1 /*
2 * Copyright (c) 1999, 2019, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/systemDictionary.hpp"
29 #include "classfile/vmSymbols.hpp"
30 #include "compiler/compileBroker.hpp"
31 #include "compiler/compileLog.hpp"
32 #include "gc/shared/barrierSet.hpp"
33 #include "jfr/support/jfrIntrinsics.hpp"
34 #include "memory/resourceArea.hpp"
35 #include "oops/klass.inline.hpp"
36 #include "oops/objArrayKlass.hpp"
37 #include "opto/addnode.hpp"
38 #include "opto/arraycopynode.hpp"
39 #include "opto/c2compiler.hpp"
40 #include "opto/callGenerator.hpp"
41 #include "opto/castnode.hpp"
42 #include "opto/cfgnode.hpp"
43 #include "opto/convertnode.hpp"
44 #include "opto/countbitsnode.hpp"
45 #include "opto/intrinsicnode.hpp"
46 #include "opto/idealKit.hpp"
47 #include "opto/mathexactnode.hpp"
48 #include "opto/movenode.hpp"
49 #include "opto/mulnode.hpp"
50 #include "opto/narrowptrnode.hpp"
51 #include "opto/opaquenode.hpp"
52 #include "opto/parse.hpp"
53 #include "opto/runtime.hpp"
54 #include "opto/rootnode.hpp"
55 #include "opto/subnode.hpp"
56 #include "prims/nativeLookup.hpp"
57 #include "prims/unsafe.hpp"
58 #include "runtime/objectMonitor.hpp"
59 #include "runtime/sharedRuntime.hpp"
60 #include "utilities/macros.hpp"
61
62
63 class LibraryIntrinsic : public InlineCallGenerator {
64 // Extend the set of intrinsics known to the runtime:
65 public:
66 private:
67 bool _is_virtual;
68 bool _does_virtual_dispatch;
69 int8_t _predicates_count; // Intrinsic is predicated by several conditions
70 int8_t _last_predicate; // Last generated predicate
71 vmIntrinsics::ID _intrinsic_id;
72
73 public:
74 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
75 : InlineCallGenerator(m),
76 _is_virtual(is_virtual),
77 _does_virtual_dispatch(does_virtual_dispatch),
78 _predicates_count((int8_t)predicates_count),
79 _last_predicate((int8_t)-1),
80 _intrinsic_id(id)
81 {
82 }
83 virtual bool is_intrinsic() const { return true; }
84 virtual bool is_virtual() const { return _is_virtual; }
85 virtual bool is_predicated() const { return _predicates_count > 0; }
86 virtual int predicates_count() const { return _predicates_count; }
87 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; }
88 virtual JVMState* generate(JVMState* jvms);
89 virtual Node* generate_predicate(JVMState* jvms, int predicate);
90 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
91 };
92
93
94 // Local helper class for LibraryIntrinsic:
95 class LibraryCallKit : public GraphKit {
96 private:
97 LibraryIntrinsic* _intrinsic; // the library intrinsic being called
98 Node* _result; // the result node, if any
99 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted
100
101 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type);
102
103 public:
104 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
105 : GraphKit(jvms),
106 _intrinsic(intrinsic),
107 _result(NULL)
108 {
109 // Check if this is a root compile. In that case we don't have a caller.
110 if (!jvms->has_method()) {
111 _reexecute_sp = sp();
112 } else {
113 // Find out how many arguments the interpreter needs when deoptimizing
114 // and save the stack pointer value so it can used by uncommon_trap.
115 // We find the argument count by looking at the declared signature.
116 bool ignored_will_link;
117 ciSignature* declared_signature = NULL;
118 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
119 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
120 _reexecute_sp = sp() + nargs; // "push" arguments back on stack
121 }
122 }
123
124 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
125
126 ciMethod* caller() const { return jvms()->method(); }
127 int bci() const { return jvms()->bci(); }
128 LibraryIntrinsic* intrinsic() const { return _intrinsic; }
129 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); }
130 ciMethod* callee() const { return _intrinsic->method(); }
131
132 bool try_to_inline(int predicate);
133 Node* try_to_predicate(int predicate);
134
135 void push_result() {
136 // Push the result onto the stack.
137 if (!stopped() && result() != NULL) {
138 BasicType bt = result()->bottom_type()->basic_type();
139 push_node(bt, result());
140 }
141 }
142
143 private:
144 void fatal_unexpected_iid(vmIntrinsics::ID iid) {
145 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid));
146 }
147
148 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
149 void set_result(RegionNode* region, PhiNode* value);
150 Node* result() { return _result; }
151
152 virtual int reexecute_sp() { return _reexecute_sp; }
153
154 // Helper functions to inline natives
155 Node* generate_guard(Node* test, RegionNode* region, float true_prob);
156 Node* generate_slow_guard(Node* test, RegionNode* region);
157 Node* generate_fair_guard(Node* test, RegionNode* region);
158 Node* generate_negative_guard(Node* index, RegionNode* region,
159 // resulting CastII of index:
160 Node* *pos_index = NULL);
161 Node* generate_limit_guard(Node* offset, Node* subseq_length,
162 Node* array_length,
163 RegionNode* region);
164 void generate_string_range_check(Node* array, Node* offset,
165 Node* length, bool char_count);
166 Node* generate_current_thread(Node* &tls_output);
167 Node* load_mirror_from_klass(Node* klass);
168 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
169 RegionNode* region, int null_path,
170 int offset);
171 Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
172 RegionNode* region, int null_path) {
173 int offset = java_lang_Class::klass_offset_in_bytes();
174 return load_klass_from_mirror_common(mirror, never_see_null,
175 region, null_path,
176 offset);
177 }
178 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
179 RegionNode* region, int null_path) {
180 int offset = java_lang_Class::array_klass_offset_in_bytes();
181 return load_klass_from_mirror_common(mirror, never_see_null,
182 region, null_path,
183 offset);
184 }
185 Node* generate_access_flags_guard(Node* kls,
186 int modifier_mask, int modifier_bits,
187 RegionNode* region);
188 Node* generate_interface_guard(Node* kls, RegionNode* region);
189 Node* generate_array_guard(Node* kls, RegionNode* region) {
190 return generate_array_guard_common(kls, region, false, false);
191 }
192 Node* generate_non_array_guard(Node* kls, RegionNode* region) {
193 return generate_array_guard_common(kls, region, false, true);
194 }
195 Node* generate_objArray_guard(Node* kls, RegionNode* region) {
196 return generate_array_guard_common(kls, region, true, false);
197 }
198 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
199 return generate_array_guard_common(kls, region, true, true);
200 }
201 Node* generate_array_guard_common(Node* kls, RegionNode* region,
202 bool obj_array, bool not_array);
203 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
204 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
205 bool is_virtual = false, bool is_static = false);
206 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
207 return generate_method_call(method_id, false, true);
208 }
209 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
210 return generate_method_call(method_id, true, false);
211 }
212 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
213 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
214
215 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
216 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
217 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
218 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
219 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
220 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
221 bool inline_string_indexOfChar();
222 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
223 bool inline_string_toBytesU();
224 bool inline_string_getCharsU();
225 bool inline_string_copy(bool compress);
226 bool inline_string_char_access(bool is_store);
227 Node* round_double_node(Node* n);
228 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
229 bool inline_math_native(vmIntrinsics::ID id);
230 bool inline_math(vmIntrinsics::ID id);
231 bool inline_double_math(vmIntrinsics::ID id);
232 template <typename OverflowOp>
233 bool inline_math_overflow(Node* arg1, Node* arg2);
234 void inline_math_mathExact(Node* math, Node* test);
235 bool inline_math_addExactI(bool is_increment);
236 bool inline_math_addExactL(bool is_increment);
237 bool inline_math_multiplyExactI();
238 bool inline_math_multiplyExactL();
239 bool inline_math_multiplyHigh();
240 bool inline_math_negateExactI();
241 bool inline_math_negateExactL();
242 bool inline_math_subtractExactI(bool is_decrement);
243 bool inline_math_subtractExactL(bool is_decrement);
244 bool inline_min_max(vmIntrinsics::ID id);
245 bool inline_notify(vmIntrinsics::ID id);
246 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
247 // This returns Type::AnyPtr, RawPtr, or OopPtr.
248 int classify_unsafe_addr(Node* &base, Node* &offset, BasicType type);
249 Node* make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type = T_ILLEGAL, bool can_cast = false);
250
251 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
252 DecoratorSet mo_decorator_for_access_kind(AccessKind kind);
253 bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
254 static bool klass_needs_init_guard(Node* kls);
255 bool inline_unsafe_allocate();
256 bool inline_unsafe_newArray(bool uninitialized);
257 bool inline_unsafe_writeback0();
258 bool inline_unsafe_writebackSync0(bool is_pre);
259 bool inline_unsafe_copyMemory();
260 bool inline_native_currentThread();
261
262 bool inline_native_time_funcs(address method, const char* funcName);
263 #ifdef JFR_HAVE_INTRINSICS
264 bool inline_native_classID();
265 bool inline_native_getEventWriter();
266 #endif
267 bool inline_native_Class_query(vmIntrinsics::ID id);
268 bool inline_native_subtype_check();
269 bool inline_native_getLength();
270 bool inline_array_copyOf(bool is_copyOfRange);
271 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
272 bool inline_preconditions_checkIndex();
273 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array);
274 bool inline_native_clone(bool is_virtual);
275 bool inline_native_Reflection_getCallerClass();
276 // Helper function for inlining native object hash method
277 bool inline_native_hashcode(bool is_virtual, bool is_static);
278 bool inline_native_getClass();
279
280 // Helper functions for inlining arraycopy
281 bool inline_arraycopy();
282 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
283 RegionNode* slow_region);
284 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
285 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp,
286 uint new_idx);
287
288 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
289 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
290 bool inline_unsafe_fence(vmIntrinsics::ID id);
291 bool inline_onspinwait();
292 bool inline_fp_conversions(vmIntrinsics::ID id);
293 bool inline_number_methods(vmIntrinsics::ID id);
294 bool inline_reference_get();
295 bool inline_Class_cast();
296 bool inline_aescrypt_Block(vmIntrinsics::ID id);
297 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
298 bool inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id);
299 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
300 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
301 Node* inline_electronicCodeBook_AESCrypt_predicate(bool decrypting);
302 Node* inline_counterMode_AESCrypt_predicate();
303 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
304 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
305 bool inline_ghash_processBlocks();
306 bool inline_base64_encodeBlock();
307 bool inline_sha_implCompress(vmIntrinsics::ID id);
308 bool inline_digestBase_implCompressMB(int predicate);
309 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
310 bool long_state, address stubAddr, const char *stubName,
311 Node* src_start, Node* ofs, Node* limit);
312 Node* get_state_from_sha_object(Node *sha_object);
313 Node* get_state_from_sha5_object(Node *sha_object);
314 Node* inline_digestBase_implCompressMB_predicate(int predicate);
315 bool inline_encodeISOArray();
316 bool inline_updateCRC32();
317 bool inline_updateBytesCRC32();
318 bool inline_updateByteBufferCRC32();
319 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
320 bool inline_updateBytesCRC32C();
321 bool inline_updateDirectByteBufferCRC32C();
322 bool inline_updateBytesAdler32();
323 bool inline_updateByteBufferAdler32();
324 bool inline_multiplyToLen();
325 bool inline_hasNegatives();
326 bool inline_squareToLen();
327 bool inline_mulAdd();
328 bool inline_montgomeryMultiply();
329 bool inline_montgomerySquare();
330 bool inline_bigIntegerShift(bool isRightShift);
331 bool inline_vectorizedMismatch();
332 bool inline_fma(vmIntrinsics::ID id);
333 bool inline_character_compare(vmIntrinsics::ID id);
334 bool inline_fp_min_max(vmIntrinsics::ID id);
335
336 bool inline_profileBoolean();
337 bool inline_isCompileConstant();
338 void clear_upper_avx() {
339 #ifdef X86
340 if (UseAVX >= 2) {
341 C->set_clear_upper_avx(true);
342 }
343 #endif
344 }
345 };
346
347 //---------------------------make_vm_intrinsic----------------------------
348 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
349 vmIntrinsics::ID id = m->intrinsic_id();
350 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
351
352 if (!m->is_loaded()) {
353 // Do not attempt to inline unloaded methods.
354 return NULL;
355 }
356
357 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
358 bool is_available = false;
359
360 {
361 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
362 // the compiler must transition to '_thread_in_vm' state because both
363 // methods access VM-internal data.
364 VM_ENTRY_MARK;
365 methodHandle mh(THREAD, m->get_Method());
366 is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
367 !C->directive()->is_intrinsic_disabled(mh) &&
368 !vmIntrinsics::is_disabled_by_flags(mh);
369
370 }
371
372 if (is_available) {
373 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
374 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
375 return new LibraryIntrinsic(m, is_virtual,
376 vmIntrinsics::predicates_needed(id),
377 vmIntrinsics::does_virtual_dispatch(id),
378 (vmIntrinsics::ID) id);
379 } else {
380 return NULL;
381 }
382 }
383
384 //----------------------register_library_intrinsics-----------------------
385 // Initialize this file's data structures, for each Compile instance.
386 void Compile::register_library_intrinsics() {
387 // Nothing to do here.
388 }
389
390 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
391 LibraryCallKit kit(jvms, this);
392 Compile* C = kit.C;
393 int nodes = C->unique();
394 #ifndef PRODUCT
395 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
396 char buf[1000];
397 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
398 tty->print_cr("Intrinsic %s", str);
399 }
400 #endif
401 ciMethod* callee = kit.callee();
402 const int bci = kit.bci();
403
404 // Try to inline the intrinsic.
405 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
406 kit.try_to_inline(_last_predicate)) {
407 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
408 : "(intrinsic)";
409 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
410 if (C->print_intrinsics() || C->print_inlining()) {
411 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
412 }
413 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
414 if (C->log()) {
415 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
416 vmIntrinsics::name_at(intrinsic_id()),
417 (is_virtual() ? " virtual='1'" : ""),
418 C->unique() - nodes);
419 }
420 // Push the result from the inlined method onto the stack.
421 kit.push_result();
422 C->print_inlining_update(this);
423 return kit.transfer_exceptions_into_jvms();
424 }
425
426 // The intrinsic bailed out
427 if (jvms->has_method()) {
428 // Not a root compile.
429 const char* msg;
430 if (callee->intrinsic_candidate()) {
431 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
432 } else {
433 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
434 : "failed to inline (intrinsic), method not annotated";
435 }
436 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
437 if (C->print_intrinsics() || C->print_inlining()) {
438 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
439 }
440 } else {
441 // Root compile
442 ResourceMark rm;
443 stringStream msg_stream;
444 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
445 vmIntrinsics::name_at(intrinsic_id()),
446 is_virtual() ? " (virtual)" : "", bci);
447 const char *msg = msg_stream.as_string();
448 log_debug(jit, inlining)("%s", msg);
449 if (C->print_intrinsics() || C->print_inlining()) {
450 tty->print("%s", msg);
451 }
452 }
453 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
454 C->print_inlining_update(this);
455 return NULL;
456 }
457
458 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
459 LibraryCallKit kit(jvms, this);
460 Compile* C = kit.C;
461 int nodes = C->unique();
462 _last_predicate = predicate;
463 #ifndef PRODUCT
464 assert(is_predicated() && predicate < predicates_count(), "sanity");
465 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
466 char buf[1000];
467 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
468 tty->print_cr("Predicate for intrinsic %s", str);
469 }
470 #endif
471 ciMethod* callee = kit.callee();
472 const int bci = kit.bci();
473
474 Node* slow_ctl = kit.try_to_predicate(predicate);
475 if (!kit.failing()) {
476 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
477 : "(intrinsic, predicate)";
478 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
479 if (C->print_intrinsics() || C->print_inlining()) {
480 C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
481 }
482 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
483 if (C->log()) {
484 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
485 vmIntrinsics::name_at(intrinsic_id()),
486 (is_virtual() ? " virtual='1'" : ""),
487 C->unique() - nodes);
488 }
489 return slow_ctl; // Could be NULL if the check folds.
490 }
491
492 // The intrinsic bailed out
493 if (jvms->has_method()) {
494 // Not a root compile.
495 const char* msg = "failed to generate predicate for intrinsic";
496 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
497 if (C->print_intrinsics() || C->print_inlining()) {
498 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
499 }
500 } else {
501 // Root compile
502 ResourceMark rm;
503 stringStream msg_stream;
504 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
505 vmIntrinsics::name_at(intrinsic_id()),
506 is_virtual() ? " (virtual)" : "", bci);
507 const char *msg = msg_stream.as_string();
508 log_debug(jit, inlining)("%s", msg);
509 if (C->print_intrinsics() || C->print_inlining()) {
510 C->print_inlining_stream()->print("%s", msg);
511 }
512 }
513 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
514 return NULL;
515 }
516
517 bool LibraryCallKit::try_to_inline(int predicate) {
518 // Handle symbolic names for otherwise undistinguished boolean switches:
519 const bool is_store = true;
520 const bool is_compress = true;
521 const bool is_static = true;
522 const bool is_volatile = true;
523
524 if (!jvms()->has_method()) {
525 // Root JVMState has a null method.
526 assert(map()->memory()->Opcode() == Op_Parm, "");
527 // Insert the memory aliasing node
528 set_all_memory(reset_memory());
529 }
530 assert(merged_memory(), "");
531
532
533 switch (intrinsic_id()) {
534 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
535 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
536 case vmIntrinsics::_getClass: return inline_native_getClass();
537
538 case vmIntrinsics::_ceil:
539 case vmIntrinsics::_floor:
540 case vmIntrinsics::_rint:
541 case vmIntrinsics::_dsin:
542 case vmIntrinsics::_dcos:
543 case vmIntrinsics::_dtan:
544 case vmIntrinsics::_dabs:
545 case vmIntrinsics::_fabs:
546 case vmIntrinsics::_iabs:
547 case vmIntrinsics::_labs:
548 case vmIntrinsics::_datan2:
549 case vmIntrinsics::_dsqrt:
550 case vmIntrinsics::_dexp:
551 case vmIntrinsics::_dlog:
552 case vmIntrinsics::_dlog10:
553 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id());
554
555 case vmIntrinsics::_min:
556 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
557
558 case vmIntrinsics::_notify:
559 case vmIntrinsics::_notifyAll:
560 return inline_notify(intrinsic_id());
561
562 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
563 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
564 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
565 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
566 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
567 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
568 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
569 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
570 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
571 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
572 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
573 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
574 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
575
576 case vmIntrinsics::_arraycopy: return inline_arraycopy();
577
578 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
579 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
580 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
581 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
582
583 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
584 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
585 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
586 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
587 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
588 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
589 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
590
591 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
592 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
593
594 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
595 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
596 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
597 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
598
599 case vmIntrinsics::_compressStringC:
600 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
601 case vmIntrinsics::_inflateStringC:
602 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
603
604 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
605 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
606 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
607 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
608 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
609 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
610 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
611 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
612 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
613
614 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
615 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
616 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
617 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
618 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
619 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
620 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
621 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
622 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
623
624 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
625 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
626 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
627 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
628 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
629 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
630 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
631 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
632 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
633
634 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
635 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
636 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
637 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
638 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
639 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
640 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
641 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
642 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
643
644 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
645 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
646 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
647 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
648
649 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
650 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
651 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
652 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
653
654 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
655 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
656 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
657 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
658 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
659 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
660 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
661 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
662 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
663
664 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
665 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
666 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
667 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
668 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
669 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
670 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
671 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
672 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
673
674 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
675 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
676 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
677 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
678 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
679 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
680 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
681 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
682 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
683
684 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
685 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
686 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
687 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
688 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
689 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
690 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
691 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
692 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
693
694 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
695 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
696 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
697 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
698 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
699
700 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
701 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
702 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
703 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
704 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
705 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
706 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
707 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
708 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
709 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
710 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
711 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
712 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
713 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
714 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
715 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
716 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
717 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
718 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
719 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
720
721 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
722 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
723 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
724 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
725 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
726 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
727 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
728 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
729 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
730 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
731 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
732 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
733 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
734 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
735 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
736
737 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
738 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
739 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
740 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
741
742 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
743 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
744 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
745 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
746 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
747
748 case vmIntrinsics::_loadFence:
749 case vmIntrinsics::_storeFence:
750 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
751
752 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
753
754 case vmIntrinsics::_currentThread: return inline_native_currentThread();
755
756 #ifdef JFR_HAVE_INTRINSICS
757 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime");
758 case vmIntrinsics::_getClassId: return inline_native_classID();
759 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
760 #endif
761 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
762 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
763 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
764 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
765 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
766 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
767 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
768 case vmIntrinsics::_getLength: return inline_native_getLength();
769 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
770 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
771 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
772 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
773 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
774 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
775
776 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
777 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
778
779 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
780
781 case vmIntrinsics::_isInstance:
782 case vmIntrinsics::_getModifiers:
783 case vmIntrinsics::_isInterface:
784 case vmIntrinsics::_isArray:
785 case vmIntrinsics::_isPrimitive:
786 case vmIntrinsics::_getSuperclass:
787 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
788
789 case vmIntrinsics::_floatToRawIntBits:
790 case vmIntrinsics::_floatToIntBits:
791 case vmIntrinsics::_intBitsToFloat:
792 case vmIntrinsics::_doubleToRawLongBits:
793 case vmIntrinsics::_doubleToLongBits:
794 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
795
796 case vmIntrinsics::_numberOfLeadingZeros_i:
797 case vmIntrinsics::_numberOfLeadingZeros_l:
798 case vmIntrinsics::_numberOfTrailingZeros_i:
799 case vmIntrinsics::_numberOfTrailingZeros_l:
800 case vmIntrinsics::_bitCount_i:
801 case vmIntrinsics::_bitCount_l:
802 case vmIntrinsics::_reverseBytes_i:
803 case vmIntrinsics::_reverseBytes_l:
804 case vmIntrinsics::_reverseBytes_s:
805 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
806
807 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
808
809 case vmIntrinsics::_Reference_get: return inline_reference_get();
810
811 case vmIntrinsics::_Class_cast: return inline_Class_cast();
812
813 case vmIntrinsics::_aescrypt_encryptBlock:
814 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
815
816 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
817 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
818 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
819
820 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
821 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
822 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
823
824 case vmIntrinsics::_counterMode_AESCrypt:
825 return inline_counterMode_AESCrypt(intrinsic_id());
826
827 case vmIntrinsics::_sha_implCompress:
828 case vmIntrinsics::_sha2_implCompress:
829 case vmIntrinsics::_sha5_implCompress:
830 return inline_sha_implCompress(intrinsic_id());
831
832 case vmIntrinsics::_digestBase_implCompressMB:
833 return inline_digestBase_implCompressMB(predicate);
834
835 case vmIntrinsics::_multiplyToLen:
836 return inline_multiplyToLen();
837
838 case vmIntrinsics::_squareToLen:
839 return inline_squareToLen();
840
841 case vmIntrinsics::_mulAdd:
842 return inline_mulAdd();
843
844 case vmIntrinsics::_montgomeryMultiply:
845 return inline_montgomeryMultiply();
846 case vmIntrinsics::_montgomerySquare:
847 return inline_montgomerySquare();
848
849 case vmIntrinsics::_bigIntegerRightShiftWorker:
850 return inline_bigIntegerShift(true);
851 case vmIntrinsics::_bigIntegerLeftShiftWorker:
852 return inline_bigIntegerShift(false);
853
854 case vmIntrinsics::_vectorizedMismatch:
855 return inline_vectorizedMismatch();
856
857 case vmIntrinsics::_ghash_processBlocks:
858 return inline_ghash_processBlocks();
859 case vmIntrinsics::_base64_encodeBlock:
860 return inline_base64_encodeBlock();
861
862 case vmIntrinsics::_encodeISOArray:
863 case vmIntrinsics::_encodeByteISOArray:
864 return inline_encodeISOArray();
865
866 case vmIntrinsics::_updateCRC32:
867 return inline_updateCRC32();
868 case vmIntrinsics::_updateBytesCRC32:
869 return inline_updateBytesCRC32();
870 case vmIntrinsics::_updateByteBufferCRC32:
871 return inline_updateByteBufferCRC32();
872
873 case vmIntrinsics::_updateBytesCRC32C:
874 return inline_updateBytesCRC32C();
875 case vmIntrinsics::_updateDirectByteBufferCRC32C:
876 return inline_updateDirectByteBufferCRC32C();
877
878 case vmIntrinsics::_updateBytesAdler32:
879 return inline_updateBytesAdler32();
880 case vmIntrinsics::_updateByteBufferAdler32:
881 return inline_updateByteBufferAdler32();
882
883 case vmIntrinsics::_profileBoolean:
884 return inline_profileBoolean();
885 case vmIntrinsics::_isCompileConstant:
886 return inline_isCompileConstant();
887
888 case vmIntrinsics::_hasNegatives:
889 return inline_hasNegatives();
890
891 case vmIntrinsics::_fmaD:
892 case vmIntrinsics::_fmaF:
893 return inline_fma(intrinsic_id());
894
895 case vmIntrinsics::_isDigit:
896 case vmIntrinsics::_isLowerCase:
897 case vmIntrinsics::_isUpperCase:
898 case vmIntrinsics::_isWhitespace:
899 return inline_character_compare(intrinsic_id());
900
901 case vmIntrinsics::_maxF:
902 case vmIntrinsics::_minF:
903 case vmIntrinsics::_maxD:
904 case vmIntrinsics::_minD:
905 return inline_fp_min_max(intrinsic_id());
906
907 default:
908 // If you get here, it may be that someone has added a new intrinsic
909 // to the list in vmSymbols.hpp without implementing it here.
910 #ifndef PRODUCT
911 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
912 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
913 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
914 }
915 #endif
916 return false;
917 }
918 }
919
920 Node* LibraryCallKit::try_to_predicate(int predicate) {
921 if (!jvms()->has_method()) {
922 // Root JVMState has a null method.
923 assert(map()->memory()->Opcode() == Op_Parm, "");
924 // Insert the memory aliasing node
925 set_all_memory(reset_memory());
926 }
927 assert(merged_memory(), "");
928
929 switch (intrinsic_id()) {
930 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
931 return inline_cipherBlockChaining_AESCrypt_predicate(false);
932 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
933 return inline_cipherBlockChaining_AESCrypt_predicate(true);
934 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
935 return inline_electronicCodeBook_AESCrypt_predicate(false);
936 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
937 return inline_electronicCodeBook_AESCrypt_predicate(true);
938 case vmIntrinsics::_counterMode_AESCrypt:
939 return inline_counterMode_AESCrypt_predicate();
940 case vmIntrinsics::_digestBase_implCompressMB:
941 return inline_digestBase_implCompressMB_predicate(predicate);
942
943 default:
944 // If you get here, it may be that someone has added a new intrinsic
945 // to the list in vmSymbols.hpp without implementing it here.
946 #ifndef PRODUCT
947 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
948 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
949 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
950 }
951 #endif
952 Node* slow_ctl = control();
953 set_control(top()); // No fast path instrinsic
954 return slow_ctl;
955 }
956 }
957
958 //------------------------------set_result-------------------------------
959 // Helper function for finishing intrinsics.
960 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
961 record_for_igvn(region);
962 set_control(_gvn.transform(region));
963 set_result( _gvn.transform(value));
964 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
965 }
966
967 //------------------------------generate_guard---------------------------
968 // Helper function for generating guarded fast-slow graph structures.
969 // The given 'test', if true, guards a slow path. If the test fails
970 // then a fast path can be taken. (We generally hope it fails.)
971 // In all cases, GraphKit::control() is updated to the fast path.
972 // The returned value represents the control for the slow path.
973 // The return value is never 'top'; it is either a valid control
974 // or NULL if it is obvious that the slow path can never be taken.
975 // Also, if region and the slow control are not NULL, the slow edge
976 // is appended to the region.
977 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
978 if (stopped()) {
979 // Already short circuited.
980 return NULL;
981 }
982
983 // Build an if node and its projections.
984 // If test is true we take the slow path, which we assume is uncommon.
985 if (_gvn.type(test) == TypeInt::ZERO) {
986 // The slow branch is never taken. No need to build this guard.
987 return NULL;
988 }
989
990 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
991
992 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
993 if (if_slow == top()) {
994 // The slow branch is never taken. No need to build this guard.
995 return NULL;
996 }
997
998 if (region != NULL)
999 region->add_req(if_slow);
1000
1001 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
1002 set_control(if_fast);
1003
1004 return if_slow;
1005 }
1006
1007 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
1008 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
1009 }
1010 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
1011 return generate_guard(test, region, PROB_FAIR);
1012 }
1013
1014 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
1015 Node* *pos_index) {
1016 if (stopped())
1017 return NULL; // already stopped
1018 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
1019 return NULL; // index is already adequately typed
1020 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
1021 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1022 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
1023 if (is_neg != NULL && pos_index != NULL) {
1024 // Emulate effect of Parse::adjust_map_after_if.
1025 Node* ccast = new CastIINode(index, TypeInt::POS);
1026 ccast->set_req(0, control());
1027 (*pos_index) = _gvn.transform(ccast);
1028 }
1029 return is_neg;
1030 }
1031
1032 // Make sure that 'position' is a valid limit index, in [0..length].
1033 // There are two equivalent plans for checking this:
1034 // A. (offset + copyLength) unsigned<= arrayLength
1035 // B. offset <= (arrayLength - copyLength)
1036 // We require that all of the values above, except for the sum and
1037 // difference, are already known to be non-negative.
1038 // Plan A is robust in the face of overflow, if offset and copyLength
1039 // are both hugely positive.
1040 //
1041 // Plan B is less direct and intuitive, but it does not overflow at
1042 // all, since the difference of two non-negatives is always
1043 // representable. Whenever Java methods must perform the equivalent
1044 // check they generally use Plan B instead of Plan A.
1045 // For the moment we use Plan A.
1046 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
1047 Node* subseq_length,
1048 Node* array_length,
1049 RegionNode* region) {
1050 if (stopped())
1051 return NULL; // already stopped
1052 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
1053 if (zero_offset && subseq_length->eqv_uncast(array_length))
1054 return NULL; // common case of whole-array copy
1055 Node* last = subseq_length;
1056 if (!zero_offset) // last += offset
1057 last = _gvn.transform(new AddINode(last, offset));
1058 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
1059 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
1060 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
1061 return is_over;
1062 }
1063
1064 // Emit range checks for the given String.value byte array
1065 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
1066 if (stopped()) {
1067 return; // already stopped
1068 }
1069 RegionNode* bailout = new RegionNode(1);
1070 record_for_igvn(bailout);
1071 if (char_count) {
1072 // Convert char count to byte count
1073 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1074 }
1075
1076 // Offset and count must not be negative
1077 generate_negative_guard(offset, bailout);
1078 generate_negative_guard(count, bailout);
1079 // Offset + count must not exceed length of array
1080 generate_limit_guard(offset, count, load_array_length(array), bailout);
1081
1082 if (bailout->req() > 1) {
1083 PreserveJVMState pjvms(this);
1084 set_control(_gvn.transform(bailout));
1085 uncommon_trap(Deoptimization::Reason_intrinsic,
1086 Deoptimization::Action_maybe_recompile);
1087 }
1088 }
1089
1090 //--------------------------generate_current_thread--------------------
1091 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1092 ciKlass* thread_klass = env()->Thread_klass();
1093 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1094 Node* thread = _gvn.transform(new ThreadLocalNode());
1095 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1096 Node* threadObj = _gvn.transform(LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(), thread_type, T_OBJECT, MemNode::unordered));
1097 tls_output = thread;
1098 return threadObj;
1099 }
1100
1101
1102 //------------------------------make_string_method_node------------------------
1103 // Helper method for String intrinsic functions. This version is called with
1104 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1105 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1106 // containing the lengths of str1 and str2.
1107 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1108 Node* result = NULL;
1109 switch (opcode) {
1110 case Op_StrIndexOf:
1111 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1112 str1_start, cnt1, str2_start, cnt2, ae);
1113 break;
1114 case Op_StrComp:
1115 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1116 str1_start, cnt1, str2_start, cnt2, ae);
1117 break;
1118 case Op_StrEquals:
1119 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1120 // Use the constant length if there is one because optimized match rule may exist.
1121 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1122 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1123 break;
1124 default:
1125 ShouldNotReachHere();
1126 return NULL;
1127 }
1128
1129 // All these intrinsics have checks.
1130 C->set_has_split_ifs(true); // Has chance for split-if optimization
1131 clear_upper_avx();
1132
1133 return _gvn.transform(result);
1134 }
1135
1136 //------------------------------inline_string_compareTo------------------------
1137 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1138 Node* arg1 = argument(0);
1139 Node* arg2 = argument(1);
1140
1141 arg1 = must_be_not_null(arg1, true);
1142 arg2 = must_be_not_null(arg2, true);
1143
1144 arg1 = access_resolve(arg1, ACCESS_READ);
1145 arg2 = access_resolve(arg2, ACCESS_READ);
1146
1147 // Get start addr and length of first argument
1148 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1149 Node* arg1_cnt = load_array_length(arg1);
1150
1151 // Get start addr and length of second argument
1152 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1153 Node* arg2_cnt = load_array_length(arg2);
1154
1155 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1156 set_result(result);
1157 return true;
1158 }
1159
1160 //------------------------------inline_string_equals------------------------
1161 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1162 Node* arg1 = argument(0);
1163 Node* arg2 = argument(1);
1164
1165 // paths (plus control) merge
1166 RegionNode* region = new RegionNode(3);
1167 Node* phi = new PhiNode(region, TypeInt::BOOL);
1168
1169 if (!stopped()) {
1170
1171 arg1 = must_be_not_null(arg1, true);
1172 arg2 = must_be_not_null(arg2, true);
1173
1174 arg1 = access_resolve(arg1, ACCESS_READ);
1175 arg2 = access_resolve(arg2, ACCESS_READ);
1176
1177 // Get start addr and length of first argument
1178 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1179 Node* arg1_cnt = load_array_length(arg1);
1180
1181 // Get start addr and length of second argument
1182 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1183 Node* arg2_cnt = load_array_length(arg2);
1184
1185 // Check for arg1_cnt != arg2_cnt
1186 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1187 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1188 Node* if_ne = generate_slow_guard(bol, NULL);
1189 if (if_ne != NULL) {
1190 phi->init_req(2, intcon(0));
1191 region->init_req(2, if_ne);
1192 }
1193
1194 // Check for count == 0 is done by assembler code for StrEquals.
1195
1196 if (!stopped()) {
1197 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1198 phi->init_req(1, equals);
1199 region->init_req(1, control());
1200 }
1201 }
1202
1203 // post merge
1204 set_control(_gvn.transform(region));
1205 record_for_igvn(region);
1206
1207 set_result(_gvn.transform(phi));
1208 return true;
1209 }
1210
1211 //------------------------------inline_array_equals----------------------------
1212 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1213 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1214 Node* arg1 = argument(0);
1215 Node* arg2 = argument(1);
1216
1217 arg1 = access_resolve(arg1, ACCESS_READ);
1218 arg2 = access_resolve(arg2, ACCESS_READ);
1219
1220 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1221 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1222 clear_upper_avx();
1223
1224 return true;
1225 }
1226
1227 //------------------------------inline_hasNegatives------------------------------
1228 bool LibraryCallKit::inline_hasNegatives() {
1229 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1230 return false;
1231 }
1232
1233 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1234 // no receiver since it is static method
1235 Node* ba = argument(0);
1236 Node* offset = argument(1);
1237 Node* len = argument(2);
1238
1239 ba = must_be_not_null(ba, true);
1240
1241 // Range checks
1242 generate_string_range_check(ba, offset, len, false);
1243 if (stopped()) {
1244 return true;
1245 }
1246 ba = access_resolve(ba, ACCESS_READ);
1247 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1248 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1249 set_result(_gvn.transform(result));
1250 return true;
1251 }
1252
1253 bool LibraryCallKit::inline_preconditions_checkIndex() {
1254 Node* index = argument(0);
1255 Node* length = argument(1);
1256 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1257 return false;
1258 }
1259
1260 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1261 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1262
1263 {
1264 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1265 uncommon_trap(Deoptimization::Reason_intrinsic,
1266 Deoptimization::Action_make_not_entrant);
1267 }
1268
1269 if (stopped()) {
1270 return false;
1271 }
1272
1273 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1274 BoolTest::mask btest = BoolTest::lt;
1275 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1276 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1277 _gvn.set_type(rc, rc->Value(&_gvn));
1278 if (!rc_bool->is_Con()) {
1279 record_for_igvn(rc);
1280 }
1281 set_control(_gvn.transform(new IfTrueNode(rc)));
1282 {
1283 PreserveJVMState pjvms(this);
1284 set_control(_gvn.transform(new IfFalseNode(rc)));
1285 uncommon_trap(Deoptimization::Reason_range_check,
1286 Deoptimization::Action_make_not_entrant);
1287 }
1288
1289 if (stopped()) {
1290 return false;
1291 }
1292
1293 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1294 result->set_req(0, control());
1295 result = _gvn.transform(result);
1296 set_result(result);
1297 replace_in_map(index, result);
1298 clear_upper_avx();
1299 return true;
1300 }
1301
1302 //------------------------------inline_string_indexOf------------------------
1303 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1304 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1305 return false;
1306 }
1307 Node* src = argument(0);
1308 Node* tgt = argument(1);
1309
1310 // Make the merge point
1311 RegionNode* result_rgn = new RegionNode(4);
1312 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1313
1314 src = must_be_not_null(src, true);
1315 tgt = must_be_not_null(tgt, true);
1316
1317 src = access_resolve(src, ACCESS_READ);
1318 tgt = access_resolve(tgt, ACCESS_READ);
1319
1320 // Get start addr and length of source string
1321 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1322 Node* src_count = load_array_length(src);
1323
1324 // Get start addr and length of substring
1325 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1326 Node* tgt_count = load_array_length(tgt);
1327
1328 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1329 // Divide src size by 2 if String is UTF16 encoded
1330 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1331 }
1332 if (ae == StrIntrinsicNode::UU) {
1333 // Divide substring size by 2 if String is UTF16 encoded
1334 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1335 }
1336
1337 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1338 if (result != NULL) {
1339 result_phi->init_req(3, result);
1340 result_rgn->init_req(3, control());
1341 }
1342 set_control(_gvn.transform(result_rgn));
1343 record_for_igvn(result_rgn);
1344 set_result(_gvn.transform(result_phi));
1345
1346 return true;
1347 }
1348
1349 //-----------------------------inline_string_indexOf-----------------------
1350 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1351 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1352 return false;
1353 }
1354 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1355 return false;
1356 }
1357 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1358 Node* src = argument(0); // byte[]
1359 Node* src_count = argument(1); // char count
1360 Node* tgt = argument(2); // byte[]
1361 Node* tgt_count = argument(3); // char count
1362 Node* from_index = argument(4); // char index
1363
1364 src = must_be_not_null(src, true);
1365 tgt = must_be_not_null(tgt, true);
1366
1367 src = access_resolve(src, ACCESS_READ);
1368 tgt = access_resolve(tgt, ACCESS_READ);
1369
1370 // Multiply byte array index by 2 if String is UTF16 encoded
1371 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1372 src_count = _gvn.transform(new SubINode(src_count, from_index));
1373 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1374 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1375
1376 // Range checks
1377 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1378 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1379 if (stopped()) {
1380 return true;
1381 }
1382
1383 RegionNode* region = new RegionNode(5);
1384 Node* phi = new PhiNode(region, TypeInt::INT);
1385
1386 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1387 if (result != NULL) {
1388 // The result is index relative to from_index if substring was found, -1 otherwise.
1389 // Generate code which will fold into cmove.
1390 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1391 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1392
1393 Node* if_lt = generate_slow_guard(bol, NULL);
1394 if (if_lt != NULL) {
1395 // result == -1
1396 phi->init_req(3, result);
1397 region->init_req(3, if_lt);
1398 }
1399 if (!stopped()) {
1400 result = _gvn.transform(new AddINode(result, from_index));
1401 phi->init_req(4, result);
1402 region->init_req(4, control());
1403 }
1404 }
1405
1406 set_control(_gvn.transform(region));
1407 record_for_igvn(region);
1408 set_result(_gvn.transform(phi));
1409 clear_upper_avx();
1410
1411 return true;
1412 }
1413
1414 // Create StrIndexOfNode with fast path checks
1415 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1416 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1417 // Check for substr count > string count
1418 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1419 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1420 Node* if_gt = generate_slow_guard(bol, NULL);
1421 if (if_gt != NULL) {
1422 phi->init_req(1, intcon(-1));
1423 region->init_req(1, if_gt);
1424 }
1425 if (!stopped()) {
1426 // Check for substr count == 0
1427 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1428 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1429 Node* if_zero = generate_slow_guard(bol, NULL);
1430 if (if_zero != NULL) {
1431 phi->init_req(2, intcon(0));
1432 region->init_req(2, if_zero);
1433 }
1434 }
1435 if (!stopped()) {
1436 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1437 }
1438 return NULL;
1439 }
1440
1441 //-----------------------------inline_string_indexOfChar-----------------------
1442 bool LibraryCallKit::inline_string_indexOfChar() {
1443 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1444 return false;
1445 }
1446 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1447 return false;
1448 }
1449 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1450 Node* src = argument(0); // byte[]
1451 Node* tgt = argument(1); // tgt is int ch
1452 Node* from_index = argument(2);
1453 Node* max = argument(3);
1454
1455 src = must_be_not_null(src, true);
1456 src = access_resolve(src, ACCESS_READ);
1457
1458 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1459 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1460 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1461
1462 // Range checks
1463 generate_string_range_check(src, src_offset, src_count, true);
1464 if (stopped()) {
1465 return true;
1466 }
1467
1468 RegionNode* region = new RegionNode(3);
1469 Node* phi = new PhiNode(region, TypeInt::INT);
1470
1471 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1472 C->set_has_split_ifs(true); // Has chance for split-if optimization
1473 _gvn.transform(result);
1474
1475 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1476 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1477
1478 Node* if_lt = generate_slow_guard(bol, NULL);
1479 if (if_lt != NULL) {
1480 // result == -1
1481 phi->init_req(2, result);
1482 region->init_req(2, if_lt);
1483 }
1484 if (!stopped()) {
1485 result = _gvn.transform(new AddINode(result, from_index));
1486 phi->init_req(1, result);
1487 region->init_req(1, control());
1488 }
1489 set_control(_gvn.transform(region));
1490 record_for_igvn(region);
1491 set_result(_gvn.transform(phi));
1492
1493 return true;
1494 }
1495 //---------------------------inline_string_copy---------------------
1496 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1497 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1498 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1499 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1500 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1501 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1502 bool LibraryCallKit::inline_string_copy(bool compress) {
1503 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1504 return false;
1505 }
1506 int nargs = 5; // 2 oops, 3 ints
1507 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1508
1509 Node* src = argument(0);
1510 Node* src_offset = argument(1);
1511 Node* dst = argument(2);
1512 Node* dst_offset = argument(3);
1513 Node* length = argument(4);
1514
1515 // Check for allocation before we add nodes that would confuse
1516 // tightly_coupled_allocation()
1517 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1518
1519 // Figure out the size and type of the elements we will be copying.
1520 const Type* src_type = src->Value(&_gvn);
1521 const Type* dst_type = dst->Value(&_gvn);
1522 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1523 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1524 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1525 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1526 "Unsupported array types for inline_string_copy");
1527
1528 src = must_be_not_null(src, true);
1529 dst = must_be_not_null(dst, true);
1530
1531 // Convert char[] offsets to byte[] offsets
1532 bool convert_src = (compress && src_elem == T_BYTE);
1533 bool convert_dst = (!compress && dst_elem == T_BYTE);
1534 if (convert_src) {
1535 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1536 } else if (convert_dst) {
1537 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1538 }
1539
1540 // Range checks
1541 generate_string_range_check(src, src_offset, length, convert_src);
1542 generate_string_range_check(dst, dst_offset, length, convert_dst);
1543 if (stopped()) {
1544 return true;
1545 }
1546
1547 src = access_resolve(src, ACCESS_READ);
1548 dst = access_resolve(dst, ACCESS_WRITE);
1549
1550 Node* src_start = array_element_address(src, src_offset, src_elem);
1551 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1552 // 'src_start' points to src array + scaled offset
1553 // 'dst_start' points to dst array + scaled offset
1554 Node* count = NULL;
1555 if (compress) {
1556 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1557 } else {
1558 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1559 }
1560
1561 if (alloc != NULL) {
1562 if (alloc->maybe_set_complete(&_gvn)) {
1563 // "You break it, you buy it."
1564 InitializeNode* init = alloc->initialization();
1565 assert(init->is_complete(), "we just did this");
1566 init->set_complete_with_arraycopy();
1567 assert(dst->is_CheckCastPP(), "sanity");
1568 assert(dst->in(0)->in(0) == init, "dest pinned");
1569 }
1570 // Do not let stores that initialize this object be reordered with
1571 // a subsequent store that would make this object accessible by
1572 // other threads.
1573 // Record what AllocateNode this StoreStore protects so that
1574 // escape analysis can go from the MemBarStoreStoreNode to the
1575 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1576 // based on the escape status of the AllocateNode.
1577 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1578 }
1579 if (compress) {
1580 set_result(_gvn.transform(count));
1581 }
1582 clear_upper_avx();
1583
1584 return true;
1585 }
1586
1587 #ifdef _LP64
1588 #define XTOP ,top() /*additional argument*/
1589 #else //_LP64
1590 #define XTOP /*no additional argument*/
1591 #endif //_LP64
1592
1593 //------------------------inline_string_toBytesU--------------------------
1594 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1595 bool LibraryCallKit::inline_string_toBytesU() {
1596 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1597 return false;
1598 }
1599 // Get the arguments.
1600 Node* value = argument(0);
1601 Node* offset = argument(1);
1602 Node* length = argument(2);
1603
1604 Node* newcopy = NULL;
1605
1606 // Set the original stack and the reexecute bit for the interpreter to reexecute
1607 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1608 { PreserveReexecuteState preexecs(this);
1609 jvms()->set_should_reexecute(true);
1610
1611 // Check if a null path was taken unconditionally.
1612 value = null_check(value);
1613
1614 RegionNode* bailout = new RegionNode(1);
1615 record_for_igvn(bailout);
1616
1617 // Range checks
1618 generate_negative_guard(offset, bailout);
1619 generate_negative_guard(length, bailout);
1620 generate_limit_guard(offset, length, load_array_length(value), bailout);
1621 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1622 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1623
1624 if (bailout->req() > 1) {
1625 PreserveJVMState pjvms(this);
1626 set_control(_gvn.transform(bailout));
1627 uncommon_trap(Deoptimization::Reason_intrinsic,
1628 Deoptimization::Action_maybe_recompile);
1629 }
1630 if (stopped()) {
1631 return true;
1632 }
1633
1634 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1635 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1636 newcopy = new_array(klass_node, size, 0); // no arguments to push
1637 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1638
1639 // Calculate starting addresses.
1640 value = access_resolve(value, ACCESS_READ);
1641 Node* src_start = array_element_address(value, offset, T_CHAR);
1642 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1643
1644 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1645 const TypeInt* toffset = gvn().type(offset)->is_int();
1646 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1647
1648 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1649 const char* copyfunc_name = "arraycopy";
1650 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1651 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1652 OptoRuntime::fast_arraycopy_Type(),
1653 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1654 src_start, dst_start, ConvI2X(length) XTOP);
1655 // Do not let reads from the cloned object float above the arraycopy.
1656 if (alloc != NULL) {
1657 if (alloc->maybe_set_complete(&_gvn)) {
1658 // "You break it, you buy it."
1659 InitializeNode* init = alloc->initialization();
1660 assert(init->is_complete(), "we just did this");
1661 init->set_complete_with_arraycopy();
1662 assert(newcopy->is_CheckCastPP(), "sanity");
1663 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1664 }
1665 // Do not let stores that initialize this object be reordered with
1666 // a subsequent store that would make this object accessible by
1667 // other threads.
1668 // Record what AllocateNode this StoreStore protects so that
1669 // escape analysis can go from the MemBarStoreStoreNode to the
1670 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1671 // based on the escape status of the AllocateNode.
1672 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1673 } else {
1674 insert_mem_bar(Op_MemBarCPUOrder);
1675 }
1676 } // original reexecute is set back here
1677
1678 C->set_has_split_ifs(true); // Has chance for split-if optimization
1679 if (!stopped()) {
1680 set_result(newcopy);
1681 }
1682 clear_upper_avx();
1683
1684 return true;
1685 }
1686
1687 //------------------------inline_string_getCharsU--------------------------
1688 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1689 bool LibraryCallKit::inline_string_getCharsU() {
1690 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1691 return false;
1692 }
1693
1694 // Get the arguments.
1695 Node* src = argument(0);
1696 Node* src_begin = argument(1);
1697 Node* src_end = argument(2); // exclusive offset (i < src_end)
1698 Node* dst = argument(3);
1699 Node* dst_begin = argument(4);
1700
1701 // Check for allocation before we add nodes that would confuse
1702 // tightly_coupled_allocation()
1703 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1704
1705 // Check if a null path was taken unconditionally.
1706 src = null_check(src);
1707 dst = null_check(dst);
1708 if (stopped()) {
1709 return true;
1710 }
1711
1712 // Get length and convert char[] offset to byte[] offset
1713 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1714 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1715
1716 // Range checks
1717 generate_string_range_check(src, src_begin, length, true);
1718 generate_string_range_check(dst, dst_begin, length, false);
1719 if (stopped()) {
1720 return true;
1721 }
1722
1723 if (!stopped()) {
1724 src = access_resolve(src, ACCESS_READ);
1725 dst = access_resolve(dst, ACCESS_WRITE);
1726
1727 // Calculate starting addresses.
1728 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1729 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1730
1731 // Check if array addresses are aligned to HeapWordSize
1732 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1733 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1734 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1735 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1736
1737 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1738 const char* copyfunc_name = "arraycopy";
1739 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1740 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1741 OptoRuntime::fast_arraycopy_Type(),
1742 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1743 src_start, dst_start, ConvI2X(length) XTOP);
1744 // Do not let reads from the cloned object float above the arraycopy.
1745 if (alloc != NULL) {
1746 if (alloc->maybe_set_complete(&_gvn)) {
1747 // "You break it, you buy it."
1748 InitializeNode* init = alloc->initialization();
1749 assert(init->is_complete(), "we just did this");
1750 init->set_complete_with_arraycopy();
1751 assert(dst->is_CheckCastPP(), "sanity");
1752 assert(dst->in(0)->in(0) == init, "dest pinned");
1753 }
1754 // Do not let stores that initialize this object be reordered with
1755 // a subsequent store that would make this object accessible by
1756 // other threads.
1757 // Record what AllocateNode this StoreStore protects so that
1758 // escape analysis can go from the MemBarStoreStoreNode to the
1759 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1760 // based on the escape status of the AllocateNode.
1761 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1762 } else {
1763 insert_mem_bar(Op_MemBarCPUOrder);
1764 }
1765 }
1766
1767 C->set_has_split_ifs(true); // Has chance for split-if optimization
1768 return true;
1769 }
1770
1771 //----------------------inline_string_char_access----------------------------
1772 // Store/Load char to/from byte[] array.
1773 // static void StringUTF16.putChar(byte[] val, int index, int c)
1774 // static char StringUTF16.getChar(byte[] val, int index)
1775 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1776 Node* value = argument(0);
1777 Node* index = argument(1);
1778 Node* ch = is_store ? argument(2) : NULL;
1779
1780 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1781 // correctly requires matched array shapes.
1782 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1783 "sanity: byte[] and char[] bases agree");
1784 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1785 "sanity: byte[] and char[] scales agree");
1786
1787 // Bail when getChar over constants is requested: constant folding would
1788 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1789 // Java method would constant fold nicely instead.
1790 if (!is_store && value->is_Con() && index->is_Con()) {
1791 return false;
1792 }
1793
1794 value = must_be_not_null(value, true);
1795 value = access_resolve(value, is_store ? ACCESS_WRITE : ACCESS_READ);
1796
1797 Node* adr = array_element_address(value, index, T_CHAR);
1798 if (adr->is_top()) {
1799 return false;
1800 }
1801 if (is_store) {
1802 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1803 } else {
1804 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
1805 set_result(ch);
1806 }
1807 return true;
1808 }
1809
1810 //--------------------------round_double_node--------------------------------
1811 // Round a double node if necessary.
1812 Node* LibraryCallKit::round_double_node(Node* n) {
1813 if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1814 n = _gvn.transform(new RoundDoubleNode(0, n));
1815 return n;
1816 }
1817
1818 //------------------------------inline_math-----------------------------------
1819 // public static double Math.abs(double)
1820 // public static double Math.sqrt(double)
1821 // public static double Math.log(double)
1822 // public static double Math.log10(double)
1823 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1824 Node* arg = round_double_node(argument(0));
1825 Node* n = NULL;
1826 switch (id) {
1827 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1828 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1829 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1830 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1831 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1832 default: fatal_unexpected_iid(id); break;
1833 }
1834 set_result(_gvn.transform(n));
1835 return true;
1836 }
1837
1838 //------------------------------inline_math-----------------------------------
1839 // public static float Math.abs(float)
1840 // public static int Math.abs(int)
1841 // public static long Math.abs(long)
1842 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1843 Node* arg = argument(0);
1844 Node* n = NULL;
1845 switch (id) {
1846 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1847 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1848 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1849 default: fatal_unexpected_iid(id); break;
1850 }
1851 set_result(_gvn.transform(n));
1852 return true;
1853 }
1854
1855 //------------------------------runtime_math-----------------------------
1856 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1857 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1858 "must be (DD)D or (D)D type");
1859
1860 // Inputs
1861 Node* a = round_double_node(argument(0));
1862 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1863
1864 const TypePtr* no_memory_effects = NULL;
1865 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1866 no_memory_effects,
1867 a, top(), b, b ? top() : NULL);
1868 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1869 #ifdef ASSERT
1870 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1871 assert(value_top == top(), "second value must be top");
1872 #endif
1873
1874 set_result(value);
1875 return true;
1876 }
1877
1878 //------------------------------inline_math_native-----------------------------
1879 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1880 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1881 switch (id) {
1882 // These intrinsics are not properly supported on all hardware
1883 case vmIntrinsics::_dsin:
1884 return StubRoutines::dsin() != NULL ?
1885 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1886 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1887 case vmIntrinsics::_dcos:
1888 return StubRoutines::dcos() != NULL ?
1889 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1890 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1891 case vmIntrinsics::_dtan:
1892 return StubRoutines::dtan() != NULL ?
1893 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1894 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1895 case vmIntrinsics::_dlog:
1896 return StubRoutines::dlog() != NULL ?
1897 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1898 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1899 case vmIntrinsics::_dlog10:
1900 return StubRoutines::dlog10() != NULL ?
1901 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1902 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1903
1904 // These intrinsics are supported on all hardware
1905 case vmIntrinsics::_ceil:
1906 case vmIntrinsics::_floor:
1907 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1908 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1909 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1910 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1911 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1912 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1913
1914 case vmIntrinsics::_dexp:
1915 return StubRoutines::dexp() != NULL ?
1916 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1917 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1918 case vmIntrinsics::_dpow: {
1919 Node* exp = round_double_node(argument(2));
1920 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1921 if (d != NULL && d->getd() == 2.0) {
1922 // Special case: pow(x, 2.0) => x * x
1923 Node* base = round_double_node(argument(0));
1924 set_result(_gvn.transform(new MulDNode(base, base)));
1925 return true;
1926 }
1927 return StubRoutines::dpow() != NULL ?
1928 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1929 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1930 }
1931 #undef FN_PTR
1932
1933 // These intrinsics are not yet correctly implemented
1934 case vmIntrinsics::_datan2:
1935 return false;
1936
1937 default:
1938 fatal_unexpected_iid(id);
1939 return false;
1940 }
1941 }
1942
1943 static bool is_simple_name(Node* n) {
1944 return (n->req() == 1 // constant
1945 || (n->is_Type() && n->as_Type()->type()->singleton())
1946 || n->is_Proj() // parameter or return value
1947 || n->is_Phi() // local of some sort
1948 );
1949 }
1950
1951 //----------------------------inline_notify-----------------------------------*
1952 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1953 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1954 address func;
1955 if (id == vmIntrinsics::_notify) {
1956 func = OptoRuntime::monitor_notify_Java();
1957 } else {
1958 func = OptoRuntime::monitor_notifyAll_Java();
1959 }
1960 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1961 make_slow_call_ex(call, env()->Throwable_klass(), false);
1962 return true;
1963 }
1964
1965
1966 //----------------------------inline_min_max-----------------------------------
1967 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1968 set_result(generate_min_max(id, argument(0), argument(1)));
1969 return true;
1970 }
1971
1972 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1973 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1974 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1975 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1976 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1977
1978 {
1979 PreserveJVMState pjvms(this);
1980 PreserveReexecuteState preexecs(this);
1981 jvms()->set_should_reexecute(true);
1982
1983 set_control(slow_path);
1984 set_i_o(i_o());
1985
1986 uncommon_trap(Deoptimization::Reason_intrinsic,
1987 Deoptimization::Action_none);
1988 }
1989
1990 set_control(fast_path);
1991 set_result(math);
1992 }
1993
1994 template <typename OverflowOp>
1995 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1996 typedef typename OverflowOp::MathOp MathOp;
1997
1998 MathOp* mathOp = new MathOp(arg1, arg2);
1999 Node* operation = _gvn.transform( mathOp );
2000 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2001 inline_math_mathExact(operation, ofcheck);
2002 return true;
2003 }
2004
2005 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2006 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2007 }
2008
2009 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2010 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2011 }
2012
2013 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2014 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2015 }
2016
2017 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2018 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2019 }
2020
2021 bool LibraryCallKit::inline_math_negateExactI() {
2022 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2023 }
2024
2025 bool LibraryCallKit::inline_math_negateExactL() {
2026 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2027 }
2028
2029 bool LibraryCallKit::inline_math_multiplyExactI() {
2030 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2031 }
2032
2033 bool LibraryCallKit::inline_math_multiplyExactL() {
2034 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2035 }
2036
2037 bool LibraryCallKit::inline_math_multiplyHigh() {
2038 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2039 return true;
2040 }
2041
2042 Node*
2043 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2044 // These are the candidate return value:
2045 Node* xvalue = x0;
2046 Node* yvalue = y0;
2047
2048 if (xvalue == yvalue) {
2049 return xvalue;
2050 }
2051
2052 bool want_max = (id == vmIntrinsics::_max);
2053
2054 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
2055 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
2056 if (txvalue == NULL || tyvalue == NULL) return top();
2057 // This is not really necessary, but it is consistent with a
2058 // hypothetical MaxINode::Value method:
2059 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
2060
2061 // %%% This folding logic should (ideally) be in a different place.
2062 // Some should be inside IfNode, and there to be a more reliable
2063 // transformation of ?: style patterns into cmoves. We also want
2064 // more powerful optimizations around cmove and min/max.
2065
2066 // Try to find a dominating comparison of these guys.
2067 // It can simplify the index computation for Arrays.copyOf
2068 // and similar uses of System.arraycopy.
2069 // First, compute the normalized version of CmpI(x, y).
2070 int cmp_op = Op_CmpI;
2071 Node* xkey = xvalue;
2072 Node* ykey = yvalue;
2073 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
2074 if (ideal_cmpxy->is_Cmp()) {
2075 // E.g., if we have CmpI(length - offset, count),
2076 // it might idealize to CmpI(length, count + offset)
2077 cmp_op = ideal_cmpxy->Opcode();
2078 xkey = ideal_cmpxy->in(1);
2079 ykey = ideal_cmpxy->in(2);
2080 }
2081
2082 // Start by locating any relevant comparisons.
2083 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
2084 Node* cmpxy = NULL;
2085 Node* cmpyx = NULL;
2086 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
2087 Node* cmp = start_from->fast_out(k);
2088 if (cmp->outcnt() > 0 && // must have prior uses
2089 cmp->in(0) == NULL && // must be context-independent
2090 cmp->Opcode() == cmp_op) { // right kind of compare
2091 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
2092 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
2093 }
2094 }
2095
2096 const int NCMPS = 2;
2097 Node* cmps[NCMPS] = { cmpxy, cmpyx };
2098 int cmpn;
2099 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2100 if (cmps[cmpn] != NULL) break; // find a result
2101 }
2102 if (cmpn < NCMPS) {
2103 // Look for a dominating test that tells us the min and max.
2104 int depth = 0; // Limit search depth for speed
2105 Node* dom = control();
2106 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
2107 if (++depth >= 100) break;
2108 Node* ifproj = dom;
2109 if (!ifproj->is_Proj()) continue;
2110 Node* iff = ifproj->in(0);
2111 if (!iff->is_If()) continue;
2112 Node* bol = iff->in(1);
2113 if (!bol->is_Bool()) continue;
2114 Node* cmp = bol->in(1);
2115 if (cmp == NULL) continue;
2116 for (cmpn = 0; cmpn < NCMPS; cmpn++)
2117 if (cmps[cmpn] == cmp) break;
2118 if (cmpn == NCMPS) continue;
2119 BoolTest::mask btest = bol->as_Bool()->_test._test;
2120 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
2121 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2122 // At this point, we know that 'x btest y' is true.
2123 switch (btest) {
2124 case BoolTest::eq:
2125 // They are proven equal, so we can collapse the min/max.
2126 // Either value is the answer. Choose the simpler.
2127 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
2128 return yvalue;
2129 return xvalue;
2130 case BoolTest::lt: // x < y
2131 case BoolTest::le: // x <= y
2132 return (want_max ? yvalue : xvalue);
2133 case BoolTest::gt: // x > y
2134 case BoolTest::ge: // x >= y
2135 return (want_max ? xvalue : yvalue);
2136 default:
2137 break;
2138 }
2139 }
2140 }
2141
2142 // We failed to find a dominating test.
2143 // Let's pick a test that might GVN with prior tests.
2144 Node* best_bol = NULL;
2145 BoolTest::mask best_btest = BoolTest::illegal;
2146 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
2147 Node* cmp = cmps[cmpn];
2148 if (cmp == NULL) continue;
2149 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2150 Node* bol = cmp->fast_out(j);
2151 if (!bol->is_Bool()) continue;
2152 BoolTest::mask btest = bol->as_Bool()->_test._test;
2153 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
2154 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
2155 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2156 best_bol = bol->as_Bool();
2157 best_btest = btest;
2158 }
2159 }
2160 }
2161
2162 Node* answer_if_true = NULL;
2163 Node* answer_if_false = NULL;
2164 switch (best_btest) {
2165 default:
2166 if (cmpxy == NULL)
2167 cmpxy = ideal_cmpxy;
2168 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2169 // and fall through:
2170 case BoolTest::lt: // x < y
2171 case BoolTest::le: // x <= y
2172 answer_if_true = (want_max ? yvalue : xvalue);
2173 answer_if_false = (want_max ? xvalue : yvalue);
2174 break;
2175 case BoolTest::gt: // x > y
2176 case BoolTest::ge: // x >= y
2177 answer_if_true = (want_max ? xvalue : yvalue);
2178 answer_if_false = (want_max ? yvalue : xvalue);
2179 break;
2180 }
2181
2182 jint hi, lo;
2183 if (want_max) {
2184 // We can sharpen the minimum.
2185 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2186 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2187 } else {
2188 // We can sharpen the maximum.
2189 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2190 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2191 }
2192
2193 // Use a flow-free graph structure, to avoid creating excess control edges
2194 // which could hinder other optimizations.
2195 // Since Math.min/max is often used with arraycopy, we want
2196 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2197 Node* cmov = CMoveNode::make(NULL, best_bol,
2198 answer_if_false, answer_if_true,
2199 TypeInt::make(lo, hi, widen));
2200
2201 return _gvn.transform(cmov);
2202
2203 /*
2204 // This is not as desirable as it may seem, since Min and Max
2205 // nodes do not have a full set of optimizations.
2206 // And they would interfere, anyway, with 'if' optimizations
2207 // and with CMoveI canonical forms.
2208 switch (id) {
2209 case vmIntrinsics::_min:
2210 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2211 case vmIntrinsics::_max:
2212 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2213 default:
2214 ShouldNotReachHere();
2215 }
2216 */
2217 }
2218
2219 inline int
2220 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2221 const TypePtr* base_type = TypePtr::NULL_PTR;
2222 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2223 if (base_type == NULL) {
2224 // Unknown type.
2225 return Type::AnyPtr;
2226 } else if (base_type == TypePtr::NULL_PTR) {
2227 // Since this is a NULL+long form, we have to switch to a rawptr.
2228 base = _gvn.transform(new CastX2PNode(offset));
2229 offset = MakeConX(0);
2230 return Type::RawPtr;
2231 } else if (base_type->base() == Type::RawPtr) {
2232 return Type::RawPtr;
2233 } else if (base_type->isa_oopptr()) {
2234 // Base is never null => always a heap address.
2235 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2236 return Type::OopPtr;
2237 }
2238 // Offset is small => always a heap address.
2239 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2240 if (offset_type != NULL &&
2241 base_type->offset() == 0 && // (should always be?)
2242 offset_type->_lo >= 0 &&
2243 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2244 return Type::OopPtr;
2245 } else if (type == T_OBJECT) {
2246 // off heap access to an oop doesn't make any sense. Has to be on
2247 // heap.
2248 return Type::OopPtr;
2249 }
2250 // Otherwise, it might either be oop+off or NULL+addr.
2251 return Type::AnyPtr;
2252 } else {
2253 // No information:
2254 return Type::AnyPtr;
2255 }
2256 }
2257
2258 inline Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, DecoratorSet decorators, BasicType type, bool can_cast) {
2259 Node* uncasted_base = base;
2260 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2261 if (kind == Type::RawPtr) {
2262 return basic_plus_adr(top(), uncasted_base, offset);
2263 } else if (kind == Type::AnyPtr) {
2264 assert(base == uncasted_base, "unexpected base change");
2265 if (can_cast) {
2266 if (!_gvn.type(base)->speculative_maybe_null() &&
2267 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2268 // According to profiling, this access is always on
2269 // heap. Casting the base to not null and thus avoiding membars
2270 // around the access should allow better optimizations
2271 Node* null_ctl = top();
2272 base = null_check_oop(base, &null_ctl, true, true, true);
2273 assert(null_ctl->is_top(), "no null control here");
2274 return basic_plus_adr(base, offset);
2275 } else if (_gvn.type(base)->speculative_always_null() &&
2276 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2277 // According to profiling, this access is always off
2278 // heap.
2279 base = null_assert(base);
2280 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2281 offset = MakeConX(0);
2282 return basic_plus_adr(top(), raw_base, offset);
2283 }
2284 }
2285 // We don't know if it's an on heap or off heap access. Fall back
2286 // to raw memory access.
2287 base = access_resolve(base, decorators);
2288 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2289 return basic_plus_adr(top(), raw, offset);
2290 } else {
2291 assert(base == uncasted_base, "unexpected base change");
2292 // We know it's an on heap access so base can't be null
2293 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2294 base = must_be_not_null(base, true);
2295 }
2296 return basic_plus_adr(base, offset);
2297 }
2298 }
2299
2300 //--------------------------inline_number_methods-----------------------------
2301 // inline int Integer.numberOfLeadingZeros(int)
2302 // inline int Long.numberOfLeadingZeros(long)
2303 //
2304 // inline int Integer.numberOfTrailingZeros(int)
2305 // inline int Long.numberOfTrailingZeros(long)
2306 //
2307 // inline int Integer.bitCount(int)
2308 // inline int Long.bitCount(long)
2309 //
2310 // inline char Character.reverseBytes(char)
2311 // inline short Short.reverseBytes(short)
2312 // inline int Integer.reverseBytes(int)
2313 // inline long Long.reverseBytes(long)
2314 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2315 Node* arg = argument(0);
2316 Node* n = NULL;
2317 switch (id) {
2318 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2319 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2320 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2321 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2322 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2323 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2324 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2325 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2326 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2327 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2328 default: fatal_unexpected_iid(id); break;
2329 }
2330 set_result(_gvn.transform(n));
2331 return true;
2332 }
2333
2334 //----------------------------inline_unsafe_access----------------------------
2335
2336 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2337 // Attempt to infer a sharper value type from the offset and base type.
2338 ciKlass* sharpened_klass = NULL;
2339
2340 // See if it is an instance field, with an object type.
2341 if (alias_type->field() != NULL) {
2342 if (alias_type->field()->type()->is_klass()) {
2343 sharpened_klass = alias_type->field()->type()->as_klass();
2344 }
2345 }
2346
2347 // See if it is a narrow oop array.
2348 if (adr_type->isa_aryptr()) {
2349 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2350 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2351 if (elem_type != NULL) {
2352 sharpened_klass = elem_type->klass();
2353 }
2354 }
2355 }
2356
2357 // The sharpened class might be unloaded if there is no class loader
2358 // contraint in place.
2359 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2360 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2361
2362 #ifndef PRODUCT
2363 if (C->print_intrinsics() || C->print_inlining()) {
2364 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2365 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2366 }
2367 #endif
2368 // Sharpen the value type.
2369 return tjp;
2370 }
2371 return NULL;
2372 }
2373
2374 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2375 switch (kind) {
2376 case Relaxed:
2377 return MO_UNORDERED;
2378 case Opaque:
2379 return MO_RELAXED;
2380 case Acquire:
2381 return MO_ACQUIRE;
2382 case Release:
2383 return MO_RELEASE;
2384 case Volatile:
2385 return MO_SEQ_CST;
2386 default:
2387 ShouldNotReachHere();
2388 return 0;
2389 }
2390 }
2391
2392 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2393 if (callee()->is_static()) return false; // caller must have the capability!
2394 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2395 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2396 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2397 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2398
2399 if (is_reference_type(type)) {
2400 decorators |= ON_UNKNOWN_OOP_REF;
2401 }
2402
2403 if (unaligned) {
2404 decorators |= C2_UNALIGNED;
2405 }
2406
2407 #ifndef PRODUCT
2408 {
2409 ResourceMark rm;
2410 // Check the signatures.
2411 ciSignature* sig = callee()->signature();
2412 #ifdef ASSERT
2413 if (!is_store) {
2414 // Object getReference(Object base, int/long offset), etc.
2415 BasicType rtype = sig->return_type()->basic_type();
2416 assert(rtype == type, "getter must return the expected value");
2417 assert(sig->count() == 2, "oop getter has 2 arguments");
2418 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2419 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2420 } else {
2421 // void putReference(Object base, int/long offset, Object x), etc.
2422 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2423 assert(sig->count() == 3, "oop putter has 3 arguments");
2424 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2425 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2426 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2427 assert(vtype == type, "putter must accept the expected value");
2428 }
2429 #endif // ASSERT
2430 }
2431 #endif //PRODUCT
2432
2433 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2434
2435 Node* receiver = argument(0); // type: oop
2436
2437 // Build address expression.
2438 Node* adr;
2439 Node* heap_base_oop = top();
2440 Node* offset = top();
2441 Node* val;
2442
2443 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2444 Node* base = argument(1); // type: oop
2445 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2446 offset = argument(2); // type: long
2447 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2448 // to be plain byte offsets, which are also the same as those accepted
2449 // by oopDesc::field_addr.
2450 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2451 "fieldOffset must be byte-scaled");
2452 // 32-bit machines ignore the high half!
2453 offset = ConvL2X(offset);
2454 adr = make_unsafe_address(base, offset, is_store ? ACCESS_WRITE : ACCESS_READ, type, kind == Relaxed);
2455
2456 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2457 if (type != T_OBJECT) {
2458 decorators |= IN_NATIVE; // off-heap primitive access
2459 } else {
2460 return false; // off-heap oop accesses are not supported
2461 }
2462 } else {
2463 heap_base_oop = base; // on-heap or mixed access
2464 }
2465
2466 // Can base be NULL? Otherwise, always on-heap access.
2467 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2468
2469 if (!can_access_non_heap) {
2470 decorators |= IN_HEAP;
2471 }
2472
2473 val = is_store ? argument(4) : NULL;
2474
2475 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2476 if (adr_type == TypePtr::NULL_PTR) {
2477 return false; // off-heap access with zero address
2478 }
2479
2480 // Try to categorize the address.
2481 Compile::AliasType* alias_type = C->alias_type(adr_type);
2482 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2483
2484 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2485 alias_type->adr_type() == TypeAryPtr::RANGE) {
2486 return false; // not supported
2487 }
2488
2489 bool mismatched = false;
2490 BasicType bt = alias_type->basic_type();
2491 if (bt != T_ILLEGAL) {
2492 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2493 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2494 // Alias type doesn't differentiate between byte[] and boolean[]).
2495 // Use address type to get the element type.
2496 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2497 }
2498 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2499 // accessing an array field with getReference is not a mismatch
2500 bt = T_OBJECT;
2501 }
2502 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2503 // Don't intrinsify mismatched object accesses
2504 return false;
2505 }
2506 mismatched = (bt != type);
2507 } else if (alias_type->adr_type()->isa_oopptr()) {
2508 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2509 }
2510
2511 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2512
2513 if (mismatched) {
2514 decorators |= C2_MISMATCHED;
2515 }
2516
2517 // First guess at the value type.
2518 const Type *value_type = Type::get_const_basic_type(type);
2519
2520 // Figure out the memory ordering.
2521 decorators |= mo_decorator_for_access_kind(kind);
2522
2523 if (!is_store && type == T_OBJECT) {
2524 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2525 if (tjp != NULL) {
2526 value_type = tjp;
2527 }
2528 }
2529
2530 receiver = null_check(receiver);
2531 if (stopped()) {
2532 return true;
2533 }
2534 // Heap pointers get a null-check from the interpreter,
2535 // as a courtesy. However, this is not guaranteed by Unsafe,
2536 // and it is not possible to fully distinguish unintended nulls
2537 // from intended ones in this API.
2538
2539 if (!is_store) {
2540 Node* p = NULL;
2541 // Try to constant fold a load from a constant field
2542 ciField* field = alias_type->field();
2543 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2544 // final or stable field
2545 p = make_constant_from_field(field, heap_base_oop);
2546 }
2547
2548 if (p == NULL) { // Could not constant fold the load
2549 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2550 // Normalize the value returned by getBoolean in the following cases
2551 if (type == T_BOOLEAN &&
2552 (mismatched ||
2553 heap_base_oop == top() || // - heap_base_oop is NULL or
2554 (can_access_non_heap && field == NULL)) // - heap_base_oop is potentially NULL
2555 // and the unsafe access is made to large offset
2556 // (i.e., larger than the maximum offset necessary for any
2557 // field access)
2558 ) {
2559 IdealKit ideal = IdealKit(this);
2560 #define __ ideal.
2561 IdealVariable normalized_result(ideal);
2562 __ declarations_done();
2563 __ set(normalized_result, p);
2564 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2565 __ set(normalized_result, ideal.ConI(1));
2566 ideal.end_if();
2567 final_sync(ideal);
2568 p = __ value(normalized_result);
2569 #undef __
2570 }
2571 }
2572 if (type == T_ADDRESS) {
2573 p = gvn().transform(new CastP2XNode(NULL, p));
2574 p = ConvX2UL(p);
2575 }
2576 // The load node has the control of the preceding MemBarCPUOrder. All
2577 // following nodes will have the control of the MemBarCPUOrder inserted at
2578 // the end of this method. So, pushing the load onto the stack at a later
2579 // point is fine.
2580 set_result(p);
2581 } else {
2582 if (bt == T_ADDRESS) {
2583 // Repackage the long as a pointer.
2584 val = ConvL2X(val);
2585 val = gvn().transform(new CastX2PNode(val));
2586 }
2587 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2588 }
2589
2590 return true;
2591 }
2592
2593 //----------------------------inline_unsafe_load_store----------------------------
2594 // This method serves a couple of different customers (depending on LoadStoreKind):
2595 //
2596 // LS_cmp_swap:
2597 //
2598 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2599 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2600 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2601 //
2602 // LS_cmp_swap_weak:
2603 //
2604 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2605 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2606 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2607 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2608 //
2609 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2610 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2611 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2612 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2613 //
2614 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2615 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2616 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2617 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2618 //
2619 // LS_cmp_exchange:
2620 //
2621 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2622 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2623 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2624 //
2625 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2626 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2627 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2628 //
2629 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2630 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2631 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2632 //
2633 // LS_get_add:
2634 //
2635 // int getAndAddInt( Object o, long offset, int delta)
2636 // long getAndAddLong(Object o, long offset, long delta)
2637 //
2638 // LS_get_set:
2639 //
2640 // int getAndSet(Object o, long offset, int newValue)
2641 // long getAndSet(Object o, long offset, long newValue)
2642 // Object getAndSet(Object o, long offset, Object newValue)
2643 //
2644 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2645 // This basic scheme here is the same as inline_unsafe_access, but
2646 // differs in enough details that combining them would make the code
2647 // overly confusing. (This is a true fact! I originally combined
2648 // them, but even I was confused by it!) As much code/comments as
2649 // possible are retained from inline_unsafe_access though to make
2650 // the correspondences clearer. - dl
2651
2652 if (callee()->is_static()) return false; // caller must have the capability!
2653
2654 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2655 decorators |= mo_decorator_for_access_kind(access_kind);
2656
2657 #ifndef PRODUCT
2658 BasicType rtype;
2659 {
2660 ResourceMark rm;
2661 // Check the signatures.
2662 ciSignature* sig = callee()->signature();
2663 rtype = sig->return_type()->basic_type();
2664 switch(kind) {
2665 case LS_get_add:
2666 case LS_get_set: {
2667 // Check the signatures.
2668 #ifdef ASSERT
2669 assert(rtype == type, "get and set must return the expected type");
2670 assert(sig->count() == 3, "get and set has 3 arguments");
2671 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2672 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2673 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2674 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2675 #endif // ASSERT
2676 break;
2677 }
2678 case LS_cmp_swap:
2679 case LS_cmp_swap_weak: {
2680 // Check the signatures.
2681 #ifdef ASSERT
2682 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2683 assert(sig->count() == 4, "CAS has 4 arguments");
2684 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2685 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2686 #endif // ASSERT
2687 break;
2688 }
2689 case LS_cmp_exchange: {
2690 // Check the signatures.
2691 #ifdef ASSERT
2692 assert(rtype == type, "CAS must return the expected type");
2693 assert(sig->count() == 4, "CAS has 4 arguments");
2694 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2695 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2696 #endif // ASSERT
2697 break;
2698 }
2699 default:
2700 ShouldNotReachHere();
2701 }
2702 }
2703 #endif //PRODUCT
2704
2705 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2706
2707 // Get arguments:
2708 Node* receiver = NULL;
2709 Node* base = NULL;
2710 Node* offset = NULL;
2711 Node* oldval = NULL;
2712 Node* newval = NULL;
2713 switch(kind) {
2714 case LS_cmp_swap:
2715 case LS_cmp_swap_weak:
2716 case LS_cmp_exchange: {
2717 const bool two_slot_type = type2size[type] == 2;
2718 receiver = argument(0); // type: oop
2719 base = argument(1); // type: oop
2720 offset = argument(2); // type: long
2721 oldval = argument(4); // type: oop, int, or long
2722 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2723 break;
2724 }
2725 case LS_get_add:
2726 case LS_get_set: {
2727 receiver = argument(0); // type: oop
2728 base = argument(1); // type: oop
2729 offset = argument(2); // type: long
2730 oldval = NULL;
2731 newval = argument(4); // type: oop, int, or long
2732 break;
2733 }
2734 default:
2735 ShouldNotReachHere();
2736 }
2737
2738 // Build field offset expression.
2739 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2740 // to be plain byte offsets, which are also the same as those accepted
2741 // by oopDesc::field_addr.
2742 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2743 // 32-bit machines ignore the high half of long offsets
2744 offset = ConvL2X(offset);
2745 Node* adr = make_unsafe_address(base, offset, ACCESS_WRITE | ACCESS_READ, type, false);
2746 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2747
2748 Compile::AliasType* alias_type = C->alias_type(adr_type);
2749 BasicType bt = alias_type->basic_type();
2750 if (bt != T_ILLEGAL &&
2751 (is_reference_type(bt) != (type == T_OBJECT))) {
2752 // Don't intrinsify mismatched object accesses.
2753 return false;
2754 }
2755
2756 // For CAS, unlike inline_unsafe_access, there seems no point in
2757 // trying to refine types. Just use the coarse types here.
2758 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2759 const Type *value_type = Type::get_const_basic_type(type);
2760
2761 switch (kind) {
2762 case LS_get_set:
2763 case LS_cmp_exchange: {
2764 if (type == T_OBJECT) {
2765 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2766 if (tjp != NULL) {
2767 value_type = tjp;
2768 }
2769 }
2770 break;
2771 }
2772 case LS_cmp_swap:
2773 case LS_cmp_swap_weak:
2774 case LS_get_add:
2775 break;
2776 default:
2777 ShouldNotReachHere();
2778 }
2779
2780 // Null check receiver.
2781 receiver = null_check(receiver);
2782 if (stopped()) {
2783 return true;
2784 }
2785
2786 int alias_idx = C->get_alias_index(adr_type);
2787
2788 if (is_reference_type(type)) {
2789 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2790
2791 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2792 // could be delayed during Parse (for example, in adjust_map_after_if()).
2793 // Execute transformation here to avoid barrier generation in such case.
2794 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2795 newval = _gvn.makecon(TypePtr::NULL_PTR);
2796
2797 if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2798 // Refine the value to a null constant, when it is known to be null
2799 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2800 }
2801 }
2802
2803 Node* result = NULL;
2804 switch (kind) {
2805 case LS_cmp_exchange: {
2806 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2807 oldval, newval, value_type, type, decorators);
2808 break;
2809 }
2810 case LS_cmp_swap_weak:
2811 decorators |= C2_WEAK_CMPXCHG;
2812 case LS_cmp_swap: {
2813 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2814 oldval, newval, value_type, type, decorators);
2815 break;
2816 }
2817 case LS_get_set: {
2818 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2819 newval, value_type, type, decorators);
2820 break;
2821 }
2822 case LS_get_add: {
2823 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2824 newval, value_type, type, decorators);
2825 break;
2826 }
2827 default:
2828 ShouldNotReachHere();
2829 }
2830
2831 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2832 set_result(result);
2833 return true;
2834 }
2835
2836 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2837 // Regardless of form, don't allow previous ld/st to move down,
2838 // then issue acquire, release, or volatile mem_bar.
2839 insert_mem_bar(Op_MemBarCPUOrder);
2840 switch(id) {
2841 case vmIntrinsics::_loadFence:
2842 insert_mem_bar(Op_LoadFence);
2843 return true;
2844 case vmIntrinsics::_storeFence:
2845 insert_mem_bar(Op_StoreFence);
2846 return true;
2847 case vmIntrinsics::_fullFence:
2848 insert_mem_bar(Op_MemBarVolatile);
2849 return true;
2850 default:
2851 fatal_unexpected_iid(id);
2852 return false;
2853 }
2854 }
2855
2856 bool LibraryCallKit::inline_onspinwait() {
2857 insert_mem_bar(Op_OnSpinWait);
2858 return true;
2859 }
2860
2861 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2862 if (!kls->is_Con()) {
2863 return true;
2864 }
2865 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
2866 if (klsptr == NULL) {
2867 return true;
2868 }
2869 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
2870 // don't need a guard for a klass that is already initialized
2871 return !ik->is_initialized();
2872 }
2873
2874 //----------------------------inline_unsafe_writeback0-------------------------
2875 // public native void Unsafe.writeback0(long address)
2876 bool LibraryCallKit::inline_unsafe_writeback0() {
2877 if (!Matcher::has_match_rule(Op_CacheWB)) {
2878 return false;
2879 }
2880 #ifndef PRODUCT
2881 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2882 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2883 ciSignature* sig = callee()->signature();
2884 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2885 #endif
2886 null_check_receiver(); // null-check, then ignore
2887 Node *addr = argument(1);
2888 addr = new CastX2PNode(addr);
2889 addr = _gvn.transform(addr);
2890 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2891 flush = _gvn.transform(flush);
2892 set_memory(flush, TypeRawPtr::BOTTOM);
2893 return true;
2894 }
2895
2896 //----------------------------inline_unsafe_writeback0-------------------------
2897 // public native void Unsafe.writeback0(long address)
2898 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2899 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2900 return false;
2901 }
2902 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2903 return false;
2904 }
2905 #ifndef PRODUCT
2906 assert(Matcher::has_match_rule(Op_CacheWB),
2907 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2908 : "found match rule for CacheWBPostSync but not CacheWB"));
2909
2910 #endif
2911 null_check_receiver(); // null-check, then ignore
2912 Node *sync;
2913 if (is_pre) {
2914 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2915 } else {
2916 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2917 }
2918 sync = _gvn.transform(sync);
2919 set_memory(sync, TypeRawPtr::BOTTOM);
2920 return true;
2921 }
2922
2923 //----------------------------inline_unsafe_allocate---------------------------
2924 // public native Object Unsafe.allocateInstance(Class<?> cls);
2925 bool LibraryCallKit::inline_unsafe_allocate() {
2926 if (callee()->is_static()) return false; // caller must have the capability!
2927
2928 null_check_receiver(); // null-check, then ignore
2929 Node* cls = null_check(argument(1));
2930 if (stopped()) return true;
2931
2932 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2933 kls = null_check(kls);
2934 if (stopped()) return true; // argument was like int.class
2935
2936 Node* test = NULL;
2937 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2938 // Note: The argument might still be an illegal value like
2939 // Serializable.class or Object[].class. The runtime will handle it.
2940 // But we must make an explicit check for initialization.
2941 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2942 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2943 // can generate code to load it as unsigned byte.
2944 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2945 Node* bits = intcon(InstanceKlass::fully_initialized);
2946 test = _gvn.transform(new SubINode(inst, bits));
2947 // The 'test' is non-zero if we need to take a slow path.
2948 }
2949
2950 Node* obj = new_instance(kls, test);
2951 set_result(obj);
2952 return true;
2953 }
2954
2955 //------------------------inline_native_time_funcs--------------
2956 // inline code for System.currentTimeMillis() and System.nanoTime()
2957 // these have the same type and signature
2958 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2959 const TypeFunc* tf = OptoRuntime::void_long_Type();
2960 const TypePtr* no_memory_effects = NULL;
2961 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2962 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2963 #ifdef ASSERT
2964 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2965 assert(value_top == top(), "second value must be top");
2966 #endif
2967 set_result(value);
2968 return true;
2969 }
2970
2971 #ifdef JFR_HAVE_INTRINSICS
2972
2973 /*
2974 * oop -> myklass
2975 * myklass->trace_id |= USED
2976 * return myklass->trace_id & ~0x3
2977 */
2978 bool LibraryCallKit::inline_native_classID() {
2979 Node* cls = null_check(argument(0), T_OBJECT);
2980 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2981 kls = null_check(kls, T_OBJECT);
2982
2983 ByteSize offset = KLASS_TRACE_ID_OFFSET;
2984 Node* insp = basic_plus_adr(kls, in_bytes(offset));
2985 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
2986
2987 Node* clsused = longcon(0x01l); // set the class bit
2988 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
2989 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
2990 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
2991
2992 #ifdef TRACE_ID_META_BITS
2993 Node* mbits = longcon(~TRACE_ID_META_BITS);
2994 tvalue = _gvn.transform(new AndLNode(tvalue, mbits));
2995 #endif
2996 #ifdef TRACE_ID_SHIFT
2997 Node* cbits = intcon(TRACE_ID_SHIFT);
2998 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits));
2999 #endif
3000
3001 set_result(tvalue);
3002 return true;
3003
3004 }
3005
3006 bool LibraryCallKit::inline_native_getEventWriter() {
3007 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3008
3009 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
3010 in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3011
3012 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3013
3014 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
3015 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
3016
3017 IfNode* iff_jobj_null =
3018 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
3019
3020 enum { _normal_path = 1,
3021 _null_path = 2,
3022 PATH_LIMIT };
3023
3024 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3025 PhiNode* result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3026
3027 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
3028 result_rgn->init_req(_null_path, jobj_is_null);
3029 result_val->init_req(_null_path, null());
3030
3031 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
3032 set_control(jobj_is_not_null);
3033 Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT,
3034 IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3035 result_rgn->init_req(_normal_path, control());
3036 result_val->init_req(_normal_path, res);
3037
3038 set_result(result_rgn, result_val);
3039
3040 return true;
3041 }
3042
3043 #endif // JFR_HAVE_INTRINSICS
3044
3045 //------------------------inline_native_currentThread------------------
3046 bool LibraryCallKit::inline_native_currentThread() {
3047 Node* junk = NULL;
3048 set_result(generate_current_thread(junk));
3049 return true;
3050 }
3051
3052 //---------------------------load_mirror_from_klass----------------------------
3053 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3054 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3055 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3056 Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3057 // mirror = ((OopHandle)mirror)->resolve();
3058 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3059 }
3060
3061 //-----------------------load_klass_from_mirror_common-------------------------
3062 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3063 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3064 // and branch to the given path on the region.
3065 // If never_see_null, take an uncommon trap on null, so we can optimistically
3066 // compile for the non-null case.
3067 // If the region is NULL, force never_see_null = true.
3068 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3069 bool never_see_null,
3070 RegionNode* region,
3071 int null_path,
3072 int offset) {
3073 if (region == NULL) never_see_null = true;
3074 Node* p = basic_plus_adr(mirror, offset);
3075 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3076 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3077 Node* null_ctl = top();
3078 kls = null_check_oop(kls, &null_ctl, never_see_null);
3079 if (region != NULL) {
3080 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3081 region->init_req(null_path, null_ctl);
3082 } else {
3083 assert(null_ctl == top(), "no loose ends");
3084 }
3085 return kls;
3086 }
3087
3088 //--------------------(inline_native_Class_query helpers)---------------------
3089 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3090 // Fall through if (mods & mask) == bits, take the guard otherwise.
3091 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3092 // Branch around if the given klass has the given modifier bit set.
3093 // Like generate_guard, adds a new path onto the region.
3094 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3095 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3096 Node* mask = intcon(modifier_mask);
3097 Node* bits = intcon(modifier_bits);
3098 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3099 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3100 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3101 return generate_fair_guard(bol, region);
3102 }
3103 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3104 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3105 }
3106
3107 //-------------------------inline_native_Class_query-------------------
3108 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3109 const Type* return_type = TypeInt::BOOL;
3110 Node* prim_return_value = top(); // what happens if it's a primitive class?
3111 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3112 bool expect_prim = false; // most of these guys expect to work on refs
3113
3114 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3115
3116 Node* mirror = argument(0);
3117 Node* obj = top();
3118
3119 switch (id) {
3120 case vmIntrinsics::_isInstance:
3121 // nothing is an instance of a primitive type
3122 prim_return_value = intcon(0);
3123 obj = argument(1);
3124 break;
3125 case vmIntrinsics::_getModifiers:
3126 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3127 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3128 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3129 break;
3130 case vmIntrinsics::_isInterface:
3131 prim_return_value = intcon(0);
3132 break;
3133 case vmIntrinsics::_isArray:
3134 prim_return_value = intcon(0);
3135 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3136 break;
3137 case vmIntrinsics::_isPrimitive:
3138 prim_return_value = intcon(1);
3139 expect_prim = true; // obviously
3140 break;
3141 case vmIntrinsics::_getSuperclass:
3142 prim_return_value = null();
3143 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3144 break;
3145 case vmIntrinsics::_getClassAccessFlags:
3146 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3147 return_type = TypeInt::INT; // not bool! 6297094
3148 break;
3149 default:
3150 fatal_unexpected_iid(id);
3151 break;
3152 }
3153
3154 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3155 if (mirror_con == NULL) return false; // cannot happen?
3156
3157 #ifndef PRODUCT
3158 if (C->print_intrinsics() || C->print_inlining()) {
3159 ciType* k = mirror_con->java_mirror_type();
3160 if (k) {
3161 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3162 k->print_name();
3163 tty->cr();
3164 }
3165 }
3166 #endif
3167
3168 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3169 RegionNode* region = new RegionNode(PATH_LIMIT);
3170 record_for_igvn(region);
3171 PhiNode* phi = new PhiNode(region, return_type);
3172
3173 // The mirror will never be null of Reflection.getClassAccessFlags, however
3174 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3175 // if it is. See bug 4774291.
3176
3177 // For Reflection.getClassAccessFlags(), the null check occurs in
3178 // the wrong place; see inline_unsafe_access(), above, for a similar
3179 // situation.
3180 mirror = null_check(mirror);
3181 // If mirror or obj is dead, only null-path is taken.
3182 if (stopped()) return true;
3183
3184 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3185
3186 // Now load the mirror's klass metaobject, and null-check it.
3187 // Side-effects region with the control path if the klass is null.
3188 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3189 // If kls is null, we have a primitive mirror.
3190 phi->init_req(_prim_path, prim_return_value);
3191 if (stopped()) { set_result(region, phi); return true; }
3192 bool safe_for_replace = (region->in(_prim_path) == top());
3193
3194 Node* p; // handy temp
3195 Node* null_ctl;
3196
3197 // Now that we have the non-null klass, we can perform the real query.
3198 // For constant classes, the query will constant-fold in LoadNode::Value.
3199 Node* query_value = top();
3200 switch (id) {
3201 case vmIntrinsics::_isInstance:
3202 // nothing is an instance of a primitive type
3203 query_value = gen_instanceof(obj, kls, safe_for_replace);
3204 break;
3205
3206 case vmIntrinsics::_getModifiers:
3207 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3208 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3209 break;
3210
3211 case vmIntrinsics::_isInterface:
3212 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3213 if (generate_interface_guard(kls, region) != NULL)
3214 // A guard was added. If the guard is taken, it was an interface.
3215 phi->add_req(intcon(1));
3216 // If we fall through, it's a plain class.
3217 query_value = intcon(0);
3218 break;
3219
3220 case vmIntrinsics::_isArray:
3221 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3222 if (generate_array_guard(kls, region) != NULL)
3223 // A guard was added. If the guard is taken, it was an array.
3224 phi->add_req(intcon(1));
3225 // If we fall through, it's a plain class.
3226 query_value = intcon(0);
3227 break;
3228
3229 case vmIntrinsics::_isPrimitive:
3230 query_value = intcon(0); // "normal" path produces false
3231 break;
3232
3233 case vmIntrinsics::_getSuperclass:
3234 // The rules here are somewhat unfortunate, but we can still do better
3235 // with random logic than with a JNI call.
3236 // Interfaces store null or Object as _super, but must report null.
3237 // Arrays store an intermediate super as _super, but must report Object.
3238 // Other types can report the actual _super.
3239 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3240 if (generate_interface_guard(kls, region) != NULL)
3241 // A guard was added. If the guard is taken, it was an interface.
3242 phi->add_req(null());
3243 if (generate_array_guard(kls, region) != NULL)
3244 // A guard was added. If the guard is taken, it was an array.
3245 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3246 // If we fall through, it's a plain class. Get its _super.
3247 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3248 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3249 null_ctl = top();
3250 kls = null_check_oop(kls, &null_ctl);
3251 if (null_ctl != top()) {
3252 // If the guard is taken, Object.superClass is null (both klass and mirror).
3253 region->add_req(null_ctl);
3254 phi ->add_req(null());
3255 }
3256 if (!stopped()) {
3257 query_value = load_mirror_from_klass(kls);
3258 }
3259 break;
3260
3261 case vmIntrinsics::_getClassAccessFlags:
3262 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3263 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3264 break;
3265
3266 default:
3267 fatal_unexpected_iid(id);
3268 break;
3269 }
3270
3271 // Fall-through is the normal case of a query to a real class.
3272 phi->init_req(1, query_value);
3273 region->init_req(1, control());
3274
3275 C->set_has_split_ifs(true); // Has chance for split-if optimization
3276 set_result(region, phi);
3277 return true;
3278 }
3279
3280 //-------------------------inline_Class_cast-------------------
3281 bool LibraryCallKit::inline_Class_cast() {
3282 Node* mirror = argument(0); // Class
3283 Node* obj = argument(1);
3284 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3285 if (mirror_con == NULL) {
3286 return false; // dead path (mirror->is_top()).
3287 }
3288 if (obj == NULL || obj->is_top()) {
3289 return false; // dead path
3290 }
3291 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3292
3293 // First, see if Class.cast() can be folded statically.
3294 // java_mirror_type() returns non-null for compile-time Class constants.
3295 ciType* tm = mirror_con->java_mirror_type();
3296 if (tm != NULL && tm->is_klass() &&
3297 tp != NULL && tp->klass() != NULL) {
3298 if (!tp->klass()->is_loaded()) {
3299 // Don't use intrinsic when class is not loaded.
3300 return false;
3301 } else {
3302 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3303 if (static_res == Compile::SSC_always_true) {
3304 // isInstance() is true - fold the code.
3305 set_result(obj);
3306 return true;
3307 } else if (static_res == Compile::SSC_always_false) {
3308 // Don't use intrinsic, have to throw ClassCastException.
3309 // If the reference is null, the non-intrinsic bytecode will
3310 // be optimized appropriately.
3311 return false;
3312 }
3313 }
3314 }
3315
3316 // Bailout intrinsic and do normal inlining if exception path is frequent.
3317 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3318 return false;
3319 }
3320
3321 // Generate dynamic checks.
3322 // Class.cast() is java implementation of _checkcast bytecode.
3323 // Do checkcast (Parse::do_checkcast()) optimizations here.
3324
3325 mirror = null_check(mirror);
3326 // If mirror is dead, only null-path is taken.
3327 if (stopped()) {
3328 return true;
3329 }
3330
3331 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3332 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3333 RegionNode* region = new RegionNode(PATH_LIMIT);
3334 record_for_igvn(region);
3335
3336 // Now load the mirror's klass metaobject, and null-check it.
3337 // If kls is null, we have a primitive mirror and
3338 // nothing is an instance of a primitive type.
3339 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3340
3341 Node* res = top();
3342 if (!stopped()) {
3343 Node* bad_type_ctrl = top();
3344 // Do checkcast optimizations.
3345 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3346 region->init_req(_bad_type_path, bad_type_ctrl);
3347 }
3348 if (region->in(_prim_path) != top() ||
3349 region->in(_bad_type_path) != top()) {
3350 // Let Interpreter throw ClassCastException.
3351 PreserveJVMState pjvms(this);
3352 set_control(_gvn.transform(region));
3353 uncommon_trap(Deoptimization::Reason_intrinsic,
3354 Deoptimization::Action_maybe_recompile);
3355 }
3356 if (!stopped()) {
3357 set_result(res);
3358 }
3359 return true;
3360 }
3361
3362
3363 //--------------------------inline_native_subtype_check------------------------
3364 // This intrinsic takes the JNI calls out of the heart of
3365 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3366 bool LibraryCallKit::inline_native_subtype_check() {
3367 // Pull both arguments off the stack.
3368 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3369 args[0] = argument(0);
3370 args[1] = argument(1);
3371 Node* klasses[2]; // corresponding Klasses: superk, subk
3372 klasses[0] = klasses[1] = top();
3373
3374 enum {
3375 // A full decision tree on {superc is prim, subc is prim}:
3376 _prim_0_path = 1, // {P,N} => false
3377 // {P,P} & superc!=subc => false
3378 _prim_same_path, // {P,P} & superc==subc => true
3379 _prim_1_path, // {N,P} => false
3380 _ref_subtype_path, // {N,N} & subtype check wins => true
3381 _both_ref_path, // {N,N} & subtype check loses => false
3382 PATH_LIMIT
3383 };
3384
3385 RegionNode* region = new RegionNode(PATH_LIMIT);
3386 Node* phi = new PhiNode(region, TypeInt::BOOL);
3387 record_for_igvn(region);
3388
3389 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3390 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3391 int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3392
3393 // First null-check both mirrors and load each mirror's klass metaobject.
3394 int which_arg;
3395 for (which_arg = 0; which_arg <= 1; which_arg++) {
3396 Node* arg = args[which_arg];
3397 arg = null_check(arg);
3398 if (stopped()) break;
3399 args[which_arg] = arg;
3400
3401 Node* p = basic_plus_adr(arg, class_klass_offset);
3402 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3403 klasses[which_arg] = _gvn.transform(kls);
3404 }
3405
3406 // Resolve oops to stable for CmpP below.
3407 args[0] = access_resolve(args[0], 0);
3408 args[1] = access_resolve(args[1], 0);
3409
3410 // Having loaded both klasses, test each for null.
3411 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3412 for (which_arg = 0; which_arg <= 1; which_arg++) {
3413 Node* kls = klasses[which_arg];
3414 Node* null_ctl = top();
3415 kls = null_check_oop(kls, &null_ctl, never_see_null);
3416 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3417 region->init_req(prim_path, null_ctl);
3418 if (stopped()) break;
3419 klasses[which_arg] = kls;
3420 }
3421
3422 if (!stopped()) {
3423 // now we have two reference types, in klasses[0..1]
3424 Node* subk = klasses[1]; // the argument to isAssignableFrom
3425 Node* superk = klasses[0]; // the receiver
3426 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3427 // now we have a successful reference subtype check
3428 region->set_req(_ref_subtype_path, control());
3429 }
3430
3431 // If both operands are primitive (both klasses null), then
3432 // we must return true when they are identical primitives.
3433 // It is convenient to test this after the first null klass check.
3434 set_control(region->in(_prim_0_path)); // go back to first null check
3435 if (!stopped()) {
3436 // Since superc is primitive, make a guard for the superc==subc case.
3437 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3438 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3439 generate_guard(bol_eq, region, PROB_FAIR);
3440 if (region->req() == PATH_LIMIT+1) {
3441 // A guard was added. If the added guard is taken, superc==subc.
3442 region->swap_edges(PATH_LIMIT, _prim_same_path);
3443 region->del_req(PATH_LIMIT);
3444 }
3445 region->set_req(_prim_0_path, control()); // Not equal after all.
3446 }
3447
3448 // these are the only paths that produce 'true':
3449 phi->set_req(_prim_same_path, intcon(1));
3450 phi->set_req(_ref_subtype_path, intcon(1));
3451
3452 // pull together the cases:
3453 assert(region->req() == PATH_LIMIT, "sane region");
3454 for (uint i = 1; i < region->req(); i++) {
3455 Node* ctl = region->in(i);
3456 if (ctl == NULL || ctl == top()) {
3457 region->set_req(i, top());
3458 phi ->set_req(i, top());
3459 } else if (phi->in(i) == NULL) {
3460 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3461 }
3462 }
3463
3464 set_control(_gvn.transform(region));
3465 set_result(_gvn.transform(phi));
3466 return true;
3467 }
3468
3469 //---------------------generate_array_guard_common------------------------
3470 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3471 bool obj_array, bool not_array) {
3472
3473 if (stopped()) {
3474 return NULL;
3475 }
3476
3477 // If obj_array/non_array==false/false:
3478 // Branch around if the given klass is in fact an array (either obj or prim).
3479 // If obj_array/non_array==false/true:
3480 // Branch around if the given klass is not an array klass of any kind.
3481 // If obj_array/non_array==true/true:
3482 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3483 // If obj_array/non_array==true/false:
3484 // Branch around if the kls is an oop array (Object[] or subtype)
3485 //
3486 // Like generate_guard, adds a new path onto the region.
3487 jint layout_con = 0;
3488 Node* layout_val = get_layout_helper(kls, layout_con);
3489 if (layout_val == NULL) {
3490 bool query = (obj_array
3491 ? Klass::layout_helper_is_objArray(layout_con)
3492 : Klass::layout_helper_is_array(layout_con));
3493 if (query == not_array) {
3494 return NULL; // never a branch
3495 } else { // always a branch
3496 Node* always_branch = control();
3497 if (region != NULL)
3498 region->add_req(always_branch);
3499 set_control(top());
3500 return always_branch;
3501 }
3502 }
3503 // Now test the correct condition.
3504 jint nval = (obj_array
3505 ? (jint)(Klass::_lh_array_tag_type_value
3506 << Klass::_lh_array_tag_shift)
3507 : Klass::_lh_neutral_value);
3508 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3509 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3510 // invert the test if we are looking for a non-array
3511 if (not_array) btest = BoolTest(btest).negate();
3512 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3513 return generate_fair_guard(bol, region);
3514 }
3515
3516
3517 //-----------------------inline_native_newArray--------------------------
3518 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3519 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3520 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3521 Node* mirror;
3522 Node* count_val;
3523 if (uninitialized) {
3524 mirror = argument(1);
3525 count_val = argument(2);
3526 } else {
3527 mirror = argument(0);
3528 count_val = argument(1);
3529 }
3530
3531 mirror = null_check(mirror);
3532 // If mirror or obj is dead, only null-path is taken.
3533 if (stopped()) return true;
3534
3535 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3536 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3537 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3538 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3539 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3540
3541 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3542 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3543 result_reg, _slow_path);
3544 Node* normal_ctl = control();
3545 Node* no_array_ctl = result_reg->in(_slow_path);
3546
3547 // Generate code for the slow case. We make a call to newArray().
3548 set_control(no_array_ctl);
3549 if (!stopped()) {
3550 // Either the input type is void.class, or else the
3551 // array klass has not yet been cached. Either the
3552 // ensuing call will throw an exception, or else it
3553 // will cache the array klass for next time.
3554 PreserveJVMState pjvms(this);
3555 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3556 Node* slow_result = set_results_for_java_call(slow_call);
3557 // this->control() comes from set_results_for_java_call
3558 result_reg->set_req(_slow_path, control());
3559 result_val->set_req(_slow_path, slow_result);
3560 result_io ->set_req(_slow_path, i_o());
3561 result_mem->set_req(_slow_path, reset_memory());
3562 }
3563
3564 set_control(normal_ctl);
3565 if (!stopped()) {
3566 // Normal case: The array type has been cached in the java.lang.Class.
3567 // The following call works fine even if the array type is polymorphic.
3568 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3569 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3570 result_reg->init_req(_normal_path, control());
3571 result_val->init_req(_normal_path, obj);
3572 result_io ->init_req(_normal_path, i_o());
3573 result_mem->init_req(_normal_path, reset_memory());
3574
3575 if (uninitialized) {
3576 // Mark the allocation so that zeroing is skipped
3577 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3578 alloc->maybe_set_complete(&_gvn);
3579 }
3580 }
3581
3582 // Return the combined state.
3583 set_i_o( _gvn.transform(result_io) );
3584 set_all_memory( _gvn.transform(result_mem));
3585
3586 C->set_has_split_ifs(true); // Has chance for split-if optimization
3587 set_result(result_reg, result_val);
3588 return true;
3589 }
3590
3591 //----------------------inline_native_getLength--------------------------
3592 // public static native int java.lang.reflect.Array.getLength(Object array);
3593 bool LibraryCallKit::inline_native_getLength() {
3594 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3595
3596 Node* array = null_check(argument(0));
3597 // If array is dead, only null-path is taken.
3598 if (stopped()) return true;
3599
3600 // Deoptimize if it is a non-array.
3601 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3602
3603 if (non_array != NULL) {
3604 PreserveJVMState pjvms(this);
3605 set_control(non_array);
3606 uncommon_trap(Deoptimization::Reason_intrinsic,
3607 Deoptimization::Action_maybe_recompile);
3608 }
3609
3610 // If control is dead, only non-array-path is taken.
3611 if (stopped()) return true;
3612
3613 // The works fine even if the array type is polymorphic.
3614 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3615 Node* result = load_array_length(array);
3616
3617 C->set_has_split_ifs(true); // Has chance for split-if optimization
3618 set_result(result);
3619 return true;
3620 }
3621
3622 //------------------------inline_array_copyOf----------------------------
3623 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3624 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3625 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3626 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3627
3628 // Get the arguments.
3629 Node* original = argument(0);
3630 Node* start = is_copyOfRange? argument(1): intcon(0);
3631 Node* end = is_copyOfRange? argument(2): argument(1);
3632 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3633
3634 Node* newcopy = NULL;
3635
3636 // Set the original stack and the reexecute bit for the interpreter to reexecute
3637 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3638 { PreserveReexecuteState preexecs(this);
3639 jvms()->set_should_reexecute(true);
3640
3641 array_type_mirror = null_check(array_type_mirror);
3642 original = null_check(original);
3643
3644 // Check if a null path was taken unconditionally.
3645 if (stopped()) return true;
3646
3647 Node* orig_length = load_array_length(original);
3648
3649 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3650 klass_node = null_check(klass_node);
3651
3652 RegionNode* bailout = new RegionNode(1);
3653 record_for_igvn(bailout);
3654
3655 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3656 // Bail out if that is so.
3657 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3658 if (not_objArray != NULL) {
3659 // Improve the klass node's type from the new optimistic assumption:
3660 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3661 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3662 Node* cast = new CastPPNode(klass_node, akls);
3663 cast->init_req(0, control());
3664 klass_node = _gvn.transform(cast);
3665 }
3666
3667 // Bail out if either start or end is negative.
3668 generate_negative_guard(start, bailout, &start);
3669 generate_negative_guard(end, bailout, &end);
3670
3671 Node* length = end;
3672 if (_gvn.type(start) != TypeInt::ZERO) {
3673 length = _gvn.transform(new SubINode(end, start));
3674 }
3675
3676 // Bail out if length is negative.
3677 // Without this the new_array would throw
3678 // NegativeArraySizeException but IllegalArgumentException is what
3679 // should be thrown
3680 generate_negative_guard(length, bailout, &length);
3681
3682 if (bailout->req() > 1) {
3683 PreserveJVMState pjvms(this);
3684 set_control(_gvn.transform(bailout));
3685 uncommon_trap(Deoptimization::Reason_intrinsic,
3686 Deoptimization::Action_maybe_recompile);
3687 }
3688
3689 if (!stopped()) {
3690 // How many elements will we copy from the original?
3691 // The answer is MinI(orig_length - start, length).
3692 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3693 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3694
3695 original = access_resolve(original, ACCESS_READ);
3696
3697 // Generate a direct call to the right arraycopy function(s).
3698 // We know the copy is disjoint but we might not know if the
3699 // oop stores need checking.
3700 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
3701 // This will fail a store-check if x contains any non-nulls.
3702
3703 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3704 // loads/stores but it is legal only if we're sure the
3705 // Arrays.copyOf would succeed. So we need all input arguments
3706 // to the copyOf to be validated, including that the copy to the
3707 // new array won't trigger an ArrayStoreException. That subtype
3708 // check can be optimized if we know something on the type of
3709 // the input array from type speculation.
3710 if (_gvn.type(klass_node)->singleton()) {
3711 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3712 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3713
3714 int test = C->static_subtype_check(superk, subk);
3715 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3716 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3717 if (t_original->speculative_type() != NULL) {
3718 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3719 }
3720 }
3721 }
3722
3723 bool validated = false;
3724 // Reason_class_check rather than Reason_intrinsic because we
3725 // want to intrinsify even if this traps.
3726 if (!too_many_traps(Deoptimization::Reason_class_check)) {
3727 Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
3728 klass_node);
3729
3730 if (not_subtype_ctrl != top()) {
3731 PreserveJVMState pjvms(this);
3732 set_control(not_subtype_ctrl);
3733 uncommon_trap(Deoptimization::Reason_class_check,
3734 Deoptimization::Action_make_not_entrant);
3735 assert(stopped(), "Should be stopped");
3736 }
3737 validated = true;
3738 }
3739
3740 if (!stopped()) {
3741 newcopy = new_array(klass_node, length, 0); // no arguments to push
3742
3743 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
3744 load_object_klass(original), klass_node);
3745 if (!is_copyOfRange) {
3746 ac->set_copyof(validated);
3747 } else {
3748 ac->set_copyofrange(validated);
3749 }
3750 Node* n = _gvn.transform(ac);
3751 if (n == ac) {
3752 ac->connect_outputs(this);
3753 } else {
3754 assert(validated, "shouldn't transform if all arguments not validated");
3755 set_all_memory(n);
3756 }
3757 }
3758 }
3759 } // original reexecute is set back here
3760
3761 C->set_has_split_ifs(true); // Has chance for split-if optimization
3762 if (!stopped()) {
3763 set_result(newcopy);
3764 }
3765 return true;
3766 }
3767
3768
3769 //----------------------generate_virtual_guard---------------------------
3770 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
3771 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
3772 RegionNode* slow_region) {
3773 ciMethod* method = callee();
3774 int vtable_index = method->vtable_index();
3775 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3776 "bad index %d", vtable_index);
3777 // Get the Method* out of the appropriate vtable entry.
3778 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
3779 vtable_index*vtableEntry::size_in_bytes() +
3780 vtableEntry::method_offset_in_bytes();
3781 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
3782 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3783
3784 // Compare the target method with the expected method (e.g., Object.hashCode).
3785 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
3786
3787 Node* native_call = makecon(native_call_addr);
3788 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
3789 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
3790
3791 return generate_slow_guard(test_native, slow_region);
3792 }
3793
3794 //-----------------------generate_method_call----------------------------
3795 // Use generate_method_call to make a slow-call to the real
3796 // method if the fast path fails. An alternative would be to
3797 // use a stub like OptoRuntime::slow_arraycopy_Java.
3798 // This only works for expanding the current library call,
3799 // not another intrinsic. (E.g., don't use this for making an
3800 // arraycopy call inside of the copyOf intrinsic.)
3801 CallJavaNode*
3802 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
3803 // When compiling the intrinsic method itself, do not use this technique.
3804 guarantee(callee() != C->method(), "cannot make slow-call to self");
3805
3806 ciMethod* method = callee();
3807 // ensure the JVMS we have will be correct for this call
3808 guarantee(method_id == method->intrinsic_id(), "must match");
3809
3810 const TypeFunc* tf = TypeFunc::make(method);
3811 CallJavaNode* slow_call;
3812 if (is_static) {
3813 assert(!is_virtual, "");
3814 slow_call = new CallStaticJavaNode(C, tf,
3815 SharedRuntime::get_resolve_static_call_stub(),
3816 method, bci());
3817 } else if (is_virtual) {
3818 null_check_receiver();
3819 int vtable_index = Method::invalid_vtable_index;
3820 if (UseInlineCaches) {
3821 // Suppress the vtable call
3822 } else {
3823 // hashCode and clone are not a miranda methods,
3824 // so the vtable index is fixed.
3825 // No need to use the linkResolver to get it.
3826 vtable_index = method->vtable_index();
3827 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3828 "bad index %d", vtable_index);
3829 }
3830 slow_call = new CallDynamicJavaNode(tf,
3831 SharedRuntime::get_resolve_virtual_call_stub(),
3832 method, vtable_index, bci());
3833 } else { // neither virtual nor static: opt_virtual
3834 null_check_receiver();
3835 slow_call = new CallStaticJavaNode(C, tf,
3836 SharedRuntime::get_resolve_opt_virtual_call_stub(),
3837 method, bci());
3838 slow_call->set_optimized_virtual(true);
3839 }
3840 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
3841 // To be able to issue a direct call (optimized virtual or virtual)
3842 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
3843 // about the method being invoked should be attached to the call site to
3844 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
3845 slow_call->set_override_symbolic_info(true);
3846 }
3847 set_arguments_for_java_call(slow_call);
3848 set_edges_for_java_call(slow_call);
3849 return slow_call;
3850 }
3851
3852
3853 /**
3854 * Build special case code for calls to hashCode on an object. This call may
3855 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
3856 * slightly different code.
3857 */
3858 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
3859 assert(is_static == callee()->is_static(), "correct intrinsic selection");
3860 assert(!(is_virtual && is_static), "either virtual, special, or static");
3861
3862 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
3863
3864 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3865 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
3866 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3867 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3868 Node* obj = NULL;
3869 if (!is_static) {
3870 // Check for hashing null object
3871 obj = null_check_receiver();
3872 if (stopped()) return true; // unconditionally null
3873 result_reg->init_req(_null_path, top());
3874 result_val->init_req(_null_path, top());
3875 } else {
3876 // Do a null check, and return zero if null.
3877 // System.identityHashCode(null) == 0
3878 obj = argument(0);
3879 Node* null_ctl = top();
3880 obj = null_check_oop(obj, &null_ctl);
3881 result_reg->init_req(_null_path, null_ctl);
3882 result_val->init_req(_null_path, _gvn.intcon(0));
3883 }
3884
3885 // Unconditionally null? Then return right away.
3886 if (stopped()) {
3887 set_control( result_reg->in(_null_path));
3888 if (!stopped())
3889 set_result(result_val->in(_null_path));
3890 return true;
3891 }
3892
3893 // We only go to the fast case code if we pass a number of guards. The
3894 // paths which do not pass are accumulated in the slow_region.
3895 RegionNode* slow_region = new RegionNode(1);
3896 record_for_igvn(slow_region);
3897
3898 // If this is a virtual call, we generate a funny guard. We pull out
3899 // the vtable entry corresponding to hashCode() from the target object.
3900 // If the target method which we are calling happens to be the native
3901 // Object hashCode() method, we pass the guard. We do not need this
3902 // guard for non-virtual calls -- the caller is known to be the native
3903 // Object hashCode().
3904 if (is_virtual) {
3905 // After null check, get the object's klass.
3906 Node* obj_klass = load_object_klass(obj);
3907 generate_virtual_guard(obj_klass, slow_region);
3908 }
3909
3910 // Get the header out of the object, use LoadMarkNode when available
3911 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
3912 // The control of the load must be NULL. Otherwise, the load can move before
3913 // the null check after castPP removal.
3914 Node* no_ctrl = NULL;
3915 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
3916
3917 // Test the header to see if it is unlocked.
3918 Node *lock_mask = _gvn.MakeConX(markWord::biased_lock_mask_in_place);
3919 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
3920 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
3921 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
3922 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
3923
3924 generate_slow_guard(test_unlocked, slow_region);
3925
3926 // Get the hash value and check to see that it has been properly assigned.
3927 // We depend on hash_mask being at most 32 bits and avoid the use of
3928 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
3929 // vm: see markWord.hpp.
3930 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
3931 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
3932 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
3933 // This hack lets the hash bits live anywhere in the mark object now, as long
3934 // as the shift drops the relevant bits into the low 32 bits. Note that
3935 // Java spec says that HashCode is an int so there's no point in capturing
3936 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
3937 hshifted_header = ConvX2I(hshifted_header);
3938 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
3939
3940 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
3941 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
3942 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
3943
3944 generate_slow_guard(test_assigned, slow_region);
3945
3946 Node* init_mem = reset_memory();
3947 // fill in the rest of the null path:
3948 result_io ->init_req(_null_path, i_o());
3949 result_mem->init_req(_null_path, init_mem);
3950
3951 result_val->init_req(_fast_path, hash_val);
3952 result_reg->init_req(_fast_path, control());
3953 result_io ->init_req(_fast_path, i_o());
3954 result_mem->init_req(_fast_path, init_mem);
3955
3956 // Generate code for the slow case. We make a call to hashCode().
3957 set_control(_gvn.transform(slow_region));
3958 if (!stopped()) {
3959 // No need for PreserveJVMState, because we're using up the present state.
3960 set_all_memory(init_mem);
3961 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
3962 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
3963 Node* slow_result = set_results_for_java_call(slow_call);
3964 // this->control() comes from set_results_for_java_call
3965 result_reg->init_req(_slow_path, control());
3966 result_val->init_req(_slow_path, slow_result);
3967 result_io ->set_req(_slow_path, i_o());
3968 result_mem ->set_req(_slow_path, reset_memory());
3969 }
3970
3971 // Return the combined state.
3972 set_i_o( _gvn.transform(result_io) );
3973 set_all_memory( _gvn.transform(result_mem));
3974
3975 set_result(result_reg, result_val);
3976 return true;
3977 }
3978
3979 //---------------------------inline_native_getClass----------------------------
3980 // public final native Class<?> java.lang.Object.getClass();
3981 //
3982 // Build special case code for calls to getClass on an object.
3983 bool LibraryCallKit::inline_native_getClass() {
3984 Node* obj = null_check_receiver();
3985 if (stopped()) return true;
3986 set_result(load_mirror_from_klass(load_object_klass(obj)));
3987 return true;
3988 }
3989
3990 //-----------------inline_native_Reflection_getCallerClass---------------------
3991 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
3992 //
3993 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
3994 //
3995 // NOTE: This code must perform the same logic as JVM_GetCallerClass
3996 // in that it must skip particular security frames and checks for
3997 // caller sensitive methods.
3998 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
3999 #ifndef PRODUCT
4000 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4001 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4002 }
4003 #endif
4004
4005 if (!jvms()->has_method()) {
4006 #ifndef PRODUCT
4007 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4008 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4009 }
4010 #endif
4011 return false;
4012 }
4013
4014 // Walk back up the JVM state to find the caller at the required
4015 // depth.
4016 JVMState* caller_jvms = jvms();
4017
4018 // Cf. JVM_GetCallerClass
4019 // NOTE: Start the loop at depth 1 because the current JVM state does
4020 // not include the Reflection.getCallerClass() frame.
4021 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4022 ciMethod* m = caller_jvms->method();
4023 switch (n) {
4024 case 0:
4025 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4026 break;
4027 case 1:
4028 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4029 if (!m->caller_sensitive()) {
4030 #ifndef PRODUCT
4031 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4032 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4033 }
4034 #endif
4035 return false; // bail-out; let JVM_GetCallerClass do the work
4036 }
4037 break;
4038 default:
4039 if (!m->is_ignored_by_security_stack_walk()) {
4040 // We have reached the desired frame; return the holder class.
4041 // Acquire method holder as java.lang.Class and push as constant.
4042 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4043 ciInstance* caller_mirror = caller_klass->java_mirror();
4044 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4045
4046 #ifndef PRODUCT
4047 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4048 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());
4049 tty->print_cr(" JVM state at this point:");
4050 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4051 ciMethod* m = jvms()->of_depth(i)->method();
4052 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4053 }
4054 }
4055 #endif
4056 return true;
4057 }
4058 break;
4059 }
4060 }
4061
4062 #ifndef PRODUCT
4063 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4064 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4065 tty->print_cr(" JVM state at this point:");
4066 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4067 ciMethod* m = jvms()->of_depth(i)->method();
4068 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4069 }
4070 }
4071 #endif
4072
4073 return false; // bail-out; let JVM_GetCallerClass do the work
4074 }
4075
4076 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4077 Node* arg = argument(0);
4078 Node* result = NULL;
4079
4080 switch (id) {
4081 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4082 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4083 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4084 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4085
4086 case vmIntrinsics::_doubleToLongBits: {
4087 // two paths (plus control) merge in a wood
4088 RegionNode *r = new RegionNode(3);
4089 Node *phi = new PhiNode(r, TypeLong::LONG);
4090
4091 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4092 // Build the boolean node
4093 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4094
4095 // Branch either way.
4096 // NaN case is less traveled, which makes all the difference.
4097 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4098 Node *opt_isnan = _gvn.transform(ifisnan);
4099 assert( opt_isnan->is_If(), "Expect an IfNode");
4100 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4101 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4102
4103 set_control(iftrue);
4104
4105 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4106 Node *slow_result = longcon(nan_bits); // return NaN
4107 phi->init_req(1, _gvn.transform( slow_result ));
4108 r->init_req(1, iftrue);
4109
4110 // Else fall through
4111 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4112 set_control(iffalse);
4113
4114 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4115 r->init_req(2, iffalse);
4116
4117 // Post merge
4118 set_control(_gvn.transform(r));
4119 record_for_igvn(r);
4120
4121 C->set_has_split_ifs(true); // Has chance for split-if optimization
4122 result = phi;
4123 assert(result->bottom_type()->isa_long(), "must be");
4124 break;
4125 }
4126
4127 case vmIntrinsics::_floatToIntBits: {
4128 // two paths (plus control) merge in a wood
4129 RegionNode *r = new RegionNode(3);
4130 Node *phi = new PhiNode(r, TypeInt::INT);
4131
4132 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4133 // Build the boolean node
4134 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4135
4136 // Branch either way.
4137 // NaN case is less traveled, which makes all the difference.
4138 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4139 Node *opt_isnan = _gvn.transform(ifisnan);
4140 assert( opt_isnan->is_If(), "Expect an IfNode");
4141 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4142 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4143
4144 set_control(iftrue);
4145
4146 static const jint nan_bits = 0x7fc00000;
4147 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4148 phi->init_req(1, _gvn.transform( slow_result ));
4149 r->init_req(1, iftrue);
4150
4151 // Else fall through
4152 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4153 set_control(iffalse);
4154
4155 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4156 r->init_req(2, iffalse);
4157
4158 // Post merge
4159 set_control(_gvn.transform(r));
4160 record_for_igvn(r);
4161
4162 C->set_has_split_ifs(true); // Has chance for split-if optimization
4163 result = phi;
4164 assert(result->bottom_type()->isa_int(), "must be");
4165 break;
4166 }
4167
4168 default:
4169 fatal_unexpected_iid(id);
4170 break;
4171 }
4172 set_result(_gvn.transform(result));
4173 return true;
4174 }
4175
4176 //----------------------inline_unsafe_copyMemory-------------------------
4177 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4178 bool LibraryCallKit::inline_unsafe_copyMemory() {
4179 if (callee()->is_static()) return false; // caller must have the capability!
4180 null_check_receiver(); // null-check receiver
4181 if (stopped()) return true;
4182
4183 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4184
4185 Node* src_ptr = argument(1); // type: oop
4186 Node* src_off = ConvL2X(argument(2)); // type: long
4187 Node* dst_ptr = argument(4); // type: oop
4188 Node* dst_off = ConvL2X(argument(5)); // type: long
4189 Node* size = ConvL2X(argument(7)); // type: long
4190
4191 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4192 "fieldOffset must be byte-scaled");
4193
4194 src_ptr = access_resolve(src_ptr, ACCESS_READ);
4195 dst_ptr = access_resolve(dst_ptr, ACCESS_WRITE);
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); // no arguments to push
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 obj = access_resolve(obj, ACCESS_READ);
4351 // Generate a direct call to the right arraycopy function(s).
4352 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4353 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false);
4354 ac->set_clone_oop_array();
4355 Node* n = _gvn.transform(ac);
4356 assert(n == ac, "cannot disappear");
4357 ac->connect_outputs(this);
4358
4359 result_reg->init_req(_objArray_path, control());
4360 result_val->init_req(_objArray_path, alloc_obj);
4361 result_i_o ->set_req(_objArray_path, i_o());
4362 result_mem ->set_req(_objArray_path, reset_memory());
4363 }
4364 }
4365 // Otherwise, there are no barriers to worry about.
4366 // (We can dispense with card marks if we know the allocation
4367 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4368 // causes the non-eden paths to take compensating steps to
4369 // simulate a fresh allocation, so that no further
4370 // card marks are required in compiled code to initialize
4371 // the object.)
4372
4373 if (!stopped()) {
4374 copy_to_clone(obj, alloc_obj, obj_size, true);
4375
4376 // Present the results of the copy.
4377 result_reg->init_req(_array_path, control());
4378 result_val->init_req(_array_path, alloc_obj);
4379 result_i_o ->set_req(_array_path, i_o());
4380 result_mem ->set_req(_array_path, reset_memory());
4381 }
4382 }
4383
4384 // We only go to the instance fast case code if we pass a number of guards.
4385 // The paths which do not pass are accumulated in the slow_region.
4386 RegionNode* slow_region = new RegionNode(1);
4387 record_for_igvn(slow_region);
4388 if (!stopped()) {
4389 // It's an instance (we did array above). Make the slow-path tests.
4390 // If this is a virtual call, we generate a funny guard. We grab
4391 // the vtable entry corresponding to clone() from the target object.
4392 // If the target method which we are calling happens to be the
4393 // Object clone() method, we pass the guard. We do not need this
4394 // guard for non-virtual calls; the caller is known to be the native
4395 // Object clone().
4396 if (is_virtual) {
4397 generate_virtual_guard(obj_klass, slow_region);
4398 }
4399
4400 // The object must be easily cloneable and must not have a finalizer.
4401 // Both of these conditions may be checked in a single test.
4402 // We could optimize the test further, but we don't care.
4403 generate_access_flags_guard(obj_klass,
4404 // Test both conditions:
4405 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4406 // Must be cloneable but not finalizer:
4407 JVM_ACC_IS_CLONEABLE_FAST,
4408 slow_region);
4409 }
4410
4411 if (!stopped()) {
4412 // It's an instance, and it passed the slow-path tests.
4413 PreserveJVMState pjvms(this);
4414 Node* obj_size = NULL;
4415 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4416 // is reexecuted if deoptimization occurs and there could be problems when merging
4417 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4418 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4419
4420 copy_to_clone(obj, alloc_obj, obj_size, false);
4421
4422 // Present the results of the slow call.
4423 result_reg->init_req(_instance_path, control());
4424 result_val->init_req(_instance_path, alloc_obj);
4425 result_i_o ->set_req(_instance_path, i_o());
4426 result_mem ->set_req(_instance_path, reset_memory());
4427 }
4428
4429 // Generate code for the slow case. We make a call to clone().
4430 set_control(_gvn.transform(slow_region));
4431 if (!stopped()) {
4432 PreserveJVMState pjvms(this);
4433 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4434 // We need to deoptimize on exception (see comment above)
4435 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4436 // this->control() comes from set_results_for_java_call
4437 result_reg->init_req(_slow_path, control());
4438 result_val->init_req(_slow_path, slow_result);
4439 result_i_o ->set_req(_slow_path, i_o());
4440 result_mem ->set_req(_slow_path, reset_memory());
4441 }
4442
4443 // Return the combined state.
4444 set_control( _gvn.transform(result_reg));
4445 set_i_o( _gvn.transform(result_i_o));
4446 set_all_memory( _gvn.transform(result_mem));
4447 } // original reexecute is set back here
4448
4449 set_result(_gvn.transform(result_val));
4450 return true;
4451 }
4452
4453 // If we have a tightly coupled allocation, the arraycopy may take care
4454 // of the array initialization. If one of the guards we insert between
4455 // the allocation and the arraycopy causes a deoptimization, an
4456 // unitialized array will escape the compiled method. To prevent that
4457 // we set the JVM state for uncommon traps between the allocation and
4458 // the arraycopy to the state before the allocation so, in case of
4459 // deoptimization, we'll reexecute the allocation and the
4460 // initialization.
4461 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4462 if (alloc != NULL) {
4463 ciMethod* trap_method = alloc->jvms()->method();
4464 int trap_bci = alloc->jvms()->bci();
4465
4466 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4467 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4468 // Make sure there's no store between the allocation and the
4469 // arraycopy otherwise visible side effects could be rexecuted
4470 // in case of deoptimization and cause incorrect execution.
4471 bool no_interfering_store = true;
4472 Node* mem = alloc->in(TypeFunc::Memory);
4473 if (mem->is_MergeMem()) {
4474 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4475 Node* n = mms.memory();
4476 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4477 assert(n->is_Store(), "what else?");
4478 no_interfering_store = false;
4479 break;
4480 }
4481 }
4482 } else {
4483 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4484 Node* n = mms.memory();
4485 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4486 assert(n->is_Store(), "what else?");
4487 no_interfering_store = false;
4488 break;
4489 }
4490 }
4491 }
4492
4493 if (no_interfering_store) {
4494 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4495 uint size = alloc->req();
4496 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4497 old_jvms->set_map(sfpt);
4498 for (uint i = 0; i < size; i++) {
4499 sfpt->init_req(i, alloc->in(i));
4500 }
4501 // re-push array length for deoptimization
4502 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4503 old_jvms->set_sp(old_jvms->sp()+1);
4504 old_jvms->set_monoff(old_jvms->monoff()+1);
4505 old_jvms->set_scloff(old_jvms->scloff()+1);
4506 old_jvms->set_endoff(old_jvms->endoff()+1);
4507 old_jvms->set_should_reexecute(true);
4508
4509 sfpt->set_i_o(map()->i_o());
4510 sfpt->set_memory(map()->memory());
4511 sfpt->set_control(map()->control());
4512
4513 JVMState* saved_jvms = jvms();
4514 saved_reexecute_sp = _reexecute_sp;
4515
4516 set_jvms(sfpt->jvms());
4517 _reexecute_sp = jvms()->sp();
4518
4519 return saved_jvms;
4520 }
4521 }
4522 }
4523 return NULL;
4524 }
4525
4526 // In case of a deoptimization, we restart execution at the
4527 // allocation, allocating a new array. We would leave an uninitialized
4528 // array in the heap that GCs wouldn't expect. Move the allocation
4529 // after the traps so we don't allocate the array if we
4530 // deoptimize. This is possible because tightly_coupled_allocation()
4531 // guarantees there's no observer of the allocated array at this point
4532 // and the control flow is simple enough.
4533 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4534 int saved_reexecute_sp, uint new_idx) {
4535 if (saved_jvms != NULL && !stopped()) {
4536 assert(alloc != NULL, "only with a tightly coupled allocation");
4537 // restore JVM state to the state at the arraycopy
4538 saved_jvms->map()->set_control(map()->control());
4539 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4540 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4541 // If we've improved the types of some nodes (null check) while
4542 // emitting the guards, propagate them to the current state
4543 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4544 set_jvms(saved_jvms);
4545 _reexecute_sp = saved_reexecute_sp;
4546
4547 // Remove the allocation from above the guards
4548 CallProjections callprojs;
4549 alloc->extract_projections(&callprojs, true);
4550 InitializeNode* init = alloc->initialization();
4551 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4552 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4553 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4554 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4555
4556 // move the allocation here (after the guards)
4557 _gvn.hash_delete(alloc);
4558 alloc->set_req(TypeFunc::Control, control());
4559 alloc->set_req(TypeFunc::I_O, i_o());
4560 Node *mem = reset_memory();
4561 set_all_memory(mem);
4562 alloc->set_req(TypeFunc::Memory, mem);
4563 set_control(init->proj_out_or_null(TypeFunc::Control));
4564 set_i_o(callprojs.fallthrough_ioproj);
4565
4566 // Update memory as done in GraphKit::set_output_for_allocation()
4567 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4568 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4569 if (ary_type->isa_aryptr() && length_type != NULL) {
4570 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4571 }
4572 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4573 int elemidx = C->get_alias_index(telemref);
4574 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
4575 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
4576
4577 Node* allocx = _gvn.transform(alloc);
4578 assert(allocx == alloc, "where has the allocation gone?");
4579 assert(dest->is_CheckCastPP(), "not an allocation result?");
4580
4581 _gvn.hash_delete(dest);
4582 dest->set_req(0, control());
4583 Node* destx = _gvn.transform(dest);
4584 assert(destx == dest, "where has the allocation result gone?");
4585 }
4586 }
4587
4588
4589 //------------------------------inline_arraycopy-----------------------
4590 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4591 // Object dest, int destPos,
4592 // int length);
4593 bool LibraryCallKit::inline_arraycopy() {
4594 // Get the arguments.
4595 Node* src = argument(0); // type: oop
4596 Node* src_offset = argument(1); // type: int
4597 Node* dest = argument(2); // type: oop
4598 Node* dest_offset = argument(3); // type: int
4599 Node* length = argument(4); // type: int
4600
4601 uint new_idx = C->unique();
4602
4603 // Check for allocation before we add nodes that would confuse
4604 // tightly_coupled_allocation()
4605 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4606
4607 int saved_reexecute_sp = -1;
4608 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4609 // See arraycopy_restore_alloc_state() comment
4610 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4611 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4612 // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
4613 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4614
4615 // The following tests must be performed
4616 // (1) src and dest are arrays.
4617 // (2) src and dest arrays must have elements of the same BasicType
4618 // (3) src and dest must not be null.
4619 // (4) src_offset must not be negative.
4620 // (5) dest_offset must not be negative.
4621 // (6) length must not be negative.
4622 // (7) src_offset + length must not exceed length of src.
4623 // (8) dest_offset + length must not exceed length of dest.
4624 // (9) each element of an oop array must be assignable
4625
4626 // (3) src and dest must not be null.
4627 // always do this here because we need the JVM state for uncommon traps
4628 Node* null_ctl = top();
4629 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4630 assert(null_ctl->is_top(), "no null control here");
4631 dest = null_check(dest, T_ARRAY);
4632
4633 if (!can_emit_guards) {
4634 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4635 // guards but the arraycopy node could still take advantage of a
4636 // tightly allocated allocation. tightly_coupled_allocation() is
4637 // called again to make sure it takes the null check above into
4638 // account: the null check is mandatory and if it caused an
4639 // uncommon trap to be emitted then the allocation can't be
4640 // considered tightly coupled in this context.
4641 alloc = tightly_coupled_allocation(dest, NULL);
4642 }
4643
4644 bool validated = false;
4645
4646 const Type* src_type = _gvn.type(src);
4647 const Type* dest_type = _gvn.type(dest);
4648 const TypeAryPtr* top_src = src_type->isa_aryptr();
4649 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4650
4651 // Do we have the type of src?
4652 bool has_src = (top_src != NULL && top_src->klass() != NULL);
4653 // Do we have the type of dest?
4654 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4655 // Is the type for src from speculation?
4656 bool src_spec = false;
4657 // Is the type for dest from speculation?
4658 bool dest_spec = false;
4659
4660 if ((!has_src || !has_dest) && can_emit_guards) {
4661 // We don't have sufficient type information, let's see if
4662 // speculative types can help. We need to have types for both src
4663 // and dest so that it pays off.
4664
4665 // Do we already have or could we have type information for src
4666 bool could_have_src = has_src;
4667 // Do we already have or could we have type information for dest
4668 bool could_have_dest = has_dest;
4669
4670 ciKlass* src_k = NULL;
4671 if (!has_src) {
4672 src_k = src_type->speculative_type_not_null();
4673 if (src_k != NULL && src_k->is_array_klass()) {
4674 could_have_src = true;
4675 }
4676 }
4677
4678 ciKlass* dest_k = NULL;
4679 if (!has_dest) {
4680 dest_k = dest_type->speculative_type_not_null();
4681 if (dest_k != NULL && dest_k->is_array_klass()) {
4682 could_have_dest = true;
4683 }
4684 }
4685
4686 if (could_have_src && could_have_dest) {
4687 // This is going to pay off so emit the required guards
4688 if (!has_src) {
4689 src = maybe_cast_profiled_obj(src, src_k, true);
4690 src_type = _gvn.type(src);
4691 top_src = src_type->isa_aryptr();
4692 has_src = (top_src != NULL && top_src->klass() != NULL);
4693 src_spec = true;
4694 }
4695 if (!has_dest) {
4696 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4697 dest_type = _gvn.type(dest);
4698 top_dest = dest_type->isa_aryptr();
4699 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4700 dest_spec = true;
4701 }
4702 }
4703 }
4704
4705 if (has_src && has_dest && can_emit_guards) {
4706 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
4707 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
4708 if (is_reference_type(src_elem)) src_elem = T_OBJECT;
4709 if (is_reference_type(dest_elem)) dest_elem = T_OBJECT;
4710
4711 if (src_elem == dest_elem && src_elem == T_OBJECT) {
4712 // If both arrays are object arrays then having the exact types
4713 // for both will remove the need for a subtype check at runtime
4714 // before the call and may make it possible to pick a faster copy
4715 // routine (without a subtype check on every element)
4716 // Do we have the exact type of src?
4717 bool could_have_src = src_spec;
4718 // Do we have the exact type of dest?
4719 bool could_have_dest = dest_spec;
4720 ciKlass* src_k = top_src->klass();
4721 ciKlass* dest_k = top_dest->klass();
4722 if (!src_spec) {
4723 src_k = src_type->speculative_type_not_null();
4724 if (src_k != NULL && src_k->is_array_klass()) {
4725 could_have_src = true;
4726 }
4727 }
4728 if (!dest_spec) {
4729 dest_k = dest_type->speculative_type_not_null();
4730 if (dest_k != NULL && dest_k->is_array_klass()) {
4731 could_have_dest = true;
4732 }
4733 }
4734 if (could_have_src && could_have_dest) {
4735 // If we can have both exact types, emit the missing guards
4736 if (could_have_src && !src_spec) {
4737 src = maybe_cast_profiled_obj(src, src_k, true);
4738 }
4739 if (could_have_dest && !dest_spec) {
4740 dest = maybe_cast_profiled_obj(dest, dest_k, true);
4741 }
4742 }
4743 }
4744 }
4745
4746 ciMethod* trap_method = method();
4747 int trap_bci = bci();
4748 if (saved_jvms != NULL) {
4749 trap_method = alloc->jvms()->method();
4750 trap_bci = alloc->jvms()->bci();
4751 }
4752
4753 bool negative_length_guard_generated = false;
4754
4755 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4756 can_emit_guards &&
4757 !src->is_top() && !dest->is_top()) {
4758 // validate arguments: enables transformation the ArrayCopyNode
4759 validated = true;
4760
4761 RegionNode* slow_region = new RegionNode(1);
4762 record_for_igvn(slow_region);
4763
4764 // (1) src and dest are arrays.
4765 generate_non_array_guard(load_object_klass(src), slow_region);
4766 generate_non_array_guard(load_object_klass(dest), slow_region);
4767
4768 // (2) src and dest arrays must have elements of the same BasicType
4769 // done at macro expansion or at Ideal transformation time
4770
4771 // (4) src_offset must not be negative.
4772 generate_negative_guard(src_offset, slow_region);
4773
4774 // (5) dest_offset must not be negative.
4775 generate_negative_guard(dest_offset, slow_region);
4776
4777 // (7) src_offset + length must not exceed length of src.
4778 generate_limit_guard(src_offset, length,
4779 load_array_length(src),
4780 slow_region);
4781
4782 // (8) dest_offset + length must not exceed length of dest.
4783 generate_limit_guard(dest_offset, length,
4784 load_array_length(dest),
4785 slow_region);
4786
4787 // (6) length must not be negative.
4788 // This is also checked in generate_arraycopy() during macro expansion, but
4789 // we also have to check it here for the case where the ArrayCopyNode will
4790 // be eliminated by Escape Analysis.
4791 if (EliminateAllocations) {
4792 generate_negative_guard(length, slow_region);
4793 negative_length_guard_generated = true;
4794 }
4795
4796 // (9) each element of an oop array must be assignable
4797 Node* src_klass = load_object_klass(src);
4798 Node* dest_klass = load_object_klass(dest);
4799 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
4800
4801 if (not_subtype_ctrl != top()) {
4802 PreserveJVMState pjvms(this);
4803 set_control(not_subtype_ctrl);
4804 uncommon_trap(Deoptimization::Reason_intrinsic,
4805 Deoptimization::Action_make_not_entrant);
4806 assert(stopped(), "Should be stopped");
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 Node* new_src = access_resolve(src, ACCESS_READ);
4828 Node* new_dest = access_resolve(dest, ACCESS_WRITE);
4829
4830 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, new_src, src_offset, new_dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
4831 // Create LoadRange and LoadKlass nodes for use during macro expansion here
4832 // so the compiler has a chance to eliminate them: during macro expansion,
4833 // we have to set their control (CastPP nodes are eliminated).
4834 load_object_klass(src), load_object_klass(dest),
4835 load_array_length(src), load_array_length(dest));
4836
4837 ac->set_arraycopy(validated);
4838
4839 Node* n = _gvn.transform(ac);
4840 if (n == ac) {
4841 ac->connect_outputs(this);
4842 } else {
4843 assert(validated, "shouldn't transform if all arguments not validated");
4844 set_all_memory(n);
4845 }
4846 clear_upper_avx();
4847
4848
4849 return true;
4850 }
4851
4852
4853 // Helper function which determines if an arraycopy immediately follows
4854 // an allocation, with no intervening tests or other escapes for the object.
4855 AllocateArrayNode*
4856 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
4857 RegionNode* slow_region) {
4858 if (stopped()) return NULL; // no fast path
4859 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
4860
4861 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
4862 if (alloc == NULL) return NULL;
4863
4864 Node* rawmem = memory(Compile::AliasIdxRaw);
4865 // Is the allocation's memory state untouched?
4866 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
4867 // Bail out if there have been raw-memory effects since the allocation.
4868 // (Example: There might have been a call or safepoint.)
4869 return NULL;
4870 }
4871 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
4872 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
4873 return NULL;
4874 }
4875
4876 // There must be no unexpected observers of this allocation.
4877 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
4878 Node* obs = ptr->fast_out(i);
4879 if (obs != this->map()) {
4880 return NULL;
4881 }
4882 }
4883
4884 // This arraycopy must unconditionally follow the allocation of the ptr.
4885 Node* alloc_ctl = ptr->in(0);
4886 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
4887
4888 Node* ctl = control();
4889 while (ctl != alloc_ctl) {
4890 // There may be guards which feed into the slow_region.
4891 // Any other control flow means that we might not get a chance
4892 // to finish initializing the allocated object.
4893 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
4894 IfNode* iff = ctl->in(0)->as_If();
4895 Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
4896 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
4897 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
4898 ctl = iff->in(0); // This test feeds the known slow_region.
4899 continue;
4900 }
4901 // One more try: Various low-level checks bottom out in
4902 // uncommon traps. If the debug-info of the trap omits
4903 // any reference to the allocation, as we've already
4904 // observed, then there can be no objection to the trap.
4905 bool found_trap = false;
4906 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
4907 Node* obs = not_ctl->fast_out(j);
4908 if (obs->in(0) == not_ctl && obs->is_Call() &&
4909 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
4910 found_trap = true; break;
4911 }
4912 }
4913 if (found_trap) {
4914 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
4915 continue;
4916 }
4917 }
4918 return NULL;
4919 }
4920
4921 // If we get this far, we have an allocation which immediately
4922 // precedes the arraycopy, and we can take over zeroing the new object.
4923 // The arraycopy will finish the initialization, and provide
4924 // a new control state to which we will anchor the destination pointer.
4925
4926 return alloc;
4927 }
4928
4929 //-------------inline_encodeISOArray-----------------------------------
4930 // encode char[] to byte[] in ISO_8859_1
4931 bool LibraryCallKit::inline_encodeISOArray() {
4932 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
4933 // no receiver since it is static method
4934 Node *src = argument(0);
4935 Node *src_offset = argument(1);
4936 Node *dst = argument(2);
4937 Node *dst_offset = argument(3);
4938 Node *length = argument(4);
4939
4940 src = must_be_not_null(src, true);
4941 dst = must_be_not_null(dst, true);
4942
4943 src = access_resolve(src, ACCESS_READ);
4944 dst = access_resolve(dst, ACCESS_WRITE);
4945
4946 const Type* src_type = src->Value(&_gvn);
4947 const Type* dst_type = dst->Value(&_gvn);
4948 const TypeAryPtr* top_src = src_type->isa_aryptr();
4949 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
4950 if (top_src == NULL || top_src->klass() == NULL ||
4951 top_dest == NULL || top_dest->klass() == NULL) {
4952 // failed array check
4953 return false;
4954 }
4955
4956 // Figure out the size and type of the elements we will be copying.
4957 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4958 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4959 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
4960 return false;
4961 }
4962
4963 Node* src_start = array_element_address(src, src_offset, T_CHAR);
4964 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
4965 // 'src_start' points to src array + scaled offset
4966 // 'dst_start' points to dst array + scaled offset
4967
4968 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
4969 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
4970 enc = _gvn.transform(enc);
4971 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
4972 set_memory(res_mem, mtype);
4973 set_result(enc);
4974 clear_upper_avx();
4975
4976 return true;
4977 }
4978
4979 //-------------inline_multiplyToLen-----------------------------------
4980 bool LibraryCallKit::inline_multiplyToLen() {
4981 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
4982
4983 address stubAddr = StubRoutines::multiplyToLen();
4984 if (stubAddr == NULL) {
4985 return false; // Intrinsic's stub is not implemented on this platform
4986 }
4987 const char* stubName = "multiplyToLen";
4988
4989 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
4990
4991 // no receiver because it is a static method
4992 Node* x = argument(0);
4993 Node* xlen = argument(1);
4994 Node* y = argument(2);
4995 Node* ylen = argument(3);
4996 Node* z = argument(4);
4997
4998 x = must_be_not_null(x, true);
4999 y = must_be_not_null(y, true);
5000
5001 x = access_resolve(x, ACCESS_READ);
5002 y = access_resolve(y, ACCESS_READ);
5003 z = access_resolve(z, ACCESS_WRITE);
5004
5005 const Type* x_type = x->Value(&_gvn);
5006 const Type* y_type = y->Value(&_gvn);
5007 const TypeAryPtr* top_x = x_type->isa_aryptr();
5008 const TypeAryPtr* top_y = y_type->isa_aryptr();
5009 if (top_x == NULL || top_x->klass() == NULL ||
5010 top_y == NULL || top_y->klass() == NULL) {
5011 // failed array check
5012 return false;
5013 }
5014
5015 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5016 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5017 if (x_elem != T_INT || y_elem != T_INT) {
5018 return false;
5019 }
5020
5021 // Set the original stack and the reexecute bit for the interpreter to reexecute
5022 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5023 // on the return from z array allocation in runtime.
5024 { PreserveReexecuteState preexecs(this);
5025 jvms()->set_should_reexecute(true);
5026
5027 Node* x_start = array_element_address(x, intcon(0), x_elem);
5028 Node* y_start = array_element_address(y, intcon(0), y_elem);
5029 // 'x_start' points to x array + scaled xlen
5030 // 'y_start' points to y array + scaled ylen
5031
5032 // Allocate the result array
5033 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5034 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5035 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5036
5037 IdealKit ideal(this);
5038
5039 #define __ ideal.
5040 Node* one = __ ConI(1);
5041 Node* zero = __ ConI(0);
5042 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5043 __ set(need_alloc, zero);
5044 __ set(z_alloc, z);
5045 __ if_then(z, BoolTest::eq, null()); {
5046 __ increment (need_alloc, one);
5047 } __ else_(); {
5048 // Update graphKit memory and control from IdealKit.
5049 sync_kit(ideal);
5050 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5051 cast->init_req(0, control());
5052 _gvn.set_type(cast, cast->bottom_type());
5053 C->record_for_igvn(cast);
5054
5055 Node* zlen_arg = load_array_length(cast);
5056 // Update IdealKit memory and control from graphKit.
5057 __ sync_kit(this);
5058 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5059 __ increment (need_alloc, one);
5060 } __ end_if();
5061 } __ end_if();
5062
5063 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5064 // Update graphKit memory and control from IdealKit.
5065 sync_kit(ideal);
5066 Node * narr = new_array(klass_node, zlen, 1);
5067 // Update IdealKit memory and control from graphKit.
5068 __ sync_kit(this);
5069 __ set(z_alloc, narr);
5070 } __ end_if();
5071
5072 sync_kit(ideal);
5073 z = __ value(z_alloc);
5074 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5075 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5076 // Final sync IdealKit and GraphKit.
5077 final_sync(ideal);
5078 #undef __
5079
5080 Node* z_start = array_element_address(z, intcon(0), T_INT);
5081
5082 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5083 OptoRuntime::multiplyToLen_Type(),
5084 stubAddr, stubName, TypePtr::BOTTOM,
5085 x_start, xlen, y_start, ylen, z_start, zlen);
5086 } // original reexecute is set back here
5087
5088 C->set_has_split_ifs(true); // Has chance for split-if optimization
5089 set_result(z);
5090 return true;
5091 }
5092
5093 //-------------inline_squareToLen------------------------------------
5094 bool LibraryCallKit::inline_squareToLen() {
5095 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5096
5097 address stubAddr = StubRoutines::squareToLen();
5098 if (stubAddr == NULL) {
5099 return false; // Intrinsic's stub is not implemented on this platform
5100 }
5101 const char* stubName = "squareToLen";
5102
5103 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5104
5105 Node* x = argument(0);
5106 Node* len = argument(1);
5107 Node* z = argument(2);
5108 Node* zlen = argument(3);
5109
5110 x = must_be_not_null(x, true);
5111 z = must_be_not_null(z, true);
5112
5113 x = access_resolve(x, ACCESS_READ);
5114 z = access_resolve(z, ACCESS_WRITE);
5115
5116 const Type* x_type = x->Value(&_gvn);
5117 const Type* z_type = z->Value(&_gvn);
5118 const TypeAryPtr* top_x = x_type->isa_aryptr();
5119 const TypeAryPtr* top_z = z_type->isa_aryptr();
5120 if (top_x == NULL || top_x->klass() == NULL ||
5121 top_z == NULL || top_z->klass() == NULL) {
5122 // failed array check
5123 return false;
5124 }
5125
5126 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5127 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5128 if (x_elem != T_INT || z_elem != T_INT) {
5129 return false;
5130 }
5131
5132
5133 Node* x_start = array_element_address(x, intcon(0), x_elem);
5134 Node* z_start = array_element_address(z, intcon(0), z_elem);
5135
5136 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5137 OptoRuntime::squareToLen_Type(),
5138 stubAddr, stubName, TypePtr::BOTTOM,
5139 x_start, len, z_start, zlen);
5140
5141 set_result(z);
5142 return true;
5143 }
5144
5145 //-------------inline_mulAdd------------------------------------------
5146 bool LibraryCallKit::inline_mulAdd() {
5147 assert(UseMulAddIntrinsic, "not implemented on this platform");
5148
5149 address stubAddr = StubRoutines::mulAdd();
5150 if (stubAddr == NULL) {
5151 return false; // Intrinsic's stub is not implemented on this platform
5152 }
5153 const char* stubName = "mulAdd";
5154
5155 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5156
5157 Node* out = argument(0);
5158 Node* in = argument(1);
5159 Node* offset = argument(2);
5160 Node* len = argument(3);
5161 Node* k = argument(4);
5162
5163 out = must_be_not_null(out, true);
5164
5165 in = access_resolve(in, ACCESS_READ);
5166 out = access_resolve(out, ACCESS_WRITE);
5167
5168 const Type* out_type = out->Value(&_gvn);
5169 const Type* in_type = in->Value(&_gvn);
5170 const TypeAryPtr* top_out = out_type->isa_aryptr();
5171 const TypeAryPtr* top_in = in_type->isa_aryptr();
5172 if (top_out == NULL || top_out->klass() == NULL ||
5173 top_in == NULL || top_in->klass() == NULL) {
5174 // failed array check
5175 return false;
5176 }
5177
5178 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5179 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5180 if (out_elem != T_INT || in_elem != T_INT) {
5181 return false;
5182 }
5183
5184 Node* outlen = load_array_length(out);
5185 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5186 Node* out_start = array_element_address(out, intcon(0), out_elem);
5187 Node* in_start = array_element_address(in, intcon(0), in_elem);
5188
5189 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5190 OptoRuntime::mulAdd_Type(),
5191 stubAddr, stubName, TypePtr::BOTTOM,
5192 out_start,in_start, new_offset, len, k);
5193 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5194 set_result(result);
5195 return true;
5196 }
5197
5198 //-------------inline_montgomeryMultiply-----------------------------------
5199 bool LibraryCallKit::inline_montgomeryMultiply() {
5200 address stubAddr = StubRoutines::montgomeryMultiply();
5201 if (stubAddr == NULL) {
5202 return false; // Intrinsic's stub is not implemented on this platform
5203 }
5204
5205 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5206 const char* stubName = "montgomery_multiply";
5207
5208 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5209
5210 Node* a = argument(0);
5211 Node* b = argument(1);
5212 Node* n = argument(2);
5213 Node* len = argument(3);
5214 Node* inv = argument(4);
5215 Node* m = argument(6);
5216
5217 a = access_resolve(a, ACCESS_READ);
5218 b = access_resolve(b, ACCESS_READ);
5219 n = access_resolve(n, ACCESS_READ);
5220 m = access_resolve(m, ACCESS_WRITE);
5221
5222 const Type* a_type = a->Value(&_gvn);
5223 const TypeAryPtr* top_a = a_type->isa_aryptr();
5224 const Type* b_type = b->Value(&_gvn);
5225 const TypeAryPtr* top_b = b_type->isa_aryptr();
5226 const Type* n_type = a->Value(&_gvn);
5227 const TypeAryPtr* top_n = n_type->isa_aryptr();
5228 const Type* m_type = a->Value(&_gvn);
5229 const TypeAryPtr* top_m = m_type->isa_aryptr();
5230 if (top_a == NULL || top_a->klass() == NULL ||
5231 top_b == NULL || top_b->klass() == NULL ||
5232 top_n == NULL || top_n->klass() == NULL ||
5233 top_m == NULL || top_m->klass() == NULL) {
5234 // failed array check
5235 return false;
5236 }
5237
5238 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5239 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5240 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5241 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5242 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5243 return false;
5244 }
5245
5246 // Make the call
5247 {
5248 Node* a_start = array_element_address(a, intcon(0), a_elem);
5249 Node* b_start = array_element_address(b, intcon(0), b_elem);
5250 Node* n_start = array_element_address(n, intcon(0), n_elem);
5251 Node* m_start = array_element_address(m, intcon(0), m_elem);
5252
5253 Node* call = make_runtime_call(RC_LEAF,
5254 OptoRuntime::montgomeryMultiply_Type(),
5255 stubAddr, stubName, TypePtr::BOTTOM,
5256 a_start, b_start, n_start, len, inv, top(),
5257 m_start);
5258 set_result(m);
5259 }
5260
5261 return true;
5262 }
5263
5264 bool LibraryCallKit::inline_montgomerySquare() {
5265 address stubAddr = StubRoutines::montgomerySquare();
5266 if (stubAddr == NULL) {
5267 return false; // Intrinsic's stub is not implemented on this platform
5268 }
5269
5270 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5271 const char* stubName = "montgomery_square";
5272
5273 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5274
5275 Node* a = argument(0);
5276 Node* n = argument(1);
5277 Node* len = argument(2);
5278 Node* inv = argument(3);
5279 Node* m = argument(5);
5280
5281 a = access_resolve(a, ACCESS_READ);
5282 n = access_resolve(n, ACCESS_READ);
5283 m = access_resolve(m, ACCESS_WRITE);
5284
5285 const Type* a_type = a->Value(&_gvn);
5286 const TypeAryPtr* top_a = a_type->isa_aryptr();
5287 const Type* n_type = a->Value(&_gvn);
5288 const TypeAryPtr* top_n = n_type->isa_aryptr();
5289 const Type* m_type = a->Value(&_gvn);
5290 const TypeAryPtr* top_m = m_type->isa_aryptr();
5291 if (top_a == NULL || top_a->klass() == NULL ||
5292 top_n == NULL || top_n->klass() == NULL ||
5293 top_m == NULL || top_m->klass() == NULL) {
5294 // failed array check
5295 return false;
5296 }
5297
5298 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5299 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5300 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5301 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5302 return false;
5303 }
5304
5305 // Make the call
5306 {
5307 Node* a_start = array_element_address(a, intcon(0), a_elem);
5308 Node* n_start = array_element_address(n, intcon(0), n_elem);
5309 Node* m_start = array_element_address(m, intcon(0), m_elem);
5310
5311 Node* call = make_runtime_call(RC_LEAF,
5312 OptoRuntime::montgomerySquare_Type(),
5313 stubAddr, stubName, TypePtr::BOTTOM,
5314 a_start, n_start, len, inv, top(),
5315 m_start);
5316 set_result(m);
5317 }
5318
5319 return true;
5320 }
5321
5322 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
5323 address stubAddr = NULL;
5324 const char* stubName = NULL;
5325
5326 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
5327 if (stubAddr == NULL) {
5328 return false; // Intrinsic's stub is not implemented on this platform
5329 }
5330
5331 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
5332
5333 assert(callee()->signature()->size() == 5, "expected 5 arguments");
5334
5335 Node* newArr = argument(0);
5336 Node* oldArr = argument(1);
5337 Node* newIdx = argument(2);
5338 Node* shiftCount = argument(3);
5339 Node* numIter = argument(4);
5340
5341 const Type* newArr_type = newArr->Value(&_gvn);
5342 const TypeAryPtr* top_newArr = newArr_type->isa_aryptr();
5343 const Type* oldArr_type = oldArr->Value(&_gvn);
5344 const TypeAryPtr* top_oldArr = oldArr_type->isa_aryptr();
5345 if (top_newArr == NULL || top_newArr->klass() == NULL || top_oldArr == NULL
5346 || top_oldArr->klass() == NULL) {
5347 return false;
5348 }
5349
5350 BasicType newArr_elem = newArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5351 BasicType oldArr_elem = oldArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5352 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
5353 return false;
5354 }
5355
5356 // Make the call
5357 {
5358 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
5359 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
5360
5361 Node* call = make_runtime_call(RC_LEAF,
5362 OptoRuntime::bigIntegerShift_Type(),
5363 stubAddr,
5364 stubName,
5365 TypePtr::BOTTOM,
5366 newArr_start,
5367 oldArr_start,
5368 newIdx,
5369 shiftCount,
5370 numIter);
5371 }
5372
5373 return true;
5374 }
5375
5376 //-------------inline_vectorizedMismatch------------------------------
5377 bool LibraryCallKit::inline_vectorizedMismatch() {
5378 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5379
5380 address stubAddr = StubRoutines::vectorizedMismatch();
5381 if (stubAddr == NULL) {
5382 return false; // Intrinsic's stub is not implemented on this platform
5383 }
5384 const char* stubName = "vectorizedMismatch";
5385 int size_l = callee()->signature()->size();
5386 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5387
5388 Node* obja = argument(0);
5389 Node* aoffset = argument(1);
5390 Node* objb = argument(3);
5391 Node* boffset = argument(4);
5392 Node* length = argument(6);
5393 Node* scale = argument(7);
5394
5395 const Type* a_type = obja->Value(&_gvn);
5396 const Type* b_type = objb->Value(&_gvn);
5397 const TypeAryPtr* top_a = a_type->isa_aryptr();
5398 const TypeAryPtr* top_b = b_type->isa_aryptr();
5399 if (top_a == NULL || top_a->klass() == NULL ||
5400 top_b == NULL || top_b->klass() == NULL) {
5401 // failed array check
5402 return false;
5403 }
5404
5405 Node* call;
5406 jvms()->set_should_reexecute(true);
5407
5408 obja = access_resolve(obja, ACCESS_READ);
5409 objb = access_resolve(objb, ACCESS_READ);
5410 Node* obja_adr = make_unsafe_address(obja, aoffset, ACCESS_READ);
5411 Node* objb_adr = make_unsafe_address(objb, boffset, ACCESS_READ);
5412
5413 call = make_runtime_call(RC_LEAF,
5414 OptoRuntime::vectorizedMismatch_Type(),
5415 stubAddr, stubName, TypePtr::BOTTOM,
5416 obja_adr, objb_adr, length, scale);
5417
5418 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5419 set_result(result);
5420 return true;
5421 }
5422
5423 /**
5424 * Calculate CRC32 for byte.
5425 * int java.util.zip.CRC32.update(int crc, int b)
5426 */
5427 bool LibraryCallKit::inline_updateCRC32() {
5428 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5429 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5430 // no receiver since it is static method
5431 Node* crc = argument(0); // type: int
5432 Node* b = argument(1); // type: int
5433
5434 /*
5435 * int c = ~ crc;
5436 * b = timesXtoThe32[(b ^ c) & 0xFF];
5437 * b = b ^ (c >>> 8);
5438 * crc = ~b;
5439 */
5440
5441 Node* M1 = intcon(-1);
5442 crc = _gvn.transform(new XorINode(crc, M1));
5443 Node* result = _gvn.transform(new XorINode(crc, b));
5444 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5445
5446 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5447 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5448 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5449 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5450
5451 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5452 result = _gvn.transform(new XorINode(crc, result));
5453 result = _gvn.transform(new XorINode(result, M1));
5454 set_result(result);
5455 return true;
5456 }
5457
5458 /**
5459 * Calculate CRC32 for byte[] array.
5460 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5461 */
5462 bool LibraryCallKit::inline_updateBytesCRC32() {
5463 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5464 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5465 // no receiver since it is static method
5466 Node* crc = argument(0); // type: int
5467 Node* src = argument(1); // type: oop
5468 Node* offset = argument(2); // type: int
5469 Node* length = argument(3); // type: int
5470
5471 const Type* src_type = src->Value(&_gvn);
5472 const TypeAryPtr* top_src = src_type->isa_aryptr();
5473 if (top_src == NULL || top_src->klass() == NULL) {
5474 // failed array check
5475 return false;
5476 }
5477
5478 // Figure out the size and type of the elements we will be copying.
5479 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5480 if (src_elem != T_BYTE) {
5481 return false;
5482 }
5483
5484 // 'src_start' points to src array + scaled offset
5485 src = must_be_not_null(src, true);
5486 src = access_resolve(src, ACCESS_READ);
5487 Node* src_start = array_element_address(src, offset, src_elem);
5488
5489 // We assume that range check is done by caller.
5490 // TODO: generate range check (offset+length < src.length) in debug VM.
5491
5492 // Call the stub.
5493 address stubAddr = StubRoutines::updateBytesCRC32();
5494 const char *stubName = "updateBytesCRC32";
5495
5496 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5497 stubAddr, stubName, TypePtr::BOTTOM,
5498 crc, src_start, length);
5499 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5500 set_result(result);
5501 return true;
5502 }
5503
5504 /**
5505 * Calculate CRC32 for ByteBuffer.
5506 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5507 */
5508 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5509 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5510 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5511 // no receiver since it is static method
5512 Node* crc = argument(0); // type: int
5513 Node* src = argument(1); // type: long
5514 Node* offset = argument(3); // type: int
5515 Node* length = argument(4); // type: int
5516
5517 src = ConvL2X(src); // adjust Java long to machine word
5518 Node* base = _gvn.transform(new CastX2PNode(src));
5519 offset = ConvI2X(offset);
5520
5521 // 'src_start' points to src array + scaled offset
5522 Node* src_start = basic_plus_adr(top(), base, offset);
5523
5524 // Call the stub.
5525 address stubAddr = StubRoutines::updateBytesCRC32();
5526 const char *stubName = "updateBytesCRC32";
5527
5528 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5529 stubAddr, stubName, TypePtr::BOTTOM,
5530 crc, src_start, length);
5531 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5532 set_result(result);
5533 return true;
5534 }
5535
5536 //------------------------------get_table_from_crc32c_class-----------------------
5537 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5538 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5539 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5540
5541 return table;
5542 }
5543
5544 //------------------------------inline_updateBytesCRC32C-----------------------
5545 //
5546 // Calculate CRC32C for byte[] array.
5547 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5548 //
5549 bool LibraryCallKit::inline_updateBytesCRC32C() {
5550 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5551 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5552 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5553 // no receiver since it is a static method
5554 Node* crc = argument(0); // type: int
5555 Node* src = argument(1); // type: oop
5556 Node* offset = argument(2); // type: int
5557 Node* end = argument(3); // type: int
5558
5559 Node* length = _gvn.transform(new SubINode(end, offset));
5560
5561 const Type* src_type = src->Value(&_gvn);
5562 const TypeAryPtr* top_src = src_type->isa_aryptr();
5563 if (top_src == NULL || top_src->klass() == NULL) {
5564 // failed array check
5565 return false;
5566 }
5567
5568 // Figure out the size and type of the elements we will be copying.
5569 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5570 if (src_elem != T_BYTE) {
5571 return false;
5572 }
5573
5574 // 'src_start' points to src array + scaled offset
5575 src = must_be_not_null(src, true);
5576 src = access_resolve(src, ACCESS_READ);
5577 Node* src_start = array_element_address(src, offset, src_elem);
5578
5579 // static final int[] byteTable in class CRC32C
5580 Node* table = get_table_from_crc32c_class(callee()->holder());
5581 table = must_be_not_null(table, true);
5582 table = access_resolve(table, ACCESS_READ);
5583 Node* table_start = array_element_address(table, intcon(0), T_INT);
5584
5585 // We assume that range check is done by caller.
5586 // TODO: generate range check (offset+length < src.length) in debug VM.
5587
5588 // Call the stub.
5589 address stubAddr = StubRoutines::updateBytesCRC32C();
5590 const char *stubName = "updateBytesCRC32C";
5591
5592 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5593 stubAddr, stubName, TypePtr::BOTTOM,
5594 crc, src_start, length, table_start);
5595 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5596 set_result(result);
5597 return true;
5598 }
5599
5600 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5601 //
5602 // Calculate CRC32C for DirectByteBuffer.
5603 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5604 //
5605 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5606 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5607 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5608 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5609 // no receiver since it is a static method
5610 Node* crc = argument(0); // type: int
5611 Node* src = argument(1); // type: long
5612 Node* offset = argument(3); // type: int
5613 Node* end = argument(4); // type: int
5614
5615 Node* length = _gvn.transform(new SubINode(end, offset));
5616
5617 src = ConvL2X(src); // adjust Java long to machine word
5618 Node* base = _gvn.transform(new CastX2PNode(src));
5619 offset = ConvI2X(offset);
5620
5621 // 'src_start' points to src array + scaled offset
5622 Node* src_start = basic_plus_adr(top(), base, offset);
5623
5624 // static final int[] byteTable in class CRC32C
5625 Node* table = get_table_from_crc32c_class(callee()->holder());
5626 table = must_be_not_null(table, true);
5627 table = access_resolve(table, ACCESS_READ);
5628 Node* table_start = array_element_address(table, intcon(0), T_INT);
5629
5630 // Call the stub.
5631 address stubAddr = StubRoutines::updateBytesCRC32C();
5632 const char *stubName = "updateBytesCRC32C";
5633
5634 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5635 stubAddr, stubName, TypePtr::BOTTOM,
5636 crc, src_start, length, table_start);
5637 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5638 set_result(result);
5639 return true;
5640 }
5641
5642 //------------------------------inline_updateBytesAdler32----------------------
5643 //
5644 // Calculate Adler32 checksum for byte[] array.
5645 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5646 //
5647 bool LibraryCallKit::inline_updateBytesAdler32() {
5648 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5649 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5650 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5651 // no receiver since it is static method
5652 Node* crc = argument(0); // type: int
5653 Node* src = argument(1); // type: oop
5654 Node* offset = argument(2); // type: int
5655 Node* length = argument(3); // type: int
5656
5657 const Type* src_type = src->Value(&_gvn);
5658 const TypeAryPtr* top_src = src_type->isa_aryptr();
5659 if (top_src == NULL || top_src->klass() == NULL) {
5660 // failed array check
5661 return false;
5662 }
5663
5664 // Figure out the size and type of the elements we will be copying.
5665 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5666 if (src_elem != T_BYTE) {
5667 return false;
5668 }
5669
5670 // 'src_start' points to src array + scaled offset
5671 src = access_resolve(src, ACCESS_READ);
5672 Node* src_start = array_element_address(src, offset, src_elem);
5673
5674 // We assume that range check is done by caller.
5675 // TODO: generate range check (offset+length < src.length) in debug VM.
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 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5685 set_result(result);
5686 return true;
5687 }
5688
5689 //------------------------------inline_updateByteBufferAdler32---------------
5690 //
5691 // Calculate Adler32 checksum for DirectByteBuffer.
5692 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5693 //
5694 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5695 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5696 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5697 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5698 // no receiver since it is static method
5699 Node* crc = argument(0); // type: int
5700 Node* src = argument(1); // type: long
5701 Node* offset = argument(3); // type: int
5702 Node* length = argument(4); // type: int
5703
5704 src = ConvL2X(src); // adjust Java long to machine word
5705 Node* base = _gvn.transform(new CastX2PNode(src));
5706 offset = ConvI2X(offset);
5707
5708 // 'src_start' points to src array + scaled offset
5709 Node* src_start = basic_plus_adr(top(), base, offset);
5710
5711 // Call the stub.
5712 address stubAddr = StubRoutines::updateBytesAdler32();
5713 const char *stubName = "updateBytesAdler32";
5714
5715 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5716 stubAddr, stubName, TypePtr::BOTTOM,
5717 crc, src_start, length);
5718
5719 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5720 set_result(result);
5721 return true;
5722 }
5723
5724 //----------------------------inline_reference_get----------------------------
5725 // public T java.lang.ref.Reference.get();
5726 bool LibraryCallKit::inline_reference_get() {
5727 const int referent_offset = java_lang_ref_Reference::referent_offset;
5728 guarantee(referent_offset > 0, "should have already been set");
5729
5730 // Get the argument:
5731 Node* reference_obj = null_check_receiver();
5732 if (stopped()) return true;
5733
5734 const TypeInstPtr* tinst = _gvn.type(reference_obj)->isa_instptr();
5735 assert(tinst != NULL, "obj is null");
5736 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5737 ciInstanceKlass* referenceKlass = tinst->klass()->as_instance_klass();
5738 ciField* field = referenceKlass->get_field_by_name(ciSymbol::make("referent"),
5739 ciSymbol::make("Ljava/lang/Object;"),
5740 false);
5741 assert (field != NULL, "undefined field");
5742
5743 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5744 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5745
5746 ciInstanceKlass* klass = env()->Object_klass();
5747 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5748
5749 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
5750 Node* result = access_load_at(reference_obj, adr, adr_type, object_type, T_OBJECT, decorators);
5751 // Add memory barrier to prevent commoning reads from this field
5752 // across safepoint since GC can change its value.
5753 insert_mem_bar(Op_MemBarCPUOrder);
5754
5755 set_result(result);
5756 return true;
5757 }
5758
5759
5760 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5761 bool is_exact=true, bool is_static=false,
5762 ciInstanceKlass * fromKls=NULL) {
5763 if (fromKls == NULL) {
5764 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5765 assert(tinst != NULL, "obj is null");
5766 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5767 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5768 fromKls = tinst->klass()->as_instance_klass();
5769 } else {
5770 assert(is_static, "only for static field access");
5771 }
5772 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5773 ciSymbol::make(fieldTypeString),
5774 is_static);
5775
5776 assert (field != NULL, "undefined field");
5777 if (field == NULL) return (Node *) NULL;
5778
5779 if (is_static) {
5780 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5781 fromObj = makecon(tip);
5782 }
5783
5784 // Next code copied from Parse::do_get_xxx():
5785
5786 // Compute address and memory type.
5787 int offset = field->offset_in_bytes();
5788 bool is_vol = field->is_volatile();
5789 ciType* field_klass = field->type();
5790 assert(field_klass->is_loaded(), "should be loaded");
5791 const TypePtr* adr_type = C->alias_type(field)->adr_type();
5792 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5793 BasicType bt = field->layout_type();
5794
5795 // Build the resultant type of the load
5796 const Type *type;
5797 if (bt == T_OBJECT) {
5798 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5799 } else {
5800 type = Type::get_const_basic_type(bt);
5801 }
5802
5803 DecoratorSet decorators = IN_HEAP;
5804
5805 if (is_vol) {
5806 decorators |= MO_SEQ_CST;
5807 }
5808
5809 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
5810 }
5811
5812 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5813 bool is_exact = true, bool is_static = false,
5814 ciInstanceKlass * fromKls = NULL) {
5815 if (fromKls == NULL) {
5816 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5817 assert(tinst != NULL, "obj is null");
5818 assert(tinst->klass()->is_loaded(), "obj is not loaded");
5819 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5820 fromKls = tinst->klass()->as_instance_klass();
5821 }
5822 else {
5823 assert(is_static, "only for static field access");
5824 }
5825 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5826 ciSymbol::make(fieldTypeString),
5827 is_static);
5828
5829 assert(field != NULL, "undefined field");
5830 assert(!field->is_volatile(), "not defined for volatile fields");
5831
5832 if (is_static) {
5833 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5834 fromObj = makecon(tip);
5835 }
5836
5837 // Next code copied from Parse::do_get_xxx():
5838
5839 // Compute address and memory type.
5840 int offset = field->offset_in_bytes();
5841 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5842
5843 return adr;
5844 }
5845
5846 //------------------------------inline_aescrypt_Block-----------------------
5847 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
5848 address stubAddr = NULL;
5849 const char *stubName;
5850 assert(UseAES, "need AES instruction support");
5851
5852 switch(id) {
5853 case vmIntrinsics::_aescrypt_encryptBlock:
5854 stubAddr = StubRoutines::aescrypt_encryptBlock();
5855 stubName = "aescrypt_encryptBlock";
5856 break;
5857 case vmIntrinsics::_aescrypt_decryptBlock:
5858 stubAddr = StubRoutines::aescrypt_decryptBlock();
5859 stubName = "aescrypt_decryptBlock";
5860 break;
5861 default:
5862 break;
5863 }
5864 if (stubAddr == NULL) return false;
5865
5866 Node* aescrypt_object = argument(0);
5867 Node* src = argument(1);
5868 Node* src_offset = argument(2);
5869 Node* dest = argument(3);
5870 Node* dest_offset = argument(4);
5871
5872 src = must_be_not_null(src, true);
5873 dest = must_be_not_null(dest, true);
5874
5875 src = access_resolve(src, ACCESS_READ);
5876 dest = access_resolve(dest, ACCESS_WRITE);
5877
5878 // (1) src and dest are arrays.
5879 const Type* src_type = src->Value(&_gvn);
5880 const Type* dest_type = dest->Value(&_gvn);
5881 const TypeAryPtr* top_src = src_type->isa_aryptr();
5882 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5883 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5884
5885 // for the quick and dirty code we will skip all the checks.
5886 // we are just trying to get the call to be generated.
5887 Node* src_start = src;
5888 Node* dest_start = dest;
5889 if (src_offset != NULL || dest_offset != NULL) {
5890 assert(src_offset != NULL && dest_offset != NULL, "");
5891 src_start = array_element_address(src, src_offset, T_BYTE);
5892 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5893 }
5894
5895 // now need to get the start of its expanded key array
5896 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5897 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5898 if (k_start == NULL) return false;
5899
5900 if (Matcher::pass_original_key_for_aes()) {
5901 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
5902 // compatibility issues between Java key expansion and SPARC crypto instructions
5903 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
5904 if (original_k_start == NULL) return false;
5905
5906 // Call the stub.
5907 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5908 stubAddr, stubName, TypePtr::BOTTOM,
5909 src_start, dest_start, k_start, original_k_start);
5910 } else {
5911 // Call the stub.
5912 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5913 stubAddr, stubName, TypePtr::BOTTOM,
5914 src_start, dest_start, k_start);
5915 }
5916
5917 return true;
5918 }
5919
5920 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
5921 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
5922 address stubAddr = NULL;
5923 const char *stubName = NULL;
5924
5925 assert(UseAES, "need AES instruction support");
5926
5927 switch(id) {
5928 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
5929 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
5930 stubName = "cipherBlockChaining_encryptAESCrypt";
5931 break;
5932 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
5933 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
5934 stubName = "cipherBlockChaining_decryptAESCrypt";
5935 break;
5936 default:
5937 break;
5938 }
5939 if (stubAddr == NULL) return false;
5940
5941 Node* cipherBlockChaining_object = argument(0);
5942 Node* src = argument(1);
5943 Node* src_offset = argument(2);
5944 Node* len = argument(3);
5945 Node* dest = argument(4);
5946 Node* dest_offset = argument(5);
5947
5948 src = must_be_not_null(src, false);
5949 dest = must_be_not_null(dest, false);
5950
5951 src = access_resolve(src, ACCESS_READ);
5952 dest = access_resolve(dest, ACCESS_WRITE);
5953
5954 // (1) src and dest are arrays.
5955 const Type* src_type = src->Value(&_gvn);
5956 const Type* dest_type = dest->Value(&_gvn);
5957 const TypeAryPtr* top_src = src_type->isa_aryptr();
5958 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5959 assert (top_src != NULL && top_src->klass() != NULL
5960 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5961
5962 // checks are the responsibility of the caller
5963 Node* src_start = src;
5964 Node* dest_start = dest;
5965 if (src_offset != NULL || dest_offset != NULL) {
5966 assert(src_offset != NULL && dest_offset != NULL, "");
5967 src_start = array_element_address(src, src_offset, T_BYTE);
5968 dest_start = array_element_address(dest, dest_offset, T_BYTE);
5969 }
5970
5971 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
5972 // (because of the predicated logic executed earlier).
5973 // so we cast it here safely.
5974 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5975
5976 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
5977 if (embeddedCipherObj == NULL) return false;
5978
5979 // cast it to what we know it will be at runtime
5980 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
5981 assert(tinst != NULL, "CBC obj is null");
5982 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
5983 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
5984 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
5985
5986 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
5987 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
5988 const TypeOopPtr* xtype = aklass->as_instance_type();
5989 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
5990 aescrypt_object = _gvn.transform(aescrypt_object);
5991
5992 // we need to get the start of the aescrypt_object's expanded key array
5993 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5994 if (k_start == NULL) return false;
5995
5996 // similarly, get the start address of the r vector
5997 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
5998 if (objRvec == NULL) return false;
5999 objRvec = access_resolve(objRvec, ACCESS_WRITE);
6000 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6001
6002 Node* cbcCrypt;
6003 if (Matcher::pass_original_key_for_aes()) {
6004 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6005 // compatibility issues between Java key expansion and SPARC crypto instructions
6006 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6007 if (original_k_start == NULL) return false;
6008
6009 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
6010 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6011 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6012 stubAddr, stubName, TypePtr::BOTTOM,
6013 src_start, dest_start, k_start, r_start, len, original_k_start);
6014 } else {
6015 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6016 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6017 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6018 stubAddr, stubName, TypePtr::BOTTOM,
6019 src_start, dest_start, k_start, r_start, len);
6020 }
6021
6022 // return cipher length (int)
6023 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6024 set_result(retvalue);
6025 return true;
6026 }
6027
6028 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
6029 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
6030 address stubAddr = NULL;
6031 const char *stubName = NULL;
6032
6033 assert(UseAES, "need AES instruction support");
6034
6035 switch (id) {
6036 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
6037 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
6038 stubName = "electronicCodeBook_encryptAESCrypt";
6039 break;
6040 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
6041 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
6042 stubName = "electronicCodeBook_decryptAESCrypt";
6043 break;
6044 default:
6045 break;
6046 }
6047
6048 if (stubAddr == NULL) return false;
6049
6050 Node* electronicCodeBook_object = argument(0);
6051 Node* src = argument(1);
6052 Node* src_offset = argument(2);
6053 Node* len = argument(3);
6054 Node* dest = argument(4);
6055 Node* dest_offset = argument(5);
6056
6057 // (1) src and dest are arrays.
6058 const Type* src_type = src->Value(&_gvn);
6059 const Type* dest_type = dest->Value(&_gvn);
6060 const TypeAryPtr* top_src = src_type->isa_aryptr();
6061 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6062 assert(top_src != NULL && top_src->klass() != NULL
6063 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6064
6065 // checks are the responsibility of the caller
6066 Node* src_start = src;
6067 Node* dest_start = dest;
6068 if (src_offset != NULL || dest_offset != NULL) {
6069 assert(src_offset != NULL && dest_offset != NULL, "");
6070 src_start = array_element_address(src, src_offset, T_BYTE);
6071 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6072 }
6073
6074 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6075 // (because of the predicated logic executed earlier).
6076 // so we cast it here safely.
6077 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6078
6079 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6080 if (embeddedCipherObj == NULL) return false;
6081
6082 // cast it to what we know it will be at runtime
6083 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6084 assert(tinst != NULL, "ECB obj is null");
6085 assert(tinst->klass()->is_loaded(), "ECB obj is not loaded");
6086 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6087 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6088
6089 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6090 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6091 const TypeOopPtr* xtype = aklass->as_instance_type();
6092 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6093 aescrypt_object = _gvn.transform(aescrypt_object);
6094
6095 // we need to get the start of the aescrypt_object's expanded key array
6096 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6097 if (k_start == NULL) return false;
6098
6099 Node* ecbCrypt;
6100 if (Matcher::pass_original_key_for_aes()) {
6101 // no SPARC version for AES/ECB intrinsics now.
6102 return false;
6103 }
6104 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6105 ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6106 OptoRuntime::electronicCodeBook_aescrypt_Type(),
6107 stubAddr, stubName, TypePtr::BOTTOM,
6108 src_start, dest_start, k_start, len);
6109
6110 // return cipher length (int)
6111 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6112 set_result(retvalue);
6113 return true;
6114 }
6115
6116 //------------------------------inline_counterMode_AESCrypt-----------------------
6117 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6118 assert(UseAES, "need AES instruction support");
6119 if (!UseAESCTRIntrinsics) return false;
6120
6121 address stubAddr = NULL;
6122 const char *stubName = NULL;
6123 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6124 stubAddr = StubRoutines::counterMode_AESCrypt();
6125 stubName = "counterMode_AESCrypt";
6126 }
6127 if (stubAddr == NULL) return false;
6128
6129 Node* counterMode_object = argument(0);
6130 Node* src = argument(1);
6131 Node* src_offset = argument(2);
6132 Node* len = argument(3);
6133 Node* dest = argument(4);
6134 Node* dest_offset = argument(5);
6135
6136 src = access_resolve(src, ACCESS_READ);
6137 dest = access_resolve(dest, ACCESS_WRITE);
6138 counterMode_object = access_resolve(counterMode_object, ACCESS_WRITE);
6139
6140 // (1) src and dest are arrays.
6141 const Type* src_type = src->Value(&_gvn);
6142 const Type* dest_type = dest->Value(&_gvn);
6143 const TypeAryPtr* top_src = src_type->isa_aryptr();
6144 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6145 assert(top_src != NULL && top_src->klass() != NULL &&
6146 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6147
6148 // checks are the responsibility of the caller
6149 Node* src_start = src;
6150 Node* dest_start = dest;
6151 if (src_offset != NULL || dest_offset != NULL) {
6152 assert(src_offset != NULL && dest_offset != NULL, "");
6153 src_start = array_element_address(src, src_offset, T_BYTE);
6154 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6155 }
6156
6157 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6158 // (because of the predicated logic executed earlier).
6159 // so we cast it here safely.
6160 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6161 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6162 if (embeddedCipherObj == NULL) return false;
6163 // cast it to what we know it will be at runtime
6164 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6165 assert(tinst != NULL, "CTR obj is null");
6166 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6167 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6168 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6169 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6170 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6171 const TypeOopPtr* xtype = aklass->as_instance_type();
6172 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6173 aescrypt_object = _gvn.transform(aescrypt_object);
6174 // we need to get the start of the aescrypt_object's expanded key array
6175 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6176 if (k_start == NULL) return false;
6177 // similarly, get the start address of the r vector
6178 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6179 if (obj_counter == NULL) return false;
6180 obj_counter = access_resolve(obj_counter, ACCESS_WRITE);
6181 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6182
6183 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6184 if (saved_encCounter == NULL) return false;
6185 saved_encCounter = access_resolve(saved_encCounter, ACCESS_WRITE);
6186 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6187 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6188
6189 Node* ctrCrypt;
6190 if (Matcher::pass_original_key_for_aes()) {
6191 // no SPARC version for AES/CTR intrinsics now.
6192 return false;
6193 }
6194 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6195 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6196 OptoRuntime::counterMode_aescrypt_Type(),
6197 stubAddr, stubName, TypePtr::BOTTOM,
6198 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6199
6200 // return cipher length (int)
6201 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6202 set_result(retvalue);
6203 return true;
6204 }
6205
6206 //------------------------------get_key_start_from_aescrypt_object-----------------------
6207 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6208 #if defined(PPC64) || defined(S390)
6209 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6210 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6211 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6212 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6213 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6214 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6215 if (objSessionK == NULL) {
6216 return (Node *) NULL;
6217 }
6218 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6219 #else
6220 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6221 #endif // PPC64
6222 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6223 if (objAESCryptKey == NULL) return (Node *) NULL;
6224
6225 // now have the array, need to get the start address of the K array
6226 objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ);
6227 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6228 return k_start;
6229 }
6230
6231 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6232 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6233 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6234 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6235 if (objAESCryptKey == NULL) return (Node *) NULL;
6236
6237 // now have the array, need to get the start address of the lastKey array
6238 objAESCryptKey = access_resolve(objAESCryptKey, ACCESS_READ);
6239 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6240 return original_k_start;
6241 }
6242
6243 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6244 // Return node representing slow path of predicate check.
6245 // the pseudo code we want to emulate with this predicate is:
6246 // for encryption:
6247 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6248 // for decryption:
6249 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6250 // note cipher==plain is more conservative than the original java code but that's OK
6251 //
6252 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6253 // The receiver was checked for NULL already.
6254 Node* objCBC = argument(0);
6255
6256 Node* src = argument(1);
6257 Node* dest = argument(4);
6258
6259 // Load embeddedCipher field of CipherBlockChaining object.
6260 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6261
6262 // get AESCrypt klass for instanceOf check
6263 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6264 // will have same classloader as CipherBlockChaining object
6265 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6266 assert(tinst != NULL, "CBCobj is null");
6267 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6268
6269 // we want to do an instanceof comparison against the AESCrypt class
6270 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6271 if (!klass_AESCrypt->is_loaded()) {
6272 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6273 Node* ctrl = control();
6274 set_control(top()); // no regular fast path
6275 return ctrl;
6276 }
6277
6278 src = must_be_not_null(src, true);
6279 dest = must_be_not_null(dest, true);
6280
6281 // Resolve oops to stable for CmpP below.
6282 src = access_resolve(src, 0);
6283 dest = access_resolve(dest, 0);
6284
6285 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6286
6287 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6288 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6289 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6290
6291 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6292
6293 // for encryption, we are done
6294 if (!decrypting)
6295 return instof_false; // even if it is NULL
6296
6297 // for decryption, we need to add a further check to avoid
6298 // taking the intrinsic path when cipher and plain are the same
6299 // see the original java code for why.
6300 RegionNode* region = new RegionNode(3);
6301 region->init_req(1, instof_false);
6302
6303 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6304 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6305 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6306 region->init_req(2, src_dest_conjoint);
6307
6308 record_for_igvn(region);
6309 return _gvn.transform(region);
6310 }
6311
6312 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
6313 // Return node representing slow path of predicate check.
6314 // the pseudo code we want to emulate with this predicate is:
6315 // for encryption:
6316 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6317 // for decryption:
6318 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6319 // note cipher==plain is more conservative than the original java code but that's OK
6320 //
6321 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
6322 // The receiver was checked for NULL already.
6323 Node* objECB = argument(0);
6324
6325 // Load embeddedCipher field of ElectronicCodeBook object.
6326 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6327
6328 // get AESCrypt klass for instanceOf check
6329 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6330 // will have same classloader as ElectronicCodeBook object
6331 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
6332 assert(tinst != NULL, "ECBobj is null");
6333 assert(tinst->klass()->is_loaded(), "ECBobj is not loaded");
6334
6335 // we want to do an instanceof comparison against the AESCrypt class
6336 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6337 if (!klass_AESCrypt->is_loaded()) {
6338 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6339 Node* ctrl = control();
6340 set_control(top()); // no regular fast path
6341 return ctrl;
6342 }
6343 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6344
6345 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6346 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6347 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6348
6349 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6350
6351 // for encryption, we are done
6352 if (!decrypting)
6353 return instof_false; // even if it is NULL
6354
6355 // for decryption, we need to add a further check to avoid
6356 // taking the intrinsic path when cipher and plain are the same
6357 // see the original java code for why.
6358 RegionNode* region = new RegionNode(3);
6359 region->init_req(1, instof_false);
6360 Node* src = argument(1);
6361 Node* dest = argument(4);
6362 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6363 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6364 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6365 region->init_req(2, src_dest_conjoint);
6366
6367 record_for_igvn(region);
6368 return _gvn.transform(region);
6369 }
6370
6371 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6372 // Return node representing slow path of predicate check.
6373 // the pseudo code we want to emulate with this predicate is:
6374 // for encryption:
6375 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6376 // for decryption:
6377 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6378 // note cipher==plain is more conservative than the original java code but that's OK
6379 //
6380
6381 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6382 // The receiver was checked for NULL already.
6383 Node* objCTR = argument(0);
6384
6385 // Load embeddedCipher field of CipherBlockChaining object.
6386 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6387
6388 // get AESCrypt klass for instanceOf check
6389 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6390 // will have same classloader as CipherBlockChaining object
6391 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6392 assert(tinst != NULL, "CTRobj is null");
6393 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6394
6395 // we want to do an instanceof comparison against the AESCrypt class
6396 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6397 if (!klass_AESCrypt->is_loaded()) {
6398 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6399 Node* ctrl = control();
6400 set_control(top()); // no regular fast path
6401 return ctrl;
6402 }
6403
6404 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6405 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6406 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6407 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6408 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6409
6410 return instof_false; // even if it is NULL
6411 }
6412
6413 //------------------------------inline_ghash_processBlocks
6414 bool LibraryCallKit::inline_ghash_processBlocks() {
6415 address stubAddr;
6416 const char *stubName;
6417 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6418
6419 stubAddr = StubRoutines::ghash_processBlocks();
6420 stubName = "ghash_processBlocks";
6421
6422 Node* data = argument(0);
6423 Node* offset = argument(1);
6424 Node* len = argument(2);
6425 Node* state = argument(3);
6426 Node* subkeyH = argument(4);
6427
6428 state = must_be_not_null(state, true);
6429 subkeyH = must_be_not_null(subkeyH, true);
6430 data = must_be_not_null(data, true);
6431
6432 state = access_resolve(state, ACCESS_WRITE);
6433 subkeyH = access_resolve(subkeyH, ACCESS_READ);
6434 data = access_resolve(data, ACCESS_READ);
6435
6436 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6437 assert(state_start, "state is NULL");
6438 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6439 assert(subkeyH_start, "subkeyH is NULL");
6440 Node* data_start = array_element_address(data, offset, T_BYTE);
6441 assert(data_start, "data is NULL");
6442
6443 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6444 OptoRuntime::ghash_processBlocks_Type(),
6445 stubAddr, stubName, TypePtr::BOTTOM,
6446 state_start, subkeyH_start, data_start, len);
6447 return true;
6448 }
6449
6450 bool LibraryCallKit::inline_base64_encodeBlock() {
6451 address stubAddr;
6452 const char *stubName;
6453 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6454 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6455 stubAddr = StubRoutines::base64_encodeBlock();
6456 stubName = "encodeBlock";
6457
6458 if (!stubAddr) return false;
6459 Node* base64obj = argument(0);
6460 Node* src = argument(1);
6461 Node* offset = argument(2);
6462 Node* len = argument(3);
6463 Node* dest = argument(4);
6464 Node* dp = argument(5);
6465 Node* isURL = argument(6);
6466
6467 src = must_be_not_null(src, true);
6468 src = access_resolve(src, ACCESS_READ);
6469 dest = must_be_not_null(dest, true);
6470 dest = access_resolve(dest, ACCESS_WRITE);
6471
6472 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6473 assert(src_start, "source array is NULL");
6474 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6475 assert(dest_start, "destination array is NULL");
6476
6477 Node* base64 = make_runtime_call(RC_LEAF,
6478 OptoRuntime::base64_encodeBlock_Type(),
6479 stubAddr, stubName, TypePtr::BOTTOM,
6480 src_start, offset, len, dest_start, dp, isURL);
6481 return true;
6482 }
6483
6484 //------------------------------inline_sha_implCompress-----------------------
6485 //
6486 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6487 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6488 //
6489 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6490 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6491 //
6492 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6493 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6494 //
6495 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6496 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6497
6498 Node* sha_obj = argument(0);
6499 Node* src = argument(1); // type oop
6500 Node* ofs = argument(2); // type int
6501
6502 const Type* src_type = src->Value(&_gvn);
6503 const TypeAryPtr* top_src = src_type->isa_aryptr();
6504 if (top_src == NULL || top_src->klass() == NULL) {
6505 // failed array check
6506 return false;
6507 }
6508 // Figure out the size and type of the elements we will be copying.
6509 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6510 if (src_elem != T_BYTE) {
6511 return false;
6512 }
6513 // 'src_start' points to src array + offset
6514 src = must_be_not_null(src, true);
6515 src = access_resolve(src, ACCESS_READ);
6516 Node* src_start = array_element_address(src, ofs, src_elem);
6517 Node* state = NULL;
6518 address stubAddr;
6519 const char *stubName;
6520
6521 switch(id) {
6522 case vmIntrinsics::_sha_implCompress:
6523 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6524 state = get_state_from_sha_object(sha_obj);
6525 stubAddr = StubRoutines::sha1_implCompress();
6526 stubName = "sha1_implCompress";
6527 break;
6528 case vmIntrinsics::_sha2_implCompress:
6529 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6530 state = get_state_from_sha_object(sha_obj);
6531 stubAddr = StubRoutines::sha256_implCompress();
6532 stubName = "sha256_implCompress";
6533 break;
6534 case vmIntrinsics::_sha5_implCompress:
6535 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6536 state = get_state_from_sha5_object(sha_obj);
6537 stubAddr = StubRoutines::sha512_implCompress();
6538 stubName = "sha512_implCompress";
6539 break;
6540 default:
6541 fatal_unexpected_iid(id);
6542 return false;
6543 }
6544 if (state == NULL) return false;
6545
6546 assert(stubAddr != NULL, "Stub is generated");
6547 if (stubAddr == NULL) return false;
6548
6549 // Call the stub.
6550 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6551 stubAddr, stubName, TypePtr::BOTTOM,
6552 src_start, state);
6553
6554 return true;
6555 }
6556
6557 //------------------------------inline_digestBase_implCompressMB-----------------------
6558 //
6559 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6560 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6561 //
6562 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6563 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6564 "need SHA1/SHA256/SHA512 instruction support");
6565 assert((uint)predicate < 3, "sanity");
6566 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6567
6568 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6569 Node* src = argument(1); // byte[] array
6570 Node* ofs = argument(2); // type int
6571 Node* limit = argument(3); // type int
6572
6573 const Type* src_type = src->Value(&_gvn);
6574 const TypeAryPtr* top_src = src_type->isa_aryptr();
6575 if (top_src == NULL || top_src->klass() == NULL) {
6576 // failed array check
6577 return false;
6578 }
6579 // Figure out the size and type of the elements we will be copying.
6580 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6581 if (src_elem != T_BYTE) {
6582 return false;
6583 }
6584 // 'src_start' points to src array + offset
6585 src = must_be_not_null(src, false);
6586 src = access_resolve(src, ACCESS_READ);
6587 Node* src_start = array_element_address(src, ofs, src_elem);
6588
6589 const char* klass_SHA_name = NULL;
6590 const char* stub_name = NULL;
6591 address stub_addr = NULL;
6592 bool long_state = false;
6593
6594 switch (predicate) {
6595 case 0:
6596 if (UseSHA1Intrinsics) {
6597 klass_SHA_name = "sun/security/provider/SHA";
6598 stub_name = "sha1_implCompressMB";
6599 stub_addr = StubRoutines::sha1_implCompressMB();
6600 }
6601 break;
6602 case 1:
6603 if (UseSHA256Intrinsics) {
6604 klass_SHA_name = "sun/security/provider/SHA2";
6605 stub_name = "sha256_implCompressMB";
6606 stub_addr = StubRoutines::sha256_implCompressMB();
6607 }
6608 break;
6609 case 2:
6610 if (UseSHA512Intrinsics) {
6611 klass_SHA_name = "sun/security/provider/SHA5";
6612 stub_name = "sha512_implCompressMB";
6613 stub_addr = StubRoutines::sha512_implCompressMB();
6614 long_state = true;
6615 }
6616 break;
6617 default:
6618 fatal("unknown SHA intrinsic predicate: %d", predicate);
6619 }
6620 if (klass_SHA_name != NULL) {
6621 assert(stub_addr != NULL, "Stub is generated");
6622 if (stub_addr == NULL) return false;
6623
6624 // get DigestBase klass to lookup for SHA klass
6625 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6626 assert(tinst != NULL, "digestBase_obj is not instance???");
6627 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6628
6629 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6630 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6631 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6632 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6633 }
6634 return false;
6635 }
6636 //------------------------------inline_sha_implCompressMB-----------------------
6637 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6638 bool long_state, address stubAddr, const char *stubName,
6639 Node* src_start, Node* ofs, Node* limit) {
6640 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6641 const TypeOopPtr* xtype = aklass->as_instance_type();
6642 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6643 sha_obj = _gvn.transform(sha_obj);
6644
6645 Node* state;
6646 if (long_state) {
6647 state = get_state_from_sha5_object(sha_obj);
6648 } else {
6649 state = get_state_from_sha_object(sha_obj);
6650 }
6651 if (state == NULL) return false;
6652
6653 // Call the stub.
6654 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6655 OptoRuntime::digestBase_implCompressMB_Type(),
6656 stubAddr, stubName, TypePtr::BOTTOM,
6657 src_start, state, ofs, limit);
6658 // return ofs (int)
6659 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6660 set_result(result);
6661
6662 return true;
6663 }
6664
6665 //------------------------------get_state_from_sha_object-----------------------
6666 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6667 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6668 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6669 if (sha_state == NULL) return (Node *) NULL;
6670
6671 // now have the array, need to get the start address of the state array
6672 sha_state = access_resolve(sha_state, ACCESS_WRITE);
6673 Node* state = array_element_address(sha_state, intcon(0), T_INT);
6674 return state;
6675 }
6676
6677 //------------------------------get_state_from_sha5_object-----------------------
6678 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6679 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6680 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6681 if (sha_state == NULL) return (Node *) NULL;
6682
6683 // now have the array, need to get the start address of the state array
6684 sha_state = access_resolve(sha_state, ACCESS_WRITE);
6685 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6686 return state;
6687 }
6688
6689 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6690 // Return node representing slow path of predicate check.
6691 // the pseudo code we want to emulate with this predicate is:
6692 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6693 //
6694 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6695 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6696 "need SHA1/SHA256/SHA512 instruction support");
6697 assert((uint)predicate < 3, "sanity");
6698
6699 // The receiver was checked for NULL already.
6700 Node* digestBaseObj = argument(0);
6701
6702 // get DigestBase klass for instanceOf check
6703 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6704 assert(tinst != NULL, "digestBaseObj is null");
6705 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6706
6707 const char* klass_SHA_name = NULL;
6708 switch (predicate) {
6709 case 0:
6710 if (UseSHA1Intrinsics) {
6711 // we want to do an instanceof comparison against the SHA class
6712 klass_SHA_name = "sun/security/provider/SHA";
6713 }
6714 break;
6715 case 1:
6716 if (UseSHA256Intrinsics) {
6717 // we want to do an instanceof comparison against the SHA2 class
6718 klass_SHA_name = "sun/security/provider/SHA2";
6719 }
6720 break;
6721 case 2:
6722 if (UseSHA512Intrinsics) {
6723 // we want to do an instanceof comparison against the SHA5 class
6724 klass_SHA_name = "sun/security/provider/SHA5";
6725 }
6726 break;
6727 default:
6728 fatal("unknown SHA intrinsic predicate: %d", predicate);
6729 }
6730
6731 ciKlass* klass_SHA = NULL;
6732 if (klass_SHA_name != NULL) {
6733 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6734 }
6735 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6736 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6737 Node* ctrl = control();
6738 set_control(top()); // no intrinsic path
6739 return ctrl;
6740 }
6741 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6742
6743 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6744 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6745 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6746 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6747
6748 return instof_false; // even if it is NULL
6749 }
6750
6751 //-------------inline_fma-----------------------------------
6752 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
6753 Node *a = NULL;
6754 Node *b = NULL;
6755 Node *c = NULL;
6756 Node* result = NULL;
6757 switch (id) {
6758 case vmIntrinsics::_fmaD:
6759 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
6760 // no receiver since it is static method
6761 a = round_double_node(argument(0));
6762 b = round_double_node(argument(2));
6763 c = round_double_node(argument(4));
6764 result = _gvn.transform(new FmaDNode(control(), a, b, c));
6765 break;
6766 case vmIntrinsics::_fmaF:
6767 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
6768 a = argument(0);
6769 b = argument(1);
6770 c = argument(2);
6771 result = _gvn.transform(new FmaFNode(control(), a, b, c));
6772 break;
6773 default:
6774 fatal_unexpected_iid(id); break;
6775 }
6776 set_result(result);
6777 return true;
6778 }
6779
6780 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
6781 // argument(0) is receiver
6782 Node* codePoint = argument(1);
6783 Node* n = NULL;
6784
6785 switch (id) {
6786 case vmIntrinsics::_isDigit :
6787 n = new DigitNode(control(), codePoint);
6788 break;
6789 case vmIntrinsics::_isLowerCase :
6790 n = new LowerCaseNode(control(), codePoint);
6791 break;
6792 case vmIntrinsics::_isUpperCase :
6793 n = new UpperCaseNode(control(), codePoint);
6794 break;
6795 case vmIntrinsics::_isWhitespace :
6796 n = new WhitespaceNode(control(), codePoint);
6797 break;
6798 default:
6799 fatal_unexpected_iid(id);
6800 }
6801
6802 set_result(_gvn.transform(n));
6803 return true;
6804 }
6805
6806 //------------------------------inline_fp_min_max------------------------------
6807 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
6808 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
6809
6810 // The intrinsic should be used only when the API branches aren't predictable,
6811 // the last one performing the most important comparison. The following heuristic
6812 // uses the branch statistics to eventually bail out if necessary.
6813
6814 ciMethodData *md = callee()->method_data();
6815
6816 if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
6817 ciCallProfile cp = caller()->call_profile_at_bci(bci());
6818
6819 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
6820 // Bail out if the call-site didn't contribute enough to the statistics.
6821 return false;
6822 }
6823
6824 uint taken = 0, not_taken = 0;
6825
6826 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
6827 if (p->is_BranchData()) {
6828 taken = ((ciBranchData*)p)->taken();
6829 not_taken = ((ciBranchData*)p)->not_taken();
6830 }
6831 }
6832
6833 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
6834 balance = balance < 0 ? -balance : balance;
6835 if ( balance > 0.2 ) {
6836 // Bail out if the most important branch is predictable enough.
6837 return false;
6838 }
6839 }
6840 */
6841
6842 Node *a = NULL;
6843 Node *b = NULL;
6844 Node *n = NULL;
6845 switch (id) {
6846 case vmIntrinsics::_maxF:
6847 case vmIntrinsics::_minF:
6848 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
6849 a = argument(0);
6850 b = argument(1);
6851 break;
6852 case vmIntrinsics::_maxD:
6853 case vmIntrinsics::_minD:
6854 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
6855 a = round_double_node(argument(0));
6856 b = round_double_node(argument(2));
6857 break;
6858 default:
6859 fatal_unexpected_iid(id);
6860 break;
6861 }
6862 switch (id) {
6863 case vmIntrinsics::_maxF: n = new MaxFNode(a, b); break;
6864 case vmIntrinsics::_minF: n = new MinFNode(a, b); break;
6865 case vmIntrinsics::_maxD: n = new MaxDNode(a, b); break;
6866 case vmIntrinsics::_minD: n = new MinDNode(a, b); break;
6867 default: fatal_unexpected_iid(id); break;
6868 }
6869 set_result(_gvn.transform(n));
6870 return true;
6871 }
6872
6873 bool LibraryCallKit::inline_profileBoolean() {
6874 Node* counts = argument(1);
6875 const TypeAryPtr* ary = NULL;
6876 ciArray* aobj = NULL;
6877 if (counts->is_Con()
6878 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6879 && (aobj = ary->const_oop()->as_array()) != NULL
6880 && (aobj->length() == 2)) {
6881 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6882 jint false_cnt = aobj->element_value(0).as_int();
6883 jint true_cnt = aobj->element_value(1).as_int();
6884
6885 if (C->log() != NULL) {
6886 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6887 false_cnt, true_cnt);
6888 }
6889
6890 if (false_cnt + true_cnt == 0) {
6891 // According to profile, never executed.
6892 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6893 Deoptimization::Action_reinterpret);
6894 return true;
6895 }
6896
6897 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6898 // is a number of each value occurrences.
6899 Node* result = argument(0);
6900 if (false_cnt == 0 || true_cnt == 0) {
6901 // According to profile, one value has been never seen.
6902 int expected_val = (false_cnt == 0) ? 1 : 0;
6903
6904 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6905 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6906
6907 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6908 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6909 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6910
6911 { // Slow path: uncommon trap for never seen value and then reexecute
6912 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6913 // the value has been seen at least once.
6914 PreserveJVMState pjvms(this);
6915 PreserveReexecuteState preexecs(this);
6916 jvms()->set_should_reexecute(true);
6917
6918 set_control(slow_path);
6919 set_i_o(i_o());
6920
6921 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6922 Deoptimization::Action_reinterpret);
6923 }
6924 // The guard for never seen value enables sharpening of the result and
6925 // returning a constant. It allows to eliminate branches on the same value
6926 // later on.
6927 set_control(fast_path);
6928 result = intcon(expected_val);
6929 }
6930 // Stop profiling.
6931 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6932 // By replacing method body with profile data (represented as ProfileBooleanNode
6933 // on IR level) we effectively disable profiling.
6934 // It enables full speed execution once optimized code is generated.
6935 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6936 C->record_for_igvn(profile);
6937 set_result(profile);
6938 return true;
6939 } else {
6940 // Continue profiling.
6941 // Profile data isn't available at the moment. So, execute method's bytecode version.
6942 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6943 // is compiled and counters aren't available since corresponding MethodHandle
6944 // isn't a compile-time constant.
6945 return false;
6946 }
6947 }
6948
6949 bool LibraryCallKit::inline_isCompileConstant() {
6950 Node* n = argument(0);
6951 set_result(n->is_Con() ? intcon(1) : intcon(0));
6952 return true;
6953 }