(region));
1401 record_for_igvn(region);
1402 set_result(_gvn.transform(phi));
1403 clear_upper_avx();
1404
1405 return true;
1406 }
1407
1408 // Create StrIndexOfNode with fast path checks
1409 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1410 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1411 // Check for substr count > string count
1412 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1413 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1414 Node* if_gt = generate_slow_guard(bol, NULL);
1415 if (if_gt != NULL) {
1416 phi->init_req(1, intcon(-1));
1417 region->init_req(1, if_gt);
1418 }
1419 if (!stopped()) {
1420 // Check for substr count == 0
1421 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1422 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1423 Node* if_zero = generate_slow_guard(bol, NULL);
1424 if (if_zero != NULL) {
1425 phi->init_req(2, intcon(0));
1426 region->init_req(2, if_zero);
1427 }
1428 }
1429 if (!stopped()) {
1430 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1431 }
1432 return NULL;
1433 }
1434
1435 //-----------------------------inline_string_indexOfChar-----------------------
1436 bool LibraryCallKit::inline_string_indexOfChar() {
1437 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1438 return false;
1439 }
1440 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1441 return false;
1442 }
1443 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1444 Node* src = argument(0); // byte[]
1445 Node* tgt = argument(1); // tgt is int ch
1446 Node* from_index = argument(2);
1447 Node* max = argument(3);
1448
1449 src = must_be_not_null(src, true);
1450 src = access_resolve(src, ACCESS_READ);
1451
1452 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1453 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1454 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1455
1456 // Range checks
1457 generate_string_range_check(src, src_offset, src_count, true);
1458 if (stopped()) {
1459 return true;
1460 }
1461
1462 RegionNode* region = new RegionNode(3);
1463 Node* phi = new PhiNode(region, TypeInt::INT);
1464
1465 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1466 C->set_has_split_ifs(true); // Has chance for split-if optimization
1467 _gvn.transform(result);
1468
1469 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1470 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1471
1472 Node* if_lt = generate_slow_guard(bol, NULL);
1473 if (if_lt != NULL) {
1474 // result == -1
1475 phi->init_req(2, result);
1476 region->init_req(2, if_lt);
1477 }
1478 if (!stopped()) {
1479 result = _gvn.transform(new AddINode(result, from_index));
1480 phi->init_req(1, result);
1481 region->init_req(1, control());
1482 }
1483 set_control(_gvn.transform(region));
1484 record_for_igvn(region);
1485 set_result(_gvn.transform(phi));
1486
1487 return true;
1488 }
1489 //---------------------------inline_string_copy---------------------
1490 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1491 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1492 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1493 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1494 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1495 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1496 bool LibraryCallKit::inline_string_copy(bool compress) {
1497 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1498 return false;
1499 }
1500 int nargs = 5; // 2 oops, 3 ints
1501 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1502
1503 Node* src = argument(0);
1504 Node* src_offset = argument(1);
1505 Node* dst = argument(2);
1506 Node* dst_offset = argument(3);
1507 Node* length = argument(4);
1508
1509 // Check for allocation before we add nodes that would confuse
1510 // tightly_coupled_allocation()
1511 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1512
1513 // Figure out the size and type of the elements we will be copying.
1514 const Type* src_type = src->Value(&_gvn);
1515 const Type* dst_type = dst->Value(&_gvn);
1516 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1517 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1518 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1519 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1520 "Unsupported array types for inline_string_copy");
1521
1522 src = must_be_not_null(src, true);
1523 dst = must_be_not_null(dst, true);
1524
1525 // Convert char[] offsets to byte[] offsets
1526 bool convert_src = (compress && src_elem == T_BYTE);
1527 bool convert_dst = (!compress && dst_elem == T_BYTE);
1528 if (convert_src) {
1529 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1530 } else if (convert_dst) {
1531 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1532 }
1533
1534 // Range checks
1535 generate_string_range_check(src, src_offset, length, convert_src);
1536 generate_string_range_check(dst, dst_offset, length, convert_dst);
1537 if (stopped()) {
1538 return true;
1539 }
1540
1541 src = access_resolve(src, ACCESS_READ);
1542 dst = access_resolve(dst, ACCESS_WRITE);
1543
1544 Node* src_start = array_element_address(src, src_offset, src_elem);
1545 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1546 // 'src_start' points to src array + scaled offset
1547 // 'dst_start' points to dst array + scaled offset
1548 Node* count = NULL;
1549 if (compress) {
1550 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1551 } else {
1552 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1553 }
1554
1555 if (alloc != NULL) {
1556 if (alloc->maybe_set_complete(&_gvn)) {
1557 // "You break it, you buy it."
1558 InitializeNode* init = alloc->initialization();
1559 assert(init->is_complete(), "we just did this");
1560 init->set_complete_with_arraycopy();
1561 assert(dst->is_CheckCastPP(), "sanity");
1562 assert(dst->in(0)->in(0) == init, "dest pinned");
1563 }
1564 // Do not let stores that initialize this object be reordered with
1565 // a subsequent store that would make this object accessible by
1566 // other threads.
1567 // Record what AllocateNode this StoreStore protects so that
1568 // escape analysis can go from the MemBarStoreStoreNode to the
1569 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1570 // based on the escape status of the AllocateNode.
1571 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1572 }
1573 if (compress) {
1574 set_result(_gvn.transform(count));
1575 }
1576 clear_upper_avx();
1577
1578 return true;
1579 }
1580
1581 #ifdef _LP64
1582 #define XTOP ,top() /*additional argument*/
1583 #else //_LP64
1584 #define XTOP /*no additional argument*/
1585 #endif //_LP64
1586
1587 //------------------------inline_string_toBytesU--------------------------
1588 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1589 bool LibraryCallKit::inline_string_toBytesU() {
1590 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1591 return false;
1592 }
1593 // Get the arguments.
1594 Node* value = argument(0);
1595 Node* offset = argument(1);
1596 Node* length = argument(2);
1597
1598 Node* newcopy = NULL;
1599
1600 // Set the original stack and the reexecute bit for the interpreter to reexecute
1601 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1602 { PreserveReexecuteState preexecs(this);
1603 jvms()->set_should_reexecute(true);
1604
1605 // Check if a null path was taken unconditionally.
1606 value = null_check(value);
1607
1608 RegionNode* bailout = new RegionNode(1);
1609 record_for_igvn(bailout);
1610
1611 // Range checks
1612 generate_negative_guard(offset, bailout);
1613 generate_negative_guard(length, bailout);
1614 generate_limit_guard(offset, length, load_array_length(value), bailout);
1615 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1616 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1617
1618 if (bailout->req() > 1) {
1619 PreserveJVMState pjvms(this);
1620 set_control(_gvn.transform(bailout));
1621 uncommon_trap(Deoptimization::Reason_intrinsic,
1622 Deoptimization::Action_maybe_recompile);
1623 }
1624 if (stopped()) {
1625 return true;
1626 }
1627
1628 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1629 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1630 newcopy = new_array(klass_node, size, 0); // no arguments to push
1631 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1632
1633 // Calculate starting addresses.
1634 value = access_resolve(value, ACCESS_READ);
1635 Node* src_start = array_element_address(value, offset, T_CHAR);
1636 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1637
1638 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1639 const TypeInt* toffset = gvn().type(offset)->is_int();
1640 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1641
1642 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1643 const char* copyfunc_name = "arraycopy";
1644 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1645 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1646 OptoRuntime::fast_arraycopy_Type(),
1647 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1648 src_start, dst_start, ConvI2X(length) XTOP);
1649 // Do not let reads from the cloned object float above the arraycopy.
1650 if (alloc != NULL) {
1651 if (alloc->maybe_set_complete(&_gvn)) {
1652 // "You break it, you buy it."
1653 InitializeNode* init = alloc->initialization();
1654 assert(init->is_complete(), "we just did this");
1655 init->set_complete_with_arraycopy();
1656 assert(newcopy->is_CheckCastPP(), "sanity");
1657 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1658 }
1659 // Do not let stores that initialize this object be reordered with
1660 // a subsequent store that would make this object accessible by
1661 // other threads.
1662 // Record what AllocateNode this StoreStore protects so that
1663 // escape analysis can go from the MemBarStoreStoreNode to the
1664 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1665 // based on the escape status of the AllocateNode.
1666 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1667 } else {
1668 insert_mem_bar(Op_MemBarCPUOrder);
1669 }
1670 } // original reexecute is set back here
1671
1672 C->set_has_split_ifs(true); // Has chance for split-if optimization
1673 if (!stopped()) {
1674 set_result(newcopy);
1675 }
1676 clear_upper_avx();
1677
1678 return true;
1679 }
1680
1681 //------------------------inline_string_getCharsU--------------------------
1682 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1683 bool LibraryCallKit::inline_string_getCharsU() {
1684 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1685 return false;
1686 }
1687
1688 // Get the arguments.
1689 Node* src = argument(0);
1690 Node* src_begin = argument(1);
1691 Node* src_end = argument(2); // exclusive offset (i < src_end)
1692 Node* dst = argument(3);
1693 Node* dst_begin = argument(4);
1694
1695 // Check for allocation before we add nodes that would confuse
1696 // tightly_coupled_allocation()
1697 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1698
1699 // Check if a null path was taken unconditionally.
1700 src = null_check(src);
1701 dst = null_check(dst);
1702 if (stopped()) {
1703 return true;
1704 }
1705
1706 // Get length and convert char[] offset to byte[] offset
1707 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1708 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1709
1710 // Range checks
1711 generate_string_range_check(src, src_begin, length, true);
1712 generate_string_range_check(dst, dst_begin, length, false);
1713 if (stopped()) {
1714 return true;
1715 }
1716
1717 if (!stopped()) {
1718 src = access_resolve(src, ACCESS_READ);
1719 dst = access_resolve(dst, ACCESS_WRITE);
1720
1721 // Calculate starting addresses.
1722 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1723 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1724
1725 // Check if array addresses are aligned to HeapWordSize
1726 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1727 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1728 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1729 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1730
1731 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1732 const char* copyfunc_name = "arraycopy";
1733 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1734 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1735 OptoRuntime::fast_arraycopy_Type(),
1736 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1737 src_start, dst_start, ConvI2X(length) XTOP);
1738 // Do not let reads from the cloned object float above the arraycopy.
1739 if (alloc != NULL) {
1740 if (alloc->maybe_set_complete(&_gvn)) {
1741 // "You break it, you buy it."
1742 InitializeNode* init = alloc->initialization();
1743 assert(init->is_complete(), "we just did this");
1744 init->set_complete_with_arraycopy();
1745 assert(dst->is_CheckCastPP(), "sanity");
1746 assert(dst->in(0)->in(0) == init, "dest pinned");
1747 }
1748 // Do not let stores that initialize this object be reordered with
1749 // a subsequent store that would make this object accessible by
1750 // other threads.
1751 // Record what AllocateNode this StoreStore protects so that
1752 // escape analysis can go from the MemBarStoreStoreNode to the
1753 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1754 // based on the escape status of the AllocateNode.
1755 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1756 } else {
1757 insert_mem_bar(Op_MemBarCPUOrder);
1758 }
1759 }
1760
1761 C->set_has_split_ifs(true); // Has chance for split-if optimization
1762 return true;
1763 }
1764
1765 //----------------------inline_string_char_access----------------------------
1766 // Store/Load char to/from byte[] array.
1767 // static void StringUTF16.putChar(byte[] val, int index, int c)
1768 // static char StringUTF16.getChar(byte[] val, int index)
1769 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1770 Node* value = argument(0);
1771 Node* index = argument(1);
1772 Node* ch = is_store ? argument(2) : NULL;
1773
1774 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1775 // correctly requires matched array shapes.
1776 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1777 "sanity: byte[] and char[] bases agree");
1778 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1779 "sanity: byte[] and char[] scales agree");
1780
1781 // Bail when getChar over constants is requested: constant folding would
1782 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1783 // Java method would constant fold nicely instead.
1784 if (!is_store && value->is_Con() && index->is_Con()) {
1785 return false;
1786 }
1787
1788 value = must_be_not_null(value, true);
1789 value = access_resolve(value, is_store ? ACCESS_WRITE : ACCESS_READ);
1790
1791 Node* adr = array_element_address(value, index, T_CHAR);
1792 if (adr->is_top()) {
1793 return false;
1794 }
1795 if (is_store) {
1796 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1797 } else {
1798 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
1799 set_result(ch);
1800 }
1801 return true;
1802 }
1803
1804 //--------------------------round_double_node--------------------------------
1805 // Round a double node if necessary.
1806 Node* LibraryCallKit::round_double_node(Node* n) {
1807 if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1808 n = _gvn.transform(new RoundDoubleNode(0, n));
1809 return n;
1810 }
1811
1812 //------------------------------inline_math-----------------------------------
1813 // public static double Math.abs(double)
1814 // public static double Math.sqrt(double)
1815 // public static double Math.log(double)
1816 // public static double Math.log10(double)
1817 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1818 Node* arg = round_double_node(argument(0));
1819 Node* n = NULL;
1820 switch (id) {
1821 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1822 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1823 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1824 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1825 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1826 default: fatal_unexpected_iid(id); break;
1827 }
1828 set_result(_gvn.transform(n));
1829 return true;
1830 }
1831
1832 //------------------------------inline_math-----------------------------------
1833 // public static float Math.abs(float)
1834 // public static int Math.abs(int)
1835 // public static long Math.abs(long)
1836 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1837 Node* arg = argument(0);
1838 Node* n = NULL;
1839 switch (id) {
1840 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1841 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1842 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1843 default: fatal_unexpected_iid(id); break;
1844 }
1845 set_result(_gvn.transform(n));
1846 return true;
1847 }
1848
1849 //------------------------------runtime_math-----------------------------
1850 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1851 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1852 "must be (DD)D or (D)D type");
1853
1854 // Inputs
1855 Node* a = round_double_node(argument(0));
1856 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1857
1858 const TypePtr* no_memory_effects = NULL;
1859 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1860 no_memory_effects,
1861 a, top(), b, b ? top() : NULL);
1862 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1863 #ifdef ASSERT
1864 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1865 assert(value_top == top(), "second value must be top");
1866 #endif
1867
1868 set_result(value);
1869 return true;
1870 }
1871
1872 //------------------------------inline_math_native-----------------------------
1873 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1874 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1875 switch (id) {
1876 // These intrinsics are not properly supported on all hardware
1877 case vmIntrinsics::_dsin:
1878 return StubRoutines::dsin() != NULL ?
1879 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1880 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1881 case vmIntrinsics::_dcos:
1882 return StubRoutines::dcos() != NULL ?
1883 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1884 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1885 case vmIntrinsics::_dtan:
1886 return StubRoutines::dtan() != NULL ?
1887 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1888 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1889 case vmIntrinsics::_dlog:
1890 return StubRoutines::dlog() != NULL ?
1891 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1892 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1893 case vmIntrinsics::_dlog10:
1894 return StubRoutines::dlog10() != NULL ?
1895 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1896 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1897
1898 // These intrinsics are supported on all hardware
1899 case vmIntrinsics::_ceil:
1900 case vmIntrinsics::_floor:
1901 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1902 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1903 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1904 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1905 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1906 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1907
1908 case vmIntrinsics::_dexp:
1909 return StubRoutines::dexp() != NULL ?
1910 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1911 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1912 case vmIntrinsics::_dpow: {
1913 Node* exp = round_double_node(argument(2));
1914 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1915 if (d != NULL && d->getd() == 2.0) {
1916 // Special case: pow(x, 2.0) => x * x
1917 Node* base = round_double_node(argument(0));
1918 set_result(_gvn.transform(new MulDNode(base, base)));
1919 return true;
1920 }
1921 return StubRoutines::dpow() != NULL ?
1922 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1923 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1924 }
1925 #undef FN_PTR
1926
1927 // These intrinsics are not yet correctly implemented
1928 case vmIntrinsics::_datan2:
1929 return false;
1930
1931 default:
1932 fatal_unexpected_iid(id);
1933 return false;
1934 }
1935 }
1936
1937 static bool is_simple_name(Node* n) {
1938 return (n->req() == 1 // constant
1939 || (n->is_Type() && n->as_Type()->type()->singleton())
1940 || n->is_Proj() // parameter or return value
1941 || n->is_Phi() // local of some sort
1942 );
1943 }
1944
1945 //----------------------------inline_notify-----------------------------------*
1946 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1947 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1948 address func;
1949 if (id == vmIntrinsics::_notify) {
1950 func = OptoRuntime::monitor_notify_Java();
1951 } else {
1952 func = OptoRuntime::monitor_notifyAll_Java();
1953 }
1954 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1955 make_slow_call_ex(call, env()->Throwable_klass(), false);
1956 return true;
1957 }
1958
1959
1960 //----------------------------inline_min_max-----------------------------------
1961 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1962 set_result(generate_min_max(id, argument(0), argument(1)));
1963 return true;
1964 }
1965
1966 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1967 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1968 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1969 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1970 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1971
1972 {
1973 PreserveJVMState pjvms(this);
1974 PreserveReexecuteState preexecs(this);
1975 jvms()->set_should_reexecute(true);
1976
1977 set_control(slow_path);
1978 set_i_o(i_o());
1979
1980 uncommon_trap(Deoptimization::Reason_intrinsic,
1981 Deoptimization::Action_none);
1982 }
1983
1984 set_control(fast_path);
1985 set_result(math);
1986 }
1987
1988 template <typename OverflowOp>
1989 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1990 typedef typename OverflowOp::MathOp MathOp;
1991
1992 MathOp* mathOp = new MathOp(arg1, arg2);
1993 Node* operation = _gvn.transform( mathOp );
1994 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1995 inline_math_mathExact(operation, ofcheck);
1996 return true;
1997 }
1998
1999 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2000 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2001 }
2002
2003 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2004 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2005 }
2006
2007 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2008 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2009 }
2010
2011 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2012 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2013 }
2014
2015 bool LibraryCallKit::inline_math_negateExactI() {
2016 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2017 }
2018
2019 bool LibraryCallKit::inline_math_negateExactL() {
2020 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2021 }
2022
2023 bool LibraryCallKit::inline_math_multiplyExactI() {
2024 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2025 }
2026
2027 bool LibraryCallKit::inline_math_multiplyExactL() {
2028 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2029 }
2030
2031 bool LibraryCallKit::inline_math_multiplyHigh() {
2032 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2033 return true;
2034 }
2035
2036 Node*
2037 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2038 // These are the candidate return value:
2039 Node* xvalue = x0;
2040 Node* yvalue = y0;
2041
2042 if (xvalue == yvalue) {
2043 return xvalue;
2044 }
2045
2046 bool want_max = (id == vmIntrinsics::_max);
2047
2048 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
2049 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
2050 if (txvalue == NULL || tyvalue == NULL) return top();
2051 // This is not really necessary, but it is consistent with a
2052 // hypothetical MaxINode::Value method:
2053 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
2054
2055 // %%% This folding logic should (ideally) be in a different place.
2056 // Some should be inside IfNode, and there to be a more reliable
2057 // transformation of ?: style patterns into cmoves. We also want
2058 // more powerful optimizations around cmove and min/max.
2059
2060 // Try to find a dominating comparison of these guys.
2061 // It can simplify the index computation for Arrays.copyOf
2062 // and similar uses of System.arraycopy.
2063 // First, compute the normalized version of CmpI(x, y).
2064 int cmp_op = Op_CmpI;
2065 Node* xkey = xvalue;
2066 Node* ykey = yvalue;
2067 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
2068 if (ideal_cmpxy->is_Cmp()) {
2069 // E.g., if we have CmpI(length - offset, count),
2070 // it might idealize to CmpI(length, count + offset)
2071 cmp_op = ideal_cmpxy->Opcode();
2072 xkey = ideal_cmpxy->in(1);
2073 ykey = ideal_cmpxy->in(2);
2074 }
2075
2076 // Start by locating any relevant comparisons.
2077 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
2078 Node* cmpxy = NULL;
2079 Node* cmpyx = NULL;
2080 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
2081 Node* cmp = start_from->fast_out(k);
2082 if (cmp->outcnt() > 0 && // must have prior uses
2083 cmp->in(0) == NULL && // must be context-independent
2084 cmp->Opcode() == cmp_op) { // right kind of compare
2085 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
2086 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
2087 }
2088 }
2089
2090 const int NCMPS = 2;
2091 Node* cmps[NCMPS] = { cmpxy, cmpyx };
2092 int cmpn;
2093 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2094 if (cmps[cmpn] != NULL) break; // find a result
2095 }
2096 if (cmpn < NCMPS) {
2097 // Look for a dominating test that tells us the min and max.
2098 int depth = 0; // Limit search depth for speed
2099 Node* dom = control();
2100 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2101 if (++depth >= 100) break;
2102 Node* ifproj = dom;
2103 if (!ifproj->is_Proj()) continue;
2104 Node* iff = ifproj->in(0);
2105 if (!iff->is_If()) continue;
2106 Node* bol = iff->in(1);
2107 if (!bol->is_Bool()) continue;
2108 Node* cmp = bol->in(1);
2109 if (cmp == NULL) continue;
2110 for (cmpn = 0; cmpn < NCMPS; cmpn++)
2111 if (cmps[cmpn] == cmp) break;
2112 if (cmpn == NCMPS) continue;
2113 BoolTest::mask btest = bol->as_Bool()->_test._test;
2114 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
2115 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2116 // At this point, we know that 'x btest y' is true.
2117 switch (btest) {
2118 case BoolTest::eq:
2119 // They are proven equal, so we can collapse the min/max.
2120 // Either value is the answer. Choose the simpler.
2121 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2122 return yvalue;
2123 return xvalue;
2124 case BoolTest::lt: // x < y
2125 case BoolTest::le: // x <= y
2126 return (want_max ? yvalue : xvalue);
2127 case BoolTest::gt: // x > y
2128 case BoolTest::ge: // x >= y
2129 return (want_max ? xvalue : yvalue);
2130 default:
2131 break;
2132 }
2133 }
2134 }
2135
2136 // We failed to find a dominating test.
2137 // Let's pick a test that might GVN with prior tests.
2138 Node* best_bol = NULL;
2139 BoolTest::mask best_btest = BoolTest::illegal;
2140 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2141 Node* cmp = cmps[cmpn];
2142 if (cmp == NULL) continue;
2143 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2144 Node* bol = cmp->fast_out(j);
2145 if (!bol->is_Bool()) continue;
2146 BoolTest::mask btest = bol->as_Bool()->_test._test;
2147 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2148 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2149 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2150 best_bol = bol->as_Bool();
2151 best_btest = btest;
2152 }
2153 }
2154 }
2155
2156 Node* answer_if_true = NULL;
2157 Node* answer_if_false = NULL;
2158 switch (best_btest) {
2159 default:
2160 if (cmpxy == NULL)
2161 cmpxy = ideal_cmpxy;
2162 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2163 // and fall through:
2164 case BoolTest::lt: // x < y
2165 case BoolTest::le: // x <= y
2166 answer_if_true = (want_max ? yvalue : xvalue);
2167 answer_if_false = (want_max ? xvalue : yvalue);
2168 break;
2169 case BoolTest::gt: // x > y
2170 case BoolTest::ge: // x >= y
2171 answer_if_true = (want_max ? xvalue : yvalue);
2172 answer_if_false = (want_max ? yvalue : xvalue);
2173 break;
2174 }
2175
2176 jint hi, lo;
2177 if (want_max) {
2178 // We can sharpen the minimum.
2179 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2180 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2181 } else {
2182 // We can sharpen the maximum.
2183 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2184 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2185 }
2186
2187 // Use a flow-free graph structure, to avoid creating excess control edges
2188 // which could hinder other optimizations.
2189 // Since Math.min/max is often used with arraycopy, we want
2190 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2191 Node* cmov = CMoveNode::make(NULL, best_bol,
2192 answer_if_false, answer_if_true,
2193 TypeInt::make(lo, hi, widen));
2194
2195 return _gvn.transform(cmov);
2196
2197 /*
2198 // This is not as desirable as it may seem, since Min and Max
2199 // nodes do not have a full set of optimizations.
2200 // And they would interfere, anyway, with 'if' optimizations
2201 // and with CMoveI canonical forms.
2202 switch (id) {
2203 case vmIntrinsics::_min:
2204 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2205 case vmIntrinsics::_max:
2206 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2207 default:
2208 ShouldNotReachHere();
2209 }
2210 */
2211 }
2212
2213 inline int
2214 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2215 const TypePtr* base_type = TypePtr::NULL_PTR;
2216 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2217 if (base_type == NULL) {
2218 // Unknown type.
2219 return Type::AnyPtr;
2220 } else if (base_type == TypePtr::NULL_PTR) {
2221 // Since this is a NULL+long form, we have to switch to a rawptr.
2222 base = _gvn.transform(new CastX2PNode(offset));
2223 offset = MakeConX(0);
2224 return Type::RawPtr;
2225 } else if (base_type->base() == Type::RawPtr) {
2226 return Type::RawPtr;
2227 } else if (base_type->isa_oopptr()) {
2228 // Base is never null => always a heap address.
2229 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2230 return Type::OopPtr;
2231 }
2232 // Offset is small => always a heap address.
2233 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2234 if (offset_type != NULL &&
2235 base_type->offset() == 0 && // (should always be?)
2236 offset_type->_lo >= 0 &&
2237 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2238 return Type::OopPtr;
2239 } else if (type == T_OBJECT) {
2240 // off heap access to an oop doesn't make any sense. Has to be on
2241 // heap.
2242 return Type::OopPtr;
2243 }
2244 // Otherwise, it might either be oop+off or NULL+addr.
2245 return Type::AnyPtr;
2246 } else {
2247 // No information:
2248 return Type::AnyPtr;
2249 }
2250 }
2251
2252 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type, bool can_cast) {
2253 Node* uncasted_base = base;
2254 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2255 if (kind == Type::RawPtr) {
2256 return basic_plus_adr(top(), uncasted_base, offset);
2257 } else if (kind == Type::AnyPtr) {
2258 assert(base == uncasted_base, "unexpected base change");
2259 if (can_cast) {
2260 if (!_gvn.type(base)->speculative_maybe_null() &&
2261 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2262 // According to profiling, this access is always on
2263 // heap. Casting the base to not null and thus avoiding membars
2264 // around the access should allow better optimizations
2265 Node* null_ctl = top();
2266 base = null_check_oop(base, &null_ctl, true, true, true);
2267 assert(null_ctl->is_top(), "no null control here");
2268 return basic_plus_adr(base, offset);
2269 } else if (_gvn.type(base)->speculative_always_null() &&
2270 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2271 // According to profiling, this access is always off
2272 // heap.
2273 base = null_assert(base);
2274 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2275 offset = MakeConX(0);
2276 return basic_plus_adr(top(), raw_base, offset);
2277 }
2278 }
2279 // We don't know if it's an on heap or off heap access. Fall back
2280 // to raw memory access.
2281 base = access_resolve(base, decorators);
2282 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2283 return basic_plus_adr(top(), raw, offset);
2284 } else {
2285 assert(base == uncasted_base, "unexpected base change");
2286 // We know it's an on heap access so base can't be null
2287 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2288 base = must_be_not_null(base, true);
2289 }
2290 return basic_plus_adr(base, offset);
2291 }
2292 }
2293
2294 //--------------------------inline_number_methods-----------------------------
2295 // inline int Integer.numberOfLeadingZeros(int)
2296 // inline int Long.numberOfLeadingZeros(long)
2297 //
2298 // inline int Integer.numberOfTrailingZeros(int)
2299 // inline int Long.numberOfTrailingZeros(long)
2300 //
2301 // inline int Integer.bitCount(int)
2302 // inline int Long.bitCount(long)
2303 //
2304 // inline char Character.reverseBytes(char)
2305 // inline short Short.reverseBytes(short)
2306 // inline int Integer.reverseBytes(int)
2307 // inline long Long.reverseBytes(long)
2308 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2309 Node* arg = argument(0);
2310 Node* n = NULL;
2311 switch (id) {
2312 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2313 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2314 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2315 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2316 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2317 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2318 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2319 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2320 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2321 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2322 default: fatal_unexpected_iid(id); break;
2323 }
2324 set_result(_gvn.transform(n));
2325 return true;
2326 }
2327
2328 //----------------------------inline_unsafe_access----------------------------
2329
2330 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2331 // Attempt to infer a sharper value type from the offset and base type.
2332 ciKlass* sharpened_klass = NULL;
2333
2334 // See if it is an instance field, with an object type.
2335 if (alias_type->field() != NULL) {
2336 if (alias_type->field()->type()->is_klass()) {
2337 sharpened_klass = alias_type->field()->type()->as_klass();
2338 }
2339 }
2340
2341 // See if it is a narrow oop array.
2342 if (adr_type->isa_aryptr()) {
2343 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2344 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2345 if (elem_type != NULL) {
2346 sharpened_klass = elem_type->klass();
2347 }
2348 }
2349 }
2350
2351 // The sharpened class might be unloaded if there is no class loader
2352 // contraint in place.
2353 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2354 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2355
2356 #ifndef PRODUCT
2357 if (C->print_intrinsics() || C->print_inlining()) {
2358 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2359 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2360 }
2361 #endif
2362 // Sharpen the value type.
2363 return tjp;
2364 }
2365 return NULL;
2366 }
2367
2368 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2369 switch (kind) {
2370 case Relaxed:
2371 return MO_UNORDERED;
2372 case Opaque:
2373 return MO_RELAXED;
2374 case Acquire:
2375 return MO_ACQUIRE;
2376 case Release:
2377 return MO_RELEASE;
2378 case Volatile:
2379 return MO_SEQ_CST;
2380 default:
2381 ShouldNotReachHere();
2382 return 0;
2383 }
2384 }
2385
2386 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2387 if (callee()->is_static()) return false; // caller must have the capability!
2388 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2389 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2390 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2391 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2392
2393 if (is_reference_type(type)) {
2394 decorators |= ON_UNKNOWN_OOP_REF;
2395 }
2396
2397 if (unaligned) {
2398 decorators |= C2_UNALIGNED;
2399 }
2400
2401 #ifndef PRODUCT
2402 {
2403 ResourceMark rm;
2404 // Check the signatures.
2405 ciSignature* sig = callee()->signature();
2406 #ifdef ASSERT
2407 if (!is_store) {
2408 // Object getReference(Object base, int/long offset), etc.
2409 BasicType rtype = sig->return_type()->basic_type();
2410 assert(rtype == type, "getter must return the expected value");
2411 assert(sig->count() == 2, "oop getter has 2 arguments");
2412 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2413 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2414 } else {
2415 // void putReference(Object base, int/long offset, Object x), etc.
2416 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2417 assert(sig->count() == 3, "oop putter has 3 arguments");
2418 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2419 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2420 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2421 assert(vtype == type, "putter must accept the expected value");
2422 }
2423 #endif // ASSERT
2424 }
2425 #endif //PRODUCT
2426
2427 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2428
2429 Node* receiver = argument(0); // type: oop
2430
2431 // Build address expression.
2432 Node* adr;
2433 Node* heap_base_oop = top();
2434 Node* offset = top();
2435 Node* val;
2436
2437 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2438 Node* base = argument(1); // type: oop
2439 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2440 offset = argument(2); // type: long
2441 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2442 // to be plain byte offsets, which are also the same as those accepted
2443 // by oopDesc::field_addr.
2444 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2445 "fieldOffset must be byte-scaled");
2446 // 32-bit machines ignore the high half!
2447 offset = ConvL2X(offset);
2448 adr = make_unsafe_address(base, offset, is_store ? ACCESS_WRITE : ACCESS_READ, type, kind == Relaxed);
2449
2450 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2451 if (type != T_OBJECT) {
2452 decorators |= IN_NATIVE; // off-heap primitive access
2453 } else {
2454 return false; // off-heap oop accesses are not supported
2455 }
2456 } else {
2457 heap_base_oop = base; // on-heap or mixed access
2458 }
2459
2460 // Can base be NULL? Otherwise, always on-heap access.
2461 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2462
2463 if (!can_access_non_heap) {
2464 decorators |= IN_HEAP;
2465 }
2466
2467 val = is_store ? argument(4) : NULL;
2468
2469 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2470 if (adr_type == TypePtr::NULL_PTR) {
2471 return false; // off-heap access with zero address
2472 }
2473
2474 // Try to categorize the address.
2475 Compile::AliasType* alias_type = C->alias_type(adr_type);
2476 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2477
2478 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2479 alias_type->adr_type() == TypeAryPtr::RANGE) {
2480 return false; // not supported
2481 }
2482
2483 bool mismatched = false;
2484 BasicType bt = alias_type->basic_type();
2485 if (bt != T_ILLEGAL) {
2486 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2487 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2488 // Alias type doesn't differentiate between byte[] and boolean[]).
2489 // Use address type to get the element type.
2490 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2491 }
2492 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2493 // accessing an array field with getReference is not a mismatch
2494 bt = T_OBJECT;
2495 }
2496 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2497 // Don't intrinsify mismatched object accesses
2498 return false;
2499 }
2500 mismatched = (bt != type);
2501 } else if (alias_type->adr_type()->isa_oopptr()) {
2502 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2503 }
2504
2505 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2506
2507 if (mismatched) {
2508 decorators |= C2_MISMATCHED;
2509 }
2510
2511 // First guess at the value type.
2512 const Type *value_type = Type::get_const_basic_type(type);
2513
2514 // Figure out the memory ordering.
2515 decorators |= mo_decorator_for_access_kind(kind);
2516
2517 if (!is_store && type == T_OBJECT) {
2518 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2519 if (tjp != NULL) {
2520 value_type = tjp;
2521 }
2522 }
2523
2524 receiver = null_check(receiver);
2525 if (stopped()) {
2526 return true;
2527 }
2528 // Heap pointers get a null-check from the interpreter,
2529 // as a courtesy. However, this is not guaranteed by Unsafe,
2530 // and it is not possible to fully distinguish unintended nulls
2531 // from intended ones in this API.
2532
2533 if (!is_store) {
2534 Node* p = NULL;
2535 // Try to constant fold a load from a constant field
2536 ciField* field = alias_type->field();
2537 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2538 // final or stable field
2539 p = make_constant_from_field(field, heap_base_oop);
2540 }
2541
2542 if (p == NULL) { // Could not constant fold the load
2543 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2544 // Normalize the value returned by getBoolean in the following cases
2545 if (type == T_BOOLEAN &&
2546 (mismatched ||
2547 heap_base_oop == top() || // - heap_base_oop is NULL or
2548 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2549 // and the unsafe access is made to large offset
2550 // (i.e., larger than the maximum offset necessary for any
2551 // field access)
2552 ) {
2553 IdealKit ideal = IdealKit(this);
2554 #define __ ideal.
2555 IdealVariable normalized_result(ideal);
2556 __ declarations_done();
2557 __ set(normalized_result, p);
2558 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2559 __ set(normalized_result, ideal.ConI(1));
2560 ideal.end_if();
2561 final_sync(ideal);
2562 p = __ value(normalized_result);
2563 #undef __
2564 }
2565 }
2566 if (type == T_ADDRESS) {
2567 p = gvn().transform(new CastP2XNode(NULL, p));
2568 p = ConvX2UL(p);
2569 }
2570 // The load node has the control of the preceding MemBarCPUOrder. All
2571 // following nodes will have the control of the MemBarCPUOrder inserted at
2572 // the end of this method. So, pushing the load onto the stack at a later
2573 // point is fine.
2574 set_result(p);
2575 } else {
2576 if (bt == T_ADDRESS) {
2577 // Repackage the long as a pointer.
2578 val = ConvL2X(val);
2579 val = gvn().transform(new CastX2PNode(val));
2580 }
2581 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2582 }
2583
2584 return true;
2585 }
2586
2587 //----------------------------inline_unsafe_load_store----------------------------
2588 // This method serves a couple of different customers (depending on LoadStoreKind):
2589 //
2590 // LS_cmp_swap:
2591 //
2592 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2593 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2594 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2595 //
2596 // LS_cmp_swap_weak:
2597 //
2598 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2599 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2600 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2601 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2602 //
2603 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2604 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2605 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2606 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2607 //
2608 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2609 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2610 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2611 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2612 //
2613 // LS_cmp_exchange:
2614 //
2615 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2616 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2617 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2618 //
2619 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2620 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2621 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2622 //
2623 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2624 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2625 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2626 //
2627 // LS_get_add:
2628 //
2629 // int getAndAddInt( Object o, long offset, int delta)
2630 // long getAndAddLong(Object o, long offset, long delta)
2631 //
2632 // LS_get_set:
2633 //
2634 // int getAndSet(Object o, long offset, int newValue)
2635 // long getAndSet(Object o, long offset, long newValue)
2636 // Object getAndSet(Object o, long offset, Object newValue)
2637 //
2638 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2639 // This basic scheme here is the same as inline_unsafe_access, but
2640 // differs in enough details that combining them would make the code
2641 // overly confusing. (This is a true fact! I originally combined
2642 // them, but even I was confused by it!) As much code/comments as
2643 // possible are retained from inline_unsafe_access though to make
2644 // the correspondences clearer. - dl
2645
2646 if (callee()->is_static()) return false; // caller must have the capability!
2647
2648 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2649 decorators |= mo_decorator_for_access_kind(access_kind);
2650
2651 #ifndef PRODUCT
2652 BasicType rtype;
2653 {
2654 ResourceMark rm;
2655 // Check the signatures.
2656 ciSignature* sig = callee()->signature();
2657 rtype = sig->return_type()->basic_type();
2658 switch(kind) {
2659 case LS_get_add:
2660 case LS_get_set: {
2661 // Check the signatures.
2662 #ifdef ASSERT
2663 assert(rtype == type, "get and set must return the expected type");
2664 assert(sig->count() == 3, "get and set has 3 arguments");
2665 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2666 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2667 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2668 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2669 #endif // ASSERT
2670 break;
2671 }
2672 case LS_cmp_swap:
2673 case LS_cmp_swap_weak: {
2674 // Check the signatures.
2675 #ifdef ASSERT
2676 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2677 assert(sig->count() == 4, "CAS has 4 arguments");
2678 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2679 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2680 #endif // ASSERT
2681 break;
2682 }
2683 case LS_cmp_exchange: {
2684 // Check the signatures.
2685 #ifdef ASSERT
2686 assert(rtype == type, "CAS must return the expected type");
2687 assert(sig->count() == 4, "CAS has 4 arguments");
2688 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2689 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2690 #endif // ASSERT
2691 break;
2692 }
2693 default:
2694 ShouldNotReachHere();
2695 }
2696 }
2697 #endif //PRODUCT
2698
2699 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2700
2701 // Get arguments:
2702 Node* receiver = NULL;
2703 Node* base = NULL;
2704 Node* offset = NULL;
2705 Node* oldval = NULL;
2706 Node* newval = NULL;
2707 switch(kind) {
2708 case LS_cmp_swap:
2709 case LS_cmp_swap_weak:
2710 case LS_cmp_exchange: {
2711 const bool two_slot_type = type2size[type] == 2;
2712 receiver = argument(0); // type: oop
2713 base = argument(1); // type: oop
2714 offset = argument(2); // type: long
2715 oldval = argument(4); // type: oop, int, or long
2716 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2717 break;
2718 }
2719 case LS_get_add:
2720 case LS_get_set: {
2721 receiver = argument(0); // type: oop
2722 base = argument(1); // type: oop
2723 offset = argument(2); // type: long
2724 oldval = NULL;
2725 newval = argument(4); // type: oop, int, or long
2726 break;
2727 }
2728 default:
2729 ShouldNotReachHere();
2730 }
2731
2732 // Build field offset expression.
2733 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2734 // to be plain byte offsets, which are also the same as those accepted
2735 // by oopDesc::field_addr.
2736 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2737 // 32-bit machines ignore the high half of long offsets
2738 offset = ConvL2X(offset);
2739 Node* adr = make_unsafe_address(base, offset, ACCESS_WRITE | ACCESS_READ, type, false);
2740 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2741
2742 Compile::AliasType* alias_type = C->alias_type(adr_type);
2743 BasicType bt = alias_type->basic_type();
2744 if (bt != T_ILLEGAL &&
2745 (is_reference_type(bt) != (type == T_OBJECT))) {
2746 // Don't intrinsify mismatched object accesses.
2747 return false;
2748 }
2749
2750 // For CAS, unlike inline_unsafe_access, there seems no point in
2751 // trying to refine types. Just use the coarse types here.
2752 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2753 const Type *value_type = Type::get_const_basic_type(type);
2754
2755 switch (kind) {
2756 case LS_get_set:
2757 case LS_cmp_exchange: {
2758 if (type == T_OBJECT) {
2759 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2760 if (tjp != NULL) {
2761 value_type = tjp;
2762 }
2763 }
2764 break;
2765 }
2766 case LS_cmp_swap:
2767 case LS_cmp_swap_weak:
2768 case LS_get_add:
2769 break;
2770 default:
2771 ShouldNotReachHere();
2772 }
2773
2774 // Null check receiver.
2775 receiver = null_check(receiver);
2776 if (stopped()) {
2777 return true;
2778 }
2779
2780 int alias_idx = C->get_alias_index(adr_type);
2781
2782 if (is_reference_type(type)) {
2783 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2784
2785 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2786 // could be delayed during Parse (for example, in adjust_map_after_if()).
2787 // Execute transformation here to avoid barrier generation in such case.
2788 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2789 newval = _gvn.makecon(TypePtr::NULL_PTR);
2790
2791 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2792 // Refine the value to a null constant, when it is known to be null
2793 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2794 }
2795 }
2796
2797 Node* result = NULL;
2798 switch (kind) {
2799 case LS_cmp_exchange: {
2800 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2801 oldval, newval, value_type, type, decorators);
2802 break;
2803 }
2804 case LS_cmp_swap_weak:
2805 decorators |= C2_WEAK_CMPXCHG;
2806 case LS_cmp_swap: {
2807 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2808 oldval, newval, value_type, type, decorators);
2809 break;
2810 }
2811 case LS_get_set: {
2812 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2813 newval, value_type, type, decorators);
2814 break;
2815 }
2816 case LS_get_add: {
2817 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2818 newval, value_type, type, decorators);
2819 break;
2820 }
2821 default:
2822 ShouldNotReachHere();
2823 }
2824
2825 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2826 set_result(result);
2827 return true;
2828 }
2829
2830 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2831 // Regardless of form, don't allow previous ld/st to move down,
2832 // then issue acquire, release, or volatile mem_bar.
2833 insert_mem_bar(Op_MemBarCPUOrder);
2834 switch(id) {
2835 case vmIntrinsics::_loadFence:
2836 insert_mem_bar(Op_LoadFence);
2837 return true;
2838 case vmIntrinsics::_storeFence:
2839 insert_mem_bar(Op_StoreFence);
2840 return true;
2841 case vmIntrinsics::_fullFence:
2842 insert_mem_bar(Op_MemBarVolatile);
2843 return true;
2844 default:
2845 fatal_unexpected_iid(id);
2846 return false;
2847 }
2848 }
2849
2850 bool LibraryCallKit::inline_onspinwait() {
2851 insert_mem_bar(Op_OnSpinWait);
2852 return true;
2853 }
2854
2855 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2856 if (!kls->is_Con()) {
2857 return true;
2858 }
2859 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
2860 if (klsptr == NULL) {
2861 return true;
2862 }
2863 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
2864 // don't need a guard for a klass that is already initialized
2865 return !ik->is_initialized();
2866 }
2867
2868 //----------------------------inline_unsafe_writeback0-------------------------
2869 // public native void Unsafe.writeback0(long address)
2870 bool LibraryCallKit::inline_unsafe_writeback0() {
2871 if (!Matcher::has_match_rule(Op_CacheWB)) {
2872 return false;
2873 }
2874 #ifndef PRODUCT
2875 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2876 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2877 ciSignature* sig = callee()->signature();
2878 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2879 #endif
2880 null_check_receiver(); // null-check, then ignore
2881 Node *addr = argument(1);
2882 addr = new CastX2PNode(addr);
2883 addr = _gvn.transform(addr);
2884 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2885 flush = _gvn.transform(flush);
2886 set_memory(flush, TypeRawPtr::BOTTOM);
2887 return true;
2888 }
2889
2890 //----------------------------inline_unsafe_writeback0-------------------------
2891 // public native void Unsafe.writeback0(long address)
2892 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2893 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2894 return false;
2895 }
2896 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2897 return false;
2898 }
2899 #ifndef PRODUCT
2900 assert(Matcher::has_match_rule(Op_CacheWB),
2901 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2902 : "found match rule for CacheWBPostSync but not CacheWB"));
2903
2904 #endif
2905 null_check_receiver(); // null-check, then ignore
2906 Node *sync;
2907 if (is_pre) {
2908 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2909 } else {
2910 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2911 }
2912 sync = _gvn.transform(sync);
2913 set_memory(sync, TypeRawPtr::BOTTOM);
2914 return true;
2915 }
2916
2917 //----------------------------inline_unsafe_allocate---------------------------
2918 // public native Object Unsafe.allocateInstance(Class<?> cls);
2919 bool LibraryCallKit::inline_unsafe_allocate() {
2920 if (callee()->is_static()) return false; // caller must have the capability!
2921
2922 null_check_receiver(); // null-check, then ignore
2923 Node* cls = null_check(argument(1));
2924 if (stopped()) return true;
2925
2926 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2927 kls = null_check(kls);
2928 if (stopped()) return true; // argument was like int.class
2929
2930 Node* test = NULL;
2931 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2932 // Note: The argument might still be an illegal value like
2933 // Serializable.class or Object[].class. The runtime will handle it.
2934 // But we must make an explicit check for initialization.
2935 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2936 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2937 // can generate code to load it as unsigned byte.
2938 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2939 Node* bits = intcon(InstanceKlass::fully_initialized);
2940 test = _gvn.transform(new SubINode(inst, bits));
2941 // The 'test' is non-zero if we need to take a slow path.
2942 }
2943
2944 Node* obj = new_instance(kls, test);
2945 set_result(obj);
2946 return true;
2947 }
2948
2949 //------------------------inline_native_time_funcs--------------
2950 // inline code for System.currentTimeMillis() and System.nanoTime()
2951 // these have the same type and signature
2952 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2953 const TypeFunc* tf = OptoRuntime::void_long_Type();
2954 const TypePtr* no_memory_effects = NULL;
2955 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2956 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2957 #ifdef ASSERT
2958 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2959 assert(value_top == top(), "second value must be top");
2960 #endif
2961 set_result(value);
2962 return true;
2963 }
2964
2965 #ifdef JFR_HAVE_INTRINSICS
2966
2967 /*
2968 * oop -> myklass
2969 * myklass->trace_id |= USED
2970 * return myklass->trace_id & ~0x3
2971 */
2972 bool LibraryCallKit::inline_native_classID() {
2973 Node* cls = null_check(argument(0), T_OBJECT);
2974 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2975 kls = null_check(kls, T_OBJECT);
2976
2977 ByteSize offset = KLASS_TRACE_ID_OFFSET;
2978 Node* insp = basic_plus_adr(kls, in_bytes(offset));
2979 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
2980
2981 Node* clsused = longcon(0x01l); // set the class bit
2982 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
2983 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
2984 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
2985
2986 #ifdef TRACE_ID_META_BITS
2987 Node* mbits = longcon(~TRACE_ID_META_BITS);
2988 tvalue = _gvn.transform(new AndLNode(tvalue, mbits));
2989 #endif
2990 #ifdef TRACE_ID_SHIFT
2991 Node* cbits = intcon(TRACE_ID_SHIFT);
2992 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits));
2993 #endif
2994
2995 set_result(tvalue);
2996 return true;
2997
2998 }
2999
3000 bool LibraryCallKit::inline_native_getEventWriter() {
3001 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3002
3003 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
3004 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3005
3006 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3007
3008 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
3009 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
3010
3011 IfNode* iff_jobj_null =
3012 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
3013
3014 enum { _normal_path = 1,
3015 _null_path = 2,
3016 PATH_LIMIT };
3017
3018 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3019 PhiNode* result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3020
3021 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
3022 result_rgn->init_req(_null_path, jobj_is_null);
3023 result_val->init_req(_null_path, null());
3024
3025 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
3026 set_control(jobj_is_not_null);
3027 Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT,
3028 IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3029 result_rgn->init_req(_normal_path, control());
3030 result_val->init_req(_normal_path, res);
3031
3032 set_result(result_rgn, result_val);
3033
3034 return true;
3035 }
3036
3037 #endif // JFR_HAVE_INTRINSICS
3038
3039 //------------------------inline_native_currentThread------------------
3040 bool LibraryCallKit::inline_native_currentThread() {
3041 Node* junk = NULL;
3042 set_result(generate_current_thread(junk));
3043 return true;
3044 }
3045
3046 //---------------------------load_mirror_from_klass----------------------------
3047 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3048 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3049 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3050 Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3051 // mirror = ((OopHandle)mirror)->resolve();
3052 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3053 }
3054
3055 //-----------------------load_klass_from_mirror_common-------------------------
3056 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3057 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3058 // and branch to the given path on the region.
3059 // If never_see_null, take an uncommon trap on null, so we can optimistically
3060 // compile for the non-null case.
3061 // If the region is NULL, force never_see_null = true.
3062 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3063 bool never_see_null,
3064 RegionNode* region,
3065 int null_path,
3066 int offset) {
3067 if (region == NULL) never_see_null = true;
3068 Node* p = basic_plus_adr(mirror, offset);
3069 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3070 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3071 Node* null_ctl = top();
3072 kls = null_check_oop(kls, &null_ctl, never_see_null);
3073 if (region != NULL) {
3074 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3075 region->init_req(null_path, null_ctl);
3076 } else {
3077 assert(null_ctl == top(), "no loose ends");
3078 }
3079 return kls;
3080 }
3081
3082 //--------------------(inline_native_Class_query helpers)---------------------
3083 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3084 // Fall through if (mods & mask) == bits, take the guard otherwise.
3085 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3086 // Branch around if the given klass has the given modifier bit set.
3087 // Like generate_guard, adds a new path onto the region.
3088 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3089 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3090 Node* mask = intcon(modifier_mask);
3091 Node* bits = intcon(modifier_bits);
3092 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3093 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3094 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3095 return generate_fair_guard(bol, region);
3096 }
3097 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3098 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3099 }
3100
3101 //-------------------------inline_native_Class_query-------------------
3102 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3103 const Type* return_type = TypeInt::BOOL;
3104 Node* prim_return_value = top(); // what happens if it's a primitive class?
3105 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3106 bool expect_prim = false; // most of these guys expect to work on refs
3107
3108 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3109
3110 Node* mirror = argument(0);
3111 Node* obj = top();
3112
3113 switch (id) {
3114 case vmIntrinsics::_isInstance:
3115 // nothing is an instance of a primitive type
3116 prim_return_value = intcon(0);
3117 obj = argument(1);
3118 break;
3119 case vmIntrinsics::_getModifiers:
3120 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3121 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3122 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3123 break;
3124 case vmIntrinsics::_isInterface:
3125 prim_return_value = intcon(0);
3126 break;
3127 case vmIntrinsics::_isArray:
3128 prim_return_value = intcon(0);
3129 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3130 break;
3131 case vmIntrinsics::_isPrimitive:
3132 prim_return_value = intcon(1);
3133 expect_prim = true; // obviously
3134 break;
3135 case vmIntrinsics::_getSuperclass:
3136 prim_return_value = null();
3137 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3138 break;
3139 case vmIntrinsics::_getClassAccessFlags:
3140 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3141 return_type = TypeInt::INT; // not bool! 6297094
3142 break;
3143 default:
3144 fatal_unexpected_iid(id);
3145 break;
3146 }
3147
3148 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3149 if (mirror_con == NULL) return false; // cannot happen?
3150
3151 #ifndef PRODUCT
3152 if (C->print_intrinsics() || C->print_inlining()) {
3153 ciType* k = mirror_con->java_mirror_type();
3154 if (k) {
3155 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3156 k->print_name();
3157 tty->cr();
3158 }
3159 }
3160 #endif
3161
3162 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3163 RegionNode* region = new RegionNode(PATH_LIMIT);
3164 record_for_igvn(region);
3165 PhiNode* phi = new PhiNode(region, return_type);
3166
3167 // The mirror will never be null of Reflection.getClassAccessFlags, however
3168 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3169 // if it is. See bug 4774291.
3170
3171 // For Reflection.getClassAccessFlags(), the null check occurs in
3172 // the wrong place; see inline_unsafe_access(), above, for a similar
3173 // situation.
3174 mirror = null_check(mirror);
3175 // If mirror or obj is dead, only null-path is taken.
3176 if (stopped()) return true;
3177
3178 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3179
3180 // Now load the mirror's klass metaobject, and null-check it.
3181 // Side-effects region with the control path if the klass is null.
3182 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3183 // If kls is null, we have a primitive mirror.
3184 phi->init_req(_prim_path, prim_return_value);
3185 if (stopped()) { set_result(region, phi); return true; }
3186 bool safe_for_replace = (region->in(_prim_path) == top());
3187
3188 Node* p; // handy temp
3189 Node* null_ctl;
3190
3191 // Now that we have the non-null klass, we can perform the real query.
3192 // For constant classes, the query will constant-fold in LoadNode::Value.
3193 Node* query_value = top();
3194 switch (id) {
3195 case vmIntrinsics::_isInstance:
3196 // nothing is an instance of a primitive type
3197 query_value = gen_instanceof(obj, kls, safe_for_replace);
3198 break;
3199
3200 case vmIntrinsics::_getModifiers:
3201 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3202 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3203 break;
3204
3205 case vmIntrinsics::_isInterface:
3206 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3207 if (generate_interface_guard(kls, region) != NULL)
3208 // A guard was added. If the guard is taken, it was an interface.
3209 phi->add_req(intcon(1));
3210 // If we fall through, it's a plain class.
3211 query_value = intcon(0);
3212 break;
3213
3214 case vmIntrinsics::_isArray:
3215 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3216 if (generate_array_guard(kls, region) != NULL)
3217 // A guard was added. If the guard is taken, it was an array.
3218 phi->add_req(intcon(1));
3219 // If we fall through, it's a plain class.
3220 query_value = intcon(0);
3221 break;
3222
3223 case vmIntrinsics::_isPrimitive:
3224 query_value = intcon(0); // "normal" path produces false
3225 break;
3226
3227 case vmIntrinsics::_getSuperclass:
3228 // The rules here are somewhat unfortunate, but we can still do better
3229 // with random logic than with a JNI call.
3230 // Interfaces store null or Object as _super, but must report null.
3231 // Arrays store an intermediate super as _super, but must report Object.
3232 // Other types can report the actual _super.
3233 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3234 if (generate_interface_guard(kls, region) != NULL)
3235 // A guard was added. If the guard is taken, it was an interface.
3236 phi->add_req(null());
3237 if (generate_array_guard(kls, region) != NULL)
3238 // A guard was added. If the guard is taken, it was an array.
3239 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3240 // If we fall through, it's a plain class. Get its _super.
3241 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3242 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3243 null_ctl = top();
3244 kls = null_check_oop(kls, &null_ctl);
3245 if (null_ctl != top()) {
3246 // If the guard is taken, Object.superClass is null (both klass and mirror).
3247 region->add_req(null_ctl);
3248 phi ->add_req(null());
3249 }
3250 if (!stopped()) {
3251 query_value = load_mirror_from_klass(kls);
3252 }
3253 break;
3254
3255 case vmIntrinsics::_getClassAccessFlags:
3256 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3257 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3258 break;
3259
3260 default:
3261 fatal_unexpected_iid(id);
3262 break;
3263 }
3264
3265 // Fall-through is the normal case of a query to a real class.
3266 phi->init_req(1, query_value);
3267 region->init_req(1, control());
3268
3269 C->set_has_split_ifs(true); // Has chance for split-if optimization
3270 set_result(region, phi);
3271 return true;
3272 }
3273
3274 //-------------------------inline_Class_cast-------------------
3275 bool LibraryCallKit::inline_Class_cast() {
3276 Node* mirror = argument(0); // Class
3277 Node* obj = argument(1);
3278 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3279 if (mirror_con == NULL) {
3280 return false; // dead path (mirror->is_top()).
3281 }
3282 if (obj == NULL || obj->is_top()) {
3283 return false; // dead path
3284 }
3285 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3286
3287 // First, see if Class.cast() can be folded statically.
3288 // java_mirror_type() returns non-null for compile-time Class constants.
3289 ciType* tm = mirror_con->java_mirror_type();
3290 if (tm != NULL && tm->is_klass() &&
3291 tp != NULL && tp->klass() != NULL) {
3292 if (!tp->klass()->is_loaded()) {
3293 // Don't use intrinsic when class is not loaded.
3294 return false;
3295 } else {
3296 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3297 if (static_res == Compile::SSC_always_true) {
3298 // isInstance() is true - fold the code.
3299 set_result(obj);
3300 return true;
3301 } else if (static_res == Compile::SSC_always_false) {
3302 // Don't use intrinsic, have to throw ClassCastException.
3303 // If the reference is null, the non-intrinsic bytecode will
3304 // be optimized appropriately.
3305 return false;
3306 }
3307 }
3308 }
3309
3310 // Bailout intrinsic and do normal inlining if exception path is frequent.
3311 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3312 return false;
3313 }
3314
3315 // Generate dynamic checks.
3316 // Class.cast() is java implementation of _checkcast bytecode.
3317 // Do checkcast (Parse::do_checkcast()) optimizations here.
3318
3319 mirror = null_check(mirror);
3320 // If mirror is dead, only null-path is taken.
3321 if (stopped()) {
3322 return true;
3323 }
3324
3325 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3326 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3327 RegionNode* region = new RegionNode(PATH_LIMIT);
3328 record_for_igvn(region);
3329
3330 // Now load the mirror's klass metaobject, and null-check it.
3331 // If kls is null, we have a primitive mirror and
3332 // nothing is an instance of a primitive type.
3333 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3334
3335 Node* res = top();
3336 if (!stopped()) {
3337 Node* bad_type_ctrl = top();
3338 // Do checkcast optimizations.
3339 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3340 region->init_req(_bad_type_path, bad_type_ctrl);
3341 }
3342 if (region->in(_prim_path) != top() ||
3343 region->in(_bad_type_path) != top()) {
3344 // Let Interpreter throw ClassCastException.
3345 PreserveJVMState pjvms(this);
3346 set_control(_gvn.transform(region));
3347 uncommon_trap(Deoptimization::Reason_intrinsic,
3348 Deoptimization::Action_maybe_recompile);
3349 }
3350 if (!stopped()) {
3351 set_result(res);
3352 }
3353 return true;
3354 }
3355
3356
3357 //--------------------------inline_native_subtype_check------------------------
3358 // This intrinsic takes the JNI calls out of the heart of
3359 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3360 bool LibraryCallKit::inline_native_subtype_check() {
3361 // Pull both arguments off the stack.
3362 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3363 args[0] = argument(0);
3364 args[1] = argument(1);
3365 Node* klasses[2]; // corresponding Klasses: superk, subk
3366 klasses[0] = klasses[1] = top();
3367
3368 enum {
3369 // A full decision tree on {superc is prim, subc is prim}:
3370 _prim_0_path = 1, // {P,N} => false
3371 // {P,P} & superc!=subc => false
3372 _prim_same_path, // {P,P} & superc==subc => true
3373 _prim_1_path, // {N,P} => false
3374 _ref_subtype_path, // {N,N} & subtype check wins => true
3375 _both_ref_path, // {N,N} & subtype check loses => false
3376 PATH_LIMIT
3377 };
3378
3379 RegionNode* region = new RegionNode(PATH_LIMIT);
3380 Node* phi = new PhiNode(region, TypeInt::BOOL);
3381 record_for_igvn(region);
3382
3383 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3384 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3385 int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3386
3387 // First null-check both mirrors and load each mirror's klass metaobject.
3388 int which_arg;
3389 for (which_arg = 0; which_arg <= 1; which_arg++) {
3390 Node* arg = args[which_arg];
3391 arg = null_check(arg);
3392 if (stopped()) break;
3393 args[which_arg] = arg;
3394
3395 Node* p = basic_plus_adr(arg, class_klass_offset);
3396 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3397 klasses[which_arg] = _gvn.transform(kls);
3398 }
3399
3400 // Resolve oops to stable for CmpP below.
3401 args[0] = access_resolve(args[0], 0);
3402 args[1] = access_resolve(args[1], 0);
3403
3404 // Having loaded both klasses, test each for null.
3405 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3406 for (which_arg = 0; which_arg <= 1; which_arg++) {
3407 Node* kls = klasses[which_arg];
3408 Node* null_ctl = top();
3409 kls = null_check_oop(kls, &null_ctl, never_see_null);
3410 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3411 region->init_req(prim_path, null_ctl);
3412 if (stopped()) break;
3413 klasses[which_arg] = kls;
3414 }
3415
3416 if (!stopped()) {
3417 // now we have two reference types, in klasses[0..1]
3418 Node* subk = klasses[1]; // the argument to isAssignableFrom
3419 Node* superk = klasses[0]; // the receiver
3420 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3421 // now we have a successful reference subtype check
3422 region->set_req(_ref_subtype_path, control());
3423 }
3424
3425 // If both operands are primitive (both klasses null), then
3426 // we must return true when they are identical primitives.
3427 // It is convenient to test this after the first null klass check.
3428 set_control(region->in(_prim_0_path)); // go back to first null check
3429 if (!stopped()) {
3430 // Since superc is primitive, make a guard for the superc==subc case.
3431 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3432 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3433 generate_guard(bol_eq, region, PROB_FAIR);
3434 if (region->req() == PATH_LIMIT+1) {
3435 // A guard was added. If the added guard is taken, superc==subc.
3436 region->swap_edges(PATH_LIMIT, _prim_same_path);
3437 region->del_req(PATH_LIMIT);
3438 }
3439 region->set_req(_prim_0_path, control()); // Not equal after all.
3440 }
3441
3442 // these are the only paths that produce 'true':
3443 phi->set_req(_prim_same_path, intcon(1));
3444 phi->set_req(_ref_subtype_path, intcon(1));
3445
3446 // pull together the cases:
3447 assert(region->req() == PATH_LIMIT, "sane region");
3448 for (uint i = 1; i < region->req(); i++) {
3449 Node* ctl = region->in(i);
3450 if (ctl == NULL || ctl == top()) {
3451 region->set_req(i, top());
3452 phi ->set_req(i, top());
3453 } else if (phi->in(i) == NULL) {
3454 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3455 }
3456 }
3457
3458 set_control(_gvn.transform(region));
3459 set_result(_gvn.transform(phi));
3460 return true;
3461 }
3462
3463 //---------------------generate_array_guard_common------------------------
3464 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3465 bool obj_array, bool not_array) {
3466
3467 if (stopped()) {
3468 return NULL;
3469 }
3470
3471 // If obj_array/non_array==false/false:
3472 // Branch around if the given klass is in fact an array (either obj or prim).
3473 // If obj_array/non_array==false/true:
3474 // Branch around if the given klass is not an array klass of any kind.
3475 // If obj_array/non_array==true/true:
3476 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3477 // If obj_array/non_array==true/false:
3478 // Branch around if the kls is an oop array (Object[] or subtype)
3479 //
3480 // Like generate_guard, adds a new path onto the region.
3481 jint layout_con = 0;
3482 Node* layout_val = get_layout_helper(kls, layout_con);
3483 if (layout_val == NULL) {
3484 bool query = (obj_array
3485 ? Klass::layout_helper_is_objArray(layout_con)
3486 : Klass::layout_helper_is_array(layout_con));
3487 if (query == not_array) {
3488 return NULL; // never a branch
3489 } else { // always a branch
3490 Node* always_branch = control();
3491 if (region != NULL)
3492 region->add_req(always_branch);
3493 set_control(top());
3494 return always_branch;
3495 }
3496 }
3497 // Now test the correct condition.
3498 jint nval = (obj_array
3499 ? (jint)(Klass::_lh_array_tag_type_value
3500 << Klass::_lh_array_tag_shift)
3501 : Klass::_lh_neutral_value);
3502 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3503 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3504 // invert the test if we are looking for a non-array
3505 if (not_array) btest = BoolTest(btest).negate();
3506 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3507 return generate_fair_guard(bol, region);
3508 }
3509
3510
3511 //-----------------------inline_native_newArray--------------------------
3512 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3513 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3514 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3515 Node* mirror;
3516 Node* count_val;
3517 if (uninitialized) {
3518 mirror = argument(1);
3519 count_val = argument(2);
3520 } else {
3521 mirror = argument(0);
3522 count_val = argument(1);
3523 }
3524
3525 mirror = null_check(mirror);
3526 // If mirror or obj is dead, only null-path is taken.
3527 if (stopped()) return true;
3528
3529 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3530 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3531 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3532 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3533 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3534
3535 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3536 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3537 result_reg, _slow_path);
3538 Node* normal_ctl = control();
3539 Node* no_array_ctl = result_reg->in(_slow_path);
3540
3541 // Generate code for the slow case. We make a call to newArray().
3542 set_control(no_array_ctl);
3543 if (!stopped()) {
3544 // Either the input type is void.class, or else the
3545 // array klass has not yet been cached. Either the
3546 // ensuing call will throw an exception, or else it
3547 // will cache the array klass for next time.
3548 PreserveJVMState pjvms(this);
3549 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3550 Node* slow_result = set_results_for_java_call(slow_call);
3551 // this->control() comes from set_results_for_java_call
3552 result_reg->set_req(_slow_path, control());
3553 result_val->set_req(_slow_path, slow_result);
3554 result_io ->set_req(_slow_path, i_o());
3555 result_mem->set_req(_slow_path, reset_memory());
3556 }
3557
3558 set_control(normal_ctl);
3559 if (!stopped()) {
3560 // Normal case: The array type has been cached in the java.lang.Class.
3561 // The following call works fine even if the array type is polymorphic.
3562 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3563 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3564 result_reg->init_req(_normal_path, control());
3565 result_val->init_req(_normal_path, obj);
3566 result_io ->init_req(_normal_path, i_o());
3567 result_mem->init_req(_normal_path, reset_memory());
3568
3569 if (uninitialized) {
3570 // Mark the allocation so that zeroing is skipped
3571 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3572 alloc->maybe_set_complete(&_gvn);
3573 }
3574 }
3575
3576 // Return the combined state.
3577 set_i_o( _gvn.transform(result_io) );
3578 set_all_memory( _gvn.transform(result_mem));
3579
3580 C->set_has_split_ifs(true); // Has chance for split-if optimization
3581 set_result(result_reg, result_val);
3582 return true;
3583 }
3584
3585 //----------------------inline_native_getLength--------------------------
3586 // public static native int java.lang.reflect.Array.getLength(Object array);
3587 bool LibraryCallKit::inline_native_getLength() {
3588 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3589
3590 Node* array = null_check(argument(0));
3591 // If array is dead, only null-path is taken.
3592 if (stopped()) return true;
3593
3594 // Deoptimize if it is a non-array.
3595 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3596
3597 if (non_array != NULL) {
3598 PreserveJVMState pjvms(this);
3599 set_control(non_array);
3600 uncommon_trap(Deoptimization::Reason_intrinsic,
3601 Deoptimization::Action_maybe_recompile);
3602 }
3603
3604 // If control is dead, only non-array-path is taken.
3605 if (stopped()) return true;
3606
3607 // The works fine even if the array type is polymorphic.
3608 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3609 Node* result = load_array_length(array);
3610
3611 C->set_has_split_ifs(true); // Has chance for split-if optimization
3612 set_result(result);
3613 return true;
3614 }
3615
3616 //------------------------inline_array_copyOf----------------------------
3617 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3618 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3619 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3620 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3621
3622 // Get the arguments.
3623 Node* original = argument(0);
3624 Node* start = is_copyOfRange? argument(1): intcon(0);
3625 Node* end = is_copyOfRange? argument(2): argument(1);
3626 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3627
3628 Node* newcopy = NULL;
3629
3630 // Set the original stack and the reexecute bit for the interpreter to reexecute
3631 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3632 { PreserveReexecuteState preexecs(this);
3633 jvms()->set_should_reexecute(true);
3634
3635 array_type_mirror = null_check(array_type_mirror);
3636 original = null_check(original);
3637
3638 // Check if a null path was taken unconditionally.
3639 if (stopped()) return true;
3640
3641 Node* orig_length = load_array_length(original);
3642
3643 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3644 klass_node = null_check(klass_node);
3645
3646 RegionNode* bailout = new RegionNode(1);
3647 record_for_igvn(bailout);
3648
3649 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3650 // Bail out if that is so.
3651 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3652 if (not_objArray != NULL) {
3653 // Improve the klass node's type from the new optimistic assumption:
3654 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3655 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3656 Node* cast = new CastPPNode(klass_node, akls);
3657 cast->init_req(0, control());
3658 klass_node = _gvn.transform(cast);
3659 }
3660
3661 // Bail out if either start or end is negative.
3662 generate_negative_guard(start, bailout, &start);
3663 generate_negative_guard(end, bailout, &end);
3664
3665 Node* length = end;
3666 if (_gvn.type(start) != TypeInt::ZERO) {
3667 length = _gvn.transform(new SubINode(end, start));
3668 }
3669
3670 // Bail out if length is negative.
3671 // Without this the new_array would throw
3672 // NegativeArraySizeException but IllegalArgumentException is what
3673 // should be thrown
3674 generate_negative_guard(length, bailout, &length);
3675
3676 if (bailout->req() > 1) {
3677 PreserveJVMState pjvms(this);
3678 set_control(_gvn.transform(bailout));
3679 uncommon_trap(Deoptimization::Reason_intrinsic,
3680 Deoptimization::Action_maybe_recompile);
3681 }
3682
3683 if (!stopped()) {
3684 // How many elements will we copy from the original?
3685 // The answer is MinI(orig_length - start, length).
3686 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3687 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3688
3689 original = access_resolve(original, ACCESS_READ);
3690
3691 // Generate a direct call to the right arraycopy function(s).
3692 // We know the copy is disjoint but we might not know if the
3693 // oop stores need checking.
3694 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3695 // This will fail a store-check if x contains any non-nulls.
3696
3697 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3698 // loads/stores but it is legal only if we're sure the
3699 // Arrays.copyOf would succeed. So we need all input arguments
3700 // to the copyOf to be validated, including that the copy to the
3701 // new array won't trigger an ArrayStoreException. That subtype
3702 // check can be optimized if we know something on the type of
3703 // the input array from type speculation.
3704 if (_gvn.type(klass_node)->singleton()) {
3705 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3706 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3707
3708 int test = C->static_subtype_check(superk, subk);
3709 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3710 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3711 if (t_original->speculative_type() != NULL) {
3712 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3713 }
3714 }
3715 }
3716
3717 bool validated = false;
3718 // Reason_class_check rather than Reason_intrinsic because we
3719 // want to intrinsify even if this traps.
3720 if (!too_many_traps(Deoptimization::Reason_class_check)) {
3721 Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
3722 klass_node);
3723
3724 if (not_subtype_ctrl != top()) {
3725 PreserveJVMState pjvms(this);
3726 set_control(not_subtype_ctrl);
3727 uncommon_trap(Deoptimization::Reason_class_check,
3728 Deoptimization::Action_make_not_entrant);
3729 assert(stopped(), "Should be stopped");
3730 }
3731 validated = true;
3732 }
3733
3734 if (!stopped()) {
3735 newcopy = new_array(klass_node, length, 0); // no arguments to push
3736
3737 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
3738 load_object_klass(original), klass_node);
3739 if (!is_copyOfRange) {
3740 ac->set_copyof(validated);
3741 } else {
3742 ac->set_copyofrange(validated);
3743 }
3744 Node* n = _gvn.transform(ac);
3745 if (n == ac) {
3746 ac->connect_outputs(this);
3747 } else {
3748 assert(validated, "shouldn't transform if all arguments not validated");
3749 set_all_memory(n);
3750 }
3751 }
3752 }
3753 } // original reexecute is set back here
3754
3755 C->set_has_split_ifs(true); // Has chance for split-if optimization
3756 if (!stopped()) {
3757 set_result(newcopy);
3758 }
3759 return true;
3760 }
3761
3762
3763 //----------------------generate_virtual_guard---------------------------
3764 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
3765 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
3766 RegionNode* slow_region) {
3767 ciMethod* method = callee();
3768 int vtable_index = method->vtable_index();
3769 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3770 "bad index %d", vtable_index);
3771 // Get the Method* out of the appropriate vtable entry.
3772 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
3773 vtable_index*vtableEntry::size_in_bytes() +
3774 vtableEntry::method_offset_in_bytes();
3775 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
3776 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3777
3778 // Compare the target method with the expected method (e.g., Object.hashCode).
3779 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
3780
3781 Node* native_call = makecon(native_call_addr);
3782 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
3783 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
3784
3785 return generate_slow_guard(test_native, slow_region);
3786 }
3787
3788 //-----------------------generate_method_call----------------------------
3789 // Use generate_method_call to make a slow-call to the real
3790 // method if the fast path fails. An alternative would be to
3791 // use a stub like OptoRuntime::slow_arraycopy_Java.
3792 // This only works for expanding the current library call,
3793 // not another intrinsic. (E.g., don't use this for making an
3794 // arraycopy call inside of the copyOf intrinsic.)
3795 CallJavaNode*
3796 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
3797 // When compiling the intrinsic method itself, do not use this technique.
3798 guarantee(callee() != C->method(), "cannot make slow-call to self");
3799
3800 ciMethod* method = callee();
3801 // ensure the JVMS we have will be correct for this call
3802 guarantee(method_id == method->intrinsic_id(), "must match");
3803
3804 const TypeFunc* tf = TypeFunc::make(method);
3805 CallJavaNode* slow_call;
3806 if (is_static) {
3807 assert(!is_virtual, "");
3808 slow_call = new CallStaticJavaNode(C, tf,
3809 SharedRuntime::get_resolve_static_call_stub(),
3810 method, bci());
3811 } else if (is_virtual) {
3812 null_check_receiver();
3813 int vtable_index = Method::invalid_vtable_index;
3814 if (UseInlineCaches) {
3815 // Suppress the vtable call
3816 } else {
3817 // hashCode and clone are not a miranda methods,
3818 // so the vtable index is fixed.
3819 // No need to use the linkResolver to get it.
3820 vtable_index = method->vtable_index();
3821 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3822 "bad index %d", vtable_index);
3823 }
3824 slow_call = new CallDynamicJavaNode(tf,
3825 SharedRuntime::get_resolve_virtual_call_stub(),
3826 method, vtable_index, bci());
3827 } else { // neither virtual nor static: opt_virtual
3828 null_check_receiver();
3829 slow_call = new CallStaticJavaNode(C, tf,
3830 SharedRuntime::get_resolve_opt_virtual_call_stub(),
3831 method, bci());
3832 slow_call->set_optimized_virtual(true);
3833 }
3834 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
3835 // To be able to issue a direct call (optimized virtual or virtual)
3836 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
3837 // about the method being invoked should be attached to the call site to
3838 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
3839 slow_call->set_override_symbolic_info(true);
3840 }
3841 set_arguments_for_java_call(slow_call);
3842 set_edges_for_java_call(slow_call);
3843 return slow_call;
3844 }
3845
3846
3847 /**
3848 * Build special case code for calls to hashCode on an object. This call may
3849 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
3850 * slightly different code.
3851 */
3852 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
3853 assert(is_static == callee()->is_static(), "correct intrinsic selection");
3854 assert(!(is_virtual && is_static), "either virtual, special, or static");
3855
3856 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
3857
3858 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3859 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
3860 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3861 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3862 Node* obj = NULL;
3863 if (!is_static) {
3864 // Check for hashing null object
3865 obj = null_check_receiver();
3866 if (stopped()) return true; // unconditionally null
3867 result_reg->init_req(_null_path, top());
3868 result_val->init_req(_null_path, top());
3869 } else {
3870 // Do a null check, and return zero if null.
3871 // System.identityHashCode(null) == 0
3872 obj = argument(0);
3873 Node* null_ctl = top();
3874 obj = null_check_oop(obj, &null_ctl);
3875 result_reg->init_req(_null_path, null_ctl);
3876 result_val->init_req(_null_path, _gvn.intcon(0));
3877 }
3878
3879 // Unconditionally null? Then return right away.
3880 if (stopped()) {
3881 set_control( result_reg->in(_null_path));
3882 if (!stopped())
3883 set_result(result_val->in(_null_path));
3884 return true;
3885 }
3886
3887 // We only go to the fast case code if we pass a number of guards. The
3888 // paths which do not pass are accumulated in the slow_region.
3889 RegionNode* slow_region = new RegionNode(1);
3890 record_for_igvn(slow_region);
3891
3892 // If this is a virtual call, we generate a funny guard. We pull out
3893 // the vtable entry corresponding to hashCode() from the target object.
3894 // If the target method which we are calling happens to be the native
3895 // Object hashCode() method, we pass the guard. We do not need this
3896 // guard for non-virtual calls -- the caller is known to be the native
3897 // Object hashCode().
3898 if (is_virtual) {
3899 // After null check, get the object's klass.
3900 Node* obj_klass = load_object_klass(obj);
3901 generate_virtual_guard(obj_klass, slow_region);
3902 }
3903
3904 // Get the header out of the object, use LoadMarkNode when available
3905 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
3906 // The control of the load must be NULL. Otherwise, the load can move before
3907 // the null check after castPP removal.
3908 Node* no_ctrl = NULL;
3909 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
3910
3911 // Test the header to see if it is unlocked.
3912 Node *lock_mask = _gvn.MakeConX(markWord::biased_lock_mask_in_place);
3913 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
3914 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
3915 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
3916 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
3917
3918 generate_slow_guard(test_unlocked, slow_region);
3919
3920 // Get the hash value and check to see that it has been properly assigned.
3921 // We depend on hash_mask being at most 32 bits and avoid the use of
3922 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
3923 // vm: see markWord.hpp.
3924 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
3925 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
3926 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
3927 // This hack lets the hash bits live anywhere in the mark object now, as long
3928 // as the shift drops the relevant bits into the low 32 bits. Note that
3929 // Java spec says that HashCode is an int so there's no point in capturing
3930 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
3931 hshifted_header = ConvX2I(hshifted_header);
3932 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
3933
3934 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
3935 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
3936 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
3937
3938 generate_slow_guard(test_assigned, slow_region);
3939
3940 Node* init_mem = reset_memory();
3941 // fill in the rest of the null path:
3942 result_io ->init_req(_null_path, i_o());
3943 result_mem->init_req(_null_path, init_mem);
3944
3945 result_val->init_req(_fast_path, hash_val);
3946 result_reg->init_req(_fast_path, control());
3947 result_io ->init_req(_fast_path, i_o());
3948 result_mem->init_req(_fast_path, init_mem);
3949
3950 // Generate code for the slow case. We make a call to hashCode().
3951 set_control(_gvn.transform(slow_region));
3952 if (!stopped()) {
3953 // No need for PreserveJVMState, because we're using up the present state.
3954 set_all_memory(init_mem);
3955 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
3956 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
3957 Node* slow_result = set_results_for_java_call(slow_call);
3958 // this->control() comes from set_results_for_java_call
3959 result_reg->init_req(_slow_path, control());
3960 result_val->init_req(_slow_path, slow_result);
3961 result_io ->set_req(_slow_path, i_o());
3962 result_mem ->set_req(_slow_path, reset_memory());
3963 }
3964
3965 // Return the combined state.
3966 set_i_o( _gvn.transform(result_io) );
3967 set_all_memory( _gvn.transform(result_mem));
3968
3969 set_result(result_reg, result_val);
3970 return true;
3971 }
3972
3973 //---------------------------inline_native_getClass----------------------------
3974 // public final native Class<?> java.lang.Object.getClass();
3975 //
3976 // Build special case code for calls to getClass on an object.
3977 bool LibraryCallKit::inline_native_getClass() {
3978 Node* obj = null_check_receiver();
3979 if (stopped()) return true;
3980 set_result(load_mirror_from_klass(load_object_klass(obj)));
3981 return true;
3982 }
3983
3984 //-----------------inline_native_Reflection_getCallerClass---------------------
3985 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
3986 //
3987 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
3988 //
3989 // NOTE: This code must perform the same logic as JVM_GetCallerClass
3990 // in that it must skip particular security frames and checks for
3991 // caller sensitive methods.
3992 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
3993 #ifndef PRODUCT
3994 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3995 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
3996 }
3997 #endif
3998
3999 if (!jvms()->has_method()) {
4000 #ifndef PRODUCT
4001 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4002 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4003 }
4004 #endif
4005 return false;
4006 }
4007
4008 // Walk back up the JVM state to find the caller at the required
4009 // depth.
4010 JVMState* caller_jvms = jvms();
4011
4012 // Cf. JVM_GetCallerClass
4013 // NOTE: Start the loop at depth 1 because the current JVM state does
4014 // not include the Reflection.getCallerClass() frame.
4015 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4016 ciMethod* m = caller_jvms->method();
4017 switch (n) {
4018 case 0:
4019 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4020 break;
4021 case 1:
4022 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4023 if (!m->caller_sensitive()) {
4024 #ifndef PRODUCT
4025 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4026 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4027 }
4028 #endif
4029 return false; // bail-out; let JVM_GetCallerClass do the work
4030 }
4031 break;
4032 default:
4033 if (!m->is_ignored_by_security_stack_walk()) {
4034 // We have reached the desired frame; return the holder class.
4035 // Acquire method holder as java.lang.Class and push as constant.
4036 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4037 ciInstance* caller_mirror = caller_klass->java_mirror();
4038 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4039
4040 #ifndef PRODUCT
4041 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4042 tty->print_cr(" Succeeded: caller = %d) %s.%s, JVMS depth = %d", n, caller_klass->name()->as_utf8(), caller_jvms->method()->name()->as_utf8(), jvms()->depth());
4043 tty->print_cr(" JVM state at this point:");
4044 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4045 ciMethod* m = jvms()->of_depth(i)->method();
4046 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4047 }
4048 }
4049 #endif
4050 return true;
4051 }
4052 break;
4053 }
4054 }
4055
4056 #ifndef PRODUCT
4057 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4058 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4059 tty->print_cr(" JVM state at this point:");
4060 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4061 ciMethod* m = jvms()->of_depth(i)->method();
4062 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4063 }
4064 }
4065 #endif
4066
4067 return false; // bail-out; let JVM_GetCallerClass do the work
4068 }
4069
4070 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4071 Node* arg = argument(0);
4072 Node* result = NULL;
4073
4074 switch (id) {
4075 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4076 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4077 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4078 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4079
4080 case vmIntrinsics::_doubleToLongBits: {
4081 // two paths (plus control) merge in a wood
4082 RegionNode *r = new RegionNode(3);
4083 Node *phi = new PhiNode(r, TypeLong::LONG);
4084
4085 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4086 // Build the boolean node
4087 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4088
4089 // Branch either way.
4090 // NaN case is less traveled, which makes all the difference.
4091 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4092 Node *opt_isnan = _gvn.transform(ifisnan);
4093 assert( opt_isnan->is_If(), "Expect an IfNode");
4094 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4095 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4096
4097 set_control(iftrue);
4098
4099 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4100 Node *slow_result = longcon(nan_bits); // return NaN
4101 phi->init_req(1, _gvn.transform( slow_result ));
4102 r->init_req(1, iftrue);
4103
4104 // Else fall through
4105 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4106 set_control(iffalse);
4107
4108 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4109 r->init_req(2, iffalse);
4110
4111 // Post merge
4112 set_control(_gvn.transform(r));
4113 record_for_igvn(r);
4114
4115 C->set_has_split_ifs(true); // Has chance for split-if optimization
4116 result = phi;
4117 assert(result->bottom_type()->isa_long(), "must be");
4118 break;
4119 }
4120
4121 case vmIntrinsics::_floatToIntBits: {
4122 // two paths (plus control) merge in a wood
4123 RegionNode *r = new RegionNode(3);
4124 Node *phi = new PhiNode(r, TypeInt::INT);
4125
4126 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4127 // Build the boolean node
4128 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4129
4130 // Branch either way.
4131 // NaN case is less traveled, which makes all the difference.
4132 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4133 Node *opt_isnan = _gvn.transform(ifisnan);
4134 assert( opt_isnan->is_If(), "Expect an IfNode");
4135 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4136 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4137
4138 set_control(iftrue);
4139
4140 static const jint nan_bits = 0x7fc00000;
4141 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4142 phi->init_req(1, _gvn.transform( slow_result ));
4143 r->init_req(1, iftrue);
4144
4145 // Else fall through
4146 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4147 set_control(iffalse);
4148
4149 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4150 r->init_req(2, iffalse);
4151
4152 // Post merge
4153 set_control(_gvn.transform(r));
4154 record_for_igvn(r);
4155
4156 C->set_has_split_ifs(true); // Has chance for split-if optimization
4157 result = phi;
4158 assert(result->bottom_type()->isa_int(), "must be");
4159 break;
4160 }
4161
4162 default:
4163 fatal_unexpected_iid(id);
4164 break;
4165 }
4166 set_result(_gvn.transform(result));
4167 return true;
4168 }
4169
4170 //----------------------inline_unsafe_copyMemory-------------------------
4171 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4172 bool LibraryCallKit::inline_unsafe_copyMemory() {
4173 if (callee()->is_static()) return false; // caller must have the capability!
4174 null_check_receiver(); // null-check receiver
4175 if (stopped()) return true;
4176
4177 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4178
4179 Node* src_ptr = argument(1); // type: oop
4180 Node* src_off = ConvL2X(argument(2)); // type: long
4181 Node* dst_ptr = argument(4); // type: oop
4182 Node* dst_off = ConvL2X(argument(5)); // type: long
4183 Node* size = ConvL2X(argument(7)); // type: long
4184
4185 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4186 "fieldOffset must be byte-scaled");
4187
4188 src_ptr = access_resolve(src_ptr, ACCESS_READ);
4189 dst_ptr = access_resolve(dst_ptr, ACCESS_WRITE);
4190 Node* src = make_unsafe_address(src_ptr, src_off, ACCESS_READ);
4191 Node* dst = make_unsafe_address(dst_ptr, dst_off, ACCESS_WRITE);
4192
4193 // Conservatively insert a memory barrier on all memory slices.
4194 // Do not let writes of the copy source or destination float below the copy.
4195 insert_mem_bar(Op_MemBarCPUOrder);
4196
4197 Node* thread = _gvn.transform(new ThreadLocalNode());
4198 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4199 BasicType doing_unsafe_access_bt = T_BYTE;
4200 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4201
4202 // update volatile field
4203 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4204
4205 // Call it. Note that the length argument is not scaled.
4206 make_runtime_call(RC_LEAF|RC_NO_FP,
4207 OptoRuntime::fast_arraycopy_Type(),
4208 StubRoutines::unsafe_arraycopy(),
4209 "unsafe_arraycopy",
4210 TypeRawPtr::BOTTOM,
4211 src, dst, size XTOP);
4212
4213 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4214
4215 // Do not let reads of the copy destination float above the copy.
4216 insert_mem_bar(Op_MemBarCPUOrder);
4217
4218 return true;
4219 }
4220
4221 //------------------------clone_coping-----------------------------------
4222 // Helper function for inline_native_clone.
4223 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4224 assert(obj_size != NULL, "");
4225 Node* raw_obj = alloc_obj->in(1);
4226 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4227
4228 AllocateNode* alloc = NULL;
4229 if (ReduceBulkZeroing) {
4230 // We will be completely responsible for initializing this object -
4231 // mark Initialize node as complete.
4232 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4233 // The object was just allocated - there should be no any stores!
4234 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4235 // Mark as complete_with_arraycopy so that on AllocateNode
4236 // expansion, we know this AllocateNode is initialized by an array
4237 // copy and a StoreStore barrier exists after the array copy.
4238 alloc->initialization()->set_complete_with_arraycopy();
4239 }
4240
4241 Node* size = _gvn.transform(obj_size);
4242 access_clone(obj, alloc_obj, size, is_array);
4243
4244 // Do not let reads from the cloned object float above the arraycopy.
4245 if (alloc != NULL) {
4246 // Do not let stores that initialize this object be reordered with
4247 // a subsequent store that would make this object accessible by
4248 // other threads.
4249 // Record what AllocateNode this StoreStore protects so that
4250 // escape analysis can go from the MemBarStoreStoreNode to the
4251 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4252 // based on the escape status of the AllocateNode.
4253 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4254 } else {
4255 insert_mem_bar(Op_MemBarCPUOrder);
4256 }
4257 }
4258
4259 //------------------------inline_native_clone----------------------------
4260 // protected native Object java.lang.Object.clone();
4261 //
4262 // Here are the simple edge cases:
4263 // null receiver => normal trap
4264 // virtual and clone was overridden => slow path to out-of-line clone
4265 // not cloneable or finalizer => slow path to out-of-line Object.clone
4266 //
4267 // The general case has two steps, allocation and copying.
4268 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4269 //
4270 // Copying also has two cases, oop arrays and everything else.
4271 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4272 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4273 //
4274 // These steps fold up nicely if and when the cloned object's klass
4275 // can be sharply typed as an object array, a type array, or an instance.
4276 //
4277 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4278 PhiNode* result_val;
4279
4280 // Set the reexecute bit for the interpreter to reexecute
4281 // the bytecode that invokes Object.clone if deoptimization happens.
4282 { PreserveReexecuteState preexecs(this);
4283 jvms()->set_should_reexecute(true);
4284
4285 Node* obj = null_check_receiver();
4286 if (stopped()) return true;
4287
4288 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4289
4290 // If we are going to clone an instance, we need its exact type to
4291 // know the number and types of fields to convert the clone to
4292 // loads/stores. Maybe a speculative type can help us.
4293 if (!obj_type->klass_is_exact() &&
4294 obj_type->speculative_type() != NULL &&
4295 obj_type->speculative_type()->is_instance_klass()) {
4296 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4297 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4298 !spec_ik->has_injected_fields()) {
4299 ciKlass* k = obj_type->klass();
4300 if (!k->is_instance_klass() ||
4301 k->as_instance_klass()->is_interface() ||
4302 k->as_instance_klass()->has_subklass()) {
4303 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4304 }
4305 }
4306 }
4307
4308 // Conservatively insert a memory barrier on all memory slices.
4309 // Do not let writes into the original float below the clone.
4310 insert_mem_bar(Op_MemBarCPUOrder);
4311
4312 // paths into result_reg:
4313 enum {
4314 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4315 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4316 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4317 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4318 PATH_LIMIT
4319 };
4320 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4321 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4322 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4323 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4324 record_for_igvn(result_reg);
4325
4326 Node* obj_klass = load_object_klass(obj);
4327 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4328 if (array_ctl != NULL) {
4329 // It's an array.
4330 PreserveJVMState pjvms(this);
4331 set_control(array_ctl);
4332 Node* obj_length = load_array_length(obj);
4333 Node* obj_size = NULL;
4334 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size); // no arguments to push
4335
4336 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4337 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, BarrierSetC2::Parsing)) {
4338 // If it is an oop array, it requires very special treatment,
4339 // because gc barriers are required when accessing the array.
4340 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4341 if (is_obja != NULL) {
4342 PreserveJVMState pjvms2(this);
4343 set_control(is_obja);
4344 obj = access_resolve(obj, ACCESS_READ);
4345 // Generate a direct call to the right arraycopy function(s).
4346 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4347 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false);
4348 ac->set_clone_oop_array();
4349 Node* n = _gvn.transform(ac);
4350 assert(n == ac, "cannot disappear");
4351 ac->connect_outputs(this);
4352
4353 result_reg->init_req(_objArray_path, control());
4354 result_val->init_req(_objArray_path, alloc_obj);
4355 result_i_o ->set_req(_objArray_path, i_o());
4356 result_mem ->set_req(_objArray_path, reset_memory());
4357 }
4358 }
4359 // Otherwise, there are no barriers to worry about.
4360 // (We can dispense with card marks if we know the allocation
4361 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4362 // causes the non-eden paths to take compensating steps to
4363 // simulate a fresh allocation, so that no further
4364 // card marks are required in compiled code to initialize
4365 // the object.)
4366
4367 if (!stopped()) {
4368 copy_to_clone(obj, alloc_obj, obj_size, true);
4369
4370 // Present the results of the copy.
4371 result_reg->init_req(_array_path, control());
4372 result_val->init_req(_array_path, alloc_obj);
4373 result_i_o ->set_req(_array_path, i_o());
4374 result_mem ->set_req(_array_path, reset_memory());
4375 }
4376 }
4377
4378 // We only go to the instance fast case code if we pass a number of guards.
4379 // The paths which do not pass are accumulated in the slow_region.
4380 RegionNode* slow_region = new RegionNode(1);
4381 record_for_igvn(slow_region);
4382 if (!stopped()) {
4383 // It's an instance (we did array above). Make the slow-path tests.
4384 // If this is a virtual call, we generate a funny guard. We grab
4385 // the vtable entry corresponding to clone() from the target object.
4386 // If the target method which we are calling happens to be the
4387 // Object clone() method, we pass the guard. We do not need this
4388 // guard for non-virtual calls; the caller is known to be the native
4389 // Object clone().
4390 if (is_virtual) {
4391 generate_virtual_guard(obj_klass, slow_region);
4392 }
4393
4394 // The object must be easily cloneable and must not have a finalizer.
4395 // Both of these conditions may be checked in a single test.
4396 // We could optimize the test further, but we don't care.
4397 generate_access_flags_guard(obj_klass,
4398 // Test both conditions:
4399 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4400 // Must be cloneable but not finalizer:
4401 JVM_ACC_IS_CLONEABLE_FAST,
4402 slow_region);
4403 }
4404
4405 if (!stopped()) {
4406 // It's an instance, and it passed the slow-path tests.
4407 PreserveJVMState pjvms(this);
4408 Node* obj_size = NULL;
4409 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4410 // is reexecuted if deoptimization occurs and there could be problems when merging
4411 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4412 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4413
4414 copy_to_clone(obj, alloc_obj, obj_size, false);
4415
4416 // Present the results of the slow call.
4417 result_reg->init_req(_instance_path, control());
4418 result_val->init_req(_instance_path, alloc_obj);
4419 result_i_o ->set_req(_instance_path, i_o());
4420 result_mem ->set_req(_instance_path, reset_memory());
4421 }
4422
4423 // Generate code for the slow case. We make a call to clone().
4424 set_control(_gvn.transform(slow_region));
4425 if (!stopped()) {
4426 PreserveJVMState pjvms(this);
4427 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4428 // We need to deoptimize on exception (see comment above)
4429 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4430 // this->control() comes from set_results_for_java_call
4431 result_reg->init_req(_slow_path, control());
4432 result_val->init_req(_slow_path, slow_result);
4433 result_i_o ->set_req(_slow_path, i_o());
4434 result_mem ->set_req(_slow_path, reset_memory());
4435 }
4436
4437 // Return the combined state.
4438 set_control( _gvn.transform(result_reg));
4439 set_i_o( _gvn.transform(result_i_o));
4440 set_all_memory( _gvn.transform(result_mem));
4441 } // original reexecute is set back here
4442
4443 set_result(_gvn.transform(result_val));
4444 return true;
4445 }
4446
4447 // If we have a tightly coupled allocation, the arraycopy may take care
4448 // of the array initialization. If one of the guards we insert between
4449 // the allocation and the arraycopy causes a deoptimization, an
4450 // unitialized array will escape the compiled method. To prevent that
4451 // we set the JVM state for uncommon traps between the allocation and
4452 // the arraycopy to the state before the allocation so, in case of
4453 // deoptimization, we'll reexecute the allocation and the
4454 // initialization.
4455 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4456 if (alloc != NULL) {
4457 ciMethod* trap_method = alloc->jvms()->method();
4458 int trap_bci = alloc->jvms()->bci();
4459
4460 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4461 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4462 // Make sure there's no store between the allocation and the
4463 // arraycopy otherwise visible side effects could be rexecuted
4464 // in case of deoptimization and cause incorrect execution.
4465 bool no_interfering_store = true;
4466 Node* mem = alloc->in(TypeFunc::Memory);
4467 if (mem->is_MergeMem()) {
4468 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4469 Node* n = mms.memory();
4470 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4471 assert(n->is_Store(), "what else?");
4472 no_interfering_store = false;
4473 break;
4474 }
4475 }
4476 } else {
4477 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4478 Node* n = mms.memory();
4479 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4480 assert(n->is_Store(), "what else?");
4481 no_interfering_store = false;
4482 break;
4483 }
4484 }
4485 }
4486
4487 if (no_interfering_store) {
4488 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4489 uint size = alloc->req();
4490 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4491 old_jvms->set_map(sfpt);
4492 for (uint i = 0; i < size; i++) {
4493 sfpt->init_req(i, alloc->in(i));
4494 }
4495 // re-push array length for deoptimization
4496 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4497 old_jvms->set_sp(old_jvms->sp()+1);
4498 old_jvms->set_monoff(old_jvms->monoff()+1);
4499 old_jvms->set_scloff(old_jvms->scloff()+1);
4500 old_jvms->set_endoff(old_jvms->endoff()+1);
4501 old_jvms->set_should_reexecute(true);
4502
4503 sfpt->set_i_o(map()->i_o());
4504 sfpt->set_memory(map()->memory());
4505 sfpt->set_control(map()->control());
4506
4507 JVMState* saved_jvms = jvms();
4508 saved_reexecute_sp = _reexecute_sp;
4509
4510 set_jvms(sfpt->jvms());
4511 _reexecute_sp = jvms()->sp();
4512
4513 return saved_jvms;
4514 }
4515 }
4516 }
4517 return NULL;
4518 }
4519
4520 // In case of a deoptimization, we restart execution at the
4521 // allocation, allocating a new array. We would leave an uninitialized
4522 // array in the heap that GCs wouldn't expect. Move the allocation
4523 // after the traps so we don't allocate the array if we
4524 // deoptimize. This is possible because tightly_coupled_allocation()
4525 // guarantees there's no observer of the allocated array at this point
4526 // and the control flow is simple enough.
4527 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4528 int saved_reexecute_sp, uint new_idx) {
4529 if (saved_jvms != NULL && !stopped()) {
4530 assert(alloc != NULL, "only with a tightly coupled allocation");
4531 // restore JVM state to the state at the arraycopy
4532 saved_jvms->map()->set_control(map()->control());
4533 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4534 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4535 // If we've improved the types of some nodes (null check) while
4536 // emitting the guards, propagate them to the current state
4537 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4538 set_jvms(saved_jvms);
4539 _reexecute_sp = saved_reexecute_sp;
4540
4541 // Remove the allocation from above the guards
4542 CallProjections callprojs;
4543 alloc->extract_projections(&callprojs, true);
4544 InitializeNode* init = alloc->initialization();
4545 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4546 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4547 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4548 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4549
4550 // move the allocation here (after the guards)
4551 _gvn.hash_delete(alloc);
4552 alloc->set_req(TypeFunc::Control, control());
4553 alloc->set_req(TypeFunc::I_O, i_o());
4554 Node *mem = reset_memory();
4555 set_all_memory(mem);
4556 alloc->set_req(TypeFunc::Memory, mem);
4557 set_control(init->proj_out_or_null(TypeFunc::Control));
4558 set_i_o(callprojs.fallthrough_ioproj);
4559
4560 // Update memory as done in GraphKit::set_output_for_allocation()
4561 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4562 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4563 if (ary_type->isa_aryptr() && length_type != NULL) {
4564 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4565 }
4566 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4567 int elemidx = C->get_alias_index(telemref);
4568 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
4569 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
4570
4571 Node* allocx = _gvn.transform(alloc);
4572 assert(allocx == alloc, "where has the allocation gone?");
4573 assert(dest->is_CheckCastPP(), "not an allocation result?");
4574
4575 _gvn.hash_delete(dest);
4576 dest->set_req(0, control());
4577 Node* destx = _gvn.transform(dest);
4578 assert(destx == dest, "where has the allocation result gone?");
4579 }
4580 }
4581
4582
4583 //------------------------------inline_arraycopy-----------------------
4584 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4585 // Object dest, int destPos,
4586 // int length);
4587 bool LibraryCallKit::inline_arraycopy() {
4588 // Get the arguments.
4589 Node* src = argument(0); // type: oop
4590 Node* src_offset = argument(1); // type: int
4591 Node* dest = argument(2); // type: oop
4592 Node* dest_offset = argument(3); // type: int
4593 Node* length = argument(4); // type: int
4594
4595 uint new_idx = C->unique();
4596
4597 // Check for allocation before we add nodes that would confuse
4598 // tightly_coupled_allocation()
4599 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4600
4601 int saved_reexecute_sp = -1;
4602 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4603 // See arraycopy_restore_alloc_state() comment
4604 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4605 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4606 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
4607 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4608
4609 // The following tests must be performed
4610 // (1) src and dest are arrays.
4611 // (2) src and dest arrays must have elements of the same BasicType
4612 // (3) src and dest must not be null.
4613 // (4) src_offset must not be negative.
4614 // (5) dest_offset must not be negative.
4615 // (6) length must not be negative.
4616 // (7) src_offset + length must not exceed length of src.
4617 // (8) dest_offset + length must not exceed length of dest.
4618 // (9) each element of an oop array must be assignable
4619
4620 // (3) src and dest must not be null.
4621 // always do this here because we need the JVM state for uncommon traps
4622 Node* null_ctl = top();
4623 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4624 assert(null_ctl->is_top(), "no null control here");
4625 dest = null_check(dest, T_ARRAY);
4626
4627 if (!can_emit_guards) {
4628 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4629 // guards but the arraycopy node could still take advantage of a
4630 // tightly allocated allocation. tightly_coupled_allocation() is
4631 // called again to make sure it takes the null check above into
4632 // account: the null check is mandatory and if it caused an
4633 // uncommon trap to be emitted then the allocation can't be
4634 // considered tightly coupled in this context.
4635 alloc = tightly_coupled_allocation(dest, NULL);
4636 }
4637
4638 bool validated = false;
4639
4640 const Type* src_type = _gvn.type(src);
4641 const Type* dest_type = _gvn.type(dest);
4642 const TypeAryPtr* top_src = src_type->isa_aryptr();
4643 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4644
4645 // Do we have the type of src?
4646 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4647 // Do we have the type of dest?
4648 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4649 // Is the type for src from speculation?
4650 bool src_spec = false;
4651 // Is the type for dest from speculation?
4652 bool dest_spec = false;
4653
4654 if ((!has_src || !has_dest) && can_emit_guards) {
4655 // We don't have sufficient type information, let's see if
4656 // speculative types can help. We need to have types for both src
4657 // and dest so that it pays off.
4658
4659 // Do we already have or could we have type information for src
4660 bool could_have_src = has_src;
4661 // Do we already have or could we have type information for dest
4662 bool could_have_dest = has_dest;
4663
4664 ciKlass* src_k = NULL;
4665 if (!has_src) {
4666 src_k = src_type->speculative_type_not_null();
4667 if (src_k != NULL && src_k->is_array_klass()) {
4668 could_have_src = true;
4669 }
4670 }
4671
4672 ciKlass* dest_k = NULL;
4673 if (!has_dest) {
4674 dest_k = dest_type->speculative_type_not_null();
4675 if (dest_k != NULL && dest_k->is_array_klass()) {
4676 could_have_dest = true;
4677 }
4678 }
4679
4680 if (could_have_src && could_have_dest) {
4681 // This is going to pay off so emit the required guards
4682 if (!has_src) {
4683 src = maybe_cast_profiled_obj(src, src_k, true);
4684 src_type = _gvn.type(src);
4685 top_src = src_type->isa_aryptr();
4686 has_src = (top_src != NULL && top_src->klass() != NULL);
4687 src_spec = true;
4688 }
4689 if (!has_dest) {
4690 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4691 dest_type = _gvn.type(dest);
4692 top_dest = dest_type->isa_aryptr();
4693 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4694 dest_spec = true;
4695 }
4696 }
4697 }
4698
4699 if (has_src && has_dest && can_emit_guards) {
4700 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
4701 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
4702 if (is_reference_type(src_elem)) src_elem = T_OBJECT;
4703 if (is_reference_type(dest_elem)) dest_elem = T_OBJECT;
4704
4705 if (src_elem == dest_elem && src_elem == T_OBJECT) {
4706 // If both arrays are object arrays then having the exact types
4707 // for both will remove the need for a subtype check at runtime
4708 // before the call and may make it possible to pick a faster copy
4709 // routine (without a subtype check on every element)
4710 // Do we have the exact type of src?
4711 bool could_have_src = src_spec;
4712 // Do we have the exact type of dest?
4713 bool could_have_dest = dest_spec;
4714 ciKlass* src_k = top_src->klass();
4715 ciKlass* dest_k = top_dest->klass();
4716 if (!src_spec) {
4717 src_k = src_type->speculative_type_not_null();
4718 if (src_k != NULL && src_k->is_array_klass()) {
4719 could_have_src = true;
4720 }
4721 }
4722 if (!dest_spec) {
4723 dest_k = dest_type->speculative_type_not_null();
4724 if (dest_k != NULL && dest_k->is_array_klass()) {
4725 could_have_dest = true;
4726 }
4727 }
4728 if (could_have_src && could_have_dest) {
4729 // If we can have both exact types, emit the missing guards
4730 if (could_have_src && !src_spec) {
4731 src = maybe_cast_profiled_obj(src, src_k, true);
4732 }
4733 if (could_have_dest && !dest_spec) {
4734 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4735 }
4736 }
4737 }
4738 }
4739
4740 ciMethod* trap_method = method();
4741 int trap_bci = bci();
4742 if (saved_jvms != NULL) {
4743 trap_method = alloc->jvms()->method();
4744 trap_bci = alloc->jvms()->bci();
4745 }
4746
4747 bool negative_length_guard_generated = false;
4748
4749 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4750 can_emit_guards &&
4751 !src->is_top() && !dest->is_top()) {
4752 // validate arguments: enables transformation the ArrayCopyNode
4753 validated = true;
4754
4755 RegionNode* slow_region = new RegionNode(1);
4756 record_for_igvn(slow_region);
4757
4758 // (1) src and dest are arrays.
4759 generate_non_array_guard(load_object_klass(src), slow_region);
4760 generate_non_array_guard(load_object_klass(dest), slow_region);
4761
4762 // (2) src and dest arrays must have elements of the same BasicType
4763 // done at macro expansion or at Ideal transformation time
4764
4765 // (4) src_offset must not be negative.
4766 generate_negative_guard(src_offset, slow_region);
4767
4768 // (5) dest_offset must not be negative.
4769 generate_negative_guard(dest_offset, slow_region);
4770
4771 // (7) src_offset + length must not exceed length of src.
4772 generate_limit_guard(src_offset, length,
4773 load_array_length(src),
4774 slow_region);
4775
4776 // (8) dest_offset + length must not exceed length of dest.
4777 generate_limit_guard(dest_offset, length,
4778 load_array_length(dest),
4779 slow_region);
4780
4781 // (6) length must not be negative.
4782 // This is also checked in generate_arraycopy() during macro expansion, but
4783 // we also have to check it here for the case where the ArrayCopyNode will
4784 // be eliminated by Escape Analysis.
4785 if (EliminateAllocations) {
4786 generate_negative_guard(length, slow_region);
4787 negative_length_guard_generated = true;
4788 }
4789
4790 // (9) each element of an oop array must be assignable
4791 Node* src_klass = load_object_klass(src);
4792 Node* dest_klass = load_object_klass(dest);
4793 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
4794
4795 if (not_subtype_ctrl != top()) {
4796 PreserveJVMState pjvms(this);
4797 set_control(not_subtype_ctrl);
4798 uncommon_trap(Deoptimization::Reason_intrinsic,
4799 Deoptimization::Action_make_not_entrant);
4800 assert(stopped(), "Should be stopped");
4801 }
4802 {
4803 PreserveJVMState pjvms(this);
4804 set_control(_gvn.transform(slow_region));
4805 uncommon_trap(Deoptimization::Reason_intrinsic,
4806 Deoptimization::Action_make_not_entrant);
4807 assert(stopped(), "Should be stopped");
4808 }
4809
4810 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
4811 const Type *toop = TypeOopPtr::make_from_klass(dest_klass_t->klass());
4812 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
4813 }
4814
4815 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
4816
4817 if (stopped()) {
4818 return true;
4819 }
4820
4821 Node* new_src = access_resolve(src, ACCESS_READ);
4822 Node* new_dest = access_resolve(dest, ACCESS_WRITE);
4823
4824 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, new_src, src_offset, new_dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
4825 // Create LoadRange and LoadKlass nodes for use during macro expansion here
4826 // so the compiler has a chance to eliminate them: during macro expansion,
4827 // we have to set their control (CastPP nodes are eliminated).
4828 load_object_klass(src), load_object_klass(dest),
4829 load_array_length(src), load_array_length(dest));
4830
4831 ac->set_arraycopy(validated);
4832
4833 Node* n = _gvn.transform(ac);
4834 if (n == ac) {
4835 ac->connect_outputs(this);
4836 } else {
4837 assert(validated, "shouldn't transform if all arguments not validated");
4838 set_all_memory(n);
4839 }
4840 clear_upper_avx();
4841
4842
4843 return true;
4844 }
4845
4846
4847 // Helper function which determines if an arraycopy immediately follows
4848 // an allocation, with no intervening tests or other escapes for the object.
4849 AllocateArrayNode*
4850 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
4851 RegionNode* slow_region) {
4852 if (stopped()) return NULL; // no fast path
4853 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
4854
4855 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
4856 if (alloc == NULL) return NULL;
4857
4858 Node* rawmem = memory(Compile::AliasIdxRaw);
4859 // Is the allocation's memory state untouched?
4860 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
4861 // Bail out if there have been raw-memory effects since the allocation.
4862 // (Example: There might have been a call or safepoint.)
4863 return NULL;
4864 }
4865 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
4866 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
4867 return NULL;
4868 }
4869
4870 // There must be no unexpected observers of this allocation.
4871 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
4872 Node* obs = ptr->fast_out(i);
4873 if (obs != this->map()) {
4874 return NULL;
4875 }
4876 }
4877
4878 // This arraycopy must unconditionally follow the allocation of the ptr.
4879 Node* alloc_ctl = ptr->in(0);
4880 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
4881
4882 Node* ctl = control();
4883 while (ctl != alloc_ctl) {
4884 // There may be guards which feed into the slow_region.
4885 // Any other control flow means that we might not get a chance
4886 // to finish initializing the allocated object.
4887 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
4888 IfNode* iff = ctl->in(0)->as_If();
4889 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
4890 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
4891 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
4892 ctl = iff->in(0); // This test feeds the known slow_region.
4893 continue;
4894 }
4895 // One more try: Various low-level checks bottom out in
4896 // uncommon traps. If the debug-info of the trap omits
4897 // any reference to the allocation, as we've already
4898 // observed, then there can be no objection to the trap.
4899 bool found_trap = false;
4900 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
4901 Node* obs = not_ctl->fast_out(j);
4902 if (obs->in(0) == not_ctl && obs->is_Call() &&
4903 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
4904 found_trap = true; break;
4905 }
4906 }
4907 if (found_trap) {
4908 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
4909 continue;
4910 }
4911 }
4912 return NULL;
4913 }
4914
4915 // If we get this far, we have an allocation which immediately
4916 // precedes the arraycopy, and we can take over zeroing the new object.
4917 // The arraycopy will finish the initialization, and provide
4918 // a new control state to which we will anchor the destination pointer.
4919
4920 return alloc;
4921 }
4922
4923 //-------------inline_encodeISOArray-----------------------------------
4924 // encode char[] to byte[] in ISO_8859_1
4925 bool LibraryCallKit::inline_encodeISOArray() {
4926 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
4927 // no receiver since it is static method
4928 Node *src = argument(0);
4929 Node *src_offset = argument(1);
4930 Node *dst = argument(2);
4931 Node *dst_offset = argument(3);
4932 Node *length = argument(4);
4933
4934 src = must_be_not_null(src, true);
4935 dst = must_be_not_null(dst, true);
4936
4937 src = access_resolve(src, ACCESS_READ);
4938 dst = access_resolve(dst, ACCESS_WRITE);
4939
4940 const Type* src_type = src->Value(&_gvn);
4941 const Type* dst_type = dst->Value(&_gvn);
4942 const TypeAryPtr* top_src = src_type->isa_aryptr();
4943 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
4944 if (top_src == NULL || top_src->klass() == NULL ||
4945 top_dest == NULL || top_dest->klass() == NULL) {
4946 // failed array check
4947 return false;
4948 }
4949
4950 // Figure out the size and type of the elements we will be copying.
4951 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4952 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4953 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
4954 return false;
4955 }
4956
4957 Node* src_start = array_element_address(src, src_offset, T_CHAR);
4958 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
4959 // 'src_start' points to src array + scaled offset
4960 // 'dst_start' points to dst array + scaled offset
4961
4962 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
4963 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
4964 enc = _gvn.transform(enc);
4965 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
4966 set_memory(res_mem, mtype);
4967 set_result(enc);
4968 clear_upper_avx();
4969
4970 return true;
4971 }
4972
4973 //-------------inline_multiplyToLen-----------------------------------
4974 bool LibraryCallKit::inline_multiplyToLen() {
4975 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
4976
4977 address stubAddr = StubRoutines::multiplyToLen();
4978 if (stubAddr == NULL) {
4979 return false; // Intrinsic's stub is not implemented on this platform
4980 }
4981 const char* stubName = "multiplyToLen";
4982
4983 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
4984
4985 // no receiver because it is a static method
4986 Node* x = argument(0);
4987 Node* xlen = argument(1);
4988 Node* y = argument(2);
4989 Node* ylen = argument(3);
4990 Node* z = argument(4);
4991
4992 x = must_be_not_null(x, true);
4993 y = must_be_not_null(y, true);
4994
4995 x = access_resolve(x, ACCESS_READ);
4996 y = access_resolve(y, ACCESS_READ);
4997 z = access_resolve(z, ACCESS_WRITE);
4998
4999 const Type* x_type = x->Value(&_gvn);
5000 const Type* y_type = y->Value(&_gvn);
5001 const TypeAryPtr* top_x = x_type->isa_aryptr();
5002 const TypeAryPtr* top_y = y_type->isa_aryptr();
5003 if (top_x == NULL || top_x->klass() == NULL ||
5004 top_y == NULL || top_y->klass() == NULL) {
5005 // failed array check
5006 return false;
5007 }
5008
5009 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5010 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5011 if (x_elem != T_INT || y_elem != T_INT) {
5012 return false;
5013 }
5014
5015 // Set the original stack and the reexecute bit for the interpreter to reexecute
5016 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5017 // on the return from z array allocation in runtime.
5018 { PreserveReexecuteState preexecs(this);
5019 jvms()->set_should_reexecute(true);
5020
5021 Node* x_start = array_element_address(x, intcon(0), x_elem);
5022 Node* y_start = array_element_address(y, intcon(0), y_elem);
5023 // 'x_start' points to x array + scaled xlen
5024 // 'y_start' points to y array + scaled ylen
5025
5026 // Allocate the result array
5027 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5028 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5029 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5030
5031 IdealKit ideal(this);
5032
5033 #define __ ideal.
5034 Node* one = __ ConI(1);
5035 Node* zero = __ ConI(0);
5036 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5037 __ set(need_alloc, zero);
5038 __ set(z_alloc, z);
5039 __ if_then(z, BoolTest::eq, null()); {
5040 __ increment (need_alloc, one);
5041 } __ else_(); {
5042 // Update graphKit memory and control from IdealKit.
5043 sync_kit(ideal);
5044 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5045 cast->init_req(0, control());
5046 _gvn.set_type(cast, cast->bottom_type());
5047 C->record_for_igvn(cast);
5048
5049 Node* zlen_arg = load_array_length(cast);
5050 // Update IdealKit memory and control from graphKit.
5051 __ sync_kit(this);
5052 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5053 __ increment (need_alloc, one);
5054 } __ end_if();
5055 } __ end_if();
5056
5057 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5058 // Update graphKit memory and control from IdealKit.
5059 sync_kit(ideal);
5060 Node * narr = new_array(klass_node, zlen, 1);
5061 // Update IdealKit memory and control from graphKit.
5062 __ sync_kit(this);
5063 __ set(z_alloc, narr);
5064 } __ end_if();
5065
5066 sync_kit(ideal);
5067 z = __ value(z_alloc);
5068 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5069 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5070 // Final sync IdealKit and GraphKit.
5071 final_sync(ideal);
5072 #undef __
5073
5074 Node* z_start = array_element_address(z, intcon(0), T_INT);
5075
5076 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5077 OptoRuntime::multiplyToLen_Type(),
5078 stubAddr, stubName, TypePtr::BOTTOM,
5079 x_start, xlen, y_start, ylen, z_start, zlen);
5080 } // original reexecute is set back here
5081
5082 C->set_has_split_ifs(true); // Has chance for split-if optimization
5083 set_result(z);
5084 return true;
5085 }
5086
5087 //-------------inline_squareToLen------------------------------------
5088 bool LibraryCallKit::inline_squareToLen() {
5089 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5090
5091 address stubAddr = StubRoutines::squareToLen();
5092 if (stubAddr == NULL) {
5093 return false; // Intrinsic's stub is not implemented on this platform
5094 }
5095 const char* stubName = "squareToLen";
5096
5097 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5098
5099 Node* x = argument(0);
5100 Node* len = argument(1);
5101 Node* z = argument(2);
5102 Node* zlen = argument(3);
5103
5104 x = must_be_not_null(x, true);
5105 z = must_be_not_null(z, true);
5106
5107 x = access_resolve(x, ACCESS_READ);
5108 z = access_resolve(z, ACCESS_WRITE);
5109
5110 const Type* x_type = x->Value(&_gvn);
5111 const Type* z_type = z->Value(&_gvn);
5112 const TypeAryPtr* top_x = x_type->isa_aryptr();
5113 const TypeAryPtr* top_z = z_type->isa_aryptr();
5114 if (top_x == NULL || top_x->klass() == NULL ||
5115 top_z == NULL || top_z->klass() == NULL) {
5116 // failed array check
5117 return false;
5118 }
5119
5120 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5121 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5122 if (x_elem != T_INT || z_elem != T_INT) {
5123 return false;
5124 }
5125
5126
5127 Node* x_start = array_element_address(x, intcon(0), x_elem);
5128 Node* z_start = array_element_address(z, intcon(0), z_elem);
5129
5130 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5131 OptoRuntime::squareToLen_Type(),
5132 stubAddr, stubName, TypePtr::BOTTOM,
5133 x_start, len, z_start, zlen);
5134
5135 set_result(z);
5136 return true;
5137 }
5138
5139 //-------------inline_mulAdd------------------------------------------
5140 bool LibraryCallKit::inline_mulAdd() {
5141 assert(UseMulAddIntrinsic, "not implemented on this platform");
5142
5143 address stubAddr = StubRoutines::mulAdd();
5144 if (stubAddr == NULL) {
5145 return false; // Intrinsic's stub is not implemented on this platform
5146 }
5147 const char* stubName = "mulAdd";
5148
5149 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5150
5151 Node* out = argument(0);
5152 Node* in = argument(1);
5153 Node* offset = argument(2);
5154 Node* len = argument(3);
5155 Node* k = argument(4);
5156
5157 out = must_be_not_null(out, true);
5158
5159 in = access_resolve(in, ACCESS_READ);
5160 out = access_resolve(out, ACCESS_WRITE);
5161
5162 const Type* out_type = out->Value(&_gvn);
5163 const Type* in_type = in->Value(&_gvn);
5164 const TypeAryPtr* top_out = out_type->isa_aryptr();
5165 const TypeAryPtr* top_in = in_type->isa_aryptr();
5166 if (top_out == NULL || top_out->klass() == NULL ||
5167 top_in == NULL || top_in->klass() == NULL) {
5168 // failed array check
5169 return false;
5170 }
5171
5172 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5173 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5174 if (out_elem != T_INT || in_elem != T_INT) {
5175 return false;
5176 }
5177
5178 Node* outlen = load_array_length(out);
5179 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5180 Node* out_start = array_element_address(out, intcon(0), out_elem);
5181 Node* in_start = array_element_address(in, intcon(0), in_elem);
5182
5183 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5184 OptoRuntime::mulAdd_Type(),
5185 stubAddr, stubName, TypePtr::BOTTOM,
5186 out_start,in_start, new_offset, len, k);
5187 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5188 set_result(result);
5189 return true;
5190 }
5191
5192 //-------------inline_montgomeryMultiply-----------------------------------
5193 bool LibraryCallKit::inline_montgomeryMultiply() {
5194 address stubAddr = StubRoutines::montgomeryMultiply();
5195 if (stubAddr == NULL) {
5196 return false; // Intrinsic's stub is not implemented on this platform
5197 }
5198
5199 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5200 const char* stubName = "montgomery_multiply";
5201
5202 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5203
5204 Node* a = argument(0);
5205 Node* b = argument(1);
5206 Node* n = argument(2);
5207 Node* len = argument(3);
5208 Node* inv = argument(4);
5209 Node* m = argument(6);
5210
5211 a = access_resolve(a, ACCESS_READ);
5212 b = access_resolve(b, ACCESS_READ);
5213 n = access_resolve(n, ACCESS_READ);
5214 m = access_resolve(m, ACCESS_WRITE);
5215
5216 const Type* a_type = a->Value(&_gvn);
5217 const TypeAryPtr* top_a = a_type->isa_aryptr();
5218 const Type* b_type = b->Value(&_gvn);
5219 const TypeAryPtr* top_b = b_type->isa_aryptr();
5220 const Type* n_type = a->Value(&_gvn);
5221 const TypeAryPtr* top_n = n_type->isa_aryptr();
5222 const Type* m_type = a->Value(&_gvn);
5223 const TypeAryPtr* top_m = m_type->isa_aryptr();
5224 if (top_a == NULL || top_a->klass() == NULL ||
5225 top_b == NULL || top_b->klass() == NULL ||
5226 top_n == NULL || top_n->klass() == NULL ||
5227 top_m == NULL || top_m->klass() == NULL) {
5228 // failed array check
5229 return false;
5230 }
5231
5232 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5233 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5234 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5235 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5236 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5237 return false;
5238 }
5239
5240 // Make the call
5241 {
5242 Node* a_start = array_element_address(a, intcon(0), a_elem);
5243 Node* b_start = array_element_address(b, intcon(0), b_elem);
5244 Node* n_start = array_element_address(n, intcon(0), n_elem);
5245 Node* m_start = array_element_address(m, intcon(0), m_elem);
5246
5247 Node* call = make_runtime_call(RC_LEAF,
5248 OptoRuntime::montgomeryMultiply_Type(),
5249 stubAddr, stubName, TypePtr::BOTTOM,
5250 a_start, b_start, n_start, len, inv, top(),
5251 m_start);
5252 set_result(m);
5253 }
5254
5255 return true;
5256 }
5257
5258 bool LibraryCallKit::inline_montgomerySquare() {
5259 address stubAddr = StubRoutines::montgomerySquare();
5260 if (stubAddr == NULL) {
5261 return false; // Intrinsic's stub is not implemented on this platform
5262 }
5263
5264 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5265 const char* stubName = "montgomery_square";
5266
5267 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5268
5269 Node* a = argument(0);
5270 Node* n = argument(1);
5271 Node* len = argument(2);
5272 Node* inv = argument(3);
5273 Node* m = argument(5);
5274
5275 a = access_resolve(a, ACCESS_READ);
5276 n = access_resolve(n, ACCESS_READ);
5277 m = access_resolve(m, ACCESS_WRITE);
5278
5279 const Type* a_type = a->Value(&_gvn);
5280 const TypeAryPtr* top_a = a_type->isa_aryptr();
5281 const Type* n_type = a->Value(&_gvn);
5282 const TypeAryPtr* top_n = n_type->isa_aryptr();
5283 const Type* m_type = a->Value(&_gvn);
5284 const TypeAryPtr* top_m = m_type->isa_aryptr();
5285 if (top_a == NULL || top_a->klass() == NULL ||
5286 top_n == NULL || top_n->klass() == NULL ||
5287 top_m == NULL || top_m->klass() == NULL) {
5288 // failed array check
5289 return false;
5290 }
5291
5292 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5293 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5294 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5295 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5296 return false;
5297 }
5298
5299 // Make the call
5300 {
5301 Node* a_start = array_element_address(a, intcon(0), a_elem);
5302 Node* n_start = array_element_address(n, intcon(0), n_elem);
5303 Node* m_start = array_element_address(m, intcon(0), m_elem);
5304
5305 Node* call = make_runtime_call(RC_LEAF,
5306 OptoRuntime::montgomerySquare_Type(),
5307 stubAddr, stubName, TypePtr::BOTTOM,
5308 a_start, n_start, len, inv, top(),
5309 m_start);
5310 set_result(m);
5311 }
5312
5313 return true;
5314 }
5315
5316 //-------------inline_vectorizedMismatch------------------------------
5317 bool LibraryCallKit::inline_vectorizedMismatch() {
5318 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5319
5320 address stubAddr = StubRoutines::vectorizedMismatch();
5321 if (stubAddr == NULL) {
5322 return false; // Intrinsic's stub is not implemented on this platform
5323 }
5324 const char* stubName = "vectorizedMismatch";
5325 int size_l = callee()->signature()->size();
5326 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5327
5328 Node* obja = argument(0);
5329 Node* aoffset = argument(1);
5330 Node* objb = argument(3);
5331 Node* boffset = argument(4);
5332 Node* length = argument(6);
5333 Node* scale = argument(7);
5334
5335 const Type* a_type = obja->Value(&_gvn);
5336 const Type* b_type = objb->Value(&_gvn);
5337 const TypeAryPtr* top_a = a_type->isa_aryptr();
5338 const TypeAryPtr* top_b = b_type->isa_aryptr();
5339 if (top_a == NULL || top_a->klass() == NULL ||
5340 top_b == NULL || top_b->klass() == NULL) {
5341 // failed array check
5342 return false;
5343 }
5344
5345 Node* call;
5346 jvms()->set_should_reexecute(true);
5347
5348 obja = access_resolve(obja, ACCESS_READ);
5349 objb = access_resolve(objb, ACCESS_READ);
5350 Node* obja_adr = make_unsafe_address(obja, aoffset, ACCESS_READ);
5351 Node* objb_adr = make_unsafe_address(objb, boffset, ACCESS_READ);
5352
5353 call = make_runtime_call(RC_LEAF,
5354 OptoRuntime::vectorizedMismatch_Type(),
5355 stubAddr, stubName, TypePtr::BOTTOM,
5356 obja_adr, objb_adr, length, scale);
5357
5358 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5359 set_result(result);
5360 return true;
5361 }
5362
5363 /**
5364 * Calculate CRC32 for byte.
5365 * int java.util.zip.CRC32.update(int crc, int b)
5366 */
5367 bool LibraryCallKit::inline_updateCRC32() {
5368 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5369 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5370 // no receiver since it is static method
5371 Node* crc = argument(0); // type: int
5372 Node* b = argument(1); // type: int
5373
5374 /*
5375 * int c = ~ crc;
5376 * b = timesXtoThe32[(b ^ c) & 0xFF];
5377 * b = b ^ (c >>> 8);
5378 * crc = ~b;
5379 */
5380
5381 Node* M1 = intcon(-1);
5382 crc = _gvn.transform(new XorINode(crc, M1));
5383 Node* result = _gvn.transform(new XorINode(crc, b));
5384 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5385
5386 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5387 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5388 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5389 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5390
5391 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5392 result = _gvn.transform(new XorINode(crc, result));
5393 result = _gvn.transform(new XorINode(result, M1));
5394 set_result(result);
5395 return true;
5396 }
5397
5398 /**
5399 * Calculate CRC32 for byte[] array.
5400 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5401 */
5402 bool LibraryCallKit::inline_updateBytesCRC32() {
5403 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5404 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5405 // no receiver since it is static method
5406 Node* crc = argument(0); // type: int
5407 Node* src = argument(1); // type: oop
5408 Node* offset = argument(2); // type: int
5409 Node* length = argument(3); // type: int
5410
5411 const Type* src_type = src->Value(&_gvn);
5412 const TypeAryPtr* top_src = src_type->isa_aryptr();
5413 if (top_src == NULL || top_src->klass() == NULL) {
5414 // failed array check
5415 return false;
5416 }
5417
5418 // Figure out the size and type of the elements we will be copying.
5419 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5420 if (src_elem != T_BYTE) {
5421 return false;
5422 }
5423
5424 // 'src_start' points to src array + scaled offset
5425 src = must_be_not_null(src, true);
5426 src = access_resolve(src, ACCESS_READ);
5427 Node* src_start = array_element_address(src, offset, src_elem);
5428
5429 // We assume that range check is done by caller.
5430 // TODO: generate range check (offset+length < src.length) in debug VM.
5431
5432 // Call the stub.
5433 address stubAddr = StubRoutines::updateBytesCRC32();
5434 const char *stubName = "updateBytesCRC32";
5435
5436 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5437 stubAddr, stubName, TypePtr::BOTTOM,
5438 crc, src_start, length);
5439 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5440 set_result(result);
5441 return true;
5442 }
5443
5444 /**
5445 * Calculate CRC32 for ByteBuffer.
5446 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5447 */
5448 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5449 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5450 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5451 // no receiver since it is static method
5452 Node* crc = argument(0); // type: int
5453 Node* src = argument(1); // type: long
5454 Node* offset = argument(3); // type: int
5455 Node* length = argument(4); // type: int
5456
5457 src = ConvL2X(src); // adjust Java long to machine word
5458 Node* base = _gvn.transform(new CastX2PNode(src));
5459 offset = ConvI2X(offset);
5460
5461 // 'src_start' points to src array + scaled offset
5462 Node* src_start = basic_plus_adr(top(), base, offset);
5463
5464 // Call the stub.
5465 address stubAddr = StubRoutines::updateBytesCRC32();
5466 const char *stubName = "updateBytesCRC32";
5467
5468 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5469 stubAddr, stubName, TypePtr::BOTTOM,
5470 crc, src_start, length);
5471 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5472 set_result(result);
5473 return true;
5474 }
5475
5476 //------------------------------get_table_from_crc32c_class-----------------------
5477 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5478 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5479 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5480
5481 return table;
5482 }
5483
5484 //------------------------------inline_updateBytesCRC32C-----------------------
5485 //
5486 // Calculate CRC32C for byte[] array.
5487 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5488 //
5489 bool LibraryCallKit::inline_updateBytesCRC32C() {
5490 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5491 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5492 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5493 // no receiver since it is a static method
5494 Node* crc = argument(0); // type: int
5495 Node* src = argument(1); // type: oop
5496 Node* offset = argument(2); // type: int
5497 Node* end = argument(3); // type: int
5498
5499 Node* length = _gvn.transform(new SubINode(end, offset));
5500
5501 const Type* src_type = src->Value(&_gvn);
5502 const TypeAryPtr* top_src = src_type->isa_aryptr();
5503 if (top_src == NULL || top_src->klass() == NULL) {
5504 // failed array check
5505 return false;
5506 }
5507
5508 // Figure out the size and type of the elements we will be copying.
5509 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5510 if (src_elem != T_BYTE) {
5511 return false;
5512 }
5513
5514 // 'src_start' points to src array + scaled offset
5515 src = must_be_not_null(src, true);
5516 src = access_resolve(src, ACCESS_READ);
5517 Node* src_start = array_element_address(src, offset, src_elem);
5518
5519 // static final int[] byteTable in class CRC32C
5520 Node* table = get_table_from_crc32c_class(callee()->holder());
5521 table = must_be_not_null(table, true);
5522 table = access_resolve(table, ACCESS_READ);
5523 Node* table_start = array_element_address(table, intcon(0), T_INT);
5524
5525 // We assume that range check is done by caller.
5526 // TODO: generate range check (offset+length < src.length) in debug VM.
5527
5528 // Call the stub.
5529 address stubAddr = StubRoutines::updateBytesCRC32C();
5530 const char *stubName = "updateBytesCRC32C";
5531
5532 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5533 stubAddr, stubName, TypePtr::BOTTOM,
5534 crc, src_start, length, table_start);
5535 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5536 set_result(result);
5537 return true;
5538 }
5539
5540 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5541 //
5542 // Calculate CRC32C for DirectByteBuffer.
5543 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5544 //
5545 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5546 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5547 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5548 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5549 // no receiver since it is a static method
5550 Node* crc = argument(0); // type: int
5551 Node* src = argument(1); // type: long
5552 Node* offset = argument(3); // type: int
5553 Node* end = argument(4); // type: int
5554
5555 Node* length = _gvn.transform(new SubINode(end, offset));
5556
5557 src = ConvL2X(src); // adjust Java long to machine word
5558 Node* base = _gvn.transform(new CastX2PNode(src));
5559 offset = ConvI2X(offset);
5560
5561 // 'src_start' points to src array + scaled offset
5562 Node* src_start = basic_plus_adr(top(), base, offset);
5563
5564 // static final int[] byteTable in class CRC32C
5565 Node* table = get_table_from_crc32c_class(callee()->holder());
5566 table = must_be_not_null(table, true);
5567 table = access_resolve(table, ACCESS_READ);
5568 Node* table_start = array_element_address(table, intcon(0), T_INT);
5569
5570 // Call the stub.
5571 address stubAddr = StubRoutines::updateBytesCRC32C();
5572 const char *stubName = "updateBytesCRC32C";
5573
5574 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5575 stubAddr, stubName, TypePtr::BOTTOM,
5576 crc, src_start, length, table_start);
5577 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5578 set_result(result);
5579 return true;
5580 }
5581
5582 //------------------------------inline_updateBytesAdler32----------------------
5583 //
5584 // Calculate Adler32 checksum for byte[] array.
5585 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5586 //
5587 bool LibraryCallKit::inline_updateBytesAdler32() {
5588 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5589 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5590 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5591 // no receiver since it is static method
5592 Node* crc = argument(0); // type: int
5593 Node* src = argument(1); // type: oop
5594 Node* offset = argument(2); // type: int
5595 Node* length = argument(3); // type: int
5596
5597 const Type* src_type = src->Value(&_gvn);
5598 const TypeAryPtr* top_src = src_type->isa_aryptr();
5599 if (top_src == NULL || top_src->klass() == NULL) {
5600 // failed array check
5601 return false;
5602 }
5603
5604 // Figure out the size and type of the elements we will be copying.
5605 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5606 if (src_elem != T_BYTE) {
5607 return false;
5608 }
5609
5610 // 'src_start' points to src array + scaled offset
5611 src = access_resolve(src, ACCESS_READ);
5612 Node* src_start = array_element_address(src, offset, src_elem);
5613
5614 // We assume that range check is done by caller.
5615 // TODO: generate range check (offset+length < src.length) in debug VM.
5616
5617 // Call the stub.
5618 address stubAddr = StubRoutines::updateBytesAdler32();
5619 const char *stubName = "updateBytesAdler32";
5620
5621 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5622 stubAddr, stubName, TypePtr::BOTTOM,
5623 crc, src_start, length);
5624 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5625 set_result(result);
5626 return true;
5627 }
5628
5629 //------------------------------inline_updateByteBufferAdler32---------------
5630 //
5631 // Calculate Adler32 checksum for DirectByteBuffer.
5632 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5633 //
5634 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5635 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5636 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5637 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5638 // no receiver since it is static method
5639 Node* crc = argument(0); // type: int
5640 Node* src = argument(1); // type: long
5641 Node* offset = argument(3); // type: int
5642 Node* length = argument(4); // type: int
5643
5644 src = ConvL2X(src); // adjust Java long to machine word
5645 Node* base = _gvn.transform(new CastX2PNode(src));
5646 offset = ConvI2X(offset);
5647
5648 // 'src_start' points to src array + scaled offset
5649 Node* src_start = basic_plus_adr(top(), base, offset);
5650
5651 // Call the stub.
5652 address stubAddr = StubRoutines::updateBytesAdler32();
5653 const char *stubName = "updateBytesAdler32";
5654
5655 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5656 stubAddr, stubName, TypePtr::BOTTOM,
5657 crc, src_start, length);
5658
5659 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5660 set_result(result);
5661 return true;
5662 }
5663
5664 //----------------------------inline_reference_get----------------------------
5665 // public T java.lang.ref.Reference.get();
5666 bool LibraryCallKit::inline_reference_get() {
5667 const int referent_offset = java_lang_ref_Reference::referent_offset;
5668 guarantee(referent_offset > 0, "should have already been set");
5669
5670 // Get the argument:
5671 Node* reference_obj = null_check_receiver();
5672 if (stopped()) return true;
5673
5674 const TypeInstPtr* tinst = _gvn.type(reference_obj)->isa_instptr();
5675 assert(tinst != NULL, "obj is null");
5676 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5677 ciInstanceKlass* referenceKlass = tinst->klass()->as_instance_klass();
5678 ciField* field = referenceKlass->get_field_by_name(ciSymbol::make("referent"),
5679 ciSymbol::make("Ljava/lang/Object;"),
5680 false);
5681 assert (field != NULL, "undefined field");
5682
5683 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5684 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5685
5686 ciInstanceKlass* klass = env()->Object_klass();
5687 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5688
5689 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
5690 Node* result = access_load_at(reference_obj, adr, adr_type, object_type, T_OBJECT, decorators);
5691 // Add memory barrier to prevent commoning reads from this field
5692 // across safepoint since GC can change its value.
5693 insert_mem_bar(Op_MemBarCPUOrder);
5694
5695 set_result(result);
5696 return true;
5697 }
5698
5699
5700 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5701 bool is_exact=true, bool is_static=false,
5702 ciInstanceKlass * fromKls=NULL) {
5703 if (fromKls == NULL) {
5704 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5705 assert(tinst != NULL, "obj is null");
5706 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5707 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5708 fromKls = tinst->klass()->as_instance_klass();
5709 } else {
5710 assert(is_static, "only for static field access");
5711 }
5712 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5713 ciSymbol::make(fieldTypeString),
5714 is_static);
5715
5716 assert (field != NULL, "undefined field");
5717 if (field == NULL) return (Node *) NULL;
5718
5719 if (is_static) {
5720 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5721 fromObj = makecon(tip);
5722 }
5723
5724 // Next code copied from Parse::do_get_xxx():
5725
5726 // Compute address and memory type.
5727 int offset = field->offset_in_bytes();
5728 bool is_vol = field->is_volatile();
5729 ciType* field_klass = field->type();
5730 assert(field_klass->is_loaded(), "should be loaded");
5731 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5732 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5733 BasicType bt = field->layout_type();
5734
5735 // Build the resultant type of the load
5736 const Type *type;
5737 if (bt == T_OBJECT) {
5738 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5739 } else {
5740 type = Type::get_const_basic_type(bt);
5741 }
5742
5743 DecoratorSet decorators = IN_HEAP;
5744
5745 if (is_vol) {
5746 decorators |= MO_SEQ_CST;
5747 }
5748
5749 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
5750 }
5751
5752 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5753 bool is_exact = true, bool is_static = false,
5754 ciInstanceKlass * fromKls = NULL) {
5755 if (fromKls == NULL) {
5756 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5757 assert(tinst != NULL, "obj is null");
5758 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5759 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5760 fromKls = tinst->klass()->as_instance_klass();
5761 }
5762 else {
5763 assert(is_static, "only for static field access");
5764 }
5765 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5766 ciSymbol::make(fieldTypeString),
5767 is_static);
5768
5769 assert(field != NULL, "undefined field");
5770 assert(!field->is_volatile(), "not defined for volatile fields");
5771
5772 if (is_static) {
5773 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5774 fromObj = makecon(tip);
5775 }
5776
5777 // Next code copied from Parse::do_get_xxx():
5778
5779 // Compute address and memory type.
5780 int offset = field->offset_in_bytes();
5781 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5782
5783 return adr;
5784 }
5785
5786 //------------------------------inline_aescrypt_Block-----------------------
5787 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
5788 address stubAddr = NULL;
5789 const char *stubName;
5790 assert(UseAES, "need AES instruction support");
5791
5792 switch(id) {
5793 case vmIntrinsics::_aescrypt_encryptBlock:
5794 stubAddr = StubRoutines::aescrypt_encryptBlock();
5795 stubName = "aescrypt_encryptBlock";
5796 break;
5797 case vmIntrinsics::_aescrypt_decryptBlock:
5798 stubAddr = StubRoutines::aescrypt_decryptBlock();
5799 stubName = "aescrypt_decryptBlock";
5800 break;
5801 default:
5802 break;
5803 }
5804 if (stubAddr == NULL) return false;
5805
5806 Node* aescrypt_object = argument(0);
5807 Node* src = argument(1);
5808 Node* src_offset = argument(2);
5809 Node* dest = argument(3);
5810 Node* dest_offset = argument(4);
5811
5812 src = must_be_not_null(src, true);
5813 dest = must_be_not_null(dest, true);
5814
5815 src = access_resolve(src, ACCESS_READ);
5816 dest = access_resolve(dest, ACCESS_WRITE);
5817
5818 // (1) src and dest are arrays.
5819 const Type* src_type = src->Value(&_gvn);
5820 const Type* dest_type = dest->Value(&_gvn);
5821 const TypeAryPtr* top_src = src_type->isa_aryptr();
5822 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5823 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5824
5825 // for the quick and dirty code we will skip all the checks.
5826 // we are just trying to get the call to be generated.
5827 Node* src_start = src;
5828 Node* dest_start = dest;
5829 if (src_offset != NULL || dest_offset != NULL) {
5830 assert(src_offset != NULL && dest_offset != NULL, "");
5831 src_start = array_element_address(src, src_offset, T_BYTE);
5832 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5833 }
5834
5835 // now need to get the start of its expanded key array
5836 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5837 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5838 if (k_start == NULL) return false;
5839
5840 if (Matcher::pass_original_key_for_aes()) {
5841 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5842 // compatibility issues between Java key expansion and SPARC crypto instructions
5843 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5844 if (original_k_start == NULL) return false;
5845
5846 // Call the stub.
5847 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5848 stubAddr, stubName, TypePtr::BOTTOM,
5849 src_start, dest_start, k_start, original_k_start);
5850 } else {
5851 // Call the stub.
5852 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5853 stubAddr, stubName, TypePtr::BOTTOM,
5854 src_start, dest_start, k_start);
5855 }
5856
5857 return true;
5858 }
5859
5860 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
5861 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
5862 address stubAddr = NULL;
5863 const char *stubName = NULL;
5864
5865 assert(UseAES, "need AES instruction support");
5866
5867 switch(id) {
5868 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
5869 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
5870 stubName = "cipherBlockChaining_encryptAESCrypt";
5871 break;
5872 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
5873 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
5874 stubName = "cipherBlockChaining_decryptAESCrypt";
5875 break;
5876 default:
5877 break;
5878 }
5879 if (stubAddr == NULL) return false;
5880
5881 Node* cipherBlockChaining_object = argument(0);
5882 Node* src = argument(1);
5883 Node* src_offset = argument(2);
5884 Node* len = argument(3);
5885 Node* dest = argument(4);
5886 Node* dest_offset = argument(5);
5887
5888 src = must_be_not_null(src, false);
5889 dest = must_be_not_null(dest, false);
5890
5891 src = access_resolve(src, ACCESS_READ);
5892 dest = access_resolve(dest, ACCESS_WRITE);
5893
5894 // (1) src and dest are arrays.
5895 const Type* src_type = src->Value(&_gvn);
5896 const Type* dest_type = dest->Value(&_gvn);
5897 const TypeAryPtr* top_src = src_type->isa_aryptr();
5898 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5899 assert (top_src != NULL && top_src->klass() != NULL
5900 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5901
5902 // checks are the responsibility of the caller
5903 Node* src_start = src;
5904 Node* dest_start = dest;
5905 if (src_offset != NULL || dest_offset != NULL) {
5906 assert(src_offset != NULL && dest_offset != NULL, "");
5907 src_start = array_element_address(src, src_offset, T_BYTE);
5908 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5909 }
5910
5911 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
5912 // (because of the predicated logic executed earlier).
5913 // so we cast it here safely.
5914 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5915
5916 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
5917 if (embeddedCipherObj == NULL) return false;
5918
5919 // cast it to what we know it will be at runtime
5920 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
5921 assert(tinst != NULL, "CBC obj is null");
5922 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
5923 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5924 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
5925
5926 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
5927 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
5928 const TypeOopPtr* xtype = aklass->as_instance_type();
5929 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
5930 aescrypt_object = _gvn.transform(aescrypt_object);
5931
5932 // we need to get the start of the aescrypt_object's expanded key array
5933 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5934 if (k_start == NULL) return false;
5935
5936 // similarly, get the start address of the r vector
5937 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
5938 if (objRvec == NULL) return false;
5939 objRvec = access_resolve(objRvec, ACCESS_WRITE);
5940 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
5941
5942 Node* cbcCrypt;
5943 if (Matcher::pass_original_key_for_aes()) {
5944 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5945 // compatibility issues between Java key expansion and SPARC crypto instructions
5946 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5947 if (original_k_start == NULL) return false;
5948
5949 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
5950 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5951 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5952 stubAddr, stubName, TypePtr::BOTTOM,
5953 src_start, dest_start, k_start, r_start, len, original_k_start);
5954 } else {
5955 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
5956 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5957 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5958 stubAddr, stubName, TypePtr::BOTTOM,
5959 src_start, dest_start, k_start, r_start, len);
5960 }
5961
5962 // return cipher length (int)
5963 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
5964 set_result(retvalue);
5965 return true;
5966 }
5967
5968 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
5969 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
5970 address stubAddr = NULL;
5971 const char *stubName = NULL;
5972
5973 assert(UseAES, "need AES instruction support");
5974
5975 switch (id) {
5976 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
5977 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
5978 stubName = "electronicCodeBook_encryptAESCrypt";
5979 break;
5980 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
5981 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
5982 stubName = "electronicCodeBook_decryptAESCrypt";
5983 break;
5984 default:
5985 break;
5986 }
5987
5988 if (stubAddr == NULL) return false;
5989
5990 Node* electronicCodeBook_object = argument(0);
5991 Node* src = argument(1);
5992 Node* src_offset = argument(2);
5993 Node* len = argument(3);
5994 Node* dest = argument(4);
5995 Node* dest_offset = argument(5);
5996
5997 // (1) src and dest are arrays.
5998 const Type* src_type = src->Value(&_gvn);
5999 const Type* dest_type = dest->Value(&_gvn);
6000 const TypeAryPtr* top_src = src_type->isa_aryptr();
6001 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6002 assert(top_src != NULL && top_src->klass() != NULL
6003 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6004
6005 // checks are the responsibility of the caller
6006 Node* src_start = src;
6007 Node* dest_start = dest;
6008 if (src_offset != NULL || dest_offset != NULL) {
6009 assert(src_offset != NULL && dest_offset != NULL, "");
6010 src_start = array_element_address(src, src_offset, T_BYTE);
6011 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6012 }
6013
6014 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6015 // (because of the predicated logic executed earlier).
6016 // so we cast it here safely.
6017 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6018
6019 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6020 if (embeddedCipherObj == NULL) return false;
6021
6022 // cast it to what we know it will be at runtime
6023 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6024 assert(tinst != NULL, "ECB obj is null");
6025 assert(tinst->klass()->is_loaded(), "ECB obj is not loaded");
6026 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6027 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6028
6029 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6030 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6031 const TypeOopPtr* xtype = aklass->as_instance_type();
6032 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6033 aescrypt_object = _gvn.transform(aescrypt_object);
6034
6035 // we need to get the start of the aescrypt_object's expanded key array
6036 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6037 if (k_start == NULL) return false;
6038
6039 Node* ecbCrypt;
6040 if (Matcher::pass_original_key_for_aes()) {
6041 // no SPARC version for AES/ECB intrinsics now.
6042 return false;
6043 }
6044 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6045 ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6046 OptoRuntime::electronicCodeBook_aescrypt_Type(),
6047 stubAddr, stubName, TypePtr::BOTTOM,
6048 src_start, dest_start, k_start, len);
6049
6050 // return cipher length (int)
6051 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6052 set_result(retvalue);
6053 return true;
6054 }
6055
6056 //------------------------------inline_counterMode_AESCrypt-----------------------
6057 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6058 assert(UseAES, "need AES instruction support");
6059 if (!UseAESCTRIntrinsics) return false;
6060
6061 address stubAddr = NULL;
6062 const char *stubName = NULL;
6063 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6064 stubAddr = StubRoutines::counterMode_AESCrypt();
6065 stubName = "counterMode_AESCrypt";
6066 }
6067 if (stubAddr == NULL) return false;
6068
6069 Node* counterMode_object = argument(0);
6070 Node* src = argument(1);
6071 Node* src_offset = argument(2);
6072 Node* len = argument(3);
6073 Node* dest = argument(4);
6074 Node* dest_offset = argument(5);
6075
6076 src = access_resolve(src, ACCESS_READ);
6077 dest = access_resolve(dest, ACCESS_WRITE);
6078 counterMode_object = access_resolve(counterMode_object, ACCESS_WRITE);
6079
6080 // (1) src and dest are arrays.
6081 const Type* src_type = src->Value(&_gvn);
6082 const Type* dest_type = dest->Value(&_gvn);
6083 const TypeAryPtr* top_src = src_type->isa_aryptr();
6084 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6085 assert(top_src != NULL && top_src->klass() != NULL &&
6086 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6087
6088 // checks are the responsibility of the caller
6089 Node* src_start = src;
6090 Node* dest_start = dest;
6091 if (src_offset != NULL || dest_offset != NULL) {
6092 assert(src_offset != NULL && dest_offset != NULL, "");
6093 src_start = array_element_address(src, src_offset, T_BYTE);
6094 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6095 }
6096
6097 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6098 // (because of the predicated logic executed earlier).
6099 // so we cast it here safely.
6100 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6101 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6102 if (embeddedCipherObj == NULL) return false;
6103 // cast it to what we know it will be at runtime
6104 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6105 assert(tinst != NULL, "CTR obj is null");
6106 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6107 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6108 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6109 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6110 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6111 const TypeOopPtr* xtype = aklass->as_instance_type();
6112 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6113 aescrypt_object = _gvn.transform(aescrypt_object);
6114 // we need to get the start of the aescrypt_object's expanded key array
6115 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6116 if (k_start == NULL) return false;
6117 // similarly, get the start address of the r vector
6118 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6119 if (obj_counter == NULL) return false;
6120 obj_counter = access_resolve(obj_counter, ACCESS_WRITE);
6121 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6122
6123 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6124 if (saved_encCounter == NULL) return false;
6125 saved_encCounter = access_resolve(saved_encCounter, ACCESS_WRITE);
6126 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6127 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6128
6129 Node* ctrCrypt;
6130 if (Matcher::pass_original_key_for_aes()) {
6131 // no SPARC version for AES/CTR intrinsics now.
6132 return false;
6133 }
6134 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6135 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6136 OptoRuntime::counterMode_aescrypt_Type(),
6137 stubAddr, stubName, TypePtr::BOTTOM,
6138 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6139
6140 // return cipher length (int)
6141 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6142 set_result(retvalue);
6143 return true;
6144 }
6145
6146 //------------------------------get_key_start_from_aescrypt_object-----------------------
6147 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6148 #if defined(PPC64) || defined(S390)
6149 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6150 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6151 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6152 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6153 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6154 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6155 if (objSessionK == NULL) {
6156 return (Node *) NULL;
6157 }
6158 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6159 #else
6160 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6161 #endif // PPC64
6162 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6163 if (objAESCryptKey == NULL) return (Node *) NULL;
6164
6165 // now have the array, need to get the start address of the K array
6166 objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ);
6167 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6168 return k_start;
6169 }
6170
6171 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6172 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6173 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6174 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6175 if (objAESCryptKey == NULL) return (Node *) NULL;
6176
6177 // now have the array, need to get the start address of the lastKey array
6178 objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ);
6179 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6180 return original_k_start;
6181 }
6182
6183 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6184 // Return node representing slow path of predicate check.
6185 // the pseudo code we want to emulate with this predicate is:
6186 // for encryption:
6187 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6188 // for decryption:
6189 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6190 // note cipher==plain is more conservative than the original java code but that's OK
6191 //
6192 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6193 // The receiver was checked for NULL already.
6194 Node* objCBC = argument(0);
6195
6196 Node* src = argument(1);
6197 Node* dest = argument(4);
6198
6199 // Load embeddedCipher field of CipherBlockChaining object.
6200 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6201
6202 // get AESCrypt klass for instanceOf check
6203 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6204 // will have same classloader as CipherBlockChaining object
6205 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6206 assert(tinst != NULL, "CBCobj is null");
6207 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6208
6209 // we want to do an instanceof comparison against the AESCrypt class
6210 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6211 if (!klass_AESCrypt->is_loaded()) {
6212 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6213 Node* ctrl = control();
6214 set_control(top()); // no regular fast path
6215 return ctrl;
6216 }
6217
6218 src = must_be_not_null(src, true);
6219 dest = must_be_not_null(dest, true);
6220
6221 // Resolve oops to stable for CmpP below.
6222 src = access_resolve(src, 0);
6223 dest = access_resolve(dest, 0);
6224
6225 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6226
6227 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6228 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6229 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6230
6231 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6232
6233 // for encryption, we are done
6234 if (!decrypting)
6235 return instof_false; // even if it is NULL
6236
6237 // for decryption, we need to add a further check to avoid
6238 // taking the intrinsic path when cipher and plain are the same
6239 // see the original java code for why.
6240 RegionNode* region = new RegionNode(3);
6241 region->init_req(1, instof_false);
6242
6243 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6244 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6245 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6246 region->init_req(2, src_dest_conjoint);
6247
6248 record_for_igvn(region);
6249 return _gvn.transform(region);
6250 }
6251
6252 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
6253 // Return node representing slow path of predicate check.
6254 // the pseudo code we want to emulate with this predicate is:
6255 // for encryption:
6256 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6257 // for decryption:
6258 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6259 // note cipher==plain is more conservative than the original java code but that's OK
6260 //
6261 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
6262 // The receiver was checked for NULL already.
6263 Node* objECB = argument(0);
6264
6265 // Load embeddedCipher field of ElectronicCodeBook object.
6266 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6267
6268 // get AESCrypt klass for instanceOf check
6269 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6270 // will have same classloader as ElectronicCodeBook object
6271 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
6272 assert(tinst != NULL, "ECBobj is null");
6273 assert(tinst->klass()->is_loaded(), "ECBobj is not loaded");
6274
6275 // we want to do an instanceof comparison against the AESCrypt class
6276 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6277 if (!klass_AESCrypt->is_loaded()) {
6278 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6279 Node* ctrl = control();
6280 set_control(top()); // no regular fast path
6281 return ctrl;
6282 }
6283 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6284
6285 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6286 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6287 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6288
6289 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6290
6291 // for encryption, we are done
6292 if (!decrypting)
6293 return instof_false; // even if it is NULL
6294
6295 // for decryption, we need to add a further check to avoid
6296 // taking the intrinsic path when cipher and plain are the same
6297 // see the original java code for why.
6298 RegionNode* region = new RegionNode(3);
6299 region->init_req(1, instof_false);
6300 Node* src = argument(1);
6301 Node* dest = argument(4);
6302 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6303 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6304 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6305 region->init_req(2, src_dest_conjoint);
6306
6307 record_for_igvn(region);
6308 return _gvn.transform(region);
6309 }
6310
6311 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6312 // Return node representing slow path of predicate check.
6313 // the pseudo code we want to emulate with this predicate is:
6314 // for encryption:
6315 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6316 // for decryption:
6317 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6318 // note cipher==plain is more conservative than the original java code but that's OK
6319 //
6320
6321 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6322 // The receiver was checked for NULL already.
6323 Node* objCTR = argument(0);
6324
6325 // Load embeddedCipher field of CipherBlockChaining object.
6326 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6327
6328 // get AESCrypt klass for instanceOf check
6329 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6330 // will have same classloader as CipherBlockChaining object
6331 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6332 assert(tinst != NULL, "CTRobj is null");
6333 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6334
6335 // we want to do an instanceof comparison against the AESCrypt class
6336 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6337 if (!klass_AESCrypt->is_loaded()) {
6338 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6339 Node* ctrl = control();
6340 set_control(top()); // no regular fast path
6341 return ctrl;
6342 }
6343
6344 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6345 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6346 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6347 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6348 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6349
6350 return instof_false; // even if it is NULL
6351 }
6352
6353 //------------------------------inline_ghash_processBlocks
6354 bool LibraryCallKit::inline_ghash_processBlocks() {
6355 address stubAddr;
6356 const char *stubName;
6357 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6358
6359 stubAddr = StubRoutines::ghash_processBlocks();
6360 stubName = "ghash_processBlocks";
6361
6362 Node* data = argument(0);
6363 Node* offset = argument(1);
6364 Node* len = argument(2);
6365 Node* state = argument(3);
6366 Node* subkeyH = argument(4);
6367
6368 state = must_be_not_null(state, true);
6369 subkeyH = must_be_not_null(subkeyH, true);
6370 data = must_be_not_null(data, true);
6371
6372 state = access_resolve(state, ACCESS_WRITE);
6373 subkeyH = access_resolve(subkeyH, ACCESS_READ);
6374 data = access_resolve(data, ACCESS_READ);
6375
6376 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6377 assert(state_start, "state is NULL");
6378 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6379 assert(subkeyH_start, "subkeyH is NULL");
6380 Node* data_start = array_element_address(data, offset, T_BYTE);
6381 assert(data_start, "data is NULL");
6382
6383 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6384 OptoRuntime::ghash_processBlocks_Type(),
6385 stubAddr, stubName, TypePtr::BOTTOM,
6386 state_start, subkeyH_start, data_start, len);
6387 return true;
6388 }
6389
6390 bool LibraryCallKit::inline_base64_encodeBlock() {
6391 address stubAddr;
6392 const char *stubName;
6393 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6394 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6395 stubAddr = StubRoutines::base64_encodeBlock();
6396 stubName = "encodeBlock";
6397
6398 if (!stubAddr) return false;
6399 Node* base64obj = argument(0);
6400 Node* src = argument(1);
6401 Node* offset = argument(2);
6402 Node* len = argument(3);
6403 Node* dest = argument(4);
6404 Node* dp = argument(5);
6405 Node* isURL = argument(6);
6406
6407 src = must_be_not_null(src, true);
6408 src = access_resolve(src, ACCESS_READ);
6409 dest = must_be_not_null(dest, true);
6410 dest = access_resolve(dest, ACCESS_WRITE);
6411
6412 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6413 assert(src_start, "source array is NULL");
6414 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6415 assert(dest_start, "destination array is NULL");
6416
6417 Node* base64 = make_runtime_call(RC_LEAF,
6418 OptoRuntime::base64_encodeBlock_Type(),
6419 stubAddr, stubName, TypePtr::BOTTOM,
6420 src_start, offset, len, dest_start, dp, isURL);
6421 return true;
6422 }
6423
6424 //------------------------------inline_sha_implCompress-----------------------
6425 //
6426 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6427 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6428 //
6429 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6430 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6431 //
6432 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6433 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6434 //
6435 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6436 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6437
6438 Node* sha_obj = argument(0);
6439 Node* src = argument(1); // type oop
6440 Node* ofs = argument(2); // type int
6441
6442 const Type* src_type = src->Value(&_gvn);
6443 const TypeAryPtr* top_src = src_type->isa_aryptr();
6444 if (top_src == NULL || top_src->klass() == NULL) {
6445 // failed array check
6446 return false;
6447 }
6448 // Figure out the size and type of the elements we will be copying.
6449 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6450 if (src_elem != T_BYTE) {
6451 return false;
6452 }
6453 // 'src_start' points to src array + offset
6454 src = must_be_not_null(src, true);
6455 src = access_resolve(src, ACCESS_READ);
6456 Node* src_start = array_element_address(src, ofs, src_elem);
6457 Node* state = NULL;
6458 address stubAddr;
6459 const char *stubName;
6460
6461 switch(id) {
6462 case vmIntrinsics::_sha_implCompress:
6463 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6464 state = get_state_from_sha_object(sha_obj);
6465 stubAddr = StubRoutines::sha1_implCompress();
6466 stubName = "sha1_implCompress";
6467 break;
6468 case vmIntrinsics::_sha2_implCompress:
6469 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6470 state = get_state_from_sha_object(sha_obj);
6471 stubAddr = StubRoutines::sha256_implCompress();
6472 stubName = "sha256_implCompress";
6473 break;
6474 case vmIntrinsics::_sha5_implCompress:
6475 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6476 state = get_state_from_sha5_object(sha_obj);
6477 stubAddr = StubRoutines::sha512_implCompress();
6478 stubName = "sha512_implCompress";
6479 break;
6480 default:
6481 fatal_unexpected_iid(id);
6482 return false;
6483 }
6484 if (state == NULL) return false;
6485
6486 assert(stubAddr != NULL, "Stub is generated");
6487 if (stubAddr == NULL) return false;
6488
6489 // Call the stub.
6490 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6491 stubAddr, stubName, TypePtr::BOTTOM,
6492 src_start, state);
6493
6494 return true;
6495 }
6496
6497 //------------------------------inline_digestBase_implCompressMB-----------------------
6498 //
6499 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6500 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6501 //
6502 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6503 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6504 "need SHA1/SHA256/SHA512 instruction support");
6505 assert((uint)predicate < 3, "sanity");
6506 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6507
6508 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6509 Node* src = argument(1); // byte[] array
6510 Node* ofs = argument(2); // type int
6511 Node* limit = argument(3); // type int
6512
6513 const Type* src_type = src->Value(&_gvn);
6514 const TypeAryPtr* top_src = src_type->isa_aryptr();
6515 if (top_src == NULL || top_src->klass() == NULL) {
6516 // failed array check
6517 return false;
6518 }
6519 // Figure out the size and type of the elements we will be copying.
6520 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6521 if (src_elem != T_BYTE) {
6522 return false;
6523 }
6524 // 'src_start' points to src array + offset
6525 src = must_be_not_null(src, false);
6526 src = access_resolve(src, ACCESS_READ);
6527 Node* src_start = array_element_address(src, ofs, src_elem);
6528
6529 const char* klass_SHA_name = NULL;
6530 const char* stub_name = NULL;
6531 address stub_addr = NULL;
6532 bool long_state = false;
6533
6534 switch (predicate) {
6535 case 0:
6536 if (UseSHA1Intrinsics) {
6537 klass_SHA_name = "sun/security/provider/SHA";
6538 stub_name = "sha1_implCompressMB";
6539 stub_addr = StubRoutines::sha1_implCompressMB();
6540 }
6541 break;
6542 case 1:
6543 if (UseSHA256Intrinsics) {
6544 klass_SHA_name = "sun/security/provider/SHA2";
6545 stub_name = "sha256_implCompressMB";
6546 stub_addr = StubRoutines::sha256_implCompressMB();
6547 }
6548 break;
6549 case 2:
6550 if (UseSHA512Intrinsics) {
6551 klass_SHA_name = "sun/security/provider/SHA5";
6552 stub_name = "sha512_implCompressMB";
6553 stub_addr = StubRoutines::sha512_implCompressMB();
6554 long_state = true;
6555 }
6556 break;
6557 default:
6558 fatal("unknown SHA intrinsic predicate: %d", predicate);
6559 }
6560 if (klass_SHA_name != NULL) {
6561 assert(stub_addr != NULL, "Stub is generated");
6562 if (stub_addr == NULL) return false;
6563
6564 // get DigestBase klass to lookup for SHA klass
6565 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6566 assert(tinst != NULL, "digestBase_obj is not instance???");
6567 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6568
6569 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6570 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6571 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6572 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6573 }
6574 return false;
6575 }
6576 //------------------------------inline_sha_implCompressMB-----------------------
6577 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6578 bool long_state, address stubAddr, const char *stubName,
6579 Node* src_start, Node* ofs, Node* limit) {
6580 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6581 const TypeOopPtr* xtype = aklass->as_instance_type();
6582 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6583 sha_obj = _gvn.transform(sha_obj);
6584
6585 Node* state;
6586 if (long_state) {
6587 state = get_state_from_sha5_object(sha_obj);
6588 } else {
6589 state = get_state_from_sha_object(sha_obj);
6590 }
6591 if (state == NULL) return false;
6592
6593 // Call the stub.
6594 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6595 OptoRuntime::digestBase_implCompressMB_Type(),
6596 stubAddr, stubName, TypePtr::BOTTOM,
6597 src_start, state, ofs, limit);
6598 // return ofs (int)
6599 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6600 set_result(result);
6601
6602 return true;
6603 }
6604
6605 //------------------------------get_state_from_sha_object-----------------------
6606 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6607 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6608 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6609 if (sha_state == NULL) return (Node *) NULL;
6610
6611 // now have the array, need to get the start address of the state array
6612 sha_state = access_resolve(sha_state, ACCESS_WRITE);
6613 Node* state = array_element_address(sha_state, intcon(0), T_INT);
6614 return state;
6615 }
6616
6617 //------------------------------get_state_from_sha5_object-----------------------
6618 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6619 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6620 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6621 if (sha_state == NULL) return (Node *) NULL;
6622
6623 // now have the array, need to get the start address of the state array
6624 sha_state = access_resolve(sha_state, ACCESS_WRITE);
6625 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6626 return state;
6627 }
6628
6629 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6630 // Return node representing slow path of predicate check.
6631 // the pseudo code we want to emulate with this predicate is:
6632 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6633 //
6634 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6635 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6636 "need SHA1/SHA256/SHA512 instruction support");
6637 assert((uint)predicate < 3, "sanity");
6638
6639 // The receiver was checked for NULL already.
6640 Node* digestBaseObj = argument(0);
6641
6642 // get DigestBase klass for instanceOf check
6643 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6644 assert(tinst != NULL, "digestBaseObj is null");
6645 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6646
6647 const char* klass_SHA_name = NULL;
6648 switch (predicate) {
6649 case 0:
6650 if (UseSHA1Intrinsics) {
6651 // we want to do an instanceof comparison against the SHA class
6652 klass_SHA_name = "sun/security/provider/SHA";
6653 }
6654 break;
6655 case 1:
6656 if (UseSHA256Intrinsics) {
6657 // we want to do an instanceof comparison against the SHA2 class
6658 klass_SHA_name = "sun/security/provider/SHA2";
6659 }
6660 break;
6661 case 2:
6662 if (UseSHA512Intrinsics) {
6663 // we want to do an instanceof comparison against the SHA5 class
6664 klass_SHA_name = "sun/security/provider/SHA5";
6665 }
6666 break;
6667 default:
6668 fatal("unknown SHA intrinsic predicate: %d", predicate);
6669 }
6670
6671 ciKlass* klass_SHA = NULL;
6672 if (klass_SHA_name != NULL) {
6673 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6674 }
6675 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6676 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6677 Node* ctrl = control();
6678 set_control(top()); // no intrinsic path
6679 return ctrl;
6680 }
6681 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6682
6683 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6684 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6685 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6686 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6687
6688 return instof_false; // even if it is NULL
6689 }
6690
6691 //-------------inline_fma-----------------------------------
6692 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
6693 Node *a = NULL;
6694 Node *b = NULL;
6695 Node *c = NULL;
6696 Node* result = NULL;
6697 switch (id) {
6698 case vmIntrinsics::_fmaD:
6699 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
6700 // no receiver since it is static method
6701 a = round_double_node(argument(0));
6702 b = round_double_node(argument(2));
6703 c = round_double_node(argument(4));
6704 result = _gvn.transform(new FmaDNode(control(), a, b, c));
6705 break;
6706 case vmIntrinsics::_fmaF:
6707 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
6708 a = argument(0);
6709 b = argument(1);
6710 c = argument(2);
6711 result = _gvn.transform(new FmaFNode(control(), a, b, c));
6712 break;
6713 default:
6714 fatal_unexpected_iid(id); break;
6715 }
6716 set_result(result);
6717 return true;
6718 }
6719
6720 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
6721 // argument(0) is receiver
6722 Node* codePoint = argument(1);
6723 Node* n = NULL;
6724
6725 switch (id) {
6726 case vmIntrinsics::_isDigit :
6727 n = new DigitNode(control(), codePoint);
6728 break;
6729 case vmIntrinsics::_isLowerCase :
6730 n = new LowerCaseNode(control(), codePoint);
6731 break;
6732 case vmIntrinsics::_isUpperCase :
6733 n = new UpperCaseNode(control(), codePoint);
6734 break;
6735 case vmIntrinsics::_isWhitespace :
6736 n = new WhitespaceNode(control(), codePoint);
6737 break;
6738 default:
6739 fatal_unexpected_iid(id);
6740 }
6741
6742 set_result(_gvn.transform(n));
6743 return true;
6744 }
6745
6746 //------------------------------inline_fp_min_max------------------------------
6747 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
6748 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
6749
6750 // The intrinsic should be used only when the API branches aren't predictable,
6751 // the last one performing the most important comparison. The following heuristic
6752 // uses the branch statistics to eventually bail out if necessary.
6753
6754 ciMethodData *md = callee()->method_data();
6755
6756 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
6757 ciCallProfile cp = caller()->call_profile_at_bci(bci());
6758
6759 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
6760 // Bail out if the call-site didn't contribute enough to the statistics.
6761 return false;
6762 }
6763
6764 uint taken = 0, not_taken = 0;
6765
6766 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
6767 if (p->is_BranchData()) {
6768 taken = ((ciBranchData*)p)->taken();
6769 not_taken = ((ciBranchData*)p)->not_taken();
6770 }
6771 }
6772
6773 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
6774 balance = balance < 0 ? -balance : balance;
6775 if ( balance > 0.2 ) {
6776 // Bail out if the most important branch is predictable enough.
6777 return false;
6778 }
6779 }
6780 */
6781
6782 Node *a = NULL;
6783 Node *b = NULL;
6784 Node *n = NULL;
6785 switch (id) {
6786 case vmIntrinsics::_maxF:
6787 case vmIntrinsics::_minF:
6788 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
6789 a = argument(0);
6790 b = argument(1);
6791 break;
6792 case vmIntrinsics::_maxD:
6793 case vmIntrinsics::_minD:
6794 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
6795 a = round_double_node(argument(0));
6796 b = round_double_node(argument(2));
6797 break;
6798 default:
6799 fatal_unexpected_iid(id);
6800 break;
6801 }
6802 switch (id) {
6803 case vmIntrinsics::_maxF: n = new MaxFNode(a, b); break;
6804 case vmIntrinsics::_minF: n = new MinFNode(a, b); break;
6805 case vmIntrinsics::_maxD: n = new MaxDNode(a, b); break;
6806 case vmIntrinsics::_minD: n = new MinDNode(a, b); break;
6807 default: fatal_unexpected_iid(id); break;
6808 }
6809 set_result(_gvn.transform(n));
6810 return true;
6811 }
6812
6813 bool LibraryCallKit::inline_profileBoolean() {
6814 Node* counts = argument(1);
6815 const TypeAryPtr* ary = NULL;
6816 ciArray* aobj = NULL;
6817 if (counts->is_Con()
6818 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6819 && (aobj = ary->const_oop()->as_array()) != NULL
6820 && (aobj->length() == 2)) {
6821 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6822 jint false_cnt = aobj->element_value(0).as_int();
6823 jint true_cnt = aobj->element_value(1).as_int();
6824
6825 if (C->log() != NULL) {
6826 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6827 false_cnt, true_cnt);
6828 }
6829
6830 if (false_cnt + true_cnt == 0) {
6831 // According to profile, never executed.
6832 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6833 Deoptimization::Action_reinterpret);
6834 return true;
6835 }
6836
6837 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6838 // is a number of each value occurrences.
6839 Node* result = argument(0);
6840 if (false_cnt == 0 || true_cnt == 0) {
6841 // According to profile, one value has been never seen.
6842 int expected_val = (false_cnt == 0) ? 1 : 0;
6843
6844 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6845 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6846
6847 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6848 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6849 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6850
6851 { // Slow path: uncommon trap for never seen value and then reexecute
6852 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6853 // the value has been seen at least once.
6854 PreserveJVMState pjvms(this);
6855 PreserveReexecuteState preexecs(this);
6856 jvms()->set_should_reexecute(true);
6857
6858 set_control(slow_path);
6859 set_i_o(i_o());
6860
6861 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6862 Deoptimization::Action_reinterpret);
6863 }
6864 // The guard for never seen value enables sharpening of the result and
6865 // returning a constant. It allows to eliminate branches on the same value
6866 // later on.
6867 set_control(fast_path);
6868 result = intcon(expected_val);
6869 }
6870 // Stop profiling.
6871 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6872 // By replacing method body with profile data (represented as ProfileBooleanNode
6873 // on IR level) we effectively disable profiling.
6874 // It enables full speed execution once optimized code is generated.
6875 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6876 C->record_for_igvn(profile);
6877 set_result(profile);
6878 return true;
6879 } else {
6880 // Continue profiling.
6881 // Profile data isn't available at the moment. So, execute method's bytecode version.
6882 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6883 // is compiled and counters aren't available since corresponding MethodHandle
6884 // isn't a compile-time constant.
6885 return false;
6886 }
6887 }
6888
6889 bool LibraryCallKit::inline_isCompileConstant() {
6890 Node* n = argument(0);
6891 set_result(n->is_Con() ? intcon(1) : intcon(0));
6892 return true;
6893 }