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