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 #include "utilities/powerOfTwo.hpp"
62
63 class LibraryIntrinsic : public InlineCallGenerator {
64 // Extend the set of intrinsics known to the runtime:
65 public:
66 private:
67 bool _is_virtual;
68 bool _does_virtual_dispatch;
69 int8_t _predicates_count; // Intrinsic is predicated by several conditions
70 int8_t _last_predicate; // Last generated predicate
71 vmIntrinsics::ID _intrinsic_id;
72
73 public:
74 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
75 : InlineCallGenerator(m),
76 _is_virtual(is_virtual),
77 _does_virtual_dispatch(does_virtual_dispatch),
78 _predicates_count((int8_t)predicates_count),
79 _last_predicate((int8_t)-1),
80 _intrinsic_id(id)
81 {
82 }
83 virtual bool is_intrinsic() const { return true; }
84 virtual bool is_virtual() const { return _is_virtual; }
85 virtual bool is_predicated() const { return _predicates_count > 0; }
86 virtual int predicates_count() const { return _predicates_count; }
87 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; }
88 virtual JVMState* generate(JVMState* jvms);
89 virtual Node* generate_predicate(JVMState* jvms, int predicate);
90 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
91 };
92
93
94 // Local helper class for LibraryIntrinsic:
95 class LibraryCallKit : public GraphKit {
96 private:
97 LibraryIntrinsic* _intrinsic; // the library intrinsic being called
98 Node* _result; // the result node, if any
99 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted
100
101 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type);
102
103 public:
104 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
105 : GraphKit(jvms),
106 _intrinsic(intrinsic),
107 _result(NULL)
108 {
109 // Check if this is a root compile. In that case we don't have a caller.
110 if (!jvms->has_method()) {
111 _reexecute_sp = sp();
112 } else {
113 // Find out how many arguments the interpreter needs when deoptimizing
114 // and save the stack pointer value so it can used by uncommon_trap.
115 // We find the argument count by looking at the declared signature.
116 bool ignored_will_link;
117 ciSignature* declared_signature = NULL;
118 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
119 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
120 _reexecute_sp = sp() + nargs; // "push" arguments back on stack
121 }
122 }
123
124 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
125
126 ciMethod* caller() const { return jvms()->method(); }
127 int bci() const { return jvms()->bci(); }
128 LibraryIntrinsic* intrinsic() const { return _intrinsic; }
129 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); }
130 ciMethod* callee() const { return _intrinsic->method(); }
131
132 bool try_to_inline(int predicate);
133 Node* try_to_predicate(int predicate);
134
135 void push_result() {
136 // Push the result onto the stack.
137 if (!stopped() && result() != NULL) {
138 BasicType bt = result()->bottom_type()->basic_type();
139 push_node(bt, result());
140 }
141 }
142
143 private:
144 void fatal_unexpected_iid(vmIntrinsics::ID iid) {
145 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid));
146 }
147
148 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
149 void set_result(RegionNode* region, PhiNode* value);
150 Node* result() { return _result; }
151
152 virtual int reexecute_sp() { return _reexecute_sp; }
153
154 // Helper functions to inline natives
155 Node* generate_guard(Node* test, RegionNode* region, float true_prob);
156 Node* generate_slow_guard(Node* test, RegionNode* region);
157 Node* generate_fair_guard(Node* test, RegionNode* region);
158 Node* generate_negative_guard(Node* index, RegionNode* region,
159 // resulting CastII of index:
160 Node* *pos_index = NULL);
161 Node* generate_limit_guard(Node* offset, Node* subseq_length,
162 Node* array_length,
163 RegionNode* region);
164 void generate_string_range_check(Node* array, Node* offset,
165 Node* length, bool char_count);
166 Node* generate_current_thread(Node* &tls_output);
167 Node* load_mirror_from_klass(Node* klass);
168 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
169 RegionNode* region, int null_path,
170 int offset);
171 Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
172 RegionNode* region, int null_path) {
173 int offset = java_lang_Class::klass_offset_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_hidden_class_guard(Node* kls, RegionNode* region);
190 Node* generate_array_guard(Node* kls, RegionNode* region) {
191 return generate_array_guard_common(kls, region, false, false);
192 }
193 Node* generate_non_array_guard(Node* kls, RegionNode* region) {
194 return generate_array_guard_common(kls, region, false, true);
195 }
196 Node* generate_objArray_guard(Node* kls, RegionNode* region) {
197 return generate_array_guard_common(kls, region, true, false);
198 }
199 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
200 return generate_array_guard_common(kls, region, true, true);
201 }
202 Node* generate_array_guard_common(Node* kls, RegionNode* region,
203 bool obj_array, bool not_array);
204 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
205 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
206 bool is_virtual = false, bool is_static = false);
207 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
208 return generate_method_call(method_id, false, true);
209 }
210 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
211 return generate_method_call(method_id, true, false);
212 }
213 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
214 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
215
216 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
217 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
218 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
219 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
220 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
221 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
222 bool inline_string_indexOfChar();
223 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
224 bool inline_string_toBytesU();
225 bool inline_string_getCharsU();
226 bool inline_string_copy(bool compress);
227 bool inline_string_char_access(bool is_store);
228 Node* round_double_node(Node* n);
229 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
230 bool inline_math_native(vmIntrinsics::ID id);
231 bool inline_math(vmIntrinsics::ID id);
232 bool inline_double_math(vmIntrinsics::ID id);
233 template <typename OverflowOp>
234 bool inline_math_overflow(Node* arg1, Node* arg2);
235 void inline_math_mathExact(Node* math, Node* test);
236 bool inline_math_addExactI(bool is_increment);
237 bool inline_math_addExactL(bool is_increment);
238 bool inline_math_multiplyExactI();
239 bool inline_math_multiplyExactL();
240 bool inline_math_multiplyHigh();
241 bool inline_math_negateExactI();
242 bool inline_math_negateExactL();
243 bool inline_math_subtractExactI(bool is_decrement);
244 bool inline_math_subtractExactL(bool is_decrement);
245 bool inline_min_max(vmIntrinsics::ID id);
246 bool inline_notify(vmIntrinsics::ID id);
247 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
248 // This returns Type::AnyPtr, RawPtr, or OopPtr.
249 int classify_unsafe_addr(Node* &base, Node* &offset, BasicType type);
250 Node* make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type = T_ILLEGAL, bool can_cast = false);
251
252 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
253 DecoratorSet mo_decorator_for_access_kind(AccessKind kind);
254 bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
255 static bool klass_needs_init_guard(Node* kls);
256 bool inline_unsafe_allocate();
257 bool inline_unsafe_newArray(bool uninitialized);
258 bool inline_unsafe_writeback0();
259 bool inline_unsafe_writebackSync0(bool is_pre);
260 bool inline_unsafe_copyMemory();
261 bool inline_native_currentThread();
262
263 bool inline_native_time_funcs(address method, const char* funcName);
264 #ifdef JFR_HAVE_INTRINSICS
265 bool inline_native_classID();
266 bool inline_native_getEventWriter();
267 #endif
268 bool inline_native_Class_query(vmIntrinsics::ID id);
269 bool inline_native_subtype_check();
270 bool inline_native_getLength();
271 bool inline_array_copyOf(bool is_copyOfRange);
272 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
273 bool inline_preconditions_checkIndex();
274 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array);
275 bool inline_native_clone(bool is_virtual);
276 bool inline_native_Reflection_getCallerClass();
277 // Helper function for inlining native object hash method
278 bool inline_native_hashcode(bool is_virtual, bool is_static);
279 bool inline_native_getClass();
280
281 // Helper functions for inlining arraycopy
282 bool inline_arraycopy();
283 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
284 RegionNode* slow_region);
285 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
286 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp,
287 uint new_idx);
288
289 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
290 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
291 bool inline_unsafe_fence(vmIntrinsics::ID id);
292 bool inline_onspinwait();
293 bool inline_fp_conversions(vmIntrinsics::ID id);
294 bool inline_number_methods(vmIntrinsics::ID id);
295 bool inline_reference_get();
296 bool inline_Class_cast();
297 bool inline_aescrypt_Block(vmIntrinsics::ID id);
298 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
299 bool inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id);
300 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
301 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
302 Node* inline_electronicCodeBook_AESCrypt_predicate(bool decrypting);
303 Node* inline_counterMode_AESCrypt_predicate();
304 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
305 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
306 bool inline_ghash_processBlocks();
307 bool inline_base64_encodeBlock();
308 bool inline_sha_implCompress(vmIntrinsics::ID id);
309 bool inline_digestBase_implCompressMB(int predicate);
310 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
311 bool long_state, address stubAddr, const char *stubName,
312 Node* src_start, Node* ofs, Node* limit);
313 Node* get_state_from_sha_object(Node *sha_object);
314 Node* get_state_from_sha5_object(Node *sha_object);
315 Node* inline_digestBase_implCompressMB_predicate(int predicate);
316 bool inline_encodeISOArray();
317 bool inline_updateCRC32();
318 bool inline_updateBytesCRC32();
319 bool inline_updateByteBufferCRC32();
320 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
321 bool inline_updateBytesCRC32C();
322 bool inline_updateDirectByteBufferCRC32C();
323 bool inline_updateBytesAdler32();
324 bool inline_updateByteBufferAdler32();
325 bool inline_multiplyToLen();
326 bool inline_hasNegatives();
327 bool inline_squareToLen();
328 bool inline_mulAdd();
329 bool inline_montgomeryMultiply();
330 bool inline_montgomerySquare();
331 bool inline_bigIntegerShift(bool isRightShift);
332 bool inline_vectorizedMismatch();
333 bool inline_fma(vmIntrinsics::ID id);
334 bool inline_character_compare(vmIntrinsics::ID id);
335 bool inline_fp_min_max(vmIntrinsics::ID id);
336
337 bool inline_profileBoolean();
338 bool inline_isCompileConstant();
339 void clear_upper_avx() {
340 #ifdef X86
341 if (UseAVX >= 2) {
342 C->set_clear_upper_avx(true);
343 }
344 #endif
345 }
346 };
347
348 //---------------------------make_vm_intrinsic----------------------------
349 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
350 vmIntrinsics::ID id = m->intrinsic_id();
351 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
352
353 if (!m->is_loaded()) {
354 // Do not attempt to inline unloaded methods.
355 return NULL;
356 }
357
358 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
359 bool is_available = false;
360
361 {
362 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
363 // the compiler must transition to '_thread_in_vm' state because both
364 // methods access VM-internal data.
365 VM_ENTRY_MARK;
366 methodHandle mh(THREAD, m->get_Method());
367 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
368 !C->directive()->is_intrinsic_disabled(mh) &&
369 !vmIntrinsics::is_disabled_by_flags(mh);
370
371 }
372
373 if (is_available) {
374 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
375 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
376 return new LibraryIntrinsic(m, is_virtual,
377 vmIntrinsics::predicates_needed(id),
378 vmIntrinsics::does_virtual_dispatch(id),
379 (vmIntrinsics::ID) id);
380 } else {
381 return NULL;
382 }
383 }
384
385 //----------------------register_library_intrinsics-----------------------
386 // Initialize this file's data structures, for each Compile instance.
387 void Compile::register_library_intrinsics() {
388 // Nothing to do here.
389 }
390
391 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
392 LibraryCallKit kit(jvms, this);
393 Compile* C = kit.C;
394 int nodes = C->unique();
395 #ifndef PRODUCT
396 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
397 char buf[1000];
398 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
399 tty->print_cr("Intrinsic %s", str);
400 }
401 #endif
402 ciMethod* callee = kit.callee();
403 const int bci = kit.bci();
404
405 // Try to inline the intrinsic.
406 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
407 kit.try_to_inline(_last_predicate)) {
408 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
409 : "(intrinsic)";
410 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
411 if (C->print_intrinsics() || C->print_inlining()) {
412 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
413 }
414 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
415 if (C->log()) {
416 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
417 vmIntrinsics::name_at(intrinsic_id()),
418 (is_virtual() ? " virtual='1'" : ""),
419 C->unique() - nodes);
420 }
421 // Push the result from the inlined method onto the stack.
422 kit.push_result();
423 C->print_inlining_update(this);
424 return kit.transfer_exceptions_into_jvms();
425 }
426
427 // The intrinsic bailed out
428 if (jvms->has_method()) {
429 // Not a root compile.
430 const char* msg;
431 if (callee->intrinsic_candidate()) {
432 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
433 } else {
434 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
435 : "failed to inline (intrinsic), method not annotated";
436 }
437 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
438 if (C->print_intrinsics() || C->print_inlining()) {
439 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
440 }
441 } else {
442 // Root compile
443 ResourceMark rm;
444 stringStream msg_stream;
445 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
446 vmIntrinsics::name_at(intrinsic_id()),
447 is_virtual() ? " (virtual)" : "", bci);
448 const char *msg = msg_stream.as_string();
449 log_debug(jit, inlining)("%s", msg);
450 if (C->print_intrinsics() || C->print_inlining()) {
451 tty->print("%s", msg);
452 }
453 }
454 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
455 C->print_inlining_update(this);
456 return NULL;
457 }
458
459 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
460 LibraryCallKit kit(jvms, this);
461 Compile* C = kit.C;
462 int nodes = C->unique();
463 _last_predicate = predicate;
464 #ifndef PRODUCT
465 assert(is_predicated() && predicate < predicates_count(), "sanity");
466 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
467 char buf[1000];
468 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
469 tty->print_cr("Predicate for intrinsic %s", str);
470 }
471 #endif
472 ciMethod* callee = kit.callee();
473 const int bci = kit.bci();
474
475 Node* slow_ctl = kit.try_to_predicate(predicate);
476 if (!kit.failing()) {
477 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
478 : "(intrinsic, predicate)";
479 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
480 if (C->print_intrinsics() || C->print_inlining()) {
481 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
482 }
483 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
484 if (C->log()) {
485 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
486 vmIntrinsics::name_at(intrinsic_id()),
487 (is_virtual() ? " virtual='1'" : ""),
488 C->unique() - nodes);
489 }
490 return slow_ctl; // Could be NULL if the check folds.
491 }
492
493 // The intrinsic bailed out
494 if (jvms->has_method()) {
495 // Not a root compile.
496 const char* msg = "failed to generate predicate for intrinsic";
497 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
498 if (C->print_intrinsics() || C->print_inlining()) {
499 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
500 }
501 } else {
502 // Root compile
503 ResourceMark rm;
504 stringStream msg_stream;
505 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
506 vmIntrinsics::name_at(intrinsic_id()),
507 is_virtual() ? " (virtual)" : "", bci);
508 const char *msg = msg_stream.as_string();
509 log_debug(jit, inlining)("%s", msg);
510 if (C->print_intrinsics() || C->print_inlining()) {
511 C->print_inlining_stream()->print("%s", msg);
512 }
513 }
514 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
515 return NULL;
516 }
517
518 bool LibraryCallKit::try_to_inline(int predicate) {
519 // Handle symbolic names for otherwise undistinguished boolean switches:
520 const bool is_store = true;
521 const bool is_compress = true;
522 const bool is_static = true;
523 const bool is_volatile = true;
524
525 if (!jvms()->has_method()) {
526 // Root JVMState has a null method.
527 assert(map()->memory()->Opcode() == Op_Parm, "");
528 // Insert the memory aliasing node
529 set_all_memory(reset_memory());
530 }
531 assert(merged_memory(), "");
532
533
534 switch (intrinsic_id()) {
535 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
536 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
537 case vmIntrinsics::_getClass: return inline_native_getClass();
538
539 case vmIntrinsics::_ceil:
540 case vmIntrinsics::_floor:
541 case vmIntrinsics::_rint:
542 case vmIntrinsics::_dsin:
543 case vmIntrinsics::_dcos:
544 case vmIntrinsics::_dtan:
545 case vmIntrinsics::_dabs:
546 case vmIntrinsics::_fabs:
547 case vmIntrinsics::_iabs:
548 case vmIntrinsics::_labs:
549 case vmIntrinsics::_datan2:
550 case vmIntrinsics::_dsqrt:
551 case vmIntrinsics::_dexp:
552 case vmIntrinsics::_dlog:
553 case vmIntrinsics::_dlog10:
554 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id());
555
556 case vmIntrinsics::_min:
557 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
558
559 case vmIntrinsics::_notify:
560 case vmIntrinsics::_notifyAll:
561 return inline_notify(intrinsic_id());
562
563 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
564 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
565 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
566 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
567 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
568 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
569 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
570 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
571 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
572 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
573 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
574 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
575 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
576
577 case vmIntrinsics::_arraycopy: return inline_arraycopy();
578
579 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
580 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
581 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
582 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
583
584 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
585 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
586 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
587 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
588 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
589 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
590 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
591
592 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
593 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
594
595 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
596 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
597 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
598 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
599
600 case vmIntrinsics::_compressStringC:
601 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
602 case vmIntrinsics::_inflateStringC:
603 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
604
605 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
606 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
607 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
608 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
609 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
610 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
611 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
612 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
613 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
614
615 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
616 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
617 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
618 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
619 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
620 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
621 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
622 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
623 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
624
625 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
626 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
627 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
628 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
629 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
630 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
631 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
632 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
633 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
634
635 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
636 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
637 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
638 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
639 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
640 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
641 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
642 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
643 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
644
645 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
646 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
647 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
648 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
649
650 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
651 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
652 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
653 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
654
655 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
656 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
657 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
658 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
659 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
660 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
661 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
662 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
663 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
664
665 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
666 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
667 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
668 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
669 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
670 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
671 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
672 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
673 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
674
675 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
676 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
677 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
678 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
679 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
680 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
681 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
682 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
683 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
684
685 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
686 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
687 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
688 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
689 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
690 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
691 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
692 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
693 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
694
695 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
696 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
697 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
698 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
699 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
700
701 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
702 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
703 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
704 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
705 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
706 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
707 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
708 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
709 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
710 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
711 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
712 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
713 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
714 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
715 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
716 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
717 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
718 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
719 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
720 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
721
722 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
723 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
724 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
725 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
726 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
727 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
728 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
729 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
730 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
731 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
732 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
733 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
734 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
735 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
736 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
737
738 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
739 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
740 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
741 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
742
743 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
744 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
745 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
746 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
747 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
748
749 case vmIntrinsics::_loadFence:
750 case vmIntrinsics::_storeFence:
751 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
752
753 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
754
755 case vmIntrinsics::_currentThread: return inline_native_currentThread();
756
757 #ifdef JFR_HAVE_INTRINSICS
758 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime");
759 case vmIntrinsics::_getClassId: return inline_native_classID();
760 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
761 #endif
762 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
763 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
764 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
765 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
766 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
767 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
768 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
769 case vmIntrinsics::_getLength: return inline_native_getLength();
770 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
771 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
772 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
773 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
774 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
775 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
776
777 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
778 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
779
780 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
781
782 case vmIntrinsics::_isInstance:
783 case vmIntrinsics::_getModifiers:
784 case vmIntrinsics::_isInterface:
785 case vmIntrinsics::_isArray:
786 case vmIntrinsics::_isPrimitive:
787 case vmIntrinsics::_isHiddenClass:
788 case vmIntrinsics::_getSuperclass:
789 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
790
791 case vmIntrinsics::_floatToRawIntBits:
792 case vmIntrinsics::_floatToIntBits:
793 case vmIntrinsics::_intBitsToFloat:
794 case vmIntrinsics::_doubleToRawLongBits:
795 case vmIntrinsics::_doubleToLongBits:
796 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
797
798 case vmIntrinsics::_numberOfLeadingZeros_i:
799 case vmIntrinsics::_numberOfLeadingZeros_l:
800 case vmIntrinsics::_numberOfTrailingZeros_i:
801 case vmIntrinsics::_numberOfTrailingZeros_l:
802 case vmIntrinsics::_bitCount_i:
803 case vmIntrinsics::_bitCount_l:
804 case vmIntrinsics::_reverseBytes_i:
805 case vmIntrinsics::_reverseBytes_l:
806 case vmIntrinsics::_reverseBytes_s:
807 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
808
809 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
810
811 case vmIntrinsics::_Reference_get: return inline_reference_get();
812
813 case vmIntrinsics::_Class_cast: return inline_Class_cast();
814
815 case vmIntrinsics::_aescrypt_encryptBlock:
816 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
817
818 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
819 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
820 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
821
822 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
823 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
824 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
825
826 case vmIntrinsics::_counterMode_AESCrypt:
827 return inline_counterMode_AESCrypt(intrinsic_id());
828
829 case vmIntrinsics::_sha_implCompress:
830 case vmIntrinsics::_sha2_implCompress:
831 case vmIntrinsics::_sha5_implCompress:
832 return inline_sha_implCompress(intrinsic_id());
833
834 case vmIntrinsics::_digestBase_implCompressMB:
835 return inline_digestBase_implCompressMB(predicate);
836
837 case vmIntrinsics::_multiplyToLen:
838 return inline_multiplyToLen();
839
840 case vmIntrinsics::_squareToLen:
841 return inline_squareToLen();
842
843 case vmIntrinsics::_mulAdd:
844 return inline_mulAdd();
845
846 case vmIntrinsics::_montgomeryMultiply:
847 return inline_montgomeryMultiply();
848 case vmIntrinsics::_montgomerySquare:
849 return inline_montgomerySquare();
850
851 case vmIntrinsics::_bigIntegerRightShiftWorker:
852 return inline_bigIntegerShift(true);
853 case vmIntrinsics::_bigIntegerLeftShiftWorker:
854 return inline_bigIntegerShift(false);
855
856 case vmIntrinsics::_vectorizedMismatch:
857 return inline_vectorizedMismatch();
858
859 case vmIntrinsics::_ghash_processBlocks:
860 return inline_ghash_processBlocks();
861 case vmIntrinsics::_base64_encodeBlock:
862 return inline_base64_encodeBlock();
863
864 case vmIntrinsics::_encodeISOArray:
865 case vmIntrinsics::_encodeByteISOArray:
866 return inline_encodeISOArray();
867
868 case vmIntrinsics::_updateCRC32:
869 return inline_updateCRC32();
870 case vmIntrinsics::_updateBytesCRC32:
871 return inline_updateBytesCRC32();
872 case vmIntrinsics::_updateByteBufferCRC32:
873 return inline_updateByteBufferCRC32();
874
875 case vmIntrinsics::_updateBytesCRC32C:
876 return inline_updateBytesCRC32C();
877 case vmIntrinsics::_updateDirectByteBufferCRC32C:
878 return inline_updateDirectByteBufferCRC32C();
879
880 case vmIntrinsics::_updateBytesAdler32:
881 return inline_updateBytesAdler32();
882 case vmIntrinsics::_updateByteBufferAdler32:
883 return inline_updateByteBufferAdler32();
884
885 case vmIntrinsics::_profileBoolean:
886 return inline_profileBoolean();
887 case vmIntrinsics::_isCompileConstant:
888 return inline_isCompileConstant();
889
890 case vmIntrinsics::_hasNegatives:
891 return inline_hasNegatives();
892
893 case vmIntrinsics::_fmaD:
894 case vmIntrinsics::_fmaF:
895 return inline_fma(intrinsic_id());
896
897 case vmIntrinsics::_isDigit:
898 case vmIntrinsics::_isLowerCase:
899 case vmIntrinsics::_isUpperCase:
900 case vmIntrinsics::_isWhitespace:
901 return inline_character_compare(intrinsic_id());
902
903 case vmIntrinsics::_maxF:
904 case vmIntrinsics::_minF:
905 case vmIntrinsics::_maxD:
906 case vmIntrinsics::_minD:
907 return inline_fp_min_max(intrinsic_id());
908
909 default:
910 // If you get here, it may be that someone has added a new intrinsic
911 // to the list in vmSymbols.hpp without implementing it here.
912 #ifndef PRODUCT
913 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
914 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
915 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
916 }
917 #endif
918 return false;
919 }
920 }
921
922 Node* LibraryCallKit::try_to_predicate(int predicate) {
923 if (!jvms()->has_method()) {
924 // Root JVMState has a null method.
925 assert(map()->memory()->Opcode() == Op_Parm, "");
926 // Insert the memory aliasing node
927 set_all_memory(reset_memory());
928 }
929 assert(merged_memory(), "");
930
931 switch (intrinsic_id()) {
932 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
933 return inline_cipherBlockChaining_AESCrypt_predicate(false);
934 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
935 return inline_cipherBlockChaining_AESCrypt_predicate(true);
936 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
937 return inline_electronicCodeBook_AESCrypt_predicate(false);
938 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
939 return inline_electronicCodeBook_AESCrypt_predicate(true);
940 case vmIntrinsics::_counterMode_AESCrypt:
941 return inline_counterMode_AESCrypt_predicate();
942 case vmIntrinsics::_digestBase_implCompressMB:
943 return inline_digestBase_implCompressMB_predicate(predicate);
944
945 default:
946 // If you get here, it may be that someone has added a new intrinsic
947 // to the list in vmSymbols.hpp without implementing it here.
948 #ifndef PRODUCT
949 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
950 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
951 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
952 }
953 #endif
954 Node* slow_ctl = control();
955 set_control(top()); // No fast path instrinsic
956 return slow_ctl;
957 }
958 }
959
960 //------------------------------set_result-------------------------------
961 // Helper function for finishing intrinsics.
962 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
963 record_for_igvn(region);
964 set_control(_gvn.transform(region));
965 set_result( _gvn.transform(value));
966 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
967 }
968
969 //------------------------------generate_guard---------------------------
970 // Helper function for generating guarded fast-slow graph structures.
971 // The given 'test', if true, guards a slow path. If the test fails
972 // then a fast path can be taken. (We generally hope it fails.)
973 // In all cases, GraphKit::control() is updated to the fast path.
974 // The returned value represents the control for the slow path.
975 // The return value is never 'top'; it is either a valid control
976 // or NULL if it is obvious that the slow path can never be taken.
977 // Also, if region and the slow control are not NULL, the slow edge
978 // is appended to the region.
979 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
980 if (stopped()) {
981 // Already short circuited.
982 return NULL;
983 }
984
985 // Build an if node and its projections.
986 // If test is true we take the slow path, which we assume is uncommon.
987 if (_gvn.type(test) == TypeInt::ZERO) {
988 // The slow branch is never taken. No need to build this guard.
989 return NULL;
990 }
991
992 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
993
994 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
995 if (if_slow == top()) {
996 // The slow branch is never taken. No need to build this guard.
997 return NULL;
998 }
999
1000 if (region != NULL)
1001 region->add_req(if_slow);
1002
1003 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
1004 set_control(if_fast);
1005
1006 return if_slow;
1007 }
1008
1009 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
1010 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
1011 }
1012 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
1013 return generate_guard(test, region, PROB_FAIR);
1014 }
1015
1016 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
1017 Node* *pos_index) {
1018 if (stopped())
1019 return NULL; // already stopped
1020 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
1021 return NULL; // index is already adequately typed
1022 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
1023 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1024 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
1025 if (is_neg != NULL && pos_index != NULL) {
1026 // Emulate effect of Parse::adjust_map_after_if.
1027 Node* ccast = new CastIINode(index, TypeInt::POS);
1028 ccast->set_req(0, control());
1029 (*pos_index) = _gvn.transform(ccast);
1030 }
1031 return is_neg;
1032 }
1033
1034 // Make sure that 'position' is a valid limit index, in [0..length].
1035 // There are two equivalent plans for checking this:
1036 // A. (offset + copyLength) unsigned<= arrayLength
1037 // B. offset <= (arrayLength - copyLength)
1038 // We require that all of the values above, except for the sum and
1039 // difference, are already known to be non-negative.
1040 // Plan A is robust in the face of overflow, if offset and copyLength
1041 // are both hugely positive.
1042 //
1043 // Plan B is less direct and intuitive, but it does not overflow at
1044 // all, since the difference of two non-negatives is always
1045 // representable. Whenever Java methods must perform the equivalent
1046 // check they generally use Plan B instead of Plan A.
1047 // For the moment we use Plan A.
1048 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
1049 Node* subseq_length,
1050 Node* array_length,
1051 RegionNode* region) {
1052 if (stopped())
1053 return NULL; // already stopped
1054 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
1055 if (zero_offset && subseq_length->eqv_uncast(array_length))
1056 return NULL; // common case of whole-array copy
1057 Node* last = subseq_length;
1058 if (!zero_offset) // last += offset
1059 last = _gvn.transform(new AddINode(last, offset));
1060 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
1061 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1062 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
1063 return is_over;
1064 }
1065
1066 // Emit range checks for the given String.value byte array
1067 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
1068 if (stopped()) {
1069 return; // already stopped
1070 }
1071 RegionNode* bailout = new RegionNode(1);
1072 record_for_igvn(bailout);
1073 if (char_count) {
1074 // Convert char count to byte count
1075 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1076 }
1077
1078 // Offset and count must not be negative
1079 generate_negative_guard(offset, bailout);
1080 generate_negative_guard(count, bailout);
1081 // Offset + count must not exceed length of array
1082 generate_limit_guard(offset, count, load_array_length(array), bailout);
1083
1084 if (bailout->req() > 1) {
1085 PreserveJVMState pjvms(this);
1086 set_control(_gvn.transform(bailout));
1087 uncommon_trap(Deoptimization::Reason_intrinsic,
1088 Deoptimization::Action_maybe_recompile);
1089 }
1090 }
1091
1092 //--------------------------generate_current_thread--------------------
1093 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1094 ciKlass* thread_klass = env()->Thread_klass();
1095 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1096 Node* thread = _gvn.transform(new ThreadLocalNode());
1097 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1098 Node* threadObj = _gvn.transform(LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(), thread_type, T_OBJECT, MemNode::unordered));
1099 tls_output = thread;
1100 return threadObj;
1101 }
1102
1103
1104 //------------------------------make_string_method_node------------------------
1105 // Helper method for String intrinsic functions. This version is called with
1106 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1107 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1108 // containing the lengths of str1 and str2.
1109 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1110 Node* result = NULL;
1111 switch (opcode) {
1112 case Op_StrIndexOf:
1113 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1114 str1_start, cnt1, str2_start, cnt2, ae);
1115 break;
1116 case Op_StrComp:
1117 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1118 str1_start, cnt1, str2_start, cnt2, ae);
1119 break;
1120 case Op_StrEquals:
1121 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1122 // Use the constant length if there is one because optimized match rule may exist.
1123 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1124 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1125 break;
1126 default:
1127 ShouldNotReachHere();
1128 return NULL;
1129 }
1130
1131 // All these intrinsics have checks.
1132 C->set_has_split_ifs(true); // Has chance for split-if optimization
1133 clear_upper_avx();
1134
1135 return _gvn.transform(result);
1136 }
1137
1138 //------------------------------inline_string_compareTo------------------------
1139 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1140 Node* arg1 = argument(0);
1141 Node* arg2 = argument(1);
1142
1143 arg1 = must_be_not_null(arg1, true);
1144 arg2 = must_be_not_null(arg2, true);
1145
1146 // Get start addr and length of first argument
1147 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1148 Node* arg1_cnt = load_array_length(arg1);
1149
1150 // Get start addr and length of second argument
1151 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1152 Node* arg2_cnt = load_array_length(arg2);
1153
1154 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1155 set_result(result);
1156 return true;
1157 }
1158
1159 //------------------------------inline_string_equals------------------------
1160 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1161 Node* arg1 = argument(0);
1162 Node* arg2 = argument(1);
1163
1164 // paths (plus control) merge
1165 RegionNode* region = new RegionNode(3);
1166 Node* phi = new PhiNode(region, TypeInt::BOOL);
1167
1168 if (!stopped()) {
1169
1170 arg1 = must_be_not_null(arg1, true);
1171 arg2 = must_be_not_null(arg2, true);
1172
1173 // Get start addr and length of first argument
1174 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1175 Node* arg1_cnt = load_array_length(arg1);
1176
1177 // Get start addr and length of second argument
1178 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1179 Node* arg2_cnt = load_array_length(arg2);
1180
1181 // Check for arg1_cnt != arg2_cnt
1182 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1183 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1184 Node* if_ne = generate_slow_guard(bol, NULL);
1185 if (if_ne != NULL) {
1186 phi->init_req(2, intcon(0));
1187 region->init_req(2, if_ne);
1188 }
1189
1190 // Check for count == 0 is done by assembler code for StrEquals.
1191
1192 if (!stopped()) {
1193 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1194 phi->init_req(1, equals);
1195 region->init_req(1, control());
1196 }
1197 }
1198
1199 // post merge
1200 set_control(_gvn.transform(region));
1201 record_for_igvn(region);
1202
1203 set_result(_gvn.transform(phi));
1204 return true;
1205 }
1206
1207 //------------------------------inline_array_equals----------------------------
1208 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1209 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1210 Node* arg1 = argument(0);
1211 Node* arg2 = argument(1);
1212
1213 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1214 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1215 clear_upper_avx();
1216
1217 return true;
1218 }
1219
1220 //------------------------------inline_hasNegatives------------------------------
1221 bool LibraryCallKit::inline_hasNegatives() {
1222 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1223 return false;
1224 }
1225
1226 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1227 // no receiver since it is static method
1228 Node* ba = argument(0);
1229 Node* offset = argument(1);
1230 Node* len = argument(2);
1231
1232 ba = must_be_not_null(ba, true);
1233
1234 // Range checks
1235 generate_string_range_check(ba, offset, len, false);
1236 if (stopped()) {
1237 return true;
1238 }
1239 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1240 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1241 set_result(_gvn.transform(result));
1242 return true;
1243 }
1244
1245 bool LibraryCallKit::inline_preconditions_checkIndex() {
1246 Node* index = argument(0);
1247 Node* length = argument(1);
1248 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1249 return false;
1250 }
1251
1252 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1253 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1254
1255 {
1256 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1257 uncommon_trap(Deoptimization::Reason_intrinsic,
1258 Deoptimization::Action_make_not_entrant);
1259 }
1260
1261 if (stopped()) {
1262 return false;
1263 }
1264
1265 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1266 BoolTest::mask btest = BoolTest::lt;
1267 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1268 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1269 _gvn.set_type(rc, rc->Value(&_gvn));
1270 if (!rc_bool->is_Con()) {
1271 record_for_igvn(rc);
1272 }
1273 set_control(_gvn.transform(new IfTrueNode(rc)));
1274 {
1275 PreserveJVMState pjvms(this);
1276 set_control(_gvn.transform(new IfFalseNode(rc)));
1277 uncommon_trap(Deoptimization::Reason_range_check,
1278 Deoptimization::Action_make_not_entrant);
1279 }
1280
1281 if (stopped()) {
1282 return false;
1283 }
1284
1285 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1286 result->set_req(0, control());
1287 result = _gvn.transform(result);
1288 set_result(result);
1289 replace_in_map(index, result);
1290 clear_upper_avx();
1291 return true;
1292 }
1293
1294 //------------------------------inline_string_indexOf------------------------
1295 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1296 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1297 return false;
1298 }
1299 Node* src = argument(0);
1300 Node* tgt = argument(1);
1301
1302 // Make the merge point
1303 RegionNode* result_rgn = new RegionNode(4);
1304 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1305
1306 src = must_be_not_null(src, true);
1307 tgt = must_be_not_null(tgt, true);
1308
1309 // Get start addr and length of source string
1310 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1311 Node* src_count = load_array_length(src);
1312
1313 // Get start addr and length of substring
1314 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1315 Node* tgt_count = load_array_length(tgt);
1316
1317 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1318 // Divide src size by 2 if String is UTF16 encoded
1319 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1320 }
1321 if (ae == StrIntrinsicNode::UU) {
1322 // Divide substring size by 2 if String is UTF16 encoded
1323 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1324 }
1325
1326 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1327 if (result != NULL) {
1328 result_phi->init_req(3, result);
1329 result_rgn->init_req(3, control());
1330 }
1331 set_control(_gvn.transform(result_rgn));
1332 record_for_igvn(result_rgn);
1333 set_result(_gvn.transform(result_phi));
1334
1335 return true;
1336 }
1337
1338 //-----------------------------inline_string_indexOf-----------------------
1339 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1340 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1341 return false;
1342 }
1343 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1344 return false;
1345 }
1346 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1347 Node* src = argument(0); // byte[]
1348 Node* src_count = argument(1); // char count
1349 Node* tgt = argument(2); // byte[]
1350 Node* tgt_count = argument(3); // char count
1351 Node* from_index = argument(4); // char index
1352
1353 src = must_be_not_null(src, true);
1354 tgt = must_be_not_null(tgt, true);
1355
1356 // Multiply byte array index by 2 if String is UTF16 encoded
1357 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1358 src_count = _gvn.transform(new SubINode(src_count, from_index));
1359 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1360 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1361
1362 // Range checks
1363 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1364 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1365 if (stopped()) {
1366 return true;
1367 }
1368
1369 RegionNode* region = new RegionNode(5);
1370 Node* phi = new PhiNode(region, TypeInt::INT);
1371
1372 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1373 if (result != NULL) {
1374 // The result is index relative to from_index if substring was found, -1 otherwise.
1375 // Generate code which will fold into cmove.
1376 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1377 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1378
1379 Node* if_lt = generate_slow_guard(bol, NULL);
1380 if (if_lt != NULL) {
1381 // result == -1
1382 phi->init_req(3, result);
1383 region->init_req(3, if_lt);
1384 }
1385 if (!stopped()) {
1386 result = _gvn.transform(new AddINode(result, from_index));
1387 phi->init_req(4, result);
1388 region->init_req(4, control());
1389 }
1390 }
1391
1392 set_control(_gvn.transform(region));
1393 record_for_igvn(region);
1394 set_result(_gvn.transform(phi));
1395 clear_upper_avx();
1396
1397 return true;
1398 }
1399
1400 // Create StrIndexOfNode with fast path checks
1401 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1402 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1403 // Check for substr count > string count
1404 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1405 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1406 Node* if_gt = generate_slow_guard(bol, NULL);
1407 if (if_gt != NULL) {
1408 phi->init_req(1, intcon(-1));
1409 region->init_req(1, if_gt);
1410 }
1411 if (!stopped()) {
1412 // Check for substr count == 0
1413 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1414 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1415 Node* if_zero = generate_slow_guard(bol, NULL);
1416 if (if_zero != NULL) {
1417 phi->init_req(2, intcon(0));
1418 region->init_req(2, if_zero);
1419 }
1420 }
1421 if (!stopped()) {
1422 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1423 }
1424 return NULL;
1425 }
1426
1427 //-----------------------------inline_string_indexOfChar-----------------------
1428 bool LibraryCallKit::inline_string_indexOfChar() {
1429 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1430 return false;
1431 }
1432 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1433 return false;
1434 }
1435 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1436 Node* src = argument(0); // byte[]
1437 Node* tgt = argument(1); // tgt is int ch
1438 Node* from_index = argument(2);
1439 Node* max = argument(3);
1440
1441 src = must_be_not_null(src, true);
1442
1443 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1444 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1445 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1446
1447 // Range checks
1448 generate_string_range_check(src, src_offset, src_count, true);
1449 if (stopped()) {
1450 return true;
1451 }
1452
1453 RegionNode* region = new RegionNode(3);
1454 Node* phi = new PhiNode(region, TypeInt::INT);
1455
1456 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1457 C->set_has_split_ifs(true); // Has chance for split-if optimization
1458 _gvn.transform(result);
1459
1460 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1461 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1462
1463 Node* if_lt = generate_slow_guard(bol, NULL);
1464 if (if_lt != NULL) {
1465 // result == -1
1466 phi->init_req(2, result);
1467 region->init_req(2, if_lt);
1468 }
1469 if (!stopped()) {
1470 result = _gvn.transform(new AddINode(result, from_index));
1471 phi->init_req(1, result);
1472 region->init_req(1, control());
1473 }
1474 set_control(_gvn.transform(region));
1475 record_for_igvn(region);
1476 set_result(_gvn.transform(phi));
1477
1478 return true;
1479 }
1480 //---------------------------inline_string_copy---------------------
1481 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1482 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1483 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1484 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1485 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1486 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1487 bool LibraryCallKit::inline_string_copy(bool compress) {
1488 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1489 return false;
1490 }
1491 int nargs = 5; // 2 oops, 3 ints
1492 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1493
1494 Node* src = argument(0);
1495 Node* src_offset = argument(1);
1496 Node* dst = argument(2);
1497 Node* dst_offset = argument(3);
1498 Node* length = argument(4);
1499
1500 // Check for allocation before we add nodes that would confuse
1501 // tightly_coupled_allocation()
1502 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1503
1504 // Figure out the size and type of the elements we will be copying.
1505 const Type* src_type = src->Value(&_gvn);
1506 const Type* dst_type = dst->Value(&_gvn);
1507 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1508 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1509 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1510 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1511 "Unsupported array types for inline_string_copy");
1512
1513 src = must_be_not_null(src, true);
1514 dst = must_be_not_null(dst, true);
1515
1516 // Convert char[] offsets to byte[] offsets
1517 bool convert_src = (compress && src_elem == T_BYTE);
1518 bool convert_dst = (!compress && dst_elem == T_BYTE);
1519 if (convert_src) {
1520 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1521 } else if (convert_dst) {
1522 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1523 }
1524
1525 // Range checks
1526 generate_string_range_check(src, src_offset, length, convert_src);
1527 generate_string_range_check(dst, dst_offset, length, convert_dst);
1528 if (stopped()) {
1529 return true;
1530 }
1531
1532 Node* src_start = array_element_address(src, src_offset, src_elem);
1533 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1534 // 'src_start' points to src array + scaled offset
1535 // 'dst_start' points to dst array + scaled offset
1536 Node* count = NULL;
1537 if (compress) {
1538 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1539 } else {
1540 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1541 }
1542
1543 if (alloc != NULL) {
1544 if (alloc->maybe_set_complete(&_gvn)) {
1545 // "You break it, you buy it."
1546 InitializeNode* init = alloc->initialization();
1547 assert(init->is_complete(), "we just did this");
1548 init->set_complete_with_arraycopy();
1549 assert(dst->is_CheckCastPP(), "sanity");
1550 assert(dst->in(0)->in(0) == init, "dest pinned");
1551 }
1552 // Do not let stores that initialize this object be reordered with
1553 // a subsequent store that would make this object accessible by
1554 // other threads.
1555 // Record what AllocateNode this StoreStore protects so that
1556 // escape analysis can go from the MemBarStoreStoreNode to the
1557 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1558 // based on the escape status of the AllocateNode.
1559 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1560 }
1561 if (compress) {
1562 set_result(_gvn.transform(count));
1563 }
1564 clear_upper_avx();
1565
1566 return true;
1567 }
1568
1569 #ifdef _LP64
1570 #define XTOP ,top() /*additional argument*/
1571 #else //_LP64
1572 #define XTOP /*no additional argument*/
1573 #endif //_LP64
1574
1575 //------------------------inline_string_toBytesU--------------------------
1576 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1577 bool LibraryCallKit::inline_string_toBytesU() {
1578 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1579 return false;
1580 }
1581 // Get the arguments.
1582 Node* value = argument(0);
1583 Node* offset = argument(1);
1584 Node* length = argument(2);
1585
1586 Node* newcopy = NULL;
1587
1588 // Set the original stack and the reexecute bit for the interpreter to reexecute
1589 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1590 { PreserveReexecuteState preexecs(this);
1591 jvms()->set_should_reexecute(true);
1592
1593 // Check if a null path was taken unconditionally.
1594 value = null_check(value);
1595
1596 RegionNode* bailout = new RegionNode(1);
1597 record_for_igvn(bailout);
1598
1599 // Range checks
1600 generate_negative_guard(offset, bailout);
1601 generate_negative_guard(length, bailout);
1602 generate_limit_guard(offset, length, load_array_length(value), bailout);
1603 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1604 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1605
1606 if (bailout->req() > 1) {
1607 PreserveJVMState pjvms(this);
1608 set_control(_gvn.transform(bailout));
1609 uncommon_trap(Deoptimization::Reason_intrinsic,
1610 Deoptimization::Action_maybe_recompile);
1611 }
1612 if (stopped()) {
1613 return true;
1614 }
1615
1616 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1617 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1618 newcopy = new_array(klass_node, size, 0); // no arguments to push
1619 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1620
1621 // Calculate starting addresses.
1622 Node* src_start = array_element_address(value, offset, T_CHAR);
1623 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1624
1625 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1626 const TypeInt* toffset = gvn().type(offset)->is_int();
1627 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1628
1629 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1630 const char* copyfunc_name = "arraycopy";
1631 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1632 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1633 OptoRuntime::fast_arraycopy_Type(),
1634 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1635 src_start, dst_start, ConvI2X(length) XTOP);
1636 // Do not let reads from the cloned object float above the arraycopy.
1637 if (alloc != NULL) {
1638 if (alloc->maybe_set_complete(&_gvn)) {
1639 // "You break it, you buy it."
1640 InitializeNode* init = alloc->initialization();
1641 assert(init->is_complete(), "we just did this");
1642 init->set_complete_with_arraycopy();
1643 assert(newcopy->is_CheckCastPP(), "sanity");
1644 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1645 }
1646 // Do not let stores that initialize this object be reordered with
1647 // a subsequent store that would make this object accessible by
1648 // other threads.
1649 // Record what AllocateNode this StoreStore protects so that
1650 // escape analysis can go from the MemBarStoreStoreNode to the
1651 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1652 // based on the escape status of the AllocateNode.
1653 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1654 } else {
1655 insert_mem_bar(Op_MemBarCPUOrder);
1656 }
1657 } // original reexecute is set back here
1658
1659 C->set_has_split_ifs(true); // Has chance for split-if optimization
1660 if (!stopped()) {
1661 set_result(newcopy);
1662 }
1663 clear_upper_avx();
1664
1665 return true;
1666 }
1667
1668 //------------------------inline_string_getCharsU--------------------------
1669 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1670 bool LibraryCallKit::inline_string_getCharsU() {
1671 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1672 return false;
1673 }
1674
1675 // Get the arguments.
1676 Node* src = argument(0);
1677 Node* src_begin = argument(1);
1678 Node* src_end = argument(2); // exclusive offset (i < src_end)
1679 Node* dst = argument(3);
1680 Node* dst_begin = argument(4);
1681
1682 // Check for allocation before we add nodes that would confuse
1683 // tightly_coupled_allocation()
1684 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1685
1686 // Check if a null path was taken unconditionally.
1687 src = null_check(src);
1688 dst = null_check(dst);
1689 if (stopped()) {
1690 return true;
1691 }
1692
1693 // Get length and convert char[] offset to byte[] offset
1694 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1695 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1696
1697 // Range checks
1698 generate_string_range_check(src, src_begin, length, true);
1699 generate_string_range_check(dst, dst_begin, length, false);
1700 if (stopped()) {
1701 return true;
1702 }
1703
1704 if (!stopped()) {
1705 // Calculate starting addresses.
1706 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1707 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1708
1709 // Check if array addresses are aligned to HeapWordSize
1710 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1711 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1712 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1713 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1714
1715 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1716 const char* copyfunc_name = "arraycopy";
1717 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1718 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1719 OptoRuntime::fast_arraycopy_Type(),
1720 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1721 src_start, dst_start, ConvI2X(length) XTOP);
1722 // Do not let reads from the cloned object float above the arraycopy.
1723 if (alloc != NULL) {
1724 if (alloc->maybe_set_complete(&_gvn)) {
1725 // "You break it, you buy it."
1726 InitializeNode* init = alloc->initialization();
1727 assert(init->is_complete(), "we just did this");
1728 init->set_complete_with_arraycopy();
1729 assert(dst->is_CheckCastPP(), "sanity");
1730 assert(dst->in(0)->in(0) == init, "dest pinned");
1731 }
1732 // Do not let stores that initialize this object be reordered with
1733 // a subsequent store that would make this object accessible by
1734 // other threads.
1735 // Record what AllocateNode this StoreStore protects so that
1736 // escape analysis can go from the MemBarStoreStoreNode to the
1737 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1738 // based on the escape status of the AllocateNode.
1739 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1740 } else {
1741 insert_mem_bar(Op_MemBarCPUOrder);
1742 }
1743 }
1744
1745 C->set_has_split_ifs(true); // Has chance for split-if optimization
1746 return true;
1747 }
1748
1749 //----------------------inline_string_char_access----------------------------
1750 // Store/Load char to/from byte[] array.
1751 // static void StringUTF16.putChar(byte[] val, int index, int c)
1752 // static char StringUTF16.getChar(byte[] val, int index)
1753 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1754 Node* value = argument(0);
1755 Node* index = argument(1);
1756 Node* ch = is_store ? argument(2) : NULL;
1757
1758 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1759 // correctly requires matched array shapes.
1760 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1761 "sanity: byte[] and char[] bases agree");
1762 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1763 "sanity: byte[] and char[] scales agree");
1764
1765 // Bail when getChar over constants is requested: constant folding would
1766 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1767 // Java method would constant fold nicely instead.
1768 if (!is_store && value->is_Con() && index->is_Con()) {
1769 return false;
1770 }
1771
1772 value = must_be_not_null(value, true);
1773
1774 Node* adr = array_element_address(value, index, T_CHAR);
1775 if (adr->is_top()) {
1776 return false;
1777 }
1778 if (is_store) {
1779 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1780 } else {
1781 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
1782 set_result(ch);
1783 }
1784 return true;
1785 }
1786
1787 //--------------------------round_double_node--------------------------------
1788 // Round a double node if necessary.
1789 Node* LibraryCallKit::round_double_node(Node* n) {
1790 if (Matcher::strict_fp_requires_explicit_rounding) {
1791 #ifdef IA32
1792 if (UseSSE < 2) {
1793 n = _gvn.transform(new RoundDoubleNode(NULL, n));
1794 }
1795 #else
1796 Unimplemented();
1797 #endif // IA32
1798 }
1799 return n;
1800 }
1801
1802 //------------------------------inline_math-----------------------------------
1803 // public static double Math.abs(double)
1804 // public static double Math.sqrt(double)
1805 // public static double Math.log(double)
1806 // public static double Math.log10(double)
1807 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1808 Node* arg = round_double_node(argument(0));
1809 Node* n = NULL;
1810 switch (id) {
1811 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1812 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1813 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1814 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1815 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1816 default: fatal_unexpected_iid(id); break;
1817 }
1818 set_result(_gvn.transform(n));
1819 return true;
1820 }
1821
1822 //------------------------------inline_math-----------------------------------
1823 // public static float Math.abs(float)
1824 // public static int Math.abs(int)
1825 // public static long Math.abs(long)
1826 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1827 Node* arg = argument(0);
1828 Node* n = NULL;
1829 switch (id) {
1830 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1831 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1832 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1833 default: fatal_unexpected_iid(id); break;
1834 }
1835 set_result(_gvn.transform(n));
1836 return true;
1837 }
1838
1839 //------------------------------runtime_math-----------------------------
1840 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1841 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1842 "must be (DD)D or (D)D type");
1843
1844 // Inputs
1845 Node* a = round_double_node(argument(0));
1846 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1847
1848 const TypePtr* no_memory_effects = NULL;
1849 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1850 no_memory_effects,
1851 a, top(), b, b ? top() : NULL);
1852 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1853 #ifdef ASSERT
1854 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1855 assert(value_top == top(), "second value must be top");
1856 #endif
1857
1858 set_result(value);
1859 return true;
1860 }
1861
1862 //------------------------------inline_math_native-----------------------------
1863 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1864 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1865 switch (id) {
1866 // These intrinsics are not properly supported on all hardware
1867 case vmIntrinsics::_dsin:
1868 return StubRoutines::dsin() != NULL ?
1869 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1870 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1871 case vmIntrinsics::_dcos:
1872 return StubRoutines::dcos() != NULL ?
1873 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1874 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1875 case vmIntrinsics::_dtan:
1876 return StubRoutines::dtan() != NULL ?
1877 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1878 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1879 case vmIntrinsics::_dlog:
1880 return StubRoutines::dlog() != NULL ?
1881 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1882 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1883 case vmIntrinsics::_dlog10:
1884 return StubRoutines::dlog10() != NULL ?
1885 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1886 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1887
1888 // These intrinsics are supported on all hardware
1889 case vmIntrinsics::_ceil:
1890 case vmIntrinsics::_floor:
1891 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1892 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1893 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1894 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1895 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1896 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1897
1898 case vmIntrinsics::_dexp:
1899 return StubRoutines::dexp() != NULL ?
1900 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1901 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1902 case vmIntrinsics::_dpow: {
1903 Node* exp = round_double_node(argument(2));
1904 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1905 if (d != NULL && d->getd() == 2.0) {
1906 // Special case: pow(x, 2.0) => x * x
1907 Node* base = round_double_node(argument(0));
1908 set_result(_gvn.transform(new MulDNode(base, base)));
1909 return true;
1910 }
1911 return StubRoutines::dpow() != NULL ?
1912 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1913 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1914 }
1915 #undef FN_PTR
1916
1917 // These intrinsics are not yet correctly implemented
1918 case vmIntrinsics::_datan2:
1919 return false;
1920
1921 default:
1922 fatal_unexpected_iid(id);
1923 return false;
1924 }
1925 }
1926
1927 static bool is_simple_name(Node* n) {
1928 return (n->req() == 1 // constant
1929 || (n->is_Type() && n->as_Type()->type()->singleton())
1930 || n->is_Proj() // parameter or return value
1931 || n->is_Phi() // local of some sort
1932 );
1933 }
1934
1935 //----------------------------inline_notify-----------------------------------*
1936 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1937 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1938 address func;
1939 if (id == vmIntrinsics::_notify) {
1940 func = OptoRuntime::monitor_notify_Java();
1941 } else {
1942 func = OptoRuntime::monitor_notifyAll_Java();
1943 }
1944 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1945 make_slow_call_ex(call, env()->Throwable_klass(), false);
1946 return true;
1947 }
1948
1949
1950 //----------------------------inline_min_max-----------------------------------
1951 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1952 set_result(generate_min_max(id, argument(0), argument(1)));
1953 return true;
1954 }
1955
1956 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1957 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1958 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1959 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1960 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1961
1962 {
1963 PreserveJVMState pjvms(this);
1964 PreserveReexecuteState preexecs(this);
1965 jvms()->set_should_reexecute(true);
1966
1967 set_control(slow_path);
1968 set_i_o(i_o());
1969
1970 uncommon_trap(Deoptimization::Reason_intrinsic,
1971 Deoptimization::Action_none);
1972 }
1973
1974 set_control(fast_path);
1975 set_result(math);
1976 }
1977
1978 template <typename OverflowOp>
1979 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1980 typedef typename OverflowOp::MathOp MathOp;
1981
1982 MathOp* mathOp = new MathOp(arg1, arg2);
1983 Node* operation = _gvn.transform( mathOp );
1984 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1985 inline_math_mathExact(operation, ofcheck);
1986 return true;
1987 }
1988
1989 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1990 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1991 }
1992
1993 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1994 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1995 }
1996
1997 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1998 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1999 }
2000
2001 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2002 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2003 }
2004
2005 bool LibraryCallKit::inline_math_negateExactI() {
2006 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2007 }
2008
2009 bool LibraryCallKit::inline_math_negateExactL() {
2010 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2011 }
2012
2013 bool LibraryCallKit::inline_math_multiplyExactI() {
2014 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2015 }
2016
2017 bool LibraryCallKit::inline_math_multiplyExactL() {
2018 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2019 }
2020
2021 bool LibraryCallKit::inline_math_multiplyHigh() {
2022 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2023 return true;
2024 }
2025
2026 Node*
2027 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2028 // These are the candidate return value:
2029 Node* xvalue = x0;
2030 Node* yvalue = y0;
2031
2032 if (xvalue == yvalue) {
2033 return xvalue;
2034 }
2035
2036 bool want_max = (id == vmIntrinsics::_max);
2037
2038 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
2039 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
2040 if (txvalue == NULL || tyvalue == NULL) return top();
2041 // This is not really necessary, but it is consistent with a
2042 // hypothetical MaxINode::Value method:
2043 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
2044
2045 // %%% This folding logic should (ideally) be in a different place.
2046 // Some should be inside IfNode, and there to be a more reliable
2047 // transformation of ?: style patterns into cmoves. We also want
2048 // more powerful optimizations around cmove and min/max.
2049
2050 // Try to find a dominating comparison of these guys.
2051 // It can simplify the index computation for Arrays.copyOf
2052 // and similar uses of System.arraycopy.
2053 // First, compute the normalized version of CmpI(x, y).
2054 int cmp_op = Op_CmpI;
2055 Node* xkey = xvalue;
2056 Node* ykey = yvalue;
2057 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
2058 if (ideal_cmpxy->is_Cmp()) {
2059 // E.g., if we have CmpI(length - offset, count),
2060 // it might idealize to CmpI(length, count + offset)
2061 cmp_op = ideal_cmpxy->Opcode();
2062 xkey = ideal_cmpxy->in(1);
2063 ykey = ideal_cmpxy->in(2);
2064 }
2065
2066 // Start by locating any relevant comparisons.
2067 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
2068 Node* cmpxy = NULL;
2069 Node* cmpyx = NULL;
2070 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
2071 Node* cmp = start_from->fast_out(k);
2072 if (cmp->outcnt() > 0 && // must have prior uses
2073 cmp->in(0) == NULL && // must be context-independent
2074 cmp->Opcode() == cmp_op) { // right kind of compare
2075 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
2076 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
2077 }
2078 }
2079
2080 const int NCMPS = 2;
2081 Node* cmps[NCMPS] = { cmpxy, cmpyx };
2082 int cmpn;
2083 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2084 if (cmps[cmpn] != NULL) break; // find a result
2085 }
2086 if (cmpn < NCMPS) {
2087 // Look for a dominating test that tells us the min and max.
2088 int depth = 0; // Limit search depth for speed
2089 Node* dom = control();
2090 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2091 if (++depth >= 100) break;
2092 Node* ifproj = dom;
2093 if (!ifproj->is_Proj()) continue;
2094 Node* iff = ifproj->in(0);
2095 if (!iff->is_If()) continue;
2096 Node* bol = iff->in(1);
2097 if (!bol->is_Bool()) continue;
2098 Node* cmp = bol->in(1);
2099 if (cmp == NULL) continue;
2100 for (cmpn = 0; cmpn < NCMPS; cmpn++)
2101 if (cmps[cmpn] == cmp) break;
2102 if (cmpn == NCMPS) continue;
2103 BoolTest::mask btest = bol->as_Bool()->_test._test;
2104 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
2105 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2106 // At this point, we know that 'x btest y' is true.
2107 switch (btest) {
2108 case BoolTest::eq:
2109 // They are proven equal, so we can collapse the min/max.
2110 // Either value is the answer. Choose the simpler.
2111 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2112 return yvalue;
2113 return xvalue;
2114 case BoolTest::lt: // x < y
2115 case BoolTest::le: // x <= y
2116 return (want_max ? yvalue : xvalue);
2117 case BoolTest::gt: // x > y
2118 case BoolTest::ge: // x >= y
2119 return (want_max ? xvalue : yvalue);
2120 default:
2121 break;
2122 }
2123 }
2124 }
2125
2126 // We failed to find a dominating test.
2127 // Let's pick a test that might GVN with prior tests.
2128 Node* best_bol = NULL;
2129 BoolTest::mask best_btest = BoolTest::illegal;
2130 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2131 Node* cmp = cmps[cmpn];
2132 if (cmp == NULL) continue;
2133 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2134 Node* bol = cmp->fast_out(j);
2135 if (!bol->is_Bool()) continue;
2136 BoolTest::mask btest = bol->as_Bool()->_test._test;
2137 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2138 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2139 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2140 best_bol = bol->as_Bool();
2141 best_btest = btest;
2142 }
2143 }
2144 }
2145
2146 Node* answer_if_true = NULL;
2147 Node* answer_if_false = NULL;
2148 switch (best_btest) {
2149 default:
2150 if (cmpxy == NULL)
2151 cmpxy = ideal_cmpxy;
2152 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2153 // and fall through:
2154 case BoolTest::lt: // x < y
2155 case BoolTest::le: // x <= y
2156 answer_if_true = (want_max ? yvalue : xvalue);
2157 answer_if_false = (want_max ? xvalue : yvalue);
2158 break;
2159 case BoolTest::gt: // x > y
2160 case BoolTest::ge: // x >= y
2161 answer_if_true = (want_max ? xvalue : yvalue);
2162 answer_if_false = (want_max ? yvalue : xvalue);
2163 break;
2164 }
2165
2166 jint hi, lo;
2167 if (want_max) {
2168 // We can sharpen the minimum.
2169 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2170 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2171 } else {
2172 // We can sharpen the maximum.
2173 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2174 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2175 }
2176
2177 // Use a flow-free graph structure, to avoid creating excess control edges
2178 // which could hinder other optimizations.
2179 // Since Math.min/max is often used with arraycopy, we want
2180 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2181 Node* cmov = CMoveNode::make(NULL, best_bol,
2182 answer_if_false, answer_if_true,
2183 TypeInt::make(lo, hi, widen));
2184
2185 return _gvn.transform(cmov);
2186
2187 /*
2188 // This is not as desirable as it may seem, since Min and Max
2189 // nodes do not have a full set of optimizations.
2190 // And they would interfere, anyway, with 'if' optimizations
2191 // and with CMoveI canonical forms.
2192 switch (id) {
2193 case vmIntrinsics::_min:
2194 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2195 case vmIntrinsics::_max:
2196 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2197 default:
2198 ShouldNotReachHere();
2199 }
2200 */
2201 }
2202
2203 inline int
2204 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2205 const TypePtr* base_type = TypePtr::NULL_PTR;
2206 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2207 if (base_type == NULL) {
2208 // Unknown type.
2209 return Type::AnyPtr;
2210 } else if (base_type == TypePtr::NULL_PTR) {
2211 // Since this is a NULL+long form, we have to switch to a rawptr.
2212 base = _gvn.transform(new CastX2PNode(offset));
2213 offset = MakeConX(0);
2214 return Type::RawPtr;
2215 } else if (base_type->base() == Type::RawPtr) {
2216 return Type::RawPtr;
2217 } else if (base_type->isa_oopptr()) {
2218 // Base is never null => always a heap address.
2219 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2220 return Type::OopPtr;
2221 }
2222 // Offset is small => always a heap address.
2223 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2224 if (offset_type != NULL &&
2225 base_type->offset() == 0 && // (should always be?)
2226 offset_type->_lo >= 0 &&
2227 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2228 return Type::OopPtr;
2229 } else if (type == T_OBJECT) {
2230 // off heap access to an oop doesn't make any sense. Has to be on
2231 // heap.
2232 return Type::OopPtr;
2233 }
2234 // Otherwise, it might either be oop+off or NULL+addr.
2235 return Type::AnyPtr;
2236 } else {
2237 // No information:
2238 return Type::AnyPtr;
2239 }
2240 }
2241
2242 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type, bool can_cast) {
2243 Node* uncasted_base = base;
2244 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2245 if (kind == Type::RawPtr) {
2246 return basic_plus_adr(top(), uncasted_base, offset);
2247 } else if (kind == Type::AnyPtr) {
2248 assert(base == uncasted_base, "unexpected base change");
2249 if (can_cast) {
2250 if (!_gvn.type(base)->speculative_maybe_null() &&
2251 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2252 // According to profiling, this access is always on
2253 // heap. Casting the base to not null and thus avoiding membars
2254 // around the access should allow better optimizations
2255 Node* null_ctl = top();
2256 base = null_check_oop(base, &null_ctl, true, true, true);
2257 assert(null_ctl->is_top(), "no null control here");
2258 return basic_plus_adr(base, offset);
2259 } else if (_gvn.type(base)->speculative_always_null() &&
2260 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2261 // According to profiling, this access is always off
2262 // heap.
2263 base = null_assert(base);
2264 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2265 offset = MakeConX(0);
2266 return basic_plus_adr(top(), raw_base, offset);
2267 }
2268 }
2269 // We don't know if it's an on heap or off heap access. Fall back
2270 // to raw memory access.
2271 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2272 return basic_plus_adr(top(), raw, offset);
2273 } else {
2274 assert(base == uncasted_base, "unexpected base change");
2275 // We know it's an on heap access so base can't be null
2276 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2277 base = must_be_not_null(base, true);
2278 }
2279 return basic_plus_adr(base, offset);
2280 }
2281 }
2282
2283 //--------------------------inline_number_methods-----------------------------
2284 // inline int Integer.numberOfLeadingZeros(int)
2285 // inline int Long.numberOfLeadingZeros(long)
2286 //
2287 // inline int Integer.numberOfTrailingZeros(int)
2288 // inline int Long.numberOfTrailingZeros(long)
2289 //
2290 // inline int Integer.bitCount(int)
2291 // inline int Long.bitCount(long)
2292 //
2293 // inline char Character.reverseBytes(char)
2294 // inline short Short.reverseBytes(short)
2295 // inline int Integer.reverseBytes(int)
2296 // inline long Long.reverseBytes(long)
2297 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2298 Node* arg = argument(0);
2299 Node* n = NULL;
2300 switch (id) {
2301 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2302 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2303 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2304 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2305 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2306 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2307 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2308 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2309 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2310 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2311 default: fatal_unexpected_iid(id); break;
2312 }
2313 set_result(_gvn.transform(n));
2314 return true;
2315 }
2316
2317 //----------------------------inline_unsafe_access----------------------------
2318
2319 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2320 // Attempt to infer a sharper value type from the offset and base type.
2321 ciKlass* sharpened_klass = NULL;
2322
2323 // See if it is an instance field, with an object type.
2324 if (alias_type->field() != NULL) {
2325 if (alias_type->field()->type()->is_klass()) {
2326 sharpened_klass = alias_type->field()->type()->as_klass();
2327 }
2328 }
2329
2330 // See if it is a narrow oop array.
2331 if (adr_type->isa_aryptr()) {
2332 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2333 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2334 if (elem_type != NULL) {
2335 sharpened_klass = elem_type->klass();
2336 }
2337 }
2338 }
2339
2340 // The sharpened class might be unloaded if there is no class loader
2341 // contraint in place.
2342 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2343 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2344
2345 #ifndef PRODUCT
2346 if (C->print_intrinsics() || C->print_inlining()) {
2347 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2348 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2349 }
2350 #endif
2351 // Sharpen the value type.
2352 return tjp;
2353 }
2354 return NULL;
2355 }
2356
2357 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2358 switch (kind) {
2359 case Relaxed:
2360 return MO_UNORDERED;
2361 case Opaque:
2362 return MO_RELAXED;
2363 case Acquire:
2364 return MO_ACQUIRE;
2365 case Release:
2366 return MO_RELEASE;
2367 case Volatile:
2368 return MO_SEQ_CST;
2369 default:
2370 ShouldNotReachHere();
2371 return 0;
2372 }
2373 }
2374
2375 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2376 if (callee()->is_static()) return false; // caller must have the capability!
2377 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2378 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2379 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2380 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2381
2382 if (is_reference_type(type)) {
2383 decorators |= ON_UNKNOWN_OOP_REF;
2384 }
2385
2386 if (unaligned) {
2387 decorators |= C2_UNALIGNED;
2388 }
2389
2390 #ifndef PRODUCT
2391 {
2392 ResourceMark rm;
2393 // Check the signatures.
2394 ciSignature* sig = callee()->signature();
2395 #ifdef ASSERT
2396 if (!is_store) {
2397 // Object getReference(Object base, int/long offset), etc.
2398 BasicType rtype = sig->return_type()->basic_type();
2399 assert(rtype == type, "getter must return the expected value");
2400 assert(sig->count() == 2, "oop getter has 2 arguments");
2401 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2402 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2403 } else {
2404 // void putReference(Object base, int/long offset, Object x), etc.
2405 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2406 assert(sig->count() == 3, "oop putter has 3 arguments");
2407 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2408 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2409 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2410 assert(vtype == type, "putter must accept the expected value");
2411 }
2412 #endif // ASSERT
2413 }
2414 #endif //PRODUCT
2415
2416 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2417
2418 Node* receiver = argument(0); // type: oop
2419
2420 // Build address expression.
2421 Node* adr;
2422 Node* heap_base_oop = top();
2423 Node* offset = top();
2424 Node* val;
2425
2426 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2427 Node* base = argument(1); // type: oop
2428 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2429 offset = argument(2); // type: long
2430 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2431 // to be plain byte offsets, which are also the same as those accepted
2432 // by oopDesc::field_addr.
2433 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2434 "fieldOffset must be byte-scaled");
2435 // 32-bit machines ignore the high half!
2436 offset = ConvL2X(offset);
2437 adr = make_unsafe_address(base, offset, is_store ? ACCESS_WRITE : ACCESS_READ, type, kind == Relaxed);
2438
2439 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2440 if (type != T_OBJECT) {
2441 decorators |= IN_NATIVE; // off-heap primitive access
2442 } else {
2443 return false; // off-heap oop accesses are not supported
2444 }
2445 } else {
2446 heap_base_oop = base; // on-heap or mixed access
2447 }
2448
2449 // Can base be NULL? Otherwise, always on-heap access.
2450 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2451
2452 if (!can_access_non_heap) {
2453 decorators |= IN_HEAP;
2454 }
2455
2456 val = is_store ? argument(4) : NULL;
2457
2458 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2459 if (adr_type == TypePtr::NULL_PTR) {
2460 return false; // off-heap access with zero address
2461 }
2462
2463 // Try to categorize the address.
2464 Compile::AliasType* alias_type = C->alias_type(adr_type);
2465 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2466
2467 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2468 alias_type->adr_type() == TypeAryPtr::RANGE) {
2469 return false; // not supported
2470 }
2471
2472 bool mismatched = false;
2473 BasicType bt = alias_type->basic_type();
2474 if (bt != T_ILLEGAL) {
2475 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2476 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2477 // Alias type doesn't differentiate between byte[] and boolean[]).
2478 // Use address type to get the element type.
2479 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2480 }
2481 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2482 // accessing an array field with getReference is not a mismatch
2483 bt = T_OBJECT;
2484 }
2485 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2486 // Don't intrinsify mismatched object accesses
2487 return false;
2488 }
2489 mismatched = (bt != type);
2490 } else if (alias_type->adr_type()->isa_oopptr()) {
2491 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2492 }
2493
2494 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2495
2496 if (mismatched) {
2497 decorators |= C2_MISMATCHED;
2498 }
2499
2500 // First guess at the value type.
2501 const Type *value_type = Type::get_const_basic_type(type);
2502
2503 // Figure out the memory ordering.
2504 decorators |= mo_decorator_for_access_kind(kind);
2505
2506 if (!is_store && type == T_OBJECT) {
2507 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2508 if (tjp != NULL) {
2509 value_type = tjp;
2510 }
2511 }
2512
2513 receiver = null_check(receiver);
2514 if (stopped()) {
2515 return true;
2516 }
2517 // Heap pointers get a null-check from the interpreter,
2518 // as a courtesy. However, this is not guaranteed by Unsafe,
2519 // and it is not possible to fully distinguish unintended nulls
2520 // from intended ones in this API.
2521
2522 if (!is_store) {
2523 Node* p = NULL;
2524 // Try to constant fold a load from a constant field
2525 ciField* field = alias_type->field();
2526 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2527 // final or stable field
2528 p = make_constant_from_field(field, heap_base_oop);
2529 }
2530
2531 if (p == NULL) { // Could not constant fold the load
2532 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2533 // Normalize the value returned by getBoolean in the following cases
2534 if (type == T_BOOLEAN &&
2535 (mismatched ||
2536 heap_base_oop == top() || // - heap_base_oop is NULL or
2537 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2538 // and the unsafe access is made to large offset
2539 // (i.e., larger than the maximum offset necessary for any
2540 // field access)
2541 ) {
2542 IdealKit ideal = IdealKit(this);
2543 #define __ ideal.
2544 IdealVariable normalized_result(ideal);
2545 __ declarations_done();
2546 __ set(normalized_result, p);
2547 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2548 __ set(normalized_result, ideal.ConI(1));
2549 ideal.end_if();
2550 final_sync(ideal);
2551 p = __ value(normalized_result);
2552 #undef __
2553 }
2554 }
2555 if (type == T_ADDRESS) {
2556 p = gvn().transform(new CastP2XNode(NULL, p));
2557 p = ConvX2UL(p);
2558 }
2559 // The load node has the control of the preceding MemBarCPUOrder. All
2560 // following nodes will have the control of the MemBarCPUOrder inserted at
2561 // the end of this method. So, pushing the load onto the stack at a later
2562 // point is fine.
2563 set_result(p);
2564 } else {
2565 if (bt == T_ADDRESS) {
2566 // Repackage the long as a pointer.
2567 val = ConvL2X(val);
2568 val = gvn().transform(new CastX2PNode(val));
2569 }
2570 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2571 }
2572
2573 return true;
2574 }
2575
2576 //----------------------------inline_unsafe_load_store----------------------------
2577 // This method serves a couple of different customers (depending on LoadStoreKind):
2578 //
2579 // LS_cmp_swap:
2580 //
2581 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2582 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2583 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2584 //
2585 // LS_cmp_swap_weak:
2586 //
2587 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2588 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2589 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2590 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2591 //
2592 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2593 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2594 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2595 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2596 //
2597 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2598 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2599 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2600 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2601 //
2602 // LS_cmp_exchange:
2603 //
2604 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2605 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2606 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2607 //
2608 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2609 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2610 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2611 //
2612 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2613 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2614 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2615 //
2616 // LS_get_add:
2617 //
2618 // int getAndAddInt( Object o, long offset, int delta)
2619 // long getAndAddLong(Object o, long offset, long delta)
2620 //
2621 // LS_get_set:
2622 //
2623 // int getAndSet(Object o, long offset, int newValue)
2624 // long getAndSet(Object o, long offset, long newValue)
2625 // Object getAndSet(Object o, long offset, Object newValue)
2626 //
2627 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2628 // This basic scheme here is the same as inline_unsafe_access, but
2629 // differs in enough details that combining them would make the code
2630 // overly confusing. (This is a true fact! I originally combined
2631 // them, but even I was confused by it!) As much code/comments as
2632 // possible are retained from inline_unsafe_access though to make
2633 // the correspondences clearer. - dl
2634
2635 if (callee()->is_static()) return false; // caller must have the capability!
2636
2637 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2638 decorators |= mo_decorator_for_access_kind(access_kind);
2639
2640 #ifndef PRODUCT
2641 BasicType rtype;
2642 {
2643 ResourceMark rm;
2644 // Check the signatures.
2645 ciSignature* sig = callee()->signature();
2646 rtype = sig->return_type()->basic_type();
2647 switch(kind) {
2648 case LS_get_add:
2649 case LS_get_set: {
2650 // Check the signatures.
2651 #ifdef ASSERT
2652 assert(rtype == type, "get and set must return the expected type");
2653 assert(sig->count() == 3, "get and set has 3 arguments");
2654 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2655 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2656 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2657 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2658 #endif // ASSERT
2659 break;
2660 }
2661 case LS_cmp_swap:
2662 case LS_cmp_swap_weak: {
2663 // Check the signatures.
2664 #ifdef ASSERT
2665 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2666 assert(sig->count() == 4, "CAS has 4 arguments");
2667 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2668 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2669 #endif // ASSERT
2670 break;
2671 }
2672 case LS_cmp_exchange: {
2673 // Check the signatures.
2674 #ifdef ASSERT
2675 assert(rtype == type, "CAS must return the expected type");
2676 assert(sig->count() == 4, "CAS has 4 arguments");
2677 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2678 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2679 #endif // ASSERT
2680 break;
2681 }
2682 default:
2683 ShouldNotReachHere();
2684 }
2685 }
2686 #endif //PRODUCT
2687
2688 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2689
2690 // Get arguments:
2691 Node* receiver = NULL;
2692 Node* base = NULL;
2693 Node* offset = NULL;
2694 Node* oldval = NULL;
2695 Node* newval = NULL;
2696 switch(kind) {
2697 case LS_cmp_swap:
2698 case LS_cmp_swap_weak:
2699 case LS_cmp_exchange: {
2700 const bool two_slot_type = type2size[type] == 2;
2701 receiver = argument(0); // type: oop
2702 base = argument(1); // type: oop
2703 offset = argument(2); // type: long
2704 oldval = argument(4); // type: oop, int, or long
2705 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2706 break;
2707 }
2708 case LS_get_add:
2709 case LS_get_set: {
2710 receiver = argument(0); // type: oop
2711 base = argument(1); // type: oop
2712 offset = argument(2); // type: long
2713 oldval = NULL;
2714 newval = argument(4); // type: oop, int, or long
2715 break;
2716 }
2717 default:
2718 ShouldNotReachHere();
2719 }
2720
2721 // Build field offset expression.
2722 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2723 // to be plain byte offsets, which are also the same as those accepted
2724 // by oopDesc::field_addr.
2725 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2726 // 32-bit machines ignore the high half of long offsets
2727 offset = ConvL2X(offset);
2728 Node* adr = make_unsafe_address(base, offset, ACCESS_WRITE | ACCESS_READ, type, false);
2729 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2730
2731 Compile::AliasType* alias_type = C->alias_type(adr_type);
2732 BasicType bt = alias_type->basic_type();
2733 if (bt != T_ILLEGAL &&
2734 (is_reference_type(bt) != (type == T_OBJECT))) {
2735 // Don't intrinsify mismatched object accesses.
2736 return false;
2737 }
2738
2739 // For CAS, unlike inline_unsafe_access, there seems no point in
2740 // trying to refine types. Just use the coarse types here.
2741 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2742 const Type *value_type = Type::get_const_basic_type(type);
2743
2744 switch (kind) {
2745 case LS_get_set:
2746 case LS_cmp_exchange: {
2747 if (type == T_OBJECT) {
2748 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2749 if (tjp != NULL) {
2750 value_type = tjp;
2751 }
2752 }
2753 break;
2754 }
2755 case LS_cmp_swap:
2756 case LS_cmp_swap_weak:
2757 case LS_get_add:
2758 break;
2759 default:
2760 ShouldNotReachHere();
2761 }
2762
2763 // Null check receiver.
2764 receiver = null_check(receiver);
2765 if (stopped()) {
2766 return true;
2767 }
2768
2769 int alias_idx = C->get_alias_index(adr_type);
2770
2771 if (is_reference_type(type)) {
2772 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2773
2774 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2775 // could be delayed during Parse (for example, in adjust_map_after_if()).
2776 // Execute transformation here to avoid barrier generation in such case.
2777 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2778 newval = _gvn.makecon(TypePtr::NULL_PTR);
2779
2780 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2781 // Refine the value to a null constant, when it is known to be null
2782 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2783 }
2784 }
2785
2786 Node* result = NULL;
2787 switch (kind) {
2788 case LS_cmp_exchange: {
2789 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2790 oldval, newval, value_type, type, decorators);
2791 break;
2792 }
2793 case LS_cmp_swap_weak:
2794 decorators |= C2_WEAK_CMPXCHG;
2795 case LS_cmp_swap: {
2796 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2797 oldval, newval, value_type, type, decorators);
2798 break;
2799 }
2800 case LS_get_set: {
2801 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2802 newval, value_type, type, decorators);
2803 break;
2804 }
2805 case LS_get_add: {
2806 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2807 newval, value_type, type, decorators);
2808 break;
2809 }
2810 default:
2811 ShouldNotReachHere();
2812 }
2813
2814 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2815 set_result(result);
2816 return true;
2817 }
2818
2819 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2820 // Regardless of form, don't allow previous ld/st to move down,
2821 // then issue acquire, release, or volatile mem_bar.
2822 insert_mem_bar(Op_MemBarCPUOrder);
2823 switch(id) {
2824 case vmIntrinsics::_loadFence:
2825 insert_mem_bar(Op_LoadFence);
2826 return true;
2827 case vmIntrinsics::_storeFence:
2828 insert_mem_bar(Op_StoreFence);
2829 return true;
2830 case vmIntrinsics::_fullFence:
2831 insert_mem_bar(Op_MemBarVolatile);
2832 return true;
2833 default:
2834 fatal_unexpected_iid(id);
2835 return false;
2836 }
2837 }
2838
2839 bool LibraryCallKit::inline_onspinwait() {
2840 insert_mem_bar(Op_OnSpinWait);
2841 return true;
2842 }
2843
2844 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2845 if (!kls->is_Con()) {
2846 return true;
2847 }
2848 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
2849 if (klsptr == NULL) {
2850 return true;
2851 }
2852 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
2853 // don't need a guard for a klass that is already initialized
2854 return !ik->is_initialized();
2855 }
2856
2857 //----------------------------inline_unsafe_writeback0-------------------------
2858 // public native void Unsafe.writeback0(long address)
2859 bool LibraryCallKit::inline_unsafe_writeback0() {
2860 if (!Matcher::has_match_rule(Op_CacheWB)) {
2861 return false;
2862 }
2863 #ifndef PRODUCT
2864 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2865 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2866 ciSignature* sig = callee()->signature();
2867 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2868 #endif
2869 null_check_receiver(); // null-check, then ignore
2870 Node *addr = argument(1);
2871 addr = new CastX2PNode(addr);
2872 addr = _gvn.transform(addr);
2873 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2874 flush = _gvn.transform(flush);
2875 set_memory(flush, TypeRawPtr::BOTTOM);
2876 return true;
2877 }
2878
2879 //----------------------------inline_unsafe_writeback0-------------------------
2880 // public native void Unsafe.writeback0(long address)
2881 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2882 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2883 return false;
2884 }
2885 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2886 return false;
2887 }
2888 #ifndef PRODUCT
2889 assert(Matcher::has_match_rule(Op_CacheWB),
2890 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2891 : "found match rule for CacheWBPostSync but not CacheWB"));
2892
2893 #endif
2894 null_check_receiver(); // null-check, then ignore
2895 Node *sync;
2896 if (is_pre) {
2897 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2898 } else {
2899 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2900 }
2901 sync = _gvn.transform(sync);
2902 set_memory(sync, TypeRawPtr::BOTTOM);
2903 return true;
2904 }
2905
2906 //----------------------------inline_unsafe_allocate---------------------------
2907 // public native Object Unsafe.allocateInstance(Class<?> cls);
2908 bool LibraryCallKit::inline_unsafe_allocate() {
2909 if (callee()->is_static()) return false; // caller must have the capability!
2910
2911 null_check_receiver(); // null-check, then ignore
2912 Node* cls = null_check(argument(1));
2913 if (stopped()) return true;
2914
2915 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2916 kls = null_check(kls);
2917 if (stopped()) return true; // argument was like int.class
2918
2919 Node* test = NULL;
2920 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2921 // Note: The argument might still be an illegal value like
2922 // Serializable.class or Object[].class. The runtime will handle it.
2923 // But we must make an explicit check for initialization.
2924 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2925 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2926 // can generate code to load it as unsigned byte.
2927 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2928 Node* bits = intcon(InstanceKlass::fully_initialized);
2929 test = _gvn.transform(new SubINode(inst, bits));
2930 // The 'test' is non-zero if we need to take a slow path.
2931 }
2932
2933 Node* obj = new_instance(kls, test);
2934 set_result(obj);
2935 return true;
2936 }
2937
2938 //------------------------inline_native_time_funcs--------------
2939 // inline code for System.currentTimeMillis() and System.nanoTime()
2940 // these have the same type and signature
2941 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2942 const TypeFunc* tf = OptoRuntime::void_long_Type();
2943 const TypePtr* no_memory_effects = NULL;
2944 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2945 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2946 #ifdef ASSERT
2947 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2948 assert(value_top == top(), "second value must be top");
2949 #endif
2950 set_result(value);
2951 return true;
2952 }
2953
2954 #ifdef JFR_HAVE_INTRINSICS
2955
2956 /*
2957 * oop -> myklass
2958 * myklass->trace_id |= USED
2959 * return myklass->trace_id & ~0x3
2960 */
2961 bool LibraryCallKit::inline_native_classID() {
2962 Node* cls = null_check(argument(0), T_OBJECT);
2963 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2964 kls = null_check(kls, T_OBJECT);
2965
2966 ByteSize offset = KLASS_TRACE_ID_OFFSET;
2967 Node* insp = basic_plus_adr(kls, in_bytes(offset));
2968 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
2969
2970 Node* clsused = longcon(0x01l); // set the class bit
2971 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
2972 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
2973 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
2974
2975 #ifdef TRACE_ID_META_BITS
2976 Node* mbits = longcon(~TRACE_ID_META_BITS);
2977 tvalue = _gvn.transform(new AndLNode(tvalue, mbits));
2978 #endif
2979 #ifdef TRACE_ID_SHIFT
2980 Node* cbits = intcon(TRACE_ID_SHIFT);
2981 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits));
2982 #endif
2983
2984 set_result(tvalue);
2985 return true;
2986
2987 }
2988
2989 bool LibraryCallKit::inline_native_getEventWriter() {
2990 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
2991
2992 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
2993 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
2994
2995 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
2996
2997 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
2998 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
2999
3000 IfNode* iff_jobj_null =
3001 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
3002
3003 enum { _normal_path = 1,
3004 _null_path = 2,
3005 PATH_LIMIT };
3006
3007 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3008 PhiNode* result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3009
3010 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
3011 result_rgn->init_req(_null_path, jobj_is_null);
3012 result_val->init_req(_null_path, null());
3013
3014 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
3015 set_control(jobj_is_not_null);
3016 Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT,
3017 IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3018 result_rgn->init_req(_normal_path, control());
3019 result_val->init_req(_normal_path, res);
3020
3021 set_result(result_rgn, result_val);
3022
3023 return true;
3024 }
3025
3026 #endif // JFR_HAVE_INTRINSICS
3027
3028 //------------------------inline_native_currentThread------------------
3029 bool LibraryCallKit::inline_native_currentThread() {
3030 Node* junk = NULL;
3031 set_result(generate_current_thread(junk));
3032 return true;
3033 }
3034
3035 //---------------------------load_mirror_from_klass----------------------------
3036 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3037 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3038 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3039 Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3040 // mirror = ((OopHandle)mirror)->resolve();
3041 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3042 }
3043
3044 //-----------------------load_klass_from_mirror_common-------------------------
3045 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3046 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3047 // and branch to the given path on the region.
3048 // If never_see_null, take an uncommon trap on null, so we can optimistically
3049 // compile for the non-null case.
3050 // If the region is NULL, force never_see_null = true.
3051 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3052 bool never_see_null,
3053 RegionNode* region,
3054 int null_path,
3055 int offset) {
3056 if (region == NULL) never_see_null = true;
3057 Node* p = basic_plus_adr(mirror, offset);
3058 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3059 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3060 Node* null_ctl = top();
3061 kls = null_check_oop(kls, &null_ctl, never_see_null);
3062 if (region != NULL) {
3063 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3064 region->init_req(null_path, null_ctl);
3065 } else {
3066 assert(null_ctl == top(), "no loose ends");
3067 }
3068 return kls;
3069 }
3070
3071 //--------------------(inline_native_Class_query helpers)---------------------
3072 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3073 // Fall through if (mods & mask) == bits, take the guard otherwise.
3074 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3075 // Branch around if the given klass has the given modifier bit set.
3076 // Like generate_guard, adds a new path onto the region.
3077 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3078 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3079 Node* mask = intcon(modifier_mask);
3080 Node* bits = intcon(modifier_bits);
3081 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3082 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3083 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3084 return generate_fair_guard(bol, region);
3085 }
3086 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3087 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3088 }
3089 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3090 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
3091 }
3092
3093 //-------------------------inline_native_Class_query-------------------
3094 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3095 const Type* return_type = TypeInt::BOOL;
3096 Node* prim_return_value = top(); // what happens if it's a primitive class?
3097 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3098 bool expect_prim = false; // most of these guys expect to work on refs
3099
3100 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3101
3102 Node* mirror = argument(0);
3103 Node* obj = top();
3104
3105 switch (id) {
3106 case vmIntrinsics::_isInstance:
3107 // nothing is an instance of a primitive type
3108 prim_return_value = intcon(0);
3109 obj = argument(1);
3110 break;
3111 case vmIntrinsics::_getModifiers:
3112 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3113 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3114 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3115 break;
3116 case vmIntrinsics::_isInterface:
3117 prim_return_value = intcon(0);
3118 break;
3119 case vmIntrinsics::_isArray:
3120 prim_return_value = intcon(0);
3121 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3122 break;
3123 case vmIntrinsics::_isPrimitive:
3124 prim_return_value = intcon(1);
3125 expect_prim = true; // obviously
3126 break;
3127 case vmIntrinsics::_isHiddenClass:
3128 prim_return_value = intcon(0);
3129 break;
3130 case vmIntrinsics::_getSuperclass:
3131 prim_return_value = null();
3132 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3133 break;
3134 case vmIntrinsics::_getClassAccessFlags:
3135 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3136 return_type = TypeInt::INT; // not bool! 6297094
3137 break;
3138 default:
3139 fatal_unexpected_iid(id);
3140 break;
3141 }
3142
3143 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3144 if (mirror_con == NULL) return false; // cannot happen?
3145
3146 #ifndef PRODUCT
3147 if (C->print_intrinsics() || C->print_inlining()) {
3148 ciType* k = mirror_con->java_mirror_type();
3149 if (k) {
3150 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3151 k->print_name();
3152 tty->cr();
3153 }
3154 }
3155 #endif
3156
3157 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3158 RegionNode* region = new RegionNode(PATH_LIMIT);
3159 record_for_igvn(region);
3160 PhiNode* phi = new PhiNode(region, return_type);
3161
3162 // The mirror will never be null of Reflection.getClassAccessFlags, however
3163 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3164 // if it is. See bug 4774291.
3165
3166 // For Reflection.getClassAccessFlags(), the null check occurs in
3167 // the wrong place; see inline_unsafe_access(), above, for a similar
3168 // situation.
3169 mirror = null_check(mirror);
3170 // If mirror or obj is dead, only null-path is taken.
3171 if (stopped()) return true;
3172
3173 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3174
3175 // Now load the mirror's klass metaobject, and null-check it.
3176 // Side-effects region with the control path if the klass is null.
3177 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3178 // If kls is null, we have a primitive mirror.
3179 phi->init_req(_prim_path, prim_return_value);
3180 if (stopped()) { set_result(region, phi); return true; }
3181 bool safe_for_replace = (region->in(_prim_path) == top());
3182
3183 Node* p; // handy temp
3184 Node* null_ctl;
3185
3186 // Now that we have the non-null klass, we can perform the real query.
3187 // For constant classes, the query will constant-fold in LoadNode::Value.
3188 Node* query_value = top();
3189 switch (id) {
3190 case vmIntrinsics::_isInstance:
3191 // nothing is an instance of a primitive type
3192 query_value = gen_instanceof(obj, kls, safe_for_replace);
3193 break;
3194
3195 case vmIntrinsics::_getModifiers:
3196 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3197 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3198 break;
3199
3200 case vmIntrinsics::_isInterface:
3201 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3202 if (generate_interface_guard(kls, region) != NULL)
3203 // A guard was added. If the guard is taken, it was an interface.
3204 phi->add_req(intcon(1));
3205 // If we fall through, it's a plain class.
3206 query_value = intcon(0);
3207 break;
3208
3209 case vmIntrinsics::_isArray:
3210 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3211 if (generate_array_guard(kls, region) != NULL)
3212 // A guard was added. If the guard is taken, it was an array.
3213 phi->add_req(intcon(1));
3214 // If we fall through, it's a plain class.
3215 query_value = intcon(0);
3216 break;
3217
3218 case vmIntrinsics::_isPrimitive:
3219 query_value = intcon(0); // "normal" path produces false
3220 break;
3221
3222 case vmIntrinsics::_isHiddenClass:
3223 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
3224 if (generate_hidden_class_guard(kls, region) != NULL)
3225 // A guard was added. If the guard is taken, it was an hidden class.
3226 phi->add_req(intcon(1));
3227 // If we fall through, it's a plain class.
3228 query_value = intcon(0);
3229 break;
3230
3231
3232 case vmIntrinsics::_getSuperclass:
3233 // The rules here are somewhat unfortunate, but we can still do better
3234 // with random logic than with a JNI call.
3235 // Interfaces store null or Object as _super, but must report null.
3236 // Arrays store an intermediate super as _super, but must report Object.
3237 // Other types can report the actual _super.
3238 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3239 if (generate_interface_guard(kls, region) != NULL)
3240 // A guard was added. If the guard is taken, it was an interface.
3241 phi->add_req(null());
3242 if (generate_array_guard(kls, region) != NULL)
3243 // A guard was added. If the guard is taken, it was an array.
3244 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3245 // If we fall through, it's a plain class. Get its _super.
3246 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3247 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3248 null_ctl = top();
3249 kls = null_check_oop(kls, &null_ctl);
3250 if (null_ctl != top()) {
3251 // If the guard is taken, Object.superClass is null (both klass and mirror).
3252 region->add_req(null_ctl);
3253 phi ->add_req(null());
3254 }
3255 if (!stopped()) {
3256 query_value = load_mirror_from_klass(kls);
3257 }
3258 break;
3259
3260 case vmIntrinsics::_getClassAccessFlags:
3261 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3262 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3263 break;
3264
3265 default:
3266 fatal_unexpected_iid(id);
3267 break;
3268 }
3269
3270 // Fall-through is the normal case of a query to a real class.
3271 phi->init_req(1, query_value);
3272 region->init_req(1, control());
3273
3274 C->set_has_split_ifs(true); // Has chance for split-if optimization
3275 set_result(region, phi);
3276 return true;
3277 }
3278
3279 //-------------------------inline_Class_cast-------------------
3280 bool LibraryCallKit::inline_Class_cast() {
3281 Node* mirror = argument(0); // Class
3282 Node* obj = argument(1);
3283 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3284 if (mirror_con == NULL) {
3285 return false; // dead path (mirror->is_top()).
3286 }
3287 if (obj == NULL || obj->is_top()) {
3288 return false; // dead path
3289 }
3290 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3291
3292 // First, see if Class.cast() can be folded statically.
3293 // java_mirror_type() returns non-null for compile-time Class constants.
3294 ciType* tm = mirror_con->java_mirror_type();
3295 if (tm != NULL && tm->is_klass() &&
3296 tp != NULL && tp->klass() != NULL) {
3297 if (!tp->klass()->is_loaded()) {
3298 // Don't use intrinsic when class is not loaded.
3299 return false;
3300 } else {
3301 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3302 if (static_res == Compile::SSC_always_true) {
3303 // isInstance() is true - fold the code.
3304 set_result(obj);
3305 return true;
3306 } else if (static_res == Compile::SSC_always_false) {
3307 // Don't use intrinsic, have to throw ClassCastException.
3308 // If the reference is null, the non-intrinsic bytecode will
3309 // be optimized appropriately.
3310 return false;
3311 }
3312 }
3313 }
3314
3315 // Bailout intrinsic and do normal inlining if exception path is frequent.
3316 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3317 return false;
3318 }
3319
3320 // Generate dynamic checks.
3321 // Class.cast() is java implementation of _checkcast bytecode.
3322 // Do checkcast (Parse::do_checkcast()) optimizations here.
3323
3324 mirror = null_check(mirror);
3325 // If mirror is dead, only null-path is taken.
3326 if (stopped()) {
3327 return true;
3328 }
3329
3330 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3331 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3332 RegionNode* region = new RegionNode(PATH_LIMIT);
3333 record_for_igvn(region);
3334
3335 // Now load the mirror's klass metaobject, and null-check it.
3336 // If kls is null, we have a primitive mirror and
3337 // nothing is an instance of a primitive type.
3338 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3339
3340 Node* res = top();
3341 if (!stopped()) {
3342 Node* bad_type_ctrl = top();
3343 // Do checkcast optimizations.
3344 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3345 region->init_req(_bad_type_path, bad_type_ctrl);
3346 }
3347 if (region->in(_prim_path) != top() ||
3348 region->in(_bad_type_path) != top()) {
3349 // Let Interpreter throw ClassCastException.
3350 PreserveJVMState pjvms(this);
3351 set_control(_gvn.transform(region));
3352 uncommon_trap(Deoptimization::Reason_intrinsic,
3353 Deoptimization::Action_maybe_recompile);
3354 }
3355 if (!stopped()) {
3356 set_result(res);
3357 }
3358 return true;
3359 }
3360
3361
3362 //--------------------------inline_native_subtype_check------------------------
3363 // This intrinsic takes the JNI calls out of the heart of
3364 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3365 bool LibraryCallKit::inline_native_subtype_check() {
3366 // Pull both arguments off the stack.
3367 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3368 args[0] = argument(0);
3369 args[1] = argument(1);
3370 Node* klasses[2]; // corresponding Klasses: superk, subk
3371 klasses[0] = klasses[1] = top();
3372
3373 enum {
3374 // A full decision tree on {superc is prim, subc is prim}:
3375 _prim_0_path = 1, // {P,N} => false
3376 // {P,P} & superc!=subc => false
3377 _prim_same_path, // {P,P} & superc==subc => true
3378 _prim_1_path, // {N,P} => false
3379 _ref_subtype_path, // {N,N} & subtype check wins => true
3380 _both_ref_path, // {N,N} & subtype check loses => false
3381 PATH_LIMIT
3382 };
3383
3384 RegionNode* region = new RegionNode(PATH_LIMIT);
3385 Node* phi = new PhiNode(region, TypeInt::BOOL);
3386 record_for_igvn(region);
3387
3388 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3389 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3390 int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3391
3392 // First null-check both mirrors and load each mirror's klass metaobject.
3393 int which_arg;
3394 for (which_arg = 0; which_arg <= 1; which_arg++) {
3395 Node* arg = args[which_arg];
3396 arg = null_check(arg);
3397 if (stopped()) break;
3398 args[which_arg] = arg;
3399
3400 Node* p = basic_plus_adr(arg, class_klass_offset);
3401 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3402 klasses[which_arg] = _gvn.transform(kls);
3403 }
3404
3405 // Having loaded both klasses, test each for null.
3406 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3407 for (which_arg = 0; which_arg <= 1; which_arg++) {
3408 Node* kls = klasses[which_arg];
3409 Node* null_ctl = top();
3410 kls = null_check_oop(kls, &null_ctl, never_see_null);
3411 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3412 region->init_req(prim_path, null_ctl);
3413 if (stopped()) break;
3414 klasses[which_arg] = kls;
3415 }
3416
3417 if (!stopped()) {
3418 // now we have two reference types, in klasses[0..1]
3419 Node* subk = klasses[1]; // the argument to isAssignableFrom
3420 Node* superk = klasses[0]; // the receiver
3421 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3422 // now we have a successful reference subtype check
3423 region->set_req(_ref_subtype_path, control());
3424 }
3425
3426 // If both operands are primitive (both klasses null), then
3427 // we must return true when they are identical primitives.
3428 // It is convenient to test this after the first null klass check.
3429 set_control(region->in(_prim_0_path)); // go back to first null check
3430 if (!stopped()) {
3431 // Since superc is primitive, make a guard for the superc==subc case.
3432 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3433 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3434 generate_guard(bol_eq, region, PROB_FAIR);
3435 if (region->req() == PATH_LIMIT+1) {
3436 // A guard was added. If the added guard is taken, superc==subc.
3437 region->swap_edges(PATH_LIMIT, _prim_same_path);
3438 region->del_req(PATH_LIMIT);
3439 }
3440 region->set_req(_prim_0_path, control()); // Not equal after all.
3441 }
3442
3443 // these are the only paths that produce 'true':
3444 phi->set_req(_prim_same_path, intcon(1));
3445 phi->set_req(_ref_subtype_path, intcon(1));
3446
3447 // pull together the cases:
3448 assert(region->req() == PATH_LIMIT, "sane region");
3449 for (uint i = 1; i < region->req(); i++) {
3450 Node* ctl = region->in(i);
3451 if (ctl == NULL || ctl == top()) {
3452 region->set_req(i, top());
3453 phi ->set_req(i, top());
3454 } else if (phi->in(i) == NULL) {
3455 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3456 }
3457 }
3458
3459 set_control(_gvn.transform(region));
3460 set_result(_gvn.transform(phi));
3461 return true;
3462 }
3463
3464 //---------------------generate_array_guard_common------------------------
3465 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3466 bool obj_array, bool not_array) {
3467
3468 if (stopped()) {
3469 return NULL;
3470 }
3471
3472 // If obj_array/non_array==false/false:
3473 // Branch around if the given klass is in fact an array (either obj or prim).
3474 // If obj_array/non_array==false/true:
3475 // Branch around if the given klass is not an array klass of any kind.
3476 // If obj_array/non_array==true/true:
3477 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3478 // If obj_array/non_array==true/false:
3479 // Branch around if the kls is an oop array (Object[] or subtype)
3480 //
3481 // Like generate_guard, adds a new path onto the region.
3482 jint layout_con = 0;
3483 Node* layout_val = get_layout_helper(kls, layout_con);
3484 if (layout_val == NULL) {
3485 bool query = (obj_array
3486 ? Klass::layout_helper_is_objArray(layout_con)
3487 : Klass::layout_helper_is_array(layout_con));
3488 if (query == not_array) {
3489 return NULL; // never a branch
3490 } else { // always a branch
3491 Node* always_branch = control();
3492 if (region != NULL)
3493 region->add_req(always_branch);
3494 set_control(top());
3495 return always_branch;
3496 }
3497 }
3498 // Now test the correct condition.
3499 jint nval = (obj_array
3500 ? (jint)(Klass::_lh_array_tag_type_value
3501 << Klass::_lh_array_tag_shift)
3502 : Klass::_lh_neutral_value);
3503 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3504 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3505 // invert the test if we are looking for a non-array
3506 if (not_array) btest = BoolTest(btest).negate();
3507 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3508 return generate_fair_guard(bol, region);
3509 }
3510
3511
3512 //-----------------------inline_native_newArray--------------------------
3513 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3514 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3515 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3516 Node* mirror;
3517 Node* count_val;
3518 if (uninitialized) {
3519 mirror = argument(1);
3520 count_val = argument(2);
3521 } else {
3522 mirror = argument(0);
3523 count_val = argument(1);
3524 }
3525
3526 mirror = null_check(mirror);
3527 // If mirror or obj is dead, only null-path is taken.
3528 if (stopped()) return true;
3529
3530 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3531 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3532 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3533 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3534 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3535
3536 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3537 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3538 result_reg, _slow_path);
3539 Node* normal_ctl = control();
3540 Node* no_array_ctl = result_reg->in(_slow_path);
3541
3542 // Generate code for the slow case. We make a call to newArray().
3543 set_control(no_array_ctl);
3544 if (!stopped()) {
3545 // Either the input type is void.class, or else the
3546 // array klass has not yet been cached. Either the
3547 // ensuing call will throw an exception, or else it
3548 // will cache the array klass for next time.
3549 PreserveJVMState pjvms(this);
3550 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3551 Node* slow_result = set_results_for_java_call(slow_call);
3552 // this->control() comes from set_results_for_java_call
3553 result_reg->set_req(_slow_path, control());
3554 result_val->set_req(_slow_path, slow_result);
3555 result_io ->set_req(_slow_path, i_o());
3556 result_mem->set_req(_slow_path, reset_memory());
3557 }
3558
3559 set_control(normal_ctl);
3560 if (!stopped()) {
3561 // Normal case: The array type has been cached in the java.lang.Class.
3562 // The following call works fine even if the array type is polymorphic.
3563 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3564 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3565 result_reg->init_req(_normal_path, control());
3566 result_val->init_req(_normal_path, obj);
3567 result_io ->init_req(_normal_path, i_o());
3568 result_mem->init_req(_normal_path, reset_memory());
3569
3570 if (uninitialized) {
3571 // Mark the allocation so that zeroing is skipped
3572 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3573 alloc->maybe_set_complete(&_gvn);
3574 }
3575 }
3576
3577 // Return the combined state.
3578 set_i_o( _gvn.transform(result_io) );
3579 set_all_memory( _gvn.transform(result_mem));
3580
3581 C->set_has_split_ifs(true); // Has chance for split-if optimization
3582 set_result(result_reg, result_val);
3583 return true;
3584 }
3585
3586 //----------------------inline_native_getLength--------------------------
3587 // public static native int java.lang.reflect.Array.getLength(Object array);
3588 bool LibraryCallKit::inline_native_getLength() {
3589 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3590
3591 Node* array = null_check(argument(0));
3592 // If array is dead, only null-path is taken.
3593 if (stopped()) return true;
3594
3595 // Deoptimize if it is a non-array.
3596 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3597
3598 if (non_array != NULL) {
3599 PreserveJVMState pjvms(this);
3600 set_control(non_array);
3601 uncommon_trap(Deoptimization::Reason_intrinsic,
3602 Deoptimization::Action_maybe_recompile);
3603 }
3604
3605 // If control is dead, only non-array-path is taken.
3606 if (stopped()) return true;
3607
3608 // The works fine even if the array type is polymorphic.
3609 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3610 Node* result = load_array_length(array);
3611
3612 C->set_has_split_ifs(true); // Has chance for split-if optimization
3613 set_result(result);
3614 return true;
3615 }
3616
3617 //------------------------inline_array_copyOf----------------------------
3618 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3619 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3620 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3621 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3622
3623 // Get the arguments.
3624 Node* original = argument(0);
3625 Node* start = is_copyOfRange? argument(1): intcon(0);
3626 Node* end = is_copyOfRange? argument(2): argument(1);
3627 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3628
3629 Node* newcopy = NULL;
3630
3631 // Set the original stack and the reexecute bit for the interpreter to reexecute
3632 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3633 { PreserveReexecuteState preexecs(this);
3634 jvms()->set_should_reexecute(true);
3635
3636 array_type_mirror = null_check(array_type_mirror);
3637 original = null_check(original);
3638
3639 // Check if a null path was taken unconditionally.
3640 if (stopped()) return true;
3641
3642 Node* orig_length = load_array_length(original);
3643
3644 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3645 klass_node = null_check(klass_node);
3646
3647 RegionNode* bailout = new RegionNode(1);
3648 record_for_igvn(bailout);
3649
3650 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3651 // Bail out if that is so.
3652 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3653 if (not_objArray != NULL) {
3654 // Improve the klass node's type from the new optimistic assumption:
3655 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3656 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3657 Node* cast = new CastPPNode(klass_node, akls);
3658 cast->init_req(0, control());
3659 klass_node = _gvn.transform(cast);
3660 }
3661
3662 // Bail out if either start or end is negative.
3663 generate_negative_guard(start, bailout, &start);
3664 generate_negative_guard(end, bailout, &end);
3665
3666 Node* length = end;
3667 if (_gvn.type(start) != TypeInt::ZERO) {
3668 length = _gvn.transform(new SubINode(end, start));
3669 }
3670
3671 // Bail out if length is negative.
3672 // Without this the new_array would throw
3673 // NegativeArraySizeException but IllegalArgumentException is what
3674 // should be thrown
3675 generate_negative_guard(length, bailout, &length);
3676
3677 if (bailout->req() > 1) {
3678 PreserveJVMState pjvms(this);
3679 set_control(_gvn.transform(bailout));
3680 uncommon_trap(Deoptimization::Reason_intrinsic,
3681 Deoptimization::Action_maybe_recompile);
3682 }
3683
3684 if (!stopped()) {
3685 // How many elements will we copy from the original?
3686 // The answer is MinI(orig_length - start, length).
3687 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3688 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3689
3690 // Generate a direct call to the right arraycopy function(s).
3691 // We know the copy is disjoint but we might not know if the
3692 // oop stores need checking.
3693 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3694 // This will fail a store-check if x contains any non-nulls.
3695
3696 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3697 // loads/stores but it is legal only if we're sure the
3698 // Arrays.copyOf would succeed. So we need all input arguments
3699 // to the copyOf to be validated, including that the copy to the
3700 // new array won't trigger an ArrayStoreException. That subtype
3701 // check can be optimized if we know something on the type of
3702 // the input array from type speculation.
3703 if (_gvn.type(klass_node)->singleton()) {
3704 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3705 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3706
3707 int test = C->static_subtype_check(superk, subk);
3708 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3709 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3710 if (t_original->speculative_type() != NULL) {
3711 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3712 }
3713 }
3714 }
3715
3716 bool validated = false;
3717 // Reason_class_check rather than Reason_intrinsic because we
3718 // want to intrinsify even if this traps.
3719 if (!too_many_traps(Deoptimization::Reason_class_check)) {
3720 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
3721
3722 if (not_subtype_ctrl != top()) {
3723 PreserveJVMState pjvms(this);
3724 set_control(not_subtype_ctrl);
3725 uncommon_trap(Deoptimization::Reason_class_check,
3726 Deoptimization::Action_make_not_entrant);
3727 assert(stopped(), "Should be stopped");
3728 }
3729 validated = true;
3730 }
3731
3732 if (!stopped()) {
3733 newcopy = new_array(klass_node, length, 0); // no arguments to push
3734
3735 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
3736 load_object_klass(original), klass_node);
3737 if (!is_copyOfRange) {
3738 ac->set_copyof(validated);
3739 } else {
3740 ac->set_copyofrange(validated);
3741 }
3742 Node* n = _gvn.transform(ac);
3743 if (n == ac) {
3744 ac->connect_outputs(this);
3745 } else {
3746 assert(validated, "shouldn't transform if all arguments not validated");
3747 set_all_memory(n);
3748 }
3749 }
3750 }
3751 } // original reexecute is set back here
3752
3753 C->set_has_split_ifs(true); // Has chance for split-if optimization
3754 if (!stopped()) {
3755 set_result(newcopy);
3756 }
3757 return true;
3758 }
3759
3760
3761 //----------------------generate_virtual_guard---------------------------
3762 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
3763 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
3764 RegionNode* slow_region) {
3765 ciMethod* method = callee();
3766 int vtable_index = method->vtable_index();
3767 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3768 "bad index %d", vtable_index);
3769 // Get the Method* out of the appropriate vtable entry.
3770 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
3771 vtable_index*vtableEntry::size_in_bytes() +
3772 vtableEntry::method_offset_in_bytes();
3773 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
3774 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3775
3776 // Compare the target method with the expected method (e.g., Object.hashCode).
3777 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
3778
3779 Node* native_call = makecon(native_call_addr);
3780 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
3781 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
3782
3783 return generate_slow_guard(test_native, slow_region);
3784 }
3785
3786 //-----------------------generate_method_call----------------------------
3787 // Use generate_method_call to make a slow-call to the real
3788 // method if the fast path fails. An alternative would be to
3789 // use a stub like OptoRuntime::slow_arraycopy_Java.
3790 // This only works for expanding the current library call,
3791 // not another intrinsic. (E.g., don't use this for making an
3792 // arraycopy call inside of the copyOf intrinsic.)
3793 CallJavaNode*
3794 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
3795 // When compiling the intrinsic method itself, do not use this technique.
3796 guarantee(callee() != C->method(), "cannot make slow-call to self");
3797
3798 ciMethod* method = callee();
3799 // ensure the JVMS we have will be correct for this call
3800 guarantee(method_id == method->intrinsic_id(), "must match");
3801
3802 const TypeFunc* tf = TypeFunc::make(method);
3803 CallJavaNode* slow_call;
3804 if (is_static) {
3805 assert(!is_virtual, "");
3806 slow_call = new CallStaticJavaNode(C, tf,
3807 SharedRuntime::get_resolve_static_call_stub(),
3808 method, bci());
3809 } else if (is_virtual) {
3810 null_check_receiver();
3811 int vtable_index = Method::invalid_vtable_index;
3812 if (UseInlineCaches) {
3813 // Suppress the vtable call
3814 } else {
3815 // hashCode and clone are not a miranda methods,
3816 // so the vtable index is fixed.
3817 // No need to use the linkResolver to get it.
3818 vtable_index = method->vtable_index();
3819 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3820 "bad index %d", vtable_index);
3821 }
3822 slow_call = new CallDynamicJavaNode(tf,
3823 SharedRuntime::get_resolve_virtual_call_stub(),
3824 method, vtable_index, bci());
3825 } else { // neither virtual nor static: opt_virtual
3826 null_check_receiver();
3827 slow_call = new CallStaticJavaNode(C, tf,
3828 SharedRuntime::get_resolve_opt_virtual_call_stub(),
3829 method, bci());
3830 slow_call->set_optimized_virtual(true);
3831 }
3832 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
3833 // To be able to issue a direct call (optimized virtual or virtual)
3834 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
3835 // about the method being invoked should be attached to the call site to
3836 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
3837 slow_call->set_override_symbolic_info(true);
3838 }
3839 set_arguments_for_java_call(slow_call);
3840 set_edges_for_java_call(slow_call);
3841 return slow_call;
3842 }
3843
3844
3845 /**
3846 * Build special case code for calls to hashCode on an object. This call may
3847 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
3848 * slightly different code.
3849 */
3850 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
3851 assert(is_static == callee()->is_static(), "correct intrinsic selection");
3852 assert(!(is_virtual && is_static), "either virtual, special, or static");
3853
3854 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
3855
3856 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3857 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
3858 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3859 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3860 Node* obj = NULL;
3861 if (!is_static) {
3862 // Check for hashing null object
3863 obj = null_check_receiver();
3864 if (stopped()) return true; // unconditionally null
3865 result_reg->init_req(_null_path, top());
3866 result_val->init_req(_null_path, top());
3867 } else {
3868 // Do a null check, and return zero if null.
3869 // System.identityHashCode(null) == 0
3870 obj = argument(0);
3871 Node* null_ctl = top();
3872 obj = null_check_oop(obj, &null_ctl);
3873 result_reg->init_req(_null_path, null_ctl);
3874 result_val->init_req(_null_path, _gvn.intcon(0));
3875 }
3876
3877 // Unconditionally null? Then return right away.
3878 if (stopped()) {
3879 set_control( result_reg->in(_null_path));
3880 if (!stopped())
3881 set_result(result_val->in(_null_path));
3882 return true;
3883 }
3884
3885 // We only go to the fast case code if we pass a number of guards. The
3886 // paths which do not pass are accumulated in the slow_region.
3887 RegionNode* slow_region = new RegionNode(1);
3888 record_for_igvn(slow_region);
3889
3890 // If this is a virtual call, we generate a funny guard. We pull out
3891 // the vtable entry corresponding to hashCode() from the target object.
3892 // If the target method which we are calling happens to be the native
3893 // Object hashCode() method, we pass the guard. We do not need this
3894 // guard for non-virtual calls -- the caller is known to be the native
3895 // Object hashCode().
3896 if (is_virtual) {
3897 // After null check, get the object's klass.
3898 Node* obj_klass = load_object_klass(obj);
3899 generate_virtual_guard(obj_klass, slow_region);
3900 }
3901
3902 // Get the header out of the object, use LoadMarkNode when available
3903 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
3904 // The control of the load must be NULL. Otherwise, the load can move before
3905 // the null check after castPP removal.
3906 Node* no_ctrl = NULL;
3907 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
3908
3909 // Test the header to see if it is unlocked.
3910 Node *lock_mask = _gvn.MakeConX(markWord::biased_lock_mask_in_place);
3911 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
3912 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
3913 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
3914 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
3915
3916 generate_slow_guard(test_unlocked, slow_region);
3917
3918 // Get the hash value and check to see that it has been properly assigned.
3919 // We depend on hash_mask being at most 32 bits and avoid the use of
3920 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
3921 // vm: see markWord.hpp.
3922 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
3923 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
3924 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
3925 // This hack lets the hash bits live anywhere in the mark object now, as long
3926 // as the shift drops the relevant bits into the low 32 bits. Note that
3927 // Java spec says that HashCode is an int so there's no point in capturing
3928 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
3929 hshifted_header = ConvX2I(hshifted_header);
3930 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
3931
3932 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
3933 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
3934 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
3935
3936 generate_slow_guard(test_assigned, slow_region);
3937
3938 Node* init_mem = reset_memory();
3939 // fill in the rest of the null path:
3940 result_io ->init_req(_null_path, i_o());
3941 result_mem->init_req(_null_path, init_mem);
3942
3943 result_val->init_req(_fast_path, hash_val);
3944 result_reg->init_req(_fast_path, control());
3945 result_io ->init_req(_fast_path, i_o());
3946 result_mem->init_req(_fast_path, init_mem);
3947
3948 // Generate code for the slow case. We make a call to hashCode().
3949 set_control(_gvn.transform(slow_region));
3950 if (!stopped()) {
3951 // No need for PreserveJVMState, because we're using up the present state.
3952 set_all_memory(init_mem);
3953 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
3954 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
3955 Node* slow_result = set_results_for_java_call(slow_call);
3956 // this->control() comes from set_results_for_java_call
3957 result_reg->init_req(_slow_path, control());
3958 result_val->init_req(_slow_path, slow_result);
3959 result_io ->set_req(_slow_path, i_o());
3960 result_mem ->set_req(_slow_path, reset_memory());
3961 }
3962
3963 // Return the combined state.
3964 set_i_o( _gvn.transform(result_io) );
3965 set_all_memory( _gvn.transform(result_mem));
3966
3967 set_result(result_reg, result_val);
3968 return true;
3969 }
3970
3971 //---------------------------inline_native_getClass----------------------------
3972 // public final native Class<?> java.lang.Object.getClass();
3973 //
3974 // Build special case code for calls to getClass on an object.
3975 bool LibraryCallKit::inline_native_getClass() {
3976 Node* obj = null_check_receiver();
3977 if (stopped()) return true;
3978 set_result(load_mirror_from_klass(load_object_klass(obj)));
3979 return true;
3980 }
3981
3982 //-----------------inline_native_Reflection_getCallerClass---------------------
3983 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
3984 //
3985 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
3986 //
3987 // NOTE: This code must perform the same logic as JVM_GetCallerClass
3988 // in that it must skip particular security frames and checks for
3989 // caller sensitive methods.
3990 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
3991 #ifndef PRODUCT
3992 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3993 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
3994 }
3995 #endif
3996
3997 if (!jvms()->has_method()) {
3998 #ifndef PRODUCT
3999 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4000 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4001 }
4002 #endif
4003 return false;
4004 }
4005
4006 // Walk back up the JVM state to find the caller at the required
4007 // depth.
4008 JVMState* caller_jvms = jvms();
4009
4010 // Cf. JVM_GetCallerClass
4011 // NOTE: Start the loop at depth 1 because the current JVM state does
4012 // not include the Reflection.getCallerClass() frame.
4013 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4014 ciMethod* m = caller_jvms->method();
4015 switch (n) {
4016 case 0:
4017 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4018 break;
4019 case 1:
4020 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4021 if (!m->caller_sensitive()) {
4022 #ifndef PRODUCT
4023 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4024 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4025 }
4026 #endif
4027 return false; // bail-out; let JVM_GetCallerClass do the work
4028 }
4029 break;
4030 default:
4031 if (!m->is_ignored_by_security_stack_walk()) {
4032 // We have reached the desired frame; return the holder class.
4033 // Acquire method holder as java.lang.Class and push as constant.
4034 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4035 ciInstance* caller_mirror = caller_klass->java_mirror();
4036 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4037
4038 #ifndef PRODUCT
4039 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4040 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());
4041 tty->print_cr(" JVM state at this point:");
4042 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4043 ciMethod* m = jvms()->of_depth(i)->method();
4044 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4045 }
4046 }
4047 #endif
4048 return true;
4049 }
4050 break;
4051 }
4052 }
4053
4054 #ifndef PRODUCT
4055 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4056 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4057 tty->print_cr(" JVM state at this point:");
4058 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4059 ciMethod* m = jvms()->of_depth(i)->method();
4060 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4061 }
4062 }
4063 #endif
4064
4065 return false; // bail-out; let JVM_GetCallerClass do the work
4066 }
4067
4068 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4069 Node* arg = argument(0);
4070 Node* result = NULL;
4071
4072 switch (id) {
4073 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4074 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4075 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4076 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4077
4078 case vmIntrinsics::_doubleToLongBits: {
4079 // two paths (plus control) merge in a wood
4080 RegionNode *r = new RegionNode(3);
4081 Node *phi = new PhiNode(r, TypeLong::LONG);
4082
4083 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4084 // Build the boolean node
4085 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4086
4087 // Branch either way.
4088 // NaN case is less traveled, which makes all the difference.
4089 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4090 Node *opt_isnan = _gvn.transform(ifisnan);
4091 assert( opt_isnan->is_If(), "Expect an IfNode");
4092 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4093 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4094
4095 set_control(iftrue);
4096
4097 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4098 Node *slow_result = longcon(nan_bits); // return NaN
4099 phi->init_req(1, _gvn.transform( slow_result ));
4100 r->init_req(1, iftrue);
4101
4102 // Else fall through
4103 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4104 set_control(iffalse);
4105
4106 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4107 r->init_req(2, iffalse);
4108
4109 // Post merge
4110 set_control(_gvn.transform(r));
4111 record_for_igvn(r);
4112
4113 C->set_has_split_ifs(true); // Has chance for split-if optimization
4114 result = phi;
4115 assert(result->bottom_type()->isa_long(), "must be");
4116 break;
4117 }
4118
4119 case vmIntrinsics::_floatToIntBits: {
4120 // two paths (plus control) merge in a wood
4121 RegionNode *r = new RegionNode(3);
4122 Node *phi = new PhiNode(r, TypeInt::INT);
4123
4124 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4125 // Build the boolean node
4126 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4127
4128 // Branch either way.
4129 // NaN case is less traveled, which makes all the difference.
4130 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4131 Node *opt_isnan = _gvn.transform(ifisnan);
4132 assert( opt_isnan->is_If(), "Expect an IfNode");
4133 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4134 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4135
4136 set_control(iftrue);
4137
4138 static const jint nan_bits = 0x7fc00000;
4139 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4140 phi->init_req(1, _gvn.transform( slow_result ));
4141 r->init_req(1, iftrue);
4142
4143 // Else fall through
4144 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4145 set_control(iffalse);
4146
4147 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4148 r->init_req(2, iffalse);
4149
4150 // Post merge
4151 set_control(_gvn.transform(r));
4152 record_for_igvn(r);
4153
4154 C->set_has_split_ifs(true); // Has chance for split-if optimization
4155 result = phi;
4156 assert(result->bottom_type()->isa_int(), "must be");
4157 break;
4158 }
4159
4160 default:
4161 fatal_unexpected_iid(id);
4162 break;
4163 }
4164 set_result(_gvn.transform(result));
4165 return true;
4166 }
4167
4168 //----------------------inline_unsafe_copyMemory-------------------------
4169 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4170 bool LibraryCallKit::inline_unsafe_copyMemory() {
4171 if (callee()->is_static()) return false; // caller must have the capability!
4172 null_check_receiver(); // null-check receiver
4173 if (stopped()) return true;
4174
4175 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4176
4177 Node* src_ptr = argument(1); // type: oop
4178 Node* src_off = ConvL2X(argument(2)); // type: long
4179 Node* dst_ptr = argument(4); // type: oop
4180 Node* dst_off = ConvL2X(argument(5)); // type: long
4181 Node* size = ConvL2X(argument(7)); // type: long
4182
4183 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4184 "fieldOffset must be byte-scaled");
4185
4186 Node* src = make_unsafe_address(src_ptr, src_off, ACCESS_READ);
4187 Node* dst = make_unsafe_address(dst_ptr, dst_off, ACCESS_WRITE);
4188
4189 // Conservatively insert a memory barrier on all memory slices.
4190 // Do not let writes of the copy source or destination float below the copy.
4191 insert_mem_bar(Op_MemBarCPUOrder);
4192
4193 Node* thread = _gvn.transform(new ThreadLocalNode());
4194 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4195 BasicType doing_unsafe_access_bt = T_BYTE;
4196 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4197
4198 // update volatile field
4199 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4200
4201 // Call it. Note that the length argument is not scaled.
4202 make_runtime_call(RC_LEAF|RC_NO_FP,
4203 OptoRuntime::fast_arraycopy_Type(),
4204 StubRoutines::unsafe_arraycopy(),
4205 "unsafe_arraycopy",
4206 TypeRawPtr::BOTTOM,
4207 src, dst, size XTOP);
4208
4209 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4210
4211 // Do not let reads of the copy destination float above the copy.
4212 insert_mem_bar(Op_MemBarCPUOrder);
4213
4214 return true;
4215 }
4216
4217 //------------------------clone_coping-----------------------------------
4218 // Helper function for inline_native_clone.
4219 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4220 assert(obj_size != NULL, "");
4221 Node* raw_obj = alloc_obj->in(1);
4222 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4223
4224 AllocateNode* alloc = NULL;
4225 if (ReduceBulkZeroing) {
4226 // We will be completely responsible for initializing this object -
4227 // mark Initialize node as complete.
4228 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4229 // The object was just allocated - there should be no any stores!
4230 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4231 // Mark as complete_with_arraycopy so that on AllocateNode
4232 // expansion, we know this AllocateNode is initialized by an array
4233 // copy and a StoreStore barrier exists after the array copy.
4234 alloc->initialization()->set_complete_with_arraycopy();
4235 }
4236
4237 Node* size = _gvn.transform(obj_size);
4238 access_clone(obj, alloc_obj, size, is_array);
4239
4240 // Do not let reads from the cloned object float above the arraycopy.
4241 if (alloc != NULL) {
4242 // Do not let stores that initialize this object be reordered with
4243 // a subsequent store that would make this object accessible by
4244 // other threads.
4245 // Record what AllocateNode this StoreStore protects so that
4246 // escape analysis can go from the MemBarStoreStoreNode to the
4247 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4248 // based on the escape status of the AllocateNode.
4249 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4250 } else {
4251 insert_mem_bar(Op_MemBarCPUOrder);
4252 }
4253 }
4254
4255 //------------------------inline_native_clone----------------------------
4256 // protected native Object java.lang.Object.clone();
4257 //
4258 // Here are the simple edge cases:
4259 // null receiver => normal trap
4260 // virtual and clone was overridden => slow path to out-of-line clone
4261 // not cloneable or finalizer => slow path to out-of-line Object.clone
4262 //
4263 // The general case has two steps, allocation and copying.
4264 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4265 //
4266 // Copying also has two cases, oop arrays and everything else.
4267 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4268 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4269 //
4270 // These steps fold up nicely if and when the cloned object's klass
4271 // can be sharply typed as an object array, a type array, or an instance.
4272 //
4273 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4274 PhiNode* result_val;
4275
4276 // Set the reexecute bit for the interpreter to reexecute
4277 // the bytecode that invokes Object.clone if deoptimization happens.
4278 { PreserveReexecuteState preexecs(this);
4279 jvms()->set_should_reexecute(true);
4280
4281 Node* obj = null_check_receiver();
4282 if (stopped()) return true;
4283
4284 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4285
4286 // If we are going to clone an instance, we need its exact type to
4287 // know the number and types of fields to convert the clone to
4288 // loads/stores. Maybe a speculative type can help us.
4289 if (!obj_type->klass_is_exact() &&
4290 obj_type->speculative_type() != NULL &&
4291 obj_type->speculative_type()->is_instance_klass()) {
4292 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4293 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4294 !spec_ik->has_injected_fields()) {
4295 ciKlass* k = obj_type->klass();
4296 if (!k->is_instance_klass() ||
4297 k->as_instance_klass()->is_interface() ||
4298 k->as_instance_klass()->has_subklass()) {
4299 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4300 }
4301 }
4302 }
4303
4304 // Conservatively insert a memory barrier on all memory slices.
4305 // Do not let writes into the original float below the clone.
4306 insert_mem_bar(Op_MemBarCPUOrder);
4307
4308 // paths into result_reg:
4309 enum {
4310 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4311 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4312 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4313 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4314 PATH_LIMIT
4315 };
4316 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4317 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4318 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4319 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4320 record_for_igvn(result_reg);
4321
4322 Node* obj_klass = load_object_klass(obj);
4323 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4324 if (array_ctl != NULL) {
4325 // It's an array.
4326 PreserveJVMState pjvms(this);
4327 set_control(array_ctl);
4328 Node* obj_length = load_array_length(obj);
4329 Node* obj_size = NULL;
4330 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size); // no arguments to push
4331
4332 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4333 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, BarrierSetC2::Parsing)) {
4334 // If it is an oop array, it requires very special treatment,
4335 // because gc barriers are required when accessing the array.
4336 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4337 if (is_obja != NULL) {
4338 PreserveJVMState pjvms2(this);
4339 set_control(is_obja);
4340 // Generate a direct call to the right arraycopy function(s).
4341 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4342 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false);
4343 ac->set_clone_oop_array();
4344 Node* n = _gvn.transform(ac);
4345 assert(n == ac, "cannot disappear");
4346 ac->connect_outputs(this);
4347
4348 result_reg->init_req(_objArray_path, control());
4349 result_val->init_req(_objArray_path, alloc_obj);
4350 result_i_o ->set_req(_objArray_path, i_o());
4351 result_mem ->set_req(_objArray_path, reset_memory());
4352 }
4353 }
4354 // Otherwise, there are no barriers to worry about.
4355 // (We can dispense with card marks if we know the allocation
4356 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4357 // causes the non-eden paths to take compensating steps to
4358 // simulate a fresh allocation, so that no further
4359 // card marks are required in compiled code to initialize
4360 // the object.)
4361
4362 if (!stopped()) {
4363 copy_to_clone(obj, alloc_obj, obj_size, true);
4364
4365 // Present the results of the copy.
4366 result_reg->init_req(_array_path, control());
4367 result_val->init_req(_array_path, alloc_obj);
4368 result_i_o ->set_req(_array_path, i_o());
4369 result_mem ->set_req(_array_path, reset_memory());
4370 }
4371 }
4372
4373 // We only go to the instance fast case code if we pass a number of guards.
4374 // The paths which do not pass are accumulated in the slow_region.
4375 RegionNode* slow_region = new RegionNode(1);
4376 record_for_igvn(slow_region);
4377 if (!stopped()) {
4378 // It's an instance (we did array above). Make the slow-path tests.
4379 // If this is a virtual call, we generate a funny guard. We grab
4380 // the vtable entry corresponding to clone() from the target object.
4381 // If the target method which we are calling happens to be the
4382 // Object clone() method, we pass the guard. We do not need this
4383 // guard for non-virtual calls; the caller is known to be the native
4384 // Object clone().
4385 if (is_virtual) {
4386 generate_virtual_guard(obj_klass, slow_region);
4387 }
4388
4389 // The object must be easily cloneable and must not have a finalizer.
4390 // Both of these conditions may be checked in a single test.
4391 // We could optimize the test further, but we don't care.
4392 generate_access_flags_guard(obj_klass,
4393 // Test both conditions:
4394 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4395 // Must be cloneable but not finalizer:
4396 JVM_ACC_IS_CLONEABLE_FAST,
4397 slow_region);
4398 }
4399
4400 if (!stopped()) {
4401 // It's an instance, and it passed the slow-path tests.
4402 PreserveJVMState pjvms(this);
4403 Node* obj_size = NULL;
4404 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4405 // is reexecuted if deoptimization occurs and there could be problems when merging
4406 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4407 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4408
4409 copy_to_clone(obj, alloc_obj, obj_size, false);
4410
4411 // Present the results of the slow call.
4412 result_reg->init_req(_instance_path, control());
4413 result_val->init_req(_instance_path, alloc_obj);
4414 result_i_o ->set_req(_instance_path, i_o());
4415 result_mem ->set_req(_instance_path, reset_memory());
4416 }
4417
4418 // Generate code for the slow case. We make a call to clone().
4419 set_control(_gvn.transform(slow_region));
4420 if (!stopped()) {
4421 PreserveJVMState pjvms(this);
4422 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4423 // We need to deoptimize on exception (see comment above)
4424 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4425 // this->control() comes from set_results_for_java_call
4426 result_reg->init_req(_slow_path, control());
4427 result_val->init_req(_slow_path, slow_result);
4428 result_i_o ->set_req(_slow_path, i_o());
4429 result_mem ->set_req(_slow_path, reset_memory());
4430 }
4431
4432 // Return the combined state.
4433 set_control( _gvn.transform(result_reg));
4434 set_i_o( _gvn.transform(result_i_o));
4435 set_all_memory( _gvn.transform(result_mem));
4436 } // original reexecute is set back here
4437
4438 set_result(_gvn.transform(result_val));
4439 return true;
4440 }
4441
4442 // If we have a tightly coupled allocation, the arraycopy may take care
4443 // of the array initialization. If one of the guards we insert between
4444 // the allocation and the arraycopy causes a deoptimization, an
4445 // unitialized array will escape the compiled method. To prevent that
4446 // we set the JVM state for uncommon traps between the allocation and
4447 // the arraycopy to the state before the allocation so, in case of
4448 // deoptimization, we'll reexecute the allocation and the
4449 // initialization.
4450 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4451 if (alloc != NULL) {
4452 ciMethod* trap_method = alloc->jvms()->method();
4453 int trap_bci = alloc->jvms()->bci();
4454
4455 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4456 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4457 // Make sure there's no store between the allocation and the
4458 // arraycopy otherwise visible side effects could be rexecuted
4459 // in case of deoptimization and cause incorrect execution.
4460 bool no_interfering_store = true;
4461 Node* mem = alloc->in(TypeFunc::Memory);
4462 if (mem->is_MergeMem()) {
4463 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4464 Node* n = mms.memory();
4465 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4466 assert(n->is_Store(), "what else?");
4467 no_interfering_store = false;
4468 break;
4469 }
4470 }
4471 } else {
4472 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4473 Node* n = mms.memory();
4474 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4475 assert(n->is_Store(), "what else?");
4476 no_interfering_store = false;
4477 break;
4478 }
4479 }
4480 }
4481
4482 if (no_interfering_store) {
4483 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4484 uint size = alloc->req();
4485 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4486 old_jvms->set_map(sfpt);
4487 for (uint i = 0; i < size; i++) {
4488 sfpt->init_req(i, alloc->in(i));
4489 }
4490 // re-push array length for deoptimization
4491 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4492 old_jvms->set_sp(old_jvms->sp()+1);
4493 old_jvms->set_monoff(old_jvms->monoff()+1);
4494 old_jvms->set_scloff(old_jvms->scloff()+1);
4495 old_jvms->set_endoff(old_jvms->endoff()+1);
4496 old_jvms->set_should_reexecute(true);
4497
4498 sfpt->set_i_o(map()->i_o());
4499 sfpt->set_memory(map()->memory());
4500 sfpt->set_control(map()->control());
4501
4502 JVMState* saved_jvms = jvms();
4503 saved_reexecute_sp = _reexecute_sp;
4504
4505 set_jvms(sfpt->jvms());
4506 _reexecute_sp = jvms()->sp();
4507
4508 return saved_jvms;
4509 }
4510 }
4511 }
4512 return NULL;
4513 }
4514
4515 // In case of a deoptimization, we restart execution at the
4516 // allocation, allocating a new array. We would leave an uninitialized
4517 // array in the heap that GCs wouldn't expect. Move the allocation
4518 // after the traps so we don't allocate the array if we
4519 // deoptimize. This is possible because tightly_coupled_allocation()
4520 // guarantees there's no observer of the allocated array at this point
4521 // and the control flow is simple enough.
4522 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4523 int saved_reexecute_sp, uint new_idx) {
4524 if (saved_jvms != NULL && !stopped()) {
4525 assert(alloc != NULL, "only with a tightly coupled allocation");
4526 // restore JVM state to the state at the arraycopy
4527 saved_jvms->map()->set_control(map()->control());
4528 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4529 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4530 // If we've improved the types of some nodes (null check) while
4531 // emitting the guards, propagate them to the current state
4532 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4533 set_jvms(saved_jvms);
4534 _reexecute_sp = saved_reexecute_sp;
4535
4536 // Remove the allocation from above the guards
4537 CallProjections callprojs;
4538 alloc->extract_projections(&callprojs, true);
4539 InitializeNode* init = alloc->initialization();
4540 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4541 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4542 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4543 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4544
4545 // move the allocation here (after the guards)
4546 _gvn.hash_delete(alloc);
4547 alloc->set_req(TypeFunc::Control, control());
4548 alloc->set_req(TypeFunc::I_O, i_o());
4549 Node *mem = reset_memory();
4550 set_all_memory(mem);
4551 alloc->set_req(TypeFunc::Memory, mem);
4552 set_control(init->proj_out_or_null(TypeFunc::Control));
4553 set_i_o(callprojs.fallthrough_ioproj);
4554
4555 // Update memory as done in GraphKit::set_output_for_allocation()
4556 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4557 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4558 if (ary_type->isa_aryptr() && length_type != NULL) {
4559 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4560 }
4561 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4562 int elemidx = C->get_alias_index(telemref);
4563 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
4564 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
4565
4566 Node* allocx = _gvn.transform(alloc);
4567 assert(allocx == alloc, "where has the allocation gone?");
4568 assert(dest->is_CheckCastPP(), "not an allocation result?");
4569
4570 _gvn.hash_delete(dest);
4571 dest->set_req(0, control());
4572 Node* destx = _gvn.transform(dest);
4573 assert(destx == dest, "where has the allocation result gone?");
4574 }
4575 }
4576
4577
4578 //------------------------------inline_arraycopy-----------------------
4579 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4580 // Object dest, int destPos,
4581 // int length);
4582 bool LibraryCallKit::inline_arraycopy() {
4583 // Get the arguments.
4584 Node* src = argument(0); // type: oop
4585 Node* src_offset = argument(1); // type: int
4586 Node* dest = argument(2); // type: oop
4587 Node* dest_offset = argument(3); // type: int
4588 Node* length = argument(4); // type: int
4589
4590 uint new_idx = C->unique();
4591
4592 // Check for allocation before we add nodes that would confuse
4593 // tightly_coupled_allocation()
4594 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4595
4596 int saved_reexecute_sp = -1;
4597 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4598 // See arraycopy_restore_alloc_state() comment
4599 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4600 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4601 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
4602 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4603
4604 // The following tests must be performed
4605 // (1) src and dest are arrays.
4606 // (2) src and dest arrays must have elements of the same BasicType
4607 // (3) src and dest must not be null.
4608 // (4) src_offset must not be negative.
4609 // (5) dest_offset must not be negative.
4610 // (6) length must not be negative.
4611 // (7) src_offset + length must not exceed length of src.
4612 // (8) dest_offset + length must not exceed length of dest.
4613 // (9) each element of an oop array must be assignable
4614
4615 // (3) src and dest must not be null.
4616 // always do this here because we need the JVM state for uncommon traps
4617 Node* null_ctl = top();
4618 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4619 assert(null_ctl->is_top(), "no null control here");
4620 dest = null_check(dest, T_ARRAY);
4621
4622 if (!can_emit_guards) {
4623 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4624 // guards but the arraycopy node could still take advantage of a
4625 // tightly allocated allocation. tightly_coupled_allocation() is
4626 // called again to make sure it takes the null check above into
4627 // account: the null check is mandatory and if it caused an
4628 // uncommon trap to be emitted then the allocation can't be
4629 // considered tightly coupled in this context.
4630 alloc = tightly_coupled_allocation(dest, NULL);
4631 }
4632
4633 bool validated = false;
4634
4635 const Type* src_type = _gvn.type(src);
4636 const Type* dest_type = _gvn.type(dest);
4637 const TypeAryPtr* top_src = src_type->isa_aryptr();
4638 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4639
4640 // Do we have the type of src?
4641 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4642 // Do we have the type of dest?
4643 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4644 // Is the type for src from speculation?
4645 bool src_spec = false;
4646 // Is the type for dest from speculation?
4647 bool dest_spec = false;
4648
4649 if ((!has_src || !has_dest) && can_emit_guards) {
4650 // We don't have sufficient type information, let's see if
4651 // speculative types can help. We need to have types for both src
4652 // and dest so that it pays off.
4653
4654 // Do we already have or could we have type information for src
4655 bool could_have_src = has_src;
4656 // Do we already have or could we have type information for dest
4657 bool could_have_dest = has_dest;
4658
4659 ciKlass* src_k = NULL;
4660 if (!has_src) {
4661 src_k = src_type->speculative_type_not_null();
4662 if (src_k != NULL && src_k->is_array_klass()) {
4663 could_have_src = true;
4664 }
4665 }
4666
4667 ciKlass* dest_k = NULL;
4668 if (!has_dest) {
4669 dest_k = dest_type->speculative_type_not_null();
4670 if (dest_k != NULL && dest_k->is_array_klass()) {
4671 could_have_dest = true;
4672 }
4673 }
4674
4675 if (could_have_src && could_have_dest) {
4676 // This is going to pay off so emit the required guards
4677 if (!has_src) {
4678 src = maybe_cast_profiled_obj(src, src_k, true);
4679 src_type = _gvn.type(src);
4680 top_src = src_type->isa_aryptr();
4681 has_src = (top_src != NULL && top_src->klass() != NULL);
4682 src_spec = true;
4683 }
4684 if (!has_dest) {
4685 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4686 dest_type = _gvn.type(dest);
4687 top_dest = dest_type->isa_aryptr();
4688 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4689 dest_spec = true;
4690 }
4691 }
4692 }
4693
4694 if (has_src && has_dest && can_emit_guards) {
4695 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
4696 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
4697 if (is_reference_type(src_elem)) src_elem = T_OBJECT;
4698 if (is_reference_type(dest_elem)) dest_elem = T_OBJECT;
4699
4700 if (src_elem == dest_elem && src_elem == T_OBJECT) {
4701 // If both arrays are object arrays then having the exact types
4702 // for both will remove the need for a subtype check at runtime
4703 // before the call and may make it possible to pick a faster copy
4704 // routine (without a subtype check on every element)
4705 // Do we have the exact type of src?
4706 bool could_have_src = src_spec;
4707 // Do we have the exact type of dest?
4708 bool could_have_dest = dest_spec;
4709 ciKlass* src_k = top_src->klass();
4710 ciKlass* dest_k = top_dest->klass();
4711 if (!src_spec) {
4712 src_k = src_type->speculative_type_not_null();
4713 if (src_k != NULL && src_k->is_array_klass()) {
4714 could_have_src = true;
4715 }
4716 }
4717 if (!dest_spec) {
4718 dest_k = dest_type->speculative_type_not_null();
4719 if (dest_k != NULL && dest_k->is_array_klass()) {
4720 could_have_dest = true;
4721 }
4722 }
4723 if (could_have_src && could_have_dest) {
4724 // If we can have both exact types, emit the missing guards
4725 if (could_have_src && !src_spec) {
4726 src = maybe_cast_profiled_obj(src, src_k, true);
4727 }
4728 if (could_have_dest && !dest_spec) {
4729 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4730 }
4731 }
4732 }
4733 }
4734
4735 ciMethod* trap_method = method();
4736 int trap_bci = bci();
4737 if (saved_jvms != NULL) {
4738 trap_method = alloc->jvms()->method();
4739 trap_bci = alloc->jvms()->bci();
4740 }
4741
4742 bool negative_length_guard_generated = false;
4743
4744 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4745 can_emit_guards &&
4746 !src->is_top() && !dest->is_top()) {
4747 // validate arguments: enables transformation the ArrayCopyNode
4748 validated = true;
4749
4750 RegionNode* slow_region = new RegionNode(1);
4751 record_for_igvn(slow_region);
4752
4753 // (1) src and dest are arrays.
4754 generate_non_array_guard(load_object_klass(src), slow_region);
4755 generate_non_array_guard(load_object_klass(dest), slow_region);
4756
4757 // (2) src and dest arrays must have elements of the same BasicType
4758 // done at macro expansion or at Ideal transformation time
4759
4760 // (4) src_offset must not be negative.
4761 generate_negative_guard(src_offset, slow_region);
4762
4763 // (5) dest_offset must not be negative.
4764 generate_negative_guard(dest_offset, slow_region);
4765
4766 // (7) src_offset + length must not exceed length of src.
4767 generate_limit_guard(src_offset, length,
4768 load_array_length(src),
4769 slow_region);
4770
4771 // (8) dest_offset + length must not exceed length of dest.
4772 generate_limit_guard(dest_offset, length,
4773 load_array_length(dest),
4774 slow_region);
4775
4776 // (6) length must not be negative.
4777 // This is also checked in generate_arraycopy() during macro expansion, but
4778 // we also have to check it here for the case where the ArrayCopyNode will
4779 // be eliminated by Escape Analysis.
4780 if (EliminateAllocations) {
4781 generate_negative_guard(length, slow_region);
4782 negative_length_guard_generated = true;
4783 }
4784
4785 // (9) each element of an oop array must be assignable
4786 Node* dest_klass = load_object_klass(dest);
4787 if (src != dest) {
4788 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
4789
4790 if (not_subtype_ctrl != top()) {
4791 PreserveJVMState pjvms(this);
4792 set_control(not_subtype_ctrl);
4793 uncommon_trap(Deoptimization::Reason_intrinsic,
4794 Deoptimization::Action_make_not_entrant);
4795 assert(stopped(), "Should be stopped");
4796 }
4797 }
4798 {
4799 PreserveJVMState pjvms(this);
4800 set_control(_gvn.transform(slow_region));
4801 uncommon_trap(Deoptimization::Reason_intrinsic,
4802 Deoptimization::Action_make_not_entrant);
4803 assert(stopped(), "Should be stopped");
4804 }
4805
4806 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
4807 const Type *toop = TypeOopPtr::make_from_klass(dest_klass_t->klass());
4808 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
4809 }
4810
4811 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
4812
4813 if (stopped()) {
4814 return true;
4815 }
4816
4817 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
4818 // Create LoadRange and LoadKlass nodes for use during macro expansion here
4819 // so the compiler has a chance to eliminate them: during macro expansion,
4820 // we have to set their control (CastPP nodes are eliminated).
4821 load_object_klass(src), load_object_klass(dest),
4822 load_array_length(src), load_array_length(dest));
4823
4824 ac->set_arraycopy(validated);
4825
4826 Node* n = _gvn.transform(ac);
4827 if (n == ac) {
4828 ac->connect_outputs(this);
4829 } else {
4830 assert(validated, "shouldn't transform if all arguments not validated");
4831 set_all_memory(n);
4832 }
4833 clear_upper_avx();
4834
4835
4836 return true;
4837 }
4838
4839
4840 // Helper function which determines if an arraycopy immediately follows
4841 // an allocation, with no intervening tests or other escapes for the object.
4842 AllocateArrayNode*
4843 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
4844 RegionNode* slow_region) {
4845 if (stopped()) return NULL; // no fast path
4846 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
4847
4848 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
4849 if (alloc == NULL) return NULL;
4850
4851 Node* rawmem = memory(Compile::AliasIdxRaw);
4852 // Is the allocation's memory state untouched?
4853 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
4854 // Bail out if there have been raw-memory effects since the allocation.
4855 // (Example: There might have been a call or safepoint.)
4856 return NULL;
4857 }
4858 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
4859 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
4860 return NULL;
4861 }
4862
4863 // There must be no unexpected observers of this allocation.
4864 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
4865 Node* obs = ptr->fast_out(i);
4866 if (obs != this->map()) {
4867 return NULL;
4868 }
4869 }
4870
4871 // This arraycopy must unconditionally follow the allocation of the ptr.
4872 Node* alloc_ctl = ptr->in(0);
4873 Node* ctl = control();
4874 while (ctl != alloc_ctl) {
4875 // There may be guards which feed into the slow_region.
4876 // Any other control flow means that we might not get a chance
4877 // to finish initializing the allocated object.
4878 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
4879 IfNode* iff = ctl->in(0)->as_If();
4880 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
4881 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
4882 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
4883 ctl = iff->in(0); // This test feeds the known slow_region.
4884 continue;
4885 }
4886 // One more try: Various low-level checks bottom out in
4887 // uncommon traps. If the debug-info of the trap omits
4888 // any reference to the allocation, as we've already
4889 // observed, then there can be no objection to the trap.
4890 bool found_trap = false;
4891 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
4892 Node* obs = not_ctl->fast_out(j);
4893 if (obs->in(0) == not_ctl && obs->is_Call() &&
4894 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
4895 found_trap = true; break;
4896 }
4897 }
4898 if (found_trap) {
4899 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
4900 continue;
4901 }
4902 }
4903 return NULL;
4904 }
4905
4906 // If we get this far, we have an allocation which immediately
4907 // precedes the arraycopy, and we can take over zeroing the new object.
4908 // The arraycopy will finish the initialization, and provide
4909 // a new control state to which we will anchor the destination pointer.
4910
4911 return alloc;
4912 }
4913
4914 //-------------inline_encodeISOArray-----------------------------------
4915 // encode char[] to byte[] in ISO_8859_1
4916 bool LibraryCallKit::inline_encodeISOArray() {
4917 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
4918 // no receiver since it is static method
4919 Node *src = argument(0);
4920 Node *src_offset = argument(1);
4921 Node *dst = argument(2);
4922 Node *dst_offset = argument(3);
4923 Node *length = argument(4);
4924
4925 src = must_be_not_null(src, true);
4926 dst = must_be_not_null(dst, true);
4927
4928 const Type* src_type = src->Value(&_gvn);
4929 const Type* dst_type = dst->Value(&_gvn);
4930 const TypeAryPtr* top_src = src_type->isa_aryptr();
4931 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
4932 if (top_src == NULL || top_src->klass() == NULL ||
4933 top_dest == NULL || top_dest->klass() == NULL) {
4934 // failed array check
4935 return false;
4936 }
4937
4938 // Figure out the size and type of the elements we will be copying.
4939 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4940 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4941 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
4942 return false;
4943 }
4944
4945 Node* src_start = array_element_address(src, src_offset, T_CHAR);
4946 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
4947 // 'src_start' points to src array + scaled offset
4948 // 'dst_start' points to dst array + scaled offset
4949
4950 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
4951 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
4952 enc = _gvn.transform(enc);
4953 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
4954 set_memory(res_mem, mtype);
4955 set_result(enc);
4956 clear_upper_avx();
4957
4958 return true;
4959 }
4960
4961 //-------------inline_multiplyToLen-----------------------------------
4962 bool LibraryCallKit::inline_multiplyToLen() {
4963 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
4964
4965 address stubAddr = StubRoutines::multiplyToLen();
4966 if (stubAddr == NULL) {
4967 return false; // Intrinsic's stub is not implemented on this platform
4968 }
4969 const char* stubName = "multiplyToLen";
4970
4971 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
4972
4973 // no receiver because it is a static method
4974 Node* x = argument(0);
4975 Node* xlen = argument(1);
4976 Node* y = argument(2);
4977 Node* ylen = argument(3);
4978 Node* z = argument(4);
4979
4980 x = must_be_not_null(x, true);
4981 y = must_be_not_null(y, true);
4982
4983 const Type* x_type = x->Value(&_gvn);
4984 const Type* y_type = y->Value(&_gvn);
4985 const TypeAryPtr* top_x = x_type->isa_aryptr();
4986 const TypeAryPtr* top_y = y_type->isa_aryptr();
4987 if (top_x == NULL || top_x->klass() == NULL ||
4988 top_y == NULL || top_y->klass() == NULL) {
4989 // failed array check
4990 return false;
4991 }
4992
4993 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4994 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4995 if (x_elem != T_INT || y_elem != T_INT) {
4996 return false;
4997 }
4998
4999 // Set the original stack and the reexecute bit for the interpreter to reexecute
5000 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5001 // on the return from z array allocation in runtime.
5002 { PreserveReexecuteState preexecs(this);
5003 jvms()->set_should_reexecute(true);
5004
5005 Node* x_start = array_element_address(x, intcon(0), x_elem);
5006 Node* y_start = array_element_address(y, intcon(0), y_elem);
5007 // 'x_start' points to x array + scaled xlen
5008 // 'y_start' points to y array + scaled ylen
5009
5010 // Allocate the result array
5011 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5012 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5013 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5014
5015 IdealKit ideal(this);
5016
5017 #define __ ideal.
5018 Node* one = __ ConI(1);
5019 Node* zero = __ ConI(0);
5020 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5021 __ set(need_alloc, zero);
5022 __ set(z_alloc, z);
5023 __ if_then(z, BoolTest::eq, null()); {
5024 __ increment (need_alloc, one);
5025 } __ else_(); {
5026 // Update graphKit memory and control from IdealKit.
5027 sync_kit(ideal);
5028 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5029 cast->init_req(0, control());
5030 _gvn.set_type(cast, cast->bottom_type());
5031 C->record_for_igvn(cast);
5032
5033 Node* zlen_arg = load_array_length(cast);
5034 // Update IdealKit memory and control from graphKit.
5035 __ sync_kit(this);
5036 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5037 __ increment (need_alloc, one);
5038 } __ end_if();
5039 } __ end_if();
5040
5041 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5042 // Update graphKit memory and control from IdealKit.
5043 sync_kit(ideal);
5044 Node * narr = new_array(klass_node, zlen, 1);
5045 // Update IdealKit memory and control from graphKit.
5046 __ sync_kit(this);
5047 __ set(z_alloc, narr);
5048 } __ end_if();
5049
5050 sync_kit(ideal);
5051 z = __ value(z_alloc);
5052 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5053 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5054 // Final sync IdealKit and GraphKit.
5055 final_sync(ideal);
5056 #undef __
5057
5058 Node* z_start = array_element_address(z, intcon(0), T_INT);
5059
5060 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5061 OptoRuntime::multiplyToLen_Type(),
5062 stubAddr, stubName, TypePtr::BOTTOM,
5063 x_start, xlen, y_start, ylen, z_start, zlen);
5064 } // original reexecute is set back here
5065
5066 C->set_has_split_ifs(true); // Has chance for split-if optimization
5067 set_result(z);
5068 return true;
5069 }
5070
5071 //-------------inline_squareToLen------------------------------------
5072 bool LibraryCallKit::inline_squareToLen() {
5073 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5074
5075 address stubAddr = StubRoutines::squareToLen();
5076 if (stubAddr == NULL) {
5077 return false; // Intrinsic's stub is not implemented on this platform
5078 }
5079 const char* stubName = "squareToLen";
5080
5081 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5082
5083 Node* x = argument(0);
5084 Node* len = argument(1);
5085 Node* z = argument(2);
5086 Node* zlen = argument(3);
5087
5088 x = must_be_not_null(x, true);
5089 z = must_be_not_null(z, true);
5090
5091 const Type* x_type = x->Value(&_gvn);
5092 const Type* z_type = z->Value(&_gvn);
5093 const TypeAryPtr* top_x = x_type->isa_aryptr();
5094 const TypeAryPtr* top_z = z_type->isa_aryptr();
5095 if (top_x == NULL || top_x->klass() == NULL ||
5096 top_z == NULL || top_z->klass() == NULL) {
5097 // failed array check
5098 return false;
5099 }
5100
5101 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5102 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5103 if (x_elem != T_INT || z_elem != T_INT) {
5104 return false;
5105 }
5106
5107
5108 Node* x_start = array_element_address(x, intcon(0), x_elem);
5109 Node* z_start = array_element_address(z, intcon(0), z_elem);
5110
5111 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5112 OptoRuntime::squareToLen_Type(),
5113 stubAddr, stubName, TypePtr::BOTTOM,
5114 x_start, len, z_start, zlen);
5115
5116 set_result(z);
5117 return true;
5118 }
5119
5120 //-------------inline_mulAdd------------------------------------------
5121 bool LibraryCallKit::inline_mulAdd() {
5122 assert(UseMulAddIntrinsic, "not implemented on this platform");
5123
5124 address stubAddr = StubRoutines::mulAdd();
5125 if (stubAddr == NULL) {
5126 return false; // Intrinsic's stub is not implemented on this platform
5127 }
5128 const char* stubName = "mulAdd";
5129
5130 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5131
5132 Node* out = argument(0);
5133 Node* in = argument(1);
5134 Node* offset = argument(2);
5135 Node* len = argument(3);
5136 Node* k = argument(4);
5137
5138 out = must_be_not_null(out, true);
5139
5140 const Type* out_type = out->Value(&_gvn);
5141 const Type* in_type = in->Value(&_gvn);
5142 const TypeAryPtr* top_out = out_type->isa_aryptr();
5143 const TypeAryPtr* top_in = in_type->isa_aryptr();
5144 if (top_out == NULL || top_out->klass() == NULL ||
5145 top_in == NULL || top_in->klass() == NULL) {
5146 // failed array check
5147 return false;
5148 }
5149
5150 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5151 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5152 if (out_elem != T_INT || in_elem != T_INT) {
5153 return false;
5154 }
5155
5156 Node* outlen = load_array_length(out);
5157 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5158 Node* out_start = array_element_address(out, intcon(0), out_elem);
5159 Node* in_start = array_element_address(in, intcon(0), in_elem);
5160
5161 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5162 OptoRuntime::mulAdd_Type(),
5163 stubAddr, stubName, TypePtr::BOTTOM,
5164 out_start,in_start, new_offset, len, k);
5165 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5166 set_result(result);
5167 return true;
5168 }
5169
5170 //-------------inline_montgomeryMultiply-----------------------------------
5171 bool LibraryCallKit::inline_montgomeryMultiply() {
5172 address stubAddr = StubRoutines::montgomeryMultiply();
5173 if (stubAddr == NULL) {
5174 return false; // Intrinsic's stub is not implemented on this platform
5175 }
5176
5177 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5178 const char* stubName = "montgomery_multiply";
5179
5180 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5181
5182 Node* a = argument(0);
5183 Node* b = argument(1);
5184 Node* n = argument(2);
5185 Node* len = argument(3);
5186 Node* inv = argument(4);
5187 Node* m = argument(6);
5188
5189 const Type* a_type = a->Value(&_gvn);
5190 const TypeAryPtr* top_a = a_type->isa_aryptr();
5191 const Type* b_type = b->Value(&_gvn);
5192 const TypeAryPtr* top_b = b_type->isa_aryptr();
5193 const Type* n_type = a->Value(&_gvn);
5194 const TypeAryPtr* top_n = n_type->isa_aryptr();
5195 const Type* m_type = a->Value(&_gvn);
5196 const TypeAryPtr* top_m = m_type->isa_aryptr();
5197 if (top_a == NULL || top_a->klass() == NULL ||
5198 top_b == NULL || top_b->klass() == NULL ||
5199 top_n == NULL || top_n->klass() == NULL ||
5200 top_m == NULL || top_m->klass() == NULL) {
5201 // failed array check
5202 return false;
5203 }
5204
5205 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5206 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5207 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5208 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5209 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5210 return false;
5211 }
5212
5213 // Make the call
5214 {
5215 Node* a_start = array_element_address(a, intcon(0), a_elem);
5216 Node* b_start = array_element_address(b, intcon(0), b_elem);
5217 Node* n_start = array_element_address(n, intcon(0), n_elem);
5218 Node* m_start = array_element_address(m, intcon(0), m_elem);
5219
5220 Node* call = make_runtime_call(RC_LEAF,
5221 OptoRuntime::montgomeryMultiply_Type(),
5222 stubAddr, stubName, TypePtr::BOTTOM,
5223 a_start, b_start, n_start, len, inv, top(),
5224 m_start);
5225 set_result(m);
5226 }
5227
5228 return true;
5229 }
5230
5231 bool LibraryCallKit::inline_montgomerySquare() {
5232 address stubAddr = StubRoutines::montgomerySquare();
5233 if (stubAddr == NULL) {
5234 return false; // Intrinsic's stub is not implemented on this platform
5235 }
5236
5237 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5238 const char* stubName = "montgomery_square";
5239
5240 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5241
5242 Node* a = argument(0);
5243 Node* n = argument(1);
5244 Node* len = argument(2);
5245 Node* inv = argument(3);
5246 Node* m = argument(5);
5247
5248 const Type* a_type = a->Value(&_gvn);
5249 const TypeAryPtr* top_a = a_type->isa_aryptr();
5250 const Type* n_type = a->Value(&_gvn);
5251 const TypeAryPtr* top_n = n_type->isa_aryptr();
5252 const Type* m_type = a->Value(&_gvn);
5253 const TypeAryPtr* top_m = m_type->isa_aryptr();
5254 if (top_a == NULL || top_a->klass() == NULL ||
5255 top_n == NULL || top_n->klass() == NULL ||
5256 top_m == NULL || top_m->klass() == NULL) {
5257 // failed array check
5258 return false;
5259 }
5260
5261 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5262 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5263 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5264 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5265 return false;
5266 }
5267
5268 // Make the call
5269 {
5270 Node* a_start = array_element_address(a, intcon(0), a_elem);
5271 Node* n_start = array_element_address(n, intcon(0), n_elem);
5272 Node* m_start = array_element_address(m, intcon(0), m_elem);
5273
5274 Node* call = make_runtime_call(RC_LEAF,
5275 OptoRuntime::montgomerySquare_Type(),
5276 stubAddr, stubName, TypePtr::BOTTOM,
5277 a_start, n_start, len, inv, top(),
5278 m_start);
5279 set_result(m);
5280 }
5281
5282 return true;
5283 }
5284
5285 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
5286 address stubAddr = NULL;
5287 const char* stubName = NULL;
5288
5289 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
5290 if (stubAddr == NULL) {
5291 return false; // Intrinsic's stub is not implemented on this platform
5292 }
5293
5294 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
5295
5296 assert(callee()->signature()->size() == 5, "expected 5 arguments");
5297
5298 Node* newArr = argument(0);
5299 Node* oldArr = argument(1);
5300 Node* newIdx = argument(2);
5301 Node* shiftCount = argument(3);
5302 Node* numIter = argument(4);
5303
5304 const Type* newArr_type = newArr->Value(&_gvn);
5305 const TypeAryPtr* top_newArr = newArr_type->isa_aryptr();
5306 const Type* oldArr_type = oldArr->Value(&_gvn);
5307 const TypeAryPtr* top_oldArr = oldArr_type->isa_aryptr();
5308 if (top_newArr == NULL || top_newArr->klass() == NULL || top_oldArr == NULL
5309 || top_oldArr->klass() == NULL) {
5310 return false;
5311 }
5312
5313 BasicType newArr_elem = newArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5314 BasicType oldArr_elem = oldArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5315 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
5316 return false;
5317 }
5318
5319 // Make the call
5320 {
5321 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
5322 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
5323
5324 Node* call = make_runtime_call(RC_LEAF,
5325 OptoRuntime::bigIntegerShift_Type(),
5326 stubAddr,
5327 stubName,
5328 TypePtr::BOTTOM,
5329 newArr_start,
5330 oldArr_start,
5331 newIdx,
5332 shiftCount,
5333 numIter);
5334 }
5335
5336 return true;
5337 }
5338
5339 //-------------inline_vectorizedMismatch------------------------------
5340 bool LibraryCallKit::inline_vectorizedMismatch() {
5341 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5342
5343 address stubAddr = StubRoutines::vectorizedMismatch();
5344 if (stubAddr == NULL) {
5345 return false; // Intrinsic's stub is not implemented on this platform
5346 }
5347 const char* stubName = "vectorizedMismatch";
5348 int size_l = callee()->signature()->size();
5349 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5350
5351 Node* obja = argument(0);
5352 Node* aoffset = argument(1);
5353 Node* objb = argument(3);
5354 Node* boffset = argument(4);
5355 Node* length = argument(6);
5356 Node* scale = argument(7);
5357
5358 const Type* a_type = obja->Value(&_gvn);
5359 const Type* b_type = objb->Value(&_gvn);
5360 const TypeAryPtr* top_a = a_type->isa_aryptr();
5361 const TypeAryPtr* top_b = b_type->isa_aryptr();
5362 if (top_a == NULL || top_a->klass() == NULL ||
5363 top_b == NULL || top_b->klass() == NULL) {
5364 // failed array check
5365 return false;
5366 }
5367
5368 Node* call;
5369 jvms()->set_should_reexecute(true);
5370
5371 Node* obja_adr = make_unsafe_address(obja, aoffset, ACCESS_READ);
5372 Node* objb_adr = make_unsafe_address(objb, boffset, ACCESS_READ);
5373
5374 call = make_runtime_call(RC_LEAF,
5375 OptoRuntime::vectorizedMismatch_Type(),
5376 stubAddr, stubName, TypePtr::BOTTOM,
5377 obja_adr, objb_adr, length, scale);
5378
5379 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5380 set_result(result);
5381 return true;
5382 }
5383
5384 /**
5385 * Calculate CRC32 for byte.
5386 * int java.util.zip.CRC32.update(int crc, int b)
5387 */
5388 bool LibraryCallKit::inline_updateCRC32() {
5389 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5390 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5391 // no receiver since it is static method
5392 Node* crc = argument(0); // type: int
5393 Node* b = argument(1); // type: int
5394
5395 /*
5396 * int c = ~ crc;
5397 * b = timesXtoThe32[(b ^ c) & 0xFF];
5398 * b = b ^ (c >>> 8);
5399 * crc = ~b;
5400 */
5401
5402 Node* M1 = intcon(-1);
5403 crc = _gvn.transform(new XorINode(crc, M1));
5404 Node* result = _gvn.transform(new XorINode(crc, b));
5405 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5406
5407 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5408 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5409 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5410 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5411
5412 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5413 result = _gvn.transform(new XorINode(crc, result));
5414 result = _gvn.transform(new XorINode(result, M1));
5415 set_result(result);
5416 return true;
5417 }
5418
5419 /**
5420 * Calculate CRC32 for byte[] array.
5421 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5422 */
5423 bool LibraryCallKit::inline_updateBytesCRC32() {
5424 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5425 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5426 // no receiver since it is static method
5427 Node* crc = argument(0); // type: int
5428 Node* src = argument(1); // type: oop
5429 Node* offset = argument(2); // type: int
5430 Node* length = argument(3); // type: int
5431
5432 const Type* src_type = src->Value(&_gvn);
5433 const TypeAryPtr* top_src = src_type->isa_aryptr();
5434 if (top_src == NULL || top_src->klass() == NULL) {
5435 // failed array check
5436 return false;
5437 }
5438
5439 // Figure out the size and type of the elements we will be copying.
5440 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5441 if (src_elem != T_BYTE) {
5442 return false;
5443 }
5444
5445 // 'src_start' points to src array + scaled offset
5446 src = must_be_not_null(src, true);
5447 Node* src_start = array_element_address(src, offset, src_elem);
5448
5449 // We assume that range check is done by caller.
5450 // TODO: generate range check (offset+length < src.length) in debug VM.
5451
5452 // Call the stub.
5453 address stubAddr = StubRoutines::updateBytesCRC32();
5454 const char *stubName = "updateBytesCRC32";
5455
5456 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5457 stubAddr, stubName, TypePtr::BOTTOM,
5458 crc, src_start, length);
5459 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5460 set_result(result);
5461 return true;
5462 }
5463
5464 /**
5465 * Calculate CRC32 for ByteBuffer.
5466 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5467 */
5468 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5469 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5470 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5471 // no receiver since it is static method
5472 Node* crc = argument(0); // type: int
5473 Node* src = argument(1); // type: long
5474 Node* offset = argument(3); // type: int
5475 Node* length = argument(4); // type: int
5476
5477 src = ConvL2X(src); // adjust Java long to machine word
5478 Node* base = _gvn.transform(new CastX2PNode(src));
5479 offset = ConvI2X(offset);
5480
5481 // 'src_start' points to src array + scaled offset
5482 Node* src_start = basic_plus_adr(top(), base, offset);
5483
5484 // Call the stub.
5485 address stubAddr = StubRoutines::updateBytesCRC32();
5486 const char *stubName = "updateBytesCRC32";
5487
5488 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5489 stubAddr, stubName, TypePtr::BOTTOM,
5490 crc, src_start, length);
5491 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5492 set_result(result);
5493 return true;
5494 }
5495
5496 //------------------------------get_table_from_crc32c_class-----------------------
5497 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5498 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5499 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5500
5501 return table;
5502 }
5503
5504 //------------------------------inline_updateBytesCRC32C-----------------------
5505 //
5506 // Calculate CRC32C for byte[] array.
5507 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5508 //
5509 bool LibraryCallKit::inline_updateBytesCRC32C() {
5510 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5511 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5512 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5513 // no receiver since it is a static method
5514 Node* crc = argument(0); // type: int
5515 Node* src = argument(1); // type: oop
5516 Node* offset = argument(2); // type: int
5517 Node* end = argument(3); // type: int
5518
5519 Node* length = _gvn.transform(new SubINode(end, offset));
5520
5521 const Type* src_type = src->Value(&_gvn);
5522 const TypeAryPtr* top_src = src_type->isa_aryptr();
5523 if (top_src == NULL || top_src->klass() == NULL) {
5524 // failed array check
5525 return false;
5526 }
5527
5528 // Figure out the size and type of the elements we will be copying.
5529 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5530 if (src_elem != T_BYTE) {
5531 return false;
5532 }
5533
5534 // 'src_start' points to src array + scaled offset
5535 src = must_be_not_null(src, true);
5536 Node* src_start = array_element_address(src, offset, src_elem);
5537
5538 // static final int[] byteTable in class CRC32C
5539 Node* table = get_table_from_crc32c_class(callee()->holder());
5540 table = must_be_not_null(table, true);
5541 Node* table_start = array_element_address(table, intcon(0), T_INT);
5542
5543 // We assume that range check is done by caller.
5544 // TODO: generate range check (offset+length < src.length) in debug VM.
5545
5546 // Call the stub.
5547 address stubAddr = StubRoutines::updateBytesCRC32C();
5548 const char *stubName = "updateBytesCRC32C";
5549
5550 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5551 stubAddr, stubName, TypePtr::BOTTOM,
5552 crc, src_start, length, table_start);
5553 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5554 set_result(result);
5555 return true;
5556 }
5557
5558 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5559 //
5560 // Calculate CRC32C for DirectByteBuffer.
5561 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5562 //
5563 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5564 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5565 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5566 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5567 // no receiver since it is a static method
5568 Node* crc = argument(0); // type: int
5569 Node* src = argument(1); // type: long
5570 Node* offset = argument(3); // type: int
5571 Node* end = argument(4); // type: int
5572
5573 Node* length = _gvn.transform(new SubINode(end, offset));
5574
5575 src = ConvL2X(src); // adjust Java long to machine word
5576 Node* base = _gvn.transform(new CastX2PNode(src));
5577 offset = ConvI2X(offset);
5578
5579 // 'src_start' points to src array + scaled offset
5580 Node* src_start = basic_plus_adr(top(), base, offset);
5581
5582 // static final int[] byteTable in class CRC32C
5583 Node* table = get_table_from_crc32c_class(callee()->holder());
5584 table = must_be_not_null(table, true);
5585 Node* table_start = array_element_address(table, intcon(0), T_INT);
5586
5587 // Call the stub.
5588 address stubAddr = StubRoutines::updateBytesCRC32C();
5589 const char *stubName = "updateBytesCRC32C";
5590
5591 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5592 stubAddr, stubName, TypePtr::BOTTOM,
5593 crc, src_start, length, table_start);
5594 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5595 set_result(result);
5596 return true;
5597 }
5598
5599 //------------------------------inline_updateBytesAdler32----------------------
5600 //
5601 // Calculate Adler32 checksum for byte[] array.
5602 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5603 //
5604 bool LibraryCallKit::inline_updateBytesAdler32() {
5605 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5606 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5607 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5608 // no receiver since it is static method
5609 Node* crc = argument(0); // type: int
5610 Node* src = argument(1); // type: oop
5611 Node* offset = argument(2); // type: int
5612 Node* length = argument(3); // type: int
5613
5614 const Type* src_type = src->Value(&_gvn);
5615 const TypeAryPtr* top_src = src_type->isa_aryptr();
5616 if (top_src == NULL || top_src->klass() == NULL) {
5617 // failed array check
5618 return false;
5619 }
5620
5621 // Figure out the size and type of the elements we will be copying.
5622 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5623 if (src_elem != T_BYTE) {
5624 return false;
5625 }
5626
5627 // 'src_start' points to src array + scaled offset
5628 Node* src_start = array_element_address(src, offset, src_elem);
5629
5630 // We assume that range check is done by caller.
5631 // TODO: generate range check (offset+length < src.length) in debug VM.
5632
5633 // Call the stub.
5634 address stubAddr = StubRoutines::updateBytesAdler32();
5635 const char *stubName = "updateBytesAdler32";
5636
5637 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5638 stubAddr, stubName, TypePtr::BOTTOM,
5639 crc, src_start, length);
5640 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5641 set_result(result);
5642 return true;
5643 }
5644
5645 //------------------------------inline_updateByteBufferAdler32---------------
5646 //
5647 // Calculate Adler32 checksum for DirectByteBuffer.
5648 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5649 //
5650 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5651 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5652 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5653 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5654 // no receiver since it is static method
5655 Node* crc = argument(0); // type: int
5656 Node* src = argument(1); // type: long
5657 Node* offset = argument(3); // type: int
5658 Node* length = argument(4); // type: int
5659
5660 src = ConvL2X(src); // adjust Java long to machine word
5661 Node* base = _gvn.transform(new CastX2PNode(src));
5662 offset = ConvI2X(offset);
5663
5664 // 'src_start' points to src array + scaled offset
5665 Node* src_start = basic_plus_adr(top(), base, offset);
5666
5667 // Call the stub.
5668 address stubAddr = StubRoutines::updateBytesAdler32();
5669 const char *stubName = "updateBytesAdler32";
5670
5671 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5672 stubAddr, stubName, TypePtr::BOTTOM,
5673 crc, src_start, length);
5674
5675 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5676 set_result(result);
5677 return true;
5678 }
5679
5680 //----------------------------inline_reference_get----------------------------
5681 // public T java.lang.ref.Reference.get();
5682 bool LibraryCallKit::inline_reference_get() {
5683 const int referent_offset = java_lang_ref_Reference::referent_offset;
5684 guarantee(referent_offset > 0, "should have already been set");
5685
5686 // Get the argument:
5687 Node* reference_obj = null_check_receiver();
5688 if (stopped()) return true;
5689
5690 const TypeInstPtr* tinst = _gvn.type(reference_obj)->isa_instptr();
5691 assert(tinst != NULL, "obj is null");
5692 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5693 ciInstanceKlass* referenceKlass = tinst->klass()->as_instance_klass();
5694 ciField* field = referenceKlass->get_field_by_name(ciSymbol::make("referent"),
5695 ciSymbol::make("Ljava/lang/Object;"),
5696 false);
5697 assert (field != NULL, "undefined field");
5698
5699 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5700 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5701
5702 ciInstanceKlass* klass = env()->Object_klass();
5703 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5704
5705 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
5706 Node* result = access_load_at(reference_obj, adr, adr_type, object_type, T_OBJECT, decorators);
5707 // Add memory barrier to prevent commoning reads from this field
5708 // across safepoint since GC can change its value.
5709 insert_mem_bar(Op_MemBarCPUOrder);
5710
5711 set_result(result);
5712 return true;
5713 }
5714
5715
5716 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5717 bool is_exact=true, bool is_static=false,
5718 ciInstanceKlass * fromKls=NULL) {
5719 if (fromKls == NULL) {
5720 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5721 assert(tinst != NULL, "obj is null");
5722 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5723 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5724 fromKls = tinst->klass()->as_instance_klass();
5725 } else {
5726 assert(is_static, "only for static field access");
5727 }
5728 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5729 ciSymbol::make(fieldTypeString),
5730 is_static);
5731
5732 assert (field != NULL, "undefined field");
5733 if (field == NULL) return (Node *) NULL;
5734
5735 if (is_static) {
5736 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5737 fromObj = makecon(tip);
5738 }
5739
5740 // Next code copied from Parse::do_get_xxx():
5741
5742 // Compute address and memory type.
5743 int offset = field->offset_in_bytes();
5744 bool is_vol = field->is_volatile();
5745 ciType* field_klass = field->type();
5746 assert(field_klass->is_loaded(), "should be loaded");
5747 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5748 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5749 BasicType bt = field->layout_type();
5750
5751 // Build the resultant type of the load
5752 const Type *type;
5753 if (bt == T_OBJECT) {
5754 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5755 } else {
5756 type = Type::get_const_basic_type(bt);
5757 }
5758
5759 DecoratorSet decorators = IN_HEAP;
5760
5761 if (is_vol) {
5762 decorators |= MO_SEQ_CST;
5763 }
5764
5765 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
5766 }
5767
5768 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5769 bool is_exact = true, bool is_static = false,
5770 ciInstanceKlass * fromKls = NULL) {
5771 if (fromKls == NULL) {
5772 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5773 assert(tinst != NULL, "obj is null");
5774 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5775 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5776 fromKls = tinst->klass()->as_instance_klass();
5777 }
5778 else {
5779 assert(is_static, "only for static field access");
5780 }
5781 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5782 ciSymbol::make(fieldTypeString),
5783 is_static);
5784
5785 assert(field != NULL, "undefined field");
5786 assert(!field->is_volatile(), "not defined for volatile fields");
5787
5788 if (is_static) {
5789 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5790 fromObj = makecon(tip);
5791 }
5792
5793 // Next code copied from Parse::do_get_xxx():
5794
5795 // Compute address and memory type.
5796 int offset = field->offset_in_bytes();
5797 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5798
5799 return adr;
5800 }
5801
5802 //------------------------------inline_aescrypt_Block-----------------------
5803 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
5804 address stubAddr = NULL;
5805 const char *stubName;
5806 assert(UseAES, "need AES instruction support");
5807
5808 switch(id) {
5809 case vmIntrinsics::_aescrypt_encryptBlock:
5810 stubAddr = StubRoutines::aescrypt_encryptBlock();
5811 stubName = "aescrypt_encryptBlock";
5812 break;
5813 case vmIntrinsics::_aescrypt_decryptBlock:
5814 stubAddr = StubRoutines::aescrypt_decryptBlock();
5815 stubName = "aescrypt_decryptBlock";
5816 break;
5817 default:
5818 break;
5819 }
5820 if (stubAddr == NULL) return false;
5821
5822 Node* aescrypt_object = argument(0);
5823 Node* src = argument(1);
5824 Node* src_offset = argument(2);
5825 Node* dest = argument(3);
5826 Node* dest_offset = argument(4);
5827
5828 src = must_be_not_null(src, true);
5829 dest = must_be_not_null(dest, true);
5830
5831 // (1) src and dest are arrays.
5832 const Type* src_type = src->Value(&_gvn);
5833 const Type* dest_type = dest->Value(&_gvn);
5834 const TypeAryPtr* top_src = src_type->isa_aryptr();
5835 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5836 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5837
5838 // for the quick and dirty code we will skip all the checks.
5839 // we are just trying to get the call to be generated.
5840 Node* src_start = src;
5841 Node* dest_start = dest;
5842 if (src_offset != NULL || dest_offset != NULL) {
5843 assert(src_offset != NULL && dest_offset != NULL, "");
5844 src_start = array_element_address(src, src_offset, T_BYTE);
5845 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5846 }
5847
5848 // now need to get the start of its expanded key array
5849 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5850 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5851 if (k_start == NULL) return false;
5852
5853 if (Matcher::pass_original_key_for_aes()) {
5854 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5855 // compatibility issues between Java key expansion and SPARC crypto instructions
5856 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5857 if (original_k_start == NULL) return false;
5858
5859 // Call the stub.
5860 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5861 stubAddr, stubName, TypePtr::BOTTOM,
5862 src_start, dest_start, k_start, original_k_start);
5863 } else {
5864 // Call the stub.
5865 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5866 stubAddr, stubName, TypePtr::BOTTOM,
5867 src_start, dest_start, k_start);
5868 }
5869
5870 return true;
5871 }
5872
5873 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
5874 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
5875 address stubAddr = NULL;
5876 const char *stubName = NULL;
5877
5878 assert(UseAES, "need AES instruction support");
5879
5880 switch(id) {
5881 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
5882 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
5883 stubName = "cipherBlockChaining_encryptAESCrypt";
5884 break;
5885 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
5886 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
5887 stubName = "cipherBlockChaining_decryptAESCrypt";
5888 break;
5889 default:
5890 break;
5891 }
5892 if (stubAddr == NULL) return false;
5893
5894 Node* cipherBlockChaining_object = argument(0);
5895 Node* src = argument(1);
5896 Node* src_offset = argument(2);
5897 Node* len = argument(3);
5898 Node* dest = argument(4);
5899 Node* dest_offset = argument(5);
5900
5901 src = must_be_not_null(src, false);
5902 dest = must_be_not_null(dest, false);
5903
5904 // (1) src and dest are arrays.
5905 const Type* src_type = src->Value(&_gvn);
5906 const Type* dest_type = dest->Value(&_gvn);
5907 const TypeAryPtr* top_src = src_type->isa_aryptr();
5908 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5909 assert (top_src != NULL && top_src->klass() != NULL
5910 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5911
5912 // checks are the responsibility of the caller
5913 Node* src_start = src;
5914 Node* dest_start = dest;
5915 if (src_offset != NULL || dest_offset != NULL) {
5916 assert(src_offset != NULL && dest_offset != NULL, "");
5917 src_start = array_element_address(src, src_offset, T_BYTE);
5918 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5919 }
5920
5921 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
5922 // (because of the predicated logic executed earlier).
5923 // so we cast it here safely.
5924 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5925
5926 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
5927 if (embeddedCipherObj == NULL) return false;
5928
5929 // cast it to what we know it will be at runtime
5930 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
5931 assert(tinst != NULL, "CBC obj is null");
5932 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
5933 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5934 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
5935
5936 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
5937 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
5938 const TypeOopPtr* xtype = aklass->as_instance_type();
5939 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
5940 aescrypt_object = _gvn.transform(aescrypt_object);
5941
5942 // we need to get the start of the aescrypt_object's expanded key array
5943 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5944 if (k_start == NULL) return false;
5945
5946 // similarly, get the start address of the r vector
5947 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
5948 if (objRvec == NULL) return false;
5949 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
5950
5951 Node* cbcCrypt;
5952 if (Matcher::pass_original_key_for_aes()) {
5953 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5954 // compatibility issues between Java key expansion and SPARC crypto instructions
5955 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5956 if (original_k_start == NULL) return false;
5957
5958 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
5959 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5960 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5961 stubAddr, stubName, TypePtr::BOTTOM,
5962 src_start, dest_start, k_start, r_start, len, original_k_start);
5963 } else {
5964 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
5965 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5966 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5967 stubAddr, stubName, TypePtr::BOTTOM,
5968 src_start, dest_start, k_start, r_start, len);
5969 }
5970
5971 // return cipher length (int)
5972 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
5973 set_result(retvalue);
5974 return true;
5975 }
5976
5977 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
5978 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
5979 address stubAddr = NULL;
5980 const char *stubName = NULL;
5981
5982 assert(UseAES, "need AES instruction support");
5983
5984 switch (id) {
5985 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
5986 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
5987 stubName = "electronicCodeBook_encryptAESCrypt";
5988 break;
5989 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
5990 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
5991 stubName = "electronicCodeBook_decryptAESCrypt";
5992 break;
5993 default:
5994 break;
5995 }
5996
5997 if (stubAddr == NULL) return false;
5998
5999 Node* electronicCodeBook_object = argument(0);
6000 Node* src = argument(1);
6001 Node* src_offset = argument(2);
6002 Node* len = argument(3);
6003 Node* dest = argument(4);
6004 Node* dest_offset = argument(5);
6005
6006 // (1) src and dest are arrays.
6007 const Type* src_type = src->Value(&_gvn);
6008 const Type* dest_type = dest->Value(&_gvn);
6009 const TypeAryPtr* top_src = src_type->isa_aryptr();
6010 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6011 assert(top_src != NULL && top_src->klass() != NULL
6012 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6013
6014 // checks are the responsibility of the caller
6015 Node* src_start = src;
6016 Node* dest_start = dest;
6017 if (src_offset != NULL || dest_offset != NULL) {
6018 assert(src_offset != NULL && dest_offset != NULL, "");
6019 src_start = array_element_address(src, src_offset, T_BYTE);
6020 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6021 }
6022
6023 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6024 // (because of the predicated logic executed earlier).
6025 // so we cast it here safely.
6026 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6027
6028 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6029 if (embeddedCipherObj == NULL) return false;
6030
6031 // cast it to what we know it will be at runtime
6032 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6033 assert(tinst != NULL, "ECB obj is null");
6034 assert(tinst->klass()->is_loaded(), "ECB obj is not loaded");
6035 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6036 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6037
6038 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6039 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6040 const TypeOopPtr* xtype = aklass->as_instance_type();
6041 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6042 aescrypt_object = _gvn.transform(aescrypt_object);
6043
6044 // we need to get the start of the aescrypt_object's expanded key array
6045 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6046 if (k_start == NULL) return false;
6047
6048 Node* ecbCrypt;
6049 if (Matcher::pass_original_key_for_aes()) {
6050 // no SPARC version for AES/ECB intrinsics now.
6051 return false;
6052 }
6053 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6054 ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6055 OptoRuntime::electronicCodeBook_aescrypt_Type(),
6056 stubAddr, stubName, TypePtr::BOTTOM,
6057 src_start, dest_start, k_start, len);
6058
6059 // return cipher length (int)
6060 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6061 set_result(retvalue);
6062 return true;
6063 }
6064
6065 //------------------------------inline_counterMode_AESCrypt-----------------------
6066 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6067 assert(UseAES, "need AES instruction support");
6068 if (!UseAESCTRIntrinsics) return false;
6069
6070 address stubAddr = NULL;
6071 const char *stubName = NULL;
6072 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6073 stubAddr = StubRoutines::counterMode_AESCrypt();
6074 stubName = "counterMode_AESCrypt";
6075 }
6076 if (stubAddr == NULL) return false;
6077
6078 Node* counterMode_object = argument(0);
6079 Node* src = argument(1);
6080 Node* src_offset = argument(2);
6081 Node* len = argument(3);
6082 Node* dest = argument(4);
6083 Node* dest_offset = argument(5);
6084
6085 // (1) src and dest are arrays.
6086 const Type* src_type = src->Value(&_gvn);
6087 const Type* dest_type = dest->Value(&_gvn);
6088 const TypeAryPtr* top_src = src_type->isa_aryptr();
6089 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6090 assert(top_src != NULL && top_src->klass() != NULL &&
6091 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6092
6093 // checks are the responsibility of the caller
6094 Node* src_start = src;
6095 Node* dest_start = dest;
6096 if (src_offset != NULL || dest_offset != NULL) {
6097 assert(src_offset != NULL && dest_offset != NULL, "");
6098 src_start = array_element_address(src, src_offset, T_BYTE);
6099 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6100 }
6101
6102 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6103 // (because of the predicated logic executed earlier).
6104 // so we cast it here safely.
6105 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6106 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6107 if (embeddedCipherObj == NULL) return false;
6108 // cast it to what we know it will be at runtime
6109 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6110 assert(tinst != NULL, "CTR obj is null");
6111 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6112 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6113 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6114 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6115 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6116 const TypeOopPtr* xtype = aklass->as_instance_type();
6117 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6118 aescrypt_object = _gvn.transform(aescrypt_object);
6119 // we need to get the start of the aescrypt_object's expanded key array
6120 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6121 if (k_start == NULL) return false;
6122 // similarly, get the start address of the r vector
6123 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6124 if (obj_counter == NULL) return false;
6125 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6126
6127 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6128 if (saved_encCounter == NULL) return false;
6129 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6130 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6131
6132 Node* ctrCrypt;
6133 if (Matcher::pass_original_key_for_aes()) {
6134 // no SPARC version for AES/CTR intrinsics now.
6135 return false;
6136 }
6137 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6138 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6139 OptoRuntime::counterMode_aescrypt_Type(),
6140 stubAddr, stubName, TypePtr::BOTTOM,
6141 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6142
6143 // return cipher length (int)
6144 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6145 set_result(retvalue);
6146 return true;
6147 }
6148
6149 //------------------------------get_key_start_from_aescrypt_object-----------------------
6150 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6151 #if defined(PPC64) || defined(S390)
6152 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6153 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6154 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6155 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6156 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6157 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6158 if (objSessionK == NULL) {
6159 return (Node *) NULL;
6160 }
6161 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6162 #else
6163 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6164 #endif // PPC64
6165 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6166 if (objAESCryptKey == NULL) return (Node *) NULL;
6167
6168 // now have the array, need to get the start address of the K array
6169 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6170 return k_start;
6171 }
6172
6173 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6174 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6175 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6176 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6177 if (objAESCryptKey == NULL) return (Node *) NULL;
6178
6179 // now have the array, need to get the start address of the lastKey array
6180 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6181 return original_k_start;
6182 }
6183
6184 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6185 // Return node representing slow path of predicate check.
6186 // the pseudo code we want to emulate with this predicate is:
6187 // for encryption:
6188 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6189 // for decryption:
6190 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6191 // note cipher==plain is more conservative than the original java code but that's OK
6192 //
6193 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6194 // The receiver was checked for NULL already.
6195 Node* objCBC = argument(0);
6196
6197 Node* src = argument(1);
6198 Node* dest = argument(4);
6199
6200 // Load embeddedCipher field of CipherBlockChaining object.
6201 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6202
6203 // get AESCrypt klass for instanceOf check
6204 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6205 // will have same classloader as CipherBlockChaining object
6206 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6207 assert(tinst != NULL, "CBCobj is null");
6208 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6209
6210 // we want to do an instanceof comparison against the AESCrypt class
6211 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6212 if (!klass_AESCrypt->is_loaded()) {
6213 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6214 Node* ctrl = control();
6215 set_control(top()); // no regular fast path
6216 return ctrl;
6217 }
6218
6219 src = must_be_not_null(src, true);
6220 dest = must_be_not_null(dest, true);
6221
6222 // Resolve oops to stable for CmpP below.
6223 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6224
6225 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6226 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6227 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6228
6229 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6230
6231 // for encryption, we are done
6232 if (!decrypting)
6233 return instof_false; // even if it is NULL
6234
6235 // for decryption, we need to add a further check to avoid
6236 // taking the intrinsic path when cipher and plain are the same
6237 // see the original java code for why.
6238 RegionNode* region = new RegionNode(3);
6239 region->init_req(1, instof_false);
6240
6241 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6242 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6243 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6244 region->init_req(2, src_dest_conjoint);
6245
6246 record_for_igvn(region);
6247 return _gvn.transform(region);
6248 }
6249
6250 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
6251 // Return node representing slow path of predicate check.
6252 // the pseudo code we want to emulate with this predicate is:
6253 // for encryption:
6254 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6255 // for decryption:
6256 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6257 // note cipher==plain is more conservative than the original java code but that's OK
6258 //
6259 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
6260 // The receiver was checked for NULL already.
6261 Node* objECB = argument(0);
6262
6263 // Load embeddedCipher field of ElectronicCodeBook object.
6264 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6265
6266 // get AESCrypt klass for instanceOf check
6267 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6268 // will have same classloader as ElectronicCodeBook object
6269 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
6270 assert(tinst != NULL, "ECBobj is null");
6271 assert(tinst->klass()->is_loaded(), "ECBobj is not loaded");
6272
6273 // we want to do an instanceof comparison against the AESCrypt class
6274 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6275 if (!klass_AESCrypt->is_loaded()) {
6276 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6277 Node* ctrl = control();
6278 set_control(top()); // no regular fast path
6279 return ctrl;
6280 }
6281 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6282
6283 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6284 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6285 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6286
6287 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6288
6289 // for encryption, we are done
6290 if (!decrypting)
6291 return instof_false; // even if it is NULL
6292
6293 // for decryption, we need to add a further check to avoid
6294 // taking the intrinsic path when cipher and plain are the same
6295 // see the original java code for why.
6296 RegionNode* region = new RegionNode(3);
6297 region->init_req(1, instof_false);
6298 Node* src = argument(1);
6299 Node* dest = argument(4);
6300 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6301 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6302 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6303 region->init_req(2, src_dest_conjoint);
6304
6305 record_for_igvn(region);
6306 return _gvn.transform(region);
6307 }
6308
6309 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6310 // Return node representing slow path of predicate check.
6311 // the pseudo code we want to emulate with this predicate is:
6312 // for encryption:
6313 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6314 // for decryption:
6315 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6316 // note cipher==plain is more conservative than the original java code but that's OK
6317 //
6318
6319 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6320 // The receiver was checked for NULL already.
6321 Node* objCTR = argument(0);
6322
6323 // Load embeddedCipher field of CipherBlockChaining object.
6324 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6325
6326 // get AESCrypt klass for instanceOf check
6327 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6328 // will have same classloader as CipherBlockChaining object
6329 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6330 assert(tinst != NULL, "CTRobj is null");
6331 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6332
6333 // we want to do an instanceof comparison against the AESCrypt class
6334 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6335 if (!klass_AESCrypt->is_loaded()) {
6336 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6337 Node* ctrl = control();
6338 set_control(top()); // no regular fast path
6339 return ctrl;
6340 }
6341
6342 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6343 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6344 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6345 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6346 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6347
6348 return instof_false; // even if it is NULL
6349 }
6350
6351 //------------------------------inline_ghash_processBlocks
6352 bool LibraryCallKit::inline_ghash_processBlocks() {
6353 address stubAddr;
6354 const char *stubName;
6355 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6356
6357 stubAddr = StubRoutines::ghash_processBlocks();
6358 stubName = "ghash_processBlocks";
6359
6360 Node* data = argument(0);
6361 Node* offset = argument(1);
6362 Node* len = argument(2);
6363 Node* state = argument(3);
6364 Node* subkeyH = argument(4);
6365
6366 state = must_be_not_null(state, true);
6367 subkeyH = must_be_not_null(subkeyH, true);
6368 data = must_be_not_null(data, true);
6369
6370 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6371 assert(state_start, "state is NULL");
6372 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6373 assert(subkeyH_start, "subkeyH is NULL");
6374 Node* data_start = array_element_address(data, offset, T_BYTE);
6375 assert(data_start, "data is NULL");
6376
6377 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6378 OptoRuntime::ghash_processBlocks_Type(),
6379 stubAddr, stubName, TypePtr::BOTTOM,
6380 state_start, subkeyH_start, data_start, len);
6381 return true;
6382 }
6383
6384 bool LibraryCallKit::inline_base64_encodeBlock() {
6385 address stubAddr;
6386 const char *stubName;
6387 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6388 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6389 stubAddr = StubRoutines::base64_encodeBlock();
6390 stubName = "encodeBlock";
6391
6392 if (!stubAddr) return false;
6393 Node* base64obj = argument(0);
6394 Node* src = argument(1);
6395 Node* offset = argument(2);
6396 Node* len = argument(3);
6397 Node* dest = argument(4);
6398 Node* dp = argument(5);
6399 Node* isURL = argument(6);
6400
6401 src = must_be_not_null(src, true);
6402 dest = must_be_not_null(dest, true);
6403
6404 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6405 assert(src_start, "source array is NULL");
6406 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6407 assert(dest_start, "destination array is NULL");
6408
6409 Node* base64 = make_runtime_call(RC_LEAF,
6410 OptoRuntime::base64_encodeBlock_Type(),
6411 stubAddr, stubName, TypePtr::BOTTOM,
6412 src_start, offset, len, dest_start, dp, isURL);
6413 return true;
6414 }
6415
6416 //------------------------------inline_sha_implCompress-----------------------
6417 //
6418 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6419 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6420 //
6421 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6422 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6423 //
6424 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6425 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6426 //
6427 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6428 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6429
6430 Node* sha_obj = argument(0);
6431 Node* src = argument(1); // type oop
6432 Node* ofs = argument(2); // type int
6433
6434 const Type* src_type = src->Value(&_gvn);
6435 const TypeAryPtr* top_src = src_type->isa_aryptr();
6436 if (top_src == NULL || top_src->klass() == NULL) {
6437 // failed array check
6438 return false;
6439 }
6440 // Figure out the size and type of the elements we will be copying.
6441 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6442 if (src_elem != T_BYTE) {
6443 return false;
6444 }
6445 // 'src_start' points to src array + offset
6446 src = must_be_not_null(src, true);
6447 Node* src_start = array_element_address(src, ofs, src_elem);
6448 Node* state = NULL;
6449 address stubAddr;
6450 const char *stubName;
6451
6452 switch(id) {
6453 case vmIntrinsics::_sha_implCompress:
6454 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6455 state = get_state_from_sha_object(sha_obj);
6456 stubAddr = StubRoutines::sha1_implCompress();
6457 stubName = "sha1_implCompress";
6458 break;
6459 case vmIntrinsics::_sha2_implCompress:
6460 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6461 state = get_state_from_sha_object(sha_obj);
6462 stubAddr = StubRoutines::sha256_implCompress();
6463 stubName = "sha256_implCompress";
6464 break;
6465 case vmIntrinsics::_sha5_implCompress:
6466 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6467 state = get_state_from_sha5_object(sha_obj);
6468 stubAddr = StubRoutines::sha512_implCompress();
6469 stubName = "sha512_implCompress";
6470 break;
6471 default:
6472 fatal_unexpected_iid(id);
6473 return false;
6474 }
6475 if (state == NULL) return false;
6476
6477 assert(stubAddr != NULL, "Stub is generated");
6478 if (stubAddr == NULL) return false;
6479
6480 // Call the stub.
6481 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6482 stubAddr, stubName, TypePtr::BOTTOM,
6483 src_start, state);
6484
6485 return true;
6486 }
6487
6488 //------------------------------inline_digestBase_implCompressMB-----------------------
6489 //
6490 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6491 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6492 //
6493 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6494 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6495 "need SHA1/SHA256/SHA512 instruction support");
6496 assert((uint)predicate < 3, "sanity");
6497 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6498
6499 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6500 Node* src = argument(1); // byte[] array
6501 Node* ofs = argument(2); // type int
6502 Node* limit = argument(3); // type int
6503
6504 const Type* src_type = src->Value(&_gvn);
6505 const TypeAryPtr* top_src = src_type->isa_aryptr();
6506 if (top_src == NULL || top_src->klass() == NULL) {
6507 // failed array check
6508 return false;
6509 }
6510 // Figure out the size and type of the elements we will be copying.
6511 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6512 if (src_elem != T_BYTE) {
6513 return false;
6514 }
6515 // 'src_start' points to src array + offset
6516 src = must_be_not_null(src, false);
6517 Node* src_start = array_element_address(src, ofs, src_elem);
6518
6519 const char* klass_SHA_name = NULL;
6520 const char* stub_name = NULL;
6521 address stub_addr = NULL;
6522 bool long_state = false;
6523
6524 switch (predicate) {
6525 case 0:
6526 if (UseSHA1Intrinsics) {
6527 klass_SHA_name = "sun/security/provider/SHA";
6528 stub_name = "sha1_implCompressMB";
6529 stub_addr = StubRoutines::sha1_implCompressMB();
6530 }
6531 break;
6532 case 1:
6533 if (UseSHA256Intrinsics) {
6534 klass_SHA_name = "sun/security/provider/SHA2";
6535 stub_name = "sha256_implCompressMB";
6536 stub_addr = StubRoutines::sha256_implCompressMB();
6537 }
6538 break;
6539 case 2:
6540 if (UseSHA512Intrinsics) {
6541 klass_SHA_name = "sun/security/provider/SHA5";
6542 stub_name = "sha512_implCompressMB";
6543 stub_addr = StubRoutines::sha512_implCompressMB();
6544 long_state = true;
6545 }
6546 break;
6547 default:
6548 fatal("unknown SHA intrinsic predicate: %d", predicate);
6549 }
6550 if (klass_SHA_name != NULL) {
6551 assert(stub_addr != NULL, "Stub is generated");
6552 if (stub_addr == NULL) return false;
6553
6554 // get DigestBase klass to lookup for SHA klass
6555 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6556 assert(tinst != NULL, "digestBase_obj is not instance???");
6557 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6558
6559 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6560 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6561 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6562 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6563 }
6564 return false;
6565 }
6566 //------------------------------inline_sha_implCompressMB-----------------------
6567 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6568 bool long_state, address stubAddr, const char *stubName,
6569 Node* src_start, Node* ofs, Node* limit) {
6570 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6571 const TypeOopPtr* xtype = aklass->as_instance_type();
6572 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6573 sha_obj = _gvn.transform(sha_obj);
6574
6575 Node* state;
6576 if (long_state) {
6577 state = get_state_from_sha5_object(sha_obj);
6578 } else {
6579 state = get_state_from_sha_object(sha_obj);
6580 }
6581 if (state == NULL) return false;
6582
6583 // Call the stub.
6584 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6585 OptoRuntime::digestBase_implCompressMB_Type(),
6586 stubAddr, stubName, TypePtr::BOTTOM,
6587 src_start, state, ofs, limit);
6588 // return ofs (int)
6589 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6590 set_result(result);
6591
6592 return true;
6593 }
6594
6595 //------------------------------get_state_from_sha_object-----------------------
6596 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6597 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6598 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6599 if (sha_state == NULL) return (Node *) NULL;
6600
6601 // now have the array, need to get the start address of the state array
6602 Node* state = array_element_address(sha_state, intcon(0), T_INT);
6603 return state;
6604 }
6605
6606 //------------------------------get_state_from_sha5_object-----------------------
6607 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6608 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6609 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6610 if (sha_state == NULL) return (Node *) NULL;
6611
6612 // now have the array, need to get the start address of the state array
6613 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6614 return state;
6615 }
6616
6617 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6618 // Return node representing slow path of predicate check.
6619 // the pseudo code we want to emulate with this predicate is:
6620 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6621 //
6622 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6623 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6624 "need SHA1/SHA256/SHA512 instruction support");
6625 assert((uint)predicate < 3, "sanity");
6626
6627 // The receiver was checked for NULL already.
6628 Node* digestBaseObj = argument(0);
6629
6630 // get DigestBase klass for instanceOf check
6631 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6632 assert(tinst != NULL, "digestBaseObj is null");
6633 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6634
6635 const char* klass_SHA_name = NULL;
6636 switch (predicate) {
6637 case 0:
6638 if (UseSHA1Intrinsics) {
6639 // we want to do an instanceof comparison against the SHA class
6640 klass_SHA_name = "sun/security/provider/SHA";
6641 }
6642 break;
6643 case 1:
6644 if (UseSHA256Intrinsics) {
6645 // we want to do an instanceof comparison against the SHA2 class
6646 klass_SHA_name = "sun/security/provider/SHA2";
6647 }
6648 break;
6649 case 2:
6650 if (UseSHA512Intrinsics) {
6651 // we want to do an instanceof comparison against the SHA5 class
6652 klass_SHA_name = "sun/security/provider/SHA5";
6653 }
6654 break;
6655 default:
6656 fatal("unknown SHA intrinsic predicate: %d", predicate);
6657 }
6658
6659 ciKlass* klass_SHA = NULL;
6660 if (klass_SHA_name != NULL) {
6661 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6662 }
6663 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6664 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6665 Node* ctrl = control();
6666 set_control(top()); // no intrinsic path
6667 return ctrl;
6668 }
6669 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6670
6671 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6672 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6673 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6674 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6675
6676 return instof_false; // even if it is NULL
6677 }
6678
6679 //-------------inline_fma-----------------------------------
6680 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
6681 Node *a = NULL;
6682 Node *b = NULL;
6683 Node *c = NULL;
6684 Node* result = NULL;
6685 switch (id) {
6686 case vmIntrinsics::_fmaD:
6687 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
6688 // no receiver since it is static method
6689 a = round_double_node(argument(0));
6690 b = round_double_node(argument(2));
6691 c = round_double_node(argument(4));
6692 result = _gvn.transform(new FmaDNode(control(), a, b, c));
6693 break;
6694 case vmIntrinsics::_fmaF:
6695 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
6696 a = argument(0);
6697 b = argument(1);
6698 c = argument(2);
6699 result = _gvn.transform(new FmaFNode(control(), a, b, c));
6700 break;
6701 default:
6702 fatal_unexpected_iid(id); break;
6703 }
6704 set_result(result);
6705 return true;
6706 }
6707
6708 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
6709 // argument(0) is receiver
6710 Node* codePoint = argument(1);
6711 Node* n = NULL;
6712
6713 switch (id) {
6714 case vmIntrinsics::_isDigit :
6715 n = new DigitNode(control(), codePoint);
6716 break;
6717 case vmIntrinsics::_isLowerCase :
6718 n = new LowerCaseNode(control(), codePoint);
6719 break;
6720 case vmIntrinsics::_isUpperCase :
6721 n = new UpperCaseNode(control(), codePoint);
6722 break;
6723 case vmIntrinsics::_isWhitespace :
6724 n = new WhitespaceNode(control(), codePoint);
6725 break;
6726 default:
6727 fatal_unexpected_iid(id);
6728 }
6729
6730 set_result(_gvn.transform(n));
6731 return true;
6732 }
6733
6734 //------------------------------inline_fp_min_max------------------------------
6735 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
6736 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
6737
6738 // The intrinsic should be used only when the API branches aren't predictable,
6739 // the last one performing the most important comparison. The following heuristic
6740 // uses the branch statistics to eventually bail out if necessary.
6741
6742 ciMethodData *md = callee()->method_data();
6743
6744 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
6745 ciCallProfile cp = caller()->call_profile_at_bci(bci());
6746
6747 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
6748 // Bail out if the call-site didn't contribute enough to the statistics.
6749 return false;
6750 }
6751
6752 uint taken = 0, not_taken = 0;
6753
6754 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
6755 if (p->is_BranchData()) {
6756 taken = ((ciBranchData*)p)->taken();
6757 not_taken = ((ciBranchData*)p)->not_taken();
6758 }
6759 }
6760
6761 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
6762 balance = balance < 0 ? -balance : balance;
6763 if ( balance > 0.2 ) {
6764 // Bail out if the most important branch is predictable enough.
6765 return false;
6766 }
6767 }
6768 */
6769
6770 Node *a = NULL;
6771 Node *b = NULL;
6772 Node *n = NULL;
6773 switch (id) {
6774 case vmIntrinsics::_maxF:
6775 case vmIntrinsics::_minF:
6776 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
6777 a = argument(0);
6778 b = argument(1);
6779 break;
6780 case vmIntrinsics::_maxD:
6781 case vmIntrinsics::_minD:
6782 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
6783 a = round_double_node(argument(0));
6784 b = round_double_node(argument(2));
6785 break;
6786 default:
6787 fatal_unexpected_iid(id);
6788 break;
6789 }
6790 switch (id) {
6791 case vmIntrinsics::_maxF: n = new MaxFNode(a, b); break;
6792 case vmIntrinsics::_minF: n = new MinFNode(a, b); break;
6793 case vmIntrinsics::_maxD: n = new MaxDNode(a, b); break;
6794 case vmIntrinsics::_minD: n = new MinDNode(a, b); break;
6795 default: fatal_unexpected_iid(id); break;
6796 }
6797 set_result(_gvn.transform(n));
6798 return true;
6799 }
6800
6801 bool LibraryCallKit::inline_profileBoolean() {
6802 Node* counts = argument(1);
6803 const TypeAryPtr* ary = NULL;
6804 ciArray* aobj = NULL;
6805 if (counts->is_Con()
6806 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6807 && (aobj = ary->const_oop()->as_array()) != NULL
6808 && (aobj->length() == 2)) {
6809 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6810 jint false_cnt = aobj->element_value(0).as_int();
6811 jint true_cnt = aobj->element_value(1).as_int();
6812
6813 if (C->log() != NULL) {
6814 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6815 false_cnt, true_cnt);
6816 }
6817
6818 if (false_cnt + true_cnt == 0) {
6819 // According to profile, never executed.
6820 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6821 Deoptimization::Action_reinterpret);
6822 return true;
6823 }
6824
6825 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6826 // is a number of each value occurrences.
6827 Node* result = argument(0);
6828 if (false_cnt == 0 || true_cnt == 0) {
6829 // According to profile, one value has been never seen.
6830 int expected_val = (false_cnt == 0) ? 1 : 0;
6831
6832 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6833 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6834
6835 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6836 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6837 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6838
6839 { // Slow path: uncommon trap for never seen value and then reexecute
6840 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6841 // the value has been seen at least once.
6842 PreserveJVMState pjvms(this);
6843 PreserveReexecuteState preexecs(this);
6844 jvms()->set_should_reexecute(true);
6845
6846 set_control(slow_path);
6847 set_i_o(i_o());
6848
6849 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6850 Deoptimization::Action_reinterpret);
6851 }
6852 // The guard for never seen value enables sharpening of the result and
6853 // returning a constant. It allows to eliminate branches on the same value
6854 // later on.
6855 set_control(fast_path);
6856 result = intcon(expected_val);
6857 }
6858 // Stop profiling.
6859 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6860 // By replacing method body with profile data (represented as ProfileBooleanNode
6861 // on IR level) we effectively disable profiling.
6862 // It enables full speed execution once optimized code is generated.
6863 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6864 C->record_for_igvn(profile);
6865 set_result(profile);
6866 return true;
6867 } else {
6868 // Continue profiling.
6869 // Profile data isn't available at the moment. So, execute method's bytecode version.
6870 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6871 // is compiled and counters aren't available since corresponding MethodHandle
6872 // isn't a compile-time constant.
6873 return false;
6874 }
6875 }
6876
6877 bool LibraryCallKit::inline_isCompileConstant() {
6878 Node* n = argument(0);
6879 set_result(n->is_Con() ? intcon(1) : intcon(0));
6880 return true;
6881 }