1 /*
2 * Copyright (c) 1999, 2023, 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/vmIntrinsics.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "gc/shared/barrierSet.hpp"
32 #include "jfr/support/jfrIntrinsics.hpp"
33 #include "memory/resourceArea.hpp"
34 #include "oops/klass.inline.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/castnode.hpp"
40 #include "opto/cfgnode.hpp"
41 #include "opto/convertnode.hpp"
42 #include "opto/countbitsnode.hpp"
43 #include "opto/idealKit.hpp"
44 #include "opto/library_call.hpp"
45 #include "opto/mathexactnode.hpp"
46 #include "opto/mulnode.hpp"
47 #include "opto/narrowptrnode.hpp"
48 #include "opto/opaquenode.hpp"
49 #include "opto/parse.hpp"
50 #include "opto/runtime.hpp"
51 #include "opto/rootnode.hpp"
52 #include "opto/subnode.hpp"
53 #include "prims/jvmtiExport.hpp"
54 #include "prims/jvmtiThreadState.hpp"
55 #include "prims/unsafe.hpp"
56 #include "runtime/jniHandles.inline.hpp"
57 #include "runtime/objectMonitor.hpp"
58 #include "runtime/sharedRuntime.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "utilities/macros.hpp"
61 #include "utilities/powerOfTwo.hpp"
62
63 //---------------------------make_vm_intrinsic----------------------------
64 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
65 vmIntrinsicID id = m->intrinsic_id();
66 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
67
68 if (!m->is_loaded()) {
69 // Do not attempt to inline unloaded methods.
70 return nullptr;
71 }
72
73 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
74 bool is_available = false;
75
76 {
77 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
78 // the compiler must transition to '_thread_in_vm' state because both
79 // methods access VM-internal data.
80 VM_ENTRY_MARK;
81 methodHandle mh(THREAD, m->get_Method());
82 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
83 if (is_available && is_virtual) {
84 is_available = vmIntrinsics::does_virtual_dispatch(id);
85 }
86 }
87
88 if (is_available) {
89 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
90 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
91 return new LibraryIntrinsic(m, is_virtual,
92 vmIntrinsics::predicates_needed(id),
93 vmIntrinsics::does_virtual_dispatch(id),
94 id);
95 } else {
96 return nullptr;
97 }
98 }
99
100 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
101 LibraryCallKit kit(jvms, this);
102 Compile* C = kit.C;
103 int nodes = C->unique();
104 #ifndef PRODUCT
105 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
106 char buf[1000];
107 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
108 tty->print_cr("Intrinsic %s", str);
109 }
110 #endif
111 ciMethod* callee = kit.callee();
112 const int bci = kit.bci();
113 #ifdef ASSERT
114 Node* ctrl = kit.control();
115 #endif
116 // Try to inline the intrinsic.
117 if (callee->check_intrinsic_candidate() &&
118 kit.try_to_inline(_last_predicate)) {
119 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
120 : "(intrinsic)";
121 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
122 if (C->print_intrinsics() || C->print_inlining()) {
123 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
124 }
125 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
126 if (C->log()) {
127 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
128 vmIntrinsics::name_at(intrinsic_id()),
129 (is_virtual() ? " virtual='1'" : ""),
130 C->unique() - nodes);
131 }
132 // Push the result from the inlined method onto the stack.
133 kit.push_result();
134 C->print_inlining_update(this);
135 return kit.transfer_exceptions_into_jvms();
136 }
137
138 // The intrinsic bailed out
139 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
140 if (jvms->has_method()) {
141 // Not a root compile.
142 const char* msg;
143 if (callee->intrinsic_candidate()) {
144 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
145 } else {
146 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
147 : "failed to inline (intrinsic), method not annotated";
148 }
149 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
150 if (C->print_intrinsics() || C->print_inlining()) {
151 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
152 }
153 } else {
154 // Root compile
155 ResourceMark rm;
156 stringStream msg_stream;
157 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
158 vmIntrinsics::name_at(intrinsic_id()),
159 is_virtual() ? " (virtual)" : "", bci);
160 const char *msg = msg_stream.freeze();
161 log_debug(jit, inlining)("%s", msg);
162 if (C->print_intrinsics() || C->print_inlining()) {
163 tty->print("%s", msg);
164 }
165 }
166 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
167 C->print_inlining_update(this);
168
169 return nullptr;
170 }
171
172 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
173 LibraryCallKit kit(jvms, this);
174 Compile* C = kit.C;
175 int nodes = C->unique();
176 _last_predicate = predicate;
177 #ifndef PRODUCT
178 assert(is_predicated() && predicate < predicates_count(), "sanity");
179 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
180 char buf[1000];
181 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
182 tty->print_cr("Predicate for intrinsic %s", str);
183 }
184 #endif
185 ciMethod* callee = kit.callee();
186 const int bci = kit.bci();
187
188 Node* slow_ctl = kit.try_to_predicate(predicate);
189 if (!kit.failing()) {
190 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
191 : "(intrinsic, predicate)";
192 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
193 if (C->print_intrinsics() || C->print_inlining()) {
194 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
195 }
196 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
197 if (C->log()) {
198 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
199 vmIntrinsics::name_at(intrinsic_id()),
200 (is_virtual() ? " virtual='1'" : ""),
201 C->unique() - nodes);
202 }
203 return slow_ctl; // Could be null if the check folds.
204 }
205
206 // The intrinsic bailed out
207 if (jvms->has_method()) {
208 // Not a root compile.
209 const char* msg = "failed to generate predicate for intrinsic";
210 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
211 if (C->print_intrinsics() || C->print_inlining()) {
212 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
213 }
214 } else {
215 // Root compile
216 ResourceMark rm;
217 stringStream msg_stream;
218 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
219 vmIntrinsics::name_at(intrinsic_id()),
220 is_virtual() ? " (virtual)" : "", bci);
221 const char *msg = msg_stream.freeze();
222 log_debug(jit, inlining)("%s", msg);
223 if (C->print_intrinsics() || C->print_inlining()) {
224 C->print_inlining_stream()->print("%s", msg);
225 }
226 }
227 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
228 return nullptr;
229 }
230
231 bool LibraryCallKit::try_to_inline(int predicate) {
232 // Handle symbolic names for otherwise undistinguished boolean switches:
233 const bool is_store = true;
234 const bool is_compress = true;
235 const bool is_static = true;
236 const bool is_volatile = true;
237
238 if (!jvms()->has_method()) {
239 // Root JVMState has a null method.
240 assert(map()->memory()->Opcode() == Op_Parm, "");
241 // Insert the memory aliasing node
242 set_all_memory(reset_memory());
243 }
244 assert(merged_memory(), "");
245
246 switch (intrinsic_id()) {
247 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
248 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
249 case vmIntrinsics::_getClass: return inline_native_getClass();
250
251 case vmIntrinsics::_ceil:
252 case vmIntrinsics::_floor:
253 case vmIntrinsics::_rint:
254 case vmIntrinsics::_dsin:
255 case vmIntrinsics::_dcos:
256 case vmIntrinsics::_dtan:
257 case vmIntrinsics::_dabs:
258 case vmIntrinsics::_fabs:
259 case vmIntrinsics::_iabs:
260 case vmIntrinsics::_labs:
261 case vmIntrinsics::_datan2:
262 case vmIntrinsics::_dsqrt:
263 case vmIntrinsics::_dsqrt_strict:
264 case vmIntrinsics::_dexp:
265 case vmIntrinsics::_dlog:
266 case vmIntrinsics::_dlog10:
267 case vmIntrinsics::_dpow:
268 case vmIntrinsics::_dcopySign:
269 case vmIntrinsics::_fcopySign:
270 case vmIntrinsics::_dsignum:
271 case vmIntrinsics::_roundF:
272 case vmIntrinsics::_roundD:
273 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
274
275 case vmIntrinsics::_notify:
276 case vmIntrinsics::_notifyAll:
277 return inline_notify(intrinsic_id());
278
279 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
280 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
281 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
282 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
283 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
284 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
285 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
286 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
287 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
288 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
289 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
290 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
291 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
292 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
293
294 case vmIntrinsics::_arraycopy: return inline_arraycopy();
295
296 case vmIntrinsics::_arraySort: return inline_array_sort();
297 case vmIntrinsics::_arrayPartition: return inline_array_partition();
298
299 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
300 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
301 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
302 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
303
304 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
305 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
306 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
307 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
308 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
309 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
310 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
311 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
312
313 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
314 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
315
316 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
317
318 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
319 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
320 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
321 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
322
323 case vmIntrinsics::_compressStringC:
324 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
325 case vmIntrinsics::_inflateStringC:
326 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
327
328 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
329 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
330 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
331 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
332 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
333 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
334 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
335 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
336 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
337
338 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
339 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
340 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
341 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
342 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
343 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
344 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
345 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
346 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
347
348 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
349 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
350 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
351 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
352 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
353 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
354 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
355 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
356 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
357
358 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
359 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
360 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
361 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
362 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
363 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
364 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
365 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
366 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
367
368 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
369 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
370 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
371 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
372
373 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
374 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
375 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
376 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
377
378 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
379 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
380 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
381 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
382 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
383 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
384 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
385 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
386 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
387
388 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
389 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
390 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
391 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
392 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
393 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
394 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
395 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
396 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
397
398 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
399 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
400 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
401 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
402 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
403 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
404 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
405 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
406 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
407
408 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
409 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
410 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
411 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
412 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
413 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
414 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
415 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
416 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
417
418 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
419 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
420 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
421 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
422 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
423
424 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
425 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
426 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
427 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
428 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
429 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
430 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
431 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
432 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
433 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
434 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
435 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
436 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
437 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
438 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
439 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
440 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
441 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
442 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
443 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
444
445 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
446 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
447 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
448 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
449 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
450 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
451 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
452 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
453 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
454 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
455 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
456 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
457 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
458 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
459 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
460
461 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
462 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
463 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
464 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
465
466 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
467 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
468 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
469 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
470 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
471
472 case vmIntrinsics::_loadFence:
473 case vmIntrinsics::_storeFence:
474 case vmIntrinsics::_storeStoreFence:
475 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
476
477 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
478
479 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
480 case vmIntrinsics::_currentThread: return inline_native_currentThread();
481 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
482
483 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
484 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
485
486 #if INCLUDE_JVMTI
487 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
488 "notifyJvmtiStart", true, false);
489 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
490 "notifyJvmtiEnd", false, true);
491 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
492 "notifyJvmtiMount", false, false);
493 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
494 "notifyJvmtiUnmount", false, false);
495 case vmIntrinsics::_notifyJvmtiVThreadHideFrames: return inline_native_notify_jvmti_hide();
496 #endif
497
498 #ifdef JFR_HAVE_INTRINSICS
499 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
500 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
501 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
502 #endif
503 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
504 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
505 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
506 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
507 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
508 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
509 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
510 case vmIntrinsics::_getLength: return inline_native_getLength();
511 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
512 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
513 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
514 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
515 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
516 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
517 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
518
519 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
520 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
521
522 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
523
524 case vmIntrinsics::_isInstance:
525 case vmIntrinsics::_getModifiers:
526 case vmIntrinsics::_isInterface:
527 case vmIntrinsics::_isArray:
528 case vmIntrinsics::_isPrimitive:
529 case vmIntrinsics::_isHidden:
530 case vmIntrinsics::_getSuperclass:
531 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
532
533 case vmIntrinsics::_floatToRawIntBits:
534 case vmIntrinsics::_floatToIntBits:
535 case vmIntrinsics::_intBitsToFloat:
536 case vmIntrinsics::_doubleToRawLongBits:
537 case vmIntrinsics::_doubleToLongBits:
538 case vmIntrinsics::_longBitsToDouble:
539 case vmIntrinsics::_floatToFloat16:
540 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
541
542 case vmIntrinsics::_floatIsFinite:
543 case vmIntrinsics::_floatIsInfinite:
544 case vmIntrinsics::_doubleIsFinite:
545 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
546
547 case vmIntrinsics::_numberOfLeadingZeros_i:
548 case vmIntrinsics::_numberOfLeadingZeros_l:
549 case vmIntrinsics::_numberOfTrailingZeros_i:
550 case vmIntrinsics::_numberOfTrailingZeros_l:
551 case vmIntrinsics::_bitCount_i:
552 case vmIntrinsics::_bitCount_l:
553 case vmIntrinsics::_reverse_i:
554 case vmIntrinsics::_reverse_l:
555 case vmIntrinsics::_reverseBytes_i:
556 case vmIntrinsics::_reverseBytes_l:
557 case vmIntrinsics::_reverseBytes_s:
558 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
559
560 case vmIntrinsics::_compress_i:
561 case vmIntrinsics::_compress_l:
562 case vmIntrinsics::_expand_i:
563 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
564
565 case vmIntrinsics::_compareUnsigned_i:
566 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
567
568 case vmIntrinsics::_divideUnsigned_i:
569 case vmIntrinsics::_divideUnsigned_l:
570 case vmIntrinsics::_remainderUnsigned_i:
571 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
572
573 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
574
575 case vmIntrinsics::_Reference_get: return inline_reference_get();
576 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
577 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
578
579 case vmIntrinsics::_Class_cast: return inline_Class_cast();
580
581 case vmIntrinsics::_aescrypt_encryptBlock:
582 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
583
584 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
585 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
586 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
587
588 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
589 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
590 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
591
592 case vmIntrinsics::_counterMode_AESCrypt:
593 return inline_counterMode_AESCrypt(intrinsic_id());
594
595 case vmIntrinsics::_galoisCounterMode_AESCrypt:
596 return inline_galoisCounterMode_AESCrypt();
597
598 case vmIntrinsics::_md5_implCompress:
599 case vmIntrinsics::_sha_implCompress:
600 case vmIntrinsics::_sha2_implCompress:
601 case vmIntrinsics::_sha5_implCompress:
602 case vmIntrinsics::_sha3_implCompress:
603 return inline_digestBase_implCompress(intrinsic_id());
604
605 case vmIntrinsics::_digestBase_implCompressMB:
606 return inline_digestBase_implCompressMB(predicate);
607
608 case vmIntrinsics::_multiplyToLen:
609 return inline_multiplyToLen();
610
611 case vmIntrinsics::_squareToLen:
612 return inline_squareToLen();
613
614 case vmIntrinsics::_mulAdd:
615 return inline_mulAdd();
616
617 case vmIntrinsics::_montgomeryMultiply:
618 return inline_montgomeryMultiply();
619 case vmIntrinsics::_montgomerySquare:
620 return inline_montgomerySquare();
621
622 case vmIntrinsics::_bigIntegerRightShiftWorker:
623 return inline_bigIntegerShift(true);
624 case vmIntrinsics::_bigIntegerLeftShiftWorker:
625 return inline_bigIntegerShift(false);
626
627 case vmIntrinsics::_vectorizedMismatch:
628 return inline_vectorizedMismatch();
629
630 case vmIntrinsics::_ghash_processBlocks:
631 return inline_ghash_processBlocks();
632 case vmIntrinsics::_chacha20Block:
633 return inline_chacha20Block();
634 case vmIntrinsics::_base64_encodeBlock:
635 return inline_base64_encodeBlock();
636 case vmIntrinsics::_base64_decodeBlock:
637 return inline_base64_decodeBlock();
638 case vmIntrinsics::_poly1305_processBlocks:
639 return inline_poly1305_processBlocks();
640
641 case vmIntrinsics::_encodeISOArray:
642 case vmIntrinsics::_encodeByteISOArray:
643 return inline_encodeISOArray(false);
644 case vmIntrinsics::_encodeAsciiArray:
645 return inline_encodeISOArray(true);
646
647 case vmIntrinsics::_updateCRC32:
648 return inline_updateCRC32();
649 case vmIntrinsics::_updateBytesCRC32:
650 return inline_updateBytesCRC32();
651 case vmIntrinsics::_updateByteBufferCRC32:
652 return inline_updateByteBufferCRC32();
653
654 case vmIntrinsics::_updateBytesCRC32C:
655 return inline_updateBytesCRC32C();
656 case vmIntrinsics::_updateDirectByteBufferCRC32C:
657 return inline_updateDirectByteBufferCRC32C();
658
659 case vmIntrinsics::_updateBytesAdler32:
660 return inline_updateBytesAdler32();
661 case vmIntrinsics::_updateByteBufferAdler32:
662 return inline_updateByteBufferAdler32();
663
664 case vmIntrinsics::_profileBoolean:
665 return inline_profileBoolean();
666 case vmIntrinsics::_isCompileConstant:
667 return inline_isCompileConstant();
668
669 case vmIntrinsics::_countPositives:
670 return inline_countPositives();
671
672 case vmIntrinsics::_fmaD:
673 case vmIntrinsics::_fmaF:
674 return inline_fma(intrinsic_id());
675
676 case vmIntrinsics::_isDigit:
677 case vmIntrinsics::_isLowerCase:
678 case vmIntrinsics::_isUpperCase:
679 case vmIntrinsics::_isWhitespace:
680 return inline_character_compare(intrinsic_id());
681
682 case vmIntrinsics::_min:
683 case vmIntrinsics::_max:
684 case vmIntrinsics::_min_strict:
685 case vmIntrinsics::_max_strict:
686 return inline_min_max(intrinsic_id());
687
688 case vmIntrinsics::_maxF:
689 case vmIntrinsics::_minF:
690 case vmIntrinsics::_maxD:
691 case vmIntrinsics::_minD:
692 case vmIntrinsics::_maxF_strict:
693 case vmIntrinsics::_minF_strict:
694 case vmIntrinsics::_maxD_strict:
695 case vmIntrinsics::_minD_strict:
696 return inline_fp_min_max(intrinsic_id());
697
698 case vmIntrinsics::_VectorUnaryOp:
699 return inline_vector_nary_operation(1);
700 case vmIntrinsics::_VectorBinaryOp:
701 return inline_vector_nary_operation(2);
702 case vmIntrinsics::_VectorTernaryOp:
703 return inline_vector_nary_operation(3);
704 case vmIntrinsics::_VectorFromBitsCoerced:
705 return inline_vector_frombits_coerced();
706 case vmIntrinsics::_VectorShuffleIota:
707 return inline_vector_shuffle_iota();
708 case vmIntrinsics::_VectorMaskOp:
709 return inline_vector_mask_operation();
710 case vmIntrinsics::_VectorShuffleToVector:
711 return inline_vector_shuffle_to_vector();
712 case vmIntrinsics::_VectorLoadOp:
713 return inline_vector_mem_operation(/*is_store=*/false);
714 case vmIntrinsics::_VectorLoadMaskedOp:
715 return inline_vector_mem_masked_operation(/*is_store*/false);
716 case vmIntrinsics::_VectorStoreOp:
717 return inline_vector_mem_operation(/*is_store=*/true);
718 case vmIntrinsics::_VectorStoreMaskedOp:
719 return inline_vector_mem_masked_operation(/*is_store=*/true);
720 case vmIntrinsics::_VectorGatherOp:
721 return inline_vector_gather_scatter(/*is_scatter*/ false);
722 case vmIntrinsics::_VectorScatterOp:
723 return inline_vector_gather_scatter(/*is_scatter*/ true);
724 case vmIntrinsics::_VectorReductionCoerced:
725 return inline_vector_reduction();
726 case vmIntrinsics::_VectorTest:
727 return inline_vector_test();
728 case vmIntrinsics::_VectorBlend:
729 return inline_vector_blend();
730 case vmIntrinsics::_VectorRearrange:
731 return inline_vector_rearrange();
732 case vmIntrinsics::_VectorCompare:
733 return inline_vector_compare();
734 case vmIntrinsics::_VectorBroadcastInt:
735 return inline_vector_broadcast_int();
736 case vmIntrinsics::_VectorConvert:
737 return inline_vector_convert();
738 case vmIntrinsics::_VectorInsert:
739 return inline_vector_insert();
740 case vmIntrinsics::_VectorExtract:
741 return inline_vector_extract();
742 case vmIntrinsics::_VectorCompressExpand:
743 return inline_vector_compress_expand();
744 case vmIntrinsics::_IndexVector:
745 return inline_index_vector();
746 case vmIntrinsics::_IndexPartiallyInUpperRange:
747 return inline_index_partially_in_upper_range();
748
749 case vmIntrinsics::_getObjectSize:
750 return inline_getObjectSize();
751
752 case vmIntrinsics::_blackhole:
753 return inline_blackhole();
754
755 default:
756 // If you get here, it may be that someone has added a new intrinsic
757 // to the list in vmIntrinsics.hpp without implementing it here.
758 #ifndef PRODUCT
759 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
760 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
761 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
762 }
763 #endif
764 return false;
765 }
766 }
767
768 Node* LibraryCallKit::try_to_predicate(int predicate) {
769 if (!jvms()->has_method()) {
770 // Root JVMState has a null method.
771 assert(map()->memory()->Opcode() == Op_Parm, "");
772 // Insert the memory aliasing node
773 set_all_memory(reset_memory());
774 }
775 assert(merged_memory(), "");
776
777 switch (intrinsic_id()) {
778 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
779 return inline_cipherBlockChaining_AESCrypt_predicate(false);
780 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
781 return inline_cipherBlockChaining_AESCrypt_predicate(true);
782 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
783 return inline_electronicCodeBook_AESCrypt_predicate(false);
784 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
785 return inline_electronicCodeBook_AESCrypt_predicate(true);
786 case vmIntrinsics::_counterMode_AESCrypt:
787 return inline_counterMode_AESCrypt_predicate();
788 case vmIntrinsics::_digestBase_implCompressMB:
789 return inline_digestBase_implCompressMB_predicate(predicate);
790 case vmIntrinsics::_galoisCounterMode_AESCrypt:
791 return inline_galoisCounterMode_AESCrypt_predicate();
792
793 default:
794 // If you get here, it may be that someone has added a new intrinsic
795 // to the list in vmIntrinsics.hpp without implementing it here.
796 #ifndef PRODUCT
797 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
798 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
799 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
800 }
801 #endif
802 Node* slow_ctl = control();
803 set_control(top()); // No fast path intrinsic
804 return slow_ctl;
805 }
806 }
807
808 //------------------------------set_result-------------------------------
809 // Helper function for finishing intrinsics.
810 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
811 record_for_igvn(region);
812 set_control(_gvn.transform(region));
813 set_result( _gvn.transform(value));
814 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
815 }
816
817 //------------------------------generate_guard---------------------------
818 // Helper function for generating guarded fast-slow graph structures.
819 // The given 'test', if true, guards a slow path. If the test fails
820 // then a fast path can be taken. (We generally hope it fails.)
821 // In all cases, GraphKit::control() is updated to the fast path.
822 // The returned value represents the control for the slow path.
823 // The return value is never 'top'; it is either a valid control
824 // or null if it is obvious that the slow path can never be taken.
825 // Also, if region and the slow control are not null, the slow edge
826 // is appended to the region.
827 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
828 if (stopped()) {
829 // Already short circuited.
830 return nullptr;
831 }
832
833 // Build an if node and its projections.
834 // If test is true we take the slow path, which we assume is uncommon.
835 if (_gvn.type(test) == TypeInt::ZERO) {
836 // The slow branch is never taken. No need to build this guard.
837 return nullptr;
838 }
839
840 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
841
842 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
843 if (if_slow == top()) {
844 // The slow branch is never taken. No need to build this guard.
845 return nullptr;
846 }
847
848 if (region != nullptr)
849 region->add_req(if_slow);
850
851 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
852 set_control(if_fast);
853
854 return if_slow;
855 }
856
857 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
858 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
859 }
860 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
861 return generate_guard(test, region, PROB_FAIR);
862 }
863
864 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
865 Node* *pos_index) {
866 if (stopped())
867 return nullptr; // already stopped
868 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
869 return nullptr; // index is already adequately typed
870 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
871 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
872 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
873 if (is_neg != nullptr && pos_index != nullptr) {
874 // Emulate effect of Parse::adjust_map_after_if.
875 Node* ccast = new CastIINode(index, TypeInt::POS);
876 ccast->set_req(0, control());
877 (*pos_index) = _gvn.transform(ccast);
878 }
879 return is_neg;
880 }
881
882 // Make sure that 'position' is a valid limit index, in [0..length].
883 // There are two equivalent plans for checking this:
884 // A. (offset + copyLength) unsigned<= arrayLength
885 // B. offset <= (arrayLength - copyLength)
886 // We require that all of the values above, except for the sum and
887 // difference, are already known to be non-negative.
888 // Plan A is robust in the face of overflow, if offset and copyLength
889 // are both hugely positive.
890 //
891 // Plan B is less direct and intuitive, but it does not overflow at
892 // all, since the difference of two non-negatives is always
893 // representable. Whenever Java methods must perform the equivalent
894 // check they generally use Plan B instead of Plan A.
895 // For the moment we use Plan A.
896 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
897 Node* subseq_length,
898 Node* array_length,
899 RegionNode* region) {
900 if (stopped())
901 return nullptr; // already stopped
902 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
903 if (zero_offset && subseq_length->eqv_uncast(array_length))
904 return nullptr; // common case of whole-array copy
905 Node* last = subseq_length;
906 if (!zero_offset) // last += offset
907 last = _gvn.transform(new AddINode(last, offset));
908 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
909 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
910 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
911 return is_over;
912 }
913
914 // Emit range checks for the given String.value byte array
915 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
916 if (stopped()) {
917 return; // already stopped
918 }
919 RegionNode* bailout = new RegionNode(1);
920 record_for_igvn(bailout);
921 if (char_count) {
922 // Convert char count to byte count
923 count = _gvn.transform(new LShiftINode(count, intcon(1)));
924 }
925
926 // Offset and count must not be negative
927 generate_negative_guard(offset, bailout);
928 generate_negative_guard(count, bailout);
929 // Offset + count must not exceed length of array
930 generate_limit_guard(offset, count, load_array_length(array), bailout);
931
932 if (bailout->req() > 1) {
933 PreserveJVMState pjvms(this);
934 set_control(_gvn.transform(bailout));
935 uncommon_trap(Deoptimization::Reason_intrinsic,
936 Deoptimization::Action_maybe_recompile);
937 }
938 }
939
940 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
941 bool is_immutable) {
942 ciKlass* thread_klass = env()->Thread_klass();
943 const Type* thread_type
944 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
945
946 Node* thread = _gvn.transform(new ThreadLocalNode());
947 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
948 tls_output = thread;
949
950 Node* thread_obj_handle
951 = (is_immutable
952 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
953 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
954 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
955 thread_obj_handle = _gvn.transform(thread_obj_handle);
956
957 DecoratorSet decorators = IN_NATIVE;
958 if (is_immutable) {
959 decorators |= C2_IMMUTABLE_MEMORY;
960 }
961 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
962 }
963
964 //--------------------------generate_current_thread--------------------
965 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
966 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
967 /*is_immutable*/false);
968 }
969
970 //--------------------------generate_virtual_thread--------------------
971 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
972 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
973 !C->method()->changes_current_thread());
974 }
975
976 //------------------------------make_string_method_node------------------------
977 // Helper method for String intrinsic functions. This version is called with
978 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
979 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
980 // containing the lengths of str1 and str2.
981 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
982 Node* result = nullptr;
983 switch (opcode) {
984 case Op_StrIndexOf:
985 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
986 str1_start, cnt1, str2_start, cnt2, ae);
987 break;
988 case Op_StrComp:
989 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
990 str1_start, cnt1, str2_start, cnt2, ae);
991 break;
992 case Op_StrEquals:
993 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
994 // Use the constant length if there is one because optimized match rule may exist.
995 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
996 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
997 break;
998 default:
999 ShouldNotReachHere();
1000 return nullptr;
1001 }
1002
1003 // All these intrinsics have checks.
1004 C->set_has_split_ifs(true); // Has chance for split-if optimization
1005 clear_upper_avx();
1006
1007 return _gvn.transform(result);
1008 }
1009
1010 //------------------------------inline_string_compareTo------------------------
1011 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1012 Node* arg1 = argument(0);
1013 Node* arg2 = argument(1);
1014
1015 arg1 = must_be_not_null(arg1, true);
1016 arg2 = must_be_not_null(arg2, true);
1017
1018 // Get start addr and length of first argument
1019 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1020 Node* arg1_cnt = load_array_length(arg1);
1021
1022 // Get start addr and length of second argument
1023 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1024 Node* arg2_cnt = load_array_length(arg2);
1025
1026 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1027 set_result(result);
1028 return true;
1029 }
1030
1031 //------------------------------inline_string_equals------------------------
1032 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1033 Node* arg1 = argument(0);
1034 Node* arg2 = argument(1);
1035
1036 // paths (plus control) merge
1037 RegionNode* region = new RegionNode(3);
1038 Node* phi = new PhiNode(region, TypeInt::BOOL);
1039
1040 if (!stopped()) {
1041
1042 arg1 = must_be_not_null(arg1, true);
1043 arg2 = must_be_not_null(arg2, true);
1044
1045 // Get start addr and length of first argument
1046 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1047 Node* arg1_cnt = load_array_length(arg1);
1048
1049 // Get start addr and length of second argument
1050 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1051 Node* arg2_cnt = load_array_length(arg2);
1052
1053 // Check for arg1_cnt != arg2_cnt
1054 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1055 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1056 Node* if_ne = generate_slow_guard(bol, nullptr);
1057 if (if_ne != nullptr) {
1058 phi->init_req(2, intcon(0));
1059 region->init_req(2, if_ne);
1060 }
1061
1062 // Check for count == 0 is done by assembler code for StrEquals.
1063
1064 if (!stopped()) {
1065 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1066 phi->init_req(1, equals);
1067 region->init_req(1, control());
1068 }
1069 }
1070
1071 // post merge
1072 set_control(_gvn.transform(region));
1073 record_for_igvn(region);
1074
1075 set_result(_gvn.transform(phi));
1076 return true;
1077 }
1078
1079 //------------------------------inline_array_equals----------------------------
1080 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1081 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1082 Node* arg1 = argument(0);
1083 Node* arg2 = argument(1);
1084
1085 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1086 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1087 clear_upper_avx();
1088
1089 return true;
1090 }
1091
1092
1093 //------------------------------inline_countPositives------------------------------
1094 bool LibraryCallKit::inline_countPositives() {
1095 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1096 return false;
1097 }
1098
1099 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1100 // no receiver since it is static method
1101 Node* ba = argument(0);
1102 Node* offset = argument(1);
1103 Node* len = argument(2);
1104
1105 ba = must_be_not_null(ba, true);
1106
1107 // Range checks
1108 generate_string_range_check(ba, offset, len, false);
1109 if (stopped()) {
1110 return true;
1111 }
1112 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1113 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1114 set_result(_gvn.transform(result));
1115 clear_upper_avx();
1116 return true;
1117 }
1118
1119 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1120 Node* index = argument(0);
1121 Node* length = bt == T_INT ? argument(1) : argument(2);
1122 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1123 return false;
1124 }
1125
1126 // check that length is positive
1127 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1128 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1129
1130 {
1131 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1132 uncommon_trap(Deoptimization::Reason_intrinsic,
1133 Deoptimization::Action_make_not_entrant);
1134 }
1135
1136 if (stopped()) {
1137 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1138 return true;
1139 }
1140
1141 // length is now known positive, add a cast node to make this explicit
1142 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1143 Node* casted_length = ConstraintCastNode::make(control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), ConstraintCastNode::RegularDependency, bt);
1144 casted_length = _gvn.transform(casted_length);
1145 replace_in_map(length, casted_length);
1146 length = casted_length;
1147
1148 // Use an unsigned comparison for the range check itself
1149 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1150 BoolTest::mask btest = BoolTest::lt;
1151 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1152 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1153 _gvn.set_type(rc, rc->Value(&_gvn));
1154 if (!rc_bool->is_Con()) {
1155 record_for_igvn(rc);
1156 }
1157 set_control(_gvn.transform(new IfTrueNode(rc)));
1158 {
1159 PreserveJVMState pjvms(this);
1160 set_control(_gvn.transform(new IfFalseNode(rc)));
1161 uncommon_trap(Deoptimization::Reason_range_check,
1162 Deoptimization::Action_make_not_entrant);
1163 }
1164
1165 if (stopped()) {
1166 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1167 return true;
1168 }
1169
1170 // index is now known to be >= 0 and < length, cast it
1171 Node* result = ConstraintCastNode::make(control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), ConstraintCastNode::RegularDependency, bt);
1172 result = _gvn.transform(result);
1173 set_result(result);
1174 replace_in_map(index, result);
1175 return true;
1176 }
1177
1178 //------------------------------inline_string_indexOf------------------------
1179 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1180 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1181 return false;
1182 }
1183 Node* src = argument(0);
1184 Node* tgt = argument(1);
1185
1186 // Make the merge point
1187 RegionNode* result_rgn = new RegionNode(4);
1188 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1189
1190 src = must_be_not_null(src, true);
1191 tgt = must_be_not_null(tgt, true);
1192
1193 // Get start addr and length of source string
1194 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1195 Node* src_count = load_array_length(src);
1196
1197 // Get start addr and length of substring
1198 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1199 Node* tgt_count = load_array_length(tgt);
1200
1201 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1202 // Divide src size by 2 if String is UTF16 encoded
1203 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1204 }
1205 if (ae == StrIntrinsicNode::UU) {
1206 // Divide substring size by 2 if String is UTF16 encoded
1207 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1208 }
1209
1210 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1211 if (result != nullptr) {
1212 result_phi->init_req(3, result);
1213 result_rgn->init_req(3, control());
1214 }
1215 set_control(_gvn.transform(result_rgn));
1216 record_for_igvn(result_rgn);
1217 set_result(_gvn.transform(result_phi));
1218
1219 return true;
1220 }
1221
1222 //-----------------------------inline_string_indexOf-----------------------
1223 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1224 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1225 return false;
1226 }
1227 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1228 return false;
1229 }
1230 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1231 Node* src = argument(0); // byte[]
1232 Node* src_count = argument(1); // char count
1233 Node* tgt = argument(2); // byte[]
1234 Node* tgt_count = argument(3); // char count
1235 Node* from_index = argument(4); // char index
1236
1237 src = must_be_not_null(src, true);
1238 tgt = must_be_not_null(tgt, true);
1239
1240 // Multiply byte array index by 2 if String is UTF16 encoded
1241 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1242 src_count = _gvn.transform(new SubINode(src_count, from_index));
1243 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1244 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1245
1246 // Range checks
1247 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1248 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1249 if (stopped()) {
1250 return true;
1251 }
1252
1253 RegionNode* region = new RegionNode(5);
1254 Node* phi = new PhiNode(region, TypeInt::INT);
1255
1256 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1257 if (result != nullptr) {
1258 // The result is index relative to from_index if substring was found, -1 otherwise.
1259 // Generate code which will fold into cmove.
1260 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1261 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1262
1263 Node* if_lt = generate_slow_guard(bol, nullptr);
1264 if (if_lt != nullptr) {
1265 // result == -1
1266 phi->init_req(3, result);
1267 region->init_req(3, if_lt);
1268 }
1269 if (!stopped()) {
1270 result = _gvn.transform(new AddINode(result, from_index));
1271 phi->init_req(4, result);
1272 region->init_req(4, control());
1273 }
1274 }
1275
1276 set_control(_gvn.transform(region));
1277 record_for_igvn(region);
1278 set_result(_gvn.transform(phi));
1279 clear_upper_avx();
1280
1281 return true;
1282 }
1283
1284 // Create StrIndexOfNode with fast path checks
1285 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1286 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1287 // Check for substr count > string count
1288 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1289 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1290 Node* if_gt = generate_slow_guard(bol, nullptr);
1291 if (if_gt != nullptr) {
1292 phi->init_req(1, intcon(-1));
1293 region->init_req(1, if_gt);
1294 }
1295 if (!stopped()) {
1296 // Check for substr count == 0
1297 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1298 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1299 Node* if_zero = generate_slow_guard(bol, nullptr);
1300 if (if_zero != nullptr) {
1301 phi->init_req(2, intcon(0));
1302 region->init_req(2, if_zero);
1303 }
1304 }
1305 if (!stopped()) {
1306 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1307 }
1308 return nullptr;
1309 }
1310
1311 //-----------------------------inline_string_indexOfChar-----------------------
1312 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1313 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1314 return false;
1315 }
1316 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1317 return false;
1318 }
1319 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1320 Node* src = argument(0); // byte[]
1321 Node* int_ch = argument(1);
1322 Node* from_index = argument(2);
1323 Node* max = argument(3);
1324
1325 src = must_be_not_null(src, true);
1326
1327 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1328 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1329 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1330
1331 // Range checks
1332 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1333
1334 // Check for int_ch >= 0
1335 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1336 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1337 {
1338 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1339 uncommon_trap(Deoptimization::Reason_intrinsic,
1340 Deoptimization::Action_maybe_recompile);
1341 }
1342 if (stopped()) {
1343 return true;
1344 }
1345
1346 RegionNode* region = new RegionNode(3);
1347 Node* phi = new PhiNode(region, TypeInt::INT);
1348
1349 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1350 C->set_has_split_ifs(true); // Has chance for split-if optimization
1351 _gvn.transform(result);
1352
1353 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1354 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1355
1356 Node* if_lt = generate_slow_guard(bol, nullptr);
1357 if (if_lt != nullptr) {
1358 // result == -1
1359 phi->init_req(2, result);
1360 region->init_req(2, if_lt);
1361 }
1362 if (!stopped()) {
1363 result = _gvn.transform(new AddINode(result, from_index));
1364 phi->init_req(1, result);
1365 region->init_req(1, control());
1366 }
1367 set_control(_gvn.transform(region));
1368 record_for_igvn(region);
1369 set_result(_gvn.transform(phi));
1370 clear_upper_avx();
1371
1372 return true;
1373 }
1374 //---------------------------inline_string_copy---------------------
1375 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1376 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1377 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1378 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1379 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1380 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1381 bool LibraryCallKit::inline_string_copy(bool compress) {
1382 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1383 return false;
1384 }
1385 int nargs = 5; // 2 oops, 3 ints
1386 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1387
1388 Node* src = argument(0);
1389 Node* src_offset = argument(1);
1390 Node* dst = argument(2);
1391 Node* dst_offset = argument(3);
1392 Node* length = argument(4);
1393
1394 // Check for allocation before we add nodes that would confuse
1395 // tightly_coupled_allocation()
1396 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1397
1398 // Figure out the size and type of the elements we will be copying.
1399 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1400 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1401 if (src_type == nullptr || dst_type == nullptr) {
1402 return false;
1403 }
1404 BasicType src_elem = src_type->elem()->array_element_basic_type();
1405 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1406 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1407 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1408 "Unsupported array types for inline_string_copy");
1409
1410 src = must_be_not_null(src, true);
1411 dst = must_be_not_null(dst, true);
1412
1413 // Convert char[] offsets to byte[] offsets
1414 bool convert_src = (compress && src_elem == T_BYTE);
1415 bool convert_dst = (!compress && dst_elem == T_BYTE);
1416 if (convert_src) {
1417 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1418 } else if (convert_dst) {
1419 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1420 }
1421
1422 // Range checks
1423 generate_string_range_check(src, src_offset, length, convert_src);
1424 generate_string_range_check(dst, dst_offset, length, convert_dst);
1425 if (stopped()) {
1426 return true;
1427 }
1428
1429 Node* src_start = array_element_address(src, src_offset, src_elem);
1430 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1431 // 'src_start' points to src array + scaled offset
1432 // 'dst_start' points to dst array + scaled offset
1433 Node* count = nullptr;
1434 if (compress) {
1435 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1436 } else {
1437 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1438 }
1439
1440 if (alloc != nullptr) {
1441 if (alloc->maybe_set_complete(&_gvn)) {
1442 // "You break it, you buy it."
1443 InitializeNode* init = alloc->initialization();
1444 assert(init->is_complete(), "we just did this");
1445 init->set_complete_with_arraycopy();
1446 assert(dst->is_CheckCastPP(), "sanity");
1447 assert(dst->in(0)->in(0) == init, "dest pinned");
1448 }
1449 // Do not let stores that initialize this object be reordered with
1450 // a subsequent store that would make this object accessible by
1451 // other threads.
1452 // Record what AllocateNode this StoreStore protects so that
1453 // escape analysis can go from the MemBarStoreStoreNode to the
1454 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1455 // based on the escape status of the AllocateNode.
1456 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1457 }
1458 if (compress) {
1459 set_result(_gvn.transform(count));
1460 }
1461 clear_upper_avx();
1462
1463 return true;
1464 }
1465
1466 #ifdef _LP64
1467 #define XTOP ,top() /*additional argument*/
1468 #else //_LP64
1469 #define XTOP /*no additional argument*/
1470 #endif //_LP64
1471
1472 //------------------------inline_string_toBytesU--------------------------
1473 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1474 bool LibraryCallKit::inline_string_toBytesU() {
1475 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1476 return false;
1477 }
1478 // Get the arguments.
1479 Node* value = argument(0);
1480 Node* offset = argument(1);
1481 Node* length = argument(2);
1482
1483 Node* newcopy = nullptr;
1484
1485 // Set the original stack and the reexecute bit for the interpreter to reexecute
1486 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1487 { PreserveReexecuteState preexecs(this);
1488 jvms()->set_should_reexecute(true);
1489
1490 // Check if a null path was taken unconditionally.
1491 value = null_check(value);
1492
1493 RegionNode* bailout = new RegionNode(1);
1494 record_for_igvn(bailout);
1495
1496 // Range checks
1497 generate_negative_guard(offset, bailout);
1498 generate_negative_guard(length, bailout);
1499 generate_limit_guard(offset, length, load_array_length(value), bailout);
1500 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1501 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1502
1503 if (bailout->req() > 1) {
1504 PreserveJVMState pjvms(this);
1505 set_control(_gvn.transform(bailout));
1506 uncommon_trap(Deoptimization::Reason_intrinsic,
1507 Deoptimization::Action_maybe_recompile);
1508 }
1509 if (stopped()) {
1510 return true;
1511 }
1512
1513 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1514 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1515 newcopy = new_array(klass_node, size, 0); // no arguments to push
1516 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1517 guarantee(alloc != nullptr, "created above");
1518
1519 // Calculate starting addresses.
1520 Node* src_start = array_element_address(value, offset, T_CHAR);
1521 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1522
1523 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1524 const TypeInt* toffset = gvn().type(offset)->is_int();
1525 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1526
1527 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1528 const char* copyfunc_name = "arraycopy";
1529 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1530 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1531 OptoRuntime::fast_arraycopy_Type(),
1532 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1533 src_start, dst_start, ConvI2X(length) XTOP);
1534 // Do not let reads from the cloned object float above the arraycopy.
1535 if (alloc->maybe_set_complete(&_gvn)) {
1536 // "You break it, you buy it."
1537 InitializeNode* init = alloc->initialization();
1538 assert(init->is_complete(), "we just did this");
1539 init->set_complete_with_arraycopy();
1540 assert(newcopy->is_CheckCastPP(), "sanity");
1541 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1542 }
1543 // Do not let stores that initialize this object be reordered with
1544 // a subsequent store that would make this object accessible by
1545 // other threads.
1546 // Record what AllocateNode this StoreStore protects so that
1547 // escape analysis can go from the MemBarStoreStoreNode to the
1548 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1549 // based on the escape status of the AllocateNode.
1550 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1551 } // original reexecute is set back here
1552
1553 C->set_has_split_ifs(true); // Has chance for split-if optimization
1554 if (!stopped()) {
1555 set_result(newcopy);
1556 }
1557 clear_upper_avx();
1558
1559 return true;
1560 }
1561
1562 //------------------------inline_string_getCharsU--------------------------
1563 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1564 bool LibraryCallKit::inline_string_getCharsU() {
1565 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1566 return false;
1567 }
1568
1569 // Get the arguments.
1570 Node* src = argument(0);
1571 Node* src_begin = argument(1);
1572 Node* src_end = argument(2); // exclusive offset (i < src_end)
1573 Node* dst = argument(3);
1574 Node* dst_begin = argument(4);
1575
1576 // Check for allocation before we add nodes that would confuse
1577 // tightly_coupled_allocation()
1578 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1579
1580 // Check if a null path was taken unconditionally.
1581 src = null_check(src);
1582 dst = null_check(dst);
1583 if (stopped()) {
1584 return true;
1585 }
1586
1587 // Get length and convert char[] offset to byte[] offset
1588 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1589 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1590
1591 // Range checks
1592 generate_string_range_check(src, src_begin, length, true);
1593 generate_string_range_check(dst, dst_begin, length, false);
1594 if (stopped()) {
1595 return true;
1596 }
1597
1598 if (!stopped()) {
1599 // Calculate starting addresses.
1600 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1601 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1602
1603 // Check if array addresses are aligned to HeapWordSize
1604 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1605 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1606 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1607 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1608
1609 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1610 const char* copyfunc_name = "arraycopy";
1611 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1612 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1613 OptoRuntime::fast_arraycopy_Type(),
1614 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1615 src_start, dst_start, ConvI2X(length) XTOP);
1616 // Do not let reads from the cloned object float above the arraycopy.
1617 if (alloc != nullptr) {
1618 if (alloc->maybe_set_complete(&_gvn)) {
1619 // "You break it, you buy it."
1620 InitializeNode* init = alloc->initialization();
1621 assert(init->is_complete(), "we just did this");
1622 init->set_complete_with_arraycopy();
1623 assert(dst->is_CheckCastPP(), "sanity");
1624 assert(dst->in(0)->in(0) == init, "dest pinned");
1625 }
1626 // Do not let stores that initialize this object be reordered with
1627 // a subsequent store that would make this object accessible by
1628 // other threads.
1629 // Record what AllocateNode this StoreStore protects so that
1630 // escape analysis can go from the MemBarStoreStoreNode to the
1631 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1632 // based on the escape status of the AllocateNode.
1633 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1634 } else {
1635 insert_mem_bar(Op_MemBarCPUOrder);
1636 }
1637 }
1638
1639 C->set_has_split_ifs(true); // Has chance for split-if optimization
1640 return true;
1641 }
1642
1643 //----------------------inline_string_char_access----------------------------
1644 // Store/Load char to/from byte[] array.
1645 // static void StringUTF16.putChar(byte[] val, int index, int c)
1646 // static char StringUTF16.getChar(byte[] val, int index)
1647 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1648 Node* value = argument(0);
1649 Node* index = argument(1);
1650 Node* ch = is_store ? argument(2) : nullptr;
1651
1652 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1653 // correctly requires matched array shapes.
1654 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1655 "sanity: byte[] and char[] bases agree");
1656 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1657 "sanity: byte[] and char[] scales agree");
1658
1659 // Bail when getChar over constants is requested: constant folding would
1660 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1661 // Java method would constant fold nicely instead.
1662 if (!is_store && value->is_Con() && index->is_Con()) {
1663 return false;
1664 }
1665
1666 // Save state and restore on bailout
1667 uint old_sp = sp();
1668 SafePointNode* old_map = clone_map();
1669
1670 value = must_be_not_null(value, true);
1671
1672 Node* adr = array_element_address(value, index, T_CHAR);
1673 if (adr->is_top()) {
1674 set_map(old_map);
1675 set_sp(old_sp);
1676 return false;
1677 }
1678 destruct_map_clone(old_map);
1679 if (is_store) {
1680 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1681 } else {
1682 ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD | C2_UNKNOWN_CONTROL_LOAD);
1683 set_result(ch);
1684 }
1685 return true;
1686 }
1687
1688 //--------------------------round_double_node--------------------------------
1689 // Round a double node if necessary.
1690 Node* LibraryCallKit::round_double_node(Node* n) {
1691 if (Matcher::strict_fp_requires_explicit_rounding) {
1692 #ifdef IA32
1693 if (UseSSE < 2) {
1694 n = _gvn.transform(new RoundDoubleNode(nullptr, n));
1695 }
1696 #else
1697 Unimplemented();
1698 #endif // IA32
1699 }
1700 return n;
1701 }
1702
1703 //------------------------------inline_math-----------------------------------
1704 // public static double Math.abs(double)
1705 // public static double Math.sqrt(double)
1706 // public static double Math.log(double)
1707 // public static double Math.log10(double)
1708 // public static double Math.round(double)
1709 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1710 Node* arg = round_double_node(argument(0));
1711 Node* n = nullptr;
1712 switch (id) {
1713 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1714 case vmIntrinsics::_dsqrt:
1715 case vmIntrinsics::_dsqrt_strict:
1716 n = new SqrtDNode(C, control(), arg); break;
1717 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1718 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1719 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1720 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1721 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1722 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1723 default: fatal_unexpected_iid(id); break;
1724 }
1725 set_result(_gvn.transform(n));
1726 return true;
1727 }
1728
1729 //------------------------------inline_math-----------------------------------
1730 // public static float Math.abs(float)
1731 // public static int Math.abs(int)
1732 // public static long Math.abs(long)
1733 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1734 Node* arg = argument(0);
1735 Node* n = nullptr;
1736 switch (id) {
1737 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1738 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1739 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1740 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1741 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1742 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1743 default: fatal_unexpected_iid(id); break;
1744 }
1745 set_result(_gvn.transform(n));
1746 return true;
1747 }
1748
1749 //------------------------------runtime_math-----------------------------
1750 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1751 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1752 "must be (DD)D or (D)D type");
1753
1754 // Inputs
1755 Node* a = round_double_node(argument(0));
1756 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : nullptr;
1757
1758 const TypePtr* no_memory_effects = nullptr;
1759 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1760 no_memory_effects,
1761 a, top(), b, b ? top() : nullptr);
1762 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1763 #ifdef ASSERT
1764 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1765 assert(value_top == top(), "second value must be top");
1766 #endif
1767
1768 set_result(value);
1769 return true;
1770 }
1771
1772 //------------------------------inline_math_pow-----------------------------
1773 bool LibraryCallKit::inline_math_pow() {
1774 Node* exp = round_double_node(argument(2));
1775 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1776 if (d != nullptr) {
1777 if (d->getd() == 2.0) {
1778 // Special case: pow(x, 2.0) => x * x
1779 Node* base = round_double_node(argument(0));
1780 set_result(_gvn.transform(new MulDNode(base, base)));
1781 return true;
1782 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1783 // Special case: pow(x, 0.5) => sqrt(x)
1784 Node* base = round_double_node(argument(0));
1785 Node* zero = _gvn.zerocon(T_DOUBLE);
1786
1787 RegionNode* region = new RegionNode(3);
1788 Node* phi = new PhiNode(region, Type::DOUBLE);
1789
1790 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1791 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1792 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1793 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1794 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1795
1796 Node* if_pow = generate_slow_guard(test, nullptr);
1797 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1798 phi->init_req(1, value_sqrt);
1799 region->init_req(1, control());
1800
1801 if (if_pow != nullptr) {
1802 set_control(if_pow);
1803 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1804 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1805 const TypePtr* no_memory_effects = nullptr;
1806 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1807 no_memory_effects, base, top(), exp, top());
1808 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1809 #ifdef ASSERT
1810 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1811 assert(value_top == top(), "second value must be top");
1812 #endif
1813 phi->init_req(2, value_pow);
1814 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1815 }
1816
1817 C->set_has_split_ifs(true); // Has chance for split-if optimization
1818 set_control(_gvn.transform(region));
1819 record_for_igvn(region);
1820 set_result(_gvn.transform(phi));
1821
1822 return true;
1823 }
1824 }
1825
1826 return StubRoutines::dpow() != nullptr ?
1827 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1828 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1829 }
1830
1831 //------------------------------inline_math_native-----------------------------
1832 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1833 switch (id) {
1834 case vmIntrinsics::_dsin:
1835 return StubRoutines::dsin() != nullptr ?
1836 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1837 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1838 case vmIntrinsics::_dcos:
1839 return StubRoutines::dcos() != nullptr ?
1840 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1841 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1842 case vmIntrinsics::_dtan:
1843 return StubRoutines::dtan() != nullptr ?
1844 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1845 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1846 case vmIntrinsics::_dexp:
1847 return StubRoutines::dexp() != nullptr ?
1848 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1849 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1850 case vmIntrinsics::_dlog:
1851 return StubRoutines::dlog() != nullptr ?
1852 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1853 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1854 case vmIntrinsics::_dlog10:
1855 return StubRoutines::dlog10() != nullptr ?
1856 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1857 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1858
1859 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1860 case vmIntrinsics::_ceil:
1861 case vmIntrinsics::_floor:
1862 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1863
1864 case vmIntrinsics::_dsqrt:
1865 case vmIntrinsics::_dsqrt_strict:
1866 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1867 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1868 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1869 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1870 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1871
1872 case vmIntrinsics::_dpow: return inline_math_pow();
1873 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1874 case vmIntrinsics::_fcopySign: return inline_math(id);
1875 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1876 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1877 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1878
1879 // These intrinsics are not yet correctly implemented
1880 case vmIntrinsics::_datan2:
1881 return false;
1882
1883 default:
1884 fatal_unexpected_iid(id);
1885 return false;
1886 }
1887 }
1888
1889 //----------------------------inline_notify-----------------------------------*
1890 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1891 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1892 address func;
1893 if (id == vmIntrinsics::_notify) {
1894 func = OptoRuntime::monitor_notify_Java();
1895 } else {
1896 func = OptoRuntime::monitor_notifyAll_Java();
1897 }
1898 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1899 make_slow_call_ex(call, env()->Throwable_klass(), false);
1900 return true;
1901 }
1902
1903
1904 //----------------------------inline_min_max-----------------------------------
1905 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1906 set_result(generate_min_max(id, argument(0), argument(1)));
1907 return true;
1908 }
1909
1910 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1911 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1912 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1913 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1914 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1915
1916 {
1917 PreserveJVMState pjvms(this);
1918 PreserveReexecuteState preexecs(this);
1919 jvms()->set_should_reexecute(true);
1920
1921 set_control(slow_path);
1922 set_i_o(i_o());
1923
1924 uncommon_trap(Deoptimization::Reason_intrinsic,
1925 Deoptimization::Action_none);
1926 }
1927
1928 set_control(fast_path);
1929 set_result(math);
1930 }
1931
1932 template <typename OverflowOp>
1933 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1934 typedef typename OverflowOp::MathOp MathOp;
1935
1936 MathOp* mathOp = new MathOp(arg1, arg2);
1937 Node* operation = _gvn.transform( mathOp );
1938 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1939 inline_math_mathExact(operation, ofcheck);
1940 return true;
1941 }
1942
1943 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1944 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1945 }
1946
1947 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1948 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1949 }
1950
1951 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1952 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1953 }
1954
1955 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1956 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1957 }
1958
1959 bool LibraryCallKit::inline_math_negateExactI() {
1960 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1961 }
1962
1963 bool LibraryCallKit::inline_math_negateExactL() {
1964 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1965 }
1966
1967 bool LibraryCallKit::inline_math_multiplyExactI() {
1968 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1969 }
1970
1971 bool LibraryCallKit::inline_math_multiplyExactL() {
1972 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1973 }
1974
1975 bool LibraryCallKit::inline_math_multiplyHigh() {
1976 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
1977 return true;
1978 }
1979
1980 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
1981 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
1982 return true;
1983 }
1984
1985 Node*
1986 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1987 Node* result_val = nullptr;
1988 switch (id) {
1989 case vmIntrinsics::_min:
1990 case vmIntrinsics::_min_strict:
1991 result_val = _gvn.transform(new MinINode(x0, y0));
1992 break;
1993 case vmIntrinsics::_max:
1994 case vmIntrinsics::_max_strict:
1995 result_val = _gvn.transform(new MaxINode(x0, y0));
1996 break;
1997 default:
1998 fatal_unexpected_iid(id);
1999 break;
2000 }
2001 return result_val;
2002 }
2003
2004 inline int
2005 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2006 const TypePtr* base_type = TypePtr::NULL_PTR;
2007 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2008 if (base_type == nullptr) {
2009 // Unknown type.
2010 return Type::AnyPtr;
2011 } else if (base_type == TypePtr::NULL_PTR) {
2012 // Since this is a null+long form, we have to switch to a rawptr.
2013 base = _gvn.transform(new CastX2PNode(offset));
2014 offset = MakeConX(0);
2015 return Type::RawPtr;
2016 } else if (base_type->base() == Type::RawPtr) {
2017 return Type::RawPtr;
2018 } else if (base_type->isa_oopptr()) {
2019 // Base is never null => always a heap address.
2020 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2021 return Type::OopPtr;
2022 }
2023 // Offset is small => always a heap address.
2024 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2025 if (offset_type != nullptr &&
2026 base_type->offset() == 0 && // (should always be?)
2027 offset_type->_lo >= 0 &&
2028 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2029 return Type::OopPtr;
2030 } else if (type == T_OBJECT) {
2031 // off heap access to an oop doesn't make any sense. Has to be on
2032 // heap.
2033 return Type::OopPtr;
2034 }
2035 // Otherwise, it might either be oop+off or null+addr.
2036 return Type::AnyPtr;
2037 } else {
2038 // No information:
2039 return Type::AnyPtr;
2040 }
2041 }
2042
2043 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2044 Node* uncasted_base = base;
2045 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2046 if (kind == Type::RawPtr) {
2047 return basic_plus_adr(top(), uncasted_base, offset);
2048 } else if (kind == Type::AnyPtr) {
2049 assert(base == uncasted_base, "unexpected base change");
2050 if (can_cast) {
2051 if (!_gvn.type(base)->speculative_maybe_null() &&
2052 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2053 // According to profiling, this access is always on
2054 // heap. Casting the base to not null and thus avoiding membars
2055 // around the access should allow better optimizations
2056 Node* null_ctl = top();
2057 base = null_check_oop(base, &null_ctl, true, true, true);
2058 assert(null_ctl->is_top(), "no null control here");
2059 return basic_plus_adr(base, offset);
2060 } else if (_gvn.type(base)->speculative_always_null() &&
2061 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2062 // According to profiling, this access is always off
2063 // heap.
2064 base = null_assert(base);
2065 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2066 offset = MakeConX(0);
2067 return basic_plus_adr(top(), raw_base, offset);
2068 }
2069 }
2070 // We don't know if it's an on heap or off heap access. Fall back
2071 // to raw memory access.
2072 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2073 return basic_plus_adr(top(), raw, offset);
2074 } else {
2075 assert(base == uncasted_base, "unexpected base change");
2076 // We know it's an on heap access so base can't be null
2077 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2078 base = must_be_not_null(base, true);
2079 }
2080 return basic_plus_adr(base, offset);
2081 }
2082 }
2083
2084 //--------------------------inline_number_methods-----------------------------
2085 // inline int Integer.numberOfLeadingZeros(int)
2086 // inline int Long.numberOfLeadingZeros(long)
2087 //
2088 // inline int Integer.numberOfTrailingZeros(int)
2089 // inline int Long.numberOfTrailingZeros(long)
2090 //
2091 // inline int Integer.bitCount(int)
2092 // inline int Long.bitCount(long)
2093 //
2094 // inline char Character.reverseBytes(char)
2095 // inline short Short.reverseBytes(short)
2096 // inline int Integer.reverseBytes(int)
2097 // inline long Long.reverseBytes(long)
2098 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2099 Node* arg = argument(0);
2100 Node* n = nullptr;
2101 switch (id) {
2102 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2103 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2104 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2105 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2106 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2107 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2108 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2109 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2110 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2111 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2112 case vmIntrinsics::_reverse_i: n = new ReverseINode(0, arg); break;
2113 case vmIntrinsics::_reverse_l: n = new ReverseLNode(0, arg); break;
2114 default: fatal_unexpected_iid(id); break;
2115 }
2116 set_result(_gvn.transform(n));
2117 return true;
2118 }
2119
2120 //--------------------------inline_bitshuffle_methods-----------------------------
2121 // inline int Integer.compress(int, int)
2122 // inline int Integer.expand(int, int)
2123 // inline long Long.compress(long, long)
2124 // inline long Long.expand(long, long)
2125 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2126 Node* n = nullptr;
2127 switch (id) {
2128 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2129 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2130 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2131 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2132 default: fatal_unexpected_iid(id); break;
2133 }
2134 set_result(_gvn.transform(n));
2135 return true;
2136 }
2137
2138 //--------------------------inline_number_methods-----------------------------
2139 // inline int Integer.compareUnsigned(int, int)
2140 // inline int Long.compareUnsigned(long, long)
2141 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2142 Node* arg1 = argument(0);
2143 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2144 Node* n = nullptr;
2145 switch (id) {
2146 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2147 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2148 default: fatal_unexpected_iid(id); break;
2149 }
2150 set_result(_gvn.transform(n));
2151 return true;
2152 }
2153
2154 //--------------------------inline_unsigned_divmod_methods-----------------------------
2155 // inline int Integer.divideUnsigned(int, int)
2156 // inline int Integer.remainderUnsigned(int, int)
2157 // inline long Long.divideUnsigned(long, long)
2158 // inline long Long.remainderUnsigned(long, long)
2159 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2160 Node* n = nullptr;
2161 switch (id) {
2162 case vmIntrinsics::_divideUnsigned_i: {
2163 zero_check_int(argument(1));
2164 // Compile-time detect of null-exception
2165 if (stopped()) {
2166 return true; // keep the graph constructed so far
2167 }
2168 n = new UDivINode(control(), argument(0), argument(1));
2169 break;
2170 }
2171 case vmIntrinsics::_divideUnsigned_l: {
2172 zero_check_long(argument(2));
2173 // Compile-time detect of null-exception
2174 if (stopped()) {
2175 return true; // keep the graph constructed so far
2176 }
2177 n = new UDivLNode(control(), argument(0), argument(2));
2178 break;
2179 }
2180 case vmIntrinsics::_remainderUnsigned_i: {
2181 zero_check_int(argument(1));
2182 // Compile-time detect of null-exception
2183 if (stopped()) {
2184 return true; // keep the graph constructed so far
2185 }
2186 n = new UModINode(control(), argument(0), argument(1));
2187 break;
2188 }
2189 case vmIntrinsics::_remainderUnsigned_l: {
2190 zero_check_long(argument(2));
2191 // Compile-time detect of null-exception
2192 if (stopped()) {
2193 return true; // keep the graph constructed so far
2194 }
2195 n = new UModLNode(control(), argument(0), argument(2));
2196 break;
2197 }
2198 default: fatal_unexpected_iid(id); break;
2199 }
2200 set_result(_gvn.transform(n));
2201 return true;
2202 }
2203
2204 //----------------------------inline_unsafe_access----------------------------
2205
2206 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2207 // Attempt to infer a sharper value type from the offset and base type.
2208 ciKlass* sharpened_klass = nullptr;
2209
2210 // See if it is an instance field, with an object type.
2211 if (alias_type->field() != nullptr) {
2212 if (alias_type->field()->type()->is_klass()) {
2213 sharpened_klass = alias_type->field()->type()->as_klass();
2214 }
2215 }
2216
2217 const TypeOopPtr* result = nullptr;
2218 // See if it is a narrow oop array.
2219 if (adr_type->isa_aryptr()) {
2220 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2221 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2222 if (elem_type != nullptr && elem_type->is_loaded()) {
2223 // Sharpen the value type.
2224 result = elem_type;
2225 }
2226 }
2227 }
2228
2229 // The sharpened class might be unloaded if there is no class loader
2230 // contraint in place.
2231 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2232 // Sharpen the value type.
2233 result = TypeOopPtr::make_from_klass(sharpened_klass);
2234 }
2235 if (result != nullptr) {
2236 #ifndef PRODUCT
2237 if (C->print_intrinsics() || C->print_inlining()) {
2238 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2239 tty->print(" sharpened value: "); result->dump(); tty->cr();
2240 }
2241 #endif
2242 }
2243 return result;
2244 }
2245
2246 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2247 switch (kind) {
2248 case Relaxed:
2249 return MO_UNORDERED;
2250 case Opaque:
2251 return MO_RELAXED;
2252 case Acquire:
2253 return MO_ACQUIRE;
2254 case Release:
2255 return MO_RELEASE;
2256 case Volatile:
2257 return MO_SEQ_CST;
2258 default:
2259 ShouldNotReachHere();
2260 return 0;
2261 }
2262 }
2263
2264 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2265 if (callee()->is_static()) return false; // caller must have the capability!
2266 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2267 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2268 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2269 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2270
2271 if (is_reference_type(type)) {
2272 decorators |= ON_UNKNOWN_OOP_REF;
2273 }
2274
2275 if (unaligned) {
2276 decorators |= C2_UNALIGNED;
2277 }
2278
2279 #ifndef PRODUCT
2280 {
2281 ResourceMark rm;
2282 // Check the signatures.
2283 ciSignature* sig = callee()->signature();
2284 #ifdef ASSERT
2285 if (!is_store) {
2286 // Object getReference(Object base, int/long offset), etc.
2287 BasicType rtype = sig->return_type()->basic_type();
2288 assert(rtype == type, "getter must return the expected value");
2289 assert(sig->count() == 2, "oop getter has 2 arguments");
2290 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2291 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2292 } else {
2293 // void putReference(Object base, int/long offset, Object x), etc.
2294 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2295 assert(sig->count() == 3, "oop putter has 3 arguments");
2296 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2297 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2298 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2299 assert(vtype == type, "putter must accept the expected value");
2300 }
2301 #endif // ASSERT
2302 }
2303 #endif //PRODUCT
2304
2305 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2306
2307 Node* receiver = argument(0); // type: oop
2308
2309 // Build address expression.
2310 Node* heap_base_oop = top();
2311
2312 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2313 Node* base = argument(1); // type: oop
2314 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2315 Node* offset = argument(2); // type: long
2316 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2317 // to be plain byte offsets, which are also the same as those accepted
2318 // by oopDesc::field_addr.
2319 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2320 "fieldOffset must be byte-scaled");
2321 // 32-bit machines ignore the high half!
2322 offset = ConvL2X(offset);
2323
2324 // Save state and restore on bailout
2325 uint old_sp = sp();
2326 SafePointNode* old_map = clone_map();
2327
2328 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2329
2330 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2331 if (type != T_OBJECT) {
2332 decorators |= IN_NATIVE; // off-heap primitive access
2333 } else {
2334 set_map(old_map);
2335 set_sp(old_sp);
2336 return false; // off-heap oop accesses are not supported
2337 }
2338 } else {
2339 heap_base_oop = base; // on-heap or mixed access
2340 }
2341
2342 // Can base be null? Otherwise, always on-heap access.
2343 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2344
2345 if (!can_access_non_heap) {
2346 decorators |= IN_HEAP;
2347 }
2348
2349 Node* val = is_store ? argument(4) : nullptr;
2350
2351 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2352 if (adr_type == TypePtr::NULL_PTR) {
2353 set_map(old_map);
2354 set_sp(old_sp);
2355 return false; // off-heap access with zero address
2356 }
2357
2358 // Try to categorize the address.
2359 Compile::AliasType* alias_type = C->alias_type(adr_type);
2360 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2361
2362 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2363 alias_type->adr_type() == TypeAryPtr::RANGE) {
2364 set_map(old_map);
2365 set_sp(old_sp);
2366 return false; // not supported
2367 }
2368
2369 bool mismatched = false;
2370 BasicType bt = alias_type->basic_type();
2371 if (bt != T_ILLEGAL) {
2372 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2373 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2374 // Alias type doesn't differentiate between byte[] and boolean[]).
2375 // Use address type to get the element type.
2376 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2377 }
2378 if (is_reference_type(bt, true)) {
2379 // accessing an array field with getReference is not a mismatch
2380 bt = T_OBJECT;
2381 }
2382 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2383 // Don't intrinsify mismatched object accesses
2384 set_map(old_map);
2385 set_sp(old_sp);
2386 return false;
2387 }
2388 mismatched = (bt != type);
2389 } else if (alias_type->adr_type()->isa_oopptr()) {
2390 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2391 }
2392
2393 destruct_map_clone(old_map);
2394 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2395
2396 if (mismatched) {
2397 decorators |= C2_MISMATCHED;
2398 }
2399
2400 // First guess at the value type.
2401 const Type *value_type = Type::get_const_basic_type(type);
2402
2403 // Figure out the memory ordering.
2404 decorators |= mo_decorator_for_access_kind(kind);
2405
2406 if (!is_store && type == T_OBJECT) {
2407 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2408 if (tjp != nullptr) {
2409 value_type = tjp;
2410 }
2411 }
2412
2413 receiver = null_check(receiver);
2414 if (stopped()) {
2415 return true;
2416 }
2417 // Heap pointers get a null-check from the interpreter,
2418 // as a courtesy. However, this is not guaranteed by Unsafe,
2419 // and it is not possible to fully distinguish unintended nulls
2420 // from intended ones in this API.
2421
2422 if (!is_store) {
2423 Node* p = nullptr;
2424 // Try to constant fold a load from a constant field
2425 ciField* field = alias_type->field();
2426 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2427 // final or stable field
2428 p = make_constant_from_field(field, heap_base_oop);
2429 }
2430
2431 if (p == nullptr) { // Could not constant fold the load
2432 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2433 // Normalize the value returned by getBoolean in the following cases
2434 if (type == T_BOOLEAN &&
2435 (mismatched ||
2436 heap_base_oop == top() || // - heap_base_oop is null or
2437 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2438 // and the unsafe access is made to large offset
2439 // (i.e., larger than the maximum offset necessary for any
2440 // field access)
2441 ) {
2442 IdealKit ideal = IdealKit(this);
2443 #define __ ideal.
2444 IdealVariable normalized_result(ideal);
2445 __ declarations_done();
2446 __ set(normalized_result, p);
2447 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2448 __ set(normalized_result, ideal.ConI(1));
2449 ideal.end_if();
2450 final_sync(ideal);
2451 p = __ value(normalized_result);
2452 #undef __
2453 }
2454 }
2455 if (type == T_ADDRESS) {
2456 p = gvn().transform(new CastP2XNode(nullptr, p));
2457 p = ConvX2UL(p);
2458 }
2459 // The load node has the control of the preceding MemBarCPUOrder. All
2460 // following nodes will have the control of the MemBarCPUOrder inserted at
2461 // the end of this method. So, pushing the load onto the stack at a later
2462 // point is fine.
2463 set_result(p);
2464 } else {
2465 if (bt == T_ADDRESS) {
2466 // Repackage the long as a pointer.
2467 val = ConvL2X(val);
2468 val = gvn().transform(new CastX2PNode(val));
2469 }
2470 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2471 }
2472
2473 return true;
2474 }
2475
2476 //----------------------------inline_unsafe_load_store----------------------------
2477 // This method serves a couple of different customers (depending on LoadStoreKind):
2478 //
2479 // LS_cmp_swap:
2480 //
2481 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2482 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2483 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2484 //
2485 // LS_cmp_swap_weak:
2486 //
2487 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2488 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2489 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2490 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2491 //
2492 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2493 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2494 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2495 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2496 //
2497 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2498 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2499 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2500 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2501 //
2502 // LS_cmp_exchange:
2503 //
2504 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2505 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2506 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2507 //
2508 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2509 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2510 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2511 //
2512 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2513 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2514 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2515 //
2516 // LS_get_add:
2517 //
2518 // int getAndAddInt( Object o, long offset, int delta)
2519 // long getAndAddLong(Object o, long offset, long delta)
2520 //
2521 // LS_get_set:
2522 //
2523 // int getAndSet(Object o, long offset, int newValue)
2524 // long getAndSet(Object o, long offset, long newValue)
2525 // Object getAndSet(Object o, long offset, Object newValue)
2526 //
2527 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2528 // This basic scheme here is the same as inline_unsafe_access, but
2529 // differs in enough details that combining them would make the code
2530 // overly confusing. (This is a true fact! I originally combined
2531 // them, but even I was confused by it!) As much code/comments as
2532 // possible are retained from inline_unsafe_access though to make
2533 // the correspondences clearer. - dl
2534
2535 if (callee()->is_static()) return false; // caller must have the capability!
2536
2537 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2538 decorators |= mo_decorator_for_access_kind(access_kind);
2539
2540 #ifndef PRODUCT
2541 BasicType rtype;
2542 {
2543 ResourceMark rm;
2544 // Check the signatures.
2545 ciSignature* sig = callee()->signature();
2546 rtype = sig->return_type()->basic_type();
2547 switch(kind) {
2548 case LS_get_add:
2549 case LS_get_set: {
2550 // Check the signatures.
2551 #ifdef ASSERT
2552 assert(rtype == type, "get and set must return the expected type");
2553 assert(sig->count() == 3, "get and set has 3 arguments");
2554 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2555 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2556 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2557 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2558 #endif // ASSERT
2559 break;
2560 }
2561 case LS_cmp_swap:
2562 case LS_cmp_swap_weak: {
2563 // Check the signatures.
2564 #ifdef ASSERT
2565 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2566 assert(sig->count() == 4, "CAS has 4 arguments");
2567 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2568 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2569 #endif // ASSERT
2570 break;
2571 }
2572 case LS_cmp_exchange: {
2573 // Check the signatures.
2574 #ifdef ASSERT
2575 assert(rtype == type, "CAS must return the expected type");
2576 assert(sig->count() == 4, "CAS has 4 arguments");
2577 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2578 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2579 #endif // ASSERT
2580 break;
2581 }
2582 default:
2583 ShouldNotReachHere();
2584 }
2585 }
2586 #endif //PRODUCT
2587
2588 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2589
2590 // Get arguments:
2591 Node* receiver = nullptr;
2592 Node* base = nullptr;
2593 Node* offset = nullptr;
2594 Node* oldval = nullptr;
2595 Node* newval = nullptr;
2596 switch(kind) {
2597 case LS_cmp_swap:
2598 case LS_cmp_swap_weak:
2599 case LS_cmp_exchange: {
2600 const bool two_slot_type = type2size[type] == 2;
2601 receiver = argument(0); // type: oop
2602 base = argument(1); // type: oop
2603 offset = argument(2); // type: long
2604 oldval = argument(4); // type: oop, int, or long
2605 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2606 break;
2607 }
2608 case LS_get_add:
2609 case LS_get_set: {
2610 receiver = argument(0); // type: oop
2611 base = argument(1); // type: oop
2612 offset = argument(2); // type: long
2613 oldval = nullptr;
2614 newval = argument(4); // type: oop, int, or long
2615 break;
2616 }
2617 default:
2618 ShouldNotReachHere();
2619 }
2620
2621 // Build field offset expression.
2622 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2623 // to be plain byte offsets, which are also the same as those accepted
2624 // by oopDesc::field_addr.
2625 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2626 // 32-bit machines ignore the high half of long offsets
2627 offset = ConvL2X(offset);
2628 // Save state and restore on bailout
2629 uint old_sp = sp();
2630 SafePointNode* old_map = clone_map();
2631 Node* adr = make_unsafe_address(base, offset,type, false);
2632 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2633
2634 Compile::AliasType* alias_type = C->alias_type(adr_type);
2635 BasicType bt = alias_type->basic_type();
2636 if (bt != T_ILLEGAL &&
2637 (is_reference_type(bt) != (type == T_OBJECT))) {
2638 // Don't intrinsify mismatched object accesses.
2639 set_map(old_map);
2640 set_sp(old_sp);
2641 return false;
2642 }
2643
2644 destruct_map_clone(old_map);
2645
2646 // For CAS, unlike inline_unsafe_access, there seems no point in
2647 // trying to refine types. Just use the coarse types here.
2648 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2649 const Type *value_type = Type::get_const_basic_type(type);
2650
2651 switch (kind) {
2652 case LS_get_set:
2653 case LS_cmp_exchange: {
2654 if (type == T_OBJECT) {
2655 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2656 if (tjp != nullptr) {
2657 value_type = tjp;
2658 }
2659 }
2660 break;
2661 }
2662 case LS_cmp_swap:
2663 case LS_cmp_swap_weak:
2664 case LS_get_add:
2665 break;
2666 default:
2667 ShouldNotReachHere();
2668 }
2669
2670 // Null check receiver.
2671 receiver = null_check(receiver);
2672 if (stopped()) {
2673 return true;
2674 }
2675
2676 int alias_idx = C->get_alias_index(adr_type);
2677
2678 if (is_reference_type(type)) {
2679 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2680
2681 // Transformation of a value which could be null pointer (CastPP #null)
2682 // could be delayed during Parse (for example, in adjust_map_after_if()).
2683 // Execute transformation here to avoid barrier generation in such case.
2684 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2685 newval = _gvn.makecon(TypePtr::NULL_PTR);
2686
2687 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2688 // Refine the value to a null constant, when it is known to be null
2689 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2690 }
2691 }
2692
2693 Node* result = nullptr;
2694 switch (kind) {
2695 case LS_cmp_exchange: {
2696 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2697 oldval, newval, value_type, type, decorators);
2698 break;
2699 }
2700 case LS_cmp_swap_weak:
2701 decorators |= C2_WEAK_CMPXCHG;
2702 case LS_cmp_swap: {
2703 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2704 oldval, newval, value_type, type, decorators);
2705 break;
2706 }
2707 case LS_get_set: {
2708 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2709 newval, value_type, type, decorators);
2710 break;
2711 }
2712 case LS_get_add: {
2713 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2714 newval, value_type, type, decorators);
2715 break;
2716 }
2717 default:
2718 ShouldNotReachHere();
2719 }
2720
2721 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2722 set_result(result);
2723 return true;
2724 }
2725
2726 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2727 // Regardless of form, don't allow previous ld/st to move down,
2728 // then issue acquire, release, or volatile mem_bar.
2729 insert_mem_bar(Op_MemBarCPUOrder);
2730 switch(id) {
2731 case vmIntrinsics::_loadFence:
2732 insert_mem_bar(Op_LoadFence);
2733 return true;
2734 case vmIntrinsics::_storeFence:
2735 insert_mem_bar(Op_StoreFence);
2736 return true;
2737 case vmIntrinsics::_storeStoreFence:
2738 insert_mem_bar(Op_StoreStoreFence);
2739 return true;
2740 case vmIntrinsics::_fullFence:
2741 insert_mem_bar(Op_MemBarVolatile);
2742 return true;
2743 default:
2744 fatal_unexpected_iid(id);
2745 return false;
2746 }
2747 }
2748
2749 bool LibraryCallKit::inline_onspinwait() {
2750 insert_mem_bar(Op_OnSpinWait);
2751 return true;
2752 }
2753
2754 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2755 if (!kls->is_Con()) {
2756 return true;
2757 }
2758 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2759 if (klsptr == nullptr) {
2760 return true;
2761 }
2762 ciInstanceKlass* ik = klsptr->instance_klass();
2763 // don't need a guard for a klass that is already initialized
2764 return !ik->is_initialized();
2765 }
2766
2767 //----------------------------inline_unsafe_writeback0-------------------------
2768 // public native void Unsafe.writeback0(long address)
2769 bool LibraryCallKit::inline_unsafe_writeback0() {
2770 if (!Matcher::has_match_rule(Op_CacheWB)) {
2771 return false;
2772 }
2773 #ifndef PRODUCT
2774 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2775 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2776 ciSignature* sig = callee()->signature();
2777 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2778 #endif
2779 null_check_receiver(); // null-check, then ignore
2780 Node *addr = argument(1);
2781 addr = new CastX2PNode(addr);
2782 addr = _gvn.transform(addr);
2783 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2784 flush = _gvn.transform(flush);
2785 set_memory(flush, TypeRawPtr::BOTTOM);
2786 return true;
2787 }
2788
2789 //----------------------------inline_unsafe_writeback0-------------------------
2790 // public native void Unsafe.writeback0(long address)
2791 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2792 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2793 return false;
2794 }
2795 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2796 return false;
2797 }
2798 #ifndef PRODUCT
2799 assert(Matcher::has_match_rule(Op_CacheWB),
2800 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2801 : "found match rule for CacheWBPostSync but not CacheWB"));
2802
2803 #endif
2804 null_check_receiver(); // null-check, then ignore
2805 Node *sync;
2806 if (is_pre) {
2807 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2808 } else {
2809 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2810 }
2811 sync = _gvn.transform(sync);
2812 set_memory(sync, TypeRawPtr::BOTTOM);
2813 return true;
2814 }
2815
2816 //----------------------------inline_unsafe_allocate---------------------------
2817 // public native Object Unsafe.allocateInstance(Class<?> cls);
2818 bool LibraryCallKit::inline_unsafe_allocate() {
2819
2820 #if INCLUDE_JVMTI
2821 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2822 return false;
2823 }
2824 #endif //INCLUDE_JVMTI
2825
2826 if (callee()->is_static()) return false; // caller must have the capability!
2827
2828 null_check_receiver(); // null-check, then ignore
2829 Node* cls = null_check(argument(1));
2830 if (stopped()) return true;
2831
2832 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2833 kls = null_check(kls);
2834 if (stopped()) return true; // argument was like int.class
2835
2836 #if INCLUDE_JVMTI
2837 // Don't try to access new allocated obj in the intrinsic.
2838 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2839 // Deoptimize and allocate in interpreter instead.
2840 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2841 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2842 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2843 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2844 {
2845 BuildCutout unless(this, tst, PROB_MAX);
2846 uncommon_trap(Deoptimization::Reason_intrinsic,
2847 Deoptimization::Action_make_not_entrant);
2848 }
2849 if (stopped()) {
2850 return true;
2851 }
2852 #endif //INCLUDE_JVMTI
2853
2854 Node* test = nullptr;
2855 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2856 // Note: The argument might still be an illegal value like
2857 // Serializable.class or Object[].class. The runtime will handle it.
2858 // But we must make an explicit check for initialization.
2859 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2860 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2861 // can generate code to load it as unsigned byte.
2862 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2863 Node* bits = intcon(InstanceKlass::fully_initialized);
2864 test = _gvn.transform(new SubINode(inst, bits));
2865 // The 'test' is non-zero if we need to take a slow path.
2866 }
2867
2868 Node* obj = new_instance(kls, test);
2869 set_result(obj);
2870 return true;
2871 }
2872
2873 //------------------------inline_native_time_funcs--------------
2874 // inline code for System.currentTimeMillis() and System.nanoTime()
2875 // these have the same type and signature
2876 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2877 const TypeFunc* tf = OptoRuntime::void_long_Type();
2878 const TypePtr* no_memory_effects = nullptr;
2879 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2880 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2881 #ifdef ASSERT
2882 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2883 assert(value_top == top(), "second value must be top");
2884 #endif
2885 set_result(value);
2886 return true;
2887 }
2888
2889
2890 #if INCLUDE_JVMTI
2891
2892 // When notifications are disabled then just update the VTMS transition bit and return.
2893 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
2894 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
2895 if (!DoJVMTIVirtualThreadTransitions) {
2896 return true;
2897 }
2898 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
2899 IdealKit ideal(this);
2900
2901 Node* ONE = ideal.ConI(1);
2902 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
2903 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
2904 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2905
2906 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
2907 sync_kit(ideal);
2908 // if notifyJvmti enabled then make a call to the given SharedRuntime function
2909 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
2910 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
2911 ideal.sync_kit(this);
2912 } ideal.else_(); {
2913 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
2914 Node* thread = ideal.thread();
2915 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
2916 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
2917 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
2918
2919 sync_kit(ideal);
2920 access_store_at(nullptr, jt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2921 access_store_at(nullptr, vt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2922
2923 ideal.sync_kit(this);
2924 } ideal.end_if();
2925 final_sync(ideal);
2926
2927 return true;
2928 }
2929
2930 // Always update the temporary VTMS transition bit.
2931 bool LibraryCallKit::inline_native_notify_jvmti_hide() {
2932 if (!DoJVMTIVirtualThreadTransitions) {
2933 return true;
2934 }
2935 IdealKit ideal(this);
2936
2937 {
2938 // unconditionally update the temporary VTMS transition bit in current JavaThread
2939 Node* thread = ideal.thread();
2940 Node* hide = _gvn.transform(argument(1)); // hide argument for temporary VTMS transition notification
2941 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_tmp_VTMS_transition_offset()));
2942 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
2943
2944 sync_kit(ideal);
2945 access_store_at(nullptr, addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2946 ideal.sync_kit(this);
2947 }
2948 final_sync(ideal);
2949
2950 return true;
2951 }
2952
2953 #endif // INCLUDE_JVMTI
2954
2955 #ifdef JFR_HAVE_INTRINSICS
2956
2957 /**
2958 * if oop->klass != null
2959 * // normal class
2960 * epoch = _epoch_state ? 2 : 1
2961 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
2962 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
2963 * }
2964 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
2965 * else
2966 * // primitive class
2967 * if oop->array_klass != null
2968 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
2969 * else
2970 * id = LAST_TYPE_ID + 1 // void class path
2971 * if (!signaled)
2972 * signaled = true
2973 */
2974 bool LibraryCallKit::inline_native_classID() {
2975 Node* cls = argument(0);
2976
2977 IdealKit ideal(this);
2978 #define __ ideal.
2979 IdealVariable result(ideal); __ declarations_done();
2980 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
2981 basic_plus_adr(cls, java_lang_Class::klass_offset()),
2982 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
2983
2984
2985 __ if_then(kls, BoolTest::ne, null()); {
2986 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
2987 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
2988
2989 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
2990 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2991 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
2992 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
2993 mask = _gvn.transform(new OrLNode(mask, epoch));
2994 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
2995
2996 float unlikely = PROB_UNLIKELY(0.999);
2997 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
2998 sync_kit(ideal);
2999 make_runtime_call(RC_LEAF,
3000 OptoRuntime::class_id_load_barrier_Type(),
3001 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3002 "class id load barrier",
3003 TypePtr::BOTTOM,
3004 kls);
3005 ideal.sync_kit(this);
3006 } __ end_if();
3007
3008 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3009 } __ else_(); {
3010 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3011 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3012 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3013 __ if_then(array_kls, BoolTest::ne, null()); {
3014 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3015 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3016 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3017 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3018 } __ else_(); {
3019 // void class case
3020 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3021 } __ end_if();
3022
3023 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3024 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3025 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3026 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3027 } __ end_if();
3028 } __ end_if();
3029
3030 final_sync(ideal);
3031 set_result(ideal.value(result));
3032 #undef __
3033 return true;
3034 }
3035
3036 //------------------------inline_native_jvm_commit------------------
3037 bool LibraryCallKit::inline_native_jvm_commit() {
3038 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3039
3040 // Save input memory and i_o state.
3041 Node* input_memory_state = reset_memory();
3042 set_all_memory(input_memory_state);
3043 Node* input_io_state = i_o();
3044
3045 // TLS.
3046 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3047 // Jfr java buffer.
3048 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3049 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3050 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3051
3052 // Load the current value of the notified field in the JfrThreadLocal.
3053 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3054 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3055
3056 // Test for notification.
3057 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3058 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3059 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3060
3061 // True branch, is notified.
3062 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3063 set_control(is_notified);
3064
3065 // Reset notified state.
3066 Node* notified_reset_memory = store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered);
3067
3068 // Iff notified, the return address of the commit method is the current position of the backing java buffer. This is used to reset the event writer.
3069 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3070 // Convert the machine-word to a long.
3071 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3072
3073 // False branch, not notified.
3074 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3075 set_control(not_notified);
3076 set_all_memory(input_memory_state);
3077
3078 // Arg is the next position as a long.
3079 Node* arg = argument(0);
3080 // Convert long to machine-word.
3081 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3082
3083 // Store the next_position to the underlying jfr java buffer.
3084 Node* commit_memory;
3085 #ifdef _LP64
3086 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_LONG, Compile::AliasIdxRaw, MemNode::release);
3087 #else
3088 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_INT, Compile::AliasIdxRaw, MemNode::release);
3089 #endif
3090
3091 // Now load the flags from off the java buffer and decide if the buffer is a lease. If so, it needs to be returned post-commit.
3092 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3093 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3094 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3095
3096 // And flags with lease constant.
3097 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3098
3099 // Branch on lease to conditionalize returning the leased java buffer.
3100 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3101 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3102 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3103
3104 // False branch, not a lease.
3105 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3106
3107 // True branch, is lease.
3108 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3109 set_control(is_lease);
3110
3111 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3112 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3113 OptoRuntime::void_void_Type(),
3114 StubRoutines::jfr_return_lease(),
3115 "return_lease", TypePtr::BOTTOM);
3116 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3117
3118 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3119 record_for_igvn(lease_compare_rgn);
3120 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3121 record_for_igvn(lease_compare_mem);
3122 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3123 record_for_igvn(lease_compare_io);
3124 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3125 record_for_igvn(lease_result_value);
3126
3127 // Update control and phi nodes.
3128 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3129 lease_compare_rgn->init_req(_false_path, not_lease);
3130
3131 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3132 lease_compare_mem->init_req(_false_path, commit_memory);
3133
3134 lease_compare_io->init_req(_true_path, i_o());
3135 lease_compare_io->init_req(_false_path, input_io_state);
3136
3137 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3138 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3139
3140 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3141 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3142 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3143 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3144
3145 // Update control and phi nodes.
3146 result_rgn->init_req(_true_path, is_notified);
3147 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3148
3149 result_mem->init_req(_true_path, notified_reset_memory);
3150 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3151
3152 result_io->init_req(_true_path, input_io_state);
3153 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3154
3155 result_value->init_req(_true_path, current_pos);
3156 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3157
3158 // Set output state.
3159 set_control(_gvn.transform(result_rgn));
3160 set_all_memory(_gvn.transform(result_mem));
3161 set_i_o(_gvn.transform(result_io));
3162 set_result(result_rgn, result_value);
3163 return true;
3164 }
3165
3166 /*
3167 * The intrinsic is a model of this pseudo-code:
3168 *
3169 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3170 * jobject h_event_writer = tl->java_event_writer();
3171 * if (h_event_writer == nullptr) {
3172 * return nullptr;
3173 * }
3174 * oop threadObj = Thread::threadObj();
3175 * oop vthread = java_lang_Thread::vthread(threadObj);
3176 * traceid tid;
3177 * bool excluded;
3178 * if (vthread != threadObj) { // i.e. current thread is virtual
3179 * tid = java_lang_Thread::tid(vthread);
3180 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3181 * excluded = vthread_epoch_raw & excluded_mask;
3182 * if (!excluded) {
3183 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3184 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3185 * if (vthread_epoch != current_epoch) {
3186 * write_checkpoint();
3187 * }
3188 * }
3189 * } else {
3190 * tid = java_lang_Thread::tid(threadObj);
3191 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3192 * excluded = thread_epoch_raw & excluded_mask;
3193 * }
3194 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3195 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3196 * if (tid_in_event_writer != tid) {
3197 * setField(event_writer, "threadID", tid);
3198 * setField(event_writer, "excluded", excluded);
3199 * }
3200 * return event_writer
3201 */
3202 bool LibraryCallKit::inline_native_getEventWriter() {
3203 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3204
3205 // Save input memory and i_o state.
3206 Node* input_memory_state = reset_memory();
3207 set_all_memory(input_memory_state);
3208 Node* input_io_state = i_o();
3209
3210 Node* excluded_mask = _gvn.intcon(32768);
3211 Node* epoch_mask = _gvn.intcon(32767);
3212
3213 // TLS
3214 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3215
3216 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3217 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3218
3219 // Load the eventwriter jobject handle.
3220 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3221
3222 // Null check the jobject handle.
3223 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3224 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3225 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3226
3227 // False path, jobj is null.
3228 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3229
3230 // True path, jobj is not null.
3231 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3232
3233 set_control(jobj_is_not_null);
3234
3235 // Load the threadObj for the CarrierThread.
3236 Node* threadObj = generate_current_thread(tls_ptr);
3237
3238 // Load the vthread.
3239 Node* vthread = generate_virtual_thread(tls_ptr);
3240
3241 // If vthread != threadObj, this is a virtual thread.
3242 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3243 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3244 IfNode* iff_vthread_not_equal_threadObj =
3245 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3246
3247 // False branch, fallback to threadObj.
3248 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3249 set_control(vthread_equal_threadObj);
3250
3251 // Load the tid field from the vthread object.
3252 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3253
3254 // Load the raw epoch value from the threadObj.
3255 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3256 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3257 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3258
3259 // Mask off the excluded information from the epoch.
3260 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3261
3262 // True branch, this is a virtual thread.
3263 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3264 set_control(vthread_not_equal_threadObj);
3265
3266 // Load the tid field from the vthread object.
3267 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3268
3269 // Load the raw epoch value from the vthread.
3270 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3271 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3272 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3273
3274 // Mask off the excluded information from the epoch.
3275 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3276
3277 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3278 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3279 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3280 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3281
3282 // False branch, vthread is excluded, no need to write epoch info.
3283 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3284
3285 // True branch, vthread is included, update epoch info.
3286 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3287 set_control(included);
3288
3289 // Get epoch value.
3290 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3291
3292 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3293 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3294 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3295
3296 // Compare the epoch in the vthread to the current epoch generation.
3297 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3298 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3299 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3300
3301 // False path, epoch is equal, checkpoint information is valid.
3302 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3303
3304 // True path, epoch is not equal, write a checkpoint for the vthread.
3305 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3306
3307 set_control(epoch_is_not_equal);
3308
3309 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3310 // The call also updates the native thread local thread id and the vthread with the current epoch.
3311 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3312 OptoRuntime::jfr_write_checkpoint_Type(),
3313 StubRoutines::jfr_write_checkpoint(),
3314 "write_checkpoint", TypePtr::BOTTOM);
3315 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3316
3317 // vthread epoch != current epoch
3318 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3319 record_for_igvn(epoch_compare_rgn);
3320 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3321 record_for_igvn(epoch_compare_mem);
3322 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3323 record_for_igvn(epoch_compare_io);
3324
3325 // Update control and phi nodes.
3326 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3327 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3328 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3329 epoch_compare_mem->init_req(_false_path, input_memory_state);
3330 epoch_compare_io->init_req(_true_path, i_o());
3331 epoch_compare_io->init_req(_false_path, input_io_state);
3332
3333 // excluded != true
3334 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3335 record_for_igvn(exclude_compare_rgn);
3336 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3337 record_for_igvn(exclude_compare_mem);
3338 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3339 record_for_igvn(exclude_compare_io);
3340
3341 // Update control and phi nodes.
3342 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3343 exclude_compare_rgn->init_req(_false_path, excluded);
3344 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3345 exclude_compare_mem->init_req(_false_path, input_memory_state);
3346 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3347 exclude_compare_io->init_req(_false_path, input_io_state);
3348
3349 // vthread != threadObj
3350 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3351 record_for_igvn(vthread_compare_rgn);
3352 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3353 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3354 record_for_igvn(vthread_compare_io);
3355 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3356 record_for_igvn(tid);
3357 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3358 record_for_igvn(exclusion);
3359
3360 // Update control and phi nodes.
3361 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3362 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3363 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3364 vthread_compare_mem->init_req(_false_path, input_memory_state);
3365 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3366 vthread_compare_io->init_req(_false_path, input_io_state);
3367 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3368 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3369 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3370 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3371
3372 // Update branch state.
3373 set_control(_gvn.transform(vthread_compare_rgn));
3374 set_all_memory(_gvn.transform(vthread_compare_mem));
3375 set_i_o(_gvn.transform(vthread_compare_io));
3376
3377 // Load the event writer oop by dereferencing the jobject handle.
3378 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3379 assert(klass_EventWriter->is_loaded(), "invariant");
3380 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3381 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3382 const TypeOopPtr* const xtype = aklass->as_instance_type();
3383 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3384 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3385
3386 // Load the current thread id from the event writer object.
3387 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3388 // Get the field offset to, conditionally, store an updated tid value later.
3389 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3390 const TypePtr* event_writer_tid_field_type = _gvn.type(event_writer_tid_field)->isa_ptr();
3391 // Get the field offset to, conditionally, store an updated exclusion value later.
3392 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3393 const TypePtr* event_writer_excluded_field_type = _gvn.type(event_writer_excluded_field)->isa_ptr();
3394
3395 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3396 record_for_igvn(event_writer_tid_compare_rgn);
3397 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3398 record_for_igvn(event_writer_tid_compare_mem);
3399 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3400 record_for_igvn(event_writer_tid_compare_io);
3401
3402 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3403 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3404 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3405 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3406
3407 // False path, tids are the same.
3408 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3409
3410 // True path, tid is not equal, need to update the tid in the event writer.
3411 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3412 record_for_igvn(tid_is_not_equal);
3413
3414 // Store the exclusion state to the event writer.
3415 store_to_memory(tid_is_not_equal, event_writer_excluded_field, _gvn.transform(exclusion), T_BOOLEAN, event_writer_excluded_field_type, MemNode::unordered);
3416
3417 // Store the tid to the event writer.
3418 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, event_writer_tid_field_type, MemNode::unordered);
3419
3420 // Update control and phi nodes.
3421 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3422 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3423 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3424 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3425 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3426 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3427
3428 // Result of top level CFG, Memory, IO and Value.
3429 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3430 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3431 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3432 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3433
3434 // Result control.
3435 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3436 result_rgn->init_req(_false_path, jobj_is_null);
3437
3438 // Result memory.
3439 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3440 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3441
3442 // Result IO.
3443 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3444 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3445
3446 // Result value.
3447 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3448 result_value->init_req(_false_path, null()); // return null
3449
3450 // Set output state.
3451 set_control(_gvn.transform(result_rgn));
3452 set_all_memory(_gvn.transform(result_mem));
3453 set_i_o(_gvn.transform(result_io));
3454 set_result(result_rgn, result_value);
3455 return true;
3456 }
3457
3458 /*
3459 * The intrinsic is a model of this pseudo-code:
3460 *
3461 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3462 * if (carrierThread != thread) { // is virtual thread
3463 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3464 * bool excluded = vthread_epoch_raw & excluded_mask;
3465 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3466 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3467 * if (!excluded) {
3468 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3469 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3470 * }
3471 * Atomic::release_store(&tl->_vthread, true);
3472 * return;
3473 * }
3474 * Atomic::release_store(&tl->_vthread, false);
3475 */
3476 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3477 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3478
3479 Node* input_memory_state = reset_memory();
3480 set_all_memory(input_memory_state);
3481
3482 Node* excluded_mask = _gvn.intcon(32768);
3483 Node* epoch_mask = _gvn.intcon(32767);
3484
3485 Node* const carrierThread = generate_current_thread(jt);
3486 // If thread != carrierThread, this is a virtual thread.
3487 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3488 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3489 IfNode* iff_thread_not_equal_carrierThread =
3490 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3491
3492 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3493
3494 // False branch, is carrierThread.
3495 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3496 // Store release
3497 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3498
3499 set_all_memory(input_memory_state);
3500
3501 // True branch, is virtual thread.
3502 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3503 set_control(thread_not_equal_carrierThread);
3504
3505 // Load the raw epoch value from the vthread.
3506 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3507 Node* epoch_raw = access_load_at(thread, epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3508 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3509
3510 // Mask off the excluded information from the epoch.
3511 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3512
3513 // Load the tid field from the thread.
3514 Node* tid = load_field_from_object(thread, "tid", "J");
3515
3516 // Store the vthread tid to the jfr thread local.
3517 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3518 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, Compile::AliasIdxRaw, MemNode::unordered, true);
3519
3520 // Branch is_excluded to conditionalize updating the epoch .
3521 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3522 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3523 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3524
3525 // True branch, vthread is excluded, no need to write epoch info.
3526 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3527 set_control(excluded);
3528 Node* vthread_is_excluded = _gvn.intcon(1);
3529
3530 // False branch, vthread is included, update epoch info.
3531 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3532 set_control(included);
3533 Node* vthread_is_included = _gvn.intcon(0);
3534
3535 // Get epoch value.
3536 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3537
3538 // Store the vthread epoch to the jfr thread local.
3539 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3540 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, Compile::AliasIdxRaw, MemNode::unordered, true);
3541
3542 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3543 record_for_igvn(excluded_rgn);
3544 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3545 record_for_igvn(excluded_mem);
3546 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3547 record_for_igvn(exclusion);
3548
3549 // Merge the excluded control and memory.
3550 excluded_rgn->init_req(_true_path, excluded);
3551 excluded_rgn->init_req(_false_path, included);
3552 excluded_mem->init_req(_true_path, tid_memory);
3553 excluded_mem->init_req(_false_path, included_memory);
3554 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3555 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3556
3557 // Set intermediate state.
3558 set_control(_gvn.transform(excluded_rgn));
3559 set_all_memory(excluded_mem);
3560
3561 // Store the vthread exclusion state to the jfr thread local.
3562 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3563 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered, true);
3564
3565 // Store release
3566 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3567
3568 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3569 record_for_igvn(thread_compare_rgn);
3570 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3571 record_for_igvn(thread_compare_mem);
3572 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3573 record_for_igvn(vthread);
3574
3575 // Merge the thread_compare control and memory.
3576 thread_compare_rgn->init_req(_true_path, control());
3577 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3578 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3579 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3580
3581 // Set output state.
3582 set_control(_gvn.transform(thread_compare_rgn));
3583 set_all_memory(_gvn.transform(thread_compare_mem));
3584 }
3585
3586 #endif // JFR_HAVE_INTRINSICS
3587
3588 //------------------------inline_native_currentCarrierThread------------------
3589 bool LibraryCallKit::inline_native_currentCarrierThread() {
3590 Node* junk = nullptr;
3591 set_result(generate_current_thread(junk));
3592 return true;
3593 }
3594
3595 //------------------------inline_native_currentThread------------------
3596 bool LibraryCallKit::inline_native_currentThread() {
3597 Node* junk = nullptr;
3598 set_result(generate_virtual_thread(junk));
3599 return true;
3600 }
3601
3602 //------------------------inline_native_setVthread------------------
3603 bool LibraryCallKit::inline_native_setCurrentThread() {
3604 assert(C->method()->changes_current_thread(),
3605 "method changes current Thread but is not annotated ChangesCurrentThread");
3606 Node* arr = argument(1);
3607 Node* thread = _gvn.transform(new ThreadLocalNode());
3608 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3609 Node* thread_obj_handle
3610 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3611 thread_obj_handle = _gvn.transform(thread_obj_handle);
3612 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3613 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3614 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3615 return true;
3616 }
3617
3618 const Type* LibraryCallKit::scopedValueCache_type() {
3619 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3620 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3621 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3622
3623 // Because we create the scopedValue cache lazily we have to make the
3624 // type of the result BotPTR.
3625 bool xk = etype->klass_is_exact();
3626 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3627 return objects_type;
3628 }
3629
3630 Node* LibraryCallKit::scopedValueCache_helper() {
3631 Node* thread = _gvn.transform(new ThreadLocalNode());
3632 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3633 // We cannot use immutable_memory() because we might flip onto a
3634 // different carrier thread, at which point we'll need to use that
3635 // carrier thread's cache.
3636 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3637 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3638 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3639 }
3640
3641 //------------------------inline_native_scopedValueCache------------------
3642 bool LibraryCallKit::inline_native_scopedValueCache() {
3643 Node* cache_obj_handle = scopedValueCache_helper();
3644 const Type* objects_type = scopedValueCache_type();
3645 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3646
3647 return true;
3648 }
3649
3650 //------------------------inline_native_setScopedValueCache------------------
3651 bool LibraryCallKit::inline_native_setScopedValueCache() {
3652 Node* arr = argument(0);
3653 Node* cache_obj_handle = scopedValueCache_helper();
3654 const Type* objects_type = scopedValueCache_type();
3655
3656 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3657 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3658
3659 return true;
3660 }
3661
3662 //---------------------------load_mirror_from_klass----------------------------
3663 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3664 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3665 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3666 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3667 // mirror = ((OopHandle)mirror)->resolve();
3668 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3669 }
3670
3671 //-----------------------load_klass_from_mirror_common-------------------------
3672 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3673 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3674 // and branch to the given path on the region.
3675 // If never_see_null, take an uncommon trap on null, so we can optimistically
3676 // compile for the non-null case.
3677 // If the region is null, force never_see_null = true.
3678 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3679 bool never_see_null,
3680 RegionNode* region,
3681 int null_path,
3682 int offset) {
3683 if (region == nullptr) never_see_null = true;
3684 Node* p = basic_plus_adr(mirror, offset);
3685 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3686 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3687 Node* null_ctl = top();
3688 kls = null_check_oop(kls, &null_ctl, never_see_null);
3689 if (region != nullptr) {
3690 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3691 region->init_req(null_path, null_ctl);
3692 } else {
3693 assert(null_ctl == top(), "no loose ends");
3694 }
3695 return kls;
3696 }
3697
3698 //--------------------(inline_native_Class_query helpers)---------------------
3699 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3700 // Fall through if (mods & mask) == bits, take the guard otherwise.
3701 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3702 // Branch around if the given klass has the given modifier bit set.
3703 // Like generate_guard, adds a new path onto the region.
3704 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3705 Node* mods = make_load(nullptr, modp, TypeInt::INT, T_INT, MemNode::unordered);
3706 Node* mask = intcon(modifier_mask);
3707 Node* bits = intcon(modifier_bits);
3708 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3709 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3710 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3711 return generate_fair_guard(bol, region);
3712 }
3713 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3714 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3715 }
3716 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3717 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
3718 }
3719
3720 //-------------------------inline_native_Class_query-------------------
3721 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3722 const Type* return_type = TypeInt::BOOL;
3723 Node* prim_return_value = top(); // what happens if it's a primitive class?
3724 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3725 bool expect_prim = false; // most of these guys expect to work on refs
3726
3727 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3728
3729 Node* mirror = argument(0);
3730 Node* obj = top();
3731
3732 switch (id) {
3733 case vmIntrinsics::_isInstance:
3734 // nothing is an instance of a primitive type
3735 prim_return_value = intcon(0);
3736 obj = argument(1);
3737 break;
3738 case vmIntrinsics::_getModifiers:
3739 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3740 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3741 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3742 break;
3743 case vmIntrinsics::_isInterface:
3744 prim_return_value = intcon(0);
3745 break;
3746 case vmIntrinsics::_isArray:
3747 prim_return_value = intcon(0);
3748 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3749 break;
3750 case vmIntrinsics::_isPrimitive:
3751 prim_return_value = intcon(1);
3752 expect_prim = true; // obviously
3753 break;
3754 case vmIntrinsics::_isHidden:
3755 prim_return_value = intcon(0);
3756 break;
3757 case vmIntrinsics::_getSuperclass:
3758 prim_return_value = null();
3759 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3760 break;
3761 case vmIntrinsics::_getClassAccessFlags:
3762 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3763 return_type = TypeInt::INT; // not bool! 6297094
3764 break;
3765 default:
3766 fatal_unexpected_iid(id);
3767 break;
3768 }
3769
3770 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3771 if (mirror_con == nullptr) return false; // cannot happen?
3772
3773 #ifndef PRODUCT
3774 if (C->print_intrinsics() || C->print_inlining()) {
3775 ciType* k = mirror_con->java_mirror_type();
3776 if (k) {
3777 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3778 k->print_name();
3779 tty->cr();
3780 }
3781 }
3782 #endif
3783
3784 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3785 RegionNode* region = new RegionNode(PATH_LIMIT);
3786 record_for_igvn(region);
3787 PhiNode* phi = new PhiNode(region, return_type);
3788
3789 // The mirror will never be null of Reflection.getClassAccessFlags, however
3790 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3791 // if it is. See bug 4774291.
3792
3793 // For Reflection.getClassAccessFlags(), the null check occurs in
3794 // the wrong place; see inline_unsafe_access(), above, for a similar
3795 // situation.
3796 mirror = null_check(mirror);
3797 // If mirror or obj is dead, only null-path is taken.
3798 if (stopped()) return true;
3799
3800 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3801
3802 // Now load the mirror's klass metaobject, and null-check it.
3803 // Side-effects region with the control path if the klass is null.
3804 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3805 // If kls is null, we have a primitive mirror.
3806 phi->init_req(_prim_path, prim_return_value);
3807 if (stopped()) { set_result(region, phi); return true; }
3808 bool safe_for_replace = (region->in(_prim_path) == top());
3809
3810 Node* p; // handy temp
3811 Node* null_ctl;
3812
3813 // Now that we have the non-null klass, we can perform the real query.
3814 // For constant classes, the query will constant-fold in LoadNode::Value.
3815 Node* query_value = top();
3816 switch (id) {
3817 case vmIntrinsics::_isInstance:
3818 // nothing is an instance of a primitive type
3819 query_value = gen_instanceof(obj, kls, safe_for_replace);
3820 break;
3821
3822 case vmIntrinsics::_getModifiers:
3823 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3824 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
3825 break;
3826
3827 case vmIntrinsics::_isInterface:
3828 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3829 if (generate_interface_guard(kls, region) != nullptr)
3830 // A guard was added. If the guard is taken, it was an interface.
3831 phi->add_req(intcon(1));
3832 // If we fall through, it's a plain class.
3833 query_value = intcon(0);
3834 break;
3835
3836 case vmIntrinsics::_isArray:
3837 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3838 if (generate_array_guard(kls, region) != nullptr)
3839 // A guard was added. If the guard is taken, it was an array.
3840 phi->add_req(intcon(1));
3841 // If we fall through, it's a plain class.
3842 query_value = intcon(0);
3843 break;
3844
3845 case vmIntrinsics::_isPrimitive:
3846 query_value = intcon(0); // "normal" path produces false
3847 break;
3848
3849 case vmIntrinsics::_isHidden:
3850 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
3851 if (generate_hidden_class_guard(kls, region) != nullptr)
3852 // A guard was added. If the guard is taken, it was an hidden class.
3853 phi->add_req(intcon(1));
3854 // If we fall through, it's a plain class.
3855 query_value = intcon(0);
3856 break;
3857
3858
3859 case vmIntrinsics::_getSuperclass:
3860 // The rules here are somewhat unfortunate, but we can still do better
3861 // with random logic than with a JNI call.
3862 // Interfaces store null or Object as _super, but must report null.
3863 // Arrays store an intermediate super as _super, but must report Object.
3864 // Other types can report the actual _super.
3865 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3866 if (generate_interface_guard(kls, region) != nullptr)
3867 // A guard was added. If the guard is taken, it was an interface.
3868 phi->add_req(null());
3869 if (generate_array_guard(kls, region) != nullptr)
3870 // A guard was added. If the guard is taken, it was an array.
3871 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3872 // If we fall through, it's a plain class. Get its _super.
3873 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3874 kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3875 null_ctl = top();
3876 kls = null_check_oop(kls, &null_ctl);
3877 if (null_ctl != top()) {
3878 // If the guard is taken, Object.superClass is null (both klass and mirror).
3879 region->add_req(null_ctl);
3880 phi ->add_req(null());
3881 }
3882 if (!stopped()) {
3883 query_value = load_mirror_from_klass(kls);
3884 }
3885 break;
3886
3887 case vmIntrinsics::_getClassAccessFlags:
3888 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3889 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
3890 break;
3891
3892 default:
3893 fatal_unexpected_iid(id);
3894 break;
3895 }
3896
3897 // Fall-through is the normal case of a query to a real class.
3898 phi->init_req(1, query_value);
3899 region->init_req(1, control());
3900
3901 C->set_has_split_ifs(true); // Has chance for split-if optimization
3902 set_result(region, phi);
3903 return true;
3904 }
3905
3906 //-------------------------inline_Class_cast-------------------
3907 bool LibraryCallKit::inline_Class_cast() {
3908 Node* mirror = argument(0); // Class
3909 Node* obj = argument(1);
3910 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3911 if (mirror_con == nullptr) {
3912 return false; // dead path (mirror->is_top()).
3913 }
3914 if (obj == nullptr || obj->is_top()) {
3915 return false; // dead path
3916 }
3917 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3918
3919 // First, see if Class.cast() can be folded statically.
3920 // java_mirror_type() returns non-null for compile-time Class constants.
3921 ciType* tm = mirror_con->java_mirror_type();
3922 if (tm != nullptr && tm->is_klass() &&
3923 tp != nullptr) {
3924 if (!tp->is_loaded()) {
3925 // Don't use intrinsic when class is not loaded.
3926 return false;
3927 } else {
3928 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
3929 if (static_res == Compile::SSC_always_true) {
3930 // isInstance() is true - fold the code.
3931 set_result(obj);
3932 return true;
3933 } else if (static_res == Compile::SSC_always_false) {
3934 // Don't use intrinsic, have to throw ClassCastException.
3935 // If the reference is null, the non-intrinsic bytecode will
3936 // be optimized appropriately.
3937 return false;
3938 }
3939 }
3940 }
3941
3942 // Bailout intrinsic and do normal inlining if exception path is frequent.
3943 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3944 return false;
3945 }
3946
3947 // Generate dynamic checks.
3948 // Class.cast() is java implementation of _checkcast bytecode.
3949 // Do checkcast (Parse::do_checkcast()) optimizations here.
3950
3951 mirror = null_check(mirror);
3952 // If mirror is dead, only null-path is taken.
3953 if (stopped()) {
3954 return true;
3955 }
3956
3957 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
3958 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3959 RegionNode* region = new RegionNode(PATH_LIMIT);
3960 record_for_igvn(region);
3961
3962 // Now load the mirror's klass metaobject, and null-check it.
3963 // If kls is null, we have a primitive mirror and
3964 // nothing is an instance of a primitive type.
3965 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3966
3967 Node* res = top();
3968 if (!stopped()) {
3969 Node* bad_type_ctrl = top();
3970 // Do checkcast optimizations.
3971 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3972 region->init_req(_bad_type_path, bad_type_ctrl);
3973 }
3974 if (region->in(_prim_path) != top() ||
3975 region->in(_bad_type_path) != top()) {
3976 // Let Interpreter throw ClassCastException.
3977 PreserveJVMState pjvms(this);
3978 set_control(_gvn.transform(region));
3979 uncommon_trap(Deoptimization::Reason_intrinsic,
3980 Deoptimization::Action_maybe_recompile);
3981 }
3982 if (!stopped()) {
3983 set_result(res);
3984 }
3985 return true;
3986 }
3987
3988
3989 //--------------------------inline_native_subtype_check------------------------
3990 // This intrinsic takes the JNI calls out of the heart of
3991 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3992 bool LibraryCallKit::inline_native_subtype_check() {
3993 // Pull both arguments off the stack.
3994 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3995 args[0] = argument(0);
3996 args[1] = argument(1);
3997 Node* klasses[2]; // corresponding Klasses: superk, subk
3998 klasses[0] = klasses[1] = top();
3999
4000 enum {
4001 // A full decision tree on {superc is prim, subc is prim}:
4002 _prim_0_path = 1, // {P,N} => false
4003 // {P,P} & superc!=subc => false
4004 _prim_same_path, // {P,P} & superc==subc => true
4005 _prim_1_path, // {N,P} => false
4006 _ref_subtype_path, // {N,N} & subtype check wins => true
4007 _both_ref_path, // {N,N} & subtype check loses => false
4008 PATH_LIMIT
4009 };
4010
4011 RegionNode* region = new RegionNode(PATH_LIMIT);
4012 Node* phi = new PhiNode(region, TypeInt::BOOL);
4013 record_for_igvn(region);
4014
4015 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4016 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4017 int class_klass_offset = java_lang_Class::klass_offset();
4018
4019 // First null-check both mirrors and load each mirror's klass metaobject.
4020 int which_arg;
4021 for (which_arg = 0; which_arg <= 1; which_arg++) {
4022 Node* arg = args[which_arg];
4023 arg = null_check(arg);
4024 if (stopped()) break;
4025 args[which_arg] = arg;
4026
4027 Node* p = basic_plus_adr(arg, class_klass_offset);
4028 Node* kls = LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, adr_type, kls_type);
4029 klasses[which_arg] = _gvn.transform(kls);
4030 }
4031
4032 // Having loaded both klasses, test each for null.
4033 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4034 for (which_arg = 0; which_arg <= 1; which_arg++) {
4035 Node* kls = klasses[which_arg];
4036 Node* null_ctl = top();
4037 kls = null_check_oop(kls, &null_ctl, never_see_null);
4038 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4039 region->init_req(prim_path, null_ctl);
4040 if (stopped()) break;
4041 klasses[which_arg] = kls;
4042 }
4043
4044 if (!stopped()) {
4045 // now we have two reference types, in klasses[0..1]
4046 Node* subk = klasses[1]; // the argument to isAssignableFrom
4047 Node* superk = klasses[0]; // the receiver
4048 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4049 // now we have a successful reference subtype check
4050 region->set_req(_ref_subtype_path, control());
4051 }
4052
4053 // If both operands are primitive (both klasses null), then
4054 // we must return true when they are identical primitives.
4055 // It is convenient to test this after the first null klass check.
4056 set_control(region->in(_prim_0_path)); // go back to first null check
4057 if (!stopped()) {
4058 // Since superc is primitive, make a guard for the superc==subc case.
4059 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4060 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4061 generate_guard(bol_eq, region, PROB_FAIR);
4062 if (region->req() == PATH_LIMIT+1) {
4063 // A guard was added. If the added guard is taken, superc==subc.
4064 region->swap_edges(PATH_LIMIT, _prim_same_path);
4065 region->del_req(PATH_LIMIT);
4066 }
4067 region->set_req(_prim_0_path, control()); // Not equal after all.
4068 }
4069
4070 // these are the only paths that produce 'true':
4071 phi->set_req(_prim_same_path, intcon(1));
4072 phi->set_req(_ref_subtype_path, intcon(1));
4073
4074 // pull together the cases:
4075 assert(region->req() == PATH_LIMIT, "sane region");
4076 for (uint i = 1; i < region->req(); i++) {
4077 Node* ctl = region->in(i);
4078 if (ctl == nullptr || ctl == top()) {
4079 region->set_req(i, top());
4080 phi ->set_req(i, top());
4081 } else if (phi->in(i) == nullptr) {
4082 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4083 }
4084 }
4085
4086 set_control(_gvn.transform(region));
4087 set_result(_gvn.transform(phi));
4088 return true;
4089 }
4090
4091 //---------------------generate_array_guard_common------------------------
4092 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4093 bool obj_array, bool not_array) {
4094
4095 if (stopped()) {
4096 return nullptr;
4097 }
4098
4099 // If obj_array/non_array==false/false:
4100 // Branch around if the given klass is in fact an array (either obj or prim).
4101 // If obj_array/non_array==false/true:
4102 // Branch around if the given klass is not an array klass of any kind.
4103 // If obj_array/non_array==true/true:
4104 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4105 // If obj_array/non_array==true/false:
4106 // Branch around if the kls is an oop array (Object[] or subtype)
4107 //
4108 // Like generate_guard, adds a new path onto the region.
4109 jint layout_con = 0;
4110 Node* layout_val = get_layout_helper(kls, layout_con);
4111 if (layout_val == nullptr) {
4112 bool query = (obj_array
4113 ? Klass::layout_helper_is_objArray(layout_con)
4114 : Klass::layout_helper_is_array(layout_con));
4115 if (query == not_array) {
4116 return nullptr; // never a branch
4117 } else { // always a branch
4118 Node* always_branch = control();
4119 if (region != nullptr)
4120 region->add_req(always_branch);
4121 set_control(top());
4122 return always_branch;
4123 }
4124 }
4125 // Now test the correct condition.
4126 jint nval = (obj_array
4127 ? (jint)(Klass::_lh_array_tag_type_value
4128 << Klass::_lh_array_tag_shift)
4129 : Klass::_lh_neutral_value);
4130 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4131 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4132 // invert the test if we are looking for a non-array
4133 if (not_array) btest = BoolTest(btest).negate();
4134 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4135 return generate_fair_guard(bol, region);
4136 }
4137
4138
4139 //-----------------------inline_native_newArray--------------------------
4140 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4141 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4142 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4143 Node* mirror;
4144 Node* count_val;
4145 if (uninitialized) {
4146 null_check_receiver();
4147 mirror = argument(1);
4148 count_val = argument(2);
4149 } else {
4150 mirror = argument(0);
4151 count_val = argument(1);
4152 }
4153
4154 mirror = null_check(mirror);
4155 // If mirror or obj is dead, only null-path is taken.
4156 if (stopped()) return true;
4157
4158 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4159 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4160 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4161 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4162 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4163
4164 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4165 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4166 result_reg, _slow_path);
4167 Node* normal_ctl = control();
4168 Node* no_array_ctl = result_reg->in(_slow_path);
4169
4170 // Generate code for the slow case. We make a call to newArray().
4171 set_control(no_array_ctl);
4172 if (!stopped()) {
4173 // Either the input type is void.class, or else the
4174 // array klass has not yet been cached. Either the
4175 // ensuing call will throw an exception, or else it
4176 // will cache the array klass for next time.
4177 PreserveJVMState pjvms(this);
4178 CallJavaNode* slow_call = nullptr;
4179 if (uninitialized) {
4180 // Generate optimized virtual call (holder class 'Unsafe' is final)
4181 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4182 } else {
4183 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4184 }
4185 Node* slow_result = set_results_for_java_call(slow_call);
4186 // this->control() comes from set_results_for_java_call
4187 result_reg->set_req(_slow_path, control());
4188 result_val->set_req(_slow_path, slow_result);
4189 result_io ->set_req(_slow_path, i_o());
4190 result_mem->set_req(_slow_path, reset_memory());
4191 }
4192
4193 set_control(normal_ctl);
4194 if (!stopped()) {
4195 // Normal case: The array type has been cached in the java.lang.Class.
4196 // The following call works fine even if the array type is polymorphic.
4197 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4198 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4199 result_reg->init_req(_normal_path, control());
4200 result_val->init_req(_normal_path, obj);
4201 result_io ->init_req(_normal_path, i_o());
4202 result_mem->init_req(_normal_path, reset_memory());
4203
4204 if (uninitialized) {
4205 // Mark the allocation so that zeroing is skipped
4206 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4207 alloc->maybe_set_complete(&_gvn);
4208 }
4209 }
4210
4211 // Return the combined state.
4212 set_i_o( _gvn.transform(result_io) );
4213 set_all_memory( _gvn.transform(result_mem));
4214
4215 C->set_has_split_ifs(true); // Has chance for split-if optimization
4216 set_result(result_reg, result_val);
4217 return true;
4218 }
4219
4220 //----------------------inline_native_getLength--------------------------
4221 // public static native int java.lang.reflect.Array.getLength(Object array);
4222 bool LibraryCallKit::inline_native_getLength() {
4223 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4224
4225 Node* array = null_check(argument(0));
4226 // If array is dead, only null-path is taken.
4227 if (stopped()) return true;
4228
4229 // Deoptimize if it is a non-array.
4230 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr);
4231
4232 if (non_array != nullptr) {
4233 PreserveJVMState pjvms(this);
4234 set_control(non_array);
4235 uncommon_trap(Deoptimization::Reason_intrinsic,
4236 Deoptimization::Action_maybe_recompile);
4237 }
4238
4239 // If control is dead, only non-array-path is taken.
4240 if (stopped()) return true;
4241
4242 // The works fine even if the array type is polymorphic.
4243 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4244 Node* result = load_array_length(array);
4245
4246 C->set_has_split_ifs(true); // Has chance for split-if optimization
4247 set_result(result);
4248 return true;
4249 }
4250
4251 //------------------------inline_array_copyOf----------------------------
4252 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4253 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4254 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4255 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4256
4257 // Get the arguments.
4258 Node* original = argument(0);
4259 Node* start = is_copyOfRange? argument(1): intcon(0);
4260 Node* end = is_copyOfRange? argument(2): argument(1);
4261 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4262
4263 Node* newcopy = nullptr;
4264
4265 // Set the original stack and the reexecute bit for the interpreter to reexecute
4266 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4267 { PreserveReexecuteState preexecs(this);
4268 jvms()->set_should_reexecute(true);
4269
4270 array_type_mirror = null_check(array_type_mirror);
4271 original = null_check(original);
4272
4273 // Check if a null path was taken unconditionally.
4274 if (stopped()) return true;
4275
4276 Node* orig_length = load_array_length(original);
4277
4278 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4279 klass_node = null_check(klass_node);
4280
4281 RegionNode* bailout = new RegionNode(1);
4282 record_for_igvn(bailout);
4283
4284 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4285 // Bail out if that is so.
4286 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4287 if (not_objArray != nullptr) {
4288 // Improve the klass node's type from the new optimistic assumption:
4289 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4290 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4291 Node* cast = new CastPPNode(klass_node, akls);
4292 cast->init_req(0, control());
4293 klass_node = _gvn.transform(cast);
4294 }
4295
4296 // Bail out if either start or end is negative.
4297 generate_negative_guard(start, bailout, &start);
4298 generate_negative_guard(end, bailout, &end);
4299
4300 Node* length = end;
4301 if (_gvn.type(start) != TypeInt::ZERO) {
4302 length = _gvn.transform(new SubINode(end, start));
4303 }
4304
4305 // Bail out if length is negative.
4306 // Without this the new_array would throw
4307 // NegativeArraySizeException but IllegalArgumentException is what
4308 // should be thrown
4309 generate_negative_guard(length, bailout, &length);
4310
4311 if (bailout->req() > 1) {
4312 PreserveJVMState pjvms(this);
4313 set_control(_gvn.transform(bailout));
4314 uncommon_trap(Deoptimization::Reason_intrinsic,
4315 Deoptimization::Action_maybe_recompile);
4316 }
4317
4318 if (!stopped()) {
4319 // How many elements will we copy from the original?
4320 // The answer is MinI(orig_length - start, length).
4321 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4322 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4323
4324 // Generate a direct call to the right arraycopy function(s).
4325 // We know the copy is disjoint but we might not know if the
4326 // oop stores need checking.
4327 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4328 // This will fail a store-check if x contains any non-nulls.
4329
4330 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4331 // loads/stores but it is legal only if we're sure the
4332 // Arrays.copyOf would succeed. So we need all input arguments
4333 // to the copyOf to be validated, including that the copy to the
4334 // new array won't trigger an ArrayStoreException. That subtype
4335 // check can be optimized if we know something on the type of
4336 // the input array from type speculation.
4337 if (_gvn.type(klass_node)->singleton()) {
4338 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4339 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4340
4341 int test = C->static_subtype_check(superk, subk);
4342 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4343 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4344 if (t_original->speculative_type() != nullptr) {
4345 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4346 }
4347 }
4348 }
4349
4350 bool validated = false;
4351 // Reason_class_check rather than Reason_intrinsic because we
4352 // want to intrinsify even if this traps.
4353 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4354 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4355
4356 if (not_subtype_ctrl != top()) {
4357 PreserveJVMState pjvms(this);
4358 set_control(not_subtype_ctrl);
4359 uncommon_trap(Deoptimization::Reason_class_check,
4360 Deoptimization::Action_make_not_entrant);
4361 assert(stopped(), "Should be stopped");
4362 }
4363 validated = true;
4364 }
4365
4366 if (!stopped()) {
4367 newcopy = new_array(klass_node, length, 0); // no arguments to push
4368
4369 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
4370 load_object_klass(original), klass_node);
4371 if (!is_copyOfRange) {
4372 ac->set_copyof(validated);
4373 } else {
4374 ac->set_copyofrange(validated);
4375 }
4376 Node* n = _gvn.transform(ac);
4377 if (n == ac) {
4378 ac->connect_outputs(this);
4379 } else {
4380 assert(validated, "shouldn't transform if all arguments not validated");
4381 set_all_memory(n);
4382 }
4383 }
4384 }
4385 } // original reexecute is set back here
4386
4387 C->set_has_split_ifs(true); // Has chance for split-if optimization
4388 if (!stopped()) {
4389 set_result(newcopy);
4390 }
4391 return true;
4392 }
4393
4394
4395 //----------------------generate_virtual_guard---------------------------
4396 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4397 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4398 RegionNode* slow_region) {
4399 ciMethod* method = callee();
4400 int vtable_index = method->vtable_index();
4401 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4402 "bad index %d", vtable_index);
4403 // Get the Method* out of the appropriate vtable entry.
4404 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4405 vtable_index*vtableEntry::size_in_bytes() +
4406 in_bytes(vtableEntry::method_offset());
4407 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4408 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4409
4410 // Compare the target method with the expected method (e.g., Object.hashCode).
4411 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4412
4413 Node* native_call = makecon(native_call_addr);
4414 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4415 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4416
4417 return generate_slow_guard(test_native, slow_region);
4418 }
4419
4420 //-----------------------generate_method_call----------------------------
4421 // Use generate_method_call to make a slow-call to the real
4422 // method if the fast path fails. An alternative would be to
4423 // use a stub like OptoRuntime::slow_arraycopy_Java.
4424 // This only works for expanding the current library call,
4425 // not another intrinsic. (E.g., don't use this for making an
4426 // arraycopy call inside of the copyOf intrinsic.)
4427 CallJavaNode*
4428 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4429 // When compiling the intrinsic method itself, do not use this technique.
4430 guarantee(callee() != C->method(), "cannot make slow-call to self");
4431
4432 ciMethod* method = callee();
4433 // ensure the JVMS we have will be correct for this call
4434 guarantee(method_id == method->intrinsic_id(), "must match");
4435
4436 const TypeFunc* tf = TypeFunc::make(method);
4437 if (res_not_null) {
4438 assert(tf->return_type() == T_OBJECT, "");
4439 const TypeTuple* range = tf->range();
4440 const Type** fields = TypeTuple::fields(range->cnt());
4441 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4442 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4443 tf = TypeFunc::make(tf->domain(), new_range);
4444 }
4445 CallJavaNode* slow_call;
4446 if (is_static) {
4447 assert(!is_virtual, "");
4448 slow_call = new CallStaticJavaNode(C, tf,
4449 SharedRuntime::get_resolve_static_call_stub(), method);
4450 } else if (is_virtual) {
4451 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4452 int vtable_index = Method::invalid_vtable_index;
4453 if (UseInlineCaches) {
4454 // Suppress the vtable call
4455 } else {
4456 // hashCode and clone are not a miranda methods,
4457 // so the vtable index is fixed.
4458 // No need to use the linkResolver to get it.
4459 vtable_index = method->vtable_index();
4460 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4461 "bad index %d", vtable_index);
4462 }
4463 slow_call = new CallDynamicJavaNode(tf,
4464 SharedRuntime::get_resolve_virtual_call_stub(),
4465 method, vtable_index);
4466 } else { // neither virtual nor static: opt_virtual
4467 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4468 slow_call = new CallStaticJavaNode(C, tf,
4469 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4470 slow_call->set_optimized_virtual(true);
4471 }
4472 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4473 // To be able to issue a direct call (optimized virtual or virtual)
4474 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4475 // about the method being invoked should be attached to the call site to
4476 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4477 slow_call->set_override_symbolic_info(true);
4478 }
4479 set_arguments_for_java_call(slow_call);
4480 set_edges_for_java_call(slow_call);
4481 return slow_call;
4482 }
4483
4484
4485 /**
4486 * Build special case code for calls to hashCode on an object. This call may
4487 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4488 * slightly different code.
4489 */
4490 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4491 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4492 assert(!(is_virtual && is_static), "either virtual, special, or static");
4493
4494 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4495
4496 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4497 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4498 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4499 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4500 Node* obj = nullptr;
4501 if (!is_static) {
4502 // Check for hashing null object
4503 obj = null_check_receiver();
4504 if (stopped()) return true; // unconditionally null
4505 result_reg->init_req(_null_path, top());
4506 result_val->init_req(_null_path, top());
4507 } else {
4508 // Do a null check, and return zero if null.
4509 // System.identityHashCode(null) == 0
4510 obj = argument(0);
4511 Node* null_ctl = top();
4512 obj = null_check_oop(obj, &null_ctl);
4513 result_reg->init_req(_null_path, null_ctl);
4514 result_val->init_req(_null_path, _gvn.intcon(0));
4515 }
4516
4517 // Unconditionally null? Then return right away.
4518 if (stopped()) {
4519 set_control( result_reg->in(_null_path));
4520 if (!stopped())
4521 set_result(result_val->in(_null_path));
4522 return true;
4523 }
4524
4525 // We only go to the fast case code if we pass a number of guards. The
4526 // paths which do not pass are accumulated in the slow_region.
4527 RegionNode* slow_region = new RegionNode(1);
4528 record_for_igvn(slow_region);
4529
4530 // If this is a virtual call, we generate a funny guard. We pull out
4531 // the vtable entry corresponding to hashCode() from the target object.
4532 // If the target method which we are calling happens to be the native
4533 // Object hashCode() method, we pass the guard. We do not need this
4534 // guard for non-virtual calls -- the caller is known to be the native
4535 // Object hashCode().
4536 if (is_virtual) {
4537 // After null check, get the object's klass.
4538 Node* obj_klass = load_object_klass(obj);
4539 generate_virtual_guard(obj_klass, slow_region);
4540 }
4541
4542 // Get the header out of the object, use LoadMarkNode when available
4543 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4544 // The control of the load must be null. Otherwise, the load can move before
4545 // the null check after castPP removal.
4546 Node* no_ctrl = nullptr;
4547 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4548
4549 // Test the header to see if it is unlocked.
4550 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4551 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4552 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4553 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4554 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4555
4556 generate_slow_guard(test_unlocked, slow_region);
4557
4558 // Get the hash value and check to see that it has been properly assigned.
4559 // We depend on hash_mask being at most 32 bits and avoid the use of
4560 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4561 // vm: see markWord.hpp.
4562 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4563 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4564 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4565 // This hack lets the hash bits live anywhere in the mark object now, as long
4566 // as the shift drops the relevant bits into the low 32 bits. Note that
4567 // Java spec says that HashCode is an int so there's no point in capturing
4568 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4569 hshifted_header = ConvX2I(hshifted_header);
4570 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4571
4572 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4573 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4574 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4575
4576 generate_slow_guard(test_assigned, slow_region);
4577
4578 Node* init_mem = reset_memory();
4579 // fill in the rest of the null path:
4580 result_io ->init_req(_null_path, i_o());
4581 result_mem->init_req(_null_path, init_mem);
4582
4583 result_val->init_req(_fast_path, hash_val);
4584 result_reg->init_req(_fast_path, control());
4585 result_io ->init_req(_fast_path, i_o());
4586 result_mem->init_req(_fast_path, init_mem);
4587
4588 // Generate code for the slow case. We make a call to hashCode().
4589 set_control(_gvn.transform(slow_region));
4590 if (!stopped()) {
4591 // No need for PreserveJVMState, because we're using up the present state.
4592 set_all_memory(init_mem);
4593 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4594 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4595 Node* slow_result = set_results_for_java_call(slow_call);
4596 // this->control() comes from set_results_for_java_call
4597 result_reg->init_req(_slow_path, control());
4598 result_val->init_req(_slow_path, slow_result);
4599 result_io ->set_req(_slow_path, i_o());
4600 result_mem ->set_req(_slow_path, reset_memory());
4601 }
4602
4603 // Return the combined state.
4604 set_i_o( _gvn.transform(result_io) );
4605 set_all_memory( _gvn.transform(result_mem));
4606
4607 set_result(result_reg, result_val);
4608 return true;
4609 }
4610
4611 //---------------------------inline_native_getClass----------------------------
4612 // public final native Class<?> java.lang.Object.getClass();
4613 //
4614 // Build special case code for calls to getClass on an object.
4615 bool LibraryCallKit::inline_native_getClass() {
4616 Node* obj = null_check_receiver();
4617 if (stopped()) return true;
4618 set_result(load_mirror_from_klass(load_object_klass(obj)));
4619 return true;
4620 }
4621
4622 //-----------------inline_native_Reflection_getCallerClass---------------------
4623 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4624 //
4625 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4626 //
4627 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4628 // in that it must skip particular security frames and checks for
4629 // caller sensitive methods.
4630 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4631 #ifndef PRODUCT
4632 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4633 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4634 }
4635 #endif
4636
4637 if (!jvms()->has_method()) {
4638 #ifndef PRODUCT
4639 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4640 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4641 }
4642 #endif
4643 return false;
4644 }
4645
4646 // Walk back up the JVM state to find the caller at the required
4647 // depth.
4648 JVMState* caller_jvms = jvms();
4649
4650 // Cf. JVM_GetCallerClass
4651 // NOTE: Start the loop at depth 1 because the current JVM state does
4652 // not include the Reflection.getCallerClass() frame.
4653 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4654 ciMethod* m = caller_jvms->method();
4655 switch (n) {
4656 case 0:
4657 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4658 break;
4659 case 1:
4660 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4661 if (!m->caller_sensitive()) {
4662 #ifndef PRODUCT
4663 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4664 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4665 }
4666 #endif
4667 return false; // bail-out; let JVM_GetCallerClass do the work
4668 }
4669 break;
4670 default:
4671 if (!m->is_ignored_by_security_stack_walk()) {
4672 // We have reached the desired frame; return the holder class.
4673 // Acquire method holder as java.lang.Class and push as constant.
4674 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4675 ciInstance* caller_mirror = caller_klass->java_mirror();
4676 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4677
4678 #ifndef PRODUCT
4679 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4680 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());
4681 tty->print_cr(" JVM state at this point:");
4682 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4683 ciMethod* m = jvms()->of_depth(i)->method();
4684 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4685 }
4686 }
4687 #endif
4688 return true;
4689 }
4690 break;
4691 }
4692 }
4693
4694 #ifndef PRODUCT
4695 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4696 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4697 tty->print_cr(" JVM state at this point:");
4698 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4699 ciMethod* m = jvms()->of_depth(i)->method();
4700 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4701 }
4702 }
4703 #endif
4704
4705 return false; // bail-out; let JVM_GetCallerClass do the work
4706 }
4707
4708 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4709 Node* arg = argument(0);
4710 Node* result = nullptr;
4711
4712 switch (id) {
4713 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4714 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4715 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4716 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4717 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4718 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4719
4720 case vmIntrinsics::_doubleToLongBits: {
4721 // two paths (plus control) merge in a wood
4722 RegionNode *r = new RegionNode(3);
4723 Node *phi = new PhiNode(r, TypeLong::LONG);
4724
4725 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4726 // Build the boolean node
4727 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4728
4729 // Branch either way.
4730 // NaN case is less traveled, which makes all the difference.
4731 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4732 Node *opt_isnan = _gvn.transform(ifisnan);
4733 assert( opt_isnan->is_If(), "Expect an IfNode");
4734 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4735 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4736
4737 set_control(iftrue);
4738
4739 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4740 Node *slow_result = longcon(nan_bits); // return NaN
4741 phi->init_req(1, _gvn.transform( slow_result ));
4742 r->init_req(1, iftrue);
4743
4744 // Else fall through
4745 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4746 set_control(iffalse);
4747
4748 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4749 r->init_req(2, iffalse);
4750
4751 // Post merge
4752 set_control(_gvn.transform(r));
4753 record_for_igvn(r);
4754
4755 C->set_has_split_ifs(true); // Has chance for split-if optimization
4756 result = phi;
4757 assert(result->bottom_type()->isa_long(), "must be");
4758 break;
4759 }
4760
4761 case vmIntrinsics::_floatToIntBits: {
4762 // two paths (plus control) merge in a wood
4763 RegionNode *r = new RegionNode(3);
4764 Node *phi = new PhiNode(r, TypeInt::INT);
4765
4766 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4767 // Build the boolean node
4768 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4769
4770 // Branch either way.
4771 // NaN case is less traveled, which makes all the difference.
4772 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4773 Node *opt_isnan = _gvn.transform(ifisnan);
4774 assert( opt_isnan->is_If(), "Expect an IfNode");
4775 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4776 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4777
4778 set_control(iftrue);
4779
4780 static const jint nan_bits = 0x7fc00000;
4781 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4782 phi->init_req(1, _gvn.transform( slow_result ));
4783 r->init_req(1, iftrue);
4784
4785 // Else fall through
4786 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4787 set_control(iffalse);
4788
4789 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4790 r->init_req(2, iffalse);
4791
4792 // Post merge
4793 set_control(_gvn.transform(r));
4794 record_for_igvn(r);
4795
4796 C->set_has_split_ifs(true); // Has chance for split-if optimization
4797 result = phi;
4798 assert(result->bottom_type()->isa_int(), "must be");
4799 break;
4800 }
4801
4802 default:
4803 fatal_unexpected_iid(id);
4804 break;
4805 }
4806 set_result(_gvn.transform(result));
4807 return true;
4808 }
4809
4810 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
4811 Node* arg = argument(0);
4812 Node* result = nullptr;
4813
4814 switch (id) {
4815 case vmIntrinsics::_floatIsInfinite:
4816 result = new IsInfiniteFNode(arg);
4817 break;
4818 case vmIntrinsics::_floatIsFinite:
4819 result = new IsFiniteFNode(arg);
4820 break;
4821 case vmIntrinsics::_doubleIsInfinite:
4822 result = new IsInfiniteDNode(arg);
4823 break;
4824 case vmIntrinsics::_doubleIsFinite:
4825 result = new IsFiniteDNode(arg);
4826 break;
4827 default:
4828 fatal_unexpected_iid(id);
4829 break;
4830 }
4831 set_result(_gvn.transform(result));
4832 return true;
4833 }
4834
4835 //----------------------inline_unsafe_copyMemory-------------------------
4836 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4837
4838 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
4839 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
4840 const Type* base_t = gvn.type(base);
4841
4842 bool in_native = (base_t == TypePtr::NULL_PTR);
4843 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
4844 bool is_mixed = !in_heap && !in_native;
4845
4846 if (is_mixed) {
4847 return true; // mixed accesses can touch both on-heap and off-heap memory
4848 }
4849 if (in_heap) {
4850 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
4851 if (!is_prim_array) {
4852 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
4853 // there's not enough type information available to determine proper memory slice for it.
4854 return true;
4855 }
4856 }
4857 return false;
4858 }
4859
4860 bool LibraryCallKit::inline_unsafe_copyMemory() {
4861 if (callee()->is_static()) return false; // caller must have the capability!
4862 null_check_receiver(); // null-check receiver
4863 if (stopped()) return true;
4864
4865 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4866
4867 Node* src_base = argument(1); // type: oop
4868 Node* src_off = ConvL2X(argument(2)); // type: long
4869 Node* dst_base = argument(4); // type: oop
4870 Node* dst_off = ConvL2X(argument(5)); // type: long
4871 Node* size = ConvL2X(argument(7)); // type: long
4872
4873 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4874 "fieldOffset must be byte-scaled");
4875
4876 Node* src_addr = make_unsafe_address(src_base, src_off);
4877 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
4878
4879 Node* thread = _gvn.transform(new ThreadLocalNode());
4880 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4881 BasicType doing_unsafe_access_bt = T_BYTE;
4882 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4883
4884 // update volatile field
4885 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4886
4887 int flags = RC_LEAF | RC_NO_FP;
4888
4889 const TypePtr* dst_type = TypePtr::BOTTOM;
4890
4891 // Adjust memory effects of the runtime call based on input values.
4892 if (!has_wide_mem(_gvn, src_addr, src_base) &&
4893 !has_wide_mem(_gvn, dst_addr, dst_base)) {
4894 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
4895
4896 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
4897 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
4898 flags |= RC_NARROW_MEM; // narrow in memory
4899 }
4900 }
4901
4902 // Call it. Note that the length argument is not scaled.
4903 make_runtime_call(flags,
4904 OptoRuntime::fast_arraycopy_Type(),
4905 StubRoutines::unsafe_arraycopy(),
4906 "unsafe_arraycopy",
4907 dst_type,
4908 src_addr, dst_addr, size XTOP);
4909
4910 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4911
4912 return true;
4913 }
4914
4915 #undef XTOP
4916
4917 //------------------------clone_coping-----------------------------------
4918 // Helper function for inline_native_clone.
4919 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4920 assert(obj_size != nullptr, "");
4921 Node* raw_obj = alloc_obj->in(1);
4922 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4923
4924 AllocateNode* alloc = nullptr;
4925 if (ReduceBulkZeroing) {
4926 // We will be completely responsible for initializing this object -
4927 // mark Initialize node as complete.
4928 alloc = AllocateNode::Ideal_allocation(alloc_obj);
4929 // The object was just allocated - there should be no any stores!
4930 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
4931 // Mark as complete_with_arraycopy so that on AllocateNode
4932 // expansion, we know this AllocateNode is initialized by an array
4933 // copy and a StoreStore barrier exists after the array copy.
4934 alloc->initialization()->set_complete_with_arraycopy();
4935 }
4936
4937 Node* size = _gvn.transform(obj_size);
4938 access_clone(obj, alloc_obj, size, is_array);
4939
4940 // Do not let reads from the cloned object float above the arraycopy.
4941 if (alloc != nullptr) {
4942 // Do not let stores that initialize this object be reordered with
4943 // a subsequent store that would make this object accessible by
4944 // other threads.
4945 // Record what AllocateNode this StoreStore protects so that
4946 // escape analysis can go from the MemBarStoreStoreNode to the
4947 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4948 // based on the escape status of the AllocateNode.
4949 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4950 } else {
4951 insert_mem_bar(Op_MemBarCPUOrder);
4952 }
4953 }
4954
4955 //------------------------inline_native_clone----------------------------
4956 // protected native Object java.lang.Object.clone();
4957 //
4958 // Here are the simple edge cases:
4959 // null receiver => normal trap
4960 // virtual and clone was overridden => slow path to out-of-line clone
4961 // not cloneable or finalizer => slow path to out-of-line Object.clone
4962 //
4963 // The general case has two steps, allocation and copying.
4964 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4965 //
4966 // Copying also has two cases, oop arrays and everything else.
4967 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4968 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4969 //
4970 // These steps fold up nicely if and when the cloned object's klass
4971 // can be sharply typed as an object array, a type array, or an instance.
4972 //
4973 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4974 PhiNode* result_val;
4975
4976 // Set the reexecute bit for the interpreter to reexecute
4977 // the bytecode that invokes Object.clone if deoptimization happens.
4978 { PreserveReexecuteState preexecs(this);
4979 jvms()->set_should_reexecute(true);
4980
4981 Node* obj = null_check_receiver();
4982 if (stopped()) return true;
4983
4984 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4985
4986 // If we are going to clone an instance, we need its exact type to
4987 // know the number and types of fields to convert the clone to
4988 // loads/stores. Maybe a speculative type can help us.
4989 if (!obj_type->klass_is_exact() &&
4990 obj_type->speculative_type() != nullptr &&
4991 obj_type->speculative_type()->is_instance_klass()) {
4992 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4993 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4994 !spec_ik->has_injected_fields()) {
4995 if (!obj_type->isa_instptr() ||
4996 obj_type->is_instptr()->instance_klass()->has_subklass()) {
4997 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4998 }
4999 }
5000 }
5001
5002 // Conservatively insert a memory barrier on all memory slices.
5003 // Do not let writes into the original float below the clone.
5004 insert_mem_bar(Op_MemBarCPUOrder);
5005
5006 // paths into result_reg:
5007 enum {
5008 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5009 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5010 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5011 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5012 PATH_LIMIT
5013 };
5014 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5015 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5016 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5017 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5018 record_for_igvn(result_reg);
5019
5020 Node* obj_klass = load_object_klass(obj);
5021 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr);
5022 if (array_ctl != nullptr) {
5023 // It's an array.
5024 PreserveJVMState pjvms(this);
5025 set_control(array_ctl);
5026 Node* obj_length = load_array_length(obj);
5027 Node* array_size = nullptr; // Size of the array without object alignment padding.
5028 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5029
5030 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5031 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5032 // If it is an oop array, it requires very special treatment,
5033 // because gc barriers are required when accessing the array.
5034 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5035 if (is_obja != nullptr) {
5036 PreserveJVMState pjvms2(this);
5037 set_control(is_obja);
5038 // Generate a direct call to the right arraycopy function(s).
5039 // Clones are always tightly coupled.
5040 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5041 ac->set_clone_oop_array();
5042 Node* n = _gvn.transform(ac);
5043 assert(n == ac, "cannot disappear");
5044 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5045
5046 result_reg->init_req(_objArray_path, control());
5047 result_val->init_req(_objArray_path, alloc_obj);
5048 result_i_o ->set_req(_objArray_path, i_o());
5049 result_mem ->set_req(_objArray_path, reset_memory());
5050 }
5051 }
5052 // Otherwise, there are no barriers to worry about.
5053 // (We can dispense with card marks if we know the allocation
5054 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5055 // causes the non-eden paths to take compensating steps to
5056 // simulate a fresh allocation, so that no further
5057 // card marks are required in compiled code to initialize
5058 // the object.)
5059
5060 if (!stopped()) {
5061 copy_to_clone(obj, alloc_obj, array_size, true);
5062
5063 // Present the results of the copy.
5064 result_reg->init_req(_array_path, control());
5065 result_val->init_req(_array_path, alloc_obj);
5066 result_i_o ->set_req(_array_path, i_o());
5067 result_mem ->set_req(_array_path, reset_memory());
5068 }
5069 }
5070
5071 // We only go to the instance fast case code if we pass a number of guards.
5072 // The paths which do not pass are accumulated in the slow_region.
5073 RegionNode* slow_region = new RegionNode(1);
5074 record_for_igvn(slow_region);
5075 if (!stopped()) {
5076 // It's an instance (we did array above). Make the slow-path tests.
5077 // If this is a virtual call, we generate a funny guard. We grab
5078 // the vtable entry corresponding to clone() from the target object.
5079 // If the target method which we are calling happens to be the
5080 // Object clone() method, we pass the guard. We do not need this
5081 // guard for non-virtual calls; the caller is known to be the native
5082 // Object clone().
5083 if (is_virtual) {
5084 generate_virtual_guard(obj_klass, slow_region);
5085 }
5086
5087 // The object must be easily cloneable and must not have a finalizer.
5088 // Both of these conditions may be checked in a single test.
5089 // We could optimize the test further, but we don't care.
5090 generate_access_flags_guard(obj_klass,
5091 // Test both conditions:
5092 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
5093 // Must be cloneable but not finalizer:
5094 JVM_ACC_IS_CLONEABLE_FAST,
5095 slow_region);
5096 }
5097
5098 if (!stopped()) {
5099 // It's an instance, and it passed the slow-path tests.
5100 PreserveJVMState pjvms(this);
5101 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5102 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5103 // is reexecuted if deoptimization occurs and there could be problems when merging
5104 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5105 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5106
5107 copy_to_clone(obj, alloc_obj, obj_size, false);
5108
5109 // Present the results of the slow call.
5110 result_reg->init_req(_instance_path, control());
5111 result_val->init_req(_instance_path, alloc_obj);
5112 result_i_o ->set_req(_instance_path, i_o());
5113 result_mem ->set_req(_instance_path, reset_memory());
5114 }
5115
5116 // Generate code for the slow case. We make a call to clone().
5117 set_control(_gvn.transform(slow_region));
5118 if (!stopped()) {
5119 PreserveJVMState pjvms(this);
5120 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5121 // We need to deoptimize on exception (see comment above)
5122 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5123 // this->control() comes from set_results_for_java_call
5124 result_reg->init_req(_slow_path, control());
5125 result_val->init_req(_slow_path, slow_result);
5126 result_i_o ->set_req(_slow_path, i_o());
5127 result_mem ->set_req(_slow_path, reset_memory());
5128 }
5129
5130 // Return the combined state.
5131 set_control( _gvn.transform(result_reg));
5132 set_i_o( _gvn.transform(result_i_o));
5133 set_all_memory( _gvn.transform(result_mem));
5134 } // original reexecute is set back here
5135
5136 set_result(_gvn.transform(result_val));
5137 return true;
5138 }
5139
5140 // If we have a tightly coupled allocation, the arraycopy may take care
5141 // of the array initialization. If one of the guards we insert between
5142 // the allocation and the arraycopy causes a deoptimization, an
5143 // uninitialized array will escape the compiled method. To prevent that
5144 // we set the JVM state for uncommon traps between the allocation and
5145 // the arraycopy to the state before the allocation so, in case of
5146 // deoptimization, we'll reexecute the allocation and the
5147 // initialization.
5148 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5149 if (alloc != nullptr) {
5150 ciMethod* trap_method = alloc->jvms()->method();
5151 int trap_bci = alloc->jvms()->bci();
5152
5153 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5154 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5155 // Make sure there's no store between the allocation and the
5156 // arraycopy otherwise visible side effects could be rexecuted
5157 // in case of deoptimization and cause incorrect execution.
5158 bool no_interfering_store = true;
5159 Node* mem = alloc->in(TypeFunc::Memory);
5160 if (mem->is_MergeMem()) {
5161 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5162 Node* n = mms.memory();
5163 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5164 assert(n->is_Store(), "what else?");
5165 no_interfering_store = false;
5166 break;
5167 }
5168 }
5169 } else {
5170 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5171 Node* n = mms.memory();
5172 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5173 assert(n->is_Store(), "what else?");
5174 no_interfering_store = false;
5175 break;
5176 }
5177 }
5178 }
5179
5180 if (no_interfering_store) {
5181 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5182
5183 JVMState* saved_jvms = jvms();
5184 saved_reexecute_sp = _reexecute_sp;
5185
5186 set_jvms(sfpt->jvms());
5187 _reexecute_sp = jvms()->sp();
5188
5189 return saved_jvms;
5190 }
5191 }
5192 }
5193 return nullptr;
5194 }
5195
5196 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5197 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5198 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5199 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5200 uint size = alloc->req();
5201 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5202 old_jvms->set_map(sfpt);
5203 for (uint i = 0; i < size; i++) {
5204 sfpt->init_req(i, alloc->in(i));
5205 }
5206 // re-push array length for deoptimization
5207 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5208 old_jvms->set_sp(old_jvms->sp()+1);
5209 old_jvms->set_monoff(old_jvms->monoff()+1);
5210 old_jvms->set_scloff(old_jvms->scloff()+1);
5211 old_jvms->set_endoff(old_jvms->endoff()+1);
5212 old_jvms->set_should_reexecute(true);
5213
5214 sfpt->set_i_o(map()->i_o());
5215 sfpt->set_memory(map()->memory());
5216 sfpt->set_control(map()->control());
5217 return sfpt;
5218 }
5219
5220 // In case of a deoptimization, we restart execution at the
5221 // allocation, allocating a new array. We would leave an uninitialized
5222 // array in the heap that GCs wouldn't expect. Move the allocation
5223 // after the traps so we don't allocate the array if we
5224 // deoptimize. This is possible because tightly_coupled_allocation()
5225 // guarantees there's no observer of the allocated array at this point
5226 // and the control flow is simple enough.
5227 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5228 int saved_reexecute_sp, uint new_idx) {
5229 if (saved_jvms_before_guards != nullptr && !stopped()) {
5230 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5231
5232 assert(alloc != nullptr, "only with a tightly coupled allocation");
5233 // restore JVM state to the state at the arraycopy
5234 saved_jvms_before_guards->map()->set_control(map()->control());
5235 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5236 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5237 // If we've improved the types of some nodes (null check) while
5238 // emitting the guards, propagate them to the current state
5239 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5240 set_jvms(saved_jvms_before_guards);
5241 _reexecute_sp = saved_reexecute_sp;
5242
5243 // Remove the allocation from above the guards
5244 CallProjections callprojs;
5245 alloc->extract_projections(&callprojs, true);
5246 InitializeNode* init = alloc->initialization();
5247 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5248 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5249 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5250
5251 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5252 // the allocation (i.e. is only valid if the allocation succeeds):
5253 // 1) replace CastIINode with AllocateArrayNode's length here
5254 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5255 //
5256 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5257 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5258 Node* init_control = init->proj_out(TypeFunc::Control);
5259 Node* alloc_length = alloc->Ideal_length();
5260 #ifdef ASSERT
5261 Node* prev_cast = nullptr;
5262 #endif
5263 for (uint i = 0; i < init_control->outcnt(); i++) {
5264 Node* init_out = init_control->raw_out(i);
5265 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5266 #ifdef ASSERT
5267 if (prev_cast == nullptr) {
5268 prev_cast = init_out;
5269 } else {
5270 if (prev_cast->cmp(*init_out) == false) {
5271 prev_cast->dump();
5272 init_out->dump();
5273 assert(false, "not equal CastIINode");
5274 }
5275 }
5276 #endif
5277 C->gvn_replace_by(init_out, alloc_length);
5278 }
5279 }
5280 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5281
5282 // move the allocation here (after the guards)
5283 _gvn.hash_delete(alloc);
5284 alloc->set_req(TypeFunc::Control, control());
5285 alloc->set_req(TypeFunc::I_O, i_o());
5286 Node *mem = reset_memory();
5287 set_all_memory(mem);
5288 alloc->set_req(TypeFunc::Memory, mem);
5289 set_control(init->proj_out_or_null(TypeFunc::Control));
5290 set_i_o(callprojs.fallthrough_ioproj);
5291
5292 // Update memory as done in GraphKit::set_output_for_allocation()
5293 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5294 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5295 if (ary_type->isa_aryptr() && length_type != nullptr) {
5296 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5297 }
5298 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5299 int elemidx = C->get_alias_index(telemref);
5300 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5301 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5302
5303 Node* allocx = _gvn.transform(alloc);
5304 assert(allocx == alloc, "where has the allocation gone?");
5305 assert(dest->is_CheckCastPP(), "not an allocation result?");
5306
5307 _gvn.hash_delete(dest);
5308 dest->set_req(0, control());
5309 Node* destx = _gvn.transform(dest);
5310 assert(destx == dest, "where has the allocation result gone?");
5311
5312 array_ideal_length(alloc, ary_type, true);
5313 }
5314 }
5315
5316 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5317 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5318 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5319 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5320 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5321 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5322 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5323 JVMState* saved_jvms_before_guards) {
5324 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5325 // There is at least one unrelated uncommon trap which needs to be replaced.
5326 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5327
5328 JVMState* saved_jvms = jvms();
5329 const int saved_reexecute_sp = _reexecute_sp;
5330 set_jvms(sfpt->jvms());
5331 _reexecute_sp = jvms()->sp();
5332
5333 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5334
5335 // Restore state
5336 set_jvms(saved_jvms);
5337 _reexecute_sp = saved_reexecute_sp;
5338 }
5339 }
5340
5341 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5342 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5343 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5344 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5345 while (if_proj->is_IfProj()) {
5346 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5347 if (uncommon_trap != nullptr) {
5348 create_new_uncommon_trap(uncommon_trap);
5349 }
5350 assert(if_proj->in(0)->is_If(), "must be If");
5351 if_proj = if_proj->in(0)->in(0);
5352 }
5353 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5354 "must have reached control projection of init node");
5355 }
5356
5357 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5358 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5359 assert(trap_request != 0, "no valid UCT trap request");
5360 PreserveJVMState pjvms(this);
5361 set_control(uncommon_trap_call->in(0));
5362 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5363 Deoptimization::trap_request_action(trap_request));
5364 assert(stopped(), "Should be stopped");
5365 _gvn.hash_delete(uncommon_trap_call);
5366 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5367 }
5368
5369 //------------------------------inline_array_partition-----------------------
5370 bool LibraryCallKit::inline_array_partition() {
5371
5372 Node* elementType = null_check(argument(0));
5373 Node* obj = argument(1);
5374 Node* offset = argument(2);
5375 Node* fromIndex = argument(4);
5376 Node* toIndex = argument(5);
5377 Node* indexPivot1 = argument(6);
5378 Node* indexPivot2 = argument(7);
5379
5380 Node* pivotIndices = nullptr;
5381
5382 // Set the original stack and the reexecute bit for the interpreter to reexecute
5383 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5384 { PreserveReexecuteState preexecs(this);
5385 jvms()->set_should_reexecute(true);
5386
5387 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5388 ciType* elem_type = elem_klass->const_oop()->as_instance()->java_mirror_type();
5389 BasicType bt = elem_type->basic_type();
5390 address stubAddr = nullptr;
5391 stubAddr = StubRoutines::select_array_partition_function();
5392 // stub not loaded
5393 if (stubAddr == nullptr) {
5394 return false;
5395 }
5396 // get the address of the array
5397 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5398 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM ) {
5399 return false; // failed input validation
5400 }
5401 Node* obj_adr = make_unsafe_address(obj, offset);
5402
5403 // create the pivotIndices array of type int and size = 2
5404 Node* size = intcon(2);
5405 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5406 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5407 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5408 guarantee(alloc != nullptr, "created above");
5409 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5410
5411 // pass the basic type enum to the stub
5412 Node* elemType = intcon(bt);
5413
5414 // Call the stub
5415 const char *stubName = "array_partition_stub";
5416 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5417 stubAddr, stubName, TypePtr::BOTTOM,
5418 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5419 indexPivot1, indexPivot2);
5420
5421 } // original reexecute is set back here
5422
5423 if (!stopped()) {
5424 set_result(pivotIndices);
5425 }
5426
5427 return true;
5428 }
5429
5430
5431 //------------------------------inline_array_sort-----------------------
5432 bool LibraryCallKit::inline_array_sort() {
5433
5434 Node* elementType = null_check(argument(0));
5435 Node* obj = argument(1);
5436 Node* offset = argument(2);
5437 Node* fromIndex = argument(4);
5438 Node* toIndex = argument(5);
5439
5440 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5441 ciType* elem_type = elem_klass->const_oop()->as_instance()->java_mirror_type();
5442 BasicType bt = elem_type->basic_type();
5443 address stubAddr = nullptr;
5444 stubAddr = StubRoutines::select_arraysort_function();
5445 //stub not loaded
5446 if (stubAddr == nullptr) {
5447 return false;
5448 }
5449
5450 // get address of the array
5451 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5452 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM ) {
5453 return false; // failed input validation
5454 }
5455 Node* obj_adr = make_unsafe_address(obj, offset);
5456
5457 // pass the basic type enum to the stub
5458 Node* elemType = intcon(bt);
5459
5460 // Call the stub.
5461 const char *stubName = "arraysort_stub";
5462 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5463 stubAddr, stubName, TypePtr::BOTTOM,
5464 obj_adr, elemType, fromIndex, toIndex);
5465
5466 return true;
5467 }
5468
5469
5470 //------------------------------inline_arraycopy-----------------------
5471 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5472 // Object dest, int destPos,
5473 // int length);
5474 bool LibraryCallKit::inline_arraycopy() {
5475 // Get the arguments.
5476 Node* src = argument(0); // type: oop
5477 Node* src_offset = argument(1); // type: int
5478 Node* dest = argument(2); // type: oop
5479 Node* dest_offset = argument(3); // type: int
5480 Node* length = argument(4); // type: int
5481
5482 uint new_idx = C->unique();
5483
5484 // Check for allocation before we add nodes that would confuse
5485 // tightly_coupled_allocation()
5486 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5487
5488 int saved_reexecute_sp = -1;
5489 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5490 // See arraycopy_restore_alloc_state() comment
5491 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5492 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5493 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5494 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5495
5496 // The following tests must be performed
5497 // (1) src and dest are arrays.
5498 // (2) src and dest arrays must have elements of the same BasicType
5499 // (3) src and dest must not be null.
5500 // (4) src_offset must not be negative.
5501 // (5) dest_offset must not be negative.
5502 // (6) length must not be negative.
5503 // (7) src_offset + length must not exceed length of src.
5504 // (8) dest_offset + length must not exceed length of dest.
5505 // (9) each element of an oop array must be assignable
5506
5507 // (3) src and dest must not be null.
5508 // always do this here because we need the JVM state for uncommon traps
5509 Node* null_ctl = top();
5510 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5511 assert(null_ctl->is_top(), "no null control here");
5512 dest = null_check(dest, T_ARRAY);
5513
5514 if (!can_emit_guards) {
5515 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5516 // guards but the arraycopy node could still take advantage of a
5517 // tightly allocated allocation. tightly_coupled_allocation() is
5518 // called again to make sure it takes the null check above into
5519 // account: the null check is mandatory and if it caused an
5520 // uncommon trap to be emitted then the allocation can't be
5521 // considered tightly coupled in this context.
5522 alloc = tightly_coupled_allocation(dest);
5523 }
5524
5525 bool validated = false;
5526
5527 const Type* src_type = _gvn.type(src);
5528 const Type* dest_type = _gvn.type(dest);
5529 const TypeAryPtr* top_src = src_type->isa_aryptr();
5530 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5531
5532 // Do we have the type of src?
5533 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5534 // Do we have the type of dest?
5535 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5536 // Is the type for src from speculation?
5537 bool src_spec = false;
5538 // Is the type for dest from speculation?
5539 bool dest_spec = false;
5540
5541 if ((!has_src || !has_dest) && can_emit_guards) {
5542 // We don't have sufficient type information, let's see if
5543 // speculative types can help. We need to have types for both src
5544 // and dest so that it pays off.
5545
5546 // Do we already have or could we have type information for src
5547 bool could_have_src = has_src;
5548 // Do we already have or could we have type information for dest
5549 bool could_have_dest = has_dest;
5550
5551 ciKlass* src_k = nullptr;
5552 if (!has_src) {
5553 src_k = src_type->speculative_type_not_null();
5554 if (src_k != nullptr && src_k->is_array_klass()) {
5555 could_have_src = true;
5556 }
5557 }
5558
5559 ciKlass* dest_k = nullptr;
5560 if (!has_dest) {
5561 dest_k = dest_type->speculative_type_not_null();
5562 if (dest_k != nullptr && dest_k->is_array_klass()) {
5563 could_have_dest = true;
5564 }
5565 }
5566
5567 if (could_have_src && could_have_dest) {
5568 // This is going to pay off so emit the required guards
5569 if (!has_src) {
5570 src = maybe_cast_profiled_obj(src, src_k, true);
5571 src_type = _gvn.type(src);
5572 top_src = src_type->isa_aryptr();
5573 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5574 src_spec = true;
5575 }
5576 if (!has_dest) {
5577 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5578 dest_type = _gvn.type(dest);
5579 top_dest = dest_type->isa_aryptr();
5580 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5581 dest_spec = true;
5582 }
5583 }
5584 }
5585
5586 if (has_src && has_dest && can_emit_guards) {
5587 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5588 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5589 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5590 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5591
5592 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5593 // If both arrays are object arrays then having the exact types
5594 // for both will remove the need for a subtype check at runtime
5595 // before the call and may make it possible to pick a faster copy
5596 // routine (without a subtype check on every element)
5597 // Do we have the exact type of src?
5598 bool could_have_src = src_spec;
5599 // Do we have the exact type of dest?
5600 bool could_have_dest = dest_spec;
5601 ciKlass* src_k = nullptr;
5602 ciKlass* dest_k = nullptr;
5603 if (!src_spec) {
5604 src_k = src_type->speculative_type_not_null();
5605 if (src_k != nullptr && src_k->is_array_klass()) {
5606 could_have_src = true;
5607 }
5608 }
5609 if (!dest_spec) {
5610 dest_k = dest_type->speculative_type_not_null();
5611 if (dest_k != nullptr && dest_k->is_array_klass()) {
5612 could_have_dest = true;
5613 }
5614 }
5615 if (could_have_src && could_have_dest) {
5616 // If we can have both exact types, emit the missing guards
5617 if (could_have_src && !src_spec) {
5618 src = maybe_cast_profiled_obj(src, src_k, true);
5619 }
5620 if (could_have_dest && !dest_spec) {
5621 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5622 }
5623 }
5624 }
5625 }
5626
5627 ciMethod* trap_method = method();
5628 int trap_bci = bci();
5629 if (saved_jvms_before_guards != nullptr) {
5630 trap_method = alloc->jvms()->method();
5631 trap_bci = alloc->jvms()->bci();
5632 }
5633
5634 bool negative_length_guard_generated = false;
5635
5636 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5637 can_emit_guards &&
5638 !src->is_top() && !dest->is_top()) {
5639 // validate arguments: enables transformation the ArrayCopyNode
5640 validated = true;
5641
5642 RegionNode* slow_region = new RegionNode(1);
5643 record_for_igvn(slow_region);
5644
5645 // (1) src and dest are arrays.
5646 generate_non_array_guard(load_object_klass(src), slow_region);
5647 generate_non_array_guard(load_object_klass(dest), slow_region);
5648
5649 // (2) src and dest arrays must have elements of the same BasicType
5650 // done at macro expansion or at Ideal transformation time
5651
5652 // (4) src_offset must not be negative.
5653 generate_negative_guard(src_offset, slow_region);
5654
5655 // (5) dest_offset must not be negative.
5656 generate_negative_guard(dest_offset, slow_region);
5657
5658 // (7) src_offset + length must not exceed length of src.
5659 generate_limit_guard(src_offset, length,
5660 load_array_length(src),
5661 slow_region);
5662
5663 // (8) dest_offset + length must not exceed length of dest.
5664 generate_limit_guard(dest_offset, length,
5665 load_array_length(dest),
5666 slow_region);
5667
5668 // (6) length must not be negative.
5669 // This is also checked in generate_arraycopy() during macro expansion, but
5670 // we also have to check it here for the case where the ArrayCopyNode will
5671 // be eliminated by Escape Analysis.
5672 if (EliminateAllocations) {
5673 generate_negative_guard(length, slow_region);
5674 negative_length_guard_generated = true;
5675 }
5676
5677 // (9) each element of an oop array must be assignable
5678 Node* dest_klass = load_object_klass(dest);
5679 if (src != dest) {
5680 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5681
5682 if (not_subtype_ctrl != top()) {
5683 PreserveJVMState pjvms(this);
5684 set_control(not_subtype_ctrl);
5685 uncommon_trap(Deoptimization::Reason_intrinsic,
5686 Deoptimization::Action_make_not_entrant);
5687 assert(stopped(), "Should be stopped");
5688 }
5689 }
5690 {
5691 PreserveJVMState pjvms(this);
5692 set_control(_gvn.transform(slow_region));
5693 uncommon_trap(Deoptimization::Reason_intrinsic,
5694 Deoptimization::Action_make_not_entrant);
5695 assert(stopped(), "Should be stopped");
5696 }
5697
5698 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5699 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
5700 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5701 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
5702 }
5703
5704 if (stopped()) {
5705 return true;
5706 }
5707
5708 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
5709 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5710 // so the compiler has a chance to eliminate them: during macro expansion,
5711 // we have to set their control (CastPP nodes are eliminated).
5712 load_object_klass(src), load_object_klass(dest),
5713 load_array_length(src), load_array_length(dest));
5714
5715 ac->set_arraycopy(validated);
5716
5717 Node* n = _gvn.transform(ac);
5718 if (n == ac) {
5719 ac->connect_outputs(this);
5720 } else {
5721 assert(validated, "shouldn't transform if all arguments not validated");
5722 set_all_memory(n);
5723 }
5724 clear_upper_avx();
5725
5726
5727 return true;
5728 }
5729
5730
5731 // Helper function which determines if an arraycopy immediately follows
5732 // an allocation, with no intervening tests or other escapes for the object.
5733 AllocateArrayNode*
5734 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
5735 if (stopped()) return nullptr; // no fast path
5736 if (!C->do_aliasing()) return nullptr; // no MergeMems around
5737
5738 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
5739 if (alloc == nullptr) return nullptr;
5740
5741 Node* rawmem = memory(Compile::AliasIdxRaw);
5742 // Is the allocation's memory state untouched?
5743 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5744 // Bail out if there have been raw-memory effects since the allocation.
5745 // (Example: There might have been a call or safepoint.)
5746 return nullptr;
5747 }
5748 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5749 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5750 return nullptr;
5751 }
5752
5753 // There must be no unexpected observers of this allocation.
5754 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5755 Node* obs = ptr->fast_out(i);
5756 if (obs != this->map()) {
5757 return nullptr;
5758 }
5759 }
5760
5761 // This arraycopy must unconditionally follow the allocation of the ptr.
5762 Node* alloc_ctl = ptr->in(0);
5763 Node* ctl = control();
5764 while (ctl != alloc_ctl) {
5765 // There may be guards which feed into the slow_region.
5766 // Any other control flow means that we might not get a chance
5767 // to finish initializing the allocated object.
5768 // Various low-level checks bottom out in uncommon traps. These
5769 // are considered safe since we've already checked above that
5770 // there is no unexpected observer of this allocation.
5771 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
5772 assert(ctl->in(0)->is_If(), "must be If");
5773 ctl = ctl->in(0)->in(0);
5774 } else {
5775 return nullptr;
5776 }
5777 }
5778
5779 // If we get this far, we have an allocation which immediately
5780 // precedes the arraycopy, and we can take over zeroing the new object.
5781 // The arraycopy will finish the initialization, and provide
5782 // a new control state to which we will anchor the destination pointer.
5783
5784 return alloc;
5785 }
5786
5787 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
5788 if (node->is_IfProj()) {
5789 Node* other_proj = node->as_IfProj()->other_if_proj();
5790 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
5791 Node* obs = other_proj->fast_out(j);
5792 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
5793 (obs->as_CallStaticJava()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5794 return obs->as_CallStaticJava();
5795 }
5796 }
5797 }
5798 return nullptr;
5799 }
5800
5801 //-------------inline_encodeISOArray-----------------------------------
5802 // encode char[] to byte[] in ISO_8859_1 or ASCII
5803 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
5804 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5805 // no receiver since it is static method
5806 Node *src = argument(0);
5807 Node *src_offset = argument(1);
5808 Node *dst = argument(2);
5809 Node *dst_offset = argument(3);
5810 Node *length = argument(4);
5811
5812 src = must_be_not_null(src, true);
5813 dst = must_be_not_null(dst, true);
5814
5815 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
5816 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
5817 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
5818 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
5819 // failed array check
5820 return false;
5821 }
5822
5823 // Figure out the size and type of the elements we will be copying.
5824 BasicType src_elem = src_type->elem()->array_element_basic_type();
5825 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
5826 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5827 return false;
5828 }
5829
5830 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5831 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5832 // 'src_start' points to src array + scaled offset
5833 // 'dst_start' points to dst array + scaled offset
5834
5835 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5836 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
5837 enc = _gvn.transform(enc);
5838 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5839 set_memory(res_mem, mtype);
5840 set_result(enc);
5841 clear_upper_avx();
5842
5843 return true;
5844 }
5845
5846 //-------------inline_multiplyToLen-----------------------------------
5847 bool LibraryCallKit::inline_multiplyToLen() {
5848 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5849
5850 address stubAddr = StubRoutines::multiplyToLen();
5851 if (stubAddr == nullptr) {
5852 return false; // Intrinsic's stub is not implemented on this platform
5853 }
5854 const char* stubName = "multiplyToLen";
5855
5856 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5857
5858 // no receiver because it is a static method
5859 Node* x = argument(0);
5860 Node* xlen = argument(1);
5861 Node* y = argument(2);
5862 Node* ylen = argument(3);
5863 Node* z = argument(4);
5864
5865 x = must_be_not_null(x, true);
5866 y = must_be_not_null(y, true);
5867
5868 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
5869 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
5870 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
5871 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
5872 // failed array check
5873 return false;
5874 }
5875
5876 BasicType x_elem = x_type->elem()->array_element_basic_type();
5877 BasicType y_elem = y_type->elem()->array_element_basic_type();
5878 if (x_elem != T_INT || y_elem != T_INT) {
5879 return false;
5880 }
5881
5882 // Set the original stack and the reexecute bit for the interpreter to reexecute
5883 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5884 // on the return from z array allocation in runtime.
5885 { PreserveReexecuteState preexecs(this);
5886 jvms()->set_should_reexecute(true);
5887
5888 Node* x_start = array_element_address(x, intcon(0), x_elem);
5889 Node* y_start = array_element_address(y, intcon(0), y_elem);
5890 // 'x_start' points to x array + scaled xlen
5891 // 'y_start' points to y array + scaled ylen
5892
5893 // Allocate the result array
5894 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5895 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5896 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5897
5898 IdealKit ideal(this);
5899
5900 #define __ ideal.
5901 Node* one = __ ConI(1);
5902 Node* zero = __ ConI(0);
5903 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5904 __ set(need_alloc, zero);
5905 __ set(z_alloc, z);
5906 __ if_then(z, BoolTest::eq, null()); {
5907 __ increment (need_alloc, one);
5908 } __ else_(); {
5909 // Update graphKit memory and control from IdealKit.
5910 sync_kit(ideal);
5911 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5912 cast->init_req(0, control());
5913 _gvn.set_type(cast, cast->bottom_type());
5914 C->record_for_igvn(cast);
5915
5916 Node* zlen_arg = load_array_length(cast);
5917 // Update IdealKit memory and control from graphKit.
5918 __ sync_kit(this);
5919 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5920 __ increment (need_alloc, one);
5921 } __ end_if();
5922 } __ end_if();
5923
5924 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5925 // Update graphKit memory and control from IdealKit.
5926 sync_kit(ideal);
5927 Node * narr = new_array(klass_node, zlen, 1);
5928 // Update IdealKit memory and control from graphKit.
5929 __ sync_kit(this);
5930 __ set(z_alloc, narr);
5931 } __ end_if();
5932
5933 sync_kit(ideal);
5934 z = __ value(z_alloc);
5935 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5936 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5937 // Final sync IdealKit and GraphKit.
5938 final_sync(ideal);
5939 #undef __
5940
5941 Node* z_start = array_element_address(z, intcon(0), T_INT);
5942
5943 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5944 OptoRuntime::multiplyToLen_Type(),
5945 stubAddr, stubName, TypePtr::BOTTOM,
5946 x_start, xlen, y_start, ylen, z_start, zlen);
5947 } // original reexecute is set back here
5948
5949 C->set_has_split_ifs(true); // Has chance for split-if optimization
5950 set_result(z);
5951 return true;
5952 }
5953
5954 //-------------inline_squareToLen------------------------------------
5955 bool LibraryCallKit::inline_squareToLen() {
5956 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5957
5958 address stubAddr = StubRoutines::squareToLen();
5959 if (stubAddr == nullptr) {
5960 return false; // Intrinsic's stub is not implemented on this platform
5961 }
5962 const char* stubName = "squareToLen";
5963
5964 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5965
5966 Node* x = argument(0);
5967 Node* len = argument(1);
5968 Node* z = argument(2);
5969 Node* zlen = argument(3);
5970
5971 x = must_be_not_null(x, true);
5972 z = must_be_not_null(z, true);
5973
5974 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
5975 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
5976 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
5977 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
5978 // failed array check
5979 return false;
5980 }
5981
5982 BasicType x_elem = x_type->elem()->array_element_basic_type();
5983 BasicType z_elem = z_type->elem()->array_element_basic_type();
5984 if (x_elem != T_INT || z_elem != T_INT) {
5985 return false;
5986 }
5987
5988
5989 Node* x_start = array_element_address(x, intcon(0), x_elem);
5990 Node* z_start = array_element_address(z, intcon(0), z_elem);
5991
5992 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5993 OptoRuntime::squareToLen_Type(),
5994 stubAddr, stubName, TypePtr::BOTTOM,
5995 x_start, len, z_start, zlen);
5996
5997 set_result(z);
5998 return true;
5999 }
6000
6001 //-------------inline_mulAdd------------------------------------------
6002 bool LibraryCallKit::inline_mulAdd() {
6003 assert(UseMulAddIntrinsic, "not implemented on this platform");
6004
6005 address stubAddr = StubRoutines::mulAdd();
6006 if (stubAddr == nullptr) {
6007 return false; // Intrinsic's stub is not implemented on this platform
6008 }
6009 const char* stubName = "mulAdd";
6010
6011 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6012
6013 Node* out = argument(0);
6014 Node* in = argument(1);
6015 Node* offset = argument(2);
6016 Node* len = argument(3);
6017 Node* k = argument(4);
6018
6019 in = must_be_not_null(in, true);
6020 out = must_be_not_null(out, true);
6021
6022 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6023 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6024 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6025 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6026 // failed array check
6027 return false;
6028 }
6029
6030 BasicType out_elem = out_type->elem()->array_element_basic_type();
6031 BasicType in_elem = in_type->elem()->array_element_basic_type();
6032 if (out_elem != T_INT || in_elem != T_INT) {
6033 return false;
6034 }
6035
6036 Node* outlen = load_array_length(out);
6037 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6038 Node* out_start = array_element_address(out, intcon(0), out_elem);
6039 Node* in_start = array_element_address(in, intcon(0), in_elem);
6040
6041 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6042 OptoRuntime::mulAdd_Type(),
6043 stubAddr, stubName, TypePtr::BOTTOM,
6044 out_start,in_start, new_offset, len, k);
6045 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6046 set_result(result);
6047 return true;
6048 }
6049
6050 //-------------inline_montgomeryMultiply-----------------------------------
6051 bool LibraryCallKit::inline_montgomeryMultiply() {
6052 address stubAddr = StubRoutines::montgomeryMultiply();
6053 if (stubAddr == nullptr) {
6054 return false; // Intrinsic's stub is not implemented on this platform
6055 }
6056
6057 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6058 const char* stubName = "montgomery_multiply";
6059
6060 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6061
6062 Node* a = argument(0);
6063 Node* b = argument(1);
6064 Node* n = argument(2);
6065 Node* len = argument(3);
6066 Node* inv = argument(4);
6067 Node* m = argument(6);
6068
6069 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6070 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6071 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6072 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6073 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6074 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6075 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6076 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6077 // failed array check
6078 return false;
6079 }
6080
6081 BasicType a_elem = a_type->elem()->array_element_basic_type();
6082 BasicType b_elem = b_type->elem()->array_element_basic_type();
6083 BasicType n_elem = n_type->elem()->array_element_basic_type();
6084 BasicType m_elem = m_type->elem()->array_element_basic_type();
6085 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6086 return false;
6087 }
6088
6089 // Make the call
6090 {
6091 Node* a_start = array_element_address(a, intcon(0), a_elem);
6092 Node* b_start = array_element_address(b, intcon(0), b_elem);
6093 Node* n_start = array_element_address(n, intcon(0), n_elem);
6094 Node* m_start = array_element_address(m, intcon(0), m_elem);
6095
6096 Node* call = make_runtime_call(RC_LEAF,
6097 OptoRuntime::montgomeryMultiply_Type(),
6098 stubAddr, stubName, TypePtr::BOTTOM,
6099 a_start, b_start, n_start, len, inv, top(),
6100 m_start);
6101 set_result(m);
6102 }
6103
6104 return true;
6105 }
6106
6107 bool LibraryCallKit::inline_montgomerySquare() {
6108 address stubAddr = StubRoutines::montgomerySquare();
6109 if (stubAddr == nullptr) {
6110 return false; // Intrinsic's stub is not implemented on this platform
6111 }
6112
6113 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6114 const char* stubName = "montgomery_square";
6115
6116 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6117
6118 Node* a = argument(0);
6119 Node* n = argument(1);
6120 Node* len = argument(2);
6121 Node* inv = argument(3);
6122 Node* m = argument(5);
6123
6124 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6125 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6126 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6127 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6128 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6129 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6130 // failed array check
6131 return false;
6132 }
6133
6134 BasicType a_elem = a_type->elem()->array_element_basic_type();
6135 BasicType n_elem = n_type->elem()->array_element_basic_type();
6136 BasicType m_elem = m_type->elem()->array_element_basic_type();
6137 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6138 return false;
6139 }
6140
6141 // Make the call
6142 {
6143 Node* a_start = array_element_address(a, intcon(0), a_elem);
6144 Node* n_start = array_element_address(n, intcon(0), n_elem);
6145 Node* m_start = array_element_address(m, intcon(0), m_elem);
6146
6147 Node* call = make_runtime_call(RC_LEAF,
6148 OptoRuntime::montgomerySquare_Type(),
6149 stubAddr, stubName, TypePtr::BOTTOM,
6150 a_start, n_start, len, inv, top(),
6151 m_start);
6152 set_result(m);
6153 }
6154
6155 return true;
6156 }
6157
6158 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6159 address stubAddr = nullptr;
6160 const char* stubName = nullptr;
6161
6162 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6163 if (stubAddr == nullptr) {
6164 return false; // Intrinsic's stub is not implemented on this platform
6165 }
6166
6167 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6168
6169 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6170
6171 Node* newArr = argument(0);
6172 Node* oldArr = argument(1);
6173 Node* newIdx = argument(2);
6174 Node* shiftCount = argument(3);
6175 Node* numIter = argument(4);
6176
6177 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6178 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6179 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6180 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6181 return false;
6182 }
6183
6184 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6185 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6186 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6187 return false;
6188 }
6189
6190 // Make the call
6191 {
6192 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6193 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6194
6195 Node* call = make_runtime_call(RC_LEAF,
6196 OptoRuntime::bigIntegerShift_Type(),
6197 stubAddr,
6198 stubName,
6199 TypePtr::BOTTOM,
6200 newArr_start,
6201 oldArr_start,
6202 newIdx,
6203 shiftCount,
6204 numIter);
6205 }
6206
6207 return true;
6208 }
6209
6210 //-------------inline_vectorizedMismatch------------------------------
6211 bool LibraryCallKit::inline_vectorizedMismatch() {
6212 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6213
6214 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6215 Node* obja = argument(0); // Object
6216 Node* aoffset = argument(1); // long
6217 Node* objb = argument(3); // Object
6218 Node* boffset = argument(4); // long
6219 Node* length = argument(6); // int
6220 Node* scale = argument(7); // int
6221
6222 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6223 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6224 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6225 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6226 scale == top()) {
6227 return false; // failed input validation
6228 }
6229
6230 Node* obja_adr = make_unsafe_address(obja, aoffset);
6231 Node* objb_adr = make_unsafe_address(objb, boffset);
6232
6233 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6234 //
6235 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6236 // if (length <= inline_limit) {
6237 // inline_path:
6238 // vmask = VectorMaskGen length
6239 // vload1 = LoadVectorMasked obja, vmask
6240 // vload2 = LoadVectorMasked objb, vmask
6241 // result1 = VectorCmpMasked vload1, vload2, vmask
6242 // } else {
6243 // call_stub_path:
6244 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6245 // }
6246 // exit_block:
6247 // return Phi(result1, result2);
6248 //
6249 enum { inline_path = 1, // input is small enough to process it all at once
6250 stub_path = 2, // input is too large; call into the VM
6251 PATH_LIMIT = 3
6252 };
6253
6254 Node* exit_block = new RegionNode(PATH_LIMIT);
6255 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6256 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6257
6258 Node* call_stub_path = control();
6259
6260 BasicType elem_bt = T_ILLEGAL;
6261
6262 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6263 if (scale_t->is_con()) {
6264 switch (scale_t->get_con()) {
6265 case 0: elem_bt = T_BYTE; break;
6266 case 1: elem_bt = T_SHORT; break;
6267 case 2: elem_bt = T_INT; break;
6268 case 3: elem_bt = T_LONG; break;
6269
6270 default: elem_bt = T_ILLEGAL; break; // not supported
6271 }
6272 }
6273
6274 int inline_limit = 0;
6275 bool do_partial_inline = false;
6276
6277 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6278 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6279 do_partial_inline = inline_limit >= 16;
6280 }
6281
6282 if (do_partial_inline) {
6283 assert(elem_bt != T_ILLEGAL, "sanity");
6284
6285 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6286 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6287 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6288
6289 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6290 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6291 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6292
6293 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6294
6295 if (!stopped()) {
6296 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6297
6298 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6299 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6300 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6301 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6302
6303 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6304 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6305 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6306 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6307
6308 exit_block->init_req(inline_path, control());
6309 memory_phi->init_req(inline_path, map()->memory());
6310 result_phi->init_req(inline_path, result);
6311
6312 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6313 clear_upper_avx();
6314 }
6315 }
6316 }
6317
6318 if (call_stub_path != nullptr) {
6319 set_control(call_stub_path);
6320
6321 Node* call = make_runtime_call(RC_LEAF,
6322 OptoRuntime::vectorizedMismatch_Type(),
6323 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6324 obja_adr, objb_adr, length, scale);
6325
6326 exit_block->init_req(stub_path, control());
6327 memory_phi->init_req(stub_path, map()->memory());
6328 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6329 }
6330
6331 exit_block = _gvn.transform(exit_block);
6332 memory_phi = _gvn.transform(memory_phi);
6333 result_phi = _gvn.transform(result_phi);
6334
6335 set_control(exit_block);
6336 set_all_memory(memory_phi);
6337 set_result(result_phi);
6338
6339 return true;
6340 }
6341
6342 //------------------------------inline_vectorizedHashcode----------------------------
6343 bool LibraryCallKit::inline_vectorizedHashCode() {
6344 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6345
6346 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6347 Node* array = argument(0);
6348 Node* offset = argument(1);
6349 Node* length = argument(2);
6350 Node* initialValue = argument(3);
6351 Node* basic_type = argument(4);
6352
6353 if (basic_type == top()) {
6354 return false; // failed input validation
6355 }
6356
6357 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6358 if (!basic_type_t->is_con()) {
6359 return false; // Only intrinsify if mode argument is constant
6360 }
6361
6362 array = must_be_not_null(array, true);
6363
6364 BasicType bt = (BasicType)basic_type_t->get_con();
6365
6366 // Resolve address of first element
6367 Node* array_start = array_element_address(array, offset, bt);
6368
6369 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6370 array_start, length, initialValue, basic_type)));
6371 clear_upper_avx();
6372
6373 return true;
6374 }
6375
6376 /**
6377 * Calculate CRC32 for byte.
6378 * int java.util.zip.CRC32.update(int crc, int b)
6379 */
6380 bool LibraryCallKit::inline_updateCRC32() {
6381 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6382 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6383 // no receiver since it is static method
6384 Node* crc = argument(0); // type: int
6385 Node* b = argument(1); // type: int
6386
6387 /*
6388 * int c = ~ crc;
6389 * b = timesXtoThe32[(b ^ c) & 0xFF];
6390 * b = b ^ (c >>> 8);
6391 * crc = ~b;
6392 */
6393
6394 Node* M1 = intcon(-1);
6395 crc = _gvn.transform(new XorINode(crc, M1));
6396 Node* result = _gvn.transform(new XorINode(crc, b));
6397 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6398
6399 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6400 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6401 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6402 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6403
6404 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6405 result = _gvn.transform(new XorINode(crc, result));
6406 result = _gvn.transform(new XorINode(result, M1));
6407 set_result(result);
6408 return true;
6409 }
6410
6411 /**
6412 * Calculate CRC32 for byte[] array.
6413 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6414 */
6415 bool LibraryCallKit::inline_updateBytesCRC32() {
6416 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6417 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6418 // no receiver since it is static method
6419 Node* crc = argument(0); // type: int
6420 Node* src = argument(1); // type: oop
6421 Node* offset = argument(2); // type: int
6422 Node* length = argument(3); // type: int
6423
6424 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6425 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6426 // failed array check
6427 return false;
6428 }
6429
6430 // Figure out the size and type of the elements we will be copying.
6431 BasicType src_elem = src_type->elem()->array_element_basic_type();
6432 if (src_elem != T_BYTE) {
6433 return false;
6434 }
6435
6436 // 'src_start' points to src array + scaled offset
6437 src = must_be_not_null(src, true);
6438 Node* src_start = array_element_address(src, offset, src_elem);
6439
6440 // We assume that range check is done by caller.
6441 // TODO: generate range check (offset+length < src.length) in debug VM.
6442
6443 // Call the stub.
6444 address stubAddr = StubRoutines::updateBytesCRC32();
6445 const char *stubName = "updateBytesCRC32";
6446
6447 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6448 stubAddr, stubName, TypePtr::BOTTOM,
6449 crc, src_start, length);
6450 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6451 set_result(result);
6452 return true;
6453 }
6454
6455 /**
6456 * Calculate CRC32 for ByteBuffer.
6457 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6458 */
6459 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6460 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6461 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6462 // no receiver since it is static method
6463 Node* crc = argument(0); // type: int
6464 Node* src = argument(1); // type: long
6465 Node* offset = argument(3); // type: int
6466 Node* length = argument(4); // type: int
6467
6468 src = ConvL2X(src); // adjust Java long to machine word
6469 Node* base = _gvn.transform(new CastX2PNode(src));
6470 offset = ConvI2X(offset);
6471
6472 // 'src_start' points to src array + scaled offset
6473 Node* src_start = basic_plus_adr(top(), base, offset);
6474
6475 // Call the stub.
6476 address stubAddr = StubRoutines::updateBytesCRC32();
6477 const char *stubName = "updateBytesCRC32";
6478
6479 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6480 stubAddr, stubName, TypePtr::BOTTOM,
6481 crc, src_start, length);
6482 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6483 set_result(result);
6484 return true;
6485 }
6486
6487 //------------------------------get_table_from_crc32c_class-----------------------
6488 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6489 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6490 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6491
6492 return table;
6493 }
6494
6495 //------------------------------inline_updateBytesCRC32C-----------------------
6496 //
6497 // Calculate CRC32C for byte[] array.
6498 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6499 //
6500 bool LibraryCallKit::inline_updateBytesCRC32C() {
6501 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6502 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6503 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6504 // no receiver since it is a static method
6505 Node* crc = argument(0); // type: int
6506 Node* src = argument(1); // type: oop
6507 Node* offset = argument(2); // type: int
6508 Node* end = argument(3); // type: int
6509
6510 Node* length = _gvn.transform(new SubINode(end, offset));
6511
6512 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6513 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6514 // failed array check
6515 return false;
6516 }
6517
6518 // Figure out the size and type of the elements we will be copying.
6519 BasicType src_elem = src_type->elem()->array_element_basic_type();
6520 if (src_elem != T_BYTE) {
6521 return false;
6522 }
6523
6524 // 'src_start' points to src array + scaled offset
6525 src = must_be_not_null(src, true);
6526 Node* src_start = array_element_address(src, offset, src_elem);
6527
6528 // static final int[] byteTable in class CRC32C
6529 Node* table = get_table_from_crc32c_class(callee()->holder());
6530 table = must_be_not_null(table, true);
6531 Node* table_start = array_element_address(table, intcon(0), T_INT);
6532
6533 // We assume that range check is done by caller.
6534 // TODO: generate range check (offset+length < src.length) in debug VM.
6535
6536 // Call the stub.
6537 address stubAddr = StubRoutines::updateBytesCRC32C();
6538 const char *stubName = "updateBytesCRC32C";
6539
6540 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6541 stubAddr, stubName, TypePtr::BOTTOM,
6542 crc, src_start, length, table_start);
6543 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6544 set_result(result);
6545 return true;
6546 }
6547
6548 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6549 //
6550 // Calculate CRC32C for DirectByteBuffer.
6551 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6552 //
6553 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6554 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6555 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6556 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6557 // no receiver since it is a static method
6558 Node* crc = argument(0); // type: int
6559 Node* src = argument(1); // type: long
6560 Node* offset = argument(3); // type: int
6561 Node* end = argument(4); // type: int
6562
6563 Node* length = _gvn.transform(new SubINode(end, offset));
6564
6565 src = ConvL2X(src); // adjust Java long to machine word
6566 Node* base = _gvn.transform(new CastX2PNode(src));
6567 offset = ConvI2X(offset);
6568
6569 // 'src_start' points to src array + scaled offset
6570 Node* src_start = basic_plus_adr(top(), base, offset);
6571
6572 // static final int[] byteTable in class CRC32C
6573 Node* table = get_table_from_crc32c_class(callee()->holder());
6574 table = must_be_not_null(table, true);
6575 Node* table_start = array_element_address(table, intcon(0), T_INT);
6576
6577 // Call the stub.
6578 address stubAddr = StubRoutines::updateBytesCRC32C();
6579 const char *stubName = "updateBytesCRC32C";
6580
6581 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6582 stubAddr, stubName, TypePtr::BOTTOM,
6583 crc, src_start, length, table_start);
6584 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6585 set_result(result);
6586 return true;
6587 }
6588
6589 //------------------------------inline_updateBytesAdler32----------------------
6590 //
6591 // Calculate Adler32 checksum for byte[] array.
6592 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6593 //
6594 bool LibraryCallKit::inline_updateBytesAdler32() {
6595 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6596 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6597 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6598 // no receiver since it is static method
6599 Node* crc = argument(0); // type: int
6600 Node* src = argument(1); // type: oop
6601 Node* offset = argument(2); // type: int
6602 Node* length = argument(3); // type: int
6603
6604 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6605 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6606 // failed array check
6607 return false;
6608 }
6609
6610 // Figure out the size and type of the elements we will be copying.
6611 BasicType src_elem = src_type->elem()->array_element_basic_type();
6612 if (src_elem != T_BYTE) {
6613 return false;
6614 }
6615
6616 // 'src_start' points to src array + scaled offset
6617 Node* src_start = array_element_address(src, offset, src_elem);
6618
6619 // We assume that range check is done by caller.
6620 // TODO: generate range check (offset+length < src.length) in debug VM.
6621
6622 // Call the stub.
6623 address stubAddr = StubRoutines::updateBytesAdler32();
6624 const char *stubName = "updateBytesAdler32";
6625
6626 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6627 stubAddr, stubName, TypePtr::BOTTOM,
6628 crc, src_start, length);
6629 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6630 set_result(result);
6631 return true;
6632 }
6633
6634 //------------------------------inline_updateByteBufferAdler32---------------
6635 //
6636 // Calculate Adler32 checksum for DirectByteBuffer.
6637 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6638 //
6639 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6640 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6641 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6642 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6643 // no receiver since it is static method
6644 Node* crc = argument(0); // type: int
6645 Node* src = argument(1); // type: long
6646 Node* offset = argument(3); // type: int
6647 Node* length = argument(4); // type: int
6648
6649 src = ConvL2X(src); // adjust Java long to machine word
6650 Node* base = _gvn.transform(new CastX2PNode(src));
6651 offset = ConvI2X(offset);
6652
6653 // 'src_start' points to src array + scaled offset
6654 Node* src_start = basic_plus_adr(top(), base, offset);
6655
6656 // Call the stub.
6657 address stubAddr = StubRoutines::updateBytesAdler32();
6658 const char *stubName = "updateBytesAdler32";
6659
6660 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6661 stubAddr, stubName, TypePtr::BOTTOM,
6662 crc, src_start, length);
6663
6664 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6665 set_result(result);
6666 return true;
6667 }
6668
6669 //----------------------------inline_reference_get----------------------------
6670 // public T java.lang.ref.Reference.get();
6671 bool LibraryCallKit::inline_reference_get() {
6672 const int referent_offset = java_lang_ref_Reference::referent_offset();
6673
6674 // Get the argument:
6675 Node* reference_obj = null_check_receiver();
6676 if (stopped()) return true;
6677
6678 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6679 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6680 decorators, /*is_static*/ false, nullptr);
6681 if (result == nullptr) return false;
6682
6683 // Add memory barrier to prevent commoning reads from this field
6684 // across safepoint since GC can change its value.
6685 insert_mem_bar(Op_MemBarCPUOrder);
6686
6687 set_result(result);
6688 return true;
6689 }
6690
6691 //----------------------------inline_reference_refersTo0----------------------------
6692 // bool java.lang.ref.Reference.refersTo0();
6693 // bool java.lang.ref.PhantomReference.refersTo0();
6694 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6695 // Get arguments:
6696 Node* reference_obj = null_check_receiver();
6697 Node* other_obj = argument(1);
6698 if (stopped()) return true;
6699
6700 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6701 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6702 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6703 decorators, /*is_static*/ false, nullptr);
6704 if (referent == nullptr) return false;
6705
6706 // Add memory barrier to prevent commoning reads from this field
6707 // across safepoint since GC can change its value.
6708 insert_mem_bar(Op_MemBarCPUOrder);
6709
6710 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6711 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6712 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6713
6714 RegionNode* region = new RegionNode(3);
6715 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6716
6717 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6718 region->init_req(1, if_true);
6719 phi->init_req(1, intcon(1));
6720
6721 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6722 region->init_req(2, if_false);
6723 phi->init_req(2, intcon(0));
6724
6725 set_control(_gvn.transform(region));
6726 record_for_igvn(region);
6727 set_result(_gvn.transform(phi));
6728 return true;
6729 }
6730
6731
6732 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
6733 DecoratorSet decorators, bool is_static,
6734 ciInstanceKlass* fromKls) {
6735 if (fromKls == nullptr) {
6736 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6737 assert(tinst != nullptr, "obj is null");
6738 assert(tinst->is_loaded(), "obj is not loaded");
6739 fromKls = tinst->instance_klass();
6740 } else {
6741 assert(is_static, "only for static field access");
6742 }
6743 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6744 ciSymbol::make(fieldTypeString),
6745 is_static);
6746
6747 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
6748 if (field == nullptr) return (Node *) nullptr;
6749
6750 if (is_static) {
6751 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6752 fromObj = makecon(tip);
6753 }
6754
6755 // Next code copied from Parse::do_get_xxx():
6756
6757 // Compute address and memory type.
6758 int offset = field->offset_in_bytes();
6759 bool is_vol = field->is_volatile();
6760 ciType* field_klass = field->type();
6761 assert(field_klass->is_loaded(), "should be loaded");
6762 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6763 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6764 BasicType bt = field->layout_type();
6765
6766 // Build the resultant type of the load
6767 const Type *type;
6768 if (bt == T_OBJECT) {
6769 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6770 } else {
6771 type = Type::get_const_basic_type(bt);
6772 }
6773
6774 if (is_vol) {
6775 decorators |= MO_SEQ_CST;
6776 }
6777
6778 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
6779 }
6780
6781 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6782 bool is_exact /* true */, bool is_static /* false */,
6783 ciInstanceKlass * fromKls /* nullptr */) {
6784 if (fromKls == nullptr) {
6785 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6786 assert(tinst != nullptr, "obj is null");
6787 assert(tinst->is_loaded(), "obj is not loaded");
6788 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6789 fromKls = tinst->instance_klass();
6790 }
6791 else {
6792 assert(is_static, "only for static field access");
6793 }
6794 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6795 ciSymbol::make(fieldTypeString),
6796 is_static);
6797
6798 assert(field != nullptr, "undefined field");
6799 assert(!field->is_volatile(), "not defined for volatile fields");
6800
6801 if (is_static) {
6802 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6803 fromObj = makecon(tip);
6804 }
6805
6806 // Next code copied from Parse::do_get_xxx():
6807
6808 // Compute address and memory type.
6809 int offset = field->offset_in_bytes();
6810 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6811
6812 return adr;
6813 }
6814
6815 //------------------------------inline_aescrypt_Block-----------------------
6816 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6817 address stubAddr = nullptr;
6818 const char *stubName;
6819 assert(UseAES, "need AES instruction support");
6820
6821 switch(id) {
6822 case vmIntrinsics::_aescrypt_encryptBlock:
6823 stubAddr = StubRoutines::aescrypt_encryptBlock();
6824 stubName = "aescrypt_encryptBlock";
6825 break;
6826 case vmIntrinsics::_aescrypt_decryptBlock:
6827 stubAddr = StubRoutines::aescrypt_decryptBlock();
6828 stubName = "aescrypt_decryptBlock";
6829 break;
6830 default:
6831 break;
6832 }
6833 if (stubAddr == nullptr) return false;
6834
6835 Node* aescrypt_object = argument(0);
6836 Node* src = argument(1);
6837 Node* src_offset = argument(2);
6838 Node* dest = argument(3);
6839 Node* dest_offset = argument(4);
6840
6841 src = must_be_not_null(src, true);
6842 dest = must_be_not_null(dest, true);
6843
6844 // (1) src and dest are arrays.
6845 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6846 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
6847 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
6848 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
6849
6850 // for the quick and dirty code we will skip all the checks.
6851 // we are just trying to get the call to be generated.
6852 Node* src_start = src;
6853 Node* dest_start = dest;
6854 if (src_offset != nullptr || dest_offset != nullptr) {
6855 assert(src_offset != nullptr && dest_offset != nullptr, "");
6856 src_start = array_element_address(src, src_offset, T_BYTE);
6857 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6858 }
6859
6860 // now need to get the start of its expanded key array
6861 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6862 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6863 if (k_start == nullptr) return false;
6864
6865 // Call the stub.
6866 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6867 stubAddr, stubName, TypePtr::BOTTOM,
6868 src_start, dest_start, k_start);
6869
6870 return true;
6871 }
6872
6873 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6874 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6875 address stubAddr = nullptr;
6876 const char *stubName = nullptr;
6877
6878 assert(UseAES, "need AES instruction support");
6879
6880 switch(id) {
6881 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6882 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6883 stubName = "cipherBlockChaining_encryptAESCrypt";
6884 break;
6885 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6886 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6887 stubName = "cipherBlockChaining_decryptAESCrypt";
6888 break;
6889 default:
6890 break;
6891 }
6892 if (stubAddr == nullptr) return false;
6893
6894 Node* cipherBlockChaining_object = argument(0);
6895 Node* src = argument(1);
6896 Node* src_offset = argument(2);
6897 Node* len = argument(3);
6898 Node* dest = argument(4);
6899 Node* dest_offset = argument(5);
6900
6901 src = must_be_not_null(src, false);
6902 dest = must_be_not_null(dest, false);
6903
6904 // (1) src and dest are arrays.
6905 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6906 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
6907 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
6908 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
6909
6910 // checks are the responsibility of the caller
6911 Node* src_start = src;
6912 Node* dest_start = dest;
6913 if (src_offset != nullptr || dest_offset != nullptr) {
6914 assert(src_offset != nullptr && dest_offset != nullptr, "");
6915 src_start = array_element_address(src, src_offset, T_BYTE);
6916 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6917 }
6918
6919 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6920 // (because of the predicated logic executed earlier).
6921 // so we cast it here safely.
6922 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6923
6924 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6925 if (embeddedCipherObj == nullptr) return false;
6926
6927 // cast it to what we know it will be at runtime
6928 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6929 assert(tinst != nullptr, "CBC obj is null");
6930 assert(tinst->is_loaded(), "CBC obj is not loaded");
6931 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6932 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6933
6934 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6935 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6936 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6937 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6938 aescrypt_object = _gvn.transform(aescrypt_object);
6939
6940 // we need to get the start of the aescrypt_object's expanded key array
6941 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6942 if (k_start == nullptr) return false;
6943
6944 // similarly, get the start address of the r vector
6945 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
6946 if (objRvec == nullptr) return false;
6947 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6948
6949 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6950 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6951 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6952 stubAddr, stubName, TypePtr::BOTTOM,
6953 src_start, dest_start, k_start, r_start, len);
6954
6955 // return cipher length (int)
6956 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6957 set_result(retvalue);
6958 return true;
6959 }
6960
6961 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
6962 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
6963 address stubAddr = nullptr;
6964 const char *stubName = nullptr;
6965
6966 assert(UseAES, "need AES instruction support");
6967
6968 switch (id) {
6969 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
6970 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
6971 stubName = "electronicCodeBook_encryptAESCrypt";
6972 break;
6973 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
6974 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
6975 stubName = "electronicCodeBook_decryptAESCrypt";
6976 break;
6977 default:
6978 break;
6979 }
6980
6981 if (stubAddr == nullptr) return false;
6982
6983 Node* electronicCodeBook_object = argument(0);
6984 Node* src = argument(1);
6985 Node* src_offset = argument(2);
6986 Node* len = argument(3);
6987 Node* dest = argument(4);
6988 Node* dest_offset = argument(5);
6989
6990 // (1) src and dest are arrays.
6991 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6992 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
6993 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
6994 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
6995
6996 // checks are the responsibility of the caller
6997 Node* src_start = src;
6998 Node* dest_start = dest;
6999 if (src_offset != nullptr || dest_offset != nullptr) {
7000 assert(src_offset != nullptr && dest_offset != nullptr, "");
7001 src_start = array_element_address(src, src_offset, T_BYTE);
7002 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7003 }
7004
7005 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7006 // (because of the predicated logic executed earlier).
7007 // so we cast it here safely.
7008 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7009
7010 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7011 if (embeddedCipherObj == nullptr) return false;
7012
7013 // cast it to what we know it will be at runtime
7014 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7015 assert(tinst != nullptr, "ECB obj is null");
7016 assert(tinst->is_loaded(), "ECB obj is not loaded");
7017 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7018 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7019
7020 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7021 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7022 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7023 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7024 aescrypt_object = _gvn.transform(aescrypt_object);
7025
7026 // we need to get the start of the aescrypt_object's expanded key array
7027 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7028 if (k_start == nullptr) return false;
7029
7030 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7031 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7032 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7033 stubAddr, stubName, TypePtr::BOTTOM,
7034 src_start, dest_start, k_start, len);
7035
7036 // return cipher length (int)
7037 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7038 set_result(retvalue);
7039 return true;
7040 }
7041
7042 //------------------------------inline_counterMode_AESCrypt-----------------------
7043 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7044 assert(UseAES, "need AES instruction support");
7045 if (!UseAESCTRIntrinsics) return false;
7046
7047 address stubAddr = nullptr;
7048 const char *stubName = nullptr;
7049 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7050 stubAddr = StubRoutines::counterMode_AESCrypt();
7051 stubName = "counterMode_AESCrypt";
7052 }
7053 if (stubAddr == nullptr) return false;
7054
7055 Node* counterMode_object = argument(0);
7056 Node* src = argument(1);
7057 Node* src_offset = argument(2);
7058 Node* len = argument(3);
7059 Node* dest = argument(4);
7060 Node* dest_offset = argument(5);
7061
7062 // (1) src and dest are arrays.
7063 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7064 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7065 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7066 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7067
7068 // checks are the responsibility of the caller
7069 Node* src_start = src;
7070 Node* dest_start = dest;
7071 if (src_offset != nullptr || dest_offset != nullptr) {
7072 assert(src_offset != nullptr && dest_offset != nullptr, "");
7073 src_start = array_element_address(src, src_offset, T_BYTE);
7074 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7075 }
7076
7077 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7078 // (because of the predicated logic executed earlier).
7079 // so we cast it here safely.
7080 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7081 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7082 if (embeddedCipherObj == nullptr) return false;
7083 // cast it to what we know it will be at runtime
7084 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7085 assert(tinst != nullptr, "CTR obj is null");
7086 assert(tinst->is_loaded(), "CTR obj is not loaded");
7087 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7088 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7089 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7090 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7091 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7092 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7093 aescrypt_object = _gvn.transform(aescrypt_object);
7094 // we need to get the start of the aescrypt_object's expanded key array
7095 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7096 if (k_start == nullptr) return false;
7097 // similarly, get the start address of the r vector
7098 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7099 if (obj_counter == nullptr) return false;
7100 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7101
7102 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7103 if (saved_encCounter == nullptr) return false;
7104 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7105 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7106
7107 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7108 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7109 OptoRuntime::counterMode_aescrypt_Type(),
7110 stubAddr, stubName, TypePtr::BOTTOM,
7111 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7112
7113 // return cipher length (int)
7114 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7115 set_result(retvalue);
7116 return true;
7117 }
7118
7119 //------------------------------get_key_start_from_aescrypt_object-----------------------
7120 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7121 #if defined(PPC64) || defined(S390)
7122 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7123 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7124 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7125 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7126 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7127 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7128 if (objSessionK == nullptr) {
7129 return (Node *) nullptr;
7130 }
7131 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7132 #else
7133 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7134 #endif // PPC64
7135 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7136 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7137
7138 // now have the array, need to get the start address of the K array
7139 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7140 return k_start;
7141 }
7142
7143 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7144 // Return node representing slow path of predicate check.
7145 // the pseudo code we want to emulate with this predicate is:
7146 // for encryption:
7147 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7148 // for decryption:
7149 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7150 // note cipher==plain is more conservative than the original java code but that's OK
7151 //
7152 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7153 // The receiver was checked for null already.
7154 Node* objCBC = argument(0);
7155
7156 Node* src = argument(1);
7157 Node* dest = argument(4);
7158
7159 // Load embeddedCipher field of CipherBlockChaining object.
7160 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7161
7162 // get AESCrypt klass for instanceOf check
7163 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7164 // will have same classloader as CipherBlockChaining object
7165 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7166 assert(tinst != nullptr, "CBCobj is null");
7167 assert(tinst->is_loaded(), "CBCobj is not loaded");
7168
7169 // we want to do an instanceof comparison against the AESCrypt class
7170 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7171 if (!klass_AESCrypt->is_loaded()) {
7172 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7173 Node* ctrl = control();
7174 set_control(top()); // no regular fast path
7175 return ctrl;
7176 }
7177
7178 src = must_be_not_null(src, true);
7179 dest = must_be_not_null(dest, true);
7180
7181 // Resolve oops to stable for CmpP below.
7182 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7183
7184 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7185 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7186 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7187
7188 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7189
7190 // for encryption, we are done
7191 if (!decrypting)
7192 return instof_false; // even if it is null
7193
7194 // for decryption, we need to add a further check to avoid
7195 // taking the intrinsic path when cipher and plain are the same
7196 // see the original java code for why.
7197 RegionNode* region = new RegionNode(3);
7198 region->init_req(1, instof_false);
7199
7200 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7201 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7202 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7203 region->init_req(2, src_dest_conjoint);
7204
7205 record_for_igvn(region);
7206 return _gvn.transform(region);
7207 }
7208
7209 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7210 // Return node representing slow path of predicate check.
7211 // the pseudo code we want to emulate with this predicate is:
7212 // for encryption:
7213 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7214 // for decryption:
7215 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7216 // note cipher==plain is more conservative than the original java code but that's OK
7217 //
7218 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7219 // The receiver was checked for null already.
7220 Node* objECB = argument(0);
7221
7222 // Load embeddedCipher field of ElectronicCodeBook object.
7223 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7224
7225 // get AESCrypt klass for instanceOf check
7226 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7227 // will have same classloader as ElectronicCodeBook object
7228 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7229 assert(tinst != nullptr, "ECBobj is null");
7230 assert(tinst->is_loaded(), "ECBobj is not loaded");
7231
7232 // we want to do an instanceof comparison against the AESCrypt class
7233 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7234 if (!klass_AESCrypt->is_loaded()) {
7235 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7236 Node* ctrl = control();
7237 set_control(top()); // no regular fast path
7238 return ctrl;
7239 }
7240 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7241
7242 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7243 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7244 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7245
7246 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7247
7248 // for encryption, we are done
7249 if (!decrypting)
7250 return instof_false; // even if it is null
7251
7252 // for decryption, we need to add a further check to avoid
7253 // taking the intrinsic path when cipher and plain are the same
7254 // see the original java code for why.
7255 RegionNode* region = new RegionNode(3);
7256 region->init_req(1, instof_false);
7257 Node* src = argument(1);
7258 Node* dest = argument(4);
7259 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7260 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7261 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7262 region->init_req(2, src_dest_conjoint);
7263
7264 record_for_igvn(region);
7265 return _gvn.transform(region);
7266 }
7267
7268 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7269 // Return node representing slow path of predicate check.
7270 // the pseudo code we want to emulate with this predicate is:
7271 // for encryption:
7272 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7273 // for decryption:
7274 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7275 // note cipher==plain is more conservative than the original java code but that's OK
7276 //
7277
7278 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7279 // The receiver was checked for null already.
7280 Node* objCTR = argument(0);
7281
7282 // Load embeddedCipher field of CipherBlockChaining object.
7283 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7284
7285 // get AESCrypt klass for instanceOf check
7286 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7287 // will have same classloader as CipherBlockChaining object
7288 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7289 assert(tinst != nullptr, "CTRobj is null");
7290 assert(tinst->is_loaded(), "CTRobj is not loaded");
7291
7292 // we want to do an instanceof comparison against the AESCrypt class
7293 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7294 if (!klass_AESCrypt->is_loaded()) {
7295 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7296 Node* ctrl = control();
7297 set_control(top()); // no regular fast path
7298 return ctrl;
7299 }
7300
7301 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7302 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7303 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7304 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7305 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7306
7307 return instof_false; // even if it is null
7308 }
7309
7310 //------------------------------inline_ghash_processBlocks
7311 bool LibraryCallKit::inline_ghash_processBlocks() {
7312 address stubAddr;
7313 const char *stubName;
7314 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7315
7316 stubAddr = StubRoutines::ghash_processBlocks();
7317 stubName = "ghash_processBlocks";
7318
7319 Node* data = argument(0);
7320 Node* offset = argument(1);
7321 Node* len = argument(2);
7322 Node* state = argument(3);
7323 Node* subkeyH = argument(4);
7324
7325 state = must_be_not_null(state, true);
7326 subkeyH = must_be_not_null(subkeyH, true);
7327 data = must_be_not_null(data, true);
7328
7329 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7330 assert(state_start, "state is null");
7331 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7332 assert(subkeyH_start, "subkeyH is null");
7333 Node* data_start = array_element_address(data, offset, T_BYTE);
7334 assert(data_start, "data is null");
7335
7336 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7337 OptoRuntime::ghash_processBlocks_Type(),
7338 stubAddr, stubName, TypePtr::BOTTOM,
7339 state_start, subkeyH_start, data_start, len);
7340 return true;
7341 }
7342
7343 //------------------------------inline_chacha20Block
7344 bool LibraryCallKit::inline_chacha20Block() {
7345 address stubAddr;
7346 const char *stubName;
7347 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7348
7349 stubAddr = StubRoutines::chacha20Block();
7350 stubName = "chacha20Block";
7351
7352 Node* state = argument(0);
7353 Node* result = argument(1);
7354
7355 state = must_be_not_null(state, true);
7356 result = must_be_not_null(result, true);
7357
7358 Node* state_start = array_element_address(state, intcon(0), T_INT);
7359 assert(state_start, "state is null");
7360 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7361 assert(result_start, "result is null");
7362
7363 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7364 OptoRuntime::chacha20Block_Type(),
7365 stubAddr, stubName, TypePtr::BOTTOM,
7366 state_start, result_start);
7367 // return key stream length (int)
7368 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7369 set_result(retvalue);
7370 return true;
7371 }
7372
7373 bool LibraryCallKit::inline_base64_encodeBlock() {
7374 address stubAddr;
7375 const char *stubName;
7376 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7377 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
7378 stubAddr = StubRoutines::base64_encodeBlock();
7379 stubName = "encodeBlock";
7380
7381 if (!stubAddr) return false;
7382 Node* base64obj = argument(0);
7383 Node* src = argument(1);
7384 Node* offset = argument(2);
7385 Node* len = argument(3);
7386 Node* dest = argument(4);
7387 Node* dp = argument(5);
7388 Node* isURL = argument(6);
7389
7390 src = must_be_not_null(src, true);
7391 dest = must_be_not_null(dest, true);
7392
7393 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7394 assert(src_start, "source array is null");
7395 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7396 assert(dest_start, "destination array is null");
7397
7398 Node* base64 = make_runtime_call(RC_LEAF,
7399 OptoRuntime::base64_encodeBlock_Type(),
7400 stubAddr, stubName, TypePtr::BOTTOM,
7401 src_start, offset, len, dest_start, dp, isURL);
7402 return true;
7403 }
7404
7405 bool LibraryCallKit::inline_base64_decodeBlock() {
7406 address stubAddr;
7407 const char *stubName;
7408 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7409 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
7410 stubAddr = StubRoutines::base64_decodeBlock();
7411 stubName = "decodeBlock";
7412
7413 if (!stubAddr) return false;
7414 Node* base64obj = argument(0);
7415 Node* src = argument(1);
7416 Node* src_offset = argument(2);
7417 Node* len = argument(3);
7418 Node* dest = argument(4);
7419 Node* dest_offset = argument(5);
7420 Node* isURL = argument(6);
7421 Node* isMIME = argument(7);
7422
7423 src = must_be_not_null(src, true);
7424 dest = must_be_not_null(dest, true);
7425
7426 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7427 assert(src_start, "source array is null");
7428 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7429 assert(dest_start, "destination array is null");
7430
7431 Node* call = make_runtime_call(RC_LEAF,
7432 OptoRuntime::base64_decodeBlock_Type(),
7433 stubAddr, stubName, TypePtr::BOTTOM,
7434 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
7435 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7436 set_result(result);
7437 return true;
7438 }
7439
7440 bool LibraryCallKit::inline_poly1305_processBlocks() {
7441 address stubAddr;
7442 const char *stubName;
7443 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
7444 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
7445 stubAddr = StubRoutines::poly1305_processBlocks();
7446 stubName = "poly1305_processBlocks";
7447
7448 if (!stubAddr) return false;
7449 null_check_receiver(); // null-check receiver
7450 if (stopped()) return true;
7451
7452 Node* input = argument(1);
7453 Node* input_offset = argument(2);
7454 Node* len = argument(3);
7455 Node* alimbs = argument(4);
7456 Node* rlimbs = argument(5);
7457
7458 input = must_be_not_null(input, true);
7459 alimbs = must_be_not_null(alimbs, true);
7460 rlimbs = must_be_not_null(rlimbs, true);
7461
7462 Node* input_start = array_element_address(input, input_offset, T_BYTE);
7463 assert(input_start, "input array is null");
7464 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
7465 assert(acc_start, "acc array is null");
7466 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
7467 assert(r_start, "r array is null");
7468
7469 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7470 OptoRuntime::poly1305_processBlocks_Type(),
7471 stubAddr, stubName, TypePtr::BOTTOM,
7472 input_start, len, acc_start, r_start);
7473 return true;
7474 }
7475
7476 //------------------------------inline_digestBase_implCompress-----------------------
7477 //
7478 // Calculate MD5 for single-block byte[] array.
7479 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
7480 //
7481 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
7482 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
7483 //
7484 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
7485 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
7486 //
7487 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
7488 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
7489 //
7490 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
7491 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
7492 //
7493 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
7494 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
7495
7496 Node* digestBase_obj = argument(0);
7497 Node* src = argument(1); // type oop
7498 Node* ofs = argument(2); // type int
7499
7500 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7501 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7502 // failed array check
7503 return false;
7504 }
7505 // Figure out the size and type of the elements we will be copying.
7506 BasicType src_elem = src_type->elem()->array_element_basic_type();
7507 if (src_elem != T_BYTE) {
7508 return false;
7509 }
7510 // 'src_start' points to src array + offset
7511 src = must_be_not_null(src, true);
7512 Node* src_start = array_element_address(src, ofs, src_elem);
7513 Node* state = nullptr;
7514 Node* block_size = nullptr;
7515 address stubAddr;
7516 const char *stubName;
7517
7518 switch(id) {
7519 case vmIntrinsics::_md5_implCompress:
7520 assert(UseMD5Intrinsics, "need MD5 instruction support");
7521 state = get_state_from_digest_object(digestBase_obj, T_INT);
7522 stubAddr = StubRoutines::md5_implCompress();
7523 stubName = "md5_implCompress";
7524 break;
7525 case vmIntrinsics::_sha_implCompress:
7526 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
7527 state = get_state_from_digest_object(digestBase_obj, T_INT);
7528 stubAddr = StubRoutines::sha1_implCompress();
7529 stubName = "sha1_implCompress";
7530 break;
7531 case vmIntrinsics::_sha2_implCompress:
7532 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
7533 state = get_state_from_digest_object(digestBase_obj, T_INT);
7534 stubAddr = StubRoutines::sha256_implCompress();
7535 stubName = "sha256_implCompress";
7536 break;
7537 case vmIntrinsics::_sha5_implCompress:
7538 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
7539 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7540 stubAddr = StubRoutines::sha512_implCompress();
7541 stubName = "sha512_implCompress";
7542 break;
7543 case vmIntrinsics::_sha3_implCompress:
7544 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
7545 state = get_state_from_digest_object(digestBase_obj, T_BYTE);
7546 stubAddr = StubRoutines::sha3_implCompress();
7547 stubName = "sha3_implCompress";
7548 block_size = get_block_size_from_digest_object(digestBase_obj);
7549 if (block_size == nullptr) return false;
7550 break;
7551 default:
7552 fatal_unexpected_iid(id);
7553 return false;
7554 }
7555 if (state == nullptr) return false;
7556
7557 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
7558 if (stubAddr == nullptr) return false;
7559
7560 // Call the stub.
7561 Node* call;
7562 if (block_size == nullptr) {
7563 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
7564 stubAddr, stubName, TypePtr::BOTTOM,
7565 src_start, state);
7566 } else {
7567 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
7568 stubAddr, stubName, TypePtr::BOTTOM,
7569 src_start, state, block_size);
7570 }
7571
7572 return true;
7573 }
7574
7575 //------------------------------inline_digestBase_implCompressMB-----------------------
7576 //
7577 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
7578 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
7579 //
7580 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
7581 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7582 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7583 assert((uint)predicate < 5, "sanity");
7584 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7585
7586 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
7587 Node* src = argument(1); // byte[] array
7588 Node* ofs = argument(2); // type int
7589 Node* limit = argument(3); // type int
7590
7591 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7592 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7593 // failed array check
7594 return false;
7595 }
7596 // Figure out the size and type of the elements we will be copying.
7597 BasicType src_elem = src_type->elem()->array_element_basic_type();
7598 if (src_elem != T_BYTE) {
7599 return false;
7600 }
7601 // 'src_start' points to src array + offset
7602 src = must_be_not_null(src, false);
7603 Node* src_start = array_element_address(src, ofs, src_elem);
7604
7605 const char* klass_digestBase_name = nullptr;
7606 const char* stub_name = nullptr;
7607 address stub_addr = nullptr;
7608 BasicType elem_type = T_INT;
7609
7610 switch (predicate) {
7611 case 0:
7612 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
7613 klass_digestBase_name = "sun/security/provider/MD5";
7614 stub_name = "md5_implCompressMB";
7615 stub_addr = StubRoutines::md5_implCompressMB();
7616 }
7617 break;
7618 case 1:
7619 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
7620 klass_digestBase_name = "sun/security/provider/SHA";
7621 stub_name = "sha1_implCompressMB";
7622 stub_addr = StubRoutines::sha1_implCompressMB();
7623 }
7624 break;
7625 case 2:
7626 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
7627 klass_digestBase_name = "sun/security/provider/SHA2";
7628 stub_name = "sha256_implCompressMB";
7629 stub_addr = StubRoutines::sha256_implCompressMB();
7630 }
7631 break;
7632 case 3:
7633 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
7634 klass_digestBase_name = "sun/security/provider/SHA5";
7635 stub_name = "sha512_implCompressMB";
7636 stub_addr = StubRoutines::sha512_implCompressMB();
7637 elem_type = T_LONG;
7638 }
7639 break;
7640 case 4:
7641 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
7642 klass_digestBase_name = "sun/security/provider/SHA3";
7643 stub_name = "sha3_implCompressMB";
7644 stub_addr = StubRoutines::sha3_implCompressMB();
7645 elem_type = T_BYTE;
7646 }
7647 break;
7648 default:
7649 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
7650 }
7651 if (klass_digestBase_name != nullptr) {
7652 assert(stub_addr != nullptr, "Stub is generated");
7653 if (stub_addr == nullptr) return false;
7654
7655 // get DigestBase klass to lookup for SHA klass
7656 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7657 assert(tinst != nullptr, "digestBase_obj is not instance???");
7658 assert(tinst->is_loaded(), "DigestBase is not loaded");
7659
7660 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
7661 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
7662 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
7663 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
7664 }
7665 return false;
7666 }
7667
7668 //------------------------------inline_digestBase_implCompressMB-----------------------
7669 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
7670 BasicType elem_type, address stubAddr, const char *stubName,
7671 Node* src_start, Node* ofs, Node* limit) {
7672 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
7673 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7674 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
7675 digest_obj = _gvn.transform(digest_obj);
7676
7677 Node* state = get_state_from_digest_object(digest_obj, elem_type);
7678 if (state == nullptr) return false;
7679
7680 Node* block_size = nullptr;
7681 if (strcmp("sha3_implCompressMB", stubName) == 0) {
7682 block_size = get_block_size_from_digest_object(digest_obj);
7683 if (block_size == nullptr) return false;
7684 }
7685
7686 // Call the stub.
7687 Node* call;
7688 if (block_size == nullptr) {
7689 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7690 OptoRuntime::digestBase_implCompressMB_Type(false),
7691 stubAddr, stubName, TypePtr::BOTTOM,
7692 src_start, state, ofs, limit);
7693 } else {
7694 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7695 OptoRuntime::digestBase_implCompressMB_Type(true),
7696 stubAddr, stubName, TypePtr::BOTTOM,
7697 src_start, state, block_size, ofs, limit);
7698 }
7699
7700 // return ofs (int)
7701 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7702 set_result(result);
7703
7704 return true;
7705 }
7706
7707 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
7708 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
7709 assert(UseAES, "need AES instruction support");
7710 address stubAddr = nullptr;
7711 const char *stubName = nullptr;
7712 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
7713 stubName = "galoisCounterMode_AESCrypt";
7714
7715 if (stubAddr == nullptr) return false;
7716
7717 Node* in = argument(0);
7718 Node* inOfs = argument(1);
7719 Node* len = argument(2);
7720 Node* ct = argument(3);
7721 Node* ctOfs = argument(4);
7722 Node* out = argument(5);
7723 Node* outOfs = argument(6);
7724 Node* gctr_object = argument(7);
7725 Node* ghash_object = argument(8);
7726
7727 // (1) in, ct and out are arrays.
7728 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
7729 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
7730 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
7731 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
7732 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
7733 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
7734
7735 // checks are the responsibility of the caller
7736 Node* in_start = in;
7737 Node* ct_start = ct;
7738 Node* out_start = out;
7739 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
7740 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
7741 in_start = array_element_address(in, inOfs, T_BYTE);
7742 ct_start = array_element_address(ct, ctOfs, T_BYTE);
7743 out_start = array_element_address(out, outOfs, T_BYTE);
7744 }
7745
7746 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7747 // (because of the predicated logic executed earlier).
7748 // so we cast it here safely.
7749 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7750 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7751 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
7752 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
7753 Node* state = load_field_from_object(ghash_object, "state", "[J");
7754
7755 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
7756 return false;
7757 }
7758 // cast it to what we know it will be at runtime
7759 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
7760 assert(tinst != nullptr, "GCTR obj is null");
7761 assert(tinst->is_loaded(), "GCTR obj is not loaded");
7762 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7763 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7764 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7765 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7766 const TypeOopPtr* xtype = aklass->as_instance_type();
7767 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7768 aescrypt_object = _gvn.transform(aescrypt_object);
7769 // we need to get the start of the aescrypt_object's expanded key array
7770 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7771 if (k_start == nullptr) return false;
7772 // similarly, get the start address of the r vector
7773 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
7774 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7775 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
7776
7777
7778 // Call the stub, passing params
7779 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7780 OptoRuntime::galoisCounterMode_aescrypt_Type(),
7781 stubAddr, stubName, TypePtr::BOTTOM,
7782 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
7783
7784 // return cipher length (int)
7785 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
7786 set_result(retvalue);
7787
7788 return true;
7789 }
7790
7791 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
7792 // Return node representing slow path of predicate check.
7793 // the pseudo code we want to emulate with this predicate is:
7794 // for encryption:
7795 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7796 // for decryption:
7797 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7798 // note cipher==plain is more conservative than the original java code but that's OK
7799 //
7800
7801 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
7802 // The receiver was checked for null already.
7803 Node* objGCTR = argument(7);
7804 // Load embeddedCipher field of GCTR object.
7805 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7806 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
7807
7808 // get AESCrypt klass for instanceOf check
7809 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7810 // will have same classloader as CipherBlockChaining object
7811 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
7812 assert(tinst != nullptr, "GCTR obj is null");
7813 assert(tinst->is_loaded(), "GCTR obj is not loaded");
7814
7815 // we want to do an instanceof comparison against the AESCrypt class
7816 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7817 if (!klass_AESCrypt->is_loaded()) {
7818 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7819 Node* ctrl = control();
7820 set_control(top()); // no regular fast path
7821 return ctrl;
7822 }
7823
7824 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7825 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7826 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7827 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7828 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7829
7830 return instof_false; // even if it is null
7831 }
7832
7833 //------------------------------get_state_from_digest_object-----------------------
7834 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
7835 const char* state_type;
7836 switch (elem_type) {
7837 case T_BYTE: state_type = "[B"; break;
7838 case T_INT: state_type = "[I"; break;
7839 case T_LONG: state_type = "[J"; break;
7840 default: ShouldNotReachHere();
7841 }
7842 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
7843 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
7844 if (digest_state == nullptr) return (Node *) nullptr;
7845
7846 // now have the array, need to get the start address of the state array
7847 Node* state = array_element_address(digest_state, intcon(0), elem_type);
7848 return state;
7849 }
7850
7851 //------------------------------get_block_size_from_sha3_object----------------------------------
7852 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
7853 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
7854 assert (block_size != nullptr, "sanity");
7855 return block_size;
7856 }
7857
7858 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
7859 // Return node representing slow path of predicate check.
7860 // the pseudo code we want to emulate with this predicate is:
7861 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
7862 //
7863 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
7864 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7865 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7866 assert((uint)predicate < 5, "sanity");
7867
7868 // The receiver was checked for null already.
7869 Node* digestBaseObj = argument(0);
7870
7871 // get DigestBase klass for instanceOf check
7872 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
7873 assert(tinst != nullptr, "digestBaseObj is null");
7874 assert(tinst->is_loaded(), "DigestBase is not loaded");
7875
7876 const char* klass_name = nullptr;
7877 switch (predicate) {
7878 case 0:
7879 if (UseMD5Intrinsics) {
7880 // we want to do an instanceof comparison against the MD5 class
7881 klass_name = "sun/security/provider/MD5";
7882 }
7883 break;
7884 case 1:
7885 if (UseSHA1Intrinsics) {
7886 // we want to do an instanceof comparison against the SHA class
7887 klass_name = "sun/security/provider/SHA";
7888 }
7889 break;
7890 case 2:
7891 if (UseSHA256Intrinsics) {
7892 // we want to do an instanceof comparison against the SHA2 class
7893 klass_name = "sun/security/provider/SHA2";
7894 }
7895 break;
7896 case 3:
7897 if (UseSHA512Intrinsics) {
7898 // we want to do an instanceof comparison against the SHA5 class
7899 klass_name = "sun/security/provider/SHA5";
7900 }
7901 break;
7902 case 4:
7903 if (UseSHA3Intrinsics) {
7904 // we want to do an instanceof comparison against the SHA3 class
7905 klass_name = "sun/security/provider/SHA3";
7906 }
7907 break;
7908 default:
7909 fatal("unknown SHA intrinsic predicate: %d", predicate);
7910 }
7911
7912 ciKlass* klass = nullptr;
7913 if (klass_name != nullptr) {
7914 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
7915 }
7916 if ((klass == nullptr) || !klass->is_loaded()) {
7917 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
7918 Node* ctrl = control();
7919 set_control(top()); // no intrinsic path
7920 return ctrl;
7921 }
7922 ciInstanceKlass* instklass = klass->as_instance_klass();
7923
7924 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
7925 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7926 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7927 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7928
7929 return instof_false; // even if it is null
7930 }
7931
7932 //-------------inline_fma-----------------------------------
7933 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
7934 Node *a = nullptr;
7935 Node *b = nullptr;
7936 Node *c = nullptr;
7937 Node* result = nullptr;
7938 switch (id) {
7939 case vmIntrinsics::_fmaD:
7940 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
7941 // no receiver since it is static method
7942 a = round_double_node(argument(0));
7943 b = round_double_node(argument(2));
7944 c = round_double_node(argument(4));
7945 result = _gvn.transform(new FmaDNode(control(), a, b, c));
7946 break;
7947 case vmIntrinsics::_fmaF:
7948 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
7949 a = argument(0);
7950 b = argument(1);
7951 c = argument(2);
7952 result = _gvn.transform(new FmaFNode(control(), a, b, c));
7953 break;
7954 default:
7955 fatal_unexpected_iid(id); break;
7956 }
7957 set_result(result);
7958 return true;
7959 }
7960
7961 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
7962 // argument(0) is receiver
7963 Node* codePoint = argument(1);
7964 Node* n = nullptr;
7965
7966 switch (id) {
7967 case vmIntrinsics::_isDigit :
7968 n = new DigitNode(control(), codePoint);
7969 break;
7970 case vmIntrinsics::_isLowerCase :
7971 n = new LowerCaseNode(control(), codePoint);
7972 break;
7973 case vmIntrinsics::_isUpperCase :
7974 n = new UpperCaseNode(control(), codePoint);
7975 break;
7976 case vmIntrinsics::_isWhitespace :
7977 n = new WhitespaceNode(control(), codePoint);
7978 break;
7979 default:
7980 fatal_unexpected_iid(id);
7981 }
7982
7983 set_result(_gvn.transform(n));
7984 return true;
7985 }
7986
7987 //------------------------------inline_fp_min_max------------------------------
7988 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
7989 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
7990
7991 // The intrinsic should be used only when the API branches aren't predictable,
7992 // the last one performing the most important comparison. The following heuristic
7993 // uses the branch statistics to eventually bail out if necessary.
7994
7995 ciMethodData *md = callee()->method_data();
7996
7997 if ( md != nullptr && md->is_mature() && md->invocation_count() > 0 ) {
7998 ciCallProfile cp = caller()->call_profile_at_bci(bci());
7999
8000 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
8001 // Bail out if the call-site didn't contribute enough to the statistics.
8002 return false;
8003 }
8004
8005 uint taken = 0, not_taken = 0;
8006
8007 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
8008 if (p->is_BranchData()) {
8009 taken = ((ciBranchData*)p)->taken();
8010 not_taken = ((ciBranchData*)p)->not_taken();
8011 }
8012 }
8013
8014 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
8015 balance = balance < 0 ? -balance : balance;
8016 if ( balance > 0.2 ) {
8017 // Bail out if the most important branch is predictable enough.
8018 return false;
8019 }
8020 }
8021 */
8022
8023 Node *a = nullptr;
8024 Node *b = nullptr;
8025 Node *n = nullptr;
8026 switch (id) {
8027 case vmIntrinsics::_maxF:
8028 case vmIntrinsics::_minF:
8029 case vmIntrinsics::_maxF_strict:
8030 case vmIntrinsics::_minF_strict:
8031 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
8032 a = argument(0);
8033 b = argument(1);
8034 break;
8035 case vmIntrinsics::_maxD:
8036 case vmIntrinsics::_minD:
8037 case vmIntrinsics::_maxD_strict:
8038 case vmIntrinsics::_minD_strict:
8039 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
8040 a = round_double_node(argument(0));
8041 b = round_double_node(argument(2));
8042 break;
8043 default:
8044 fatal_unexpected_iid(id);
8045 break;
8046 }
8047 switch (id) {
8048 case vmIntrinsics::_maxF:
8049 case vmIntrinsics::_maxF_strict:
8050 n = new MaxFNode(a, b);
8051 break;
8052 case vmIntrinsics::_minF:
8053 case vmIntrinsics::_minF_strict:
8054 n = new MinFNode(a, b);
8055 break;
8056 case vmIntrinsics::_maxD:
8057 case vmIntrinsics::_maxD_strict:
8058 n = new MaxDNode(a, b);
8059 break;
8060 case vmIntrinsics::_minD:
8061 case vmIntrinsics::_minD_strict:
8062 n = new MinDNode(a, b);
8063 break;
8064 default:
8065 fatal_unexpected_iid(id);
8066 break;
8067 }
8068 set_result(_gvn.transform(n));
8069 return true;
8070 }
8071
8072 bool LibraryCallKit::inline_profileBoolean() {
8073 Node* counts = argument(1);
8074 const TypeAryPtr* ary = nullptr;
8075 ciArray* aobj = nullptr;
8076 if (counts->is_Con()
8077 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8078 && (aobj = ary->const_oop()->as_array()) != nullptr
8079 && (aobj->length() == 2)) {
8080 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8081 jint false_cnt = aobj->element_value(0).as_int();
8082 jint true_cnt = aobj->element_value(1).as_int();
8083
8084 if (C->log() != nullptr) {
8085 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8086 false_cnt, true_cnt);
8087 }
8088
8089 if (false_cnt + true_cnt == 0) {
8090 // According to profile, never executed.
8091 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8092 Deoptimization::Action_reinterpret);
8093 return true;
8094 }
8095
8096 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8097 // is a number of each value occurrences.
8098 Node* result = argument(0);
8099 if (false_cnt == 0 || true_cnt == 0) {
8100 // According to profile, one value has been never seen.
8101 int expected_val = (false_cnt == 0) ? 1 : 0;
8102
8103 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8104 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8105
8106 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8107 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8108 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8109
8110 { // Slow path: uncommon trap for never seen value and then reexecute
8111 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8112 // the value has been seen at least once.
8113 PreserveJVMState pjvms(this);
8114 PreserveReexecuteState preexecs(this);
8115 jvms()->set_should_reexecute(true);
8116
8117 set_control(slow_path);
8118 set_i_o(i_o());
8119
8120 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8121 Deoptimization::Action_reinterpret);
8122 }
8123 // The guard for never seen value enables sharpening of the result and
8124 // returning a constant. It allows to eliminate branches on the same value
8125 // later on.
8126 set_control(fast_path);
8127 result = intcon(expected_val);
8128 }
8129 // Stop profiling.
8130 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8131 // By replacing method body with profile data (represented as ProfileBooleanNode
8132 // on IR level) we effectively disable profiling.
8133 // It enables full speed execution once optimized code is generated.
8134 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8135 C->record_for_igvn(profile);
8136 set_result(profile);
8137 return true;
8138 } else {
8139 // Continue profiling.
8140 // Profile data isn't available at the moment. So, execute method's bytecode version.
8141 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8142 // is compiled and counters aren't available since corresponding MethodHandle
8143 // isn't a compile-time constant.
8144 return false;
8145 }
8146 }
8147
8148 bool LibraryCallKit::inline_isCompileConstant() {
8149 Node* n = argument(0);
8150 set_result(n->is_Con() ? intcon(1) : intcon(0));
8151 return true;
8152 }
8153
8154 //------------------------------- inline_getObjectSize --------------------------------------
8155 //
8156 // Calculate the runtime size of the object/array.
8157 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8158 //
8159 bool LibraryCallKit::inline_getObjectSize() {
8160 Node* obj = argument(3);
8161 Node* klass_node = load_object_klass(obj);
8162
8163 jint layout_con = Klass::_lh_neutral_value;
8164 Node* layout_val = get_layout_helper(klass_node, layout_con);
8165 int layout_is_con = (layout_val == nullptr);
8166
8167 if (layout_is_con) {
8168 // Layout helper is constant, can figure out things at compile time.
8169
8170 if (Klass::layout_helper_is_instance(layout_con)) {
8171 // Instance case: layout_con contains the size itself.
8172 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8173 set_result(size);
8174 } else {
8175 // Array case: size is round(header + element_size*arraylength).
8176 // Since arraylength is different for every array instance, we have to
8177 // compute the whole thing at runtime.
8178
8179 Node* arr_length = load_array_length(obj);
8180
8181 int round_mask = MinObjAlignmentInBytes - 1;
8182 int hsize = Klass::layout_helper_header_size(layout_con);
8183 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8184
8185 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8186 round_mask = 0; // strength-reduce it if it goes away completely
8187 }
8188 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8189 Node* header_size = intcon(hsize + round_mask);
8190
8191 Node* lengthx = ConvI2X(arr_length);
8192 Node* headerx = ConvI2X(header_size);
8193
8194 Node* abody = lengthx;
8195 if (eshift != 0) {
8196 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8197 }
8198 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8199 if (round_mask != 0) {
8200 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8201 }
8202 size = ConvX2L(size);
8203 set_result(size);
8204 }
8205 } else {
8206 // Layout helper is not constant, need to test for array-ness at runtime.
8207
8208 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8209 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8210 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8211 record_for_igvn(result_reg);
8212
8213 Node* array_ctl = generate_array_guard(klass_node, nullptr);
8214 if (array_ctl != nullptr) {
8215 // Array case: size is round(header + element_size*arraylength).
8216 // Since arraylength is different for every array instance, we have to
8217 // compute the whole thing at runtime.
8218
8219 PreserveJVMState pjvms(this);
8220 set_control(array_ctl);
8221 Node* arr_length = load_array_length(obj);
8222
8223 int round_mask = MinObjAlignmentInBytes - 1;
8224 Node* mask = intcon(round_mask);
8225
8226 Node* hss = intcon(Klass::_lh_header_size_shift);
8227 Node* hsm = intcon(Klass::_lh_header_size_mask);
8228 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8229 header_size = _gvn.transform(new AndINode(header_size, hsm));
8230 header_size = _gvn.transform(new AddINode(header_size, mask));
8231
8232 // There is no need to mask or shift this value.
8233 // The semantics of LShiftINode include an implicit mask to 0x1F.
8234 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8235 Node* elem_shift = layout_val;
8236
8237 Node* lengthx = ConvI2X(arr_length);
8238 Node* headerx = ConvI2X(header_size);
8239
8240 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8241 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8242 if (round_mask != 0) {
8243 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8244 }
8245 size = ConvX2L(size);
8246
8247 result_reg->init_req(_array_path, control());
8248 result_val->init_req(_array_path, size);
8249 }
8250
8251 if (!stopped()) {
8252 // Instance case: the layout helper gives us instance size almost directly,
8253 // but we need to mask out the _lh_instance_slow_path_bit.
8254 Node* size = ConvI2X(layout_val);
8255 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8256 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8257 size = _gvn.transform(new AndXNode(size, mask));
8258 size = ConvX2L(size);
8259
8260 result_reg->init_req(_instance_path, control());
8261 result_val->init_req(_instance_path, size);
8262 }
8263
8264 set_result(result_reg, result_val);
8265 }
8266
8267 return true;
8268 }
8269
8270 //------------------------------- inline_blackhole --------------------------------------
8271 //
8272 // Make sure all arguments to this node are alive.
8273 // This matches methods that were requested to be blackholed through compile commands.
8274 //
8275 bool LibraryCallKit::inline_blackhole() {
8276 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8277 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8278 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8279
8280 // Blackhole node pinches only the control, not memory. This allows
8281 // the blackhole to be pinned in the loop that computes blackholed
8282 // values, but have no other side effects, like breaking the optimizations
8283 // across the blackhole.
8284
8285 Node* bh = _gvn.transform(new BlackholeNode(control()));
8286 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
8287
8288 // Bind call arguments as blackhole arguments to keep them alive
8289 uint nargs = callee()->arg_size();
8290 for (uint i = 0; i < nargs; i++) {
8291 bh->add_req(argument(i));
8292 }
8293
8294 return true;
8295 }