1 /*
2 * Copyright (c) 1999, 2020, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/systemDictionary.hpp"
29 #include "classfile/vmSymbols.hpp"
30 #include "compiler/compileBroker.hpp"
31 #include "compiler/compileLog.hpp"
32 #include "gc/shared/barrierSet.hpp"
33 #include "jfr/support/jfrIntrinsics.hpp"
34 #include "memory/resourceArea.hpp"
35 #include "oops/klass.inline.hpp"
36 #include "oops/objArrayKlass.hpp"
37 #include "opto/addnode.hpp"
38 #include "opto/arraycopynode.hpp"
39 #include "opto/c2compiler.hpp"
40 #include "opto/callGenerator.hpp"
41 #include "opto/castnode.hpp"
42 #include "opto/cfgnode.hpp"
43 #include "opto/convertnode.hpp"
44 #include "opto/countbitsnode.hpp"
45 #include "opto/intrinsicnode.hpp"
46 #include "opto/idealKit.hpp"
47 #include "opto/mathexactnode.hpp"
48 #include "opto/movenode.hpp"
49 #include "opto/mulnode.hpp"
50 #include "opto/narrowptrnode.hpp"
51 #include "opto/opaquenode.hpp"
52 #include "opto/parse.hpp"
53 #include "opto/runtime.hpp"
54 #include "opto/rootnode.hpp"
55 #include "opto/signum.hpp"
56 #include "opto/subnode.hpp"
57 #include "prims/nativeLookup.hpp"
58 #include "prims/unsafe.hpp"
59 #include "runtime/objectMonitor.hpp"
60 #include "runtime/sharedRuntime.hpp"
61 #include "utilities/macros.hpp"
62 #include "utilities/powerOfTwo.hpp"
63
64 class LibraryIntrinsic : public InlineCallGenerator {
65 // Extend the set of intrinsics known to the runtime:
66 public:
67 private:
68 bool _is_virtual;
69 bool _does_virtual_dispatch;
70 int8_t _predicates_count; // Intrinsic is predicated by several conditions
71 int8_t _last_predicate; // Last generated predicate
72 vmIntrinsics::ID _intrinsic_id;
73
74 public:
75 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
76 : InlineCallGenerator(m),
77 _is_virtual(is_virtual),
78 _does_virtual_dispatch(does_virtual_dispatch),
79 _predicates_count((int8_t)predicates_count),
80 _last_predicate((int8_t)-1),
81 _intrinsic_id(id)
82 {
83 }
84 virtual bool is_intrinsic() const { return true; }
85 virtual bool is_virtual() const { return _is_virtual; }
86 virtual bool is_predicated() const { return _predicates_count > 0; }
87 virtual int predicates_count() const { return _predicates_count; }
88 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; }
89 virtual JVMState* generate(JVMState* jvms);
90 virtual Node* generate_predicate(JVMState* jvms, int predicate);
91 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
92 };
93
94
95 // Local helper class for LibraryIntrinsic:
96 class LibraryCallKit : public GraphKit {
97 private:
98 LibraryIntrinsic* _intrinsic; // the library intrinsic being called
99 Node* _result; // the result node, if any
100 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted
101
102 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type);
103
104 public:
105 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
106 : GraphKit(jvms),
107 _intrinsic(intrinsic),
108 _result(NULL)
109 {
110 // Check if this is a root compile. In that case we don't have a caller.
111 if (!jvms->has_method()) {
112 _reexecute_sp = sp();
113 } else {
114 // Find out how many arguments the interpreter needs when deoptimizing
115 // and save the stack pointer value so it can used by uncommon_trap.
116 // We find the argument count by looking at the declared signature.
117 bool ignored_will_link;
118 ciSignature* declared_signature = NULL;
119 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
120 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
121 _reexecute_sp = sp() + nargs; // "push" arguments back on stack
122 }
123 }
124
125 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
126
127 ciMethod* caller() const { return jvms()->method(); }
128 int bci() const { return jvms()->bci(); }
129 LibraryIntrinsic* intrinsic() const { return _intrinsic; }
130 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); }
131 ciMethod* callee() const { return _intrinsic->method(); }
132
133 bool try_to_inline(int predicate);
134 Node* try_to_predicate(int predicate);
135
136 void push_result() {
137 // Push the result onto the stack.
138 if (!stopped() && result() != NULL) {
139 BasicType bt = result()->bottom_type()->basic_type();
140 push_node(bt, result());
141 }
142 }
143
144 private:
145 void fatal_unexpected_iid(vmIntrinsics::ID iid) {
146 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid));
147 }
148
149 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
150 void set_result(RegionNode* region, PhiNode* value);
151 Node* result() { return _result; }
152
153 virtual int reexecute_sp() { return _reexecute_sp; }
154
155 // Helper functions to inline natives
156 Node* generate_guard(Node* test, RegionNode* region, float true_prob);
157 Node* generate_slow_guard(Node* test, RegionNode* region);
158 Node* generate_fair_guard(Node* test, RegionNode* region);
159 Node* generate_negative_guard(Node* index, RegionNode* region,
160 // resulting CastII of index:
161 Node* *pos_index = NULL);
162 Node* generate_limit_guard(Node* offset, Node* subseq_length,
163 Node* array_length,
164 RegionNode* region);
165 void generate_string_range_check(Node* array, Node* offset,
166 Node* length, bool char_count);
167 Node* generate_current_thread(Node* &tls_output);
168 Node* load_mirror_from_klass(Node* klass);
169 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
170 RegionNode* region, int null_path,
171 int offset);
172 Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
173 RegionNode* region, int null_path) {
174 int offset = java_lang_Class::klass_offset();
175 return load_klass_from_mirror_common(mirror, never_see_null,
176 region, null_path,
177 offset);
178 }
179 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
180 RegionNode* region, int null_path) {
181 int offset = java_lang_Class::array_klass_offset();
182 return load_klass_from_mirror_common(mirror, never_see_null,
183 region, null_path,
184 offset);
185 }
186 Node* generate_access_flags_guard(Node* kls,
187 int modifier_mask, int modifier_bits,
188 RegionNode* region);
189 Node* generate_interface_guard(Node* kls, RegionNode* region);
190 Node* generate_hidden_class_guard(Node* kls, RegionNode* region);
191 Node* generate_array_guard(Node* kls, RegionNode* region) {
192 return generate_array_guard_common(kls, region, false, false);
193 }
194 Node* generate_non_array_guard(Node* kls, RegionNode* region) {
195 return generate_array_guard_common(kls, region, false, true);
196 }
197 Node* generate_objArray_guard(Node* kls, RegionNode* region) {
198 return generate_array_guard_common(kls, region, true, false);
199 }
200 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
201 return generate_array_guard_common(kls, region, true, true);
202 }
203 Node* generate_array_guard_common(Node* kls, RegionNode* region,
204 bool obj_array, bool not_array);
205 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
206 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
207 bool is_virtual = false, bool is_static = false);
208 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
209 return generate_method_call(method_id, false, true);
210 }
211 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
212 return generate_method_call(method_id, true, false);
213 }
214 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
215 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
216
217 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
218 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
219 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
220 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
221 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
222 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
223 bool inline_string_indexOfChar();
224 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
225 bool inline_string_toBytesU();
226 bool inline_string_getCharsU();
227 bool inline_string_copy(bool compress);
228 bool inline_string_char_access(bool is_store);
229 Node* round_double_node(Node* n);
230 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
231 bool inline_math_native(vmIntrinsics::ID id);
232 bool inline_math(vmIntrinsics::ID id);
233 bool inline_double_math(vmIntrinsics::ID id);
234 template <typename OverflowOp>
235 bool inline_math_overflow(Node* arg1, Node* arg2);
236 void inline_math_mathExact(Node* math, Node* test);
237 bool inline_math_addExactI(bool is_increment);
238 bool inline_math_addExactL(bool is_increment);
239 bool inline_math_multiplyExactI();
240 bool inline_math_multiplyExactL();
241 bool inline_math_multiplyHigh();
242 bool inline_math_negateExactI();
243 bool inline_math_negateExactL();
244 bool inline_math_subtractExactI(bool is_decrement);
245 bool inline_math_subtractExactL(bool is_decrement);
246 bool inline_min_max(vmIntrinsics::ID id);
247 bool inline_notify(vmIntrinsics::ID id);
248 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
249 // This returns Type::AnyPtr, RawPtr, or OopPtr.
250 int classify_unsafe_addr(Node* &base, Node* &offset, BasicType type);
251 Node* make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type = T_ILLEGAL, bool can_cast = false);
252
253 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
254 DecoratorSet mo_decorator_for_access_kind(AccessKind kind);
255 bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
256 static bool klass_needs_init_guard(Node* kls);
257 bool inline_unsafe_allocate();
258 bool inline_unsafe_newArray(bool uninitialized);
259 bool inline_unsafe_writeback0();
260 bool inline_unsafe_writebackSync0(bool is_pre);
261 bool inline_unsafe_copyMemory();
262 bool inline_native_currentThread();
263
264 bool inline_native_time_funcs(address method, const char* funcName);
265 #ifdef JFR_HAVE_INTRINSICS
266 bool inline_native_classID();
267 bool inline_native_getEventWriter();
268 #endif
269 bool inline_native_Class_query(vmIntrinsics::ID id);
270 bool inline_native_subtype_check();
271 bool inline_native_getLength();
272 bool inline_array_copyOf(bool is_copyOfRange);
273 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
274 bool inline_preconditions_checkIndex();
275 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array);
276 bool inline_native_clone(bool is_virtual);
277 bool inline_native_Reflection_getCallerClass();
278 // Helper function for inlining native object hash method
279 bool inline_native_hashcode(bool is_virtual, bool is_static);
280 bool inline_native_getClass();
281
282 // Helper functions for inlining arraycopy
283 bool inline_arraycopy();
284 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
285 RegionNode* slow_region);
286 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
287 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp,
288 uint new_idx);
289
290 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
291 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
292 bool inline_unsafe_fence(vmIntrinsics::ID id);
293 bool inline_onspinwait();
294 bool inline_fp_conversions(vmIntrinsics::ID id);
295 bool inline_number_methods(vmIntrinsics::ID id);
296 bool inline_reference_get();
297 bool inline_Class_cast();
298 bool inline_aescrypt_Block(vmIntrinsics::ID id);
299 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
300 bool inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id);
301 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
302 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
303 Node* inline_electronicCodeBook_AESCrypt_predicate(bool decrypting);
304 Node* inline_counterMode_AESCrypt_predicate();
305 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
306 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
307 bool inline_ghash_processBlocks();
308 bool inline_base64_encodeBlock();
309 bool inline_sha_implCompress(vmIntrinsics::ID id);
310 bool inline_digestBase_implCompressMB(int predicate);
311 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
312 bool long_state, address stubAddr, const char *stubName,
313 Node* src_start, Node* ofs, Node* limit);
314 Node* get_state_from_sha_object(Node *sha_object);
315 Node* get_state_from_sha5_object(Node *sha_object);
316 Node* inline_digestBase_implCompressMB_predicate(int predicate);
317 bool inline_encodeISOArray();
318 bool inline_updateCRC32();
319 bool inline_updateBytesCRC32();
320 bool inline_updateByteBufferCRC32();
321 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
322 bool inline_updateBytesCRC32C();
323 bool inline_updateDirectByteBufferCRC32C();
324 bool inline_updateBytesAdler32();
325 bool inline_updateByteBufferAdler32();
326 bool inline_multiplyToLen();
327 bool inline_hasNegatives();
328 bool inline_squareToLen();
329 bool inline_mulAdd();
330 bool inline_montgomeryMultiply();
331 bool inline_montgomerySquare();
332 bool inline_bigIntegerShift(bool isRightShift);
333 bool inline_vectorizedMismatch();
334 bool inline_fma(vmIntrinsics::ID id);
335 bool inline_character_compare(vmIntrinsics::ID id);
336 bool inline_fp_min_max(vmIntrinsics::ID id);
337
338 bool inline_profileBoolean();
339 bool inline_isCompileConstant();
340 void clear_upper_avx() {
341 #ifdef X86
342 if (UseAVX >= 2) {
343 C->set_clear_upper_avx(true);
344 }
345 #endif
346 }
347 };
348
349 //---------------------------make_vm_intrinsic----------------------------
350 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
351 vmIntrinsics::ID id = m->intrinsic_id();
352 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
353
354 if (!m->is_loaded()) {
355 // Do not attempt to inline unloaded methods.
356 return NULL;
357 }
358
359 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
360 bool is_available = false;
361
362 {
363 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
364 // the compiler must transition to '_thread_in_vm' state because both
365 // methods access VM-internal data.
366 VM_ENTRY_MARK;
367 methodHandle mh(THREAD, m->get_Method());
368 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
369 !C->directive()->is_intrinsic_disabled(mh) &&
370 !vmIntrinsics::is_disabled_by_flags(mh);
371
372 }
373
374 if (is_available) {
375 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
376 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
377 return new LibraryIntrinsic(m, is_virtual,
378 vmIntrinsics::predicates_needed(id),
379 vmIntrinsics::does_virtual_dispatch(id),
380 (vmIntrinsics::ID) id);
381 } else {
382 return NULL;
383 }
384 }
385
386 //----------------------register_library_intrinsics-----------------------
387 // Initialize this file's data structures, for each Compile instance.
388 void Compile::register_library_intrinsics() {
389 // Nothing to do here.
390 }
391
392 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
393 LibraryCallKit kit(jvms, this);
394 Compile* C = kit.C;
395 int nodes = C->unique();
396 #ifndef PRODUCT
397 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
398 char buf[1000];
399 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
400 tty->print_cr("Intrinsic %s", str);
401 }
402 #endif
403 ciMethod* callee = kit.callee();
404 const int bci = kit.bci();
405
406 // Try to inline the intrinsic.
407 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
408 kit.try_to_inline(_last_predicate)) {
409 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
410 : "(intrinsic)";
411 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
412 if (C->print_intrinsics() || C->print_inlining()) {
413 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
414 }
415 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
416 if (C->log()) {
417 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
418 vmIntrinsics::name_at(intrinsic_id()),
419 (is_virtual() ? " virtual='1'" : ""),
420 C->unique() - nodes);
421 }
422 // Push the result from the inlined method onto the stack.
423 kit.push_result();
424 C->print_inlining_update(this);
425 return kit.transfer_exceptions_into_jvms();
426 }
427
428 // The intrinsic bailed out
429 if (jvms->has_method()) {
430 // Not a root compile.
431 const char* msg;
432 if (callee->intrinsic_candidate()) {
433 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
434 } else {
435 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
436 : "failed to inline (intrinsic), method not annotated";
437 }
438 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
439 if (C->print_intrinsics() || C->print_inlining()) {
440 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
441 }
442 } else {
443 // Root compile
444 ResourceMark rm;
445 stringStream msg_stream;
446 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
447 vmIntrinsics::name_at(intrinsic_id()),
448 is_virtual() ? " (virtual)" : "", bci);
449 const char *msg = msg_stream.as_string();
450 log_debug(jit, inlining)("%s", msg);
451 if (C->print_intrinsics() || C->print_inlining()) {
452 tty->print("%s", msg);
453 }
454 }
455 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
456 C->print_inlining_update(this);
457 return NULL;
458 }
459
460 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
461 LibraryCallKit kit(jvms, this);
462 Compile* C = kit.C;
463 int nodes = C->unique();
464 _last_predicate = predicate;
465 #ifndef PRODUCT
466 assert(is_predicated() && predicate < predicates_count(), "sanity");
467 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
468 char buf[1000];
469 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
470 tty->print_cr("Predicate for intrinsic %s", str);
471 }
472 #endif
473 ciMethod* callee = kit.callee();
474 const int bci = kit.bci();
475
476 Node* slow_ctl = kit.try_to_predicate(predicate);
477 if (!kit.failing()) {
478 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
479 : "(intrinsic, predicate)";
480 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
481 if (C->print_intrinsics() || C->print_inlining()) {
482 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
483 }
484 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
485 if (C->log()) {
486 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
487 vmIntrinsics::name_at(intrinsic_id()),
488 (is_virtual() ? " virtual='1'" : ""),
489 C->unique() - nodes);
490 }
491 return slow_ctl; // Could be NULL if the check folds.
492 }
493
494 // The intrinsic bailed out
495 if (jvms->has_method()) {
496 // Not a root compile.
497 const char* msg = "failed to generate predicate for intrinsic";
498 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
499 if (C->print_intrinsics() || C->print_inlining()) {
500 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
501 }
502 } else {
503 // Root compile
504 ResourceMark rm;
505 stringStream msg_stream;
506 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
507 vmIntrinsics::name_at(intrinsic_id()),
508 is_virtual() ? " (virtual)" : "", bci);
509 const char *msg = msg_stream.as_string();
510 log_debug(jit, inlining)("%s", msg);
511 if (C->print_intrinsics() || C->print_inlining()) {
512 C->print_inlining_stream()->print("%s", msg);
513 }
514 }
515 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
516 return NULL;
517 }
518
519 bool LibraryCallKit::try_to_inline(int predicate) {
520 // Handle symbolic names for otherwise undistinguished boolean switches:
521 const bool is_store = true;
522 const bool is_compress = true;
523 const bool is_static = true;
524 const bool is_volatile = true;
525
526 if (!jvms()->has_method()) {
527 // Root JVMState has a null method.
528 assert(map()->memory()->Opcode() == Op_Parm, "");
529 // Insert the memory aliasing node
530 set_all_memory(reset_memory());
531 }
532 assert(merged_memory(), "");
533
534
535 switch (intrinsic_id()) {
536 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
537 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
538 case vmIntrinsics::_getClass: return inline_native_getClass();
539
540 case vmIntrinsics::_ceil:
541 case vmIntrinsics::_floor:
542 case vmIntrinsics::_rint:
543 case vmIntrinsics::_dsin:
544 case vmIntrinsics::_dcos:
545 case vmIntrinsics::_dtan:
546 case vmIntrinsics::_dabs:
547 case vmIntrinsics::_fabs:
548 case vmIntrinsics::_iabs:
549 case vmIntrinsics::_labs:
550 case vmIntrinsics::_datan2:
551 case vmIntrinsics::_dsqrt:
552 case vmIntrinsics::_dexp:
553 case vmIntrinsics::_dlog:
554 case vmIntrinsics::_dlog10:
555 case vmIntrinsics::_dpow:
556 case vmIntrinsics::_dsignum:
557 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
558
559 case vmIntrinsics::_min:
560 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
561
562 case vmIntrinsics::_notify:
563 case vmIntrinsics::_notifyAll:
564 return inline_notify(intrinsic_id());
565
566 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
567 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
568 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
569 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
570 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
571 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
572 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
573 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
574 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
575 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
576 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
577 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
578 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
579
580 case vmIntrinsics::_arraycopy: return inline_arraycopy();
581
582 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
583 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
584 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
585 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
586
587 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
588 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
589 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
590 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
591 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
592 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
593 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
594
595 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
596 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
597
598 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
599 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
600 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
601 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
602
603 case vmIntrinsics::_compressStringC:
604 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
605 case vmIntrinsics::_inflateStringC:
606 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
607
608 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
609 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
610 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
611 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
612 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
613 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
614 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
615 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
616 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
617
618 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
619 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
620 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
621 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
622 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
623 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
624 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
625 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
626 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
627
628 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
629 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
630 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
631 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
632 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
633 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
634 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
635 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
636 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
637
638 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
639 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
640 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
641 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
642 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
643 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
644 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
645 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
646 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
647
648 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
649 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
650 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
651 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
652
653 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
654 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
655 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
656 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
657
658 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
659 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
660 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
661 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
662 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
663 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
664 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
665 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
666 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
667
668 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
669 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
670 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
671 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
672 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
673 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
674 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
675 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
676 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
677
678 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
679 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
680 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
681 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
682 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
683 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
684 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
685 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
686 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
687
688 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
689 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
690 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
691 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
692 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
693 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
694 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
695 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
696 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
697
698 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
699 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
700 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
701 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
702 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
703
704 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
705 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
706 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
707 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
708 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
709 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
710 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
711 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
712 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
713 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
714 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
715 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
716 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
717 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
718 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
719 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
720 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
721 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
722 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
723 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
724
725 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
726 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
727 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
728 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
729 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
730 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
731 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
732 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
733 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
734 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
735 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
736 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
737 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
738 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
739 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
740
741 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
742 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
743 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
744 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
745
746 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
747 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
748 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
749 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
750 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
751
752 case vmIntrinsics::_loadFence:
753 case vmIntrinsics::_storeFence:
754 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
755
756 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
757
758 case vmIntrinsics::_currentThread: return inline_native_currentThread();
759
760 #ifdef JFR_HAVE_INTRINSICS
761 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime");
762 case vmIntrinsics::_getClassId: return inline_native_classID();
763 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
764 #endif
765 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
766 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
767 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
768 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
769 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
770 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
771 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
772 case vmIntrinsics::_getLength: return inline_native_getLength();
773 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
774 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
775 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
776 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
777 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
778 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
779
780 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
781 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
782
783 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
784
785 case vmIntrinsics::_isInstance:
786 case vmIntrinsics::_getModifiers:
787 case vmIntrinsics::_isInterface:
788 case vmIntrinsics::_isArray:
789 case vmIntrinsics::_isPrimitive:
790 case vmIntrinsics::_isHidden:
791 case vmIntrinsics::_getSuperclass:
792 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
793
794 case vmIntrinsics::_floatToRawIntBits:
795 case vmIntrinsics::_floatToIntBits:
796 case vmIntrinsics::_intBitsToFloat:
797 case vmIntrinsics::_doubleToRawLongBits:
798 case vmIntrinsics::_doubleToLongBits:
799 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
800
801 case vmIntrinsics::_numberOfLeadingZeros_i:
802 case vmIntrinsics::_numberOfLeadingZeros_l:
803 case vmIntrinsics::_numberOfTrailingZeros_i:
804 case vmIntrinsics::_numberOfTrailingZeros_l:
805 case vmIntrinsics::_bitCount_i:
806 case vmIntrinsics::_bitCount_l:
807 case vmIntrinsics::_reverseBytes_i:
808 case vmIntrinsics::_reverseBytes_l:
809 case vmIntrinsics::_reverseBytes_s:
810 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
811
812 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
813
814 case vmIntrinsics::_Reference_get: return inline_reference_get();
815
816 case vmIntrinsics::_Class_cast: return inline_Class_cast();
817
818 case vmIntrinsics::_aescrypt_encryptBlock:
819 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
820
821 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
822 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
823 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
824
825 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
826 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
827 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
828
829 case vmIntrinsics::_counterMode_AESCrypt:
830 return inline_counterMode_AESCrypt(intrinsic_id());
831
832 case vmIntrinsics::_sha_implCompress:
833 case vmIntrinsics::_sha2_implCompress:
834 case vmIntrinsics::_sha5_implCompress:
835 return inline_sha_implCompress(intrinsic_id());
836
837 case vmIntrinsics::_digestBase_implCompressMB:
838 return inline_digestBase_implCompressMB(predicate);
839
840 case vmIntrinsics::_multiplyToLen:
841 return inline_multiplyToLen();
842
843 case vmIntrinsics::_squareToLen:
844 return inline_squareToLen();
845
846 case vmIntrinsics::_mulAdd:
847 return inline_mulAdd();
848
849 case vmIntrinsics::_montgomeryMultiply:
850 return inline_montgomeryMultiply();
851 case vmIntrinsics::_montgomerySquare:
852 return inline_montgomerySquare();
853
854 case vmIntrinsics::_bigIntegerRightShiftWorker:
855 return inline_bigIntegerShift(true);
856 case vmIntrinsics::_bigIntegerLeftShiftWorker:
857 return inline_bigIntegerShift(false);
858
859 case vmIntrinsics::_vectorizedMismatch:
860 return inline_vectorizedMismatch();
861
862 case vmIntrinsics::_ghash_processBlocks:
863 return inline_ghash_processBlocks();
864 case vmIntrinsics::_base64_encodeBlock:
865 return inline_base64_encodeBlock();
866
867 case vmIntrinsics::_encodeISOArray:
868 case vmIntrinsics::_encodeByteISOArray:
869 return inline_encodeISOArray();
870
871 case vmIntrinsics::_updateCRC32:
872 return inline_updateCRC32();
873 case vmIntrinsics::_updateBytesCRC32:
874 return inline_updateBytesCRC32();
875 case vmIntrinsics::_updateByteBufferCRC32:
876 return inline_updateByteBufferCRC32();
877
878 case vmIntrinsics::_updateBytesCRC32C:
879 return inline_updateBytesCRC32C();
880 case vmIntrinsics::_updateDirectByteBufferCRC32C:
881 return inline_updateDirectByteBufferCRC32C();
882
883 case vmIntrinsics::_updateBytesAdler32:
884 return inline_updateBytesAdler32();
885 case vmIntrinsics::_updateByteBufferAdler32:
886 return inline_updateByteBufferAdler32();
887
888 case vmIntrinsics::_profileBoolean:
889 return inline_profileBoolean();
890 case vmIntrinsics::_isCompileConstant:
891 return inline_isCompileConstant();
892
893 case vmIntrinsics::_hasNegatives:
894 return inline_hasNegatives();
895
896 case vmIntrinsics::_fmaD:
897 case vmIntrinsics::_fmaF:
898 return inline_fma(intrinsic_id());
899
900 case vmIntrinsics::_isDigit:
901 case vmIntrinsics::_isLowerCase:
902 case vmIntrinsics::_isUpperCase:
903 case vmIntrinsics::_isWhitespace:
904 return inline_character_compare(intrinsic_id());
905
906 case vmIntrinsics::_maxF:
907 case vmIntrinsics::_minF:
908 case vmIntrinsics::_maxD:
909 case vmIntrinsics::_minD:
910 return inline_fp_min_max(intrinsic_id());
911
912 default:
913 // If you get here, it may be that someone has added a new intrinsic
914 // to the list in vmSymbols.hpp without implementing it here.
915 #ifndef PRODUCT
916 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
917 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
918 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
919 }
920 #endif
921 return false;
922 }
923 }
924
925 Node* LibraryCallKit::try_to_predicate(int predicate) {
926 if (!jvms()->has_method()) {
927 // Root JVMState has a null method.
928 assert(map()->memory()->Opcode() == Op_Parm, "");
929 // Insert the memory aliasing node
930 set_all_memory(reset_memory());
931 }
932 assert(merged_memory(), "");
933
934 switch (intrinsic_id()) {
935 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
936 return inline_cipherBlockChaining_AESCrypt_predicate(false);
937 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
938 return inline_cipherBlockChaining_AESCrypt_predicate(true);
939 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
940 return inline_electronicCodeBook_AESCrypt_predicate(false);
941 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
942 return inline_electronicCodeBook_AESCrypt_predicate(true);
943 case vmIntrinsics::_counterMode_AESCrypt:
944 return inline_counterMode_AESCrypt_predicate();
945 case vmIntrinsics::_digestBase_implCompressMB:
946 return inline_digestBase_implCompressMB_predicate(predicate);
947
948 default:
949 // If you get here, it may be that someone has added a new intrinsic
950 // to the list in vmSymbols.hpp without implementing it here.
951 #ifndef PRODUCT
952 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
953 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
954 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
955 }
956 #endif
957 Node* slow_ctl = control();
958 set_control(top()); // No fast path instrinsic
959 return slow_ctl;
960 }
961 }
962
963 //------------------------------set_result-------------------------------
964 // Helper function for finishing intrinsics.
965 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
966 record_for_igvn(region);
967 set_control(_gvn.transform(region));
968 set_result( _gvn.transform(value));
969 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
970 }
971
972 //------------------------------generate_guard---------------------------
973 // Helper function for generating guarded fast-slow graph structures.
974 // The given 'test', if true, guards a slow path. If the test fails
975 // then a fast path can be taken. (We generally hope it fails.)
976 // In all cases, GraphKit::control() is updated to the fast path.
977 // The returned value represents the control for the slow path.
978 // The return value is never 'top'; it is either a valid control
979 // or NULL if it is obvious that the slow path can never be taken.
980 // Also, if region and the slow control are not NULL, the slow edge
981 // is appended to the region.
982 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
983 if (stopped()) {
984 // Already short circuited.
985 return NULL;
986 }
987
988 // Build an if node and its projections.
989 // If test is true we take the slow path, which we assume is uncommon.
990 if (_gvn.type(test) == TypeInt::ZERO) {
991 // The slow branch is never taken. No need to build this guard.
992 return NULL;
993 }
994
995 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
996
997 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
998 if (if_slow == top()) {
999 // The slow branch is never taken. No need to build this guard.
1000 return NULL;
1001 }
1002
1003 if (region != NULL)
1004 region->add_req(if_slow);
1005
1006 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
1007 set_control(if_fast);
1008
1009 return if_slow;
1010 }
1011
1012 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
1013 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
1014 }
1015 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
1016 return generate_guard(test, region, PROB_FAIR);
1017 }
1018
1019 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
1020 Node* *pos_index) {
1021 if (stopped())
1022 return NULL; // already stopped
1023 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
1024 return NULL; // index is already adequately typed
1025 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
1026 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1027 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
1028 if (is_neg != NULL && pos_index != NULL) {
1029 // Emulate effect of Parse::adjust_map_after_if.
1030 Node* ccast = new CastIINode(index, TypeInt::POS);
1031 ccast->set_req(0, control());
1032 (*pos_index) = _gvn.transform(ccast);
1033 }
1034 return is_neg;
1035 }
1036
1037 // Make sure that 'position' is a valid limit index, in [0..length].
1038 // There are two equivalent plans for checking this:
1039 // A. (offset + copyLength) unsigned<= arrayLength
1040 // B. offset <= (arrayLength - copyLength)
1041 // We require that all of the values above, except for the sum and
1042 // difference, are already known to be non-negative.
1043 // Plan A is robust in the face of overflow, if offset and copyLength
1044 // are both hugely positive.
1045 //
1046 // Plan B is less direct and intuitive, but it does not overflow at
1047 // all, since the difference of two non-negatives is always
1048 // representable. Whenever Java methods must perform the equivalent
1049 // check they generally use Plan B instead of Plan A.
1050 // For the moment we use Plan A.
1051 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
1052 Node* subseq_length,
1053 Node* array_length,
1054 RegionNode* region) {
1055 if (stopped())
1056 return NULL; // already stopped
1057 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
1058 if (zero_offset && subseq_length->eqv_uncast(array_length))
1059 return NULL; // common case of whole-array copy
1060 Node* last = subseq_length;
1061 if (!zero_offset) // last += offset
1062 last = _gvn.transform(new AddINode(last, offset));
1063 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
1064 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1065 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
1066 return is_over;
1067 }
1068
1069 // Emit range checks for the given String.value byte array
1070 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
1071 if (stopped()) {
1072 return; // already stopped
1073 }
1074 RegionNode* bailout = new RegionNode(1);
1075 record_for_igvn(bailout);
1076 if (char_count) {
1077 // Convert char count to byte count
1078 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1079 }
1080
1081 // Offset and count must not be negative
1082 generate_negative_guard(offset, bailout);
1083 generate_negative_guard(count, bailout);
1084 // Offset + count must not exceed length of array
1085 generate_limit_guard(offset, count, load_array_length(array), bailout);
1086
1087 if (bailout->req() > 1) {
1088 PreserveJVMState pjvms(this);
1089 set_control(_gvn.transform(bailout));
1090 uncommon_trap(Deoptimization::Reason_intrinsic,
1091 Deoptimization::Action_maybe_recompile);
1092 }
1093 }
1094
1095 //--------------------------generate_current_thread--------------------
1096 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1097 ciKlass* thread_klass = env()->Thread_klass();
1098 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1099 Node* thread = _gvn.transform(new ThreadLocalNode());
1100 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1101 Node* threadObj = _gvn.transform(LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(), thread_type, T_OBJECT, MemNode::unordered));
1102 tls_output = thread;
1103 return threadObj;
1104 }
1105
1106
1107 //------------------------------make_string_method_node------------------------
1108 // Helper method for String intrinsic functions. This version is called with
1109 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1110 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1111 // containing the lengths of str1 and str2.
1112 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1113 Node* result = NULL;
1114 switch (opcode) {
1115 case Op_StrIndexOf:
1116 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1117 str1_start, cnt1, str2_start, cnt2, ae);
1118 break;
1119 case Op_StrComp:
1120 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1121 str1_start, cnt1, str2_start, cnt2, ae);
1122 break;
1123 case Op_StrEquals:
1124 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1125 // Use the constant length if there is one because optimized match rule may exist.
1126 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1127 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1128 break;
1129 default:
1130 ShouldNotReachHere();
1131 return NULL;
1132 }
1133
1134 // All these intrinsics have checks.
1135 C->set_has_split_ifs(true); // Has chance for split-if optimization
1136 clear_upper_avx();
1137
1138 return _gvn.transform(result);
1139 }
1140
1141 //------------------------------inline_string_compareTo------------------------
1142 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1143 Node* arg1 = argument(0);
1144 Node* arg2 = argument(1);
1145
1146 arg1 = must_be_not_null(arg1, true);
1147 arg2 = must_be_not_null(arg2, true);
1148
1149 // Get start addr and length of first argument
1150 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1151 Node* arg1_cnt = load_array_length(arg1);
1152
1153 // Get start addr and length of second argument
1154 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1155 Node* arg2_cnt = load_array_length(arg2);
1156
1157 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1158 set_result(result);
1159 return true;
1160 }
1161
1162 //------------------------------inline_string_equals------------------------
1163 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1164 Node* arg1 = argument(0);
1165 Node* arg2 = argument(1);
1166
1167 // paths (plus control) merge
1168 RegionNode* region = new RegionNode(3);
1169 Node* phi = new PhiNode(region, TypeInt::BOOL);
1170
1171 if (!stopped()) {
1172
1173 arg1 = must_be_not_null(arg1, true);
1174 arg2 = must_be_not_null(arg2, true);
1175
1176 // Get start addr and length of first argument
1177 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1178 Node* arg1_cnt = load_array_length(arg1);
1179
1180 // Get start addr and length of second argument
1181 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1182 Node* arg2_cnt = load_array_length(arg2);
1183
1184 // Check for arg1_cnt != arg2_cnt
1185 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1186 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1187 Node* if_ne = generate_slow_guard(bol, NULL);
1188 if (if_ne != NULL) {
1189 phi->init_req(2, intcon(0));
1190 region->init_req(2, if_ne);
1191 }
1192
1193 // Check for count == 0 is done by assembler code for StrEquals.
1194
1195 if (!stopped()) {
1196 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1197 phi->init_req(1, equals);
1198 region->init_req(1, control());
1199 }
1200 }
1201
1202 // post merge
1203 set_control(_gvn.transform(region));
1204 record_for_igvn(region);
1205
1206 set_result(_gvn.transform(phi));
1207 return true;
1208 }
1209
1210 //------------------------------inline_array_equals----------------------------
1211 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1212 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1213 Node* arg1 = argument(0);
1214 Node* arg2 = argument(1);
1215
1216 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1217 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1218 clear_upper_avx();
1219
1220 return true;
1221 }
1222
1223 //------------------------------inline_hasNegatives------------------------------
1224 bool LibraryCallKit::inline_hasNegatives() {
1225 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1226 return false;
1227 }
1228
1229 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1230 // no receiver since it is static method
1231 Node* ba = argument(0);
1232 Node* offset = argument(1);
1233 Node* len = argument(2);
1234
1235 ba = must_be_not_null(ba, true);
1236
1237 // Range checks
1238 generate_string_range_check(ba, offset, len, false);
1239 if (stopped()) {
1240 return true;
1241 }
1242 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1243 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1244 set_result(_gvn.transform(result));
1245 return true;
1246 }
1247
1248 bool LibraryCallKit::inline_preconditions_checkIndex() {
1249 Node* index = argument(0);
1250 Node* length = argument(1);
1251 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1252 return false;
1253 }
1254
1255 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1256 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1257
1258 {
1259 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1260 uncommon_trap(Deoptimization::Reason_intrinsic,
1261 Deoptimization::Action_make_not_entrant);
1262 }
1263
1264 if (stopped()) {
1265 return false;
1266 }
1267
1268 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1269 BoolTest::mask btest = BoolTest::lt;
1270 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1271 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1272 _gvn.set_type(rc, rc->Value(&_gvn));
1273 if (!rc_bool->is_Con()) {
1274 record_for_igvn(rc);
1275 }
1276 set_control(_gvn.transform(new IfTrueNode(rc)));
1277 {
1278 PreserveJVMState pjvms(this);
1279 set_control(_gvn.transform(new IfFalseNode(rc)));
1280 uncommon_trap(Deoptimization::Reason_range_check,
1281 Deoptimization::Action_make_not_entrant);
1282 }
1283
1284 if (stopped()) {
1285 return false;
1286 }
1287
1288 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1289 result->set_req(0, control());
1290 result = _gvn.transform(result);
1291 set_result(result);
1292 replace_in_map(index, result);
1293 clear_upper_avx();
1294 return true;
1295 }
1296
1297 //------------------------------inline_string_indexOf------------------------
1298 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1299 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1300 return false;
1301 }
1302 Node* src = argument(0);
1303 Node* tgt = argument(1);
1304
1305 // Make the merge point
1306 RegionNode* result_rgn = new RegionNode(4);
1307 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1308
1309 src = must_be_not_null(src, true);
1310 tgt = must_be_not_null(tgt, true);
1311
1312 // Get start addr and length of source string
1313 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1314 Node* src_count = load_array_length(src);
1315
1316 // Get start addr and length of substring
1317 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1318 Node* tgt_count = load_array_length(tgt);
1319
1320 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1321 // Divide src size by 2 if String is UTF16 encoded
1322 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1323 }
1324 if (ae == StrIntrinsicNode::UU) {
1325 // Divide substring size by 2 if String is UTF16 encoded
1326 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1327 }
1328
1329 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1330 if (result != NULL) {
1331 result_phi->init_req(3, result);
1332 result_rgn->init_req(3, control());
1333 }
1334 set_control(_gvn.transform(result_rgn));
1335 record_for_igvn(result_rgn);
1336 set_result(_gvn.transform(result_phi));
1337
1338 return true;
1339 }
1340
1341 //-----------------------------inline_string_indexOf-----------------------
1342 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1343 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1344 return false;
1345 }
1346 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1347 return false;
1348 }
1349 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1350 Node* src = argument(0); // byte[]
1351 Node* src_count = argument(1); // char count
1352 Node* tgt = argument(2); // byte[]
1353 Node* tgt_count = argument(3); // char count
1354 Node* from_index = argument(4); // char index
1355
1356 src = must_be_not_null(src, true);
1357 tgt = must_be_not_null(tgt, true);
1358
1359 // Multiply byte array index by 2 if String is UTF16 encoded
1360 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1361 src_count = _gvn.transform(new SubINode(src_count, from_index));
1362 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1363 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1364
1365 // Range checks
1366 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1367 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1368 if (stopped()) {
1369 return true;
1370 }
1371
1372 RegionNode* region = new RegionNode(5);
1373 Node* phi = new PhiNode(region, TypeInt::INT);
1374
1375 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1376 if (result != NULL) {
1377 // The result is index relative to from_index if substring was found, -1 otherwise.
1378 // Generate code which will fold into cmove.
1379 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1380 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1381
1382 Node* if_lt = generate_slow_guard(bol, NULL);
1383 if (if_lt != NULL) {
1384 // result == -1
1385 phi->init_req(3, result);
1386 region->init_req(3, if_lt);
1387 }
1388 if (!stopped()) {
1389 result = _gvn.transform(new AddINode(result, from_index));
1390 phi->init_req(4, result);
1391 region->init_req(4, control());
1392 }
1393 }
1394
1395 set_control(_gvn.transform(region));
1396 record_for_igvn(region);
1397 set_result(_gvn.transform(phi));
1398 clear_upper_avx();
1399
1400 return true;
1401 }
1402
1403 // Create StrIndexOfNode with fast path checks
1404 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1405 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1406 // Check for substr count > string count
1407 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1408 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1409 Node* if_gt = generate_slow_guard(bol, NULL);
1410 if (if_gt != NULL) {
1411 phi->init_req(1, intcon(-1));
1412 region->init_req(1, if_gt);
1413 }
1414 if (!stopped()) {
1415 // Check for substr count == 0
1416 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1417 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1418 Node* if_zero = generate_slow_guard(bol, NULL);
1419 if (if_zero != NULL) {
1420 phi->init_req(2, intcon(0));
1421 region->init_req(2, if_zero);
1422 }
1423 }
1424 if (!stopped()) {
1425 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1426 }
1427 return NULL;
1428 }
1429
1430 //-----------------------------inline_string_indexOfChar-----------------------
1431 bool LibraryCallKit::inline_string_indexOfChar() {
1432 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1433 return false;
1434 }
1435 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1436 return false;
1437 }
1438 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1439 Node* src = argument(0); // byte[]
1440 Node* tgt = argument(1); // tgt is int ch
1441 Node* from_index = argument(2);
1442 Node* max = argument(3);
1443
1444 src = must_be_not_null(src, true);
1445
1446 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1447 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1448 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1449
1450 // Range checks
1451 generate_string_range_check(src, src_offset, src_count, true);
1452 if (stopped()) {
1453 return true;
1454 }
1455
1456 RegionNode* region = new RegionNode(3);
1457 Node* phi = new PhiNode(region, TypeInt::INT);
1458
1459 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1460 C->set_has_split_ifs(true); // Has chance for split-if optimization
1461 _gvn.transform(result);
1462
1463 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1464 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1465
1466 Node* if_lt = generate_slow_guard(bol, NULL);
1467 if (if_lt != NULL) {
1468 // result == -1
1469 phi->init_req(2, result);
1470 region->init_req(2, if_lt);
1471 }
1472 if (!stopped()) {
1473 result = _gvn.transform(new AddINode(result, from_index));
1474 phi->init_req(1, result);
1475 region->init_req(1, control());
1476 }
1477 set_control(_gvn.transform(region));
1478 record_for_igvn(region);
1479 set_result(_gvn.transform(phi));
1480
1481 return true;
1482 }
1483 //---------------------------inline_string_copy---------------------
1484 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1485 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1486 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1487 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1488 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1489 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1490 bool LibraryCallKit::inline_string_copy(bool compress) {
1491 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1492 return false;
1493 }
1494 int nargs = 5; // 2 oops, 3 ints
1495 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1496
1497 Node* src = argument(0);
1498 Node* src_offset = argument(1);
1499 Node* dst = argument(2);
1500 Node* dst_offset = argument(3);
1501 Node* length = argument(4);
1502
1503 // Check for allocation before we add nodes that would confuse
1504 // tightly_coupled_allocation()
1505 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1506
1507 // Figure out the size and type of the elements we will be copying.
1508 const Type* src_type = src->Value(&_gvn);
1509 const Type* dst_type = dst->Value(&_gvn);
1510 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1511 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1512 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1513 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1514 "Unsupported array types for inline_string_copy");
1515
1516 src = must_be_not_null(src, true);
1517 dst = must_be_not_null(dst, true);
1518
1519 // Convert char[] offsets to byte[] offsets
1520 bool convert_src = (compress && src_elem == T_BYTE);
1521 bool convert_dst = (!compress && dst_elem == T_BYTE);
1522 if (convert_src) {
1523 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1524 } else if (convert_dst) {
1525 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1526 }
1527
1528 // Range checks
1529 generate_string_range_check(src, src_offset, length, convert_src);
1530 generate_string_range_check(dst, dst_offset, length, convert_dst);
1531 if (stopped()) {
1532 return true;
1533 }
1534
1535 Node* src_start = array_element_address(src, src_offset, src_elem);
1536 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1537 // 'src_start' points to src array + scaled offset
1538 // 'dst_start' points to dst array + scaled offset
1539 Node* count = NULL;
1540 if (compress) {
1541 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1542 } else {
1543 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1544 }
1545
1546 if (alloc != NULL) {
1547 if (alloc->maybe_set_complete(&_gvn)) {
1548 // "You break it, you buy it."
1549 InitializeNode* init = alloc->initialization();
1550 assert(init->is_complete(), "we just did this");
1551 init->set_complete_with_arraycopy();
1552 assert(dst->is_CheckCastPP(), "sanity");
1553 assert(dst->in(0)->in(0) == init, "dest pinned");
1554 }
1555 // Do not let stores that initialize this object be reordered with
1556 // a subsequent store that would make this object accessible by
1557 // other threads.
1558 // Record what AllocateNode this StoreStore protects so that
1559 // escape analysis can go from the MemBarStoreStoreNode to the
1560 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1561 // based on the escape status of the AllocateNode.
1562 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1563 }
1564 if (compress) {
1565 set_result(_gvn.transform(count));
1566 }
1567 clear_upper_avx();
1568
1569 return true;
1570 }
1571
1572 #ifdef _LP64
1573 #define XTOP ,top() /*additional argument*/
1574 #else //_LP64
1575 #define XTOP /*no additional argument*/
1576 #endif //_LP64
1577
1578 //------------------------inline_string_toBytesU--------------------------
1579 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1580 bool LibraryCallKit::inline_string_toBytesU() {
1581 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1582 return false;
1583 }
1584 // Get the arguments.
1585 Node* value = argument(0);
1586 Node* offset = argument(1);
1587 Node* length = argument(2);
1588
1589 Node* newcopy = NULL;
1590
1591 // Set the original stack and the reexecute bit for the interpreter to reexecute
1592 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1593 { PreserveReexecuteState preexecs(this);
1594 jvms()->set_should_reexecute(true);
1595
1596 // Check if a null path was taken unconditionally.
1597 value = null_check(value);
1598
1599 RegionNode* bailout = new RegionNode(1);
1600 record_for_igvn(bailout);
1601
1602 // Range checks
1603 generate_negative_guard(offset, bailout);
1604 generate_negative_guard(length, bailout);
1605 generate_limit_guard(offset, length, load_array_length(value), bailout);
1606 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1607 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1608
1609 if (bailout->req() > 1) {
1610 PreserveJVMState pjvms(this);
1611 set_control(_gvn.transform(bailout));
1612 uncommon_trap(Deoptimization::Reason_intrinsic,
1613 Deoptimization::Action_maybe_recompile);
1614 }
1615 if (stopped()) {
1616 return true;
1617 }
1618
1619 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1620 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1621 newcopy = new_array(klass_node, size, 0); // no arguments to push
1622 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1623
1624 // Calculate starting addresses.
1625 Node* src_start = array_element_address(value, offset, T_CHAR);
1626 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1627
1628 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1629 const TypeInt* toffset = gvn().type(offset)->is_int();
1630 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1631
1632 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1633 const char* copyfunc_name = "arraycopy";
1634 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1635 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1636 OptoRuntime::fast_arraycopy_Type(),
1637 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1638 src_start, dst_start, ConvI2X(length) XTOP);
1639 // Do not let reads from the cloned object float above the arraycopy.
1640 if (alloc != NULL) {
1641 if (alloc->maybe_set_complete(&_gvn)) {
1642 // "You break it, you buy it."
1643 InitializeNode* init = alloc->initialization();
1644 assert(init->is_complete(), "we just did this");
1645 init->set_complete_with_arraycopy();
1646 assert(newcopy->is_CheckCastPP(), "sanity");
1647 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1648 }
1649 // Do not let stores that initialize this object be reordered with
1650 // a subsequent store that would make this object accessible by
1651 // other threads.
1652 // Record what AllocateNode this StoreStore protects so that
1653 // escape analysis can go from the MemBarStoreStoreNode to the
1654 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1655 // based on the escape status of the AllocateNode.
1656 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1657 } else {
1658 insert_mem_bar(Op_MemBarCPUOrder);
1659 }
1660 } // original reexecute is set back here
1661
1662 C->set_has_split_ifs(true); // Has chance for split-if optimization
1663 if (!stopped()) {
1664 set_result(newcopy);
1665 }
1666 clear_upper_avx();
1667
1668 return true;
1669 }
1670
1671 //------------------------inline_string_getCharsU--------------------------
1672 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1673 bool LibraryCallKit::inline_string_getCharsU() {
1674 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1675 return false;
1676 }
1677
1678 // Get the arguments.
1679 Node* src = argument(0);
1680 Node* src_begin = argument(1);
1681 Node* src_end = argument(2); // exclusive offset (i < src_end)
1682 Node* dst = argument(3);
1683 Node* dst_begin = argument(4);
1684
1685 // Check for allocation before we add nodes that would confuse
1686 // tightly_coupled_allocation()
1687 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1688
1689 // Check if a null path was taken unconditionally.
1690 src = null_check(src);
1691 dst = null_check(dst);
1692 if (stopped()) {
1693 return true;
1694 }
1695
1696 // Get length and convert char[] offset to byte[] offset
1697 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1698 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1699
1700 // Range checks
1701 generate_string_range_check(src, src_begin, length, true);
1702 generate_string_range_check(dst, dst_begin, length, false);
1703 if (stopped()) {
1704 return true;
1705 }
1706
1707 if (!stopped()) {
1708 // Calculate starting addresses.
1709 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1710 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1711
1712 // Check if array addresses are aligned to HeapWordSize
1713 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1714 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1715 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1716 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1717
1718 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1719 const char* copyfunc_name = "arraycopy";
1720 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1721 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1722 OptoRuntime::fast_arraycopy_Type(),
1723 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1724 src_start, dst_start, ConvI2X(length) XTOP);
1725 // Do not let reads from the cloned object float above the arraycopy.
1726 if (alloc != NULL) {
1727 if (alloc->maybe_set_complete(&_gvn)) {
1728 // "You break it, you buy it."
1729 InitializeNode* init = alloc->initialization();
1730 assert(init->is_complete(), "we just did this");
1731 init->set_complete_with_arraycopy();
1732 assert(dst->is_CheckCastPP(), "sanity");
1733 assert(dst->in(0)->in(0) == init, "dest pinned");
1734 }
1735 // Do not let stores that initialize this object be reordered with
1736 // a subsequent store that would make this object accessible by
1737 // other threads.
1738 // Record what AllocateNode this StoreStore protects so that
1739 // escape analysis can go from the MemBarStoreStoreNode to the
1740 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1741 // based on the escape status of the AllocateNode.
1742 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1743 } else {
1744 insert_mem_bar(Op_MemBarCPUOrder);
1745 }
1746 }
1747
1748 C->set_has_split_ifs(true); // Has chance for split-if optimization
1749 return true;
1750 }
1751
1752 //----------------------inline_string_char_access----------------------------
1753 // Store/Load char to/from byte[] array.
1754 // static void StringUTF16.putChar(byte[] val, int index, int c)
1755 // static char StringUTF16.getChar(byte[] val, int index)
1756 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1757 Node* value = argument(0);
1758 Node* index = argument(1);
1759 Node* ch = is_store ? argument(2) : NULL;
1760
1761 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1762 // correctly requires matched array shapes.
1763 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1764 "sanity: byte[] and char[] bases agree");
1765 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1766 "sanity: byte[] and char[] scales agree");
1767
1768 // Bail when getChar over constants is requested: constant folding would
1769 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1770 // Java method would constant fold nicely instead.
1771 if (!is_store && value->is_Con() && index->is_Con()) {
1772 return false;
1773 }
1774
1775 value = must_be_not_null(value, true);
1776
1777 Node* adr = array_element_address(value, index, T_CHAR);
1778 if (adr->is_top()) {
1779 return false;
1780 }
1781 if (is_store) {
1782 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1783 } else {
1784 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
1785 set_result(ch);
1786 }
1787 return true;
1788 }
1789
1790 //--------------------------round_double_node--------------------------------
1791 // Round a double node if necessary.
1792 Node* LibraryCallKit::round_double_node(Node* n) {
1793 if (Matcher::strict_fp_requires_explicit_rounding) {
1794 #ifdef IA32
1795 if (UseSSE < 2) {
1796 n = _gvn.transform(new RoundDoubleNode(NULL, n));
1797 }
1798 #else
1799 Unimplemented();
1800 #endif // IA32
1801 }
1802 return n;
1803 }
1804
1805 //------------------------------inline_math-----------------------------------
1806 // public static double Math.abs(double)
1807 // public static double Math.sqrt(double)
1808 // public static double Math.log(double)
1809 // public static double Math.log10(double)
1810 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1811 Node* arg = round_double_node(argument(0));
1812 Node* n = NULL;
1813 switch (id) {
1814 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1815 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1816 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1817 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1818 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1819 case vmIntrinsics::_dsignum: n = new SignumDNode(arg); break;
1820 default: fatal_unexpected_iid(id); break;
1821 }
1822 set_result(_gvn.transform(n));
1823 return true;
1824 }
1825
1826 //------------------------------inline_math-----------------------------------
1827 // public static float Math.abs(float)
1828 // public static int Math.abs(int)
1829 // public static long Math.abs(long)
1830 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1831 Node* arg = argument(0);
1832 Node* n = NULL;
1833 switch (id) {
1834 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1835 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1836 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1837 case vmIntrinsics::_fsignum: n = new SignumFNode(arg); break;
1838 default: fatal_unexpected_iid(id); break;
1839 }
1840 set_result(_gvn.transform(n));
1841 return true;
1842 }
1843
1844 //------------------------------runtime_math-----------------------------
1845 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1846 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1847 "must be (DD)D or (D)D type");
1848
1849 // Inputs
1850 Node* a = round_double_node(argument(0));
1851 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1852
1853 const TypePtr* no_memory_effects = NULL;
1854 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1855 no_memory_effects,
1856 a, top(), b, b ? top() : NULL);
1857 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1858 #ifdef ASSERT
1859 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1860 assert(value_top == top(), "second value must be top");
1861 #endif
1862
1863 set_result(value);
1864 return true;
1865 }
1866
1867 //------------------------------inline_math_native-----------------------------
1868 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1869 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1870 switch (id) {
1871 // These intrinsics are not properly supported on all hardware
1872 case vmIntrinsics::_dsin:
1873 return StubRoutines::dsin() != NULL ?
1874 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1875 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1876 case vmIntrinsics::_dcos:
1877 return StubRoutines::dcos() != NULL ?
1878 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1879 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1880 case vmIntrinsics::_dtan:
1881 return StubRoutines::dtan() != NULL ?
1882 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1883 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1884 case vmIntrinsics::_dlog:
1885 return StubRoutines::dlog() != NULL ?
1886 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1887 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1888 case vmIntrinsics::_dlog10:
1889 return StubRoutines::dlog10() != NULL ?
1890 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1891 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1892
1893 // These intrinsics are supported on all hardware
1894 case vmIntrinsics::_ceil:
1895 case vmIntrinsics::_floor:
1896 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1897 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1898 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1899 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1900 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1901 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1902
1903 case vmIntrinsics::_dexp:
1904 return StubRoutines::dexp() != NULL ?
1905 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1906 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1907 case vmIntrinsics::_dpow: {
1908 Node* exp = round_double_node(argument(2));
1909 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1910 if (d != NULL && d->getd() == 2.0) {
1911 // Special case: pow(x, 2.0) => x * x
1912 Node* base = round_double_node(argument(0));
1913 set_result(_gvn.transform(new MulDNode(base, base)));
1914 return true;
1915 }
1916 return StubRoutines::dpow() != NULL ?
1917 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1918 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1919 }
1920 #undef FN_PTR
1921
1922 case vmIntrinsics::_dsignum:
1923 return UseSignumIntrinsic && (Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false);
1924 case vmIntrinsics::_fsignum:
1925 return UseSignumIntrinsic && (Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false);
1926
1927 // These intrinsics are not yet correctly implemented
1928 case vmIntrinsics::_datan2:
1929 return false;
1930
1931 default:
1932 fatal_unexpected_iid(id);
1933 return false;
1934 }
1935 }
1936
1937 static bool is_simple_name(Node* n) {
1938 return (n->req() == 1 // constant
1939 || (n->is_Type() && n->as_Type()->type()->singleton())
1940 || n->is_Proj() // parameter or return value
1941 || n->is_Phi() // local of some sort
1942 );
1943 }
1944
1945 //----------------------------inline_notify-----------------------------------*
1946 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1947 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1948 address func;
1949 if (id == vmIntrinsics::_notify) {
1950 func = OptoRuntime::monitor_notify_Java();
1951 } else {
1952 func = OptoRuntime::monitor_notifyAll_Java();
1953 }
1954 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1955 make_slow_call_ex(call, env()->Throwable_klass(), false);
1956 return true;
1957 }
1958
1959
1960 //----------------------------inline_min_max-----------------------------------
1961 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1962 set_result(generate_min_max(id, argument(0), argument(1)));
1963 return true;
1964 }
1965
1966 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1967 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1968 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1969 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1970 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1971
1972 {
1973 PreserveJVMState pjvms(this);
1974 PreserveReexecuteState preexecs(this);
1975 jvms()->set_should_reexecute(true);
1976
1977 set_control(slow_path);
1978 set_i_o(i_o());
1979
1980 uncommon_trap(Deoptimization::Reason_intrinsic,
1981 Deoptimization::Action_none);
1982 }
1983
1984 set_control(fast_path);
1985 set_result(math);
1986 }
1987
1988 template <typename OverflowOp>
1989 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1990 typedef typename OverflowOp::MathOp MathOp;
1991
1992 MathOp* mathOp = new MathOp(arg1, arg2);
1993 Node* operation = _gvn.transform( mathOp );
1994 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1995 inline_math_mathExact(operation, ofcheck);
1996 return true;
1997 }
1998
1999 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2000 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2001 }
2002
2003 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2004 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2005 }
2006
2007 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2008 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2009 }
2010
2011 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2012 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2013 }
2014
2015 bool LibraryCallKit::inline_math_negateExactI() {
2016 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2017 }
2018
2019 bool LibraryCallKit::inline_math_negateExactL() {
2020 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2021 }
2022
2023 bool LibraryCallKit::inline_math_multiplyExactI() {
2024 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2025 }
2026
2027 bool LibraryCallKit::inline_math_multiplyExactL() {
2028 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2029 }
2030
2031 bool LibraryCallKit::inline_math_multiplyHigh() {
2032 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2033 return true;
2034 }
2035
2036 Node*
2037 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2038 // These are the candidate return value:
2039 Node* xvalue = x0;
2040 Node* yvalue = y0;
2041
2042 if (xvalue == yvalue) {
2043 return xvalue;
2044 }
2045
2046 bool want_max = (id == vmIntrinsics::_max);
2047
2048 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
2049 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
2050 if (txvalue == NULL || tyvalue == NULL) return top();
2051 // This is not really necessary, but it is consistent with a
2052 // hypothetical MaxINode::Value method:
2053 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
2054
2055 // %%% This folding logic should (ideally) be in a different place.
2056 // Some should be inside IfNode, and there to be a more reliable
2057 // transformation of ?: style patterns into cmoves. We also want
2058 // more powerful optimizations around cmove and min/max.
2059
2060 // Try to find a dominating comparison of these guys.
2061 // It can simplify the index computation for Arrays.copyOf
2062 // and similar uses of System.arraycopy.
2063 // First, compute the normalized version of CmpI(x, y).
2064 int cmp_op = Op_CmpI;
2065 Node* xkey = xvalue;
2066 Node* ykey = yvalue;
2067 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
2068 if (ideal_cmpxy->is_Cmp()) {
2069 // E.g., if we have CmpI(length - offset, count),
2070 // it might idealize to CmpI(length, count + offset)
2071 cmp_op = ideal_cmpxy->Opcode();
2072 xkey = ideal_cmpxy->in(1);
2073 ykey = ideal_cmpxy->in(2);
2074 }
2075
2076 // Start by locating any relevant comparisons.
2077 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
2078 Node* cmpxy = NULL;
2079 Node* cmpyx = NULL;
2080 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
2081 Node* cmp = start_from->fast_out(k);
2082 if (cmp->outcnt() > 0 && // must have prior uses
2083 cmp->in(0) == NULL && // must be context-independent
2084 cmp->Opcode() == cmp_op) { // right kind of compare
2085 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
2086 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
2087 }
2088 }
2089
2090 const int NCMPS = 2;
2091 Node* cmps[NCMPS] = { cmpxy, cmpyx };
2092 int cmpn;
2093 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2094 if (cmps[cmpn] != NULL) break; // find a result
2095 }
2096 if (cmpn < NCMPS) {
2097 // Look for a dominating test that tells us the min and max.
2098 int depth = 0; // Limit search depth for speed
2099 Node* dom = control();
2100 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2101 if (++depth >= 100) break;
2102 Node* ifproj = dom;
2103 if (!ifproj->is_Proj()) continue;
2104 Node* iff = ifproj->in(0);
2105 if (!iff->is_If()) continue;
2106 Node* bol = iff->in(1);
2107 if (!bol->is_Bool()) continue;
2108 Node* cmp = bol->in(1);
2109 if (cmp == NULL) continue;
2110 for (cmpn = 0; cmpn < NCMPS; cmpn++)
2111 if (cmps[cmpn] == cmp) break;
2112 if (cmpn == NCMPS) continue;
2113 BoolTest::mask btest = bol->as_Bool()->_test._test;
2114 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
2115 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2116 // At this point, we know that 'x btest y' is true.
2117 switch (btest) {
2118 case BoolTest::eq:
2119 // They are proven equal, so we can collapse the min/max.
2120 // Either value is the answer. Choose the simpler.
2121 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2122 return yvalue;
2123 return xvalue;
2124 case BoolTest::lt: // x < y
2125 case BoolTest::le: // x <= y
2126 return (want_max ? yvalue : xvalue);
2127 case BoolTest::gt: // x > y
2128 case BoolTest::ge: // x >= y
2129 return (want_max ? xvalue : yvalue);
2130 default:
2131 break;
2132 }
2133 }
2134 }
2135
2136 // We failed to find a dominating test.
2137 // Let's pick a test that might GVN with prior tests.
2138 Node* best_bol = NULL;
2139 BoolTest::mask best_btest = BoolTest::illegal;
2140 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2141 Node* cmp = cmps[cmpn];
2142 if (cmp == NULL) continue;
2143 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2144 Node* bol = cmp->fast_out(j);
2145 if (!bol->is_Bool()) continue;
2146 BoolTest::mask btest = bol->as_Bool()->_test._test;
2147 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2148 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2149 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2150 best_bol = bol->as_Bool();
2151 best_btest = btest;
2152 }
2153 }
2154 }
2155
2156 Node* answer_if_true = NULL;
2157 Node* answer_if_false = NULL;
2158 switch (best_btest) {
2159 default:
2160 if (cmpxy == NULL)
2161 cmpxy = ideal_cmpxy;
2162 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2163 // and fall through:
2164 case BoolTest::lt: // x < y
2165 case BoolTest::le: // x <= y
2166 answer_if_true = (want_max ? yvalue : xvalue);
2167 answer_if_false = (want_max ? xvalue : yvalue);
2168 break;
2169 case BoolTest::gt: // x > y
2170 case BoolTest::ge: // x >= y
2171 answer_if_true = (want_max ? xvalue : yvalue);
2172 answer_if_false = (want_max ? yvalue : xvalue);
2173 break;
2174 }
2175
2176 jint hi, lo;
2177 if (want_max) {
2178 // We can sharpen the minimum.
2179 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2180 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2181 } else {
2182 // We can sharpen the maximum.
2183 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2184 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2185 }
2186
2187 // Use a flow-free graph structure, to avoid creating excess control edges
2188 // which could hinder other optimizations.
2189 // Since Math.min/max is often used with arraycopy, we want
2190 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2191 Node* cmov = CMoveNode::make(NULL, best_bol,
2192 answer_if_false, answer_if_true,
2193 TypeInt::make(lo, hi, widen));
2194
2195 return _gvn.transform(cmov);
2196
2197 /*
2198 // This is not as desirable as it may seem, since Min and Max
2199 // nodes do not have a full set of optimizations.
2200 // And they would interfere, anyway, with 'if' optimizations
2201 // and with CMoveI canonical forms.
2202 switch (id) {
2203 case vmIntrinsics::_min:
2204 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2205 case vmIntrinsics::_max:
2206 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2207 default:
2208 ShouldNotReachHere();
2209 }
2210 */
2211 }
2212
2213 inline int
2214 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2215 const TypePtr* base_type = TypePtr::NULL_PTR;
2216 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2217 if (base_type == NULL) {
2218 // Unknown type.
2219 return Type::AnyPtr;
2220 } else if (base_type == TypePtr::NULL_PTR) {
2221 // Since this is a NULL+long form, we have to switch to a rawptr.
2222 base = _gvn.transform(new CastX2PNode(offset));
2223 offset = MakeConX(0);
2224 return Type::RawPtr;
2225 } else if (base_type->base() == Type::RawPtr) {
2226 return Type::RawPtr;
2227 } else if (base_type->isa_oopptr()) {
2228 // Base is never null => always a heap address.
2229 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2230 return Type::OopPtr;
2231 }
2232 // Offset is small => always a heap address.
2233 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2234 if (offset_type != NULL &&
2235 base_type->offset() == 0 && // (should always be?)
2236 offset_type->_lo >= 0 &&
2237 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2238 return Type::OopPtr;
2239 } else if (type == T_OBJECT) {
2240 // off heap access to an oop doesn't make any sense. Has to be on
2241 // heap.
2242 return Type::OopPtr;
2243 }
2244 // Otherwise, it might either be oop+off or NULL+addr.
2245 return Type::AnyPtr;
2246 } else {
2247 // No information:
2248 return Type::AnyPtr;
2249 }
2250 }
2251
2252 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type, bool can_cast) {
2253 Node* uncasted_base = base;
2254 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2255 if (kind == Type::RawPtr) {
2256 return basic_plus_adr(top(), uncasted_base, offset);
2257 } else if (kind == Type::AnyPtr) {
2258 assert(base == uncasted_base, "unexpected base change");
2259 if (can_cast) {
2260 if (!_gvn.type(base)->speculative_maybe_null() &&
2261 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2262 // According to profiling, this access is always on
2263 // heap. Casting the base to not null and thus avoiding membars
2264 // around the access should allow better optimizations
2265 Node* null_ctl = top();
2266 base = null_check_oop(base, &null_ctl, true, true, true);
2267 assert(null_ctl->is_top(), "no null control here");
2268 return basic_plus_adr(base, offset);
2269 } else if (_gvn.type(base)->speculative_always_null() &&
2270 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2271 // According to profiling, this access is always off
2272 // heap.
2273 base = null_assert(base);
2274 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2275 offset = MakeConX(0);
2276 return basic_plus_adr(top(), raw_base, offset);
2277 }
2278 }
2279 // We don't know if it's an on heap or off heap access. Fall back
2280 // to raw memory access.
2281 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2282 return basic_plus_adr(top(), raw, offset);
2283 } else {
2284 assert(base == uncasted_base, "unexpected base change");
2285 // We know it's an on heap access so base can't be null
2286 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2287 base = must_be_not_null(base, true);
2288 }
2289 return basic_plus_adr(base, offset);
2290 }
2291 }
2292
2293 //--------------------------inline_number_methods-----------------------------
2294 // inline int Integer.numberOfLeadingZeros(int)
2295 // inline int Long.numberOfLeadingZeros(long)
2296 //
2297 // inline int Integer.numberOfTrailingZeros(int)
2298 // inline int Long.numberOfTrailingZeros(long)
2299 //
2300 // inline int Integer.bitCount(int)
2301 // inline int Long.bitCount(long)
2302 //
2303 // inline char Character.reverseBytes(char)
2304 // inline short Short.reverseBytes(short)
2305 // inline int Integer.reverseBytes(int)
2306 // inline long Long.reverseBytes(long)
2307 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2308 Node* arg = argument(0);
2309 Node* n = NULL;
2310 switch (id) {
2311 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2312 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2313 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2314 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2315 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2316 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2317 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2318 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2319 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2320 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2321 default: fatal_unexpected_iid(id); break;
2322 }
2323 set_result(_gvn.transform(n));
2324 return true;
2325 }
2326
2327 //----------------------------inline_unsafe_access----------------------------
2328
2329 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2330 // Attempt to infer a sharper value type from the offset and base type.
2331 ciKlass* sharpened_klass = NULL;
2332
2333 // See if it is an instance field, with an object type.
2334 if (alias_type->field() != NULL) {
2335 if (alias_type->field()->type()->is_klass()) {
2336 sharpened_klass = alias_type->field()->type()->as_klass();
2337 }
2338 }
2339
2340 // See if it is a narrow oop array.
2341 if (adr_type->isa_aryptr()) {
2342 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2343 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2344 if (elem_type != NULL) {
2345 sharpened_klass = elem_type->klass();
2346 }
2347 }
2348 }
2349
2350 // The sharpened class might be unloaded if there is no class loader
2351 // contraint in place.
2352 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2353 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2354
2355 #ifndef PRODUCT
2356 if (C->print_intrinsics() || C->print_inlining()) {
2357 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2358 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2359 }
2360 #endif
2361 // Sharpen the value type.
2362 return tjp;
2363 }
2364 return NULL;
2365 }
2366
2367 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2368 switch (kind) {
2369 case Relaxed:
2370 return MO_UNORDERED;
2371 case Opaque:
2372 return MO_RELAXED;
2373 case Acquire:
2374 return MO_ACQUIRE;
2375 case Release:
2376 return MO_RELEASE;
2377 case Volatile:
2378 return MO_SEQ_CST;
2379 default:
2380 ShouldNotReachHere();
2381 return 0;
2382 }
2383 }
2384
2385 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2386 if (callee()->is_static()) return false; // caller must have the capability!
2387 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2388 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2389 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2390 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2391
2392 if (is_reference_type(type)) {
2393 decorators |= ON_UNKNOWN_OOP_REF;
2394 }
2395
2396 if (unaligned) {
2397 decorators |= C2_UNALIGNED;
2398 }
2399
2400 #ifndef PRODUCT
2401 {
2402 ResourceMark rm;
2403 // Check the signatures.
2404 ciSignature* sig = callee()->signature();
2405 #ifdef ASSERT
2406 if (!is_store) {
2407 // Object getReference(Object base, int/long offset), etc.
2408 BasicType rtype = sig->return_type()->basic_type();
2409 assert(rtype == type, "getter must return the expected value");
2410 assert(sig->count() == 2, "oop getter has 2 arguments");
2411 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2412 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2413 } else {
2414 // void putReference(Object base, int/long offset, Object x), etc.
2415 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2416 assert(sig->count() == 3, "oop putter has 3 arguments");
2417 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2418 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2419 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2420 assert(vtype == type, "putter must accept the expected value");
2421 }
2422 #endif // ASSERT
2423 }
2424 #endif //PRODUCT
2425
2426 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2427
2428 Node* receiver = argument(0); // type: oop
2429
2430 // Build address expression.
2431 Node* adr;
2432 Node* heap_base_oop = top();
2433 Node* offset = top();
2434 Node* val;
2435
2436 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2437 Node* base = argument(1); // type: oop
2438 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2439 offset = argument(2); // type: long
2440 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2441 // to be plain byte offsets, which are also the same as those accepted
2442 // by oopDesc::field_addr.
2443 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2444 "fieldOffset must be byte-scaled");
2445 // 32-bit machines ignore the high half!
2446 offset = ConvL2X(offset);
2447 adr = make_unsafe_address(base, offset, is_store ? ACCESS_WRITE : ACCESS_READ, type, kind == Relaxed);
2448
2449 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2450 if (type != T_OBJECT) {
2451 decorators |= IN_NATIVE; // off-heap primitive access
2452 } else {
2453 return false; // off-heap oop accesses are not supported
2454 }
2455 } else {
2456 heap_base_oop = base; // on-heap or mixed access
2457 }
2458
2459 // Can base be NULL? Otherwise, always on-heap access.
2460 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2461
2462 if (!can_access_non_heap) {
2463 decorators |= IN_HEAP;
2464 }
2465
2466 val = is_store ? argument(4) : NULL;
2467
2468 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2469 if (adr_type == TypePtr::NULL_PTR) {
2470 return false; // off-heap access with zero address
2471 }
2472
2473 // Try to categorize the address.
2474 Compile::AliasType* alias_type = C->alias_type(adr_type);
2475 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2476
2477 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2478 alias_type->adr_type() == TypeAryPtr::RANGE) {
2479 return false; // not supported
2480 }
2481
2482 bool mismatched = false;
2483 BasicType bt = alias_type->basic_type();
2484 if (bt != T_ILLEGAL) {
2485 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2486 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2487 // Alias type doesn't differentiate between byte[] and boolean[]).
2488 // Use address type to get the element type.
2489 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2490 }
2491 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2492 // accessing an array field with getReference is not a mismatch
2493 bt = T_OBJECT;
2494 }
2495 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2496 // Don't intrinsify mismatched object accesses
2497 return false;
2498 }
2499 mismatched = (bt != type);
2500 } else if (alias_type->adr_type()->isa_oopptr()) {
2501 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2502 }
2503
2504 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2505
2506 if (mismatched) {
2507 decorators |= C2_MISMATCHED;
2508 }
2509
2510 // First guess at the value type.
2511 const Type *value_type = Type::get_const_basic_type(type);
2512
2513 // Figure out the memory ordering.
2514 decorators |= mo_decorator_for_access_kind(kind);
2515
2516 if (!is_store && type == T_OBJECT) {
2517 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2518 if (tjp != NULL) {
2519 value_type = tjp;
2520 }
2521 }
2522
2523 receiver = null_check(receiver);
2524 if (stopped()) {
2525 return true;
2526 }
2527 // Heap pointers get a null-check from the interpreter,
2528 // as a courtesy. However, this is not guaranteed by Unsafe,
2529 // and it is not possible to fully distinguish unintended nulls
2530 // from intended ones in this API.
2531
2532 if (!is_store) {
2533 Node* p = NULL;
2534 // Try to constant fold a load from a constant field
2535 ciField* field = alias_type->field();
2536 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2537 // final or stable field
2538 p = make_constant_from_field(field, heap_base_oop);
2539 }
2540
2541 if (p == NULL) { // Could not constant fold the load
2542 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2543 // Normalize the value returned by getBoolean in the following cases
2544 if (type == T_BOOLEAN &&
2545 (mismatched ||
2546 heap_base_oop == top() || // - heap_base_oop is NULL or
2547 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2548 // and the unsafe access is made to large offset
2549 // (i.e., larger than the maximum offset necessary for any
2550 // field access)
2551 ) {
2552 IdealKit ideal = IdealKit(this);
2553 #define __ ideal.
2554 IdealVariable normalized_result(ideal);
2555 __ declarations_done();
2556 __ set(normalized_result, p);
2557 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2558 __ set(normalized_result, ideal.ConI(1));
2559 ideal.end_if();
2560 final_sync(ideal);
2561 p = __ value(normalized_result);
2562 #undef __
2563 }
2564 }
2565 if (type == T_ADDRESS) {
2566 p = gvn().transform(new CastP2XNode(NULL, p));
2567 p = ConvX2UL(p);
2568 }
2569 // The load node has the control of the preceding MemBarCPUOrder. All
2570 // following nodes will have the control of the MemBarCPUOrder inserted at
2571 // the end of this method. So, pushing the load onto the stack at a later
2572 // point is fine.
2573 set_result(p);
2574 } else {
2575 if (bt == T_ADDRESS) {
2576 // Repackage the long as a pointer.
2577 val = ConvL2X(val);
2578 val = gvn().transform(new CastX2PNode(val));
2579 }
2580 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2581 }
2582
2583 return true;
2584 }
2585
2586 //----------------------------inline_unsafe_load_store----------------------------
2587 // This method serves a couple of different customers (depending on LoadStoreKind):
2588 //
2589 // LS_cmp_swap:
2590 //
2591 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2592 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2593 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2594 //
2595 // LS_cmp_swap_weak:
2596 //
2597 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2598 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2599 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2600 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2601 //
2602 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2603 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2604 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2605 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2606 //
2607 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2608 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2609 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2610 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2611 //
2612 // LS_cmp_exchange:
2613 //
2614 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2615 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2616 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2617 //
2618 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2619 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2620 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2621 //
2622 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2623 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2624 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2625 //
2626 // LS_get_add:
2627 //
2628 // int getAndAddInt( Object o, long offset, int delta)
2629 // long getAndAddLong(Object o, long offset, long delta)
2630 //
2631 // LS_get_set:
2632 //
2633 // int getAndSet(Object o, long offset, int newValue)
2634 // long getAndSet(Object o, long offset, long newValue)
2635 // Object getAndSet(Object o, long offset, Object newValue)
2636 //
2637 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2638 // This basic scheme here is the same as inline_unsafe_access, but
2639 // differs in enough details that combining them would make the code
2640 // overly confusing. (This is a true fact! I originally combined
2641 // them, but even I was confused by it!) As much code/comments as
2642 // possible are retained from inline_unsafe_access though to make
2643 // the correspondences clearer. - dl
2644
2645 if (callee()->is_static()) return false; // caller must have the capability!
2646
2647 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2648 decorators |= mo_decorator_for_access_kind(access_kind);
2649
2650 #ifndef PRODUCT
2651 BasicType rtype;
2652 {
2653 ResourceMark rm;
2654 // Check the signatures.
2655 ciSignature* sig = callee()->signature();
2656 rtype = sig->return_type()->basic_type();
2657 switch(kind) {
2658 case LS_get_add:
2659 case LS_get_set: {
2660 // Check the signatures.
2661 #ifdef ASSERT
2662 assert(rtype == type, "get and set must return the expected type");
2663 assert(sig->count() == 3, "get and set has 3 arguments");
2664 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2665 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2666 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2667 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2668 #endif // ASSERT
2669 break;
2670 }
2671 case LS_cmp_swap:
2672 case LS_cmp_swap_weak: {
2673 // Check the signatures.
2674 #ifdef ASSERT
2675 assert(rtype == T_BOOLEAN, "CAS must return boolean");
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 case LS_cmp_exchange: {
2683 // Check the signatures.
2684 #ifdef ASSERT
2685 assert(rtype == type, "CAS must return the expected type");
2686 assert(sig->count() == 4, "CAS has 4 arguments");
2687 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2688 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2689 #endif // ASSERT
2690 break;
2691 }
2692 default:
2693 ShouldNotReachHere();
2694 }
2695 }
2696 #endif //PRODUCT
2697
2698 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2699
2700 // Get arguments:
2701 Node* receiver = NULL;
2702 Node* base = NULL;
2703 Node* offset = NULL;
2704 Node* oldval = NULL;
2705 Node* newval = NULL;
2706 switch(kind) {
2707 case LS_cmp_swap:
2708 case LS_cmp_swap_weak:
2709 case LS_cmp_exchange: {
2710 const bool two_slot_type = type2size[type] == 2;
2711 receiver = argument(0); // type: oop
2712 base = argument(1); // type: oop
2713 offset = argument(2); // type: long
2714 oldval = argument(4); // type: oop, int, or long
2715 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2716 break;
2717 }
2718 case LS_get_add:
2719 case LS_get_set: {
2720 receiver = argument(0); // type: oop
2721 base = argument(1); // type: oop
2722 offset = argument(2); // type: long
2723 oldval = NULL;
2724 newval = argument(4); // type: oop, int, or long
2725 break;
2726 }
2727 default:
2728 ShouldNotReachHere();
2729 }
2730
2731 // Build field offset expression.
2732 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2733 // to be plain byte offsets, which are also the same as those accepted
2734 // by oopDesc::field_addr.
2735 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2736 // 32-bit machines ignore the high half of long offsets
2737 offset = ConvL2X(offset);
2738 Node* adr = make_unsafe_address(base, offset, ACCESS_WRITE | ACCESS_READ, type, false);
2739 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2740
2741 Compile::AliasType* alias_type = C->alias_type(adr_type);
2742 BasicType bt = alias_type->basic_type();
2743 if (bt != T_ILLEGAL &&
2744 (is_reference_type(bt) != (type == T_OBJECT))) {
2745 // Don't intrinsify mismatched object accesses.
2746 return false;
2747 }
2748
2749 // For CAS, unlike inline_unsafe_access, there seems no point in
2750 // trying to refine types. Just use the coarse types here.
2751 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2752 const Type *value_type = Type::get_const_basic_type(type);
2753
2754 switch (kind) {
2755 case LS_get_set:
2756 case LS_cmp_exchange: {
2757 if (type == T_OBJECT) {
2758 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2759 if (tjp != NULL) {
2760 value_type = tjp;
2761 }
2762 }
2763 break;
2764 }
2765 case LS_cmp_swap:
2766 case LS_cmp_swap_weak:
2767 case LS_get_add:
2768 break;
2769 default:
2770 ShouldNotReachHere();
2771 }
2772
2773 // Null check receiver.
2774 receiver = null_check(receiver);
2775 if (stopped()) {
2776 return true;
2777 }
2778
2779 int alias_idx = C->get_alias_index(adr_type);
2780
2781 if (is_reference_type(type)) {
2782 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2783
2784 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2785 // could be delayed during Parse (for example, in adjust_map_after_if()).
2786 // Execute transformation here to avoid barrier generation in such case.
2787 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2788 newval = _gvn.makecon(TypePtr::NULL_PTR);
2789
2790 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2791 // Refine the value to a null constant, when it is known to be null
2792 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2793 }
2794 }
2795
2796 Node* result = NULL;
2797 switch (kind) {
2798 case LS_cmp_exchange: {
2799 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2800 oldval, newval, value_type, type, decorators);
2801 break;
2802 }
2803 case LS_cmp_swap_weak:
2804 decorators |= C2_WEAK_CMPXCHG;
2805 case LS_cmp_swap: {
2806 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2807 oldval, newval, value_type, type, decorators);
2808 break;
2809 }
2810 case LS_get_set: {
2811 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2812 newval, value_type, type, decorators);
2813 break;
2814 }
2815 case LS_get_add: {
2816 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2817 newval, value_type, type, decorators);
2818 break;
2819 }
2820 default:
2821 ShouldNotReachHere();
2822 }
2823
2824 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2825 set_result(result);
2826 return true;
2827 }
2828
2829 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2830 // Regardless of form, don't allow previous ld/st to move down,
2831 // then issue acquire, release, or volatile mem_bar.
2832 insert_mem_bar(Op_MemBarCPUOrder);
2833 switch(id) {
2834 case vmIntrinsics::_loadFence:
2835 insert_mem_bar(Op_LoadFence);
2836 return true;
2837 case vmIntrinsics::_storeFence:
2838 insert_mem_bar(Op_StoreFence);
2839 return true;
2840 case vmIntrinsics::_fullFence:
2841 insert_mem_bar(Op_MemBarVolatile);
2842 return true;
2843 default:
2844 fatal_unexpected_iid(id);
2845 return false;
2846 }
2847 }
2848
2849 bool LibraryCallKit::inline_onspinwait() {
2850 insert_mem_bar(Op_OnSpinWait);
2851 return true;
2852 }
2853
2854 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2855 if (!kls->is_Con()) {
2856 return true;
2857 }
2858 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
2859 if (klsptr == NULL) {
2860 return true;
2861 }
2862 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
2863 // don't need a guard for a klass that is already initialized
2864 return !ik->is_initialized();
2865 }
2866
2867 //----------------------------inline_unsafe_writeback0-------------------------
2868 // public native void Unsafe.writeback0(long address)
2869 bool LibraryCallKit::inline_unsafe_writeback0() {
2870 if (!Matcher::has_match_rule(Op_CacheWB)) {
2871 return false;
2872 }
2873 #ifndef PRODUCT
2874 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2875 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2876 ciSignature* sig = callee()->signature();
2877 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2878 #endif
2879 null_check_receiver(); // null-check, then ignore
2880 Node *addr = argument(1);
2881 addr = new CastX2PNode(addr);
2882 addr = _gvn.transform(addr);
2883 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2884 flush = _gvn.transform(flush);
2885 set_memory(flush, TypeRawPtr::BOTTOM);
2886 return true;
2887 }
2888
2889 //----------------------------inline_unsafe_writeback0-------------------------
2890 // public native void Unsafe.writeback0(long address)
2891 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2892 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2893 return false;
2894 }
2895 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2896 return false;
2897 }
2898 #ifndef PRODUCT
2899 assert(Matcher::has_match_rule(Op_CacheWB),
2900 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2901 : "found match rule for CacheWBPostSync but not CacheWB"));
2902
2903 #endif
2904 null_check_receiver(); // null-check, then ignore
2905 Node *sync;
2906 if (is_pre) {
2907 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2908 } else {
2909 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2910 }
2911 sync = _gvn.transform(sync);
2912 set_memory(sync, TypeRawPtr::BOTTOM);
2913 return true;
2914 }
2915
2916 //----------------------------inline_unsafe_allocate---------------------------
2917 // public native Object Unsafe.allocateInstance(Class<?> cls);
2918 bool LibraryCallKit::inline_unsafe_allocate() {
2919 if (callee()->is_static()) return false; // caller must have the capability!
2920
2921 null_check_receiver(); // null-check, then ignore
2922 Node* cls = null_check(argument(1));
2923 if (stopped()) return true;
2924
2925 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2926 kls = null_check(kls);
2927 if (stopped()) return true; // argument was like int.class
2928
2929 Node* test = NULL;
2930 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2931 // Note: The argument might still be an illegal value like
2932 // Serializable.class or Object[].class. The runtime will handle it.
2933 // But we must make an explicit check for initialization.
2934 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2935 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2936 // can generate code to load it as unsigned byte.
2937 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2938 Node* bits = intcon(InstanceKlass::fully_initialized);
2939 test = _gvn.transform(new SubINode(inst, bits));
2940 // The 'test' is non-zero if we need to take a slow path.
2941 }
2942
2943 Node* obj = new_instance(kls, test);
2944 set_result(obj);
2945 return true;
2946 }
2947
2948 //------------------------inline_native_time_funcs--------------
2949 // inline code for System.currentTimeMillis() and System.nanoTime()
2950 // these have the same type and signature
2951 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2952 const TypeFunc* tf = OptoRuntime::void_long_Type();
2953 const TypePtr* no_memory_effects = NULL;
2954 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2955 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2956 #ifdef ASSERT
2957 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2958 assert(value_top == top(), "second value must be top");
2959 #endif
2960 set_result(value);
2961 return true;
2962 }
2963
2964 #ifdef JFR_HAVE_INTRINSICS
2965
2966 /*
2967 * oop -> myklass
2968 * myklass->trace_id |= USED
2969 * return myklass->trace_id & ~0x3
2970 */
2971 bool LibraryCallKit::inline_native_classID() {
2972 Node* cls = null_check(argument(0), T_OBJECT);
2973 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2974 kls = null_check(kls, T_OBJECT);
2975
2976 ByteSize offset = KLASS_TRACE_ID_OFFSET;
2977 Node* insp = basic_plus_adr(kls, in_bytes(offset));
2978 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
2979
2980 Node* clsused = longcon(0x01l); // set the class bit
2981 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
2982 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
2983 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
2984
2985 #ifdef TRACE_ID_META_BITS
2986 Node* mbits = longcon(~TRACE_ID_META_BITS);
2987 tvalue = _gvn.transform(new AndLNode(tvalue, mbits));
2988 #endif
2989 #ifdef TRACE_ID_SHIFT
2990 Node* cbits = intcon(TRACE_ID_SHIFT);
2991 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits));
2992 #endif
2993
2994 set_result(tvalue);
2995 return true;
2996
2997 }
2998
2999 bool LibraryCallKit::inline_native_getEventWriter() {
3000 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3001
3002 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
3003 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3004
3005 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3006
3007 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
3008 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
3009
3010 IfNode* iff_jobj_null =
3011 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
3012
3013 enum { _normal_path = 1,
3014 _null_path = 2,
3015 PATH_LIMIT };
3016
3017 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3018 PhiNode* result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3019
3020 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
3021 result_rgn->init_req(_null_path, jobj_is_null);
3022 result_val->init_req(_null_path, null());
3023
3024 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
3025 set_control(jobj_is_not_null);
3026 Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT,
3027 IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3028 result_rgn->init_req(_normal_path, control());
3029 result_val->init_req(_normal_path, res);
3030
3031 set_result(result_rgn, result_val);
3032
3033 return true;
3034 }
3035
3036 #endif // JFR_HAVE_INTRINSICS
3037
3038 //------------------------inline_native_currentThread------------------
3039 bool LibraryCallKit::inline_native_currentThread() {
3040 Node* junk = NULL;
3041 set_result(generate_current_thread(junk));
3042 return true;
3043 }
3044
3045 //---------------------------load_mirror_from_klass----------------------------
3046 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3047 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3048 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3049 Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3050 // mirror = ((OopHandle)mirror)->resolve();
3051 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3052 }
3053
3054 //-----------------------load_klass_from_mirror_common-------------------------
3055 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3056 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3057 // and branch to the given path on the region.
3058 // If never_see_null, take an uncommon trap on null, so we can optimistically
3059 // compile for the non-null case.
3060 // If the region is NULL, force never_see_null = true.
3061 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3062 bool never_see_null,
3063 RegionNode* region,
3064 int null_path,
3065 int offset) {
3066 if (region == NULL) never_see_null = true;
3067 Node* p = basic_plus_adr(mirror, offset);
3068 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3069 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3070 Node* null_ctl = top();
3071 kls = null_check_oop(kls, &null_ctl, never_see_null);
3072 if (region != NULL) {
3073 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3074 region->init_req(null_path, null_ctl);
3075 } else {
3076 assert(null_ctl == top(), "no loose ends");
3077 }
3078 return kls;
3079 }
3080
3081 //--------------------(inline_native_Class_query helpers)---------------------
3082 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3083 // Fall through if (mods & mask) == bits, take the guard otherwise.
3084 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3085 // Branch around if the given klass has the given modifier bit set.
3086 // Like generate_guard, adds a new path onto the region.
3087 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3088 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3089 Node* mask = intcon(modifier_mask);
3090 Node* bits = intcon(modifier_bits);
3091 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3092 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3093 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3094 return generate_fair_guard(bol, region);
3095 }
3096 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3097 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3098 }
3099 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3100 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
3101 }
3102
3103 //-------------------------inline_native_Class_query-------------------
3104 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3105 const Type* return_type = TypeInt::BOOL;
3106 Node* prim_return_value = top(); // what happens if it's a primitive class?
3107 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3108 bool expect_prim = false; // most of these guys expect to work on refs
3109
3110 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3111
3112 Node* mirror = argument(0);
3113 Node* obj = top();
3114
3115 switch (id) {
3116 case vmIntrinsics::_isInstance:
3117 // nothing is an instance of a primitive type
3118 prim_return_value = intcon(0);
3119 obj = argument(1);
3120 break;
3121 case vmIntrinsics::_getModifiers:
3122 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3123 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3124 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3125 break;
3126 case vmIntrinsics::_isInterface:
3127 prim_return_value = intcon(0);
3128 break;
3129 case vmIntrinsics::_isArray:
3130 prim_return_value = intcon(0);
3131 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3132 break;
3133 case vmIntrinsics::_isPrimitive:
3134 prim_return_value = intcon(1);
3135 expect_prim = true; // obviously
3136 break;
3137 case vmIntrinsics::_isHidden:
3138 prim_return_value = intcon(0);
3139 break;
3140 case vmIntrinsics::_getSuperclass:
3141 prim_return_value = null();
3142 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3143 break;
3144 case vmIntrinsics::_getClassAccessFlags:
3145 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3146 return_type = TypeInt::INT; // not bool! 6297094
3147 break;
3148 default:
3149 fatal_unexpected_iid(id);
3150 break;
3151 }
3152
3153 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3154 if (mirror_con == NULL) return false; // cannot happen?
3155
3156 #ifndef PRODUCT
3157 if (C->print_intrinsics() || C->print_inlining()) {
3158 ciType* k = mirror_con->java_mirror_type();
3159 if (k) {
3160 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3161 k->print_name();
3162 tty->cr();
3163 }
3164 }
3165 #endif
3166
3167 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3168 RegionNode* region = new RegionNode(PATH_LIMIT);
3169 record_for_igvn(region);
3170 PhiNode* phi = new PhiNode(region, return_type);
3171
3172 // The mirror will never be null of Reflection.getClassAccessFlags, however
3173 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3174 // if it is. See bug 4774291.
3175
3176 // For Reflection.getClassAccessFlags(), the null check occurs in
3177 // the wrong place; see inline_unsafe_access(), above, for a similar
3178 // situation.
3179 mirror = null_check(mirror);
3180 // If mirror or obj is dead, only null-path is taken.
3181 if (stopped()) return true;
3182
3183 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3184
3185 // Now load the mirror's klass metaobject, and null-check it.
3186 // Side-effects region with the control path if the klass is null.
3187 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3188 // If kls is null, we have a primitive mirror.
3189 phi->init_req(_prim_path, prim_return_value);
3190 if (stopped()) { set_result(region, phi); return true; }
3191 bool safe_for_replace = (region->in(_prim_path) == top());
3192
3193 Node* p; // handy temp
3194 Node* null_ctl;
3195
3196 // Now that we have the non-null klass, we can perform the real query.
3197 // For constant classes, the query will constant-fold in LoadNode::Value.
3198 Node* query_value = top();
3199 switch (id) {
3200 case vmIntrinsics::_isInstance:
3201 // nothing is an instance of a primitive type
3202 query_value = gen_instanceof(obj, kls, safe_for_replace);
3203 break;
3204
3205 case vmIntrinsics::_getModifiers:
3206 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3207 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3208 break;
3209
3210 case vmIntrinsics::_isInterface:
3211 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3212 if (generate_interface_guard(kls, region) != NULL)
3213 // A guard was added. If the guard is taken, it was an interface.
3214 phi->add_req(intcon(1));
3215 // If we fall through, it's a plain class.
3216 query_value = intcon(0);
3217 break;
3218
3219 case vmIntrinsics::_isArray:
3220 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3221 if (generate_array_guard(kls, region) != NULL)
3222 // A guard was added. If the guard is taken, it was an array.
3223 phi->add_req(intcon(1));
3224 // If we fall through, it's a plain class.
3225 query_value = intcon(0);
3226 break;
3227
3228 case vmIntrinsics::_isPrimitive:
3229 query_value = intcon(0); // "normal" path produces false
3230 break;
3231
3232 case vmIntrinsics::_isHidden:
3233 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
3234 if (generate_hidden_class_guard(kls, region) != NULL)
3235 // A guard was added. If the guard is taken, it was an hidden class.
3236 phi->add_req(intcon(1));
3237 // If we fall through, it's a plain class.
3238 query_value = intcon(0);
3239 break;
3240
3241
3242 case vmIntrinsics::_getSuperclass:
3243 // The rules here are somewhat unfortunate, but we can still do better
3244 // with random logic than with a JNI call.
3245 // Interfaces store null or Object as _super, but must report null.
3246 // Arrays store an intermediate super as _super, but must report Object.
3247 // Other types can report the actual _super.
3248 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3249 if (generate_interface_guard(kls, region) != NULL)
3250 // A guard was added. If the guard is taken, it was an interface.
3251 phi->add_req(null());
3252 if (generate_array_guard(kls, region) != NULL)
3253 // A guard was added. If the guard is taken, it was an array.
3254 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3255 // If we fall through, it's a plain class. Get its _super.
3256 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3257 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3258 null_ctl = top();
3259 kls = null_check_oop(kls, &null_ctl);
3260 if (null_ctl != top()) {
3261 // If the guard is taken, Object.superClass is null (both klass and mirror).
3262 region->add_req(null_ctl);
3263 phi ->add_req(null());
3264 }
3265 if (!stopped()) {
3266 query_value = load_mirror_from_klass(kls);
3267 }
3268 break;
3269
3270 case vmIntrinsics::_getClassAccessFlags:
3271 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3272 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3273 break;
3274
3275 default:
3276 fatal_unexpected_iid(id);
3277 break;
3278 }
3279
3280 // Fall-through is the normal case of a query to a real class.
3281 phi->init_req(1, query_value);
3282 region->init_req(1, control());
3283
3284 C->set_has_split_ifs(true); // Has chance for split-if optimization
3285 set_result(region, phi);
3286 return true;
3287 }
3288
3289 //-------------------------inline_Class_cast-------------------
3290 bool LibraryCallKit::inline_Class_cast() {
3291 Node* mirror = argument(0); // Class
3292 Node* obj = argument(1);
3293 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3294 if (mirror_con == NULL) {
3295 return false; // dead path (mirror->is_top()).
3296 }
3297 if (obj == NULL || obj->is_top()) {
3298 return false; // dead path
3299 }
3300 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3301
3302 // First, see if Class.cast() can be folded statically.
3303 // java_mirror_type() returns non-null for compile-time Class constants.
3304 ciType* tm = mirror_con->java_mirror_type();
3305 if (tm != NULL && tm->is_klass() &&
3306 tp != NULL && tp->klass() != NULL) {
3307 if (!tp->klass()->is_loaded()) {
3308 // Don't use intrinsic when class is not loaded.
3309 return false;
3310 } else {
3311 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3312 if (static_res == Compile::SSC_always_true) {
3313 // isInstance() is true - fold the code.
3314 set_result(obj);
3315 return true;
3316 } else if (static_res == Compile::SSC_always_false) {
3317 // Don't use intrinsic, have to throw ClassCastException.
3318 // If the reference is null, the non-intrinsic bytecode will
3319 // be optimized appropriately.
3320 return false;
3321 }
3322 }
3323 }
3324
3325 // Bailout intrinsic and do normal inlining if exception path is frequent.
3326 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3327 return false;
3328 }
3329
3330 // Generate dynamic checks.
3331 // Class.cast() is java implementation of _checkcast bytecode.
3332 // Do checkcast (Parse::do_checkcast()) optimizations here.
3333
3334 mirror = null_check(mirror);
3335 // If mirror is dead, only null-path is taken.
3336 if (stopped()) {
3337 return true;
3338 }
3339
3340 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3341 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3342 RegionNode* region = new RegionNode(PATH_LIMIT);
3343 record_for_igvn(region);
3344
3345 // Now load the mirror's klass metaobject, and null-check it.
3346 // If kls is null, we have a primitive mirror and
3347 // nothing is an instance of a primitive type.
3348 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3349
3350 Node* res = top();
3351 if (!stopped()) {
3352 Node* bad_type_ctrl = top();
3353 // Do checkcast optimizations.
3354 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3355 region->init_req(_bad_type_path, bad_type_ctrl);
3356 }
3357 if (region->in(_prim_path) != top() ||
3358 region->in(_bad_type_path) != top()) {
3359 // Let Interpreter throw ClassCastException.
3360 PreserveJVMState pjvms(this);
3361 set_control(_gvn.transform(region));
3362 uncommon_trap(Deoptimization::Reason_intrinsic,
3363 Deoptimization::Action_maybe_recompile);
3364 }
3365 if (!stopped()) {
3366 set_result(res);
3367 }
3368 return true;
3369 }
3370
3371
3372 //--------------------------inline_native_subtype_check------------------------
3373 // This intrinsic takes the JNI calls out of the heart of
3374 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3375 bool LibraryCallKit::inline_native_subtype_check() {
3376 // Pull both arguments off the stack.
3377 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3378 args[0] = argument(0);
3379 args[1] = argument(1);
3380 Node* klasses[2]; // corresponding Klasses: superk, subk
3381 klasses[0] = klasses[1] = top();
3382
3383 enum {
3384 // A full decision tree on {superc is prim, subc is prim}:
3385 _prim_0_path = 1, // {P,N} => false
3386 // {P,P} & superc!=subc => false
3387 _prim_same_path, // {P,P} & superc==subc => true
3388 _prim_1_path, // {N,P} => false
3389 _ref_subtype_path, // {N,N} & subtype check wins => true
3390 _both_ref_path, // {N,N} & subtype check loses => false
3391 PATH_LIMIT
3392 };
3393
3394 RegionNode* region = new RegionNode(PATH_LIMIT);
3395 Node* phi = new PhiNode(region, TypeInt::BOOL);
3396 record_for_igvn(region);
3397
3398 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3399 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3400 int class_klass_offset = java_lang_Class::klass_offset();
3401
3402 // First null-check both mirrors and load each mirror's klass metaobject.
3403 int which_arg;
3404 for (which_arg = 0; which_arg <= 1; which_arg++) {
3405 Node* arg = args[which_arg];
3406 arg = null_check(arg);
3407 if (stopped()) break;
3408 args[which_arg] = arg;
3409
3410 Node* p = basic_plus_adr(arg, class_klass_offset);
3411 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3412 klasses[which_arg] = _gvn.transform(kls);
3413 }
3414
3415 // Having loaded both klasses, test each for null.
3416 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3417 for (which_arg = 0; which_arg <= 1; which_arg++) {
3418 Node* kls = klasses[which_arg];
3419 Node* null_ctl = top();
3420 kls = null_check_oop(kls, &null_ctl, never_see_null);
3421 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3422 region->init_req(prim_path, null_ctl);
3423 if (stopped()) break;
3424 klasses[which_arg] = kls;
3425 }
3426
3427 if (!stopped()) {
3428 // now we have two reference types, in klasses[0..1]
3429 Node* subk = klasses[1]; // the argument to isAssignableFrom
3430 Node* superk = klasses[0]; // the receiver
3431 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3432 // now we have a successful reference subtype check
3433 region->set_req(_ref_subtype_path, control());
3434 }
3435
3436 // If both operands are primitive (both klasses null), then
3437 // we must return true when they are identical primitives.
3438 // It is convenient to test this after the first null klass check.
3439 set_control(region->in(_prim_0_path)); // go back to first null check
3440 if (!stopped()) {
3441 // Since superc is primitive, make a guard for the superc==subc case.
3442 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3443 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3444 generate_guard(bol_eq, region, PROB_FAIR);
3445 if (region->req() == PATH_LIMIT+1) {
3446 // A guard was added. If the added guard is taken, superc==subc.
3447 region->swap_edges(PATH_LIMIT, _prim_same_path);
3448 region->del_req(PATH_LIMIT);
3449 }
3450 region->set_req(_prim_0_path, control()); // Not equal after all.
3451 }
3452
3453 // these are the only paths that produce 'true':
3454 phi->set_req(_prim_same_path, intcon(1));
3455 phi->set_req(_ref_subtype_path, intcon(1));
3456
3457 // pull together the cases:
3458 assert(region->req() == PATH_LIMIT, "sane region");
3459 for (uint i = 1; i < region->req(); i++) {
3460 Node* ctl = region->in(i);
3461 if (ctl == NULL || ctl == top()) {
3462 region->set_req(i, top());
3463 phi ->set_req(i, top());
3464 } else if (phi->in(i) == NULL) {
3465 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3466 }
3467 }
3468
3469 set_control(_gvn.transform(region));
3470 set_result(_gvn.transform(phi));
3471 return true;
3472 }
3473
3474 //---------------------generate_array_guard_common------------------------
3475 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3476 bool obj_array, bool not_array) {
3477
3478 if (stopped()) {
3479 return NULL;
3480 }
3481
3482 // If obj_array/non_array==false/false:
3483 // Branch around if the given klass is in fact an array (either obj or prim).
3484 // If obj_array/non_array==false/true:
3485 // Branch around if the given klass is not an array klass of any kind.
3486 // If obj_array/non_array==true/true:
3487 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3488 // If obj_array/non_array==true/false:
3489 // Branch around if the kls is an oop array (Object[] or subtype)
3490 //
3491 // Like generate_guard, adds a new path onto the region.
3492 jint layout_con = 0;
3493 Node* layout_val = get_layout_helper(kls, layout_con);
3494 if (layout_val == NULL) {
3495 bool query = (obj_array
3496 ? Klass::layout_helper_is_objArray(layout_con)
3497 : Klass::layout_helper_is_array(layout_con));
3498 if (query == not_array) {
3499 return NULL; // never a branch
3500 } else { // always a branch
3501 Node* always_branch = control();
3502 if (region != NULL)
3503 region->add_req(always_branch);
3504 set_control(top());
3505 return always_branch;
3506 }
3507 }
3508 // Now test the correct condition.
3509 jint nval = (obj_array
3510 ? (jint)(Klass::_lh_array_tag_type_value
3511 << Klass::_lh_array_tag_shift)
3512 : Klass::_lh_neutral_value);
3513 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3514 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3515 // invert the test if we are looking for a non-array
3516 if (not_array) btest = BoolTest(btest).negate();
3517 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3518 return generate_fair_guard(bol, region);
3519 }
3520
3521
3522 //-----------------------inline_native_newArray--------------------------
3523 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3524 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3525 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3526 Node* mirror;
3527 Node* count_val;
3528 if (uninitialized) {
3529 mirror = argument(1);
3530 count_val = argument(2);
3531 } else {
3532 mirror = argument(0);
3533 count_val = argument(1);
3534 }
3535
3536 mirror = null_check(mirror);
3537 // If mirror or obj is dead, only null-path is taken.
3538 if (stopped()) return true;
3539
3540 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3541 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3542 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3543 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3544 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3545
3546 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3547 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3548 result_reg, _slow_path);
3549 Node* normal_ctl = control();
3550 Node* no_array_ctl = result_reg->in(_slow_path);
3551
3552 // Generate code for the slow case. We make a call to newArray().
3553 set_control(no_array_ctl);
3554 if (!stopped()) {
3555 // Either the input type is void.class, or else the
3556 // array klass has not yet been cached. Either the
3557 // ensuing call will throw an exception, or else it
3558 // will cache the array klass for next time.
3559 PreserveJVMState pjvms(this);
3560 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3561 Node* slow_result = set_results_for_java_call(slow_call);
3562 // this->control() comes from set_results_for_java_call
3563 result_reg->set_req(_slow_path, control());
3564 result_val->set_req(_slow_path, slow_result);
3565 result_io ->set_req(_slow_path, i_o());
3566 result_mem->set_req(_slow_path, reset_memory());
3567 }
3568
3569 set_control(normal_ctl);
3570 if (!stopped()) {
3571 // Normal case: The array type has been cached in the java.lang.Class.
3572 // The following call works fine even if the array type is polymorphic.
3573 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3574 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3575 result_reg->init_req(_normal_path, control());
3576 result_val->init_req(_normal_path, obj);
3577 result_io ->init_req(_normal_path, i_o());
3578 result_mem->init_req(_normal_path, reset_memory());
3579
3580 if (uninitialized) {
3581 // Mark the allocation so that zeroing is skipped
3582 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3583 alloc->maybe_set_complete(&_gvn);
3584 }
3585 }
3586
3587 // Return the combined state.
3588 set_i_o( _gvn.transform(result_io) );
3589 set_all_memory( _gvn.transform(result_mem));
3590
3591 C->set_has_split_ifs(true); // Has chance for split-if optimization
3592 set_result(result_reg, result_val);
3593 return true;
3594 }
3595
3596 //----------------------inline_native_getLength--------------------------
3597 // public static native int java.lang.reflect.Array.getLength(Object array);
3598 bool LibraryCallKit::inline_native_getLength() {
3599 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3600
3601 Node* array = null_check(argument(0));
3602 // If array is dead, only null-path is taken.
3603 if (stopped()) return true;
3604
3605 // Deoptimize if it is a non-array.
3606 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3607
3608 if (non_array != NULL) {
3609 PreserveJVMState pjvms(this);
3610 set_control(non_array);
3611 uncommon_trap(Deoptimization::Reason_intrinsic,
3612 Deoptimization::Action_maybe_recompile);
3613 }
3614
3615 // If control is dead, only non-array-path is taken.
3616 if (stopped()) return true;
3617
3618 // The works fine even if the array type is polymorphic.
3619 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3620 Node* result = load_array_length(array);
3621
3622 C->set_has_split_ifs(true); // Has chance for split-if optimization
3623 set_result(result);
3624 return true;
3625 }
3626
3627 //------------------------inline_array_copyOf----------------------------
3628 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3629 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3630 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3631 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3632
3633 // Get the arguments.
3634 Node* original = argument(0);
3635 Node* start = is_copyOfRange? argument(1): intcon(0);
3636 Node* end = is_copyOfRange? argument(2): argument(1);
3637 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3638
3639 Node* newcopy = NULL;
3640
3641 // Set the original stack and the reexecute bit for the interpreter to reexecute
3642 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3643 { PreserveReexecuteState preexecs(this);
3644 jvms()->set_should_reexecute(true);
3645
3646 array_type_mirror = null_check(array_type_mirror);
3647 original = null_check(original);
3648
3649 // Check if a null path was taken unconditionally.
3650 if (stopped()) return true;
3651
3652 Node* orig_length = load_array_length(original);
3653
3654 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3655 klass_node = null_check(klass_node);
3656
3657 RegionNode* bailout = new RegionNode(1);
3658 record_for_igvn(bailout);
3659
3660 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3661 // Bail out if that is so.
3662 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3663 if (not_objArray != NULL) {
3664 // Improve the klass node's type from the new optimistic assumption:
3665 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3666 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3667 Node* cast = new CastPPNode(klass_node, akls);
3668 cast->init_req(0, control());
3669 klass_node = _gvn.transform(cast);
3670 }
3671
3672 // Bail out if either start or end is negative.
3673 generate_negative_guard(start, bailout, &start);
3674 generate_negative_guard(end, bailout, &end);
3675
3676 Node* length = end;
3677 if (_gvn.type(start) != TypeInt::ZERO) {
3678 length = _gvn.transform(new SubINode(end, start));
3679 }
3680
3681 // Bail out if length is negative.
3682 // Without this the new_array would throw
3683 // NegativeArraySizeException but IllegalArgumentException is what
3684 // should be thrown
3685 generate_negative_guard(length, bailout, &length);
3686
3687 if (bailout->req() > 1) {
3688 PreserveJVMState pjvms(this);
3689 set_control(_gvn.transform(bailout));
3690 uncommon_trap(Deoptimization::Reason_intrinsic,
3691 Deoptimization::Action_maybe_recompile);
3692 }
3693
3694 if (!stopped()) {
3695 // How many elements will we copy from the original?
3696 // The answer is MinI(orig_length - start, length).
3697 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3698 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3699
3700 // Generate a direct call to the right arraycopy function(s).
3701 // We know the copy is disjoint but we might not know if the
3702 // oop stores need checking.
3703 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3704 // This will fail a store-check if x contains any non-nulls.
3705
3706 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3707 // loads/stores but it is legal only if we're sure the
3708 // Arrays.copyOf would succeed. So we need all input arguments
3709 // to the copyOf to be validated, including that the copy to the
3710 // new array won't trigger an ArrayStoreException. That subtype
3711 // check can be optimized if we know something on the type of
3712 // the input array from type speculation.
3713 if (_gvn.type(klass_node)->singleton()) {
3714 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3715 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3716
3717 int test = C->static_subtype_check(superk, subk);
3718 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3719 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3720 if (t_original->speculative_type() != NULL) {
3721 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3722 }
3723 }
3724 }
3725
3726 bool validated = false;
3727 // Reason_class_check rather than Reason_intrinsic because we
3728 // want to intrinsify even if this traps.
3729 if (!too_many_traps(Deoptimization::Reason_class_check)) {
3730 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
3731
3732 if (not_subtype_ctrl != top()) {
3733 PreserveJVMState pjvms(this);
3734 set_control(not_subtype_ctrl);
3735 uncommon_trap(Deoptimization::Reason_class_check,
3736 Deoptimization::Action_make_not_entrant);
3737 assert(stopped(), "Should be stopped");
3738 }
3739 validated = true;
3740 }
3741
3742 if (!stopped()) {
3743 newcopy = new_array(klass_node, length, 0); // no arguments to push
3744
3745 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
3746 load_object_klass(original), klass_node);
3747 if (!is_copyOfRange) {
3748 ac->set_copyof(validated);
3749 } else {
3750 ac->set_copyofrange(validated);
3751 }
3752 Node* n = _gvn.transform(ac);
3753 if (n == ac) {
3754 ac->connect_outputs(this);
3755 } else {
3756 assert(validated, "shouldn't transform if all arguments not validated");
3757 set_all_memory(n);
3758 }
3759 }
3760 }
3761 } // original reexecute is set back here
3762
3763 C->set_has_split_ifs(true); // Has chance for split-if optimization
3764 if (!stopped()) {
3765 set_result(newcopy);
3766 }
3767 return true;
3768 }
3769
3770
3771 //----------------------generate_virtual_guard---------------------------
3772 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
3773 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
3774 RegionNode* slow_region) {
3775 ciMethod* method = callee();
3776 int vtable_index = method->vtable_index();
3777 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3778 "bad index %d", vtable_index);
3779 // Get the Method* out of the appropriate vtable entry.
3780 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
3781 vtable_index*vtableEntry::size_in_bytes() +
3782 vtableEntry::method_offset_in_bytes();
3783 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
3784 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3785
3786 // Compare the target method with the expected method (e.g., Object.hashCode).
3787 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
3788
3789 Node* native_call = makecon(native_call_addr);
3790 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
3791 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
3792
3793 return generate_slow_guard(test_native, slow_region);
3794 }
3795
3796 //-----------------------generate_method_call----------------------------
3797 // Use generate_method_call to make a slow-call to the real
3798 // method if the fast path fails. An alternative would be to
3799 // use a stub like OptoRuntime::slow_arraycopy_Java.
3800 // This only works for expanding the current library call,
3801 // not another intrinsic. (E.g., don't use this for making an
3802 // arraycopy call inside of the copyOf intrinsic.)
3803 CallJavaNode*
3804 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
3805 // When compiling the intrinsic method itself, do not use this technique.
3806 guarantee(callee() != C->method(), "cannot make slow-call to self");
3807
3808 ciMethod* method = callee();
3809 // ensure the JVMS we have will be correct for this call
3810 guarantee(method_id == method->intrinsic_id(), "must match");
3811
3812 const TypeFunc* tf = TypeFunc::make(method);
3813 CallJavaNode* slow_call;
3814 if (is_static) {
3815 assert(!is_virtual, "");
3816 slow_call = new CallStaticJavaNode(C, tf,
3817 SharedRuntime::get_resolve_static_call_stub(),
3818 method, bci());
3819 } else if (is_virtual) {
3820 null_check_receiver();
3821 int vtable_index = Method::invalid_vtable_index;
3822 if (UseInlineCaches) {
3823 // Suppress the vtable call
3824 } else {
3825 // hashCode and clone are not a miranda methods,
3826 // so the vtable index is fixed.
3827 // No need to use the linkResolver to get it.
3828 vtable_index = method->vtable_index();
3829 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3830 "bad index %d", vtable_index);
3831 }
3832 slow_call = new CallDynamicJavaNode(tf,
3833 SharedRuntime::get_resolve_virtual_call_stub(),
3834 method, vtable_index, bci());
3835 } else { // neither virtual nor static: opt_virtual
3836 null_check_receiver();
3837 slow_call = new CallStaticJavaNode(C, tf,
3838 SharedRuntime::get_resolve_opt_virtual_call_stub(),
3839 method, bci());
3840 slow_call->set_optimized_virtual(true);
3841 }
3842 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
3843 // To be able to issue a direct call (optimized virtual or virtual)
3844 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
3845 // about the method being invoked should be attached to the call site to
3846 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
3847 slow_call->set_override_symbolic_info(true);
3848 }
3849 set_arguments_for_java_call(slow_call);
3850 set_edges_for_java_call(slow_call);
3851 return slow_call;
3852 }
3853
3854
3855 /**
3856 * Build special case code for calls to hashCode on an object. This call may
3857 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
3858 * slightly different code.
3859 */
3860 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
3861 assert(is_static == callee()->is_static(), "correct intrinsic selection");
3862 assert(!(is_virtual && is_static), "either virtual, special, or static");
3863
3864 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
3865
3866 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3867 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
3868 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3869 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3870 Node* obj = NULL;
3871 if (!is_static) {
3872 // Check for hashing null object
3873 obj = null_check_receiver();
3874 if (stopped()) return true; // unconditionally null
3875 result_reg->init_req(_null_path, top());
3876 result_val->init_req(_null_path, top());
3877 } else {
3878 // Do a null check, and return zero if null.
3879 // System.identityHashCode(null) == 0
3880 obj = argument(0);
3881 Node* null_ctl = top();
3882 obj = null_check_oop(obj, &null_ctl);
3883 result_reg->init_req(_null_path, null_ctl);
3884 result_val->init_req(_null_path, _gvn.intcon(0));
3885 }
3886
3887 // Unconditionally null? Then return right away.
3888 if (stopped()) {
3889 set_control( result_reg->in(_null_path));
3890 if (!stopped())
3891 set_result(result_val->in(_null_path));
3892 return true;
3893 }
3894
3895 // We only go to the fast case code if we pass a number of guards. The
3896 // paths which do not pass are accumulated in the slow_region.
3897 RegionNode* slow_region = new RegionNode(1);
3898 record_for_igvn(slow_region);
3899
3900 // If this is a virtual call, we generate a funny guard. We pull out
3901 // the vtable entry corresponding to hashCode() from the target object.
3902 // If the target method which we are calling happens to be the native
3903 // Object hashCode() method, we pass the guard. We do not need this
3904 // guard for non-virtual calls -- the caller is known to be the native
3905 // Object hashCode().
3906 if (is_virtual) {
3907 // After null check, get the object's klass.
3908 Node* obj_klass = load_object_klass(obj);
3909 generate_virtual_guard(obj_klass, slow_region);
3910 }
3911
3912 // Get the header out of the object, use LoadMarkNode when available
3913 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
3914 // The control of the load must be NULL. Otherwise, the load can move before
3915 // the null check after castPP removal.
3916 Node* no_ctrl = NULL;
3917 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
3918
3919 // Test the header to see if it is unlocked.
3920 Node *lock_mask = _gvn.MakeConX(markWord::biased_lock_mask_in_place);
3921 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
3922 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
3923 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
3924 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
3925
3926 generate_slow_guard(test_unlocked, slow_region);
3927
3928 // Get the hash value and check to see that it has been properly assigned.
3929 // We depend on hash_mask being at most 32 bits and avoid the use of
3930 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
3931 // vm: see markWord.hpp.
3932 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
3933 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
3934 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
3935 // This hack lets the hash bits live anywhere in the mark object now, as long
3936 // as the shift drops the relevant bits into the low 32 bits. Note that
3937 // Java spec says that HashCode is an int so there's no point in capturing
3938 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
3939 hshifted_header = ConvX2I(hshifted_header);
3940 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
3941
3942 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
3943 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
3944 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
3945
3946 generate_slow_guard(test_assigned, slow_region);
3947
3948 Node* init_mem = reset_memory();
3949 // fill in the rest of the null path:
3950 result_io ->init_req(_null_path, i_o());
3951 result_mem->init_req(_null_path, init_mem);
3952
3953 result_val->init_req(_fast_path, hash_val);
3954 result_reg->init_req(_fast_path, control());
3955 result_io ->init_req(_fast_path, i_o());
3956 result_mem->init_req(_fast_path, init_mem);
3957
3958 // Generate code for the slow case. We make a call to hashCode().
3959 set_control(_gvn.transform(slow_region));
3960 if (!stopped()) {
3961 // No need for PreserveJVMState, because we're using up the present state.
3962 set_all_memory(init_mem);
3963 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
3964 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
3965 Node* slow_result = set_results_for_java_call(slow_call);
3966 // this->control() comes from set_results_for_java_call
3967 result_reg->init_req(_slow_path, control());
3968 result_val->init_req(_slow_path, slow_result);
3969 result_io ->set_req(_slow_path, i_o());
3970 result_mem ->set_req(_slow_path, reset_memory());
3971 }
3972
3973 // Return the combined state.
3974 set_i_o( _gvn.transform(result_io) );
3975 set_all_memory( _gvn.transform(result_mem));
3976
3977 set_result(result_reg, result_val);
3978 return true;
3979 }
3980
3981 //---------------------------inline_native_getClass----------------------------
3982 // public final native Class<?> java.lang.Object.getClass();
3983 //
3984 // Build special case code for calls to getClass on an object.
3985 bool LibraryCallKit::inline_native_getClass() {
3986 Node* obj = null_check_receiver();
3987 if (stopped()) return true;
3988 set_result(load_mirror_from_klass(load_object_klass(obj)));
3989 return true;
3990 }
3991
3992 //-----------------inline_native_Reflection_getCallerClass---------------------
3993 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
3994 //
3995 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
3996 //
3997 // NOTE: This code must perform the same logic as JVM_GetCallerClass
3998 // in that it must skip particular security frames and checks for
3999 // caller sensitive methods.
4000 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4001 #ifndef PRODUCT
4002 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4003 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4004 }
4005 #endif
4006
4007 if (!jvms()->has_method()) {
4008 #ifndef PRODUCT
4009 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4010 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4011 }
4012 #endif
4013 return false;
4014 }
4015
4016 // Walk back up the JVM state to find the caller at the required
4017 // depth.
4018 JVMState* caller_jvms = jvms();
4019
4020 // Cf. JVM_GetCallerClass
4021 // NOTE: Start the loop at depth 1 because the current JVM state does
4022 // not include the Reflection.getCallerClass() frame.
4023 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4024 ciMethod* m = caller_jvms->method();
4025 switch (n) {
4026 case 0:
4027 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4028 break;
4029 case 1:
4030 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4031 if (!m->caller_sensitive()) {
4032 #ifndef PRODUCT
4033 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4034 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4035 }
4036 #endif
4037 return false; // bail-out; let JVM_GetCallerClass do the work
4038 }
4039 break;
4040 default:
4041 if (!m->is_ignored_by_security_stack_walk()) {
4042 // We have reached the desired frame; return the holder class.
4043 // Acquire method holder as java.lang.Class and push as constant.
4044 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4045 ciInstance* caller_mirror = caller_klass->java_mirror();
4046 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4047
4048 #ifndef PRODUCT
4049 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4050 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());
4051 tty->print_cr(" JVM state at this point:");
4052 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4053 ciMethod* m = jvms()->of_depth(i)->method();
4054 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4055 }
4056 }
4057 #endif
4058 return true;
4059 }
4060 break;
4061 }
4062 }
4063
4064 #ifndef PRODUCT
4065 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4066 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4067 tty->print_cr(" JVM state at this point:");
4068 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4069 ciMethod* m = jvms()->of_depth(i)->method();
4070 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4071 }
4072 }
4073 #endif
4074
4075 return false; // bail-out; let JVM_GetCallerClass do the work
4076 }
4077
4078 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4079 Node* arg = argument(0);
4080 Node* result = NULL;
4081
4082 switch (id) {
4083 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4084 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4085 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4086 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4087
4088 case vmIntrinsics::_doubleToLongBits: {
4089 // two paths (plus control) merge in a wood
4090 RegionNode *r = new RegionNode(3);
4091 Node *phi = new PhiNode(r, TypeLong::LONG);
4092
4093 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4094 // Build the boolean node
4095 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4096
4097 // Branch either way.
4098 // NaN case is less traveled, which makes all the difference.
4099 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4100 Node *opt_isnan = _gvn.transform(ifisnan);
4101 assert( opt_isnan->is_If(), "Expect an IfNode");
4102 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4103 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4104
4105 set_control(iftrue);
4106
4107 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4108 Node *slow_result = longcon(nan_bits); // return NaN
4109 phi->init_req(1, _gvn.transform( slow_result ));
4110 r->init_req(1, iftrue);
4111
4112 // Else fall through
4113 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4114 set_control(iffalse);
4115
4116 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4117 r->init_req(2, iffalse);
4118
4119 // Post merge
4120 set_control(_gvn.transform(r));
4121 record_for_igvn(r);
4122
4123 C->set_has_split_ifs(true); // Has chance for split-if optimization
4124 result = phi;
4125 assert(result->bottom_type()->isa_long(), "must be");
4126 break;
4127 }
4128
4129 case vmIntrinsics::_floatToIntBits: {
4130 // two paths (plus control) merge in a wood
4131 RegionNode *r = new RegionNode(3);
4132 Node *phi = new PhiNode(r, TypeInt::INT);
4133
4134 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4135 // Build the boolean node
4136 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4137
4138 // Branch either way.
4139 // NaN case is less traveled, which makes all the difference.
4140 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4141 Node *opt_isnan = _gvn.transform(ifisnan);
4142 assert( opt_isnan->is_If(), "Expect an IfNode");
4143 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4144 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4145
4146 set_control(iftrue);
4147
4148 static const jint nan_bits = 0x7fc00000;
4149 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4150 phi->init_req(1, _gvn.transform( slow_result ));
4151 r->init_req(1, iftrue);
4152
4153 // Else fall through
4154 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4155 set_control(iffalse);
4156
4157 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4158 r->init_req(2, iffalse);
4159
4160 // Post merge
4161 set_control(_gvn.transform(r));
4162 record_for_igvn(r);
4163
4164 C->set_has_split_ifs(true); // Has chance for split-if optimization
4165 result = phi;
4166 assert(result->bottom_type()->isa_int(), "must be");
4167 break;
4168 }
4169
4170 default:
4171 fatal_unexpected_iid(id);
4172 break;
4173 }
4174 set_result(_gvn.transform(result));
4175 return true;
4176 }
4177
4178 //----------------------inline_unsafe_copyMemory-------------------------
4179 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4180 bool LibraryCallKit::inline_unsafe_copyMemory() {
4181 if (callee()->is_static()) return false; // caller must have the capability!
4182 null_check_receiver(); // null-check receiver
4183 if (stopped()) return true;
4184
4185 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4186
4187 Node* src_ptr = argument(1); // type: oop
4188 Node* src_off = ConvL2X(argument(2)); // type: long
4189 Node* dst_ptr = argument(4); // type: oop
4190 Node* dst_off = ConvL2X(argument(5)); // type: long
4191 Node* size = ConvL2X(argument(7)); // type: long
4192
4193 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4194 "fieldOffset must be byte-scaled");
4195
4196 Node* src = make_unsafe_address(src_ptr, src_off, ACCESS_READ);
4197 Node* dst = make_unsafe_address(dst_ptr, dst_off, ACCESS_WRITE);
4198
4199 // Conservatively insert a memory barrier on all memory slices.
4200 // Do not let writes of the copy source or destination float below the copy.
4201 insert_mem_bar(Op_MemBarCPUOrder);
4202
4203 Node* thread = _gvn.transform(new ThreadLocalNode());
4204 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4205 BasicType doing_unsafe_access_bt = T_BYTE;
4206 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4207
4208 // update volatile field
4209 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4210
4211 // Call it. Note that the length argument is not scaled.
4212 make_runtime_call(RC_LEAF|RC_NO_FP,
4213 OptoRuntime::fast_arraycopy_Type(),
4214 StubRoutines::unsafe_arraycopy(),
4215 "unsafe_arraycopy",
4216 TypeRawPtr::BOTTOM,
4217 src, dst, size XTOP);
4218
4219 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4220
4221 // Do not let reads of the copy destination float above the copy.
4222 insert_mem_bar(Op_MemBarCPUOrder);
4223
4224 return true;
4225 }
4226
4227 //------------------------clone_coping-----------------------------------
4228 // Helper function for inline_native_clone.
4229 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4230 assert(obj_size != NULL, "");
4231 Node* raw_obj = alloc_obj->in(1);
4232 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4233
4234 AllocateNode* alloc = NULL;
4235 if (ReduceBulkZeroing) {
4236 // We will be completely responsible for initializing this object -
4237 // mark Initialize node as complete.
4238 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4239 // The object was just allocated - there should be no any stores!
4240 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4241 // Mark as complete_with_arraycopy so that on AllocateNode
4242 // expansion, we know this AllocateNode is initialized by an array
4243 // copy and a StoreStore barrier exists after the array copy.
4244 alloc->initialization()->set_complete_with_arraycopy();
4245 }
4246
4247 Node* size = _gvn.transform(obj_size);
4248 access_clone(obj, alloc_obj, size, is_array);
4249
4250 // Do not let reads from the cloned object float above the arraycopy.
4251 if (alloc != NULL) {
4252 // Do not let stores that initialize this object be reordered with
4253 // a subsequent store that would make this object accessible by
4254 // other threads.
4255 // Record what AllocateNode this StoreStore protects so that
4256 // escape analysis can go from the MemBarStoreStoreNode to the
4257 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4258 // based on the escape status of the AllocateNode.
4259 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4260 } else {
4261 insert_mem_bar(Op_MemBarCPUOrder);
4262 }
4263 }
4264
4265 //------------------------inline_native_clone----------------------------
4266 // protected native Object java.lang.Object.clone();
4267 //
4268 // Here are the simple edge cases:
4269 // null receiver => normal trap
4270 // virtual and clone was overridden => slow path to out-of-line clone
4271 // not cloneable or finalizer => slow path to out-of-line Object.clone
4272 //
4273 // The general case has two steps, allocation and copying.
4274 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4275 //
4276 // Copying also has two cases, oop arrays and everything else.
4277 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4278 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4279 //
4280 // These steps fold up nicely if and when the cloned object's klass
4281 // can be sharply typed as an object array, a type array, or an instance.
4282 //
4283 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4284 PhiNode* result_val;
4285
4286 // Set the reexecute bit for the interpreter to reexecute
4287 // the bytecode that invokes Object.clone if deoptimization happens.
4288 { PreserveReexecuteState preexecs(this);
4289 jvms()->set_should_reexecute(true);
4290
4291 Node* obj = null_check_receiver();
4292 if (stopped()) return true;
4293
4294 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4295
4296 // If we are going to clone an instance, we need its exact type to
4297 // know the number and types of fields to convert the clone to
4298 // loads/stores. Maybe a speculative type can help us.
4299 if (!obj_type->klass_is_exact() &&
4300 obj_type->speculative_type() != NULL &&
4301 obj_type->speculative_type()->is_instance_klass()) {
4302 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4303 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4304 !spec_ik->has_injected_fields()) {
4305 ciKlass* k = obj_type->klass();
4306 if (!k->is_instance_klass() ||
4307 k->as_instance_klass()->is_interface() ||
4308 k->as_instance_klass()->has_subklass()) {
4309 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4310 }
4311 }
4312 }
4313
4314 // Conservatively insert a memory barrier on all memory slices.
4315 // Do not let writes into the original float below the clone.
4316 insert_mem_bar(Op_MemBarCPUOrder);
4317
4318 // paths into result_reg:
4319 enum {
4320 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4321 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4322 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4323 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4324 PATH_LIMIT
4325 };
4326 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4327 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4328 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4329 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4330 record_for_igvn(result_reg);
4331
4332 Node* obj_klass = load_object_klass(obj);
4333 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4334 if (array_ctl != NULL) {
4335 // It's an array.
4336 PreserveJVMState pjvms(this);
4337 set_control(array_ctl);
4338 Node* obj_length = load_array_length(obj);
4339 Node* obj_size = NULL;
4340 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, /*deoptimize_on_exception=*/true);
4341
4342 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4343 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, BarrierSetC2::Parsing)) {
4344 // If it is an oop array, it requires very special treatment,
4345 // because gc barriers are required when accessing the array.
4346 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4347 if (is_obja != NULL) {
4348 PreserveJVMState pjvms2(this);
4349 set_control(is_obja);
4350 // Generate a direct call to the right arraycopy function(s).
4351 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4352 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false);
4353 ac->set_clone_oop_array();
4354 Node* n = _gvn.transform(ac);
4355 assert(n == ac, "cannot disappear");
4356 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
4357
4358 result_reg->init_req(_objArray_path, control());
4359 result_val->init_req(_objArray_path, alloc_obj);
4360 result_i_o ->set_req(_objArray_path, i_o());
4361 result_mem ->set_req(_objArray_path, reset_memory());
4362 }
4363 }
4364 // Otherwise, there are no barriers to worry about.
4365 // (We can dispense with card marks if we know the allocation
4366 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4367 // causes the non-eden paths to take compensating steps to
4368 // simulate a fresh allocation, so that no further
4369 // card marks are required in compiled code to initialize
4370 // the object.)
4371
4372 if (!stopped()) {
4373 copy_to_clone(obj, alloc_obj, obj_size, true);
4374
4375 // Present the results of the copy.
4376 result_reg->init_req(_array_path, control());
4377 result_val->init_req(_array_path, alloc_obj);
4378 result_i_o ->set_req(_array_path, i_o());
4379 result_mem ->set_req(_array_path, reset_memory());
4380 }
4381 }
4382
4383 // We only go to the instance fast case code if we pass a number of guards.
4384 // The paths which do not pass are accumulated in the slow_region.
4385 RegionNode* slow_region = new RegionNode(1);
4386 record_for_igvn(slow_region);
4387 if (!stopped()) {
4388 // It's an instance (we did array above). Make the slow-path tests.
4389 // If this is a virtual call, we generate a funny guard. We grab
4390 // the vtable entry corresponding to clone() from the target object.
4391 // If the target method which we are calling happens to be the
4392 // Object clone() method, we pass the guard. We do not need this
4393 // guard for non-virtual calls; the caller is known to be the native
4394 // Object clone().
4395 if (is_virtual) {
4396 generate_virtual_guard(obj_klass, slow_region);
4397 }
4398
4399 // The object must be easily cloneable and must not have a finalizer.
4400 // Both of these conditions may be checked in a single test.
4401 // We could optimize the test further, but we don't care.
4402 generate_access_flags_guard(obj_klass,
4403 // Test both conditions:
4404 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4405 // Must be cloneable but not finalizer:
4406 JVM_ACC_IS_CLONEABLE_FAST,
4407 slow_region);
4408 }
4409
4410 if (!stopped()) {
4411 // It's an instance, and it passed the slow-path tests.
4412 PreserveJVMState pjvms(this);
4413 Node* obj_size = NULL;
4414 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4415 // is reexecuted if deoptimization occurs and there could be problems when merging
4416 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4417 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4418
4419 copy_to_clone(obj, alloc_obj, obj_size, false);
4420
4421 // Present the results of the slow call.
4422 result_reg->init_req(_instance_path, control());
4423 result_val->init_req(_instance_path, alloc_obj);
4424 result_i_o ->set_req(_instance_path, i_o());
4425 result_mem ->set_req(_instance_path, reset_memory());
4426 }
4427
4428 // Generate code for the slow case. We make a call to clone().
4429 set_control(_gvn.transform(slow_region));
4430 if (!stopped()) {
4431 PreserveJVMState pjvms(this);
4432 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4433 // We need to deoptimize on exception (see comment above)
4434 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4435 // this->control() comes from set_results_for_java_call
4436 result_reg->init_req(_slow_path, control());
4437 result_val->init_req(_slow_path, slow_result);
4438 result_i_o ->set_req(_slow_path, i_o());
4439 result_mem ->set_req(_slow_path, reset_memory());
4440 }
4441
4442 // Return the combined state.
4443 set_control( _gvn.transform(result_reg));
4444 set_i_o( _gvn.transform(result_i_o));
4445 set_all_memory( _gvn.transform(result_mem));
4446 } // original reexecute is set back here
4447
4448 set_result(_gvn.transform(result_val));
4449 return true;
4450 }
4451
4452 // If we have a tightly coupled allocation, the arraycopy may take care
4453 // of the array initialization. If one of the guards we insert between
4454 // the allocation and the arraycopy causes a deoptimization, an
4455 // unitialized array will escape the compiled method. To prevent that
4456 // we set the JVM state for uncommon traps between the allocation and
4457 // the arraycopy to the state before the allocation so, in case of
4458 // deoptimization, we'll reexecute the allocation and the
4459 // initialization.
4460 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4461 if (alloc != NULL) {
4462 ciMethod* trap_method = alloc->jvms()->method();
4463 int trap_bci = alloc->jvms()->bci();
4464
4465 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4466 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4467 // Make sure there's no store between the allocation and the
4468 // arraycopy otherwise visible side effects could be rexecuted
4469 // in case of deoptimization and cause incorrect execution.
4470 bool no_interfering_store = true;
4471 Node* mem = alloc->in(TypeFunc::Memory);
4472 if (mem->is_MergeMem()) {
4473 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4474 Node* n = mms.memory();
4475 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4476 assert(n->is_Store(), "what else?");
4477 no_interfering_store = false;
4478 break;
4479 }
4480 }
4481 } else {
4482 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4483 Node* n = mms.memory();
4484 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4485 assert(n->is_Store(), "what else?");
4486 no_interfering_store = false;
4487 break;
4488 }
4489 }
4490 }
4491
4492 if (no_interfering_store) {
4493 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4494 uint size = alloc->req();
4495 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4496 old_jvms->set_map(sfpt);
4497 for (uint i = 0; i < size; i++) {
4498 sfpt->init_req(i, alloc->in(i));
4499 }
4500 // re-push array length for deoptimization
4501 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4502 old_jvms->set_sp(old_jvms->sp()+1);
4503 old_jvms->set_monoff(old_jvms->monoff()+1);
4504 old_jvms->set_scloff(old_jvms->scloff()+1);
4505 old_jvms->set_endoff(old_jvms->endoff()+1);
4506 old_jvms->set_should_reexecute(true);
4507
4508 sfpt->set_i_o(map()->i_o());
4509 sfpt->set_memory(map()->memory());
4510 sfpt->set_control(map()->control());
4511
4512 JVMState* saved_jvms = jvms();
4513 saved_reexecute_sp = _reexecute_sp;
4514
4515 set_jvms(sfpt->jvms());
4516 _reexecute_sp = jvms()->sp();
4517
4518 return saved_jvms;
4519 }
4520 }
4521 }
4522 return NULL;
4523 }
4524
4525 // In case of a deoptimization, we restart execution at the
4526 // allocation, allocating a new array. We would leave an uninitialized
4527 // array in the heap that GCs wouldn't expect. Move the allocation
4528 // after the traps so we don't allocate the array if we
4529 // deoptimize. This is possible because tightly_coupled_allocation()
4530 // guarantees there's no observer of the allocated array at this point
4531 // and the control flow is simple enough.
4532 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4533 int saved_reexecute_sp, uint new_idx) {
4534 if (saved_jvms != NULL && !stopped()) {
4535 assert(alloc != NULL, "only with a tightly coupled allocation");
4536 // restore JVM state to the state at the arraycopy
4537 saved_jvms->map()->set_control(map()->control());
4538 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4539 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4540 // If we've improved the types of some nodes (null check) while
4541 // emitting the guards, propagate them to the current state
4542 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4543 set_jvms(saved_jvms);
4544 _reexecute_sp = saved_reexecute_sp;
4545
4546 // Remove the allocation from above the guards
4547 CallProjections callprojs;
4548 alloc->extract_projections(&callprojs, true);
4549 InitializeNode* init = alloc->initialization();
4550 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4551 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4552 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4553 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4554
4555 // move the allocation here (after the guards)
4556 _gvn.hash_delete(alloc);
4557 alloc->set_req(TypeFunc::Control, control());
4558 alloc->set_req(TypeFunc::I_O, i_o());
4559 Node *mem = reset_memory();
4560 set_all_memory(mem);
4561 alloc->set_req(TypeFunc::Memory, mem);
4562 set_control(init->proj_out_or_null(TypeFunc::Control));
4563 set_i_o(callprojs.fallthrough_ioproj);
4564
4565 // Update memory as done in GraphKit::set_output_for_allocation()
4566 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4567 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4568 if (ary_type->isa_aryptr() && length_type != NULL) {
4569 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4570 }
4571 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4572 int elemidx = C->get_alias_index(telemref);
4573 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
4574 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
4575
4576 Node* allocx = _gvn.transform(alloc);
4577 assert(allocx == alloc, "where has the allocation gone?");
4578 assert(dest->is_CheckCastPP(), "not an allocation result?");
4579
4580 _gvn.hash_delete(dest);
4581 dest->set_req(0, control());
4582 Node* destx = _gvn.transform(dest);
4583 assert(destx == dest, "where has the allocation result gone?");
4584 }
4585 }
4586
4587
4588 //------------------------------inline_arraycopy-----------------------
4589 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4590 // Object dest, int destPos,
4591 // int length);
4592 bool LibraryCallKit::inline_arraycopy() {
4593 // Get the arguments.
4594 Node* src = argument(0); // type: oop
4595 Node* src_offset = argument(1); // type: int
4596 Node* dest = argument(2); // type: oop
4597 Node* dest_offset = argument(3); // type: int
4598 Node* length = argument(4); // type: int
4599
4600 uint new_idx = C->unique();
4601
4602 // Check for allocation before we add nodes that would confuse
4603 // tightly_coupled_allocation()
4604 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4605
4606 int saved_reexecute_sp = -1;
4607 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4608 // See arraycopy_restore_alloc_state() comment
4609 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4610 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4611 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
4612 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4613
4614 // The following tests must be performed
4615 // (1) src and dest are arrays.
4616 // (2) src and dest arrays must have elements of the same BasicType
4617 // (3) src and dest must not be null.
4618 // (4) src_offset must not be negative.
4619 // (5) dest_offset must not be negative.
4620 // (6) length must not be negative.
4621 // (7) src_offset + length must not exceed length of src.
4622 // (8) dest_offset + length must not exceed length of dest.
4623 // (9) each element of an oop array must be assignable
4624
4625 // (3) src and dest must not be null.
4626 // always do this here because we need the JVM state for uncommon traps
4627 Node* null_ctl = top();
4628 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4629 assert(null_ctl->is_top(), "no null control here");
4630 dest = null_check(dest, T_ARRAY);
4631
4632 if (!can_emit_guards) {
4633 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4634 // guards but the arraycopy node could still take advantage of a
4635 // tightly allocated allocation. tightly_coupled_allocation() is
4636 // called again to make sure it takes the null check above into
4637 // account: the null check is mandatory and if it caused an
4638 // uncommon trap to be emitted then the allocation can't be
4639 // considered tightly coupled in this context.
4640 alloc = tightly_coupled_allocation(dest, NULL);
4641 }
4642
4643 bool validated = false;
4644
4645 const Type* src_type = _gvn.type(src);
4646 const Type* dest_type = _gvn.type(dest);
4647 const TypeAryPtr* top_src = src_type->isa_aryptr();
4648 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4649
4650 // Do we have the type of src?
4651 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4652 // Do we have the type of dest?
4653 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4654 // Is the type for src from speculation?
4655 bool src_spec = false;
4656 // Is the type for dest from speculation?
4657 bool dest_spec = false;
4658
4659 if ((!has_src || !has_dest) && can_emit_guards) {
4660 // We don't have sufficient type information, let's see if
4661 // speculative types can help. We need to have types for both src
4662 // and dest so that it pays off.
4663
4664 // Do we already have or could we have type information for src
4665 bool could_have_src = has_src;
4666 // Do we already have or could we have type information for dest
4667 bool could_have_dest = has_dest;
4668
4669 ciKlass* src_k = NULL;
4670 if (!has_src) {
4671 src_k = src_type->speculative_type_not_null();
4672 if (src_k != NULL && src_k->is_array_klass()) {
4673 could_have_src = true;
4674 }
4675 }
4676
4677 ciKlass* dest_k = NULL;
4678 if (!has_dest) {
4679 dest_k = dest_type->speculative_type_not_null();
4680 if (dest_k != NULL && dest_k->is_array_klass()) {
4681 could_have_dest = true;
4682 }
4683 }
4684
4685 if (could_have_src && could_have_dest) {
4686 // This is going to pay off so emit the required guards
4687 if (!has_src) {
4688 src = maybe_cast_profiled_obj(src, src_k, true);
4689 src_type = _gvn.type(src);
4690 top_src = src_type->isa_aryptr();
4691 has_src = (top_src != NULL && top_src->klass() != NULL);
4692 src_spec = true;
4693 }
4694 if (!has_dest) {
4695 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4696 dest_type = _gvn.type(dest);
4697 top_dest = dest_type->isa_aryptr();
4698 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4699 dest_spec = true;
4700 }
4701 }
4702 }
4703
4704 if (has_src && has_dest && can_emit_guards) {
4705 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
4706 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
4707 if (is_reference_type(src_elem)) src_elem = T_OBJECT;
4708 if (is_reference_type(dest_elem)) dest_elem = T_OBJECT;
4709
4710 if (src_elem == dest_elem && src_elem == T_OBJECT) {
4711 // If both arrays are object arrays then having the exact types
4712 // for both will remove the need for a subtype check at runtime
4713 // before the call and may make it possible to pick a faster copy
4714 // routine (without a subtype check on every element)
4715 // Do we have the exact type of src?
4716 bool could_have_src = src_spec;
4717 // Do we have the exact type of dest?
4718 bool could_have_dest = dest_spec;
4719 ciKlass* src_k = top_src->klass();
4720 ciKlass* dest_k = top_dest->klass();
4721 if (!src_spec) {
4722 src_k = src_type->speculative_type_not_null();
4723 if (src_k != NULL && src_k->is_array_klass()) {
4724 could_have_src = true;
4725 }
4726 }
4727 if (!dest_spec) {
4728 dest_k = dest_type->speculative_type_not_null();
4729 if (dest_k != NULL && dest_k->is_array_klass()) {
4730 could_have_dest = true;
4731 }
4732 }
4733 if (could_have_src && could_have_dest) {
4734 // If we can have both exact types, emit the missing guards
4735 if (could_have_src && !src_spec) {
4736 src = maybe_cast_profiled_obj(src, src_k, true);
4737 }
4738 if (could_have_dest && !dest_spec) {
4739 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4740 }
4741 }
4742 }
4743 }
4744
4745 ciMethod* trap_method = method();
4746 int trap_bci = bci();
4747 if (saved_jvms != NULL) {
4748 trap_method = alloc->jvms()->method();
4749 trap_bci = alloc->jvms()->bci();
4750 }
4751
4752 bool negative_length_guard_generated = false;
4753
4754 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4755 can_emit_guards &&
4756 !src->is_top() && !dest->is_top()) {
4757 // validate arguments: enables transformation the ArrayCopyNode
4758 validated = true;
4759
4760 RegionNode* slow_region = new RegionNode(1);
4761 record_for_igvn(slow_region);
4762
4763 // (1) src and dest are arrays.
4764 generate_non_array_guard(load_object_klass(src), slow_region);
4765 generate_non_array_guard(load_object_klass(dest), slow_region);
4766
4767 // (2) src and dest arrays must have elements of the same BasicType
4768 // done at macro expansion or at Ideal transformation time
4769
4770 // (4) src_offset must not be negative.
4771 generate_negative_guard(src_offset, slow_region);
4772
4773 // (5) dest_offset must not be negative.
4774 generate_negative_guard(dest_offset, slow_region);
4775
4776 // (7) src_offset + length must not exceed length of src.
4777 generate_limit_guard(src_offset, length,
4778 load_array_length(src),
4779 slow_region);
4780
4781 // (8) dest_offset + length must not exceed length of dest.
4782 generate_limit_guard(dest_offset, length,
4783 load_array_length(dest),
4784 slow_region);
4785
4786 // (6) length must not be negative.
4787 // This is also checked in generate_arraycopy() during macro expansion, but
4788 // we also have to check it here for the case where the ArrayCopyNode will
4789 // be eliminated by Escape Analysis.
4790 if (EliminateAllocations) {
4791 generate_negative_guard(length, slow_region);
4792 negative_length_guard_generated = true;
4793 }
4794
4795 // (9) each element of an oop array must be assignable
4796 Node* dest_klass = load_object_klass(dest);
4797 if (src != dest) {
4798 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
4799
4800 if (not_subtype_ctrl != top()) {
4801 PreserveJVMState pjvms(this);
4802 set_control(not_subtype_ctrl);
4803 uncommon_trap(Deoptimization::Reason_intrinsic,
4804 Deoptimization::Action_make_not_entrant);
4805 assert(stopped(), "Should be stopped");
4806 }
4807 }
4808 {
4809 PreserveJVMState pjvms(this);
4810 set_control(_gvn.transform(slow_region));
4811 uncommon_trap(Deoptimization::Reason_intrinsic,
4812 Deoptimization::Action_make_not_entrant);
4813 assert(stopped(), "Should be stopped");
4814 }
4815
4816 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
4817 const Type *toop = TypeOopPtr::make_from_klass(dest_klass_t->klass());
4818 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
4819 }
4820
4821 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
4822
4823 if (stopped()) {
4824 return true;
4825 }
4826
4827 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
4828 // Create LoadRange and LoadKlass nodes for use during macro expansion here
4829 // so the compiler has a chance to eliminate them: during macro expansion,
4830 // we have to set their control (CastPP nodes are eliminated).
4831 load_object_klass(src), load_object_klass(dest),
4832 load_array_length(src), load_array_length(dest));
4833
4834 ac->set_arraycopy(validated);
4835
4836 Node* n = _gvn.transform(ac);
4837 if (n == ac) {
4838 ac->connect_outputs(this);
4839 } else {
4840 assert(validated, "shouldn't transform if all arguments not validated");
4841 set_all_memory(n);
4842 }
4843 clear_upper_avx();
4844
4845
4846 return true;
4847 }
4848
4849
4850 // Helper function which determines if an arraycopy immediately follows
4851 // an allocation, with no intervening tests or other escapes for the object.
4852 AllocateArrayNode*
4853 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
4854 RegionNode* slow_region) {
4855 if (stopped()) return NULL; // no fast path
4856 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
4857
4858 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
4859 if (alloc == NULL) return NULL;
4860
4861 Node* rawmem = memory(Compile::AliasIdxRaw);
4862 // Is the allocation's memory state untouched?
4863 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
4864 // Bail out if there have been raw-memory effects since the allocation.
4865 // (Example: There might have been a call or safepoint.)
4866 return NULL;
4867 }
4868 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
4869 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
4870 return NULL;
4871 }
4872
4873 // There must be no unexpected observers of this allocation.
4874 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
4875 Node* obs = ptr->fast_out(i);
4876 if (obs != this->map()) {
4877 return NULL;
4878 }
4879 }
4880
4881 // This arraycopy must unconditionally follow the allocation of the ptr.
4882 Node* alloc_ctl = ptr->in(0);
4883 Node* ctl = control();
4884 while (ctl != alloc_ctl) {
4885 // There may be guards which feed into the slow_region.
4886 // Any other control flow means that we might not get a chance
4887 // to finish initializing the allocated object.
4888 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
4889 IfNode* iff = ctl->in(0)->as_If();
4890 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
4891 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
4892 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
4893 ctl = iff->in(0); // This test feeds the known slow_region.
4894 continue;
4895 }
4896 // One more try: Various low-level checks bottom out in
4897 // uncommon traps. If the debug-info of the trap omits
4898 // any reference to the allocation, as we've already
4899 // observed, then there can be no objection to the trap.
4900 bool found_trap = false;
4901 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
4902 Node* obs = not_ctl->fast_out(j);
4903 if (obs->in(0) == not_ctl && obs->is_Call() &&
4904 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
4905 found_trap = true; break;
4906 }
4907 }
4908 if (found_trap) {
4909 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
4910 continue;
4911 }
4912 }
4913 return NULL;
4914 }
4915
4916 // If we get this far, we have an allocation which immediately
4917 // precedes the arraycopy, and we can take over zeroing the new object.
4918 // The arraycopy will finish the initialization, and provide
4919 // a new control state to which we will anchor the destination pointer.
4920
4921 return alloc;
4922 }
4923
4924 //-------------inline_encodeISOArray-----------------------------------
4925 // encode char[] to byte[] in ISO_8859_1
4926 bool LibraryCallKit::inline_encodeISOArray() {
4927 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
4928 // no receiver since it is static method
4929 Node *src = argument(0);
4930 Node *src_offset = argument(1);
4931 Node *dst = argument(2);
4932 Node *dst_offset = argument(3);
4933 Node *length = argument(4);
4934
4935 src = must_be_not_null(src, true);
4936 dst = must_be_not_null(dst, true);
4937
4938 const Type* src_type = src->Value(&_gvn);
4939 const Type* dst_type = dst->Value(&_gvn);
4940 const TypeAryPtr* top_src = src_type->isa_aryptr();
4941 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
4942 if (top_src == NULL || top_src->klass() == NULL ||
4943 top_dest == NULL || top_dest->klass() == NULL) {
4944 // failed array check
4945 return false;
4946 }
4947
4948 // Figure out the size and type of the elements we will be copying.
4949 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4950 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4951 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
4952 return false;
4953 }
4954
4955 Node* src_start = array_element_address(src, src_offset, T_CHAR);
4956 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
4957 // 'src_start' points to src array + scaled offset
4958 // 'dst_start' points to dst array + scaled offset
4959
4960 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
4961 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
4962 enc = _gvn.transform(enc);
4963 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
4964 set_memory(res_mem, mtype);
4965 set_result(enc);
4966 clear_upper_avx();
4967
4968 return true;
4969 }
4970
4971 //-------------inline_multiplyToLen-----------------------------------
4972 bool LibraryCallKit::inline_multiplyToLen() {
4973 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
4974
4975 address stubAddr = StubRoutines::multiplyToLen();
4976 if (stubAddr == NULL) {
4977 return false; // Intrinsic's stub is not implemented on this platform
4978 }
4979 const char* stubName = "multiplyToLen";
4980
4981 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
4982
4983 // no receiver because it is a static method
4984 Node* x = argument(0);
4985 Node* xlen = argument(1);
4986 Node* y = argument(2);
4987 Node* ylen = argument(3);
4988 Node* z = argument(4);
4989
4990 x = must_be_not_null(x, true);
4991 y = must_be_not_null(y, true);
4992
4993 const Type* x_type = x->Value(&_gvn);
4994 const Type* y_type = y->Value(&_gvn);
4995 const TypeAryPtr* top_x = x_type->isa_aryptr();
4996 const TypeAryPtr* top_y = y_type->isa_aryptr();
4997 if (top_x == NULL || top_x->klass() == NULL ||
4998 top_y == NULL || top_y->klass() == NULL) {
4999 // failed array check
5000 return false;
5001 }
5002
5003 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5004 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5005 if (x_elem != T_INT || y_elem != T_INT) {
5006 return false;
5007 }
5008
5009 // Set the original stack and the reexecute bit for the interpreter to reexecute
5010 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5011 // on the return from z array allocation in runtime.
5012 { PreserveReexecuteState preexecs(this);
5013 jvms()->set_should_reexecute(true);
5014
5015 Node* x_start = array_element_address(x, intcon(0), x_elem);
5016 Node* y_start = array_element_address(y, intcon(0), y_elem);
5017 // 'x_start' points to x array + scaled xlen
5018 // 'y_start' points to y array + scaled ylen
5019
5020 // Allocate the result array
5021 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5022 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5023 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5024
5025 IdealKit ideal(this);
5026
5027 #define __ ideal.
5028 Node* one = __ ConI(1);
5029 Node* zero = __ ConI(0);
5030 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5031 __ set(need_alloc, zero);
5032 __ set(z_alloc, z);
5033 __ if_then(z, BoolTest::eq, null()); {
5034 __ increment (need_alloc, one);
5035 } __ else_(); {
5036 // Update graphKit memory and control from IdealKit.
5037 sync_kit(ideal);
5038 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5039 cast->init_req(0, control());
5040 _gvn.set_type(cast, cast->bottom_type());
5041 C->record_for_igvn(cast);
5042
5043 Node* zlen_arg = load_array_length(cast);
5044 // Update IdealKit memory and control from graphKit.
5045 __ sync_kit(this);
5046 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5047 __ increment (need_alloc, one);
5048 } __ end_if();
5049 } __ end_if();
5050
5051 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5052 // Update graphKit memory and control from IdealKit.
5053 sync_kit(ideal);
5054 Node * narr = new_array(klass_node, zlen, 1);
5055 // Update IdealKit memory and control from graphKit.
5056 __ sync_kit(this);
5057 __ set(z_alloc, narr);
5058 } __ end_if();
5059
5060 sync_kit(ideal);
5061 z = __ value(z_alloc);
5062 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5063 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5064 // Final sync IdealKit and GraphKit.
5065 final_sync(ideal);
5066 #undef __
5067
5068 Node* z_start = array_element_address(z, intcon(0), T_INT);
5069
5070 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5071 OptoRuntime::multiplyToLen_Type(),
5072 stubAddr, stubName, TypePtr::BOTTOM,
5073 x_start, xlen, y_start, ylen, z_start, zlen);
5074 } // original reexecute is set back here
5075
5076 C->set_has_split_ifs(true); // Has chance for split-if optimization
5077 set_result(z);
5078 return true;
5079 }
5080
5081 //-------------inline_squareToLen------------------------------------
5082 bool LibraryCallKit::inline_squareToLen() {
5083 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5084
5085 address stubAddr = StubRoutines::squareToLen();
5086 if (stubAddr == NULL) {
5087 return false; // Intrinsic's stub is not implemented on this platform
5088 }
5089 const char* stubName = "squareToLen";
5090
5091 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5092
5093 Node* x = argument(0);
5094 Node* len = argument(1);
5095 Node* z = argument(2);
5096 Node* zlen = argument(3);
5097
5098 x = must_be_not_null(x, true);
5099 z = must_be_not_null(z, true);
5100
5101 const Type* x_type = x->Value(&_gvn);
5102 const Type* z_type = z->Value(&_gvn);
5103 const TypeAryPtr* top_x = x_type->isa_aryptr();
5104 const TypeAryPtr* top_z = z_type->isa_aryptr();
5105 if (top_x == NULL || top_x->klass() == NULL ||
5106 top_z == NULL || top_z->klass() == NULL) {
5107 // failed array check
5108 return false;
5109 }
5110
5111 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5112 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5113 if (x_elem != T_INT || z_elem != T_INT) {
5114 return false;
5115 }
5116
5117
5118 Node* x_start = array_element_address(x, intcon(0), x_elem);
5119 Node* z_start = array_element_address(z, intcon(0), z_elem);
5120
5121 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5122 OptoRuntime::squareToLen_Type(),
5123 stubAddr, stubName, TypePtr::BOTTOM,
5124 x_start, len, z_start, zlen);
5125
5126 set_result(z);
5127 return true;
5128 }
5129
5130 //-------------inline_mulAdd------------------------------------------
5131 bool LibraryCallKit::inline_mulAdd() {
5132 assert(UseMulAddIntrinsic, "not implemented on this platform");
5133
5134 address stubAddr = StubRoutines::mulAdd();
5135 if (stubAddr == NULL) {
5136 return false; // Intrinsic's stub is not implemented on this platform
5137 }
5138 const char* stubName = "mulAdd";
5139
5140 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5141
5142 Node* out = argument(0);
5143 Node* in = argument(1);
5144 Node* offset = argument(2);
5145 Node* len = argument(3);
5146 Node* k = argument(4);
5147
5148 out = must_be_not_null(out, true);
5149
5150 const Type* out_type = out->Value(&_gvn);
5151 const Type* in_type = in->Value(&_gvn);
5152 const TypeAryPtr* top_out = out_type->isa_aryptr();
5153 const TypeAryPtr* top_in = in_type->isa_aryptr();
5154 if (top_out == NULL || top_out->klass() == NULL ||
5155 top_in == NULL || top_in->klass() == NULL) {
5156 // failed array check
5157 return false;
5158 }
5159
5160 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5161 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5162 if (out_elem != T_INT || in_elem != T_INT) {
5163 return false;
5164 }
5165
5166 Node* outlen = load_array_length(out);
5167 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5168 Node* out_start = array_element_address(out, intcon(0), out_elem);
5169 Node* in_start = array_element_address(in, intcon(0), in_elem);
5170
5171 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5172 OptoRuntime::mulAdd_Type(),
5173 stubAddr, stubName, TypePtr::BOTTOM,
5174 out_start,in_start, new_offset, len, k);
5175 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5176 set_result(result);
5177 return true;
5178 }
5179
5180 //-------------inline_montgomeryMultiply-----------------------------------
5181 bool LibraryCallKit::inline_montgomeryMultiply() {
5182 address stubAddr = StubRoutines::montgomeryMultiply();
5183 if (stubAddr == NULL) {
5184 return false; // Intrinsic's stub is not implemented on this platform
5185 }
5186
5187 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5188 const char* stubName = "montgomery_multiply";
5189
5190 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5191
5192 Node* a = argument(0);
5193 Node* b = argument(1);
5194 Node* n = argument(2);
5195 Node* len = argument(3);
5196 Node* inv = argument(4);
5197 Node* m = argument(6);
5198
5199 const Type* a_type = a->Value(&_gvn);
5200 const TypeAryPtr* top_a = a_type->isa_aryptr();
5201 const Type* b_type = b->Value(&_gvn);
5202 const TypeAryPtr* top_b = b_type->isa_aryptr();
5203 const Type* n_type = a->Value(&_gvn);
5204 const TypeAryPtr* top_n = n_type->isa_aryptr();
5205 const Type* m_type = a->Value(&_gvn);
5206 const TypeAryPtr* top_m = m_type->isa_aryptr();
5207 if (top_a == NULL || top_a->klass() == NULL ||
5208 top_b == NULL || top_b->klass() == NULL ||
5209 top_n == NULL || top_n->klass() == NULL ||
5210 top_m == NULL || top_m->klass() == NULL) {
5211 // failed array check
5212 return false;
5213 }
5214
5215 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5216 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5217 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5218 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5219 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5220 return false;
5221 }
5222
5223 // Make the call
5224 {
5225 Node* a_start = array_element_address(a, intcon(0), a_elem);
5226 Node* b_start = array_element_address(b, intcon(0), b_elem);
5227 Node* n_start = array_element_address(n, intcon(0), n_elem);
5228 Node* m_start = array_element_address(m, intcon(0), m_elem);
5229
5230 Node* call = make_runtime_call(RC_LEAF,
5231 OptoRuntime::montgomeryMultiply_Type(),
5232 stubAddr, stubName, TypePtr::BOTTOM,
5233 a_start, b_start, n_start, len, inv, top(),
5234 m_start);
5235 set_result(m);
5236 }
5237
5238 return true;
5239 }
5240
5241 bool LibraryCallKit::inline_montgomerySquare() {
5242 address stubAddr = StubRoutines::montgomerySquare();
5243 if (stubAddr == NULL) {
5244 return false; // Intrinsic's stub is not implemented on this platform
5245 }
5246
5247 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5248 const char* stubName = "montgomery_square";
5249
5250 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5251
5252 Node* a = argument(0);
5253 Node* n = argument(1);
5254 Node* len = argument(2);
5255 Node* inv = argument(3);
5256 Node* m = argument(5);
5257
5258 const Type* a_type = a->Value(&_gvn);
5259 const TypeAryPtr* top_a = a_type->isa_aryptr();
5260 const Type* n_type = a->Value(&_gvn);
5261 const TypeAryPtr* top_n = n_type->isa_aryptr();
5262 const Type* m_type = a->Value(&_gvn);
5263 const TypeAryPtr* top_m = m_type->isa_aryptr();
5264 if (top_a == NULL || top_a->klass() == NULL ||
5265 top_n == NULL || top_n->klass() == NULL ||
5266 top_m == NULL || top_m->klass() == NULL) {
5267 // failed array check
5268 return false;
5269 }
5270
5271 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5272 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5273 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5274 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5275 return false;
5276 }
5277
5278 // Make the call
5279 {
5280 Node* a_start = array_element_address(a, intcon(0), a_elem);
5281 Node* n_start = array_element_address(n, intcon(0), n_elem);
5282 Node* m_start = array_element_address(m, intcon(0), m_elem);
5283
5284 Node* call = make_runtime_call(RC_LEAF,
5285 OptoRuntime::montgomerySquare_Type(),
5286 stubAddr, stubName, TypePtr::BOTTOM,
5287 a_start, n_start, len, inv, top(),
5288 m_start);
5289 set_result(m);
5290 }
5291
5292 return true;
5293 }
5294
5295 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
5296 address stubAddr = NULL;
5297 const char* stubName = NULL;
5298
5299 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
5300 if (stubAddr == NULL) {
5301 return false; // Intrinsic's stub is not implemented on this platform
5302 }
5303
5304 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
5305
5306 assert(callee()->signature()->size() == 5, "expected 5 arguments");
5307
5308 Node* newArr = argument(0);
5309 Node* oldArr = argument(1);
5310 Node* newIdx = argument(2);
5311 Node* shiftCount = argument(3);
5312 Node* numIter = argument(4);
5313
5314 const Type* newArr_type = newArr->Value(&_gvn);
5315 const TypeAryPtr* top_newArr = newArr_type->isa_aryptr();
5316 const Type* oldArr_type = oldArr->Value(&_gvn);
5317 const TypeAryPtr* top_oldArr = oldArr_type->isa_aryptr();
5318 if (top_newArr == NULL || top_newArr->klass() == NULL || top_oldArr == NULL
5319 || top_oldArr->klass() == NULL) {
5320 return false;
5321 }
5322
5323 BasicType newArr_elem = newArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5324 BasicType oldArr_elem = oldArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5325 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
5326 return false;
5327 }
5328
5329 // Make the call
5330 {
5331 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
5332 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
5333
5334 Node* call = make_runtime_call(RC_LEAF,
5335 OptoRuntime::bigIntegerShift_Type(),
5336 stubAddr,
5337 stubName,
5338 TypePtr::BOTTOM,
5339 newArr_start,
5340 oldArr_start,
5341 newIdx,
5342 shiftCount,
5343 numIter);
5344 }
5345
5346 return true;
5347 }
5348
5349 //-------------inline_vectorizedMismatch------------------------------
5350 bool LibraryCallKit::inline_vectorizedMismatch() {
5351 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5352
5353 address stubAddr = StubRoutines::vectorizedMismatch();
5354 if (stubAddr == NULL) {
5355 return false; // Intrinsic's stub is not implemented on this platform
5356 }
5357 const char* stubName = "vectorizedMismatch";
5358 int size_l = callee()->signature()->size();
5359 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5360
5361 Node* obja = argument(0);
5362 Node* aoffset = argument(1);
5363 Node* objb = argument(3);
5364 Node* boffset = argument(4);
5365 Node* length = argument(6);
5366 Node* scale = argument(7);
5367
5368 const Type* a_type = obja->Value(&_gvn);
5369 const Type* b_type = objb->Value(&_gvn);
5370 const TypeAryPtr* top_a = a_type->isa_aryptr();
5371 const TypeAryPtr* top_b = b_type->isa_aryptr();
5372 if (top_a == NULL || top_a->klass() == NULL ||
5373 top_b == NULL || top_b->klass() == NULL) {
5374 // failed array check
5375 return false;
5376 }
5377
5378 Node* call;
5379 jvms()->set_should_reexecute(true);
5380
5381 Node* obja_adr = make_unsafe_address(obja, aoffset, ACCESS_READ);
5382 Node* objb_adr = make_unsafe_address(objb, boffset, ACCESS_READ);
5383
5384 call = make_runtime_call(RC_LEAF,
5385 OptoRuntime::vectorizedMismatch_Type(),
5386 stubAddr, stubName, TypePtr::BOTTOM,
5387 obja_adr, objb_adr, length, scale);
5388
5389 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5390 set_result(result);
5391 return true;
5392 }
5393
5394 /**
5395 * Calculate CRC32 for byte.
5396 * int java.util.zip.CRC32.update(int crc, int b)
5397 */
5398 bool LibraryCallKit::inline_updateCRC32() {
5399 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5400 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5401 // no receiver since it is static method
5402 Node* crc = argument(0); // type: int
5403 Node* b = argument(1); // type: int
5404
5405 /*
5406 * int c = ~ crc;
5407 * b = timesXtoThe32[(b ^ c) & 0xFF];
5408 * b = b ^ (c >>> 8);
5409 * crc = ~b;
5410 */
5411
5412 Node* M1 = intcon(-1);
5413 crc = _gvn.transform(new XorINode(crc, M1));
5414 Node* result = _gvn.transform(new XorINode(crc, b));
5415 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5416
5417 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5418 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5419 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5420 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5421
5422 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5423 result = _gvn.transform(new XorINode(crc, result));
5424 result = _gvn.transform(new XorINode(result, M1));
5425 set_result(result);
5426 return true;
5427 }
5428
5429 /**
5430 * Calculate CRC32 for byte[] array.
5431 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5432 */
5433 bool LibraryCallKit::inline_updateBytesCRC32() {
5434 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5435 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5436 // no receiver since it is static method
5437 Node* crc = argument(0); // type: int
5438 Node* src = argument(1); // type: oop
5439 Node* offset = argument(2); // type: int
5440 Node* length = argument(3); // type: int
5441
5442 const Type* src_type = src->Value(&_gvn);
5443 const TypeAryPtr* top_src = src_type->isa_aryptr();
5444 if (top_src == NULL || top_src->klass() == NULL) {
5445 // failed array check
5446 return false;
5447 }
5448
5449 // Figure out the size and type of the elements we will be copying.
5450 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5451 if (src_elem != T_BYTE) {
5452 return false;
5453 }
5454
5455 // 'src_start' points to src array + scaled offset
5456 src = must_be_not_null(src, true);
5457 Node* src_start = array_element_address(src, offset, src_elem);
5458
5459 // We assume that range check is done by caller.
5460 // TODO: generate range check (offset+length < src.length) in debug VM.
5461
5462 // Call the stub.
5463 address stubAddr = StubRoutines::updateBytesCRC32();
5464 const char *stubName = "updateBytesCRC32";
5465
5466 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5467 stubAddr, stubName, TypePtr::BOTTOM,
5468 crc, src_start, length);
5469 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5470 set_result(result);
5471 return true;
5472 }
5473
5474 /**
5475 * Calculate CRC32 for ByteBuffer.
5476 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5477 */
5478 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5479 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5480 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5481 // no receiver since it is static method
5482 Node* crc = argument(0); // type: int
5483 Node* src = argument(1); // type: long
5484 Node* offset = argument(3); // type: int
5485 Node* length = argument(4); // type: int
5486
5487 src = ConvL2X(src); // adjust Java long to machine word
5488 Node* base = _gvn.transform(new CastX2PNode(src));
5489 offset = ConvI2X(offset);
5490
5491 // 'src_start' points to src array + scaled offset
5492 Node* src_start = basic_plus_adr(top(), base, offset);
5493
5494 // Call the stub.
5495 address stubAddr = StubRoutines::updateBytesCRC32();
5496 const char *stubName = "updateBytesCRC32";
5497
5498 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5499 stubAddr, stubName, TypePtr::BOTTOM,
5500 crc, src_start, length);
5501 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5502 set_result(result);
5503 return true;
5504 }
5505
5506 //------------------------------get_table_from_crc32c_class-----------------------
5507 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5508 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5509 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5510
5511 return table;
5512 }
5513
5514 //------------------------------inline_updateBytesCRC32C-----------------------
5515 //
5516 // Calculate CRC32C for byte[] array.
5517 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5518 //
5519 bool LibraryCallKit::inline_updateBytesCRC32C() {
5520 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5521 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5522 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5523 // no receiver since it is a static method
5524 Node* crc = argument(0); // type: int
5525 Node* src = argument(1); // type: oop
5526 Node* offset = argument(2); // type: int
5527 Node* end = argument(3); // type: int
5528
5529 Node* length = _gvn.transform(new SubINode(end, offset));
5530
5531 const Type* src_type = src->Value(&_gvn);
5532 const TypeAryPtr* top_src = src_type->isa_aryptr();
5533 if (top_src == NULL || top_src->klass() == NULL) {
5534 // failed array check
5535 return false;
5536 }
5537
5538 // Figure out the size and type of the elements we will be copying.
5539 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5540 if (src_elem != T_BYTE) {
5541 return false;
5542 }
5543
5544 // 'src_start' points to src array + scaled offset
5545 src = must_be_not_null(src, true);
5546 Node* src_start = array_element_address(src, offset, src_elem);
5547
5548 // static final int[] byteTable in class CRC32C
5549 Node* table = get_table_from_crc32c_class(callee()->holder());
5550 table = must_be_not_null(table, true);
5551 Node* table_start = array_element_address(table, intcon(0), T_INT);
5552
5553 // We assume that range check is done by caller.
5554 // TODO: generate range check (offset+length < src.length) in debug VM.
5555
5556 // Call the stub.
5557 address stubAddr = StubRoutines::updateBytesCRC32C();
5558 const char *stubName = "updateBytesCRC32C";
5559
5560 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5561 stubAddr, stubName, TypePtr::BOTTOM,
5562 crc, src_start, length, table_start);
5563 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5564 set_result(result);
5565 return true;
5566 }
5567
5568 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5569 //
5570 // Calculate CRC32C for DirectByteBuffer.
5571 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5572 //
5573 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5574 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5575 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5576 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5577 // no receiver since it is a static method
5578 Node* crc = argument(0); // type: int
5579 Node* src = argument(1); // type: long
5580 Node* offset = argument(3); // type: int
5581 Node* end = argument(4); // type: int
5582
5583 Node* length = _gvn.transform(new SubINode(end, offset));
5584
5585 src = ConvL2X(src); // adjust Java long to machine word
5586 Node* base = _gvn.transform(new CastX2PNode(src));
5587 offset = ConvI2X(offset);
5588
5589 // 'src_start' points to src array + scaled offset
5590 Node* src_start = basic_plus_adr(top(), base, offset);
5591
5592 // static final int[] byteTable in class CRC32C
5593 Node* table = get_table_from_crc32c_class(callee()->holder());
5594 table = must_be_not_null(table, true);
5595 Node* table_start = array_element_address(table, intcon(0), T_INT);
5596
5597 // Call the stub.
5598 address stubAddr = StubRoutines::updateBytesCRC32C();
5599 const char *stubName = "updateBytesCRC32C";
5600
5601 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5602 stubAddr, stubName, TypePtr::BOTTOM,
5603 crc, src_start, length, table_start);
5604 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5605 set_result(result);
5606 return true;
5607 }
5608
5609 //------------------------------inline_updateBytesAdler32----------------------
5610 //
5611 // Calculate Adler32 checksum for byte[] array.
5612 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5613 //
5614 bool LibraryCallKit::inline_updateBytesAdler32() {
5615 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5616 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5617 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5618 // no receiver since it is static method
5619 Node* crc = argument(0); // type: int
5620 Node* src = argument(1); // type: oop
5621 Node* offset = argument(2); // type: int
5622 Node* length = argument(3); // type: int
5623
5624 const Type* src_type = src->Value(&_gvn);
5625 const TypeAryPtr* top_src = src_type->isa_aryptr();
5626 if (top_src == NULL || top_src->klass() == NULL) {
5627 // failed array check
5628 return false;
5629 }
5630
5631 // Figure out the size and type of the elements we will be copying.
5632 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5633 if (src_elem != T_BYTE) {
5634 return false;
5635 }
5636
5637 // 'src_start' points to src array + scaled offset
5638 Node* src_start = array_element_address(src, offset, src_elem);
5639
5640 // We assume that range check is done by caller.
5641 // TODO: generate range check (offset+length < src.length) in debug VM.
5642
5643 // Call the stub.
5644 address stubAddr = StubRoutines::updateBytesAdler32();
5645 const char *stubName = "updateBytesAdler32";
5646
5647 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5648 stubAddr, stubName, TypePtr::BOTTOM,
5649 crc, src_start, length);
5650 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5651 set_result(result);
5652 return true;
5653 }
5654
5655 //------------------------------inline_updateByteBufferAdler32---------------
5656 //
5657 // Calculate Adler32 checksum for DirectByteBuffer.
5658 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5659 //
5660 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5661 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5662 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5663 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5664 // no receiver since it is static method
5665 Node* crc = argument(0); // type: int
5666 Node* src = argument(1); // type: long
5667 Node* offset = argument(3); // type: int
5668 Node* length = argument(4); // type: int
5669
5670 src = ConvL2X(src); // adjust Java long to machine word
5671 Node* base = _gvn.transform(new CastX2PNode(src));
5672 offset = ConvI2X(offset);
5673
5674 // 'src_start' points to src array + scaled offset
5675 Node* src_start = basic_plus_adr(top(), base, offset);
5676
5677 // Call the stub.
5678 address stubAddr = StubRoutines::updateBytesAdler32();
5679 const char *stubName = "updateBytesAdler32";
5680
5681 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5682 stubAddr, stubName, TypePtr::BOTTOM,
5683 crc, src_start, length);
5684
5685 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5686 set_result(result);
5687 return true;
5688 }
5689
5690 //----------------------------inline_reference_get----------------------------
5691 // public T java.lang.ref.Reference.get();
5692 bool LibraryCallKit::inline_reference_get() {
5693 const int referent_offset = java_lang_ref_Reference::referent_offset();
5694
5695 // Get the argument:
5696 Node* reference_obj = null_check_receiver();
5697 if (stopped()) return true;
5698
5699 const TypeInstPtr* tinst = _gvn.type(reference_obj)->isa_instptr();
5700 assert(tinst != NULL, "obj is null");
5701 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5702 ciInstanceKlass* referenceKlass = tinst->klass()->as_instance_klass();
5703 ciField* field = referenceKlass->get_field_by_name(ciSymbol::make("referent"),
5704 ciSymbol::make("Ljava/lang/Object;"),
5705 false);
5706 assert (field != NULL, "undefined field");
5707
5708 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5709 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5710
5711 ciInstanceKlass* klass = env()->Object_klass();
5712 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5713
5714 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
5715 Node* result = access_load_at(reference_obj, adr, adr_type, object_type, T_OBJECT, decorators);
5716 // Add memory barrier to prevent commoning reads from this field
5717 // across safepoint since GC can change its value.
5718 insert_mem_bar(Op_MemBarCPUOrder);
5719
5720 set_result(result);
5721 return true;
5722 }
5723
5724
5725 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5726 bool is_exact=true, bool is_static=false,
5727 ciInstanceKlass * fromKls=NULL) {
5728 if (fromKls == NULL) {
5729 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5730 assert(tinst != NULL, "obj is null");
5731 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5732 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5733 fromKls = tinst->klass()->as_instance_klass();
5734 } else {
5735 assert(is_static, "only for static field access");
5736 }
5737 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5738 ciSymbol::make(fieldTypeString),
5739 is_static);
5740
5741 assert (field != NULL, "undefined field");
5742 if (field == NULL) return (Node *) NULL;
5743
5744 if (is_static) {
5745 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5746 fromObj = makecon(tip);
5747 }
5748
5749 // Next code copied from Parse::do_get_xxx():
5750
5751 // Compute address and memory type.
5752 int offset = field->offset_in_bytes();
5753 bool is_vol = field->is_volatile();
5754 ciType* field_klass = field->type();
5755 assert(field_klass->is_loaded(), "should be loaded");
5756 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5757 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5758 BasicType bt = field->layout_type();
5759
5760 // Build the resultant type of the load
5761 const Type *type;
5762 if (bt == T_OBJECT) {
5763 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5764 } else {
5765 type = Type::get_const_basic_type(bt);
5766 }
5767
5768 DecoratorSet decorators = IN_HEAP;
5769
5770 if (is_vol) {
5771 decorators |= MO_SEQ_CST;
5772 }
5773
5774 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
5775 }
5776
5777 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5778 bool is_exact = true, bool is_static = false,
5779 ciInstanceKlass * fromKls = NULL) {
5780 if (fromKls == NULL) {
5781 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5782 assert(tinst != NULL, "obj is null");
5783 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5784 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5785 fromKls = tinst->klass()->as_instance_klass();
5786 }
5787 else {
5788 assert(is_static, "only for static field access");
5789 }
5790 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5791 ciSymbol::make(fieldTypeString),
5792 is_static);
5793
5794 assert(field != NULL, "undefined field");
5795 assert(!field->is_volatile(), "not defined for volatile fields");
5796
5797 if (is_static) {
5798 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5799 fromObj = makecon(tip);
5800 }
5801
5802 // Next code copied from Parse::do_get_xxx():
5803
5804 // Compute address and memory type.
5805 int offset = field->offset_in_bytes();
5806 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5807
5808 return adr;
5809 }
5810
5811 //------------------------------inline_aescrypt_Block-----------------------
5812 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
5813 address stubAddr = NULL;
5814 const char *stubName;
5815 assert(UseAES, "need AES instruction support");
5816
5817 switch(id) {
5818 case vmIntrinsics::_aescrypt_encryptBlock:
5819 stubAddr = StubRoutines::aescrypt_encryptBlock();
5820 stubName = "aescrypt_encryptBlock";
5821 break;
5822 case vmIntrinsics::_aescrypt_decryptBlock:
5823 stubAddr = StubRoutines::aescrypt_decryptBlock();
5824 stubName = "aescrypt_decryptBlock";
5825 break;
5826 default:
5827 break;
5828 }
5829 if (stubAddr == NULL) return false;
5830
5831 Node* aescrypt_object = argument(0);
5832 Node* src = argument(1);
5833 Node* src_offset = argument(2);
5834 Node* dest = argument(3);
5835 Node* dest_offset = argument(4);
5836
5837 src = must_be_not_null(src, true);
5838 dest = must_be_not_null(dest, true);
5839
5840 // (1) src and dest are arrays.
5841 const Type* src_type = src->Value(&_gvn);
5842 const Type* dest_type = dest->Value(&_gvn);
5843 const TypeAryPtr* top_src = src_type->isa_aryptr();
5844 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5845 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5846
5847 // for the quick and dirty code we will skip all the checks.
5848 // we are just trying to get the call to be generated.
5849 Node* src_start = src;
5850 Node* dest_start = dest;
5851 if (src_offset != NULL || dest_offset != NULL) {
5852 assert(src_offset != NULL && dest_offset != NULL, "");
5853 src_start = array_element_address(src, src_offset, T_BYTE);
5854 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5855 }
5856
5857 // now need to get the start of its expanded key array
5858 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5859 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5860 if (k_start == NULL) return false;
5861
5862 if (Matcher::pass_original_key_for_aes()) {
5863 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5864 // compatibility issues between Java key expansion and SPARC crypto instructions
5865 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5866 if (original_k_start == NULL) return false;
5867
5868 // Call the stub.
5869 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5870 stubAddr, stubName, TypePtr::BOTTOM,
5871 src_start, dest_start, k_start, original_k_start);
5872 } else {
5873 // Call the stub.
5874 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5875 stubAddr, stubName, TypePtr::BOTTOM,
5876 src_start, dest_start, k_start);
5877 }
5878
5879 return true;
5880 }
5881
5882 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
5883 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
5884 address stubAddr = NULL;
5885 const char *stubName = NULL;
5886
5887 assert(UseAES, "need AES instruction support");
5888
5889 switch(id) {
5890 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
5891 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
5892 stubName = "cipherBlockChaining_encryptAESCrypt";
5893 break;
5894 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
5895 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
5896 stubName = "cipherBlockChaining_decryptAESCrypt";
5897 break;
5898 default:
5899 break;
5900 }
5901 if (stubAddr == NULL) return false;
5902
5903 Node* cipherBlockChaining_object = argument(0);
5904 Node* src = argument(1);
5905 Node* src_offset = argument(2);
5906 Node* len = argument(3);
5907 Node* dest = argument(4);
5908 Node* dest_offset = argument(5);
5909
5910 src = must_be_not_null(src, false);
5911 dest = must_be_not_null(dest, false);
5912
5913 // (1) src and dest are arrays.
5914 const Type* src_type = src->Value(&_gvn);
5915 const Type* dest_type = dest->Value(&_gvn);
5916 const TypeAryPtr* top_src = src_type->isa_aryptr();
5917 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5918 assert (top_src != NULL && top_src->klass() != NULL
5919 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5920
5921 // checks are the responsibility of the caller
5922 Node* src_start = src;
5923 Node* dest_start = dest;
5924 if (src_offset != NULL || dest_offset != NULL) {
5925 assert(src_offset != NULL && dest_offset != NULL, "");
5926 src_start = array_element_address(src, src_offset, T_BYTE);
5927 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5928 }
5929
5930 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
5931 // (because of the predicated logic executed earlier).
5932 // so we cast it here safely.
5933 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5934
5935 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
5936 if (embeddedCipherObj == NULL) return false;
5937
5938 // cast it to what we know it will be at runtime
5939 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
5940 assert(tinst != NULL, "CBC obj is null");
5941 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
5942 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5943 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
5944
5945 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
5946 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
5947 const TypeOopPtr* xtype = aklass->as_instance_type();
5948 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
5949 aescrypt_object = _gvn.transform(aescrypt_object);
5950
5951 // we need to get the start of the aescrypt_object's expanded key array
5952 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5953 if (k_start == NULL) return false;
5954
5955 // similarly, get the start address of the r vector
5956 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
5957 if (objRvec == NULL) return false;
5958 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
5959
5960 Node* cbcCrypt;
5961 if (Matcher::pass_original_key_for_aes()) {
5962 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5963 // compatibility issues between Java key expansion and SPARC crypto instructions
5964 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5965 if (original_k_start == NULL) return false;
5966
5967 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
5968 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5969 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5970 stubAddr, stubName, TypePtr::BOTTOM,
5971 src_start, dest_start, k_start, r_start, len, original_k_start);
5972 } else {
5973 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
5974 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
5975 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
5976 stubAddr, stubName, TypePtr::BOTTOM,
5977 src_start, dest_start, k_start, r_start, len);
5978 }
5979
5980 // return cipher length (int)
5981 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
5982 set_result(retvalue);
5983 return true;
5984 }
5985
5986 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
5987 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
5988 address stubAddr = NULL;
5989 const char *stubName = NULL;
5990
5991 assert(UseAES, "need AES instruction support");
5992
5993 switch (id) {
5994 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
5995 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
5996 stubName = "electronicCodeBook_encryptAESCrypt";
5997 break;
5998 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
5999 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
6000 stubName = "electronicCodeBook_decryptAESCrypt";
6001 break;
6002 default:
6003 break;
6004 }
6005
6006 if (stubAddr == NULL) return false;
6007
6008 Node* electronicCodeBook_object = argument(0);
6009 Node* src = argument(1);
6010 Node* src_offset = argument(2);
6011 Node* len = argument(3);
6012 Node* dest = argument(4);
6013 Node* dest_offset = argument(5);
6014
6015 // (1) src and dest are arrays.
6016 const Type* src_type = src->Value(&_gvn);
6017 const Type* dest_type = dest->Value(&_gvn);
6018 const TypeAryPtr* top_src = src_type->isa_aryptr();
6019 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6020 assert(top_src != NULL && top_src->klass() != NULL
6021 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6022
6023 // checks are the responsibility of the caller
6024 Node* src_start = src;
6025 Node* dest_start = dest;
6026 if (src_offset != NULL || dest_offset != NULL) {
6027 assert(src_offset != NULL && dest_offset != NULL, "");
6028 src_start = array_element_address(src, src_offset, T_BYTE);
6029 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6030 }
6031
6032 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6033 // (because of the predicated logic executed earlier).
6034 // so we cast it here safely.
6035 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6036
6037 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6038 if (embeddedCipherObj == NULL) return false;
6039
6040 // cast it to what we know it will be at runtime
6041 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6042 assert(tinst != NULL, "ECB obj is null");
6043 assert(tinst->klass()->is_loaded(), "ECB obj is not loaded");
6044 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6045 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6046
6047 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6048 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6049 const TypeOopPtr* xtype = aklass->as_instance_type();
6050 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6051 aescrypt_object = _gvn.transform(aescrypt_object);
6052
6053 // we need to get the start of the aescrypt_object's expanded key array
6054 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6055 if (k_start == NULL) return false;
6056
6057 Node* ecbCrypt;
6058 if (Matcher::pass_original_key_for_aes()) {
6059 // no SPARC version for AES/ECB intrinsics now.
6060 return false;
6061 }
6062 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6063 ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6064 OptoRuntime::electronicCodeBook_aescrypt_Type(),
6065 stubAddr, stubName, TypePtr::BOTTOM,
6066 src_start, dest_start, k_start, len);
6067
6068 // return cipher length (int)
6069 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6070 set_result(retvalue);
6071 return true;
6072 }
6073
6074 //------------------------------inline_counterMode_AESCrypt-----------------------
6075 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6076 assert(UseAES, "need AES instruction support");
6077 if (!UseAESCTRIntrinsics) return false;
6078
6079 address stubAddr = NULL;
6080 const char *stubName = NULL;
6081 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6082 stubAddr = StubRoutines::counterMode_AESCrypt();
6083 stubName = "counterMode_AESCrypt";
6084 }
6085 if (stubAddr == NULL) return false;
6086
6087 Node* counterMode_object = argument(0);
6088 Node* src = argument(1);
6089 Node* src_offset = argument(2);
6090 Node* len = argument(3);
6091 Node* dest = argument(4);
6092 Node* dest_offset = argument(5);
6093
6094 // (1) src and dest are arrays.
6095 const Type* src_type = src->Value(&_gvn);
6096 const Type* dest_type = dest->Value(&_gvn);
6097 const TypeAryPtr* top_src = src_type->isa_aryptr();
6098 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6099 assert(top_src != NULL && top_src->klass() != NULL &&
6100 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6101
6102 // checks are the responsibility of the caller
6103 Node* src_start = src;
6104 Node* dest_start = dest;
6105 if (src_offset != NULL || dest_offset != NULL) {
6106 assert(src_offset != NULL && dest_offset != NULL, "");
6107 src_start = array_element_address(src, src_offset, T_BYTE);
6108 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6109 }
6110
6111 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6112 // (because of the predicated logic executed earlier).
6113 // so we cast it here safely.
6114 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6115 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6116 if (embeddedCipherObj == NULL) return false;
6117 // cast it to what we know it will be at runtime
6118 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6119 assert(tinst != NULL, "CTR obj is null");
6120 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6121 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6122 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6123 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6124 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6125 const TypeOopPtr* xtype = aklass->as_instance_type();
6126 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6127 aescrypt_object = _gvn.transform(aescrypt_object);
6128 // we need to get the start of the aescrypt_object's expanded key array
6129 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6130 if (k_start == NULL) return false;
6131 // similarly, get the start address of the r vector
6132 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6133 if (obj_counter == NULL) return false;
6134 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6135
6136 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6137 if (saved_encCounter == NULL) return false;
6138 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6139 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6140
6141 Node* ctrCrypt;
6142 if (Matcher::pass_original_key_for_aes()) {
6143 // no SPARC version for AES/CTR intrinsics now.
6144 return false;
6145 }
6146 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6147 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6148 OptoRuntime::counterMode_aescrypt_Type(),
6149 stubAddr, stubName, TypePtr::BOTTOM,
6150 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6151
6152 // return cipher length (int)
6153 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6154 set_result(retvalue);
6155 return true;
6156 }
6157
6158 //------------------------------get_key_start_from_aescrypt_object-----------------------
6159 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6160 #if defined(PPC64) || defined(S390)
6161 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6162 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6163 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6164 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6165 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6166 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6167 if (objSessionK == NULL) {
6168 return (Node *) NULL;
6169 }
6170 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6171 #else
6172 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6173 #endif // PPC64
6174 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6175 if (objAESCryptKey == NULL) return (Node *) NULL;
6176
6177 // now have the array, need to get the start address of the K array
6178 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6179 return k_start;
6180 }
6181
6182 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6183 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6184 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6185 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6186 if (objAESCryptKey == NULL) return (Node *) NULL;
6187
6188 // now have the array, need to get the start address of the lastKey array
6189 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6190 return original_k_start;
6191 }
6192
6193 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6194 // Return node representing slow path of predicate check.
6195 // the pseudo code we want to emulate with this predicate is:
6196 // for encryption:
6197 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6198 // for decryption:
6199 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6200 // note cipher==plain is more conservative than the original java code but that's OK
6201 //
6202 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6203 // The receiver was checked for NULL already.
6204 Node* objCBC = argument(0);
6205
6206 Node* src = argument(1);
6207 Node* dest = argument(4);
6208
6209 // Load embeddedCipher field of CipherBlockChaining object.
6210 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6211
6212 // get AESCrypt klass for instanceOf check
6213 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6214 // will have same classloader as CipherBlockChaining object
6215 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6216 assert(tinst != NULL, "CBCobj is null");
6217 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6218
6219 // we want to do an instanceof comparison against the AESCrypt class
6220 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6221 if (!klass_AESCrypt->is_loaded()) {
6222 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6223 Node* ctrl = control();
6224 set_control(top()); // no regular fast path
6225 return ctrl;
6226 }
6227
6228 src = must_be_not_null(src, true);
6229 dest = must_be_not_null(dest, true);
6230
6231 // Resolve oops to stable for CmpP below.
6232 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6233
6234 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6235 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6236 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6237
6238 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6239
6240 // for encryption, we are done
6241 if (!decrypting)
6242 return instof_false; // even if it is NULL
6243
6244 // for decryption, we need to add a further check to avoid
6245 // taking the intrinsic path when cipher and plain are the same
6246 // see the original java code for why.
6247 RegionNode* region = new RegionNode(3);
6248 region->init_req(1, instof_false);
6249
6250 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6251 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6252 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6253 region->init_req(2, src_dest_conjoint);
6254
6255 record_for_igvn(region);
6256 return _gvn.transform(region);
6257 }
6258
6259 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
6260 // Return node representing slow path of predicate check.
6261 // the pseudo code we want to emulate with this predicate is:
6262 // for encryption:
6263 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6264 // for decryption:
6265 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6266 // note cipher==plain is more conservative than the original java code but that's OK
6267 //
6268 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
6269 // The receiver was checked for NULL already.
6270 Node* objECB = argument(0);
6271
6272 // Load embeddedCipher field of ElectronicCodeBook object.
6273 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6274
6275 // get AESCrypt klass for instanceOf check
6276 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6277 // will have same classloader as ElectronicCodeBook object
6278 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
6279 assert(tinst != NULL, "ECBobj is null");
6280 assert(tinst->klass()->is_loaded(), "ECBobj is not loaded");
6281
6282 // we want to do an instanceof comparison against the AESCrypt class
6283 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6284 if (!klass_AESCrypt->is_loaded()) {
6285 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6286 Node* ctrl = control();
6287 set_control(top()); // no regular fast path
6288 return ctrl;
6289 }
6290 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6291
6292 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6293 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6294 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6295
6296 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6297
6298 // for encryption, we are done
6299 if (!decrypting)
6300 return instof_false; // even if it is NULL
6301
6302 // for decryption, we need to add a further check to avoid
6303 // taking the intrinsic path when cipher and plain are the same
6304 // see the original java code for why.
6305 RegionNode* region = new RegionNode(3);
6306 region->init_req(1, instof_false);
6307 Node* src = argument(1);
6308 Node* dest = argument(4);
6309 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6310 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6311 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6312 region->init_req(2, src_dest_conjoint);
6313
6314 record_for_igvn(region);
6315 return _gvn.transform(region);
6316 }
6317
6318 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6319 // Return node representing slow path of predicate check.
6320 // the pseudo code we want to emulate with this predicate is:
6321 // for encryption:
6322 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6323 // for decryption:
6324 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6325 // note cipher==plain is more conservative than the original java code but that's OK
6326 //
6327
6328 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6329 // The receiver was checked for NULL already.
6330 Node* objCTR = argument(0);
6331
6332 // Load embeddedCipher field of CipherBlockChaining object.
6333 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6334
6335 // get AESCrypt klass for instanceOf check
6336 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6337 // will have same classloader as CipherBlockChaining object
6338 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6339 assert(tinst != NULL, "CTRobj is null");
6340 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6341
6342 // we want to do an instanceof comparison against the AESCrypt class
6343 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6344 if (!klass_AESCrypt->is_loaded()) {
6345 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6346 Node* ctrl = control();
6347 set_control(top()); // no regular fast path
6348 return ctrl;
6349 }
6350
6351 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6352 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6353 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6354 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6355 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6356
6357 return instof_false; // even if it is NULL
6358 }
6359
6360 //------------------------------inline_ghash_processBlocks
6361 bool LibraryCallKit::inline_ghash_processBlocks() {
6362 address stubAddr;
6363 const char *stubName;
6364 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6365
6366 stubAddr = StubRoutines::ghash_processBlocks();
6367 stubName = "ghash_processBlocks";
6368
6369 Node* data = argument(0);
6370 Node* offset = argument(1);
6371 Node* len = argument(2);
6372 Node* state = argument(3);
6373 Node* subkeyH = argument(4);
6374
6375 state = must_be_not_null(state, true);
6376 subkeyH = must_be_not_null(subkeyH, true);
6377 data = must_be_not_null(data, true);
6378
6379 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6380 assert(state_start, "state is NULL");
6381 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6382 assert(subkeyH_start, "subkeyH is NULL");
6383 Node* data_start = array_element_address(data, offset, T_BYTE);
6384 assert(data_start, "data is NULL");
6385
6386 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6387 OptoRuntime::ghash_processBlocks_Type(),
6388 stubAddr, stubName, TypePtr::BOTTOM,
6389 state_start, subkeyH_start, data_start, len);
6390 return true;
6391 }
6392
6393 bool LibraryCallKit::inline_base64_encodeBlock() {
6394 address stubAddr;
6395 const char *stubName;
6396 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6397 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6398 stubAddr = StubRoutines::base64_encodeBlock();
6399 stubName = "encodeBlock";
6400
6401 if (!stubAddr) return false;
6402 Node* base64obj = argument(0);
6403 Node* src = argument(1);
6404 Node* offset = argument(2);
6405 Node* len = argument(3);
6406 Node* dest = argument(4);
6407 Node* dp = argument(5);
6408 Node* isURL = argument(6);
6409
6410 src = must_be_not_null(src, true);
6411 dest = must_be_not_null(dest, true);
6412
6413 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6414 assert(src_start, "source array is NULL");
6415 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6416 assert(dest_start, "destination array is NULL");
6417
6418 Node* base64 = make_runtime_call(RC_LEAF,
6419 OptoRuntime::base64_encodeBlock_Type(),
6420 stubAddr, stubName, TypePtr::BOTTOM,
6421 src_start, offset, len, dest_start, dp, isURL);
6422 return true;
6423 }
6424
6425 //------------------------------inline_sha_implCompress-----------------------
6426 //
6427 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6428 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6429 //
6430 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6431 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6432 //
6433 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6434 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6435 //
6436 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6437 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6438
6439 Node* sha_obj = argument(0);
6440 Node* src = argument(1); // type oop
6441 Node* ofs = argument(2); // type int
6442
6443 const Type* src_type = src->Value(&_gvn);
6444 const TypeAryPtr* top_src = src_type->isa_aryptr();
6445 if (top_src == NULL || top_src->klass() == NULL) {
6446 // failed array check
6447 return false;
6448 }
6449 // Figure out the size and type of the elements we will be copying.
6450 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6451 if (src_elem != T_BYTE) {
6452 return false;
6453 }
6454 // 'src_start' points to src array + offset
6455 src = must_be_not_null(src, true);
6456 Node* src_start = array_element_address(src, ofs, src_elem);
6457 Node* state = NULL;
6458 address stubAddr;
6459 const char *stubName;
6460
6461 switch(id) {
6462 case vmIntrinsics::_sha_implCompress:
6463 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6464 state = get_state_from_sha_object(sha_obj);
6465 stubAddr = StubRoutines::sha1_implCompress();
6466 stubName = "sha1_implCompress";
6467 break;
6468 case vmIntrinsics::_sha2_implCompress:
6469 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6470 state = get_state_from_sha_object(sha_obj);
6471 stubAddr = StubRoutines::sha256_implCompress();
6472 stubName = "sha256_implCompress";
6473 break;
6474 case vmIntrinsics::_sha5_implCompress:
6475 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6476 state = get_state_from_sha5_object(sha_obj);
6477 stubAddr = StubRoutines::sha512_implCompress();
6478 stubName = "sha512_implCompress";
6479 break;
6480 default:
6481 fatal_unexpected_iid(id);
6482 return false;
6483 }
6484 if (state == NULL) return false;
6485
6486 assert(stubAddr != NULL, "Stub is generated");
6487 if (stubAddr == NULL) return false;
6488
6489 // Call the stub.
6490 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6491 stubAddr, stubName, TypePtr::BOTTOM,
6492 src_start, state);
6493
6494 return true;
6495 }
6496
6497 //------------------------------inline_digestBase_implCompressMB-----------------------
6498 //
6499 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6500 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6501 //
6502 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6503 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6504 "need SHA1/SHA256/SHA512 instruction support");
6505 assert((uint)predicate < 3, "sanity");
6506 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6507
6508 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6509 Node* src = argument(1); // byte[] array
6510 Node* ofs = argument(2); // type int
6511 Node* limit = argument(3); // type int
6512
6513 const Type* src_type = src->Value(&_gvn);
6514 const TypeAryPtr* top_src = src_type->isa_aryptr();
6515 if (top_src == NULL || top_src->klass() == NULL) {
6516 // failed array check
6517 return false;
6518 }
6519 // Figure out the size and type of the elements we will be copying.
6520 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6521 if (src_elem != T_BYTE) {
6522 return false;
6523 }
6524 // 'src_start' points to src array + offset
6525 src = must_be_not_null(src, false);
6526 Node* src_start = array_element_address(src, ofs, src_elem);
6527
6528 const char* klass_SHA_name = NULL;
6529 const char* stub_name = NULL;
6530 address stub_addr = NULL;
6531 bool long_state = false;
6532
6533 switch (predicate) {
6534 case 0:
6535 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
6536 klass_SHA_name = "sun/security/provider/SHA";
6537 stub_name = "sha1_implCompressMB";
6538 stub_addr = StubRoutines::sha1_implCompressMB();
6539 }
6540 break;
6541 case 1:
6542 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
6543 klass_SHA_name = "sun/security/provider/SHA2";
6544 stub_name = "sha256_implCompressMB";
6545 stub_addr = StubRoutines::sha256_implCompressMB();
6546 }
6547 break;
6548 case 2:
6549 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
6550 klass_SHA_name = "sun/security/provider/SHA5";
6551 stub_name = "sha512_implCompressMB";
6552 stub_addr = StubRoutines::sha512_implCompressMB();
6553 long_state = true;
6554 }
6555 break;
6556 default:
6557 fatal("unknown SHA intrinsic predicate: %d", predicate);
6558 }
6559 if (klass_SHA_name != NULL) {
6560 assert(stub_addr != NULL, "Stub is generated");
6561 if (stub_addr == NULL) return false;
6562
6563 // get DigestBase klass to lookup for SHA klass
6564 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6565 assert(tinst != NULL, "digestBase_obj is not instance???");
6566 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6567
6568 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6569 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6570 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6571 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6572 }
6573 return false;
6574 }
6575 //------------------------------inline_sha_implCompressMB-----------------------
6576 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6577 bool long_state, address stubAddr, const char *stubName,
6578 Node* src_start, Node* ofs, Node* limit) {
6579 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6580 const TypeOopPtr* xtype = aklass->as_instance_type();
6581 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6582 sha_obj = _gvn.transform(sha_obj);
6583
6584 Node* state;
6585 if (long_state) {
6586 state = get_state_from_sha5_object(sha_obj);
6587 } else {
6588 state = get_state_from_sha_object(sha_obj);
6589 }
6590 if (state == NULL) return false;
6591
6592 // Call the stub.
6593 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6594 OptoRuntime::digestBase_implCompressMB_Type(),
6595 stubAddr, stubName, TypePtr::BOTTOM,
6596 src_start, state, ofs, limit);
6597 // return ofs (int)
6598 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6599 set_result(result);
6600
6601 return true;
6602 }
6603
6604 //------------------------------get_state_from_sha_object-----------------------
6605 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6606 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6607 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6608 if (sha_state == NULL) return (Node *) NULL;
6609
6610 // now have the array, need to get the start address of the state array
6611 Node* state = array_element_address(sha_state, intcon(0), T_INT);
6612 return state;
6613 }
6614
6615 //------------------------------get_state_from_sha5_object-----------------------
6616 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6617 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6618 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6619 if (sha_state == NULL) return (Node *) NULL;
6620
6621 // now have the array, need to get the start address of the state array
6622 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6623 return state;
6624 }
6625
6626 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6627 // Return node representing slow path of predicate check.
6628 // the pseudo code we want to emulate with this predicate is:
6629 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6630 //
6631 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6632 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6633 "need SHA1/SHA256/SHA512 instruction support");
6634 assert((uint)predicate < 3, "sanity");
6635
6636 // The receiver was checked for NULL already.
6637 Node* digestBaseObj = argument(0);
6638
6639 // get DigestBase klass for instanceOf check
6640 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6641 assert(tinst != NULL, "digestBaseObj is null");
6642 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6643
6644 const char* klass_SHA_name = NULL;
6645 switch (predicate) {
6646 case 0:
6647 if (UseSHA1Intrinsics) {
6648 // we want to do an instanceof comparison against the SHA class
6649 klass_SHA_name = "sun/security/provider/SHA";
6650 }
6651 break;
6652 case 1:
6653 if (UseSHA256Intrinsics) {
6654 // we want to do an instanceof comparison against the SHA2 class
6655 klass_SHA_name = "sun/security/provider/SHA2";
6656 }
6657 break;
6658 case 2:
6659 if (UseSHA512Intrinsics) {
6660 // we want to do an instanceof comparison against the SHA5 class
6661 klass_SHA_name = "sun/security/provider/SHA5";
6662 }
6663 break;
6664 default:
6665 fatal("unknown SHA intrinsic predicate: %d", predicate);
6666 }
6667
6668 ciKlass* klass_SHA = NULL;
6669 if (klass_SHA_name != NULL) {
6670 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6671 }
6672 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6673 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6674 Node* ctrl = control();
6675 set_control(top()); // no intrinsic path
6676 return ctrl;
6677 }
6678 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6679
6680 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6681 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6682 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6683 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6684
6685 return instof_false; // even if it is NULL
6686 }
6687
6688 //-------------inline_fma-----------------------------------
6689 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
6690 Node *a = NULL;
6691 Node *b = NULL;
6692 Node *c = NULL;
6693 Node* result = NULL;
6694 switch (id) {
6695 case vmIntrinsics::_fmaD:
6696 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
6697 // no receiver since it is static method
6698 a = round_double_node(argument(0));
6699 b = round_double_node(argument(2));
6700 c = round_double_node(argument(4));
6701 result = _gvn.transform(new FmaDNode(control(), a, b, c));
6702 break;
6703 case vmIntrinsics::_fmaF:
6704 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
6705 a = argument(0);
6706 b = argument(1);
6707 c = argument(2);
6708 result = _gvn.transform(new FmaFNode(control(), a, b, c));
6709 break;
6710 default:
6711 fatal_unexpected_iid(id); break;
6712 }
6713 set_result(result);
6714 return true;
6715 }
6716
6717 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
6718 // argument(0) is receiver
6719 Node* codePoint = argument(1);
6720 Node* n = NULL;
6721
6722 switch (id) {
6723 case vmIntrinsics::_isDigit :
6724 n = new DigitNode(control(), codePoint);
6725 break;
6726 case vmIntrinsics::_isLowerCase :
6727 n = new LowerCaseNode(control(), codePoint);
6728 break;
6729 case vmIntrinsics::_isUpperCase :
6730 n = new UpperCaseNode(control(), codePoint);
6731 break;
6732 case vmIntrinsics::_isWhitespace :
6733 n = new WhitespaceNode(control(), codePoint);
6734 break;
6735 default:
6736 fatal_unexpected_iid(id);
6737 }
6738
6739 set_result(_gvn.transform(n));
6740 return true;
6741 }
6742
6743 //------------------------------inline_fp_min_max------------------------------
6744 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
6745 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
6746
6747 // The intrinsic should be used only when the API branches aren't predictable,
6748 // the last one performing the most important comparison. The following heuristic
6749 // uses the branch statistics to eventually bail out if necessary.
6750
6751 ciMethodData *md = callee()->method_data();
6752
6753 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
6754 ciCallProfile cp = caller()->call_profile_at_bci(bci());
6755
6756 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
6757 // Bail out if the call-site didn't contribute enough to the statistics.
6758 return false;
6759 }
6760
6761 uint taken = 0, not_taken = 0;
6762
6763 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
6764 if (p->is_BranchData()) {
6765 taken = ((ciBranchData*)p)->taken();
6766 not_taken = ((ciBranchData*)p)->not_taken();
6767 }
6768 }
6769
6770 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
6771 balance = balance < 0 ? -balance : balance;
6772 if ( balance > 0.2 ) {
6773 // Bail out if the most important branch is predictable enough.
6774 return false;
6775 }
6776 }
6777 */
6778
6779 Node *a = NULL;
6780 Node *b = NULL;
6781 Node *n = NULL;
6782 switch (id) {
6783 case vmIntrinsics::_maxF:
6784 case vmIntrinsics::_minF:
6785 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
6786 a = argument(0);
6787 b = argument(1);
6788 break;
6789 case vmIntrinsics::_maxD:
6790 case vmIntrinsics::_minD:
6791 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
6792 a = round_double_node(argument(0));
6793 b = round_double_node(argument(2));
6794 break;
6795 default:
6796 fatal_unexpected_iid(id);
6797 break;
6798 }
6799 switch (id) {
6800 case vmIntrinsics::_maxF: n = new MaxFNode(a, b); break;
6801 case vmIntrinsics::_minF: n = new MinFNode(a, b); break;
6802 case vmIntrinsics::_maxD: n = new MaxDNode(a, b); break;
6803 case vmIntrinsics::_minD: n = new MinDNode(a, b); break;
6804 default: fatal_unexpected_iid(id); break;
6805 }
6806 set_result(_gvn.transform(n));
6807 return true;
6808 }
6809
6810 bool LibraryCallKit::inline_profileBoolean() {
6811 Node* counts = argument(1);
6812 const TypeAryPtr* ary = NULL;
6813 ciArray* aobj = NULL;
6814 if (counts->is_Con()
6815 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6816 && (aobj = ary->const_oop()->as_array()) != NULL
6817 && (aobj->length() == 2)) {
6818 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6819 jint false_cnt = aobj->element_value(0).as_int();
6820 jint true_cnt = aobj->element_value(1).as_int();
6821
6822 if (C->log() != NULL) {
6823 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6824 false_cnt, true_cnt);
6825 }
6826
6827 if (false_cnt + true_cnt == 0) {
6828 // According to profile, never executed.
6829 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6830 Deoptimization::Action_reinterpret);
6831 return true;
6832 }
6833
6834 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6835 // is a number of each value occurrences.
6836 Node* result = argument(0);
6837 if (false_cnt == 0 || true_cnt == 0) {
6838 // According to profile, one value has been never seen.
6839 int expected_val = (false_cnt == 0) ? 1 : 0;
6840
6841 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6842 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6843
6844 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6845 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6846 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6847
6848 { // Slow path: uncommon trap for never seen value and then reexecute
6849 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6850 // the value has been seen at least once.
6851 PreserveJVMState pjvms(this);
6852 PreserveReexecuteState preexecs(this);
6853 jvms()->set_should_reexecute(true);
6854
6855 set_control(slow_path);
6856 set_i_o(i_o());
6857
6858 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6859 Deoptimization::Action_reinterpret);
6860 }
6861 // The guard for never seen value enables sharpening of the result and
6862 // returning a constant. It allows to eliminate branches on the same value
6863 // later on.
6864 set_control(fast_path);
6865 result = intcon(expected_val);
6866 }
6867 // Stop profiling.
6868 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6869 // By replacing method body with profile data (represented as ProfileBooleanNode
6870 // on IR level) we effectively disable profiling.
6871 // It enables full speed execution once optimized code is generated.
6872 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6873 C->record_for_igvn(profile);
6874 set_result(profile);
6875 return true;
6876 } else {
6877 // Continue profiling.
6878 // Profile data isn't available at the moment. So, execute method's bytecode version.
6879 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6880 // is compiled and counters aren't available since corresponding MethodHandle
6881 // isn't a compile-time constant.
6882 return false;
6883 }
6884 }
6885
6886 bool LibraryCallKit::inline_isCompileConstant() {
6887 Node* n = argument(0);
6888 set_result(n->is_Con() ? intcon(1) : intcon(0));
6889 return true;
6890 }