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/ciFlatArrayKlass.hpp"
28 #include "ci/ciUtilities.inline.hpp"
29 #include "classfile/vmIntrinsics.hpp"
30 #include "compiler/compileBroker.hpp"
31 #include "compiler/compileLog.hpp"
32 #include "gc/shared/barrierSet.hpp"
33 #include "jfr/support/jfrIntrinsics.hpp"
34 #include "memory/resourceArea.hpp"
35 #include "oops/klass.inline.hpp"
36 #include "oops/objArrayKlass.hpp"
37 #include "opto/addnode.hpp"
38 #include "opto/arraycopynode.hpp"
39 #include "opto/c2compiler.hpp"
40 #include "opto/castnode.hpp"
41 #include "opto/cfgnode.hpp"
42 #include "opto/convertnode.hpp"
43 #include "opto/countbitsnode.hpp"
44 #include "opto/idealKit.hpp"
45 #include "opto/library_call.hpp"
46 #include "opto/mathexactnode.hpp"
47 #include "opto/mulnode.hpp"
48 #include "opto/narrowptrnode.hpp"
49 #include "opto/opaquenode.hpp"
50 #include "opto/parse.hpp"
51 #include "opto/runtime.hpp"
52 #include "opto/rootnode.hpp"
53 #include "opto/subnode.hpp"
54 #include "prims/jvmtiExport.hpp"
55 #include "prims/jvmtiThreadState.hpp"
56 #include "prims/unsafe.hpp"
57 #include "runtime/jniHandles.inline.hpp"
58 #include "runtime/objectMonitor.hpp"
59 #include "runtime/sharedRuntime.hpp"
60 #include "runtime/stubRoutines.hpp"
61 #include "utilities/macros.hpp"
62 #include "utilities/powerOfTwo.hpp"
63
64 //---------------------------make_vm_intrinsic----------------------------
65 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
66 vmIntrinsicID id = m->intrinsic_id();
67 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
68
69 if (!m->is_loaded()) {
70 // Do not attempt to inline unloaded methods.
71 return nullptr;
72 }
73
74 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
75 bool is_available = false;
76
77 {
78 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
79 // the compiler must transition to '_thread_in_vm' state because both
80 // methods access VM-internal data.
81 VM_ENTRY_MARK;
82 methodHandle mh(THREAD, m->get_Method());
83 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
84 if (is_available && is_virtual) {
85 is_available = vmIntrinsics::does_virtual_dispatch(id);
86 }
87 }
88
89 if (is_available) {
90 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
91 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
92 return new LibraryIntrinsic(m, is_virtual,
93 vmIntrinsics::predicates_needed(id),
94 vmIntrinsics::does_virtual_dispatch(id),
95 id);
96 } else {
97 return nullptr;
98 }
99 }
100
101 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
102 LibraryCallKit kit(jvms, this);
103 Compile* C = kit.C;
104 int nodes = C->unique();
105 #ifndef PRODUCT
106 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
107 char buf[1000];
108 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
109 tty->print_cr("Intrinsic %s", str);
110 }
111 #endif
112 ciMethod* callee = kit.callee();
113 const int bci = kit.bci();
114 #ifdef ASSERT
115 Node* ctrl = kit.control();
116 #endif
117 // Try to inline the intrinsic.
118 if (callee->check_intrinsic_candidate() &&
119 kit.try_to_inline(_last_predicate)) {
120 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
121 : "(intrinsic)";
122 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
123 if (C->print_intrinsics() || C->print_inlining()) {
124 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
125 }
126 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
127 if (C->log()) {
128 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
129 vmIntrinsics::name_at(intrinsic_id()),
130 (is_virtual() ? " virtual='1'" : ""),
131 C->unique() - nodes);
132 }
133 // Push the result from the inlined method onto the stack.
134 kit.push_result();
135 C->print_inlining_update(this);
136 return kit.transfer_exceptions_into_jvms();
137 }
138
139 // The intrinsic bailed out
140 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
141 if (jvms->has_method()) {
142 // Not a root compile.
143 const char* msg;
144 if (callee->intrinsic_candidate()) {
145 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
146 } else {
147 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
148 : "failed to inline (intrinsic), method not annotated";
149 }
150 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
151 if (C->print_intrinsics() || C->print_inlining()) {
152 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
153 }
154 } else {
155 // Root compile
156 ResourceMark rm;
157 stringStream msg_stream;
158 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
159 vmIntrinsics::name_at(intrinsic_id()),
160 is_virtual() ? " (virtual)" : "", bci);
161 const char *msg = msg_stream.freeze();
162 log_debug(jit, inlining)("%s", msg);
163 if (C->print_intrinsics() || C->print_inlining()) {
164 tty->print("%s", msg);
165 }
166 }
167 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
168 C->print_inlining_update(this);
169
170 return nullptr;
171 }
172
173 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
174 LibraryCallKit kit(jvms, this);
175 Compile* C = kit.C;
176 int nodes = C->unique();
177 _last_predicate = predicate;
178 #ifndef PRODUCT
179 assert(is_predicated() && predicate < predicates_count(), "sanity");
180 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
181 char buf[1000];
182 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
183 tty->print_cr("Predicate for intrinsic %s", str);
184 }
185 #endif
186 ciMethod* callee = kit.callee();
187 const int bci = kit.bci();
188
189 Node* slow_ctl = kit.try_to_predicate(predicate);
190 if (!kit.failing()) {
191 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
192 : "(intrinsic, predicate)";
193 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
194 if (C->print_intrinsics() || C->print_inlining()) {
195 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
196 }
197 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
198 if (C->log()) {
199 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
200 vmIntrinsics::name_at(intrinsic_id()),
201 (is_virtual() ? " virtual='1'" : ""),
202 C->unique() - nodes);
203 }
204 return slow_ctl; // Could be null if the check folds.
205 }
206
207 // The intrinsic bailed out
208 if (jvms->has_method()) {
209 // Not a root compile.
210 const char* msg = "failed to generate predicate for intrinsic";
211 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
212 if (C->print_intrinsics() || C->print_inlining()) {
213 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
214 }
215 } else {
216 // Root compile
217 ResourceMark rm;
218 stringStream msg_stream;
219 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
220 vmIntrinsics::name_at(intrinsic_id()),
221 is_virtual() ? " (virtual)" : "", bci);
222 const char *msg = msg_stream.freeze();
223 log_debug(jit, inlining)("%s", msg);
224 if (C->print_intrinsics() || C->print_inlining()) {
225 C->print_inlining_stream()->print("%s", msg);
226 }
227 }
228 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
229 return nullptr;
230 }
231
232 bool LibraryCallKit::try_to_inline(int predicate) {
233 // Handle symbolic names for otherwise undistinguished boolean switches:
234 const bool is_store = true;
235 const bool is_compress = true;
236 const bool is_static = true;
237 const bool is_volatile = true;
238
239 if (!jvms()->has_method()) {
240 // Root JVMState has a null method.
241 assert(map()->memory()->Opcode() == Op_Parm, "");
242 // Insert the memory aliasing node
243 set_all_memory(reset_memory());
244 }
245 assert(merged_memory(), "");
246
247 switch (intrinsic_id()) {
248 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
249 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
250 case vmIntrinsics::_getClass: return inline_native_getClass();
251
252 case vmIntrinsics::_ceil:
253 case vmIntrinsics::_floor:
254 case vmIntrinsics::_rint:
255 case vmIntrinsics::_dsin:
256 case vmIntrinsics::_dcos:
257 case vmIntrinsics::_dtan:
258 case vmIntrinsics::_dabs:
259 case vmIntrinsics::_fabs:
260 case vmIntrinsics::_iabs:
261 case vmIntrinsics::_labs:
262 case vmIntrinsics::_datan2:
263 case vmIntrinsics::_dsqrt:
264 case vmIntrinsics::_dsqrt_strict:
265 case vmIntrinsics::_dexp:
266 case vmIntrinsics::_dlog:
267 case vmIntrinsics::_dlog10:
268 case vmIntrinsics::_dpow:
269 case vmIntrinsics::_dcopySign:
270 case vmIntrinsics::_fcopySign:
271 case vmIntrinsics::_dsignum:
272 case vmIntrinsics::_roundF:
273 case vmIntrinsics::_roundD:
274 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
275
276 case vmIntrinsics::_notify:
277 case vmIntrinsics::_notifyAll:
278 return inline_notify(intrinsic_id());
279
280 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
281 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
282 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
283 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
284 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
285 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
286 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
287 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
288 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
289 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
290 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
291 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
292 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
293 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
294
295 case vmIntrinsics::_arraycopy: return inline_arraycopy();
296
297 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
298 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
299 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
300 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
301
302 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
303 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
304 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
305 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
306 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
307 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
308 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
309 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
310
311 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
312 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
313
314 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
315
316 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
317 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
318 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
319 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
320
321 case vmIntrinsics::_compressStringC:
322 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
323 case vmIntrinsics::_inflateStringC:
324 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
325
326 case vmIntrinsics::_makePrivateBuffer: return inline_unsafe_make_private_buffer();
327 case vmIntrinsics::_finishPrivateBuffer: return inline_unsafe_finish_private_buffer();
328 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
329 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
330 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
331 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
332 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
333 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
334 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
335 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
336 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
337 case vmIntrinsics::_getValue: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false, true);
338
339 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
340 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
341 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
342 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
343 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
344 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
345 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
346 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
347 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
348 case vmIntrinsics::_putValue: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false, true);
349
350 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
351 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
352 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
353 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
354 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
355 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
356 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
357 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
358 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
359
360 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
361 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
362 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
363 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
364 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
365 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
366 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
367 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
368 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
369
370 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
371 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
372 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
373 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
374
375 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
376 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
377 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
378 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
379
380 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
381 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
382 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
383 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
384 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
385 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
386 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
387 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
388 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
389
390 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
391 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
392 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
393 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
394 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
395 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
396 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
397 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
398 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
399
400 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
401 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
402 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
403 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
404 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
405 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
406 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
407 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
408 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
409
410 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
411 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
412 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
413 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
414 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
415 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
416 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
417 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
418 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
419
420 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
421 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
422 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
423 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
424 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
425
426 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
427 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
428 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
429 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
430 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
431 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
432 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
433 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
434 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
435 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
436 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
437 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
438 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
439 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
440 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
441 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
442 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
443 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
444 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
445 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
446
447 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
448 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
449 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
450 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
451 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
452 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
453 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
454 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
455 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
456 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
457 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
458 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
459 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
460 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
461 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
462
463 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
464 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
465 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
466 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
467
468 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
469 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
470 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
471 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
472 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
473
474 case vmIntrinsics::_loadFence:
475 case vmIntrinsics::_storeFence:
476 case vmIntrinsics::_storeStoreFence:
477 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
478
479 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
480
481 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
482 case vmIntrinsics::_currentThread: return inline_native_currentThread();
483 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
484
485 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
486 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
487
488 #if INCLUDE_JVMTI
489 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
490 "notifyJvmtiStart", true, false);
491 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
492 "notifyJvmtiEnd", false, true);
493 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
494 "notifyJvmtiMount", false, false);
495 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
496 "notifyJvmtiUnmount", false, false);
497 case vmIntrinsics::_notifyJvmtiVThreadHideFrames: return inline_native_notify_jvmti_hide();
498 #endif
499
500 #ifdef JFR_HAVE_INTRINSICS
501 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
502 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
503 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
504 #endif
505 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
506 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
507 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
508 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
509 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
510 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
511 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
512 case vmIntrinsics::_isFlattenedArray: return inline_unsafe_isFlattenedArray();
513 case vmIntrinsics::_getLength: return inline_native_getLength();
514 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
515 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
516 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
517 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
518 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
519 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
520 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
521
522 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
523 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
524
525 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
526
527 case vmIntrinsics::_isInstance:
528 case vmIntrinsics::_getModifiers:
529 case vmIntrinsics::_isInterface:
530 case vmIntrinsics::_isArray:
531 case vmIntrinsics::_isPrimitive:
532 case vmIntrinsics::_isHidden:
533 case vmIntrinsics::_getSuperclass:
534 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
535
536 case vmIntrinsics::_asPrimaryType:
537 case vmIntrinsics::_asPrimaryTypeArg:
538 case vmIntrinsics::_asValueType:
539 case vmIntrinsics::_asValueTypeArg: return inline_primitive_Class_conversion(intrinsic_id());
540
541 case vmIntrinsics::_floatToRawIntBits:
542 case vmIntrinsics::_floatToIntBits:
543 case vmIntrinsics::_intBitsToFloat:
544 case vmIntrinsics::_doubleToRawLongBits:
545 case vmIntrinsics::_doubleToLongBits:
546 case vmIntrinsics::_longBitsToDouble:
547 case vmIntrinsics::_floatToFloat16:
548 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
549
550 case vmIntrinsics::_floatIsFinite:
551 case vmIntrinsics::_floatIsInfinite:
552 case vmIntrinsics::_doubleIsFinite:
553 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
554
555 case vmIntrinsics::_numberOfLeadingZeros_i:
556 case vmIntrinsics::_numberOfLeadingZeros_l:
557 case vmIntrinsics::_numberOfTrailingZeros_i:
558 case vmIntrinsics::_numberOfTrailingZeros_l:
559 case vmIntrinsics::_bitCount_i:
560 case vmIntrinsics::_bitCount_l:
561 case vmIntrinsics::_reverse_i:
562 case vmIntrinsics::_reverse_l:
563 case vmIntrinsics::_reverseBytes_i:
564 case vmIntrinsics::_reverseBytes_l:
565 case vmIntrinsics::_reverseBytes_s:
566 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
567
568 case vmIntrinsics::_compress_i:
569 case vmIntrinsics::_compress_l:
570 case vmIntrinsics::_expand_i:
571 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
572
573 case vmIntrinsics::_compareUnsigned_i:
574 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
575
576 case vmIntrinsics::_divideUnsigned_i:
577 case vmIntrinsics::_divideUnsigned_l:
578 case vmIntrinsics::_remainderUnsigned_i:
579 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
580
581 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
582
583 case vmIntrinsics::_Reference_get: return inline_reference_get();
584 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
585 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
586
587 case vmIntrinsics::_Class_cast: return inline_Class_cast();
588
589 case vmIntrinsics::_aescrypt_encryptBlock:
590 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
591
592 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
593 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
594 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
595
596 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
597 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
598 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
599
600 case vmIntrinsics::_counterMode_AESCrypt:
601 return inline_counterMode_AESCrypt(intrinsic_id());
602
603 case vmIntrinsics::_galoisCounterMode_AESCrypt:
604 return inline_galoisCounterMode_AESCrypt();
605
606 case vmIntrinsics::_md5_implCompress:
607 case vmIntrinsics::_sha_implCompress:
608 case vmIntrinsics::_sha2_implCompress:
609 case vmIntrinsics::_sha5_implCompress:
610 case vmIntrinsics::_sha3_implCompress:
611 return inline_digestBase_implCompress(intrinsic_id());
612
613 case vmIntrinsics::_digestBase_implCompressMB:
614 return inline_digestBase_implCompressMB(predicate);
615
616 case vmIntrinsics::_multiplyToLen:
617 return inline_multiplyToLen();
618
619 case vmIntrinsics::_squareToLen:
620 return inline_squareToLen();
621
622 case vmIntrinsics::_mulAdd:
623 return inline_mulAdd();
624
625 case vmIntrinsics::_montgomeryMultiply:
626 return inline_montgomeryMultiply();
627 case vmIntrinsics::_montgomerySquare:
628 return inline_montgomerySquare();
629
630 case vmIntrinsics::_bigIntegerRightShiftWorker:
631 return inline_bigIntegerShift(true);
632 case vmIntrinsics::_bigIntegerLeftShiftWorker:
633 return inline_bigIntegerShift(false);
634
635 case vmIntrinsics::_vectorizedMismatch:
636 return inline_vectorizedMismatch();
637
638 case vmIntrinsics::_ghash_processBlocks:
639 return inline_ghash_processBlocks();
640 case vmIntrinsics::_chacha20Block:
641 return inline_chacha20Block();
642 case vmIntrinsics::_base64_encodeBlock:
643 return inline_base64_encodeBlock();
644 case vmIntrinsics::_base64_decodeBlock:
645 return inline_base64_decodeBlock();
646 case vmIntrinsics::_poly1305_processBlocks:
647 return inline_poly1305_processBlocks();
648
649 case vmIntrinsics::_encodeISOArray:
650 case vmIntrinsics::_encodeByteISOArray:
651 return inline_encodeISOArray(false);
652 case vmIntrinsics::_encodeAsciiArray:
653 return inline_encodeISOArray(true);
654
655 case vmIntrinsics::_updateCRC32:
656 return inline_updateCRC32();
657 case vmIntrinsics::_updateBytesCRC32:
658 return inline_updateBytesCRC32();
659 case vmIntrinsics::_updateByteBufferCRC32:
660 return inline_updateByteBufferCRC32();
661
662 case vmIntrinsics::_updateBytesCRC32C:
663 return inline_updateBytesCRC32C();
664 case vmIntrinsics::_updateDirectByteBufferCRC32C:
665 return inline_updateDirectByteBufferCRC32C();
666
667 case vmIntrinsics::_updateBytesAdler32:
668 return inline_updateBytesAdler32();
669 case vmIntrinsics::_updateByteBufferAdler32:
670 return inline_updateByteBufferAdler32();
671
672 case vmIntrinsics::_profileBoolean:
673 return inline_profileBoolean();
674 case vmIntrinsics::_isCompileConstant:
675 return inline_isCompileConstant();
676
677 case vmIntrinsics::_countPositives:
678 return inline_countPositives();
679
680 case vmIntrinsics::_fmaD:
681 case vmIntrinsics::_fmaF:
682 return inline_fma(intrinsic_id());
683
684 case vmIntrinsics::_isDigit:
685 case vmIntrinsics::_isLowerCase:
686 case vmIntrinsics::_isUpperCase:
687 case vmIntrinsics::_isWhitespace:
688 return inline_character_compare(intrinsic_id());
689
690 case vmIntrinsics::_min:
691 case vmIntrinsics::_max:
692 case vmIntrinsics::_min_strict:
693 case vmIntrinsics::_max_strict:
694 return inline_min_max(intrinsic_id());
695
696 case vmIntrinsics::_maxF:
697 case vmIntrinsics::_minF:
698 case vmIntrinsics::_maxD:
699 case vmIntrinsics::_minD:
700 case vmIntrinsics::_maxF_strict:
701 case vmIntrinsics::_minF_strict:
702 case vmIntrinsics::_maxD_strict:
703 case vmIntrinsics::_minD_strict:
704 return inline_fp_min_max(intrinsic_id());
705
706 case vmIntrinsics::_VectorUnaryOp:
707 return inline_vector_nary_operation(1);
708 case vmIntrinsics::_VectorBinaryOp:
709 return inline_vector_nary_operation(2);
710 case vmIntrinsics::_VectorTernaryOp:
711 return inline_vector_nary_operation(3);
712 case vmIntrinsics::_VectorFromBitsCoerced:
713 return inline_vector_frombits_coerced();
714 case vmIntrinsics::_VectorShuffleIota:
715 return inline_vector_shuffle_iota();
716 case vmIntrinsics::_VectorMaskOp:
717 return inline_vector_mask_operation();
718 case vmIntrinsics::_VectorShuffleToVector:
719 return inline_vector_shuffle_to_vector();
720 case vmIntrinsics::_VectorLoadOp:
721 return inline_vector_mem_operation(/*is_store=*/false);
722 case vmIntrinsics::_VectorLoadMaskedOp:
723 return inline_vector_mem_masked_operation(/*is_store*/false);
724 case vmIntrinsics::_VectorStoreOp:
725 return inline_vector_mem_operation(/*is_store=*/true);
726 case vmIntrinsics::_VectorStoreMaskedOp:
727 return inline_vector_mem_masked_operation(/*is_store=*/true);
728 case vmIntrinsics::_VectorGatherOp:
729 return inline_vector_gather_scatter(/*is_scatter*/ false);
730 case vmIntrinsics::_VectorScatterOp:
731 return inline_vector_gather_scatter(/*is_scatter*/ true);
732 case vmIntrinsics::_VectorReductionCoerced:
733 return inline_vector_reduction();
734 case vmIntrinsics::_VectorTest:
735 return inline_vector_test();
736 case vmIntrinsics::_VectorBlend:
737 return inline_vector_blend();
738 case vmIntrinsics::_VectorRearrange:
739 return inline_vector_rearrange();
740 case vmIntrinsics::_VectorCompare:
741 return inline_vector_compare();
742 case vmIntrinsics::_VectorBroadcastInt:
743 return inline_vector_broadcast_int();
744 case vmIntrinsics::_VectorConvert:
745 return inline_vector_convert();
746 case vmIntrinsics::_VectorInsert:
747 return inline_vector_insert();
748 case vmIntrinsics::_VectorExtract:
749 return inline_vector_extract();
750 case vmIntrinsics::_VectorCompressExpand:
751 return inline_vector_compress_expand();
752 case vmIntrinsics::_IndexVector:
753 return inline_index_vector();
754 case vmIntrinsics::_IndexPartiallyInUpperRange:
755 return inline_index_partially_in_upper_range();
756
757 case vmIntrinsics::_getObjectSize:
758 return inline_getObjectSize();
759
760 case vmIntrinsics::_blackhole:
761 return inline_blackhole();
762
763 default:
764 // If you get here, it may be that someone has added a new intrinsic
765 // to the list in vmIntrinsics.hpp without implementing it here.
766 #ifndef PRODUCT
767 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
768 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
769 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
770 }
771 #endif
772 return false;
773 }
774 }
775
776 Node* LibraryCallKit::try_to_predicate(int predicate) {
777 if (!jvms()->has_method()) {
778 // Root JVMState has a null method.
779 assert(map()->memory()->Opcode() == Op_Parm, "");
780 // Insert the memory aliasing node
781 set_all_memory(reset_memory());
782 }
783 assert(merged_memory(), "");
784
785 switch (intrinsic_id()) {
786 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
787 return inline_cipherBlockChaining_AESCrypt_predicate(false);
788 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
789 return inline_cipherBlockChaining_AESCrypt_predicate(true);
790 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
791 return inline_electronicCodeBook_AESCrypt_predicate(false);
792 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
793 return inline_electronicCodeBook_AESCrypt_predicate(true);
794 case vmIntrinsics::_counterMode_AESCrypt:
795 return inline_counterMode_AESCrypt_predicate();
796 case vmIntrinsics::_digestBase_implCompressMB:
797 return inline_digestBase_implCompressMB_predicate(predicate);
798 case vmIntrinsics::_galoisCounterMode_AESCrypt:
799 return inline_galoisCounterMode_AESCrypt_predicate();
800
801 default:
802 // If you get here, it may be that someone has added a new intrinsic
803 // to the list in vmIntrinsics.hpp without implementing it here.
804 #ifndef PRODUCT
805 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
806 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
807 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
808 }
809 #endif
810 Node* slow_ctl = control();
811 set_control(top()); // No fast path intrinsic
812 return slow_ctl;
813 }
814 }
815
816 //------------------------------set_result-------------------------------
817 // Helper function for finishing intrinsics.
818 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
819 record_for_igvn(region);
820 set_control(_gvn.transform(region));
821 set_result( _gvn.transform(value));
822 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
823 }
824
825 //------------------------------generate_guard---------------------------
826 // Helper function for generating guarded fast-slow graph structures.
827 // The given 'test', if true, guards a slow path. If the test fails
828 // then a fast path can be taken. (We generally hope it fails.)
829 // In all cases, GraphKit::control() is updated to the fast path.
830 // The returned value represents the control for the slow path.
831 // The return value is never 'top'; it is either a valid control
832 // or null if it is obvious that the slow path can never be taken.
833 // Also, if region and the slow control are not null, the slow edge
834 // is appended to the region.
835 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
836 if (stopped()) {
837 // Already short circuited.
838 return nullptr;
839 }
840
841 // Build an if node and its projections.
842 // If test is true we take the slow path, which we assume is uncommon.
843 if (_gvn.type(test) == TypeInt::ZERO) {
844 // The slow branch is never taken. No need to build this guard.
845 return nullptr;
846 }
847
848 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
849
850 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
851 if (if_slow == top()) {
852 // The slow branch is never taken. No need to build this guard.
853 return nullptr;
854 }
855
856 if (region != nullptr)
857 region->add_req(if_slow);
858
859 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
860 set_control(if_fast);
861
862 return if_slow;
863 }
864
865 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
866 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
867 }
868 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
869 return generate_guard(test, region, PROB_FAIR);
870 }
871
872 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
873 Node* *pos_index) {
874 if (stopped())
875 return nullptr; // already stopped
876 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
877 return nullptr; // index is already adequately typed
878 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
879 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
880 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
881 if (is_neg != nullptr && pos_index != nullptr) {
882 // Emulate effect of Parse::adjust_map_after_if.
883 Node* ccast = new CastIINode(index, TypeInt::POS);
884 ccast->set_req(0, control());
885 (*pos_index) = _gvn.transform(ccast);
886 }
887 return is_neg;
888 }
889
890 // Make sure that 'position' is a valid limit index, in [0..length].
891 // There are two equivalent plans for checking this:
892 // A. (offset + copyLength) unsigned<= arrayLength
893 // B. offset <= (arrayLength - copyLength)
894 // We require that all of the values above, except for the sum and
895 // difference, are already known to be non-negative.
896 // Plan A is robust in the face of overflow, if offset and copyLength
897 // are both hugely positive.
898 //
899 // Plan B is less direct and intuitive, but it does not overflow at
900 // all, since the difference of two non-negatives is always
901 // representable. Whenever Java methods must perform the equivalent
902 // check they generally use Plan B instead of Plan A.
903 // For the moment we use Plan A.
904 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
905 Node* subseq_length,
906 Node* array_length,
907 RegionNode* region) {
908 if (stopped())
909 return nullptr; // already stopped
910 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
911 if (zero_offset && subseq_length->eqv_uncast(array_length))
912 return nullptr; // common case of whole-array copy
913 Node* last = subseq_length;
914 if (!zero_offset) // last += offset
915 last = _gvn.transform(new AddINode(last, offset));
916 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
917 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
918 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
919 return is_over;
920 }
921
922 // Emit range checks for the given String.value byte array
923 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
924 if (stopped()) {
925 return; // already stopped
926 }
927 RegionNode* bailout = new RegionNode(1);
928 record_for_igvn(bailout);
929 if (char_count) {
930 // Convert char count to byte count
931 count = _gvn.transform(new LShiftINode(count, intcon(1)));
932 }
933
934 // Offset and count must not be negative
935 generate_negative_guard(offset, bailout);
936 generate_negative_guard(count, bailout);
937 // Offset + count must not exceed length of array
938 generate_limit_guard(offset, count, load_array_length(array), bailout);
939
940 if (bailout->req() > 1) {
941 PreserveJVMState pjvms(this);
942 set_control(_gvn.transform(bailout));
943 uncommon_trap(Deoptimization::Reason_intrinsic,
944 Deoptimization::Action_maybe_recompile);
945 }
946 }
947
948 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
949 bool is_immutable) {
950 ciKlass* thread_klass = env()->Thread_klass();
951 const Type* thread_type
952 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
953
954 Node* thread = _gvn.transform(new ThreadLocalNode());
955 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
956 tls_output = thread;
957
958 Node* thread_obj_handle
959 = (is_immutable
960 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
961 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
962 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
963 thread_obj_handle = _gvn.transform(thread_obj_handle);
964
965 DecoratorSet decorators = IN_NATIVE;
966 if (is_immutable) {
967 decorators |= C2_IMMUTABLE_MEMORY;
968 }
969 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
970 }
971
972 //--------------------------generate_current_thread--------------------
973 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
974 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
975 /*is_immutable*/false);
976 }
977
978 //--------------------------generate_virtual_thread--------------------
979 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
980 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
981 !C->method()->changes_current_thread());
982 }
983
984 //------------------------------make_string_method_node------------------------
985 // Helper method for String intrinsic functions. This version is called with
986 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
987 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
988 // containing the lengths of str1 and str2.
989 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
990 Node* result = nullptr;
991 switch (opcode) {
992 case Op_StrIndexOf:
993 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
994 str1_start, cnt1, str2_start, cnt2, ae);
995 break;
996 case Op_StrComp:
997 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
998 str1_start, cnt1, str2_start, cnt2, ae);
999 break;
1000 case Op_StrEquals:
1001 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1002 // Use the constant length if there is one because optimized match rule may exist.
1003 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1004 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1005 break;
1006 default:
1007 ShouldNotReachHere();
1008 return nullptr;
1009 }
1010
1011 // All these intrinsics have checks.
1012 C->set_has_split_ifs(true); // Has chance for split-if optimization
1013 clear_upper_avx();
1014
1015 return _gvn.transform(result);
1016 }
1017
1018 //------------------------------inline_string_compareTo------------------------
1019 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1020 Node* arg1 = argument(0);
1021 Node* arg2 = argument(1);
1022
1023 arg1 = must_be_not_null(arg1, true);
1024 arg2 = must_be_not_null(arg2, true);
1025
1026 // Get start addr and length of first argument
1027 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1028 Node* arg1_cnt = load_array_length(arg1);
1029
1030 // Get start addr and length of second argument
1031 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1032 Node* arg2_cnt = load_array_length(arg2);
1033
1034 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1035 set_result(result);
1036 return true;
1037 }
1038
1039 //------------------------------inline_string_equals------------------------
1040 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1041 Node* arg1 = argument(0);
1042 Node* arg2 = argument(1);
1043
1044 // paths (plus control) merge
1045 RegionNode* region = new RegionNode(3);
1046 Node* phi = new PhiNode(region, TypeInt::BOOL);
1047
1048 if (!stopped()) {
1049
1050 arg1 = must_be_not_null(arg1, true);
1051 arg2 = must_be_not_null(arg2, true);
1052
1053 // Get start addr and length of first argument
1054 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1055 Node* arg1_cnt = load_array_length(arg1);
1056
1057 // Get start addr and length of second argument
1058 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1059 Node* arg2_cnt = load_array_length(arg2);
1060
1061 // Check for arg1_cnt != arg2_cnt
1062 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1063 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1064 Node* if_ne = generate_slow_guard(bol, nullptr);
1065 if (if_ne != nullptr) {
1066 phi->init_req(2, intcon(0));
1067 region->init_req(2, if_ne);
1068 }
1069
1070 // Check for count == 0 is done by assembler code for StrEquals.
1071
1072 if (!stopped()) {
1073 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1074 phi->init_req(1, equals);
1075 region->init_req(1, control());
1076 }
1077 }
1078
1079 // post merge
1080 set_control(_gvn.transform(region));
1081 record_for_igvn(region);
1082
1083 set_result(_gvn.transform(phi));
1084 return true;
1085 }
1086
1087 //------------------------------inline_array_equals----------------------------
1088 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1089 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1090 Node* arg1 = argument(0);
1091 Node* arg2 = argument(1);
1092
1093 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1094 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1095 clear_upper_avx();
1096
1097 return true;
1098 }
1099
1100
1101 //------------------------------inline_countPositives------------------------------
1102 bool LibraryCallKit::inline_countPositives() {
1103 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1104 return false;
1105 }
1106
1107 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1108 // no receiver since it is static method
1109 Node* ba = argument(0);
1110 Node* offset = argument(1);
1111 Node* len = argument(2);
1112
1113 ba = must_be_not_null(ba, true);
1114
1115 // Range checks
1116 generate_string_range_check(ba, offset, len, false);
1117 if (stopped()) {
1118 return true;
1119 }
1120 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1121 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1122 set_result(_gvn.transform(result));
1123 return true;
1124 }
1125
1126 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1127 Node* index = argument(0);
1128 Node* length = bt == T_INT ? argument(1) : argument(2);
1129 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1130 return false;
1131 }
1132
1133 // check that length is positive
1134 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1135 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1136
1137 {
1138 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1139 uncommon_trap(Deoptimization::Reason_intrinsic,
1140 Deoptimization::Action_make_not_entrant);
1141 }
1142
1143 if (stopped()) {
1144 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1145 return true;
1146 }
1147
1148 // length is now known positive, add a cast node to make this explicit
1149 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1150 Node* casted_length = ConstraintCastNode::make(control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), ConstraintCastNode::RegularDependency, bt);
1151 casted_length = _gvn.transform(casted_length);
1152 replace_in_map(length, casted_length);
1153 length = casted_length;
1154
1155 // Use an unsigned comparison for the range check itself
1156 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1157 BoolTest::mask btest = BoolTest::lt;
1158 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1159 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1160 _gvn.set_type(rc, rc->Value(&_gvn));
1161 if (!rc_bool->is_Con()) {
1162 record_for_igvn(rc);
1163 }
1164 set_control(_gvn.transform(new IfTrueNode(rc)));
1165 {
1166 PreserveJVMState pjvms(this);
1167 set_control(_gvn.transform(new IfFalseNode(rc)));
1168 uncommon_trap(Deoptimization::Reason_range_check,
1169 Deoptimization::Action_make_not_entrant);
1170 }
1171
1172 if (stopped()) {
1173 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1174 return true;
1175 }
1176
1177 // index is now known to be >= 0 and < length, cast it
1178 Node* result = ConstraintCastNode::make(control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), ConstraintCastNode::RegularDependency, bt);
1179 result = _gvn.transform(result);
1180 set_result(result);
1181 replace_in_map(index, result);
1182 return true;
1183 }
1184
1185 //------------------------------inline_string_indexOf------------------------
1186 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1187 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1188 return false;
1189 }
1190 Node* src = argument(0);
1191 Node* tgt = argument(1);
1192
1193 // Make the merge point
1194 RegionNode* result_rgn = new RegionNode(4);
1195 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1196
1197 src = must_be_not_null(src, true);
1198 tgt = must_be_not_null(tgt, true);
1199
1200 // Get start addr and length of source string
1201 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1202 Node* src_count = load_array_length(src);
1203
1204 // Get start addr and length of substring
1205 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1206 Node* tgt_count = load_array_length(tgt);
1207
1208 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1209 // Divide src size by 2 if String is UTF16 encoded
1210 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1211 }
1212 if (ae == StrIntrinsicNode::UU) {
1213 // Divide substring size by 2 if String is UTF16 encoded
1214 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1215 }
1216
1217 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1218 if (result != nullptr) {
1219 result_phi->init_req(3, result);
1220 result_rgn->init_req(3, control());
1221 }
1222 set_control(_gvn.transform(result_rgn));
1223 record_for_igvn(result_rgn);
1224 set_result(_gvn.transform(result_phi));
1225
1226 return true;
1227 }
1228
1229 //-----------------------------inline_string_indexOf-----------------------
1230 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1231 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1232 return false;
1233 }
1234 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1235 return false;
1236 }
1237 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1238 Node* src = argument(0); // byte[]
1239 Node* src_count = argument(1); // char count
1240 Node* tgt = argument(2); // byte[]
1241 Node* tgt_count = argument(3); // char count
1242 Node* from_index = argument(4); // char index
1243
1244 src = must_be_not_null(src, true);
1245 tgt = must_be_not_null(tgt, true);
1246
1247 // Multiply byte array index by 2 if String is UTF16 encoded
1248 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1249 src_count = _gvn.transform(new SubINode(src_count, from_index));
1250 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1251 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1252
1253 // Range checks
1254 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1255 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1256 if (stopped()) {
1257 return true;
1258 }
1259
1260 RegionNode* region = new RegionNode(5);
1261 Node* phi = new PhiNode(region, TypeInt::INT);
1262
1263 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1264 if (result != nullptr) {
1265 // The result is index relative to from_index if substring was found, -1 otherwise.
1266 // Generate code which will fold into cmove.
1267 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1268 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1269
1270 Node* if_lt = generate_slow_guard(bol, nullptr);
1271 if (if_lt != nullptr) {
1272 // result == -1
1273 phi->init_req(3, result);
1274 region->init_req(3, if_lt);
1275 }
1276 if (!stopped()) {
1277 result = _gvn.transform(new AddINode(result, from_index));
1278 phi->init_req(4, result);
1279 region->init_req(4, control());
1280 }
1281 }
1282
1283 set_control(_gvn.transform(region));
1284 record_for_igvn(region);
1285 set_result(_gvn.transform(phi));
1286 clear_upper_avx();
1287
1288 return true;
1289 }
1290
1291 // Create StrIndexOfNode with fast path checks
1292 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1293 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1294 // Check for substr count > string count
1295 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1296 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1297 Node* if_gt = generate_slow_guard(bol, nullptr);
1298 if (if_gt != nullptr) {
1299 phi->init_req(1, intcon(-1));
1300 region->init_req(1, if_gt);
1301 }
1302 if (!stopped()) {
1303 // Check for substr count == 0
1304 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1305 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1306 Node* if_zero = generate_slow_guard(bol, nullptr);
1307 if (if_zero != nullptr) {
1308 phi->init_req(2, intcon(0));
1309 region->init_req(2, if_zero);
1310 }
1311 }
1312 if (!stopped()) {
1313 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1314 }
1315 return nullptr;
1316 }
1317
1318 //-----------------------------inline_string_indexOfChar-----------------------
1319 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1320 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1321 return false;
1322 }
1323 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1324 return false;
1325 }
1326 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1327 Node* src = argument(0); // byte[]
1328 Node* int_ch = argument(1);
1329 Node* from_index = argument(2);
1330 Node* max = argument(3);
1331
1332 src = must_be_not_null(src, true);
1333
1334 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1335 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1336 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1337
1338 // Range checks
1339 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1340
1341 // Check for int_ch >= 0
1342 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1343 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1344 {
1345 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1346 uncommon_trap(Deoptimization::Reason_intrinsic,
1347 Deoptimization::Action_maybe_recompile);
1348 }
1349 if (stopped()) {
1350 return true;
1351 }
1352
1353 RegionNode* region = new RegionNode(3);
1354 Node* phi = new PhiNode(region, TypeInt::INT);
1355
1356 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1357 C->set_has_split_ifs(true); // Has chance for split-if optimization
1358 _gvn.transform(result);
1359
1360 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1361 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1362
1363 Node* if_lt = generate_slow_guard(bol, nullptr);
1364 if (if_lt != nullptr) {
1365 // result == -1
1366 phi->init_req(2, result);
1367 region->init_req(2, if_lt);
1368 }
1369 if (!stopped()) {
1370 result = _gvn.transform(new AddINode(result, from_index));
1371 phi->init_req(1, result);
1372 region->init_req(1, control());
1373 }
1374 set_control(_gvn.transform(region));
1375 record_for_igvn(region);
1376 set_result(_gvn.transform(phi));
1377
1378 return true;
1379 }
1380 //---------------------------inline_string_copy---------------------
1381 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1382 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1383 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1384 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1385 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1386 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1387 bool LibraryCallKit::inline_string_copy(bool compress) {
1388 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1389 return false;
1390 }
1391 int nargs = 5; // 2 oops, 3 ints
1392 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1393
1394 Node* src = argument(0);
1395 Node* src_offset = argument(1);
1396 Node* dst = argument(2);
1397 Node* dst_offset = argument(3);
1398 Node* length = argument(4);
1399
1400 // Check for allocation before we add nodes that would confuse
1401 // tightly_coupled_allocation()
1402 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1403
1404 // Figure out the size and type of the elements we will be copying.
1405 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1406 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1407 if (src_type == nullptr || dst_type == nullptr) {
1408 return false;
1409 }
1410 BasicType src_elem = src_type->elem()->array_element_basic_type();
1411 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1412 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1413 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1414 "Unsupported array types for inline_string_copy");
1415
1416 src = must_be_not_null(src, true);
1417 dst = must_be_not_null(dst, true);
1418
1419 // Convert char[] offsets to byte[] offsets
1420 bool convert_src = (compress && src_elem == T_BYTE);
1421 bool convert_dst = (!compress && dst_elem == T_BYTE);
1422 if (convert_src) {
1423 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1424 } else if (convert_dst) {
1425 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1426 }
1427
1428 // Range checks
1429 generate_string_range_check(src, src_offset, length, convert_src);
1430 generate_string_range_check(dst, dst_offset, length, convert_dst);
1431 if (stopped()) {
1432 return true;
1433 }
1434
1435 Node* src_start = array_element_address(src, src_offset, src_elem);
1436 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1437 // 'src_start' points to src array + scaled offset
1438 // 'dst_start' points to dst array + scaled offset
1439 Node* count = nullptr;
1440 if (compress) {
1441 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1442 } else {
1443 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1444 }
1445
1446 if (alloc != nullptr) {
1447 if (alloc->maybe_set_complete(&_gvn)) {
1448 // "You break it, you buy it."
1449 InitializeNode* init = alloc->initialization();
1450 assert(init->is_complete(), "we just did this");
1451 init->set_complete_with_arraycopy();
1452 assert(dst->is_CheckCastPP(), "sanity");
1453 assert(dst->in(0)->in(0) == init, "dest pinned");
1454 }
1455 // Do not let stores that initialize this object be reordered with
1456 // a subsequent store that would make this object accessible by
1457 // other threads.
1458 // Record what AllocateNode this StoreStore protects so that
1459 // escape analysis can go from the MemBarStoreStoreNode to the
1460 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1461 // based on the escape status of the AllocateNode.
1462 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1463 }
1464 if (compress) {
1465 set_result(_gvn.transform(count));
1466 }
1467 clear_upper_avx();
1468
1469 return true;
1470 }
1471
1472 #ifdef _LP64
1473 #define XTOP ,top() /*additional argument*/
1474 #else //_LP64
1475 #define XTOP /*no additional argument*/
1476 #endif //_LP64
1477
1478 //------------------------inline_string_toBytesU--------------------------
1479 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1480 bool LibraryCallKit::inline_string_toBytesU() {
1481 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1482 return false;
1483 }
1484 // Get the arguments.
1485 Node* value = argument(0);
1486 Node* offset = argument(1);
1487 Node* length = argument(2);
1488
1489 Node* newcopy = nullptr;
1490
1491 // Set the original stack and the reexecute bit for the interpreter to reexecute
1492 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1493 { PreserveReexecuteState preexecs(this);
1494 jvms()->set_should_reexecute(true);
1495
1496 // Check if a null path was taken unconditionally.
1497 value = null_check(value);
1498
1499 RegionNode* bailout = new RegionNode(1);
1500 record_for_igvn(bailout);
1501
1502 // Range checks
1503 generate_negative_guard(offset, bailout);
1504 generate_negative_guard(length, bailout);
1505 generate_limit_guard(offset, length, load_array_length(value), bailout);
1506 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1507 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1508
1509 if (bailout->req() > 1) {
1510 PreserveJVMState pjvms(this);
1511 set_control(_gvn.transform(bailout));
1512 uncommon_trap(Deoptimization::Reason_intrinsic,
1513 Deoptimization::Action_maybe_recompile);
1514 }
1515 if (stopped()) {
1516 return true;
1517 }
1518
1519 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1520 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1521 newcopy = new_array(klass_node, size, 0); // no arguments to push
1522 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1523 guarantee(alloc != nullptr, "created above");
1524
1525 // Calculate starting addresses.
1526 Node* src_start = array_element_address(value, offset, T_CHAR);
1527 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1528
1529 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1530 const TypeInt* toffset = gvn().type(offset)->is_int();
1531 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1532
1533 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1534 const char* copyfunc_name = "arraycopy";
1535 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1536 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1537 OptoRuntime::fast_arraycopy_Type(),
1538 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1539 src_start, dst_start, ConvI2X(length) XTOP);
1540 // Do not let reads from the cloned object float above the arraycopy.
1541 if (alloc->maybe_set_complete(&_gvn)) {
1542 // "You break it, you buy it."
1543 InitializeNode* init = alloc->initialization();
1544 assert(init->is_complete(), "we just did this");
1545 init->set_complete_with_arraycopy();
1546 assert(newcopy->is_CheckCastPP(), "sanity");
1547 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1548 }
1549 // Do not let stores that initialize this object be reordered with
1550 // a subsequent store that would make this object accessible by
1551 // other threads.
1552 // Record what AllocateNode this StoreStore protects so that
1553 // escape analysis can go from the MemBarStoreStoreNode to the
1554 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1555 // based on the escape status of the AllocateNode.
1556 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1557 } // original reexecute is set back here
1558
1559 C->set_has_split_ifs(true); // Has chance for split-if optimization
1560 if (!stopped()) {
1561 set_result(newcopy);
1562 }
1563 clear_upper_avx();
1564
1565 return true;
1566 }
1567
1568 //------------------------inline_string_getCharsU--------------------------
1569 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1570 bool LibraryCallKit::inline_string_getCharsU() {
1571 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1572 return false;
1573 }
1574
1575 // Get the arguments.
1576 Node* src = argument(0);
1577 Node* src_begin = argument(1);
1578 Node* src_end = argument(2); // exclusive offset (i < src_end)
1579 Node* dst = argument(3);
1580 Node* dst_begin = argument(4);
1581
1582 // Check for allocation before we add nodes that would confuse
1583 // tightly_coupled_allocation()
1584 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1585
1586 // Check if a null path was taken unconditionally.
1587 src = null_check(src);
1588 dst = null_check(dst);
1589 if (stopped()) {
1590 return true;
1591 }
1592
1593 // Get length and convert char[] offset to byte[] offset
1594 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1595 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1596
1597 // Range checks
1598 generate_string_range_check(src, src_begin, length, true);
1599 generate_string_range_check(dst, dst_begin, length, false);
1600 if (stopped()) {
1601 return true;
1602 }
1603
1604 if (!stopped()) {
1605 // Calculate starting addresses.
1606 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1607 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1608
1609 // Check if array addresses are aligned to HeapWordSize
1610 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1611 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1612 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1613 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1614
1615 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1616 const char* copyfunc_name = "arraycopy";
1617 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1618 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1619 OptoRuntime::fast_arraycopy_Type(),
1620 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1621 src_start, dst_start, ConvI2X(length) XTOP);
1622 // Do not let reads from the cloned object float above the arraycopy.
1623 if (alloc != nullptr) {
1624 if (alloc->maybe_set_complete(&_gvn)) {
1625 // "You break it, you buy it."
1626 InitializeNode* init = alloc->initialization();
1627 assert(init->is_complete(), "we just did this");
1628 init->set_complete_with_arraycopy();
1629 assert(dst->is_CheckCastPP(), "sanity");
1630 assert(dst->in(0)->in(0) == init, "dest pinned");
1631 }
1632 // Do not let stores that initialize this object be reordered with
1633 // a subsequent store that would make this object accessible by
1634 // other threads.
1635 // Record what AllocateNode this StoreStore protects so that
1636 // escape analysis can go from the MemBarStoreStoreNode to the
1637 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1638 // based on the escape status of the AllocateNode.
1639 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1640 } else {
1641 insert_mem_bar(Op_MemBarCPUOrder);
1642 }
1643 }
1644
1645 C->set_has_split_ifs(true); // Has chance for split-if optimization
1646 return true;
1647 }
1648
1649 //----------------------inline_string_char_access----------------------------
1650 // Store/Load char to/from byte[] array.
1651 // static void StringUTF16.putChar(byte[] val, int index, int c)
1652 // static char StringUTF16.getChar(byte[] val, int index)
1653 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1654 Node* value = argument(0);
1655 Node* index = argument(1);
1656 Node* ch = is_store ? argument(2) : nullptr;
1657
1658 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1659 // correctly requires matched array shapes.
1660 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1661 "sanity: byte[] and char[] bases agree");
1662 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1663 "sanity: byte[] and char[] scales agree");
1664
1665 // Bail when getChar over constants is requested: constant folding would
1666 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1667 // Java method would constant fold nicely instead.
1668 if (!is_store && value->is_Con() && index->is_Con()) {
1669 return false;
1670 }
1671
1672 // Save state and restore on bailout
1673 uint old_sp = sp();
1674 SafePointNode* old_map = clone_map();
1675
1676 value = must_be_not_null(value, true);
1677
1678 Node* adr = array_element_address(value, index, T_CHAR);
1679 if (adr->is_top()) {
1680 set_map(old_map);
1681 set_sp(old_sp);
1682 return false;
1683 }
1684 destruct_map_clone(old_map);
1685 if (is_store) {
1686 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1687 } else {
1688 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);
1689 set_result(ch);
1690 }
1691 return true;
1692 }
1693
1694 //--------------------------round_double_node--------------------------------
1695 // Round a double node if necessary.
1696 Node* LibraryCallKit::round_double_node(Node* n) {
1697 if (Matcher::strict_fp_requires_explicit_rounding) {
1698 #ifdef IA32
1699 if (UseSSE < 2) {
1700 n = _gvn.transform(new RoundDoubleNode(nullptr, n));
1701 }
1702 #else
1703 Unimplemented();
1704 #endif // IA32
1705 }
1706 return n;
1707 }
1708
1709 //------------------------------inline_math-----------------------------------
1710 // public static double Math.abs(double)
1711 // public static double Math.sqrt(double)
1712 // public static double Math.log(double)
1713 // public static double Math.log10(double)
1714 // public static double Math.round(double)
1715 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1716 Node* arg = round_double_node(argument(0));
1717 Node* n = nullptr;
1718 switch (id) {
1719 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1720 case vmIntrinsics::_dsqrt:
1721 case vmIntrinsics::_dsqrt_strict:
1722 n = new SqrtDNode(C, control(), arg); break;
1723 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1724 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1725 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1726 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1727 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1728 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1729 default: fatal_unexpected_iid(id); break;
1730 }
1731 set_result(_gvn.transform(n));
1732 return true;
1733 }
1734
1735 //------------------------------inline_math-----------------------------------
1736 // public static float Math.abs(float)
1737 // public static int Math.abs(int)
1738 // public static long Math.abs(long)
1739 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1740 Node* arg = argument(0);
1741 Node* n = nullptr;
1742 switch (id) {
1743 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1744 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1745 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1746 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1747 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1748 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1749 default: fatal_unexpected_iid(id); break;
1750 }
1751 set_result(_gvn.transform(n));
1752 return true;
1753 }
1754
1755 //------------------------------runtime_math-----------------------------
1756 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1757 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1758 "must be (DD)D or (D)D type");
1759
1760 // Inputs
1761 Node* a = round_double_node(argument(0));
1762 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : nullptr;
1763
1764 const TypePtr* no_memory_effects = nullptr;
1765 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1766 no_memory_effects,
1767 a, top(), b, b ? top() : nullptr);
1768 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1769 #ifdef ASSERT
1770 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1771 assert(value_top == top(), "second value must be top");
1772 #endif
1773
1774 set_result(value);
1775 return true;
1776 }
1777
1778 //------------------------------inline_math_pow-----------------------------
1779 bool LibraryCallKit::inline_math_pow() {
1780 Node* exp = round_double_node(argument(2));
1781 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1782 if (d != nullptr) {
1783 if (d->getd() == 2.0) {
1784 // Special case: pow(x, 2.0) => x * x
1785 Node* base = round_double_node(argument(0));
1786 set_result(_gvn.transform(new MulDNode(base, base)));
1787 return true;
1788 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1789 // Special case: pow(x, 0.5) => sqrt(x)
1790 Node* base = round_double_node(argument(0));
1791 Node* zero = _gvn.zerocon(T_DOUBLE);
1792
1793 RegionNode* region = new RegionNode(3);
1794 Node* phi = new PhiNode(region, Type::DOUBLE);
1795
1796 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1797 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1798 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1799 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1800 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1801
1802 Node* if_pow = generate_slow_guard(test, nullptr);
1803 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1804 phi->init_req(1, value_sqrt);
1805 region->init_req(1, control());
1806
1807 if (if_pow != nullptr) {
1808 set_control(if_pow);
1809 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1810 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1811 const TypePtr* no_memory_effects = nullptr;
1812 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1813 no_memory_effects, base, top(), exp, top());
1814 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1815 #ifdef ASSERT
1816 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1817 assert(value_top == top(), "second value must be top");
1818 #endif
1819 phi->init_req(2, value_pow);
1820 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1821 }
1822
1823 C->set_has_split_ifs(true); // Has chance for split-if optimization
1824 set_control(_gvn.transform(region));
1825 record_for_igvn(region);
1826 set_result(_gvn.transform(phi));
1827
1828 return true;
1829 }
1830 }
1831
1832 return StubRoutines::dpow() != nullptr ?
1833 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1834 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1835 }
1836
1837 //------------------------------inline_math_native-----------------------------
1838 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1839 switch (id) {
1840 case vmIntrinsics::_dsin:
1841 return StubRoutines::dsin() != nullptr ?
1842 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1843 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1844 case vmIntrinsics::_dcos:
1845 return StubRoutines::dcos() != nullptr ?
1846 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1847 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1848 case vmIntrinsics::_dtan:
1849 return StubRoutines::dtan() != nullptr ?
1850 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1851 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1852 case vmIntrinsics::_dexp:
1853 return StubRoutines::dexp() != nullptr ?
1854 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1855 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1856 case vmIntrinsics::_dlog:
1857 return StubRoutines::dlog() != nullptr ?
1858 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1859 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1860 case vmIntrinsics::_dlog10:
1861 return StubRoutines::dlog10() != nullptr ?
1862 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1863 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1864
1865 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1866 case vmIntrinsics::_ceil:
1867 case vmIntrinsics::_floor:
1868 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1869
1870 case vmIntrinsics::_dsqrt:
1871 case vmIntrinsics::_dsqrt_strict:
1872 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1873 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1874 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1875 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1876 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1877
1878 case vmIntrinsics::_dpow: return inline_math_pow();
1879 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1880 case vmIntrinsics::_fcopySign: return inline_math(id);
1881 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1882 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1883 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1884
1885 // These intrinsics are not yet correctly implemented
1886 case vmIntrinsics::_datan2:
1887 return false;
1888
1889 default:
1890 fatal_unexpected_iid(id);
1891 return false;
1892 }
1893 }
1894
1895 //----------------------------inline_notify-----------------------------------*
1896 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1897 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1898 address func;
1899 if (id == vmIntrinsics::_notify) {
1900 func = OptoRuntime::monitor_notify_Java();
1901 } else {
1902 func = OptoRuntime::monitor_notifyAll_Java();
1903 }
1904 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1905 make_slow_call_ex(call, env()->Throwable_klass(), false);
1906 return true;
1907 }
1908
1909
1910 //----------------------------inline_min_max-----------------------------------
1911 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1912 set_result(generate_min_max(id, argument(0), argument(1)));
1913 return true;
1914 }
1915
1916 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1917 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1918 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1919 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1920 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1921
1922 {
1923 PreserveJVMState pjvms(this);
1924 PreserveReexecuteState preexecs(this);
1925 jvms()->set_should_reexecute(true);
1926
1927 set_control(slow_path);
1928 set_i_o(i_o());
1929
1930 uncommon_trap(Deoptimization::Reason_intrinsic,
1931 Deoptimization::Action_none);
1932 }
1933
1934 set_control(fast_path);
1935 set_result(math);
1936 }
1937
1938 template <typename OverflowOp>
1939 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1940 typedef typename OverflowOp::MathOp MathOp;
1941
1942 MathOp* mathOp = new MathOp(arg1, arg2);
1943 Node* operation = _gvn.transform( mathOp );
1944 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1945 inline_math_mathExact(operation, ofcheck);
1946 return true;
1947 }
1948
1949 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1950 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1951 }
1952
1953 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1954 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1955 }
1956
1957 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1958 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1959 }
1960
1961 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1962 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1963 }
1964
1965 bool LibraryCallKit::inline_math_negateExactI() {
1966 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1967 }
1968
1969 bool LibraryCallKit::inline_math_negateExactL() {
1970 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1971 }
1972
1973 bool LibraryCallKit::inline_math_multiplyExactI() {
1974 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1975 }
1976
1977 bool LibraryCallKit::inline_math_multiplyExactL() {
1978 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1979 }
1980
1981 bool LibraryCallKit::inline_math_multiplyHigh() {
1982 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
1983 return true;
1984 }
1985
1986 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
1987 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
1988 return true;
1989 }
1990
1991 Node*
1992 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1993 Node* result_val = nullptr;
1994 switch (id) {
1995 case vmIntrinsics::_min:
1996 case vmIntrinsics::_min_strict:
1997 result_val = _gvn.transform(new MinINode(x0, y0));
1998 break;
1999 case vmIntrinsics::_max:
2000 case vmIntrinsics::_max_strict:
2001 result_val = _gvn.transform(new MaxINode(x0, y0));
2002 break;
2003 default:
2004 fatal_unexpected_iid(id);
2005 break;
2006 }
2007 return result_val;
2008 }
2009
2010 inline int
2011 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2012 const TypePtr* base_type = TypePtr::NULL_PTR;
2013 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2014 if (base_type == nullptr) {
2015 // Unknown type.
2016 return Type::AnyPtr;
2017 } else if (base_type == TypePtr::NULL_PTR) {
2018 // Since this is a null+long form, we have to switch to a rawptr.
2019 base = _gvn.transform(new CastX2PNode(offset));
2020 offset = MakeConX(0);
2021 return Type::RawPtr;
2022 } else if (base_type->base() == Type::RawPtr) {
2023 return Type::RawPtr;
2024 } else if (base_type->isa_oopptr()) {
2025 // Base is never null => always a heap address.
2026 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2027 return Type::OopPtr;
2028 }
2029 // Offset is small => always a heap address.
2030 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2031 if (offset_type != nullptr &&
2032 base_type->offset() == 0 && // (should always be?)
2033 offset_type->_lo >= 0 &&
2034 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2035 return Type::OopPtr;
2036 } else if (type == T_OBJECT) {
2037 // off heap access to an oop doesn't make any sense. Has to be on
2038 // heap.
2039 return Type::OopPtr;
2040 }
2041 // Otherwise, it might either be oop+off or null+addr.
2042 return Type::AnyPtr;
2043 } else {
2044 // No information:
2045 return Type::AnyPtr;
2046 }
2047 }
2048
2049 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2050 Node* uncasted_base = base;
2051 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2052 if (kind == Type::RawPtr) {
2053 return basic_plus_adr(top(), uncasted_base, offset);
2054 } else if (kind == Type::AnyPtr) {
2055 assert(base == uncasted_base, "unexpected base change");
2056 if (can_cast) {
2057 if (!_gvn.type(base)->speculative_maybe_null() &&
2058 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2059 // According to profiling, this access is always on
2060 // heap. Casting the base to not null and thus avoiding membars
2061 // around the access should allow better optimizations
2062 Node* null_ctl = top();
2063 base = null_check_oop(base, &null_ctl, true, true, true);
2064 assert(null_ctl->is_top(), "no null control here");
2065 return basic_plus_adr(base, offset);
2066 } else if (_gvn.type(base)->speculative_always_null() &&
2067 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2068 // According to profiling, this access is always off
2069 // heap.
2070 base = null_assert(base);
2071 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2072 offset = MakeConX(0);
2073 return basic_plus_adr(top(), raw_base, offset);
2074 }
2075 }
2076 // We don't know if it's an on heap or off heap access. Fall back
2077 // to raw memory access.
2078 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2079 return basic_plus_adr(top(), raw, offset);
2080 } else {
2081 assert(base == uncasted_base, "unexpected base change");
2082 // We know it's an on heap access so base can't be null
2083 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2084 base = must_be_not_null(base, true);
2085 }
2086 return basic_plus_adr(base, offset);
2087 }
2088 }
2089
2090 //--------------------------inline_number_methods-----------------------------
2091 // inline int Integer.numberOfLeadingZeros(int)
2092 // inline int Long.numberOfLeadingZeros(long)
2093 //
2094 // inline int Integer.numberOfTrailingZeros(int)
2095 // inline int Long.numberOfTrailingZeros(long)
2096 //
2097 // inline int Integer.bitCount(int)
2098 // inline int Long.bitCount(long)
2099 //
2100 // inline char Character.reverseBytes(char)
2101 // inline short Short.reverseBytes(short)
2102 // inline int Integer.reverseBytes(int)
2103 // inline long Long.reverseBytes(long)
2104 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2105 Node* arg = argument(0);
2106 Node* n = nullptr;
2107 switch (id) {
2108 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2109 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2110 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2111 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2112 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2113 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2114 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2115 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2116 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2117 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2118 case vmIntrinsics::_reverse_i: n = new ReverseINode(0, arg); break;
2119 case vmIntrinsics::_reverse_l: n = new ReverseLNode(0, arg); break;
2120 default: fatal_unexpected_iid(id); break;
2121 }
2122 set_result(_gvn.transform(n));
2123 return true;
2124 }
2125
2126 //--------------------------inline_bitshuffle_methods-----------------------------
2127 // inline int Integer.compress(int, int)
2128 // inline int Integer.expand(int, int)
2129 // inline long Long.compress(long, long)
2130 // inline long Long.expand(long, long)
2131 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2132 Node* n = nullptr;
2133 switch (id) {
2134 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2135 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2136 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2137 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2138 default: fatal_unexpected_iid(id); break;
2139 }
2140 set_result(_gvn.transform(n));
2141 return true;
2142 }
2143
2144 //--------------------------inline_number_methods-----------------------------
2145 // inline int Integer.compareUnsigned(int, int)
2146 // inline int Long.compareUnsigned(long, long)
2147 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2148 Node* arg1 = argument(0);
2149 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2150 Node* n = nullptr;
2151 switch (id) {
2152 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2153 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2154 default: fatal_unexpected_iid(id); break;
2155 }
2156 set_result(_gvn.transform(n));
2157 return true;
2158 }
2159
2160 //--------------------------inline_unsigned_divmod_methods-----------------------------
2161 // inline int Integer.divideUnsigned(int, int)
2162 // inline int Integer.remainderUnsigned(int, int)
2163 // inline long Long.divideUnsigned(long, long)
2164 // inline long Long.remainderUnsigned(long, long)
2165 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2166 Node* n = nullptr;
2167 switch (id) {
2168 case vmIntrinsics::_divideUnsigned_i: {
2169 zero_check_int(argument(1));
2170 // Compile-time detect of null-exception
2171 if (stopped()) {
2172 return true; // keep the graph constructed so far
2173 }
2174 n = new UDivINode(control(), argument(0), argument(1));
2175 break;
2176 }
2177 case vmIntrinsics::_divideUnsigned_l: {
2178 zero_check_long(argument(2));
2179 // Compile-time detect of null-exception
2180 if (stopped()) {
2181 return true; // keep the graph constructed so far
2182 }
2183 n = new UDivLNode(control(), argument(0), argument(2));
2184 break;
2185 }
2186 case vmIntrinsics::_remainderUnsigned_i: {
2187 zero_check_int(argument(1));
2188 // Compile-time detect of null-exception
2189 if (stopped()) {
2190 return true; // keep the graph constructed so far
2191 }
2192 n = new UModINode(control(), argument(0), argument(1));
2193 break;
2194 }
2195 case vmIntrinsics::_remainderUnsigned_l: {
2196 zero_check_long(argument(2));
2197 // Compile-time detect of null-exception
2198 if (stopped()) {
2199 return true; // keep the graph constructed so far
2200 }
2201 n = new UModLNode(control(), argument(0), argument(2));
2202 break;
2203 }
2204 default: fatal_unexpected_iid(id); break;
2205 }
2206 set_result(_gvn.transform(n));
2207 return true;
2208 }
2209
2210 //----------------------------inline_unsafe_access----------------------------
2211
2212 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2213 // Attempt to infer a sharper value type from the offset and base type.
2214 ciKlass* sharpened_klass = nullptr;
2215 bool null_free = false;
2216
2217 // See if it is an instance field, with an object type.
2218 if (alias_type->field() != nullptr) {
2219 if (alias_type->field()->type()->is_klass()) {
2220 sharpened_klass = alias_type->field()->type()->as_klass();
2221 null_free = alias_type->field()->is_null_free();
2222 }
2223 }
2224
2225 const TypeOopPtr* result = nullptr;
2226 // See if it is a narrow oop array.
2227 if (adr_type->isa_aryptr()) {
2228 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2229 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2230 null_free = adr_type->is_aryptr()->is_null_free();
2231 if (elem_type != nullptr && elem_type->is_loaded()) {
2232 // Sharpen the value type.
2233 result = elem_type;
2234 }
2235 }
2236 }
2237
2238 // The sharpened class might be unloaded if there is no class loader
2239 // contraint in place.
2240 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2241 // Sharpen the value type.
2242 result = TypeOopPtr::make_from_klass(sharpened_klass);
2243 if (null_free) {
2244 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2245 }
2246 }
2247 if (result != nullptr) {
2248 #ifndef PRODUCT
2249 if (C->print_intrinsics() || C->print_inlining()) {
2250 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2251 tty->print(" sharpened value: "); result->dump(); tty->cr();
2252 }
2253 #endif
2254 }
2255 return result;
2256 }
2257
2258 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2259 switch (kind) {
2260 case Relaxed:
2261 return MO_UNORDERED;
2262 case Opaque:
2263 return MO_RELAXED;
2264 case Acquire:
2265 return MO_ACQUIRE;
2266 case Release:
2267 return MO_RELEASE;
2268 case Volatile:
2269 return MO_SEQ_CST;
2270 default:
2271 ShouldNotReachHere();
2272 return 0;
2273 }
2274 }
2275
2276 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned, const bool is_flat) {
2277 if (callee()->is_static()) return false; // caller must have the capability!
2278 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2279 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2280 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2281 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2282
2283 if (is_reference_type(type)) {
2284 decorators |= ON_UNKNOWN_OOP_REF;
2285 }
2286
2287 if (unaligned) {
2288 decorators |= C2_UNALIGNED;
2289 }
2290
2291 #ifndef PRODUCT
2292 {
2293 ResourceMark rm;
2294 // Check the signatures.
2295 ciSignature* sig = callee()->signature();
2296 #ifdef ASSERT
2297 if (!is_store) {
2298 // Object getReference(Object base, int/long offset), etc.
2299 BasicType rtype = sig->return_type()->basic_type();
2300 assert(rtype == type, "getter must return the expected value");
2301 assert(sig->count() == 2 || (is_flat && sig->count() == 3), "oop getter has 2 or 3 arguments");
2302 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2303 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2304 } else {
2305 // void putReference(Object base, int/long offset, Object x), etc.
2306 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2307 assert(sig->count() == 3 || (is_flat && sig->count() == 4), "oop putter has 3 arguments");
2308 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2309 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2310 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2311 assert(vtype == type, "putter must accept the expected value");
2312 }
2313 #endif // ASSERT
2314 }
2315 #endif //PRODUCT
2316
2317 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2318
2319 Node* receiver = argument(0); // type: oop
2320
2321 // Build address expression.
2322 Node* heap_base_oop = top();
2323
2324 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2325 Node* base = argument(1); // type: oop
2326 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2327 Node* offset = argument(2); // type: long
2328 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2329 // to be plain byte offsets, which are also the same as those accepted
2330 // by oopDesc::field_addr.
2331 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2332 "fieldOffset must be byte-scaled");
2333
2334 ciInlineKlass* inline_klass = nullptr;
2335 if (is_flat) {
2336 const TypeInstPtr* cls = _gvn.type(argument(4))->isa_instptr();
2337 if (cls == nullptr || cls->const_oop() == nullptr) {
2338 return false;
2339 }
2340 ciType* mirror_type = cls->const_oop()->as_instance()->java_mirror_type();
2341 if (!mirror_type->is_inlinetype()) {
2342 return false;
2343 }
2344 inline_klass = mirror_type->as_inline_klass();
2345 }
2346
2347 if (base->is_InlineType()) {
2348 InlineTypeNode* vt = base->as_InlineType();
2349 if (is_store) {
2350 if (!vt->is_allocated(&_gvn)) {
2351 return false;
2352 }
2353 base = vt->get_oop();
2354 } else {
2355 if (offset->is_Con()) {
2356 long off = find_long_con(offset, 0);
2357 ciInlineKlass* vk = vt->type()->inline_klass();
2358 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2359 return false;
2360 }
2361
2362 ciField* field = vk->get_non_flat_field_by_offset(off);
2363 if (field != nullptr) {
2364 BasicType bt = type2field[field->type()->basic_type()];
2365 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2366 bt = T_OBJECT;
2367 }
2368 if (bt == type && (!field->is_flat() || field->type() == inline_klass)) {
2369 Node* value = vt->field_value_by_offset(off, false);
2370 if (value->is_InlineType()) {
2371 value = value->as_InlineType()->adjust_scalarization_depth(this);
2372 }
2373 set_result(value);
2374 return true;
2375 }
2376 }
2377 }
2378 {
2379 // Re-execute the unsafe access if allocation triggers deoptimization.
2380 PreserveReexecuteState preexecs(this);
2381 jvms()->set_should_reexecute(true);
2382 vt = vt->buffer(this);
2383 }
2384 base = vt->get_oop();
2385 }
2386 }
2387
2388 // 32-bit machines ignore the high half!
2389 offset = ConvL2X(offset);
2390
2391 // Save state and restore on bailout
2392 uint old_sp = sp();
2393 SafePointNode* old_map = clone_map();
2394
2395 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2396
2397 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2398 if (type != T_OBJECT && (inline_klass == nullptr || !inline_klass->has_object_fields())) {
2399 decorators |= IN_NATIVE; // off-heap primitive access
2400 } else {
2401 set_map(old_map);
2402 set_sp(old_sp);
2403 return false; // off-heap oop accesses are not supported
2404 }
2405 } else {
2406 heap_base_oop = base; // on-heap or mixed access
2407 }
2408
2409 // Can base be null? Otherwise, always on-heap access.
2410 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2411
2412 if (!can_access_non_heap) {
2413 decorators |= IN_HEAP;
2414 }
2415
2416 Node* val = is_store ? argument(4 + (is_flat ? 1 : 0)) : nullptr;
2417
2418 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2419 if (adr_type == TypePtr::NULL_PTR) {
2420 set_map(old_map);
2421 set_sp(old_sp);
2422 return false; // off-heap access with zero address
2423 }
2424
2425 // Try to categorize the address.
2426 Compile::AliasType* alias_type = C->alias_type(adr_type);
2427 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2428
2429 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2430 alias_type->adr_type() == TypeAryPtr::RANGE) {
2431 set_map(old_map);
2432 set_sp(old_sp);
2433 return false; // not supported
2434 }
2435
2436 bool mismatched = false;
2437 BasicType bt = T_ILLEGAL;
2438 ciField* field = nullptr;
2439 if (adr_type->isa_instptr()) {
2440 const TypeInstPtr* instptr = adr_type->is_instptr();
2441 ciInstanceKlass* k = instptr->instance_klass();
2442 int off = instptr->offset();
2443 if (instptr->const_oop() != nullptr &&
2444 k == ciEnv::current()->Class_klass() &&
2445 instptr->offset() >= (k->size_helper() * wordSize)) {
2446 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2447 field = k->get_field_by_offset(off, true);
2448 } else {
2449 field = k->get_non_flat_field_by_offset(off);
2450 }
2451 if (field != nullptr) {
2452 bt = type2field[field->type()->basic_type()];
2453 }
2454 assert(bt == alias_type->basic_type() || is_flat, "should match");
2455 } else {
2456 bt = alias_type->basic_type();
2457 }
2458
2459 if (bt != T_ILLEGAL) {
2460 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2461 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2462 // Alias type doesn't differentiate between byte[] and boolean[]).
2463 // Use address type to get the element type.
2464 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2465 }
2466 if (is_reference_type(bt, true)) {
2467 // accessing an array field with getReference is not a mismatch
2468 bt = T_OBJECT;
2469 }
2470 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2471 // Don't intrinsify mismatched object accesses
2472 set_map(old_map);
2473 set_sp(old_sp);
2474 return false;
2475 }
2476 mismatched = (bt != type);
2477 } else if (alias_type->adr_type()->isa_oopptr()) {
2478 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2479 }
2480
2481 if (is_flat) {
2482 if (adr_type->isa_instptr()) {
2483 if (field == nullptr || field->type() != inline_klass) {
2484 mismatched = true;
2485 }
2486 } else if (adr_type->isa_aryptr()) {
2487 const Type* elem = adr_type->is_aryptr()->elem();
2488 if (!adr_type->is_flat() || elem->inline_klass() != inline_klass) {
2489 mismatched = true;
2490 }
2491 } else {
2492 mismatched = true;
2493 }
2494 if (is_store) {
2495 const Type* val_t = _gvn.type(val);
2496 if (!val_t->is_inlinetypeptr() || val_t->inline_klass() != inline_klass) {
2497 set_map(old_map);
2498 set_sp(old_sp);
2499 return false;
2500 }
2501 }
2502 }
2503
2504 destruct_map_clone(old_map);
2505 assert(!mismatched || is_flat || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2506
2507 if (mismatched) {
2508 decorators |= C2_MISMATCHED;
2509 }
2510
2511 // First guess at the value type.
2512 const Type *value_type = Type::get_const_basic_type(type);
2513
2514 // Figure out the memory ordering.
2515 decorators |= mo_decorator_for_access_kind(kind);
2516
2517 if (!is_store) {
2518 if (type == T_OBJECT && !is_flat) {
2519 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2520 if (tjp != nullptr) {
2521 value_type = tjp;
2522 }
2523 }
2524 }
2525
2526 receiver = null_check(receiver);
2527 if (stopped()) {
2528 return true;
2529 }
2530 // Heap pointers get a null-check from the interpreter,
2531 // as a courtesy. However, this is not guaranteed by Unsafe,
2532 // and it is not possible to fully distinguish unintended nulls
2533 // from intended ones in this API.
2534
2535 if (!is_store) {
2536 Node* p = nullptr;
2537 // Try to constant fold a load from a constant field
2538
2539 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) {
2540 // final or stable field
2541 p = make_constant_from_field(field, heap_base_oop);
2542 }
2543
2544 if (p == nullptr) { // Could not constant fold the load
2545 if (is_flat) {
2546 if (adr_type->isa_instptr() && !mismatched) {
2547 ciInstanceKlass* holder = adr_type->is_instptr()->instance_klass();
2548 int offset = adr_type->is_instptr()->offset();
2549 p = InlineTypeNode::make_from_flat(this, inline_klass, base, base, holder, offset, decorators);
2550 } else {
2551 p = InlineTypeNode::make_from_flat(this, inline_klass, base, adr, nullptr, 0, decorators);
2552 }
2553 } else {
2554 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2555 const TypeOopPtr* ptr = value_type->make_oopptr();
2556 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2557 // Load a non-flattened inline type from memory
2558 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass(), !ptr->maybe_null());
2559 }
2560 }
2561 // Normalize the value returned by getBoolean in the following cases
2562 if (type == T_BOOLEAN &&
2563 (mismatched ||
2564 heap_base_oop == top() || // - heap_base_oop is null or
2565 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2566 // and the unsafe access is made to large offset
2567 // (i.e., larger than the maximum offset necessary for any
2568 // field access)
2569 ) {
2570 IdealKit ideal = IdealKit(this);
2571 #define __ ideal.
2572 IdealVariable normalized_result(ideal);
2573 __ declarations_done();
2574 __ set(normalized_result, p);
2575 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2576 __ set(normalized_result, ideal.ConI(1));
2577 ideal.end_if();
2578 final_sync(ideal);
2579 p = __ value(normalized_result);
2580 #undef __
2581 }
2582 }
2583 if (type == T_ADDRESS) {
2584 p = gvn().transform(new CastP2XNode(nullptr, p));
2585 p = ConvX2UL(p);
2586 }
2587 // The load node has the control of the preceding MemBarCPUOrder. All
2588 // following nodes will have the control of the MemBarCPUOrder inserted at
2589 // the end of this method. So, pushing the load onto the stack at a later
2590 // point is fine.
2591 set_result(p);
2592 } else {
2593 if (bt == T_ADDRESS) {
2594 // Repackage the long as a pointer.
2595 val = ConvL2X(val);
2596 val = gvn().transform(new CastX2PNode(val));
2597 }
2598 if (is_flat) {
2599 if (adr_type->isa_instptr() && !mismatched) {
2600 ciInstanceKlass* holder = adr_type->is_instptr()->instance_klass();
2601 int offset = adr_type->is_instptr()->offset();
2602 val->as_InlineType()->store_flat(this, base, base, holder, offset, decorators);
2603 } else {
2604 val->as_InlineType()->store_flat(this, base, adr, nullptr, 0, decorators);
2605 }
2606 } else {
2607 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2608 }
2609 }
2610
2611 if (argument(1)->is_InlineType() && is_store) {
2612 InlineTypeNode* value = InlineTypeNode::make_from_oop(this, base, _gvn.type(argument(1))->inline_klass());
2613 value = value->make_larval(this, false);
2614 replace_in_map(argument(1), value);
2615 }
2616
2617 return true;
2618 }
2619
2620 bool LibraryCallKit::inline_unsafe_make_private_buffer() {
2621 Node* receiver = argument(0);
2622 Node* value = argument(1);
2623 if (!value->is_InlineType()) {
2624 return false;
2625 }
2626
2627 receiver = null_check(receiver);
2628 if (stopped()) {
2629 return true;
2630 }
2631
2632 set_result(value->as_InlineType()->make_larval(this, true));
2633 return true;
2634 }
2635
2636 bool LibraryCallKit::inline_unsafe_finish_private_buffer() {
2637 Node* receiver = argument(0);
2638 Node* buffer = argument(1);
2639 if (!buffer->is_InlineType()) {
2640 return false;
2641 }
2642 InlineTypeNode* vt = buffer->as_InlineType();
2643 if (!vt->is_allocated(&_gvn)) {
2644 return false;
2645 }
2646 // TODO 8239003 Why is this needed?
2647 if (AllocateNode::Ideal_allocation(vt->get_oop()) == nullptr) {
2648 return false;
2649 }
2650
2651 receiver = null_check(receiver);
2652 if (stopped()) {
2653 return true;
2654 }
2655
2656 set_result(vt->finish_larval(this));
2657 return true;
2658 }
2659
2660 //----------------------------inline_unsafe_load_store----------------------------
2661 // This method serves a couple of different customers (depending on LoadStoreKind):
2662 //
2663 // LS_cmp_swap:
2664 //
2665 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2666 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2667 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2668 //
2669 // LS_cmp_swap_weak:
2670 //
2671 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2672 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2673 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2674 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2675 //
2676 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2677 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2678 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2679 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2680 //
2681 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2682 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2683 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2684 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2685 //
2686 // LS_cmp_exchange:
2687 //
2688 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2689 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2690 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2691 //
2692 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2693 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2694 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2695 //
2696 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2697 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2698 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2699 //
2700 // LS_get_add:
2701 //
2702 // int getAndAddInt( Object o, long offset, int delta)
2703 // long getAndAddLong(Object o, long offset, long delta)
2704 //
2705 // LS_get_set:
2706 //
2707 // int getAndSet(Object o, long offset, int newValue)
2708 // long getAndSet(Object o, long offset, long newValue)
2709 // Object getAndSet(Object o, long offset, Object newValue)
2710 //
2711 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2712 // This basic scheme here is the same as inline_unsafe_access, but
2713 // differs in enough details that combining them would make the code
2714 // overly confusing. (This is a true fact! I originally combined
2715 // them, but even I was confused by it!) As much code/comments as
2716 // possible are retained from inline_unsafe_access though to make
2717 // the correspondences clearer. - dl
2718
2719 if (callee()->is_static()) return false; // caller must have the capability!
2720
2721 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2722 decorators |= mo_decorator_for_access_kind(access_kind);
2723
2724 #ifndef PRODUCT
2725 BasicType rtype;
2726 {
2727 ResourceMark rm;
2728 // Check the signatures.
2729 ciSignature* sig = callee()->signature();
2730 rtype = sig->return_type()->basic_type();
2731 switch(kind) {
2732 case LS_get_add:
2733 case LS_get_set: {
2734 // Check the signatures.
2735 #ifdef ASSERT
2736 assert(rtype == type, "get and set must return the expected type");
2737 assert(sig->count() == 3, "get and set has 3 arguments");
2738 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2739 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2740 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2741 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2742 #endif // ASSERT
2743 break;
2744 }
2745 case LS_cmp_swap:
2746 case LS_cmp_swap_weak: {
2747 // Check the signatures.
2748 #ifdef ASSERT
2749 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2750 assert(sig->count() == 4, "CAS has 4 arguments");
2751 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2752 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2753 #endif // ASSERT
2754 break;
2755 }
2756 case LS_cmp_exchange: {
2757 // Check the signatures.
2758 #ifdef ASSERT
2759 assert(rtype == type, "CAS must return the expected type");
2760 assert(sig->count() == 4, "CAS has 4 arguments");
2761 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2762 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2763 #endif // ASSERT
2764 break;
2765 }
2766 default:
2767 ShouldNotReachHere();
2768 }
2769 }
2770 #endif //PRODUCT
2771
2772 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2773
2774 // Get arguments:
2775 Node* receiver = nullptr;
2776 Node* base = nullptr;
2777 Node* offset = nullptr;
2778 Node* oldval = nullptr;
2779 Node* newval = nullptr;
2780 switch(kind) {
2781 case LS_cmp_swap:
2782 case LS_cmp_swap_weak:
2783 case LS_cmp_exchange: {
2784 const bool two_slot_type = type2size[type] == 2;
2785 receiver = argument(0); // type: oop
2786 base = argument(1); // type: oop
2787 offset = argument(2); // type: long
2788 oldval = argument(4); // type: oop, int, or long
2789 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2790 break;
2791 }
2792 case LS_get_add:
2793 case LS_get_set: {
2794 receiver = argument(0); // type: oop
2795 base = argument(1); // type: oop
2796 offset = argument(2); // type: long
2797 oldval = nullptr;
2798 newval = argument(4); // type: oop, int, or long
2799 break;
2800 }
2801 default:
2802 ShouldNotReachHere();
2803 }
2804
2805 // Build field offset expression.
2806 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2807 // to be plain byte offsets, which are also the same as those accepted
2808 // by oopDesc::field_addr.
2809 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2810 // 32-bit machines ignore the high half of long offsets
2811 offset = ConvL2X(offset);
2812 // Save state and restore on bailout
2813 uint old_sp = sp();
2814 SafePointNode* old_map = clone_map();
2815 Node* adr = make_unsafe_address(base, offset,type, false);
2816 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2817
2818 Compile::AliasType* alias_type = C->alias_type(adr_type);
2819 BasicType bt = alias_type->basic_type();
2820 if (bt != T_ILLEGAL &&
2821 (is_reference_type(bt) != (type == T_OBJECT))) {
2822 // Don't intrinsify mismatched object accesses.
2823 set_map(old_map);
2824 set_sp(old_sp);
2825 return false;
2826 }
2827
2828 destruct_map_clone(old_map);
2829
2830 // For CAS, unlike inline_unsafe_access, there seems no point in
2831 // trying to refine types. Just use the coarse types here.
2832 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2833 const Type *value_type = Type::get_const_basic_type(type);
2834
2835 switch (kind) {
2836 case LS_get_set:
2837 case LS_cmp_exchange: {
2838 if (type == T_OBJECT) {
2839 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2840 if (tjp != nullptr) {
2841 value_type = tjp;
2842 }
2843 }
2844 break;
2845 }
2846 case LS_cmp_swap:
2847 case LS_cmp_swap_weak:
2848 case LS_get_add:
2849 break;
2850 default:
2851 ShouldNotReachHere();
2852 }
2853
2854 // Null check receiver.
2855 receiver = null_check(receiver);
2856 if (stopped()) {
2857 return true;
2858 }
2859
2860 int alias_idx = C->get_alias_index(adr_type);
2861
2862 if (is_reference_type(type)) {
2863 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2864
2865 if (oldval != nullptr && oldval->is_InlineType()) {
2866 // Re-execute the unsafe access if allocation triggers deoptimization.
2867 PreserveReexecuteState preexecs(this);
2868 jvms()->set_should_reexecute(true);
2869 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
2870 }
2871 if (newval != nullptr && newval->is_InlineType()) {
2872 // Re-execute the unsafe access if allocation triggers deoptimization.
2873 PreserveReexecuteState preexecs(this);
2874 jvms()->set_should_reexecute(true);
2875 newval = newval->as_InlineType()->buffer(this)->get_oop();
2876 }
2877
2878 // Transformation of a value which could be null pointer (CastPP #null)
2879 // could be delayed during Parse (for example, in adjust_map_after_if()).
2880 // Execute transformation here to avoid barrier generation in such case.
2881 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2882 newval = _gvn.makecon(TypePtr::NULL_PTR);
2883
2884 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2885 // Refine the value to a null constant, when it is known to be null
2886 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2887 }
2888 }
2889
2890 Node* result = nullptr;
2891 switch (kind) {
2892 case LS_cmp_exchange: {
2893 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2894 oldval, newval, value_type, type, decorators);
2895 break;
2896 }
2897 case LS_cmp_swap_weak:
2898 decorators |= C2_WEAK_CMPXCHG;
2899 case LS_cmp_swap: {
2900 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2901 oldval, newval, value_type, type, decorators);
2902 break;
2903 }
2904 case LS_get_set: {
2905 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2906 newval, value_type, type, decorators);
2907 break;
2908 }
2909 case LS_get_add: {
2910 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2911 newval, value_type, type, decorators);
2912 break;
2913 }
2914 default:
2915 ShouldNotReachHere();
2916 }
2917
2918 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2919 set_result(result);
2920 return true;
2921 }
2922
2923 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2924 // Regardless of form, don't allow previous ld/st to move down,
2925 // then issue acquire, release, or volatile mem_bar.
2926 insert_mem_bar(Op_MemBarCPUOrder);
2927 switch(id) {
2928 case vmIntrinsics::_loadFence:
2929 insert_mem_bar(Op_LoadFence);
2930 return true;
2931 case vmIntrinsics::_storeFence:
2932 insert_mem_bar(Op_StoreFence);
2933 return true;
2934 case vmIntrinsics::_storeStoreFence:
2935 insert_mem_bar(Op_StoreStoreFence);
2936 return true;
2937 case vmIntrinsics::_fullFence:
2938 insert_mem_bar(Op_MemBarVolatile);
2939 return true;
2940 default:
2941 fatal_unexpected_iid(id);
2942 return false;
2943 }
2944 }
2945
2946 bool LibraryCallKit::inline_onspinwait() {
2947 insert_mem_bar(Op_OnSpinWait);
2948 return true;
2949 }
2950
2951 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2952 if (!kls->is_Con()) {
2953 return true;
2954 }
2955 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2956 if (klsptr == nullptr) {
2957 return true;
2958 }
2959 ciInstanceKlass* ik = klsptr->instance_klass();
2960 // don't need a guard for a klass that is already initialized
2961 return !ik->is_initialized();
2962 }
2963
2964 //----------------------------inline_unsafe_writeback0-------------------------
2965 // public native void Unsafe.writeback0(long address)
2966 bool LibraryCallKit::inline_unsafe_writeback0() {
2967 if (!Matcher::has_match_rule(Op_CacheWB)) {
2968 return false;
2969 }
2970 #ifndef PRODUCT
2971 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2972 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2973 ciSignature* sig = callee()->signature();
2974 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2975 #endif
2976 null_check_receiver(); // null-check, then ignore
2977 Node *addr = argument(1);
2978 addr = new CastX2PNode(addr);
2979 addr = _gvn.transform(addr);
2980 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2981 flush = _gvn.transform(flush);
2982 set_memory(flush, TypeRawPtr::BOTTOM);
2983 return true;
2984 }
2985
2986 //----------------------------inline_unsafe_writeback0-------------------------
2987 // public native void Unsafe.writeback0(long address)
2988 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2989 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2990 return false;
2991 }
2992 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2993 return false;
2994 }
2995 #ifndef PRODUCT
2996 assert(Matcher::has_match_rule(Op_CacheWB),
2997 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2998 : "found match rule for CacheWBPostSync but not CacheWB"));
2999
3000 #endif
3001 null_check_receiver(); // null-check, then ignore
3002 Node *sync;
3003 if (is_pre) {
3004 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3005 } else {
3006 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3007 }
3008 sync = _gvn.transform(sync);
3009 set_memory(sync, TypeRawPtr::BOTTOM);
3010 return true;
3011 }
3012
3013 //----------------------------inline_unsafe_allocate---------------------------
3014 // public native Object Unsafe.allocateInstance(Class<?> cls);
3015 bool LibraryCallKit::inline_unsafe_allocate() {
3016
3017 #if INCLUDE_JVMTI
3018 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3019 return false;
3020 }
3021 #endif //INCLUDE_JVMTI
3022
3023 if (callee()->is_static()) return false; // caller must have the capability!
3024
3025 null_check_receiver(); // null-check, then ignore
3026 Node* cls = null_check(argument(1));
3027 if (stopped()) return true;
3028
3029 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3030 kls = null_check(kls);
3031 if (stopped()) return true; // argument was like int.class
3032
3033 #if INCLUDE_JVMTI
3034 // Don't try to access new allocated obj in the intrinsic.
3035 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3036 // Deoptimize and allocate in interpreter instead.
3037 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3038 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3039 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3040 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3041 {
3042 BuildCutout unless(this, tst, PROB_MAX);
3043 uncommon_trap(Deoptimization::Reason_intrinsic,
3044 Deoptimization::Action_make_not_entrant);
3045 }
3046 if (stopped()) {
3047 return true;
3048 }
3049 #endif //INCLUDE_JVMTI
3050
3051 Node* test = nullptr;
3052 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3053 // Note: The argument might still be an illegal value like
3054 // Serializable.class or Object[].class. The runtime will handle it.
3055 // But we must make an explicit check for initialization.
3056 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3057 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3058 // can generate code to load it as unsigned byte.
3059 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3060 Node* bits = intcon(InstanceKlass::fully_initialized);
3061 test = _gvn.transform(new SubINode(inst, bits));
3062 // The 'test' is non-zero if we need to take a slow path.
3063 }
3064 Node* obj = nullptr;
3065 const TypeInstKlassPtr* tkls = _gvn.type(kls)->isa_instklassptr();
3066 if (tkls != nullptr && tkls->instance_klass()->is_inlinetype()) {
3067 obj = InlineTypeNode::make_default(_gvn, tkls->instance_klass()->as_inline_klass())->buffer(this);
3068 } else {
3069 obj = new_instance(kls, test);
3070 }
3071 set_result(obj);
3072 return true;
3073 }
3074
3075 //------------------------inline_native_time_funcs--------------
3076 // inline code for System.currentTimeMillis() and System.nanoTime()
3077 // these have the same type and signature
3078 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3079 const TypeFunc* tf = OptoRuntime::void_long_Type();
3080 const TypePtr* no_memory_effects = nullptr;
3081 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3082 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3083 #ifdef ASSERT
3084 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3085 assert(value_top == top(), "second value must be top");
3086 #endif
3087 set_result(value);
3088 return true;
3089 }
3090
3091
3092 #if INCLUDE_JVMTI
3093
3094 // When notifications are disabled then just update the VTMS transition bit and return.
3095 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
3096 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
3097 if (!DoJVMTIVirtualThreadTransitions) {
3098 return true;
3099 }
3100 IdealKit ideal(this);
3101
3102 Node* ONE = ideal.ConI(1);
3103 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
3104 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
3105 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3106
3107 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
3108 // if notifyJvmti enabled then make a call to the given SharedRuntime function
3109 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
3110 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3111
3112 sync_kit(ideal);
3113 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
3114 ideal.sync_kit(this);
3115 } ideal.else_(); {
3116 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
3117 Node* vt_oop = _gvn.transform(argument(0)); // this argument - VirtualThread oop
3118 Node* thread = ideal.thread();
3119 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
3120 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
3121 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3122
3123 sync_kit(ideal);
3124 access_store_at(nullptr, jt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3125 access_store_at(nullptr, vt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3126
3127 ideal.sync_kit(this);
3128 } ideal.end_if();
3129 final_sync(ideal);
3130
3131 return true;
3132 }
3133
3134 // Always update the temporary VTMS transition bit.
3135 bool LibraryCallKit::inline_native_notify_jvmti_hide() {
3136 if (!DoJVMTIVirtualThreadTransitions) {
3137 return true;
3138 }
3139 IdealKit ideal(this);
3140
3141 {
3142 // unconditionally update the temporary VTMS transition bit in current JavaThread
3143 Node* thread = ideal.thread();
3144 Node* hide = _gvn.transform(argument(1)); // hide argument for temporary VTMS transition notification
3145 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_tmp_VTMS_transition_offset()));
3146 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3147
3148 sync_kit(ideal);
3149 access_store_at(nullptr, addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3150 ideal.sync_kit(this);
3151 }
3152 final_sync(ideal);
3153
3154 return true;
3155 }
3156
3157 #endif // INCLUDE_JVMTI
3158
3159 #ifdef JFR_HAVE_INTRINSICS
3160
3161 /**
3162 * if oop->klass != null
3163 * // normal class
3164 * epoch = _epoch_state ? 2 : 1
3165 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3166 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3167 * }
3168 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3169 * else
3170 * // primitive class
3171 * if oop->array_klass != null
3172 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3173 * else
3174 * id = LAST_TYPE_ID + 1 // void class path
3175 * if (!signaled)
3176 * signaled = true
3177 */
3178 bool LibraryCallKit::inline_native_classID() {
3179 Node* cls = argument(0);
3180
3181 IdealKit ideal(this);
3182 #define __ ideal.
3183 IdealVariable result(ideal); __ declarations_done();
3184 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3185 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3186 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3187
3188
3189 __ if_then(kls, BoolTest::ne, null()); {
3190 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3191 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3192
3193 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3194 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3195 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3196 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3197 mask = _gvn.transform(new OrLNode(mask, epoch));
3198 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3199
3200 float unlikely = PROB_UNLIKELY(0.999);
3201 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3202 sync_kit(ideal);
3203 make_runtime_call(RC_LEAF,
3204 OptoRuntime::class_id_load_barrier_Type(),
3205 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3206 "class id load barrier",
3207 TypePtr::BOTTOM,
3208 kls);
3209 ideal.sync_kit(this);
3210 } __ end_if();
3211
3212 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3213 } __ else_(); {
3214 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3215 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3216 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3217 __ if_then(array_kls, BoolTest::ne, null()); {
3218 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3219 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3220 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3221 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3222 } __ else_(); {
3223 // void class case
3224 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3225 } __ end_if();
3226
3227 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3228 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3229 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3230 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3231 } __ end_if();
3232 } __ end_if();
3233
3234 final_sync(ideal);
3235 set_result(ideal.value(result));
3236 #undef __
3237 return true;
3238 }
3239
3240 //------------------------inline_native_jvm_commit------------------
3241 bool LibraryCallKit::inline_native_jvm_commit() {
3242 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3243
3244 // Save input memory and i_o state.
3245 Node* input_memory_state = reset_memory();
3246 set_all_memory(input_memory_state);
3247 Node* input_io_state = i_o();
3248
3249 // TLS.
3250 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3251 // Jfr java buffer.
3252 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3253 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3254 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3255
3256 // Load the current value of the notified field in the JfrThreadLocal.
3257 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3258 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3259
3260 // Test for notification.
3261 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3262 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3263 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3264
3265 // True branch, is notified.
3266 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3267 set_control(is_notified);
3268
3269 // Reset notified state.
3270 Node* notified_reset_memory = store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered);
3271
3272 // Iff notified, the return address of the commit method is the current position of the backing java buffer. This is used to reset the event writer.
3273 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3274 // Convert the machine-word to a long.
3275 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3276
3277 // False branch, not notified.
3278 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3279 set_control(not_notified);
3280 set_all_memory(input_memory_state);
3281
3282 // Arg is the next position as a long.
3283 Node* arg = argument(0);
3284 // Convert long to machine-word.
3285 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3286
3287 // Store the next_position to the underlying jfr java buffer.
3288 Node* commit_memory;
3289 #ifdef _LP64
3290 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_LONG, Compile::AliasIdxRaw, MemNode::release);
3291 #else
3292 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_INT, Compile::AliasIdxRaw, MemNode::release);
3293 #endif
3294
3295 // Now load the flags from off the java buffer and decide if the buffer is a lease. If so, it needs to be returned post-commit.
3296 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3297 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3298 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3299
3300 // And flags with lease constant.
3301 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3302
3303 // Branch on lease to conditionalize returning the leased java buffer.
3304 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3305 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3306 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3307
3308 // False branch, not a lease.
3309 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3310
3311 // True branch, is lease.
3312 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3313 set_control(is_lease);
3314
3315 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3316 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3317 OptoRuntime::void_void_Type(),
3318 StubRoutines::jfr_return_lease(),
3319 "return_lease", TypePtr::BOTTOM);
3320 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3321
3322 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3323 record_for_igvn(lease_compare_rgn);
3324 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3325 record_for_igvn(lease_compare_mem);
3326 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3327 record_for_igvn(lease_compare_io);
3328 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3329 record_for_igvn(lease_result_value);
3330
3331 // Update control and phi nodes.
3332 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3333 lease_compare_rgn->init_req(_false_path, not_lease);
3334
3335 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3336 lease_compare_mem->init_req(_false_path, commit_memory);
3337
3338 lease_compare_io->init_req(_true_path, i_o());
3339 lease_compare_io->init_req(_false_path, input_io_state);
3340
3341 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3342 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3343
3344 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3345 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3346 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3347 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3348
3349 // Update control and phi nodes.
3350 result_rgn->init_req(_true_path, is_notified);
3351 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3352
3353 result_mem->init_req(_true_path, notified_reset_memory);
3354 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3355
3356 result_io->init_req(_true_path, input_io_state);
3357 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3358
3359 result_value->init_req(_true_path, current_pos);
3360 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3361
3362 // Set output state.
3363 set_control(_gvn.transform(result_rgn));
3364 set_all_memory(_gvn.transform(result_mem));
3365 set_i_o(_gvn.transform(result_io));
3366 set_result(result_rgn, result_value);
3367 return true;
3368 }
3369
3370 /*
3371 * The intrinsic is a model of this pseudo-code:
3372 *
3373 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3374 * jobject h_event_writer = tl->java_event_writer();
3375 * if (h_event_writer == nullptr) {
3376 * return nullptr;
3377 * }
3378 * oop threadObj = Thread::threadObj();
3379 * oop vthread = java_lang_Thread::vthread(threadObj);
3380 * traceid tid;
3381 * bool excluded;
3382 * if (vthread != threadObj) { // i.e. current thread is virtual
3383 * tid = java_lang_Thread::tid(vthread);
3384 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3385 * excluded = vthread_epoch_raw & excluded_mask;
3386 * if (!excluded) {
3387 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3388 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3389 * if (vthread_epoch != current_epoch) {
3390 * write_checkpoint();
3391 * }
3392 * }
3393 * } else {
3394 * tid = java_lang_Thread::tid(threadObj);
3395 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3396 * excluded = thread_epoch_raw & excluded_mask;
3397 * }
3398 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3399 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3400 * if (tid_in_event_writer != tid) {
3401 * setField(event_writer, "threadID", tid);
3402 * setField(event_writer, "excluded", excluded);
3403 * }
3404 * return event_writer
3405 */
3406 bool LibraryCallKit::inline_native_getEventWriter() {
3407 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3408
3409 // Save input memory and i_o state.
3410 Node* input_memory_state = reset_memory();
3411 set_all_memory(input_memory_state);
3412 Node* input_io_state = i_o();
3413
3414 Node* excluded_mask = _gvn.intcon(32768);
3415 Node* epoch_mask = _gvn.intcon(32767);
3416
3417 // TLS
3418 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3419
3420 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3421 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3422
3423 // Load the eventwriter jobject handle.
3424 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3425
3426 // Null check the jobject handle.
3427 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3428 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3429 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3430
3431 // False path, jobj is null.
3432 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3433
3434 // True path, jobj is not null.
3435 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3436
3437 set_control(jobj_is_not_null);
3438
3439 // Load the threadObj for the CarrierThread.
3440 Node* threadObj = generate_current_thread(tls_ptr);
3441
3442 // Load the vthread.
3443 Node* vthread = generate_virtual_thread(tls_ptr);
3444
3445 // If vthread != threadObj, this is a virtual thread.
3446 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3447 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3448 IfNode* iff_vthread_not_equal_threadObj =
3449 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3450
3451 // False branch, fallback to threadObj.
3452 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3453 set_control(vthread_equal_threadObj);
3454
3455 // Load the tid field from the vthread object.
3456 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3457
3458 // Load the raw epoch value from the threadObj.
3459 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3460 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3461 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3462
3463 // Mask off the excluded information from the epoch.
3464 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3465
3466 // True branch, this is a virtual thread.
3467 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3468 set_control(vthread_not_equal_threadObj);
3469
3470 // Load the tid field from the vthread object.
3471 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3472
3473 // Load the raw epoch value from the vthread.
3474 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3475 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3476 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3477
3478 // Mask off the excluded information from the epoch.
3479 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3480
3481 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3482 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3483 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3484 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3485
3486 // False branch, vthread is excluded, no need to write epoch info.
3487 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3488
3489 // True branch, vthread is included, update epoch info.
3490 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3491 set_control(included);
3492
3493 // Get epoch value.
3494 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3495
3496 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3497 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3498 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3499
3500 // Compare the epoch in the vthread to the current epoch generation.
3501 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3502 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3503 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3504
3505 // False path, epoch is equal, checkpoint information is valid.
3506 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3507
3508 // True path, epoch is not equal, write a checkpoint for the vthread.
3509 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3510
3511 set_control(epoch_is_not_equal);
3512
3513 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3514 // The call also updates the native thread local thread id and the vthread with the current epoch.
3515 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3516 OptoRuntime::jfr_write_checkpoint_Type(),
3517 StubRoutines::jfr_write_checkpoint(),
3518 "write_checkpoint", TypePtr::BOTTOM);
3519 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3520
3521 // vthread epoch != current epoch
3522 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3523 record_for_igvn(epoch_compare_rgn);
3524 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3525 record_for_igvn(epoch_compare_mem);
3526 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3527 record_for_igvn(epoch_compare_io);
3528
3529 // Update control and phi nodes.
3530 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3531 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3532 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3533 epoch_compare_mem->init_req(_false_path, input_memory_state);
3534 epoch_compare_io->init_req(_true_path, i_o());
3535 epoch_compare_io->init_req(_false_path, input_io_state);
3536
3537 // excluded != true
3538 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3539 record_for_igvn(exclude_compare_rgn);
3540 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3541 record_for_igvn(exclude_compare_mem);
3542 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3543 record_for_igvn(exclude_compare_io);
3544
3545 // Update control and phi nodes.
3546 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3547 exclude_compare_rgn->init_req(_false_path, excluded);
3548 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3549 exclude_compare_mem->init_req(_false_path, input_memory_state);
3550 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3551 exclude_compare_io->init_req(_false_path, input_io_state);
3552
3553 // vthread != threadObj
3554 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3555 record_for_igvn(vthread_compare_rgn);
3556 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3557 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3558 record_for_igvn(vthread_compare_io);
3559 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3560 record_for_igvn(tid);
3561 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3562 record_for_igvn(exclusion);
3563
3564 // Update control and phi nodes.
3565 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3566 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3567 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3568 vthread_compare_mem->init_req(_false_path, input_memory_state);
3569 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3570 vthread_compare_io->init_req(_false_path, input_io_state);
3571 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3572 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3573 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3574 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3575
3576 // Update branch state.
3577 set_control(_gvn.transform(vthread_compare_rgn));
3578 set_all_memory(_gvn.transform(vthread_compare_mem));
3579 set_i_o(_gvn.transform(vthread_compare_io));
3580
3581 // Load the event writer oop by dereferencing the jobject handle.
3582 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3583 assert(klass_EventWriter->is_loaded(), "invariant");
3584 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3585 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3586 const TypeOopPtr* const xtype = aklass->as_instance_type();
3587 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3588 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3589
3590 // Load the current thread id from the event writer object.
3591 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3592 // Get the field offset to, conditionally, store an updated tid value later.
3593 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3594 const TypePtr* event_writer_tid_field_type = _gvn.type(event_writer_tid_field)->isa_ptr();
3595 // Get the field offset to, conditionally, store an updated exclusion value later.
3596 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3597 const TypePtr* event_writer_excluded_field_type = _gvn.type(event_writer_excluded_field)->isa_ptr();
3598
3599 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3600 record_for_igvn(event_writer_tid_compare_rgn);
3601 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3602 record_for_igvn(event_writer_tid_compare_mem);
3603 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3604 record_for_igvn(event_writer_tid_compare_io);
3605
3606 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3607 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3608 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3609 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3610
3611 // False path, tids are the same.
3612 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3613
3614 // True path, tid is not equal, need to update the tid in the event writer.
3615 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3616 record_for_igvn(tid_is_not_equal);
3617
3618 // Store the exclusion state to the event writer.
3619 store_to_memory(tid_is_not_equal, event_writer_excluded_field, _gvn.transform(exclusion), T_BOOLEAN, event_writer_excluded_field_type, MemNode::unordered);
3620
3621 // Store the tid to the event writer.
3622 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, event_writer_tid_field_type, MemNode::unordered);
3623
3624 // Update control and phi nodes.
3625 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3626 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3627 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3628 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3629 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3630 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3631
3632 // Result of top level CFG, Memory, IO and Value.
3633 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3634 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3635 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3636 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3637
3638 // Result control.
3639 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3640 result_rgn->init_req(_false_path, jobj_is_null);
3641
3642 // Result memory.
3643 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3644 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3645
3646 // Result IO.
3647 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3648 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3649
3650 // Result value.
3651 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3652 result_value->init_req(_false_path, null()); // return null
3653
3654 // Set output state.
3655 set_control(_gvn.transform(result_rgn));
3656 set_all_memory(_gvn.transform(result_mem));
3657 set_i_o(_gvn.transform(result_io));
3658 set_result(result_rgn, result_value);
3659 return true;
3660 }
3661
3662 /*
3663 * The intrinsic is a model of this pseudo-code:
3664 *
3665 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3666 * if (carrierThread != thread) { // is virtual thread
3667 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3668 * bool excluded = vthread_epoch_raw & excluded_mask;
3669 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3670 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3671 * if (!excluded) {
3672 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3673 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3674 * }
3675 * Atomic::release_store(&tl->_vthread, true);
3676 * return;
3677 * }
3678 * Atomic::release_store(&tl->_vthread, false);
3679 */
3680 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3681 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3682
3683 Node* input_memory_state = reset_memory();
3684 set_all_memory(input_memory_state);
3685
3686 Node* excluded_mask = _gvn.intcon(32768);
3687 Node* epoch_mask = _gvn.intcon(32767);
3688
3689 Node* const carrierThread = generate_current_thread(jt);
3690 // If thread != carrierThread, this is a virtual thread.
3691 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3692 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3693 IfNode* iff_thread_not_equal_carrierThread =
3694 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3695
3696 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3697
3698 // False branch, is carrierThread.
3699 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3700 // Store release
3701 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3702
3703 set_all_memory(input_memory_state);
3704
3705 // True branch, is virtual thread.
3706 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3707 set_control(thread_not_equal_carrierThread);
3708
3709 // Load the raw epoch value from the vthread.
3710 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3711 Node* epoch_raw = access_load_at(thread, epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3712 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3713
3714 // Mask off the excluded information from the epoch.
3715 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3716
3717 // Load the tid field from the thread.
3718 Node* tid = load_field_from_object(thread, "tid", "J");
3719
3720 // Store the vthread tid to the jfr thread local.
3721 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3722 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, Compile::AliasIdxRaw, MemNode::unordered, true);
3723
3724 // Branch is_excluded to conditionalize updating the epoch .
3725 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3726 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3727 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3728
3729 // True branch, vthread is excluded, no need to write epoch info.
3730 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3731 set_control(excluded);
3732 Node* vthread_is_excluded = _gvn.intcon(1);
3733
3734 // False branch, vthread is included, update epoch info.
3735 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3736 set_control(included);
3737 Node* vthread_is_included = _gvn.intcon(0);
3738
3739 // Get epoch value.
3740 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3741
3742 // Store the vthread epoch to the jfr thread local.
3743 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3744 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, Compile::AliasIdxRaw, MemNode::unordered, true);
3745
3746 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3747 record_for_igvn(excluded_rgn);
3748 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3749 record_for_igvn(excluded_mem);
3750 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3751 record_for_igvn(exclusion);
3752
3753 // Merge the excluded control and memory.
3754 excluded_rgn->init_req(_true_path, excluded);
3755 excluded_rgn->init_req(_false_path, included);
3756 excluded_mem->init_req(_true_path, tid_memory);
3757 excluded_mem->init_req(_false_path, included_memory);
3758 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3759 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3760
3761 // Set intermediate state.
3762 set_control(_gvn.transform(excluded_rgn));
3763 set_all_memory(excluded_mem);
3764
3765 // Store the vthread exclusion state to the jfr thread local.
3766 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3767 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered, true);
3768
3769 // Store release
3770 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3771
3772 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3773 record_for_igvn(thread_compare_rgn);
3774 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3775 record_for_igvn(thread_compare_mem);
3776 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3777 record_for_igvn(vthread);
3778
3779 // Merge the thread_compare control and memory.
3780 thread_compare_rgn->init_req(_true_path, control());
3781 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3782 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3783 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3784
3785 // Set output state.
3786 set_control(_gvn.transform(thread_compare_rgn));
3787 set_all_memory(_gvn.transform(thread_compare_mem));
3788 }
3789
3790 #endif // JFR_HAVE_INTRINSICS
3791
3792 //------------------------inline_native_currentCarrierThread------------------
3793 bool LibraryCallKit::inline_native_currentCarrierThread() {
3794 Node* junk = nullptr;
3795 set_result(generate_current_thread(junk));
3796 return true;
3797 }
3798
3799 //------------------------inline_native_currentThread------------------
3800 bool LibraryCallKit::inline_native_currentThread() {
3801 Node* junk = nullptr;
3802 set_result(generate_virtual_thread(junk));
3803 return true;
3804 }
3805
3806 //------------------------inline_native_setVthread------------------
3807 bool LibraryCallKit::inline_native_setCurrentThread() {
3808 assert(C->method()->changes_current_thread(),
3809 "method changes current Thread but is not annotated ChangesCurrentThread");
3810 Node* arr = argument(1);
3811 Node* thread = _gvn.transform(new ThreadLocalNode());
3812 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3813 Node* thread_obj_handle
3814 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3815 thread_obj_handle = _gvn.transform(thread_obj_handle);
3816 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3817 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3818 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3819 return true;
3820 }
3821
3822 const Type* LibraryCallKit::scopedValueCache_type() {
3823 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3824 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3825 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true);
3826
3827 // Because we create the scopedValue cache lazily we have to make the
3828 // type of the result BotPTR.
3829 bool xk = etype->klass_is_exact();
3830 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
3831 return objects_type;
3832 }
3833
3834 Node* LibraryCallKit::scopedValueCache_helper() {
3835 Node* thread = _gvn.transform(new ThreadLocalNode());
3836 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3837 // We cannot use immutable_memory() because we might flip onto a
3838 // different carrier thread, at which point we'll need to use that
3839 // carrier thread's cache.
3840 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3841 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3842 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3843 }
3844
3845 //------------------------inline_native_scopedValueCache------------------
3846 bool LibraryCallKit::inline_native_scopedValueCache() {
3847 Node* cache_obj_handle = scopedValueCache_helper();
3848 const Type* objects_type = scopedValueCache_type();
3849 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3850
3851 return true;
3852 }
3853
3854 //------------------------inline_native_setScopedValueCache------------------
3855 bool LibraryCallKit::inline_native_setScopedValueCache() {
3856 Node* arr = argument(0);
3857 Node* cache_obj_handle = scopedValueCache_helper();
3858 const Type* objects_type = scopedValueCache_type();
3859
3860 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3861 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3862
3863 return true;
3864 }
3865
3866 //-----------------------load_klass_from_mirror_common-------------------------
3867 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3868 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3869 // and branch to the given path on the region.
3870 // If never_see_null, take an uncommon trap on null, so we can optimistically
3871 // compile for the non-null case.
3872 // If the region is null, force never_see_null = true.
3873 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3874 bool never_see_null,
3875 RegionNode* region,
3876 int null_path,
3877 int offset) {
3878 if (region == nullptr) never_see_null = true;
3879 Node* p = basic_plus_adr(mirror, offset);
3880 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3881 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3882 Node* null_ctl = top();
3883 kls = null_check_oop(kls, &null_ctl, never_see_null);
3884 if (region != nullptr) {
3885 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3886 region->init_req(null_path, null_ctl);
3887 } else {
3888 assert(null_ctl == top(), "no loose ends");
3889 }
3890 return kls;
3891 }
3892
3893 //--------------------(inline_native_Class_query helpers)---------------------
3894 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3895 // Fall through if (mods & mask) == bits, take the guard otherwise.
3896 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3897 // Branch around if the given klass has the given modifier bit set.
3898 // Like generate_guard, adds a new path onto the region.
3899 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3900 Node* mods = make_load(nullptr, modp, TypeInt::INT, T_INT, MemNode::unordered);
3901 Node* mask = intcon(modifier_mask);
3902 Node* bits = intcon(modifier_bits);
3903 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3904 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3905 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3906 return generate_fair_guard(bol, region);
3907 }
3908
3909 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3910 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3911 }
3912 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3913 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
3914 }
3915
3916 //-------------------------inline_native_Class_query-------------------
3917 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3918 const Type* return_type = TypeInt::BOOL;
3919 Node* prim_return_value = top(); // what happens if it's a primitive class?
3920 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3921 bool expect_prim = false; // most of these guys expect to work on refs
3922
3923 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3924
3925 Node* mirror = argument(0);
3926 Node* obj = top();
3927
3928 switch (id) {
3929 case vmIntrinsics::_isInstance:
3930 // nothing is an instance of a primitive type
3931 prim_return_value = intcon(0);
3932 obj = argument(1);
3933 break;
3934 case vmIntrinsics::_getModifiers:
3935 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3936 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3937 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3938 break;
3939 case vmIntrinsics::_isInterface:
3940 prim_return_value = intcon(0);
3941 break;
3942 case vmIntrinsics::_isArray:
3943 prim_return_value = intcon(0);
3944 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3945 break;
3946 case vmIntrinsics::_isPrimitive:
3947 prim_return_value = intcon(1);
3948 expect_prim = true; // obviously
3949 break;
3950 case vmIntrinsics::_isHidden:
3951 prim_return_value = intcon(0);
3952 break;
3953 case vmIntrinsics::_getSuperclass:
3954 prim_return_value = null();
3955 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3956 break;
3957 case vmIntrinsics::_getClassAccessFlags:
3958 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3959 return_type = TypeInt::INT; // not bool! 6297094
3960 break;
3961 default:
3962 fatal_unexpected_iid(id);
3963 break;
3964 }
3965
3966 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3967 if (mirror_con == nullptr) return false; // cannot happen?
3968
3969 #ifndef PRODUCT
3970 if (C->print_intrinsics() || C->print_inlining()) {
3971 ciType* k = mirror_con->java_mirror_type();
3972 if (k) {
3973 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3974 k->print_name();
3975 tty->cr();
3976 }
3977 }
3978 #endif
3979
3980 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3981 RegionNode* region = new RegionNode(PATH_LIMIT);
3982 record_for_igvn(region);
3983 PhiNode* phi = new PhiNode(region, return_type);
3984
3985 // The mirror will never be null of Reflection.getClassAccessFlags, however
3986 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3987 // if it is. See bug 4774291.
3988
3989 // For Reflection.getClassAccessFlags(), the null check occurs in
3990 // the wrong place; see inline_unsafe_access(), above, for a similar
3991 // situation.
3992 mirror = null_check(mirror);
3993 // If mirror or obj is dead, only null-path is taken.
3994 if (stopped()) return true;
3995
3996 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3997
3998 // Now load the mirror's klass metaobject, and null-check it.
3999 // Side-effects region with the control path if the klass is null.
4000 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4001 // If kls is null, we have a primitive mirror.
4002 phi->init_req(_prim_path, prim_return_value);
4003 if (stopped()) { set_result(region, phi); return true; }
4004 bool safe_for_replace = (region->in(_prim_path) == top());
4005
4006 Node* p; // handy temp
4007 Node* null_ctl;
4008
4009 // Now that we have the non-null klass, we can perform the real query.
4010 // For constant classes, the query will constant-fold in LoadNode::Value.
4011 Node* query_value = top();
4012 switch (id) {
4013 case vmIntrinsics::_isInstance:
4014 // nothing is an instance of a primitive type
4015 query_value = gen_instanceof(obj, kls, safe_for_replace);
4016 break;
4017
4018 case vmIntrinsics::_getModifiers:
4019 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
4020 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
4021 break;
4022
4023 case vmIntrinsics::_isInterface:
4024 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4025 if (generate_interface_guard(kls, region) != nullptr)
4026 // A guard was added. If the guard is taken, it was an interface.
4027 phi->add_req(intcon(1));
4028 // If we fall through, it's a plain class.
4029 query_value = intcon(0);
4030 break;
4031
4032 case vmIntrinsics::_isArray:
4033 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
4034 if (generate_array_guard(kls, region) != nullptr)
4035 // A guard was added. If the guard is taken, it was an array.
4036 phi->add_req(intcon(1));
4037 // If we fall through, it's a plain class.
4038 query_value = intcon(0);
4039 break;
4040
4041 case vmIntrinsics::_isPrimitive:
4042 query_value = intcon(0); // "normal" path produces false
4043 break;
4044
4045 case vmIntrinsics::_isHidden:
4046 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4047 if (generate_hidden_class_guard(kls, region) != nullptr)
4048 // A guard was added. If the guard is taken, it was an hidden class.
4049 phi->add_req(intcon(1));
4050 // If we fall through, it's a plain class.
4051 query_value = intcon(0);
4052 break;
4053
4054
4055 case vmIntrinsics::_getSuperclass:
4056 // The rules here are somewhat unfortunate, but we can still do better
4057 // with random logic than with a JNI call.
4058 // Interfaces store null or Object as _super, but must report null.
4059 // Arrays store an intermediate super as _super, but must report Object.
4060 // Other types can report the actual _super.
4061 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4062 if (generate_interface_guard(kls, region) != nullptr)
4063 // A guard was added. If the guard is taken, it was an interface.
4064 phi->add_req(null());
4065 if (generate_array_guard(kls, region) != nullptr)
4066 // A guard was added. If the guard is taken, it was an array.
4067 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4068 // If we fall through, it's a plain class. Get its _super.
4069 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4070 kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4071 null_ctl = top();
4072 kls = null_check_oop(kls, &null_ctl);
4073 if (null_ctl != top()) {
4074 // If the guard is taken, Object.superClass is null (both klass and mirror).
4075 region->add_req(null_ctl);
4076 phi ->add_req(null());
4077 }
4078 if (!stopped()) {
4079 query_value = load_mirror_from_klass(kls);
4080 }
4081 break;
4082
4083 case vmIntrinsics::_getClassAccessFlags:
4084 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4085 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
4086 break;
4087
4088 default:
4089 fatal_unexpected_iid(id);
4090 break;
4091 }
4092
4093 // Fall-through is the normal case of a query to a real class.
4094 phi->init_req(1, query_value);
4095 region->init_req(1, control());
4096
4097 C->set_has_split_ifs(true); // Has chance for split-if optimization
4098 set_result(region, phi);
4099 return true;
4100 }
4101
4102 //-------------------------inline_primitive_Class_conversion-------------------
4103 // Class<T> java.lang.Class .asPrimaryType()
4104 // public static Class<T> jdk.internal.value.PrimitiveClass.asPrimaryType(Class<T>)
4105 // Class<T> java.lang.Class .asValueType()
4106 // public static Class<T> jdk.internal.value.PrimitiveClass.asValueType(Class<T>)
4107 bool LibraryCallKit::inline_primitive_Class_conversion(vmIntrinsics::ID id) {
4108 Node* mirror = argument(0); // Receiver/argument Class
4109 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4110 if (mirror_con == nullptr) {
4111 return false;
4112 }
4113
4114 bool is_val_mirror = true;
4115 ciType* tm = mirror_con->java_mirror_type(&is_val_mirror);
4116 if (tm != nullptr) {
4117 Node* result = mirror;
4118 if ((id == vmIntrinsics::_asPrimaryType || id == vmIntrinsics::_asPrimaryTypeArg) && is_val_mirror) {
4119 result = _gvn.makecon(TypeInstPtr::make(tm->as_inline_klass()->ref_mirror()));
4120 } else if (id == vmIntrinsics::_asValueType || id == vmIntrinsics::_asValueTypeArg) {
4121 if (!tm->is_inlinetype()) {
4122 return false; // Throw UnsupportedOperationException
4123 } else if (!is_val_mirror) {
4124 result = _gvn.makecon(TypeInstPtr::make(tm->as_inline_klass()->val_mirror()));
4125 }
4126 }
4127 set_result(result);
4128 return true;
4129 }
4130 return false;
4131 }
4132
4133 //-------------------------inline_Class_cast-------------------
4134 bool LibraryCallKit::inline_Class_cast() {
4135 Node* mirror = argument(0); // Class
4136 Node* obj = argument(1);
4137 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4138 if (mirror_con == nullptr) {
4139 return false; // dead path (mirror->is_top()).
4140 }
4141 if (obj == nullptr || obj->is_top()) {
4142 return false; // dead path
4143 }
4144 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4145
4146 // First, see if Class.cast() can be folded statically.
4147 // java_mirror_type() returns non-null for compile-time Class constants.
4148 bool requires_null_check = false;
4149 ciType* tm = mirror_con->java_mirror_type(&requires_null_check);
4150 if (tm != nullptr && tm->is_klass() &&
4151 tp != nullptr) {
4152 if (!tp->is_loaded()) {
4153 // Don't use intrinsic when class is not loaded.
4154 return false;
4155 } else {
4156 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4157 if (static_res == Compile::SSC_always_true) {
4158 // isInstance() is true - fold the code.
4159 if (requires_null_check) {
4160 obj = null_check(obj);
4161 }
4162 set_result(obj);
4163 return true;
4164 } else if (static_res == Compile::SSC_always_false) {
4165 // Don't use intrinsic, have to throw ClassCastException.
4166 // If the reference is null, the non-intrinsic bytecode will
4167 // be optimized appropriately.
4168 return false;
4169 }
4170 }
4171 }
4172
4173 // Bailout intrinsic and do normal inlining if exception path is frequent.
4174 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4175 return false;
4176 }
4177
4178 // Generate dynamic checks.
4179 // Class.cast() is java implementation of _checkcast bytecode.
4180 // Do checkcast (Parse::do_checkcast()) optimizations here.
4181
4182 if (requires_null_check) {
4183 obj = null_check(obj);
4184 }
4185 mirror = null_check(mirror);
4186 // If mirror is dead, only null-path is taken.
4187 if (stopped()) {
4188 return true;
4189 }
4190
4191 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4192 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4193 RegionNode* region = new RegionNode(PATH_LIMIT);
4194 record_for_igvn(region);
4195
4196 // Now load the mirror's klass metaobject, and null-check it.
4197 // If kls is null, we have a primitive mirror and
4198 // nothing is an instance of a primitive type.
4199 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4200
4201 Node* res = top();
4202 Node* io = i_o();
4203 Node* mem = merged_memory();
4204 if (!stopped()) {
4205 if (EnableValhalla && !requires_null_check) {
4206 // Check if we are casting to QMyValue
4207 Node* ctrl_val_mirror = generate_fair_guard(is_val_mirror(mirror), nullptr);
4208 if (ctrl_val_mirror != nullptr) {
4209 RegionNode* r = new RegionNode(3);
4210 record_for_igvn(r);
4211 r->init_req(1, control());
4212
4213 // Casting to QMyValue, check for null
4214 set_control(ctrl_val_mirror);
4215 { // PreserveJVMState because null check replaces obj in map
4216 PreserveJVMState pjvms(this);
4217 Node* null_ctr = top();
4218 null_check_oop(obj, &null_ctr);
4219 region->init_req(_npe_path, null_ctr);
4220 r->init_req(2, control());
4221 }
4222 set_control(_gvn.transform(r));
4223 }
4224 }
4225
4226 Node* bad_type_ctrl = top();
4227 // Do checkcast optimizations.
4228 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4229 region->init_req(_bad_type_path, bad_type_ctrl);
4230 }
4231 if (region->in(_prim_path) != top() ||
4232 region->in(_bad_type_path) != top() ||
4233 region->in(_npe_path) != top()) {
4234 // Let Interpreter throw ClassCastException.
4235 PreserveJVMState pjvms(this);
4236 set_control(_gvn.transform(region));
4237 // Set IO and memory because gen_checkcast may override them when buffering inline types
4238 set_i_o(io);
4239 set_all_memory(mem);
4240 uncommon_trap(Deoptimization::Reason_intrinsic,
4241 Deoptimization::Action_maybe_recompile);
4242 }
4243 if (!stopped()) {
4244 set_result(res);
4245 }
4246 return true;
4247 }
4248
4249
4250 //--------------------------inline_native_subtype_check------------------------
4251 // This intrinsic takes the JNI calls out of the heart of
4252 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4253 bool LibraryCallKit::inline_native_subtype_check() {
4254 // Pull both arguments off the stack.
4255 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4256 args[0] = argument(0);
4257 args[1] = argument(1);
4258 Node* klasses[2]; // corresponding Klasses: superk, subk
4259 klasses[0] = klasses[1] = top();
4260
4261 enum {
4262 // A full decision tree on {superc is prim, subc is prim}:
4263 _prim_0_path = 1, // {P,N} => false
4264 // {P,P} & superc!=subc => false
4265 _prim_same_path, // {P,P} & superc==subc => true
4266 _prim_1_path, // {N,P} => false
4267 _ref_subtype_path, // {N,N} & subtype check wins => true
4268 _both_ref_path, // {N,N} & subtype check loses => false
4269 PATH_LIMIT
4270 };
4271
4272 RegionNode* region = new RegionNode(PATH_LIMIT);
4273 RegionNode* prim_region = new RegionNode(2);
4274 Node* phi = new PhiNode(region, TypeInt::BOOL);
4275 record_for_igvn(region);
4276 record_for_igvn(prim_region);
4277
4278 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4279 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4280 int class_klass_offset = java_lang_Class::klass_offset();
4281
4282 // First null-check both mirrors and load each mirror's klass metaobject.
4283 int which_arg;
4284 for (which_arg = 0; which_arg <= 1; which_arg++) {
4285 Node* arg = args[which_arg];
4286 arg = null_check(arg);
4287 if (stopped()) break;
4288 args[which_arg] = arg;
4289
4290 Node* p = basic_plus_adr(arg, class_klass_offset);
4291 Node* kls = LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, adr_type, kls_type);
4292 klasses[which_arg] = _gvn.transform(kls);
4293 }
4294
4295 // Having loaded both klasses, test each for null.
4296 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4297 for (which_arg = 0; which_arg <= 1; which_arg++) {
4298 Node* kls = klasses[which_arg];
4299 Node* null_ctl = top();
4300 kls = null_check_oop(kls, &null_ctl, never_see_null);
4301 if (which_arg == 0) {
4302 prim_region->init_req(1, null_ctl);
4303 } else {
4304 region->init_req(_prim_1_path, null_ctl);
4305 }
4306 if (stopped()) break;
4307 klasses[which_arg] = kls;
4308 }
4309
4310 if (!stopped()) {
4311 // now we have two reference types, in klasses[0..1]
4312 Node* subk = klasses[1]; // the argument to isAssignableFrom
4313 Node* superk = klasses[0]; // the receiver
4314 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4315 // If superc is an inline mirror, we also need to check if superc == subc because LMyValue
4316 // is not a subtype of QMyValue but due to subk == superk the subtype check will pass.
4317 generate_fair_guard(is_val_mirror(args[0]), prim_region);
4318 // now we have a successful reference subtype check
4319 region->set_req(_ref_subtype_path, control());
4320 }
4321
4322 // If both operands are primitive (both klasses null), then
4323 // we must return true when they are identical primitives.
4324 // It is convenient to test this after the first null klass check.
4325 // This path is also used if superc is a value mirror.
4326 set_control(_gvn.transform(prim_region));
4327 if (!stopped()) {
4328 // Since superc is primitive, make a guard for the superc==subc case.
4329 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4330 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4331 generate_fair_guard(bol_eq, region);
4332 if (region->req() == PATH_LIMIT+1) {
4333 // A guard was added. If the added guard is taken, superc==subc.
4334 region->swap_edges(PATH_LIMIT, _prim_same_path);
4335 region->del_req(PATH_LIMIT);
4336 }
4337 region->set_req(_prim_0_path, control()); // Not equal after all.
4338 }
4339
4340 // these are the only paths that produce 'true':
4341 phi->set_req(_prim_same_path, intcon(1));
4342 phi->set_req(_ref_subtype_path, intcon(1));
4343
4344 // pull together the cases:
4345 assert(region->req() == PATH_LIMIT, "sane region");
4346 for (uint i = 1; i < region->req(); i++) {
4347 Node* ctl = region->in(i);
4348 if (ctl == nullptr || ctl == top()) {
4349 region->set_req(i, top());
4350 phi ->set_req(i, top());
4351 } else if (phi->in(i) == nullptr) {
4352 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4353 }
4354 }
4355
4356 set_control(_gvn.transform(region));
4357 set_result(_gvn.transform(phi));
4358 return true;
4359 }
4360
4361 //---------------------generate_array_guard_common------------------------
4362 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind) {
4363
4364 if (stopped()) {
4365 return nullptr;
4366 }
4367
4368 // Like generate_guard, adds a new path onto the region.
4369 jint layout_con = 0;
4370 Node* layout_val = get_layout_helper(kls, layout_con);
4371 if (layout_val == nullptr) {
4372 bool query = 0;
4373 switch(kind) {
4374 case ObjectArray: query = Klass::layout_helper_is_objArray(layout_con); break;
4375 case NonObjectArray: query = !Klass::layout_helper_is_objArray(layout_con); break;
4376 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4377 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4378 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4379 default:
4380 ShouldNotReachHere();
4381 }
4382 if (!query) {
4383 return nullptr; // never a branch
4384 } else { // always a branch
4385 Node* always_branch = control();
4386 if (region != nullptr)
4387 region->add_req(always_branch);
4388 set_control(top());
4389 return always_branch;
4390 }
4391 }
4392 unsigned int value = 0;
4393 BoolTest::mask btest = BoolTest::illegal;
4394 switch(kind) {
4395 case ObjectArray:
4396 case NonObjectArray: {
4397 value = Klass::_lh_array_tag_obj_value;
4398 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4399 btest = (kind == ObjectArray) ? BoolTest::eq : BoolTest::ne;
4400 break;
4401 }
4402 case TypeArray: {
4403 value = Klass::_lh_array_tag_type_value;
4404 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4405 btest = BoolTest::eq;
4406 break;
4407 }
4408 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4409 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4410 default:
4411 ShouldNotReachHere();
4412 }
4413 // Now test the correct condition.
4414 jint nval = (jint)value;
4415 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4416 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4417 return generate_fair_guard(bol, region);
4418 }
4419
4420
4421 //-----------------------inline_native_newArray--------------------------
4422 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
4423 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4424 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4425 Node* mirror;
4426 Node* count_val;
4427 if (uninitialized) {
4428 null_check_receiver();
4429 mirror = argument(1);
4430 count_val = argument(2);
4431 } else {
4432 mirror = argument(0);
4433 count_val = argument(1);
4434 }
4435
4436 mirror = null_check(mirror);
4437 // If mirror or obj is dead, only null-path is taken.
4438 if (stopped()) return true;
4439
4440 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4441 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4442 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4443 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4444 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4445
4446 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4447 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4448 result_reg, _slow_path);
4449 Node* normal_ctl = control();
4450 Node* no_array_ctl = result_reg->in(_slow_path);
4451
4452 // Generate code for the slow case. We make a call to newArray().
4453 set_control(no_array_ctl);
4454 if (!stopped()) {
4455 // Either the input type is void.class, or else the
4456 // array klass has not yet been cached. Either the
4457 // ensuing call will throw an exception, or else it
4458 // will cache the array klass for next time.
4459 PreserveJVMState pjvms(this);
4460 CallJavaNode* slow_call = nullptr;
4461 if (uninitialized) {
4462 // Generate optimized virtual call (holder class 'Unsafe' is final)
4463 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4464 } else {
4465 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4466 }
4467 Node* slow_result = set_results_for_java_call(slow_call);
4468 // this->control() comes from set_results_for_java_call
4469 result_reg->set_req(_slow_path, control());
4470 result_val->set_req(_slow_path, slow_result);
4471 result_io ->set_req(_slow_path, i_o());
4472 result_mem->set_req(_slow_path, reset_memory());
4473 }
4474
4475 set_control(normal_ctl);
4476 if (!stopped()) {
4477 // Normal case: The array type has been cached in the java.lang.Class.
4478 // The following call works fine even if the array type is polymorphic.
4479 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4480 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4481 result_reg->init_req(_normal_path, control());
4482 result_val->init_req(_normal_path, obj);
4483 result_io ->init_req(_normal_path, i_o());
4484 result_mem->init_req(_normal_path, reset_memory());
4485
4486 if (uninitialized) {
4487 // Mark the allocation so that zeroing is skipped
4488 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4489 alloc->maybe_set_complete(&_gvn);
4490 }
4491 }
4492
4493 // Return the combined state.
4494 set_i_o( _gvn.transform(result_io) );
4495 set_all_memory( _gvn.transform(result_mem));
4496
4497 C->set_has_split_ifs(true); // Has chance for split-if optimization
4498 set_result(result_reg, result_val);
4499 return true;
4500 }
4501
4502 //----------------------inline_native_getLength--------------------------
4503 // public static native int java.lang.reflect.Array.getLength(Object array);
4504 bool LibraryCallKit::inline_native_getLength() {
4505 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4506
4507 Node* array = null_check(argument(0));
4508 // If array is dead, only null-path is taken.
4509 if (stopped()) return true;
4510
4511 // Deoptimize if it is a non-array.
4512 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr);
4513
4514 if (non_array != nullptr) {
4515 PreserveJVMState pjvms(this);
4516 set_control(non_array);
4517 uncommon_trap(Deoptimization::Reason_intrinsic,
4518 Deoptimization::Action_maybe_recompile);
4519 }
4520
4521 // If control is dead, only non-array-path is taken.
4522 if (stopped()) return true;
4523
4524 // The works fine even if the array type is polymorphic.
4525 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4526 Node* result = load_array_length(array);
4527
4528 C->set_has_split_ifs(true); // Has chance for split-if optimization
4529 set_result(result);
4530 return true;
4531 }
4532
4533 //------------------------inline_array_copyOf----------------------------
4534 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4535 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4536 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4537 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4538
4539 // Get the arguments.
4540 Node* original = argument(0);
4541 Node* start = is_copyOfRange? argument(1): intcon(0);
4542 Node* end = is_copyOfRange? argument(2): argument(1);
4543 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4544
4545 Node* newcopy = nullptr;
4546
4547 // Set the original stack and the reexecute bit for the interpreter to reexecute
4548 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4549 { PreserveReexecuteState preexecs(this);
4550 jvms()->set_should_reexecute(true);
4551
4552 array_type_mirror = null_check(array_type_mirror);
4553 original = null_check(original);
4554
4555 // Check if a null path was taken unconditionally.
4556 if (stopped()) return true;
4557
4558 Node* orig_length = load_array_length(original);
4559
4560 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4561 klass_node = null_check(klass_node);
4562
4563 RegionNode* bailout = new RegionNode(1);
4564 record_for_igvn(bailout);
4565
4566 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4567 // Bail out if that is so.
4568 // Inline type array may have object field that would require a
4569 // write barrier. Conservatively, go to slow path.
4570 // TODO 8251971: Optimize for the case when flat src/dst are later found
4571 // to not contain oops (i.e., move this check to the macro expansion phase).
4572 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4573 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
4574 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
4575 bool exclude_flat = UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
4576 // Can src array be flat and contain oops?
4577 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
4578 // Can dest array be flat and contain oops?
4579 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
4580 Node* not_objArray = exclude_flat ? generate_non_objArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
4581 if (not_objArray != nullptr) {
4582 // Improve the klass node's type from the new optimistic assumption:
4583 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4584 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0));
4585 Node* cast = new CastPPNode(klass_node, akls);
4586 cast->init_req(0, control());
4587 klass_node = _gvn.transform(cast);
4588 }
4589
4590 // Bail out if either start or end is negative.
4591 generate_negative_guard(start, bailout, &start);
4592 generate_negative_guard(end, bailout, &end);
4593
4594 Node* length = end;
4595 if (_gvn.type(start) != TypeInt::ZERO) {
4596 length = _gvn.transform(new SubINode(end, start));
4597 }
4598
4599 // Bail out if length is negative.
4600 // Without this the new_array would throw
4601 // NegativeArraySizeException but IllegalArgumentException is what
4602 // should be thrown
4603 generate_negative_guard(length, bailout, &length);
4604
4605 // Handle inline type arrays
4606 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
4607 if (!stopped()) {
4608 orig_t = _gvn.type(original)->isa_aryptr();
4609 if (orig_t != nullptr && orig_t->is_flat()) {
4610 // Src is flat, check that dest is flat as well
4611 if (exclude_flat) {
4612 // Dest can't be flat, bail out
4613 bailout->add_req(control());
4614 set_control(top());
4615 } else {
4616 generate_fair_guard(flat_array_test(klass_node, /* flat = */ false), bailout);
4617 }
4618 } else if (UseFlatArray && (orig_t == nullptr || !orig_t->is_not_flat()) &&
4619 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
4620 ((!tklass->is_flat() && tklass->can_be_inline_array()) || !can_validate)) {
4621 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
4622 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
4623 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
4624 if (orig_t != nullptr) {
4625 orig_t = orig_t->cast_to_not_flat();
4626 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
4627 }
4628 }
4629 if (!can_validate) {
4630 // No validation. The subtype check emitted at macro expansion time will not go to the slow
4631 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
4632 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
4633 generate_fair_guard(null_free_array_test(klass_node), bailout);
4634 }
4635 }
4636
4637 if (bailout->req() > 1) {
4638 PreserveJVMState pjvms(this);
4639 set_control(_gvn.transform(bailout));
4640 uncommon_trap(Deoptimization::Reason_intrinsic,
4641 Deoptimization::Action_maybe_recompile);
4642 }
4643
4644 if (!stopped()) {
4645 // How many elements will we copy from the original?
4646 // The answer is MinI(orig_length - start, length).
4647 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4648 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4649
4650 // Generate a direct call to the right arraycopy function(s).
4651 // We know the copy is disjoint but we might not know if the
4652 // oop stores need checking.
4653 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4654 // This will fail a store-check if x contains any non-nulls.
4655
4656 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4657 // loads/stores but it is legal only if we're sure the
4658 // Arrays.copyOf would succeed. So we need all input arguments
4659 // to the copyOf to be validated, including that the copy to the
4660 // new array won't trigger an ArrayStoreException. That subtype
4661 // check can be optimized if we know something on the type of
4662 // the input array from type speculation.
4663 if (_gvn.type(klass_node)->singleton()) {
4664 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4665 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4666
4667 int test = C->static_subtype_check(superk, subk);
4668 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4669 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4670 if (t_original->speculative_type() != nullptr) {
4671 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4672 }
4673 }
4674 }
4675
4676 bool validated = false;
4677 // Reason_class_check rather than Reason_intrinsic because we
4678 // want to intrinsify even if this traps.
4679 if (can_validate) {
4680 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4681
4682 if (not_subtype_ctrl != top()) {
4683 PreserveJVMState pjvms(this);
4684 set_control(not_subtype_ctrl);
4685 uncommon_trap(Deoptimization::Reason_class_check,
4686 Deoptimization::Action_make_not_entrant);
4687 assert(stopped(), "Should be stopped");
4688 }
4689 validated = true;
4690 }
4691
4692 if (!stopped()) {
4693 newcopy = new_array(klass_node, length, 0); // no arguments to push
4694
4695 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
4696 load_object_klass(original), klass_node);
4697 if (!is_copyOfRange) {
4698 ac->set_copyof(validated);
4699 } else {
4700 ac->set_copyofrange(validated);
4701 }
4702 Node* n = _gvn.transform(ac);
4703 if (n == ac) {
4704 ac->connect_outputs(this);
4705 } else {
4706 assert(validated, "shouldn't transform if all arguments not validated");
4707 set_all_memory(n);
4708 }
4709 }
4710 }
4711 } // original reexecute is set back here
4712
4713 C->set_has_split_ifs(true); // Has chance for split-if optimization
4714 if (!stopped()) {
4715 set_result(newcopy);
4716 }
4717 return true;
4718 }
4719
4720
4721 //----------------------generate_virtual_guard---------------------------
4722 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4723 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4724 RegionNode* slow_region) {
4725 ciMethod* method = callee();
4726 int vtable_index = method->vtable_index();
4727 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4728 "bad index %d", vtable_index);
4729 // Get the Method* out of the appropriate vtable entry.
4730 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4731 vtable_index*vtableEntry::size_in_bytes() +
4732 in_bytes(vtableEntry::method_offset());
4733 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4734 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4735
4736 // Compare the target method with the expected method (e.g., Object.hashCode).
4737 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4738
4739 Node* native_call = makecon(native_call_addr);
4740 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4741 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4742
4743 return generate_slow_guard(test_native, slow_region);
4744 }
4745
4746 //-----------------------generate_method_call----------------------------
4747 // Use generate_method_call to make a slow-call to the real
4748 // method if the fast path fails. An alternative would be to
4749 // use a stub like OptoRuntime::slow_arraycopy_Java.
4750 // This only works for expanding the current library call,
4751 // not another intrinsic. (E.g., don't use this for making an
4752 // arraycopy call inside of the copyOf intrinsic.)
4753 CallJavaNode*
4754 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4755 // When compiling the intrinsic method itself, do not use this technique.
4756 guarantee(callee() != C->method(), "cannot make slow-call to self");
4757
4758 ciMethod* method = callee();
4759 // ensure the JVMS we have will be correct for this call
4760 guarantee(method_id == method->intrinsic_id(), "must match");
4761
4762 const TypeFunc* tf = TypeFunc::make(method);
4763 if (res_not_null) {
4764 assert(tf->return_type() == T_OBJECT, "");
4765 const TypeTuple* range = tf->range_cc();
4766 const Type** fields = TypeTuple::fields(range->cnt());
4767 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4768 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4769 tf = TypeFunc::make(tf->domain_cc(), new_range);
4770 }
4771 CallJavaNode* slow_call;
4772 if (is_static) {
4773 assert(!is_virtual, "");
4774 slow_call = new CallStaticJavaNode(C, tf,
4775 SharedRuntime::get_resolve_static_call_stub(), method);
4776 } else if (is_virtual) {
4777 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4778 int vtable_index = Method::invalid_vtable_index;
4779 if (UseInlineCaches) {
4780 // Suppress the vtable call
4781 } else {
4782 // hashCode and clone are not a miranda methods,
4783 // so the vtable index is fixed.
4784 // No need to use the linkResolver to get it.
4785 vtable_index = method->vtable_index();
4786 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4787 "bad index %d", vtable_index);
4788 }
4789 slow_call = new CallDynamicJavaNode(tf,
4790 SharedRuntime::get_resolve_virtual_call_stub(),
4791 method, vtable_index);
4792 } else { // neither virtual nor static: opt_virtual
4793 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4794 slow_call = new CallStaticJavaNode(C, tf,
4795 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4796 slow_call->set_optimized_virtual(true);
4797 }
4798 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4799 // To be able to issue a direct call (optimized virtual or virtual)
4800 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4801 // about the method being invoked should be attached to the call site to
4802 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4803 slow_call->set_override_symbolic_info(true);
4804 }
4805 set_arguments_for_java_call(slow_call);
4806 set_edges_for_java_call(slow_call);
4807 return slow_call;
4808 }
4809
4810
4811 /**
4812 * Build special case code for calls to hashCode on an object. This call may
4813 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4814 * slightly different code.
4815 */
4816 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4817 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4818 assert(!(is_virtual && is_static), "either virtual, special, or static");
4819
4820 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4821
4822 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4823 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4824 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4825 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4826 Node* obj = argument(0);
4827
4828 if (gvn().type(obj)->is_inlinetypeptr()) {
4829 return false;
4830 }
4831
4832 if (!is_static) {
4833 // Check for hashing null object
4834 obj = null_check_receiver();
4835 if (stopped()) return true; // unconditionally null
4836 result_reg->init_req(_null_path, top());
4837 result_val->init_req(_null_path, top());
4838 } else {
4839 // Do a null check, and return zero if null.
4840 // System.identityHashCode(null) == 0
4841 Node* null_ctl = top();
4842 obj = null_check_oop(obj, &null_ctl);
4843 result_reg->init_req(_null_path, null_ctl);
4844 result_val->init_req(_null_path, _gvn.intcon(0));
4845 }
4846
4847 // Unconditionally null? Then return right away.
4848 if (stopped()) {
4849 set_control( result_reg->in(_null_path));
4850 if (!stopped())
4851 set_result(result_val->in(_null_path));
4852 return true;
4853 }
4854
4855 // We only go to the fast case code if we pass a number of guards. The
4856 // paths which do not pass are accumulated in the slow_region.
4857 RegionNode* slow_region = new RegionNode(1);
4858 record_for_igvn(slow_region);
4859
4860 // If this is a virtual call, we generate a funny guard. We pull out
4861 // the vtable entry corresponding to hashCode() from the target object.
4862 // If the target method which we are calling happens to be the native
4863 // Object hashCode() method, we pass the guard. We do not need this
4864 // guard for non-virtual calls -- the caller is known to be the native
4865 // Object hashCode().
4866 if (is_virtual) {
4867 // After null check, get the object's klass.
4868 Node* obj_klass = load_object_klass(obj);
4869 generate_virtual_guard(obj_klass, slow_region);
4870 }
4871
4872 // Get the header out of the object, use LoadMarkNode when available
4873 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4874 // The control of the load must be null. Otherwise, the load can move before
4875 // the null check after castPP removal.
4876 Node* no_ctrl = nullptr;
4877 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4878
4879 // Test the header to see if it is unlocked.
4880 // This also serves as guard against inline types
4881 Node *lock_mask = _gvn.MakeConX(markWord::inline_type_mask_in_place);
4882 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4883 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4884 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4885 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4886
4887 generate_slow_guard(test_unlocked, slow_region);
4888
4889 // Get the hash value and check to see that it has been properly assigned.
4890 // We depend on hash_mask being at most 32 bits and avoid the use of
4891 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4892 // vm: see markWord.hpp.
4893 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4894 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4895 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4896 // This hack lets the hash bits live anywhere in the mark object now, as long
4897 // as the shift drops the relevant bits into the low 32 bits. Note that
4898 // Java spec says that HashCode is an int so there's no point in capturing
4899 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4900 hshifted_header = ConvX2I(hshifted_header);
4901 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4902
4903 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4904 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4905 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4906
4907 generate_slow_guard(test_assigned, slow_region);
4908
4909 Node* init_mem = reset_memory();
4910 // fill in the rest of the null path:
4911 result_io ->init_req(_null_path, i_o());
4912 result_mem->init_req(_null_path, init_mem);
4913
4914 result_val->init_req(_fast_path, hash_val);
4915 result_reg->init_req(_fast_path, control());
4916 result_io ->init_req(_fast_path, i_o());
4917 result_mem->init_req(_fast_path, init_mem);
4918
4919 // Generate code for the slow case. We make a call to hashCode().
4920 set_control(_gvn.transform(slow_region));
4921 if (!stopped()) {
4922 // No need for PreserveJVMState, because we're using up the present state.
4923 set_all_memory(init_mem);
4924 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4925 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4926 Node* slow_result = set_results_for_java_call(slow_call);
4927 // this->control() comes from set_results_for_java_call
4928 result_reg->init_req(_slow_path, control());
4929 result_val->init_req(_slow_path, slow_result);
4930 result_io ->set_req(_slow_path, i_o());
4931 result_mem ->set_req(_slow_path, reset_memory());
4932 }
4933
4934 // Return the combined state.
4935 set_i_o( _gvn.transform(result_io) );
4936 set_all_memory( _gvn.transform(result_mem));
4937
4938 set_result(result_reg, result_val);
4939 return true;
4940 }
4941
4942 //---------------------------inline_native_getClass----------------------------
4943 // public final native Class<?> java.lang.Object.getClass();
4944 //
4945 // Build special case code for calls to getClass on an object.
4946 bool LibraryCallKit::inline_native_getClass() {
4947 Node* obj = argument(0);
4948 if (obj->is_InlineType()) {
4949 const Type* t = _gvn.type(obj);
4950 if (t->maybe_null()) {
4951 null_check(obj);
4952 }
4953 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
4954 return true;
4955 }
4956 obj = null_check_receiver();
4957 if (stopped()) return true;
4958 set_result(load_mirror_from_klass(load_object_klass(obj)));
4959 return true;
4960 }
4961
4962 //-----------------inline_native_Reflection_getCallerClass---------------------
4963 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4964 //
4965 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4966 //
4967 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4968 // in that it must skip particular security frames and checks for
4969 // caller sensitive methods.
4970 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4971 #ifndef PRODUCT
4972 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4973 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4974 }
4975 #endif
4976
4977 if (!jvms()->has_method()) {
4978 #ifndef PRODUCT
4979 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4980 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4981 }
4982 #endif
4983 return false;
4984 }
4985
4986 // Walk back up the JVM state to find the caller at the required
4987 // depth.
4988 JVMState* caller_jvms = jvms();
4989
4990 // Cf. JVM_GetCallerClass
4991 // NOTE: Start the loop at depth 1 because the current JVM state does
4992 // not include the Reflection.getCallerClass() frame.
4993 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4994 ciMethod* m = caller_jvms->method();
4995 switch (n) {
4996 case 0:
4997 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4998 break;
4999 case 1:
5000 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5001 if (!m->caller_sensitive()) {
5002 #ifndef PRODUCT
5003 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5004 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5005 }
5006 #endif
5007 return false; // bail-out; let JVM_GetCallerClass do the work
5008 }
5009 break;
5010 default:
5011 if (!m->is_ignored_by_security_stack_walk()) {
5012 // We have reached the desired frame; return the holder class.
5013 // Acquire method holder as java.lang.Class and push as constant.
5014 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5015 ciInstance* caller_mirror = caller_klass->java_mirror();
5016 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5017
5018 #ifndef PRODUCT
5019 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5020 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());
5021 tty->print_cr(" JVM state at this point:");
5022 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5023 ciMethod* m = jvms()->of_depth(i)->method();
5024 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5025 }
5026 }
5027 #endif
5028 return true;
5029 }
5030 break;
5031 }
5032 }
5033
5034 #ifndef PRODUCT
5035 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5036 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5037 tty->print_cr(" JVM state at this point:");
5038 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5039 ciMethod* m = jvms()->of_depth(i)->method();
5040 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5041 }
5042 }
5043 #endif
5044
5045 return false; // bail-out; let JVM_GetCallerClass do the work
5046 }
5047
5048 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5049 Node* arg = argument(0);
5050 Node* result = nullptr;
5051
5052 switch (id) {
5053 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5054 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5055 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5056 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5057 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5058 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5059
5060 case vmIntrinsics::_doubleToLongBits: {
5061 // two paths (plus control) merge in a wood
5062 RegionNode *r = new RegionNode(3);
5063 Node *phi = new PhiNode(r, TypeLong::LONG);
5064
5065 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5066 // Build the boolean node
5067 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5068
5069 // Branch either way.
5070 // NaN case is less traveled, which makes all the difference.
5071 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5072 Node *opt_isnan = _gvn.transform(ifisnan);
5073 assert( opt_isnan->is_If(), "Expect an IfNode");
5074 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5075 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5076
5077 set_control(iftrue);
5078
5079 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5080 Node *slow_result = longcon(nan_bits); // return NaN
5081 phi->init_req(1, _gvn.transform( slow_result ));
5082 r->init_req(1, iftrue);
5083
5084 // Else fall through
5085 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5086 set_control(iffalse);
5087
5088 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5089 r->init_req(2, iffalse);
5090
5091 // Post merge
5092 set_control(_gvn.transform(r));
5093 record_for_igvn(r);
5094
5095 C->set_has_split_ifs(true); // Has chance for split-if optimization
5096 result = phi;
5097 assert(result->bottom_type()->isa_long(), "must be");
5098 break;
5099 }
5100
5101 case vmIntrinsics::_floatToIntBits: {
5102 // two paths (plus control) merge in a wood
5103 RegionNode *r = new RegionNode(3);
5104 Node *phi = new PhiNode(r, TypeInt::INT);
5105
5106 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5107 // Build the boolean node
5108 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5109
5110 // Branch either way.
5111 // NaN case is less traveled, which makes all the difference.
5112 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5113 Node *opt_isnan = _gvn.transform(ifisnan);
5114 assert( opt_isnan->is_If(), "Expect an IfNode");
5115 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5116 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5117
5118 set_control(iftrue);
5119
5120 static const jint nan_bits = 0x7fc00000;
5121 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5122 phi->init_req(1, _gvn.transform( slow_result ));
5123 r->init_req(1, iftrue);
5124
5125 // Else fall through
5126 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5127 set_control(iffalse);
5128
5129 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5130 r->init_req(2, iffalse);
5131
5132 // Post merge
5133 set_control(_gvn.transform(r));
5134 record_for_igvn(r);
5135
5136 C->set_has_split_ifs(true); // Has chance for split-if optimization
5137 result = phi;
5138 assert(result->bottom_type()->isa_int(), "must be");
5139 break;
5140 }
5141
5142 default:
5143 fatal_unexpected_iid(id);
5144 break;
5145 }
5146 set_result(_gvn.transform(result));
5147 return true;
5148 }
5149
5150 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5151 Node* arg = argument(0);
5152 Node* result = nullptr;
5153
5154 switch (id) {
5155 case vmIntrinsics::_floatIsInfinite:
5156 result = new IsInfiniteFNode(arg);
5157 break;
5158 case vmIntrinsics::_floatIsFinite:
5159 result = new IsFiniteFNode(arg);
5160 break;
5161 case vmIntrinsics::_doubleIsInfinite:
5162 result = new IsInfiniteDNode(arg);
5163 break;
5164 case vmIntrinsics::_doubleIsFinite:
5165 result = new IsFiniteDNode(arg);
5166 break;
5167 default:
5168 fatal_unexpected_iid(id);
5169 break;
5170 }
5171 set_result(_gvn.transform(result));
5172 return true;
5173 }
5174
5175 //----------------------inline_unsafe_copyMemory-------------------------
5176 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5177
5178 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5179 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5180 const Type* base_t = gvn.type(base);
5181
5182 bool in_native = (base_t == TypePtr::NULL_PTR);
5183 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5184 bool is_mixed = !in_heap && !in_native;
5185
5186 if (is_mixed) {
5187 return true; // mixed accesses can touch both on-heap and off-heap memory
5188 }
5189 if (in_heap) {
5190 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5191 if (!is_prim_array) {
5192 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5193 // there's not enough type information available to determine proper memory slice for it.
5194 return true;
5195 }
5196 }
5197 return false;
5198 }
5199
5200 bool LibraryCallKit::inline_unsafe_copyMemory() {
5201 if (callee()->is_static()) return false; // caller must have the capability!
5202 null_check_receiver(); // null-check receiver
5203 if (stopped()) return true;
5204
5205 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5206
5207 Node* src_base = argument(1); // type: oop
5208 Node* src_off = ConvL2X(argument(2)); // type: long
5209 Node* dst_base = argument(4); // type: oop
5210 Node* dst_off = ConvL2X(argument(5)); // type: long
5211 Node* size = ConvL2X(argument(7)); // type: long
5212
5213 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5214 "fieldOffset must be byte-scaled");
5215
5216 Node* src_addr = make_unsafe_address(src_base, src_off);
5217 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5218
5219 Node* thread = _gvn.transform(new ThreadLocalNode());
5220 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5221 BasicType doing_unsafe_access_bt = T_BYTE;
5222 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5223
5224 // update volatile field
5225 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5226
5227 int flags = RC_LEAF | RC_NO_FP;
5228
5229 const TypePtr* dst_type = TypePtr::BOTTOM;
5230
5231 // Adjust memory effects of the runtime call based on input values.
5232 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5233 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5234 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5235
5236 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5237 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5238 flags |= RC_NARROW_MEM; // narrow in memory
5239 }
5240 }
5241
5242 // Call it. Note that the length argument is not scaled.
5243 make_runtime_call(flags,
5244 OptoRuntime::fast_arraycopy_Type(),
5245 StubRoutines::unsafe_arraycopy(),
5246 "unsafe_arraycopy",
5247 dst_type,
5248 src_addr, dst_addr, size XTOP);
5249
5250 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5251
5252 return true;
5253 }
5254
5255 #undef XTOP
5256
5257 //----------------------inline_unsafe_isFlattenedArray-------------------
5258 // public native boolean Unsafe.isFlattenedArray(Class<?> arrayClass);
5259 // This intrinsic exploits assumptions made by the native implementation
5260 // (arrayClass is neither null nor primitive) to avoid unnecessary null checks.
5261 bool LibraryCallKit::inline_unsafe_isFlattenedArray() {
5262 Node* cls = argument(1);
5263 Node* p = basic_plus_adr(cls, java_lang_Class::klass_offset());
5264 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p,
5265 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT));
5266 Node* result = flat_array_test(kls);
5267 set_result(result);
5268 return true;
5269 }
5270
5271 //------------------------clone_coping-----------------------------------
5272 // Helper function for inline_native_clone.
5273 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5274 assert(obj_size != nullptr, "");
5275 Node* raw_obj = alloc_obj->in(1);
5276 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5277
5278 AllocateNode* alloc = nullptr;
5279 if (ReduceBulkZeroing) {
5280 // We will be completely responsible for initializing this object -
5281 // mark Initialize node as complete.
5282 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5283 // The object was just allocated - there should be no any stores!
5284 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5285 // Mark as complete_with_arraycopy so that on AllocateNode
5286 // expansion, we know this AllocateNode is initialized by an array
5287 // copy and a StoreStore barrier exists after the array copy.
5288 alloc->initialization()->set_complete_with_arraycopy();
5289 }
5290
5291 Node* size = _gvn.transform(obj_size);
5292 access_clone(obj, alloc_obj, size, is_array);
5293
5294 // Do not let reads from the cloned object float above the arraycopy.
5295 if (alloc != nullptr) {
5296 // Do not let stores that initialize this object be reordered with
5297 // a subsequent store that would make this object accessible by
5298 // other threads.
5299 // Record what AllocateNode this StoreStore protects so that
5300 // escape analysis can go from the MemBarStoreStoreNode to the
5301 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5302 // based on the escape status of the AllocateNode.
5303 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5304 } else {
5305 insert_mem_bar(Op_MemBarCPUOrder);
5306 }
5307 }
5308
5309 //------------------------inline_native_clone----------------------------
5310 // protected native Object java.lang.Object.clone();
5311 //
5312 // Here are the simple edge cases:
5313 // null receiver => normal trap
5314 // virtual and clone was overridden => slow path to out-of-line clone
5315 // not cloneable or finalizer => slow path to out-of-line Object.clone
5316 //
5317 // The general case has two steps, allocation and copying.
5318 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5319 //
5320 // Copying also has two cases, oop arrays and everything else.
5321 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5322 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5323 //
5324 // These steps fold up nicely if and when the cloned object's klass
5325 // can be sharply typed as an object array, a type array, or an instance.
5326 //
5327 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5328 PhiNode* result_val;
5329
5330 // Set the reexecute bit for the interpreter to reexecute
5331 // the bytecode that invokes Object.clone if deoptimization happens.
5332 { PreserveReexecuteState preexecs(this);
5333 jvms()->set_should_reexecute(true);
5334
5335 Node* obj = argument(0);
5336 obj = null_check_receiver();
5337 if (stopped()) return true;
5338
5339 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5340
5341 // If we are going to clone an instance, we need its exact type to
5342 // know the number and types of fields to convert the clone to
5343 // loads/stores. Maybe a speculative type can help us.
5344 if (!obj_type->klass_is_exact() &&
5345 obj_type->speculative_type() != nullptr &&
5346 obj_type->speculative_type()->is_instance_klass() &&
5347 !obj_type->speculative_type()->is_inlinetype()) {
5348 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5349 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5350 !spec_ik->has_injected_fields()) {
5351 if (!obj_type->isa_instptr() ||
5352 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5353 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5354 }
5355 }
5356 }
5357
5358 // Conservatively insert a memory barrier on all memory slices.
5359 // Do not let writes into the original float below the clone.
5360 insert_mem_bar(Op_MemBarCPUOrder);
5361
5362 // paths into result_reg:
5363 enum {
5364 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5365 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5366 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5367 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5368 PATH_LIMIT
5369 };
5370 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5371 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5372 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5373 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5374 record_for_igvn(result_reg);
5375
5376 Node* obj_klass = load_object_klass(obj);
5377 // We only go to the fast case code if we pass a number of guards.
5378 // The paths which do not pass are accumulated in the slow_region.
5379 RegionNode* slow_region = new RegionNode(1);
5380 record_for_igvn(slow_region);
5381
5382 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr);
5383 if (array_ctl != nullptr) {
5384 // It's an array.
5385 PreserveJVMState pjvms(this);
5386 set_control(array_ctl);
5387
5388 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5389 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
5390 if (UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
5391 obj_type->can_be_inline_array() &&
5392 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
5393 // Flat inline type array may have object field that would require a
5394 // write barrier. Conservatively, go to slow path.
5395 generate_fair_guard(flat_array_test(obj_klass), slow_region);
5396 }
5397
5398 if (!stopped()) {
5399 Node* obj_length = load_array_length(obj);
5400 Node* obj_size = nullptr;
5401 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, /*deoptimize_on_exception=*/true);
5402
5403 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5404 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5405 // If it is an oop array, it requires very special treatment,
5406 // because gc barriers are required when accessing the array.
5407 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5408 if (is_obja != nullptr) {
5409 PreserveJVMState pjvms2(this);
5410 set_control(is_obja);
5411 // Generate a direct call to the right arraycopy function(s).
5412 // Clones are always tightly coupled.
5413 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5414 ac->set_clone_oop_array();
5415 Node* n = _gvn.transform(ac);
5416 assert(n == ac, "cannot disappear");
5417 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5418
5419 result_reg->init_req(_objArray_path, control());
5420 result_val->init_req(_objArray_path, alloc_obj);
5421 result_i_o ->set_req(_objArray_path, i_o());
5422 result_mem ->set_req(_objArray_path, reset_memory());
5423 }
5424 }
5425 // Otherwise, there are no barriers to worry about.
5426 // (We can dispense with card marks if we know the allocation
5427 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5428 // causes the non-eden paths to take compensating steps to
5429 // simulate a fresh allocation, so that no further
5430 // card marks are required in compiled code to initialize
5431 // the object.)
5432
5433 if (!stopped()) {
5434 copy_to_clone(obj, alloc_obj, obj_size, true);
5435
5436 // Present the results of the copy.
5437 result_reg->init_req(_array_path, control());
5438 result_val->init_req(_array_path, alloc_obj);
5439 result_i_o ->set_req(_array_path, i_o());
5440 result_mem ->set_req(_array_path, reset_memory());
5441 }
5442 }
5443 }
5444
5445 if (!stopped()) {
5446 // It's an instance (we did array above). Make the slow-path tests.
5447 // If this is a virtual call, we generate a funny guard. We grab
5448 // the vtable entry corresponding to clone() from the target object.
5449 // If the target method which we are calling happens to be the
5450 // Object clone() method, we pass the guard. We do not need this
5451 // guard for non-virtual calls; the caller is known to be the native
5452 // Object clone().
5453 if (is_virtual) {
5454 generate_virtual_guard(obj_klass, slow_region);
5455 }
5456
5457 // The object must be easily cloneable and must not have a finalizer.
5458 // Both of these conditions may be checked in a single test.
5459 // We could optimize the test further, but we don't care.
5460 generate_access_flags_guard(obj_klass,
5461 // Test both conditions:
5462 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
5463 // Must be cloneable but not finalizer:
5464 JVM_ACC_IS_CLONEABLE_FAST,
5465 slow_region);
5466 }
5467
5468 if (!stopped()) {
5469 // It's an instance, and it passed the slow-path tests.
5470 PreserveJVMState pjvms(this);
5471 Node* obj_size = nullptr;
5472 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5473 // is reexecuted if deoptimization occurs and there could be problems when merging
5474 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5475 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5476
5477 copy_to_clone(obj, alloc_obj, obj_size, false);
5478
5479 // Present the results of the slow call.
5480 result_reg->init_req(_instance_path, control());
5481 result_val->init_req(_instance_path, alloc_obj);
5482 result_i_o ->set_req(_instance_path, i_o());
5483 result_mem ->set_req(_instance_path, reset_memory());
5484 }
5485
5486 // Generate code for the slow case. We make a call to clone().
5487 set_control(_gvn.transform(slow_region));
5488 if (!stopped()) {
5489 PreserveJVMState pjvms(this);
5490 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5491 // We need to deoptimize on exception (see comment above)
5492 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5493 // this->control() comes from set_results_for_java_call
5494 result_reg->init_req(_slow_path, control());
5495 result_val->init_req(_slow_path, slow_result);
5496 result_i_o ->set_req(_slow_path, i_o());
5497 result_mem ->set_req(_slow_path, reset_memory());
5498 }
5499
5500 // Return the combined state.
5501 set_control( _gvn.transform(result_reg));
5502 set_i_o( _gvn.transform(result_i_o));
5503 set_all_memory( _gvn.transform(result_mem));
5504 } // original reexecute is set back here
5505
5506 set_result(_gvn.transform(result_val));
5507 return true;
5508 }
5509
5510 // If we have a tightly coupled allocation, the arraycopy may take care
5511 // of the array initialization. If one of the guards we insert between
5512 // the allocation and the arraycopy causes a deoptimization, an
5513 // uninitialized array will escape the compiled method. To prevent that
5514 // we set the JVM state for uncommon traps between the allocation and
5515 // the arraycopy to the state before the allocation so, in case of
5516 // deoptimization, we'll reexecute the allocation and the
5517 // initialization.
5518 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5519 if (alloc != nullptr) {
5520 ciMethod* trap_method = alloc->jvms()->method();
5521 int trap_bci = alloc->jvms()->bci();
5522
5523 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5524 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5525 // Make sure there's no store between the allocation and the
5526 // arraycopy otherwise visible side effects could be rexecuted
5527 // in case of deoptimization and cause incorrect execution.
5528 bool no_interfering_store = true;
5529 Node* mem = alloc->in(TypeFunc::Memory);
5530 if (mem->is_MergeMem()) {
5531 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5532 Node* n = mms.memory();
5533 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5534 assert(n->is_Store(), "what else?");
5535 no_interfering_store = false;
5536 break;
5537 }
5538 }
5539 } else {
5540 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5541 Node* n = mms.memory();
5542 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5543 assert(n->is_Store(), "what else?");
5544 no_interfering_store = false;
5545 break;
5546 }
5547 }
5548 }
5549
5550 if (no_interfering_store) {
5551 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5552
5553 JVMState* saved_jvms = jvms();
5554 saved_reexecute_sp = _reexecute_sp;
5555
5556 set_jvms(sfpt->jvms());
5557 _reexecute_sp = jvms()->sp();
5558
5559 return saved_jvms;
5560 }
5561 }
5562 }
5563 return nullptr;
5564 }
5565
5566 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5567 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5568 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5569 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5570 uint size = alloc->req();
5571 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5572 old_jvms->set_map(sfpt);
5573 for (uint i = 0; i < size; i++) {
5574 sfpt->init_req(i, alloc->in(i));
5575 }
5576 // re-push array length for deoptimization
5577 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5578 old_jvms->set_sp(old_jvms->sp()+1);
5579 old_jvms->set_monoff(old_jvms->monoff()+1);
5580 old_jvms->set_scloff(old_jvms->scloff()+1);
5581 old_jvms->set_endoff(old_jvms->endoff()+1);
5582 old_jvms->set_should_reexecute(true);
5583
5584 sfpt->set_i_o(map()->i_o());
5585 sfpt->set_memory(map()->memory());
5586 sfpt->set_control(map()->control());
5587 return sfpt;
5588 }
5589
5590 // In case of a deoptimization, we restart execution at the
5591 // allocation, allocating a new array. We would leave an uninitialized
5592 // array in the heap that GCs wouldn't expect. Move the allocation
5593 // after the traps so we don't allocate the array if we
5594 // deoptimize. This is possible because tightly_coupled_allocation()
5595 // guarantees there's no observer of the allocated array at this point
5596 // and the control flow is simple enough.
5597 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5598 int saved_reexecute_sp, uint new_idx) {
5599 if (saved_jvms_before_guards != nullptr && !stopped()) {
5600 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5601
5602 assert(alloc != nullptr, "only with a tightly coupled allocation");
5603 // restore JVM state to the state at the arraycopy
5604 saved_jvms_before_guards->map()->set_control(map()->control());
5605 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5606 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5607 // If we've improved the types of some nodes (null check) while
5608 // emitting the guards, propagate them to the current state
5609 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5610 set_jvms(saved_jvms_before_guards);
5611 _reexecute_sp = saved_reexecute_sp;
5612
5613 // Remove the allocation from above the guards
5614 CallProjections* callprojs = alloc->extract_projections(true);
5615 InitializeNode* init = alloc->initialization();
5616 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5617 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5618 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5619
5620 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5621 // the allocation (i.e. is only valid if the allocation succeeds):
5622 // 1) replace CastIINode with AllocateArrayNode's length here
5623 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5624 //
5625 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5626 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5627 Node* init_control = init->proj_out(TypeFunc::Control);
5628 Node* alloc_length = alloc->Ideal_length();
5629 #ifdef ASSERT
5630 Node* prev_cast = nullptr;
5631 #endif
5632 for (uint i = 0; i < init_control->outcnt(); i++) {
5633 Node* init_out = init_control->raw_out(i);
5634 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5635 #ifdef ASSERT
5636 if (prev_cast == nullptr) {
5637 prev_cast = init_out;
5638 } else {
5639 if (prev_cast->cmp(*init_out) == false) {
5640 prev_cast->dump();
5641 init_out->dump();
5642 assert(false, "not equal CastIINode");
5643 }
5644 }
5645 #endif
5646 C->gvn_replace_by(init_out, alloc_length);
5647 }
5648 }
5649 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5650
5651 // move the allocation here (after the guards)
5652 _gvn.hash_delete(alloc);
5653 alloc->set_req(TypeFunc::Control, control());
5654 alloc->set_req(TypeFunc::I_O, i_o());
5655 Node *mem = reset_memory();
5656 set_all_memory(mem);
5657 alloc->set_req(TypeFunc::Memory, mem);
5658 set_control(init->proj_out_or_null(TypeFunc::Control));
5659 set_i_o(callprojs->fallthrough_ioproj);
5660
5661 // Update memory as done in GraphKit::set_output_for_allocation()
5662 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5663 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5664 if (ary_type->isa_aryptr() && length_type != nullptr) {
5665 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5666 }
5667 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5668 int elemidx = C->get_alias_index(telemref);
5669 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5670 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5671
5672 Node* allocx = _gvn.transform(alloc);
5673 assert(allocx == alloc, "where has the allocation gone?");
5674 assert(dest->is_CheckCastPP(), "not an allocation result?");
5675
5676 _gvn.hash_delete(dest);
5677 dest->set_req(0, control());
5678 Node* destx = _gvn.transform(dest);
5679 assert(destx == dest, "where has the allocation result gone?");
5680
5681 array_ideal_length(alloc, ary_type, true);
5682 }
5683 }
5684
5685 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5686 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5687 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5688 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5689 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5690 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5691 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5692 JVMState* saved_jvms_before_guards) {
5693 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5694 // There is at least one unrelated uncommon trap which needs to be replaced.
5695 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5696
5697 JVMState* saved_jvms = jvms();
5698 const int saved_reexecute_sp = _reexecute_sp;
5699 set_jvms(sfpt->jvms());
5700 _reexecute_sp = jvms()->sp();
5701
5702 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5703
5704 // Restore state
5705 set_jvms(saved_jvms);
5706 _reexecute_sp = saved_reexecute_sp;
5707 }
5708 }
5709
5710 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5711 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5712 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5713 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5714 while (if_proj->is_IfProj()) {
5715 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5716 if (uncommon_trap != nullptr) {
5717 create_new_uncommon_trap(uncommon_trap);
5718 }
5719 assert(if_proj->in(0)->is_If(), "must be If");
5720 if_proj = if_proj->in(0)->in(0);
5721 }
5722 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5723 "must have reached control projection of init node");
5724 }
5725
5726 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5727 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5728 assert(trap_request != 0, "no valid UCT trap request");
5729 PreserveJVMState pjvms(this);
5730 set_control(uncommon_trap_call->in(0));
5731 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5732 Deoptimization::trap_request_action(trap_request));
5733 assert(stopped(), "Should be stopped");
5734 _gvn.hash_delete(uncommon_trap_call);
5735 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5736 }
5737
5738 //------------------------------inline_arraycopy-----------------------
5739 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5740 // Object dest, int destPos,
5741 // int length);
5742 bool LibraryCallKit::inline_arraycopy() {
5743 // Get the arguments.
5744 Node* src = argument(0); // type: oop
5745 Node* src_offset = argument(1); // type: int
5746 Node* dest = argument(2); // type: oop
5747 Node* dest_offset = argument(3); // type: int
5748 Node* length = argument(4); // type: int
5749
5750 uint new_idx = C->unique();
5751
5752 // Check for allocation before we add nodes that would confuse
5753 // tightly_coupled_allocation()
5754 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5755
5756 int saved_reexecute_sp = -1;
5757 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5758 // See arraycopy_restore_alloc_state() comment
5759 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5760 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5761 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5762 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5763
5764 // The following tests must be performed
5765 // (1) src and dest are arrays.
5766 // (2) src and dest arrays must have elements of the same BasicType
5767 // (3) src and dest must not be null.
5768 // (4) src_offset must not be negative.
5769 // (5) dest_offset must not be negative.
5770 // (6) length must not be negative.
5771 // (7) src_offset + length must not exceed length of src.
5772 // (8) dest_offset + length must not exceed length of dest.
5773 // (9) each element of an oop array must be assignable
5774
5775 // (3) src and dest must not be null.
5776 // always do this here because we need the JVM state for uncommon traps
5777 Node* null_ctl = top();
5778 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5779 assert(null_ctl->is_top(), "no null control here");
5780 dest = null_check(dest, T_ARRAY);
5781
5782 if (!can_emit_guards) {
5783 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5784 // guards but the arraycopy node could still take advantage of a
5785 // tightly allocated allocation. tightly_coupled_allocation() is
5786 // called again to make sure it takes the null check above into
5787 // account: the null check is mandatory and if it caused an
5788 // uncommon trap to be emitted then the allocation can't be
5789 // considered tightly coupled in this context.
5790 alloc = tightly_coupled_allocation(dest);
5791 }
5792
5793 bool validated = false;
5794
5795 const Type* src_type = _gvn.type(src);
5796 const Type* dest_type = _gvn.type(dest);
5797 const TypeAryPtr* top_src = src_type->isa_aryptr();
5798 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5799
5800 // Do we have the type of src?
5801 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5802 // Do we have the type of dest?
5803 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5804 // Is the type for src from speculation?
5805 bool src_spec = false;
5806 // Is the type for dest from speculation?
5807 bool dest_spec = false;
5808
5809 if ((!has_src || !has_dest) && can_emit_guards) {
5810 // We don't have sufficient type information, let's see if
5811 // speculative types can help. We need to have types for both src
5812 // and dest so that it pays off.
5813
5814 // Do we already have or could we have type information for src
5815 bool could_have_src = has_src;
5816 // Do we already have or could we have type information for dest
5817 bool could_have_dest = has_dest;
5818
5819 ciKlass* src_k = nullptr;
5820 if (!has_src) {
5821 src_k = src_type->speculative_type_not_null();
5822 if (src_k != nullptr && src_k->is_array_klass()) {
5823 could_have_src = true;
5824 }
5825 }
5826
5827 ciKlass* dest_k = nullptr;
5828 if (!has_dest) {
5829 dest_k = dest_type->speculative_type_not_null();
5830 if (dest_k != nullptr && dest_k->is_array_klass()) {
5831 could_have_dest = true;
5832 }
5833 }
5834
5835 if (could_have_src && could_have_dest) {
5836 // This is going to pay off so emit the required guards
5837 if (!has_src) {
5838 src = maybe_cast_profiled_obj(src, src_k, true);
5839 src_type = _gvn.type(src);
5840 top_src = src_type->isa_aryptr();
5841 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5842 src_spec = true;
5843 }
5844 if (!has_dest) {
5845 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5846 dest_type = _gvn.type(dest);
5847 top_dest = dest_type->isa_aryptr();
5848 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5849 dest_spec = true;
5850 }
5851 }
5852 }
5853
5854 if (has_src && has_dest && can_emit_guards) {
5855 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5856 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5857 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5858 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5859
5860 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
5861 // If both arrays are object arrays then having the exact types
5862 // for both will remove the need for a subtype check at runtime
5863 // before the call and may make it possible to pick a faster copy
5864 // routine (without a subtype check on every element)
5865 // Do we have the exact type of src?
5866 bool could_have_src = src_spec;
5867 // Do we have the exact type of dest?
5868 bool could_have_dest = dest_spec;
5869 ciKlass* src_k = nullptr;
5870 ciKlass* dest_k = nullptr;
5871 if (!src_spec) {
5872 src_k = src_type->speculative_type_not_null();
5873 if (src_k != nullptr && src_k->is_array_klass()) {
5874 could_have_src = true;
5875 }
5876 }
5877 if (!dest_spec) {
5878 dest_k = dest_type->speculative_type_not_null();
5879 if (dest_k != nullptr && dest_k->is_array_klass()) {
5880 could_have_dest = true;
5881 }
5882 }
5883 if (could_have_src && could_have_dest) {
5884 // If we can have both exact types, emit the missing guards
5885 if (could_have_src && !src_spec) {
5886 src = maybe_cast_profiled_obj(src, src_k, true);
5887 src_type = _gvn.type(src);
5888 top_src = src_type->isa_aryptr();
5889 }
5890 if (could_have_dest && !dest_spec) {
5891 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5892 dest_type = _gvn.type(dest);
5893 top_dest = dest_type->isa_aryptr();
5894 }
5895 }
5896 }
5897 }
5898
5899 ciMethod* trap_method = method();
5900 int trap_bci = bci();
5901 if (saved_jvms_before_guards != nullptr) {
5902 trap_method = alloc->jvms()->method();
5903 trap_bci = alloc->jvms()->bci();
5904 }
5905
5906 bool negative_length_guard_generated = false;
5907
5908 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5909 can_emit_guards && !src->is_top() && !dest->is_top()) {
5910 // validate arguments: enables transformation the ArrayCopyNode
5911 validated = true;
5912
5913 RegionNode* slow_region = new RegionNode(1);
5914 record_for_igvn(slow_region);
5915
5916 // (1) src and dest are arrays.
5917 generate_non_array_guard(load_object_klass(src), slow_region);
5918 generate_non_array_guard(load_object_klass(dest), slow_region);
5919
5920 // (2) src and dest arrays must have elements of the same BasicType
5921 // done at macro expansion or at Ideal transformation time
5922
5923 // (4) src_offset must not be negative.
5924 generate_negative_guard(src_offset, slow_region);
5925
5926 // (5) dest_offset must not be negative.
5927 generate_negative_guard(dest_offset, slow_region);
5928
5929 // (7) src_offset + length must not exceed length of src.
5930 generate_limit_guard(src_offset, length,
5931 load_array_length(src),
5932 slow_region);
5933
5934 // (8) dest_offset + length must not exceed length of dest.
5935 generate_limit_guard(dest_offset, length,
5936 load_array_length(dest),
5937 slow_region);
5938
5939 // (6) length must not be negative.
5940 // This is also checked in generate_arraycopy() during macro expansion, but
5941 // we also have to check it here for the case where the ArrayCopyNode will
5942 // be eliminated by Escape Analysis.
5943 if (EliminateAllocations) {
5944 generate_negative_guard(length, slow_region);
5945 negative_length_guard_generated = true;
5946 }
5947
5948 // (9) each element of an oop array must be assignable
5949 Node* dest_klass = load_object_klass(dest);
5950 if (src != dest) {
5951 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5952 slow_region->add_req(not_subtype_ctrl);
5953 }
5954
5955 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5956 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
5957 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5958 src_type = _gvn.type(src);
5959 top_src = src_type->isa_aryptr();
5960
5961 // Handle flat inline type arrays (null-free arrays are handled by the subtype check above)
5962 if (!stopped() && UseFlatArray) {
5963 // If dest is flat, src must be flat as well (guaranteed by src <: dest check). Handle flat src here.
5964 assert(top_dest == nullptr || !top_dest->is_flat() || top_src->is_flat(), "src array must be flat");
5965 if (top_src != nullptr && top_src->is_flat()) {
5966 // Src is flat, check that dest is flat as well
5967 if (top_dest != nullptr && !top_dest->is_flat()) {
5968 generate_fair_guard(flat_array_test(dest_klass, /* flat = */ false), slow_region);
5969 // Since dest is flat and src <: dest, dest must have the same type as src.
5970 top_dest = top_src->cast_to_exactness(false);
5971 assert(top_dest->is_flat(), "dest must be flat");
5972 dest = _gvn.transform(new CheckCastPPNode(control(), dest, top_dest));
5973 }
5974 } else if (top_src == nullptr || !top_src->is_not_flat()) {
5975 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
5976 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
5977 assert(top_dest == nullptr || !top_dest->is_flat(), "dest array must not be flat");
5978 generate_fair_guard(flat_array_test(src), slow_region);
5979 if (top_src != nullptr) {
5980 top_src = top_src->cast_to_not_flat();
5981 src = _gvn.transform(new CheckCastPPNode(control(), src, top_src));
5982 }
5983 }
5984 }
5985
5986 {
5987 PreserveJVMState pjvms(this);
5988 set_control(_gvn.transform(slow_region));
5989 uncommon_trap(Deoptimization::Reason_intrinsic,
5990 Deoptimization::Action_make_not_entrant);
5991 assert(stopped(), "Should be stopped");
5992 }
5993 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
5994 }
5995
5996 if (stopped()) {
5997 return true;
5998 }
5999
6000 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6001 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6002 // so the compiler has a chance to eliminate them: during macro expansion,
6003 // we have to set their control (CastPP nodes are eliminated).
6004 load_object_klass(src), load_object_klass(dest),
6005 load_array_length(src), load_array_length(dest));
6006
6007 ac->set_arraycopy(validated);
6008
6009 Node* n = _gvn.transform(ac);
6010 if (n == ac) {
6011 ac->connect_outputs(this);
6012 } else {
6013 assert(validated, "shouldn't transform if all arguments not validated");
6014 set_all_memory(n);
6015 }
6016 clear_upper_avx();
6017
6018
6019 return true;
6020 }
6021
6022
6023 // Helper function which determines if an arraycopy immediately follows
6024 // an allocation, with no intervening tests or other escapes for the object.
6025 AllocateArrayNode*
6026 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6027 if (stopped()) return nullptr; // no fast path
6028 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6029
6030 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6031 if (alloc == nullptr) return nullptr;
6032
6033 Node* rawmem = memory(Compile::AliasIdxRaw);
6034 // Is the allocation's memory state untouched?
6035 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6036 // Bail out if there have been raw-memory effects since the allocation.
6037 // (Example: There might have been a call or safepoint.)
6038 return nullptr;
6039 }
6040 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6041 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6042 return nullptr;
6043 }
6044
6045 // There must be no unexpected observers of this allocation.
6046 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6047 Node* obs = ptr->fast_out(i);
6048 if (obs != this->map()) {
6049 return nullptr;
6050 }
6051 }
6052
6053 // This arraycopy must unconditionally follow the allocation of the ptr.
6054 Node* alloc_ctl = ptr->in(0);
6055 Node* ctl = control();
6056 while (ctl != alloc_ctl) {
6057 // There may be guards which feed into the slow_region.
6058 // Any other control flow means that we might not get a chance
6059 // to finish initializing the allocated object.
6060 // Various low-level checks bottom out in uncommon traps. These
6061 // are considered safe since we've already checked above that
6062 // there is no unexpected observer of this allocation.
6063 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6064 assert(ctl->in(0)->is_If(), "must be If");
6065 ctl = ctl->in(0)->in(0);
6066 } else {
6067 return nullptr;
6068 }
6069 }
6070
6071 // If we get this far, we have an allocation which immediately
6072 // precedes the arraycopy, and we can take over zeroing the new object.
6073 // The arraycopy will finish the initialization, and provide
6074 // a new control state to which we will anchor the destination pointer.
6075
6076 return alloc;
6077 }
6078
6079 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6080 if (node->is_IfProj()) {
6081 Node* other_proj = node->as_IfProj()->other_if_proj();
6082 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6083 Node* obs = other_proj->fast_out(j);
6084 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6085 (obs->as_CallStaticJava()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
6086 return obs->as_CallStaticJava();
6087 }
6088 }
6089 }
6090 return nullptr;
6091 }
6092
6093 //-------------inline_encodeISOArray-----------------------------------
6094 // encode char[] to byte[] in ISO_8859_1 or ASCII
6095 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6096 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6097 // no receiver since it is static method
6098 Node *src = argument(0);
6099 Node *src_offset = argument(1);
6100 Node *dst = argument(2);
6101 Node *dst_offset = argument(3);
6102 Node *length = argument(4);
6103
6104 src = must_be_not_null(src, true);
6105 dst = must_be_not_null(dst, true);
6106
6107 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6108 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6109 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6110 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6111 // failed array check
6112 return false;
6113 }
6114
6115 // Figure out the size and type of the elements we will be copying.
6116 BasicType src_elem = src_type->elem()->array_element_basic_type();
6117 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6118 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6119 return false;
6120 }
6121
6122 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6123 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6124 // 'src_start' points to src array + scaled offset
6125 // 'dst_start' points to dst array + scaled offset
6126
6127 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6128 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6129 enc = _gvn.transform(enc);
6130 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6131 set_memory(res_mem, mtype);
6132 set_result(enc);
6133 clear_upper_avx();
6134
6135 return true;
6136 }
6137
6138 //-------------inline_multiplyToLen-----------------------------------
6139 bool LibraryCallKit::inline_multiplyToLen() {
6140 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6141
6142 address stubAddr = StubRoutines::multiplyToLen();
6143 if (stubAddr == nullptr) {
6144 return false; // Intrinsic's stub is not implemented on this platform
6145 }
6146 const char* stubName = "multiplyToLen";
6147
6148 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6149
6150 // no receiver because it is a static method
6151 Node* x = argument(0);
6152 Node* xlen = argument(1);
6153 Node* y = argument(2);
6154 Node* ylen = argument(3);
6155 Node* z = argument(4);
6156
6157 x = must_be_not_null(x, true);
6158 y = must_be_not_null(y, true);
6159
6160 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6161 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6162 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6163 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6164 // failed array check
6165 return false;
6166 }
6167
6168 BasicType x_elem = x_type->elem()->array_element_basic_type();
6169 BasicType y_elem = y_type->elem()->array_element_basic_type();
6170 if (x_elem != T_INT || y_elem != T_INT) {
6171 return false;
6172 }
6173
6174 // Set the original stack and the reexecute bit for the interpreter to reexecute
6175 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
6176 // on the return from z array allocation in runtime.
6177 { PreserveReexecuteState preexecs(this);
6178 jvms()->set_should_reexecute(true);
6179
6180 Node* x_start = array_element_address(x, intcon(0), x_elem);
6181 Node* y_start = array_element_address(y, intcon(0), y_elem);
6182 // 'x_start' points to x array + scaled xlen
6183 // 'y_start' points to y array + scaled ylen
6184
6185 // Allocate the result array
6186 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
6187 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
6188 Node* klass_node = makecon(TypeKlassPtr::make(klass));
6189
6190 IdealKit ideal(this);
6191
6192 #define __ ideal.
6193 Node* one = __ ConI(1);
6194 Node* zero = __ ConI(0);
6195 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
6196 __ set(need_alloc, zero);
6197 __ set(z_alloc, z);
6198 __ if_then(z, BoolTest::eq, null()); {
6199 __ increment (need_alloc, one);
6200 } __ else_(); {
6201 // Update graphKit memory and control from IdealKit.
6202 sync_kit(ideal);
6203 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
6204 cast->init_req(0, control());
6205 _gvn.set_type(cast, cast->bottom_type());
6206 C->record_for_igvn(cast);
6207
6208 Node* zlen_arg = load_array_length(cast);
6209 // Update IdealKit memory and control from graphKit.
6210 __ sync_kit(this);
6211 __ if_then(zlen_arg, BoolTest::lt, zlen); {
6212 __ increment (need_alloc, one);
6213 } __ end_if();
6214 } __ end_if();
6215
6216 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
6217 // Update graphKit memory and control from IdealKit.
6218 sync_kit(ideal);
6219 Node * narr = new_array(klass_node, zlen, 1);
6220 // Update IdealKit memory and control from graphKit.
6221 __ sync_kit(this);
6222 __ set(z_alloc, narr);
6223 } __ end_if();
6224
6225 sync_kit(ideal);
6226 z = __ value(z_alloc);
6227 // Can't use TypeAryPtr::INTS which uses Bottom offset.
6228 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
6229 // Final sync IdealKit and GraphKit.
6230 final_sync(ideal);
6231 #undef __
6232
6233 Node* z_start = array_element_address(z, intcon(0), T_INT);
6234
6235 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6236 OptoRuntime::multiplyToLen_Type(),
6237 stubAddr, stubName, TypePtr::BOTTOM,
6238 x_start, xlen, y_start, ylen, z_start, zlen);
6239 } // original reexecute is set back here
6240
6241 C->set_has_split_ifs(true); // Has chance for split-if optimization
6242 set_result(z);
6243 return true;
6244 }
6245
6246 //-------------inline_squareToLen------------------------------------
6247 bool LibraryCallKit::inline_squareToLen() {
6248 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6249
6250 address stubAddr = StubRoutines::squareToLen();
6251 if (stubAddr == nullptr) {
6252 return false; // Intrinsic's stub is not implemented on this platform
6253 }
6254 const char* stubName = "squareToLen";
6255
6256 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6257
6258 Node* x = argument(0);
6259 Node* len = argument(1);
6260 Node* z = argument(2);
6261 Node* zlen = argument(3);
6262
6263 x = must_be_not_null(x, true);
6264 z = must_be_not_null(z, true);
6265
6266 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6267 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6268 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6269 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6270 // failed array check
6271 return false;
6272 }
6273
6274 BasicType x_elem = x_type->elem()->array_element_basic_type();
6275 BasicType z_elem = z_type->elem()->array_element_basic_type();
6276 if (x_elem != T_INT || z_elem != T_INT) {
6277 return false;
6278 }
6279
6280
6281 Node* x_start = array_element_address(x, intcon(0), x_elem);
6282 Node* z_start = array_element_address(z, intcon(0), z_elem);
6283
6284 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6285 OptoRuntime::squareToLen_Type(),
6286 stubAddr, stubName, TypePtr::BOTTOM,
6287 x_start, len, z_start, zlen);
6288
6289 set_result(z);
6290 return true;
6291 }
6292
6293 //-------------inline_mulAdd------------------------------------------
6294 bool LibraryCallKit::inline_mulAdd() {
6295 assert(UseMulAddIntrinsic, "not implemented on this platform");
6296
6297 address stubAddr = StubRoutines::mulAdd();
6298 if (stubAddr == nullptr) {
6299 return false; // Intrinsic's stub is not implemented on this platform
6300 }
6301 const char* stubName = "mulAdd";
6302
6303 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6304
6305 Node* out = argument(0);
6306 Node* in = argument(1);
6307 Node* offset = argument(2);
6308 Node* len = argument(3);
6309 Node* k = argument(4);
6310
6311 in = must_be_not_null(in, true);
6312 out = must_be_not_null(out, true);
6313
6314 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6315 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6316 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6317 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6318 // failed array check
6319 return false;
6320 }
6321
6322 BasicType out_elem = out_type->elem()->array_element_basic_type();
6323 BasicType in_elem = in_type->elem()->array_element_basic_type();
6324 if (out_elem != T_INT || in_elem != T_INT) {
6325 return false;
6326 }
6327
6328 Node* outlen = load_array_length(out);
6329 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6330 Node* out_start = array_element_address(out, intcon(0), out_elem);
6331 Node* in_start = array_element_address(in, intcon(0), in_elem);
6332
6333 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6334 OptoRuntime::mulAdd_Type(),
6335 stubAddr, stubName, TypePtr::BOTTOM,
6336 out_start,in_start, new_offset, len, k);
6337 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6338 set_result(result);
6339 return true;
6340 }
6341
6342 //-------------inline_montgomeryMultiply-----------------------------------
6343 bool LibraryCallKit::inline_montgomeryMultiply() {
6344 address stubAddr = StubRoutines::montgomeryMultiply();
6345 if (stubAddr == nullptr) {
6346 return false; // Intrinsic's stub is not implemented on this platform
6347 }
6348
6349 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6350 const char* stubName = "montgomery_multiply";
6351
6352 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6353
6354 Node* a = argument(0);
6355 Node* b = argument(1);
6356 Node* n = argument(2);
6357 Node* len = argument(3);
6358 Node* inv = argument(4);
6359 Node* m = argument(6);
6360
6361 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6362 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6363 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6364 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6365 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6366 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6367 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6368 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6369 // failed array check
6370 return false;
6371 }
6372
6373 BasicType a_elem = a_type->elem()->array_element_basic_type();
6374 BasicType b_elem = b_type->elem()->array_element_basic_type();
6375 BasicType n_elem = n_type->elem()->array_element_basic_type();
6376 BasicType m_elem = m_type->elem()->array_element_basic_type();
6377 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6378 return false;
6379 }
6380
6381 // Make the call
6382 {
6383 Node* a_start = array_element_address(a, intcon(0), a_elem);
6384 Node* b_start = array_element_address(b, intcon(0), b_elem);
6385 Node* n_start = array_element_address(n, intcon(0), n_elem);
6386 Node* m_start = array_element_address(m, intcon(0), m_elem);
6387
6388 Node* call = make_runtime_call(RC_LEAF,
6389 OptoRuntime::montgomeryMultiply_Type(),
6390 stubAddr, stubName, TypePtr::BOTTOM,
6391 a_start, b_start, n_start, len, inv, top(),
6392 m_start);
6393 set_result(m);
6394 }
6395
6396 return true;
6397 }
6398
6399 bool LibraryCallKit::inline_montgomerySquare() {
6400 address stubAddr = StubRoutines::montgomerySquare();
6401 if (stubAddr == nullptr) {
6402 return false; // Intrinsic's stub is not implemented on this platform
6403 }
6404
6405 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6406 const char* stubName = "montgomery_square";
6407
6408 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6409
6410 Node* a = argument(0);
6411 Node* n = argument(1);
6412 Node* len = argument(2);
6413 Node* inv = argument(3);
6414 Node* m = argument(5);
6415
6416 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6417 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6418 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6419 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6420 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6421 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6422 // failed array check
6423 return false;
6424 }
6425
6426 BasicType a_elem = a_type->elem()->array_element_basic_type();
6427 BasicType n_elem = n_type->elem()->array_element_basic_type();
6428 BasicType m_elem = m_type->elem()->array_element_basic_type();
6429 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6430 return false;
6431 }
6432
6433 // Make the call
6434 {
6435 Node* a_start = array_element_address(a, intcon(0), a_elem);
6436 Node* n_start = array_element_address(n, intcon(0), n_elem);
6437 Node* m_start = array_element_address(m, intcon(0), m_elem);
6438
6439 Node* call = make_runtime_call(RC_LEAF,
6440 OptoRuntime::montgomerySquare_Type(),
6441 stubAddr, stubName, TypePtr::BOTTOM,
6442 a_start, n_start, len, inv, top(),
6443 m_start);
6444 set_result(m);
6445 }
6446
6447 return true;
6448 }
6449
6450 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6451 address stubAddr = nullptr;
6452 const char* stubName = nullptr;
6453
6454 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6455 if (stubAddr == nullptr) {
6456 return false; // Intrinsic's stub is not implemented on this platform
6457 }
6458
6459 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6460
6461 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6462
6463 Node* newArr = argument(0);
6464 Node* oldArr = argument(1);
6465 Node* newIdx = argument(2);
6466 Node* shiftCount = argument(3);
6467 Node* numIter = argument(4);
6468
6469 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6470 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6471 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6472 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6473 return false;
6474 }
6475
6476 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6477 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6478 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6479 return false;
6480 }
6481
6482 // Make the call
6483 {
6484 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6485 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6486
6487 Node* call = make_runtime_call(RC_LEAF,
6488 OptoRuntime::bigIntegerShift_Type(),
6489 stubAddr,
6490 stubName,
6491 TypePtr::BOTTOM,
6492 newArr_start,
6493 oldArr_start,
6494 newIdx,
6495 shiftCount,
6496 numIter);
6497 }
6498
6499 return true;
6500 }
6501
6502 //-------------inline_vectorizedMismatch------------------------------
6503 bool LibraryCallKit::inline_vectorizedMismatch() {
6504 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6505
6506 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6507 Node* obja = argument(0); // Object
6508 Node* aoffset = argument(1); // long
6509 Node* objb = argument(3); // Object
6510 Node* boffset = argument(4); // long
6511 Node* length = argument(6); // int
6512 Node* scale = argument(7); // int
6513
6514 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6515 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6516 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6517 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6518 scale == top()) {
6519 return false; // failed input validation
6520 }
6521
6522 Node* obja_adr = make_unsafe_address(obja, aoffset);
6523 Node* objb_adr = make_unsafe_address(objb, boffset);
6524
6525 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6526 //
6527 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6528 // if (length <= inline_limit) {
6529 // inline_path:
6530 // vmask = VectorMaskGen length
6531 // vload1 = LoadVectorMasked obja, vmask
6532 // vload2 = LoadVectorMasked objb, vmask
6533 // result1 = VectorCmpMasked vload1, vload2, vmask
6534 // } else {
6535 // call_stub_path:
6536 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6537 // }
6538 // exit_block:
6539 // return Phi(result1, result2);
6540 //
6541 enum { inline_path = 1, // input is small enough to process it all at once
6542 stub_path = 2, // input is too large; call into the VM
6543 PATH_LIMIT = 3
6544 };
6545
6546 Node* exit_block = new RegionNode(PATH_LIMIT);
6547 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6548 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6549
6550 Node* call_stub_path = control();
6551
6552 BasicType elem_bt = T_ILLEGAL;
6553
6554 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6555 if (scale_t->is_con()) {
6556 switch (scale_t->get_con()) {
6557 case 0: elem_bt = T_BYTE; break;
6558 case 1: elem_bt = T_SHORT; break;
6559 case 2: elem_bt = T_INT; break;
6560 case 3: elem_bt = T_LONG; break;
6561
6562 default: elem_bt = T_ILLEGAL; break; // not supported
6563 }
6564 }
6565
6566 int inline_limit = 0;
6567 bool do_partial_inline = false;
6568
6569 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6570 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6571 do_partial_inline = inline_limit >= 16;
6572 }
6573
6574 if (do_partial_inline) {
6575 assert(elem_bt != T_ILLEGAL, "sanity");
6576
6577 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6578 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6579 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6580
6581 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6582 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6583 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6584
6585 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6586
6587 if (!stopped()) {
6588 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6589
6590 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6591 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6592 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6593 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6594
6595 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6596 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6597 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6598 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6599
6600 exit_block->init_req(inline_path, control());
6601 memory_phi->init_req(inline_path, map()->memory());
6602 result_phi->init_req(inline_path, result);
6603
6604 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6605 clear_upper_avx();
6606 }
6607 }
6608 }
6609
6610 if (call_stub_path != nullptr) {
6611 set_control(call_stub_path);
6612
6613 Node* call = make_runtime_call(RC_LEAF,
6614 OptoRuntime::vectorizedMismatch_Type(),
6615 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6616 obja_adr, objb_adr, length, scale);
6617
6618 exit_block->init_req(stub_path, control());
6619 memory_phi->init_req(stub_path, map()->memory());
6620 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6621 }
6622
6623 exit_block = _gvn.transform(exit_block);
6624 memory_phi = _gvn.transform(memory_phi);
6625 result_phi = _gvn.transform(result_phi);
6626
6627 set_control(exit_block);
6628 set_all_memory(memory_phi);
6629 set_result(result_phi);
6630
6631 return true;
6632 }
6633
6634 //------------------------------inline_vectorizedHashcode----------------------------
6635 bool LibraryCallKit::inline_vectorizedHashCode() {
6636 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6637
6638 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6639 Node* array = argument(0);
6640 Node* offset = argument(1);
6641 Node* length = argument(2);
6642 Node* initialValue = argument(3);
6643 Node* basic_type = argument(4);
6644
6645 if (basic_type == top()) {
6646 return false; // failed input validation
6647 }
6648
6649 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6650 if (!basic_type_t->is_con()) {
6651 return false; // Only intrinsify if mode argument is constant
6652 }
6653
6654 array = must_be_not_null(array, true);
6655
6656 BasicType bt = (BasicType)basic_type_t->get_con();
6657
6658 // Resolve address of first element
6659 Node* array_start = array_element_address(array, offset, bt);
6660
6661 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6662 array_start, length, initialValue, basic_type)));
6663 clear_upper_avx();
6664
6665 return true;
6666 }
6667
6668 /**
6669 * Calculate CRC32 for byte.
6670 * int java.util.zip.CRC32.update(int crc, int b)
6671 */
6672 bool LibraryCallKit::inline_updateCRC32() {
6673 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6674 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6675 // no receiver since it is static method
6676 Node* crc = argument(0); // type: int
6677 Node* b = argument(1); // type: int
6678
6679 /*
6680 * int c = ~ crc;
6681 * b = timesXtoThe32[(b ^ c) & 0xFF];
6682 * b = b ^ (c >>> 8);
6683 * crc = ~b;
6684 */
6685
6686 Node* M1 = intcon(-1);
6687 crc = _gvn.transform(new XorINode(crc, M1));
6688 Node* result = _gvn.transform(new XorINode(crc, b));
6689 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6690
6691 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6692 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6693 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6694 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6695
6696 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6697 result = _gvn.transform(new XorINode(crc, result));
6698 result = _gvn.transform(new XorINode(result, M1));
6699 set_result(result);
6700 return true;
6701 }
6702
6703 /**
6704 * Calculate CRC32 for byte[] array.
6705 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6706 */
6707 bool LibraryCallKit::inline_updateBytesCRC32() {
6708 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6709 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6710 // no receiver since it is static method
6711 Node* crc = argument(0); // type: int
6712 Node* src = argument(1); // type: oop
6713 Node* offset = argument(2); // type: int
6714 Node* length = argument(3); // type: int
6715
6716 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6717 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6718 // failed array check
6719 return false;
6720 }
6721
6722 // Figure out the size and type of the elements we will be copying.
6723 BasicType src_elem = src_type->elem()->array_element_basic_type();
6724 if (src_elem != T_BYTE) {
6725 return false;
6726 }
6727
6728 // 'src_start' points to src array + scaled offset
6729 src = must_be_not_null(src, true);
6730 Node* src_start = array_element_address(src, offset, src_elem);
6731
6732 // We assume that range check is done by caller.
6733 // TODO: generate range check (offset+length < src.length) in debug VM.
6734
6735 // Call the stub.
6736 address stubAddr = StubRoutines::updateBytesCRC32();
6737 const char *stubName = "updateBytesCRC32";
6738
6739 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6740 stubAddr, stubName, TypePtr::BOTTOM,
6741 crc, src_start, length);
6742 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6743 set_result(result);
6744 return true;
6745 }
6746
6747 /**
6748 * Calculate CRC32 for ByteBuffer.
6749 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6750 */
6751 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6752 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6753 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6754 // no receiver since it is static method
6755 Node* crc = argument(0); // type: int
6756 Node* src = argument(1); // type: long
6757 Node* offset = argument(3); // type: int
6758 Node* length = argument(4); // type: int
6759
6760 src = ConvL2X(src); // adjust Java long to machine word
6761 Node* base = _gvn.transform(new CastX2PNode(src));
6762 offset = ConvI2X(offset);
6763
6764 // 'src_start' points to src array + scaled offset
6765 Node* src_start = basic_plus_adr(top(), base, offset);
6766
6767 // Call the stub.
6768 address stubAddr = StubRoutines::updateBytesCRC32();
6769 const char *stubName = "updateBytesCRC32";
6770
6771 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6772 stubAddr, stubName, TypePtr::BOTTOM,
6773 crc, src_start, length);
6774 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6775 set_result(result);
6776 return true;
6777 }
6778
6779 //------------------------------get_table_from_crc32c_class-----------------------
6780 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6781 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6782 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6783
6784 return table;
6785 }
6786
6787 //------------------------------inline_updateBytesCRC32C-----------------------
6788 //
6789 // Calculate CRC32C for byte[] array.
6790 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6791 //
6792 bool LibraryCallKit::inline_updateBytesCRC32C() {
6793 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6794 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6795 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6796 // no receiver since it is a static method
6797 Node* crc = argument(0); // type: int
6798 Node* src = argument(1); // type: oop
6799 Node* offset = argument(2); // type: int
6800 Node* end = argument(3); // type: int
6801
6802 Node* length = _gvn.transform(new SubINode(end, offset));
6803
6804 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6805 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6806 // failed array check
6807 return false;
6808 }
6809
6810 // Figure out the size and type of the elements we will be copying.
6811 BasicType src_elem = src_type->elem()->array_element_basic_type();
6812 if (src_elem != T_BYTE) {
6813 return false;
6814 }
6815
6816 // 'src_start' points to src array + scaled offset
6817 src = must_be_not_null(src, true);
6818 Node* src_start = array_element_address(src, offset, src_elem);
6819
6820 // static final int[] byteTable in class CRC32C
6821 Node* table = get_table_from_crc32c_class(callee()->holder());
6822 table = must_be_not_null(table, true);
6823 Node* table_start = array_element_address(table, intcon(0), T_INT);
6824
6825 // We assume that range check is done by caller.
6826 // TODO: generate range check (offset+length < src.length) in debug VM.
6827
6828 // Call the stub.
6829 address stubAddr = StubRoutines::updateBytesCRC32C();
6830 const char *stubName = "updateBytesCRC32C";
6831
6832 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6833 stubAddr, stubName, TypePtr::BOTTOM,
6834 crc, src_start, length, table_start);
6835 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6836 set_result(result);
6837 return true;
6838 }
6839
6840 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6841 //
6842 // Calculate CRC32C for DirectByteBuffer.
6843 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6844 //
6845 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6846 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6847 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6848 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6849 // no receiver since it is a static method
6850 Node* crc = argument(0); // type: int
6851 Node* src = argument(1); // type: long
6852 Node* offset = argument(3); // type: int
6853 Node* end = argument(4); // type: int
6854
6855 Node* length = _gvn.transform(new SubINode(end, offset));
6856
6857 src = ConvL2X(src); // adjust Java long to machine word
6858 Node* base = _gvn.transform(new CastX2PNode(src));
6859 offset = ConvI2X(offset);
6860
6861 // 'src_start' points to src array + scaled offset
6862 Node* src_start = basic_plus_adr(top(), base, offset);
6863
6864 // static final int[] byteTable in class CRC32C
6865 Node* table = get_table_from_crc32c_class(callee()->holder());
6866 table = must_be_not_null(table, true);
6867 Node* table_start = array_element_address(table, intcon(0), T_INT);
6868
6869 // Call the stub.
6870 address stubAddr = StubRoutines::updateBytesCRC32C();
6871 const char *stubName = "updateBytesCRC32C";
6872
6873 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6874 stubAddr, stubName, TypePtr::BOTTOM,
6875 crc, src_start, length, table_start);
6876 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6877 set_result(result);
6878 return true;
6879 }
6880
6881 //------------------------------inline_updateBytesAdler32----------------------
6882 //
6883 // Calculate Adler32 checksum for byte[] array.
6884 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6885 //
6886 bool LibraryCallKit::inline_updateBytesAdler32() {
6887 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6888 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6889 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6890 // no receiver since it is static method
6891 Node* crc = argument(0); // type: int
6892 Node* src = argument(1); // type: oop
6893 Node* offset = argument(2); // type: int
6894 Node* length = argument(3); // type: int
6895
6896 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6897 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6898 // failed array check
6899 return false;
6900 }
6901
6902 // Figure out the size and type of the elements we will be copying.
6903 BasicType src_elem = src_type->elem()->array_element_basic_type();
6904 if (src_elem != T_BYTE) {
6905 return false;
6906 }
6907
6908 // 'src_start' points to src array + scaled offset
6909 Node* src_start = array_element_address(src, offset, src_elem);
6910
6911 // We assume that range check is done by caller.
6912 // TODO: generate range check (offset+length < src.length) in debug VM.
6913
6914 // Call the stub.
6915 address stubAddr = StubRoutines::updateBytesAdler32();
6916 const char *stubName = "updateBytesAdler32";
6917
6918 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6919 stubAddr, stubName, TypePtr::BOTTOM,
6920 crc, src_start, length);
6921 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6922 set_result(result);
6923 return true;
6924 }
6925
6926 //------------------------------inline_updateByteBufferAdler32---------------
6927 //
6928 // Calculate Adler32 checksum for DirectByteBuffer.
6929 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6930 //
6931 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6932 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6933 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6934 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6935 // no receiver since it is static method
6936 Node* crc = argument(0); // type: int
6937 Node* src = argument(1); // type: long
6938 Node* offset = argument(3); // type: int
6939 Node* length = argument(4); // type: int
6940
6941 src = ConvL2X(src); // adjust Java long to machine word
6942 Node* base = _gvn.transform(new CastX2PNode(src));
6943 offset = ConvI2X(offset);
6944
6945 // 'src_start' points to src array + scaled offset
6946 Node* src_start = basic_plus_adr(top(), base, offset);
6947
6948 // Call the stub.
6949 address stubAddr = StubRoutines::updateBytesAdler32();
6950 const char *stubName = "updateBytesAdler32";
6951
6952 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6953 stubAddr, stubName, TypePtr::BOTTOM,
6954 crc, src_start, length);
6955
6956 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6957 set_result(result);
6958 return true;
6959 }
6960
6961 //----------------------------inline_reference_get----------------------------
6962 // public T java.lang.ref.Reference.get();
6963 bool LibraryCallKit::inline_reference_get() {
6964 const int referent_offset = java_lang_ref_Reference::referent_offset();
6965
6966 // Get the argument:
6967 Node* reference_obj = null_check_receiver();
6968 if (stopped()) return true;
6969
6970 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6971 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6972 decorators, /*is_static*/ false, nullptr);
6973 if (result == nullptr) return false;
6974
6975 // Add memory barrier to prevent commoning reads from this field
6976 // across safepoint since GC can change its value.
6977 insert_mem_bar(Op_MemBarCPUOrder);
6978
6979 set_result(result);
6980 return true;
6981 }
6982
6983 //----------------------------inline_reference_refersTo0----------------------------
6984 // bool java.lang.ref.Reference.refersTo0();
6985 // bool java.lang.ref.PhantomReference.refersTo0();
6986 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6987 // Get arguments:
6988 Node* reference_obj = null_check_receiver();
6989 Node* other_obj = argument(1);
6990 if (stopped()) return true;
6991
6992 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6993 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6994 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6995 decorators, /*is_static*/ false, nullptr);
6996 if (referent == nullptr) return false;
6997
6998 // Add memory barrier to prevent commoning reads from this field
6999 // across safepoint since GC can change its value.
7000 insert_mem_bar(Op_MemBarCPUOrder);
7001
7002 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7003 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7004 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7005
7006 RegionNode* region = new RegionNode(3);
7007 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7008
7009 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7010 region->init_req(1, if_true);
7011 phi->init_req(1, intcon(1));
7012
7013 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7014 region->init_req(2, if_false);
7015 phi->init_req(2, intcon(0));
7016
7017 set_control(_gvn.transform(region));
7018 record_for_igvn(region);
7019 set_result(_gvn.transform(phi));
7020 return true;
7021 }
7022
7023
7024 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7025 DecoratorSet decorators, bool is_static,
7026 ciInstanceKlass* fromKls) {
7027 if (fromKls == nullptr) {
7028 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7029 assert(tinst != nullptr, "obj is null");
7030 assert(tinst->is_loaded(), "obj is not loaded");
7031 fromKls = tinst->instance_klass();
7032 } else {
7033 assert(is_static, "only for static field access");
7034 }
7035 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7036 ciSymbol::make(fieldTypeString),
7037 is_static);
7038
7039 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7040 if (field == nullptr) return (Node *) nullptr;
7041
7042 if (is_static) {
7043 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7044 fromObj = makecon(tip);
7045 }
7046
7047 // Next code copied from Parse::do_get_xxx():
7048
7049 // Compute address and memory type.
7050 int offset = field->offset_in_bytes();
7051 bool is_vol = field->is_volatile();
7052 ciType* field_klass = field->type();
7053 assert(field_klass->is_loaded(), "should be loaded");
7054 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7055 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7056 BasicType bt = field->layout_type();
7057
7058 // Build the resultant type of the load
7059 const Type *type;
7060 if (bt == T_OBJECT) {
7061 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7062 } else {
7063 type = Type::get_const_basic_type(bt);
7064 }
7065
7066 if (is_vol) {
7067 decorators |= MO_SEQ_CST;
7068 }
7069
7070 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7071 }
7072
7073 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7074 bool is_exact /* true */, bool is_static /* false */,
7075 ciInstanceKlass * fromKls /* nullptr */) {
7076 if (fromKls == nullptr) {
7077 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7078 assert(tinst != nullptr, "obj is null");
7079 assert(tinst->is_loaded(), "obj is not loaded");
7080 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7081 fromKls = tinst->instance_klass();
7082 }
7083 else {
7084 assert(is_static, "only for static field access");
7085 }
7086 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7087 ciSymbol::make(fieldTypeString),
7088 is_static);
7089
7090 assert(field != nullptr, "undefined field");
7091 assert(!field->is_volatile(), "not defined for volatile fields");
7092
7093 if (is_static) {
7094 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7095 fromObj = makecon(tip);
7096 }
7097
7098 // Next code copied from Parse::do_get_xxx():
7099
7100 // Compute address and memory type.
7101 int offset = field->offset_in_bytes();
7102 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7103
7104 return adr;
7105 }
7106
7107 //------------------------------inline_aescrypt_Block-----------------------
7108 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7109 address stubAddr = nullptr;
7110 const char *stubName;
7111 assert(UseAES, "need AES instruction support");
7112
7113 switch(id) {
7114 case vmIntrinsics::_aescrypt_encryptBlock:
7115 stubAddr = StubRoutines::aescrypt_encryptBlock();
7116 stubName = "aescrypt_encryptBlock";
7117 break;
7118 case vmIntrinsics::_aescrypt_decryptBlock:
7119 stubAddr = StubRoutines::aescrypt_decryptBlock();
7120 stubName = "aescrypt_decryptBlock";
7121 break;
7122 default:
7123 break;
7124 }
7125 if (stubAddr == nullptr) return false;
7126
7127 Node* aescrypt_object = argument(0);
7128 Node* src = argument(1);
7129 Node* src_offset = argument(2);
7130 Node* dest = argument(3);
7131 Node* dest_offset = argument(4);
7132
7133 src = must_be_not_null(src, true);
7134 dest = must_be_not_null(dest, true);
7135
7136 // (1) src and dest are arrays.
7137 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7138 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7139 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7140 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7141
7142 // for the quick and dirty code we will skip all the checks.
7143 // we are just trying to get the call to be generated.
7144 Node* src_start = src;
7145 Node* dest_start = dest;
7146 if (src_offset != nullptr || dest_offset != nullptr) {
7147 assert(src_offset != nullptr && dest_offset != nullptr, "");
7148 src_start = array_element_address(src, src_offset, T_BYTE);
7149 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7150 }
7151
7152 // now need to get the start of its expanded key array
7153 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7154 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7155 if (k_start == nullptr) return false;
7156
7157 // Call the stub.
7158 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7159 stubAddr, stubName, TypePtr::BOTTOM,
7160 src_start, dest_start, k_start);
7161
7162 return true;
7163 }
7164
7165 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7166 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7167 address stubAddr = nullptr;
7168 const char *stubName = nullptr;
7169
7170 assert(UseAES, "need AES instruction support");
7171
7172 switch(id) {
7173 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7174 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7175 stubName = "cipherBlockChaining_encryptAESCrypt";
7176 break;
7177 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7178 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7179 stubName = "cipherBlockChaining_decryptAESCrypt";
7180 break;
7181 default:
7182 break;
7183 }
7184 if (stubAddr == nullptr) return false;
7185
7186 Node* cipherBlockChaining_object = argument(0);
7187 Node* src = argument(1);
7188 Node* src_offset = argument(2);
7189 Node* len = argument(3);
7190 Node* dest = argument(4);
7191 Node* dest_offset = argument(5);
7192
7193 src = must_be_not_null(src, false);
7194 dest = must_be_not_null(dest, false);
7195
7196 // (1) src and dest are arrays.
7197 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7198 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7199 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7200 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7201
7202 // checks are the responsibility of the caller
7203 Node* src_start = src;
7204 Node* dest_start = dest;
7205 if (src_offset != nullptr || dest_offset != nullptr) {
7206 assert(src_offset != nullptr && dest_offset != nullptr, "");
7207 src_start = array_element_address(src, src_offset, T_BYTE);
7208 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7209 }
7210
7211 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7212 // (because of the predicated logic executed earlier).
7213 // so we cast it here safely.
7214 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7215
7216 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7217 if (embeddedCipherObj == nullptr) return false;
7218
7219 // cast it to what we know it will be at runtime
7220 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7221 assert(tinst != nullptr, "CBC obj is null");
7222 assert(tinst->is_loaded(), "CBC obj is not loaded");
7223 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7224 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7225
7226 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7227 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7228 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7229 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7230 aescrypt_object = _gvn.transform(aescrypt_object);
7231
7232 // we need to get the start of the aescrypt_object's expanded key array
7233 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7234 if (k_start == nullptr) return false;
7235
7236 // similarly, get the start address of the r vector
7237 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7238 if (objRvec == nullptr) return false;
7239 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7240
7241 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7242 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7243 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7244 stubAddr, stubName, TypePtr::BOTTOM,
7245 src_start, dest_start, k_start, r_start, len);
7246
7247 // return cipher length (int)
7248 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7249 set_result(retvalue);
7250 return true;
7251 }
7252
7253 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7254 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7255 address stubAddr = nullptr;
7256 const char *stubName = nullptr;
7257
7258 assert(UseAES, "need AES instruction support");
7259
7260 switch (id) {
7261 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7262 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7263 stubName = "electronicCodeBook_encryptAESCrypt";
7264 break;
7265 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7266 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7267 stubName = "electronicCodeBook_decryptAESCrypt";
7268 break;
7269 default:
7270 break;
7271 }
7272
7273 if (stubAddr == nullptr) return false;
7274
7275 Node* electronicCodeBook_object = argument(0);
7276 Node* src = argument(1);
7277 Node* src_offset = argument(2);
7278 Node* len = argument(3);
7279 Node* dest = argument(4);
7280 Node* dest_offset = argument(5);
7281
7282 // (1) src and dest are arrays.
7283 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7284 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7285 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7286 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7287
7288 // checks are the responsibility of the caller
7289 Node* src_start = src;
7290 Node* dest_start = dest;
7291 if (src_offset != nullptr || dest_offset != nullptr) {
7292 assert(src_offset != nullptr && dest_offset != nullptr, "");
7293 src_start = array_element_address(src, src_offset, T_BYTE);
7294 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7295 }
7296
7297 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7298 // (because of the predicated logic executed earlier).
7299 // so we cast it here safely.
7300 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7301
7302 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7303 if (embeddedCipherObj == nullptr) return false;
7304
7305 // cast it to what we know it will be at runtime
7306 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7307 assert(tinst != nullptr, "ECB obj is null");
7308 assert(tinst->is_loaded(), "ECB obj is not loaded");
7309 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7310 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7311
7312 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7313 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7314 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7315 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7316 aescrypt_object = _gvn.transform(aescrypt_object);
7317
7318 // we need to get the start of the aescrypt_object's expanded key array
7319 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7320 if (k_start == nullptr) return false;
7321
7322 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7323 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7324 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7325 stubAddr, stubName, TypePtr::BOTTOM,
7326 src_start, dest_start, k_start, len);
7327
7328 // return cipher length (int)
7329 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7330 set_result(retvalue);
7331 return true;
7332 }
7333
7334 //------------------------------inline_counterMode_AESCrypt-----------------------
7335 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7336 assert(UseAES, "need AES instruction support");
7337 if (!UseAESCTRIntrinsics) return false;
7338
7339 address stubAddr = nullptr;
7340 const char *stubName = nullptr;
7341 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7342 stubAddr = StubRoutines::counterMode_AESCrypt();
7343 stubName = "counterMode_AESCrypt";
7344 }
7345 if (stubAddr == nullptr) return false;
7346
7347 Node* counterMode_object = argument(0);
7348 Node* src = argument(1);
7349 Node* src_offset = argument(2);
7350 Node* len = argument(3);
7351 Node* dest = argument(4);
7352 Node* dest_offset = argument(5);
7353
7354 // (1) src and dest are arrays.
7355 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7356 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7357 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7358 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7359
7360 // checks are the responsibility of the caller
7361 Node* src_start = src;
7362 Node* dest_start = dest;
7363 if (src_offset != nullptr || dest_offset != nullptr) {
7364 assert(src_offset != nullptr && dest_offset != nullptr, "");
7365 src_start = array_element_address(src, src_offset, T_BYTE);
7366 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7367 }
7368
7369 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7370 // (because of the predicated logic executed earlier).
7371 // so we cast it here safely.
7372 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7373 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7374 if (embeddedCipherObj == nullptr) return false;
7375 // cast it to what we know it will be at runtime
7376 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7377 assert(tinst != nullptr, "CTR obj is null");
7378 assert(tinst->is_loaded(), "CTR obj is not loaded");
7379 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7380 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7381 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7382 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7383 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7384 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7385 aescrypt_object = _gvn.transform(aescrypt_object);
7386 // we need to get the start of the aescrypt_object's expanded key array
7387 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7388 if (k_start == nullptr) return false;
7389 // similarly, get the start address of the r vector
7390 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7391 if (obj_counter == nullptr) return false;
7392 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7393
7394 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7395 if (saved_encCounter == nullptr) return false;
7396 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7397 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7398
7399 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7400 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7401 OptoRuntime::counterMode_aescrypt_Type(),
7402 stubAddr, stubName, TypePtr::BOTTOM,
7403 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7404
7405 // return cipher length (int)
7406 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7407 set_result(retvalue);
7408 return true;
7409 }
7410
7411 //------------------------------get_key_start_from_aescrypt_object-----------------------
7412 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7413 #if defined(PPC64) || defined(S390)
7414 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7415 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7416 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7417 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7418 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7419 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7420 if (objSessionK == nullptr) {
7421 return (Node *) nullptr;
7422 }
7423 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7424 #else
7425 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7426 #endif // PPC64
7427 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7428 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7429
7430 // now have the array, need to get the start address of the K array
7431 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7432 return k_start;
7433 }
7434
7435 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7436 // Return node representing slow path of predicate check.
7437 // the pseudo code we want to emulate with this predicate is:
7438 // for encryption:
7439 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7440 // for decryption:
7441 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7442 // note cipher==plain is more conservative than the original java code but that's OK
7443 //
7444 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7445 // The receiver was checked for null already.
7446 Node* objCBC = argument(0);
7447
7448 Node* src = argument(1);
7449 Node* dest = argument(4);
7450
7451 // Load embeddedCipher field of CipherBlockChaining object.
7452 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7453
7454 // get AESCrypt klass for instanceOf check
7455 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7456 // will have same classloader as CipherBlockChaining object
7457 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7458 assert(tinst != nullptr, "CBCobj is null");
7459 assert(tinst->is_loaded(), "CBCobj is not loaded");
7460
7461 // we want to do an instanceof comparison against the AESCrypt class
7462 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7463 if (!klass_AESCrypt->is_loaded()) {
7464 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7465 Node* ctrl = control();
7466 set_control(top()); // no regular fast path
7467 return ctrl;
7468 }
7469
7470 src = must_be_not_null(src, true);
7471 dest = must_be_not_null(dest, true);
7472
7473 // Resolve oops to stable for CmpP below.
7474 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7475
7476 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7477 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7478 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7479
7480 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7481
7482 // for encryption, we are done
7483 if (!decrypting)
7484 return instof_false; // even if it is null
7485
7486 // for decryption, we need to add a further check to avoid
7487 // taking the intrinsic path when cipher and plain are the same
7488 // see the original java code for why.
7489 RegionNode* region = new RegionNode(3);
7490 region->init_req(1, instof_false);
7491
7492 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7493 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7494 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7495 region->init_req(2, src_dest_conjoint);
7496
7497 record_for_igvn(region);
7498 return _gvn.transform(region);
7499 }
7500
7501 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7502 // Return node representing slow path of predicate check.
7503 // the pseudo code we want to emulate with this predicate is:
7504 // for encryption:
7505 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7506 // for decryption:
7507 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7508 // note cipher==plain is more conservative than the original java code but that's OK
7509 //
7510 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7511 // The receiver was checked for null already.
7512 Node* objECB = argument(0);
7513
7514 // Load embeddedCipher field of ElectronicCodeBook object.
7515 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7516
7517 // get AESCrypt klass for instanceOf check
7518 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7519 // will have same classloader as ElectronicCodeBook object
7520 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7521 assert(tinst != nullptr, "ECBobj is null");
7522 assert(tinst->is_loaded(), "ECBobj is not loaded");
7523
7524 // we want to do an instanceof comparison against the AESCrypt class
7525 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7526 if (!klass_AESCrypt->is_loaded()) {
7527 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7528 Node* ctrl = control();
7529 set_control(top()); // no regular fast path
7530 return ctrl;
7531 }
7532 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7533
7534 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7535 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7536 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7537
7538 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7539
7540 // for encryption, we are done
7541 if (!decrypting)
7542 return instof_false; // even if it is null
7543
7544 // for decryption, we need to add a further check to avoid
7545 // taking the intrinsic path when cipher and plain are the same
7546 // see the original java code for why.
7547 RegionNode* region = new RegionNode(3);
7548 region->init_req(1, instof_false);
7549 Node* src = argument(1);
7550 Node* dest = argument(4);
7551 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7552 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7553 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7554 region->init_req(2, src_dest_conjoint);
7555
7556 record_for_igvn(region);
7557 return _gvn.transform(region);
7558 }
7559
7560 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7561 // Return node representing slow path of predicate check.
7562 // the pseudo code we want to emulate with this predicate is:
7563 // for encryption:
7564 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7565 // for decryption:
7566 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7567 // note cipher==plain is more conservative than the original java code but that's OK
7568 //
7569
7570 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7571 // The receiver was checked for null already.
7572 Node* objCTR = argument(0);
7573
7574 // Load embeddedCipher field of CipherBlockChaining object.
7575 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7576
7577 // get AESCrypt klass for instanceOf check
7578 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7579 // will have same classloader as CipherBlockChaining object
7580 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7581 assert(tinst != nullptr, "CTRobj is null");
7582 assert(tinst->is_loaded(), "CTRobj is not loaded");
7583
7584 // we want to do an instanceof comparison against the AESCrypt class
7585 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7586 if (!klass_AESCrypt->is_loaded()) {
7587 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7588 Node* ctrl = control();
7589 set_control(top()); // no regular fast path
7590 return ctrl;
7591 }
7592
7593 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7594 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7595 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7596 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7597 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7598
7599 return instof_false; // even if it is null
7600 }
7601
7602 //------------------------------inline_ghash_processBlocks
7603 bool LibraryCallKit::inline_ghash_processBlocks() {
7604 address stubAddr;
7605 const char *stubName;
7606 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7607
7608 stubAddr = StubRoutines::ghash_processBlocks();
7609 stubName = "ghash_processBlocks";
7610
7611 Node* data = argument(0);
7612 Node* offset = argument(1);
7613 Node* len = argument(2);
7614 Node* state = argument(3);
7615 Node* subkeyH = argument(4);
7616
7617 state = must_be_not_null(state, true);
7618 subkeyH = must_be_not_null(subkeyH, true);
7619 data = must_be_not_null(data, true);
7620
7621 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7622 assert(state_start, "state is null");
7623 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7624 assert(subkeyH_start, "subkeyH is null");
7625 Node* data_start = array_element_address(data, offset, T_BYTE);
7626 assert(data_start, "data is null");
7627
7628 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7629 OptoRuntime::ghash_processBlocks_Type(),
7630 stubAddr, stubName, TypePtr::BOTTOM,
7631 state_start, subkeyH_start, data_start, len);
7632 return true;
7633 }
7634
7635 //------------------------------inline_chacha20Block
7636 bool LibraryCallKit::inline_chacha20Block() {
7637 address stubAddr;
7638 const char *stubName;
7639 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7640
7641 stubAddr = StubRoutines::chacha20Block();
7642 stubName = "chacha20Block";
7643
7644 Node* state = argument(0);
7645 Node* result = argument(1);
7646
7647 state = must_be_not_null(state, true);
7648 result = must_be_not_null(result, true);
7649
7650 Node* state_start = array_element_address(state, intcon(0), T_INT);
7651 assert(state_start, "state is null");
7652 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7653 assert(result_start, "result is null");
7654
7655 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7656 OptoRuntime::chacha20Block_Type(),
7657 stubAddr, stubName, TypePtr::BOTTOM,
7658 state_start, result_start);
7659 // return key stream length (int)
7660 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7661 set_result(retvalue);
7662 return true;
7663 }
7664
7665 bool LibraryCallKit::inline_base64_encodeBlock() {
7666 address stubAddr;
7667 const char *stubName;
7668 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7669 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
7670 stubAddr = StubRoutines::base64_encodeBlock();
7671 stubName = "encodeBlock";
7672
7673 if (!stubAddr) return false;
7674 Node* base64obj = argument(0);
7675 Node* src = argument(1);
7676 Node* offset = argument(2);
7677 Node* len = argument(3);
7678 Node* dest = argument(4);
7679 Node* dp = argument(5);
7680 Node* isURL = argument(6);
7681
7682 src = must_be_not_null(src, true);
7683 dest = must_be_not_null(dest, true);
7684
7685 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7686 assert(src_start, "source array is null");
7687 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7688 assert(dest_start, "destination array is null");
7689
7690 Node* base64 = make_runtime_call(RC_LEAF,
7691 OptoRuntime::base64_encodeBlock_Type(),
7692 stubAddr, stubName, TypePtr::BOTTOM,
7693 src_start, offset, len, dest_start, dp, isURL);
7694 return true;
7695 }
7696
7697 bool LibraryCallKit::inline_base64_decodeBlock() {
7698 address stubAddr;
7699 const char *stubName;
7700 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7701 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
7702 stubAddr = StubRoutines::base64_decodeBlock();
7703 stubName = "decodeBlock";
7704
7705 if (!stubAddr) return false;
7706 Node* base64obj = argument(0);
7707 Node* src = argument(1);
7708 Node* src_offset = argument(2);
7709 Node* len = argument(3);
7710 Node* dest = argument(4);
7711 Node* dest_offset = argument(5);
7712 Node* isURL = argument(6);
7713 Node* isMIME = argument(7);
7714
7715 src = must_be_not_null(src, true);
7716 dest = must_be_not_null(dest, true);
7717
7718 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7719 assert(src_start, "source array is null");
7720 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7721 assert(dest_start, "destination array is null");
7722
7723 Node* call = make_runtime_call(RC_LEAF,
7724 OptoRuntime::base64_decodeBlock_Type(),
7725 stubAddr, stubName, TypePtr::BOTTOM,
7726 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
7727 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7728 set_result(result);
7729 return true;
7730 }
7731
7732 bool LibraryCallKit::inline_poly1305_processBlocks() {
7733 address stubAddr;
7734 const char *stubName;
7735 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
7736 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
7737 stubAddr = StubRoutines::poly1305_processBlocks();
7738 stubName = "poly1305_processBlocks";
7739
7740 if (!stubAddr) return false;
7741 null_check_receiver(); // null-check receiver
7742 if (stopped()) return true;
7743
7744 Node* input = argument(1);
7745 Node* input_offset = argument(2);
7746 Node* len = argument(3);
7747 Node* alimbs = argument(4);
7748 Node* rlimbs = argument(5);
7749
7750 input = must_be_not_null(input, true);
7751 alimbs = must_be_not_null(alimbs, true);
7752 rlimbs = must_be_not_null(rlimbs, true);
7753
7754 Node* input_start = array_element_address(input, input_offset, T_BYTE);
7755 assert(input_start, "input array is null");
7756 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
7757 assert(acc_start, "acc array is null");
7758 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
7759 assert(r_start, "r array is null");
7760
7761 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7762 OptoRuntime::poly1305_processBlocks_Type(),
7763 stubAddr, stubName, TypePtr::BOTTOM,
7764 input_start, len, acc_start, r_start);
7765 return true;
7766 }
7767
7768 //------------------------------inline_digestBase_implCompress-----------------------
7769 //
7770 // Calculate MD5 for single-block byte[] array.
7771 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
7772 //
7773 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
7774 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
7775 //
7776 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
7777 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
7778 //
7779 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
7780 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
7781 //
7782 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
7783 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
7784 //
7785 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
7786 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
7787
7788 Node* digestBase_obj = argument(0);
7789 Node* src = argument(1); // type oop
7790 Node* ofs = argument(2); // type int
7791
7792 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7793 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7794 // failed array check
7795 return false;
7796 }
7797 // Figure out the size and type of the elements we will be copying.
7798 BasicType src_elem = src_type->elem()->array_element_basic_type();
7799 if (src_elem != T_BYTE) {
7800 return false;
7801 }
7802 // 'src_start' points to src array + offset
7803 src = must_be_not_null(src, true);
7804 Node* src_start = array_element_address(src, ofs, src_elem);
7805 Node* state = nullptr;
7806 Node* block_size = nullptr;
7807 address stubAddr;
7808 const char *stubName;
7809
7810 switch(id) {
7811 case vmIntrinsics::_md5_implCompress:
7812 assert(UseMD5Intrinsics, "need MD5 instruction support");
7813 state = get_state_from_digest_object(digestBase_obj, T_INT);
7814 stubAddr = StubRoutines::md5_implCompress();
7815 stubName = "md5_implCompress";
7816 break;
7817 case vmIntrinsics::_sha_implCompress:
7818 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
7819 state = get_state_from_digest_object(digestBase_obj, T_INT);
7820 stubAddr = StubRoutines::sha1_implCompress();
7821 stubName = "sha1_implCompress";
7822 break;
7823 case vmIntrinsics::_sha2_implCompress:
7824 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
7825 state = get_state_from_digest_object(digestBase_obj, T_INT);
7826 stubAddr = StubRoutines::sha256_implCompress();
7827 stubName = "sha256_implCompress";
7828 break;
7829 case vmIntrinsics::_sha5_implCompress:
7830 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
7831 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7832 stubAddr = StubRoutines::sha512_implCompress();
7833 stubName = "sha512_implCompress";
7834 break;
7835 case vmIntrinsics::_sha3_implCompress:
7836 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
7837 state = get_state_from_digest_object(digestBase_obj, T_BYTE);
7838 stubAddr = StubRoutines::sha3_implCompress();
7839 stubName = "sha3_implCompress";
7840 block_size = get_block_size_from_digest_object(digestBase_obj);
7841 if (block_size == nullptr) return false;
7842 break;
7843 default:
7844 fatal_unexpected_iid(id);
7845 return false;
7846 }
7847 if (state == nullptr) return false;
7848
7849 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
7850 if (stubAddr == nullptr) return false;
7851
7852 // Call the stub.
7853 Node* call;
7854 if (block_size == nullptr) {
7855 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
7856 stubAddr, stubName, TypePtr::BOTTOM,
7857 src_start, state);
7858 } else {
7859 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
7860 stubAddr, stubName, TypePtr::BOTTOM,
7861 src_start, state, block_size);
7862 }
7863
7864 return true;
7865 }
7866
7867 //------------------------------inline_digestBase_implCompressMB-----------------------
7868 //
7869 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
7870 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
7871 //
7872 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
7873 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7874 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7875 assert((uint)predicate < 5, "sanity");
7876 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7877
7878 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
7879 Node* src = argument(1); // byte[] array
7880 Node* ofs = argument(2); // type int
7881 Node* limit = argument(3); // type int
7882
7883 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7884 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7885 // failed array check
7886 return false;
7887 }
7888 // Figure out the size and type of the elements we will be copying.
7889 BasicType src_elem = src_type->elem()->array_element_basic_type();
7890 if (src_elem != T_BYTE) {
7891 return false;
7892 }
7893 // 'src_start' points to src array + offset
7894 src = must_be_not_null(src, false);
7895 Node* src_start = array_element_address(src, ofs, src_elem);
7896
7897 const char* klass_digestBase_name = nullptr;
7898 const char* stub_name = nullptr;
7899 address stub_addr = nullptr;
7900 BasicType elem_type = T_INT;
7901
7902 switch (predicate) {
7903 case 0:
7904 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
7905 klass_digestBase_name = "sun/security/provider/MD5";
7906 stub_name = "md5_implCompressMB";
7907 stub_addr = StubRoutines::md5_implCompressMB();
7908 }
7909 break;
7910 case 1:
7911 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
7912 klass_digestBase_name = "sun/security/provider/SHA";
7913 stub_name = "sha1_implCompressMB";
7914 stub_addr = StubRoutines::sha1_implCompressMB();
7915 }
7916 break;
7917 case 2:
7918 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
7919 klass_digestBase_name = "sun/security/provider/SHA2";
7920 stub_name = "sha256_implCompressMB";
7921 stub_addr = StubRoutines::sha256_implCompressMB();
7922 }
7923 break;
7924 case 3:
7925 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
7926 klass_digestBase_name = "sun/security/provider/SHA5";
7927 stub_name = "sha512_implCompressMB";
7928 stub_addr = StubRoutines::sha512_implCompressMB();
7929 elem_type = T_LONG;
7930 }
7931 break;
7932 case 4:
7933 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
7934 klass_digestBase_name = "sun/security/provider/SHA3";
7935 stub_name = "sha3_implCompressMB";
7936 stub_addr = StubRoutines::sha3_implCompressMB();
7937 elem_type = T_BYTE;
7938 }
7939 break;
7940 default:
7941 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
7942 }
7943 if (klass_digestBase_name != nullptr) {
7944 assert(stub_addr != nullptr, "Stub is generated");
7945 if (stub_addr == nullptr) return false;
7946
7947 // get DigestBase klass to lookup for SHA klass
7948 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7949 assert(tinst != nullptr, "digestBase_obj is not instance???");
7950 assert(tinst->is_loaded(), "DigestBase is not loaded");
7951
7952 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
7953 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
7954 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
7955 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
7956 }
7957 return false;
7958 }
7959
7960 //------------------------------inline_digestBase_implCompressMB-----------------------
7961 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
7962 BasicType elem_type, address stubAddr, const char *stubName,
7963 Node* src_start, Node* ofs, Node* limit) {
7964 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
7965 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7966 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
7967 digest_obj = _gvn.transform(digest_obj);
7968
7969 Node* state = get_state_from_digest_object(digest_obj, elem_type);
7970 if (state == nullptr) return false;
7971
7972 Node* block_size = nullptr;
7973 if (strcmp("sha3_implCompressMB", stubName) == 0) {
7974 block_size = get_block_size_from_digest_object(digest_obj);
7975 if (block_size == nullptr) return false;
7976 }
7977
7978 // Call the stub.
7979 Node* call;
7980 if (block_size == nullptr) {
7981 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7982 OptoRuntime::digestBase_implCompressMB_Type(false),
7983 stubAddr, stubName, TypePtr::BOTTOM,
7984 src_start, state, ofs, limit);
7985 } else {
7986 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7987 OptoRuntime::digestBase_implCompressMB_Type(true),
7988 stubAddr, stubName, TypePtr::BOTTOM,
7989 src_start, state, block_size, ofs, limit);
7990 }
7991
7992 // return ofs (int)
7993 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7994 set_result(result);
7995
7996 return true;
7997 }
7998
7999 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8000 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8001 assert(UseAES, "need AES instruction support");
8002 address stubAddr = nullptr;
8003 const char *stubName = nullptr;
8004 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8005 stubName = "galoisCounterMode_AESCrypt";
8006
8007 if (stubAddr == nullptr) return false;
8008
8009 Node* in = argument(0);
8010 Node* inOfs = argument(1);
8011 Node* len = argument(2);
8012 Node* ct = argument(3);
8013 Node* ctOfs = argument(4);
8014 Node* out = argument(5);
8015 Node* outOfs = argument(6);
8016 Node* gctr_object = argument(7);
8017 Node* ghash_object = argument(8);
8018
8019 // (1) in, ct and out are arrays.
8020 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8021 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8022 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8023 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8024 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8025 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8026
8027 // checks are the responsibility of the caller
8028 Node* in_start = in;
8029 Node* ct_start = ct;
8030 Node* out_start = out;
8031 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8032 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8033 in_start = array_element_address(in, inOfs, T_BYTE);
8034 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8035 out_start = array_element_address(out, outOfs, T_BYTE);
8036 }
8037
8038 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8039 // (because of the predicated logic executed earlier).
8040 // so we cast it here safely.
8041 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8042 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8043 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8044 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8045 Node* state = load_field_from_object(ghash_object, "state", "[J");
8046
8047 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8048 return false;
8049 }
8050 // cast it to what we know it will be at runtime
8051 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8052 assert(tinst != nullptr, "GCTR obj is null");
8053 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8054 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8055 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8056 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8057 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8058 const TypeOopPtr* xtype = aklass->as_instance_type();
8059 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8060 aescrypt_object = _gvn.transform(aescrypt_object);
8061 // we need to get the start of the aescrypt_object's expanded key array
8062 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8063 if (k_start == nullptr) return false;
8064 // similarly, get the start address of the r vector
8065 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8066 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8067 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8068
8069
8070 // Call the stub, passing params
8071 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8072 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8073 stubAddr, stubName, TypePtr::BOTTOM,
8074 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8075
8076 // return cipher length (int)
8077 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8078 set_result(retvalue);
8079
8080 return true;
8081 }
8082
8083 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8084 // Return node representing slow path of predicate check.
8085 // the pseudo code we want to emulate with this predicate is:
8086 // for encryption:
8087 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8088 // for decryption:
8089 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8090 // note cipher==plain is more conservative than the original java code but that's OK
8091 //
8092
8093 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8094 // The receiver was checked for null already.
8095 Node* objGCTR = argument(7);
8096 // Load embeddedCipher field of GCTR object.
8097 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8098 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8099
8100 // get AESCrypt klass for instanceOf check
8101 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8102 // will have same classloader as CipherBlockChaining object
8103 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8104 assert(tinst != nullptr, "GCTR obj is null");
8105 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8106
8107 // we want to do an instanceof comparison against the AESCrypt class
8108 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8109 if (!klass_AESCrypt->is_loaded()) {
8110 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8111 Node* ctrl = control();
8112 set_control(top()); // no regular fast path
8113 return ctrl;
8114 }
8115
8116 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8117 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8118 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8119 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8120 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8121
8122 return instof_false; // even if it is null
8123 }
8124
8125 //------------------------------get_state_from_digest_object-----------------------
8126 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8127 const char* state_type;
8128 switch (elem_type) {
8129 case T_BYTE: state_type = "[B"; break;
8130 case T_INT: state_type = "[I"; break;
8131 case T_LONG: state_type = "[J"; break;
8132 default: ShouldNotReachHere();
8133 }
8134 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8135 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8136 if (digest_state == nullptr) return (Node *) nullptr;
8137
8138 // now have the array, need to get the start address of the state array
8139 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8140 return state;
8141 }
8142
8143 //------------------------------get_block_size_from_sha3_object----------------------------------
8144 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8145 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8146 assert (block_size != nullptr, "sanity");
8147 return block_size;
8148 }
8149
8150 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8151 // Return node representing slow path of predicate check.
8152 // the pseudo code we want to emulate with this predicate is:
8153 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8154 //
8155 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8156 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8157 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8158 assert((uint)predicate < 5, "sanity");
8159
8160 // The receiver was checked for null already.
8161 Node* digestBaseObj = argument(0);
8162
8163 // get DigestBase klass for instanceOf check
8164 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8165 assert(tinst != nullptr, "digestBaseObj is null");
8166 assert(tinst->is_loaded(), "DigestBase is not loaded");
8167
8168 const char* klass_name = nullptr;
8169 switch (predicate) {
8170 case 0:
8171 if (UseMD5Intrinsics) {
8172 // we want to do an instanceof comparison against the MD5 class
8173 klass_name = "sun/security/provider/MD5";
8174 }
8175 break;
8176 case 1:
8177 if (UseSHA1Intrinsics) {
8178 // we want to do an instanceof comparison against the SHA class
8179 klass_name = "sun/security/provider/SHA";
8180 }
8181 break;
8182 case 2:
8183 if (UseSHA256Intrinsics) {
8184 // we want to do an instanceof comparison against the SHA2 class
8185 klass_name = "sun/security/provider/SHA2";
8186 }
8187 break;
8188 case 3:
8189 if (UseSHA512Intrinsics) {
8190 // we want to do an instanceof comparison against the SHA5 class
8191 klass_name = "sun/security/provider/SHA5";
8192 }
8193 break;
8194 case 4:
8195 if (UseSHA3Intrinsics) {
8196 // we want to do an instanceof comparison against the SHA3 class
8197 klass_name = "sun/security/provider/SHA3";
8198 }
8199 break;
8200 default:
8201 fatal("unknown SHA intrinsic predicate: %d", predicate);
8202 }
8203
8204 ciKlass* klass = nullptr;
8205 if (klass_name != nullptr) {
8206 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8207 }
8208 if ((klass == nullptr) || !klass->is_loaded()) {
8209 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8210 Node* ctrl = control();
8211 set_control(top()); // no intrinsic path
8212 return ctrl;
8213 }
8214 ciInstanceKlass* instklass = klass->as_instance_klass();
8215
8216 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8217 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8218 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8219 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8220
8221 return instof_false; // even if it is null
8222 }
8223
8224 //-------------inline_fma-----------------------------------
8225 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8226 Node *a = nullptr;
8227 Node *b = nullptr;
8228 Node *c = nullptr;
8229 Node* result = nullptr;
8230 switch (id) {
8231 case vmIntrinsics::_fmaD:
8232 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8233 // no receiver since it is static method
8234 a = round_double_node(argument(0));
8235 b = round_double_node(argument(2));
8236 c = round_double_node(argument(4));
8237 result = _gvn.transform(new FmaDNode(control(), a, b, c));
8238 break;
8239 case vmIntrinsics::_fmaF:
8240 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8241 a = argument(0);
8242 b = argument(1);
8243 c = argument(2);
8244 result = _gvn.transform(new FmaFNode(control(), a, b, c));
8245 break;
8246 default:
8247 fatal_unexpected_iid(id); break;
8248 }
8249 set_result(result);
8250 return true;
8251 }
8252
8253 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8254 // argument(0) is receiver
8255 Node* codePoint = argument(1);
8256 Node* n = nullptr;
8257
8258 switch (id) {
8259 case vmIntrinsics::_isDigit :
8260 n = new DigitNode(control(), codePoint);
8261 break;
8262 case vmIntrinsics::_isLowerCase :
8263 n = new LowerCaseNode(control(), codePoint);
8264 break;
8265 case vmIntrinsics::_isUpperCase :
8266 n = new UpperCaseNode(control(), codePoint);
8267 break;
8268 case vmIntrinsics::_isWhitespace :
8269 n = new WhitespaceNode(control(), codePoint);
8270 break;
8271 default:
8272 fatal_unexpected_iid(id);
8273 }
8274
8275 set_result(_gvn.transform(n));
8276 return true;
8277 }
8278
8279 //------------------------------inline_fp_min_max------------------------------
8280 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
8281 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
8282
8283 // The intrinsic should be used only when the API branches aren't predictable,
8284 // the last one performing the most important comparison. The following heuristic
8285 // uses the branch statistics to eventually bail out if necessary.
8286
8287 ciMethodData *md = callee()->method_data();
8288
8289 if ( md != nullptr && md->is_mature() && md->invocation_count() > 0 ) {
8290 ciCallProfile cp = caller()->call_profile_at_bci(bci());
8291
8292 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
8293 // Bail out if the call-site didn't contribute enough to the statistics.
8294 return false;
8295 }
8296
8297 uint taken = 0, not_taken = 0;
8298
8299 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
8300 if (p->is_BranchData()) {
8301 taken = ((ciBranchData*)p)->taken();
8302 not_taken = ((ciBranchData*)p)->not_taken();
8303 }
8304 }
8305
8306 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
8307 balance = balance < 0 ? -balance : balance;
8308 if ( balance > 0.2 ) {
8309 // Bail out if the most important branch is predictable enough.
8310 return false;
8311 }
8312 }
8313 */
8314
8315 Node *a = nullptr;
8316 Node *b = nullptr;
8317 Node *n = nullptr;
8318 switch (id) {
8319 case vmIntrinsics::_maxF:
8320 case vmIntrinsics::_minF:
8321 case vmIntrinsics::_maxF_strict:
8322 case vmIntrinsics::_minF_strict:
8323 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
8324 a = argument(0);
8325 b = argument(1);
8326 break;
8327 case vmIntrinsics::_maxD:
8328 case vmIntrinsics::_minD:
8329 case vmIntrinsics::_maxD_strict:
8330 case vmIntrinsics::_minD_strict:
8331 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
8332 a = round_double_node(argument(0));
8333 b = round_double_node(argument(2));
8334 break;
8335 default:
8336 fatal_unexpected_iid(id);
8337 break;
8338 }
8339 switch (id) {
8340 case vmIntrinsics::_maxF:
8341 case vmIntrinsics::_maxF_strict:
8342 n = new MaxFNode(a, b);
8343 break;
8344 case vmIntrinsics::_minF:
8345 case vmIntrinsics::_minF_strict:
8346 n = new MinFNode(a, b);
8347 break;
8348 case vmIntrinsics::_maxD:
8349 case vmIntrinsics::_maxD_strict:
8350 n = new MaxDNode(a, b);
8351 break;
8352 case vmIntrinsics::_minD:
8353 case vmIntrinsics::_minD_strict:
8354 n = new MinDNode(a, b);
8355 break;
8356 default:
8357 fatal_unexpected_iid(id);
8358 break;
8359 }
8360 set_result(_gvn.transform(n));
8361 return true;
8362 }
8363
8364 bool LibraryCallKit::inline_profileBoolean() {
8365 Node* counts = argument(1);
8366 const TypeAryPtr* ary = nullptr;
8367 ciArray* aobj = nullptr;
8368 if (counts->is_Con()
8369 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8370 && (aobj = ary->const_oop()->as_array()) != nullptr
8371 && (aobj->length() == 2)) {
8372 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8373 jint false_cnt = aobj->element_value(0).as_int();
8374 jint true_cnt = aobj->element_value(1).as_int();
8375
8376 if (C->log() != nullptr) {
8377 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8378 false_cnt, true_cnt);
8379 }
8380
8381 if (false_cnt + true_cnt == 0) {
8382 // According to profile, never executed.
8383 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8384 Deoptimization::Action_reinterpret);
8385 return true;
8386 }
8387
8388 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8389 // is a number of each value occurrences.
8390 Node* result = argument(0);
8391 if (false_cnt == 0 || true_cnt == 0) {
8392 // According to profile, one value has been never seen.
8393 int expected_val = (false_cnt == 0) ? 1 : 0;
8394
8395 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8396 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8397
8398 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8399 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8400 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8401
8402 { // Slow path: uncommon trap for never seen value and then reexecute
8403 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8404 // the value has been seen at least once.
8405 PreserveJVMState pjvms(this);
8406 PreserveReexecuteState preexecs(this);
8407 jvms()->set_should_reexecute(true);
8408
8409 set_control(slow_path);
8410 set_i_o(i_o());
8411
8412 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8413 Deoptimization::Action_reinterpret);
8414 }
8415 // The guard for never seen value enables sharpening of the result and
8416 // returning a constant. It allows to eliminate branches on the same value
8417 // later on.
8418 set_control(fast_path);
8419 result = intcon(expected_val);
8420 }
8421 // Stop profiling.
8422 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8423 // By replacing method body with profile data (represented as ProfileBooleanNode
8424 // on IR level) we effectively disable profiling.
8425 // It enables full speed execution once optimized code is generated.
8426 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8427 C->record_for_igvn(profile);
8428 set_result(profile);
8429 return true;
8430 } else {
8431 // Continue profiling.
8432 // Profile data isn't available at the moment. So, execute method's bytecode version.
8433 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8434 // is compiled and counters aren't available since corresponding MethodHandle
8435 // isn't a compile-time constant.
8436 return false;
8437 }
8438 }
8439
8440 bool LibraryCallKit::inline_isCompileConstant() {
8441 Node* n = argument(0);
8442 set_result(n->is_Con() ? intcon(1) : intcon(0));
8443 return true;
8444 }
8445
8446 //------------------------------- inline_getObjectSize --------------------------------------
8447 //
8448 // Calculate the runtime size of the object/array.
8449 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8450 //
8451 bool LibraryCallKit::inline_getObjectSize() {
8452 Node* obj = argument(3);
8453 Node* klass_node = load_object_klass(obj);
8454
8455 jint layout_con = Klass::_lh_neutral_value;
8456 Node* layout_val = get_layout_helper(klass_node, layout_con);
8457 int layout_is_con = (layout_val == nullptr);
8458
8459 if (layout_is_con) {
8460 // Layout helper is constant, can figure out things at compile time.
8461
8462 if (Klass::layout_helper_is_instance(layout_con)) {
8463 // Instance case: layout_con contains the size itself.
8464 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8465 set_result(size);
8466 } else {
8467 // Array case: size is round(header + element_size*arraylength).
8468 // Since arraylength is different for every array instance, we have to
8469 // compute the whole thing at runtime.
8470
8471 Node* arr_length = load_array_length(obj);
8472
8473 int round_mask = MinObjAlignmentInBytes - 1;
8474 int hsize = Klass::layout_helper_header_size(layout_con);
8475 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8476
8477 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8478 round_mask = 0; // strength-reduce it if it goes away completely
8479 }
8480 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8481 Node* header_size = intcon(hsize + round_mask);
8482
8483 Node* lengthx = ConvI2X(arr_length);
8484 Node* headerx = ConvI2X(header_size);
8485
8486 Node* abody = lengthx;
8487 if (eshift != 0) {
8488 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8489 }
8490 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8491 if (round_mask != 0) {
8492 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8493 }
8494 size = ConvX2L(size);
8495 set_result(size);
8496 }
8497 } else {
8498 // Layout helper is not constant, need to test for array-ness at runtime.
8499
8500 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8501 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8502 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8503 record_for_igvn(result_reg);
8504
8505 Node* array_ctl = generate_array_guard(klass_node, nullptr);
8506 if (array_ctl != nullptr) {
8507 // Array case: size is round(header + element_size*arraylength).
8508 // Since arraylength is different for every array instance, we have to
8509 // compute the whole thing at runtime.
8510
8511 PreserveJVMState pjvms(this);
8512 set_control(array_ctl);
8513 Node* arr_length = load_array_length(obj);
8514
8515 int round_mask = MinObjAlignmentInBytes - 1;
8516 Node* mask = intcon(round_mask);
8517
8518 Node* hss = intcon(Klass::_lh_header_size_shift);
8519 Node* hsm = intcon(Klass::_lh_header_size_mask);
8520 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8521 header_size = _gvn.transform(new AndINode(header_size, hsm));
8522 header_size = _gvn.transform(new AddINode(header_size, mask));
8523
8524 // There is no need to mask or shift this value.
8525 // The semantics of LShiftINode include an implicit mask to 0x1F.
8526 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8527 Node* elem_shift = layout_val;
8528
8529 Node* lengthx = ConvI2X(arr_length);
8530 Node* headerx = ConvI2X(header_size);
8531
8532 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8533 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8534 if (round_mask != 0) {
8535 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8536 }
8537 size = ConvX2L(size);
8538
8539 result_reg->init_req(_array_path, control());
8540 result_val->init_req(_array_path, size);
8541 }
8542
8543 if (!stopped()) {
8544 // Instance case: the layout helper gives us instance size almost directly,
8545 // but we need to mask out the _lh_instance_slow_path_bit.
8546 Node* size = ConvI2X(layout_val);
8547 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8548 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8549 size = _gvn.transform(new AndXNode(size, mask));
8550 size = ConvX2L(size);
8551
8552 result_reg->init_req(_instance_path, control());
8553 result_val->init_req(_instance_path, size);
8554 }
8555
8556 set_result(result_reg, result_val);
8557 }
8558
8559 return true;
8560 }
8561
8562 //------------------------------- inline_blackhole --------------------------------------
8563 //
8564 // Make sure all arguments to this node are alive.
8565 // This matches methods that were requested to be blackholed through compile commands.
8566 //
8567 bool LibraryCallKit::inline_blackhole() {
8568 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8569 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8570 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8571
8572 // Blackhole node pinches only the control, not memory. This allows
8573 // the blackhole to be pinned in the loop that computes blackholed
8574 // values, but have no other side effects, like breaking the optimizations
8575 // across the blackhole.
8576
8577 Node* bh = _gvn.transform(new BlackholeNode(control()));
8578 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
8579
8580 // Bind call arguments as blackhole arguments to keep them alive
8581 uint nargs = callee()->arg_size();
8582 for (uint i = 0; i < nargs; i++) {
8583 bh->add_req(argument(i));
8584 }
8585
8586 return true;
8587 }